[PATCHSET v5 00/11] xfs: consolidate posteof and cowblocks cleanup

[PATCHSET v5 00/11] xfs: consolidate posteof and cowblocks cleanup

[Darrick J. Wong]

Hi all,

Currently, we treat the garbage collection of post-EOF preallocations
and copy-on-write preallocations as totally separate tasks -- different
incore inode tags, different workqueues, etc.  This is wasteful of radix
tree tags and workqueue resources since we effectively have parallel
code paths to do the same thing.

Therefore, consolidate both functions under one radix tree bit and one
workqueue function that scans an inode for both things at the same time.
At the end of the series we make the scanning per-AG instead of per-fs
so that the scanning can run in parallel.  This reduces locking
contention from background threads and will free up a radix tree bit for
the deferred inode inactivation series.

More recent iterations of this patchset also rework the code to avoid
static predeclarations and take advantage of the evolutions of the
three previous patchsets.

v2: clean up and rebase against 5.11.
v3: various streamlining as part of restructuring the space reclaim series
v4: move the workqueue parallelism bits to their own series
v5: expose blockgc wq via sysfs and reduce iolock cycling.

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

kernel git tree:
https://git.kernel.org/cgit/linux/kernel/git/djwong/xfs-linux.git/log/?h=eofblocks-consolidation-5.12
---
 Documentation/admin-guide/xfs.rst |    3 
 fs/xfs/scrub/common.c             |    4 -
 fs/xfs/xfs_globals.c              |    7 -
 fs/xfs/xfs_icache.c               |  295 ++++++++++++++++---------------------
 fs/xfs/xfs_icache.h               |   16 +-
 fs/xfs/xfs_ioctl.c                |    2 
 fs/xfs/xfs_linux.h                |    3 
 fs/xfs/xfs_mount.c                |    5 +
 fs/xfs/xfs_mount.h                |    9 +
 fs/xfs/xfs_super.c                |   23 +--
 fs/xfs/xfs_sysctl.c               |   15 --
 fs/xfs/xfs_sysctl.h               |    3 
 fs/xfs/xfs_trace.h                |    6 -
 13 files changed, 170 insertions(+), 221 deletions(-)

[PATCH 01/11] xfs: relocate the eofb/cowb workqueue functions

[Darrick J. Wong]

From: Darrick J. Wong <djwong@kernel.org>

Move the xfs_{eof,cow}blocks_worker and xfs_queue_{eof,cow}blocks
functions further down in the file so that the cleanups in the next
patches won't have to pre-declare static functions.  No functional
changes.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
---
 fs/xfs/xfs_icache.c |  126 ++++++++++++++++++++++++++-------------------------
 1 file changed, 63 insertions(+), 63 deletions(-)


diff --git a/fs/xfs/xfs_icache.c b/fs/xfs/xfs_icache.c
index 97c15fcdd6f7..15dcf57b4b19 100644
--- a/fs/xfs/xfs_icache.c
+++ b/fs/xfs/xfs_icache.c
@@ -915,69 +915,6 @@ xfs_inode_walk(
 	return last_error;
 }
 
-/*
- * Background scanning to trim post-EOF preallocated space. This is queued
- * based on the 'speculative_prealloc_lifetime' tunable (5m by default).
- */
-void
-xfs_queue_eofblocks(
-	struct xfs_mount *mp)
-{
-	rcu_read_lock();
-	if (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_EOFBLOCKS_TAG))
-		queue_delayed_work(mp->m_eofblocks_workqueue,
-				   &mp->m_eofblocks_work,
-				   msecs_to_jiffies(xfs_eofb_secs * 1000));
-	rcu_read_unlock();
-}
-
-void
-xfs_eofblocks_worker(
-	struct work_struct *work)
-{
-	struct xfs_mount *mp = container_of(to_delayed_work(work),
-				struct xfs_mount, m_eofblocks_work);
-
-	if (!sb_start_write_trylock(mp->m_super))
-		return;
-	xfs_icache_free_eofblocks(mp, NULL);
-	sb_end_write(mp->m_super);
-
-	xfs_queue_eofblocks(mp);
-}
-
-/*
- * Background scanning to trim preallocated CoW space. This is queued
- * based on the 'speculative_cow_prealloc_lifetime' tunable (5m by default).
- * (We'll just piggyback on the post-EOF prealloc space workqueue.)
- */
-void
-xfs_queue_cowblocks(
-	struct xfs_mount *mp)
-{
-	rcu_read_lock();
-	if (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_COWBLOCKS_TAG))
-		queue_delayed_work(mp->m_eofblocks_workqueue,
-				   &mp->m_cowblocks_work,
-				   msecs_to_jiffies(xfs_cowb_secs * 1000));
-	rcu_read_unlock();
-}
-
-void
-xfs_cowblocks_worker(
-	struct work_struct *work)
-{
-	struct xfs_mount *mp = container_of(to_delayed_work(work),
-				struct xfs_mount, m_cowblocks_work);
-
-	if (!sb_start_write_trylock(mp->m_super))
-		return;
-	xfs_icache_free_cowblocks(mp, NULL);
-	sb_end_write(mp->m_super);
-
-	xfs_queue_cowblocks(mp);
-}
-
 /*
  * Grab the inode for reclaim exclusively.
  *
@@ -1396,6 +1333,37 @@ xfs_icache_free_eofblocks(
 			XFS_ICI_EOFBLOCKS_TAG);
 }
 
+/*
+ * Background scanning to trim post-EOF preallocated space. This is queued
+ * based on the 'speculative_prealloc_lifetime' tunable (5m by default).
+ */
+void
+xfs_queue_eofblocks(
+	struct xfs_mount *mp)
+{
+	rcu_read_lock();
+	if (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_EOFBLOCKS_TAG))
+		queue_delayed_work(mp->m_eofblocks_workqueue,
+				   &mp->m_eofblocks_work,
+				   msecs_to_jiffies(xfs_eofb_secs * 1000));
+	rcu_read_unlock();
+}
+
+void
+xfs_eofblocks_worker(
+	struct work_struct *work)
+{
+	struct xfs_mount *mp = container_of(to_delayed_work(work),
+				struct xfs_mount, m_eofblocks_work);
+
+	if (!sb_start_write_trylock(mp->m_super))
+		return;
+	xfs_icache_free_eofblocks(mp, NULL);
+	sb_end_write(mp->m_super);
+
+	xfs_queue_eofblocks(mp);
+}
+
 static inline unsigned long
 xfs_iflag_for_tag(
 	int		tag)
@@ -1608,6 +1576,38 @@ xfs_icache_free_cowblocks(
 			XFS_ICI_COWBLOCKS_TAG);
 }
 
+/*
+ * Background scanning to trim preallocated CoW space. This is queued
+ * based on the 'speculative_cow_prealloc_lifetime' tunable (5m by default).
+ * (We'll just piggyback on the post-EOF prealloc space workqueue.)
+ */
+void
+xfs_queue_cowblocks(
+	struct xfs_mount *mp)
+{
+	rcu_read_lock();
+	if (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_COWBLOCKS_TAG))
+		queue_delayed_work(mp->m_eofblocks_workqueue,
+				   &mp->m_cowblocks_work,
+				   msecs_to_jiffies(xfs_cowb_secs * 1000));
+	rcu_read_unlock();
+}
+
+void
+xfs_cowblocks_worker(
+	struct work_struct *work)
+{
+	struct xfs_mount *mp = container_of(to_delayed_work(work),
+				struct xfs_mount, m_cowblocks_work);
+
+	if (!sb_start_write_trylock(mp->m_super))
+		return;
+	xfs_icache_free_cowblocks(mp, NULL);
+	sb_end_write(mp->m_super);
+
+	xfs_queue_cowblocks(mp);
+}
+
 void
 xfs_inode_set_cowblocks_tag(
 	xfs_inode_t	*ip)

