* [PATCH v5 0/9] xfs_repair: use btree bulk loading
@ 2020-05-20 1:50 Darrick J. Wong
2020-05-20 1:50 ` [PATCH 1/9] xfs_repair: port the online repair newbt structure Darrick J. Wong
` (8 more replies)
0 siblings, 9 replies; 30+ messages in thread
From: Darrick J. Wong @ 2020-05-20 1:50 UTC (permalink / raw)
To: sandeen, darrick.wong; +Cc: Eric Sandeen, linux-xfs, bfoster
Hi all,
In preparation for landing the online fs repair feature, it was
necessary to design a generic btree bulk loading module that it could
use. Fortunately, xfs_repair has four of these (for the four btree
types), so I synthesized one generic version and pushed it into the
kernel libxfs in 5.7.
That being done, port xfs_repair to use the generic btree bulk loader.
In addition to dropping a lot of code from xfs_repair, this also enables
us to control the fullness of the tree nodes in the rebuilt indices for
testing.
For v5 I rebased the support code from my kernel tree, and fixed some
of the more obvious warts that Brian found in v4.
If you're going to start using this mess, you probably ought to just
pull from my git trees, which are linked below.
This is an extraordinary way to destroy everything. Enjoy!
Comments and questions are, as always, welcome.
--D
xfsprogs git tree:
https://git.kernel.org/cgit/linux/kernel/git/djwong/xfsprogs-dev.git/log/?h=repair-bulk-load
^ permalink raw reply [flat|nested] 30+ messages in thread
* [PATCH 1/9] xfs_repair: port the online repair newbt structure
2020-05-20 1:50 [PATCH v5 0/9] xfs_repair: use btree bulk loading Darrick J. Wong
@ 2020-05-20 1:50 ` Darrick J. Wong
2020-05-27 12:15 ` Brian Foster
2020-05-20 1:50 ` [PATCH 2/9] xfs_repair: remove gratuitous code block in phase5 Darrick J. Wong
` (7 subsequent siblings)
8 siblings, 1 reply; 30+ messages in thread
From: Darrick J. Wong @ 2020-05-20 1:50 UTC (permalink / raw)
To: sandeen, darrick.wong; +Cc: linux-xfs, bfoster
From: Darrick J. Wong <darrick.wong@oracle.com>
Port the new btree staging context and related block reservation helper
code from the kernel to repair. We'll use this in subsequent patches to
implement btree bulk loading.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
---
include/libxfs.h | 1
libxfs/libxfs_api_defs.h | 2
repair/Makefile | 4 -
repair/bload.c | 303 ++++++++++++++++++++++++++++++++++++++++++++++
repair/bload.h | 77 ++++++++++++
repair/xfs_repair.c | 17 +++
6 files changed, 402 insertions(+), 2 deletions(-)
create mode 100644 repair/bload.c
create mode 100644 repair/bload.h
diff --git a/include/libxfs.h b/include/libxfs.h
index 12447835..b9370139 100644
--- a/include/libxfs.h
+++ b/include/libxfs.h
@@ -76,6 +76,7 @@ struct iomap;
#include "xfs_rmap.h"
#include "xfs_refcount_btree.h"
#include "xfs_refcount.h"
+#include "xfs_btree_staging.h"
#ifndef ARRAY_SIZE
#define ARRAY_SIZE(x) (sizeof(x) / sizeof((x)[0]))
diff --git a/libxfs/libxfs_api_defs.h b/libxfs/libxfs_api_defs.h
index be06c763..61047f8f 100644
--- a/libxfs/libxfs_api_defs.h
+++ b/libxfs/libxfs_api_defs.h
@@ -27,12 +27,14 @@
#define xfs_alloc_fix_freelist libxfs_alloc_fix_freelist
#define xfs_alloc_min_freelist libxfs_alloc_min_freelist
#define xfs_alloc_read_agf libxfs_alloc_read_agf
+#define xfs_alloc_vextent libxfs_alloc_vextent
#define xfs_attr_get libxfs_attr_get
#define xfs_attr_leaf_newentsize libxfs_attr_leaf_newentsize
#define xfs_attr_namecheck libxfs_attr_namecheck
#define xfs_attr_set libxfs_attr_set
+#define __xfs_bmap_add_free __libxfs_bmap_add_free
#define xfs_bmapi_read libxfs_bmapi_read
#define xfs_bmapi_write libxfs_bmapi_write
#define xfs_bmap_last_offset libxfs_bmap_last_offset
diff --git a/repair/Makefile b/repair/Makefile
index 0964499a..8cc1ee68 100644
--- a/repair/Makefile
+++ b/repair/Makefile
@@ -9,11 +9,11 @@ LSRCFILES = README
LTCOMMAND = xfs_repair
-HFILES = agheader.h attr_repair.h avl.h bmap.h btree.h \
+HFILES = agheader.h attr_repair.h avl.h bload.h bmap.h btree.h \
da_util.h dinode.h dir2.h err_protos.h globals.h incore.h protos.h \
rt.h progress.h scan.h versions.h prefetch.h rmap.h slab.h threads.h
-CFILES = agheader.c attr_repair.c avl.c bmap.c btree.c \
+CFILES = agheader.c attr_repair.c avl.c bload.c bmap.c btree.c \
da_util.c dino_chunks.c dinode.c dir2.c globals.c incore.c \
incore_bmc.c init.c incore_ext.c incore_ino.c phase1.c \
phase2.c phase3.c phase4.c phase5.c phase6.c phase7.c \
diff --git a/repair/bload.c b/repair/bload.c
new file mode 100644
index 00000000..9bc17468
--- /dev/null
+++ b/repair/bload.c
@@ -0,0 +1,303 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * Copyright (C) 2020 Oracle. All Rights Reserved.
+ * Author: Darrick J. Wong <darrick.wong@oracle.com>
+ */
+#include <libxfs.h>
+#include "bload.h"
+
+#define trace_xrep_newbt_claim_block(...) ((void) 0)
+#define trace_xrep_newbt_free_blocks(...) ((void) 0)
+
+int bload_leaf_slack = -1;
+int bload_node_slack = -1;
+
+/* Ported routines from fs/xfs/scrub/repair.c */
+
+/*
+ * Roll a transaction, keeping the AG headers locked and reinitializing
+ * the btree cursors.
+ */
+int
+xrep_roll_ag_trans(
+ struct repair_ctx *sc)
+{
+ int error;
+
+ /* Keep the AG header buffers locked so we can keep going. */
+ if (sc->agi_bp)
+ libxfs_trans_bhold(sc->tp, sc->agi_bp);
+ if (sc->agf_bp)
+ libxfs_trans_bhold(sc->tp, sc->agf_bp);
+ if (sc->agfl_bp)
+ libxfs_trans_bhold(sc->tp, sc->agfl_bp);
+
+ /*
+ * Roll the transaction. We still own the buffer and the buffer lock
+ * regardless of whether or not the roll succeeds. If the roll fails,
+ * the buffers will be released during teardown on our way out of the
+ * kernel. If it succeeds, we join them to the new transaction and
+ * move on.
+ */
+ error = -libxfs_trans_roll(&sc->tp);
+ if (error)
+ return error;
+
+ /* Join AG headers to the new transaction. */
+ if (sc->agi_bp)
+ libxfs_trans_bjoin(sc->tp, sc->agi_bp);
+ if (sc->agf_bp)
+ libxfs_trans_bjoin(sc->tp, sc->agf_bp);
+ if (sc->agfl_bp)
+ libxfs_trans_bjoin(sc->tp, sc->agfl_bp);
+
+ return 0;
+}
+
+/* Initialize accounting resources for staging a new AG btree. */
+void
+xrep_newbt_init_ag(
+ struct xrep_newbt *xnr,
+ struct repair_ctx *sc,
+ const struct xfs_owner_info *oinfo,
+ xfs_fsblock_t alloc_hint,
+ enum xfs_ag_resv_type resv)
+{
+ memset(xnr, 0, sizeof(struct xrep_newbt));
+ xnr->sc = sc;
+ xnr->oinfo = *oinfo; /* structure copy */
+ xnr->alloc_hint = alloc_hint;
+ xnr->resv = resv;
+ INIT_LIST_HEAD(&xnr->resv_list);
+}
+
+/* Initialize accounting resources for staging a new inode fork btree. */
+void
+xrep_newbt_init_inode(
+ struct xrep_newbt *xnr,
+ struct repair_ctx *sc,
+ int whichfork,
+ const struct xfs_owner_info *oinfo)
+{
+ xrep_newbt_init_ag(xnr, sc, oinfo,
+ XFS_INO_TO_FSB(sc->mp, sc->ip->i_ino),
+ XFS_AG_RESV_NONE);
+ xnr->ifake.if_fork = kmem_zone_zalloc(xfs_ifork_zone, 0);
+ xnr->ifake.if_fork_size = XFS_IFORK_SIZE(sc->ip, whichfork);
+}
+
+/*
+ * Initialize accounting resources for staging a new btree. Callers are
+ * expected to add their own reservations (and clean them up) manually.
+ */
+void
+xrep_newbt_init_bare(
+ struct xrep_newbt *xnr,
+ struct repair_ctx *sc)
+{
+ xrep_newbt_init_ag(xnr, sc, &XFS_RMAP_OINFO_ANY_OWNER, NULLFSBLOCK,
+ XFS_AG_RESV_NONE);
+}
+
+/* Designate specific blocks to be used to build our new btree. */
+int
+xrep_newbt_add_blocks(
+ struct xrep_newbt *xnr,
+ xfs_fsblock_t fsbno,
+ xfs_extlen_t len)
+{
+ struct xrep_newbt_resv *resv;
+
+ resv = kmem_alloc(sizeof(struct xrep_newbt_resv), KM_MAYFAIL);
+ if (!resv)
+ return ENOMEM;
+
+ INIT_LIST_HEAD(&resv->list);
+ resv->fsbno = fsbno;
+ resv->len = len;
+ resv->used = 0;
+ list_add_tail(&resv->list, &xnr->resv_list);
+ return 0;
+}
+
+/* Reserve disk space for our new btree. */
+int
+xrep_newbt_alloc_blocks(
+ struct xrep_newbt *xnr,
+ uint64_t nr_blocks)
+{
+ struct repair_ctx *sc = xnr->sc;
+ xfs_alloctype_t type;
+ xfs_fsblock_t alloc_hint = xnr->alloc_hint;
+ int error = 0;
+
+ type = sc->ip ? XFS_ALLOCTYPE_START_BNO : XFS_ALLOCTYPE_NEAR_BNO;
+
+ while (nr_blocks > 0 && !error) {
+ struct xfs_alloc_arg args = {
+ .tp = sc->tp,
+ .mp = sc->mp,
+ .type = type,
+ .fsbno = alloc_hint,
+ .oinfo = xnr->oinfo,
+ .minlen = 1,
+ .maxlen = nr_blocks,
+ .prod = 1,
+ .resv = xnr->resv,
+ };
+
+ error = -libxfs_alloc_vextent(&args);
+ if (error)
+ return error;
+ if (args.fsbno == NULLFSBLOCK)
+ return ENOSPC;
+
+ /* We don't have real EFIs here so skip that. */
+
+ error = xrep_newbt_add_blocks(xnr, args.fsbno, args.len);
+ if (error)
+ break;
+
+ nr_blocks -= args.len;
+ alloc_hint = args.fsbno + args.len - 1;
+
+ if (sc->ip)
+ error = -libxfs_trans_roll_inode(&sc->tp, sc->ip);
+ else
+ error = xrep_roll_ag_trans(sc);
+ }
+
+ return error;
+}
+
+/*
+ * Release blocks that were reserved for a btree repair. If the repair
+ * succeeded then we log deferred frees for unused blocks. Otherwise, we try
+ * to free the extents immediately to roll the filesystem back to where it was
+ * before we started.
+ */
+static inline int
+xrep_newbt_destroy_reservation(
+ struct xrep_newbt *xnr,
+ struct xrep_newbt_resv *resv,
+ bool cancel_repair)
+{
+ struct repair_ctx *sc = xnr->sc;
+
+ if (cancel_repair) {
+ int error;
+
+ /* Free the extent then roll the transaction. */
+ error = -libxfs_free_extent(sc->tp, resv->fsbno, resv->len,
+ &xnr->oinfo, xnr->resv);
+ if (error)
+ return error;
+
+ if (sc->ip)
+ return -libxfs_trans_roll_inode(&sc->tp, sc->ip);
+ return xrep_roll_ag_trans(sc);
+ }
+
+ /* We don't have EFIs here so skip the EFD. */
+
+ /*
+ * Use the deferred freeing mechanism to schedule for deletion any
+ * blocks we didn't use to rebuild the tree. This enables us to log
+ * them all in the same transaction as the root change.
+ */
+ resv->fsbno += resv->used;
+ resv->len -= resv->used;
+ resv->used = 0;
+
+ if (resv->len == 0)
+ return 0;
+
+ trace_xrep_newbt_free_blocks(sc->mp,
+ XFS_FSB_TO_AGNO(sc->mp, resv->fsbno),
+ XFS_FSB_TO_AGBNO(sc->mp, resv->fsbno),
+ resv->len, xnr->oinfo.oi_owner);
+
+ __xfs_bmap_add_free(sc->tp, resv->fsbno, resv->len, &xnr->oinfo, true);
+
+ return 0;
+}
+
+/* Free all the accounting info and disk space we reserved for a new btree. */
+void
+xrep_newbt_destroy(
+ struct xrep_newbt *xnr,
+ int error)
+{
+ struct repair_ctx *sc = xnr->sc;
+ struct xrep_newbt_resv *resv, *n;
+ int err2;
+
+ list_for_each_entry_safe(resv, n, &xnr->resv_list, list) {
+ err2 = xrep_newbt_destroy_reservation(xnr, resv, error != 0);
+ if (err2)
+ goto junkit;
+
+ list_del(&resv->list);
+ kmem_free(resv);
+ }
+
+junkit:
+ /*
+ * If we still have reservations attached to @newbt, cleanup must have
+ * failed and the filesystem is about to go down. Clean up the incore
+ * reservations.
+ */
+ list_for_each_entry_safe(resv, n, &xnr->resv_list, list) {
+ list_del(&resv->list);
+ kmem_free(resv);
+ }
+
+ if (sc->ip) {
+ kmem_cache_free(xfs_ifork_zone, xnr->ifake.if_fork);
+ xnr->ifake.if_fork = NULL;
+ }
+}
+
+/* Feed one of the reserved btree blocks to the bulk loader. */
+int
+xrep_newbt_claim_block(
+ struct xfs_btree_cur *cur,
+ struct xrep_newbt *xnr,
+ union xfs_btree_ptr *ptr)
+{
+ struct xrep_newbt_resv *resv;
+ xfs_fsblock_t fsb;
+
+ /*
+ * The first item in the list should always have a free block unless
+ * we're completely out.
+ */
+ resv = list_first_entry(&xnr->resv_list, struct xrep_newbt_resv, list);
+ if (resv->used == resv->len)
+ return ENOSPC;
+
+ /*
+ * Peel off a block from the start of the reservation. We allocate
+ * blocks in order to place blocks on disk in increasing record or key
+ * order. The block reservations tend to end up on the list in
+ * decreasing order, which hopefully results in leaf blocks ending up
+ * together.
+ */
+ fsb = resv->fsbno + resv->used;
+ resv->used++;
+
+ /* If we used all the blocks in this reservation, move it to the end. */
+ if (resv->used == resv->len)
+ list_move_tail(&resv->list, &xnr->resv_list);
+
+ trace_xrep_newbt_claim_block(cur->bc_mp,
+ XFS_FSB_TO_AGNO(cur->bc_mp, fsb),
+ XFS_FSB_TO_AGBNO(cur->bc_mp, fsb),
+ 1, xnr->oinfo.oi_owner);
+
+ if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
+ ptr->l = cpu_to_be64(fsb);
+ else
+ ptr->s = cpu_to_be32(XFS_FSB_TO_AGBNO(cur->bc_mp, fsb));
+ return 0;
+}
diff --git a/repair/bload.h b/repair/bload.h
new file mode 100644
index 00000000..020c4834
--- /dev/null
+++ b/repair/bload.h
@@ -0,0 +1,77 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * Copyright (C) 2020 Oracle. All Rights Reserved.
+ * Author: Darrick J. Wong <darrick.wong@oracle.com>
+ */
+#ifndef __XFS_REPAIR_BLOAD_H__
+#define __XFS_REPAIR_BLOAD_H__
+
+extern int bload_leaf_slack;
+extern int bload_node_slack;
+
+struct repair_ctx {
+ struct xfs_mount *mp;
+ struct xfs_inode *ip;
+ struct xfs_trans *tp;
+
+ struct xfs_buf *agi_bp;
+ struct xfs_buf *agf_bp;
+ struct xfs_buf *agfl_bp;
+};
+
+struct xrep_newbt_resv {
+ /* Link to list of extents that we've reserved. */
+ struct list_head list;
+
+ /* FSB of the block we reserved. */
+ xfs_fsblock_t fsbno;
+
+ /* Length of the reservation. */
+ xfs_extlen_t len;
+
+ /* How much of this reservation we've used. */
+ xfs_extlen_t used;
+};
+
+struct xrep_newbt {
+ struct repair_ctx *sc;
+
+ /* List of extents that we've reserved. */
+ struct list_head resv_list;
+
+ /* Fake root for new btree. */
+ union {
+ struct xbtree_afakeroot afake;
+ struct xbtree_ifakeroot ifake;
+ };
+
+ /* rmap owner of these blocks */
+ struct xfs_owner_info oinfo;
+
+ /* The last reservation we allocated from. */
+ struct xrep_newbt_resv *last_resv;
+
+ /* Allocation hint */
+ xfs_fsblock_t alloc_hint;
+
+ /* per-ag reservation type */
+ enum xfs_ag_resv_type resv;
+};
+
+#define for_each_xrep_newbt_reservation(xnr, resv, n) \
+ list_for_each_entry_safe((resv), (n), &(xnr)->resv_list, list)
+
+void xrep_newbt_init_bare(struct xrep_newbt *xnr, struct repair_ctx *sc);
+void xrep_newbt_init_ag(struct xrep_newbt *xnr, struct repair_ctx *sc,
+ const struct xfs_owner_info *oinfo, xfs_fsblock_t alloc_hint,
+ enum xfs_ag_resv_type resv);
+void xrep_newbt_init_inode(struct xrep_newbt *xnr, struct repair_ctx *sc,
+ int whichfork, const struct xfs_owner_info *oinfo);
+int xrep_newbt_add_blocks(struct xrep_newbt *xnr, xfs_fsblock_t fsbno,
+ xfs_extlen_t len);
+int xrep_newbt_alloc_blocks(struct xrep_newbt *xnr, uint64_t nr_blocks);
+void xrep_newbt_destroy(struct xrep_newbt *xnr, int error);
+int xrep_newbt_claim_block(struct xfs_btree_cur *cur, struct xrep_newbt *xnr,
+ union xfs_btree_ptr *ptr);
+
+#endif /* __XFS_REPAIR_BLOAD_H__ */
diff --git a/repair/xfs_repair.c b/repair/xfs_repair.c
index 9d72fa8e..8fbd3649 100644
--- a/repair/xfs_repair.c
+++ b/repair/xfs_repair.c
@@ -24,6 +24,7 @@
#include "rmap.h"
#include "libfrog/fsgeom.h"
#include "libfrog/platform.h"
+#include "bload.h"
/*
* option tables for getsubopt calls
@@ -39,6 +40,8 @@ enum o_opt_nums {
AG_STRIDE,
FORCE_GEO,
PHASE2_THREADS,
+ BLOAD_LEAF_SLACK,
+ BLOAD_NODE_SLACK,
O_MAX_OPTS,
};
@@ -49,6 +52,8 @@ static char *o_opts[] = {
[AG_STRIDE] = "ag_stride",
[FORCE_GEO] = "force_geometry",
[PHASE2_THREADS] = "phase2_threads",
+ [BLOAD_LEAF_SLACK] = "debug_bload_leaf_slack",
+ [BLOAD_NODE_SLACK] = "debug_bload_node_slack",
[O_MAX_OPTS] = NULL,
};
@@ -260,6 +265,18 @@ process_args(int argc, char **argv)
_("-o phase2_threads requires a parameter\n"));
phase2_threads = (int)strtol(val, NULL, 0);
break;
+ case BLOAD_LEAF_SLACK:
+ if (!val)
+ do_abort(
+ _("-o debug_bload_leaf_slack requires a parameter\n"));
+ bload_leaf_slack = (int)strtol(val, NULL, 0);
+ break;
+ case BLOAD_NODE_SLACK:
+ if (!val)
+ do_abort(
+ _("-o debug_bload_node_slack requires a parameter\n"));
+ bload_node_slack = (int)strtol(val, NULL, 0);
+ break;
default:
unknown('o', val);
break;
^ permalink raw reply related [flat|nested] 30+ messages in thread
* [PATCH 2/9] xfs_repair: remove gratuitous code block in phase5
2020-05-20 1:50 [PATCH v5 0/9] xfs_repair: use btree bulk loading Darrick J. Wong
2020-05-20 1:50 ` [PATCH 1/9] xfs_repair: port the online repair newbt structure Darrick J. Wong
@ 2020-05-20 1:50 ` Darrick J. Wong
2020-05-27 12:15 ` Brian Foster
2020-05-20 1:51 ` [PATCH 3/9] xfs_repair: create a new class of btree rebuild cursors Darrick J. Wong
` (6 subsequent siblings)
8 siblings, 1 reply; 30+ messages in thread
From: Darrick J. Wong @ 2020-05-20 1:50 UTC (permalink / raw)
To: sandeen, darrick.wong; +Cc: Eric Sandeen, linux-xfs, bfoster
From: Eric Sandeen <sandeen@redhat.com>
A commit back in 2008 removed a "for" loop ahead of this code block, but
left the indented code block in place. Remove it for clarity and reflow
comments & lines as needed.
Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
---
repair/phase5.c | 317 ++++++++++++++++++++++++++-----------------------------
1 file changed, 151 insertions(+), 166 deletions(-)
diff --git a/repair/phase5.c b/repair/phase5.c
index 677297fe..84c05a13 100644
--- a/repair/phase5.c
+++ b/repair/phase5.c
@@ -2313,201 +2313,186 @@ phase5_func(
if (verbose)
do_log(_(" - agno = %d\n"), agno);
- {
- /*
- * build up incore bno and bcnt extent btrees
- */
- num_extents = mk_incore_fstree(mp, agno);
+ /*
+ * build up incore bno and bcnt extent btrees
+ */
+ num_extents = mk_incore_fstree(mp, agno);
#ifdef XR_BLD_FREE_TRACE
- fprintf(stderr, "# of bno extents is %d\n",
- count_bno_extents(agno));
+ fprintf(stderr, "# of bno extents is %d\n", count_bno_extents(agno));
#endif
- if (num_extents == 0) {
- /*
- * XXX - what we probably should do here is pick an
- * inode for a regular file in the allocation group
- * that has space allocated and shoot it by traversing
- * the bmap list and putting all its extents on the
- * incore freespace trees, clearing the inode,
- * and clearing the in-use bit in the incore inode
- * tree. Then try mk_incore_fstree() again.
- */
- do_error(_("unable to rebuild AG %u. "
- "Not enough free space in on-disk AG.\n"),
- agno);
- }
-
- /*
- * ok, now set up the btree cursors for the
- * on-disk btrees (includs pre-allocating all
- * required blocks for the trees themselves)
- */
- init_ino_cursor(mp, agno, &ino_btree_curs, &num_inos,
- &num_free_inos, 0);
-
- if (xfs_sb_version_hasfinobt(&mp->m_sb))
- init_ino_cursor(mp, agno, &fino_btree_curs,
- &finobt_num_inos, &finobt_num_free_inos,
- 1);
-
- sb_icount_ag[agno] += num_inos;
- sb_ifree_ag[agno] += num_free_inos;
-
- /*
- * Set up the btree cursors for the on-disk rmap btrees,
- * which includes pre-allocating all required blocks.
- */
- init_rmapbt_cursor(mp, agno, &rmap_btree_curs);
-
+ if (num_extents == 0) {
/*
- * Set up the btree cursors for the on-disk refcount btrees,
- * which includes pre-allocating all required blocks.
+ * XXX - what we probably should do here is pick an inode for
+ * a regular file in the allocation group that has space
+ * allocated and shoot it by traversing the bmap list and
+ * putting all its extents on the incore freespace trees,
+ * clearing the inode, and clearing the in-use bit in the
+ * incore inode tree. Then try mk_incore_fstree() again.
*/
- init_refc_cursor(mp, agno, &refcnt_btree_curs);
-
- num_extents = count_bno_extents_blocks(agno, &num_freeblocks);
+ do_error(
+_("unable to rebuild AG %u. Not enough free space in on-disk AG.\n"),
+ agno);
+ }
+
+ /*
+ * ok, now set up the btree cursors for the on-disk btrees (includes
+ * pre-allocating all required blocks for the trees themselves)
+ */
+ init_ino_cursor(mp, agno, &ino_btree_curs, &num_inos,
+ &num_free_inos, 0);
+
+ if (xfs_sb_version_hasfinobt(&mp->m_sb))
+ init_ino_cursor(mp, agno, &fino_btree_curs, &finobt_num_inos,
+ &finobt_num_free_inos, 1);
+
+ sb_icount_ag[agno] += num_inos;
+ sb_ifree_ag[agno] += num_free_inos;
+
+ /*
+ * Set up the btree cursors for the on-disk rmap btrees, which includes
+ * pre-allocating all required blocks.
+ */
+ init_rmapbt_cursor(mp, agno, &rmap_btree_curs);
+
+ /*
+ * Set up the btree cursors for the on-disk refcount btrees,
+ * which includes pre-allocating all required blocks.
+ */
+ init_refc_cursor(mp, agno, &refcnt_btree_curs);
+
+ num_extents = count_bno_extents_blocks(agno, &num_freeblocks);
+ /*
+ * lose two blocks per AG -- the space tree roots are counted as
+ * allocated since the space trees always have roots
+ */
+ sb_fdblocks_ag[agno] += num_freeblocks - 2;
+
+ if (num_extents == 0) {
/*
- * lose two blocks per AG -- the space tree roots
- * are counted as allocated since the space trees
- * always have roots
+ * XXX - what we probably should do here is pick an inode for
+ * a regular file in the allocation group that has space
+ * allocated and shoot it by traversing the bmap list and
+ * putting all its extents on the incore freespace trees,
+ * clearing the inode, and clearing the in-use bit in the
+ * incore inode tree. Then try mk_incore_fstree() again.
*/
- sb_fdblocks_ag[agno] += num_freeblocks - 2;
-
- if (num_extents == 0) {
- /*
- * XXX - what we probably should do here is pick an
- * inode for a regular file in the allocation group
- * that has space allocated and shoot it by traversing
- * the bmap list and putting all its extents on the
- * incore freespace trees, clearing the inode,
- * and clearing the in-use bit in the incore inode
- * tree. Then try mk_incore_fstree() again.
- */
- do_error(
- _("unable to rebuild AG %u. No free space.\n"), agno);
- }
+ do_error(_("unable to rebuild AG %u. No free space.\n"), agno);
+ }
#ifdef XR_BLD_FREE_TRACE
- fprintf(stderr, "# of bno extents is %d\n", num_extents);
+ fprintf(stderr, "# of bno extents is %d\n", num_extents);
#endif
- /*
- * track blocks that we might really lose
- */
- extra_blocks = calculate_freespace_cursor(mp, agno,
- &num_extents, &bno_btree_curs);
+ /*
+ * track blocks that we might really lose
+ */
+ extra_blocks = calculate_freespace_cursor(mp, agno,
+ &num_extents, &bno_btree_curs);
- /*
- * freespace btrees live in the "free space" but
- * the filesystem treats AGFL blocks as allocated
- * since they aren't described by the freespace trees
- */
+ /*
+ * freespace btrees live in the "free space" but the filesystem treats
+ * AGFL blocks as allocated since they aren't described by the
+ * freespace trees
+ */
- /*
- * see if we can fit all the extra blocks into the AGFL
- */
- extra_blocks = (extra_blocks - libxfs_agfl_size(mp) > 0)
- ? extra_blocks - libxfs_agfl_size(mp)
- : 0;
+ /*
+ * see if we can fit all the extra blocks into the AGFL
+ */
+ extra_blocks = (extra_blocks - libxfs_agfl_size(mp) > 0) ?
+ extra_blocks - libxfs_agfl_size(mp) : 0;
- if (extra_blocks > 0)
- sb_fdblocks_ag[agno] -= extra_blocks;
+ if (extra_blocks > 0)
+ sb_fdblocks_ag[agno] -= extra_blocks;
- bcnt_btree_curs = bno_btree_curs;
+ bcnt_btree_curs = bno_btree_curs;
- bno_btree_curs.owner = XFS_RMAP_OWN_AG;
- bcnt_btree_curs.owner = XFS_RMAP_OWN_AG;
- setup_cursor(mp, agno, &bno_btree_curs);
- setup_cursor(mp, agno, &bcnt_btree_curs);
+ bno_btree_curs.owner = XFS_RMAP_OWN_AG;
+ bcnt_btree_curs.owner = XFS_RMAP_OWN_AG;
+ setup_cursor(mp, agno, &bno_btree_curs);
+ setup_cursor(mp, agno, &bcnt_btree_curs);
#ifdef XR_BLD_FREE_TRACE
- fprintf(stderr, "# of bno extents is %d\n",
- count_bno_extents(agno));
- fprintf(stderr, "# of bcnt extents is %d\n",
- count_bcnt_extents(agno));
+ fprintf(stderr, "# of bno extents is %d\n", count_bno_extents(agno));
+ fprintf(stderr, "# of bcnt extents is %d\n", count_bcnt_extents(agno));
#endif
- /*
- * now rebuild the freespace trees
- */
- freeblks1 = build_freespace_tree(mp, agno,
+ /*
+ * now rebuild the freespace trees
+ */
+ freeblks1 = build_freespace_tree(mp, agno,
&bno_btree_curs, XFS_BTNUM_BNO);
#ifdef XR_BLD_FREE_TRACE
- fprintf(stderr, "# of free blocks == %d\n", freeblks1);
+ fprintf(stderr, "# of free blocks == %d\n", freeblks1);
#endif
- write_cursor(&bno_btree_curs);
+ write_cursor(&bno_btree_curs);
#ifdef DEBUG
- freeblks2 = build_freespace_tree(mp, agno,
- &bcnt_btree_curs, XFS_BTNUM_CNT);
+ freeblks2 = build_freespace_tree(mp, agno,
+ &bcnt_btree_curs, XFS_BTNUM_CNT);
#else
- (void) build_freespace_tree(mp, agno,
- &bcnt_btree_curs, XFS_BTNUM_CNT);
+ (void) build_freespace_tree(mp, agno, &bcnt_btree_curs, XFS_BTNUM_CNT);
#endif
- write_cursor(&bcnt_btree_curs);
-
- ASSERT(freeblks1 == freeblks2);
-
- if (xfs_sb_version_hasrmapbt(&mp->m_sb)) {
- build_rmap_tree(mp, agno, &rmap_btree_curs);
- write_cursor(&rmap_btree_curs);
- sb_fdblocks_ag[agno] += (rmap_btree_curs.num_tot_blocks -
- rmap_btree_curs.num_free_blocks) - 1;
- }
-
- if (xfs_sb_version_hasreflink(&mp->m_sb)) {
- build_refcount_tree(mp, agno, &refcnt_btree_curs);
- write_cursor(&refcnt_btree_curs);
- }
-
- /*
- * set up agf and agfl
- */
- build_agf_agfl(mp, agno, &bno_btree_curs,
- &bcnt_btree_curs, freeblks1, extra_blocks,
- &rmap_btree_curs, &refcnt_btree_curs, lost_fsb);
- /*
- * build inode allocation tree.
- */
- build_ino_tree(mp, agno, &ino_btree_curs, XFS_BTNUM_INO,
- &agi_stat);
- write_cursor(&ino_btree_curs);
-
- /*
- * build free inode tree
- */
- if (xfs_sb_version_hasfinobt(&mp->m_sb)) {
- build_ino_tree(mp, agno, &fino_btree_curs,
- XFS_BTNUM_FINO, NULL);
- write_cursor(&fino_btree_curs);
- }
-
- /* build the agi */
- build_agi(mp, agno, &ino_btree_curs, &fino_btree_curs,
- &agi_stat);
-
- /*
- * tear down cursors
- */
- finish_cursor(&bno_btree_curs);
- finish_cursor(&ino_btree_curs);
- if (xfs_sb_version_hasrmapbt(&mp->m_sb))
- finish_cursor(&rmap_btree_curs);
- if (xfs_sb_version_hasreflink(&mp->m_sb))
- finish_cursor(&refcnt_btree_curs);
- if (xfs_sb_version_hasfinobt(&mp->m_sb))
- finish_cursor(&fino_btree_curs);
- finish_cursor(&bcnt_btree_curs);
-
- /*
- * release the incore per-AG bno/bcnt trees so
- * the extent nodes can be recycled
- */
- release_agbno_extent_tree(agno);
- release_agbcnt_extent_tree(agno);
+ write_cursor(&bcnt_btree_curs);
+
+ ASSERT(freeblks1 == freeblks2);
+
+ if (xfs_sb_version_hasrmapbt(&mp->m_sb)) {
+ build_rmap_tree(mp, agno, &rmap_btree_curs);
+ write_cursor(&rmap_btree_curs);
+ sb_fdblocks_ag[agno] += (rmap_btree_curs.num_tot_blocks -
+ rmap_btree_curs.num_free_blocks) - 1;
+ }
+
+ if (xfs_sb_version_hasreflink(&mp->m_sb)) {
+ build_refcount_tree(mp, agno, &refcnt_btree_curs);
+ write_cursor(&refcnt_btree_curs);
}
+
+ /*
+ * set up agf and agfl
+ */
+ build_agf_agfl(mp, agno, &bno_btree_curs,
+ &bcnt_btree_curs, freeblks1, extra_blocks,
+ &rmap_btree_curs, &refcnt_btree_curs, lost_fsb);
+ /*
+ * build inode allocation tree.
+ */
+ build_ino_tree(mp, agno, &ino_btree_curs, XFS_BTNUM_INO, &agi_stat);
+ write_cursor(&ino_btree_curs);
+
+ /*
+ * build free inode tree
+ */
+ if (xfs_sb_version_hasfinobt(&mp->m_sb)) {
+ build_ino_tree(mp, agno, &fino_btree_curs,
+ XFS_BTNUM_FINO, NULL);
+ write_cursor(&fino_btree_curs);
+ }
+
+ /* build the agi */
+ build_agi(mp, agno, &ino_btree_curs, &fino_btree_curs, &agi_stat);
+
+ /*
+ * tear down cursors
+ */
+ finish_cursor(&bno_btree_curs);
+ finish_cursor(&ino_btree_curs);
+ if (xfs_sb_version_hasrmapbt(&mp->m_sb))
+ finish_cursor(&rmap_btree_curs);
+ if (xfs_sb_version_hasreflink(&mp->m_sb))
+ finish_cursor(&refcnt_btree_curs);
+ if (xfs_sb_version_hasfinobt(&mp->m_sb))
+ finish_cursor(&fino_btree_curs);
+ finish_cursor(&bcnt_btree_curs);
+
+ /*
+ * release the incore per-AG bno/bcnt trees so the extent nodes
+ * can be recycled
+ */
+ release_agbno_extent_tree(agno);
+ release_agbcnt_extent_tree(agno);
PROG_RPT_INC(prog_rpt_done[agno], 1);
}
^ permalink raw reply related [flat|nested] 30+ messages in thread
* [PATCH 3/9] xfs_repair: create a new class of btree rebuild cursors
2020-05-20 1:50 [PATCH v5 0/9] xfs_repair: use btree bulk loading Darrick J. Wong
2020-05-20 1:50 ` [PATCH 1/9] xfs_repair: port the online repair newbt structure Darrick J. Wong
2020-05-20 1:50 ` [PATCH 2/9] xfs_repair: remove gratuitous code block in phase5 Darrick J. Wong
@ 2020-05-20 1:51 ` Darrick J. Wong
2020-05-27 12:18 ` Brian Foster
2020-05-20 1:51 ` [PATCH 4/9] xfs_repair: rebuild free space btrees with bulk loader Darrick J. Wong
` (5 subsequent siblings)
8 siblings, 1 reply; 30+ messages in thread
From: Darrick J. Wong @ 2020-05-20 1:51 UTC (permalink / raw)
To: sandeen, darrick.wong; +Cc: linux-xfs, bfoster
From: Darrick J. Wong <darrick.wong@oracle.com>
Create some new support structures and functions to assist phase5 in
using the btree bulk loader to reconstruct metadata btrees. This is the
first step in removing the open-coded rebuilding code.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
---
repair/phase5.c | 239 ++++++++++++++++++++++++++++++++++++++++++++++++++-----
1 file changed, 218 insertions(+), 21 deletions(-)
diff --git a/repair/phase5.c b/repair/phase5.c
index 84c05a13..8f5e5f59 100644
--- a/repair/phase5.c
+++ b/repair/phase5.c
@@ -18,6 +18,7 @@
#include "progress.h"
#include "slab.h"
#include "rmap.h"
+#include "bload.h"
/*
* we maintain the current slice (path from root to leaf)
@@ -65,6 +66,23 @@ typedef struct bt_status {
uint64_t owner; /* owner */
} bt_status_t;
+/* Context for rebuilding a per-AG btree. */
+struct bt_rebuild {
+ /* Fake root for staging and space preallocations. */
+ struct xrep_newbt newbt;
+
+ /* Geometry of the new btree. */
+ struct xfs_btree_bload bload;
+
+ /* Staging btree cursor for the new tree. */
+ struct xfs_btree_cur *cur;
+
+ /* Tree-specific data. */
+ union {
+ struct xfs_slab_cursor *slab_cursor;
+ };
+};
+
/*
* extra metadata for the agi
*/
@@ -306,6 +324,156 @@ _("error - not enough free space in filesystem\n"));
#endif
}
+/*
+ * Estimate proper slack values for a btree that's being reloaded.
+ *
+ * Under most circumstances, we'll take whatever default loading value the
+ * btree bulk loading code calculates for us. However, there are some
+ * exceptions to this rule:
+ *
+ * (1) If someone turned one of the debug knobs.
+ * (2) The AG has less than ~9% space free.
+ *
+ * Note that we actually use 3/32 for the comparison to avoid division.
+ */
+static void
+estimate_ag_bload_slack(
+ struct repair_ctx *sc,
+ struct xfs_btree_bload *bload,
+ unsigned int free)
+{
+ /*
+ * The global values are set to -1 (i.e. take the bload defaults)
+ * unless someone has set them otherwise, so we just pull the values
+ * here.
+ */
+ bload->leaf_slack = bload_leaf_slack;
+ bload->node_slack = bload_node_slack;
+
+ /* No further changes if there's more than 3/32ths space left. */
+ if (free >= ((sc->mp->m_sb.sb_agblocks * 3) >> 5))
+ return;
+
+ /* We're low on space; load the btrees as tightly as possible. */
+ if (bload->leaf_slack < 0)
+ bload->leaf_slack = 0;
+ if (bload->node_slack < 0)
+ bload->node_slack = 0;
+}
+
+/* Initialize a btree rebuild context. */
+static void
+init_rebuild(
+ struct repair_ctx *sc,
+ const struct xfs_owner_info *oinfo,
+ xfs_agblock_t free_space,
+ struct bt_rebuild *btr)
+{
+ memset(btr, 0, sizeof(struct bt_rebuild));
+
+ xrep_newbt_init_bare(&btr->newbt, sc);
+ btr->newbt.oinfo = *oinfo; /* struct copy */
+ estimate_ag_bload_slack(sc, &btr->bload, free_space);
+}
+
+/*
+ * Update this free space record to reflect the blocks we stole from the
+ * beginning of the record.
+ */
+static void
+consume_freespace(
+ xfs_agnumber_t agno,
+ struct extent_tree_node *ext_ptr,
+ uint32_t len)
+{
+ struct extent_tree_node *bno_ext_ptr;
+ xfs_agblock_t new_start = ext_ptr->ex_startblock + len;
+ xfs_extlen_t new_len = ext_ptr->ex_blockcount - len;
+
+ /* Delete the used-up extent from both extent trees. */
+#ifdef XR_BLD_FREE_TRACE
+ fprintf(stderr, "releasing extent: %u [%u %u]\n", agno,
+ ext_ptr->ex_startblock, ext_ptr->ex_blockcount);
+#endif
+ bno_ext_ptr = find_bno_extent(agno, ext_ptr->ex_startblock);
+ ASSERT(bno_ext_ptr != NULL);
+ get_bno_extent(agno, bno_ext_ptr);
+ release_extent_tree_node(bno_ext_ptr);
+
+ ext_ptr = get_bcnt_extent(agno, ext_ptr->ex_startblock,
+ ext_ptr->ex_blockcount);
+ release_extent_tree_node(ext_ptr);
+
+ /*
+ * If we only used part of this last extent, then we must reinsert the
+ * extent in the extent trees.
+ */
+ if (new_len > 0) {
+ add_bno_extent(agno, new_start, new_len);
+ add_bcnt_extent(agno, new_start, new_len);
+ }
+}
+
+/* Reserve blocks for the new btree. */
+static void
+setup_rebuild(
+ struct xfs_mount *mp,
+ xfs_agnumber_t agno,
+ struct bt_rebuild *btr,
+ uint32_t nr_blocks)
+{
+ struct extent_tree_node *ext_ptr;
+ uint32_t blocks_allocated = 0;
+ uint32_t len;
+ int error;
+
+ while (blocks_allocated < nr_blocks) {
+ /*
+ * Grab the smallest extent and use it up, then get the
+ * next smallest. This mimics the init_*_cursor code.
+ */
+ ext_ptr = findfirst_bcnt_extent(agno);
+ if (!ext_ptr)
+ do_error(
+_("error - not enough free space in filesystem\n"));
+
+ /* Use up the extent we've got. */
+ len = min(ext_ptr->ex_blockcount, nr_blocks - blocks_allocated);
+ error = xrep_newbt_add_blocks(&btr->newbt,
+ XFS_AGB_TO_FSB(mp, agno,
+ ext_ptr->ex_startblock),
+ len);
+ if (error)
+ do_error(_("could not set up btree reservation: %s\n"),
+ strerror(-error));
+
+ error = rmap_add_ag_rec(mp, agno, ext_ptr->ex_startblock, len,
+ btr->newbt.oinfo.oi_owner);
+ if (error)
+ do_error(_("could not set up btree rmaps: %s\n"),
+ strerror(-error));
+
+ consume_freespace(agno, ext_ptr, len);
+ blocks_allocated += len;
+ }
+#ifdef XR_BLD_FREE_TRACE
+ fprintf(stderr, "blocks_allocated = %d\n",
+ blocks_allocated);
+#endif
+}
+
+/* Feed one of the new btree blocks to the bulk loader. */
+static int
+rebuild_claim_block(
+ struct xfs_btree_cur *cur,
+ union xfs_btree_ptr *ptr,
+ void *priv)
+{
+ struct bt_rebuild *btr = priv;
+
+ return xrep_newbt_claim_block(cur, &btr->newbt, ptr);
+}
+
static void
write_cursor(bt_status_t *curs)
{
@@ -336,6 +504,34 @@ finish_cursor(bt_status_t *curs)
free(curs->btree_blocks);
}
+/*
+ * Scoop up leftovers from a rebuild cursor for later freeing, then free the
+ * rebuild context.
+ */
+static void
+finish_rebuild(
+ struct xfs_mount *mp,
+ struct bt_rebuild *btr,
+ struct xfs_slab *lost_fsb)
+{
+ struct xrep_newbt_resv *resv, *n;
+
+ for_each_xrep_newbt_reservation(&btr->newbt, resv, n) {
+ while (resv->used < resv->len) {
+ xfs_fsblock_t fsb = resv->fsbno + resv->used;
+ int error;
+
+ error = slab_add(lost_fsb, &fsb);
+ if (error)
+ do_error(
+_("Insufficient memory saving lost blocks.\n"));
+ resv->used++;
+ }
+ }
+
+ xrep_newbt_destroy(&btr->newbt, 0);
+}
+
/*
* We need to leave some free records in the tree for the corner case of
* setting up the AGFL. This may require allocation of blocks, and as
@@ -2287,28 +2483,29 @@ keep_fsinos(xfs_mount_t *mp)
static void
phase5_func(
- xfs_mount_t *mp,
- xfs_agnumber_t agno,
- struct xfs_slab *lost_fsb)
+ struct xfs_mount *mp,
+ xfs_agnumber_t agno,
+ struct xfs_slab *lost_fsb)
{
- uint64_t num_inos;
- uint64_t num_free_inos;
- uint64_t finobt_num_inos;
- uint64_t finobt_num_free_inos;
- bt_status_t bno_btree_curs;
- bt_status_t bcnt_btree_curs;
- bt_status_t ino_btree_curs;
- bt_status_t fino_btree_curs;
- bt_status_t rmap_btree_curs;
- bt_status_t refcnt_btree_curs;
- int extra_blocks = 0;
- uint num_freeblocks;
- xfs_extlen_t freeblks1;
+ struct repair_ctx sc = { .mp = mp, };
+ struct agi_stat agi_stat = {0,};
+ uint64_t num_inos;
+ uint64_t num_free_inos;
+ uint64_t finobt_num_inos;
+ uint64_t finobt_num_free_inos;
+ bt_status_t bno_btree_curs;
+ bt_status_t bcnt_btree_curs;
+ bt_status_t ino_btree_curs;
+ bt_status_t fino_btree_curs;
+ bt_status_t rmap_btree_curs;
+ bt_status_t refcnt_btree_curs;
+ int extra_blocks = 0;
+ uint num_freeblocks;
+ xfs_extlen_t freeblks1;
#ifdef DEBUG
- xfs_extlen_t freeblks2;
+ xfs_extlen_t freeblks2;
#endif
- xfs_agblock_t num_extents;
- struct agi_stat agi_stat = {0,};
+ xfs_agblock_t num_extents;
if (verbose)
do_log(_(" - agno = %d\n"), agno);
@@ -2516,8 +2713,8 @@ inject_lost_blocks(
if (error)
goto out_cancel;
- error = -libxfs_free_extent(tp, *fsb, 1, &XFS_RMAP_OINFO_AG,
- XFS_AG_RESV_NONE);
+ error = -libxfs_free_extent(tp, *fsb, 1,
+ &XFS_RMAP_OINFO_ANY_OWNER, XFS_AG_RESV_NONE);
if (error)
goto out_cancel;
^ permalink raw reply related [flat|nested] 30+ messages in thread
* [PATCH 4/9] xfs_repair: rebuild free space btrees with bulk loader
2020-05-20 1:50 [PATCH v5 0/9] xfs_repair: use btree bulk loading Darrick J. Wong
` (2 preceding siblings ...)
2020-05-20 1:51 ` [PATCH 3/9] xfs_repair: create a new class of btree rebuild cursors Darrick J. Wong
@ 2020-05-20 1:51 ` Darrick J. Wong
2020-05-28 15:10 ` Brian Foster
2020-05-20 1:51 ` [PATCH 5/9] xfs_repair: rebuild inode " Darrick J. Wong
` (4 subsequent siblings)
8 siblings, 1 reply; 30+ messages in thread
From: Darrick J. Wong @ 2020-05-20 1:51 UTC (permalink / raw)
To: sandeen, darrick.wong; +Cc: linux-xfs, bfoster
From: Darrick J. Wong <darrick.wong@oracle.com>
Use the btree bulk loading functions to rebuild the free space btrees
and drop the open-coded implementation.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
---
libxfs/libxfs_api_defs.h | 3
repair/phase5.c | 870 +++++++++++++---------------------------------
2 files changed, 254 insertions(+), 619 deletions(-)
diff --git a/libxfs/libxfs_api_defs.h b/libxfs/libxfs_api_defs.h
index 61047f8f..bace739c 100644
--- a/libxfs/libxfs_api_defs.h
+++ b/libxfs/libxfs_api_defs.h
@@ -24,6 +24,7 @@
#define xfs_alloc_ag_max_usable libxfs_alloc_ag_max_usable
#define xfs_allocbt_maxrecs libxfs_allocbt_maxrecs
+#define xfs_allocbt_stage_cursor libxfs_allocbt_stage_cursor
#define xfs_alloc_fix_freelist libxfs_alloc_fix_freelist
#define xfs_alloc_min_freelist libxfs_alloc_min_freelist
#define xfs_alloc_read_agf libxfs_alloc_read_agf
@@ -41,6 +42,8 @@
#define xfs_bmbt_maxrecs libxfs_bmbt_maxrecs
#define xfs_bmdr_maxrecs libxfs_bmdr_maxrecs
+#define xfs_btree_bload libxfs_btree_bload
+#define xfs_btree_bload_compute_geometry libxfs_btree_bload_compute_geometry
#define xfs_btree_del_cursor libxfs_btree_del_cursor
#define xfs_btree_init_block libxfs_btree_init_block
#define xfs_buf_delwri_submit libxfs_buf_delwri_submit
diff --git a/repair/phase5.c b/repair/phase5.c
index 8f5e5f59..e69b042c 100644
--- a/repair/phase5.c
+++ b/repair/phase5.c
@@ -80,6 +80,10 @@ struct bt_rebuild {
/* Tree-specific data. */
union {
struct xfs_slab_cursor *slab_cursor;
+ struct {
+ struct extent_tree_node *bno_rec;
+ xfs_agblock_t *freeblks;
+ };
};
};
@@ -97,7 +101,10 @@ static uint64_t *sb_ifree_ag; /* free inodes per ag */
static uint64_t *sb_fdblocks_ag; /* free data blocks per ag */
static int
-mk_incore_fstree(xfs_mount_t *mp, xfs_agnumber_t agno)
+mk_incore_fstree(
+ struct xfs_mount *mp,
+ xfs_agnumber_t agno,
+ unsigned int *num_freeblocks)
{
int in_extent;
int num_extents;
@@ -109,6 +116,8 @@ mk_incore_fstree(xfs_mount_t *mp, xfs_agnumber_t agno)
xfs_extlen_t blen;
int bstate;
+ *num_freeblocks = 0;
+
/*
* scan the bitmap for the ag looking for continuous
* extents of free blocks. At this point, we know
@@ -164,6 +173,7 @@ mk_incore_fstree(xfs_mount_t *mp, xfs_agnumber_t agno)
#endif
add_bno_extent(agno, extent_start, extent_len);
add_bcnt_extent(agno, extent_start, extent_len);
+ *num_freeblocks += extent_len;
}
}
}
@@ -177,6 +187,7 @@ mk_incore_fstree(xfs_mount_t *mp, xfs_agnumber_t agno)
#endif
add_bno_extent(agno, extent_start, extent_len);
add_bcnt_extent(agno, extent_start, extent_len);
+ *num_freeblocks += extent_len;
}
return(num_extents);
@@ -532,287 +543,6 @@ _("Insufficient memory saving lost blocks.\n"));
xrep_newbt_destroy(&btr->newbt, 0);
}
-/*
- * We need to leave some free records in the tree for the corner case of
- * setting up the AGFL. This may require allocation of blocks, and as
- * such can require insertion of new records into the tree (e.g. moving
- * a record in the by-count tree when a long extent is shortened). If we
- * pack the records into the leaves with no slack space, this requires a
- * leaf split to occur and a block to be allocated from the free list.
- * If we don't have any blocks on the free list (because we are setting
- * it up!), then we fail, and the filesystem will fail with the same
- * failure at runtime. Hence leave a couple of records slack space in
- * each block to allow immediate modification of the tree without
- * requiring splits to be done.
- *
- * XXX(hch): any reason we don't just look at mp->m_alloc_mxr?
- */
-#define XR_ALLOC_BLOCK_MAXRECS(mp, level) \
- (libxfs_allocbt_maxrecs((mp), (mp)->m_sb.sb_blocksize, (level) == 0) - 2)
-
-/*
- * this calculates a freespace cursor for an ag.
- * btree_curs is an in/out. returns the number of
- * blocks that will show up in the AGFL.
- */
-static int
-calculate_freespace_cursor(xfs_mount_t *mp, xfs_agnumber_t agno,
- xfs_agblock_t *extents, bt_status_t *btree_curs)
-{
- xfs_extlen_t blocks_needed; /* a running count */
- xfs_extlen_t blocks_allocated_pt; /* per tree */
- xfs_extlen_t blocks_allocated_total; /* for both trees */
- xfs_agblock_t num_extents;
- int i;
- int extents_used;
- int extra_blocks;
- bt_stat_level_t *lptr;
- bt_stat_level_t *p_lptr;
- extent_tree_node_t *ext_ptr;
- int level;
-
- num_extents = *extents;
- extents_used = 0;
-
- ASSERT(num_extents != 0);
-
- lptr = &btree_curs->level[0];
- btree_curs->init = 1;
-
- /*
- * figure out how much space we need for the leaf level
- * of the tree and set up the cursor for the leaf level
- * (note that the same code is duplicated further down)
- */
- lptr->num_blocks = howmany(num_extents, XR_ALLOC_BLOCK_MAXRECS(mp, 0));
- lptr->num_recs_pb = num_extents / lptr->num_blocks;
- lptr->modulo = num_extents % lptr->num_blocks;
- lptr->num_recs_tot = num_extents;
- level = 1;
-
-#ifdef XR_BLD_FREE_TRACE
- fprintf(stderr, "%s 0 %d %d %d %d\n", __func__,
- lptr->num_blocks,
- lptr->num_recs_pb,
- lptr->modulo,
- lptr->num_recs_tot);
-#endif
- /*
- * if we need more levels, set them up. # of records
- * per level is the # of blocks in the level below it
- */
- if (lptr->num_blocks > 1) {
- for (; btree_curs->level[level - 1].num_blocks > 1
- && level < XFS_BTREE_MAXLEVELS;
- level++) {
- lptr = &btree_curs->level[level];
- p_lptr = &btree_curs->level[level - 1];
- lptr->num_blocks = howmany(p_lptr->num_blocks,
- XR_ALLOC_BLOCK_MAXRECS(mp, level));
- lptr->modulo = p_lptr->num_blocks
- % lptr->num_blocks;
- lptr->num_recs_pb = p_lptr->num_blocks
- / lptr->num_blocks;
- lptr->num_recs_tot = p_lptr->num_blocks;
-#ifdef XR_BLD_FREE_TRACE
- fprintf(stderr, "%s %d %d %d %d %d\n", __func__,
- level,
- lptr->num_blocks,
- lptr->num_recs_pb,
- lptr->modulo,
- lptr->num_recs_tot);
-#endif
- }
- }
-
- ASSERT(lptr->num_blocks == 1);
- btree_curs->num_levels = level;
-
- /*
- * ok, now we have a hypothetical cursor that
- * will work for both the bno and bcnt trees.
- * now figure out if using up blocks to set up the
- * trees will perturb the shape of the freespace tree.
- * if so, we've over-allocated. the freespace trees
- * as they will be *after* accounting for the free space
- * we've used up will need fewer blocks to to represent
- * than we've allocated. We can use the AGFL to hold
- * xfs_agfl_size (sector/struct xfs_agfl) blocks but that's it.
- * Thus we limit things to xfs_agfl_size/2 for each of the 2 btrees.
- * if the number of extra blocks is more than that,
- * we'll have to be called again.
- */
- for (blocks_needed = 0, i = 0; i < level; i++) {
- blocks_needed += btree_curs->level[i].num_blocks;
- }
-
- /*
- * record the # of blocks we've allocated
- */
- blocks_allocated_pt = blocks_needed;
- blocks_needed *= 2;
- blocks_allocated_total = blocks_needed;
-
- /*
- * figure out how many free extents will be used up by
- * our space allocation
- */
- if ((ext_ptr = findfirst_bcnt_extent(agno)) == NULL)
- do_error(_("can't rebuild fs trees -- not enough free space "
- "on ag %u\n"), agno);
-
- while (ext_ptr != NULL && blocks_needed > 0) {
- if (ext_ptr->ex_blockcount <= blocks_needed) {
- blocks_needed -= ext_ptr->ex_blockcount;
- extents_used++;
- } else {
- blocks_needed = 0;
- }
-
- ext_ptr = findnext_bcnt_extent(agno, ext_ptr);
-
-#ifdef XR_BLD_FREE_TRACE
- if (ext_ptr != NULL) {
- fprintf(stderr, "got next extent [%u %u]\n",
- ext_ptr->ex_startblock, ext_ptr->ex_blockcount);
- } else {
- fprintf(stderr, "out of extents\n");
- }
-#endif
- }
- if (blocks_needed > 0)
- do_error(_("ag %u - not enough free space to build freespace "
- "btrees\n"), agno);
-
- ASSERT(num_extents >= extents_used);
-
- num_extents -= extents_used;
-
- /*
- * see if the number of leaf blocks will change as a result
- * of the number of extents changing
- */
- if (howmany(num_extents, XR_ALLOC_BLOCK_MAXRECS(mp, 0))
- != btree_curs->level[0].num_blocks) {
- /*
- * yes -- recalculate the cursor. If the number of
- * excess (overallocated) blocks is < xfs_agfl_size/2, we're ok.
- * we can put those into the AGFL. we don't try
- * and get things to converge exactly (reach a
- * state with zero excess blocks) because there
- * exist pathological cases which will never
- * converge. first, check for the zero-case.
- */
- if (num_extents == 0) {
- /*
- * ok, we've used up all the free blocks
- * trying to lay out the leaf level. go
- * to a one block (empty) btree and put the
- * already allocated blocks into the AGFL
- */
- if (btree_curs->level[0].num_blocks != 1) {
- /*
- * we really needed more blocks because
- * the old tree had more than one level.
- * this is bad.
- */
- do_warn(_("not enough free blocks left to "
- "describe all free blocks in AG "
- "%u\n"), agno);
- }
-#ifdef XR_BLD_FREE_TRACE
- fprintf(stderr,
- "ag %u -- no free extents, alloc'ed %d\n",
- agno, blocks_allocated_pt);
-#endif
- lptr->num_blocks = 1;
- lptr->modulo = 0;
- lptr->num_recs_pb = 0;
- lptr->num_recs_tot = 0;
-
- btree_curs->num_levels = 1;
-
- /*
- * don't reset the allocation stats, assume
- * they're all extra blocks
- * don't forget to return the total block count
- * not the per-tree block count. these are the
- * extras that will go into the AGFL. subtract
- * two for the root blocks.
- */
- btree_curs->num_tot_blocks = blocks_allocated_pt;
- btree_curs->num_free_blocks = blocks_allocated_pt;
-
- *extents = 0;
-
- return(blocks_allocated_total - 2);
- }
-
- lptr = &btree_curs->level[0];
- lptr->num_blocks = howmany(num_extents,
- XR_ALLOC_BLOCK_MAXRECS(mp, 0));
- lptr->num_recs_pb = num_extents / lptr->num_blocks;
- lptr->modulo = num_extents % lptr->num_blocks;
- lptr->num_recs_tot = num_extents;
- level = 1;
-
- /*
- * if we need more levels, set them up
- */
- if (lptr->num_blocks > 1) {
- for (level = 1; btree_curs->level[level-1].num_blocks
- > 1 && level < XFS_BTREE_MAXLEVELS;
- level++) {
- lptr = &btree_curs->level[level];
- p_lptr = &btree_curs->level[level-1];
- lptr->num_blocks = howmany(p_lptr->num_blocks,
- XR_ALLOC_BLOCK_MAXRECS(mp, level));
- lptr->modulo = p_lptr->num_blocks
- % lptr->num_blocks;
- lptr->num_recs_pb = p_lptr->num_blocks
- / lptr->num_blocks;
- lptr->num_recs_tot = p_lptr->num_blocks;
- }
- }
- ASSERT(lptr->num_blocks == 1);
- btree_curs->num_levels = level;
-
- /*
- * now figure out the number of excess blocks
- */
- for (blocks_needed = 0, i = 0; i < level; i++) {
- blocks_needed += btree_curs->level[i].num_blocks;
- }
- blocks_needed *= 2;
-
- ASSERT(blocks_allocated_total >= blocks_needed);
- extra_blocks = blocks_allocated_total - blocks_needed;
- } else {
- if (extents_used > 0) {
- /*
- * reset the leaf level geometry to account
- * for consumed extents. we can leave the
- * rest of the cursor alone since the number
- * of leaf blocks hasn't changed.
- */
- lptr = &btree_curs->level[0];
-
- lptr->num_recs_pb = num_extents / lptr->num_blocks;
- lptr->modulo = num_extents % lptr->num_blocks;
- lptr->num_recs_tot = num_extents;
- }
-
- extra_blocks = 0;
- }
-
- btree_curs->num_tot_blocks = blocks_allocated_pt;
- btree_curs->num_free_blocks = blocks_allocated_pt;
-
- *extents = num_extents;
-
- return(extra_blocks);
-}
-
/* Map btnum to buffer ops for the types that need it. */
static const struct xfs_buf_ops *
btnum_to_ops(
@@ -837,268 +567,202 @@ btnum_to_ops(
}
}
-static void
-prop_freespace_cursor(xfs_mount_t *mp, xfs_agnumber_t agno,
- bt_status_t *btree_curs, xfs_agblock_t startblock,
- xfs_extlen_t blockcount, int level, xfs_btnum_t btnum)
-{
- struct xfs_btree_block *bt_hdr;
- xfs_alloc_key_t *bt_key;
- xfs_alloc_ptr_t *bt_ptr;
- xfs_agblock_t agbno;
- bt_stat_level_t *lptr;
- const struct xfs_buf_ops *ops = btnum_to_ops(btnum);
- int error;
-
- ASSERT(btnum == XFS_BTNUM_BNO || btnum == XFS_BTNUM_CNT);
-
- level++;
-
- if (level >= btree_curs->num_levels)
- return;
-
- lptr = &btree_curs->level[level];
- bt_hdr = XFS_BUF_TO_BLOCK(lptr->buf_p);
-
- if (be16_to_cpu(bt_hdr->bb_numrecs) == 0) {
- /*
- * only happens once when initializing the
- * left-hand side of the tree.
- */
- prop_freespace_cursor(mp, agno, btree_curs, startblock,
- blockcount, level, btnum);
- }
-
- if (be16_to_cpu(bt_hdr->bb_numrecs) ==
- lptr->num_recs_pb + (lptr->modulo > 0)) {
- /*
- * write out current prev block, grab us a new block,
- * and set the rightsib pointer of current block
- */
-#ifdef XR_BLD_FREE_TRACE
- fprintf(stderr, " %d ", lptr->prev_agbno);
-#endif
- if (lptr->prev_agbno != NULLAGBLOCK) {
- ASSERT(lptr->prev_buf_p != NULL);
- libxfs_buf_mark_dirty(lptr->prev_buf_p);
- libxfs_buf_relse(lptr->prev_buf_p);
- }
- lptr->prev_agbno = lptr->agbno;;
- lptr->prev_buf_p = lptr->buf_p;
- agbno = get_next_blockaddr(agno, level, btree_curs);
-
- bt_hdr->bb_u.s.bb_rightsib = cpu_to_be32(agbno);
-
- error = -libxfs_buf_get(mp->m_dev,
- XFS_AGB_TO_DADDR(mp, agno, agbno),
- XFS_FSB_TO_BB(mp, 1), &lptr->buf_p);
- if (error)
- do_error(
- _("Cannot grab free space btree buffer, err=%d"),
- error);
- lptr->agbno = agbno;
-
- if (lptr->modulo)
- lptr->modulo--;
-
- /*
- * initialize block header
- */
- lptr->buf_p->b_ops = ops;
- bt_hdr = XFS_BUF_TO_BLOCK(lptr->buf_p);
- memset(bt_hdr, 0, mp->m_sb.sb_blocksize);
- libxfs_btree_init_block(mp, lptr->buf_p, btnum, level,
- 0, agno);
-
- bt_hdr->bb_u.s.bb_leftsib = cpu_to_be32(lptr->prev_agbno);
-
- /*
- * propagate extent record for first extent in new block up
- */
- prop_freespace_cursor(mp, agno, btree_curs, startblock,
- blockcount, level, btnum);
- }
- /*
- * add extent info to current block
- */
- be16_add_cpu(&bt_hdr->bb_numrecs, 1);
-
- bt_key = XFS_ALLOC_KEY_ADDR(mp, bt_hdr,
- be16_to_cpu(bt_hdr->bb_numrecs));
- bt_ptr = XFS_ALLOC_PTR_ADDR(mp, bt_hdr,
- be16_to_cpu(bt_hdr->bb_numrecs),
- mp->m_alloc_mxr[1]);
-
- bt_key->ar_startblock = cpu_to_be32(startblock);
- bt_key->ar_blockcount = cpu_to_be32(blockcount);
- *bt_ptr = cpu_to_be32(btree_curs->level[level-1].agbno);
-}
-
/*
- * rebuilds a freespace tree given a cursor and type
- * of tree to build (bno or bcnt). returns the number of free blocks
- * represented by the tree.
+ * Free Space Btrees
+ *
+ * We need to leave some free records in the tree for the corner case of
+ * setting up the AGFL. This may require allocation of blocks, and as
+ * such can require insertion of new records into the tree (e.g. moving
+ * a record in the by-count tree when a long extent is shortened). If we
+ * pack the records into the leaves with no slack space, this requires a
+ * leaf split to occur and a block to be allocated from the free list.