[PATCH 02/11] xfs: hide xfs_icache_free_eofblocks

[Darrick J. Wong]

From: Darrick J. Wong <djwong@kernel.org>

Change the one remaining caller of xfs_icache_free_eofblocks to use our
new combined blockgc scan function instead, since we will soon be
combining the two scans.  This introduces a slight behavior change,
since the XFS_IOC_FREE_EOFBLOCKS now clears out speculative CoW
reservations in addition to post-eof blocks.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
---
 fs/xfs/xfs_icache.c |    2 +-
 fs/xfs/xfs_icache.h |    1 -
 fs/xfs/xfs_ioctl.c  |    2 +-
 3 files changed, 2 insertions(+), 3 deletions(-)


diff --git a/fs/xfs/xfs_icache.c b/fs/xfs/xfs_icache.c
index 15dcf57b4b19..2e1b3887754a 100644
--- a/fs/xfs/xfs_icache.c
+++ b/fs/xfs/xfs_icache.c
@@ -1324,7 +1324,7 @@ xfs_inode_free_eofblocks(
 	return ret;
 }
 
-int
+static int
 xfs_icache_free_eofblocks(
 	struct xfs_mount	*mp,
 	struct xfs_eofblocks	*eofb)
diff --git a/fs/xfs/xfs_icache.h b/fs/xfs/xfs_icache.h
index 583c132ae0fb..c23e49aab683 100644
--- a/fs/xfs/xfs_icache.h
+++ b/fs/xfs/xfs_icache.h
@@ -61,7 +61,6 @@ int xfs_blockgc_free_space(struct xfs_mount *mp, struct xfs_eofblocks *eofb);
 
 void xfs_inode_set_eofblocks_tag(struct xfs_inode *ip);
 void xfs_inode_clear_eofblocks_tag(struct xfs_inode *ip);
-int xfs_icache_free_eofblocks(struct xfs_mount *, struct xfs_eofblocks *);
 void xfs_eofblocks_worker(struct work_struct *);
 void xfs_queue_eofblocks(struct xfs_mount *);
 
diff --git a/fs/xfs/xfs_ioctl.c b/fs/xfs/xfs_ioctl.c
index cc6ddc6d22a0..7e22346e8ff5 100644
--- a/fs/xfs/xfs_ioctl.c
+++ b/fs/xfs/xfs_ioctl.c
@@ -2358,7 +2358,7 @@ xfs_file_ioctl(
 		trace_xfs_ioc_free_eofblocks(mp, &keofb, _RET_IP_);
 
 		sb_start_write(mp->m_super);
-		error = xfs_icache_free_eofblocks(mp, &keofb);
+		error = xfs_blockgc_free_space(mp, &keofb);
 		sb_end_write(mp->m_super);
 		return error;
 	}

[PATCH 03/11] xfs: hide xfs_icache_free_cowblocks

[Darrick J. Wong]

From: Darrick J. Wong <djwong@kernel.org>

Change the one remaining caller of xfs_icache_free_cowblocks to use our
new combined blockgc scan function instead, since we will soon be
combining the two scans.  This introduces a slight behavior change,
since a readonly remount now clears out post-EOF preallocations and not
just CoW staging extents.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
---
 fs/xfs/xfs_icache.c |    2 +-
 fs/xfs/xfs_icache.h |    1 -
 fs/xfs/xfs_super.c  |    2 +-
 3 files changed, 2 insertions(+), 3 deletions(-)


diff --git a/fs/xfs/xfs_icache.c b/fs/xfs/xfs_icache.c
index 2e1b3887754a..ef4c0659f38f 100644
--- a/fs/xfs/xfs_icache.c
+++ b/fs/xfs/xfs_icache.c
@@ -1567,7 +1567,7 @@ xfs_inode_free_cowblocks(
 	return ret;
 }
 
-int
+static int
 xfs_icache_free_cowblocks(
 	struct xfs_mount	*mp,
 	struct xfs_eofblocks	*eofb)
diff --git a/fs/xfs/xfs_icache.h b/fs/xfs/xfs_icache.h
index c23e49aab683..8566798d42ab 100644
--- a/fs/xfs/xfs_icache.h
+++ b/fs/xfs/xfs_icache.h
@@ -66,7 +66,6 @@ void xfs_queue_eofblocks(struct xfs_mount *);
 
 void xfs_inode_set_cowblocks_tag(struct xfs_inode *ip);
 void xfs_inode_clear_cowblocks_tag(struct xfs_inode *ip);
-int xfs_icache_free_cowblocks(struct xfs_mount *, struct xfs_eofblocks *);
 void xfs_cowblocks_worker(struct work_struct *);
 void xfs_queue_cowblocks(struct xfs_mount *);
 
diff --git a/fs/xfs/xfs_super.c b/fs/xfs/xfs_super.c
index 8959561351ca..e8f71714d737 100644
--- a/fs/xfs/xfs_super.c
+++ b/fs/xfs/xfs_super.c
@@ -1713,7 +1713,7 @@ xfs_remount_ro(
 	xfs_stop_block_reaping(mp);
 
 	/* Get rid of any leftover CoW reservations... */
-	error = xfs_icache_free_cowblocks(mp, NULL);
+	error = xfs_blockgc_free_space(mp, NULL);
 	if (error) {
 		xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
 		return error;

[PATCH 04/11] xfs: remove trivial eof/cowblocks functions

[Darrick J. Wong]

From: Darrick J. Wong <djwong@kernel.org>

Get rid of these trivial helpers.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
---
 fs/xfs/xfs_icache.c |   30 ++++++++----------------------
 1 file changed, 8 insertions(+), 22 deletions(-)


diff --git a/fs/xfs/xfs_icache.c b/fs/xfs/xfs_icache.c
index ef4c0659f38f..7440ae93b204 100644
--- a/fs/xfs/xfs_icache.c
+++ b/fs/xfs/xfs_icache.c
@@ -1324,15 +1324,6 @@ xfs_inode_free_eofblocks(
 	return ret;
 }
 
-static int
-xfs_icache_free_eofblocks(
-	struct xfs_mount	*mp,
-	struct xfs_eofblocks	*eofb)
-{
-	return xfs_inode_walk(mp, 0, xfs_inode_free_eofblocks, eofb,
-			XFS_ICI_EOFBLOCKS_TAG);
-}
-
 /*
  * Background scanning to trim post-EOF preallocated space. This is queued
  * based on the 'speculative_prealloc_lifetime' tunable (5m by default).
@@ -1358,7 +1349,8 @@ xfs_eofblocks_worker(
 
 	if (!sb_start_write_trylock(mp->m_super))
 		return;
-	xfs_icache_free_eofblocks(mp, NULL);
+	xfs_inode_walk(mp, 0, xfs_inode_free_eofblocks, NULL,
+			XFS_ICI_EOFBLOCKS_TAG);
 	sb_end_write(mp->m_super);
 
 	xfs_queue_eofblocks(mp);
@@ -1567,15 +1559,6 @@ xfs_inode_free_cowblocks(
 	return ret;
 }
 
-static int
-xfs_icache_free_cowblocks(
-	struct xfs_mount	*mp,
-	struct xfs_eofblocks	*eofb)
-{
-	return xfs_inode_walk(mp, 0, xfs_inode_free_cowblocks, eofb,
-			XFS_ICI_COWBLOCKS_TAG);
-}
-
 /*
  * Background scanning to trim preallocated CoW space. This is queued
  * based on the 'speculative_cow_prealloc_lifetime' tunable (5m by default).
@@ -1602,7 +1585,8 @@ xfs_cowblocks_worker(
 
 	if (!sb_start_write_trylock(mp->m_super))
 		return;
-	xfs_icache_free_cowblocks(mp, NULL);
+	xfs_inode_walk(mp, 0, xfs_inode_free_cowblocks, NULL,
+			XFS_ICI_COWBLOCKS_TAG);
 	sb_end_write(mp->m_super);
 
 	xfs_queue_cowblocks(mp);
@@ -1653,11 +1637,13 @@ xfs_blockgc_scan(
 {
 	int			error;
 