+ * If we don't have any blocks on the free list (because we are setting
+ * it up!), then we fail, and the filesystem will fail with the same
+ * failure at runtime. Hence leave a couple of records slack space in
+ * each block to allow immediate modification of the tree without
+ * requiring splits to be done.
*/
-static xfs_extlen_t
-build_freespace_tree(xfs_mount_t *mp, xfs_agnumber_t agno,
- bt_status_t *btree_curs, xfs_btnum_t btnum)
+
+static void
+init_freespace_cursors(
+ struct repair_ctx *sc,
+ xfs_agnumber_t agno,
+ unsigned int free_space,
+ unsigned int *nr_extents,
+ int *extra_blocks,
+ struct bt_rebuild *btr_bno,
+ struct bt_rebuild *btr_cnt)
{
- xfs_agnumber_t i;
- xfs_agblock_t j;
- struct xfs_btree_block *bt_hdr;
- xfs_alloc_rec_t *bt_rec;
- int level;
- xfs_agblock_t agbno;
- extent_tree_node_t *ext_ptr;
- bt_stat_level_t *lptr;
- xfs_extlen_t freeblks;
- const struct xfs_buf_ops *ops = btnum_to_ops(btnum);
+ unsigned int bno_blocks;
+ unsigned int cnt_blocks;
int error;
- ASSERT(btnum == XFS_BTNUM_BNO || btnum == XFS_BTNUM_CNT);
-
-#ifdef XR_BLD_FREE_TRACE
- fprintf(stderr, "in build_freespace_tree, agno = %d\n", agno);
-#endif
- level = btree_curs->num_levels;
- freeblks = 0;
+ init_rebuild(sc, &XFS_RMAP_OINFO_AG, free_space, btr_bno);
+ init_rebuild(sc, &XFS_RMAP_OINFO_AG, free_space, btr_cnt);
- ASSERT(level > 0);
+ btr_bno->cur = libxfs_allocbt_stage_cursor(sc->mp,
+ &btr_bno->newbt.afake, agno, XFS_BTNUM_BNO);
+ btr_cnt->cur = libxfs_allocbt_stage_cursor(sc->mp,
+ &btr_cnt->newbt.afake, agno, XFS_BTNUM_CNT);
/*
- * initialize the first block on each btree level
+ * Now we need to allocate blocks for the free space btrees using the
+ * free space records we're about to put in them. Every record we use
+ * can change the shape of the free space trees, so we recompute the
+ * btree shape until we stop needing /more/ blocks. If we have any
+ * left over we'll stash them in the AGFL when we're done.
*/
- for (i = 0; i < level; i++) {
- lptr = &btree_curs->level[i];
+ do {
+ unsigned int num_freeblocks;
+
+ bno_blocks = btr_bno->bload.nr_blocks;
+ cnt_blocks = btr_cnt->bload.nr_blocks;
- agbno = get_next_blockaddr(agno, i, btree_curs);
- error = -libxfs_buf_get(mp->m_dev,
- XFS_AGB_TO_DADDR(mp, agno, agbno),
- XFS_FSB_TO_BB(mp, 1), &lptr->buf_p);
+ /* Compute how many bnobt blocks we'll need. */
+ error = -libxfs_btree_bload_compute_geometry(btr_bno->cur,
+ &btr_bno->bload, *nr_extents);
if (error)
do_error(
- _("Cannot grab free space btree buffer, err=%d"),
- error);
-
- if (i == btree_curs->num_levels - 1)
- btree_curs->root = agbno;
-
- lptr->agbno = agbno;
- lptr->prev_agbno = NULLAGBLOCK;
- lptr->prev_buf_p = NULL;
- /*
- * initialize block header
- */
- lptr->buf_p->b_ops = ops;
- bt_hdr = XFS_BUF_TO_BLOCK(lptr->buf_p);
- memset(bt_hdr, 0, mp->m_sb.sb_blocksize);
- libxfs_btree_init_block(mp, lptr->buf_p, btnum, i, 0, agno);
- }
- /*
- * run along leaf, setting up records. as we have to switch
- * blocks, call the prop_freespace_cursor routine to set up the new
- * pointers for the parent. that can recurse up to the root
- * if required. set the sibling pointers for leaf level here.
- */
- if (btnum == XFS_BTNUM_BNO)
- ext_ptr = findfirst_bno_extent(agno);
- else
- ext_ptr = findfirst_bcnt_extent(agno);
-
-#ifdef XR_BLD_FREE_TRACE
- fprintf(stderr, "bft, agno = %d, start = %u, count = %u\n",
- agno, ext_ptr->ex_startblock, ext_ptr->ex_blockcount);
-#endif
-
- lptr = &btree_curs->level[0];
-
- for (i = 0; i < btree_curs->level[0].num_blocks; i++) {
- /*
- * block initialization, lay in block header
- */
- lptr->buf_p->b_ops = ops;
- bt_hdr = XFS_BUF_TO_BLOCK(lptr->buf_p);
- memset(bt_hdr, 0, mp->m_sb.sb_blocksize);
- libxfs_btree_init_block(mp, lptr->buf_p, btnum, 0, 0, agno);
-
- bt_hdr->bb_u.s.bb_leftsib = cpu_to_be32(lptr->prev_agbno);
- bt_hdr->bb_numrecs = cpu_to_be16(lptr->num_recs_pb +
- (lptr->modulo > 0));
-#ifdef XR_BLD_FREE_TRACE
- fprintf(stderr, "bft, bb_numrecs = %d\n",
- be16_to_cpu(bt_hdr->bb_numrecs));
-#endif
-
- if (lptr->modulo > 0)
- lptr->modulo--;
-
- /*
- * initialize values in the path up to the root if
- * this is a multi-level btree
- */
- if (btree_curs->num_levels > 1)
- prop_freespace_cursor(mp, agno, btree_curs,
- ext_ptr->ex_startblock,
- ext_ptr->ex_blockcount,
- 0, btnum);
-
- bt_rec = (xfs_alloc_rec_t *)
- ((char *)bt_hdr + XFS_ALLOC_BLOCK_LEN(mp));
- for (j = 0; j < be16_to_cpu(bt_hdr->bb_numrecs); j++) {
- ASSERT(ext_ptr != NULL);
- bt_rec[j].ar_startblock = cpu_to_be32(
- ext_ptr->ex_startblock);
- bt_rec[j].ar_blockcount = cpu_to_be32(
- ext_ptr->ex_blockcount);
- freeblks += ext_ptr->ex_blockcount;
- if (btnum == XFS_BTNUM_BNO)
- ext_ptr = findnext_bno_extent(ext_ptr);
- else
- ext_ptr = findnext_bcnt_extent(agno, ext_ptr);
-#if 0
-#ifdef XR_BLD_FREE_TRACE
- if (ext_ptr == NULL)
- fprintf(stderr, "null extent pointer, j = %d\n",
- j);
- else
- fprintf(stderr,
- "bft, agno = %d, start = %u, count = %u\n",
- agno, ext_ptr->ex_startblock,
- ext_ptr->ex_blockcount);
-#endif
-#endif
- }
-
- if (ext_ptr != NULL) {
- /*
- * get next leaf level block
- */
- if (lptr->prev_buf_p != NULL) {
-#ifdef XR_BLD_FREE_TRACE
- fprintf(stderr, " writing fst agbno %u\n",
- lptr->prev_agbno);
-#endif
- ASSERT(lptr->prev_agbno != NULLAGBLOCK);
- libxfs_buf_mark_dirty(lptr->prev_buf_p);
- libxfs_buf_relse(lptr->prev_buf_p);
- }
- lptr->prev_buf_p = lptr->buf_p;
- lptr->prev_agbno = lptr->agbno;
- lptr->agbno = get_next_blockaddr(agno, 0, btree_curs);
- bt_hdr->bb_u.s.bb_rightsib = cpu_to_be32(lptr->agbno);
-
- error = -libxfs_buf_get(mp->m_dev,
- XFS_AGB_TO_DADDR(mp, agno, lptr->agbno),
- XFS_FSB_TO_BB(mp, 1),
- &lptr->buf_p);
- if (error)
- do_error(
- _("Cannot grab free space btree buffer, err=%d"),
- error);
- }
- }
-
- return(freeblks);
+_("Unable to compute free space by block btree geometry, error %d.\n"), -error);
+
+ /* Compute how many cntbt blocks we'll need. */
+ error = -libxfs_btree_bload_compute_geometry(btr_bno->cur,
+ &btr_cnt->bload, *nr_extents);
+ if (error)
+ do_error(
+_("Unable to compute free space by length btree geometry, error %d.\n"), -error);
+
+ /* We don't need any more blocks, so we're done. */
+ if (bno_blocks >= btr_bno->bload.nr_blocks &&
+ cnt_blocks >= btr_cnt->bload.nr_blocks)
+ break;
+
+ /* Allocate however many more blocks we need this time. */
+ if (bno_blocks < btr_bno->bload.nr_blocks)
+ setup_rebuild(sc->mp, agno, btr_bno,
+ btr_bno->bload.nr_blocks - bno_blocks);
+ if (cnt_blocks < btr_cnt->bload.nr_blocks)
+ setup_rebuild(sc->mp, agno, btr_cnt,
+ btr_cnt->bload.nr_blocks - cnt_blocks);
+
+ /* Ok, now how many free space records do we have? */
+ *nr_extents = count_bno_extents_blocks(agno, &num_freeblocks);
+ } while (1);
+
+ *extra_blocks = (bno_blocks - btr_bno->bload.nr_blocks) +
+ (cnt_blocks - btr_cnt->bload.nr_blocks);
+}
+
+static void
+get_freesp_data(
+ struct xfs_btree_cur *cur,
+ struct extent_tree_node *bno_rec,
+ xfs_agblock_t *freeblks)
+{
+ struct xfs_alloc_rec_incore *arec = &cur->bc_rec.a;
+
+ arec->ar_startblock = bno_rec->ex_startblock;
+ arec->ar_blockcount = bno_rec->ex_blockcount;
+ if (freeblks)
+ *freeblks += bno_rec->ex_blockcount;
+}
+
+/* Grab one bnobt record. */
+static int
+get_bnobt_record(
+ struct xfs_btree_cur *cur,
+ void *priv)
+{
+ struct bt_rebuild *btr = priv;
+
+ get_freesp_data(cur, btr->bno_rec, btr->freeblks);
+ btr->bno_rec = findnext_bno_extent(btr->bno_rec);
+ return 0;
+}
+
+/* Rebuild a free space by block number btree. */
+static void
+build_bnobt(
+ struct repair_ctx *sc,
+ xfs_agnumber_t agno,
+ struct bt_rebuild *btr_bno,
+ xfs_agblock_t *freeblks)
+{
+ int error;
+
+ *freeblks = 0;
+ btr_bno->bload.get_record = get_bnobt_record;
+ btr_bno->bload.claim_block = rebuild_claim_block;
+ btr_bno->bno_rec = findfirst_bno_extent(agno);
+ btr_bno->freeblks = freeblks;
+
+ error = -libxfs_trans_alloc_empty(sc->mp, &sc->tp);
+ if (error)
+ do_error(
+_("Insufficient memory to construct bnobt rebuild transaction.\n"));
+
+ /* Add all observed bnobt records. */
+ error = -libxfs_btree_bload(btr_bno->cur, &btr_bno->bload, btr_bno);
+ if (error)
+ do_error(
+_("Error %d while creating bnobt btree for AG %u.\n"), error, agno);
+
+ /* Since we're not writing the AGF yet, no need to commit the cursor */
+ libxfs_btree_del_cursor(btr_bno->cur, 0);
+ error = -libxfs_trans_commit(sc->tp);
+ if (error)
+ do_error(
+_("Error %d while writing bnobt btree for AG %u.\n"), error, agno);
+ sc->tp = NULL;
+}
+
+/* Grab one cntbt record. */
+static int
+get_cntbt_record(
+ struct xfs_btree_cur *cur,
+ void *priv)
+{
+ struct bt_rebuild *btr = priv;
+
+ get_freesp_data(cur, btr->bno_rec, btr->freeblks);
+ btr->bno_rec = findnext_bcnt_extent(cur->bc_ag.agno, btr->bno_rec);
+ return 0;
+}
+
+/* Rebuild a freespace by count btree. */
+static void
+build_cntbt(
+ struct repair_ctx *sc,
+ xfs_agnumber_t agno,
+ struct bt_rebuild *btr_cnt,
+ xfs_agblock_t *freeblks)
+{
+ int error;
+
+ *freeblks = 0;
+ btr_cnt->bload.get_record = get_cntbt_record;
+ btr_cnt->bload.claim_block = rebuild_claim_block;
+ btr_cnt->bno_rec = findfirst_bcnt_extent(agno);
+ btr_cnt->freeblks = freeblks;
+
+ error = -libxfs_trans_alloc_empty(sc->mp, &sc->tp);
+ if (error)
+ do_error(
+_("Insufficient memory to construct cntbt rebuild transaction.\n"));
+
+ /* Add all observed cntbt records. */
+ error = -libxfs_btree_bload(btr_cnt->cur, &btr_cnt->bload, btr_cnt);
+ if (error)
+ do_error(
+_("Error %d while creating cntbt btree for AG %u.\n"), error, agno);
+
+ /* Since we're not writing the AGF yet, no need to commit the cursor */
+ libxfs_btree_del_cursor(btr_cnt->cur, 0);
+ error = -libxfs_trans_commit(sc->tp);
+ if (error)
+ do_error(
+_("Error %d while writing cntbt btree for AG %u.\n"), error, agno);
+ sc->tp = NULL;
}
/*
@@ -2232,6 +1896,27 @@ _("Insufficient memory to construct refcount cursor."));
free_slab_cursor(&refc_cur);
}
+/* Fill the AGFL with any leftover bnobt rebuilder blocks. */
+static void
+fill_agfl(
+ struct bt_rebuild *btr,
+ __be32 *agfl_bnos,
+ int *i)
+{
+ struct xrep_newbt_resv *resv, *n;
+ struct xfs_mount *mp = btr->newbt.sc->mp;
+
+ for_each_xrep_newbt_reservation(&btr->newbt, resv, n) {
+ xfs_agblock_t bno;
+
+ bno = XFS_FSB_TO_AGBNO(mp, resv->fsbno + resv->used);
+ while (resv->used < resv->len && (*i) < libxfs_agfl_size(mp)) {
+ agfl_bnos[(*i)++] = cpu_to_be32(bno++);
+ resv->used++;
+ }
+ }
+}
+
/*
* build both the agf and the agfl for an agno given both
* btree cursors.
@@ -2242,8 +1927,8 @@ static void
build_agf_agfl(
struct xfs_mount *mp,
xfs_agnumber_t agno,
- struct bt_status *bno_bt,
- struct bt_status *bcnt_bt,
+ struct bt_rebuild *btr_bno,
+ struct bt_rebuild *btr_cnt,
xfs_extlen_t freeblks, /* # free blocks in tree */
int lostblocks, /* # blocks that will be lost */
struct bt_status *rmap_bt,
@@ -2255,7 +1940,6 @@ build_agf_agfl(
int i;
struct xfs_agfl *agfl;
struct xfs_agf *agf;
- xfs_fsblock_t fsb;
__be32 *freelist;
int error;
@@ -2287,10 +1971,14 @@ build_agf_agfl(
agf->agf_length = cpu_to_be32(mp->m_sb.sb_dblocks -
(xfs_rfsblock_t) mp->m_sb.sb_agblocks * agno);
- agf->agf_roots[XFS_BTNUM_BNO] = cpu_to_be32(bno_bt->root);
- agf->agf_levels[XFS_BTNUM_BNO] = cpu_to_be32(bno_bt->num_levels);
- agf->agf_roots[XFS_BTNUM_CNT] = cpu_to_be32(bcnt_bt->root);
- agf->agf_levels[XFS_BTNUM_CNT] = cpu_to_be32(bcnt_bt->num_levels);
+ agf->agf_roots[XFS_BTNUM_BNO] =
+ cpu_to_be32(btr_bno->newbt.afake.af_root);
+ agf->agf_levels[XFS_BTNUM_BNO] =
+ cpu_to_be32(btr_bno->newbt.afake.af_levels);
+ agf->agf_roots[XFS_BTNUM_CNT] =
+ cpu_to_be32(btr_cnt->newbt.afake.af_root);
+ agf->agf_levels[XFS_BTNUM_CNT] =
+ cpu_to_be32(btr_cnt->newbt.afake.af_levels);
agf->agf_roots[XFS_BTNUM_RMAP] = cpu_to_be32(rmap_bt->root);
agf->agf_levels[XFS_BTNUM_RMAP] = cpu_to_be32(rmap_bt->num_levels);
agf->agf_freeblks = cpu_to_be32(freeblks);
@@ -2310,9 +1998,8 @@ build_agf_agfl(
* Don't count the root blocks as they are already
* accounted for.
*/
- blks = (bno_bt->num_tot_blocks - bno_bt->num_free_blocks) +
- (bcnt_bt->num_tot_blocks - bcnt_bt->num_free_blocks) -
- 2;
+ blks = btr_bno->newbt.afake.af_blocks +
+ btr_cnt->newbt.afake.af_blocks - 2;
if (xfs_sb_version_hasrmapbt(&mp->m_sb))
blks += rmap_bt->num_tot_blocks - rmap_bt->num_free_blocks - 1;
agf->agf_btreeblks = cpu_to_be32(blks);
@@ -2354,48 +2041,14 @@ build_agf_agfl(
freelist[i] = cpu_to_be32(NULLAGBLOCK);
}
- /*
- * do we have left-over blocks in the btree cursors that should
- * be used to fill the AGFL?
- */
- if (bno_bt->num_free_blocks > 0 || bcnt_bt->num_free_blocks > 0) {
- /*
- * yes, now grab as many blocks as we can
- */
- i = 0;
- while (bno_bt->num_free_blocks > 0 && i < libxfs_agfl_size(mp))
- {
- freelist[i] = cpu_to_be32(
- get_next_blockaddr(agno, 0, bno_bt));
- i++;
- }
-
- while (bcnt_bt->num_free_blocks > 0 && i < libxfs_agfl_size(mp))
- {
- freelist[i] = cpu_to_be32(
- get_next_blockaddr(agno, 0, bcnt_bt));
- i++;
- }
- /*
- * now throw the rest of the blocks away and complain
- */
- while (bno_bt->num_free_blocks > 0) {
- fsb = XFS_AGB_TO_FSB(mp, agno,
- get_next_blockaddr(agno, 0, bno_bt));
- error = slab_add(lost_fsb, &fsb);
- if (error)
- do_error(
-_("Insufficient memory saving lost blocks.\n"));
- }
- while (bcnt_bt->num_free_blocks > 0) {
- fsb = XFS_AGB_TO_FSB(mp, agno,
- get_next_blockaddr(agno, 0, bcnt_bt));
- error = slab_add(lost_fsb, &fsb);
- if (error)
- do_error(
-_("Insufficient memory saving lost blocks.\n"));
- }
+ /* Fill the AGFL with leftover blocks or save them for later. */
+ i = 0;
+ freelist = xfs_buf_to_agfl_bno(agfl_buf);
+ fill_agfl(btr_bno, freelist, &i);
+ fill_agfl(btr_cnt, freelist, &i);
+ /* Set the AGF counters for the AGFL. */
+ if (i > 0) {
agf->agf_flfirst = 0;
agf->agf_fllast = cpu_to_be32(i - 1);
agf->agf_flcount = cpu_to_be32(i);
@@ -2493,8 +2146,8 @@ phase5_func(
uint64_t num_free_inos;
uint64_t finobt_num_inos;
uint64_t finobt_num_free_inos;
- bt_status_t bno_btree_curs;
- bt_status_t bcnt_btree_curs;
+ struct bt_rebuild btr_bno;
+ struct bt_rebuild btr_cnt;
bt_status_t ino_btree_curs;
bt_status_t fino_btree_curs;
bt_status_t rmap_btree_curs;
@@ -2502,9 +2155,7 @@ phase5_func(
int extra_blocks = 0;
uint num_freeblocks;
xfs_extlen_t freeblks1;
-#ifdef DEBUG
xfs_extlen_t freeblks2;
-#endif
xfs_agblock_t num_extents;
if (verbose)
@@ -2513,7 +2164,7 @@ phase5_func(
/*
* build up incore bno and bcnt extent btrees
*/
- num_extents = mk_incore_fstree(mp, agno);
+ num_extents = mk_incore_fstree(mp, agno, &num_freeblocks);
#ifdef XR_BLD_FREE_TRACE
fprintf(stderr, "# of bno extents is %d\n", count_bno_extents(agno));
@@ -2585,8 +2236,8 @@ _("unable to rebuild AG %u. Not enough free space in on-disk AG.\n"),
/*
* track blocks that we might really lose
*/
- extra_blocks = calculate_freespace_cursor(mp, agno,
- &num_extents, &bno_btree_curs);
+ init_freespace_cursors(&sc, agno, num_freeblocks, &num_extents,
+ &extra_blocks, &btr_bno, &btr_cnt);
/*
* freespace btrees live in the "free space" but the filesystem treats
@@ -2603,36 +2254,18 @@ _("unable to rebuild AG %u. Not enough free space in on-disk AG.\n"),
if (extra_blocks > 0)
sb_fdblocks_ag[agno] -= extra_blocks;
- bcnt_btree_curs = bno_btree_curs;
-
- bno_btree_curs.owner = XFS_RMAP_OWN_AG;
- bcnt_btree_curs.owner = XFS_RMAP_OWN_AG;
- setup_cursor(mp, agno, &bno_btree_curs);
- setup_cursor(mp, agno, &bcnt_btree_curs);
-
#ifdef XR_BLD_FREE_TRACE
fprintf(stderr, "# of bno extents is %d\n", count_bno_extents(agno));
fprintf(stderr, "# of bcnt extents is %d\n", count_bcnt_extents(agno));
#endif
- /*
- * now rebuild the freespace trees
- */
- freeblks1 = build_freespace_tree(mp, agno,
- &bno_btree_curs, XFS_BTNUM_BNO);
+ /* Rebuild the freespace btrees. */
+ build_bnobt(&sc, agno, &btr_bno, &freeblks1);
+ build_cntbt(&sc, agno, &btr_cnt, &freeblks2);
+
#ifdef XR_BLD_FREE_TRACE
- fprintf(stderr, "# of free blocks == %d\n", freeblks1);
+ fprintf(stderr, "# of free blocks == %d/%d\n", freeblks1, freeblks2);
#endif
- write_cursor(&bno_btree_curs);
-
-#ifdef DEBUG
- freeblks2 = build_freespace_tree(mp, agno,
- &bcnt_btree_curs, XFS_BTNUM_CNT);
-#else
- (void) build_freespace_tree(mp, agno, &bcnt_btree_curs, XFS_BTNUM_CNT);
-#endif
- write_cursor(&bcnt_btree_curs);
-
ASSERT(freeblks1 == freeblks2);
if (xfs_sb_version_hasrmapbt(&mp->m_sb)) {
@@ -2650,9 +2283,9 @@ _("unable to rebuild AG %u. Not enough free space in on-disk AG.\n"),
/*
* set up agf and agfl
*/
- build_agf_agfl(mp, agno, &bno_btree_curs,
- &bcnt_btree_curs, freeblks1, extra_blocks,
- &rmap_btree_curs, &refcnt_btree_curs, lost_fsb);
+ build_agf_agfl(mp, agno, &btr_bno, &btr_cnt, freeblks1, extra_blocks,
+ &rmap_btree_curs, &refcnt_btree_curs, lost_fsb);
+
/*
* build inode allocation tree.
*/
@@ -2674,15 +2307,14 @@ _("unable to rebuild AG %u. Not enough free space in on-disk AG.\n"),
/*
* tear down cursors
*/
- finish_cursor(&bno_btree_curs);
- finish_cursor(&ino_btree_curs);
+ finish_rebuild(mp, &btr_bno, lost_fsb);
+ finish_rebuild(mp, &btr_cnt, lost_fsb);
if (xfs_sb_version_hasrmapbt(&mp->m_sb))
finish_cursor(&rmap_btree_curs);
if (xfs_sb_version_hasreflink(&mp->m_sb))
finish_cursor(&refcnt_btree_curs);
if (xfs_sb_version_hasfinobt(&mp->m_sb))
finish_cursor(&fino_btree_curs);
- finish_cursor(&bcnt_btree_curs);
/*
* release the incore per-AG bno/bcnt trees so the extent nodes
^ permalink raw reply related [flat|nested] 30+ messages in thread
* [PATCH 5/9] xfs_repair: rebuild inode btrees with bulk loader
2020-05-20 1:50 [PATCH v5 0/9] xfs_repair: use btree bulk loading Darrick J. Wong
` (3 preceding siblings ...)
2020-05-20 1:51 ` [PATCH 4/9] xfs_repair: rebuild free space btrees with bulk loader Darrick J. Wong
@ 2020-05-20 1:51 ` Darrick J. Wong
2020-05-28 15:11 ` Brian Foster
2020-05-20 1:51 ` [PATCH 6/9] xfs_repair: rebuild reverse mapping " Darrick J. Wong
` (3 subsequent siblings)
8 siblings, 1 reply; 30+ messages in thread
From: Darrick J. Wong @ 2020-05-20 1:51 UTC (permalink / raw)
To: sandeen, darrick.wong; +Cc: linux-xfs, bfoster
From: Darrick J. Wong <darrick.wong@oracle.com>
Use the btree bulk loading functions to rebuild the inode btrees
and drop the open-coded implementation.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
---
libxfs/libxfs_api_defs.h | 1
repair/phase5.c | 642 +++++++++++++++++-----------------------------
2 files changed, 240 insertions(+), 403 deletions(-)
diff --git a/libxfs/libxfs_api_defs.h b/libxfs/libxfs_api_defs.h
index bace739c..5d0868c2 100644
--- a/libxfs/libxfs_api_defs.h
+++ b/libxfs/libxfs_api_defs.h
@@ -115,6 +115,7 @@
#define xfs_init_local_fork libxfs_init_local_fork
#define xfs_inobt_maxrecs libxfs_inobt_maxrecs
+#define xfs_inobt_stage_cursor libxfs_inobt_stage_cursor
#define xfs_inode_from_disk libxfs_inode_from_disk
#define xfs_inode_to_disk libxfs_inode_to_disk
#define xfs_inode_validate_cowextsize libxfs_inode_validate_cowextsize
diff --git a/repair/phase5.c b/repair/phase5.c
index e69b042c..38f30753 100644
--- a/repair/phase5.c
+++ b/repair/phase5.c
@@ -84,6 +84,10 @@ struct bt_rebuild {
struct extent_tree_node *bno_rec;
xfs_agblock_t *freeblks;
};
+ struct {
+ struct ino_tree_node *ino_rec;
+ struct agi_stat *agi_stat;
+ };
};
};
@@ -372,6 +376,11 @@ estimate_ag_bload_slack(
bload->node_slack = 0;
}
+static inline void skip_rebuild(struct bt_rebuild *btr)
+{
+ memset(btr, 0, sizeof(struct bt_rebuild));
+}
+
/* Initialize a btree rebuild context. */
static void
init_rebuild(
@@ -765,48 +774,38 @@ _("Error %d while writing cntbt btree for AG %u.\n"), error, agno);
sc->tp = NULL;
}
-/*
- * XXX(hch): any reason we don't just look at mp->m_inobt_mxr?
- */
-#define XR_INOBT_BLOCK_MAXRECS(mp, level) \
- libxfs_inobt_maxrecs((mp), (mp)->m_sb.sb_blocksize, \
- (level) == 0)
+/* Inode Btrees */
-/*
- * we don't have to worry here about how chewing up free extents
- * may perturb things because inode tree building happens before
- * freespace tree building.
- */
+/* Initialize both inode btree cursors as needed. */
static void
-init_ino_cursor(xfs_mount_t *mp, xfs_agnumber_t agno, bt_status_t *btree_curs,
- uint64_t *num_inos, uint64_t *num_free_inos, int finobt)
+init_ino_cursors(
+ struct repair_ctx *sc,
+ xfs_agnumber_t agno,
+ unsigned int free_space,
+ uint64_t *num_inos,
+ uint64_t *num_free_inos,
+ struct bt_rebuild *btr_ino,
+ struct bt_rebuild *btr_fino)
{
- uint64_t ninos;
- uint64_t nfinos;
- int rec_nfinos;
- int rec_ninos;
- ino_tree_node_t *ino_rec;
- int num_recs;
- int level;
- bt_stat_level_t *lptr;
- bt_stat_level_t *p_lptr;
- xfs_extlen_t blocks_allocated;
- int i;
+ struct ino_tree_node *ino_rec;
+ unsigned int ino_recs = 0;
+ unsigned int fino_recs = 0;
+ bool finobt;
+ int error;
- *num_inos = *num_free_inos = 0;
- ninos = nfinos = 0;
+ finobt = xfs_sb_version_hasfinobt(&sc->mp->m_sb);
+ init_rebuild(sc, &XFS_RMAP_OINFO_INOBT, free_space, btr_ino);
- lptr = &btree_curs->level[0];
- btree_curs->init = 1;
- btree_curs->owner = XFS_RMAP_OWN_INOBT;
+ /* Compute inode statistics. */
+ *num_free_inos = 0;
+ *num_inos = 0;
+ for (ino_rec = findfirst_inode_rec(agno);
+ ino_rec != NULL;
+ ino_rec = next_ino_rec(ino_rec)) {
+ unsigned int rec_ninos = 0;
+ unsigned int rec_nfinos = 0;
+ int i;
- /*
- * build up statistics
- */
- ino_rec = findfirst_inode_rec(agno);
- for (num_recs = 0; ino_rec != NULL; ino_rec = next_ino_rec(ino_rec)) {
- rec_ninos = 0;
- rec_nfinos = 0;
for (i = 0; i < XFS_INODES_PER_CHUNK; i++) {
ASSERT(is_inode_confirmed(ino_rec, i));
/*
@@ -820,174 +819,220 @@ init_ino_cursor(xfs_mount_t *mp, xfs_agnumber_t agno, bt_status_t *btree_curs,
rec_ninos++;
}
- /*
- * finobt only considers records with free inodes
- */
- if (finobt && !rec_nfinos)
+ *num_free_inos += rec_nfinos;
+ *num_inos += rec_ninos;
+ ino_recs++;
+
+ /* finobt only considers records with free inodes */
+ if (rec_nfinos)
+ fino_recs++;
+ }
+
+ btr_ino->cur = libxfs_inobt_stage_cursor(sc->mp, &btr_ino->newbt.afake,
+ agno, XFS_BTNUM_INO);
+
+ /* Compute how many inobt blocks we'll need. */
+ error = -libxfs_btree_bload_compute_geometry(btr_ino->cur,
+ &btr_ino->bload, ino_recs);
+ if (error)
+ do_error(
+_("Unable to compute inode btree geometry, error %d.\n"), error);
+
+ setup_rebuild(sc->mp, agno, btr_ino, btr_ino->bload.nr_blocks);
+
+ if (!finobt) {
+ skip_rebuild(btr_fino);
+ return;
+ }
+
+ init_rebuild(sc, &XFS_RMAP_OINFO_INOBT, free_space, btr_fino);
+ btr_fino->cur = libxfs_inobt_stage_cursor(sc->mp,
+ &btr_fino->newbt.afake, agno, XFS_BTNUM_FINO);
+
+ /* Compute how many finobt blocks we'll need. */
+ error = -libxfs_btree_bload_compute_geometry(btr_fino->cur,
+ &btr_fino->bload, fino_recs);
+ if (error)
+ do_error(
+_("Unable to compute free inode btree geometry, error %d.\n"), error);
+
+ setup_rebuild(sc->mp, agno, btr_fino, btr_fino->bload.nr_blocks);
+}
+
+/* Copy one incore inode record into the inobt cursor. */
+static void
+get_inode_data(
+ struct xfs_btree_cur *cur,
+ struct ino_tree_node *ino_rec,
+ struct agi_stat *agi_stat)
+{
+ struct xfs_inobt_rec_incore *irec = &cur->bc_rec.i;
+ int inocnt = 0;
+ int finocnt = 0;
+ int k;
+
+ irec->ir_startino = ino_rec->ino_startnum;
+ irec->ir_free = ino_rec->ir_free;
+
+ for (k = 0; k < sizeof(xfs_inofree_t) * NBBY; k++) {
+ ASSERT(is_inode_confirmed(ino_rec, k));
+
+ if (is_inode_sparse(ino_rec, k))
continue;
-
- nfinos += rec_nfinos;
- ninos += rec_ninos;
- num_recs++;
+ if (is_inode_free(ino_rec, k))
+ finocnt++;
+ inocnt++;
}
- if (num_recs == 0) {
- /*
- * easy corner-case -- no inode records
- */
- lptr->num_blocks = 1;
- lptr->modulo = 0;
- lptr->num_recs_pb = 0;
- lptr->num_recs_tot = 0;
-
- btree_curs->num_levels = 1;
- btree_curs->num_tot_blocks = btree_curs->num_free_blocks = 1;
-
- setup_cursor(mp, agno, btree_curs);
+ irec->ir_count = inocnt;
+ irec->ir_freecount = finocnt;
- return;
- }
+ if (xfs_sb_version_hassparseinodes(&cur->bc_mp->m_sb)) {
+ uint64_t sparse;
+ int spmask;
+ uint16_t holemask;
- blocks_allocated = lptr->num_blocks = howmany(num_recs,
- XR_INOBT_BLOCK_MAXRECS(mp, 0));
-
- lptr->modulo = num_recs % lptr->num_blocks;
- lptr->num_recs_pb = num_recs / lptr->num_blocks;
- lptr->num_recs_tot = num_recs;
- level = 1;
-
- if (lptr->num_blocks > 1) {
- for (; btree_curs->level[level-1].num_blocks > 1
- && level < XFS_BTREE_MAXLEVELS;
- level++) {
- lptr = &btree_curs->level[level];
- p_lptr = &btree_curs->level[level - 1];
- lptr->num_blocks = howmany(p_lptr->num_blocks,
- XR_INOBT_BLOCK_MAXRECS(mp, level));
- lptr->modulo = p_lptr->num_blocks % lptr->num_blocks;
- lptr->num_recs_pb = p_lptr->num_blocks
- / lptr->num_blocks;
- lptr->num_recs_tot = p_lptr->num_blocks;
-
- blocks_allocated += lptr->num_blocks;
+ /*
+ * Convert the 64-bit in-core sparse inode state to the
+ * 16-bit on-disk holemask.
+ */
+ holemask = 0;
+ spmask = (1 << XFS_INODES_PER_HOLEMASK_BIT) - 1;
+ sparse = ino_rec->ir_sparse;
+ for (k = 0; k < XFS_INOBT_HOLEMASK_BITS; k++) {
+ if (sparse & spmask) {
+ ASSERT((sparse & spmask) == spmask);
+ holemask |= (1 << k);
+ } else
+ ASSERT((sparse & spmask) == 0);
+ sparse >>= XFS_INODES_PER_HOLEMASK_BIT;
}
+
+ irec->ir_holemask = holemask;
+ } else {
+ irec->ir_holemask = 0;
}
- ASSERT(lptr->num_blocks == 1);
- btree_curs->num_levels = level;
- btree_curs->num_tot_blocks = btree_curs->num_free_blocks
- = blocks_allocated;
+ if (!agi_stat)
+ return;
- setup_cursor(mp, agno, btree_curs);
+ if (agi_stat->first_agino != NULLAGINO)
+ agi_stat->first_agino = ino_rec->ino_startnum;
+ agi_stat->freecount += finocnt;
+ agi_stat->count += inocnt;
+}
- *num_inos = ninos;
- *num_free_inos = nfinos;
+/* Grab one inobt record. */
+static int
+get_inobt_record(
+ struct xfs_btree_cur *cur,
+ void *priv)
+{
+ struct bt_rebuild *rebuild = priv;
- return;
+ get_inode_data(cur, rebuild->ino_rec, rebuild->agi_stat);
+ rebuild->ino_rec = next_ino_rec(rebuild->ino_rec);
+ return 0;
}
+/* Rebuild a inobt btree. */
static void
-prop_ino_cursor(xfs_mount_t *mp, xfs_agnumber_t agno, bt_status_t *btree_curs,
- xfs_btnum_t btnum, xfs_agino_t startino, int level)
+build_inobt(
+ struct repair_ctx *sc,
+ xfs_agnumber_t agno,
+ struct bt_rebuild *btr_ino,
+ struct agi_stat *agi_stat)
{
- struct xfs_btree_block *bt_hdr;
- xfs_inobt_key_t *bt_key;
- xfs_inobt_ptr_t *bt_ptr;
- xfs_agblock_t agbno;
- bt_stat_level_t *lptr;
- const struct xfs_buf_ops *ops = btnum_to_ops(btnum);
int error;
- level++;
-
- if (level >= btree_curs->num_levels)
- return;
-
- lptr = &btree_curs->level[level];
- bt_hdr = XFS_BUF_TO_BLOCK(lptr->buf_p);
-
- if (be16_to_cpu(bt_hdr->bb_numrecs) == 0) {
- /*
- * this only happens once to initialize the
- * first path up the left side of the tree
- * where the agbno's are already set up
- */
- prop_ino_cursor(mp, agno, btree_curs, btnum, startino, level);
- }
-
- if (be16_to_cpu(bt_hdr->bb_numrecs) ==
- lptr->num_recs_pb + (lptr->modulo > 0)) {
- /*
- * write out current prev block, grab us a new block,
- * and set the rightsib pointer of current block
- */
-#ifdef XR_BLD_INO_TRACE
- fprintf(stderr, " ino prop agbno %d ", lptr->prev_agbno);
-#endif
- if (lptr->prev_agbno != NULLAGBLOCK) {
- ASSERT(lptr->prev_buf_p != NULL);
- libxfs_buf_mark_dirty(lptr->prev_buf_p);
- libxfs_buf_relse(lptr->prev_buf_p);
- }
- lptr->prev_agbno = lptr->agbno;;
- lptr->prev_buf_p = lptr->buf_p;
- agbno = get_next_blockaddr(agno, level, btree_curs);
-
- bt_hdr->bb_u.s.bb_rightsib = cpu_to_be32(agbno);
-
- error = -libxfs_buf_get(mp->m_dev,
- XFS_AGB_TO_DADDR(mp, agno, agbno),
- XFS_FSB_TO_BB(mp, 1), &lptr->buf_p);
- if (error)
- do_error(_("Cannot grab inode btree buffer, err=%d"),
- error);
- lptr->agbno = agbno;
+ btr_ino->bload.get_record = get_inobt_record;
+ btr_ino->bload.claim_block = rebuild_claim_block;
+ agi_stat->count = agi_stat->freecount = 0;
+ agi_stat->first_agino = NULLAGINO;
+ btr_ino->agi_stat = agi_stat;
+ btr_ino->ino_rec = findfirst_inode_rec(agno);
+
+ error = -libxfs_trans_alloc_empty(sc->mp, &sc->tp);
+ if (error)
+ do_error(
+_("Insufficient memory to construct inobt rebuild transaction.\n"));
+
+ /* Add all observed inobt records. */
+ error = -libxfs_btree_bload(btr_ino->cur, &btr_ino->bload, btr_ino);
+ if (error)
+ do_error(
+_("Error %d while creating inobt btree for AG %u.\n"), error, agno);
+
+ /* Since we're not writing the AGI yet, no need to commit the cursor */
+ libxfs_btree_del_cursor(btr_ino->cur, 0);
+ error = -libxfs_trans_commit(sc->tp);
+ if (error)
+ do_error(
+_("Error %d while writing inobt btree for AG %u.\n"), error, agno);
+ sc->tp = NULL;
+}
- if (lptr->modulo)
- lptr->modulo--;
+/* Grab one finobt record. */
+static int
+get_finobt_record(
+ struct xfs_btree_cur *cur,
+ void *priv)
+{
+ struct bt_rebuild *rebuild = priv;
- /*
- * initialize block header
- */
- lptr->buf_p->b_ops = ops;
- bt_hdr = XFS_BUF_TO_BLOCK(lptr->buf_p);
- memset(bt_hdr, 0, mp->m_sb.sb_blocksize);
- libxfs_btree_init_block(mp, lptr->buf_p, btnum,
- level, 0, agno);
+ get_inode_data(cur, rebuild->ino_rec, NULL);
+ rebuild->ino_rec = next_free_ino_rec(rebuild->ino_rec);
+ return 0;
+}
- bt_hdr->bb_u.s.bb_leftsib = cpu_to_be32(lptr->prev_agbno);
+/* Rebuild a finobt btree. */
+static void
+build_finobt(
+ struct repair_ctx *sc,
+ xfs_agnumber_t agno,
+ struct bt_rebuild *btr_fino)
+{
+ int error;
- /*
- * propagate extent record for first extent in new block up
- */
- prop_ino_cursor(mp, agno, btree_curs, btnum, startino, level);
- }
- /*
- * add inode info to current block
- */
- be16_add_cpu(&bt_hdr->bb_numrecs, 1);
-
- bt_key = XFS_INOBT_KEY_ADDR(mp, bt_hdr,
- be16_to_cpu(bt_hdr->bb_numrecs));
- bt_ptr = XFS_INOBT_PTR_ADDR(mp, bt_hdr,
- be16_to_cpu(bt_hdr->bb_numrecs),
- M_IGEO(mp)->inobt_mxr[1]);
-
- bt_key->ir_startino = cpu_to_be32(startino);
- *bt_ptr = cpu_to_be32(btree_curs->level[level-1].agbno);
+ btr_fino->bload.get_record = get_finobt_record;
+ btr_fino->bload.claim_block = rebuild_claim_block;
+ btr_fino->ino_rec = findfirst_free_inode_rec(agno);
+
+ error = -libxfs_trans_alloc_empty(sc->mp, &sc->tp);
+ if (error)
+ do_error(
+_("Insufficient memory to construct finobt rebuild transaction.\n"));
+
+ /* Add all observed finobt records. */
+ error = -libxfs_btree_bload(btr_fino->cur, &btr_fino->bload, btr_fino);
+ if (error)
+ do_error(
+_("Error %d while creating finobt btree for AG %u.\n"), error, agno);
+
+ /* Since we're not writing the AGI yet, no need to commit the cursor */
+ libxfs_btree_del_cursor(btr_fino->cur, 0);
+ error = -libxfs_trans_commit(sc->tp);
+ if (error)
+ do_error(
+_("Error %d while writing finobt btree for AG %u.\n"), error, agno);
+ sc->tp = NULL;
}
/*
* XXX: yet more code that can be shared with mkfs, growfs.
*/
static void
-build_agi(xfs_mount_t *mp, xfs_agnumber_t agno, bt_status_t *btree_curs,
- bt_status_t *finobt_curs, struct agi_stat *agi_stat)
+build_agi(
+ struct xfs_mount *mp,
+ xfs_agnumber_t agno,
+ struct bt_rebuild *btr_ino,
+ struct bt_rebuild *btr_fino,
+ struct agi_stat *agi_stat)
{
- xfs_buf_t *agi_buf;
- xfs_agi_t *agi;
- int i;
- int error;
+ struct xfs_buf *agi_buf;
+ struct xfs_agi *agi;
+ int i;
+ int error;
error = -libxfs_buf_get(mp->m_dev,
XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
@@ -1008,8 +1053,8 @@ build_agi(xfs_mount_t *mp, xfs_agnumber_t agno, bt_status_t *btree_curs,
agi->agi_length = cpu_to_be32(mp->m_sb.sb_dblocks -
(xfs_rfsblock_t) mp->m_sb.sb_agblocks * agno);
agi->agi_count = cpu_to_be32(agi_stat->count);
- agi->agi_root = cpu_to_be32(btree_curs->root);
- agi->agi_level = cpu_to_be32(btree_curs->num_levels);
+ agi->agi_root = cpu_to_be32(btr_ino->newbt.afake.af_root);
+ agi->agi_level = cpu_to_be32(btr_ino->newbt.afake.af_levels);
agi->agi_freecount = cpu_to_be32(agi_stat->freecount);
agi->agi_newino = cpu_to_be32(agi_stat->first_agino);
agi->agi_dirino = cpu_to_be32(NULLAGINO);
@@ -1021,203 +1066,16 @@ build_agi(xfs_mount_t *mp, xfs_agnumber_t agno, bt_status_t *btree_curs,
platform_uuid_copy(&agi->agi_uuid, &mp->m_sb.sb_meta_uuid);
if (xfs_sb_version_hasfinobt(&mp->m_sb)) {
- agi->agi_free_root = cpu_to_be32(finobt_curs->root);
- agi->agi_free_level = cpu_to_be32(finobt_curs->num_levels);
+ agi->agi_free_root =
+ cpu_to_be32(btr_fino->newbt.afake.af_root);
+ agi->agi_free_level =
+ cpu_to_be32(btr_fino->newbt.afake.af_levels);
}
libxfs_buf_mark_dirty(agi_buf);
libxfs_buf_relse(agi_buf);
}
-/*
- * rebuilds an inode tree given a cursor. We're lazy here and call
- * the routine that builds the agi
- */
-static void
-build_ino_tree(xfs_mount_t *mp, xfs_agnumber_t agno,
- bt_status_t *btree_curs, xfs_btnum_t btnum,
- struct agi_stat *agi_stat)
-{
- xfs_agnumber_t i;
- xfs_agblock_t j;
- xfs_agblock_t agbno;
- xfs_agino_t first_agino;
- struct xfs_btree_block *bt_hdr;
- xfs_inobt_rec_t *bt_rec;
- ino_tree_node_t *ino_rec;
- bt_stat_level_t *lptr;
- const struct xfs_buf_ops *ops = btnum_to_ops(btnum);
- xfs_agino_t count = 0;
- xfs_agino_t freecount = 0;
- int inocnt;
- uint8_t finocnt;
- int k;
- int level = btree_curs->num_levels;
- int spmask;
- uint64_t sparse;
- uint16_t holemask;
- int error;
-
- ASSERT(btnum == XFS_BTNUM_INO || btnum == XFS_BTNUM_FINO);
-
- for (i = 0; i < level; i++) {
- lptr = &btree_curs->level[i];
-
- agbno = get_next_blockaddr(agno, i, btree_curs);
- error = -libxfs_buf_get(mp->m_dev,
- XFS_AGB_TO_DADDR(mp, agno, agbno),
- XFS_FSB_TO_BB(mp, 1), &lptr->buf_p);
- if (error)
- do_error(_("Cannot grab inode btree buffer, err=%d"),
- error);
-
- if (i == btree_curs->num_levels - 1)
- btree_curs->root = agbno;
-
- lptr->agbno = agbno;
- lptr->prev_agbno = NULLAGBLOCK;
- lptr->prev_buf_p = NULL;
- /*
- * initialize block header
- */
-
- lptr->buf_p->b_ops = ops;
- bt_hdr = XFS_BUF_TO_BLOCK(lptr->buf_p);
- memset(bt_hdr, 0, mp->m_sb.sb_blocksize);
- libxfs_btree_init_block(mp, lptr->buf_p, btnum, i, 0, agno);
- }
-
- /*
- * run along leaf, setting up records. as we have to switch
- * blocks, call the prop_ino_cursor routine to set up the new
- * pointers for the parent. that can recurse up to the root
- * if required. set the sibling pointers for leaf level here.
- */
- if (btnum == XFS_BTNUM_FINO)
- ino_rec = findfirst_free_inode_rec(agno);
- else
- ino_rec = findfirst_inode_rec(agno);
-
- if (ino_rec != NULL)
- first_agino = ino_rec->ino_startnum;
- else
- first_agino = NULLAGINO;
-
- lptr = &btree_curs->level[0];
-
- for (i = 0; i < lptr->num_blocks; i++) {
- /*
- * block initialization, lay in block header
- */
- lptr->buf_p->b_ops = ops;
- bt_hdr = XFS_BUF_TO_BLOCK(lptr->buf_p);
- memset(bt_hdr, 0, mp->m_sb.sb_blocksize);
- libxfs_btree_init_block(mp, lptr->buf_p, btnum, 0, 0, agno);
-
- bt_hdr->bb_u.s.bb_leftsib = cpu_to_be32(lptr->prev_agbno);
- bt_hdr->bb_numrecs = cpu_to_be16(lptr->num_recs_pb +
- (lptr->modulo > 0));
-
- if (lptr->modulo > 0)
- lptr->modulo--;
-
- if (lptr->num_recs_pb > 0)
- prop_ino_cursor(mp, agno, btree_curs, btnum,
- ino_rec->ino_startnum, 0);
-
- bt_rec = (xfs_inobt_rec_t *)
- ((char *)bt_hdr + XFS_INOBT_BLOCK_LEN(mp));
- for (j = 0; j < be16_to_cpu(bt_hdr->bb_numrecs); j++) {
- ASSERT(ino_rec != NULL);
- bt_rec[j].ir_startino =
- cpu_to_be32(ino_rec->ino_startnum);
- bt_rec[j].ir_free = cpu_to_be64(ino_rec->ir_free);
-
- inocnt = finocnt = 0;
- for (k = 0; k < sizeof(xfs_inofree_t)*NBBY; k++) {
- ASSERT(is_inode_confirmed(ino_rec, k));
-
- if (is_inode_sparse(ino_rec, k))
- continue;
- if (is_inode_free(ino_rec, k))
- finocnt++;
- inocnt++;
- }
-
- /*
- * Set the freecount and check whether we need to update
- * the sparse format fields. Otherwise, skip to the next
- * record.
- */
- inorec_set_freecount(mp, &bt_rec[j], finocnt);
- if (!xfs_sb_version_hassparseinodes(&mp->m_sb))
- goto nextrec;
-
- /*
- * Convert the 64-bit in-core sparse inode state to the
- * 16-bit on-disk holemask.
- */
- holemask = 0;
- spmask = (1 << XFS_INODES_PER_HOLEMASK_BIT) - 1;
- sparse = ino_rec->ir_sparse;
- for (k = 0; k < XFS_INOBT_HOLEMASK_BITS; k++) {
- if (sparse & spmask) {
- ASSERT((sparse & spmask) == spmask);
- holemask |= (1 << k);
- } else
- ASSERT((sparse & spmask) == 0);
- sparse >>= XFS_INODES_PER_HOLEMASK_BIT;
- }
-
- bt_rec[j].ir_u.sp.ir_count = inocnt;
- bt_rec[j].ir_u.sp.ir_holemask = cpu_to_be16(holemask);
-
-nextrec:
- freecount += finocnt;
- count += inocnt;
-
- if (btnum == XFS_BTNUM_FINO)
- ino_rec = next_free_ino_rec(ino_rec);
- else
- ino_rec = next_ino_rec(ino_rec);
- }
-
- if (ino_rec != NULL) {
- /*
- * get next leaf level block
- */
- if (lptr->prev_buf_p != NULL) {
-#ifdef XR_BLD_INO_TRACE
- fprintf(stderr, "writing inobt agbno %u\n",
- lptr->prev_agbno);
-#endif
- ASSERT(lptr->prev_agbno != NULLAGBLOCK);
- libxfs_buf_mark_dirty(lptr->prev_buf_p);
- libxfs_buf_relse(lptr->prev_buf_p);
- }
- lptr->prev_buf_p = lptr->buf_p;
- lptr->prev_agbno = lptr->agbno;
- lptr->agbno = get_next_blockaddr(agno, 0, btree_curs);
- bt_hdr->bb_u.s.bb_rightsib = cpu_to_be32(lptr->agbno);
-
- error = -libxfs_buf_get(mp->m_dev,
- XFS_AGB_TO_DADDR(mp, agno, lptr->agbno),
- XFS_FSB_TO_BB(mp, 1),
- &lptr->buf_p);
- if (error)
- do_error(
- _("Cannot grab inode btree buffer, err=%d"),
- error);
- }
- }
-
- if (agi_stat) {
- agi_stat->first_agino = first_agino;
- agi_stat->count = count;
- agi_stat->freecount = freecount;
- }
-}
-
/* rebuild the rmap tree */
/*
@@ -2142,14 +2000,10 @@ phase5_func(
{
struct repair_ctx sc = { .mp = mp, };
struct agi_stat agi_stat = {0,};
- uint64_t num_inos;
- uint64_t num_free_inos;
- uint64_t finobt_num_inos;
- uint64_t finobt_num_free_inos;
struct bt_rebuild btr_bno;
struct bt_rebuild btr_cnt;
- bt_status_t ino_btree_curs;
- bt_status_t fino_btree_curs;
+ struct bt_rebuild btr_ino;
+ struct bt_rebuild btr_fino;
bt_status_t rmap_btree_curs;
bt_status_t refcnt_btree_curs;
int extra_blocks = 0;
@@ -2184,19 +2038,8 @@ _("unable to rebuild AG %u. Not enough free space in on-disk AG.\n"),
agno);
}
- /*
- * ok, now set up the btree cursors for the on-disk btrees (includes
- * pre-allocating all required blocks for the trees themselves)
- */
- init_ino_cursor(mp, agno, &ino_btree_curs, &num_inos,
- &num_free_inos, 0);
-
- if (xfs_sb_version_hasfinobt(&mp->m_sb))
- init_ino_cursor(mp, agno, &fino_btree_curs, &finobt_num_inos,
- &finobt_num_free_inos, 1);
-
- sb_icount_ag[agno] += num_inos;
- sb_ifree_ag[agno] += num_free_inos;
+ init_ino_cursors(&sc, agno, num_freeblocks, &sb_icount_ag[agno],
+ &sb_ifree_ag[agno], &btr_ino, &btr_fino);
/*
* Set up the btree cursors for the on-disk rmap btrees, which includes
@@ -2287,34 +2130,27 @@ _("unable to rebuild AG %u. Not enough free space in on-disk AG.\n"),
&rmap_btree_curs, &refcnt_btree_curs, lost_fsb);
/*
- * build inode allocation tree.
+ * build inode allocation trees.
*/
- build_ino_tree(mp, agno, &ino_btree_curs, XFS_BTNUM_INO, &agi_stat);
- write_cursor(&ino_btree_curs);
-
- /*
- * build free inode tree
- */
- if (xfs_sb_version_hasfinobt(&mp->m_sb)) {
- build_ino_tree(mp, agno, &fino_btree_curs,
- XFS_BTNUM_FINO, NULL);
- write_cursor(&fino_btree_curs);
- }
+ build_inobt(&sc, agno, &btr_ino, &agi_stat);
+ if (xfs_sb_version_hasfinobt(&mp->m_sb))
+ build_finobt(&sc, agno, &btr_fino);
/* build the agi */
- build_agi(mp, agno, &ino_btree_curs, &fino_btree_curs, &agi_stat);
+ build_agi(mp, agno, &btr_ino, &btr_fino, &agi_stat);
/*
* tear down cursors
*/
finish_rebuild(mp, &btr_bno, lost_fsb);
finish_rebuild(mp, &btr_cnt, lost_fsb);
+ finish_rebuild(mp, &btr_ino, lost_fsb);
+ if (xfs_sb_version_hasfinobt(&mp->m_sb))
+ finish_rebuild(mp, &btr_fino, lost_fsb);
if (xfs_sb_version_hasrmapbt(&mp->m_sb))
finish_cursor(&rmap_btree_curs);
if (xfs_sb_version_hasreflink(&mp->m_sb))
finish_cursor(&refcnt_btree_curs);
- if (xfs_sb_version_hasfinobt(&mp->m_sb))
- finish_cursor(&fino_btree_curs);
/*
* release the incore per-AG bno/bcnt trees so the extent nodes
^ permalink raw reply related [flat|nested] 30+ messages in thread
* [PATCH 6/9] xfs_repair: rebuild reverse mapping btrees with bulk loader
2020-05-20 1:50 [PATCH v5 0/9] xfs_repair: use btree bulk loading Darrick J. Wong
` (4 preceding siblings ...)
2020-05-20 1:51 ` [PATCH 5/9] xfs_repair: rebuild inode " Darrick J. Wong
@ 2020-05-20 1:51 ` Darrick J. Wong
2020-05-20 1:51 ` [PATCH 7/9] xfs_repair: rebuild refcount " Darrick J. Wong
` (2 subsequent siblings)
8 siblings, 0 replies; 30+ messages in thread
From: Darrick J. Wong @ 2020-05-20 1:51 UTC (permalink / raw)
To: sandeen, darrick.wong; +Cc: linux-xfs, bfoster
From: Darrick J. Wong <darrick.wong@oracle.com>
Use the btree bulk loading functions to rebuild the reverse mapping
btrees and drop the open-coded implementation.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
---
libxfs/libxfs_api_defs.h | 1
repair/phase5.c | 425 +++++++---------------------------------------
2 files changed, 70 insertions(+), 356 deletions(-)
diff --git a/libxfs/libxfs_api_defs.h b/libxfs/libxfs_api_defs.h
index 5d0868c2..0026ca45 100644
--- a/libxfs/libxfs_api_defs.h
+++ b/libxfs/libxfs_api_defs.h
@@ -142,6 +142,7 @@
#define xfs_rmapbt_calc_reserves libxfs_rmapbt_calc_reserves
#define xfs_rmapbt_init_cursor libxfs_rmapbt_init_cursor
#define xfs_rmapbt_maxrecs libxfs_rmapbt_maxrecs
+#define xfs_rmapbt_stage_cursor libxfs_rmapbt_stage_cursor
#define xfs_rmap_compare libxfs_rmap_compare
#define xfs_rmap_get_rec libxfs_rmap_get_rec
#define xfs_rmap_irec_offset_pack libxfs_rmap_irec_offset_pack
diff --git a/repair/phase5.c b/repair/phase5.c
index 38f30753..bb32fd4f 100644
--- a/repair/phase5.c
+++ b/repair/phase5.c
@@ -1078,373 +1078,84 @@ build_agi(
/* rebuild the rmap tree */
-/*
- * we don't have to worry here about how chewing up free extents
- * may perturb things because rmap tree building happens before
- * freespace tree building.
- */
+/* Set up the rmap rebuild parameters. */
static void
init_rmapbt_cursor(
- struct xfs_mount *mp,
+ struct repair_ctx *sc,
xfs_agnumber_t agno,
- struct bt_status *btree_curs)
+ unsigned int free_space,
+ struct bt_rebuild *btr)
{
- size_t num_recs;
- int level;
- struct bt_stat_level *lptr;
- struct bt_stat_level *p_lptr;
- xfs_extlen_t blocks_allocated;
- int maxrecs;
-
- if (!xfs_sb_version_hasrmapbt(&mp->m_sb)) {
- memset(btree_curs, 0, sizeof(struct bt_status));
- return;
- }
-
- lptr = &btree_curs->level[0];
- btree_curs->init = 1;
- btree_curs->owner = XFS_RMAP_OWN_AG;
-
- /*
- * build up statistics
- */
- num_recs = rmap_record_count(mp, agno);
- if (num_recs == 0) {
- /*
- * easy corner-case -- no rmap records
- */
- lptr->num_blocks = 1;
- lptr->modulo = 0;
- lptr->num_recs_pb = 0;
- lptr->num_recs_tot = 0;
-
- btree_curs->num_levels = 1;
- btree_curs->num_tot_blocks = btree_curs->num_free_blocks = 1;
-
- setup_cursor(mp, agno, btree_curs);
-
- return;
- }
-
- /*
- * Leave enough slack in the rmapbt that we can insert the
- * metadata AG entries without too many splits.
- */
- maxrecs = mp->m_rmap_mxr[0];
- if (num_recs > maxrecs)
- maxrecs -= 10;
- blocks_allocated = lptr->num_blocks = howmany(num_recs, maxrecs);
-
- lptr->modulo = num_recs % lptr->num_blocks;
- lptr->num_recs_pb = num_recs / lptr->num_blocks;
- lptr->num_recs_tot = num_recs;
- level = 1;
-
- if (lptr->num_blocks > 1) {
- for (; btree_curs->level[level-1].num_blocks > 1
- && level < XFS_BTREE_MAXLEVELS;
- level++) {
- lptr = &btree_curs->level[level];
- p_lptr = &btree_curs->level[level - 1];
- lptr->num_blocks = howmany(p_lptr->num_blocks,
- mp->m_rmap_mxr[1]);
- lptr->modulo = p_lptr->num_blocks % lptr->num_blocks;
- lptr->num_recs_pb = p_lptr->num_blocks
- / lptr->num_blocks;
- lptr->num_recs_tot = p_lptr->num_blocks;
-
- blocks_allocated += lptr->num_blocks;
- }
- }
- ASSERT(lptr->num_blocks == 1);
- btree_curs->num_levels = level;
-
- btree_curs->num_tot_blocks = btree_curs->num_free_blocks
- = blocks_allocated;
-
- setup_cursor(mp, agno, btree_curs);
-}
-
-static void
-prop_rmap_cursor(
- struct xfs_mount *mp,
- xfs_agnumber_t agno,
- struct bt_status *btree_curs,
- struct xfs_rmap_irec *rm_rec,
- int level)
-{
- struct xfs_btree_block *bt_hdr;
- struct xfs_rmap_key *bt_key;
- xfs_rmap_ptr_t *bt_ptr;
- xfs_agblock_t agbno;
- struct bt_stat_level *lptr;
- const struct xfs_buf_ops *ops = btnum_to_ops(XFS_BTNUM_RMAP);
int error;
- level++;
-
- if (level >= btree_curs->num_levels)
+ if (!xfs_sb_version_hasrmapbt(&sc->mp->m_sb)) {
+ skip_rebuild(btr);
return;
-
- lptr = &btree_curs->level[level];
- bt_hdr = XFS_BUF_TO_BLOCK(lptr->buf_p);
-
- if (be16_to_cpu(bt_hdr->bb_numrecs) == 0) {
- /*
- * this only happens once to initialize the
- * first path up the left side of the tree
- * where the agbno's are already set up
- */
- prop_rmap_cursor(mp, agno, btree_curs, rm_rec, level);
- }
-
- if (be16_to_cpu(bt_hdr->bb_numrecs) ==
- lptr->num_recs_pb + (lptr->modulo > 0)) {
- /*
- * write out current prev block, grab us a new block,
- * and set the rightsib pointer of current block
- */
-#ifdef XR_BLD_INO_TRACE
- fprintf(stderr, " rmap prop agbno %d ", lptr->prev_agbno);
-#endif
- if (lptr->prev_agbno != NULLAGBLOCK) {
- ASSERT(lptr->prev_buf_p != NULL);
- libxfs_buf_mark_dirty(lptr->prev_buf_p);
- libxfs_buf_relse(lptr->prev_buf_p);
- }
- lptr->prev_agbno = lptr->agbno;
- lptr->prev_buf_p = lptr->buf_p;
- agbno = get_next_blockaddr(agno, level, btree_curs);
-
- bt_hdr->bb_u.s.bb_rightsib = cpu_to_be32(agbno);
-
- error = -libxfs_buf_get(mp->m_dev,
- XFS_AGB_TO_DADDR(mp, agno, agbno),
- XFS_FSB_TO_BB(mp, 1), &lptr->buf_p);
- if (error)
- do_error(_("Cannot grab rmapbt buffer, err=%d"),
- error);
- lptr->agbno = agbno;
-
- if (lptr->modulo)
- lptr->modulo--;
-
- /*
- * initialize block header
- */
- lptr->buf_p->b_ops = ops;
- bt_hdr = XFS_BUF_TO_BLOCK(lptr->buf_p);
- memset(bt_hdr, 0, mp->m_sb.sb_blocksize);
- libxfs_btree_init_block(mp, lptr->buf_p, XFS_BTNUM_RMAP,
- level, 0, agno);
-
- bt_hdr->bb_u.s.bb_leftsib = cpu_to_be32(lptr->prev_agbno);
-
- /*
- * propagate extent record for first extent in new block up
- */
- prop_rmap_cursor(mp, agno, btree_curs, rm_rec, level);
}
- /*
- * add rmap info to current block
- */
- be16_add_cpu(&bt_hdr->bb_numrecs, 1);
- bt_key = XFS_RMAP_KEY_ADDR(bt_hdr,
- be16_to_cpu(bt_hdr->bb_numrecs));
- bt_ptr = XFS_RMAP_PTR_ADDR(bt_hdr,
- be16_to_cpu(bt_hdr->bb_numrecs),
- mp->m_rmap_mxr[1]);
+ init_rebuild(sc, &XFS_RMAP_OINFO_AG, free_space, btr);
+ btr->cur = libxfs_rmapbt_stage_cursor(sc->mp, &btr->newbt.afake, agno);
- bt_key->rm_startblock = cpu_to_be32(rm_rec->rm_startblock);
- bt_key->rm_owner = cpu_to_be64(rm_rec->rm_owner);
- bt_key->rm_offset = cpu_to_be64(rm_rec->rm_offset);
+ /* Compute how many blocks we'll need. */
+ error = -libxfs_btree_bload_compute_geometry(btr->cur, &btr->bload,
+ rmap_record_count(sc->mp, agno));
+ if (error)
+ do_error(
+_("Unable to compute rmap btree geometry, error %d.\n"), error);
- *bt_ptr = cpu_to_be32(btree_curs->level[level-1].agbno);
+ setup_rebuild(sc->mp, agno, btr, btr->bload.nr_blocks);
}
-static void
-prop_rmap_highkey(
- struct xfs_mount *mp,
- xfs_agnumber_t agno,
- struct bt_status *btree_curs,
- struct xfs_rmap_irec *rm_highkey)
+/* Grab one rmap record. */
+static int
+get_rmapbt_record(
+ struct xfs_btree_cur *cur,
+ void *priv)
{
- struct xfs_btree_block *bt_hdr;
- struct xfs_rmap_key *bt_key;
- struct bt_stat_level *lptr;
- struct xfs_rmap_irec key = {0};
- struct xfs_rmap_irec high_key;
- int level;
- int i;
- int numrecs;
+ struct xfs_rmap_irec *rec;
+ struct bt_rebuild *btr = priv;
- high_key = *rm_highkey;
- for (level = 1; level < btree_curs->num_levels; level++) {
- lptr = &btree_curs->level[level];
- bt_hdr = XFS_BUF_TO_BLOCK(lptr->buf_p);
- numrecs = be16_to_cpu(bt_hdr->bb_numrecs);
- bt_key = XFS_RMAP_HIGH_KEY_ADDR(bt_hdr, numrecs);
-
- bt_key->rm_startblock = cpu_to_be32(high_key.rm_startblock);
- bt_key->rm_owner = cpu_to_be64(high_key.rm_owner);
- bt_key->rm_offset = cpu_to_be64(
- libxfs_rmap_irec_offset_pack(&high_key));
-
- for (i = 1; i <= numrecs; i++) {
- bt_key = XFS_RMAP_HIGH_KEY_ADDR(bt_hdr, i);
- key.rm_startblock = be32_to_cpu(bt_key->rm_startblock);
- key.rm_owner = be64_to_cpu(bt_key->rm_owner);
- key.rm_offset = be64_to_cpu(bt_key->rm_offset);
- if (rmap_diffkeys(&key, &high_key) > 0)
- high_key = key;
- }
- }
+ rec = pop_slab_cursor(btr->slab_cursor);
+ memcpy(&cur->bc_rec.r, rec, sizeof(struct xfs_rmap_irec));
+ return 0;
}
-/*
- * rebuilds a rmap btree given a cursor.