-	error = xfs_icache_free_eofblocks(mp, eofb);
+	error = xfs_inode_walk(mp, 0, xfs_inode_free_eofblocks, eofb,
+			XFS_ICI_EOFBLOCKS_TAG);
 	if (error)
 		return error;
 
-	error = xfs_icache_free_cowblocks(mp, eofb);
+	error = xfs_inode_walk(mp, 0, xfs_inode_free_cowblocks, eofb,
+			XFS_ICI_COWBLOCKS_TAG);
 	if (error)
 		return error;
 

[PATCH 05/11] xfs: consolidate incore inode radix tree posteof/cowblocks tags

[Darrick J. Wong]

From: Darrick J. Wong <djwong@kernel.org>

The clearing of posteof blocks and cowblocks serve the same purpose:
removing speculative block preallocations from inactive files.  We don't
need to burn two radix tree tags on this, so combine them into one.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
---
 fs/xfs/xfs_icache.c |  114 ++++++++++++++++++++++++---------------------------
 fs/xfs/xfs_icache.h |    4 +-
 fs/xfs/xfs_trace.h  |    6 +--
 3 files changed, 58 insertions(+), 66 deletions(-)


diff --git a/fs/xfs/xfs_icache.c b/fs/xfs/xfs_icache.c
index 7440ae93b204..1009556fb8b2 100644
--- a/fs/xfs/xfs_icache.c
+++ b/fs/xfs/xfs_icache.c
@@ -1291,6 +1291,9 @@ xfs_inode_free_eofblocks(
 
 	wait = eofb && (eofb->eof_flags & XFS_EOF_FLAGS_SYNC);
 
+	if (!xfs_iflags_test(ip, XFS_IEOFBLOCKS))
+		return 0;
+
 	if (!xfs_can_free_eofblocks(ip, false)) {
 		/* inode could be preallocated or append-only */
 		trace_xfs_inode_free_eofblocks_invalid(ip);
@@ -1333,7 +1336,7 @@ xfs_queue_eofblocks(
 	struct xfs_mount *mp)
 {
 	rcu_read_lock();
-	if (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_EOFBLOCKS_TAG))
+	if (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_BLOCKGC_TAG))
 		queue_delayed_work(mp->m_eofblocks_workqueue,
 				   &mp->m_eofblocks_work,
 				   msecs_to_jiffies(xfs_eofb_secs * 1000));
@@ -1350,67 +1353,54 @@ xfs_eofblocks_worker(
 	if (!sb_start_write_trylock(mp->m_super))
 		return;
 	xfs_inode_walk(mp, 0, xfs_inode_free_eofblocks, NULL,
-			XFS_ICI_EOFBLOCKS_TAG);
+			XFS_ICI_BLOCKGC_TAG);
 	sb_end_write(mp->m_super);
 
 	xfs_queue_eofblocks(mp);
 }
 
-static inline unsigned long
-xfs_iflag_for_tag(
-	int		tag)
-{
-	switch (tag) {
-	case XFS_ICI_EOFBLOCKS_TAG:
-		return XFS_IEOFBLOCKS;
-	case XFS_ICI_COWBLOCKS_TAG:
-		return XFS_ICOWBLOCKS;
-	default:
-		ASSERT(0);
-		return 0;
-	}
-}
-
 static void
-__xfs_inode_set_blocks_tag(
-	xfs_inode_t	*ip,
-	void		(*execute)(struct xfs_mount *mp),
-	void		(*set_tp)(struct xfs_mount *mp, xfs_agnumber_t agno,
-				  int error, unsigned long caller_ip),
-	int		tag)
+xfs_blockgc_set_iflag(
+	struct xfs_inode	*ip,
+	void			(*execute)(struct xfs_mount *mp),
+	unsigned long		iflag)
 {
-	struct xfs_mount *mp = ip->i_mount;
-	struct xfs_perag *pag;
-	int tagged;
+	struct xfs_mount	*mp = ip->i_mount;
+	struct xfs_perag	*pag;
+	int			tagged;
+
+	ASSERT((iflag & ~(XFS_IEOFBLOCKS | XFS_ICOWBLOCKS)) == 0);
 
 	/*
 	 * Don't bother locking the AG and looking up in the radix trees
 	 * if we already know that we have the tag set.
 	 */
-	if (ip->i_flags & xfs_iflag_for_tag(tag))
+	if (ip->i_flags & iflag)
 		return;
 	spin_lock(&ip->i_flags_lock);
-	ip->i_flags |= xfs_iflag_for_tag(tag);
+	ip->i_flags |= iflag;
 	spin_unlock(&ip->i_flags_lock);
 
 	pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
 	spin_lock(&pag->pag_ici_lock);
 
-	tagged = radix_tree_tagged(&pag->pag_ici_root, tag);
+	tagged = radix_tree_tagged(&pag->pag_ici_root, XFS_ICI_BLOCKGC_TAG);
 	radix_tree_tag_set(&pag->pag_ici_root,
-			   XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino), tag);
+			   XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino),
+			   XFS_ICI_BLOCKGC_TAG);
 	if (!tagged) {
-		/* propagate the eofblocks tag up into the perag radix tree */
+		/* propagate the blockgc tag up into the perag radix tree */
 		spin_lock(&ip->i_mount->m_perag_lock);
 		radix_tree_tag_set(&ip->i_mount->m_perag_tree,
 				   XFS_INO_TO_AGNO(ip->i_mount, ip->i_ino),
-				   tag);
+				   XFS_ICI_BLOCKGC_TAG);
 		spin_unlock(&ip->i_mount->m_perag_lock);
 
 		/* kick off background trimming */
 		execute(ip->i_mount);
 
-		set_tp(ip->i_mount, pag->pag_agno, -1, _RET_IP_);
+		trace_xfs_perag_set_blockgc(ip->i_mount, pag->pag_agno, -1,
+				_RET_IP_);
 	}
 
 	spin_unlock(&pag->pag_ici_lock);
@@ -1422,38 +1412,43 @@ xfs_inode_set_eofblocks_tag(
 	xfs_inode_t	*ip)
 {
 	trace_xfs_inode_set_eofblocks_tag(ip);
-	return __xfs_inode_set_blocks_tag(ip, xfs_queue_eofblocks,
-			trace_xfs_perag_set_eofblocks,
-			XFS_ICI_EOFBLOCKS_TAG);
+	return xfs_blockgc_set_iflag(ip, xfs_queue_eofblocks, XFS_IEOFBLOCKS);
 }
 
 static void
-__xfs_inode_clear_blocks_tag(
-	xfs_inode_t	*ip,
-	void		(*clear_tp)(struct xfs_mount *mp, xfs_agnumber_t agno,
-				    int error, unsigned long caller_ip),
-	int		tag)
+xfs_blockgc_clear_iflag(
+	struct xfs_inode	*ip,
+	unsigned long		iflag)
 {
-	struct xfs_mount *mp = ip->i_mount;
-	struct xfs_perag *pag;
+	struct xfs_mount	*mp = ip->i_mount;
+	struct xfs_perag	*pag;
+	bool			clear_tag;
+
+	ASSERT((iflag & ~(XFS_IEOFBLOCKS | XFS_ICOWBLOCKS)) == 0);
 
 	spin_lock(&ip->i_flags_lock);
-	ip->i_flags &= ~xfs_iflag_for_tag(tag);
+	ip->i_flags &= ~iflag;
+	clear_tag = (ip->i_flags & (XFS_IEOFBLOCKS | XFS_ICOWBLOCKS)) == 0;
 	spin_unlock(&ip->i_flags_lock);
 
+	if (!clear_tag)
+		return;
+
 	pag = xfs_perag_get(mp, XFS_INO_TO_AGNO(mp, ip->i_ino));
 	spin_lock(&pag->pag_ici_lock);
 
 	radix_tree_tag_clear(&pag->pag_ici_root,
-			     XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino), tag);
-	if (!radix_tree_tagged(&pag->pag_ici_root, tag)) {
-		/* clear the eofblocks tag from the perag radix tree */
+			     XFS_INO_TO_AGINO(ip->i_mount, ip->i_ino),
+			     XFS_ICI_BLOCKGC_TAG);
+	if (!radix_tree_tagged(&pag->pag_ici_root, XFS_ICI_BLOCKGC_TAG)) {
+		/* clear the blockgc tag from the perag radix tree */
 		spin_lock(&ip->i_mount->m_perag_lock);
 		radix_tree_tag_clear(&ip->i_mount->m_perag_tree,
 				     XFS_INO_TO_AGNO(ip->i_mount, ip->i_ino),
-				     tag);
+				     XFS_ICI_BLOCKGC_TAG);
 		spin_unlock(&ip->i_mount->m_perag_lock);
-		clear_tp(ip->i_mount, pag->pag_agno, -1, _RET_IP_);
+		trace_xfs_perag_clear_blockgc(ip->i_mount, pag->pag_agno, -1,
+				_RET_IP_);
 	}
 
 	spin_unlock(&pag->pag_ici_lock);
@@ -1465,8 +1460,7 @@ xfs_inode_clear_eofblocks_tag(
 	xfs_inode_t	*ip)
 {
 	trace_xfs_inode_clear_eofblocks_tag(ip);
-	return __xfs_inode_clear_blocks_tag(ip,
-			trace_xfs_perag_clear_eofblocks, XFS_ICI_EOFBLOCKS_TAG);
+	return xfs_blockgc_clear_iflag(ip, XFS_IEOFBLOCKS);
 }
 
 /*
@@ -1524,6 +1518,9 @@ xfs_inode_free_cowblocks(
 
 	wait = eofb && (eofb->eof_flags & XFS_EOF_FLAGS_SYNC);
 
+	if (!xfs_iflags_test(ip, XFS_ICOWBLOCKS))
+		return 0;
+
 	if (!xfs_prep_free_cowblocks(ip))
 		return 0;
 
@@ -1569,7 +1566,7 @@ xfs_queue_cowblocks(
 	struct xfs_mount *mp)
 {
 	rcu_read_lock();
-	if (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_COWBLOCKS_TAG))
+	if (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_BLOCKGC_TAG))
 		queue_delayed_work(mp->m_eofblocks_workqueue,
 				   &mp->m_cowblocks_work,
 				   msecs_to_jiffies(xfs_cowb_secs * 1000));
@@ -1586,7 +1583,7 @@ xfs_cowblocks_worker(
 	if (!sb_start_write_trylock(mp->m_super))
 		return;
 	xfs_inode_walk(mp, 0, xfs_inode_free_cowblocks, NULL,
-			XFS_ICI_COWBLOCKS_TAG);
+			XFS_ICI_BLOCKGC_TAG);
 	sb_end_write(mp->m_super);
 
 	xfs_queue_cowblocks(mp);
@@ -1597,9 +1594,7 @@ xfs_inode_set_cowblocks_tag(
 	xfs_inode_t	*ip)
 {
 	trace_xfs_inode_set_cowblocks_tag(ip);
-	return __xfs_inode_set_blocks_tag(ip, xfs_queue_cowblocks,
-			trace_xfs_perag_set_cowblocks,
-			XFS_ICI_COWBLOCKS_TAG);
+	return xfs_blockgc_set_iflag(ip, xfs_queue_cowblocks, XFS_ICOWBLOCKS);
 }
 
 void
@@ -1607,8 +1602,7 @@ xfs_inode_clear_cowblocks_tag(
 	xfs_inode_t	*ip)
 {
 	trace_xfs_inode_clear_cowblocks_tag(ip);
-	return __xfs_inode_clear_blocks_tag(ip,
-			trace_xfs_perag_clear_cowblocks, XFS_ICI_COWBLOCKS_TAG);
+	return xfs_blockgc_clear_iflag(ip, XFS_ICOWBLOCKS);
 }
 
 /* Disable post-EOF and CoW block auto-reclamation. */
@@ -1638,12 +1632,12 @@ xfs_blockgc_scan(
 	int			error;
 
 	error = xfs_inode_walk(mp, 0, xfs_inode_free_eofblocks, eofb,
-			XFS_ICI_EOFBLOCKS_TAG);
+			XFS_ICI_BLOCKGC_TAG);
 	if (error)
 		return error;
 
 	error = xfs_inode_walk(mp, 0, xfs_inode_free_cowblocks, eofb,
-			XFS_ICI_COWBLOCKS_TAG);
+			XFS_ICI_BLOCKGC_TAG);
 	if (error)
 		return error;
 
diff --git a/fs/xfs/xfs_icache.h b/fs/xfs/xfs_icache.h
index 8566798d42ab..d2ef1ef75e7a 100644
--- a/fs/xfs/xfs_icache.h
+++ b/fs/xfs/xfs_icache.h
@@ -23,8 +23,8 @@ struct xfs_eofblocks {
 #define XFS_ICI_NO_TAG		(-1)	/* special flag for an untagged lookup
 					   in xfs_inode_walk */
 #define XFS_ICI_RECLAIM_TAG	0	/* inode is to be reclaimed */
-#define XFS_ICI_EOFBLOCKS_TAG	1	/* inode has blocks beyond EOF */
-#define XFS_ICI_COWBLOCKS_TAG	2	/* inode can have cow blocks to gc */
+/* Inode has speculative preallocations (posteof or cow) to clean. */
+#define XFS_ICI_BLOCKGC_TAG	1
 
 /*
  * Flags for xfs_iget()
diff --git a/fs/xfs/xfs_trace.h b/fs/xfs/xfs_trace.h
index 27929c6ca43a..e74bbb648f83 100644
--- a/fs/xfs/xfs_trace.h
+++ b/fs/xfs/xfs_trace.h
@@ -155,10 +155,8 @@ DEFINE_PERAG_REF_EVENT(xfs_perag_get_tag);
 DEFINE_PERAG_REF_EVENT(xfs_perag_put);
 DEFINE_PERAG_REF_EVENT(xfs_perag_set_reclaim);
 DEFINE_PERAG_REF_EVENT(xfs_perag_clear_reclaim);
-DEFINE_PERAG_REF_EVENT(xfs_perag_set_eofblocks);
-DEFINE_PERAG_REF_EVENT(xfs_perag_clear_eofblocks);
-DEFINE_PERAG_REF_EVENT(xfs_perag_set_cowblocks);
-DEFINE_PERAG_REF_EVENT(xfs_perag_clear_cowblocks);
+DEFINE_PERAG_REF_EVENT(xfs_perag_set_blockgc);
+DEFINE_PERAG_REF_EVENT(xfs_perag_clear_blockgc);
 
 DECLARE_EVENT_CLASS(xfs_ag_class,
 	TP_PROTO(struct xfs_mount *mp, xfs_agnumber_t agno),

[PATCH 06/11] xfs: consolidate the eofblocks and cowblocks workers

[Darrick J. Wong]