- */
+/* Rebuild a rmap btree. */
static void
build_rmap_tree(
- struct xfs_mount *mp,
+ struct repair_ctx *sc,
xfs_agnumber_t agno,
- struct bt_status *btree_curs)
+ struct bt_rebuild *btr)
{
- xfs_agnumber_t i;
- xfs_agblock_t j;
- xfs_agblock_t agbno;
- struct xfs_btree_block *bt_hdr;
- struct xfs_rmap_irec *rm_rec;
- struct xfs_slab_cursor *rmap_cur;
- struct xfs_rmap_rec *bt_rec;
- struct xfs_rmap_irec highest_key = {0};
- struct xfs_rmap_irec hi_key = {0};
- struct bt_stat_level *lptr;
- const struct xfs_buf_ops *ops = btnum_to_ops(XFS_BTNUM_RMAP);
- int numrecs;
- int level = btree_curs->num_levels;
int error;
- highest_key.rm_flags = 0;
- for (i = 0; i < level; i++) {
- lptr = &btree_curs->level[i];
+ btr->bload.get_record = get_rmapbt_record;
+ btr->bload.claim_block = rebuild_claim_block;
- agbno = get_next_blockaddr(agno, i, btree_curs);
- error = -libxfs_buf_get(mp->m_dev,
- XFS_AGB_TO_DADDR(mp, agno, agbno),
- XFS_FSB_TO_BB(mp, 1), &lptr->buf_p);
- if (error)
- do_error(_("Cannot grab rmapbt buffer, err=%d"),
- error);
-
- if (i == btree_curs->num_levels - 1)
- btree_curs->root = agbno;
-
- lptr->agbno = agbno;
- lptr->prev_agbno = NULLAGBLOCK;
- lptr->prev_buf_p = NULL;
- /*
- * initialize block header
- */
-
- lptr->buf_p->b_ops = ops;
- bt_hdr = XFS_BUF_TO_BLOCK(lptr->buf_p);
- memset(bt_hdr, 0, mp->m_sb.sb_blocksize);
- libxfs_btree_init_block(mp, lptr->buf_p, XFS_BTNUM_RMAP,
- i, 0, agno);
- }
-
- /*
- * run along leaf, setting up records. as we have to switch
- * blocks, call the prop_rmap_cursor routine to set up the new
- * pointers for the parent. that can recurse up to the root
- * if required. set the sibling pointers for leaf level here.
- */
- error = rmap_init_cursor(agno, &rmap_cur);
+ error = -libxfs_trans_alloc_empty(sc->mp, &sc->tp);
if (error)
do_error(
-_("Insufficient memory to construct reverse-map cursor."));
- rm_rec = pop_slab_cursor(rmap_cur);
- lptr = &btree_curs->level[0];
-
- for (i = 0; i < lptr->num_blocks; i++) {
- numrecs = lptr->num_recs_pb + (lptr->modulo > 0);
- ASSERT(rm_rec != NULL || numrecs == 0);
-
- /*
- * block initialization, lay in block header
- */
- lptr->buf_p->b_ops = ops;
- bt_hdr = XFS_BUF_TO_BLOCK(lptr->buf_p);
- memset(bt_hdr, 0, mp->m_sb.sb_blocksize);
- libxfs_btree_init_block(mp, lptr->buf_p, XFS_BTNUM_RMAP,
- 0, 0, agno);
-
- bt_hdr->bb_u.s.bb_leftsib = cpu_to_be32(lptr->prev_agbno);
- bt_hdr->bb_numrecs = cpu_to_be16(numrecs);
-
- if (lptr->modulo > 0)
- lptr->modulo--;
-
- if (lptr->num_recs_pb > 0) {
- ASSERT(rm_rec != NULL);
- prop_rmap_cursor(mp, agno, btree_curs, rm_rec, 0);
- }
+_("Insufficient memory to construct rmap rebuild transaction.\n"));
- bt_rec = (struct xfs_rmap_rec *)
- ((char *)bt_hdr + XFS_RMAP_BLOCK_LEN);
- highest_key.rm_startblock = 0;
- highest_key.rm_owner = 0;
- highest_key.rm_offset = 0;
- for (j = 0; j < be16_to_cpu(bt_hdr->bb_numrecs); j++) {
- ASSERT(rm_rec != NULL);
- bt_rec[j].rm_startblock =
- cpu_to_be32(rm_rec->rm_startblock);
- bt_rec[j].rm_blockcount =
- cpu_to_be32(rm_rec->rm_blockcount);
- bt_rec[j].rm_owner = cpu_to_be64(rm_rec->rm_owner);
- bt_rec[j].rm_offset = cpu_to_be64(
- libxfs_rmap_irec_offset_pack(rm_rec));
- rmap_high_key_from_rec(rm_rec, &hi_key);
- if (rmap_diffkeys(&hi_key, &highest_key) > 0)
- highest_key = hi_key;
-
- rm_rec = pop_slab_cursor(rmap_cur);
- }
-
- /* Now go set the parent key */
- prop_rmap_highkey(mp, agno, btree_curs, &highest_key);
+ error = rmap_init_cursor(agno, &btr->slab_cursor);
+ if (error)
+ do_error(
+_("Insufficient memory to construct rmap cursor.\n"));
- if (rm_rec != NULL) {
- /*
- * get next leaf level block
- */
- if (lptr->prev_buf_p != NULL) {
-#ifdef XR_BLD_RL_TRACE
- fprintf(stderr, "writing rmapbt agbno %u\n",
- lptr->prev_agbno);
-#endif
- ASSERT(lptr->prev_agbno != NULLAGBLOCK);
- libxfs_buf_mark_dirty(lptr->prev_buf_p);
- libxfs_buf_relse(lptr->prev_buf_p);
- }
- lptr->prev_buf_p = lptr->buf_p;
- lptr->prev_agbno = lptr->agbno;
- lptr->agbno = get_next_blockaddr(agno, 0, btree_curs);
- bt_hdr->bb_u.s.bb_rightsib = cpu_to_be32(lptr->agbno);
+ /* Add all observed rmap records. */
+ error = -libxfs_btree_bload(btr->cur, &btr->bload, btr);
+ if (error)
+ do_error(
+_("Error %d while creating rmap btree for AG %u.\n"), error, agno);
- error = -libxfs_buf_get(mp->m_dev,
- XFS_AGB_TO_DADDR(mp, agno, lptr->agbno),
- XFS_FSB_TO_BB(mp, 1),
- &lptr->buf_p);
- if (error)
- do_error(
- _("Cannot grab rmapbt buffer, err=%d"),
- error);
- }
- }
- free_slab_cursor(&rmap_cur);
+ /* Since we're not writing the AGF yet, no need to commit the cursor */
+ libxfs_btree_del_cursor(btr->cur, 0);
+ free_slab_cursor(&btr->slab_cursor);
+ error = -libxfs_trans_commit(sc->tp);
+ if (error)
+ do_error(
+_("Error %d while writing rmap btree for AG %u.\n"), error, agno);
+ sc->tp = NULL;
}
/* rebuild the refcount tree */
@@ -1789,7 +1500,7 @@ build_agf_agfl(
struct bt_rebuild *btr_cnt,
xfs_extlen_t freeblks, /* # free blocks in tree */
int lostblocks, /* # blocks that will be lost */
- struct bt_status *rmap_bt,
+ struct bt_rebuild *btr_rmap,
struct bt_status *refcnt_bt,
struct xfs_slab *lost_fsb)
{
@@ -1837,11 +1548,17 @@ build_agf_agfl(
cpu_to_be32(btr_cnt->newbt.afake.af_root);
agf->agf_levels[XFS_BTNUM_CNT] =
cpu_to_be32(btr_cnt->newbt.afake.af_levels);
- agf->agf_roots[XFS_BTNUM_RMAP] = cpu_to_be32(rmap_bt->root);
- agf->agf_levels[XFS_BTNUM_RMAP] = cpu_to_be32(rmap_bt->num_levels);
agf->agf_freeblks = cpu_to_be32(freeblks);
- agf->agf_rmap_blocks = cpu_to_be32(rmap_bt->num_tot_blocks -
- rmap_bt->num_free_blocks);
+
+ if (xfs_sb_version_hasrmapbt(&mp->m_sb)) {
+ agf->agf_roots[XFS_BTNUM_RMAP] =
+ cpu_to_be32(btr_rmap->newbt.afake.af_root);
+ agf->agf_levels[XFS_BTNUM_RMAP] =
+ cpu_to_be32(btr_rmap->newbt.afake.af_levels);
+ agf->agf_rmap_blocks =
+ cpu_to_be32(btr_rmap->newbt.afake.af_blocks);
+ }
+
agf->agf_refcount_root = cpu_to_be32(refcnt_bt->root);
agf->agf_refcount_level = cpu_to_be32(refcnt_bt->num_levels);
agf->agf_refcount_blocks = cpu_to_be32(refcnt_bt->num_tot_blocks -
@@ -1859,7 +1576,7 @@ build_agf_agfl(
blks = btr_bno->newbt.afake.af_blocks +
btr_cnt->newbt.afake.af_blocks - 2;
if (xfs_sb_version_hasrmapbt(&mp->m_sb))
- blks += rmap_bt->num_tot_blocks - rmap_bt->num_free_blocks - 1;
+ blks += btr_rmap->newbt.afake.af_blocks - 1;
agf->agf_btreeblks = cpu_to_be32(blks);
#ifdef XR_BLD_FREE_TRACE
fprintf(stderr, "agf->agf_btreeblks = %u\n",
@@ -1904,6 +1621,8 @@ build_agf_agfl(
freelist = xfs_buf_to_agfl_bno(agfl_buf);
fill_agfl(btr_bno, freelist, &i);
fill_agfl(btr_cnt, freelist, &i);
+ if (xfs_sb_version_hasrmapbt(&mp->m_sb))
+ fill_agfl(btr_rmap, freelist, &i);
/* Set the AGF counters for the AGFL. */
if (i > 0) {
@@ -2004,7 +1723,7 @@ phase5_func(
struct bt_rebuild btr_cnt;
struct bt_rebuild btr_ino;
struct bt_rebuild btr_fino;
- bt_status_t rmap_btree_curs;
+ struct bt_rebuild btr_rmap;
bt_status_t refcnt_btree_curs;
int extra_blocks = 0;
uint num_freeblocks;
@@ -2041,11 +1760,7 @@ _("unable to rebuild AG %u. Not enough free space in on-disk AG.\n"),
init_ino_cursors(&sc, agno, num_freeblocks, &sb_icount_ag[agno],
&sb_ifree_ag[agno], &btr_ino, &btr_fino);
- /*
- * Set up the btree cursors for the on-disk rmap btrees, which includes
- * pre-allocating all required blocks.
- */
- init_rmapbt_cursor(mp, agno, &rmap_btree_curs);
+ init_rmapbt_cursor(&sc, agno, num_freeblocks, &btr_rmap);
/*
* Set up the btree cursors for the on-disk refcount btrees,
@@ -2112,10 +1827,8 @@ _("unable to rebuild AG %u. Not enough free space in on-disk AG.\n"),
ASSERT(freeblks1 == freeblks2);
if (xfs_sb_version_hasrmapbt(&mp->m_sb)) {
- build_rmap_tree(mp, agno, &rmap_btree_curs);
- write_cursor(&rmap_btree_curs);
- sb_fdblocks_ag[agno] += (rmap_btree_curs.num_tot_blocks -
- rmap_btree_curs.num_free_blocks) - 1;
+ build_rmap_tree(&sc, agno, &btr_rmap);
+ sb_fdblocks_ag[agno] += btr_rmap.newbt.afake.af_blocks - 1;
}
if (xfs_sb_version_hasreflink(&mp->m_sb)) {
@@ -2127,7 +1840,7 @@ _("unable to rebuild AG %u. Not enough free space in on-disk AG.\n"),
* set up agf and agfl
*/
build_agf_agfl(mp, agno, &btr_bno, &btr_cnt, freeblks1, extra_blocks,
- &rmap_btree_curs, &refcnt_btree_curs, lost_fsb);
+ &btr_rmap, &refcnt_btree_curs, lost_fsb);
/*
* build inode allocation trees.
@@ -2148,7 +1861,7 @@ _("unable to rebuild AG %u. Not enough free space in on-disk AG.\n"),
if (xfs_sb_version_hasfinobt(&mp->m_sb))
finish_rebuild(mp, &btr_fino, lost_fsb);
if (xfs_sb_version_hasrmapbt(&mp->m_sb))
- finish_cursor(&rmap_btree_curs);
+ finish_rebuild(mp, &btr_rmap, lost_fsb);
if (xfs_sb_version_hasreflink(&mp->m_sb))
finish_cursor(&refcnt_btree_curs);
^ permalink raw reply related [flat|nested] 30+ messages in thread
* [PATCH 7/9] xfs_repair: rebuild refcount btrees with bulk loader
2020-05-20 1:50 [PATCH v5 0/9] xfs_repair: use btree bulk loading Darrick J. Wong
` (5 preceding siblings ...)
2020-05-20 1:51 ` [PATCH 6/9] xfs_repair: rebuild reverse mapping " Darrick J. Wong
@ 2020-05-20 1:51 ` Darrick J. Wong
2020-05-20 1:51 ` [PATCH 8/9] xfs_repair: remove old btree rebuild support code Darrick J. Wong
2020-05-20 1:51 ` [PATCH 9/9] xfs_repair: track blocks lost during btree construction via extents Darrick J. Wong
8 siblings, 0 replies; 30+ messages in thread
From: Darrick J. Wong @ 2020-05-20 1:51 UTC (permalink / raw)
To: sandeen, darrick.wong; +Cc: linux-xfs, bfoster
From: Darrick J. Wong <darrick.wong@oracle.com>
Use the btree bulk loading functions to rebuild the refcount btrees
and drop the open-coded implementation.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
---
libxfs/libxfs_api_defs.h | 1
repair/phase5.c | 362 +++++++++-------------------------------------
2 files changed, 69 insertions(+), 294 deletions(-)
diff --git a/libxfs/libxfs_api_defs.h b/libxfs/libxfs_api_defs.h
index 0026ca45..1a7cdbf9 100644
--- a/libxfs/libxfs_api_defs.h
+++ b/libxfs/libxfs_api_defs.h
@@ -135,6 +135,7 @@
#define xfs_refcountbt_calc_reserves libxfs_refcountbt_calc_reserves
#define xfs_refcountbt_init_cursor libxfs_refcountbt_init_cursor
#define xfs_refcountbt_maxrecs libxfs_refcountbt_maxrecs
+#define xfs_refcountbt_stage_cursor libxfs_refcountbt_stage_cursor
#define xfs_refcount_get_rec libxfs_refcount_get_rec
#define xfs_refcount_lookup_le libxfs_refcount_lookup_le
diff --git a/repair/phase5.c b/repair/phase5.c
index bb32fd4f..797ad6c1 100644
--- a/repair/phase5.c
+++ b/repair/phase5.c
@@ -1160,309 +1160,85 @@ _("Error %d while writing rmap btree for AG %u.\n"), error, agno);
/* rebuild the refcount tree */
-/*
- * we don't have to worry here about how chewing up free extents
- * may perturb things because reflink tree building happens before
- * freespace tree building.
- */
+/* Set up the refcount rebuild parameters. */
static void
init_refc_cursor(
- struct xfs_mount *mp,
+ struct repair_ctx *sc,
xfs_agnumber_t agno,
- struct bt_status *btree_curs)
+ unsigned int free_space,
+ struct bt_rebuild *btr)
{
- size_t num_recs;
- int level;
- struct bt_stat_level *lptr;
- struct bt_stat_level *p_lptr;
- xfs_extlen_t blocks_allocated;
-
- if (!xfs_sb_version_hasreflink(&mp->m_sb)) {
- memset(btree_curs, 0, sizeof(struct bt_status));
- return;
- }
-
- lptr = &btree_curs->level[0];
- btree_curs->init = 1;
- btree_curs->owner = XFS_RMAP_OWN_REFC;
-
- /*
- * build up statistics
- */
- num_recs = refcount_record_count(mp, agno);
- if (num_recs == 0) {
- /*
- * easy corner-case -- no refcount records
- */
- lptr->num_blocks = 1;
- lptr->modulo = 0;
- lptr->num_recs_pb = 0;
- lptr->num_recs_tot = 0;
-
- btree_curs->num_levels = 1;
- btree_curs->num_tot_blocks = btree_curs->num_free_blocks = 1;
-
- setup_cursor(mp, agno, btree_curs);
-
- return;
- }
-
- blocks_allocated = lptr->num_blocks = howmany(num_recs,
- mp->m_refc_mxr[0]);
-
- lptr->modulo = num_recs % lptr->num_blocks;
- lptr->num_recs_pb = num_recs / lptr->num_blocks;
- lptr->num_recs_tot = num_recs;
- level = 1;
-
- if (lptr->num_blocks > 1) {
- for (; btree_curs->level[level-1].num_blocks > 1
- && level < XFS_BTREE_MAXLEVELS;
- level++) {
- lptr = &btree_curs->level[level];
- p_lptr = &btree_curs->level[level - 1];
- lptr->num_blocks = howmany(p_lptr->num_blocks,
- mp->m_refc_mxr[1]);
- lptr->modulo = p_lptr->num_blocks % lptr->num_blocks;
- lptr->num_recs_pb = p_lptr->num_blocks
- / lptr->num_blocks;
- lptr->num_recs_tot = p_lptr->num_blocks;
-
- blocks_allocated += lptr->num_blocks;
- }
- }
- ASSERT(lptr->num_blocks == 1);
- btree_curs->num_levels = level;
-
- btree_curs->num_tot_blocks = btree_curs->num_free_blocks
- = blocks_allocated;
-
- setup_cursor(mp, agno, btree_curs);
-}
-
-static void
-prop_refc_cursor(
- struct xfs_mount *mp,
- xfs_agnumber_t agno,
- struct bt_status *btree_curs,
- xfs_agblock_t startbno,
- int level)
-{
- struct xfs_btree_block *bt_hdr;
- struct xfs_refcount_key *bt_key;
- xfs_refcount_ptr_t *bt_ptr;
- xfs_agblock_t agbno;
- struct bt_stat_level *lptr;
- const struct xfs_buf_ops *ops = btnum_to_ops(XFS_BTNUM_REFC);
int error;
- level++;
-
- if (level >= btree_curs->num_levels)
+ if (!xfs_sb_version_hasreflink(&sc->mp->m_sb)) {
+ skip_rebuild(btr);
return;
-
- lptr = &btree_curs->level[level];
- bt_hdr = XFS_BUF_TO_BLOCK(lptr->buf_p);
-
- if (be16_to_cpu(bt_hdr->bb_numrecs) == 0) {
- /*
- * this only happens once to initialize the
- * first path up the left side of the tree
- * where the agbno's are already set up
- */
- prop_refc_cursor(mp, agno, btree_curs, startbno, level);
}
- if (be16_to_cpu(bt_hdr->bb_numrecs) ==
- lptr->num_recs_pb + (lptr->modulo > 0)) {
- /*
- * write out current prev block, grab us a new block,
- * and set the rightsib pointer of current block
- */
-#ifdef XR_BLD_INO_TRACE
- fprintf(stderr, " ino prop agbno %d ", lptr->prev_agbno);
-#endif
- if (lptr->prev_agbno != NULLAGBLOCK) {
- ASSERT(lptr->prev_buf_p != NULL);
- libxfs_buf_mark_dirty(lptr->prev_buf_p);
- libxfs_buf_relse(lptr->prev_buf_p);
- }
- lptr->prev_agbno = lptr->agbno;
- lptr->prev_buf_p = lptr->buf_p;
- agbno = get_next_blockaddr(agno, level, btree_curs);
-
- bt_hdr->bb_u.s.bb_rightsib = cpu_to_be32(agbno);
-
- error = -libxfs_buf_get(mp->m_dev,
- XFS_AGB_TO_DADDR(mp, agno, agbno),
- XFS_FSB_TO_BB(mp, 1), &lptr->buf_p);
- if (error)
- do_error(_("Cannot grab refcountbt buffer, err=%d"),
- error);
- lptr->agbno = agbno;
-
- if (lptr->modulo)
- lptr->modulo--;
+ init_rebuild(sc, &XFS_RMAP_OINFO_REFC, free_space, btr);
+ btr->cur = libxfs_refcountbt_stage_cursor(sc->mp, &btr->newbt.afake,
+ agno);
- /*
- * initialize block header
- */
- lptr->buf_p->b_ops = ops;
- bt_hdr = XFS_BUF_TO_BLOCK(lptr->buf_p);
- memset(bt_hdr, 0, mp->m_sb.sb_blocksize);
- libxfs_btree_init_block(mp, lptr->buf_p, XFS_BTNUM_REFC,
- level, 0, agno);
+ /* Compute how many blocks we'll need. */
+ error = -libxfs_btree_bload_compute_geometry(btr->cur, &btr->bload,
+ refcount_record_count(sc->mp, agno));
+ if (error)
+ do_error(
+_("Unable to compute refcount btree geometry, error %d.\n"), error);
- bt_hdr->bb_u.s.bb_leftsib = cpu_to_be32(lptr->prev_agbno);
-
- /*
- * propagate extent record for first extent in new block up
- */
- prop_refc_cursor(mp, agno, btree_curs, startbno, level);
- }
- /*
- * add inode info to current block
- */
- be16_add_cpu(&bt_hdr->bb_numrecs, 1);
-
- bt_key = XFS_REFCOUNT_KEY_ADDR(bt_hdr,
- be16_to_cpu(bt_hdr->bb_numrecs));
- bt_ptr = XFS_REFCOUNT_PTR_ADDR(bt_hdr,
- be16_to_cpu(bt_hdr->bb_numrecs),
- mp->m_refc_mxr[1]);
+ setup_rebuild(sc->mp, agno, btr, btr->bload.nr_blocks);
+}
- bt_key->rc_startblock = cpu_to_be32(startbno);
- *bt_ptr = cpu_to_be32(btree_curs->level[level-1].agbno);
+/* Grab one refcount record. */
+static int
+get_refcountbt_record(
+ struct xfs_btree_cur *cur,
+ void *priv)
+{
+ struct xfs_refcount_irec *rec;
+ struct bt_rebuild *btr = priv;
+
+ rec = pop_slab_cursor(btr->slab_cursor);
+ memcpy(&cur->bc_rec.rc, rec, sizeof(struct xfs_refcount_irec));
+ return 0;
}
-/*
- * rebuilds a refcount btree given a cursor.
- */
+/* Rebuild a refcount btree. */
static void
build_refcount_tree(
- struct xfs_mount *mp,
+ struct repair_ctx *sc,
xfs_agnumber_t agno,
- struct bt_status *btree_curs)
+ struct bt_rebuild *btr)
{
- xfs_agnumber_t i;
- xfs_agblock_t j;
- xfs_agblock_t agbno;
- struct xfs_btree_block *bt_hdr;
- struct xfs_refcount_irec *refc_rec;
- struct xfs_slab_cursor *refc_cur;
- struct xfs_refcount_rec *bt_rec;
- struct bt_stat_level *lptr;
- const struct xfs_buf_ops *ops = btnum_to_ops(XFS_BTNUM_REFC);
- int numrecs;
- int level = btree_curs->num_levels;
int error;
- for (i = 0; i < level; i++) {
- lptr = &btree_curs->level[i];
+ btr->bload.get_record = get_refcountbt_record;
+ btr->bload.claim_block = rebuild_claim_block;
- agbno = get_next_blockaddr(agno, i, btree_curs);
- error = -libxfs_buf_get(mp->m_dev,
- XFS_AGB_TO_DADDR(mp, agno, agbno),
- XFS_FSB_TO_BB(mp, 1), &lptr->buf_p);
- if (error)
- do_error(_("Cannot grab refcountbt buffer, err=%d"),
- error);
-
- if (i == btree_curs->num_levels - 1)
- btree_curs->root = agbno;
-
- lptr->agbno = agbno;
- lptr->prev_agbno = NULLAGBLOCK;
- lptr->prev_buf_p = NULL;
- /*
- * initialize block header
- */
-
- lptr->buf_p->b_ops = ops;
- bt_hdr = XFS_BUF_TO_BLOCK(lptr->buf_p);
- memset(bt_hdr, 0, mp->m_sb.sb_blocksize);
- libxfs_btree_init_block(mp, lptr->buf_p, XFS_BTNUM_REFC,
- i, 0, agno);
- }
-
- /*
- * run along leaf, setting up records. as we have to switch
- * blocks, call the prop_refc_cursor routine to set up the new
- * pointers for the parent. that can recurse up to the root
- * if required. set the sibling pointers for leaf level here.
- */
- error = init_refcount_cursor(agno, &refc_cur);
+ error = -libxfs_trans_alloc_empty(sc->mp, &sc->tp);
if (error)
do_error(
-_("Insufficient memory to construct refcount cursor."));
- refc_rec = pop_slab_cursor(refc_cur);
- lptr = &btree_curs->level[0];
-
- for (i = 0; i < lptr->num_blocks; i++) {
- numrecs = lptr->num_recs_pb + (lptr->modulo > 0);
- ASSERT(refc_rec != NULL || numrecs == 0);
-
- /*
- * block initialization, lay in block header
- */
- lptr->buf_p->b_ops = ops;
- bt_hdr = XFS_BUF_TO_BLOCK(lptr->buf_p);
- memset(bt_hdr, 0, mp->m_sb.sb_blocksize);
- libxfs_btree_init_block(mp, lptr->buf_p, XFS_BTNUM_REFC,
- 0, 0, agno);
-
- bt_hdr->bb_u.s.bb_leftsib = cpu_to_be32(lptr->prev_agbno);
- bt_hdr->bb_numrecs = cpu_to_be16(numrecs);
-
- if (lptr->modulo > 0)
- lptr->modulo--;
+_("Insufficient memory to construct refcount rebuild transaction.\n"));
- if (lptr->num_recs_pb > 0)
- prop_refc_cursor(mp, agno, btree_curs,
- refc_rec->rc_startblock, 0);
-
- bt_rec = (struct xfs_refcount_rec *)
- ((char *)bt_hdr + XFS_REFCOUNT_BLOCK_LEN);
- for (j = 0; j < be16_to_cpu(bt_hdr->bb_numrecs); j++) {
- ASSERT(refc_rec != NULL);
- bt_rec[j].rc_startblock =
- cpu_to_be32(refc_rec->rc_startblock);
- bt_rec[j].rc_blockcount =
- cpu_to_be32(refc_rec->rc_blockcount);
- bt_rec[j].rc_refcount = cpu_to_be32(refc_rec->rc_refcount);
-
- refc_rec = pop_slab_cursor(refc_cur);
- }
+ error = init_refcount_cursor(agno, &btr->slab_cursor);
+ if (error)
+ do_error(
+_("Insufficient memory to construct refcount cursor.\n"));
- if (refc_rec != NULL) {
- /*
- * get next leaf level block
- */
- if (lptr->prev_buf_p != NULL) {
-#ifdef XR_BLD_RL_TRACE
- fprintf(stderr, "writing refcntbt agbno %u\n",
- lptr->prev_agbno);
-#endif
- ASSERT(lptr->prev_agbno != NULLAGBLOCK);
- libxfs_buf_mark_dirty(lptr->prev_buf_p);
- libxfs_buf_relse(lptr->prev_buf_p);
- }
- lptr->prev_buf_p = lptr->buf_p;
- lptr->prev_agbno = lptr->agbno;
- lptr->agbno = get_next_blockaddr(agno, 0, btree_curs);
- bt_hdr->bb_u.s.bb_rightsib = cpu_to_be32(lptr->agbno);
+ /* Add all observed refcount records. */
+ error = -libxfs_btree_bload(btr->cur, &btr->bload, btr);
+ if (error)
+ do_error(
+_("Error %d while creating refcount btree for AG %u.\n"), error, agno);
- error = -libxfs_buf_get(mp->m_dev,
- XFS_AGB_TO_DADDR(mp, agno, lptr->agbno),
- XFS_FSB_TO_BB(mp, 1),
- &lptr->buf_p);
- if (error)
- do_error(
- _("Cannot grab refcountbt buffer, err=%d"),
- error);
- }
- }
- free_slab_cursor(&refc_cur);
+ /* Since we're not writing the AGF yet, no need to commit the cursor */
+ libxfs_btree_del_cursor(btr->cur, 0);
+ free_slab_cursor(&btr->slab_cursor);
+ error = -libxfs_trans_commit(sc->tp);
+ if (error)
+ do_error(
+_("Error %d while writing refcount btree for AG %u.\n"), error, agno);
+ sc->tp = NULL;
}
/* Fill the AGFL with any leftover bnobt rebuilder blocks. */
@@ -1501,7 +1277,7 @@ build_agf_agfl(
xfs_extlen_t freeblks, /* # free blocks in tree */
int lostblocks, /* # blocks that will be lost */
struct bt_rebuild *btr_rmap,
- struct bt_status *refcnt_bt,
+ struct bt_rebuild *btr_refc,
struct xfs_slab *lost_fsb)
{
struct extent_tree_node *ext_ptr;
@@ -1559,10 +1335,14 @@ build_agf_agfl(
cpu_to_be32(btr_rmap->newbt.afake.af_blocks);
}
- agf->agf_refcount_root = cpu_to_be32(refcnt_bt->root);
- agf->agf_refcount_level = cpu_to_be32(refcnt_bt->num_levels);
- agf->agf_refcount_blocks = cpu_to_be32(refcnt_bt->num_tot_blocks -
- refcnt_bt->num_free_blocks);
+ if (xfs_sb_version_hasreflink(&mp->m_sb)) {
+ agf->agf_refcount_root =
+ cpu_to_be32(btr_refc->newbt.afake.af_root);
+ agf->agf_refcount_level =
+ cpu_to_be32(btr_refc->newbt.afake.af_levels);
+ agf->agf_refcount_blocks =
+ cpu_to_be32(btr_refc->newbt.afake.af_blocks);
+ }
/*
* Count and record the number of btree blocks consumed if required.
@@ -1724,7 +1504,7 @@ phase5_func(
struct bt_rebuild btr_ino;
struct bt_rebuild btr_fino;
struct bt_rebuild btr_rmap;
- bt_status_t refcnt_btree_curs;
+ struct bt_rebuild btr_refc;
int extra_blocks = 0;
uint num_freeblocks;
xfs_extlen_t freeblks1;
@@ -1762,11 +1542,7 @@ _("unable to rebuild AG %u. Not enough free space in on-disk AG.\n"),
init_rmapbt_cursor(&sc, agno, num_freeblocks, &btr_rmap);
- /*
- * Set up the btree cursors for the on-disk refcount btrees,
- * which includes pre-allocating all required blocks.
- */
- init_refc_cursor(mp, agno, &refcnt_btree_curs);
+ init_refc_cursor(&sc, agno, num_freeblocks, &btr_refc);
num_extents = count_bno_extents_blocks(agno, &num_freeblocks);
/*
@@ -1831,16 +1607,14 @@ _("unable to rebuild AG %u. Not enough free space in on-disk AG.\n"),
sb_fdblocks_ag[agno] += btr_rmap.newbt.afake.af_blocks - 1;
}
- if (xfs_sb_version_hasreflink(&mp->m_sb)) {
- build_refcount_tree(mp, agno, &refcnt_btree_curs);
- write_cursor(&refcnt_btree_curs);
- }
+ if (xfs_sb_version_hasreflink(&mp->m_sb))
+ build_refcount_tree(&sc, agno, &btr_refc);
/*
* set up agf and agfl
*/
build_agf_agfl(mp, agno, &btr_bno, &btr_cnt, freeblks1, extra_blocks,
- &btr_rmap, &refcnt_btree_curs, lost_fsb);
+ &btr_rmap, &btr_refc, lost_fsb);
/*
* build inode allocation trees.
@@ -1863,7 +1637,7 @@ _("unable to rebuild AG %u. Not enough free space in on-disk AG.\n"),
if (xfs_sb_version_hasrmapbt(&mp->m_sb))
finish_rebuild(mp, &btr_rmap, lost_fsb);
if (xfs_sb_version_hasreflink(&mp->m_sb))
- finish_cursor(&refcnt_btree_curs);
+ finish_rebuild(mp, &btr_refc, lost_fsb);
/*
* release the incore per-AG bno/bcnt trees so the extent nodes
^ permalink raw reply related [flat|nested] 30+ messages in thread
* [PATCH 8/9] xfs_repair: remove old btree rebuild support code
2020-05-20 1:50 [PATCH v5 0/9] xfs_repair: use btree bulk loading Darrick J. Wong
` (6 preceding siblings ...)
2020-05-20 1:51 ` [PATCH 7/9] xfs_repair: rebuild refcount " Darrick J. Wong
@ 2020-05-20 1:51 ` Darrick J. Wong
2020-05-20 1:51 ` [PATCH 9/9] xfs_repair: track blocks lost during btree construction via extents Darrick J. Wong
8 siblings, 0 replies; 30+ messages in thread
From: Darrick J. Wong @ 2020-05-20 1:51 UTC (permalink / raw)
To: sandeen, darrick.wong; +Cc: linux-xfs, bfoster
From: Darrick J. Wong <darrick.wong@oracle.com>
This code isn't needed anymore, so get rid of it.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
---
repair/phase5.c | 242 -------------------------------------------------------
1 file changed, 242 deletions(-)
diff --git a/repair/phase5.c b/repair/phase5.c
index 797ad6c1..72c6908a 100644
--- a/repair/phase5.c
+++ b/repair/phase5.c
@@ -20,52 +20,6 @@
#include "rmap.h"
#include "bload.h"
-/*
- * we maintain the current slice (path from root to leaf)
- * of the btree incore. when we need a new block, we ask
- * the block allocator for the address of a block on that
- * level, map the block in, and set up the appropriate
- * pointers (child, silbing, etc.) and keys that should
- * point to the new block.
- */
-typedef struct bt_stat_level {
- /*
- * set in setup_cursor routine and maintained in the tree-building
- * routines
- */
- xfs_buf_t *buf_p; /* 2 buffer pointers to ... */
- xfs_buf_t *prev_buf_p;
- xfs_agblock_t agbno; /* current block being filled */
- xfs_agblock_t prev_agbno; /* previous block */
- /*
- * set in calculate/init cursor routines for each btree level
- */
- int num_recs_tot; /* # tree recs in level */
- int num_blocks; /* # tree blocks in level */
- int num_recs_pb; /* num_recs_tot / num_blocks */
- int modulo; /* num_recs_tot % num_blocks */
-} bt_stat_level_t;
-
-typedef struct bt_status {
- int init; /* cursor set up once? */
- int num_levels; /* # of levels in btree */
- xfs_extlen_t num_tot_blocks; /* # blocks alloc'ed for tree */
- xfs_extlen_t num_free_blocks;/* # blocks currently unused */
-
- xfs_agblock_t root; /* root block */
- /*
- * list of blocks to be used to set up this tree
- * and pointer to the first unused block on the list
- */
- xfs_agblock_t *btree_blocks; /* block list */
- xfs_agblock_t *free_btree_blocks; /* first unused block */
- /*
- * per-level status info
- */
- bt_stat_level_t level[XFS_BTREE_MAXLEVELS];
- uint64_t owner; /* owner */
-} bt_status_t;
-
/* Context for rebuilding a per-AG btree. */
struct bt_rebuild {
/* Fake root for staging and space preallocations. */
@@ -197,148 +151,6 @@ mk_incore_fstree(
return(num_extents);
}
-static xfs_agblock_t
-get_next_blockaddr(xfs_agnumber_t agno, int level, bt_status_t *curs)
-{
- ASSERT(curs->free_btree_blocks < curs->btree_blocks +
- curs->num_tot_blocks);
- ASSERT(curs->num_free_blocks > 0);
-
- curs->num_free_blocks--;
- return(*curs->free_btree_blocks++);
-}
-
-/*
- * set up the dynamically allocated block allocation data in the btree
- * cursor that depends on the info in the static portion of the cursor.
- * allocates space from the incore bno/bcnt extent trees and sets up
- * the first path up the left side of the tree. Also sets up the
- * cursor pointer to the btree root. called by init_freespace_cursor()
- * and init_ino_cursor()
- */
-static void
-setup_cursor(xfs_mount_t *mp, xfs_agnumber_t agno, bt_status_t *curs)
-{
- int j;
- unsigned int u;
- xfs_extlen_t big_extent_len;
- xfs_agblock_t big_extent_start;
- extent_tree_node_t *ext_ptr;
- extent_tree_node_t *bno_ext_ptr;
- xfs_extlen_t blocks_allocated;
- xfs_agblock_t *agb_ptr;
- int error;
-
- /*
- * get the number of blocks we need to allocate, then
- * set up block number array, set the free block pointer
- * to the first block in the array, and null the array
- */
- big_extent_len = curs->num_tot_blocks;
- blocks_allocated = 0;
-
- ASSERT(big_extent_len > 0);
-
- if ((curs->btree_blocks = malloc(sizeof(xfs_agblock_t)
- * big_extent_len)) == NULL)
- do_error(_("could not set up btree block array\n"));
-
- agb_ptr = curs->free_btree_blocks = curs->btree_blocks;
-
- for (j = 0; j < curs->num_free_blocks; j++, agb_ptr++)
- *agb_ptr = NULLAGBLOCK;
-
- /*
- * grab the smallest extent and use it up, then get the
- * next smallest. This mimics the init_*_cursor code.
- */
- ext_ptr = findfirst_bcnt_extent(agno);
-
- agb_ptr = curs->btree_blocks;
-
- /*
- * set up the free block array
- */
- while (blocks_allocated < big_extent_len) {
- if (!ext_ptr)
- do_error(
-_("error - not enough free space in filesystem\n"));
- /*
- * use up the extent we've got
- */
- for (u = 0; u < ext_ptr->ex_blockcount &&
- blocks_allocated < big_extent_len; u++) {
- ASSERT(agb_ptr < curs->btree_blocks
- + curs->num_tot_blocks);
- *agb_ptr++ = ext_ptr->ex_startblock + u;
- blocks_allocated++;
- }
-
- error = rmap_add_ag_rec(mp, agno, ext_ptr->ex_startblock, u,
- curs->owner);
- if (error)
- do_error(_("could not set up btree rmaps: %s\n"),
- strerror(-error));
-
- /*
- * if we only used part of this last extent, then we
- * need only to reset the extent in the extent
- * trees and we're done
- */
- if (u < ext_ptr->ex_blockcount) {
- big_extent_start = ext_ptr->ex_startblock + u;
- big_extent_len = ext_ptr->ex_blockcount - u;
-
- ASSERT(big_extent_len > 0);
-
- bno_ext_ptr = find_bno_extent(agno,
- ext_ptr->ex_startblock);
- ASSERT(bno_ext_ptr != NULL);
- get_bno_extent(agno, bno_ext_ptr);
- release_extent_tree_node(bno_ext_ptr);
-
- ext_ptr = get_bcnt_extent(agno, ext_ptr->ex_startblock,
- ext_ptr->ex_blockcount);
- release_extent_tree_node(ext_ptr);
-#ifdef XR_BLD_FREE_TRACE
- fprintf(stderr, "releasing extent: %u [%u %u]\n",
- agno, ext_ptr->ex_startblock,
- ext_ptr->ex_blockcount);
- fprintf(stderr, "blocks_allocated = %d\n",
- blocks_allocated);
-#endif
-
- add_bno_extent(agno, big_extent_start, big_extent_len);
- add_bcnt_extent(agno, big_extent_start, big_extent_len);
-
- return;
- }
- /*
- * delete the used-up extent from both extent trees and
- * find next biggest extent
- */
-#ifdef XR_BLD_FREE_TRACE
- fprintf(stderr, "releasing extent: %u [%u %u]\n",
- agno, ext_ptr->ex_startblock, ext_ptr->ex_blockcount);
-#endif
- bno_ext_ptr = find_bno_extent(agno, ext_ptr->ex_startblock);
- ASSERT(bno_ext_ptr != NULL);
- get_bno_extent(agno, bno_ext_ptr);
- release_extent_tree_node(bno_ext_ptr);
-
- ext_ptr = get_bcnt_extent(agno, ext_ptr->ex_startblock,
- ext_ptr->ex_blockcount);
- ASSERT(ext_ptr != NULL);
- release_extent_tree_node(ext_ptr);
-
- ext_ptr = findfirst_bcnt_extent(agno);
- }
-#ifdef XR_BLD_FREE_TRACE
- fprintf(stderr, "blocks_allocated = %d\n",
- blocks_allocated);
-#endif
-}
-
/*
* Estimate proper slack values for a btree that's being reloaded.
*
@@ -494,36 +306,6 @@ rebuild_claim_block(
return xrep_newbt_claim_block(cur, &btr->newbt, ptr);
}
-static void
-write_cursor(bt_status_t *curs)
-{
- int i;
-
- for (i = 0; i < curs->num_levels; i++) {
-#if defined(XR_BLD_FREE_TRACE) || defined(XR_BLD_INO_TRACE)
- fprintf(stderr, "writing bt block %u\n", curs->level[i].agbno);
-#endif
- if (curs->level[i].prev_buf_p != NULL) {
- ASSERT(curs->level[i].prev_agbno != NULLAGBLOCK);
-#if defined(XR_BLD_FREE_TRACE) || defined(XR_BLD_INO_TRACE)
- fprintf(stderr, "writing bt prev block %u\n",
- curs->level[i].prev_agbno);
-#endif
- libxfs_buf_mark_dirty(curs->level[i].prev_buf_p);
- libxfs_buf_relse(curs->level[i].prev_buf_p);
- }
- libxfs_buf_mark_dirty(curs->level[i].buf_p);
- libxfs_buf_relse(curs->level[i].buf_p);
- }
-}
-
-static void
-finish_cursor(bt_status_t *curs)
-{
- ASSERT(curs->num_free_blocks == 0);
- free(curs->btree_blocks);
-}
-
/*
* Scoop up leftovers from a rebuild cursor for later freeing, then free the
* rebuild context.
@@ -552,30 +334,6 @@ _("Insufficient memory saving lost blocks.\n"));
xrep_newbt_destroy(&btr->newbt, 0);
}
-/* Map btnum to buffer ops for the types that need it. */
-static const struct xfs_buf_ops *
-btnum_to_ops(
- xfs_btnum_t btnum)
-{
- switch (btnum) {
- case XFS_BTNUM_BNO:
- return &xfs_bnobt_buf_ops;
- case XFS_BTNUM_CNT:
- return &xfs_cntbt_buf_ops;
- case XFS_BTNUM_INO:
- return &xfs_inobt_buf_ops;
- case XFS_BTNUM_FINO:
- return &xfs_finobt_buf_ops;
- case XFS_BTNUM_RMAP:
- return &xfs_rmapbt_buf_ops;
- case XFS_BTNUM_REFC:
- return &xfs_refcountbt_buf_ops;
- default:
- ASSERT(0);
- return NULL;
- }
-}
-
/*
* Free Space Btrees
*
^ permalink raw reply related [flat|nested] 30+ messages in thread
* [PATCH 9/9] xfs_repair: track blocks lost during btree construction via extents
2020-05-20 1:50 [PATCH v5 0/9] xfs_repair: use btree bulk loading Darrick J. Wong
` (7 preceding siblings ...)
2020-05-20 1:51 ` [PATCH 8/9] xfs_repair: remove old btree rebuild support code Darrick J. Wong
@ 2020-05-20 1:51 ` Darrick J. Wong
2020-05-28 17:00 ` Brian Foster
8 siblings, 1 reply; 30+ messages in thread
From: Darrick J. Wong @ 2020-05-20 1:51 UTC (permalink / raw)
To: sandeen, darrick.wong; +Cc: linux-xfs, bfoster
From: Darrick J. Wong <darrick.wong@oracle.com>
Use extent records (not just raw fsbs) to track blocks that were lost
during btree construction. This makes it somewhat more efficient.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
---
repair/phase5.c | 61 ++++++++++++++++++++++++++++++++-----------------------
1 file changed, 35 insertions(+), 26 deletions(-)
diff --git a/repair/phase5.c b/repair/phase5.c
index 72c6908a..22007275 100644
--- a/repair/phase5.c
+++ b/repair/phase5.c
@@ -45,6 +45,12 @@ struct bt_rebuild {
};
};
+struct lost_fsb {
+ xfs_fsblock_t fsbno;
+ xfs_extlen_t len;
+};
+
+
/*
* extra metadata for the agi
*/
@@ -314,21 +320,24 @@ static void
finish_rebuild(
struct xfs_mount *mp,
struct bt_rebuild *btr,
- struct xfs_slab *lost_fsb)
+ struct xfs_slab *lost_fsbs)
{
struct xrep_newbt_resv *resv, *n;
for_each_xrep_newbt_reservation(&btr->newbt, resv, n) {
- while (resv->used < resv->len) {
- xfs_fsblock_t fsb = resv->fsbno + resv->used;
- int error;
+ struct lost_fsb lost;
+ int error;
- error = slab_add(lost_fsb, &fsb);
- if (error)
- do_error(
+ if (resv->used == resv->len)
+ continue;
+
+ lost.fsbno = resv->fsbno + resv->used;
+ lost.len = resv->len - resv->used;
+ error = slab_add(lost_fsbs, &lost);
+ if (error)
+ do_error(
_("Insufficient memory saving lost blocks.\n"));
- resv->used++;
- }
+ resv->used = resv->len;
}
xrep_newbt_destroy(&btr->newbt, 0);
@@ -1036,7 +1045,7 @@ build_agf_agfl(
int lostblocks, /* # blocks that will be lost */
struct bt_rebuild *btr_rmap,
struct bt_rebuild *btr_refc,
- struct xfs_slab *lost_fsb)
+ struct xfs_slab *lost_fsbs)
{
struct extent_tree_node *ext_ptr;
struct xfs_buf *agf_buf, *agfl_buf;
@@ -1253,7 +1262,7 @@ static void
phase5_func(
struct xfs_mount *mp,
xfs_agnumber_t agno,
- struct xfs_slab *lost_fsb)
+ struct xfs_slab *lost_fsbs)
{
struct repair_ctx sc = { .mp = mp, };
struct agi_stat agi_stat = {0,};
@@ -1372,7 +1381,7 @@ _("unable to rebuild AG %u. Not enough free space in on-disk AG.\n"),
* set up agf and agfl
*/
build_agf_agfl(mp, agno, &btr_bno, &btr_cnt, freeblks1, extra_blocks,
- &btr_rmap, &btr_refc, lost_fsb);
+ &btr_rmap, &btr_refc, lost_fsbs);
/*
* build inode allocation trees.
@@ -1387,15 +1396,15 @@ _("unable to rebuild AG %u. Not enough free space in on-disk AG.\n"),
/*
* tear down cursors
*/
- finish_rebuild(mp, &btr_bno, lost_fsb);
- finish_rebuild(mp, &btr_cnt, lost_fsb);
- finish_rebuild(mp, &btr_ino, lost_fsb);
+ finish_rebuild(mp, &btr_bno, lost_fsbs);
+ finish_rebuild(mp, &btr_cnt, lost_fsbs);
+ finish_rebuild(mp, &btr_ino, lost_fsbs);
if (xfs_sb_version_hasfinobt(&mp->m_sb))
- finish_rebuild(mp, &btr_fino, lost_fsb);
+ finish_rebuild(mp, &btr_fino, lost_fsbs);
if (xfs_sb_version_hasrmapbt(&mp->m_sb))
- finish_rebuild(mp, &btr_rmap, lost_fsb);
+ finish_rebuild(mp, &btr_rmap, lost_fsbs);
if (xfs_sb_version_hasreflink(&mp->m_sb))
- finish_rebuild(mp, &btr_refc, lost_fsb);
+ finish_rebuild(mp, &btr_refc, lost_fsbs);
/*
* release the incore per-AG bno/bcnt trees so the extent nodes
@@ -1414,19 +1423,19 @@ inject_lost_blocks(
{
struct xfs_trans *tp = NULL;
struct xfs_slab_cursor *cur = NULL;
- xfs_fsblock_t *fsb;
+ struct lost_fsb *lost;
int error;
error = init_slab_cursor(lost_fsbs, NULL, &cur);
if (error)
return error;
- while ((fsb = pop_slab_cursor(cur)) != NULL) {
+ while ((lost = pop_slab_cursor(cur)) != NULL) {
error = -libxfs_trans_alloc_rollable(mp, 16, &tp);
if (error)
goto out_cancel;
- error = -libxfs_free_extent(tp, *fsb, 1,
+ error = -libxfs_free_extent(tp, lost->fsbno, lost->len,
&XFS_RMAP_OINFO_ANY_OWNER, XFS_AG_RESV_NONE);
if (error)
goto out_cancel;
@@ -1447,7 +1456,7 @@ inject_lost_blocks(
void
phase5(xfs_mount_t *mp)
{
- struct xfs_slab *lost_fsb;
+ struct xfs_slab *lost_fsbs;
xfs_agnumber_t agno;
int error;
@@ -1490,12 +1499,12 @@ phase5(xfs_mount_t *mp)
if (sb_fdblocks_ag == NULL)
do_error(_("cannot alloc sb_fdblocks_ag buffers\n"));
- error = init_slab(&lost_fsb, sizeof(xfs_fsblock_t));
+ error = init_slab(&lost_fsbs, sizeof(struct lost_fsb));
if (error)
do_error(_("cannot alloc lost block slab\n"));
for (agno = 0; agno < mp->m_sb.sb_agcount; agno++)
- phase5_func(mp, agno, lost_fsb);
+ phase5_func(mp, agno, lost_fsbs);
print_final_rpt();
@@ -1538,10 +1547,10 @@ _("unable to add AG %u reverse-mapping data to btree.\n"), agno);
* Put blocks that were unnecessarily reserved for btree
* reconstruction back into the filesystem free space data.
*/
- error = inject_lost_blocks(mp, lost_fsb);
+ error = inject_lost_blocks(mp, lost_fsbs);
if (error)
do_error(_("Unable to reinsert lost blocks into filesystem.\n"));
- free_slab(&lost_fsb);
+ free_slab(&lost_fsbs);
bad_ino_btree = 0;
^ permalink raw reply related [flat|nested] 30+ messages in thread
* Re: [PATCH 1/9] xfs_repair: port the online repair newbt structure
2020-05-20 1:50 ` [PATCH 1/9] xfs_repair: port the online repair newbt structure Darrick J. Wong
@ 2020-05-27 12:15 ` Brian Foster
2020-05-27 22:34 ` Darrick J. Wong
0 siblings, 1 reply; 30+ messages in thread
From: Brian Foster @ 2020-05-27 12:15 UTC (permalink / raw)
To: Darrick J. Wong; +Cc: sandeen, linux-xfs
On Tue, May 19, 2020 at 06:50:49PM -0700, Darrick J. Wong wrote:
> From: Darrick J. Wong <darrick.wong@oracle.com>
>
> Port the new btree staging context and related block reservation helper
> code from the kernel to repair. We'll use this in subsequent patches to
> implement btree bulk loading.
>
> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
> ---
> include/libxfs.h | 1
> libxfs/libxfs_api_defs.h | 2
> repair/Makefile | 4 -
> repair/bload.c | 303 ++++++++++++++++++++++++++++++++++++++++++++++
> repair/bload.h | 77 ++++++++++++
> repair/xfs_repair.c | 17 +++
> 6 files changed, 402 insertions(+), 2 deletions(-)
> create mode 100644 repair/bload.c
> create mode 100644 repair/bload.h
>
>
...
> diff --git a/repair/bload.c b/repair/bload.c
> new file mode 100644
> index 00000000..9bc17468
> --- /dev/null
> +++ b/repair/bload.c
> @@ -0,0 +1,303 @@
> +// SPDX-License-Identifier: GPL-2.0-or-later
> +/*
> + * Copyright (C) 2020 Oracle. All Rights Reserved.
> + * Author: Darrick J. Wong <darrick.wong@oracle.com>
> + */
> +#include <libxfs.h>
> +#include "bload.h"
> +
> +#define trace_xrep_newbt_claim_block(...) ((void) 0)
> +#define trace_xrep_newbt_free_blocks(...) ((void) 0)
> +
> +int bload_leaf_slack = -1;
> +int bload_node_slack = -1;
> +
> +/* Ported routines from fs/xfs/scrub/repair.c */
> +
Looks mostly straightforward, but I'll have to come back to this as I
get to the code that uses it later in the series. In the meantime, I see
some of these helpers in scrub/repair.c while not others. Are there
references to other routines that are intended to be copies from kernel
code?
Brian
> +/*
> + * Roll a transaction, keeping the AG headers locked and reinitializing
> + * the btree cursors.
> + */
> +int
> +xrep_roll_ag_trans(
> + struct repair_ctx *sc)
> +{
> + int error;
> +
> + /* Keep the AG header buffers locked so we can keep going. */
> + if (sc->agi_bp)
> + libxfs_trans_bhold(sc->tp, sc->agi_bp);
> + if (sc->agf_bp)
> + libxfs_trans_bhold(sc->tp, sc->agf_bp);
> + if (sc->agfl_bp)
> + libxfs_trans_bhold(sc->tp, sc->agfl_bp);
> +
> + /*
> + * Roll the transaction. We still own the buffer and the buffer lock
> + * regardless of whether or not the roll succeeds. If the roll fails,
> + * the buffers will be released during teardown on our way out of the
> + * kernel. If it succeeds, we join them to the new transaction and
> + * move on.
> + */
> + error = -libxfs_trans_roll(&sc->tp);
> + if (error)
> + return error;
> +
> + /* Join AG headers to the new transaction. */
> + if (sc->agi_bp)
> + libxfs_trans_bjoin(sc->tp, sc->agi_bp);
> + if (sc->agf_bp)
> + libxfs_trans_bjoin(sc->tp, sc->agf_bp);
> + if (sc->agfl_bp)
> + libxfs_trans_bjoin(sc->tp, sc->agfl_bp);
> +
> + return 0;
> +}
> +
> +/* Initialize accounting resources for staging a new AG btree. */
> +void
> +xrep_newbt_init_ag(
> + struct xrep_newbt *xnr,
> + struct repair_ctx *sc,
> + const struct xfs_owner_info *oinfo,
> + xfs_fsblock_t alloc_hint,
> + enum xfs_ag_resv_type resv)
> +{
> + memset(xnr, 0, sizeof(struct xrep_newbt));
> + xnr->sc = sc;
> + xnr->oinfo = *oinfo; /* structure copy */
> + xnr->alloc_hint = alloc_hint;
> + xnr->resv = resv;
> + INIT_LIST_HEAD(&xnr->resv_list);
> +}
> +
> +/* Initialize accounting resources for staging a new inode fork btree. */
> +void
> +xrep_newbt_init_inode(
> + struct xrep_newbt *xnr,
> + struct repair_ctx *sc,
> + int whichfork,
> + const struct xfs_owner_info *oinfo)
> +{
> + xrep_newbt_init_ag(xnr, sc, oinfo,
> + XFS_INO_TO_FSB(sc->mp, sc->ip->i_ino),
> + XFS_AG_RESV_NONE);
> + xnr->ifake.if_fork = kmem_zone_zalloc(xfs_ifork_zone, 0);
> + xnr->ifake.if_fork_size = XFS_IFORK_SIZE(sc->ip, whichfork);
> +}
> +
> +/*
> + * Initialize accounting resources for staging a new btree. Callers are
> + * expected to add their own reservations (and clean them up) manually.
> + */
> +void
> +xrep_newbt_init_bare(
> + struct xrep_newbt *xnr,
> + struct repair_ctx *sc)
> +{
> + xrep_newbt_init_ag(xnr, sc, &XFS_RMAP_OINFO_ANY_OWNER, NULLFSBLOCK,
> + XFS_AG_RESV_NONE);
> +}
> +
> +/* Designate specific blocks to be used to build our new btree. */
> +int
> +xrep_newbt_add_blocks(
> + struct xrep_newbt *xnr,
> + xfs_fsblock_t fsbno,
> + xfs_extlen_t len)
> +{
> + struct xrep_newbt_resv *resv;
> +
> + resv = kmem_alloc(sizeof(struct xrep_newbt_resv), KM_MAYFAIL);
> + if (!resv)
> + return ENOMEM;
> +
> + INIT_LIST_HEAD(&resv->list);
> + resv->fsbno = fsbno;
> + resv->len = len;
> + resv->used = 0;
> + list_add_tail(&resv->list, &xnr->resv_list);
> + return 0;
> +}
> +
> +/* Reserve disk space for our new btree. */
> +int
> +xrep_newbt_alloc_blocks(
> + struct xrep_newbt *xnr,
> + uint64_t nr_blocks)
> +{
> + struct repair_ctx *sc = xnr->sc;
> + xfs_alloctype_t type;
> + xfs_fsblock_t alloc_hint = xnr->alloc_hint;
> + int error = 0;
> +
> + type = sc->ip ? XFS_ALLOCTYPE_START_BNO : XFS_ALLOCTYPE_NEAR_BNO;
> +
> + while (nr_blocks > 0 && !error) {
> + struct xfs_alloc_arg args = {
> + .tp = sc->tp,
> + .mp = sc->mp,
> + .type = type,
> + .fsbno = alloc_hint,
> + .oinfo = xnr->oinfo,
> + .minlen = 1,
> + .maxlen = nr_blocks,
> + .prod = 1,
> + .resv = xnr->resv,
> + };
> +
> + error = -libxfs_alloc_vextent(&args);
> + if (error)
> + return error;
> + if (args.fsbno == NULLFSBLOCK)
> + return ENOSPC;
> +
> + /* We don't have real EFIs here so skip that. */
> +
> + error = xrep_newbt_add_blocks(xnr, args.fsbno, args.len);
> + if (error)
> + break;
> +
> + nr_blocks -= args.len;
> + alloc_hint = args.fsbno + args.len - 1;
> +
> + if (sc->ip)
> + error = -libxfs_trans_roll_inode(&sc->tp, sc->ip);
> + else
> + error = xrep_roll_ag_trans(sc);
> + }
> +
> + return error;
> +}
> +
> +/*
> + * Release blocks that were reserved for a btree repair. If the repair
> + * succeeded then we log deferred frees for unused blocks. Otherwise, we try
> + * to free the extents immediately to roll the filesystem back to where it was
> + * before we started.
> + */
> +static inline int
> +xrep_newbt_destroy_reservation(
> + struct xrep_newbt *xnr,
> + struct xrep_newbt_resv *resv,
> + bool cancel_repair)
> +{
> + struct repair_ctx *sc = xnr->sc;
> +
> + if (cancel_repair) {
> + int error;
> +
> + /* Free the extent then roll the transaction. */
> + error = -libxfs_free_extent(sc->tp, resv->fsbno, resv->len,
> + &xnr->oinfo, xnr->resv);
> + if (error)
> + return error;
> +
> + if (sc->ip)
> + return -libxfs_trans_roll_inode(&sc->tp, sc->ip);
> + return xrep_roll_ag_trans(sc);
> + }
> +
> + /* We don't have EFIs here so skip the EFD. */
> +
> + /*
> + * Use the deferred freeing mechanism to schedule for deletion any
> + * blocks we didn't use to rebuild the tree. This enables us to log
> + * them all in the same transaction as the root change.
> + */
> + resv->fsbno += resv->used;
> + resv->len -= resv->used;
> + resv->used = 0;
> +
> + if (resv->len == 0)
> + return 0;
> +
> + trace_xrep_newbt_free_blocks(sc->mp,
> + XFS_FSB_TO_AGNO(sc->mp, resv->fsbno),
> + XFS_FSB_TO_AGBNO(sc->mp, resv->fsbno),
> + resv->len, xnr->oinfo.oi_owner);
> +
> + __xfs_bmap_add_free(sc->tp, resv->fsbno, resv->len, &xnr->oinfo, true);
> +
> + return 0;
> +}
> +
> +/* Free all the accounting info and disk space we reserved for a new btree. */
> +void
> +xrep_newbt_destroy(
> + struct xrep_newbt *xnr,
> + int error)
> +{
> + struct repair_ctx *sc = xnr->sc;
> + struct xrep_newbt_resv *resv, *n;
> + int err2;
> +
> + list_for_each_entry_safe(resv, n, &xnr->resv_list, list) {
> + err2 = xrep_newbt_destroy_reservation(xnr, resv, error != 0);
> + if (err2)
> + goto junkit;
> +
> + list_del(&resv->list);
> + kmem_free(resv);
> + }
> +
> +junkit:
> + /*
> + * If we still have reservations attached to @newbt, cleanup must have
> + * failed and the filesystem is about to go down. Clean up the incore
> + * reservations.
> + */
> + list_for_each_entry_safe(resv, n, &xnr->resv_list, list) {
> + list_del(&resv->list);
> + kmem_free(resv);
> + }
> +
> + if (sc->ip) {
> + kmem_cache_free(xfs_ifork_zone, xnr->ifake.if_fork);
> + xnr->ifake.if_fork = NULL;
> + }
> +}
> +
> +/* Feed one of the reserved btree blocks to the bulk loader. */
> +int
> +xrep_newbt_claim_block(
> + struct xfs_btree_cur *cur,
> + struct xrep_newbt *xnr,
> + union xfs_btree_ptr *ptr)
> +{
> + struct xrep_newbt_resv *resv;
> + xfs_fsblock_t fsb;
> +
> + /*
> + * The first item in the list should always have a free block unless
> + * we're completely out.
> + */
> + resv = list_first_entry(&xnr->resv_list, struct xrep_newbt_resv, list);
> + if (resv->used == resv->len)
> + return ENOSPC;
> +
> + /*
> + * Peel off a block from the start of the reservation. We allocate
> + * blocks in order to place blocks on disk in increasing record or key
> + * order. The block reservations tend to end up on the list in
> + * decreasing order, which hopefully results in leaf blocks ending up
> + * together.
> + */
> + fsb = resv->fsbno + resv->used;
> + resv->used++;
> +
> + /* If we used all the blocks in this reservation, move it to the end. */
> + if (resv->used == resv->len)
> + list_move_tail(&resv->list, &xnr->resv_list);
> +
> + trace_xrep_newbt_claim_block(cur->bc_mp,
> + XFS_FSB_TO_AGNO(cur->bc_mp, fsb),
> + XFS_FSB_TO_AGBNO(cur->bc_mp, fsb),
> + 1, xnr->oinfo.oi_owner);
> +
> + if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
> + ptr->l = cpu_to_be64(fsb);
> + else
> + ptr->s = cpu_to_be32(XFS_FSB_TO_AGBNO(cur->bc_mp, fsb));
> + return 0;
> +}
> diff --git a/repair/bload.h b/repair/bload.h
> new file mode 100644
> index 00000000..020c4834
> --- /dev/null
> +++ b/repair/bload.h
> @@ -0,0 +1,77 @@
> +// SPDX-License-Identifier: GPL-2.0-or-later
> +/*
> + * Copyright (C) 2020 Oracle. All Rights Reserved.