From: Darrick J. Wong <djwong@kernel.org>

Remove the separate cowblocks work items and knob so that we can control
and run everything from a single blockgc work queue.  Note that the
speculative_prealloc_lifetime sysfs knob retains its historical name
even though the functions move to prefix xfs_blockgc_*.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
---
 fs/xfs/xfs_globals.c |    7 ++--
 fs/xfs/xfs_icache.c  |   96 ++++++++++++++++----------------------------------
 fs/xfs/xfs_icache.h  |    6 +--
 fs/xfs/xfs_linux.h   |    3 +-
 fs/xfs/xfs_mount.h   |    6 +--
 fs/xfs/xfs_super.c   |   11 +++---
 fs/xfs/xfs_sysctl.c  |   15 ++------
 fs/xfs/xfs_sysctl.h  |    3 +-
 8 files changed, 48 insertions(+), 99 deletions(-)


diff --git a/fs/xfs/xfs_globals.c b/fs/xfs/xfs_globals.c
index fa55ab8b8d80..f62fa652c2fd 100644
--- a/fs/xfs/xfs_globals.c
+++ b/fs/xfs/xfs_globals.c
@@ -8,8 +8,8 @@
 /*
  * Tunable XFS parameters.  xfs_params is required even when CONFIG_SYSCTL=n,
  * other XFS code uses these values.  Times are measured in centisecs (i.e.
- * 100ths of a second) with the exception of eofb_timer and cowb_timer, which
- * are measured in seconds.
+ * 100ths of a second) with the exception of blockgc_timer, which is measured
+ * in seconds.
  */
 xfs_param_t xfs_params = {
 			  /*	MIN		DFLT		MAX	*/
@@ -28,8 +28,7 @@ xfs_param_t xfs_params = {
 	.rotorstep	= {	1,		1,		255	},
 	.inherit_nodfrg	= {	0,		1,		1	},
 	.fstrm_timer	= {	1,		30*100,		3600*100},
-	.eofb_timer	= {	1,		300,		3600*24},
-	.cowb_timer	= {	1,		1800,		3600*24},
+	.blockgc_timer	= {	1,		300,		3600*24},
 };
 
 struct xfs_globals xfs_globals = {
diff --git a/fs/xfs/xfs_icache.c b/fs/xfs/xfs_icache.c
index 1009556fb8b2..0d0462268711 100644
--- a/fs/xfs/xfs_icache.c
+++ b/fs/xfs/xfs_icache.c
@@ -1328,41 +1328,24 @@ xfs_inode_free_eofblocks(
 }
 
 /*
- * Background scanning to trim post-EOF preallocated space. This is queued
- * based on the 'speculative_prealloc_lifetime' tunable (5m by default).
+ * Background scanning to trim preallocated space. This is queued based on the
+ * 'speculative_prealloc_lifetime' tunable (5m by default).
  */
-void
-xfs_queue_eofblocks(
-	struct xfs_mount *mp)
+static inline void
+xfs_blockgc_queue(
+	struct xfs_mount	*mp)
 {
 	rcu_read_lock();
 	if (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_BLOCKGC_TAG))
-		queue_delayed_work(mp->m_eofblocks_workqueue,
-				   &mp->m_eofblocks_work,
-				   msecs_to_jiffies(xfs_eofb_secs * 1000));
+		queue_delayed_work(mp->m_blockgc_workqueue,
+				   &mp->m_blockgc_work,
+				   msecs_to_jiffies(xfs_blockgc_secs * 1000));
 	rcu_read_unlock();
 }
 
-void
-xfs_eofblocks_worker(
-	struct work_struct *work)
-{
-	struct xfs_mount *mp = container_of(to_delayed_work(work),
-				struct xfs_mount, m_eofblocks_work);
-
-	if (!sb_start_write_trylock(mp->m_super))
-		return;
-	xfs_inode_walk(mp, 0, xfs_inode_free_eofblocks, NULL,
-			XFS_ICI_BLOCKGC_TAG);
-	sb_end_write(mp->m_super);
-
-	xfs_queue_eofblocks(mp);
-}
-
 static void
 xfs_blockgc_set_iflag(
 	struct xfs_inode	*ip,
-	void			(*execute)(struct xfs_mount *mp),
 	unsigned long		iflag)
 {
 	struct xfs_mount	*mp = ip->i_mount;
@@ -1397,7 +1380,7 @@ xfs_blockgc_set_iflag(
 		spin_unlock(&ip->i_mount->m_perag_lock);
 
 		/* kick off background trimming */
-		execute(ip->i_mount);
+		xfs_blockgc_queue(ip->i_mount);
 
 		trace_xfs_perag_set_blockgc(ip->i_mount, pag->pag_agno, -1,
 				_RET_IP_);
@@ -1412,7 +1395,7 @@ xfs_inode_set_eofblocks_tag(
 	xfs_inode_t	*ip)
 {
 	trace_xfs_inode_set_eofblocks_tag(ip);
-	return xfs_blockgc_set_iflag(ip, xfs_queue_eofblocks, XFS_IEOFBLOCKS);
+	return xfs_blockgc_set_iflag(ip, XFS_IEOFBLOCKS);
 }
 
 static void
@@ -1556,45 +1539,12 @@ xfs_inode_free_cowblocks(
 	return ret;
 }
 
-/*
- * Background scanning to trim preallocated CoW space. This is queued
- * based on the 'speculative_cow_prealloc_lifetime' tunable (5m by default).
- * (We'll just piggyback on the post-EOF prealloc space workqueue.)
- */
-void
-xfs_queue_cowblocks(
-	struct xfs_mount *mp)
-{
-	rcu_read_lock();
-	if (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_BLOCKGC_TAG))
-		queue_delayed_work(mp->m_eofblocks_workqueue,
-				   &mp->m_cowblocks_work,
-				   msecs_to_jiffies(xfs_cowb_secs * 1000));
-	rcu_read_unlock();
-}
-
-void
-xfs_cowblocks_worker(
-	struct work_struct *work)
-{
-	struct xfs_mount *mp = container_of(to_delayed_work(work),
-				struct xfs_mount, m_cowblocks_work);
-
-	if (!sb_start_write_trylock(mp->m_super))
-		return;
-	xfs_inode_walk(mp, 0, xfs_inode_free_cowblocks, NULL,
-			XFS_ICI_BLOCKGC_TAG);
-	sb_end_write(mp->m_super);
-
-	xfs_queue_cowblocks(mp);
-}
-
 void
 xfs_inode_set_cowblocks_tag(
 	xfs_inode_t	*ip)
 {
 	trace_xfs_inode_set_cowblocks_tag(ip);
-	return xfs_blockgc_set_iflag(ip, xfs_queue_cowblocks, XFS_ICOWBLOCKS);
+	return xfs_blockgc_set_iflag(ip, XFS_ICOWBLOCKS);
 }
 
 void
@@ -1610,8 +1560,7 @@ void
 xfs_stop_block_reaping(
 	struct xfs_mount	*mp)
 {
-	cancel_delayed_work_sync(&mp->m_eofblocks_work);
-	cancel_delayed_work_sync(&mp->m_cowblocks_work);
+	cancel_delayed_work_sync(&mp->m_blockgc_work);
 }
 
 /* Enable post-EOF and CoW block auto-reclamation. */
@@ -1619,8 +1568,7 @@ void
 xfs_start_block_reaping(
 	struct xfs_mount	*mp)
 {
-	xfs_queue_eofblocks(mp);
-	xfs_queue_cowblocks(mp);
+	xfs_blockgc_queue(mp);
 }
 
 /* Scan all incore inodes for block preallocations that we can remove. */
@@ -1644,6 +1592,24 @@ xfs_blockgc_scan(
 	return 0;
 }
 
+/* Background worker that trims preallocated space. */
+void
+xfs_blockgc_worker(
+	struct work_struct	*work)
+{
+	struct xfs_mount	*mp = container_of(to_delayed_work(work),
+					struct xfs_mount, m_blockgc_work);
+	int			error;
+
+	if (!sb_start_write_trylock(mp->m_super))
+		return;
+	error = xfs_blockgc_scan(mp, NULL);
+	if (error)
+		xfs_info(mp, "preallocation gc worker failed, err=%d", error);
+	sb_end_write(mp->m_super);
+	xfs_blockgc_queue(mp);
+}
+
 /*
  * Try to free space in the filesystem by purging eofblocks and cowblocks.
  */
diff --git a/fs/xfs/xfs_icache.h b/fs/xfs/xfs_icache.h
index d2ef1ef75e7a..d781703af025 100644
--- a/fs/xfs/xfs_icache.h
+++ b/fs/xfs/xfs_icache.h
@@ -61,13 +61,11 @@ int xfs_blockgc_free_space(struct xfs_mount *mp, struct xfs_eofblocks *eofb);
 
 void xfs_inode_set_eofblocks_tag(struct xfs_inode *ip);
 void xfs_inode_clear_eofblocks_tag(struct xfs_inode *ip);
-void xfs_eofblocks_worker(struct work_struct *);
-void xfs_queue_eofblocks(struct xfs_mount *);
 
 void xfs_inode_set_cowblocks_tag(struct xfs_inode *ip);
 void xfs_inode_clear_cowblocks_tag(struct xfs_inode *ip);
-void xfs_cowblocks_worker(struct work_struct *);
-void xfs_queue_cowblocks(struct xfs_mount *);
+
+void xfs_blockgc_worker(struct work_struct *work);
 
 int xfs_inode_walk(struct xfs_mount *mp, int iter_flags,
 	int (*execute)(struct xfs_inode *ip, void *args),
diff --git a/fs/xfs/xfs_linux.h b/fs/xfs/xfs_linux.h
index 5b7a1e201559..af6be9b9ccdf 100644
--- a/fs/xfs/xfs_linux.h
+++ b/fs/xfs/xfs_linux.h
@@ -98,8 +98,7 @@ typedef __u32			xfs_nlink_t;
 #define xfs_rotorstep		xfs_params.rotorstep.val
 #define xfs_inherit_nodefrag	xfs_params.inherit_nodfrg.val
 #define xfs_fstrm_centisecs	xfs_params.fstrm_timer.val
-#define xfs_eofb_secs		xfs_params.eofb_timer.val
-#define xfs_cowb_secs		xfs_params.cowb_timer.val
+#define xfs_blockgc_secs	xfs_params.blockgc_timer.val
 
 #define current_cpu()		(raw_smp_processor_id())
 #define current_set_flags_nested(sp, f)		\
diff --git a/fs/xfs/xfs_mount.h b/fs/xfs/xfs_mount.h
index 452ca7654dc5..316e0d79cc40 100644
--- a/fs/xfs/xfs_mount.h
+++ b/fs/xfs/xfs_mount.h
@@ -93,7 +93,7 @@ typedef struct xfs_mount {
 	struct workqueue_struct	*m_unwritten_workqueue;
 	struct workqueue_struct	*m_cil_workqueue;
 	struct workqueue_struct	*m_reclaim_workqueue;
-	struct workqueue_struct *m_eofblocks_workqueue;
+	struct workqueue_struct *m_blockgc_workqueue;
 	struct workqueue_struct	*m_sync_workqueue;
 
 	int			m_bsize;	/* fs logical block size */
@@ -177,9 +177,7 @@ typedef struct xfs_mount {
 	uint64_t		m_resblks_avail;/* available reserved blocks */
 	uint64_t		m_resblks_save;	/* reserved blks @ remount,ro */
 	struct delayed_work	m_reclaim_work;	/* background inode reclaim */
-	struct delayed_work	m_eofblocks_work; /* background eof blocks
-						     trimming */
-	struct delayed_work	m_cowblocks_work; /* background cow blocks
+	struct delayed_work	m_blockgc_work; /* background prealloc blocks
 						     trimming */
 	struct xfs_kobj		m_kobj;
 	struct xfs_kobj		m_error_kobj;
diff --git a/fs/xfs/xfs_super.c b/fs/xfs/xfs_super.c
index e8f71714d737..471592e8dba6 100644
--- a/fs/xfs/xfs_super.c
+++ b/fs/xfs/xfs_super.c
@@ -518,10 +518,10 @@ xfs_init_mount_workqueues(
 	if (!mp->m_reclaim_workqueue)
 		goto out_destroy_cil;
 
-	mp->m_eofblocks_workqueue = alloc_workqueue("xfs-eofblocks/%s",
+	mp->m_blockgc_workqueue = alloc_workqueue("xfs-blockgc/%s",
 			XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
 			0, mp->m_super->s_id);
-	if (!mp->m_eofblocks_workqueue)
+	if (!mp->m_blockgc_workqueue)
 		goto out_destroy_reclaim;
 
 	mp->m_sync_workqueue = alloc_workqueue("xfs-sync/%s",
@@ -532,7 +532,7 @@ xfs_init_mount_workqueues(
 	return 0;
 
 out_destroy_eofb:
-	destroy_workqueue(mp->m_eofblocks_workqueue);
+	destroy_workqueue(mp->m_blockgc_workqueue);
 out_destroy_reclaim:
 	destroy_workqueue(mp->m_reclaim_workqueue);
 out_destroy_cil:
@@ -550,7 +550,7 @@ xfs_destroy_mount_workqueues(
 	struct xfs_mount	*mp)
 {
 	destroy_workqueue(mp->m_sync_workqueue);
-	destroy_workqueue(mp->m_eofblocks_workqueue);
+	destroy_workqueue(mp->m_blockgc_workqueue);
 	destroy_workqueue(mp->m_reclaim_workqueue);
 	destroy_workqueue(mp->m_cil_workqueue);
 	destroy_workqueue(mp->m_unwritten_workqueue);
@@ -1842,8 +1842,7 @@ static int xfs_init_fs_context(
 	mutex_init(&mp->m_growlock);
 	INIT_WORK(&mp->m_flush_inodes_work, xfs_flush_inodes_worker);
 	INIT_DELAYED_WORK(&mp->m_reclaim_work, xfs_reclaim_worker);
-	INIT_DELAYED_WORK(&mp->m_eofblocks_work, xfs_eofblocks_worker);
-	INIT_DELAYED_WORK(&mp->m_cowblocks_work, xfs_cowblocks_worker);
+	INIT_DELAYED_WORK(&mp->m_blockgc_work, xfs_blockgc_worker);
 	mp->m_kobj.kobject.kset = xfs_kset;
 	/*
 	 * We don't create the finobt per-ag space reservation until after log
diff --git a/fs/xfs/xfs_sysctl.c b/fs/xfs/xfs_sysctl.c
index fac9de7ee6d0..145e06c47744 100644
--- a/fs/xfs/xfs_sysctl.c
+++ b/fs/xfs/xfs_sysctl.c
@@ -194,21 +194,12 @@ static struct ctl_table xfs_table[] = {
 	},
 	{
 		.procname	= "speculative_prealloc_lifetime",
-		.data		= &xfs_params.eofb_timer.val,
+		.data		= &xfs_params.blockgc_timer.val,
 		.maxlen		= sizeof(int),
 		.mode		= 0644,
 		.proc_handler	= proc_dointvec_minmax,
-		.extra1		= &xfs_params.eofb_timer.min,
-		.extra2		= &xfs_params.eofb_timer.max,
-	},
-	{
-		.procname	= "speculative_cow_prealloc_lifetime",
-		.data		= &xfs_params.cowb_timer.val,
-		.maxlen		= sizeof(int),
-		.mode		= 0644,
-		.proc_handler	= proc_dointvec_minmax,
-		.extra1		= &xfs_params.cowb_timer.min,
-		.extra2		= &xfs_params.cowb_timer.max,
+		.extra1		= &xfs_params.blockgc_timer.min,
+		.extra2		= &xfs_params.blockgc_timer.max,
 	},
 	/* please keep this the last entry */
 #ifdef CONFIG_PROC_FS
diff --git a/fs/xfs/xfs_sysctl.h b/fs/xfs/xfs_sysctl.h
index 8abf4640f1d5..7692e76ead33 100644
--- a/fs/xfs/xfs_sysctl.h
+++ b/fs/xfs/xfs_sysctl.h
@@ -35,8 +35,7 @@ typedef struct xfs_param {
 	xfs_sysctl_val_t rotorstep;	/* inode32 AG rotoring control knob */
 	xfs_sysctl_val_t inherit_nodfrg;/* Inherit the "nodefrag" inode flag. */
 	xfs_sysctl_val_t fstrm_timer;	/* Filestream dir-AG assoc'n timeout. */
-	xfs_sysctl_val_t eofb_timer;	/* Interval between eofb scan wakeups */
-	xfs_sysctl_val_t cowb_timer;	/* Interval between cowb scan wakeups */
+	xfs_sysctl_val_t blockgc_timer;	/* Interval between blockgc scans */
 } xfs_param_t;
 