> + * Author: Darrick J. Wong <darrick.wong@oracle.com>
> + */
> +#ifndef __XFS_REPAIR_BLOAD_H__
> +#define __XFS_REPAIR_BLOAD_H__
> +
> +extern int bload_leaf_slack;
> +extern int bload_node_slack;
> +
> +struct repair_ctx {
> + struct xfs_mount *mp;
> + struct xfs_inode *ip;
> + struct xfs_trans *tp;
> +
> + struct xfs_buf *agi_bp;
> + struct xfs_buf *agf_bp;
> + struct xfs_buf *agfl_bp;
> +};
> +
> +struct xrep_newbt_resv {
> + /* Link to list of extents that we've reserved. */
> + struct list_head list;
> +
> + /* FSB of the block we reserved. */
> + xfs_fsblock_t fsbno;
> +
> + /* Length of the reservation. */
> + xfs_extlen_t len;
> +
> + /* How much of this reservation we've used. */
> + xfs_extlen_t used;
> +};
> +
> +struct xrep_newbt {
> + struct repair_ctx *sc;
> +
> + /* List of extents that we've reserved. */
> + struct list_head resv_list;
> +
> + /* Fake root for new btree. */
> + union {
> + struct xbtree_afakeroot afake;
> + struct xbtree_ifakeroot ifake;
> + };
> +
> + /* rmap owner of these blocks */
> + struct xfs_owner_info oinfo;
> +
> + /* The last reservation we allocated from. */
> + struct xrep_newbt_resv *last_resv;
> +
> + /* Allocation hint */
> + xfs_fsblock_t alloc_hint;
> +
> + /* per-ag reservation type */
> + enum xfs_ag_resv_type resv;
> +};
> +
> +#define for_each_xrep_newbt_reservation(xnr, resv, n) \
> + list_for_each_entry_safe((resv), (n), &(xnr)->resv_list, list)
> +
> +void xrep_newbt_init_bare(struct xrep_newbt *xnr, struct repair_ctx *sc);
> +void xrep_newbt_init_ag(struct xrep_newbt *xnr, struct repair_ctx *sc,
> + const struct xfs_owner_info *oinfo, xfs_fsblock_t alloc_hint,
> + enum xfs_ag_resv_type resv);
> +void xrep_newbt_init_inode(struct xrep_newbt *xnr, struct repair_ctx *sc,
> + int whichfork, const struct xfs_owner_info *oinfo);
> +int xrep_newbt_add_blocks(struct xrep_newbt *xnr, xfs_fsblock_t fsbno,
> + xfs_extlen_t len);
> +int xrep_newbt_alloc_blocks(struct xrep_newbt *xnr, uint64_t nr_blocks);
> +void xrep_newbt_destroy(struct xrep_newbt *xnr, int error);
> +int xrep_newbt_claim_block(struct xfs_btree_cur *cur, struct xrep_newbt *xnr,
> + union xfs_btree_ptr *ptr);
> +
> +#endif /* __XFS_REPAIR_BLOAD_H__ */
> diff --git a/repair/xfs_repair.c b/repair/xfs_repair.c
> index 9d72fa8e..8fbd3649 100644
> --- a/repair/xfs_repair.c
> +++ b/repair/xfs_repair.c
> @@ -24,6 +24,7 @@
> #include "rmap.h"
> #include "libfrog/fsgeom.h"
> #include "libfrog/platform.h"
> +#include "bload.h"
>
> /*
> * option tables for getsubopt calls
> @@ -39,6 +40,8 @@ enum o_opt_nums {
> AG_STRIDE,
> FORCE_GEO,
> PHASE2_THREADS,
> + BLOAD_LEAF_SLACK,
> + BLOAD_NODE_SLACK,
> O_MAX_OPTS,
> };
>
> @@ -49,6 +52,8 @@ static char *o_opts[] = {
> [AG_STRIDE] = "ag_stride",
> [FORCE_GEO] = "force_geometry",
> [PHASE2_THREADS] = "phase2_threads",
> + [BLOAD_LEAF_SLACK] = "debug_bload_leaf_slack",
> + [BLOAD_NODE_SLACK] = "debug_bload_node_slack",
> [O_MAX_OPTS] = NULL,
> };
>
> @@ -260,6 +265,18 @@ process_args(int argc, char **argv)
> _("-o phase2_threads requires a parameter\n"));
> phase2_threads = (int)strtol(val, NULL, 0);
> break;
> + case BLOAD_LEAF_SLACK:
> + if (!val)
> + do_abort(
> + _("-o debug_bload_leaf_slack requires a parameter\n"));
> + bload_leaf_slack = (int)strtol(val, NULL, 0);
> + break;
> + case BLOAD_NODE_SLACK:
> + if (!val)
> + do_abort(
> + _("-o debug_bload_node_slack requires a parameter\n"));
> + bload_node_slack = (int)strtol(val, NULL, 0);
> + break;
> default:
> unknown('o', val);
> break;
>
^ permalink raw reply [flat|nested] 30+ messages in thread
* Re: [PATCH 2/9] xfs_repair: remove gratuitous code block in phase5
2020-05-20 1:50 ` [PATCH 2/9] xfs_repair: remove gratuitous code block in phase5 Darrick J. Wong
@ 2020-05-27 12:15 ` Brian Foster
0 siblings, 0 replies; 30+ messages in thread
From: Brian Foster @ 2020-05-27 12:15 UTC (permalink / raw)
To: Darrick J. Wong; +Cc: sandeen, Eric Sandeen, linux-xfs
On Tue, May 19, 2020 at 06:50:55PM -0700, Darrick J. Wong wrote:
> From: Eric Sandeen <sandeen@redhat.com>
>
> A commit back in 2008 removed a "for" loop ahead of this code block, but
> left the indented code block in place. Remove it for clarity and reflow
> comments & lines as needed.
>
> Signed-off-by: Eric Sandeen <sandeen@redhat.com>
> Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
> ---
Reviewed-by: Brian Foster <bfoster@redhat.com>
> repair/phase5.c | 317 ++++++++++++++++++++++++++-----------------------------
> 1 file changed, 151 insertions(+), 166 deletions(-)
>
>
> diff --git a/repair/phase5.c b/repair/phase5.c
> index 677297fe..84c05a13 100644
> --- a/repair/phase5.c
> +++ b/repair/phase5.c
> @@ -2313,201 +2313,186 @@ phase5_func(
> if (verbose)
> do_log(_(" - agno = %d\n"), agno);
>
> - {
> - /*
> - * build up incore bno and bcnt extent btrees
> - */
> - num_extents = mk_incore_fstree(mp, agno);
> + /*
> + * build up incore bno and bcnt extent btrees
> + */
> + num_extents = mk_incore_fstree(mp, agno);
>
> #ifdef XR_BLD_FREE_TRACE
> - fprintf(stderr, "# of bno extents is %d\n",
> - count_bno_extents(agno));
> + fprintf(stderr, "# of bno extents is %d\n", count_bno_extents(agno));
> #endif
>
> - if (num_extents == 0) {
> - /*
> - * XXX - what we probably should do here is pick an
> - * inode for a regular file in the allocation group
> - * that has space allocated and shoot it by traversing
> - * the bmap list and putting all its extents on the
> - * incore freespace trees, clearing the inode,
> - * and clearing the in-use bit in the incore inode
> - * tree. Then try mk_incore_fstree() again.
> - */
> - do_error(_("unable to rebuild AG %u. "
> - "Not enough free space in on-disk AG.\n"),
> - agno);
> - }
> -
> - /*
> - * ok, now set up the btree cursors for the
> - * on-disk btrees (includs pre-allocating all
> - * required blocks for the trees themselves)
> - */
> - init_ino_cursor(mp, agno, &ino_btree_curs, &num_inos,
> - &num_free_inos, 0);
> -
> - if (xfs_sb_version_hasfinobt(&mp->m_sb))
> - init_ino_cursor(mp, agno, &fino_btree_curs,
> - &finobt_num_inos, &finobt_num_free_inos,
> - 1);
> -
> - sb_icount_ag[agno] += num_inos;
> - sb_ifree_ag[agno] += num_free_inos;
> -
> - /*
> - * Set up the btree cursors for the on-disk rmap btrees,
> - * which includes pre-allocating all required blocks.
> - */
> - init_rmapbt_cursor(mp, agno, &rmap_btree_curs);
> -
> + if (num_extents == 0) {
> /*
> - * Set up the btree cursors for the on-disk refcount btrees,
> - * which includes pre-allocating all required blocks.
> + * XXX - what we probably should do here is pick an inode for
> + * a regular file in the allocation group that has space
> + * allocated and shoot it by traversing the bmap list and
> + * putting all its extents on the incore freespace trees,
> + * clearing the inode, and clearing the in-use bit in the
> + * incore inode tree. Then try mk_incore_fstree() again.
> */
> - init_refc_cursor(mp, agno, &refcnt_btree_curs);
> -
> - num_extents = count_bno_extents_blocks(agno, &num_freeblocks);
> + do_error(
> +_("unable to rebuild AG %u. Not enough free space in on-disk AG.\n"),
> + agno);
> + }
> +
> + /*
> + * ok, now set up the btree cursors for the on-disk btrees (includes
> + * pre-allocating all required blocks for the trees themselves)
> + */
> + init_ino_cursor(mp, agno, &ino_btree_curs, &num_inos,
> + &num_free_inos, 0);
> +
> + if (xfs_sb_version_hasfinobt(&mp->m_sb))
> + init_ino_cursor(mp, agno, &fino_btree_curs, &finobt_num_inos,
> + &finobt_num_free_inos, 1);
> +
> + sb_icount_ag[agno] += num_inos;
> + sb_ifree_ag[agno] += num_free_inos;
> +
> + /*
> + * Set up the btree cursors for the on-disk rmap btrees, which includes
> + * pre-allocating all required blocks.
> + */
> + init_rmapbt_cursor(mp, agno, &rmap_btree_curs);
> +
> + /*
> + * Set up the btree cursors for the on-disk refcount btrees,
> + * which includes pre-allocating all required blocks.
> + */
> + init_refc_cursor(mp, agno, &refcnt_btree_curs);
> +
> + num_extents = count_bno_extents_blocks(agno, &num_freeblocks);
> + /*
> + * lose two blocks per AG -- the space tree roots are counted as
> + * allocated since the space trees always have roots
> + */
> + sb_fdblocks_ag[agno] += num_freeblocks - 2;
> +
> + if (num_extents == 0) {
> /*
> - * lose two blocks per AG -- the space tree roots
> - * are counted as allocated since the space trees
> - * always have roots
> + * XXX - what we probably should do here is pick an inode for
> + * a regular file in the allocation group that has space
> + * allocated and shoot it by traversing the bmap list and
> + * putting all its extents on the incore freespace trees,
> + * clearing the inode, and clearing the in-use bit in the
> + * incore inode tree. Then try mk_incore_fstree() again.
> */
> - sb_fdblocks_ag[agno] += num_freeblocks - 2;
> -
> - if (num_extents == 0) {
> - /*
> - * XXX - what we probably should do here is pick an
> - * inode for a regular file in the allocation group
> - * that has space allocated and shoot it by traversing
> - * the bmap list and putting all its extents on the
> - * incore freespace trees, clearing the inode,
> - * and clearing the in-use bit in the incore inode
> - * tree. Then try mk_incore_fstree() again.
> - */
> - do_error(
> - _("unable to rebuild AG %u. No free space.\n"), agno);
> - }
> + do_error(_("unable to rebuild AG %u. No free space.\n"), agno);
> + }
>
> #ifdef XR_BLD_FREE_TRACE
> - fprintf(stderr, "# of bno extents is %d\n", num_extents);
> + fprintf(stderr, "# of bno extents is %d\n", num_extents);
> #endif
>
> - /*
> - * track blocks that we might really lose
> - */
> - extra_blocks = calculate_freespace_cursor(mp, agno,
> - &num_extents, &bno_btree_curs);
> + /*
> + * track blocks that we might really lose
> + */
> + extra_blocks = calculate_freespace_cursor(mp, agno,
> + &num_extents, &bno_btree_curs);
>
> - /*
> - * freespace btrees live in the "free space" but
> - * the filesystem treats AGFL blocks as allocated
> - * since they aren't described by the freespace trees
> - */
> + /*
> + * freespace btrees live in the "free space" but the filesystem treats
> + * AGFL blocks as allocated since they aren't described by the
> + * freespace trees
> + */
>
> - /*
> - * see if we can fit all the extra blocks into the AGFL
> - */
> - extra_blocks = (extra_blocks - libxfs_agfl_size(mp) > 0)
> - ? extra_blocks - libxfs_agfl_size(mp)
> - : 0;
> + /*
> + * see if we can fit all the extra blocks into the AGFL
> + */
> + extra_blocks = (extra_blocks - libxfs_agfl_size(mp) > 0) ?
> + extra_blocks - libxfs_agfl_size(mp) : 0;
>
> - if (extra_blocks > 0)
> - sb_fdblocks_ag[agno] -= extra_blocks;
> + if (extra_blocks > 0)
> + sb_fdblocks_ag[agno] -= extra_blocks;
>
> - bcnt_btree_curs = bno_btree_curs;
> + bcnt_btree_curs = bno_btree_curs;
>
> - bno_btree_curs.owner = XFS_RMAP_OWN_AG;
> - bcnt_btree_curs.owner = XFS_RMAP_OWN_AG;
> - setup_cursor(mp, agno, &bno_btree_curs);
> - setup_cursor(mp, agno, &bcnt_btree_curs);
> + bno_btree_curs.owner = XFS_RMAP_OWN_AG;
> + bcnt_btree_curs.owner = XFS_RMAP_OWN_AG;
> + setup_cursor(mp, agno, &bno_btree_curs);
> + setup_cursor(mp, agno, &bcnt_btree_curs);
>
> #ifdef XR_BLD_FREE_TRACE
> - fprintf(stderr, "# of bno extents is %d\n",
> - count_bno_extents(agno));
> - fprintf(stderr, "# of bcnt extents is %d\n",
> - count_bcnt_extents(agno));
> + fprintf(stderr, "# of bno extents is %d\n", count_bno_extents(agno));
> + fprintf(stderr, "# of bcnt extents is %d\n", count_bcnt_extents(agno));
> #endif
>
> - /*
> - * now rebuild the freespace trees
> - */
> - freeblks1 = build_freespace_tree(mp, agno,
> + /*
> + * now rebuild the freespace trees
> + */
> + freeblks1 = build_freespace_tree(mp, agno,
> &bno_btree_curs, XFS_BTNUM_BNO);
> #ifdef XR_BLD_FREE_TRACE
> - fprintf(stderr, "# of free blocks == %d\n", freeblks1);
> + fprintf(stderr, "# of free blocks == %d\n", freeblks1);
> #endif
> - write_cursor(&bno_btree_curs);
> + write_cursor(&bno_btree_curs);
>
> #ifdef DEBUG
> - freeblks2 = build_freespace_tree(mp, agno,
> - &bcnt_btree_curs, XFS_BTNUM_CNT);
> + freeblks2 = build_freespace_tree(mp, agno,
> + &bcnt_btree_curs, XFS_BTNUM_CNT);
> #else
> - (void) build_freespace_tree(mp, agno,
> - &bcnt_btree_curs, XFS_BTNUM_CNT);
> + (void) build_freespace_tree(mp, agno, &bcnt_btree_curs, XFS_BTNUM_CNT);
> #endif
> - write_cursor(&bcnt_btree_curs);
> -
> - ASSERT(freeblks1 == freeblks2);
> -
> - if (xfs_sb_version_hasrmapbt(&mp->m_sb)) {
> - build_rmap_tree(mp, agno, &rmap_btree_curs);
> - write_cursor(&rmap_btree_curs);
> - sb_fdblocks_ag[agno] += (rmap_btree_curs.num_tot_blocks -
> - rmap_btree_curs.num_free_blocks) - 1;
> - }
> -
> - if (xfs_sb_version_hasreflink(&mp->m_sb)) {
> - build_refcount_tree(mp, agno, &refcnt_btree_curs);
> - write_cursor(&refcnt_btree_curs);
> - }
> -
> - /*
> - * set up agf and agfl
> - */
> - build_agf_agfl(mp, agno, &bno_btree_curs,
> - &bcnt_btree_curs, freeblks1, extra_blocks,
> - &rmap_btree_curs, &refcnt_btree_curs, lost_fsb);
> - /*
> - * build inode allocation tree.
> - */
> - build_ino_tree(mp, agno, &ino_btree_curs, XFS_BTNUM_INO,
> - &agi_stat);
> - write_cursor(&ino_btree_curs);
> -
> - /*
> - * build free inode tree
> - */
> - if (xfs_sb_version_hasfinobt(&mp->m_sb)) {
> - build_ino_tree(mp, agno, &fino_btree_curs,
> - XFS_BTNUM_FINO, NULL);
> - write_cursor(&fino_btree_curs);
> - }
> -
> - /* build the agi */
> - build_agi(mp, agno, &ino_btree_curs, &fino_btree_curs,
> - &agi_stat);
> -
> - /*
> - * tear down cursors
> - */
> - finish_cursor(&bno_btree_curs);
> - finish_cursor(&ino_btree_curs);
> - if (xfs_sb_version_hasrmapbt(&mp->m_sb))
> - finish_cursor(&rmap_btree_curs);
> - if (xfs_sb_version_hasreflink(&mp->m_sb))
> - finish_cursor(&refcnt_btree_curs);
> - if (xfs_sb_version_hasfinobt(&mp->m_sb))
> - finish_cursor(&fino_btree_curs);
> - finish_cursor(&bcnt_btree_curs);
> -
> - /*
> - * release the incore per-AG bno/bcnt trees so
> - * the extent nodes can be recycled
> - */
> - release_agbno_extent_tree(agno);
> - release_agbcnt_extent_tree(agno);
> + write_cursor(&bcnt_btree_curs);
> +
> + ASSERT(freeblks1 == freeblks2);
> +
> + if (xfs_sb_version_hasrmapbt(&mp->m_sb)) {
> + build_rmap_tree(mp, agno, &rmap_btree_curs);
> + write_cursor(&rmap_btree_curs);
> + sb_fdblocks_ag[agno] += (rmap_btree_curs.num_tot_blocks -
> + rmap_btree_curs.num_free_blocks) - 1;
> + }
> +
> + if (xfs_sb_version_hasreflink(&mp->m_sb)) {
> + build_refcount_tree(mp, agno, &refcnt_btree_curs);
> + write_cursor(&refcnt_btree_curs);
> }
> +
> + /*
> + * set up agf and agfl
> + */
> + build_agf_agfl(mp, agno, &bno_btree_curs,
> + &bcnt_btree_curs, freeblks1, extra_blocks,
> + &rmap_btree_curs, &refcnt_btree_curs, lost_fsb);
> + /*
> + * build inode allocation tree.
> + */
> + build_ino_tree(mp, agno, &ino_btree_curs, XFS_BTNUM_INO, &agi_stat);
> + write_cursor(&ino_btree_curs);
> +
> + /*
> + * build free inode tree
> + */
> + if (xfs_sb_version_hasfinobt(&mp->m_sb)) {
> + build_ino_tree(mp, agno, &fino_btree_curs,
> + XFS_BTNUM_FINO, NULL);
> + write_cursor(&fino_btree_curs);
> + }
> +
> + /* build the agi */
> + build_agi(mp, agno, &ino_btree_curs, &fino_btree_curs, &agi_stat);
> +
> + /*
> + * tear down cursors
> + */
> + finish_cursor(&bno_btree_curs);
> + finish_cursor(&ino_btree_curs);
> + if (xfs_sb_version_hasrmapbt(&mp->m_sb))
> + finish_cursor(&rmap_btree_curs);
> + if (xfs_sb_version_hasreflink(&mp->m_sb))
> + finish_cursor(&refcnt_btree_curs);
> + if (xfs_sb_version_hasfinobt(&mp->m_sb))
> + finish_cursor(&fino_btree_curs);
> + finish_cursor(&bcnt_btree_curs);
> +
> + /*
> + * release the incore per-AG bno/bcnt trees so the extent nodes
> + * can be recycled
> + */
> + release_agbno_extent_tree(agno);
> + release_agbcnt_extent_tree(agno);
> PROG_RPT_INC(prog_rpt_done[agno], 1);
> }
>
>
^ permalink raw reply [flat|nested] 30+ messages in thread
* Re: [PATCH 3/9] xfs_repair: create a new class of btree rebuild cursors
2020-05-20 1:51 ` [PATCH 3/9] xfs_repair: create a new class of btree rebuild cursors Darrick J. Wong
@ 2020-05-27 12:18 ` Brian Foster
2020-05-27 22:07 ` Darrick J. Wong
0 siblings, 1 reply; 30+ messages in thread
From: Brian Foster @ 2020-05-27 12:18 UTC (permalink / raw)
To: Darrick J. Wong; +Cc: sandeen, linux-xfs
On Tue, May 19, 2020 at 06:51:02PM -0700, Darrick J. Wong wrote:
> From: Darrick J. Wong <darrick.wong@oracle.com>
>
> Create some new support structures and functions to assist phase5 in
> using the btree bulk loader to reconstruct metadata btrees. This is the
> first step in removing the open-coded rebuilding code.
>
> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
> ---
> repair/phase5.c | 239 ++++++++++++++++++++++++++++++++++++++++++++++++++-----
> 1 file changed, 218 insertions(+), 21 deletions(-)
>
>
> diff --git a/repair/phase5.c b/repair/phase5.c
> index 84c05a13..8f5e5f59 100644
> --- a/repair/phase5.c
> +++ b/repair/phase5.c
> @@ -18,6 +18,7 @@
> #include "progress.h"
> #include "slab.h"
> #include "rmap.h"
> +#include "bload.h"
>
> /*
> * we maintain the current slice (path from root to leaf)
...
> @@ -306,6 +324,156 @@ _("error - not enough free space in filesystem\n"));
> #endif
> }
>
...
> +static void
> +consume_freespace(
> + xfs_agnumber_t agno,
> + struct extent_tree_node *ext_ptr,
> + uint32_t len)
> +{
> + struct extent_tree_node *bno_ext_ptr;
> + xfs_agblock_t new_start = ext_ptr->ex_startblock + len;
> + xfs_extlen_t new_len = ext_ptr->ex_blockcount - len;
> +
> + /* Delete the used-up extent from both extent trees. */
> +#ifdef XR_BLD_FREE_TRACE
> + fprintf(stderr, "releasing extent: %u [%u %u]\n", agno,
> + ext_ptr->ex_startblock, ext_ptr->ex_blockcount);
> +#endif
> + bno_ext_ptr = find_bno_extent(agno, ext_ptr->ex_startblock);
> + ASSERT(bno_ext_ptr != NULL);
> + get_bno_extent(agno, bno_ext_ptr);
> + release_extent_tree_node(bno_ext_ptr);
> +
> + ext_ptr = get_bcnt_extent(agno, ext_ptr->ex_startblock,
> + ext_ptr->ex_blockcount);
> + release_extent_tree_node(ext_ptr);
> +
Not having looked too deeply at the in-core extent tracking structures,
is there any particular reason we unconditionally remove and reinsert
new records each time around? Is it because we're basically changing the
extent index in the tree? If so, comment please (an update to the
comment below is probably fine). :)
> + /*
> + * If we only used part of this last extent, then we must reinsert the
> + * extent in the extent trees.
> + */
> + if (new_len > 0) {
> + add_bno_extent(agno, new_start, new_len);
> + add_bcnt_extent(agno, new_start, new_len);
> + }
> +}
> +
> +/* Reserve blocks for the new btree. */
> +static void
> +setup_rebuild(
> + struct xfs_mount *mp,
> + xfs_agnumber_t agno,
> + struct bt_rebuild *btr,
> + uint32_t nr_blocks)
> +{
> + struct extent_tree_node *ext_ptr;
> + uint32_t blocks_allocated = 0;
> + uint32_t len;
> + int error;
> +
> + while (blocks_allocated < nr_blocks) {
> + /*
> + * Grab the smallest extent and use it up, then get the
> + * next smallest. This mimics the init_*_cursor code.
> + */
> + ext_ptr = findfirst_bcnt_extent(agno);
Extra whitespace ^
> + if (!ext_ptr)
> + do_error(
> +_("error - not enough free space in filesystem\n"));
> +
> + /* Use up the extent we've got. */
> + len = min(ext_ptr->ex_blockcount, nr_blocks - blocks_allocated);
> + error = xrep_newbt_add_blocks(&btr->newbt,
> + XFS_AGB_TO_FSB(mp, agno,
> + ext_ptr->ex_startblock),
> + len);
Alignment.
> + if (error)
> + do_error(_("could not set up btree reservation: %s\n"),
> + strerror(-error));
> +
> + error = rmap_add_ag_rec(mp, agno, ext_ptr->ex_startblock, len,
> + btr->newbt.oinfo.oi_owner);
> + if (error)
> + do_error(_("could not set up btree rmaps: %s\n"),
> + strerror(-error));
> +
> + consume_freespace(agno, ext_ptr, len);
> + blocks_allocated += len;
> + }
> +#ifdef XR_BLD_FREE_TRACE
> + fprintf(stderr, "blocks_allocated = %d\n",
> + blocks_allocated);
> +#endif
> +}
> +
> +/* Feed one of the new btree blocks to the bulk loader. */
> +static int
> +rebuild_claim_block(
> + struct xfs_btree_cur *cur,
> + union xfs_btree_ptr *ptr,
> + void *priv)
> +{
> + struct bt_rebuild *btr = priv;
> +
> + return xrep_newbt_claim_block(cur, &btr->newbt, ptr);
> +}
> +
Seems like an unnecessary helper, unless this grows more code in later
patches..?
> static void
> write_cursor(bt_status_t *curs)
> {
...
> @@ -2287,28 +2483,29 @@ keep_fsinos(xfs_mount_t *mp)
>
> static void
> phase5_func(
> - xfs_mount_t *mp,
> - xfs_agnumber_t agno,
> - struct xfs_slab *lost_fsb)
> + struct xfs_mount *mp,
> + xfs_agnumber_t agno,
> + struct xfs_slab *lost_fsb)
> {
> - uint64_t num_inos;
> - uint64_t num_free_inos;
> - uint64_t finobt_num_inos;
> - uint64_t finobt_num_free_inos;
> - bt_status_t bno_btree_curs;
> - bt_status_t bcnt_btree_curs;
> - bt_status_t ino_btree_curs;
> - bt_status_t fino_btree_curs;
> - bt_status_t rmap_btree_curs;
> - bt_status_t refcnt_btree_curs;
> - int extra_blocks = 0;
> - uint num_freeblocks;
> - xfs_extlen_t freeblks1;
> + struct repair_ctx sc = { .mp = mp, };
I don't see any reason to add sc here when it's still unused. It's not
as if a single variable is saving complexity somewhere else. I guess
I'll defer to Eric on the approach wrt to the other unused warnings.
Also, what's the purpose of the rmap change below? I'm wondering if that
(along with all of the indentation cleanup) should be its own patch with
appropriate explanation.
Brian
> + struct agi_stat agi_stat = {0,};
> + uint64_t num_inos;
> + uint64_t num_free_inos;
> + uint64_t finobt_num_inos;
> + uint64_t finobt_num_free_inos;
> + bt_status_t bno_btree_curs;
> + bt_status_t bcnt_btree_curs;
> + bt_status_t ino_btree_curs;
> + bt_status_t fino_btree_curs;
> + bt_status_t rmap_btree_curs;
> + bt_status_t refcnt_btree_curs;
> + int extra_blocks = 0;
> + uint num_freeblocks;
> + xfs_extlen_t freeblks1;
> #ifdef DEBUG
> - xfs_extlen_t freeblks2;
> + xfs_extlen_t freeblks2;
> #endif
> - xfs_agblock_t num_extents;
> - struct agi_stat agi_stat = {0,};
> + xfs_agblock_t num_extents;
>
> if (verbose)
> do_log(_(" - agno = %d\n"), agno);
> @@ -2516,8 +2713,8 @@ inject_lost_blocks(
> if (error)
> goto out_cancel;
>
> - error = -libxfs_free_extent(tp, *fsb, 1, &XFS_RMAP_OINFO_AG,
> - XFS_AG_RESV_NONE);
> + error = -libxfs_free_extent(tp, *fsb, 1,
> + &XFS_RMAP_OINFO_ANY_OWNER, XFS_AG_RESV_NONE);
> if (error)
> goto out_cancel;
>
>
^ permalink raw reply [flat|nested] 30+ messages in thread
* Re: [PATCH 3/9] xfs_repair: create a new class of btree rebuild cursors
2020-05-27 12:18 ` Brian Foster
@ 2020-05-27 22:07 ` Darrick J. Wong
2020-05-28 15:09 ` Brian Foster
0 siblings, 1 reply; 30+ messages in thread
From: Darrick J. Wong @ 2020-05-27 22:07 UTC (permalink / raw)
To: Brian Foster; +Cc: sandeen, linux-xfs
On Wed, May 27, 2020 at 08:18:04AM -0400, Brian Foster wrote:
> On Tue, May 19, 2020 at 06:51:02PM -0700, Darrick J. Wong wrote:
> > From: Darrick J. Wong <darrick.wong@oracle.com>
> >
> > Create some new support structures and functions to assist phase5 in
> > using the btree bulk loader to reconstruct metadata btrees. This is the
> > first step in removing the open-coded rebuilding code.
> >
> > Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
> > ---
> > repair/phase5.c | 239 ++++++++++++++++++++++++++++++++++++++++++++++++++-----
> > 1 file changed, 218 insertions(+), 21 deletions(-)
> >
> >
> > diff --git a/repair/phase5.c b/repair/phase5.c
> > index 84c05a13..8f5e5f59 100644
> > --- a/repair/phase5.c
> > +++ b/repair/phase5.c
> > @@ -18,6 +18,7 @@
> > #include "progress.h"
> > #include "slab.h"
> > #include "rmap.h"
> > +#include "bload.h"
> >
> > /*
> > * we maintain the current slice (path from root to leaf)
> ...
> > @@ -306,6 +324,156 @@ _("error - not enough free space in filesystem\n"));
> > #endif
> > }
> >
> ...
> > +static void
> > +consume_freespace(
> > + xfs_agnumber_t agno,
> > + struct extent_tree_node *ext_ptr,
> > + uint32_t len)
> > +{
> > + struct extent_tree_node *bno_ext_ptr;
> > + xfs_agblock_t new_start = ext_ptr->ex_startblock + len;
> > + xfs_extlen_t new_len = ext_ptr->ex_blockcount - len;
> > +
> > + /* Delete the used-up extent from both extent trees. */
> > +#ifdef XR_BLD_FREE_TRACE
> > + fprintf(stderr, "releasing extent: %u [%u %u]\n", agno,
> > + ext_ptr->ex_startblock, ext_ptr->ex_blockcount);
> > +#endif
> > + bno_ext_ptr = find_bno_extent(agno, ext_ptr->ex_startblock);
> > + ASSERT(bno_ext_ptr != NULL);
> > + get_bno_extent(agno, bno_ext_ptr);
> > + release_extent_tree_node(bno_ext_ptr);
> > +
> > + ext_ptr = get_bcnt_extent(agno, ext_ptr->ex_startblock,
> > + ext_ptr->ex_blockcount);
> > + release_extent_tree_node(ext_ptr);
> > +
>
> Not having looked too deeply at the in-core extent tracking structures,
> is there any particular reason we unconditionally remove and reinsert
> new records each time around? Is it because we're basically changing the
> extent index in the tree? If so, comment please (an update to the
> comment below is probably fine). :)
Yes. We're changing the free space tree records, and the incore bno and
cnt trees maintain the records in sorted order. Therefore, if we want
to change a record we have to delete the record from the tree and
reinsert it.
> > + /*
> > + * If we only used part of this last extent, then we must reinsert the
> > + * extent in the extent trees.
/*
* If we only used part of this last extent, then we must reinsert the
* extent to maintain proper sorting order.
*/
How about that?
> > + */
> > + if (new_len > 0) {
> > + add_bno_extent(agno, new_start, new_len);
> > + add_bcnt_extent(agno, new_start, new_len);
> > + }
> > +}
> > +
> > +/* Reserve blocks for the new btree. */
> > +static void
> > +setup_rebuild(
> > + struct xfs_mount *mp,
> > + xfs_agnumber_t agno,
> > + struct bt_rebuild *btr,
> > + uint32_t nr_blocks)
> > +{
> > + struct extent_tree_node *ext_ptr;
> > + uint32_t blocks_allocated = 0;
> > + uint32_t len;
> > + int error;
> > +
> > + while (blocks_allocated < nr_blocks) {
> > + /*
> > + * Grab the smallest extent and use it up, then get the
> > + * next smallest. This mimics the init_*_cursor code.
> > + */
> > + ext_ptr = findfirst_bcnt_extent(agno);
>
> Extra whitespace ^
>
> > + if (!ext_ptr)
> > + do_error(
> > +_("error - not enough free space in filesystem\n"));
> > +
> > + /* Use up the extent we've got. */
> > + len = min(ext_ptr->ex_blockcount, nr_blocks - blocks_allocated);
> > + error = xrep_newbt_add_blocks(&btr->newbt,
> > + XFS_AGB_TO_FSB(mp, agno,
> > + ext_ptr->ex_startblock),
> > + len);
>
> Alignment.
Will fix both of these.
> > + if (error)
> > + do_error(_("could not set up btree reservation: %s\n"),
> > + strerror(-error));
> > +
> > + error = rmap_add_ag_rec(mp, agno, ext_ptr->ex_startblock, len,
> > + btr->newbt.oinfo.oi_owner);
> > + if (error)
> > + do_error(_("could not set up btree rmaps: %s\n"),
> > + strerror(-error));
> > +
> > + consume_freespace(agno, ext_ptr, len);
> > + blocks_allocated += len;
> > + }
> > +#ifdef XR_BLD_FREE_TRACE
> > + fprintf(stderr, "blocks_allocated = %d\n",
> > + blocks_allocated);
> > +#endif
> > +}
> > +
> > +/* Feed one of the new btree blocks to the bulk loader. */
> > +static int
> > +rebuild_claim_block(
> > + struct xfs_btree_cur *cur,
> > + union xfs_btree_ptr *ptr,
> > + void *priv)
> > +{
> > + struct bt_rebuild *btr = priv;
> > +
> > + return xrep_newbt_claim_block(cur, &btr->newbt, ptr);
> > +}
> > +
>
> Seems like an unnecessary helper, unless this grows more code in later
> patches..?
It doesn't grow any more code, but keep in mind that get_record,
claim_block, and iroot_size are all callbacks of xfs_btree_bload(). The
priv parameter passed to that function are passed unchanged to the three
callbacks. The bulk load code doesn't know anything about where the
blocks or the records come from, so this is how both repairs will pass
that information to the callbacks.
> > static void
> > write_cursor(bt_status_t *curs)
> > {
> ...
> > @@ -2287,28 +2483,29 @@ keep_fsinos(xfs_mount_t *mp)
> >
> > static void
> > phase5_func(
> > - xfs_mount_t *mp,
> > - xfs_agnumber_t agno,
> > - struct xfs_slab *lost_fsb)
> > + struct xfs_mount *mp,
> > + xfs_agnumber_t agno,
> > + struct xfs_slab *lost_fsb)
> > {
> > - uint64_t num_inos;
> > - uint64_t num_free_inos;
> > - uint64_t finobt_num_inos;
> > - uint64_t finobt_num_free_inos;
> > - bt_status_t bno_btree_curs;
> > - bt_status_t bcnt_btree_curs;
> > - bt_status_t ino_btree_curs;
> > - bt_status_t fino_btree_curs;
> > - bt_status_t rmap_btree_curs;
> > - bt_status_t refcnt_btree_curs;
> > - int extra_blocks = 0;
> > - uint num_freeblocks;
> > - xfs_extlen_t freeblks1;
> > + struct repair_ctx sc = { .mp = mp, };
>
> I don't see any reason to add sc here when it's still unused. It's not
> as if a single variable is saving complexity somewhere else. I guess
> I'll defer to Eric on the approach wrt to the other unused warnings.
<shrug> I'll ask. It seems dumb to have a prep patch that adds a bunch
of symbols that won't get used until the next patch, but OTOH combining
the two will make for a ~40K patch.
> Also, what's the purpose of the rmap change below? I'm wondering if that
> (along with all of the indentation cleanup) should be its own patch with
> appropriate explanation.
Errk, that one definitely should be separate.
> Brian
>
> > + struct agi_stat agi_stat = {0,};
> > + uint64_t num_inos;
> > + uint64_t num_free_inos;
> > + uint64_t finobt_num_inos;
> > + uint64_t finobt_num_free_inos;
> > + bt_status_t bno_btree_curs;
> > + bt_status_t bcnt_btree_curs;
> > + bt_status_t ino_btree_curs;
> > + bt_status_t fino_btree_curs;
> > + bt_status_t rmap_btree_curs;
> > + bt_status_t refcnt_btree_curs;
> > + int extra_blocks = 0;
> > + uint num_freeblocks;
> > + xfs_extlen_t freeblks1;
> > #ifdef DEBUG
> > - xfs_extlen_t freeblks2;
> > + xfs_extlen_t freeblks2;
> > #endif
> > - xfs_agblock_t num_extents;
> > - struct agi_stat agi_stat = {0,};
> > + xfs_agblock_t num_extents;
> >
> > if (verbose)
> > do_log(_(" - agno = %d\n"), agno);
> > @@ -2516,8 +2713,8 @@ inject_lost_blocks(
> > if (error)
> > goto out_cancel;
> >
> > - error = -libxfs_free_extent(tp, *fsb, 1, &XFS_RMAP_OINFO_AG,
> > - XFS_AG_RESV_NONE);
> > + error = -libxfs_free_extent(tp, *fsb, 1,
> > + &XFS_RMAP_OINFO_ANY_OWNER, XFS_AG_RESV_NONE);
> > if (error)
> > goto out_cancel;
> >
> >
>
^ permalink raw reply [flat|nested] 30+ messages in thread
* Re: [PATCH 1/9] xfs_repair: port the online repair newbt structure
2020-05-27 12:15 ` Brian Foster
@ 2020-05-27 22:34 ` Darrick J. Wong
2020-05-28 15:08 ` Brian Foster
0 siblings, 1 reply; 30+ messages in thread
From: Darrick J. Wong @ 2020-05-27 22:34 UTC (permalink / raw)
To: Brian Foster; +Cc: sandeen, linux-xfs
On Wed, May 27, 2020 at 08:15:31AM -0400, Brian Foster wrote:
> On Tue, May 19, 2020 at 06:50:49PM -0700, Darrick J. Wong wrote:
> > From: Darrick J. Wong <darrick.wong@oracle.com>
> >
> > Port the new btree staging context and related block reservation helper
> > code from the kernel to repair. We'll use this in subsequent patches to
> > implement btree bulk loading.
> >
> > Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
> > ---
> > include/libxfs.h | 1
> > libxfs/libxfs_api_defs.h | 2
> > repair/Makefile | 4 -
> > repair/bload.c | 303 ++++++++++++++++++++++++++++++++++++++++++++++
> > repair/bload.h | 77 ++++++++++++
> > repair/xfs_repair.c | 17 +++
> > 6 files changed, 402 insertions(+), 2 deletions(-)
> > create mode 100644 repair/bload.c
> > create mode 100644 repair/bload.h
> >
> >
> ...
> > diff --git a/repair/bload.c b/repair/bload.c
> > new file mode 100644
> > index 00000000..9bc17468
> > --- /dev/null
> > +++ b/repair/bload.c
> > @@ -0,0 +1,303 @@
> > +// SPDX-License-Identifier: GPL-2.0-or-later
> > +/*
> > + * Copyright (C) 2020 Oracle. All Rights Reserved.
> > + * Author: Darrick J. Wong <darrick.wong@oracle.com>
> > + */
> > +#include <libxfs.h>
> > +#include "bload.h"
> > +
> > +#define trace_xrep_newbt_claim_block(...) ((void) 0)
> > +#define trace_xrep_newbt_free_blocks(...) ((void) 0)
> > +
> > +int bload_leaf_slack = -1;
> > +int bload_node_slack = -1;
> > +
> > +/* Ported routines from fs/xfs/scrub/repair.c */
> > +
>
> Looks mostly straightforward, but I'll have to come back to this as I
> get to the code that uses it later in the series. In the meantime, I see
> some of these helpers in scrub/repair.c while not others. Are there
> references to other routines that are intended to be copies from kernel
> code?
Hm. I might not understand the question, but in general the code should
be fairly similar to the kernel functions. The biggest differences are
(a) that whole libxfs error code mess, (b) the much simpler repair_ctx
structure, and (c) the fact that repair doesn't bother with EFIs to
automatically reap blocks.
So... the ten functions you see here do the same things as their kernel
counterparts, but they get to do it in the much simpler userspace
environment.
The other functions in scrub/repair.c that didn't get ported are either
for other types of repairs or exist to support the in-kernel code and
aren't needed here.
--D
> Brian
>
> > +/*
> > + * Roll a transaction, keeping the AG headers locked and reinitializing
> > + * the btree cursors.
> > + */
> > +int
> > +xrep_roll_ag_trans(
> > + struct repair_ctx *sc)
> > +{
> > + int error;
> > +
> > + /* Keep the AG header buffers locked so we can keep going. */
> > + if (sc->agi_bp)
> > + libxfs_trans_bhold(sc->tp, sc->agi_bp);
> > + if (sc->agf_bp)
> > + libxfs_trans_bhold(sc->tp, sc->agf_bp);
> > + if (sc->agfl_bp)
> > + libxfs_trans_bhold(sc->tp, sc->agfl_bp);
> > +
> > + /*
> > + * Roll the transaction. We still own the buffer and the buffer lock
> > + * regardless of whether or not the roll succeeds. If the roll fails,
> > + * the buffers will be released during teardown on our way out of the
> > + * kernel. If it succeeds, we join them to the new transaction and
> > + * move on.
> > + */
> > + error = -libxfs_trans_roll(&sc->tp);
> > + if (error)
> > + return error;
> > +
> > + /* Join AG headers to the new transaction. */
> > + if (sc->agi_bp)
> > + libxfs_trans_bjoin(sc->tp, sc->agi_bp);
> > + if (sc->agf_bp)
> > + libxfs_trans_bjoin(sc->tp, sc->agf_bp);
> > + if (sc->agfl_bp)
> > + libxfs_trans_bjoin(sc->tp, sc->agfl_bp);
> > +
> > + return 0;
> > +}
> > +
> > +/* Initialize accounting resources for staging a new AG btree. */
> > +void
> > +xrep_newbt_init_ag(
> > + struct xrep_newbt *xnr,
> > + struct repair_ctx *sc,
> > + const struct xfs_owner_info *oinfo,
> > + xfs_fsblock_t alloc_hint,
> > + enum xfs_ag_resv_type resv)
> > +{
> > + memset(xnr, 0, sizeof(struct xrep_newbt));
> > + xnr->sc = sc;
> > + xnr->oinfo = *oinfo; /* structure copy */
> > + xnr->alloc_hint = alloc_hint;
> > + xnr->resv = resv;
> > + INIT_LIST_HEAD(&xnr->resv_list);
> > +}
> > +
> > +/* Initialize accounting resources for staging a new inode fork btree. */
> > +void
> > +xrep_newbt_init_inode(
> > + struct xrep_newbt *xnr,
> > + struct repair_ctx *sc,
> > + int whichfork,
> > + const struct xfs_owner_info *oinfo)
> > +{
> > + xrep_newbt_init_ag(xnr, sc, oinfo,
> > + XFS_INO_TO_FSB(sc->mp, sc->ip->i_ino),
> > + XFS_AG_RESV_NONE);
> > + xnr->ifake.if_fork = kmem_zone_zalloc(xfs_ifork_zone, 0);
> > + xnr->ifake.if_fork_size = XFS_IFORK_SIZE(sc->ip, whichfork);
> > +}
> > +
> > +/*
> > + * Initialize accounting resources for staging a new btree. Callers are
> > + * expected to add their own reservations (and clean them up) manually.
> > + */
> > +void
> > +xrep_newbt_init_bare(
> > + struct xrep_newbt *xnr,
> > + struct repair_ctx *sc)
> > +{
> > + xrep_newbt_init_ag(xnr, sc, &XFS_RMAP_OINFO_ANY_OWNER, NULLFSBLOCK,
> > + XFS_AG_RESV_NONE);
> > +}
> > +
> > +/* Designate specific blocks to be used to build our new btree. */
> > +int
> > +xrep_newbt_add_blocks(
> > + struct xrep_newbt *xnr,
> > + xfs_fsblock_t fsbno,
> > + xfs_extlen_t len)
> > +{
> > + struct xrep_newbt_resv *resv;
> > +
> > + resv = kmem_alloc(sizeof(struct xrep_newbt_resv), KM_MAYFAIL);
> > + if (!resv)
> > + return ENOMEM;
> > +
> > + INIT_LIST_HEAD(&resv->list);
> > + resv->fsbno = fsbno;
> > + resv->len = len;
> > + resv->used = 0;
> > + list_add_tail(&resv->list, &xnr->resv_list);
> > + return 0;
> > +}
> > +
> > +/* Reserve disk space for our new btree. */
> > +int
> > +xrep_newbt_alloc_blocks(
> > + struct xrep_newbt *xnr,
> > + uint64_t nr_blocks)
> > +{
> > + struct repair_ctx *sc = xnr->sc;
> > + xfs_alloctype_t type;
> > + xfs_fsblock_t alloc_hint = xnr->alloc_hint;
> > + int error = 0;
> > +
> > + type = sc->ip ? XFS_ALLOCTYPE_START_BNO : XFS_ALLOCTYPE_NEAR_BNO;
> > +
> > + while (nr_blocks > 0 && !error) {
> > + struct xfs_alloc_arg args = {
> > + .tp = sc->tp,
> > + .mp = sc->mp,
> > + .type = type,
> > + .fsbno = alloc_hint,
> > + .oinfo = xnr->oinfo,
> > + .minlen = 1,
> > + .maxlen = nr_blocks,
> > + .prod = 1,
> > + .resv = xnr->resv,
> > + };
> > +
> > + error = -libxfs_alloc_vextent(&args);
> > + if (error)
> > + return error;
> > + if (args.fsbno == NULLFSBLOCK)
> > + return ENOSPC;
> > +
> > + /* We don't have real EFIs here so skip that. */
> > +
> > + error = xrep_newbt_add_blocks(xnr, args.fsbno, args.len);
> > + if (error)
> > + break;
> > +
> > + nr_blocks -= args.len;
> > + alloc_hint = args.fsbno + args.len - 1;
> > +
> > + if (sc->ip)
> > + error = -libxfs_trans_roll_inode(&sc->tp, sc->ip);
> > + else
> > + error = xrep_roll_ag_trans(sc);
> > + }
> > +
> > + return error;
> > +}
> > +
> > +/*
> > + * Release blocks that were reserved for a btree repair. If the repair
> > + * succeeded then we log deferred frees for unused blocks. Otherwise, we try
> > + * to free the extents immediately to roll the filesystem back to where it was
> > + * before we started.
> > + */
> > +static inline int
> > +xrep_newbt_destroy_reservation(
> > + struct xrep_newbt *xnr,
> > + struct xrep_newbt_resv *resv,
> > + bool cancel_repair)
> > +{
> > + struct repair_ctx *sc = xnr->sc;
> > +
> > + if (cancel_repair) {
> > + int error;
> > +
> > + /* Free the extent then roll the transaction. */
> > + error = -libxfs_free_extent(sc->tp, resv->fsbno, resv->len,
> > + &xnr->oinfo, xnr->resv);
> > + if (error)
> > + return error;
> > +
> > + if (sc->ip)
> > + return -libxfs_trans_roll_inode(&sc->tp, sc->ip);
> > + return xrep_roll_ag_trans(sc);
> > + }
> > +
> > + /* We don't have EFIs here so skip the EFD. */
> > +
> > + /*
> > + * Use the deferred freeing mechanism to schedule for deletion any
> > + * blocks we didn't use to rebuild the tree. This enables us to log
> > + * them all in the same transaction as the root change.
> > + */
> > + resv->fsbno += resv->used;
> > + resv->len -= resv->used;
> > + resv->used = 0;
> > +
> > + if (resv->len == 0)
> > + return 0;
> > +
> > + trace_xrep_newbt_free_blocks(sc->mp,
> > + XFS_FSB_TO_AGNO(sc->mp, resv->fsbno),
> > + XFS_FSB_TO_AGBNO(sc->mp, resv->fsbno),
> > + resv->len, xnr->oinfo.oi_owner);
> > +
> > + __xfs_bmap_add_free(sc->tp, resv->fsbno, resv->len, &xnr->oinfo, true);
> > +
> > + return 0;
> > +}
> > +
> > +/* Free all the accounting info and disk space we reserved for a new btree. */
> > +void
> > +xrep_newbt_destroy(
> > + struct xrep_newbt *xnr,
> > + int error)
> > +{
> > + struct repair_ctx *sc = xnr->sc;
> > + struct xrep_newbt_resv *resv, *n;
> > + int err2;
> > +
> > + list_for_each_entry_safe(resv, n, &xnr->resv_list, list) {
> > + err2 = xrep_newbt_destroy_reservation(xnr, resv, error != 0);
> > + if (err2)
> > + goto junkit;
> > +
> > + list_del(&resv->list);
> > + kmem_free(resv);
> > + }
> > +
> > +junkit:
> > + /*
> > + * If we still have reservations attached to @newbt, cleanup must have
> > + * failed and the filesystem is about to go down. Clean up the incore
> > + * reservations.
> > + */
> > + list_for_each_entry_safe(resv, n, &xnr->resv_list, list) {
> > + list_del(&resv->list);
> > + kmem_free(resv);
> > + }
> > +
> > + if (sc->ip) {
> > + kmem_cache_free(xfs_ifork_zone, xnr->ifake.if_fork);
> > + xnr->ifake.if_fork = NULL;
> > + }
> > +}
> > +
> > +/* Feed one of the reserved btree blocks to the bulk loader. */
> > +int
> > +xrep_newbt_claim_block(
> > + struct xfs_btree_cur *cur,
> > + struct xrep_newbt *xnr,
> > + union xfs_btree_ptr *ptr)
> > +{
> > + struct xrep_newbt_resv *resv;
> > + xfs_fsblock_t fsb;
> > +
> > + /*
> > + * The first item in the list should always have a free block unless
> > + * we're completely out.
> > + */
> > + resv = list_first_entry(&xnr->resv_list, struct xrep_newbt_resv, list);
> > + if (resv->used == resv->len)
> > + return ENOSPC;
> > +
> > + /*
> > + * Peel off a block from the start of the reservation. We allocate
> > + * blocks in order to place blocks on disk in increasing record or key
> > + * order. The block reservations tend to end up on the list in
> > + * decreasing order, which hopefully results in leaf blocks ending up
> > + * together.
> > + */
> > + fsb = resv->fsbno + resv->used;
> > + resv->used++;
> > +
> > + /* If we used all the blocks in this reservation, move it to the end. */
> > + if (resv->used == resv->len)
> > + list_move_tail(&resv->list, &xnr->resv_list);
> > +
> > + trace_xrep_newbt_claim_block(cur->bc_mp,
> > + XFS_FSB_TO_AGNO(cur->bc_mp, fsb),
> > + XFS_FSB_TO_AGBNO(cur->bc_mp, fsb),
> > + 1, xnr->oinfo.oi_owner);
> > +
> > + if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
> > + ptr->l = cpu_to_be64(fsb);
> > + else
> > + ptr->s = cpu_to_be32(XFS_FSB_TO_AGBNO(cur->bc_mp, fsb));
> > + return 0;
> > +}
> > diff --git a/repair/bload.h b/repair/bload.h
> > new file mode 100644
> > index 00000000..020c4834
> > --- /dev/null
> > +++ b/repair/bload.h
> > @@ -0,0 +1,77 @@
> > +// SPDX-License-Identifier: GPL-2.0-or-later
> > +/*
> > + * Copyright (C) 2020 Oracle. All Rights Reserved.
> > + * Author: Darrick J. Wong <darrick.wong@oracle.com>
> > + */
> > +#ifndef __XFS_REPAIR_BLOAD_H__
> > +#define __XFS_REPAIR_BLOAD_H__
> > +
> > +extern int bload_leaf_slack;
> > +extern int bload_node_slack;
> > +
> > +struct repair_ctx {
> > + struct xfs_mount *mp;
> > + struct xfs_inode *ip;
> > + struct xfs_trans *tp;
> > +
> > + struct xfs_buf *agi_bp;
> > + struct xfs_buf *agf_bp;
> > + struct xfs_buf *agfl_bp;
> > +};
> > +
> > +struct xrep_newbt_resv {
> > + /* Link to list of extents that we've reserved. */
> > + struct list_head list;
> > +
> > + /* FSB of the block we reserved. */
> > + xfs_fsblock_t fsbno;
> > +
> > + /* Length of the reservation. */
> > + xfs_extlen_t len;
> > +
> > + /* How much of this reservation we've used. */
> > + xfs_extlen_t used;
> > +};
> > +
> > +struct xrep_newbt {
> > + struct repair_ctx *sc;
> > +
> > + /* List of extents that we've reserved. */
> > + struct list_head resv_list;
> > +
> > + /* Fake root for new btree. */
> > + union {
> > + struct xbtree_afakeroot afake;
> > + struct xbtree_ifakeroot ifake;
> > + };
> > +
> > + /* rmap owner of these blocks */
> > + struct xfs_owner_info oinfo;
> > +
> > + /* The last reservation we allocated from. */
> > + struct xrep_newbt_resv *last_resv;
> > +
> > + /* Allocation hint */
> > + xfs_fsblock_t alloc_hint;
> > +
> > + /* per-ag reservation type */
> > + enum xfs_ag_resv_type resv;
> > +};
> > +
> > +#define for_each_xrep_newbt_reservation(xnr, resv, n) \
> > + list_for_each_entry_safe((resv), (n), &(xnr)->resv_list, list)
> > +
> > +void xrep_newbt_init_bare(struct xrep_newbt *xnr, struct repair_ctx *sc);
> > +void xrep_newbt_init_ag(struct xrep_newbt *xnr, struct repair_ctx *sc,
> > + const struct xfs_owner_info *oinfo, xfs_fsblock_t alloc_hint,
> > + enum xfs_ag_resv_type resv);
> > +void xrep_newbt_init_inode(struct xrep_newbt *xnr, struct repair_ctx *sc,
> > + int whichfork, const struct xfs_owner_info *oinfo);
> > +int xrep_newbt_add_blocks(struct xrep_newbt *xnr, xfs_fsblock_t fsbno,
> > + xfs_extlen_t len);
> > +int xrep_newbt_alloc_blocks(struct xrep_newbt *xnr, uint64_t nr_blocks);
> > +void xrep_newbt_destroy(struct xrep_newbt *xnr, int error);
> > +int xrep_newbt_claim_block(struct xfs_btree_cur *cur, struct xrep_newbt *xnr,
> > + union xfs_btree_ptr *ptr);
> > +
> > +#endif /* __XFS_REPAIR_BLOAD_H__ */
> > diff --git a/repair/xfs_repair.c b/repair/xfs_repair.c
> > index 9d72fa8e..8fbd3649 100644
> > --- a/repair/xfs_repair.c
> > +++ b/repair/xfs_repair.c
> > @@ -24,6 +24,7 @@
> > #include "rmap.h"
> > #include "libfrog/fsgeom.h"
> > #include "libfrog/platform.h"
> > +#include "bload.h"
> >
> > /*
> > * option tables for getsubopt calls
> > @@ -39,6 +40,8 @@ enum o_opt_nums {
> > AG_STRIDE,
> > FORCE_GEO,
> > PHASE2_THREADS,
> > + BLOAD_LEAF_SLACK,
> > + BLOAD_NODE_SLACK,
> > O_MAX_OPTS,
> > };
> >
> > @@ -49,6 +52,8 @@ static char *o_opts[] = {
> > [AG_STRIDE] = "ag_stride",
> > [FORCE_GEO] = "force_geometry",
> > [PHASE2_THREADS] = "phase2_threads",
> > + [BLOAD_LEAF_SLACK] = "debug_bload_leaf_slack",
> > + [BLOAD_NODE_SLACK] = "debug_bload_node_slack",
> > [O_MAX_OPTS] = NULL,
> > };
> >
> > @@ -260,6 +265,18 @@ process_args(int argc, char **argv)
> > _("-o phase2_threads requires a parameter\n"));
> > phase2_threads = (int)strtol(val, NULL, 0);
> > break;
> > + case BLOAD_LEAF_SLACK:
> > + if (!val)
> > + do_abort(
> > + _("-o debug_bload_leaf_slack requires a parameter\n"));
> > + bload_leaf_slack = (int)strtol(val, NULL, 0);
> > + break;
> > + case BLOAD_NODE_SLACK:
> > + if (!val)
> > + do_abort(
> > + _("-o debug_bload_node_slack requires a parameter\n"));
> > + bload_node_slack = (int)strtol(val, NULL, 0);
> > + break;
> > default:
> > unknown('o', val);
> > break;
> >
>
^ permalink raw reply [flat|nested] 30+ messages in thread
* Re: [PATCH 1/9] xfs_repair: port the online repair newbt structure
2020-05-27 22:34 ` Darrick J. Wong
@ 2020-05-28 15:08 ` Brian Foster
2020-05-29 21:01 ` Darrick J. Wong
0 siblings, 1 reply; 30+ messages in thread
From: Brian Foster @ 2020-05-28 15:08 UTC (permalink / raw)
To: Darrick J. Wong; +Cc: sandeen, linux-xfs
On Wed, May 27, 2020 at 03:34:24PM -0700, Darrick J. Wong wrote:
> On Wed, May 27, 2020 at 08:15:31AM -0400, Brian Foster wrote:
> > On Tue, May 19, 2020 at 06:50:49PM -0700, Darrick J. Wong wrote:
> > > From: Darrick J. Wong <darrick.wong@oracle.com>
> > >
> > > Port the new btree staging context and related block reservation helper
> > > code from the kernel to repair. We'll use this in subsequent patches to
> > > implement btree bulk loading.
> > >
> > > Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
> > > ---
> > > include/libxfs.h | 1
> > > libxfs/libxfs_api_defs.h | 2
> > > repair/Makefile | 4 -
> > > repair/bload.c | 303 ++++++++++++++++++++++++++++++++++++++++++++++
> > > repair/bload.h | 77 ++++++++++++
> > > repair/xfs_repair.c | 17 +++
> > > 6 files changed, 402 insertions(+), 2 deletions(-)
> > > create mode 100644 repair/bload.c
> > > create mode 100644 repair/bload.h
> > >
> > >
> > ...
> > > diff --git a/repair/bload.c b/repair/bload.c
> > > new file mode 100644
> > > index 00000000..9bc17468
> > > --- /dev/null
> > > +++ b/repair/bload.c
> > > @@ -0,0 +1,303 @@
> > > +// SPDX-License-Identifier: GPL-2.0-or-later
> > > +/*
> > > + * Copyright (C) 2020 Oracle. All Rights Reserved.
> > > + * Author: Darrick J. Wong <darrick.wong@oracle.com>
> > > + */
> > > +#include <libxfs.h>
> > > +#include "bload.h"
> > > +
> > > +#define trace_xrep_newbt_claim_block(...) ((void) 0)
> > > +#define trace_xrep_newbt_free_blocks(...) ((void) 0)
> > > +
> > > +int bload_leaf_slack = -1;
> > > +int bload_node_slack = -1;
> > > +
> > > +/* Ported routines from fs/xfs/scrub/repair.c */
> > > +
> >
> > Looks mostly straightforward, but I'll have to come back to this as I
> > get to the code that uses it later in the series. In the meantime, I see
> > some of these helpers in scrub/repair.c while not others. Are there
> > references to other routines that are intended to be copies from kernel
> > code?
>
> Hm. I might not understand the question, but in general the code should
> be fairly similar to the kernel functions. The biggest differences are
> (a) that whole libxfs error code mess, (b) the much simpler repair_ctx
> structure, and (c) the fact that repair doesn't bother with EFIs to
> automatically reap blocks.
>
> So... the ten functions you see here do the same things as their kernel
> counterparts, but they get to do it in the much simpler userspace
> environment.
>
Right.. I was able to find the first function (xrep_roll_ag_trans())
easily in the kernel because it has the same name. The next one or two
(i.e., xrep_newbt_*()) I couldn't find and then gave up. Are they
renamed? Unmerged?
> The other functions in scrub/repair.c that didn't get ported are either
> for other types of repairs or exist to support the in-kernel code and
> aren't needed here.
>
Sure, I'm just curious how to identify the source of the ones that are.
Brian
> --D
>
> > Brian
> >
> > > +/*
> > > + * Roll a transaction, keeping the AG headers locked and reinitializing
> > > + * the btree cursors.
> > > + */
> > > +int
> > > +xrep_roll_ag_trans(
> > > + struct repair_ctx *sc)
> > > +{
> > > + int error;
> > > +
> > > + /* Keep the AG header buffers locked so we can keep going. */
> > > + if (sc->agi_bp)
> > > + libxfs_trans_bhold(sc->tp, sc->agi_bp);
> > > + if (sc->agf_bp)
> > > + libxfs_trans_bhold(sc->tp, sc->agf_bp);
> > > + if (sc->agfl_bp)
> > > + libxfs_trans_bhold(sc->tp, sc->agfl_bp);
> > > +
> > > + /*
> > > + * Roll the transaction. We still own the buffer and the buffer lock
> > > + * regardless of whether or not the roll succeeds. If the roll fails,
> > > + * the buffers will be released during teardown on our way out of the
> > > + * kernel. If it succeeds, we join them to the new transaction and
> > > + * move on.
> > > + */
> > > + error = -libxfs_trans_roll(&sc->tp);
> > > + if (error)
> > > + return error;
> > > +
> > > + /* Join AG headers to the new transaction. */
> > > + if (sc->agi_bp)
> > > + libxfs_trans_bjoin(sc->tp, sc->agi_bp);
> > > + if (sc->agf_bp)
> > > + libxfs_trans_bjoin(sc->tp, sc->agf_bp);
> > > + if (sc->agfl_bp)
> > > + libxfs_trans_bjoin(sc->tp, sc->agfl_bp);
> > > +
> > > + return 0;
> > > +}
> > > +
> > > +/* Initialize accounting resources for staging a new AG btree. */
> > > +void
> > > +xrep_newbt_init_ag(
> > > + struct xrep_newbt *xnr,
> > > + struct repair_ctx *sc,
> > > + const struct xfs_owner_info *oinfo,
> > > + xfs_fsblock_t alloc_hint,
> > > + enum xfs_ag_resv_type resv)
> > > +{
> > > + memset(xnr, 0, sizeof(struct xrep_newbt));
> > > + xnr->sc = sc;
> > > + xnr->oinfo = *oinfo; /* structure copy */
> > > + xnr->alloc_hint = alloc_hint;
> > > + xnr->resv = resv;
> > > + INIT_LIST_HEAD(&xnr->resv_list);
> > > +}
> > > +
> > > +/* Initialize accounting resources for staging a new inode fork btree. */
> > > +void
> > > +xrep_newbt_init_inode(
> > > + struct xrep_newbt *xnr,
> > > + struct repair_ctx *sc,
> > > + int whichfork,
> > > + const struct xfs_owner_info *oinfo)
> > > +{
> > > + xrep_newbt_init_ag(xnr, sc, oinfo,
> > > + XFS_INO_TO_FSB(sc->mp, sc->ip->i_ino),
> > > + XFS_AG_RESV_NONE);
> > > + xnr->ifake.if_fork = kmem_zone_zalloc(xfs_ifork_zone, 0);
> > > + xnr->ifake.if_fork_size = XFS_IFORK_SIZE(sc->ip, whichfork);
> > > +}
> > > +
> > > +/*
> > > + * Initialize accounting resources for staging a new btree. Callers are
> > > + * expected to add their own reservations (and clean them up) manually.
> > > + */
> > > +void
> > > +xrep_newbt_init_bare(
> > > + struct xrep_newbt *xnr,
> > > + struct repair_ctx *sc)
> > > +{
> > > + xrep_newbt_init_ag(xnr, sc, &XFS_RMAP_OINFO_ANY_OWNER, NULLFSBLOCK,
> > > + XFS_AG_RESV_NONE);
> > > +}
> > > +
> > > +/* Designate specific blocks to be used to build our new btree. */
> > > +int
> > > +xrep_newbt_add_blocks(
> > > + struct xrep_newbt *xnr,
> > > + xfs_fsblock_t fsbno,
> > > + xfs_extlen_t len)
> > > +{
> > > + struct xrep_newbt_resv *resv;
> > > +
> > > + resv = kmem_alloc(sizeof(struct xrep_newbt_resv), KM_MAYFAIL);
> > > + if (!resv)
> > > + return ENOMEM;
> > > +
> > > + INIT_LIST_HEAD(&resv->list);
> > > + resv->fsbno = fsbno;
> > > + resv->len = len;
> > > + resv->used = 0;
> > > + list_add_tail(&resv->list, &xnr->resv_list);
> > > + return 0;
> > > +}
> > > +
> > > +/* Reserve disk space for our new btree. */
> > > +int
> > > +xrep_newbt_alloc_blocks(
> > > + struct xrep_newbt *xnr,
> > > + uint64_t nr_blocks)
> > > +{
> > > + struct repair_ctx *sc = xnr->sc;
> > > + xfs_alloctype_t type;
> > > + xfs_fsblock_t alloc_hint = xnr->alloc_hint;
> > > + int error = 0;
> > > +
> > > + type = sc->ip ? XFS_ALLOCTYPE_START_BNO : XFS_ALLOCTYPE_NEAR_BNO;
> > > +
> > > + while (nr_blocks > 0 && !error) {
> > > + struct xfs_alloc_arg args = {
> > > + .tp = sc->tp,
> > > + .mp = sc->mp,
> > > + .type = type,
> > > + .fsbno = alloc_hint,
> > > + .oinfo = xnr->oinfo,
> > > + .minlen = 1,
> > > + .maxlen = nr_blocks,
> > > + .prod = 1,
> > > + .resv = xnr->resv,
> > > + };
> > > +
> > > + error = -libxfs_alloc_vextent(&args);
> > > + if (error)
> > > + return error;
> > > + if (args.fsbno == NULLFSBLOCK)
> > > + return ENOSPC;
> > > +
> > > + /* We don't have real EFIs here so skip that. */
> > > +
> > > + error = xrep_newbt_add_blocks(xnr, args.fsbno, args.len);
> > > + if (error)
> > > + break;
> > > +
> > > + nr_blocks -= args.len;
> > > + alloc_hint = args.fsbno + args.len - 1;
> > > +
> > > + if (sc->ip)
> > > + error = -libxfs_trans_roll_inode(&sc->tp, sc->ip);
> > > + else
> > > + error = xrep_roll_ag_trans(sc);
> > > + }
> > > +
> > > + return error;
> > > +}
> > > +
> > > +/*
> > > + * Release blocks that were reserved for a btree repair. If the repair
> > > + * succeeded then we log deferred frees for unused blocks. Otherwise, we try
> > > + * to free the extents immediately to roll the filesystem back to where it was
> > > + * before we started.