 /*

[PATCH 07/11] xfs: only walk the incore inode tree once per blockgc scan

[Darrick J. Wong]

From: Darrick J. Wong <djwong@kernel.org>

Perform background block preallocation gc scans more efficiently by
walking the incore inode tree once.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
---
 fs/xfs/xfs_icache.c |   22 +++++++++++-----------
 1 file changed, 11 insertions(+), 11 deletions(-)


diff --git a/fs/xfs/xfs_icache.c b/fs/xfs/xfs_icache.c
index 0d0462268711..b32400b8e1ee 100644
--- a/fs/xfs/xfs_icache.c
+++ b/fs/xfs/xfs_icache.c
@@ -1571,21 +1571,19 @@ xfs_start_block_reaping(
 	xfs_blockgc_queue(mp);
 }
 
-/* Scan all incore inodes for block preallocations that we can remove. */
-static inline int
-xfs_blockgc_scan(
-	struct xfs_mount	*mp,
-	struct xfs_eofblocks	*eofb)
+/* Scan one incore inode for block preallocations that we can remove. */
+static int
+xfs_blockgc_scan_inode(
+	struct xfs_inode	*ip,
+	void			*args)
 {
 	int			error;
 
-	error = xfs_inode_walk(mp, 0, xfs_inode_free_eofblocks, eofb,
-			XFS_ICI_BLOCKGC_TAG);
+	error = xfs_inode_free_eofblocks(ip, args);
 	if (error)
 		return error;
 
-	error = xfs_inode_walk(mp, 0, xfs_inode_free_cowblocks, eofb,
-			XFS_ICI_BLOCKGC_TAG);
+	error = xfs_inode_free_cowblocks(ip, args);
 	if (error)
 		return error;
 
@@ -1603,7 +1601,8 @@ xfs_blockgc_worker(
 
 	if (!sb_start_write_trylock(mp->m_super))
 		return;
-	error = xfs_blockgc_scan(mp, NULL);
+	error = xfs_inode_walk(mp, 0, xfs_blockgc_scan_inode, NULL,
+			XFS_ICI_BLOCKGC_TAG);
 	if (error)
 		xfs_info(mp, "preallocation gc worker failed, err=%d", error);
 	sb_end_write(mp->m_super);
@@ -1620,7 +1619,8 @@ xfs_blockgc_free_space(
 {
 	trace_xfs_blockgc_free_space(mp, eofb, _RET_IP_);
 
-	return xfs_blockgc_scan(mp, eofb);
+	return xfs_inode_walk(mp, 0, xfs_blockgc_scan_inode, eofb,
+			XFS_ICI_BLOCKGC_TAG);
 }
 
 /*

[PATCH 08/11] xfs: rename block gc start and stop functions

[Darrick J. Wong]

From: Darrick J. Wong <djwong@kernel.org>

Shorten the names of the two functions that start and stop block
preallocation garbage collection and move them up to the other blockgc
functions.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
---
 fs/xfs/scrub/common.c |    4 ++--
 fs/xfs/xfs_icache.c   |    4 ++--
 fs/xfs/xfs_icache.h   |    4 ++--
 fs/xfs/xfs_mount.c    |    2 +-
 fs/xfs/xfs_super.c    |    8 ++++----
 5 files changed, 11 insertions(+), 11 deletions(-)


diff --git a/fs/xfs/scrub/common.c b/fs/xfs/scrub/common.c
index 8ea6d4aa3f55..53456f3de881 100644
--- a/fs/xfs/scrub/common.c
+++ b/fs/xfs/scrub/common.c
@@ -888,7 +888,7 @@ xchk_stop_reaping(
 	struct xfs_scrub	*sc)
 {
 	sc->flags |= XCHK_REAPING_DISABLED;
-	xfs_stop_block_reaping(sc->mp);
+	xfs_blockgc_stop(sc->mp);
 }
 
 /* Restart background reaping of resources. */
@@ -896,6 +896,6 @@ void
 xchk_start_reaping(
 	struct xfs_scrub	*sc)
 {
-	xfs_start_block_reaping(sc->mp);
+	xfs_blockgc_start(sc->mp);
 	sc->flags &= ~XCHK_REAPING_DISABLED;
 }
diff --git a/fs/xfs/xfs_icache.c b/fs/xfs/xfs_icache.c
index b32400b8e1ee..675c9e17dac0 100644
--- a/fs/xfs/xfs_icache.c
+++ b/fs/xfs/xfs_icache.c
@@ -1557,7 +1557,7 @@ xfs_inode_clear_cowblocks_tag(
 
 /* Disable post-EOF and CoW block auto-reclamation. */
 void
-xfs_stop_block_reaping(
+xfs_blockgc_stop(
 	struct xfs_mount	*mp)
 {
 	cancel_delayed_work_sync(&mp->m_blockgc_work);
@@ -1565,7 +1565,7 @@ xfs_stop_block_reaping(
 
 /* Enable post-EOF and CoW block auto-reclamation. */
 void
-xfs_start_block_reaping(
+xfs_blockgc_start(
 	struct xfs_mount	*mp)
 {
 	xfs_blockgc_queue(mp);
diff --git a/fs/xfs/xfs_icache.h b/fs/xfs/xfs_icache.h
index d781703af025..6ffb2fc5e458 100644
--- a/fs/xfs/xfs_icache.h
+++ b/fs/xfs/xfs_icache.h
@@ -74,7 +74,7 @@ int xfs_inode_walk(struct xfs_mount *mp, int iter_flags,
 int xfs_icache_inode_is_allocated(struct xfs_mount *mp, struct xfs_trans *tp,
 				  xfs_ino_t ino, bool *inuse);
 
-void xfs_stop_block_reaping(struct xfs_mount *mp);
-void xfs_start_block_reaping(struct xfs_mount *mp);
+void xfs_blockgc_stop(struct xfs_mount *mp);
+void xfs_blockgc_start(struct xfs_mount *mp);
 
 #endif
diff --git a/fs/xfs/xfs_mount.c b/fs/xfs/xfs_mount.c
index 53b8ccab7235..be9ce114527f 100644
--- a/fs/xfs/xfs_mount.c
+++ b/fs/xfs/xfs_mount.c
@@ -1054,7 +1054,7 @@ xfs_unmountfs(
 	uint64_t		resblks;
 	int			error;
 