> > > + */
> > > +static inline int
> > > +xrep_newbt_destroy_reservation(
> > > + struct xrep_newbt *xnr,
> > > + struct xrep_newbt_resv *resv,
> > > + bool cancel_repair)
> > > +{
> > > + struct repair_ctx *sc = xnr->sc;
> > > +
> > > + if (cancel_repair) {
> > > + int error;
> > > +
> > > + /* Free the extent then roll the transaction. */
> > > + error = -libxfs_free_extent(sc->tp, resv->fsbno, resv->len,
> > > + &xnr->oinfo, xnr->resv);
> > > + if (error)
> > > + return error;
> > > +
> > > + if (sc->ip)
> > > + return -libxfs_trans_roll_inode(&sc->tp, sc->ip);
> > > + return xrep_roll_ag_trans(sc);
> > > + }
> > > +
> > > + /* We don't have EFIs here so skip the EFD. */
> > > +
> > > + /*
> > > + * Use the deferred freeing mechanism to schedule for deletion any
> > > + * blocks we didn't use to rebuild the tree. This enables us to log
> > > + * them all in the same transaction as the root change.
> > > + */
> > > + resv->fsbno += resv->used;
> > > + resv->len -= resv->used;
> > > + resv->used = 0;
> > > +
> > > + if (resv->len == 0)
> > > + return 0;
> > > +
> > > + trace_xrep_newbt_free_blocks(sc->mp,
> > > + XFS_FSB_TO_AGNO(sc->mp, resv->fsbno),
> > > + XFS_FSB_TO_AGBNO(sc->mp, resv->fsbno),
> > > + resv->len, xnr->oinfo.oi_owner);
> > > +
> > > + __xfs_bmap_add_free(sc->tp, resv->fsbno, resv->len, &xnr->oinfo, true);
> > > +
> > > + return 0;
> > > +}
> > > +
> > > +/* Free all the accounting info and disk space we reserved for a new btree. */
> > > +void
> > > +xrep_newbt_destroy(
> > > + struct xrep_newbt *xnr,
> > > + int error)
> > > +{
> > > + struct repair_ctx *sc = xnr->sc;
> > > + struct xrep_newbt_resv *resv, *n;
> > > + int err2;
> > > +
> > > + list_for_each_entry_safe(resv, n, &xnr->resv_list, list) {
> > > + err2 = xrep_newbt_destroy_reservation(xnr, resv, error != 0);
> > > + if (err2)
> > > + goto junkit;
> > > +
> > > + list_del(&resv->list);
> > > + kmem_free(resv);
> > > + }
> > > +
> > > +junkit:
> > > + /*
> > > + * If we still have reservations attached to @newbt, cleanup must have
> > > + * failed and the filesystem is about to go down. Clean up the incore
> > > + * reservations.
> > > + */
> > > + list_for_each_entry_safe(resv, n, &xnr->resv_list, list) {
> > > + list_del(&resv->list);
> > > + kmem_free(resv);
> > > + }
> > > +
> > > + if (sc->ip) {
> > > + kmem_cache_free(xfs_ifork_zone, xnr->ifake.if_fork);
> > > + xnr->ifake.if_fork = NULL;
> > > + }
> > > +}
> > > +
> > > +/* Feed one of the reserved btree blocks to the bulk loader. */
> > > +int
> > > +xrep_newbt_claim_block(
> > > + struct xfs_btree_cur *cur,
> > > + struct xrep_newbt *xnr,
> > > + union xfs_btree_ptr *ptr)
> > > +{
> > > + struct xrep_newbt_resv *resv;
> > > + xfs_fsblock_t fsb;
> > > +
> > > + /*
> > > + * The first item in the list should always have a free block unless
> > > + * we're completely out.
> > > + */
> > > + resv = list_first_entry(&xnr->resv_list, struct xrep_newbt_resv, list);
> > > + if (resv->used == resv->len)
> > > + return ENOSPC;
> > > +
> > > + /*
> > > + * Peel off a block from the start of the reservation. We allocate
> > > + * blocks in order to place blocks on disk in increasing record or key
> > > + * order. The block reservations tend to end up on the list in
> > > + * decreasing order, which hopefully results in leaf blocks ending up
> > > + * together.
> > > + */
> > > + fsb = resv->fsbno + resv->used;
> > > + resv->used++;
> > > +
> > > + /* If we used all the blocks in this reservation, move it to the end. */
> > > + if (resv->used == resv->len)
> > > + list_move_tail(&resv->list, &xnr->resv_list);
> > > +
> > > + trace_xrep_newbt_claim_block(cur->bc_mp,
> > > + XFS_FSB_TO_AGNO(cur->bc_mp, fsb),
> > > + XFS_FSB_TO_AGBNO(cur->bc_mp, fsb),
> > > + 1, xnr->oinfo.oi_owner);
> > > +
> > > + if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
> > > + ptr->l = cpu_to_be64(fsb);
> > > + else
> > > + ptr->s = cpu_to_be32(XFS_FSB_TO_AGBNO(cur->bc_mp, fsb));
> > > + return 0;
> > > +}
> > > diff --git a/repair/bload.h b/repair/bload.h
> > > new file mode 100644
> > > index 00000000..020c4834
> > > --- /dev/null
> > > +++ b/repair/bload.h
> > > @@ -0,0 +1,77 @@
> > > +// SPDX-License-Identifier: GPL-2.0-or-later
> > > +/*
> > > + * Copyright (C) 2020 Oracle. All Rights Reserved.
> > > + * Author: Darrick J. Wong <darrick.wong@oracle.com>
> > > + */
> > > +#ifndef __XFS_REPAIR_BLOAD_H__
> > > +#define __XFS_REPAIR_BLOAD_H__
> > > +
> > > +extern int bload_leaf_slack;
> > > +extern int bload_node_slack;
> > > +
> > > +struct repair_ctx {
> > > + struct xfs_mount *mp;
> > > + struct xfs_inode *ip;
> > > + struct xfs_trans *tp;
> > > +
> > > + struct xfs_buf *agi_bp;
> > > + struct xfs_buf *agf_bp;
> > > + struct xfs_buf *agfl_bp;
> > > +};
> > > +
> > > +struct xrep_newbt_resv {
> > > + /* Link to list of extents that we've reserved. */
> > > + struct list_head list;
> > > +
> > > + /* FSB of the block we reserved. */
> > > + xfs_fsblock_t fsbno;
> > > +
> > > + /* Length of the reservation. */
> > > + xfs_extlen_t len;
> > > +
> > > + /* How much of this reservation we've used. */
> > > + xfs_extlen_t used;
> > > +};
> > > +
> > > +struct xrep_newbt {
> > > + struct repair_ctx *sc;
> > > +
> > > + /* List of extents that we've reserved. */
> > > + struct list_head resv_list;
> > > +
> > > + /* Fake root for new btree. */
> > > + union {
> > > + struct xbtree_afakeroot afake;
> > > + struct xbtree_ifakeroot ifake;
> > > + };
> > > +
> > > + /* rmap owner of these blocks */
> > > + struct xfs_owner_info oinfo;
> > > +
> > > + /* The last reservation we allocated from. */
> > > + struct xrep_newbt_resv *last_resv;
> > > +
> > > + /* Allocation hint */
> > > + xfs_fsblock_t alloc_hint;
> > > +
> > > + /* per-ag reservation type */
> > > + enum xfs_ag_resv_type resv;
> > > +};
> > > +
> > > +#define for_each_xrep_newbt_reservation(xnr, resv, n) \
> > > + list_for_each_entry_safe((resv), (n), &(xnr)->resv_list, list)
> > > +
> > > +void xrep_newbt_init_bare(struct xrep_newbt *xnr, struct repair_ctx *sc);
> > > +void xrep_newbt_init_ag(struct xrep_newbt *xnr, struct repair_ctx *sc,
> > > + const struct xfs_owner_info *oinfo, xfs_fsblock_t alloc_hint,
> > > + enum xfs_ag_resv_type resv);
> > > +void xrep_newbt_init_inode(struct xrep_newbt *xnr, struct repair_ctx *sc,
> > > + int whichfork, const struct xfs_owner_info *oinfo);
> > > +int xrep_newbt_add_blocks(struct xrep_newbt *xnr, xfs_fsblock_t fsbno,
> > > + xfs_extlen_t len);
> > > +int xrep_newbt_alloc_blocks(struct xrep_newbt *xnr, uint64_t nr_blocks);
> > > +void xrep_newbt_destroy(struct xrep_newbt *xnr, int error);
> > > +int xrep_newbt_claim_block(struct xfs_btree_cur *cur, struct xrep_newbt *xnr,
> > > + union xfs_btree_ptr *ptr);
> > > +
> > > +#endif /* __XFS_REPAIR_BLOAD_H__ */
> > > diff --git a/repair/xfs_repair.c b/repair/xfs_repair.c
> > > index 9d72fa8e..8fbd3649 100644
> > > --- a/repair/xfs_repair.c
> > > +++ b/repair/xfs_repair.c
> > > @@ -24,6 +24,7 @@
> > > #include "rmap.h"
> > > #include "libfrog/fsgeom.h"
> > > #include "libfrog/platform.h"
> > > +#include "bload.h"
> > >
> > > /*
> > > * option tables for getsubopt calls
> > > @@ -39,6 +40,8 @@ enum o_opt_nums {
> > > AG_STRIDE,
> > > FORCE_GEO,
> > > PHASE2_THREADS,
> > > + BLOAD_LEAF_SLACK,
> > > + BLOAD_NODE_SLACK,
> > > O_MAX_OPTS,
> > > };
> > >
> > > @@ -49,6 +52,8 @@ static char *o_opts[] = {
> > > [AG_STRIDE] = "ag_stride",
> > > [FORCE_GEO] = "force_geometry",
> > > [PHASE2_THREADS] = "phase2_threads",
> > > + [BLOAD_LEAF_SLACK] = "debug_bload_leaf_slack",
> > > + [BLOAD_NODE_SLACK] = "debug_bload_node_slack",
> > > [O_MAX_OPTS] = NULL,
> > > };
> > >
> > > @@ -260,6 +265,18 @@ process_args(int argc, char **argv)
> > > _("-o phase2_threads requires a parameter\n"));
> > > phase2_threads = (int)strtol(val, NULL, 0);
> > > break;
> > > + case BLOAD_LEAF_SLACK:
> > > + if (!val)
> > > + do_abort(
> > > + _("-o debug_bload_leaf_slack requires a parameter\n"));
> > > + bload_leaf_slack = (int)strtol(val, NULL, 0);
> > > + break;
> > > + case BLOAD_NODE_SLACK:
> > > + if (!val)
> > > + do_abort(
> > > + _("-o debug_bload_node_slack requires a parameter\n"));
> > > + bload_node_slack = (int)strtol(val, NULL, 0);
> > > + break;
> > > default:
> > > unknown('o', val);
> > > break;
> > >
> >
>
^ permalink raw reply [flat|nested] 30+ messages in thread
* Re: [PATCH 3/9] xfs_repair: create a new class of btree rebuild cursors
2020-05-27 22:07 ` Darrick J. Wong
@ 2020-05-28 15:09 ` Brian Foster
2020-05-29 21:08 ` Darrick J. Wong
0 siblings, 1 reply; 30+ messages in thread
From: Brian Foster @ 2020-05-28 15:09 UTC (permalink / raw)
To: Darrick J. Wong; +Cc: sandeen, linux-xfs
On Wed, May 27, 2020 at 03:07:33PM -0700, Darrick J. Wong wrote:
> On Wed, May 27, 2020 at 08:18:04AM -0400, Brian Foster wrote:
> > On Tue, May 19, 2020 at 06:51:02PM -0700, Darrick J. Wong wrote:
> > > From: Darrick J. Wong <darrick.wong@oracle.com>
> > >
> > > Create some new support structures and functions to assist phase5 in
> > > using the btree bulk loader to reconstruct metadata btrees. This is the
> > > first step in removing the open-coded rebuilding code.
> > >
> > > Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
> > > ---
> > > repair/phase5.c | 239 ++++++++++++++++++++++++++++++++++++++++++++++++++-----
> > > 1 file changed, 218 insertions(+), 21 deletions(-)
> > >
> > >
> > > diff --git a/repair/phase5.c b/repair/phase5.c
> > > index 84c05a13..8f5e5f59 100644
> > > --- a/repair/phase5.c
> > > +++ b/repair/phase5.c
> > > @@ -18,6 +18,7 @@
> > > #include "progress.h"
> > > #include "slab.h"
> > > #include "rmap.h"
> > > +#include "bload.h"
> > >
> > > /*
> > > * we maintain the current slice (path from root to leaf)
> > ...
> > > @@ -306,6 +324,156 @@ _("error - not enough free space in filesystem\n"));
> > > #endif
> > > }
> > >
> > ...
> > > +static void
> > > +consume_freespace(
> > > + xfs_agnumber_t agno,
> > > + struct extent_tree_node *ext_ptr,
> > > + uint32_t len)
> > > +{
> > > + struct extent_tree_node *bno_ext_ptr;
> > > + xfs_agblock_t new_start = ext_ptr->ex_startblock + len;
> > > + xfs_extlen_t new_len = ext_ptr->ex_blockcount - len;
> > > +
> > > + /* Delete the used-up extent from both extent trees. */
> > > +#ifdef XR_BLD_FREE_TRACE
> > > + fprintf(stderr, "releasing extent: %u [%u %u]\n", agno,
> > > + ext_ptr->ex_startblock, ext_ptr->ex_blockcount);
> > > +#endif
> > > + bno_ext_ptr = find_bno_extent(agno, ext_ptr->ex_startblock);
> > > + ASSERT(bno_ext_ptr != NULL);
> > > + get_bno_extent(agno, bno_ext_ptr);
> > > + release_extent_tree_node(bno_ext_ptr);
> > > +
> > > + ext_ptr = get_bcnt_extent(agno, ext_ptr->ex_startblock,
> > > + ext_ptr->ex_blockcount);
> > > + release_extent_tree_node(ext_ptr);
> > > +
> >
> > Not having looked too deeply at the in-core extent tracking structures,
> > is there any particular reason we unconditionally remove and reinsert
> > new records each time around? Is it because we're basically changing the
> > extent index in the tree? If so, comment please (an update to the
> > comment below is probably fine). :)
>
> Yes. We're changing the free space tree records, and the incore bno and
> cnt trees maintain the records in sorted order. Therefore, if we want
> to change a record we have to delete the record from the tree and
> reinsert it.
>
> > > + /*
> > > + * If we only used part of this last extent, then we must reinsert the
> > > + * extent in the extent trees.
>
> /*
> * If we only used part of this last extent, then we must reinsert the
> * extent to maintain proper sorting order.
> */
>
> How about that?
>
Works for me, thanks.
> > > + */
> > > + if (new_len > 0) {
> > > + add_bno_extent(agno, new_start, new_len);
> > > + add_bcnt_extent(agno, new_start, new_len);
> > > + }
> > > +}
> > > +
> > > +/* Reserve blocks for the new btree. */
> > > +static void
> > > +setup_rebuild(
> > > + struct xfs_mount *mp,
> > > + xfs_agnumber_t agno,
> > > + struct bt_rebuild *btr,
> > > + uint32_t nr_blocks)
> > > +{
> > > + struct extent_tree_node *ext_ptr;
> > > + uint32_t blocks_allocated = 0;
> > > + uint32_t len;
> > > + int error;
> > > +
> > > + while (blocks_allocated < nr_blocks) {
> > > + /*
> > > + * Grab the smallest extent and use it up, then get the
> > > + * next smallest. This mimics the init_*_cursor code.
> > > + */
> > > + ext_ptr = findfirst_bcnt_extent(agno);
> >
> > Extra whitespace ^
> >
> > > + if (!ext_ptr)
> > > + do_error(
> > > +_("error - not enough free space in filesystem\n"));
> > > +
> > > + /* Use up the extent we've got. */
> > > + len = min(ext_ptr->ex_blockcount, nr_blocks - blocks_allocated);
> > > + error = xrep_newbt_add_blocks(&btr->newbt,
> > > + XFS_AGB_TO_FSB(mp, agno,
> > > + ext_ptr->ex_startblock),
> > > + len);
> >
> > Alignment.
>
> Will fix both of these.
>
> > > + if (error)
> > > + do_error(_("could not set up btree reservation: %s\n"),
> > > + strerror(-error));
> > > +
> > > + error = rmap_add_ag_rec(mp, agno, ext_ptr->ex_startblock, len,
> > > + btr->newbt.oinfo.oi_owner);
> > > + if (error)
> > > + do_error(_("could not set up btree rmaps: %s\n"),
> > > + strerror(-error));
> > > +
> > > + consume_freespace(agno, ext_ptr, len);
> > > + blocks_allocated += len;
> > > + }
> > > +#ifdef XR_BLD_FREE_TRACE
> > > + fprintf(stderr, "blocks_allocated = %d\n",
> > > + blocks_allocated);
> > > +#endif
> > > +}
> > > +
> > > +/* Feed one of the new btree blocks to the bulk loader. */
> > > +static int
> > > +rebuild_claim_block(
> > > + struct xfs_btree_cur *cur,
> > > + union xfs_btree_ptr *ptr,
> > > + void *priv)
> > > +{
> > > + struct bt_rebuild *btr = priv;
> > > +
> > > + return xrep_newbt_claim_block(cur, &btr->newbt, ptr);
> > > +}
> > > +
> >
> > Seems like an unnecessary helper, unless this grows more code in later
> > patches..?
>
> It doesn't grow any more code, but keep in mind that get_record,
> claim_block, and iroot_size are all callbacks of xfs_btree_bload(). The
> priv parameter passed to that function are passed unchanged to the three
> callbacks. The bulk load code doesn't know anything about where the
> blocks or the records come from, so this is how both repairs will pass
> that information to the callbacks.
>
Ok.
> > > static void
> > > write_cursor(bt_status_t *curs)
> > > {
> > ...
> > > @@ -2287,28 +2483,29 @@ keep_fsinos(xfs_mount_t *mp)
> > >
> > > static void
> > > phase5_func(
> > > - xfs_mount_t *mp,
> > > - xfs_agnumber_t agno,
> > > - struct xfs_slab *lost_fsb)
> > > + struct xfs_mount *mp,
> > > + xfs_agnumber_t agno,
> > > + struct xfs_slab *lost_fsb)
> > > {
> > > - uint64_t num_inos;
> > > - uint64_t num_free_inos;
> > > - uint64_t finobt_num_inos;
> > > - uint64_t finobt_num_free_inos;
> > > - bt_status_t bno_btree_curs;
> > > - bt_status_t bcnt_btree_curs;
> > > - bt_status_t ino_btree_curs;
> > > - bt_status_t fino_btree_curs;
> > > - bt_status_t rmap_btree_curs;
> > > - bt_status_t refcnt_btree_curs;
> > > - int extra_blocks = 0;
> > > - uint num_freeblocks;
> > > - xfs_extlen_t freeblks1;
> > > + struct repair_ctx sc = { .mp = mp, };
> >
> > I don't see any reason to add sc here when it's still unused. It's not
> > as if a single variable is saving complexity somewhere else. I guess
> > I'll defer to Eric on the approach wrt to the other unused warnings.
>
> <shrug> I'll ask. It seems dumb to have a prep patch that adds a bunch
> of symbols that won't get used until the next patch, but OTOH combining
> the two will make for a ~40K patch.
>
I've no strong preference either way in general (the single variable
thing aside) as long as each patch compiles and functions correctly and
warnings are addressed by the end of the series. I do think that if we
keep separate patches, it should probably be documented in the commit
log that unused infrastructure is introduced (i.e. warnings expected)
and users are introduced in a following patch. It's usually easier to
squash patches than separate, so the maintainer can always squash them
post review if he wanted to eliminate the warnings from the commit
history.
Brian
> > Also, what's the purpose of the rmap change below? I'm wondering if that
> > (along with all of the indentation cleanup) should be its own patch with
> > appropriate explanation.
>
> Errk, that one definitely should be separate.
>
> > Brian
> >
> > > + struct agi_stat agi_stat = {0,};
> > > + uint64_t num_inos;
> > > + uint64_t num_free_inos;
> > > + uint64_t finobt_num_inos;
> > > + uint64_t finobt_num_free_inos;
> > > + bt_status_t bno_btree_curs;
> > > + bt_status_t bcnt_btree_curs;
> > > + bt_status_t ino_btree_curs;
> > > + bt_status_t fino_btree_curs;
> > > + bt_status_t rmap_btree_curs;
> > > + bt_status_t refcnt_btree_curs;
> > > + int extra_blocks = 0;
> > > + uint num_freeblocks;
> > > + xfs_extlen_t freeblks1;
> > > #ifdef DEBUG
> > > - xfs_extlen_t freeblks2;
> > > + xfs_extlen_t freeblks2;
> > > #endif
> > > - xfs_agblock_t num_extents;
> > > - struct agi_stat agi_stat = {0,};
> > > + xfs_agblock_t num_extents;
> > >
> > > if (verbose)
> > > do_log(_(" - agno = %d\n"), agno);
> > > @@ -2516,8 +2713,8 @@ inject_lost_blocks(
> > > if (error)
> > > goto out_cancel;
> > >
> > > - error = -libxfs_free_extent(tp, *fsb, 1, &XFS_RMAP_OINFO_AG,
> > > - XFS_AG_RESV_NONE);
> > > + error = -libxfs_free_extent(tp, *fsb, 1,
> > > + &XFS_RMAP_OINFO_ANY_OWNER, XFS_AG_RESV_NONE);
> > > if (error)
> > > goto out_cancel;
> > >
> > >
> >
>
^ permalink raw reply [flat|nested] 30+ messages in thread
* Re: [PATCH 4/9] xfs_repair: rebuild free space btrees with bulk loader
2020-05-20 1:51 ` [PATCH 4/9] xfs_repair: rebuild free space btrees with bulk loader Darrick J. Wong
@ 2020-05-28 15:10 ` Brian Foster
2020-05-29 21:39 ` Darrick J. Wong
0 siblings, 1 reply; 30+ messages in thread
From: Brian Foster @ 2020-05-28 15:10 UTC (permalink / raw)
To: Darrick J. Wong; +Cc: sandeen, linux-xfs
On Tue, May 19, 2020 at 06:51:08PM -0700, Darrick J. Wong wrote:
> From: Darrick J. Wong <darrick.wong@oracle.com>
>
> Use the btree bulk loading functions to rebuild the free space btrees
> and drop the open-coded implementation.
>
> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
> ---
> libxfs/libxfs_api_defs.h | 3
> repair/phase5.c | 870 +++++++++++++---------------------------------
> 2 files changed, 254 insertions(+), 619 deletions(-)
>
>
...
> diff --git a/repair/phase5.c b/repair/phase5.c
> index 8f5e5f59..e69b042c 100644
> --- a/repair/phase5.c
> +++ b/repair/phase5.c
...
> @@ -837,268 +567,202 @@ btnum_to_ops(
...
> +static void
> +init_freespace_cursors(
> + struct repair_ctx *sc,
> + xfs_agnumber_t agno,
> + unsigned int free_space,
> + unsigned int *nr_extents,
> + int *extra_blocks,
> + struct bt_rebuild *btr_bno,
> + struct bt_rebuild *btr_cnt)
> {
> - xfs_agnumber_t i;
> - xfs_agblock_t j;
> - struct xfs_btree_block *bt_hdr;
> - xfs_alloc_rec_t *bt_rec;
> - int level;
> - xfs_agblock_t agbno;
> - extent_tree_node_t *ext_ptr;
> - bt_stat_level_t *lptr;
> - xfs_extlen_t freeblks;
> - const struct xfs_buf_ops *ops = btnum_to_ops(btnum);
> + unsigned int bno_blocks;
> + unsigned int cnt_blocks;
> int error;
>
> - ASSERT(btnum == XFS_BTNUM_BNO || btnum == XFS_BTNUM_CNT);
> -
> -#ifdef XR_BLD_FREE_TRACE
> - fprintf(stderr, "in build_freespace_tree, agno = %d\n", agno);
> -#endif
> - level = btree_curs->num_levels;
> - freeblks = 0;
> + init_rebuild(sc, &XFS_RMAP_OINFO_AG, free_space, btr_bno);
> + init_rebuild(sc, &XFS_RMAP_OINFO_AG, free_space, btr_cnt);
>
> - ASSERT(level > 0);
> + btr_bno->cur = libxfs_allocbt_stage_cursor(sc->mp,
> + &btr_bno->newbt.afake, agno, XFS_BTNUM_BNO);
> + btr_cnt->cur = libxfs_allocbt_stage_cursor(sc->mp,
> + &btr_cnt->newbt.afake, agno, XFS_BTNUM_CNT);
>
> /*
> - * initialize the first block on each btree level
> + * Now we need to allocate blocks for the free space btrees using the
> + * free space records we're about to put in them. Every record we use
> + * can change the shape of the free space trees, so we recompute the
> + * btree shape until we stop needing /more/ blocks. If we have any
> + * left over we'll stash them in the AGFL when we're done.
> */
> - for (i = 0; i < level; i++) {
> - lptr = &btree_curs->level[i];
> + do {
> + unsigned int num_freeblocks;
> +
> + bno_blocks = btr_bno->bload.nr_blocks;
> + cnt_blocks = btr_cnt->bload.nr_blocks;
>
> - agbno = get_next_blockaddr(agno, i, btree_curs);
> - error = -libxfs_buf_get(mp->m_dev,
> - XFS_AGB_TO_DADDR(mp, agno, agbno),
> - XFS_FSB_TO_BB(mp, 1), &lptr->buf_p);
> + /* Compute how many bnobt blocks we'll need. */
> + error = -libxfs_btree_bload_compute_geometry(btr_bno->cur,
> + &btr_bno->bload, *nr_extents);
> if (error)
> do_error(
...
> +_("Unable to compute free space by block btree geometry, error %d.\n"), -error);
> +
> + /* Compute how many cntbt blocks we'll need. */
> + error = -libxfs_btree_bload_compute_geometry(btr_bno->cur,
btr_cnt? ^
> + &btr_cnt->bload, *nr_extents);
> + if (error)
> + do_error(
> +_("Unable to compute free space by length btree geometry, error %d.\n"), -error);
> +
> + /* We don't need any more blocks, so we're done. */
> + if (bno_blocks >= btr_bno->bload.nr_blocks &&
> + cnt_blocks >= btr_cnt->bload.nr_blocks)
> + break;
> +
> + /* Allocate however many more blocks we need this time. */
> + if (bno_blocks < btr_bno->bload.nr_blocks)
> + setup_rebuild(sc->mp, agno, btr_bno,
> + btr_bno->bload.nr_blocks - bno_blocks);
> + if (cnt_blocks < btr_cnt->bload.nr_blocks)
> + setup_rebuild(sc->mp, agno, btr_cnt,
> + btr_cnt->bload.nr_blocks - cnt_blocks);
Hmm, I find the setup_rebuild() name a little confusing when seeing an
init_rebuild() used in the same function. The difference is not apparent
at a glance. Could we rename setup_rebuild() to something more specific?
reserve_blocks() perhaps?
> +
> + /* Ok, now how many free space records do we have? */
> + *nr_extents = count_bno_extents_blocks(agno, &num_freeblocks);
> + } while (1);
> +
> + *extra_blocks = (bno_blocks - btr_bno->bload.nr_blocks) +
> + (cnt_blocks - btr_cnt->bload.nr_blocks);
> +}
> +
> +static void
> +get_freesp_data(
> + struct xfs_btree_cur *cur,
> + struct extent_tree_node *bno_rec,
> + xfs_agblock_t *freeblks)
> +{
> + struct xfs_alloc_rec_incore *arec = &cur->bc_rec.a;
> +
> + arec->ar_startblock = bno_rec->ex_startblock;
> + arec->ar_blockcount = bno_rec->ex_blockcount;
> + if (freeblks)
> + *freeblks += bno_rec->ex_blockcount;
> +}
> +
> +/* Grab one bnobt record. */
> +static int
> +get_bnobt_record(
> + struct xfs_btree_cur *cur,
> + void *priv)
> +{
> + struct bt_rebuild *btr = priv;
> +
> + get_freesp_data(cur, btr->bno_rec, btr->freeblks);
> + btr->bno_rec = findnext_bno_extent(btr->bno_rec);
We should probably check for NULL here even if we don't expect it to
happen.
Also, this logic where we load the current pointer and get the next had
my eyes crossed for a bit. Can we just check the pointer state here to
determine whether to find the first record or the next?
Finally, I notice we duplicate the build_[bno|cnt]bt() functions for
what amounts to a different get_record() callback specific to the tree
type. If we also generalize get_record() to use the tree type (it
already has the cursor), it seems that much of the duplication can be
reduced and the logic simplified.
> + return 0;
> +}
> +
> +/* Rebuild a free space by block number btree. */
> +static void
> +build_bnobt(
> + struct repair_ctx *sc,
> + xfs_agnumber_t agno,
> + struct bt_rebuild *btr_bno,
> + xfs_agblock_t *freeblks)
> +{
> + int error;
> +
> + *freeblks = 0;
> + btr_bno->bload.get_record = get_bnobt_record;
> + btr_bno->bload.claim_block = rebuild_claim_block;
Note that one of these callbacks is introduced in this patch and the
other in the previous. I've no issue with splitting the patches as noted
in the previous patch, but these should probably be consistent one way
or the other.
> + btr_bno->bno_rec = findfirst_bno_extent(agno);
> + btr_bno->freeblks = freeblks;
> +
> + error = -libxfs_trans_alloc_empty(sc->mp, &sc->tp);
> + if (error)
> + do_error(
> +_("Insufficient memory to construct bnobt rebuild transaction.\n"));
> +
> + /* Add all observed bnobt records. */
> + error = -libxfs_btree_bload(btr_bno->cur, &btr_bno->bload, btr_bno);
> + if (error)
> + do_error(
> +_("Error %d while creating bnobt btree for AG %u.\n"), error, agno);
> +
> + /* Since we're not writing the AGF yet, no need to commit the cursor */
> + libxfs_btree_del_cursor(btr_bno->cur, 0);
> + error = -libxfs_trans_commit(sc->tp);
> + if (error)
> + do_error(
> +_("Error %d while writing bnobt btree for AG %u.\n"), error, agno);
> + sc->tp = NULL;
> +}
> +
...
> @@ -2354,48 +2041,14 @@ build_agf_agfl(
> freelist[i] = cpu_to_be32(NULLAGBLOCK);
> }
>
> - /*
> - * do we have left-over blocks in the btree cursors that should
> - * be used to fill the AGFL?
> - */
> - if (bno_bt->num_free_blocks > 0 || bcnt_bt->num_free_blocks > 0) {
> - /*
> - * yes, now grab as many blocks as we can
> - */
> - i = 0;
> - while (bno_bt->num_free_blocks > 0 && i < libxfs_agfl_size(mp))
> - {
> - freelist[i] = cpu_to_be32(
> - get_next_blockaddr(agno, 0, bno_bt));
> - i++;
> - }
> -
> - while (bcnt_bt->num_free_blocks > 0 && i < libxfs_agfl_size(mp))
> - {
> - freelist[i] = cpu_to_be32(
> - get_next_blockaddr(agno, 0, bcnt_bt));
> - i++;
> - }
> - /*
> - * now throw the rest of the blocks away and complain
> - */
> - while (bno_bt->num_free_blocks > 0) {
> - fsb = XFS_AGB_TO_FSB(mp, agno,
> - get_next_blockaddr(agno, 0, bno_bt));
> - error = slab_add(lost_fsb, &fsb);
> - if (error)
> - do_error(
> -_("Insufficient memory saving lost blocks.\n"));
> - }
> - while (bcnt_bt->num_free_blocks > 0) {
> - fsb = XFS_AGB_TO_FSB(mp, agno,
> - get_next_blockaddr(agno, 0, bcnt_bt));
> - error = slab_add(lost_fsb, &fsb);
> - if (error)
> - do_error(
> -_("Insufficient memory saving lost blocks.\n"));
> - }
> + /* Fill the AGFL with leftover blocks or save them for later. */
> + i = 0;
How about agfl_idx or something since this variable is passed down into
a function?
> + freelist = xfs_buf_to_agfl_bno(agfl_buf);
> + fill_agfl(btr_bno, freelist, &i);
> + fill_agfl(btr_cnt, freelist, &i);
>
> + /* Set the AGF counters for the AGFL. */
> + if (i > 0) {
> agf->agf_flfirst = 0;
> agf->agf_fllast = cpu_to_be32(i - 1);
> agf->agf_flcount = cpu_to_be32(i);
...
> @@ -2650,9 +2283,9 @@ _("unable to rebuild AG %u. Not enough free space in on-disk AG.\n"),
> /*
> * set up agf and agfl
> */
> - build_agf_agfl(mp, agno, &bno_btree_curs,
> - &bcnt_btree_curs, freeblks1, extra_blocks,
> - &rmap_btree_curs, &refcnt_btree_curs, lost_fsb);
> + build_agf_agfl(mp, agno, &btr_bno, &btr_cnt, freeblks1, extra_blocks,
> + &rmap_btree_curs, &refcnt_btree_curs, lost_fsb);
> +
I was trying to figure out what extra_blocks was used for and noticed
that it's not even used by this function. Perhaps a preceding cleanup
patch to drop it?
> /*
> * build inode allocation tree.
> */
> @@ -2674,15 +2307,14 @@ _("unable to rebuild AG %u. Not enough free space in on-disk AG.\n"),
> /*
> * tear down cursors
> */
> - finish_cursor(&bno_btree_curs);
> - finish_cursor(&ino_btree_curs);
Looks like this one shouldn't be going away here.
Brian
> + finish_rebuild(mp, &btr_bno, lost_fsb);
> + finish_rebuild(mp, &btr_cnt, lost_fsb);
> if (xfs_sb_version_hasrmapbt(&mp->m_sb))
> finish_cursor(&rmap_btree_curs);
> if (xfs_sb_version_hasreflink(&mp->m_sb))
> finish_cursor(&refcnt_btree_curs);
> if (xfs_sb_version_hasfinobt(&mp->m_sb))
> finish_cursor(&fino_btree_curs);
> - finish_cursor(&bcnt_btree_curs);
>
> /*
> * release the incore per-AG bno/bcnt trees so the extent nodes
>
^ permalink raw reply [flat|nested] 30+ messages in thread
* Re: [PATCH 5/9] xfs_repair: rebuild inode btrees with bulk loader
2020-05-20 1:51 ` [PATCH 5/9] xfs_repair: rebuild inode " Darrick J. Wong
@ 2020-05-28 15:11 ` Brian Foster
2020-05-29 22:18 ` Darrick J. Wong
0 siblings, 1 reply; 30+ messages in thread
From: Brian Foster @ 2020-05-28 15:11 UTC (permalink / raw)
To: Darrick J. Wong; +Cc: sandeen, linux-xfs
On Tue, May 19, 2020 at 06:51:15PM -0700, Darrick J. Wong wrote:
> From: Darrick J. Wong <darrick.wong@oracle.com>
>
> Use the btree bulk loading functions to rebuild the inode btrees
> and drop the open-coded implementation.
>
> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
> ---
> libxfs/libxfs_api_defs.h | 1
> repair/phase5.c | 642 +++++++++++++++++-----------------------------
> 2 files changed, 240 insertions(+), 403 deletions(-)
>
>
...
> diff --git a/repair/phase5.c b/repair/phase5.c
> index e69b042c..38f30753 100644
> --- a/repair/phase5.c
> +++ b/repair/phase5.c
...
> @@ -372,6 +376,11 @@ estimate_ag_bload_slack(
> bload->node_slack = 0;
> }
>
> +static inline void skip_rebuild(struct bt_rebuild *btr)
> +{
> + memset(btr, 0, sizeof(struct bt_rebuild));
> +}
> +
Is there any functional purpose to this? It looks like the memset could
be open-coded, but also seems like it could be elided if we just check
hasfinobt() before using the pointer..?
> /* Initialize a btree rebuild context. */
> static void
> init_rebuild(
> @@ -765,48 +774,38 @@ _("Error %d while writing cntbt btree for AG %u.\n"), error, agno);
...
> +
> +/* Copy one incore inode record into the inobt cursor. */
> +static void
> +get_inode_data(
> + struct xfs_btree_cur *cur,
> + struct ino_tree_node *ino_rec,
> + struct agi_stat *agi_stat)
> +{
> + struct xfs_inobt_rec_incore *irec = &cur->bc_rec.i;
> + int inocnt = 0;
> + int finocnt = 0;
> + int k;
> +
> + irec->ir_startino = ino_rec->ino_startnum;
> + irec->ir_free = ino_rec->ir_free;
> +
> + for (k = 0; k < sizeof(xfs_inofree_t) * NBBY; k++) {
> + ASSERT(is_inode_confirmed(ino_rec, k));
> +
> + if (is_inode_sparse(ino_rec, k))
> continue;
> -
> - nfinos += rec_nfinos;
> - ninos += rec_ninos;
> - num_recs++;
> + if (is_inode_free(ino_rec, k))
> + finocnt++;
> + inocnt++;
> }
>
> - if (num_recs == 0) {
> - /*
> - * easy corner-case -- no inode records
> - */
> - lptr->num_blocks = 1;
> - lptr->modulo = 0;
> - lptr->num_recs_pb = 0;
> - lptr->num_recs_tot = 0;
> -
> - btree_curs->num_levels = 1;
> - btree_curs->num_tot_blocks = btree_curs->num_free_blocks = 1;
> -
> - setup_cursor(mp, agno, btree_curs);
> + irec->ir_count = inocnt;
> + irec->ir_freecount = finocnt;
>
> - return;
> - }
> + if (xfs_sb_version_hassparseinodes(&cur->bc_mp->m_sb)) {
> + uint64_t sparse;
> + int spmask;
> + uint16_t holemask;
>
> - blocks_allocated = lptr->num_blocks = howmany(num_recs,
> - XR_INOBT_BLOCK_MAXRECS(mp, 0));
> -
> - lptr->modulo = num_recs % lptr->num_blocks;
> - lptr->num_recs_pb = num_recs / lptr->num_blocks;
> - lptr->num_recs_tot = num_recs;
> - level = 1;
> -
> - if (lptr->num_blocks > 1) {
> - for (; btree_curs->level[level-1].num_blocks > 1
> - && level < XFS_BTREE_MAXLEVELS;
> - level++) {
> - lptr = &btree_curs->level[level];
> - p_lptr = &btree_curs->level[level - 1];
> - lptr->num_blocks = howmany(p_lptr->num_blocks,
> - XR_INOBT_BLOCK_MAXRECS(mp, level));
> - lptr->modulo = p_lptr->num_blocks % lptr->num_blocks;
> - lptr->num_recs_pb = p_lptr->num_blocks
> - / lptr->num_blocks;
> - lptr->num_recs_tot = p_lptr->num_blocks;
> -
> - blocks_allocated += lptr->num_blocks;
> + /*
> + * Convert the 64-bit in-core sparse inode state to the
> + * 16-bit on-disk holemask.
> + */
> + holemask = 0;
> + spmask = (1 << XFS_INODES_PER_HOLEMASK_BIT) - 1;
> + sparse = ino_rec->ir_sparse;
> + for (k = 0; k < XFS_INOBT_HOLEMASK_BITS; k++) {
> + if (sparse & spmask) {
> + ASSERT((sparse & spmask) == spmask);
> + holemask |= (1 << k);
> + } else
> + ASSERT((sparse & spmask) == 0);
> + sparse >>= XFS_INODES_PER_HOLEMASK_BIT;
> }
> +
> + irec->ir_holemask = holemask;
> + } else {
> + irec->ir_holemask = 0;
> }
> - ASSERT(lptr->num_blocks == 1);
> - btree_curs->num_levels = level;
>
> - btree_curs->num_tot_blocks = btree_curs->num_free_blocks
> - = blocks_allocated;
> + if (!agi_stat)
> + return;
>
> - setup_cursor(mp, agno, btree_curs);
> + if (agi_stat->first_agino != NULLAGINO)
> + agi_stat->first_agino = ino_rec->ino_startnum;
This is initialized to NULLAGINO. When do we ever update it?
> + agi_stat->freecount += finocnt;
> + agi_stat->count += inocnt;
> +}
>
> - *num_inos = ninos;
> - *num_free_inos = nfinos;
> +/* Grab one inobt record. */
> +static int
> +get_inobt_record(
> + struct xfs_btree_cur *cur,
> + void *priv)
> +{
> + struct bt_rebuild *rebuild = priv;
>
> - return;
> + get_inode_data(cur, rebuild->ino_rec, rebuild->agi_stat);
> + rebuild->ino_rec = next_ino_rec(rebuild->ino_rec);
> + return 0;
> }
>
> +/* Rebuild a inobt btree. */
> static void
> -prop_ino_cursor(xfs_mount_t *mp, xfs_agnumber_t agno, bt_status_t *btree_curs,
> - xfs_btnum_t btnum, xfs_agino_t startino, int level)
> +build_inobt(
> + struct repair_ctx *sc,
> + xfs_agnumber_t agno,
> + struct bt_rebuild *btr_ino,
> + struct agi_stat *agi_stat)
> {
> - struct xfs_btree_block *bt_hdr;
> - xfs_inobt_key_t *bt_key;
> - xfs_inobt_ptr_t *bt_ptr;
> - xfs_agblock_t agbno;
> - bt_stat_level_t *lptr;
> - const struct xfs_buf_ops *ops = btnum_to_ops(btnum);
> int error;
>
> - level++;
> -
> - if (level >= btree_curs->num_levels)
> - return;
> -
> - lptr = &btree_curs->level[level];
> - bt_hdr = XFS_BUF_TO_BLOCK(lptr->buf_p);
> -
> - if (be16_to_cpu(bt_hdr->bb_numrecs) == 0) {
> - /*
> - * this only happens once to initialize the
> - * first path up the left side of the tree
> - * where the agbno's are already set up
> - */
> - prop_ino_cursor(mp, agno, btree_curs, btnum, startino, level);
> - }
> -
> - if (be16_to_cpu(bt_hdr->bb_numrecs) ==
> - lptr->num_recs_pb + (lptr->modulo > 0)) {
> - /*
> - * write out current prev block, grab us a new block,
> - * and set the rightsib pointer of current block
> - */
> -#ifdef XR_BLD_INO_TRACE
> - fprintf(stderr, " ino prop agbno %d ", lptr->prev_agbno);
> -#endif
> - if (lptr->prev_agbno != NULLAGBLOCK) {
> - ASSERT(lptr->prev_buf_p != NULL);
> - libxfs_buf_mark_dirty(lptr->prev_buf_p);
> - libxfs_buf_relse(lptr->prev_buf_p);
> - }
> - lptr->prev_agbno = lptr->agbno;;
> - lptr->prev_buf_p = lptr->buf_p;
> - agbno = get_next_blockaddr(agno, level, btree_curs);
> -
> - bt_hdr->bb_u.s.bb_rightsib = cpu_to_be32(agbno);
> -
> - error = -libxfs_buf_get(mp->m_dev,
> - XFS_AGB_TO_DADDR(mp, agno, agbno),
> - XFS_FSB_TO_BB(mp, 1), &lptr->buf_p);
> - if (error)
> - do_error(_("Cannot grab inode btree buffer, err=%d"),
> - error);
> - lptr->agbno = agbno;
> + btr_ino->bload.get_record = get_inobt_record;
> + btr_ino->bload.claim_block = rebuild_claim_block;
> + agi_stat->count = agi_stat->freecount = 0;
These are already initialized to zero by the caller. I suppose we might
as well also move the ->first_agino init to where this is allocated.
Otherwise I mostly just have the same general feedback as for the
previous patch (wrt to the get_record() logic and build_[f]inobt()
duplication.
Brian
> + agi_stat->first_agino = NULLAGINO;
> + btr_ino->agi_stat = agi_stat;
> + btr_ino->ino_rec = findfirst_inode_rec(agno);
> +
> + error = -libxfs_trans_alloc_empty(sc->mp, &sc->tp);
> + if (error)
> + do_error(
> +_("Insufficient memory to construct inobt rebuild transaction.\n"));
> +
> + /* Add all observed inobt records. */
> + error = -libxfs_btree_bload(btr_ino->cur, &btr_ino->bload, btr_ino);
> + if (error)
> + do_error(
> +_("Error %d while creating inobt btree for AG %u.\n"), error, agno);
> +
> + /* Since we're not writing the AGI yet, no need to commit the cursor */
> + libxfs_btree_del_cursor(btr_ino->cur, 0);
> + error = -libxfs_trans_commit(sc->tp);
> + if (error)
> + do_error(
> +_("Error %d while writing inobt btree for AG %u.\n"), error, agno);
> + sc->tp = NULL;
> +}
>
> - if (lptr->modulo)
> - lptr->modulo--;
> +/* Grab one finobt record. */
> +static int
> +get_finobt_record(
> + struct xfs_btree_cur *cur,
> + void *priv)
> +{
> + struct bt_rebuild *rebuild = priv;
>
> - /*
> - * initialize block header
> - */
> - lptr->buf_p->b_ops = ops;
> - bt_hdr = XFS_BUF_TO_BLOCK(lptr->buf_p);
> - memset(bt_hdr, 0, mp->m_sb.sb_blocksize);
> - libxfs_btree_init_block(mp, lptr->buf_p, btnum,
> - level, 0, agno);
> + get_inode_data(cur, rebuild->ino_rec, NULL);
> + rebuild->ino_rec = next_free_ino_rec(rebuild->ino_rec);
> + return 0;
> +}
>
> - bt_hdr->bb_u.s.bb_leftsib = cpu_to_be32(lptr->prev_agbno);
> +/* Rebuild a finobt btree. */
> +static void
> +build_finobt(
> + struct repair_ctx *sc,
> + xfs_agnumber_t agno,
> + struct bt_rebuild *btr_fino)
> +{
> + int error;
>
> - /*
> - * propagate extent record for first extent in new block up
> - */
> - prop_ino_cursor(mp, agno, btree_curs, btnum, startino, level);
> - }
> - /*
> - * add inode info to current block
> - */
> - be16_add_cpu(&bt_hdr->bb_numrecs, 1);
> -
> - bt_key = XFS_INOBT_KEY_ADDR(mp, bt_hdr,
> - be16_to_cpu(bt_hdr->bb_numrecs));
> - bt_ptr = XFS_INOBT_PTR_ADDR(mp, bt_hdr,
> - be16_to_cpu(bt_hdr->bb_numrecs),
> - M_IGEO(mp)->inobt_mxr[1]);
> -
> - bt_key->ir_startino = cpu_to_be32(startino);
> - *bt_ptr = cpu_to_be32(btree_curs->level[level-1].agbno);
> + btr_fino->bload.get_record = get_finobt_record;
> + btr_fino->bload.claim_block = rebuild_claim_block;
> + btr_fino->ino_rec = findfirst_free_inode_rec(agno);
> +
> + error = -libxfs_trans_alloc_empty(sc->mp, &sc->tp);
> + if (error)
> + do_error(
> +_("Insufficient memory to construct finobt rebuild transaction.\n"));
> +
> + /* Add all observed finobt records. */
> + error = -libxfs_btree_bload(btr_fino->cur, &btr_fino->bload, btr_fino);
> + if (error)
> + do_error(
> +_("Error %d while creating finobt btree for AG %u.\n"), error, agno);
> +
> + /* Since we're not writing the AGI yet, no need to commit the cursor */
> + libxfs_btree_del_cursor(btr_fino->cur, 0);
> + error = -libxfs_trans_commit(sc->tp);
> + if (error)
> + do_error(
> +_("Error %d while writing finobt btree for AG %u.\n"), error, agno);
> + sc->tp = NULL;
> }
>
> /*
> * XXX: yet more code that can be shared with mkfs, growfs.
> */
> static void
> -build_agi(xfs_mount_t *mp, xfs_agnumber_t agno, bt_status_t *btree_curs,
> - bt_status_t *finobt_curs, struct agi_stat *agi_stat)
> +build_agi(
> + struct xfs_mount *mp,
> + xfs_agnumber_t agno,
> + struct bt_rebuild *btr_ino,
> + struct bt_rebuild *btr_fino,
> + struct agi_stat *agi_stat)
> {
> - xfs_buf_t *agi_buf;
> - xfs_agi_t *agi;
> - int i;
> - int error;
> + struct xfs_buf *agi_buf;
> + struct xfs_agi *agi;
> + int i;
> + int error;
>
> error = -libxfs_buf_get(mp->m_dev,
> XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
> @@ -1008,8 +1053,8 @@ build_agi(xfs_mount_t *mp, xfs_agnumber_t agno, bt_status_t *btree_curs,
> agi->agi_length = cpu_to_be32(mp->m_sb.sb_dblocks -
> (xfs_rfsblock_t) mp->m_sb.sb_agblocks * agno);
> agi->agi_count = cpu_to_be32(agi_stat->count);
> - agi->agi_root = cpu_to_be32(btree_curs->root);
> - agi->agi_level = cpu_to_be32(btree_curs->num_levels);
> + agi->agi_root = cpu_to_be32(btr_ino->newbt.afake.af_root);
> + agi->agi_level = cpu_to_be32(btr_ino->newbt.afake.af_levels);
> agi->agi_freecount = cpu_to_be32(agi_stat->freecount);
> agi->agi_newino = cpu_to_be32(agi_stat->first_agino);
> agi->agi_dirino = cpu_to_be32(NULLAGINO);
> @@ -1021,203 +1066,16 @@ build_agi(xfs_mount_t *mp, xfs_agnumber_t agno, bt_status_t *btree_curs,
> platform_uuid_copy(&agi->agi_uuid, &mp->m_sb.sb_meta_uuid);
>
> if (xfs_sb_version_hasfinobt(&mp->m_sb)) {
> - agi->agi_free_root = cpu_to_be32(finobt_curs->root);
> - agi->agi_free_level = cpu_to_be32(finobt_curs->num_levels);
> + agi->agi_free_root =
> + cpu_to_be32(btr_fino->newbt.afake.af_root);
> + agi->agi_free_level =
> + cpu_to_be32(btr_fino->newbt.afake.af_levels);
> }
>
> libxfs_buf_mark_dirty(agi_buf);
> libxfs_buf_relse(agi_buf);
> }
>
> -/*
> - * rebuilds an inode tree given a cursor. We're lazy here and call
> - * the routine that builds the agi
> - */
> -static void
> -build_ino_tree(xfs_mount_t *mp, xfs_agnumber_t agno,
> - bt_status_t *btree_curs, xfs_btnum_t btnum,
> - struct agi_stat *agi_stat)
> -{
> - xfs_agnumber_t i;
> - xfs_agblock_t j;
> - xfs_agblock_t agbno;
> - xfs_agino_t first_agino;
> - struct xfs_btree_block *bt_hdr;
> - xfs_inobt_rec_t *bt_rec;
> - ino_tree_node_t *ino_rec;
> - bt_stat_level_t *lptr;
> - const struct xfs_buf_ops *ops = btnum_to_ops(btnum);
> - xfs_agino_t count = 0;
> - xfs_agino_t freecount = 0;
> - int inocnt;
> - uint8_t finocnt;
> - int k;
> - int level = btree_curs->num_levels;
> - int spmask;
> - uint64_t sparse;
> - uint16_t holemask;
> - int error;
> -
> - ASSERT(btnum == XFS_BTNUM_INO || btnum == XFS_BTNUM_FINO);
> -
> - for (i = 0; i < level; i++) {
> - lptr = &btree_curs->level[i];
> -
> - agbno = get_next_blockaddr(agno, i, btree_curs);
> - error = -libxfs_buf_get(mp->m_dev,
> - XFS_AGB_TO_DADDR(mp, agno, agbno),
> - XFS_FSB_TO_BB(mp, 1), &lptr->buf_p);
> - if (error)
> - do_error(_("Cannot grab inode btree buffer, err=%d"),
> - error);
> -
> - if (i == btree_curs->num_levels - 1)
> - btree_curs->root = agbno;
> -
> - lptr->agbno = agbno;
> - lptr->prev_agbno = NULLAGBLOCK;
> - lptr->prev_buf_p = NULL;
> - /*
> - * initialize block header
> - */
> -
> - lptr->buf_p->b_ops = ops;
> - bt_hdr = XFS_BUF_TO_BLOCK(lptr->buf_p);
> - memset(bt_hdr, 0, mp->m_sb.sb_blocksize);
> - libxfs_btree_init_block(mp, lptr->buf_p, btnum, i, 0, agno);
> - }
> -
> - /*
> - * run along leaf, setting up records. as we have to switch
> - * blocks, call the prop_ino_cursor routine to set up the new
> - * pointers for the parent. that can recurse up to the root
> - * if required. set the sibling pointers for leaf level here.
> - */
> - if (btnum == XFS_BTNUM_FINO)
> - ino_rec = findfirst_free_inode_rec(agno);
> - else
> - ino_rec = findfirst_inode_rec(agno);
> -
> - if (ino_rec != NULL)
> - first_agino = ino_rec->ino_startnum;
> - else
> - first_agino = NULLAGINO;
> -
> - lptr = &btree_curs->level[0];
> -
> - for (i = 0; i < lptr->num_blocks; i++) {
> - /*
> - * block initialization, lay in block header
> - */
> - lptr->buf_p->b_ops = ops;
> - bt_hdr = XFS_BUF_TO_BLOCK(lptr->buf_p);
> - memset(bt_hdr, 0, mp->m_sb.sb_blocksize);
> - libxfs_btree_init_block(mp, lptr->buf_p, btnum, 0, 0, agno);
> -
> - bt_hdr->bb_u.s.bb_leftsib = cpu_to_be32(lptr->prev_agbno);
> - bt_hdr->bb_numrecs = cpu_to_be16(lptr->num_recs_pb +
> - (lptr->modulo > 0));
> -
> - if (lptr->modulo > 0)
> - lptr->modulo--;
> -
> - if (lptr->num_recs_pb > 0)
> - prop_ino_cursor(mp, agno, btree_curs, btnum,
> - ino_rec->ino_startnum, 0);
> -
> - bt_rec = (xfs_inobt_rec_t *)
> - ((char *)bt_hdr + XFS_INOBT_BLOCK_LEN(mp));
> - for (j = 0; j < be16_to_cpu(bt_hdr->bb_numrecs); j++) {
> - ASSERT(ino_rec != NULL);
> - bt_rec[j].ir_startino =
> - cpu_to_be32(ino_rec->ino_startnum);
> - bt_rec[j].ir_free = cpu_to_be64(ino_rec->ir_free);
> -
> - inocnt = finocnt = 0;
> - for (k = 0; k < sizeof(xfs_inofree_t)*NBBY; k++) {
> - ASSERT(is_inode_confirmed(ino_rec, k));
> -
> - if (is_inode_sparse(ino_rec, k))
> - continue;
> - if (is_inode_free(ino_rec, k))
> - finocnt++;
> - inocnt++;
> - }
> -
> - /*
> - * Set the freecount and check whether we need to update
> - * the sparse format fields. Otherwise, skip to the next
> - * record.
> - */
> - inorec_set_freecount(mp, &bt_rec[j], finocnt);
> - if (!xfs_sb_version_hassparseinodes(&mp->m_sb))
> - goto nextrec;
> -
> - /*
> - * Convert the 64-bit in-core sparse inode state to the
> - * 16-bit on-disk holemask.
> - */
> - holemask = 0;
> - spmask = (1 << XFS_INODES_PER_HOLEMASK_BIT) - 1;
> - sparse = ino_rec->ir_sparse;
> - for (k = 0; k < XFS_INOBT_HOLEMASK_BITS; k++) {
> - if (sparse & spmask) {
> - ASSERT((sparse & spmask) == spmask);
> - holemask |= (1 << k);
> - } else
> - ASSERT((sparse & spmask) == 0);
> - sparse >>= XFS_INODES_PER_HOLEMASK_BIT;
> - }
> -
> - bt_rec[j].ir_u.sp.ir_count = inocnt;
> - bt_rec[j].ir_u.sp.ir_holemask = cpu_to_be16(holemask);
> -
> -nextrec:
> - freecount += finocnt;
> - count += inocnt;
> -
> - if (btnum == XFS_BTNUM_FINO)
> - ino_rec = next_free_ino_rec(ino_rec);
> - else
> - ino_rec = next_ino_rec(ino_rec);
> - }
> -
> - if (ino_rec != NULL) {
> - /*
> - * get next leaf level block
> - */
> - if (lptr->prev_buf_p != NULL) {
> -#ifdef XR_BLD_INO_TRACE
> - fprintf(stderr, "writing inobt agbno %u\n",
> - lptr->prev_agbno);
> -#endif
> - ASSERT(lptr->prev_agbno != NULLAGBLOCK);
> - libxfs_buf_mark_dirty(lptr->prev_buf_p);
> - libxfs_buf_relse(lptr->prev_buf_p);
> - }
> - lptr->prev_buf_p = lptr->buf_p;
> - lptr->prev_agbno = lptr->agbno;
> - lptr->agbno = get_next_blockaddr(agno, 0, btree_curs);
> - bt_hdr->bb_u.s.bb_rightsib = cpu_to_be32(lptr->agbno);
> -
> - error = -libxfs_buf_get(mp->m_dev,
> - XFS_AGB_TO_DADDR(mp, agno, lptr->agbno),
> - XFS_FSB_TO_BB(mp, 1),
> - &lptr->buf_p);
> - if (error)
> - do_error(
> - _("Cannot grab inode btree buffer, err=%d"),
> - error);
> - }
> - }
> -
> - if (agi_stat) {
> - agi_stat->first_agino = first_agino;
> - agi_stat->count = count;
> - agi_stat->freecount = freecount;
> - }
> -}
> -
> /* rebuild the rmap tree */
>
> /*
> @@ -2142,14 +2000,10 @@ phase5_func(
> {
> struct repair_ctx sc = { .mp = mp, };
> struct agi_stat agi_stat = {0,};
> - uint64_t num_inos;
> - uint64_t num_free_inos;
> - uint64_t finobt_num_inos;
> - uint64_t finobt_num_free_inos;
> struct bt_rebuild btr_bno;
> struct bt_rebuild btr_cnt;
> - bt_status_t ino_btree_curs;
> - bt_status_t fino_btree_curs;
> + struct bt_rebuild btr_ino;
> + struct bt_rebuild btr_fino;
> bt_status_t rmap_btree_curs;
> bt_status_t refcnt_btree_curs;
> int extra_blocks = 0;
> @@ -2184,19 +2038,8 @@ _("unable to rebuild AG %u. Not enough free space in on-disk AG.\n"),
> agno);
> }
>
> - /*
> - * ok, now set up the btree cursors for the on-disk btrees (includes
> - * pre-allocating all required blocks for the trees themselves)
> - */
> - init_ino_cursor(mp, agno, &ino_btree_curs, &num_inos,
> - &num_free_inos, 0);
> -
> - if (xfs_sb_version_hasfinobt(&mp->m_sb))
> - init_ino_cursor(mp, agno, &fino_btree_curs, &finobt_num_inos,
> - &finobt_num_free_inos, 1);
> -
> - sb_icount_ag[agno] += num_inos;
> - sb_ifree_ag[agno] += num_free_inos;
> + init_ino_cursors(&sc, agno, num_freeblocks, &sb_icount_ag[agno],
> + &sb_ifree_ag[agno], &btr_ino, &btr_fino);
>
> /*
> * Set up the btree cursors for the on-disk rmap btrees, which includes
> @@ -2287,34 +2130,27 @@ _("unable to rebuild AG %u. Not enough free space in on-disk AG.\n"),
> &rmap_btree_curs, &refcnt_btree_curs, lost_fsb);
>
> /*
> - * build inode allocation tree.
> + * build inode allocation trees.
> */
> - build_ino_tree(mp, agno, &ino_btree_curs, XFS_BTNUM_INO, &agi_stat);
> - write_cursor(&ino_btree_curs);
> -
> - /*
> - * build free inode tree
> - */
> - if (xfs_sb_version_hasfinobt(&mp->m_sb)) {
> - build_ino_tree(mp, agno, &fino_btree_curs,
> - XFS_BTNUM_FINO, NULL);
> - write_cursor(&fino_btree_curs);
> - }
> + build_inobt(&sc, agno, &btr_ino, &agi_stat);
> + if (xfs_sb_version_hasfinobt(&mp->m_sb))
> + build_finobt(&sc, agno, &btr_fino);
>
> /* build the agi */
> - build_agi(mp, agno, &ino_btree_curs, &fino_btree_curs, &agi_stat);
> + build_agi(mp, agno, &btr_ino, &btr_fino, &agi_stat);
>
> /*
> * tear down cursors
> */
> finish_rebuild(mp, &btr_bno, lost_fsb);
> finish_rebuild(mp, &btr_cnt, lost_fsb);
> + finish_rebuild(mp, &btr_ino, lost_fsb);
> + if (xfs_sb_version_hasfinobt(&mp->m_sb))
> + finish_rebuild(mp, &btr_fino, lost_fsb);
> if (xfs_sb_version_hasrmapbt(&mp->m_sb))
> finish_cursor(&rmap_btree_curs);
> if (xfs_sb_version_hasreflink(&mp->m_sb))
> finish_cursor(&refcnt_btree_curs);
> - if (xfs_sb_version_hasfinobt(&mp->m_sb))
> - finish_cursor(&fino_btree_curs);
>
> /*
> * release the incore per-AG bno/bcnt trees so the extent nodes
>
^ permalink raw reply [flat|nested] 30+ messages in thread
* Re: [PATCH 9/9] xfs_repair: track blocks lost during btree construction via extents
2020-05-20 1:51 ` [PATCH 9/9] xfs_repair: track blocks lost during btree construction via extents Darrick J. Wong
@ 2020-05-28 17:00 ` Brian Foster
2020-05-29 22:19 ` Darrick J. Wong
0 siblings, 1 reply; 30+ messages in thread
From: Brian Foster @ 2020-05-28 17:00 UTC (permalink / raw)
To: Darrick J. Wong; +Cc: sandeen, linux-xfs
On Tue, May 19, 2020 at 06:51:40PM -0700, Darrick J. Wong wrote:
> From: Darrick J. Wong <darrick.wong@oracle.com>
>
> Use extent records (not just raw fsbs) to track blocks that were lost
> during btree construction. This makes it somewhat more efficient.
>
> Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
> ---
> repair/phase5.c | 61 ++++++++++++++++++++++++++++++++-----------------------
> 1 file changed, 35 insertions(+), 26 deletions(-)
>
>
> diff --git a/repair/phase5.c b/repair/phase5.c
> index 72c6908a..22007275 100644
> --- a/repair/phase5.c
> +++ b/repair/phase5.c
> @@ -45,6 +45,12 @@ struct bt_rebuild {
> };
> };
>
> +struct lost_fsb {
> + xfs_fsblock_t fsbno;
> + xfs_extlen_t len;
> +};
> +
Looks reasonable at a glance, but could we call the above structure an
extent and rename lost_fsbs to lost_extents or some such?
Brian
> +
> /*
> * extra metadata for the agi
> */
> @@ -314,21 +320,24 @@ static void
> finish_rebuild(
> struct xfs_mount *mp,
> struct bt_rebuild *btr,
> - struct xfs_slab *lost_fsb)
> + struct xfs_slab *lost_fsbs)
> {
> struct xrep_newbt_resv *resv, *n;
>
> for_each_xrep_newbt_reservation(&btr->newbt, resv, n) {
> - while (resv->used < resv->len) {
> - xfs_fsblock_t fsb = resv->fsbno + resv->used;
> - int error;
> + struct lost_fsb lost;
> + int error;
>
> - error = slab_add(lost_fsb, &fsb);
> - if (error)
> - do_error(
> + if (resv->used == resv->len)
> + continue;
> +
> + lost.fsbno = resv->fsbno + resv->used;
> + lost.len = resv->len - resv->used;
> + error = slab_add(lost_fsbs, &lost);
> + if (error)
> + do_error(
> _("Insufficient memory saving lost blocks.\n"));
> - resv->used++;
> - }
> + resv->used = resv->len;
> }
>
> xrep_newbt_destroy(&btr->newbt, 0);
> @@ -1036,7 +1045,7 @@ build_agf_agfl(
> int lostblocks, /* # blocks that will be lost */
> struct bt_rebuild *btr_rmap,
> struct bt_rebuild *btr_refc,
> - struct xfs_slab *lost_fsb)
> + struct xfs_slab *lost_fsbs)
> {
> struct extent_tree_node *ext_ptr;
> struct xfs_buf *agf_buf, *agfl_buf;
> @@ -1253,7 +1262,7 @@ static void
> phase5_func(
> struct xfs_mount *mp,
> xfs_agnumber_t agno,
> - struct xfs_slab *lost_fsb)
> + struct xfs_slab *lost_fsbs)
> {
> struct repair_ctx sc = { .mp = mp, };
> struct agi_stat agi_stat = {0,};
> @@ -1372,7 +1381,7 @@ _("unable to rebuild AG %u. Not enough free space in on-disk AG.\n"),
> * set up agf and agfl
> */
> build_agf_agfl(mp, agno, &btr_bno, &btr_cnt, freeblks1, extra_blocks,
> - &btr_rmap, &btr_refc, lost_fsb);
> + &btr_rmap, &btr_refc, lost_fsbs);
>
> /*
> * build inode allocation trees.
> @@ -1387,15 +1396,15 @@ _("unable to rebuild AG %u. Not enough free space in on-disk AG.\n"),
> /*
> * tear down cursors
> */
> - finish_rebuild(mp, &btr_bno, lost_fsb);
> - finish_rebuild(mp, &btr_cnt, lost_fsb);
> - finish_rebuild(mp, &btr_ino, lost_fsb);
> + finish_rebuild(mp, &btr_bno, lost_fsbs);
> + finish_rebuild(mp, &btr_cnt, lost_fsbs);
> + finish_rebuild(mp, &btr_ino, lost_fsbs);
> if (xfs_sb_version_hasfinobt(&mp->m_sb))
> - finish_rebuild(mp, &btr_fino, lost_fsb);
> + finish_rebuild(mp, &btr_fino, lost_fsbs);
> if (xfs_sb_version_hasrmapbt(&mp->m_sb))
> - finish_rebuild(mp, &btr_rmap, lost_fsb);
> + finish_rebuild(mp, &btr_rmap, lost_fsbs);
> if (xfs_sb_version_hasreflink(&mp->m_sb))
> - finish_rebuild(mp, &btr_refc, lost_fsb);
> + finish_rebuild(mp, &btr_refc, lost_fsbs);
>
> /*
> * release the incore per-AG bno/bcnt trees so the extent nodes
> @@ -1414,19 +1423,19 @@ inject_lost_blocks(
> {
> struct xfs_trans *tp = NULL;
> struct xfs_slab_cursor *cur = NULL;
> - xfs_fsblock_t *fsb;
> + struct lost_fsb *lost;
> int error;
>
> error = init_slab_cursor(lost_fsbs, NULL, &cur);
> if (error)
> return error;
>
> - while ((fsb = pop_slab_cursor(cur)) != NULL) {
> + while ((lost = pop_slab_cursor(cur)) != NULL) {
> error = -libxfs_trans_alloc_rollable(mp, 16, &tp);
> if (error)
> goto out_cancel;
>
> - error = -libxfs_free_extent(tp, *fsb, 1,
> + error = -libxfs_free_extent(tp, lost->fsbno, lost->len,
> &XFS_RMAP_OINFO_ANY_OWNER, XFS_AG_RESV_NONE);
> if (error)
> goto out_cancel;
> @@ -1447,7 +1456,7 @@ inject_lost_blocks(
> void
> phase5(xfs_mount_t *mp)
> {
> - struct xfs_slab *lost_fsb;
> + struct xfs_slab *lost_fsbs;
> xfs_agnumber_t agno;
> int error;
>
> @@ -1490,12 +1499,12 @@ phase5(xfs_mount_t *mp)
> if (sb_fdblocks_ag == NULL)
> do_error(_("cannot alloc sb_fdblocks_ag buffers\n"));
>
> - error = init_slab(&lost_fsb, sizeof(xfs_fsblock_t));
> + error = init_slab(&lost_fsbs, sizeof(struct lost_fsb));
> if (error)
> do_error(_("cannot alloc lost block slab\n"));
>
> for (agno = 0; agno < mp->m_sb.sb_agcount; agno++)
> - phase5_func(mp, agno, lost_fsb);
> + phase5_func(mp, agno, lost_fsbs);
>
> print_final_rpt();
>
> @@ -1538,10 +1547,10 @@ _("unable to add AG %u reverse-mapping data to btree.\n"), agno);
> * Put blocks that were unnecessarily reserved for btree
> * reconstruction back into the filesystem free space data.
> */
> - error = inject_lost_blocks(mp, lost_fsb);
> + error = inject_lost_blocks(mp, lost_fsbs);
> if (error)
> do_error(_("Unable to reinsert lost blocks into filesystem.\n"));
> - free_slab(&lost_fsb);
> + free_slab(&lost_fsbs);
>
> bad_ino_btree = 0;
>
>
^ permalink raw reply [flat|nested] 30+ messages in thread
* Re: [PATCH 1/9] xfs_repair: port the online repair newbt structure
2020-05-28 15:08 ` Brian Foster
@ 2020-05-29 21:01 ` Darrick J. Wong
2020-06-01 12:03 ` Brian Foster
0 siblings, 1 reply; 30+ messages in thread
From: Darrick J. Wong @ 2020-05-29 21:01 UTC (permalink / raw)
To: Brian Foster; +Cc: sandeen, linux-xfs
On Thu, May 28, 2020 at 11:08:36AM -0400, Brian Foster wrote:
> On Wed, May 27, 2020 at 03:34:24PM -0700, Darrick J. Wong wrote:
> > On Wed, May 27, 2020 at 08:15:31AM -0400, Brian Foster wrote:
> > > On Tue, May 19, 2020 at 06:50:49PM -0700, Darrick J. Wong wrote:
> > > > From: Darrick J. Wong <darrick.wong@oracle.com>
> > > >
> > > > Port the new btree staging context and related block reservation helper
> > > > code from the kernel to repair. We'll use this in subsequent patches to
> > > > implement btree bulk loading.
> > > >
> > > > Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
> > > > ---
> > > > include/libxfs.h | 1
> > > > libxfs/libxfs_api_defs.h | 2
> > > > repair/Makefile | 4 -
> > > > repair/bload.c | 303 ++++++++++++++++++++++++++++++++++++++++++++++
> > > > repair/bload.h | 77 ++++++++++++
> > > > repair/xfs_repair.c | 17 +++
> > > > 6 files changed, 402 insertions(+), 2 deletions(-)
> > > > create mode 100644 repair/bload.c
> > > > create mode 100644 repair/bload.h
> > > >
> > > >
> > > ...
> > > > diff --git a/repair/bload.c b/repair/bload.c
> > > > new file mode 100644
> > > > index 00000000..9bc17468
> > > > --- /dev/null
> > > > +++ b/repair/bload.c
> > > > @@ -0,0 +1,303 @@
> > > > +// SPDX-License-Identifier: GPL-2.0-or-later
> > > > +/*
> > > > + * Copyright (C) 2020 Oracle. All Rights Reserved.
> > > > + * Author: Darrick J. Wong <darrick.wong@oracle.com>
> > > > + */
> > > > +#include <libxfs.h>
> > > > +#include "bload.h"
> > > > +
> > > > +#define trace_xrep_newbt_claim_block(...) ((void) 0)
> > > > +#define trace_xrep_newbt_free_blocks(...) ((void) 0)
> > > > +
> > > > +int bload_leaf_slack = -1;
> > > > +int bload_node_slack = -1;
> > > > +
> > > > +/* Ported routines from fs/xfs/scrub/repair.c */
> > > > +
> > >
> > > Looks mostly straightforward, but I'll have to come back to this as I
> > > get to the code that uses it later in the series. In the meantime, I see
> > > some of these helpers in scrub/repair.c while not others. Are there
> > > references to other routines that are intended to be copies from kernel
> > > code?
> >
> > Hm. I might not understand the question, but in general the code should
> > be fairly similar to the kernel functions. The biggest differences are
> > (a) that whole libxfs error code mess, (b) the much simpler repair_ctx
> > structure, and (c) the fact that repair doesn't bother with EFIs to
> > automatically reap blocks.
> >
> > So... the ten functions you see here do the same things as their kernel
> > counterparts, but they get to do it in the much simpler userspace
> > environment.
> >
>
> Right.. I was able to find the first function (xrep_roll_ag_trans())
> easily in the kernel because it has the same name. The next one or two
> (i.e., xrep_newbt_*()) I couldn't find and then gave up. Are they
> renamed? Unmerged?
Ahh, silly me, the rest are as yet unmerged components of the online
repair code. Maybe it makes more sense to drop the "Ported routines
from XXXX" comment, particularly since it probably will be easier to
merge the xfs_repair series first, which is /much/ smaller in scope.
> > The other functions in scrub/repair.c that didn't get ported are either
> > for other types of repairs or exist to support the in-kernel code and
> > aren't needed here.
> >
>
> Sure, I'm just curious how to identify the source of the ones that are.
https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux.git/tree/fs/xfs/scrub/repair.c?h=djwong-wtf
Is a fairly recent snapshot of what the kernel version looks like.
--D
>
> Brian
>
> > --D
> >
> > > Brian
> > >
> > > > +/*
> > > > + * Roll a transaction, keeping the AG headers locked and reinitializing
> > > > + * the btree cursors.
> > > > + */
> > > > +int
> > > > +xrep_roll_ag_trans(
> > > > + struct repair_ctx *sc)
> > > > +{
> > > > + int error;
> > > > +
> > > > + /* Keep the AG header buffers locked so we can keep going. */
> > > > + if (sc->agi_bp)
> > > > + libxfs_trans_bhold(sc->tp, sc->agi_bp);
> > > > + if (sc->agf_bp)
> > > > + libxfs_trans_bhold(sc->tp, sc->agf_bp);
> > > > + if (sc->agfl_bp)
> > > > + libxfs_trans_bhold(sc->tp, sc->agfl_bp);
> > > > +
> > > > + /*
> > > > + * Roll the transaction. We still own the buffer and the buffer lock
> > > > + * regardless of whether or not the roll succeeds. If the roll fails,
> > > > + * the buffers will be released during teardown on our way out of the
> > > > + * kernel. If it succeeds, we join them to the new transaction and
> > > > + * move on.
> > > > + */
> > > > + error = -libxfs_trans_roll(&sc->tp);
> > > > + if (error)
> > > > + return error;
> > > > +
> > > > + /* Join AG headers to the new transaction. */
> > > > + if (sc->agi_bp)
> > > > + libxfs_trans_bjoin(sc->tp, sc->agi_bp);
> > > > + if (sc->agf_bp)
> > > > + libxfs_trans_bjoin(sc->tp, sc->agf_bp);
> > > > + if (sc->agfl_bp)
> > > > + libxfs_trans_bjoin(sc->tp, sc->agfl_bp);
> > > > +
> > > > + return 0;
> > > > +}
> > > > +
> > > > +/* Initialize accounting resources for staging a new AG btree. */
> > > > +void
> > > > +xrep_newbt_init_ag(
> > > > + struct xrep_newbt *xnr,
> > > > + struct repair_ctx *sc,
> > > > + const struct xfs_owner_info *oinfo,
> > > > + xfs_fsblock_t alloc_hint,
> > > > + enum xfs_ag_resv_type resv)
> > > > +{
> > > > + memset(xnr, 0, sizeof(struct xrep_newbt));
> > > > + xnr->sc = sc;
> > > > + xnr->oinfo = *oinfo; /* structure copy */
> > > > + xnr->alloc_hint = alloc_hint;
> > > > + xnr->resv = resv;
> > > > + INIT_LIST_HEAD(&xnr->resv_list);
> > > > +}
> > > > +
> > > > +/* Initialize accounting resources for staging a new inode fork btree. */
> > > > +void
> > > > +xrep_newbt_init_inode(
> > > > + struct xrep_newbt *xnr,
> > > > + struct repair_ctx *sc,
> > > > + int whichfork,
> > > > + const struct xfs_owner_info *oinfo)
> > > > +{
> > > > + xrep_newbt_init_ag(xnr, sc, oinfo,
> > > > + XFS_INO_TO_FSB(sc->mp, sc->ip->i_ino),
> > > > + XFS_AG_RESV_NONE);
> > > > + xnr->ifake.if_fork = kmem_zone_zalloc(xfs_ifork_zone, 0);
> > > > + xnr->ifake.if_fork_size = XFS_IFORK_SIZE(sc->ip, whichfork);
> > > > +}
> > > > +
> > > > +/*
> > > > + * Initialize accounting resources for staging a new btree. Callers are
> > > > + * expected to add their own reservations (and clean them up) manually.
> > > > + */
> > > > +void
> > > > +xrep_newbt_init_bare(
> > > > + struct xrep_newbt *xnr,
> > > > + struct repair_ctx *sc)
> > > > +{
> > > > + xrep_newbt_init_ag(xnr, sc, &XFS_RMAP_OINFO_ANY_OWNER, NULLFSBLOCK,
> > > > + XFS_AG_RESV_NONE);
> > > > +}
> > > > +
> > > > +/* Designate specific blocks to be used to build our new btree. */
> > > > +int
> > > > +xrep_newbt_add_blocks(
> > > > + struct xrep_newbt *xnr,
> > > > + xfs_fsblock_t fsbno,
> > > > + xfs_extlen_t len)
> > > > +{
> > > > + struct xrep_newbt_resv *resv;
> > > > +
> > > > + resv = kmem_alloc(sizeof(struct xrep_newbt_resv), KM_MAYFAIL);
> > > > + if (!resv)
> > > > + return ENOMEM;
> > > > +
> > > > + INIT_LIST_HEAD(&resv->list);
> > > > + resv->fsbno = fsbno;
> > > > + resv->len = len;
> > > > + resv->used = 0;
> > > > + list_add_tail(&resv->list, &xnr->resv_list);
> > > > + return 0;
> > > > +}
> > > > +
> > > > +/* Reserve disk space for our new btree. */
> > > > +int
> > > > +xrep_newbt_alloc_blocks(
> > > > + struct xrep_newbt *xnr,
> > > > + uint64_t nr_blocks)
> > > > +{
> > > > + struct repair_ctx *sc = xnr->sc;
> > > > + xfs_alloctype_t type;
> > > > + xfs_fsblock_t alloc_hint = xnr->alloc_hint;
> > > > + int error = 0;
> > > > +
> > > > + type = sc->ip ? XFS_ALLOCTYPE_START_BNO : XFS_ALLOCTYPE_NEAR_BNO;
> > > > +
> > > > + while (nr_blocks > 0 && !error) {
> > > > + struct xfs_alloc_arg args = {
> > > > + .tp = sc->tp,
> > > > + .mp = sc->mp,
> > > > + .type = type,
> > > > + .fsbno = alloc_hint,
> > > > + .oinfo = xnr->oinfo,
> > > > + .minlen = 1,
> > > > + .maxlen = nr_blocks,
> > > > + .prod = 1,
> > > > + .resv = xnr->resv,
> > > > + };
> > > > +
> > > > + error = -libxfs_alloc_vextent(&args);
> > > > + if (error)
> > > > + return error;
> > > > + if (args.fsbno == NULLFSBLOCK)
> > > > + return ENOSPC;
> > > > +
> > > > + /* We don't have real EFIs here so skip that. */
> > > > +
> > > > + error = xrep_newbt_add_blocks(xnr, args.fsbno, args.len);
> > > > + if (error)
> > > > + break;
> > > > +
> > > > + nr_blocks -= args.len;
> > > > + alloc_hint = args.fsbno + args.len - 1;
> > > > +
> > > > + if (sc->ip)
> > > > + error = -libxfs_trans_roll_inode(&sc->tp, sc->ip);
> > > > + else
> > > > + error = xrep_roll_ag_trans(sc);
> > > > + }
> > > > +
> > > > + return error;
> > > > +}
> > > > +
> > > > +/*
> > > > + * Release blocks that were reserved for a btree repair. If the repair
> > > > + * succeeded then we log deferred frees for unused blocks. Otherwise, we try
> > > > + * to free the extents immediately to roll the filesystem back to where it was
> > > > + * before we started.
> > > > + */
> > > > +static inline int
> > > > +xrep_newbt_destroy_reservation(
> > > > + struct xrep_newbt *xnr,
> > > > + struct xrep_newbt_resv *resv,
> > > > + bool cancel_repair)
> > > > +{
> > > > + struct repair_ctx *sc = xnr->sc;
> > > > +
> > > > + if (cancel_repair) {
> > > > + int error;
> > > > +
> > > > + /* Free the extent then roll the transaction. */
> > > > + error = -libxfs_free_extent(sc->tp, resv->fsbno, resv->len,
> > > > + &xnr->oinfo, xnr->resv);
> > > > + if (error)
> > > > + return error;
> > > > +
> > > > + if (sc->ip)
> > > > + return -libxfs_trans_roll_inode(&sc->tp, sc->ip);
> > > > + return xrep_roll_ag_trans(sc);
> > > > + }
> > > > +
> > > > + /* We don't have EFIs here so skip the EFD. */
> > > > +
> > > > + /*
> > > > + * Use the deferred freeing mechanism to schedule for deletion any
> > > > + * blocks we didn't use to rebuild the tree. This enables us to log
> > > > + * them all in the same transaction as the root change.
> > > > + */
> > > > + resv->fsbno += resv->used;
> > > > + resv->len -= resv->used;
> > > > + resv->used = 0;
> > > > +
> > > > + if (resv->len == 0)
> > > > + return 0;
> > > > +
> > > > + trace_xrep_newbt_free_blocks(sc->mp,
> > > > + XFS_FSB_TO_AGNO(sc->mp, resv->fsbno),
> > > > + XFS_FSB_TO_AGBNO(sc->mp, resv->fsbno),
> > > > + resv->len, xnr->oinfo.oi_owner);
> > > > +
> > > > + __xfs_bmap_add_free(sc->tp, resv->fsbno, resv->len, &xnr->oinfo, true);
> > > > +
> > > > + return 0;
> > > > +}
> > > > +
> > > > +/* Free all the accounting info and disk space we reserved for a new btree. */
> > > > +void
> > > > +xrep_newbt_destroy(
> > > > + struct xrep_newbt *xnr,
> > > > + int error)
> > > > +{
> > > > + struct repair_ctx *sc = xnr->sc;
> > > > + struct xrep_newbt_resv *resv, *n;
> > > > + int err2;
> > > > +
> > > > + list_for_each_entry_safe(resv, n, &xnr->resv_list, list) {
> > > > + err2 = xrep_newbt_destroy_reservation(xnr, resv, error != 0);
> > > > + if (err2)
> > > > + goto junkit;
> > > > +
> > > > + list_del(&resv->list);
> > > > + kmem_free(resv);
> > > > + }
> > > > +
> > > > +junkit:
> > > > + /*
> > > > + * If we still have reservations attached to @newbt, cleanup must have
> > > > + * failed and the filesystem is about to go down. Clean up the incore
> > > > + * reservations.
> > > > + */
> > > > + list_for_each_entry_safe(resv, n, &xnr->resv_list, list) {
> > > > + list_del(&resv->list);
> > > > + kmem_free(resv);
> > > > + }
> > > > +
> > > > + if (sc->ip) {
> > > > + kmem_cache_free(xfs_ifork_zone, xnr->ifake.if_fork);
> > > > + xnr->ifake.if_fork = NULL;
> > > > + }
> > > > +}
> > > > +
> > > > +/* Feed one of the reserved btree blocks to the bulk loader. */
> > > > +int
> > > > +xrep_newbt_claim_block(
> > > > + struct xfs_btree_cur *cur,
> > > > + struct xrep_newbt *xnr,
> > > > + union xfs_btree_ptr *ptr)
> > > > +{
> > > > + struct xrep_newbt_resv *resv;
> > > > + xfs_fsblock_t fsb;
> > > > +
> > > > + /*
> > > > + * The first item in the list should always have a free block unless
> > > > + * we're completely out.
> > > > + */
> > > > + resv = list_first_entry(&xnr->resv_list, struct xrep_newbt_resv, list);
> > > > + if (resv->used == resv->len)
> > > > + return ENOSPC;
> > > > +
> > > > + /*
> > > > + * Peel off a block from the start of the reservation. We allocate
> > > > + * blocks in order to place blocks on disk in increasing record or key
> > > > + * order. The block reservations tend to end up on the list in
> > > > + * decreasing order, which hopefully results in leaf blocks ending up
> > > > + * together.
> > > > + */
> > > > + fsb = resv->fsbno + resv->used;
> > > > + resv->used++;
> > > > +
> > > > + /* If we used all the blocks in this reservation, move it to the end. */
> > > > + if (resv->used == resv->len)
> > > > + list_move_tail(&resv->list, &xnr->resv_list);
> > > > +
> > > > + trace_xrep_newbt_claim_block(cur->bc_mp,
> > > > + XFS_FSB_TO_AGNO(cur->bc_mp, fsb),
> > > > + XFS_FSB_TO_AGBNO(cur->bc_mp, fsb),
> > > > + 1, xnr->oinfo.oi_owner);
> > > > +
> > > > + if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
> > > > + ptr->l = cpu_to_be64(fsb);
> > > > + else
> > > > + ptr->s = cpu_to_be32(XFS_FSB_TO_AGBNO(cur->bc_mp, fsb));
> > > > + return 0;
> > > > +}
> > > > diff --git a/repair/bload.h b/repair/bload.h
> > > > new file mode 100644
> > > > index 00000000..020c4834
> > > > --- /dev/null
> > > > +++ b/repair/bload.h
> > > > @@ -0,0 +1,77 @@
> > > > +// SPDX-License-Identifier: GPL-2.0-or-later
> > > > +/*
> > > > + * Copyright (C) 2020 Oracle. All Rights Reserved.
> > > > + * Author: Darrick J. Wong <darrick.wong@oracle.com>
> > > > + */
> > > > +#ifndef __XFS_REPAIR_BLOAD_H__
> > > > +#define __XFS_REPAIR_BLOAD_H__
> > > > +
> > > > +extern int bload_leaf_slack;
> > > > +extern int bload_node_slack;
> > > > +
> > > > +struct repair_ctx {
> > > > + struct xfs_mount *mp;
> > > > + struct xfs_inode *ip;
> > > > + struct xfs_trans *tp;
> > > > +
> > > > + struct xfs_buf *agi_bp;
> > > > + struct xfs_buf *agf_bp;
> > > > + struct xfs_buf *agfl_bp;
> > > > +};
> > > > +
> > > > +struct xrep_newbt_resv {
> > > > + /* Link to list of extents that we've reserved. */
> > > > + struct list_head list;
> > > > +
> > > > + /* FSB of the block we reserved. */
> > > > + xfs_fsblock_t fsbno;
> > > > +
> > > > + /* Length of the reservation. */
> > > > + xfs_extlen_t len;
> > > > +
> > > > + /* How much of this reservation we've used. */
> > > > + xfs_extlen_t used;
> > > > +};
> > > > +
> > > > +struct xrep_newbt {
> > > > + struct repair_ctx *sc;
> > > > +
> > > > + /* List of extents that we've reserved. */
> > > > + struct list_head resv_list;
> > > > +
> > > > + /* Fake root for new btree. */
> > > > + union {
> > > > + struct xbtree_afakeroot afake;
> > > > + struct xbtree_ifakeroot ifake;
> > > > + };
> > > > +
> > > > + /* rmap owner of these blocks */
> > > > + struct xfs_owner_info oinfo;
> > > > +
> > > > + /* The last reservation we allocated from. */
> > > > + struct xrep_newbt_resv *last_resv;
> > > > +
> > > > + /* Allocation hint */
> > > > + xfs_fsblock_t alloc_hint;
> > > > +
> > > > + /* per-ag reservation type */
> > > > + enum xfs_ag_resv_type resv;
> > > > +};
> > > > +
> > > > +#define for_each_xrep_newbt_reservation(xnr, resv, n) \
> > > > + list_for_each_entry_safe((resv), (n), &(xnr)->resv_list, list)
> > > > +
> > > > +void xrep_newbt_init_bare(struct xrep_newbt *xnr, struct repair_ctx *sc);
> > > > +void xrep_newbt_init_ag(struct xrep_newbt *xnr, struct repair_ctx *sc,
> > > > + const struct xfs_owner_info *oinfo, xfs_fsblock_t alloc_hint,
> > > > + enum xfs_ag_resv_type resv);
> > > > +void xrep_newbt_init_inode(struct xrep_newbt *xnr, struct repair_ctx *sc,
> > > > + int whichfork, const struct xfs_owner_info *oinfo);
> > > > +int xrep_newbt_add_blocks(struct xrep_newbt *xnr, xfs_fsblock_t fsbno,
> > > > + xfs_extlen_t len);
> > > > +int xrep_newbt_alloc_blocks(struct xrep_newbt *xnr, uint64_t nr_blocks);
> > > > +void xrep_newbt_destroy(struct xrep_newbt *xnr, int error);
> > > > +int xrep_newbt_claim_block(struct xfs_btree_cur *cur, struct xrep_newbt *xnr,
> > > > + union xfs_btree_ptr *ptr);
> > > > +
> > > > +#endif /* __XFS_REPAIR_BLOAD_H__ */
> > > > diff --git a/repair/xfs_repair.c b/repair/xfs_repair.c
> > > > index 9d72fa8e..8fbd3649 100644
> > > > --- a/repair/xfs_repair.c
> > > > +++ b/repair/xfs_repair.c
> > > > @@ -24,6 +24,7 @@
> > > > #include "rmap.h"
> > > > #include "libfrog/fsgeom.h"
> > > > #include "libfrog/platform.h"
> > > > +#include "bload.h"
> > > >
> > > > /*
> > > > * option tables for getsubopt calls
> > > > @@ -39,6 +40,8 @@ enum o_opt_nums {
> > > > AG_STRIDE,
> > > > FORCE_GEO,
> > > > PHASE2_THREADS,
> > > > + BLOAD_LEAF_SLACK,
> > > > + BLOAD_NODE_SLACK,
> > > > O_MAX_OPTS,
> > > > };
> > > >
> > > > @@ -49,6 +52,8 @@ static char *o_opts[] = {
> > > > [AG_STRIDE] = "ag_stride",
> > > > [FORCE_GEO] = "force_geometry",
> > > > [PHASE2_THREADS] = "phase2_threads",
> > > > + [BLOAD_LEAF_SLACK] = "debug_bload_leaf_slack",
> > > > + [BLOAD_NODE_SLACK] = "debug_bload_node_slack",
> > > > [O_MAX_OPTS] = NULL,
> > > > };
> > > >
> > > > @@ -260,6 +265,18 @@ process_args(int argc, char **argv)
> > > > _("-o phase2_threads requires a parameter\n"));
> > > > phase2_threads = (int)strtol(val, NULL, 0);
> > > > break;
> > > > + case BLOAD_LEAF_SLACK:
> > > > + if (!val)
> > > > + do_abort(
> > > > + _("-o debug_bload_leaf_slack requires a parameter\n"));
> > > > + bload_leaf_slack = (int)strtol(val, NULL, 0);
> > > > + break;
> > > > + case BLOAD_NODE_SLACK:
> > > > + if (!val)
> > > > + do_abort(
> > > > + _("-o debug_bload_node_slack requires a parameter\n"));
> > > > + bload_node_slack = (int)strtol(val, NULL, 0);
> > > > + break;
> > > > default:
> > > > unknown('o', val);
> > > > break;
> > > >
> > >
> >
>
^ permalink raw reply [flat|nested] 30+ messages in thread
* Re: [PATCH 3/9] xfs_repair: create a new class of btree rebuild cursors
2020-05-28 15:09 ` Brian Foster
@ 2020-05-29 21:08 ` Darrick J. Wong
0 siblings, 0 replies; 30+ messages in thread
From: Darrick J. Wong @ 2020-05-29 21:08 UTC (permalink / raw)
To: Brian Foster; +Cc: sandeen, linux-xfs
On Thu, May 28, 2020 at 11:09:21AM -0400, Brian Foster wrote:
> On Wed, May 27, 2020 at 03:07:33PM -0700, Darrick J. Wong wrote:
> > On Wed, May 27, 2020 at 08:18:04AM -0400, Brian Foster wrote:
> > > On Tue, May 19, 2020 at 06:51:02PM -0700, Darrick J. Wong wrote:
> > > > From: Darrick J. Wong <darrick.wong@oracle.com>
> > > >
> > > > Create some new support structures and functions to assist phase5 in
> > > > using the btree bulk loader to reconstruct metadata btrees. This is the
> > > > first step in removing the open-coded rebuilding code.
> > > >
> > > > Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
> > > > ---
> > > > repair/phase5.c | 239 ++++++++++++++++++++++++++++++++++++++++++++++++++-----
> > > > 1 file changed, 218 insertions(+), 21 deletions(-)
> > > >
> > > >
> > > > diff --git a/repair/phase5.c b/repair/phase5.c
> > > > index 84c05a13..8f5e5f59 100644
> > > > --- a/repair/phase5.c
> > > > +++ b/repair/phase5.c
> > > > @@ -18,6 +18,7 @@
> > > > #include "progress.h"
> > > > #include "slab.h"
> > > > #include "rmap.h"
> > > > +#include "bload.h"
> > > >
> > > > /*
> > > > * we maintain the current slice (path from root to leaf)
> > > ...
> > > > @@ -306,6 +324,156 @@ _("error - not enough free space in filesystem\n"));
> > > > #endif
> > > > }
> > > >
> > > ...
> > > > +static void
> > > > +consume_freespace(
> > > > + xfs_agnumber_t agno,
> > > > + struct extent_tree_node *ext_ptr,
> > > > + uint32_t len)
> > > > +{
> > > > + struct extent_tree_node *bno_ext_ptr;
> > > > + xfs_agblock_t new_start = ext_ptr->ex_startblock + len;
> > > > + xfs_extlen_t new_len = ext_ptr->ex_blockcount - len;
> > > > +
> > > > + /* Delete the used-up extent from both extent trees. */
> > > > +#ifdef XR_BLD_FREE_TRACE
> > > > + fprintf(stderr, "releasing extent: %u [%u %u]\n", agno,
> > > > + ext_ptr->ex_startblock, ext_ptr->ex_blockcount);
> > > > +#endif
> > > > + bno_ext_ptr = find_bno_extent(agno, ext_ptr->ex_startblock);
> > > > + ASSERT(bno_ext_ptr != NULL);
> > > > + get_bno_extent(agno, bno_ext_ptr);
> > > > + release_extent_tree_node(bno_ext_ptr);
> > > > +
> > > > + ext_ptr = get_bcnt_extent(agno, ext_ptr->ex_startblock,
> > > > + ext_ptr->ex_blockcount);
> > > > + release_extent_tree_node(ext_ptr);
> > > > +
> > >
> > > Not having looked too deeply at the in-core extent tracking structures,
> > > is there any particular reason we unconditionally remove and reinsert
> > > new records each time around? Is it because we're basically changing the
> > > extent index in the tree? If so, comment please (an update to the
> > > comment below is probably fine). :)
> >
> > Yes. We're changing the free space tree records, and the incore bno and
> > cnt trees maintain the records in sorted order. Therefore, if we want
> > to change a record we have to delete the record from the tree and
> > reinsert it.
> >
> > > > + /*
> > > > + * If we only used part of this last extent, then we must reinsert the
> > > > + * extent in the extent trees.
> >
> > /*
> > * If we only used part of this last extent, then we must reinsert the
> > * extent to maintain proper sorting order.
> > */
> >
> > How about that?
> >
>
> Works for me, thanks.
>
> > > > + */
> > > > + if (new_len > 0) {
> > > > + add_bno_extent(agno, new_start, new_len);
> > > > + add_bcnt_extent(agno, new_start, new_len);
> > > > + }
> > > > +}
> > > > +
> > > > +/* Reserve blocks for the new btree. */
> > > > +static void
> > > > +setup_rebuild(
> > > > + struct xfs_mount *mp,
> > > > + xfs_agnumber_t agno,
> > > > + struct bt_rebuild *btr,
> > > > + uint32_t nr_blocks)
> > > > +{
> > > > + struct extent_tree_node *ext_ptr;
> > > > + uint32_t blocks_allocated = 0;
> > > > + uint32_t len;
> > > > + int error;
> > > > +
> > > > + while (blocks_allocated < nr_blocks) {
> > > > + /*
> > > > + * Grab the smallest extent and use it up, then get the
> > > > + * next smallest. This mimics the init_*_cursor code.
> > > > + */
> > > > + ext_ptr = findfirst_bcnt_extent(agno);
> > >
> > > Extra whitespace ^
> > >
> > > > + if (!ext_ptr)
> > > > + do_error(
> > > > +_("error - not enough free space in filesystem\n"));
> > > > +
> > > > + /* Use up the extent we've got. */
> > > > + len = min(ext_ptr->ex_blockcount, nr_blocks - blocks_allocated);
> > > > + error = xrep_newbt_add_blocks(&btr->newbt,
> > > > + XFS_AGB_TO_FSB(mp, agno,
> > > > + ext_ptr->ex_startblock),
> > > > + len);
> > >
> > > Alignment.
> >
> > Will fix both of these.
> >
> > > > + if (error)
> > > > + do_error(_("could not set up btree reservation: %s\n"),
> > > > + strerror(-error));
> > > > +
> > > > + error = rmap_add_ag_rec(mp, agno, ext_ptr->ex_startblock, len,
> > > > + btr->newbt.oinfo.oi_owner);
> > > > + if (error)
> > > > + do_error(_("could not set up btree rmaps: %s\n"),
> > > > + strerror(-error));
> > > > +
> > > > + consume_freespace(agno, ext_ptr, len);
> > > > + blocks_allocated += len;
> > > > + }
> > > > +#ifdef XR_BLD_FREE_TRACE
> > > > + fprintf(stderr, "blocks_allocated = %d\n",
> > > > + blocks_allocated);
> > > > +#endif
> > > > +}
> > > > +
> > > > +/* Feed one of the new btree blocks to the bulk loader. */
> > > > +static int
> > > > +rebuild_claim_block(
> > > > + struct xfs_btree_cur *cur,
> > > > + union xfs_btree_ptr *ptr,
> > > > + void *priv)
> > > > +{
> > > > + struct bt_rebuild *btr = priv;
> > > > +
> > > > + return xrep_newbt_claim_block(cur, &btr->newbt, ptr);
> > > > +}
> > > > +
> > >
> > > Seems like an unnecessary helper, unless this grows more code in later
> > > patches..?
> >
> > It doesn't grow any more code, but keep in mind that get_record,
> > claim_block, and iroot_size are all callbacks of xfs_btree_bload(). The
> > priv parameter passed to that function are passed unchanged to the three
> > callbacks. The bulk load code doesn't know anything about where the
> > blocks or the records come from, so this is how both repairs will pass
> > that information to the callbacks.
> >
>
> Ok.
>
> > > > static void
> > > > write_cursor(bt_status_t *curs)
> > > > {
> > > ...
> > > > @@ -2287,28 +2483,29 @@ keep_fsinos(xfs_mount_t *mp)
> > > >
> > > > static void
> > > > phase5_func(
> > > > - xfs_mount_t *mp,
> > > > - xfs_agnumber_t agno,
> > > > - struct xfs_slab *lost_fsb)
> > > > + struct xfs_mount *mp,
> > > > + xfs_agnumber_t agno,
> > > > + struct xfs_slab *lost_fsb)
> > > > {
> > > > - uint64_t num_inos;
> > > > - uint64_t num_free_inos;
> > > > - uint64_t finobt_num_inos;
> > > > - uint64_t finobt_num_free_inos;
> > > > - bt_status_t bno_btree_curs;
> > > > - bt_status_t bcnt_btree_curs;
> > > > - bt_status_t ino_btree_curs;
> > > > - bt_status_t fino_btree_curs;
> > > > - bt_status_t rmap_btree_curs;
> > > > - bt_status_t refcnt_btree_curs;
> > > > - int extra_blocks = 0;
> > > > - uint num_freeblocks;
> > > > - xfs_extlen_t freeblks1;
> > > > + struct repair_ctx sc = { .mp = mp, };
> > >
> > > I don't see any reason to add sc here when it's still unused. It's not
> > > as if a single variable is saving complexity somewhere else. I guess
> > > I'll defer to Eric on the approach wrt to the other unused warnings.
> >
> > <shrug> I'll ask. It seems dumb to have a prep patch that adds a bunch
> > of symbols that won't get used until the next patch, but OTOH combining
> > the two will make for a ~40K patch.
> >
>
> I've no strong preference either way in general (the single variable
> thing aside) as long as each patch compiles and functions correctly and
> warnings are addressed by the end of the series. I do think that if we
> keep separate patches, it should probably be documented in the commit
> log that unused infrastructure is introduced (i.e. warnings expected)
> and users are introduced in a following patch. It's usually easier to
> squash patches than separate, so the maintainer can always squash them
> post review if he wanted to eliminate the warnings from the commit
> history.
Ok, will do.
--D
> Brian
>
> > > Also, what's the purpose of the rmap change below? I'm wondering if that
> > > (along with all of the indentation cleanup) should be its own patch with
> > > appropriate explanation.
> >
> > Errk, that one definitely should be separate.
> >
> > > Brian
> > >
> > > > + struct agi_stat agi_stat = {0,};
> > > > + uint64_t num_inos;
> > > > + uint64_t num_free_inos;
> > > > + uint64_t finobt_num_inos;
> > > > + uint64_t finobt_num_free_inos;
> > > > + bt_status_t bno_btree_curs;
> > > > + bt_status_t bcnt_btree_curs;
> > > > + bt_status_t ino_btree_curs;
> > > > + bt_status_t fino_btree_curs;
> > > > + bt_status_t rmap_btree_curs;
> > > > + bt_status_t refcnt_btree_curs;
> > > > + int extra_blocks = 0;
> > > > + uint num_freeblocks;
> > > > + xfs_extlen_t freeblks1;
> > > > #ifdef DEBUG
> > > > - xfs_extlen_t freeblks2;
> > > > + xfs_extlen_t freeblks2;
> > > > #endif
> > > > - xfs_agblock_t num_extents;
> > > > - struct agi_stat agi_stat = {0,};
> > > > + xfs_agblock_t num_extents;
> > > >
> > > > if (verbose)
> > > > do_log(_(" - agno = %d\n"), agno);
> > > > @@ -2516,8 +2713,8 @@ inject_lost_blocks(
> > > > if (error)
> > > > goto out_cancel;
> > > >
> > > > - error = -libxfs_free_extent(tp, *fsb, 1, &XFS_RMAP_OINFO_AG,
> > > > - XFS_AG_RESV_NONE);
> > > > + error = -libxfs_free_extent(tp, *fsb, 1,
> > > > + &XFS_RMAP_OINFO_ANY_OWNER, XFS_AG_RESV_NONE);
> > > > if (error)
> > > > goto out_cancel;
> > > >
> > > >
> > >
> >
>
^ permalink raw reply [flat|nested] 30+ messages in thread
* Re: [PATCH 4/9] xfs_repair: rebuild free space btrees with bulk loader
2020-05-28 15:10 ` Brian Foster
@ 2020-05-29 21:39 ` Darrick J. Wong
2020-06-01 12:05 ` Brian Foster
0 siblings, 1 reply; 30+ messages in thread
From: Darrick J. Wong @ 2020-05-29 21:39 UTC (permalink / raw)
To: Brian Foster; +Cc: sandeen, linux-xfs
On Thu, May 28, 2020 at 11:10:44AM -0400, Brian Foster wrote:
> On Tue, May 19, 2020 at 06:51:08PM -0700, Darrick J. Wong wrote:
> > From: Darrick J. Wong <darrick.wong@oracle.com>
> >
> > Use the btree bulk loading functions to rebuild the free space btrees
> > and drop the open-coded implementation.
> >
> > Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
> > ---
> > libxfs/libxfs_api_defs.h | 3
> > repair/phase5.c | 870 +++++++++++++---------------------------------
> > 2 files changed, 254 insertions(+), 619 deletions(-)
> >
> >
> ...
> > diff --git a/repair/phase5.c b/repair/phase5.c
> > index 8f5e5f59..e69b042c 100644
> > --- a/repair/phase5.c
> > +++ b/repair/phase5.c
> ...
> > @@ -837,268 +567,202 @@ btnum_to_ops(
> ...
> > +static void
> > +init_freespace_cursors(
> > + struct repair_ctx *sc,
> > + xfs_agnumber_t agno,
> > + unsigned int free_space,
> > + unsigned int *nr_extents,
> > + int *extra_blocks,
> > + struct bt_rebuild *btr_bno,
> > + struct bt_rebuild *btr_cnt)
> > {
> > - xfs_agnumber_t i;
> > - xfs_agblock_t j;
> > - struct xfs_btree_block *bt_hdr;
> > - xfs_alloc_rec_t *bt_rec;
> > - int level;
> > - xfs_agblock_t agbno;
> > - extent_tree_node_t *ext_ptr;
> > - bt_stat_level_t *lptr;
> > - xfs_extlen_t freeblks;
> > - const struct xfs_buf_ops *ops = btnum_to_ops(btnum);
> > + unsigned int bno_blocks;
> > + unsigned int cnt_blocks;
> > int error;
> >
> > - ASSERT(btnum == XFS_BTNUM_BNO || btnum == XFS_BTNUM_CNT);
> > -
> > -#ifdef XR_BLD_FREE_TRACE
> > - fprintf(stderr, "in build_freespace_tree, agno = %d\n", agno);
> > -#endif
> > - level = btree_curs->num_levels;
> > - freeblks = 0;
> > + init_rebuild(sc, &XFS_RMAP_OINFO_AG, free_space, btr_bno);
> > + init_rebuild(sc, &XFS_RMAP_OINFO_AG, free_space, btr_cnt);
> >
> > - ASSERT(level > 0);
> > + btr_bno->cur = libxfs_allocbt_stage_cursor(sc->mp,
> > + &btr_bno->newbt.afake, agno, XFS_BTNUM_BNO);
> > + btr_cnt->cur = libxfs_allocbt_stage_cursor(sc->mp,
> > + &btr_cnt->newbt.afake, agno, XFS_BTNUM_CNT);
> >
> > /*
> > - * initialize the first block on each btree level
> > + * Now we need to allocate blocks for the free space btrees using the
> > + * free space records we're about to put in them. Every record we use
> > + * can change the shape of the free space trees, so we recompute the
> > + * btree shape until we stop needing /more/ blocks. If we have any
> > + * left over we'll stash them in the AGFL when we're done.
> > */
> > - for (i = 0; i < level; i++) {
> > - lptr = &btree_curs->level[i];
> > + do {
> > + unsigned int num_freeblocks;
> > +
> > + bno_blocks = btr_bno->bload.nr_blocks;
> > + cnt_blocks = btr_cnt->bload.nr_blocks;
> >
> > - agbno = get_next_blockaddr(agno, i, btree_curs);
> > - error = -libxfs_buf_get(mp->m_dev,
> > - XFS_AGB_TO_DADDR(mp, agno, agbno),
> > - XFS_FSB_TO_BB(mp, 1), &lptr->buf_p);
> > + /* Compute how many bnobt blocks we'll need. */
> > + error = -libxfs_btree_bload_compute_geometry(btr_bno->cur,
> > + &btr_bno->bload, *nr_extents);
> > if (error)
> > do_error(
> ...
> > +_("Unable to compute free space by block btree geometry, error %d.\n"), -error);
> > +
> > + /* Compute how many cntbt blocks we'll need. */
> > + error = -libxfs_btree_bload_compute_geometry(btr_bno->cur,
>
> btr_cnt? ^
Ooops, yes. Fixed.
> > + &btr_cnt->bload, *nr_extents);
> > + if (error)
> > + do_error(
> > +_("Unable to compute free space by length btree geometry, error %d.\n"), -error);
> > +
> > + /* We don't need any more blocks, so we're done. */
> > + if (bno_blocks >= btr_bno->bload.nr_blocks &&
> > + cnt_blocks >= btr_cnt->bload.nr_blocks)
> > + break;
> > +
> > + /* Allocate however many more blocks we need this time. */
> > + if (bno_blocks < btr_bno->bload.nr_blocks)
> > + setup_rebuild(sc->mp, agno, btr_bno,
> > + btr_bno->bload.nr_blocks - bno_blocks);
> > + if (cnt_blocks < btr_cnt->bload.nr_blocks)
> > + setup_rebuild(sc->mp, agno, btr_cnt,
> > + btr_cnt->bload.nr_blocks - cnt_blocks);
>
> Hmm, I find the setup_rebuild() name a little confusing when seeing an
> init_rebuild() used in the same function. The difference is not apparent
> at a glance. Could we rename setup_rebuild() to something more specific?
> reserve_blocks() perhaps?
Ok, renamed.
> > +
> > + /* Ok, now how many free space records do we have? */
> > + *nr_extents = count_bno_extents_blocks(agno, &num_freeblocks);
> > + } while (1);
> > +
> > + *extra_blocks = (bno_blocks - btr_bno->bload.nr_blocks) +
> > + (cnt_blocks - btr_cnt->bload.nr_blocks);
> > +}
> > +
> > +static void
> > +get_freesp_data(
> > + struct xfs_btree_cur *cur,
> > + struct extent_tree_node *bno_rec,
> > + xfs_agblock_t *freeblks)
> > +{
> > + struct xfs_alloc_rec_incore *arec = &cur->bc_rec.a;
> > +
> > + arec->ar_startblock = bno_rec->ex_startblock;
> > + arec->ar_blockcount = bno_rec->ex_blockcount;
> > + if (freeblks)
> > + *freeblks += bno_rec->ex_blockcount;
> > +}
> > +
> > +/* Grab one bnobt record. */
> > +static int
> > +get_bnobt_record(
> > + struct xfs_btree_cur *cur,
> > + void *priv)
> > +{
> > + struct bt_rebuild *btr = priv;
> > +
> > + get_freesp_data(cur, btr->bno_rec, btr->freeblks);
> > + btr->bno_rec = findnext_bno_extent(btr->bno_rec);
>
> We should probably check for NULL here even if we don't expect it to
> happen.
>
> Also, this logic where we load the current pointer and get the next had
> my eyes crossed for a bit. Can we just check the pointer state here to
> determine whether to find the first record or the next?
Hm, I did that, and the logic ends up more cluttered looking:
if (cur->bc_btnum == XFS_BTNUM_BNO) {
if (btr->bno_rec == NULL)
btr->bno_rec = findfirst_bno_extent(agno);
else
btr->bno_rec = findnext_bno_extent(btr->bno_rec);
} else {
if (btr->bno_rec == NULL)
btr->bno_rec = findfirst_bcnt_extent(agno);
else
btr->bno_rec = findnext_bcnt_extent(cur->bc_ag.agno,
btr->bno_rec);
}
OTOH I think you do have a good point that splitting the code that sets
bno_rec across three functions is not easy to understand.
> Finally, I notice we duplicate the build_[bno|cnt]bt() functions for
> what amounts to a different get_record() callback specific to the tree
> type. If we also generalize get_record() to use the tree type (it
> already has the cursor), it seems that much of the duplication can be
> reduced and the logic simplified.
Ok, done.
> > + return 0;
> > +}
> > +
> > +/* Rebuild a free space by block number btree. */
> > +static void
> > +build_bnobt(
> > + struct repair_ctx *sc,
> > + xfs_agnumber_t agno,
> > + struct bt_rebuild *btr_bno,
> > + xfs_agblock_t *freeblks)
> > +{
> > + int error;
> > +
> > + *freeblks = 0;
> > + btr_bno->bload.get_record = get_bnobt_record;
> > + btr_bno->bload.claim_block = rebuild_claim_block;
>
> Note that one of these callbacks is introduced in this patch and the
> other in the previous. I've no issue with splitting the patches as noted
> in the previous patch, but these should probably be consistent one way
> or the other.
All the btree types set ->claim_block to rebuild_claim_block, whereas
->get_record is specific to the particular type of btree.
> > + btr_bno->bno_rec = findfirst_bno_extent(agno);
> > + btr_bno->freeblks = freeblks;
> > +
> > + error = -libxfs_trans_alloc_empty(sc->mp, &sc->tp);
> > + if (error)
> > + do_error(
> > +_("Insufficient memory to construct bnobt rebuild transaction.\n"));
> > +
> > + /* Add all observed bnobt records. */
> > + error = -libxfs_btree_bload(btr_bno->cur, &btr_bno->bload, btr_bno);
> > + if (error)
> > + do_error(
> > +_("Error %d while creating bnobt btree for AG %u.\n"), error, agno);
> > +
> > + /* Since we're not writing the AGF yet, no need to commit the cursor */
> > + libxfs_btree_del_cursor(btr_bno->cur, 0);
> > + error = -libxfs_trans_commit(sc->tp);
> > + if (error)
> > + do_error(
> > +_("Error %d while writing bnobt btree for AG %u.\n"), error, agno);
> > + sc->tp = NULL;
> > +}
> > +
> ...
> > @@ -2354,48 +2041,14 @@ build_agf_agfl(
> > freelist[i] = cpu_to_be32(NULLAGBLOCK);
> > }
> >
> > - /*
> > - * do we have left-over blocks in the btree cursors that should
> > - * be used to fill the AGFL?
> > - */
> > - if (bno_bt->num_free_blocks > 0 || bcnt_bt->num_free_blocks > 0) {
> > - /*
> > - * yes, now grab as many blocks as we can
> > - */
> > - i = 0;
> > - while (bno_bt->num_free_blocks > 0 && i < libxfs_agfl_size(mp))
> > - {
> > - freelist[i] = cpu_to_be32(
> > - get_next_blockaddr(agno, 0, bno_bt));
> > - i++;
> > - }
> > -
> > - while (bcnt_bt->num_free_blocks > 0 && i < libxfs_agfl_size(mp))
> > - {
> > - freelist[i] = cpu_to_be32(
> > - get_next_blockaddr(agno, 0, bcnt_bt));
> > - i++;
> > - }
> > - /*
> > - * now throw the rest of the blocks away and complain
> > - */
> > - while (bno_bt->num_free_blocks > 0) {
> > - fsb = XFS_AGB_TO_FSB(mp, agno,
> > - get_next_blockaddr(agno, 0, bno_bt));
> > - error = slab_add(lost_fsb, &fsb);
> > - if (error)
> > - do_error(
> > -_("Insufficient memory saving lost blocks.\n"));
> > - }
> > - while (bcnt_bt->num_free_blocks > 0) {
> > - fsb = XFS_AGB_TO_FSB(mp, agno,
> > - get_next_blockaddr(agno, 0, bcnt_bt));
> > - error = slab_add(lost_fsb, &fsb);
> > - if (error)
> > - do_error(
> > -_("Insufficient memory saving lost blocks.\n"));
> > - }
> > + /* Fill the AGFL with leftover blocks or save them for later. */
> > + i = 0;
>
> How about agfl_idx or something since this variable is passed down into
> a function?
The 'i' variable existed before this patch. I can add a new cleanup to
fix the name...
> > + freelist = xfs_buf_to_agfl_bno(agfl_buf);
> > + fill_agfl(btr_bno, freelist, &i);
> > + fill_agfl(btr_cnt, freelist, &i);
> >
> > + /* Set the AGF counters for the AGFL. */
> > + if (i > 0) {
> > agf->agf_flfirst = 0;
> > agf->agf_fllast = cpu_to_be32(i - 1);
> > agf->agf_flcount = cpu_to_be32(i);
> ...
> > @@ -2650,9 +2283,9 @@ _("unable to rebuild AG %u. Not enough free space in on-disk AG.\n"),
> > /*
> > * set up agf and agfl
> > */
> > - build_agf_agfl(mp, agno, &bno_btree_curs,
> > - &bcnt_btree_curs, freeblks1, extra_blocks,
> > - &rmap_btree_curs, &refcnt_btree_curs, lost_fsb);
> > + build_agf_agfl(mp, agno, &btr_bno, &btr_cnt, freeblks1, extra_blocks,
> > + &rmap_btree_curs, &refcnt_btree_curs, lost_fsb);
> > +
>
> I was trying to figure out what extra_blocks was used for and noticed
> that it's not even used by this function. Perhaps a preceding cleanup
> patch to drop it?
Ok.
> > /*
> > * build inode allocation tree.
> > */
> > @@ -2674,15 +2307,14 @@ _("unable to rebuild AG %u. Not enough free space in on-disk AG.\n"),
> > /*
> > * tear down cursors
> > */
> > - finish_cursor(&bno_btree_curs);
> > - finish_cursor(&ino_btree_curs);
>
> Looks like this one shouldn't be going away here.
Oops, good catch.
--D
> Brian
>
> > + finish_rebuild(mp, &btr_bno, lost_fsb);
> > + finish_rebuild(mp, &btr_cnt, lost_fsb);
> > if (xfs_sb_version_hasrmapbt(&mp->m_sb))
> > finish_cursor(&rmap_btree_curs);
> > if (xfs_sb_version_hasreflink(&mp->m_sb))
> > finish_cursor(&refcnt_btree_curs);
> > if (xfs_sb_version_hasfinobt(&mp->m_sb))
> > finish_cursor(&fino_btree_curs);
> > - finish_cursor(&bcnt_btree_curs);
> >
> > /*
> > * release the incore per-AG bno/bcnt trees so the extent nodes
> >
>
^ permalink raw reply [flat|nested] 30+ messages in thread
* Re: [PATCH 5/9] xfs_repair: rebuild inode btrees with bulk loader
2020-05-28 15:11 ` Brian Foster
@ 2020-05-29 22:18 ` Darrick J. Wong
2020-05-29 22:32 ` Darrick J. Wong
0 siblings, 1 reply; 30+ messages in thread
From: Darrick J. Wong @ 2020-05-29 22:18 UTC (permalink / raw)
To: Brian Foster; +Cc: sandeen, linux-xfs
On Thu, May 28, 2020 at 11:11:21AM -0400, Brian Foster wrote:
> On Tue, May 19, 2020 at 06:51:15PM -0700, Darrick J. Wong wrote:
> > From: Darrick J. Wong <darrick.wong@oracle.com>
> >
> > Use the btree bulk loading functions to rebuild the inode btrees
> > and drop the open-coded implementation.
> >
> > Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
> > ---
> > libxfs/libxfs_api_defs.h | 1
> > repair/phase5.c | 642 +++++++++++++++++-----------------------------
> > 2 files changed, 240 insertions(+), 403 deletions(-)
> >
> >
> ...
> > diff --git a/repair/phase5.c b/repair/phase5.c
> > index e69b042c..38f30753 100644
> > --- a/repair/phase5.c
> > +++ b/repair/phase5.c
> ...
> > @@ -372,6 +376,11 @@ estimate_ag_bload_slack(
> > bload->node_slack = 0;
> > }
> >
> > +static inline void skip_rebuild(struct bt_rebuild *btr)
> > +{
> > + memset(btr, 0, sizeof(struct bt_rebuild));
> > +}
> > +
>
> Is there any functional purpose to this? It looks like the memset could
> be open-coded, but also seems like it could be elided if we just check
> hasfinobt() before using the pointer..?
Hm, yeah, ok, I'll go audit this.
> > /* Initialize a btree rebuild context. */
> > static void
> > init_rebuild(
> > @@ -765,48 +774,38 @@ _("Error %d while writing cntbt btree for AG %u.\n"), error, agno);
> ...
> > +
> > +/* Copy one incore inode record into the inobt cursor. */
> > +static void
> > +get_inode_data(
> > + struct xfs_btree_cur *cur,
> > + struct ino_tree_node *ino_rec,
> > + struct agi_stat *agi_stat)
> > +{
> > + struct xfs_inobt_rec_incore *irec = &cur->bc_rec.i;
> > + int inocnt = 0;
> > + int finocnt = 0;
> > + int k;
> > +
> > + irec->ir_startino = ino_rec->ino_startnum;
> > + irec->ir_free = ino_rec->ir_free;
> > +
> > + for (k = 0; k < sizeof(xfs_inofree_t) * NBBY; k++) {
> > + ASSERT(is_inode_confirmed(ino_rec, k));
> > +
> > + if (is_inode_sparse(ino_rec, k))
> > continue;
> > -
> > - nfinos += rec_nfinos;
> > - ninos += rec_ninos;
> > - num_recs++;
> > + if (is_inode_free(ino_rec, k))
> > + finocnt++;
> > + inocnt++;
> > }
> >
> > - if (num_recs == 0) {
> > - /*
> > - * easy corner-case -- no inode records
> > - */
> > - lptr->num_blocks = 1;
> > - lptr->modulo = 0;
> > - lptr->num_recs_pb = 0;
> > - lptr->num_recs_tot = 0;
> > -
> > - btree_curs->num_levels = 1;
> > - btree_curs->num_tot_blocks = btree_curs->num_free_blocks = 1;
> > -
> > - setup_cursor(mp, agno, btree_curs);
> > + irec->ir_count = inocnt;
> > + irec->ir_freecount = finocnt;
> >
> > - return;
> > - }
> > + if (xfs_sb_version_hassparseinodes(&cur->bc_mp->m_sb)) {
> > + uint64_t sparse;
> > + int spmask;
> > + uint16_t holemask;
> >
> > - blocks_allocated = lptr->num_blocks = howmany(num_recs,
> > - XR_INOBT_BLOCK_MAXRECS(mp, 0));
> > -
> > - lptr->modulo = num_recs % lptr->num_blocks;
> > - lptr->num_recs_pb = num_recs / lptr->num_blocks;
> > - lptr->num_recs_tot = num_recs;
> > - level = 1;
> > -
> > - if (lptr->num_blocks > 1) {
> > - for (; btree_curs->level[level-1].num_blocks > 1
> > - && level < XFS_BTREE_MAXLEVELS;
> > - level++) {
> > - lptr = &btree_curs->level[level];
> > - p_lptr = &btree_curs->level[level - 1];
> > - lptr->num_blocks = howmany(p_lptr->num_blocks,
> > - XR_INOBT_BLOCK_MAXRECS(mp, level));
> > - lptr->modulo = p_lptr->num_blocks % lptr->num_blocks;
> > - lptr->num_recs_pb = p_lptr->num_blocks
> > - / lptr->num_blocks;
> > - lptr->num_recs_tot = p_lptr->num_blocks;
> > -
> > - blocks_allocated += lptr->num_blocks;
> > + /*
> > + * Convert the 64-bit in-core sparse inode state to the
> > + * 16-bit on-disk holemask.
> > + */
> > + holemask = 0;
> > + spmask = (1 << XFS_INODES_PER_HOLEMASK_BIT) - 1;
> > + sparse = ino_rec->ir_sparse;
> > + for (k = 0; k < XFS_INOBT_HOLEMASK_BITS; k++) {
> > + if (sparse & spmask) {
> > + ASSERT((sparse & spmask) == spmask);
> > + holemask |= (1 << k);
> > + } else
> > + ASSERT((sparse & spmask) == 0);
> > + sparse >>= XFS_INODES_PER_HOLEMASK_BIT;
> > }
> > +
> > + irec->ir_holemask = holemask;
> > + } else {
> > + irec->ir_holemask = 0;
> > }
> > - ASSERT(lptr->num_blocks == 1);
> > - btree_curs->num_levels = level;
> >
> > - btree_curs->num_tot_blocks = btree_curs->num_free_blocks
> > - = blocks_allocated;
> > + if (!agi_stat)
> > + return;
> >
> > - setup_cursor(mp, agno, btree_curs);
> > + if (agi_stat->first_agino != NULLAGINO)
> > + agi_stat->first_agino = ino_rec->ino_startnum;
>
> This is initialized to NULLAGINO. When do we ever update it?
I don't understand your question. The purpose of the first_agino code
is to set agi_newino to the first inode cluster in the AG (or NULLAGINO
if there are no inodes) so we initialize first_ino to NULLAGINO and if
we process any inode records, we'll update it to ir_startino of the
first inode record that we put in the btree.
> > + agi_stat->freecount += finocnt;
> > + agi_stat->count += inocnt;
> > +}
> >
> > - *num_inos = ninos;
> > - *num_free_inos = nfinos;
> > +/* Grab one inobt record. */
> > +static int
> > +get_inobt_record(
> > + struct xfs_btree_cur *cur,
> > + void *priv)
> > +{
> > + struct bt_rebuild *rebuild = priv;
> >
> > - return;
> > + get_inode_data(cur, rebuild->ino_rec, rebuild->agi_stat);
> > + rebuild->ino_rec = next_ino_rec(rebuild->ino_rec);
> > + return 0;
> > }
> >
> > +/* Rebuild a inobt btree. */
> > static void
> > -prop_ino_cursor(xfs_mount_t *mp, xfs_agnumber_t agno, bt_status_t *btree_curs,
> > - xfs_btnum_t btnum, xfs_agino_t startino, int level)
> > +build_inobt(
> > + struct repair_ctx *sc,
> > + xfs_agnumber_t agno,
> > + struct bt_rebuild *btr_ino,
> > + struct agi_stat *agi_stat)
> > {
> > - struct xfs_btree_block *bt_hdr;
> > - xfs_inobt_key_t *bt_key;
> > - xfs_inobt_ptr_t *bt_ptr;
> > - xfs_agblock_t agbno;
> > - bt_stat_level_t *lptr;
> > - const struct xfs_buf_ops *ops = btnum_to_ops(btnum);
> > int error;
> >
> > - level++;
> > -
> > - if (level >= btree_curs->num_levels)
> > - return;
> > -
> > - lptr = &btree_curs->level[level];
> > - bt_hdr = XFS_BUF_TO_BLOCK(lptr->buf_p);
> > -
> > - if (be16_to_cpu(bt_hdr->bb_numrecs) == 0) {
> > - /*
> > - * this only happens once to initialize the
> > - * first path up the left side of the tree
> > - * where the agbno's are already set up
> > - */
> > - prop_ino_cursor(mp, agno, btree_curs, btnum, startino, level);
> > - }
> > -
> > - if (be16_to_cpu(bt_hdr->bb_numrecs) ==
> > - lptr->num_recs_pb + (lptr->modulo > 0)) {
> > - /*
> > - * write out current prev block, grab us a new block,
> > - * and set the rightsib pointer of current block
> > - */
> > -#ifdef XR_BLD_INO_TRACE
> > - fprintf(stderr, " ino prop agbno %d ", lptr->prev_agbno);
> > -#endif
> > - if (lptr->prev_agbno != NULLAGBLOCK) {
> > - ASSERT(lptr->prev_buf_p != NULL);
> > - libxfs_buf_mark_dirty(lptr->prev_buf_p);
> > - libxfs_buf_relse(lptr->prev_buf_p);
> > - }
> > - lptr->prev_agbno = lptr->agbno;;
> > - lptr->prev_buf_p = lptr->buf_p;
> > - agbno = get_next_blockaddr(agno, level, btree_curs);
> > -
> > - bt_hdr->bb_u.s.bb_rightsib = cpu_to_be32(agbno);
> > -
> > - error = -libxfs_buf_get(mp->m_dev,
> > - XFS_AGB_TO_DADDR(mp, agno, agbno),
> > - XFS_FSB_TO_BB(mp, 1), &lptr->buf_p);
> > - if (error)
> > - do_error(_("Cannot grab inode btree buffer, err=%d"),
> > - error);
> > - lptr->agbno = agbno;
> > + btr_ino->bload.get_record = get_inobt_record;
> > + btr_ino->bload.claim_block = rebuild_claim_block;
> > + agi_stat->count = agi_stat->freecount = 0;
>
> These are already initialized to zero by the caller. I suppose we might
> as well also move the ->first_agino init to where this is allocated.
TBH we don't need to (re)init agi_stat at all. I'll drop these two
lines.
> Otherwise I mostly just have the same general feedback as for the
> previous patch (wrt to the get_record() logic and build_[f]inobt()
> duplication.
<nod> Will fix that too.
> Brian
>
> > + agi_stat->first_agino = NULLAGINO;
> > + btr_ino->agi_stat = agi_stat;
> > + btr_ino->ino_rec = findfirst_inode_rec(agno);
> > +
> > + error = -libxfs_trans_alloc_empty(sc->mp, &sc->tp);
> > + if (error)
> > + do_error(
> > +_("Insufficient memory to construct inobt rebuild transaction.\n"));
> > +
> > + /* Add all observed inobt records. */
> > + error = -libxfs_btree_bload(btr_ino->cur, &btr_ino->bload, btr_ino);
> > + if (error)
> > + do_error(
> > +_("Error %d while creating inobt btree for AG %u.\n"), error, agno);
> > +
> > + /* Since we're not writing the AGI yet, no need to commit the cursor */
> > + libxfs_btree_del_cursor(btr_ino->cur, 0);
> > + error = -libxfs_trans_commit(sc->tp);
> > + if (error)
> > + do_error(
> > +_("Error %d while writing inobt btree for AG %u.\n"), error, agno);
> > + sc->tp = NULL;
> > +}
> >
> > - if (lptr->modulo)
> > - lptr->modulo--;
> > +/* Grab one finobt record. */
> > +static int
> > +get_finobt_record(
> > + struct xfs_btree_cur *cur,
> > + void *priv)
> > +{
> > + struct bt_rebuild *rebuild = priv;
> >
> > - /*
> > - * initialize block header
> > - */
> > - lptr->buf_p->b_ops = ops;
> > - bt_hdr = XFS_BUF_TO_BLOCK(lptr->buf_p);
> > - memset(bt_hdr, 0, mp->m_sb.sb_blocksize);
> > - libxfs_btree_init_block(mp, lptr->buf_p, btnum,
> > - level, 0, agno);
> > + get_inode_data(cur, rebuild->ino_rec, NULL);
> > + rebuild->ino_rec = next_free_ino_rec(rebuild->ino_rec);
> > + return 0;
> > +}
> >
> > - bt_hdr->bb_u.s.bb_leftsib = cpu_to_be32(lptr->prev_agbno);
> > +/* Rebuild a finobt btree. */
> > +static void
> > +build_finobt(
> > + struct repair_ctx *sc,
> > + xfs_agnumber_t agno,
> > + struct bt_rebuild *btr_fino)
> > +{
> > + int error;
> >
> > - /*
> > - * propagate extent record for first extent in new block up
> > - */
> > - prop_ino_cursor(mp, agno, btree_curs, btnum, startino, level);
> > - }
> > - /*
> > - * add inode info to current block
> > - */
> > - be16_add_cpu(&bt_hdr->bb_numrecs, 1);
> > -
> > - bt_key = XFS_INOBT_KEY_ADDR(mp, bt_hdr,
> > - be16_to_cpu(bt_hdr->bb_numrecs));
> > - bt_ptr = XFS_INOBT_PTR_ADDR(mp, bt_hdr,
> > - be16_to_cpu(bt_hdr->bb_numrecs),
> > - M_IGEO(mp)->inobt_mxr[1]);
> > -
> > - bt_key->ir_startino = cpu_to_be32(startino);
> > - *bt_ptr = cpu_to_be32(btree_curs->level[level-1].agbno);
> > + btr_fino->bload.get_record = get_finobt_record;
> > + btr_fino->bload.claim_block = rebuild_claim_block;
> > + btr_fino->ino_rec = findfirst_free_inode_rec(agno);
> > +
> > + error = -libxfs_trans_alloc_empty(sc->mp, &sc->tp);
> > + if (error)
> > + do_error(
> > +_("Insufficient memory to construct finobt rebuild transaction.\n"));
> > +
> > + /* Add all observed finobt records. */
> > + error = -libxfs_btree_bload(btr_fino->cur, &btr_fino->bload, btr_fino);
> > + if (error)
> > + do_error(
> > +_("Error %d while creating finobt btree for AG %u.\n"), error, agno);
> > +
> > + /* Since we're not writing the AGI yet, no need to commit the cursor */
> > + libxfs_btree_del_cursor(btr_fino->cur, 0);
> > + error = -libxfs_trans_commit(sc->tp);
> > + if (error)
> > + do_error(
> > +_("Error %d while writing finobt btree for AG %u.\n"), error, agno);
> > + sc->tp = NULL;
> > }
> >
> > /*
> > * XXX: yet more code that can be shared with mkfs, growfs.
> > */
> > static void
> > -build_agi(xfs_mount_t *mp, xfs_agnumber_t agno, bt_status_t *btree_curs,
> > - bt_status_t *finobt_curs, struct agi_stat *agi_stat)
> > +build_agi(
> > + struct xfs_mount *mp,
> > + xfs_agnumber_t agno,
> > + struct bt_rebuild *btr_ino,
> > + struct bt_rebuild *btr_fino,
> > + struct agi_stat *agi_stat)
> > {
> > - xfs_buf_t *agi_buf;
> > - xfs_agi_t *agi;
> > - int i;
> > - int error;
> > + struct xfs_buf *agi_buf;
> > + struct xfs_agi *agi;
> > + int i;
> > + int error;
> >
> > error = -libxfs_buf_get(mp->m_dev,
> > XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
> > @@ -1008,8 +1053,8 @@ build_agi(xfs_mount_t *mp, xfs_agnumber_t agno, bt_status_t *btree_curs,
> > agi->agi_length = cpu_to_be32(mp->m_sb.sb_dblocks -
> > (xfs_rfsblock_t) mp->m_sb.sb_agblocks * agno);
> > agi->agi_count = cpu_to_be32(agi_stat->count);
> > - agi->agi_root = cpu_to_be32(btree_curs->root);
> > - agi->agi_level = cpu_to_be32(btree_curs->num_levels);
> > + agi->agi_root = cpu_to_be32(btr_ino->newbt.afake.af_root);
> > + agi->agi_level = cpu_to_be32(btr_ino->newbt.afake.af_levels);
> > agi->agi_freecount = cpu_to_be32(agi_stat->freecount);
> > agi->agi_newino = cpu_to_be32(agi_stat->first_agino);
> > agi->agi_dirino = cpu_to_be32(NULLAGINO);
> > @@ -1021,203 +1066,16 @@ build_agi(xfs_mount_t *mp, xfs_agnumber_t agno, bt_status_t *btree_curs,
> > platform_uuid_copy(&agi->agi_uuid, &mp->m_sb.sb_meta_uuid);
> >
> > if (xfs_sb_version_hasfinobt(&mp->m_sb)) {
> > - agi->agi_free_root = cpu_to_be32(finobt_curs->root);
> > - agi->agi_free_level = cpu_to_be32(finobt_curs->num_levels);
> > + agi->agi_free_root =
> > + cpu_to_be32(btr_fino->newbt.afake.af_root);
> > + agi->agi_free_level =
> > + cpu_to_be32(btr_fino->newbt.afake.af_levels);
> > }
> >
> > libxfs_buf_mark_dirty(agi_buf);
> > libxfs_buf_relse(agi_buf);
> > }
> >
> > -/*
> > - * rebuilds an inode tree given a cursor. We're lazy here and call
> > - * the routine that builds the agi
> > - */
> > -static void
> > -build_ino_tree(xfs_mount_t *mp, xfs_agnumber_t agno,
> > - bt_status_t *btree_curs, xfs_btnum_t btnum,
> > - struct agi_stat *agi_stat)
> > -{
> > - xfs_agnumber_t i;
> > - xfs_agblock_t j;
> > - xfs_agblock_t agbno;
> > - xfs_agino_t first_agino;
> > - struct xfs_btree_block *bt_hdr;
> > - xfs_inobt_rec_t *bt_rec;
> > - ino_tree_node_t *ino_rec;
> > - bt_stat_level_t *lptr;
> > - const struct xfs_buf_ops *ops = btnum_to_ops(btnum);
> > - xfs_agino_t count = 0;
> > - xfs_agino_t freecount = 0;
> > - int inocnt;
> > - uint8_t finocnt;
> > - int k;
> > - int level = btree_curs->num_levels;
> > - int spmask;
> > - uint64_t sparse;
> > - uint16_t holemask;
> > - int error;
> > -
> > - ASSERT(btnum == XFS_BTNUM_INO || btnum == XFS_BTNUM_FINO);
> > -
> > - for (i = 0; i < level; i++) {
> > - lptr = &btree_curs->level[i];
> > -
> > - agbno = get_next_blockaddr(agno, i, btree_curs);
> > - error = -libxfs_buf_get(mp->m_dev,
> > - XFS_AGB_TO_DADDR(mp, agno, agbno),
> > - XFS_FSB_TO_BB(mp, 1), &lptr->buf_p);
> > - if (error)
> > - do_error(_("Cannot grab inode btree buffer, err=%d"),
> > - error);
> > -
> > - if (i == btree_curs->num_levels - 1)
> > - btree_curs->root = agbno;
> > -
> > - lptr->agbno = agbno;
> > - lptr->prev_agbno = NULLAGBLOCK;
> > - lptr->prev_buf_p = NULL;
> > - /*
> > - * initialize block header
> > - */
> > -
> > - lptr->buf_p->b_ops = ops;
> > - bt_hdr = XFS_BUF_TO_BLOCK(lptr->buf_p);
> > - memset(bt_hdr, 0, mp->m_sb.sb_blocksize);
> > - libxfs_btree_init_block(mp, lptr->buf_p, btnum, i, 0, agno);
> > - }
> > -
> > - /*
> > - * run along leaf, setting up records. as we have to switch
> > - * blocks, call the prop_ino_cursor routine to set up the new
> > - * pointers for the parent. that can recurse up to the root
> > - * if required. set the sibling pointers for leaf level here.
> > - */
> > - if (btnum == XFS_BTNUM_FINO)
> > - ino_rec = findfirst_free_inode_rec(agno);
> > - else
> > - ino_rec = findfirst_inode_rec(agno);
> > -
> > - if (ino_rec != NULL)
> > - first_agino = ino_rec->ino_startnum;
> > - else
> > - first_agino = NULLAGINO;
> > -
> > - lptr = &btree_curs->level[0];
> > -
> > - for (i = 0; i < lptr->num_blocks; i++) {
> > - /*
> > - * block initialization, lay in block header
> > - */
> > - lptr->buf_p->b_ops = ops;
> > - bt_hdr = XFS_BUF_TO_BLOCK(lptr->buf_p);
> > - memset(bt_hdr, 0, mp->m_sb.sb_blocksize);
> > - libxfs_btree_init_block(mp, lptr->buf_p, btnum, 0, 0, agno);
> > -
> > - bt_hdr->bb_u.s.bb_leftsib = cpu_to_be32(lptr->prev_agbno);
> > - bt_hdr->bb_numrecs = cpu_to_be16(lptr->num_recs_pb +
> > - (lptr->modulo > 0));
> > -
> > - if (lptr->modulo > 0)
> > - lptr->modulo--;
> > -
> > - if (lptr->num_recs_pb > 0)
> > - prop_ino_cursor(mp, agno, btree_curs, btnum,
> > - ino_rec->ino_startnum, 0);
> > -
> > - bt_rec = (xfs_inobt_rec_t *)
> > - ((char *)bt_hdr + XFS_INOBT_BLOCK_LEN(mp));
> > - for (j = 0; j < be16_to_cpu(bt_hdr->bb_numrecs); j++) {
> > - ASSERT(ino_rec != NULL);
> > - bt_rec[j].ir_startino =
> > - cpu_to_be32(ino_rec->ino_startnum);
> > - bt_rec[j].ir_free = cpu_to_be64(ino_rec->ir_free);
> > -
> > - inocnt = finocnt = 0;
> > - for (k = 0; k < sizeof(xfs_inofree_t)*NBBY; k++) {
> > - ASSERT(is_inode_confirmed(ino_rec, k));
> > -
> > - if (is_inode_sparse(ino_rec, k))
> > - continue;
> > - if (is_inode_free(ino_rec, k))
> > - finocnt++;
> > - inocnt++;
> > - }
> > -
> > - /*
> > - * Set the freecount and check whether we need to update
> > - * the sparse format fields. Otherwise, skip to the next
> > - * record.
> > - */
> > - inorec_set_freecount(mp, &bt_rec[j], finocnt);
> > - if (!xfs_sb_version_hassparseinodes(&mp->m_sb))
> > - goto nextrec;
> > -
> > - /*
> > - * Convert the 64-bit in-core sparse inode state to the
> > - * 16-bit on-disk holemask.
> > - */
> > - holemask = 0;
> > - spmask = (1 << XFS_INODES_PER_HOLEMASK_BIT) - 1;
> > - sparse = ino_rec->ir_sparse;
> > - for (k = 0; k < XFS_INOBT_HOLEMASK_BITS; k++) {
> > - if (sparse & spmask) {
> > - ASSERT((sparse & spmask) == spmask);
> > - holemask |= (1 << k);
> > - } else
> > - ASSERT((sparse & spmask) == 0);
> > - sparse >>= XFS_INODES_PER_HOLEMASK_BIT;
> > - }
> > -
> > - bt_rec[j].ir_u.sp.ir_count = inocnt;
> > - bt_rec[j].ir_u.sp.ir_holemask = cpu_to_be16(holemask);
> > -
> > -nextrec:
> > - freecount += finocnt;
> > - count += inocnt;
> > -
> > - if (btnum == XFS_BTNUM_FINO)
> > - ino_rec = next_free_ino_rec(ino_rec);
> > - else
> > - ino_rec = next_ino_rec(ino_rec);
> > - }
> > -
> > - if (ino_rec != NULL) {
> > - /*
> > - * get next leaf level block
> > - */
> > - if (lptr->prev_buf_p != NULL) {
> > -#ifdef XR_BLD_INO_TRACE
> > - fprintf(stderr, "writing inobt agbno %u\n",
> > - lptr->prev_agbno);
> > -#endif
> > - ASSERT(lptr->prev_agbno != NULLAGBLOCK);
> > - libxfs_buf_mark_dirty(lptr->prev_buf_p);
> > - libxfs_buf_relse(lptr->prev_buf_p);
> > - }
> > - lptr->prev_buf_p = lptr->buf_p;
> > - lptr->prev_agbno = lptr->agbno;
> > - lptr->agbno = get_next_blockaddr(agno, 0, btree_curs);
> > - bt_hdr->bb_u.s.bb_rightsib = cpu_to_be32(lptr->agbno);
> > -
> > - error = -libxfs_buf_get(mp->m_dev,
> > - XFS_AGB_TO_DADDR(mp, agno, lptr->agbno),
> > - XFS_FSB_TO_BB(mp, 1),
> > - &lptr->buf_p);
> > - if (error)
> > - do_error(
> > - _("Cannot grab inode btree buffer, err=%d"),
> > - error);
> > - }
> > - }
> > -
> > - if (agi_stat) {
> > - agi_stat->first_agino = first_agino;
> > - agi_stat->count = count;
> > - agi_stat->freecount = freecount;
> > - }
> > -}
> > -
> > /* rebuild the rmap tree */
> >
> > /*
> > @@ -2142,14 +2000,10 @@ phase5_func(
> > {
> > struct repair_ctx sc = { .mp = mp, };
> > struct agi_stat agi_stat = {0,};
> > - uint64_t num_inos;
> > - uint64_t num_free_inos;
> > - uint64_t finobt_num_inos;
> > - uint64_t finobt_num_free_inos;
> > struct bt_rebuild btr_bno;
> > struct bt_rebuild btr_cnt;
> > - bt_status_t ino_btree_curs;
> > - bt_status_t fino_btree_curs;
> > + struct bt_rebuild btr_ino;
> > + struct bt_rebuild btr_fino;
> > bt_status_t rmap_btree_curs;
> > bt_status_t refcnt_btree_curs;
> > int extra_blocks = 0;
> > @@ -2184,19 +2038,8 @@ _("unable to rebuild AG %u. Not enough free space in on-disk AG.\n"),
> > agno);
> > }
> >
> > - /*
> > - * ok, now set up the btree cursors for the on-disk btrees (includes
> > - * pre-allocating all required blocks for the trees themselves)
> > - */
> > - init_ino_cursor(mp, agno, &ino_btree_curs, &num_inos,
> > - &num_free_inos, 0);
> > -
> > - if (xfs_sb_version_hasfinobt(&mp->m_sb))
> > - init_ino_cursor(mp, agno, &fino_btree_curs, &finobt_num_inos,
> > - &finobt_num_free_inos, 1);
> > -
> > - sb_icount_ag[agno] += num_inos;
> > - sb_ifree_ag[agno] += num_free_inos;
> > + init_ino_cursors(&sc, agno, num_freeblocks, &sb_icount_ag[agno],
> > + &sb_ifree_ag[agno], &btr_ino, &btr_fino);
> >
> > /*
> > * Set up the btree cursors for the on-disk rmap btrees, which includes
> > @@ -2287,34 +2130,27 @@ _("unable to rebuild AG %u. Not enough free space in on-disk AG.\n"),
> > &rmap_btree_curs, &refcnt_btree_curs, lost_fsb);
> >
> > /*
> > - * build inode allocation tree.
> > + * build inode allocation trees.
> > */
> > - build_ino_tree(mp, agno, &ino_btree_curs, XFS_BTNUM_INO, &agi_stat);
> > - write_cursor(&ino_btree_curs);
> > -
> > - /*
> > - * build free inode tree
> > - */
> > - if (xfs_sb_version_hasfinobt(&mp->m_sb)) {
> > - build_ino_tree(mp, agno, &fino_btree_curs,
> > - XFS_BTNUM_FINO, NULL);
> > - write_cursor(&fino_btree_curs);
> > - }
> > + build_inobt(&sc, agno, &btr_ino, &agi_stat);
> > + if (xfs_sb_version_hasfinobt(&mp->m_sb))
> > + build_finobt(&sc, agno, &btr_fino);
> >
> > /* build the agi */
> > - build_agi(mp, agno, &ino_btree_curs, &fino_btree_curs, &agi_stat);
> > + build_agi(mp, agno, &btr_ino, &btr_fino, &agi_stat);
> >
> > /*
> > * tear down cursors
> > */
> > finish_rebuild(mp, &btr_bno, lost_fsb);
> > finish_rebuild(mp, &btr_cnt, lost_fsb);
> > + finish_rebuild(mp, &btr_ino, lost_fsb);
> > + if (xfs_sb_version_hasfinobt(&mp->m_sb))
> > + finish_rebuild(mp, &btr_fino, lost_fsb);
> > if (xfs_sb_version_hasrmapbt(&mp->m_sb))
> > finish_cursor(&rmap_btree_curs);
> > if (xfs_sb_version_hasreflink(&mp->m_sb))
> > finish_cursor(&refcnt_btree_curs);
> > - if (xfs_sb_version_hasfinobt(&mp->m_sb))
> > - finish_cursor(&fino_btree_curs);
> >
> > /*
> > * release the incore per-AG bno/bcnt trees so the extent nodes
> >
>
^ permalink raw reply [flat|nested] 30+ messages in thread
* Re: [PATCH 9/9] xfs_repair: track blocks lost during btree construction via extents
2020-05-28 17:00 ` Brian Foster
@ 2020-05-29 22:19 ` Darrick J. Wong
0 siblings, 0 replies; 30+ messages in thread
From: Darrick J. Wong @ 2020-05-29 22:19 UTC (permalink / raw)
To: Brian Foster; +Cc: sandeen, linux-xfs
On Thu, May 28, 2020 at 01:00:12PM -0400, Brian Foster wrote:
> On Tue, May 19, 2020 at 06:51:40PM -0700, Darrick J. Wong wrote:
> > From: Darrick J. Wong <darrick.wong@oracle.com>
> >
> > Use extent records (not just raw fsbs) to track blocks that were lost
> > during btree construction. This makes it somewhat more efficient.
> >
> > Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
> > ---
> > repair/phase5.c | 61 ++++++++++++++++++++++++++++++++-----------------------
> > 1 file changed, 35 insertions(+), 26 deletions(-)
> >
> >
> > diff --git a/repair/phase5.c b/repair/phase5.c
> > index 72c6908a..22007275 100644
> > --- a/repair/phase5.c
> > +++ b/repair/phase5.c
> > @@ -45,6 +45,12 @@ struct bt_rebuild {
> > };
> > };
> >
> > +struct lost_fsb {
> > + xfs_fsblock_t fsbno;
> > + xfs_extlen_t len;
> > +};
> > +
>
> Looks reasonable at a glance, but could we call the above structure an
> extent and rename lost_fsbs to lost_extents or some such?
Ok. I suspect this ought to be converted to libfrog/bitmap from the
slab cursors and whatnot that it uses now.
--D
> Brian
>
> > +
> > /*
> > * extra metadata for the agi
> > */
> > @@ -314,21 +320,24 @@ static void
> > finish_rebuild(
> > struct xfs_mount *mp,
> > struct bt_rebuild *btr,
> > - struct xfs_slab *lost_fsb)
> > + struct xfs_slab *lost_fsbs)
> > {
> > struct xrep_newbt_resv *resv, *n;
> >
> > for_each_xrep_newbt_reservation(&btr->newbt, resv, n) {
> > - while (resv->used < resv->len) {
> > - xfs_fsblock_t fsb = resv->fsbno + resv->used;
> > - int error;
> > + struct lost_fsb lost;
> > + int error;
> >
> > - error = slab_add(lost_fsb, &fsb);
> > - if (error)
> > - do_error(
> > + if (resv->used == resv->len)
> > + continue;
> > +
> > + lost.fsbno = resv->fsbno + resv->used;
> > + lost.len = resv->len - resv->used;
> > + error = slab_add(lost_fsbs, &lost);
> > + if (error)
> > + do_error(
> > _("Insufficient memory saving lost blocks.\n"));
> > - resv->used++;
> > - }
> > + resv->used = resv->len;
> > }
> >
> > xrep_newbt_destroy(&btr->newbt, 0);
> > @@ -1036,7 +1045,7 @@ build_agf_agfl(
> > int lostblocks, /* # blocks that will be lost */
> > struct bt_rebuild *btr_rmap,
> > struct bt_rebuild *btr_refc,
> > - struct xfs_slab *lost_fsb)
> > + struct xfs_slab *lost_fsbs)
> > {
> > struct extent_tree_node *ext_ptr;
> > struct xfs_buf *agf_buf, *agfl_buf;
> > @@ -1253,7 +1262,7 @@ static void
> > phase5_func(
> > struct xfs_mount *mp,
> > xfs_agnumber_t agno,
> > - struct xfs_slab *lost_fsb)
> > + struct xfs_slab *lost_fsbs)
> > {
> > struct repair_ctx sc = { .mp = mp, };
> > struct agi_stat agi_stat = {0,};
> > @@ -1372,7 +1381,7 @@ _("unable to rebuild AG %u. Not enough free space in on-disk AG.\n"),
> > * set up agf and agfl
> > */
> > build_agf_agfl(mp, agno, &btr_bno, &btr_cnt, freeblks1, extra_blocks,
> > - &btr_rmap, &btr_refc, lost_fsb);
> > + &btr_rmap, &btr_refc, lost_fsbs);
> >
> > /*
> > * build inode allocation trees.
> > @@ -1387,15 +1396,15 @@ _("unable to rebuild AG %u. Not enough free space in on-disk AG.\n"),
> > /*
> > * tear down cursors
> > */
> > - finish_rebuild(mp, &btr_bno, lost_fsb);
> > - finish_rebuild(mp, &btr_cnt, lost_fsb);
> > - finish_rebuild(mp, &btr_ino, lost_fsb);
> > + finish_rebuild(mp, &btr_bno, lost_fsbs);
> > + finish_rebuild(mp, &btr_cnt, lost_fsbs);
> > + finish_rebuild(mp, &btr_ino, lost_fsbs);
> > if (xfs_sb_version_hasfinobt(&mp->m_sb))
> > - finish_rebuild(mp, &btr_fino, lost_fsb);
> > + finish_rebuild(mp, &btr_fino, lost_fsbs);
> > if (xfs_sb_version_hasrmapbt(&mp->m_sb))
> > - finish_rebuild(mp, &btr_rmap, lost_fsb);
> > + finish_rebuild(mp, &btr_rmap, lost_fsbs);
> > if (xfs_sb_version_hasreflink(&mp->m_sb))
> > - finish_rebuild(mp, &btr_refc, lost_fsb);
> > + finish_rebuild(mp, &btr_refc, lost_fsbs);
> >
> > /*
> > * release the incore per-AG bno/bcnt trees so the extent nodes
> > @@ -1414,19 +1423,19 @@ inject_lost_blocks(
> > {
> > struct xfs_trans *tp = NULL;
> > struct xfs_slab_cursor *cur = NULL;
> > - xfs_fsblock_t *fsb;
> > + struct lost_fsb *lost;
> > int error;
> >
> > error = init_slab_cursor(lost_fsbs, NULL, &cur);
> > if (error)
> > return error;
> >
> > - while ((fsb = pop_slab_cursor(cur)) != NULL) {
> > + while ((lost = pop_slab_cursor(cur)) != NULL) {
> > error = -libxfs_trans_alloc_rollable(mp, 16, &tp);
> > if (error)
> > goto out_cancel;
> >
> > - error = -libxfs_free_extent(tp, *fsb, 1,
> > + error = -libxfs_free_extent(tp, lost->fsbno, lost->len,
> > &XFS_RMAP_OINFO_ANY_OWNER, XFS_AG_RESV_NONE);
> > if (error)
> > goto out_cancel;
> > @@ -1447,7 +1456,7 @@ inject_lost_blocks(
> > void
> > phase5(xfs_mount_t *mp)
> > {
> > - struct xfs_slab *lost_fsb;
> > + struct xfs_slab *lost_fsbs;
> > xfs_agnumber_t agno;
> > int error;
> >
> > @@ -1490,12 +1499,12 @@ phase5(xfs_mount_t *mp)
> > if (sb_fdblocks_ag == NULL)
> > do_error(_("cannot alloc sb_fdblocks_ag buffers\n"));
> >
> > - error = init_slab(&lost_fsb, sizeof(xfs_fsblock_t));
> > + error = init_slab(&lost_fsbs, sizeof(struct lost_fsb));
> > if (error)
> > do_error(_("cannot alloc lost block slab\n"));
> >
> > for (agno = 0; agno < mp->m_sb.sb_agcount; agno++)
> > - phase5_func(mp, agno, lost_fsb);
> > + phase5_func(mp, agno, lost_fsbs);
> >
> > print_final_rpt();
> >
> > @@ -1538,10 +1547,10 @@ _("unable to add AG %u reverse-mapping data to btree.\n"), agno);
> > * Put blocks that were unnecessarily reserved for btree
> > * reconstruction back into the filesystem free space data.
> > */
> > - error = inject_lost_blocks(mp, lost_fsb);
> > + error = inject_lost_blocks(mp, lost_fsbs);
> > if (error)
> > do_error(_("Unable to reinsert lost blocks into filesystem.\n"));
> > - free_slab(&lost_fsb);
> > + free_slab(&lost_fsbs);
> >
> > bad_ino_btree = 0;
> >
> >
>
^ permalink raw reply [flat|nested] 30+ messages in thread
* Re: [PATCH 5/9] xfs_repair: rebuild inode btrees with bulk loader
2020-05-29 22:18 ` Darrick J. Wong
@ 2020-05-29 22:32 ` Darrick J. Wong
0 siblings, 0 replies; 30+ messages in thread
From: Darrick J. Wong @ 2020-05-29 22:32 UTC (permalink / raw)
To: Brian Foster; +Cc: sandeen, linux-xfs
On Fri, May 29, 2020 at 03:18:40PM -0700, Darrick J. Wong wrote:
> On Thu, May 28, 2020 at 11:11:21AM -0400, Brian Foster wrote:
> > On Tue, May 19, 2020 at 06:51:15PM -0700, Darrick J. Wong wrote:
> > > From: Darrick J. Wong <darrick.wong@oracle.com>
> > >
> > > Use the btree bulk loading functions to rebuild the inode btrees
> > > and drop the open-coded implementation.
> > >
> > > Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
> > > ---
> > > libxfs/libxfs_api_defs.h | 1
> > > repair/phase5.c | 642 +++++++++++++++++-----------------------------
> > > 2 files changed, 240 insertions(+), 403 deletions(-)
> > >
> > >
> > ...
> > > diff --git a/repair/phase5.c b/repair/phase5.c
> > > index e69b042c..38f30753 100644
> > > --- a/repair/phase5.c
> > > +++ b/repair/phase5.c
> > ...
> > > @@ -372,6 +376,11 @@ estimate_ag_bload_slack(
> > > bload->node_slack = 0;
> > > }
> > >
> > > +static inline void skip_rebuild(struct bt_rebuild *btr)
> > > +{
> > > + memset(btr, 0, sizeof(struct bt_rebuild));
> > > +}
> > > +
> >
> > Is there any functional purpose to this? It looks like the memset could
> > be open-coded, but also seems like it could be elided if we just check
> > hasfinobt() before using the pointer..?
>
> Hm, yeah, ok, I'll go audit this.
>
> > > /* Initialize a btree rebuild context. */
> > > static void
> > > init_rebuild(
> > > @@ -765,48 +774,38 @@ _("Error %d while writing cntbt btree for AG %u.\n"), error, agno);
> > ...
> > > +
> > > +/* Copy one incore inode record into the inobt cursor. */
> > > +static void
> > > +get_inode_data(
> > > + struct xfs_btree_cur *cur,
> > > + struct ino_tree_node *ino_rec,
> > > + struct agi_stat *agi_stat)
> > > +{
> > > + struct xfs_inobt_rec_incore *irec = &cur->bc_rec.i;
> > > + int inocnt = 0;
> > > + int finocnt = 0;
> > > + int k;
> > > +
> > > + irec->ir_startino = ino_rec->ino_startnum;
> > > + irec->ir_free = ino_rec->ir_free;
> > > +
> > > + for (k = 0; k < sizeof(xfs_inofree_t) * NBBY; k++) {
> > > + ASSERT(is_inode_confirmed(ino_rec, k));
> > > +
> > > + if (is_inode_sparse(ino_rec, k))
> > > continue;
> > > -
> > > - nfinos += rec_nfinos;
> > > - ninos += rec_ninos;
> > > - num_recs++;
> > > + if (is_inode_free(ino_rec, k))
> > > + finocnt++;
> > > + inocnt++;
> > > }
> > >
> > > - if (num_recs == 0) {
> > > - /*
> > > - * easy corner-case -- no inode records
> > > - */
> > > - lptr->num_blocks = 1;
> > > - lptr->modulo = 0;
> > > - lptr->num_recs_pb = 0;
> > > - lptr->num_recs_tot = 0;
> > > -
> > > - btree_curs->num_levels = 1;
> > > - btree_curs->num_tot_blocks = btree_curs->num_free_blocks = 1;
> > > -
> > > - setup_cursor(mp, agno, btree_curs);
> > > + irec->ir_count = inocnt;
> > > + irec->ir_freecount = finocnt;
> > >
> > > - return;
> > > - }
> > > + if (xfs_sb_version_hassparseinodes(&cur->bc_mp->m_sb)) {
> > > + uint64_t sparse;
> > > + int spmask;
> > > + uint16_t holemask;
> > >
> > > - blocks_allocated = lptr->num_blocks = howmany(num_recs,
> > > - XR_INOBT_BLOCK_MAXRECS(mp, 0));
> > > -
> > > - lptr->modulo = num_recs % lptr->num_blocks;
> > > - lptr->num_recs_pb = num_recs / lptr->num_blocks;
> > > - lptr->num_recs_tot = num_recs;
> > > - level = 1;
> > > -
> > > - if (lptr->num_blocks > 1) {
> > > - for (; btree_curs->level[level-1].num_blocks > 1
> > > - && level < XFS_BTREE_MAXLEVELS;
> > > - level++) {
> > > - lptr = &btree_curs->level[level];
> > > - p_lptr = &btree_curs->level[level - 1];
> > > - lptr->num_blocks = howmany(p_lptr->num_blocks,
> > > - XR_INOBT_BLOCK_MAXRECS(mp, level));
> > > - lptr->modulo = p_lptr->num_blocks % lptr->num_blocks;
> > > - lptr->num_recs_pb = p_lptr->num_blocks
> > > - / lptr->num_blocks;
> > > - lptr->num_recs_tot = p_lptr->num_blocks;
> > > -
> > > - blocks_allocated += lptr->num_blocks;
> > > + /*
> > > + * Convert the 64-bit in-core sparse inode state to the
> > > + * 16-bit on-disk holemask.
> > > + */
> > > + holemask = 0;
> > > + spmask = (1 << XFS_INODES_PER_HOLEMASK_BIT) - 1;
> > > + sparse = ino_rec->ir_sparse;
> > > + for (k = 0; k < XFS_INOBT_HOLEMASK_BITS; k++) {
> > > + if (sparse & spmask) {
> > > + ASSERT((sparse & spmask) == spmask);
> > > + holemask |= (1 << k);
> > > + } else
> > > + ASSERT((sparse & spmask) == 0);
> > > + sparse >>= XFS_INODES_PER_HOLEMASK_BIT;
> > > }
> > > +
> > > + irec->ir_holemask = holemask;
> > > + } else {
> > > + irec->ir_holemask = 0;
> > > }
> > > - ASSERT(lptr->num_blocks == 1);
> > > - btree_curs->num_levels = level;
> > >
> > > - btree_curs->num_tot_blocks = btree_curs->num_free_blocks
> > > - = blocks_allocated;
> > > + if (!agi_stat)
> > > + return;
> > >
> > > - setup_cursor(mp, agno, btree_curs);
> > > + if (agi_stat->first_agino != NULLAGINO)
> > > + agi_stat->first_agino = ino_rec->ino_startnum;
> >
> > This is initialized to NULLAGINO. When do we ever update it?
>
> I don't understand your question. The purpose of the first_agino code
Aaaaha, the test here is wrong. This should be:
if (agi_stat->first_agino == NULLAGINO)
agi_stat->first_agino = ino_rec->ino_startnum;
And while I'm at it I'll get rid of agi_stat.
--D
> is to set agi_newino to the first inode cluster in the AG (or NULLAGINO
> if there are no inodes) so we initialize first_ino to NULLAGINO and if
> we process any inode records, we'll update it to ir_startino of the
> first inode record that we put in the btree.
>
> > > + agi_stat->freecount += finocnt;
> > > + agi_stat->count += inocnt;
> > > +}
> > >
> > > - *num_inos = ninos;
> > > - *num_free_inos = nfinos;
> > > +/* Grab one inobt record. */
> > > +static int
> > > +get_inobt_record(
> > > + struct xfs_btree_cur *cur,
> > > + void *priv)
> > > +{
> > > + struct bt_rebuild *rebuild = priv;
> > >
> > > - return;
> > > + get_inode_data(cur, rebuild->ino_rec, rebuild->agi_stat);
> > > + rebuild->ino_rec = next_ino_rec(rebuild->ino_rec);
> > > + return 0;
> > > }
> > >
> > > +/* Rebuild a inobt btree. */
> > > static void
> > > -prop_ino_cursor(xfs_mount_t *mp, xfs_agnumber_t agno, bt_status_t *btree_curs,
> > > - xfs_btnum_t btnum, xfs_agino_t startino, int level)
> > > +build_inobt(
> > > + struct repair_ctx *sc,
> > > + xfs_agnumber_t agno,
> > > + struct bt_rebuild *btr_ino,
> > > + struct agi_stat *agi_stat)
> > > {
> > > - struct xfs_btree_block *bt_hdr;
> > > - xfs_inobt_key_t *bt_key;
> > > - xfs_inobt_ptr_t *bt_ptr;
> > > - xfs_agblock_t agbno;
> > > - bt_stat_level_t *lptr;
> > > - const struct xfs_buf_ops *ops = btnum_to_ops(btnum);
> > > int error;
> > >
> > > - level++;
> > > -
> > > - if (level >= btree_curs->num_levels)
> > > - return;
> > > -
> > > - lptr = &btree_curs->level[level];
> > > - bt_hdr = XFS_BUF_TO_BLOCK(lptr->buf_p);
> > > -
> > > - if (be16_to_cpu(bt_hdr->bb_numrecs) == 0) {
> > > - /*
> > > - * this only happens once to initialize the
> > > - * first path up the left side of the tree
> > > - * where the agbno's are already set up
> > > - */
> > > - prop_ino_cursor(mp, agno, btree_curs, btnum, startino, level);
> > > - }
> > > -
> > > - if (be16_to_cpu(bt_hdr->bb_numrecs) ==
> > > - lptr->num_recs_pb + (lptr->modulo > 0)) {
> > > - /*
> > > - * write out current prev block, grab us a new block,
> > > - * and set the rightsib pointer of current block
> > > - */
> > > -#ifdef XR_BLD_INO_TRACE
> > > - fprintf(stderr, " ino prop agbno %d ", lptr->prev_agbno);
> > > -#endif
> > > - if (lptr->prev_agbno != NULLAGBLOCK) {
> > > - ASSERT(lptr->prev_buf_p != NULL);
> > > - libxfs_buf_mark_dirty(lptr->prev_buf_p);
> > > - libxfs_buf_relse(lptr->prev_buf_p);
> > > - }
> > > - lptr->prev_agbno = lptr->agbno;;
> > > - lptr->prev_buf_p = lptr->buf_p;
> > > - agbno = get_next_blockaddr(agno, level, btree_curs);
> > > -
> > > - bt_hdr->bb_u.s.bb_rightsib = cpu_to_be32(agbno);
> > > -
> > > - error = -libxfs_buf_get(mp->m_dev,
> > > - XFS_AGB_TO_DADDR(mp, agno, agbno),
> > > - XFS_FSB_TO_BB(mp, 1), &lptr->buf_p);
> > > - if (error)
> > > - do_error(_("Cannot grab inode btree buffer, err=%d"),
> > > - error);
> > > - lptr->agbno = agbno;
> > > + btr_ino->bload.get_record = get_inobt_record;
> > > + btr_ino->bload.claim_block = rebuild_claim_block;
> > > + agi_stat->count = agi_stat->freecount = 0;
> >
> > These are already initialized to zero by the caller. I suppose we might
> > as well also move the ->first_agino init to where this is allocated.
>
> TBH we don't need to (re)init agi_stat at all. I'll drop these two
> lines.
>
> > Otherwise I mostly just have the same general feedback as for the
> > previous patch (wrt to the get_record() logic and build_[f]inobt()
> > duplication.
>
> <nod> Will fix that too.
>
> > Brian
> >
> > > + agi_stat->first_agino = NULLAGINO;
> > > + btr_ino->agi_stat = agi_stat;
> > > + btr_ino->ino_rec = findfirst_inode_rec(agno);
> > > +
> > > + error = -libxfs_trans_alloc_empty(sc->mp, &sc->tp);
> > > + if (error)
> > > + do_error(
> > > +_("Insufficient memory to construct inobt rebuild transaction.\n"));
> > > +
> > > + /* Add all observed inobt records. */
> > > + error = -libxfs_btree_bload(btr_ino->cur, &btr_ino->bload, btr_ino);
> > > + if (error)
> > > + do_error(
> > > +_("Error %d while creating inobt btree for AG %u.\n"), error, agno);
> > > +
> > > + /* Since we're not writing the AGI yet, no need to commit the cursor */
> > > + libxfs_btree_del_cursor(btr_ino->cur, 0);
> > > + error = -libxfs_trans_commit(sc->tp);
> > > + if (error)
> > > + do_error(
> > > +_("Error %d while writing inobt btree for AG %u.\n"), error, agno);
> > > + sc->tp = NULL;
> > > +}
> > >
> > > - if (lptr->modulo)
> > > - lptr->modulo--;
> > > +/* Grab one finobt record. */
> > > +static int
> > > +get_finobt_record(
> > > + struct xfs_btree_cur *cur,
> > > + void *priv)
> > > +{
> > > + struct bt_rebuild *rebuild = priv;
> > >
> > > - /*
> > > - * initialize block header
> > > - */
> > > - lptr->buf_p->b_ops = ops;
> > > - bt_hdr = XFS_BUF_TO_BLOCK(lptr->buf_p);
> > > - memset(bt_hdr, 0, mp->m_sb.sb_blocksize);
> > > - libxfs_btree_init_block(mp, lptr->buf_p, btnum,
> > > - level, 0, agno);
> > > + get_inode_data(cur, rebuild->ino_rec, NULL);
> > > + rebuild->ino_rec = next_free_ino_rec(rebuild->ino_rec);
> > > + return 0;
> > > +}
> > >
> > > - bt_hdr->bb_u.s.bb_leftsib = cpu_to_be32(lptr->prev_agbno);
> > > +/* Rebuild a finobt btree. */
> > > +static void
> > > +build_finobt(
> > > + struct repair_ctx *sc,
> > > + xfs_agnumber_t agno,
> > > + struct bt_rebuild *btr_fino)
> > > +{
> > > + int error;
> > >
> > > - /*
> > > - * propagate extent record for first extent in new block up
> > > - */
> > > - prop_ino_cursor(mp, agno, btree_curs, btnum, startino, level);
> > > - }
> > > - /*
> > > - * add inode info to current block
> > > - */
> > > - be16_add_cpu(&bt_hdr->bb_numrecs, 1);
> > > -
> > > - bt_key = XFS_INOBT_KEY_ADDR(mp, bt_hdr,
> > > - be16_to_cpu(bt_hdr->bb_numrecs));
> > > - bt_ptr = XFS_INOBT_PTR_ADDR(mp, bt_hdr,
> > > - be16_to_cpu(bt_hdr->bb_numrecs),
> > > - M_IGEO(mp)->inobt_mxr[1]);
> > > -
> > > - bt_key->ir_startino = cpu_to_be32(startino);
> > > - *bt_ptr = cpu_to_be32(btree_curs->level[level-1].agbno);
> > > + btr_fino->bload.get_record = get_finobt_record;
> > > + btr_fino->bload.claim_block = rebuild_claim_block;
> > > + btr_fino->ino_rec = findfirst_free_inode_rec(agno);
> > > +
> > > + error = -libxfs_trans_alloc_empty(sc->mp, &sc->tp);
> > > + if (error)
> > > + do_error(
> > > +_("Insufficient memory to construct finobt rebuild transaction.\n"));
> > > +
> > > + /* Add all observed finobt records. */
> > > + error = -libxfs_btree_bload(btr_fino->cur, &btr_fino->bload, btr_fino);
> > > + if (error)
> > > + do_error(
> > > +_("Error %d while creating finobt btree for AG %u.\n"), error, agno);
> > > +
> > > + /* Since we're not writing the AGI yet, no need to commit the cursor */
> > > + libxfs_btree_del_cursor(btr_fino->cur, 0);
> > > + error = -libxfs_trans_commit(sc->tp);
> > > + if (error)
> > > + do_error(
> > > +_("Error %d while writing finobt btree for AG %u.\n"), error, agno);
> > > + sc->tp = NULL;
> > > }
> > >
> > > /*
> > > * XXX: yet more code that can be shared with mkfs, growfs.
> > > */
> > > static void
> > > -build_agi(xfs_mount_t *mp, xfs_agnumber_t agno, bt_status_t *btree_curs,
> > > - bt_status_t *finobt_curs, struct agi_stat *agi_stat)
> > > +build_agi(
> > > + struct xfs_mount *mp,
> > > + xfs_agnumber_t agno,
> > > + struct bt_rebuild *btr_ino,
> > > + struct bt_rebuild *btr_fino,
> > > + struct agi_stat *agi_stat)
> > > {
> > > - xfs_buf_t *agi_buf;
> > > - xfs_agi_t *agi;
> > > - int i;
> > > - int error;
> > > + struct xfs_buf *agi_buf;
> > > + struct xfs_agi *agi;
> > > + int i;
> > > + int error;
> > >
> > > error = -libxfs_buf_get(mp->m_dev,
> > > XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
> > > @@ -1008,8 +1053,8 @@ build_agi(xfs_mount_t *mp, xfs_agnumber_t agno, bt_status_t *btree_curs,
> > > agi->agi_length = cpu_to_be32(mp->m_sb.sb_dblocks -
> > > (xfs_rfsblock_t) mp->m_sb.sb_agblocks * agno);
> > > agi->agi_count = cpu_to_be32(agi_stat->count);
> > > - agi->agi_root = cpu_to_be32(btree_curs->root);
> > > - agi->agi_level = cpu_to_be32(btree_curs->num_levels);
> > > + agi->agi_root = cpu_to_be32(btr_ino->newbt.afake.af_root);
> > > + agi->agi_level = cpu_to_be32(btr_ino->newbt.afake.af_levels);
> > > agi->agi_freecount = cpu_to_be32(agi_stat->freecount);
> > > agi->agi_newino = cpu_to_be32(agi_stat->first_agino);
> > > agi->agi_dirino = cpu_to_be32(NULLAGINO);
> > > @@ -1021,203 +1066,16 @@ build_agi(xfs_mount_t *mp, xfs_agnumber_t agno, bt_status_t *btree_curs,
> > > platform_uuid_copy(&agi->agi_uuid, &mp->m_sb.sb_meta_uuid);
> > >
> > > if (xfs_sb_version_hasfinobt(&mp->m_sb)) {
> > > - agi->agi_free_root = cpu_to_be32(finobt_curs->root);
> > > - agi->agi_free_level = cpu_to_be32(finobt_curs->num_levels);
> > > + agi->agi_free_root =
> > > + cpu_to_be32(btr_fino->newbt.afake.af_root);
> > > + agi->agi_free_level =
> > > + cpu_to_be32(btr_fino->newbt.afake.af_levels);
> > > }
> > >
> > > libxfs_buf_mark_dirty(agi_buf);
> > > libxfs_buf_relse(agi_buf);
> > > }
> > >
> > > -/*
> > > - * rebuilds an inode tree given a cursor. We're lazy here and call
> > > - * the routine that builds the agi
> > > - */
> > > -static void
> > > -build_ino_tree(xfs_mount_t *mp, xfs_agnumber_t agno,
> > > - bt_status_t *btree_curs, xfs_btnum_t btnum,
> > > - struct agi_stat *agi_stat)
> > > -{
> > > - xfs_agnumber_t i;
> > > - xfs_agblock_t j;
> > > - xfs_agblock_t agbno;
> > > - xfs_agino_t first_agino;
> > > - struct xfs_btree_block *bt_hdr;
> > > - xfs_inobt_rec_t *bt_rec;
> > > - ino_tree_node_t *ino_rec;
> > > - bt_stat_level_t *lptr;
> > > - const struct xfs_buf_ops *ops = btnum_to_ops(btnum);
> > > - xfs_agino_t count = 0;
> > > - xfs_agino_t freecount = 0;
> > > - int inocnt;
> > > - uint8_t finocnt;
> > > - int k;
> > > - int level = btree_curs->num_levels;
> > > - int spmask;
> > > - uint64_t sparse;
> > > - uint16_t holemask;
> > > - int error;
> > > -
> > > - ASSERT(btnum == XFS_BTNUM_INO || btnum == XFS_BTNUM_FINO);
> > > -
> > > - for (i = 0; i < level; i++) {
> > > - lptr = &btree_curs->level[i];
> > > -
> > > - agbno = get_next_blockaddr(agno, i, btree_curs);
> > > - error = -libxfs_buf_get(mp->m_dev,
> > > - XFS_AGB_TO_DADDR(mp, agno, agbno),
> > > - XFS_FSB_TO_BB(mp, 1), &lptr->buf_p);
> > > - if (error)
> > > - do_error(_("Cannot grab inode btree buffer, err=%d"),
> > > - error);
> > > -
> > > - if (i == btree_curs->num_levels - 1)
> > > - btree_curs->root = agbno;
> > > -
> > > - lptr->agbno = agbno;
> > > - lptr->prev_agbno = NULLAGBLOCK;
> > > - lptr->prev_buf_p = NULL;
> > > - /*
> > > - * initialize block header
> > > - */
> > > -
> > > - lptr->buf_p->b_ops = ops;
> > > - bt_hdr = XFS_BUF_TO_BLOCK(lptr->buf_p);
> > > - memset(bt_hdr, 0, mp->m_sb.sb_blocksize);
> > > - libxfs_btree_init_block(mp, lptr->buf_p, btnum, i, 0, agno);
> > > - }
> > > -
> > > - /*
> > > - * run along leaf, setting up records. as we have to switch
> > > - * blocks, call the prop_ino_cursor routine to set up the new
> > > - * pointers for the parent. that can recurse up to the root
> > > - * if required. set the sibling pointers for leaf level here.
> > > - */
> > > - if (btnum == XFS_BTNUM_FINO)
> > > - ino_rec = findfirst_free_inode_rec(agno);
> > > - else
> > > - ino_rec = findfirst_inode_rec(agno);
> > > -
> > > - if (ino_rec != NULL)
> > > - first_agino = ino_rec->ino_startnum;
> > > - else
> > > - first_agino = NULLAGINO;
> > > -
> > > - lptr = &btree_curs->level[0];
> > > -
> > > - for (i = 0; i < lptr->num_blocks; i++) {
> > > - /*
> > > - * block initialization, lay in block header
> > > - */
> > > - lptr->buf_p->b_ops = ops;
> > > - bt_hdr = XFS_BUF_TO_BLOCK(lptr->buf_p);
> > > - memset(bt_hdr, 0, mp->m_sb.sb_blocksize);
> > > - libxfs_btree_init_block(mp, lptr->buf_p, btnum, 0, 0, agno);
> > > -
> > > - bt_hdr->bb_u.s.bb_leftsib = cpu_to_be32(lptr->prev_agbno);
> > > - bt_hdr->bb_numrecs = cpu_to_be16(lptr->num_recs_pb +
> > > - (lptr->modulo > 0));
> > > -
> > > - if (lptr->modulo > 0)
> > > - lptr->modulo--;
> > > -
> > > - if (lptr->num_recs_pb > 0)
> > > - prop_ino_cursor(mp, agno, btree_curs, btnum,
> > > - ino_rec->ino_startnum, 0);
> > > -
> > > - bt_rec = (xfs_inobt_rec_t *)
> > > - ((char *)bt_hdr + XFS_INOBT_BLOCK_LEN(mp));
> > > - for (j = 0; j < be16_to_cpu(bt_hdr->bb_numrecs); j++) {
> > > - ASSERT(ino_rec != NULL);
> > > - bt_rec[j].ir_startino =
> > > - cpu_to_be32(ino_rec->ino_startnum);
> > > - bt_rec[j].ir_free = cpu_to_be64(ino_rec->ir_free);
> > > -
> > > - inocnt = finocnt = 0;
> > > - for (k = 0; k < sizeof(xfs_inofree_t)*NBBY; k++) {
> > > - ASSERT(is_inode_confirmed(ino_rec, k));
> > > -
> > > - if (is_inode_sparse(ino_rec, k))
> > > - continue;
> > > - if (is_inode_free(ino_rec, k))
> > > - finocnt++;
> > > - inocnt++;
> > > - }
> > > -
> > > - /*
> > > - * Set the freecount and check whether we need to update
> > > - * the sparse format fields. Otherwise, skip to the next
> > > - * record.
> > > - */
> > > - inorec_set_freecount(mp, &bt_rec[j], finocnt);
> > > - if (!xfs_sb_version_hassparseinodes(&mp->m_sb))
> > > - goto nextrec;
> > > -
> > > - /*
> > > - * Convert the 64-bit in-core sparse inode state to the
> > > - * 16-bit on-disk holemask.
> > > - */
> > > - holemask = 0;
> > > - spmask = (1 << XFS_INODES_PER_HOLEMASK_BIT) - 1;
> > > - sparse = ino_rec->ir_sparse;
> > > - for (k = 0; k < XFS_INOBT_HOLEMASK_BITS; k++) {
> > > - if (sparse & spmask) {
> > > - ASSERT((sparse & spmask) == spmask);
> > > - holemask |= (1 << k);
> > > - } else
> > > - ASSERT((sparse & spmask) == 0);
> > > - sparse >>= XFS_INODES_PER_HOLEMASK_BIT;
> > > - }
> > > -
> > > - bt_rec[j].ir_u.sp.ir_count = inocnt;
> > > - bt_rec[j].ir_u.sp.ir_holemask = cpu_to_be16(holemask);
> > > -
> > > -nextrec:
> > > - freecount += finocnt;
> > > - count += inocnt;
> > > -
> > > - if (btnum == XFS_BTNUM_FINO)
> > > - ino_rec = next_free_ino_rec(ino_rec);
> > > - else
> > > - ino_rec = next_ino_rec(ino_rec);
> > > - }
> > > -
> > > - if (ino_rec != NULL) {
> > > - /*
> > > - * get next leaf level block
> > > - */
> > > - if (lptr->prev_buf_p != NULL) {
> > > -#ifdef XR_BLD_INO_TRACE
> > > - fprintf(stderr, "writing inobt agbno %u\n",
> > > - lptr->prev_agbno);
> > > -#endif
> > > - ASSERT(lptr->prev_agbno != NULLAGBLOCK);
> > > - libxfs_buf_mark_dirty(lptr->prev_buf_p);
> > > - libxfs_buf_relse(lptr->prev_buf_p);
> > > - }
> > > - lptr->prev_buf_p = lptr->buf_p;
> > > - lptr->prev_agbno = lptr->agbno;
> > > - lptr->agbno = get_next_blockaddr(agno, 0, btree_curs);
> > > - bt_hdr->bb_u.s.bb_rightsib = cpu_to_be32(lptr->agbno);
> > > -
> > > - error = -libxfs_buf_get(mp->m_dev,
> > > - XFS_AGB_TO_DADDR(mp, agno, lptr->agbno),
> > > - XFS_FSB_TO_BB(mp, 1),
> > > - &lptr->buf_p);
> > > - if (error)
> > > - do_error(
> > > - _("Cannot grab inode btree buffer, err=%d"),
> > > - error);
> > > - }
> > > - }
> > > -
> > > - if (agi_stat) {
> > > - agi_stat->first_agino = first_agino;
> > > - agi_stat->count = count;
> > > - agi_stat->freecount = freecount;
> > > - }
> > > -}
> > > -
> > > /* rebuild the rmap tree */
> > >
> > > /*
> > > @@ -2142,14 +2000,10 @@ phase5_func(
> > > {
> > > struct repair_ctx sc = { .mp = mp, };
> > > struct agi_stat agi_stat = {0,};
> > > - uint64_t num_inos;
> > > - uint64_t num_free_inos;
> > > - uint64_t finobt_num_inos;
> > > - uint64_t finobt_num_free_inos;
> > > struct bt_rebuild btr_bno;
> > > struct bt_rebuild btr_cnt;
> > > - bt_status_t ino_btree_curs;
> > > - bt_status_t fino_btree_curs;
> > > + struct bt_rebuild btr_ino;
> > > + struct bt_rebuild btr_fino;
> > > bt_status_t rmap_btree_curs;
> > > bt_status_t refcnt_btree_curs;
> > > int extra_blocks = 0;
> > > @@ -2184,19 +2038,8 @@ _("unable to rebuild AG %u. Not enough free space in on-disk AG.\n"),
> > > agno);
> > > }
> > >
> > > - /*
> > > - * ok, now set up the btree cursors for the on-disk btrees (includes
> > > - * pre-allocating all required blocks for the trees themselves)
> > > - */
> > > - init_ino_cursor(mp, agno, &ino_btree_curs, &num_inos,
> > > - &num_free_inos, 0);
> > > -
> > > - if (xfs_sb_version_hasfinobt(&mp->m_sb))
> > > - init_ino_cursor(mp, agno, &fino_btree_curs, &finobt_num_inos,
> > > - &finobt_num_free_inos, 1);
> > > -
> > > - sb_icount_ag[agno] += num_inos;
> > > - sb_ifree_ag[agno] += num_free_inos;
> > > + init_ino_cursors(&sc, agno, num_freeblocks, &sb_icount_ag[agno],
> > > + &sb_ifree_ag[agno], &btr_ino, &btr_fino);
> > >
> > > /*
> > > * Set up the btree cursors for the on-disk rmap btrees, which includes
> > > @@ -2287,34 +2130,27 @@ _("unable to rebuild AG %u. Not enough free space in on-disk AG.\n"),
> > > &rmap_btree_curs, &refcnt_btree_curs, lost_fsb);
> > >
> > > /*
> > > - * build inode allocation tree.
> > > + * build inode allocation trees.
> > > */
> > > - build_ino_tree(mp, agno, &ino_btree_curs, XFS_BTNUM_INO, &agi_stat);
> > > - write_cursor(&ino_btree_curs);
> > > -
> > > - /*
> > > - * build free inode tree
> > > - */
> > > - if (xfs_sb_version_hasfinobt(&mp->m_sb)) {
> > > - build_ino_tree(mp, agno, &fino_btree_curs,
> > > - XFS_BTNUM_FINO, NULL);
> > > - write_cursor(&fino_btree_curs);
> > > - }
> > > + build_inobt(&sc, agno, &btr_ino, &agi_stat);
> > > + if (xfs_sb_version_hasfinobt(&mp->m_sb))
> > > + build_finobt(&sc, agno, &btr_fino);
> > >
> > > /* build the agi */
> > > - build_agi(mp, agno, &ino_btree_curs, &fino_btree_curs, &agi_stat);
> > > + build_agi(mp, agno, &btr_ino, &btr_fino, &agi_stat);
> > >
> > > /*
> > > * tear down cursors
> > > */
> > > finish_rebuild(mp, &btr_bno, lost_fsb);
> > > finish_rebuild(mp, &btr_cnt, lost_fsb);
> > > + finish_rebuild(mp, &btr_ino, lost_fsb);
> > > + if (xfs_sb_version_hasfinobt(&mp->m_sb))
> > > + finish_rebuild(mp, &btr_fino, lost_fsb);
> > > if (xfs_sb_version_hasrmapbt(&mp->m_sb))
> > > finish_cursor(&rmap_btree_curs);
> > > if (xfs_sb_version_hasreflink(&mp->m_sb))
> > > finish_cursor(&refcnt_btree_curs);
> > > - if (xfs_sb_version_hasfinobt(&mp->m_sb))
> > > - finish_cursor(&fino_btree_curs);
> > >
> > > /*
> > > * release the incore per-AG bno/bcnt trees so the extent nodes
> > >
> >
^ permalink raw reply [flat|nested] 30+ messages in thread
* Re: [PATCH 1/9] xfs_repair: port the online repair newbt structure
2020-05-29 21:01 ` Darrick J. Wong
@ 2020-06-01 12:03 ` Brian Foster
2020-06-02 0:12 ` Darrick J. Wong
0 siblings, 1 reply; 30+ messages in thread
From: Brian Foster @ 2020-06-01 12:03 UTC (permalink / raw)
To: Darrick J. Wong; +Cc: sandeen, linux-xfs
On Fri, May 29, 2020 at 02:01:59PM -0700, Darrick J. Wong wrote:
> On Thu, May 28, 2020 at 11:08:36AM -0400, Brian Foster wrote:
> > On Wed, May 27, 2020 at 03:34:24PM -0700, Darrick J. Wong wrote:
> > > On Wed, May 27, 2020 at 08:15:31AM -0400, Brian Foster wrote:
> > > > On Tue, May 19, 2020 at 06:50:49PM -0700, Darrick J. Wong wrote:
> > > > > From: Darrick J. Wong <darrick.wong@oracle.com>
> > > > >
> > > > > Port the new btree staging context and related block reservation helper
> > > > > code from the kernel to repair. We'll use this in subsequent patches to
> > > > > implement btree bulk loading.
> > > > >
> > > > > Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
> > > > > ---
> > > > > include/libxfs.h | 1
> > > > > libxfs/libxfs_api_defs.h | 2
> > > > > repair/Makefile | 4 -
> > > > > repair/bload.c | 303 ++++++++++++++++++++++++++++++++++++++++++++++
> > > > > repair/bload.h | 77 ++++++++++++
> > > > > repair/xfs_repair.c | 17 +++
> > > > > 6 files changed, 402 insertions(+), 2 deletions(-)
> > > > > create mode 100644 repair/bload.c
> > > > > create mode 100644 repair/bload.h
> > > > >
> > > > >
> > > > ...
> > > > > diff --git a/repair/bload.c b/repair/bload.c
> > > > > new file mode 100644
> > > > > index 00000000..9bc17468
> > > > > --- /dev/null
> > > > > +++ b/repair/bload.c
> > > > > @@ -0,0 +1,303 @@
> > > > > +// SPDX-License-Identifier: GPL-2.0-or-later
> > > > > +/*
> > > > > + * Copyright (C) 2020 Oracle. All Rights Reserved.
> > > > > + * Author: Darrick J. Wong <darrick.wong@oracle.com>
> > > > > + */
> > > > > +#include <libxfs.h>
> > > > > +#include "bload.h"
> > > > > +
> > > > > +#define trace_xrep_newbt_claim_block(...) ((void) 0)
> > > > > +#define trace_xrep_newbt_free_blocks(...) ((void) 0)
> > > > > +
> > > > > +int bload_leaf_slack = -1;
> > > > > +int bload_node_slack = -1;
> > > > > +
> > > > > +/* Ported routines from fs/xfs/scrub/repair.c */
> > > > > +
> > > >
> > > > Looks mostly straightforward, but I'll have to come back to this as I
> > > > get to the code that uses it later in the series. In the meantime, I see
> > > > some of these helpers in scrub/repair.c while not others. Are there
> > > > references to other routines that are intended to be copies from kernel
> > > > code?
> > >
> > > Hm. I might not understand the question, but in general the code should
> > > be fairly similar to the kernel functions. The biggest differences are
> > > (a) that whole libxfs error code mess, (b) the much simpler repair_ctx
> > > structure, and (c) the fact that repair doesn't bother with EFIs to
> > > automatically reap blocks.
> > >
> > > So... the ten functions you see here do the same things as their kernel
> > > counterparts, but they get to do it in the much simpler userspace
> > > environment.
> > >
> >
> > Right.. I was able to find the first function (xrep_roll_ag_trans())
> > easily in the kernel because it has the same name. The next one or two
> > (i.e., xrep_newbt_*()) I couldn't find and then gave up. Are they
> > renamed? Unmerged?
>
> Ahh, silly me, the rest are as yet unmerged components of the online
> repair code. Maybe it makes more sense to drop the "Ported routines
> from XXXX" comment, particularly since it probably will be easier to
> merge the xfs_repair series first, which is /much/ smaller in scope.
>
Probably so. It's just going to confuse if this is not intended to land
at the same time as the kernel code.
> > > The other functions in scrub/repair.c that didn't get ported are either
> > > for other types of repairs or exist to support the in-kernel code and
> > > aren't needed here.
> > >
> >
> > Sure, I'm just curious how to identify the source of the ones that are.
>
> https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux.git/tree/fs/xfs/scrub/repair.c?h=djwong-wtf
>
> Is a fairly recent snapshot of what the kernel version looks like.
>
Thanks.
Brian
> --D
>
> >
> > Brian
> >
> > > --D
> > >
> > > > Brian
> > > >
> > > > > +/*
> > > > > + * Roll a transaction, keeping the AG headers locked and reinitializing
> > > > > + * the btree cursors.
> > > > > + */
> > > > > +int
> > > > > +xrep_roll_ag_trans(
> > > > > + struct repair_ctx *sc)
> > > > > +{
> > > > > + int error;
> > > > > +
> > > > > + /* Keep the AG header buffers locked so we can keep going. */
> > > > > + if (sc->agi_bp)
> > > > > + libxfs_trans_bhold(sc->tp, sc->agi_bp);
> > > > > + if (sc->agf_bp)
> > > > > + libxfs_trans_bhold(sc->tp, sc->agf_bp);
> > > > > + if (sc->agfl_bp)
> > > > > + libxfs_trans_bhold(sc->tp, sc->agfl_bp);
> > > > > +
> > > > > + /*
> > > > > + * Roll the transaction. We still own the buffer and the buffer lock
> > > > > + * regardless of whether or not the roll succeeds. If the roll fails,
> > > > > + * the buffers will be released during teardown on our way out of the
> > > > > + * kernel. If it succeeds, we join them to the new transaction and
> > > > > + * move on.
> > > > > + */
> > > > > + error = -libxfs_trans_roll(&sc->tp);
> > > > > + if (error)
> > > > > + return error;
> > > > > +
> > > > > + /* Join AG headers to the new transaction. */
> > > > > + if (sc->agi_bp)
> > > > > + libxfs_trans_bjoin(sc->tp, sc->agi_bp);
> > > > > + if (sc->agf_bp)
> > > > > + libxfs_trans_bjoin(sc->tp, sc->agf_bp);
> > > > > + if (sc->agfl_bp)
> > > > > + libxfs_trans_bjoin(sc->tp, sc->agfl_bp);
> > > > > +
> > > > > + return 0;
> > > > > +}
> > > > > +
> > > > > +/* Initialize accounting resources for staging a new AG btree. */
> > > > > +void
> > > > > +xrep_newbt_init_ag(
> > > > > + struct xrep_newbt *xnr,
> > > > > + struct repair_ctx *sc,
> > > > > + const struct xfs_owner_info *oinfo,
> > > > > + xfs_fsblock_t alloc_hint,
> > > > > + enum xfs_ag_resv_type resv)
> > > > > +{
> > > > > + memset(xnr, 0, sizeof(struct xrep_newbt));
> > > > > + xnr->sc = sc;
> > > > > + xnr->oinfo = *oinfo; /* structure copy */
> > > > > + xnr->alloc_hint = alloc_hint;
> > > > > + xnr->resv = resv;
> > > > > + INIT_LIST_HEAD(&xnr->resv_list);
> > > > > +}
> > > > > +
> > > > > +/* Initialize accounting resources for staging a new inode fork btree. */
> > > > > +void
> > > > > +xrep_newbt_init_inode(
> > > > > + struct xrep_newbt *xnr,
> > > > > + struct repair_ctx *sc,
> > > > > + int whichfork,
> > > > > + const struct xfs_owner_info *oinfo)
> > > > > +{
> > > > > + xrep_newbt_init_ag(xnr, sc, oinfo,
> > > > > + XFS_INO_TO_FSB(sc->mp, sc->ip->i_ino),
> > > > > + XFS_AG_RESV_NONE);
> > > > > + xnr->ifake.if_fork = kmem_zone_zalloc(xfs_ifork_zone, 0);
> > > > > + xnr->ifake.if_fork_size = XFS_IFORK_SIZE(sc->ip, whichfork);
> > > > > +}
> > > > > +
> > > > > +/*
> > > > > + * Initialize accounting resources for staging a new btree. Callers are
> > > > > + * expected to add their own reservations (and clean them up) manually.
> > > > > + */
> > > > > +void
> > > > > +xrep_newbt_init_bare(
> > > > > + struct xrep_newbt *xnr,
> > > > > + struct repair_ctx *sc)
> > > > > +{
> > > > > + xrep_newbt_init_ag(xnr, sc, &XFS_RMAP_OINFO_ANY_OWNER, NULLFSBLOCK,
> > > > > + XFS_AG_RESV_NONE);
> > > > > +}
> > > > > +
> > > > > +/* Designate specific blocks to be used to build our new btree. */
> > > > > +int
> > > > > +xrep_newbt_add_blocks(
> > > > > + struct xrep_newbt *xnr,
> > > > > + xfs_fsblock_t fsbno,
> > > > > + xfs_extlen_t len)
> > > > > +{
> > > > > + struct xrep_newbt_resv *resv;
> > > > > +
> > > > > + resv = kmem_alloc(sizeof(struct xrep_newbt_resv), KM_MAYFAIL);
> > > > > + if (!resv)
> > > > > + return ENOMEM;
> > > > > +
> > > > > + INIT_LIST_HEAD(&resv->list);
> > > > > + resv->fsbno = fsbno;
> > > > > + resv->len = len;
> > > > > + resv->used = 0;
> > > > > + list_add_tail(&resv->list, &xnr->resv_list);
> > > > > + return 0;
> > > > > +}
> > > > > +
> > > > > +/* Reserve disk space for our new btree. */
> > > > > +int
> > > > > +xrep_newbt_alloc_blocks(
> > > > > + struct xrep_newbt *xnr,
> > > > > + uint64_t nr_blocks)
> > > > > +{
> > > > > + struct repair_ctx *sc = xnr->sc;
> > > > > + xfs_alloctype_t type;
> > > > > + xfs_fsblock_t alloc_hint = xnr->alloc_hint;
> > > > > + int error = 0;
> > > > > +
> > > > > + type = sc->ip ? XFS_ALLOCTYPE_START_BNO : XFS_ALLOCTYPE_NEAR_BNO;
> > > > > +
> > > > > + while (nr_blocks > 0 && !error) {
> > > > > + struct xfs_alloc_arg args = {
> > > > > + .tp = sc->tp,
> > > > > + .mp = sc->mp,
> > > > > + .type = type,
> > > > > + .fsbno = alloc_hint,
> > > > > + .oinfo = xnr->oinfo,
> > > > > + .minlen = 1,
> > > > > + .maxlen = nr_blocks,
> > > > > + .prod = 1,
> > > > > + .resv = xnr->resv,
> > > > > + };
> > > > > +
> > > > > + error = -libxfs_alloc_vextent(&args);
> > > > > + if (error)
> > > > > + return error;
> > > > > + if (args.fsbno == NULLFSBLOCK)
> > > > > + return ENOSPC;
> > > > > +
> > > > > + /* We don't have real EFIs here so skip that. */
> > > > > +
> > > > > + error = xrep_newbt_add_blocks(xnr, args.fsbno, args.len);
> > > > > + if (error)
> > > > > + break;
> > > > > +
> > > > > + nr_blocks -= args.len;
> > > > > + alloc_hint = args.fsbno + args.len - 1;
> > > > > +
> > > > > + if (sc->ip)
> > > > > + error = -libxfs_trans_roll_inode(&sc->tp, sc->ip);
> > > > > + else
> > > > > + error = xrep_roll_ag_trans(sc);
> > > > > + }
> > > > > +
> > > > > + return error;
> > > > > +}
> > > > > +
> > > > > +/*
> > > > > + * Release blocks that were reserved for a btree repair. If the repair
> > > > > + * succeeded then we log deferred frees for unused blocks. Otherwise, we try
> > > > > + * to free the extents immediately to roll the filesystem back to where it was
> > > > > + * before we started.
> > > > > + */
> > > > > +static inline int
> > > > > +xrep_newbt_destroy_reservation(
> > > > > + struct xrep_newbt *xnr,
> > > > > + struct xrep_newbt_resv *resv,
> > > > > + bool cancel_repair)
> > > > > +{
> > > > > + struct repair_ctx *sc = xnr->sc;
> > > > > +
> > > > > + if (cancel_repair) {
> > > > > + int error;
> > > > > +
> > > > > + /* Free the extent then roll the transaction. */
> > > > > + error = -libxfs_free_extent(sc->tp, resv->fsbno, resv->len,
> > > > > + &xnr->oinfo, xnr->resv);
> > > > > + if (error)
> > > > > + return error;
> > > > > +
> > > > > + if (sc->ip)
> > > > > + return -libxfs_trans_roll_inode(&sc->tp, sc->ip);
> > > > > + return xrep_roll_ag_trans(sc);
> > > > > + }
> > > > > +
> > > > > + /* We don't have EFIs here so skip the EFD. */
> > > > > +
> > > > > + /*
> > > > > + * Use the deferred freeing mechanism to schedule for deletion any
> > > > > + * blocks we didn't use to rebuild the tree. This enables us to log
> > > > > + * them all in the same transaction as the root change.
> > > > > + */
> > > > > + resv->fsbno += resv->used;
> > > > > + resv->len -= resv->used;
> > > > > + resv->used = 0;
> > > > > +
> > > > > + if (resv->len == 0)
> > > > > + return 0;
> > > > > +
> > > > > + trace_xrep_newbt_free_blocks(sc->mp,
> > > > > + XFS_FSB_TO_AGNO(sc->mp, resv->fsbno),
> > > > > + XFS_FSB_TO_AGBNO(sc->mp, resv->fsbno),
> > > > > + resv->len, xnr->oinfo.oi_owner);
> > > > > +
> > > > > + __xfs_bmap_add_free(sc->tp, resv->fsbno, resv->len, &xnr->oinfo, true);
> > > > > +
> > > > > + return 0;
> > > > > +}
> > > > > +
> > > > > +/* Free all the accounting info and disk space we reserved for a new btree. */
> > > > > +void
> > > > > +xrep_newbt_destroy(
> > > > > + struct xrep_newbt *xnr,
> > > > > + int error)
> > > > > +{
> > > > > + struct repair_ctx *sc = xnr->sc;
> > > > > + struct xrep_newbt_resv *resv, *n;
> > > > > + int err2;
> > > > > +
> > > > > + list_for_each_entry_safe(resv, n, &xnr->resv_list, list) {
> > > > > + err2 = xrep_newbt_destroy_reservation(xnr, resv, error != 0);
> > > > > + if (err2)
> > > > > + goto junkit;
> > > > > +
> > > > > + list_del(&resv->list);
> > > > > + kmem_free(resv);
> > > > > + }
> > > > > +
> > > > > +junkit:
> > > > > + /*
> > > > > + * If we still have reservations attached to @newbt, cleanup must have
> > > > > + * failed and the filesystem is about to go down. Clean up the incore
> > > > > + * reservations.
> > > > > + */
> > > > > + list_for_each_entry_safe(resv, n, &xnr->resv_list, list) {
> > > > > + list_del(&resv->list);
> > > > > + kmem_free(resv);
> > > > > + }
> > > > > +
> > > > > + if (sc->ip) {
> > > > > + kmem_cache_free(xfs_ifork_zone, xnr->ifake.if_fork);
> > > > > + xnr->ifake.if_fork = NULL;
> > > > > + }
> > > > > +}
> > > > > +
> > > > > +/* Feed one of the reserved btree blocks to the bulk loader. */
> > > > > +int
> > > > > +xrep_newbt_claim_block(
> > > > > + struct xfs_btree_cur *cur,
> > > > > + struct xrep_newbt *xnr,
> > > > > + union xfs_btree_ptr *ptr)
> > > > > +{
> > > > > + struct xrep_newbt_resv *resv;
> > > > > + xfs_fsblock_t fsb;
> > > > > +
> > > > > + /*
> > > > > + * The first item in the list should always have a free block unless
> > > > > + * we're completely out.
> > > > > + */
> > > > > + resv = list_first_entry(&xnr->resv_list, struct xrep_newbt_resv, list);
> > > > > + if (resv->used == resv->len)
> > > > > + return ENOSPC;
> > > > > +
> > > > > + /*
> > > > > + * Peel off a block from the start of the reservation. We allocate
> > > > > + * blocks in order to place blocks on disk in increasing record or key
> > > > > + * order. The block reservations tend to end up on the list in
> > > > > + * decreasing order, which hopefully results in leaf blocks ending up
> > > > > + * together.
> > > > > + */
> > > > > + fsb = resv->fsbno + resv->used;
> > > > > + resv->used++;
> > > > > +
> > > > > + /* If we used all the blocks in this reservation, move it to the end. */
> > > > > + if (resv->used == resv->len)
> > > > > + list_move_tail(&resv->list, &xnr->resv_list);
> > > > > +
> > > > > + trace_xrep_newbt_claim_block(cur->bc_mp,
> > > > > + XFS_FSB_TO_AGNO(cur->bc_mp, fsb),
> > > > > + XFS_FSB_TO_AGBNO(cur->bc_mp, fsb),
> > > > > + 1, xnr->oinfo.oi_owner);
> > > > > +
> > > > > + if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
> > > > > + ptr->l = cpu_to_be64(fsb);
> > > > > + else
> > > > > + ptr->s = cpu_to_be32(XFS_FSB_TO_AGBNO(cur->bc_mp, fsb));
> > > > > + return 0;
> > > > > +}
> > > > > diff --git a/repair/bload.h b/repair/bload.h
> > > > > new file mode 100644
> > > > > index 00000000..020c4834
> > > > > --- /dev/null
> > > > > +++ b/repair/bload.h
> > > > > @@ -0,0 +1,77 @@
> > > > > +// SPDX-License-Identifier: GPL-2.0-or-later
> > > > > +/*
> > > > > + * Copyright (C) 2020 Oracle. All Rights Reserved.
> > > > > + * Author: Darrick J. Wong <darrick.wong@oracle.com>
> > > > > + */
> > > > > +#ifndef __XFS_REPAIR_BLOAD_H__
> > > > > +#define __XFS_REPAIR_BLOAD_H__
> > > > > +
> > > > > +extern int bload_leaf_slack;
> > > > > +extern int bload_node_slack;
> > > > > +
> > > > > +struct repair_ctx {
> > > > > + struct xfs_mount *mp;
> > > > > + struct xfs_inode *ip;
> > > > > + struct xfs_trans *tp;
> > > > > +
> > > > > + struct xfs_buf *agi_bp;
> > > > > + struct xfs_buf *agf_bp;
> > > > > + struct xfs_buf *agfl_bp;
> > > > > +};
> > > > > +
> > > > > +struct xrep_newbt_resv {
> > > > > + /* Link to list of extents that we've reserved. */
> > > > > + struct list_head list;
> > > > > +
> > > > > + /* FSB of the block we reserved. */
> > > > > + xfs_fsblock_t fsbno;
> > > > > +
> > > > > + /* Length of the reservation. */
> > > > > + xfs_extlen_t len;
> > > > > +
> > > > > + /* How much of this reservation we've used. */
> > > > > + xfs_extlen_t used;
> > > > > +};
> > > > > +
> > > > > +struct xrep_newbt {
> > > > > + struct repair_ctx *sc;
> > > > > +
> > > > > + /* List of extents that we've reserved. */
> > > > > + struct list_head resv_list;
> > > > > +
> > > > > + /* Fake root for new btree. */
> > > > > + union {
> > > > > + struct xbtree_afakeroot afake;
> > > > > + struct xbtree_ifakeroot ifake;
> > > > > + };
> > > > > +
> > > > > + /* rmap owner of these blocks */
> > > > > + struct xfs_owner_info oinfo;
> > > > > +
> > > > > + /* The last reservation we allocated from. */
> > > > > + struct xrep_newbt_resv *last_resv;
> > > > > +
> > > > > + /* Allocation hint */
> > > > > + xfs_fsblock_t alloc_hint;
> > > > > +
> > > > > + /* per-ag reservation type */
> > > > > + enum xfs_ag_resv_type resv;
> > > > > +};
> > > > > +
> > > > > +#define for_each_xrep_newbt_reservation(xnr, resv, n) \
> > > > > + list_for_each_entry_safe((resv), (n), &(xnr)->resv_list, list)
> > > > > +
> > > > > +void xrep_newbt_init_bare(struct xrep_newbt *xnr, struct repair_ctx *sc);
> > > > > +void xrep_newbt_init_ag(struct xrep_newbt *xnr, struct repair_ctx *sc,
> > > > > + const struct xfs_owner_info *oinfo, xfs_fsblock_t alloc_hint,
> > > > > + enum xfs_ag_resv_type resv);
> > > > > +void xrep_newbt_init_inode(struct xrep_newbt *xnr, struct repair_ctx *sc,
> > > > > + int whichfork, const struct xfs_owner_info *oinfo);
> > > > > +int xrep_newbt_add_blocks(struct xrep_newbt *xnr, xfs_fsblock_t fsbno,
> > > > > + xfs_extlen_t len);
> > > > > +int xrep_newbt_alloc_blocks(struct xrep_newbt *xnr, uint64_t nr_blocks);
> > > > > +void xrep_newbt_destroy(struct xrep_newbt *xnr, int error);
> > > > > +int xrep_newbt_claim_block(struct xfs_btree_cur *cur, struct xrep_newbt *xnr,
> > > > > + union xfs_btree_ptr *ptr);
> > > > > +
> > > > > +#endif /* __XFS_REPAIR_BLOAD_H__ */
> > > > > diff --git a/repair/xfs_repair.c b/repair/xfs_repair.c
> > > > > index 9d72fa8e..8fbd3649 100644
> > > > > --- a/repair/xfs_repair.c
> > > > > +++ b/repair/xfs_repair.c
> > > > > @@ -24,6 +24,7 @@
> > > > > #include "rmap.h"
> > > > > #include "libfrog/fsgeom.h"
> > > > > #include "libfrog/platform.h"
> > > > > +#include "bload.h"
> > > > >
> > > > > /*
> > > > > * option tables for getsubopt calls
> > > > > @@ -39,6 +40,8 @@ enum o_opt_nums {
> > > > > AG_STRIDE,
> > > > > FORCE_GEO,
> > > > > PHASE2_THREADS,
> > > > > + BLOAD_LEAF_SLACK,
> > > > > + BLOAD_NODE_SLACK,
> > > > > O_MAX_OPTS,
> > > > > };
> > > > >
> > > > > @@ -49,6 +52,8 @@ static char *o_opts[] = {
> > > > > [AG_STRIDE] = "ag_stride",
> > > > > [FORCE_GEO] = "force_geometry",
> > > > > [PHASE2_THREADS] = "phase2_threads",
> > > > > + [BLOAD_LEAF_SLACK] = "debug_bload_leaf_slack",
> > > > > + [BLOAD_NODE_SLACK] = "debug_bload_node_slack",
> > > > > [O_MAX_OPTS] = NULL,
> > > > > };
> > > > >
> > > > > @@ -260,6 +265,18 @@ process_args(int argc, char **argv)
> > > > > _("-o phase2_threads requires a parameter\n"));
> > > > > phase2_threads = (int)strtol(val, NULL, 0);
> > > > > break;
> > > > > + case BLOAD_LEAF_SLACK:
> > > > > + if (!val)
> > > > > + do_abort(
> > > > > + _("-o debug_bload_leaf_slack requires a parameter\n"));
> > > > > + bload_leaf_slack = (int)strtol(val, NULL, 0);
> > > > > + break;
> > > > > + case BLOAD_NODE_SLACK:
> > > > > + if (!val)
> > > > > + do_abort(
> > > > > + _("-o debug_bload_node_slack requires a parameter\n"));
> > > > > + bload_node_slack = (int)strtol(val, NULL, 0);
> > > > > + break;
> > > > > default:
> > > > > unknown('o', val);
> > > > > break;
> > > > >
> > > >
> > >
> >
>
^ permalink raw reply [flat|nested] 30+ messages in thread
* Re: [PATCH 4/9] xfs_repair: rebuild free space btrees with bulk loader
2020-05-29 21:39 ` Darrick J. Wong
@ 2020-06-01 12:05 ` Brian Foster
2020-06-02 0:21 ` Darrick J. Wong
0 siblings, 1 reply; 30+ messages in thread
From: Brian Foster @ 2020-06-01 12:05 UTC (permalink / raw)
To: Darrick J. Wong; +Cc: sandeen, linux-xfs
On Fri, May 29, 2020 at 02:39:37PM -0700, Darrick J. Wong wrote:
> On Thu, May 28, 2020 at 11:10:44AM -0400, Brian Foster wrote:
> > On Tue, May 19, 2020 at 06:51:08PM -0700, Darrick J. Wong wrote:
> > > From: Darrick J. Wong <darrick.wong@oracle.com>
> > >
> > > Use the btree bulk loading functions to rebuild the free space btrees
> > > and drop the open-coded implementation.
> > >
> > > Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
> > > ---
> > > libxfs/libxfs_api_defs.h | 3
> > > repair/phase5.c | 870 +++++++++++++---------------------------------
> > > 2 files changed, 254 insertions(+), 619 deletions(-)
> > >
> > >
> > ...
> > > diff --git a/repair/phase5.c b/repair/phase5.c
> > > index 8f5e5f59..e69b042c 100644
> > > --- a/repair/phase5.c
> > > +++ b/repair/phase5.c
> > ...
> > > @@ -837,268 +567,202 @@ btnum_to_ops(
...
> > > +
> > > + /* Ok, now how many free space records do we have? */
> > > + *nr_extents = count_bno_extents_blocks(agno, &num_freeblocks);
> > > + } while (1);
> > > +
> > > + *extra_blocks = (bno_blocks - btr_bno->bload.nr_blocks) +
> > > + (cnt_blocks - btr_cnt->bload.nr_blocks);
> > > +}
> > > +
> > > +static void
> > > +get_freesp_data(
> > > + struct xfs_btree_cur *cur,
> > > + struct extent_tree_node *bno_rec,
> > > + xfs_agblock_t *freeblks)
> > > +{
> > > + struct xfs_alloc_rec_incore *arec = &cur->bc_rec.a;
> > > +
> > > + arec->ar_startblock = bno_rec->ex_startblock;
> > > + arec->ar_blockcount = bno_rec->ex_blockcount;
> > > + if (freeblks)
> > > + *freeblks += bno_rec->ex_blockcount;
> > > +}
> > > +
> > > +/* Grab one bnobt record. */
> > > +static int
> > > +get_bnobt_record(
> > > + struct xfs_btree_cur *cur,
> > > + void *priv)
> > > +{
> > > + struct bt_rebuild *btr = priv;
> > > +
> > > + get_freesp_data(cur, btr->bno_rec, btr->freeblks);
> > > + btr->bno_rec = findnext_bno_extent(btr->bno_rec);
> >
> > We should probably check for NULL here even if we don't expect it to
> > happen.
> >
> > Also, this logic where we load the current pointer and get the next had
> > my eyes crossed for a bit. Can we just check the pointer state here to
> > determine whether to find the first record or the next?
>
> Hm, I did that, and the logic ends up more cluttered looking:
>
> if (cur->bc_btnum == XFS_BTNUM_BNO) {
> if (btr->bno_rec == NULL)
> btr->bno_rec = findfirst_bno_extent(agno);
> else
> btr->bno_rec = findnext_bno_extent(btr->bno_rec);
> } else {
> if (btr->bno_rec == NULL)
> btr->bno_rec = findfirst_bcnt_extent(agno);
> else
> btr->bno_rec = findnext_bcnt_extent(cur->bc_ag.agno,
> btr->bno_rec);
> }
>
Yeah, I figured this approach would boil down to something like this in
the record feeding code. For one, I'd stick that in a function with a
brief comment. Otherwise, despite the cluttered look, the logic itself
is simple and much easier to read IMO. I had to stare at the current
code for a few minutes before I recognized what was going on, then read
through the iteration thing again to make sure I had it correct. This
eliminates that confusion entirely and thus simplifies the broader
implementation. I think it also facilitates reducing the amount of code
even more if we can generalize the build_*() functions a bit more..
> OTOH I think you do have a good point that splitting the code that sets
> bno_rec across three functions is not easy to understand.
>
Exactly.
> > Finally, I notice we duplicate the build_[bno|cnt]bt() functions for
> > what amounts to a different get_record() callback specific to the tree
> > type. If we also generalize get_record() to use the tree type (it
> > already has the cursor), it seems that much of the duplication can be
> > reduced and the logic simplified.
>
> Ok, done.
>
> > > + return 0;
> > > +}
> > > +
> > > +/* Rebuild a free space by block number btree. */
> > > +static void
> > > +build_bnobt(
> > > + struct repair_ctx *sc,
> > > + xfs_agnumber_t agno,
> > > + struct bt_rebuild *btr_bno,
> > > + xfs_agblock_t *freeblks)
> > > +{
> > > + int error;
> > > +
> > > + *freeblks = 0;
> > > + btr_bno->bload.get_record = get_bnobt_record;
> > > + btr_bno->bload.claim_block = rebuild_claim_block;
> >
> > Note that one of these callbacks is introduced in this patch and the
> > other in the previous. I've no issue with splitting the patches as noted
> > in the previous patch, but these should probably be consistent one way
> > or the other.
>
> All the btree types set ->claim_block to rebuild_claim_block, whereas
> ->get_record is specific to the particular type of btree.
>
Ok, seems more reasonable in hindsight.
Brian
> > > + btr_bno->bno_rec = findfirst_bno_extent(agno);
> > > + btr_bno->freeblks = freeblks;
> > > +
> > > + error = -libxfs_trans_alloc_empty(sc->mp, &sc->tp);
> > > + if (error)
> > > + do_error(
> > > +_("Insufficient memory to construct bnobt rebuild transaction.\n"));
> > > +
> > > + /* Add all observed bnobt records. */
> > > + error = -libxfs_btree_bload(btr_bno->cur, &btr_bno->bload, btr_bno);
> > > + if (error)
> > > + do_error(
> > > +_("Error %d while creating bnobt btree for AG %u.\n"), error, agno);
> > > +
> > > + /* Since we're not writing the AGF yet, no need to commit the cursor */
> > > + libxfs_btree_del_cursor(btr_bno->cur, 0);
> > > + error = -libxfs_trans_commit(sc->tp);
> > > + if (error)
> > > + do_error(
> > > +_("Error %d while writing bnobt btree for AG %u.\n"), error, agno);
> > > + sc->tp = NULL;
> > > +}
> > > +
> > ...
> > > @@ -2354,48 +2041,14 @@ build_agf_agfl(
> > > freelist[i] = cpu_to_be32(NULLAGBLOCK);
> > > }
> > >
> > > - /*
> > > - * do we have left-over blocks in the btree cursors that should
> > > - * be used to fill the AGFL?
> > > - */
> > > - if (bno_bt->num_free_blocks > 0 || bcnt_bt->num_free_blocks > 0) {
> > > - /*
> > > - * yes, now grab as many blocks as we can
> > > - */
> > > - i = 0;
> > > - while (bno_bt->num_free_blocks > 0 && i < libxfs_agfl_size(mp))
> > > - {
> > > - freelist[i] = cpu_to_be32(
> > > - get_next_blockaddr(agno, 0, bno_bt));
> > > - i++;
> > > - }
> > > -
> > > - while (bcnt_bt->num_free_blocks > 0 && i < libxfs_agfl_size(mp))
> > > - {
> > > - freelist[i] = cpu_to_be32(
> > > - get_next_blockaddr(agno, 0, bcnt_bt));
> > > - i++;
> > > - }
> > > - /*
> > > - * now throw the rest of the blocks away and complain
> > > - */
> > > - while (bno_bt->num_free_blocks > 0) {
> > > - fsb = XFS_AGB_TO_FSB(mp, agno,
> > > - get_next_blockaddr(agno, 0, bno_bt));
> > > - error = slab_add(lost_fsb, &fsb);
> > > - if (error)
> > > - do_error(
> > > -_("Insufficient memory saving lost blocks.\n"));
> > > - }
> > > - while (bcnt_bt->num_free_blocks > 0) {
> > > - fsb = XFS_AGB_TO_FSB(mp, agno,
> > > - get_next_blockaddr(agno, 0, bcnt_bt));
> > > - error = slab_add(lost_fsb, &fsb);
> > > - if (error)
> > > - do_error(
> > > -_("Insufficient memory saving lost blocks.\n"));
> > > - }
> > > + /* Fill the AGFL with leftover blocks or save them for later. */
> > > + i = 0;
> >
> > How about agfl_idx or something since this variable is passed down into
> > a function?
>
> The 'i' variable existed before this patch. I can add a new cleanup to
> fix the name...
>
> > > + freelist = xfs_buf_to_agfl_bno(agfl_buf);
> > > + fill_agfl(btr_bno, freelist, &i);
> > > + fill_agfl(btr_cnt, freelist, &i);
> > >
> > > + /* Set the AGF counters for the AGFL. */
> > > + if (i > 0) {
> > > agf->agf_flfirst = 0;
> > > agf->agf_fllast = cpu_to_be32(i - 1);
> > > agf->agf_flcount = cpu_to_be32(i);
> > ...
> > > @@ -2650,9 +2283,9 @@ _("unable to rebuild AG %u. Not enough free space in on-disk AG.\n"),
> > > /*
> > > * set up agf and agfl
> > > */
> > > - build_agf_agfl(mp, agno, &bno_btree_curs,
> > > - &bcnt_btree_curs, freeblks1, extra_blocks,
> > > - &rmap_btree_curs, &refcnt_btree_curs, lost_fsb);
> > > + build_agf_agfl(mp, agno, &btr_bno, &btr_cnt, freeblks1, extra_blocks,
> > > + &rmap_btree_curs, &refcnt_btree_curs, lost_fsb);
> > > +
> >
> > I was trying to figure out what extra_blocks was used for and noticed
> > that it's not even used by this function. Perhaps a preceding cleanup
> > patch to drop it?
>
> Ok.
>
> > > /*
> > > * build inode allocation tree.
> > > */
> > > @@ -2674,15 +2307,14 @@ _("unable to rebuild AG %u. Not enough free space in on-disk AG.\n"),
> > > /*
> > > * tear down cursors
> > > */
> > > - finish_cursor(&bno_btree_curs);
> > > - finish_cursor(&ino_btree_curs);
> >
> > Looks like this one shouldn't be going away here.
>
> Oops, good catch.
>
> --D
>
> > Brian
> >
> > > + finish_rebuild(mp, &btr_bno, lost_fsb);
> > > + finish_rebuild(mp, &btr_cnt, lost_fsb);
> > > if (xfs_sb_version_hasrmapbt(&mp->m_sb))
> > > finish_cursor(&rmap_btree_curs);
> > > if (xfs_sb_version_hasreflink(&mp->m_sb))
> > > finish_cursor(&refcnt_btree_curs);
> > > if (xfs_sb_version_hasfinobt(&mp->m_sb))
> > > finish_cursor(&fino_btree_curs);
> > > - finish_cursor(&bcnt_btree_curs);
> > >
> > > /*
> > > * release the incore per-AG bno/bcnt trees so the extent nodes
> > >
> >
>
^ permalink raw reply [flat|nested] 30+ messages in thread
* Re: [PATCH 1/9] xfs_repair: port the online repair newbt structure
2020-06-01 12:03 ` Brian Foster
@ 2020-06-02 0:12 ` Darrick J. Wong
0 siblings, 0 replies; 30+ messages in thread
From: Darrick J. Wong @ 2020-06-02 0:12 UTC (permalink / raw)
To: Brian Foster; +Cc: sandeen, linux-xfs
On Mon, Jun 01, 2020 at 08:03:38AM -0400, Brian Foster wrote:
> On Fri, May 29, 2020 at 02:01:59PM -0700, Darrick J. Wong wrote:
> > On Thu, May 28, 2020 at 11:08:36AM -0400, Brian Foster wrote:
> > > On Wed, May 27, 2020 at 03:34:24PM -0700, Darrick J. Wong wrote:
> > > > On Wed, May 27, 2020 at 08:15:31AM -0400, Brian Foster wrote:
> > > > > On Tue, May 19, 2020 at 06:50:49PM -0700, Darrick J. Wong wrote:
> > > > > > From: Darrick J. Wong <darrick.wong@oracle.com>
> > > > > >
> > > > > > Port the new btree staging context and related block reservation helper
> > > > > > code from the kernel to repair. We'll use this in subsequent patches to
> > > > > > implement btree bulk loading.
> > > > > >
> > > > > > Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
> > > > > > ---
> > > > > > include/libxfs.h | 1
> > > > > > libxfs/libxfs_api_defs.h | 2
> > > > > > repair/Makefile | 4 -
> > > > > > repair/bload.c | 303 ++++++++++++++++++++++++++++++++++++++++++++++
> > > > > > repair/bload.h | 77 ++++++++++++
> > > > > > repair/xfs_repair.c | 17 +++
> > > > > > 6 files changed, 402 insertions(+), 2 deletions(-)
> > > > > > create mode 100644 repair/bload.c
> > > > > > create mode 100644 repair/bload.h
> > > > > >
> > > > > >
> > > > > ...
> > > > > > diff --git a/repair/bload.c b/repair/bload.c
> > > > > > new file mode 100644
> > > > > > index 00000000..9bc17468
> > > > > > --- /dev/null
> > > > > > +++ b/repair/bload.c
> > > > > > @@ -0,0 +1,303 @@
> > > > > > +// SPDX-License-Identifier: GPL-2.0-or-later
> > > > > > +/*
> > > > > > + * Copyright (C) 2020 Oracle. All Rights Reserved.
> > > > > > + * Author: Darrick J. Wong <darrick.wong@oracle.com>
> > > > > > + */
> > > > > > +#include <libxfs.h>
> > > > > > +#include "bload.h"
> > > > > > +
> > > > > > +#define trace_xrep_newbt_claim_block(...) ((void) 0)
> > > > > > +#define trace_xrep_newbt_free_blocks(...) ((void) 0)
> > > > > > +
> > > > > > +int bload_leaf_slack = -1;
> > > > > > +int bload_node_slack = -1;
> > > > > > +
> > > > > > +/* Ported routines from fs/xfs/scrub/repair.c */
> > > > > > +
> > > > >
> > > > > Looks mostly straightforward, but I'll have to come back to this as I
> > > > > get to the code that uses it later in the series. In the meantime, I see
> > > > > some of these helpers in scrub/repair.c while not others. Are there
> > > > > references to other routines that are intended to be copies from kernel
> > > > > code?
> > > >
> > > > Hm. I might not understand the question, but in general the code should
> > > > be fairly similar to the kernel functions. The biggest differences are
> > > > (a) that whole libxfs error code mess, (b) the much simpler repair_ctx
> > > > structure, and (c) the fact that repair doesn't bother with EFIs to
> > > > automatically reap blocks.
> > > >
> > > > So... the ten functions you see here do the same things as their kernel
> > > > counterparts, but they get to do it in the much simpler userspace
> > > > environment.
> > > >
> > >
> > > Right.. I was able to find the first function (xrep_roll_ag_trans())
> > > easily in the kernel because it has the same name. The next one or two
> > > (i.e., xrep_newbt_*()) I couldn't find and then gave up. Are they
> > > renamed? Unmerged?
> >
> > Ahh, silly me, the rest are as yet unmerged components of the online
> > repair code. Maybe it makes more sense to drop the "Ported routines
> > from XXXX" comment, particularly since it probably will be easier to
> > merge the xfs_repair series first, which is /much/ smaller in scope.
> >
>
> Probably so. It's just going to confuse if this is not intended to land
> at the same time as the kernel code.
<nod> I also realized while re-examining the two codebases that the
xrep_newbt code *isn't* going to converge. The kernel needs extra bits
for smuggling in EFIs, and userspace doesn't ever need the ability to
bhold the AG header buffers to roll a transaction.
So in effect I dropped a bunch of code and re-prefixed the surviving
code so that this patch will get quite a bit smaller. I'll try to send
a new revision tomorrow.
--D
> > > > The other functions in scrub/repair.c that didn't get ported are either
> > > > for other types of repairs or exist to support the in-kernel code and
> > > > aren't needed here.
> > > >
> > >
> > > Sure, I'm just curious how to identify the source of the ones that are.
> >
> > https://git.kernel.org/pub/scm/linux/kernel/git/djwong/xfs-linux.git/tree/fs/xfs/scrub/repair.c?h=djwong-wtf
> >
> > Is a fairly recent snapshot of what the kernel version looks like.
> >
>
> Thanks.
>
> Brian
>
> > --D
> >
> > >
> > > Brian
> > >
> > > > --D
> > > >
> > > > > Brian
> > > > >
> > > > > > +/*
> > > > > > + * Roll a transaction, keeping the AG headers locked and reinitializing
> > > > > > + * the btree cursors.
> > > > > > + */
> > > > > > +int
> > > > > > +xrep_roll_ag_trans(
> > > > > > + struct repair_ctx *sc)
> > > > > > +{
> > > > > > + int error;
> > > > > > +
> > > > > > + /* Keep the AG header buffers locked so we can keep going. */
> > > > > > + if (sc->agi_bp)
> > > > > > + libxfs_trans_bhold(sc->tp, sc->agi_bp);
> > > > > > + if (sc->agf_bp)
> > > > > > + libxfs_trans_bhold(sc->tp, sc->agf_bp);
> > > > > > + if (sc->agfl_bp)
> > > > > > + libxfs_trans_bhold(sc->tp, sc->agfl_bp);
> > > > > > +
> > > > > > + /*
> > > > > > + * Roll the transaction. We still own the buffer and the buffer lock
> > > > > > + * regardless of whether or not the roll succeeds. If the roll fails,
> > > > > > + * the buffers will be released during teardown on our way out of the
> > > > > > + * kernel. If it succeeds, we join them to the new transaction and
> > > > > > + * move on.
> > > > > > + */
> > > > > > + error = -libxfs_trans_roll(&sc->tp);
> > > > > > + if (error)
> > > > > > + return error;
> > > > > > +
> > > > > > + /* Join AG headers to the new transaction. */
> > > > > > + if (sc->agi_bp)
> > > > > > + libxfs_trans_bjoin(sc->tp, sc->agi_bp);
> > > > > > + if (sc->agf_bp)
> > > > > > + libxfs_trans_bjoin(sc->tp, sc->agf_bp);
> > > > > > + if (sc->agfl_bp)
> > > > > > + libxfs_trans_bjoin(sc->tp, sc->agfl_bp);
> > > > > > +
> > > > > > + return 0;
> > > > > > +}
> > > > > > +
> > > > > > +/* Initialize accounting resources for staging a new AG btree. */
> > > > > > +void
> > > > > > +xrep_newbt_init_ag(
> > > > > > + struct xrep_newbt *xnr,
> > > > > > + struct repair_ctx *sc,
> > > > > > + const struct xfs_owner_info *oinfo,
> > > > > > + xfs_fsblock_t alloc_hint,
> > > > > > + enum xfs_ag_resv_type resv)
> > > > > > +{
> > > > > > + memset(xnr, 0, sizeof(struct xrep_newbt));
> > > > > > + xnr->sc = sc;
> > > > > > + xnr->oinfo = *oinfo; /* structure copy */
> > > > > > + xnr->alloc_hint = alloc_hint;
> > > > > > + xnr->resv = resv;
> > > > > > + INIT_LIST_HEAD(&xnr->resv_list);
> > > > > > +}
> > > > > > +
> > > > > > +/* Initialize accounting resources for staging a new inode fork btree. */
> > > > > > +void
> > > > > > +xrep_newbt_init_inode(
> > > > > > + struct xrep_newbt *xnr,
> > > > > > + struct repair_ctx *sc,
> > > > > > + int whichfork,
> > > > > > + const struct xfs_owner_info *oinfo)
> > > > > > +{
> > > > > > + xrep_newbt_init_ag(xnr, sc, oinfo,
> > > > > > + XFS_INO_TO_FSB(sc->mp, sc->ip->i_ino),
> > > > > > + XFS_AG_RESV_NONE);
> > > > > > + xnr->ifake.if_fork = kmem_zone_zalloc(xfs_ifork_zone, 0);
> > > > > > + xnr->ifake.if_fork_size = XFS_IFORK_SIZE(sc->ip, whichfork);
> > > > > > +}
> > > > > > +
> > > > > > +/*
> > > > > > + * Initialize accounting resources for staging a new btree. Callers are
> > > > > > + * expected to add their own reservations (and clean them up) manually.
> > > > > > + */
> > > > > > +void
> > > > > > +xrep_newbt_init_bare(
> > > > > > + struct xrep_newbt *xnr,
> > > > > > + struct repair_ctx *sc)
> > > > > > +{
> > > > > > + xrep_newbt_init_ag(xnr, sc, &XFS_RMAP_OINFO_ANY_OWNER, NULLFSBLOCK,
> > > > > > + XFS_AG_RESV_NONE);
> > > > > > +}
> > > > > > +
> > > > > > +/* Designate specific blocks to be used to build our new btree. */
> > > > > > +int
> > > > > > +xrep_newbt_add_blocks(
> > > > > > + struct xrep_newbt *xnr,
> > > > > > + xfs_fsblock_t fsbno,
> > > > > > + xfs_extlen_t len)
> > > > > > +{
> > > > > > + struct xrep_newbt_resv *resv;
> > > > > > +
> > > > > > + resv = kmem_alloc(sizeof(struct xrep_newbt_resv), KM_MAYFAIL);
> > > > > > + if (!resv)
> > > > > > + return ENOMEM;
> > > > > > +
> > > > > > + INIT_LIST_HEAD(&resv->list);
> > > > > > + resv->fsbno = fsbno;
> > > > > > + resv->len = len;
> > > > > > + resv->used = 0;
> > > > > > + list_add_tail(&resv->list, &xnr->resv_list);
> > > > > > + return 0;
> > > > > > +}
> > > > > > +
> > > > > > +/* Reserve disk space for our new btree. */
> > > > > > +int
> > > > > > +xrep_newbt_alloc_blocks(
> > > > > > + struct xrep_newbt *xnr,
> > > > > > + uint64_t nr_blocks)
> > > > > > +{
> > > > > > + struct repair_ctx *sc = xnr->sc;
> > > > > > + xfs_alloctype_t type;
> > > > > > + xfs_fsblock_t alloc_hint = xnr->alloc_hint;
> > > > > > + int error = 0;
> > > > > > +
> > > > > > + type = sc->ip ? XFS_ALLOCTYPE_START_BNO : XFS_ALLOCTYPE_NEAR_BNO;
> > > > > > +
> > > > > > + while (nr_blocks > 0 && !error) {
> > > > > > + struct xfs_alloc_arg args = {
> > > > > > + .tp = sc->tp,
> > > > > > + .mp = sc->mp,
> > > > > > + .type = type,
> > > > > > + .fsbno = alloc_hint,
> > > > > > + .oinfo = xnr->oinfo,
> > > > > > + .minlen = 1,
> > > > > > + .maxlen = nr_blocks,
> > > > > > + .prod = 1,
> > > > > > + .resv = xnr->resv,
> > > > > > + };
> > > > > > +
> > > > > > + error = -libxfs_alloc_vextent(&args);
> > > > > > + if (error)
> > > > > > + return error;
> > > > > > + if (args.fsbno == NULLFSBLOCK)
> > > > > > + return ENOSPC;
> > > > > > +
> > > > > > + /* We don't have real EFIs here so skip that. */
> > > > > > +
> > > > > > + error = xrep_newbt_add_blocks(xnr, args.fsbno, args.len);
> > > > > > + if (error)
> > > > > > + break;
> > > > > > +
> > > > > > + nr_blocks -= args.len;
> > > > > > + alloc_hint = args.fsbno + args.len - 1;
> > > > > > +
> > > > > > + if (sc->ip)
> > > > > > + error = -libxfs_trans_roll_inode(&sc->tp, sc->ip);
> > > > > > + else
> > > > > > + error = xrep_roll_ag_trans(sc);
> > > > > > + }
> > > > > > +
> > > > > > + return error;
> > > > > > +}
> > > > > > +
> > > > > > +/*
> > > > > > + * Release blocks that were reserved for a btree repair. If the repair
> > > > > > + * succeeded then we log deferred frees for unused blocks. Otherwise, we try
> > > > > > + * to free the extents immediately to roll the filesystem back to where it was
> > > > > > + * before we started.
> > > > > > + */
> > > > > > +static inline int
> > > > > > +xrep_newbt_destroy_reservation(
> > > > > > + struct xrep_newbt *xnr,
> > > > > > + struct xrep_newbt_resv *resv,
> > > > > > + bool cancel_repair)
> > > > > > +{
> > > > > > + struct repair_ctx *sc = xnr->sc;
> > > > > > +
> > > > > > + if (cancel_repair) {
> > > > > > + int error;
> > > > > > +
> > > > > > + /* Free the extent then roll the transaction. */
> > > > > > + error = -libxfs_free_extent(sc->tp, resv->fsbno, resv->len,
> > > > > > + &xnr->oinfo, xnr->resv);
> > > > > > + if (error)
> > > > > > + return error;
> > > > > > +
> > > > > > + if (sc->ip)
> > > > > > + return -libxfs_trans_roll_inode(&sc->tp, sc->ip);
> > > > > > + return xrep_roll_ag_trans(sc);
> > > > > > + }
> > > > > > +
> > > > > > + /* We don't have EFIs here so skip the EFD. */
> > > > > > +
> > > > > > + /*
> > > > > > + * Use the deferred freeing mechanism to schedule for deletion any
> > > > > > + * blocks we didn't use to rebuild the tree. This enables us to log
> > > > > > + * them all in the same transaction as the root change.
> > > > > > + */
> > > > > > + resv->fsbno += resv->used;
> > > > > > + resv->len -= resv->used;
> > > > > > + resv->used = 0;
> > > > > > +
> > > > > > + if (resv->len == 0)
> > > > > > + return 0;
> > > > > > +
> > > > > > + trace_xrep_newbt_free_blocks(sc->mp,
> > > > > > + XFS_FSB_TO_AGNO(sc->mp, resv->fsbno),
> > > > > > + XFS_FSB_TO_AGBNO(sc->mp, resv->fsbno),
> > > > > > + resv->len, xnr->oinfo.oi_owner);
> > > > > > +
> > > > > > + __xfs_bmap_add_free(sc->tp, resv->fsbno, resv->len, &xnr->oinfo, true);
> > > > > > +
> > > > > > + return 0;
> > > > > > +}
> > > > > > +
> > > > > > +/* Free all the accounting info and disk space we reserved for a new btree. */
> > > > > > +void
> > > > > > +xrep_newbt_destroy(
> > > > > > + struct xrep_newbt *xnr,
> > > > > > + int error)
> > > > > > +{
> > > > > > + struct repair_ctx *sc = xnr->sc;
> > > > > > + struct xrep_newbt_resv *resv, *n;
> > > > > > + int err2;
> > > > > > +
> > > > > > + list_for_each_entry_safe(resv, n, &xnr->resv_list, list) {
> > > > > > + err2 = xrep_newbt_destroy_reservation(xnr, resv, error != 0);
> > > > > > + if (err2)
> > > > > > + goto junkit;
> > > > > > +
> > > > > > + list_del(&resv->list);
> > > > > > + kmem_free(resv);
> > > > > > + }
> > > > > > +
> > > > > > +junkit:
> > > > > > + /*
> > > > > > + * If we still have reservations attached to @newbt, cleanup must have
> > > > > > + * failed and the filesystem is about to go down. Clean up the incore
> > > > > > + * reservations.
> > > > > > + */
> > > > > > + list_for_each_entry_safe(resv, n, &xnr->resv_list, list) {
> > > > > > + list_del(&resv->list);
> > > > > > + kmem_free(resv);
> > > > > > + }
> > > > > > +
> > > > > > + if (sc->ip) {
> > > > > > + kmem_cache_free(xfs_ifork_zone, xnr->ifake.if_fork);
> > > > > > + xnr->ifake.if_fork = NULL;
> > > > > > + }
> > > > > > +}
> > > > > > +
> > > > > > +/* Feed one of the reserved btree blocks to the bulk loader. */
> > > > > > +int
> > > > > > +xrep_newbt_claim_block(
> > > > > > + struct xfs_btree_cur *cur,
> > > > > > + struct xrep_newbt *xnr,
> > > > > > + union xfs_btree_ptr *ptr)
> > > > > > +{
> > > > > > + struct xrep_newbt_resv *resv;
> > > > > > + xfs_fsblock_t fsb;
> > > > > > +
> > > > > > + /*
> > > > > > + * The first item in the list should always have a free block unless
> > > > > > + * we're completely out.
> > > > > > + */
> > > > > > + resv = list_first_entry(&xnr->resv_list, struct xrep_newbt_resv, list);
> > > > > > + if (resv->used == resv->len)
> > > > > > + return ENOSPC;
> > > > > > +
> > > > > > + /*
> > > > > > + * Peel off a block from the start of the reservation. We allocate
> > > > > > + * blocks in order to place blocks on disk in increasing record or key
> > > > > > + * order. The block reservations tend to end up on the list in
> > > > > > + * decreasing order, which hopefully results in leaf blocks ending up
> > > > > > + * together.
> > > > > > + */
> > > > > > + fsb = resv->fsbno + resv->used;
> > > > > > + resv->used++;
> > > > > > +
> > > > > > + /* If we used all the blocks in this reservation, move it to the end. */
> > > > > > + if (resv->used == resv->len)
> > > > > > + list_move_tail(&resv->list, &xnr->resv_list);
> > > > > > +
> > > > > > + trace_xrep_newbt_claim_block(cur->bc_mp,
> > > > > > + XFS_FSB_TO_AGNO(cur->bc_mp, fsb),
> > > > > > + XFS_FSB_TO_AGBNO(cur->bc_mp, fsb),
> > > > > > + 1, xnr->oinfo.oi_owner);
> > > > > > +
> > > > > > + if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
> > > > > > + ptr->l = cpu_to_be64(fsb);
> > > > > > + else
> > > > > > + ptr->s = cpu_to_be32(XFS_FSB_TO_AGBNO(cur->bc_mp, fsb));
> > > > > > + return 0;
> > > > > > +}
> > > > > > diff --git a/repair/bload.h b/repair/bload.h
> > > > > > new file mode 100644
> > > > > > index 00000000..020c4834
> > > > > > --- /dev/null
> > > > > > +++ b/repair/bload.h
> > > > > > @@ -0,0 +1,77 @@
> > > > > > +// SPDX-License-Identifier: GPL-2.0-or-later
> > > > > > +/*
> > > > > > + * Copyright (C) 2020 Oracle. All Rights Reserved.
> > > > > > + * Author: Darrick J. Wong <darrick.wong@oracle.com>
> > > > > > + */
> > > > > > +#ifndef __XFS_REPAIR_BLOAD_H__
> > > > > > +#define __XFS_REPAIR_BLOAD_H__
> > > > > > +
> > > > > > +extern int bload_leaf_slack;
> > > > > > +extern int bload_node_slack;
> > > > > > +
> > > > > > +struct repair_ctx {
> > > > > > + struct xfs_mount *mp;
> > > > > > + struct xfs_inode *ip;
> > > > > > + struct xfs_trans *tp;
> > > > > > +
> > > > > > + struct xfs_buf *agi_bp;
> > > > > > + struct xfs_buf *agf_bp;
> > > > > > + struct xfs_buf *agfl_bp;
> > > > > > +};
> > > > > > +
> > > > > > +struct xrep_newbt_resv {
> > > > > > + /* Link to list of extents that we've reserved. */
> > > > > > + struct list_head list;
> > > > > > +
> > > > > > + /* FSB of the block we reserved. */
> > > > > > + xfs_fsblock_t fsbno;
> > > > > > +
> > > > > > + /* Length of the reservation. */
> > > > > > + xfs_extlen_t len;
> > > > > > +
> > > > > > + /* How much of this reservation we've used. */
> > > > > > + xfs_extlen_t used;
> > > > > > +};
> > > > > > +
> > > > > > +struct xrep_newbt {
> > > > > > + struct repair_ctx *sc;
> > > > > > +
> > > > > > + /* List of extents that we've reserved. */
> > > > > > + struct list_head resv_list;
> > > > > > +
> > > > > > + /* Fake root for new btree. */
> > > > > > + union {
> > > > > > + struct xbtree_afakeroot afake;
> > > > > > + struct xbtree_ifakeroot ifake;
> > > > > > + };
> > > > > > +
> > > > > > + /* rmap owner of these blocks */
> > > > > > + struct xfs_owner_info oinfo;
> > > > > > +
> > > > > > + /* The last reservation we allocated from. */
> > > > > > + struct xrep_newbt_resv *last_resv;
> > > > > > +
> > > > > > + /* Allocation hint */
> > > > > > + xfs_fsblock_t alloc_hint;
> > > > > > +
> > > > > > + /* per-ag reservation type */
> > > > > > + enum xfs_ag_resv_type resv;
> > > > > > +};
> > > > > > +
> > > > > > +#define for_each_xrep_newbt_reservation(xnr, resv, n) \
> > > > > > + list_for_each_entry_safe((resv), (n), &(xnr)->resv_list, list)
> > > > > > +
> > > > > > +void xrep_newbt_init_bare(struct xrep_newbt *xnr, struct repair_ctx *sc);
> > > > > > +void xrep_newbt_init_ag(struct xrep_newbt *xnr, struct repair_ctx *sc,
> > > > > > + const struct xfs_owner_info *oinfo, xfs_fsblock_t alloc_hint,
> > > > > > + enum xfs_ag_resv_type resv);
> > > > > > +void xrep_newbt_init_inode(struct xrep_newbt *xnr, struct repair_ctx *sc,
> > > > > > + int whichfork, const struct xfs_owner_info *oinfo);
> > > > > > +int xrep_newbt_add_blocks(struct xrep_newbt *xnr, xfs_fsblock_t fsbno,
> > > > > > + xfs_extlen_t len);
> > > > > > +int xrep_newbt_alloc_blocks(struct xrep_newbt *xnr, uint64_t nr_blocks);
> > > > > > +void xrep_newbt_destroy(struct xrep_newbt *xnr, int error);
> > > > > > +int xrep_newbt_claim_block(struct xfs_btree_cur *cur, struct xrep_newbt *xnr,
> > > > > > + union xfs_btree_ptr *ptr);
> > > > > > +
> > > > > > +#endif /* __XFS_REPAIR_BLOAD_H__ */
> > > > > > diff --git a/repair/xfs_repair.c b/repair/xfs_repair.c
> > > > > > index 9d72fa8e..8fbd3649 100644
> > > > > > --- a/repair/xfs_repair.c
> > > > > > +++ b/repair/xfs_repair.c
> > > > > > @@ -24,6 +24,7 @@
> > > > > > #include "rmap.h"
> > > > > > #include "libfrog/fsgeom.h"
> > > > > > #include "libfrog/platform.h"
> > > > > > +#include "bload.h"
> > > > > >
> > > > > > /*
> > > > > > * option tables for getsubopt calls
> > > > > > @@ -39,6 +40,8 @@ enum o_opt_nums {
> > > > > > AG_STRIDE,
> > > > > > FORCE_GEO,
> > > > > > PHASE2_THREADS,
> > > > > > + BLOAD_LEAF_SLACK,
> > > > > > + BLOAD_NODE_SLACK,
> > > > > > O_MAX_OPTS,
> > > > > > };
> > > > > >
> > > > > > @@ -49,6 +52,8 @@ static char *o_opts[] = {
> > > > > > [AG_STRIDE] = "ag_stride",
> > > > > > [FORCE_GEO] = "force_geometry",
> > > > > > [PHASE2_THREADS] = "phase2_threads",
> > > > > > + [BLOAD_LEAF_SLACK] = "debug_bload_leaf_slack",
> > > > > > + [BLOAD_NODE_SLACK] = "debug_bload_node_slack",
> > > > > > [O_MAX_OPTS] = NULL,
> > > > > > };
> > > > > >
> > > > > > @@ -260,6 +265,18 @@ process_args(int argc, char **argv)
> > > > > > _("-o phase2_threads requires a parameter\n"));
> > > > > > phase2_threads = (int)strtol(val, NULL, 0);
> > > > > > break;
> > > > > > + case BLOAD_LEAF_SLACK:
> > > > > > + if (!val)
> > > > > > + do_abort(
> > > > > > + _("-o debug_bload_leaf_slack requires a parameter\n"));
> > > > > > + bload_leaf_slack = (int)strtol(val, NULL, 0);
> > > > > > + break;
> > > > > > + case BLOAD_NODE_SLACK:
> > > > > > + if (!val)
> > > > > > + do_abort(
> > > > > > + _("-o debug_bload_node_slack requires a parameter\n"));
> > > > > > + bload_node_slack = (int)strtol(val, NULL, 0);
> > > > > > + break;
> > > > > > default:
> > > > > > unknown('o', val);
> > > > > > break;
> > > > > >
> > > > >
> > > >
> > >
> >
>
^ permalink raw reply [flat|nested] 30+ messages in thread
* Re: [PATCH 4/9] xfs_repair: rebuild free space btrees with bulk loader
2020-06-01 12:05 ` Brian Foster
@ 2020-06-02 0:21 ` Darrick J. Wong
0 siblings, 0 replies; 30+ messages in thread
From: Darrick J. Wong @ 2020-06-02 0:21 UTC (permalink / raw)
To: Brian Foster; +Cc: sandeen, linux-xfs
On Mon, Jun 01, 2020 at 08:05:25AM -0400, Brian Foster wrote:
> On Fri, May 29, 2020 at 02:39:37PM -0700, Darrick J. Wong wrote:
> > On Thu, May 28, 2020 at 11:10:44AM -0400, Brian Foster wrote:
> > > On Tue, May 19, 2020 at 06:51:08PM -0700, Darrick J. Wong wrote:
> > > > From: Darrick J. Wong <darrick.wong@oracle.com>
> > > >
> > > > Use the btree bulk loading functions to rebuild the free space btrees
> > > > and drop the open-coded implementation.
> > > >
> > > > Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
> > > > ---
> > > > libxfs/libxfs_api_defs.h | 3
> > > > repair/phase5.c | 870 +++++++++++++---------------------------------
> > > > 2 files changed, 254 insertions(+), 619 deletions(-)
> > > >
> > > >
> > > ...
> > > > diff --git a/repair/phase5.c b/repair/phase5.c
> > > > index 8f5e5f59..e69b042c 100644
> > > > --- a/repair/phase5.c
> > > > +++ b/repair/phase5.c
> > > ...
> > > > @@ -837,268 +567,202 @@ btnum_to_ops(
> ...
> > > > +
> > > > + /* Ok, now how many free space records do we have? */
> > > > + *nr_extents = count_bno_extents_blocks(agno, &num_freeblocks);
> > > > + } while (1);
> > > > +
> > > > + *extra_blocks = (bno_blocks - btr_bno->bload.nr_blocks) +
> > > > + (cnt_blocks - btr_cnt->bload.nr_blocks);
> > > > +}
> > > > +
> > > > +static void
> > > > +get_freesp_data(
> > > > + struct xfs_btree_cur *cur,
> > > > + struct extent_tree_node *bno_rec,
> > > > + xfs_agblock_t *freeblks)
> > > > +{
> > > > + struct xfs_alloc_rec_incore *arec = &cur->bc_rec.a;
> > > > +
> > > > + arec->ar_startblock = bno_rec->ex_startblock;
> > > > + arec->ar_blockcount = bno_rec->ex_blockcount;
> > > > + if (freeblks)
> > > > + *freeblks += bno_rec->ex_blockcount;
> > > > +}
> > > > +
> > > > +/* Grab one bnobt record. */
> > > > +static int
> > > > +get_bnobt_record(
> > > > + struct xfs_btree_cur *cur,
> > > > + void *priv)
> > > > +{
> > > > + struct bt_rebuild *btr = priv;
> > > > +
> > > > + get_freesp_data(cur, btr->bno_rec, btr->freeblks);
> > > > + btr->bno_rec = findnext_bno_extent(btr->bno_rec);
> > >
> > > We should probably check for NULL here even if we don't expect it to
> > > happen.
> > >
> > > Also, this logic where we load the current pointer and get the next had
> > > my eyes crossed for a bit. Can we just check the pointer state here to
> > > determine whether to find the first record or the next?
> >
> > Hm, I did that, and the logic ends up more cluttered looking:
> >
> > if (cur->bc_btnum == XFS_BTNUM_BNO) {
> > if (btr->bno_rec == NULL)
> > btr->bno_rec = findfirst_bno_extent(agno);
> > else
> > btr->bno_rec = findnext_bno_extent(btr->bno_rec);
> > } else {
> > if (btr->bno_rec == NULL)
> > btr->bno_rec = findfirst_bcnt_extent(agno);
> > else
> > btr->bno_rec = findnext_bcnt_extent(cur->bc_ag.agno,
> > btr->bno_rec);
> > }
> >
>
> Yeah, I figured this approach would boil down to something like this in
> the record feeding code. For one, I'd stick that in a function with a
> brief comment. Otherwise, despite the cluttered look, the logic itself
> is simple and much easier to read IMO. I had to stare at the current
> code for a few minutes before I recognized what was going on, then read
> through the iteration thing again to make sure I had it correct. This
> eliminates that confusion entirely and thus simplifies the broader
> implementation. I think it also facilitates reducing the amount of code
> even more if we can generalize the build_*() functions a bit more..
<nod>
> > OTOH I think you do have a good point that splitting the code that sets
> > bno_rec across three functions is not easy to understand.
> >
>
> Exactly.
>
> > > Finally, I notice we duplicate the build_[bno|cnt]bt() functions for
> > > what amounts to a different get_record() callback specific to the tree
> > > type. If we also generalize get_record() to use the tree type (it
> > > already has the cursor), it seems that much of the duplication can be
> > > reduced and the logic simplified.
> >
> > Ok, done.
> >
> > > > + return 0;
> > > > +}
> > > > +
> > > > +/* Rebuild a free space by block number btree. */
> > > > +static void
> > > > +build_bnobt(
> > > > + struct repair_ctx *sc,
> > > > + xfs_agnumber_t agno,
> > > > + struct bt_rebuild *btr_bno,
> > > > + xfs_agblock_t *freeblks)
> > > > +{
> > > > + int error;
> > > > +
> > > > + *freeblks = 0;
> > > > + btr_bno->bload.get_record = get_bnobt_record;
> > > > + btr_bno->bload.claim_block = rebuild_claim_block;
> > >
> > > Note that one of these callbacks is introduced in this patch and the
> > > other in the previous. I've no issue with splitting the patches as noted
> > > in the previous patch, but these should probably be consistent one way
> > > or the other.
> >
> > All the btree types set ->claim_block to rebuild_claim_block, whereas
> > ->get_record is specific to the particular type of btree.
> >
>
> Ok, seems more reasonable in hindsight.
<nod> FWIW I also put the new code in its own file to declutter phase5.c
further.
--D
> Brian
>
> > > > + btr_bno->bno_rec = findfirst_bno_extent(agno);
> > > > + btr_bno->freeblks = freeblks;
> > > > +
> > > > + error = -libxfs_trans_alloc_empty(sc->mp, &sc->tp);
> > > > + if (error)
> > > > + do_error(
> > > > +_("Insufficient memory to construct bnobt rebuild transaction.\n"));
> > > > +
> > > > + /* Add all observed bnobt records. */
> > > > + error = -libxfs_btree_bload(btr_bno->cur, &btr_bno->bload, btr_bno);
> > > > + if (error)
> > > > + do_error(
> > > > +_("Error %d while creating bnobt btree for AG %u.\n"), error, agno);
> > > > +
> > > > + /* Since we're not writing the AGF yet, no need to commit the cursor */
> > > > + libxfs_btree_del_cursor(btr_bno->cur, 0);
> > > > + error = -libxfs_trans_commit(sc->tp);
> > > > + if (error)
> > > > + do_error(
> > > > +_("Error %d while writing bnobt btree for AG %u.\n"), error, agno);
> > > > + sc->tp = NULL;
> > > > +}
> > > > +
> > > ...
> > > > @@ -2354,48 +2041,14 @@ build_agf_agfl(
> > > > freelist[i] = cpu_to_be32(NULLAGBLOCK);
> > > > }
> > > >
> > > > - /*
> > > > - * do we have left-over blocks in the btree cursors that should
> > > > - * be used to fill the AGFL?
> > > > - */
> > > > - if (bno_bt->num_free_blocks > 0 || bcnt_bt->num_free_blocks > 0) {
> > > > - /*
> > > > - * yes, now grab as many blocks as we can
> > > > - */
> > > > - i = 0;
> > > > - while (bno_bt->num_free_blocks > 0 && i < libxfs_agfl_size(mp))
> > > > - {
> > > > - freelist[i] = cpu_to_be32(
> > > > - get_next_blockaddr(agno, 0, bno_bt));
> > > > - i++;
> > > > - }
> > > > -
> > > > - while (bcnt_bt->num_free_blocks > 0 && i < libxfs_agfl_size(mp))
> > > > - {
> > > > - freelist[i] = cpu_to_be32(
> > > > - get_next_blockaddr(agno, 0, bcnt_bt));
> > > > - i++;
> > > > - }
> > > > - /*
> > > > - * now throw the rest of the blocks away and complain
> > > > - */
> > > > - while (bno_bt->num_free_blocks > 0) {
> > > > - fsb = XFS_AGB_TO_FSB(mp, agno,
> > > > - get_next_blockaddr(agno, 0, bno_bt));
> > > > - error = slab_add(lost_fsb, &fsb);
> > > > - if (error)
> > > > - do_error(
> > > > -_("Insufficient memory saving lost blocks.\n"));
> > > > - }
> > > > - while (bcnt_bt->num_free_blocks > 0) {
> > > > - fsb = XFS_AGB_TO_FSB(mp, agno,
> > > > - get_next_blockaddr(agno, 0, bcnt_bt));
> > > > - error = slab_add(lost_fsb, &fsb);
> > > > - if (error)
> > > > - do_error(
> > > > -_("Insufficient memory saving lost blocks.\n"));
> > > > - }
> > > > + /* Fill the AGFL with leftover blocks or save them for later. */
> > > > + i = 0;
> > >
> > > How about agfl_idx or something since this variable is passed down into
> > > a function?
> >
> > The 'i' variable existed before this patch. I can add a new cleanup to
> > fix the name...
> >
> > > > + freelist = xfs_buf_to_agfl_bno(agfl_buf);
> > > > + fill_agfl(btr_bno, freelist, &i);
> > > > + fill_agfl(btr_cnt, freelist, &i);
> > > >
> > > > + /* Set the AGF counters for the AGFL. */
> > > > + if (i > 0) {
> > > > agf->agf_flfirst = 0;
> > > > agf->agf_fllast = cpu_to_be32(i - 1);
> > > > agf->agf_flcount = cpu_to_be32(i);
> > > ...
> > > > @@ -2650,9 +2283,9 @@ _("unable to rebuild AG %u. Not enough free space in on-disk AG.\n"),
> > > > /*
> > > > * set up agf and agfl
> > > > */
> > > > - build_agf_agfl(mp, agno, &bno_btree_curs,
> > > > - &bcnt_btree_curs, freeblks1, extra_blocks,
> > > > - &rmap_btree_curs, &refcnt_btree_curs, lost_fsb);
> > > > + build_agf_agfl(mp, agno, &btr_bno, &btr_cnt, freeblks1, extra_blocks,
> > > > + &rmap_btree_curs, &refcnt_btree_curs, lost_fsb);
> > > > +
> > >
> > > I was trying to figure out what extra_blocks was used for and noticed
> > > that it's not even used by this function. Perhaps a preceding cleanup
> > > patch to drop it?
> >
> > Ok.
> >
> > > > /*
> > > > * build inode allocation tree.
> > > > */
> > > > @@ -2674,15 +2307,14 @@ _("unable to rebuild AG %u. Not enough free space in on-disk AG.\n"),
> > > > /*
> > > > * tear down cursors
> > > > */
> > > > - finish_cursor(&bno_btree_curs);
> > > > - finish_cursor(&ino_btree_curs);
> > >
> > > Looks like this one shouldn't be going away here.
> >
> > Oops, good catch.
> >
> > --D
> >
> > > Brian
> > >
> > > > + finish_rebuild(mp, &btr_bno, lost_fsb);
> > > > + finish_rebuild(mp, &btr_cnt, lost_fsb);
> > > > if (xfs_sb_version_hasrmapbt(&mp->m_sb))
> > > > finish_cursor(&rmap_btree_curs);
> > > > if (xfs_sb_version_hasreflink(&mp->m_sb))
> > > > finish_cursor(&refcnt_btree_curs);
> > > > if (xfs_sb_version_hasfinobt(&mp->m_sb))
> > > > finish_cursor(&fino_btree_curs);
> > > > - finish_cursor(&bcnt_btree_curs);
> > > >
> > > > /*
> > > > * release the incore per-AG bno/bcnt trees so the extent nodes
> > > >
> > >
> >
>
^ permalink raw reply [flat|nested] 30+ messages in thread
end of thread, other threads:[~2020-06-02 0:24 UTC | newest]
Thread overview: 30+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2020-05-20 1:50 [PATCH v5 0/9] xfs_repair: use btree bulk loading Darrick J. Wong
2020-05-20 1:50 ` [PATCH 1/9] xfs_repair: port the online repair newbt structure Darrick J. Wong
2020-05-27 12:15 ` Brian Foster
2020-05-27 22:34 ` Darrick J. Wong
2020-05-28 15:08 ` Brian Foster
2020-05-29 21:01 ` Darrick J. Wong
2020-06-01 12:03 ` Brian Foster
2020-06-02 0:12 ` Darrick J. Wong
2020-05-20 1:50 ` [PATCH 2/9] xfs_repair: remove gratuitous code block in phase5 Darrick J. Wong
2020-05-27 12:15 ` Brian Foster
2020-05-20 1:51 ` [PATCH 3/9] xfs_repair: create a new class of btree rebuild cursors Darrick J. Wong
2020-05-27 12:18 ` Brian Foster
2020-05-27 22:07 ` Darrick J. Wong
2020-05-28 15:09 ` Brian Foster
2020-05-29 21:08 ` Darrick J. Wong
2020-05-20 1:51 ` [PATCH 4/9] xfs_repair: rebuild free space btrees with bulk loader Darrick J. Wong
2020-05-28 15:10 ` Brian Foster
2020-05-29 21:39 ` Darrick J. Wong
2020-06-01 12:05 ` Brian Foster
2020-06-02 0:21 ` Darrick J. Wong
2020-05-20 1:51 ` [PATCH 5/9] xfs_repair: rebuild inode " Darrick J. Wong
2020-05-28 15:11 ` Brian Foster
2020-05-29 22:18 ` Darrick J. Wong
2020-05-29 22:32 ` Darrick J. Wong
2020-05-20 1:51 ` [PATCH 6/9] xfs_repair: rebuild reverse mapping " Darrick J. Wong
2020-05-20 1:51 ` [PATCH 7/9] xfs_repair: rebuild refcount " Darrick J. Wong
2020-05-20 1:51 ` [PATCH 8/9] xfs_repair: remove old btree rebuild support code Darrick J. Wong
2020-05-20 1:51 ` [PATCH 9/9] xfs_repair: track blocks lost during btree construction via extents Darrick J. Wong
2020-05-28 17:00 ` Brian Foster
2020-05-29 22:19 ` Darrick J. Wong
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.