-	xfs_stop_block_reaping(mp);
+	xfs_blockgc_stop(mp);
 	xfs_fs_unreserve_ag_blocks(mp);
 	xfs_qm_unmount_quotas(mp);
 	xfs_rtunmount_inodes(mp);
diff --git a/fs/xfs/xfs_super.c b/fs/xfs/xfs_super.c
index 471592e8dba6..ea942089d074 100644
--- a/fs/xfs/xfs_super.c
+++ b/fs/xfs/xfs_super.c
@@ -891,7 +891,7 @@ xfs_fs_freeze(
 	 * set a GFP_NOFS context here to avoid recursion deadlocks.
 	 */
 	flags = memalloc_nofs_save();
-	xfs_stop_block_reaping(mp);
+	xfs_blockgc_stop(mp);
 	xfs_save_resvblks(mp);
 	ret = xfs_log_quiesce(mp);
 	memalloc_nofs_restore(flags);
@@ -906,7 +906,7 @@ xfs_fs_unfreeze(
 
 	xfs_restore_resvblks(mp);
 	xfs_log_work_queue(mp);
-	xfs_start_block_reaping(mp);
+	xfs_blockgc_start(mp);
 	return 0;
 }
 
@@ -1690,7 +1690,7 @@ xfs_remount_rw(
 		xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
 		return error;
 	}
-	xfs_start_block_reaping(mp);
+	xfs_blockgc_start(mp);
 
 	/* Create the per-AG metadata reservation pool .*/
 	error = xfs_fs_reserve_ag_blocks(mp);
@@ -1710,7 +1710,7 @@ xfs_remount_ro(
 	 * Cancel background eofb scanning so it cannot race with the final
 	 * log force+buftarg wait and deadlock the remount.
 	 */
-	xfs_stop_block_reaping(mp);
+	xfs_blockgc_stop(mp);
 
 	/* Get rid of any leftover CoW reservations... */
 	error = xfs_blockgc_free_space(mp, NULL);

[PATCH 09/11] xfs: parallelize block preallocation garbage collection

[Darrick J. Wong]

From: Darrick J. Wong <djwong@kernel.org>

Split the block preallocation garbage collection work into per-AG work
items so that we can take advantage of parallelization.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
---
 fs/xfs/xfs_icache.c |   42 ++++++++++++++++++++++++++++++------------
 fs/xfs/xfs_mount.c  |    3 +++
 fs/xfs/xfs_mount.h  |    5 +++--
 fs/xfs/xfs_super.c  |    4 ++--
 4 files changed, 38 insertions(+), 16 deletions(-)


diff --git a/fs/xfs/xfs_icache.c b/fs/xfs/xfs_icache.c
index 675c9e17dac0..ad881e92d6cd 100644
--- a/fs/xfs/xfs_icache.c
+++ b/fs/xfs/xfs_icache.c
@@ -1333,12 +1333,12 @@ xfs_inode_free_eofblocks(
  */
 static inline void
 xfs_blockgc_queue(
-	struct xfs_mount	*mp)
+	struct xfs_perag	*pag)
 {
 	rcu_read_lock();
-	if (radix_tree_tagged(&mp->m_perag_tree, XFS_ICI_BLOCKGC_TAG))
-		queue_delayed_work(mp->m_blockgc_workqueue,
-				   &mp->m_blockgc_work,
+	if (radix_tree_tagged(&pag->pag_ici_root, XFS_ICI_BLOCKGC_TAG))
+		queue_delayed_work(pag->pag_mount->m_blockgc_workqueue,
+				   &pag->pag_blockgc_work,
 				   msecs_to_jiffies(xfs_blockgc_secs * 1000));
 	rcu_read_unlock();
 }
@@ -1380,7 +1380,7 @@ xfs_blockgc_set_iflag(
 		spin_unlock(&ip->i_mount->m_perag_lock);
 
 		/* kick off background trimming */
-		xfs_blockgc_queue(ip->i_mount);
+		xfs_blockgc_queue(pag);
 
 		trace_xfs_perag_set_blockgc(ip->i_mount, pag->pag_agno, -1,
 				_RET_IP_);
@@ -1555,12 +1555,24 @@ xfs_inode_clear_cowblocks_tag(
 	return xfs_blockgc_clear_iflag(ip, XFS_ICOWBLOCKS);
 }
 
+#define for_each_perag_tag(mp, next_agno, pag, tag) \
+	for ((next_agno) = 0, (pag) = xfs_perag_get_tag((mp), 0, (tag)); \
+		(pag) != NULL; \
+		(next_agno) = (pag)->pag_agno + 1, \
+		xfs_perag_put(pag), \
+		(pag) = xfs_perag_get_tag((mp), (next_agno), (tag)))
+
+
 /* Disable post-EOF and CoW block auto-reclamation. */
 void
 xfs_blockgc_stop(
 	struct xfs_mount	*mp)
 {
-	cancel_delayed_work_sync(&mp->m_blockgc_work);
+	struct xfs_perag	*pag;
+	xfs_agnumber_t		agno;
+
+	for_each_perag_tag(mp, agno, pag, XFS_ICI_BLOCKGC_TAG)
+		cancel_delayed_work_sync(&pag->pag_blockgc_work);
 }
 
 /* Enable post-EOF and CoW block auto-reclamation. */
@@ -1568,7 +1580,11 @@ void
 xfs_blockgc_start(
 	struct xfs_mount	*mp)
 {
-	xfs_blockgc_queue(mp);
+	struct xfs_perag	*pag;
+	xfs_agnumber_t		agno;
+
+	for_each_perag_tag(mp, agno, pag, XFS_ICI_BLOCKGC_TAG)
+		xfs_blockgc_queue(pag);
 }
 
 /* Scan one incore inode for block preallocations that we can remove. */
@@ -1595,18 +1611,20 @@ void
 xfs_blockgc_worker(
 	struct work_struct	*work)
 {
-	struct xfs_mount	*mp = container_of(to_delayed_work(work),
-					struct xfs_mount, m_blockgc_work);
+	struct xfs_perag	*pag = container_of(to_delayed_work(work),
+					struct xfs_perag, pag_blockgc_work);
+	struct xfs_mount	*mp = pag->pag_mount;
 	int			error;
 
 	if (!sb_start_write_trylock(mp->m_super))
 		return;
-	error = xfs_inode_walk(mp, 0, xfs_blockgc_scan_inode, NULL,
+	error = xfs_inode_walk_ag(pag, 0, xfs_blockgc_scan_inode, NULL,
 			XFS_ICI_BLOCKGC_TAG);
 	if (error)
-		xfs_info(mp, "preallocation gc worker failed, err=%d", error);
+		xfs_info(mp, "AG %u preallocation gc worker failed, err=%d",
+				pag->pag_agno, error);
 	sb_end_write(mp->m_super);
-	xfs_blockgc_queue(mp);
+	xfs_blockgc_queue(pag);
 }
 
 /*
diff --git a/fs/xfs/xfs_mount.c b/fs/xfs/xfs_mount.c
index be9ce114527f..52370d0a3f43 100644
--- a/fs/xfs/xfs_mount.c
+++ b/fs/xfs/xfs_mount.c
@@ -126,6 +126,7 @@ __xfs_free_perag(
 {
 	struct xfs_perag *pag = container_of(head, struct xfs_perag, rcu_head);
 
+	ASSERT(!delayed_work_pending(&pag->pag_blockgc_work));
 	ASSERT(atomic_read(&pag->pag_ref) == 0);
 	kmem_free(pag);
 }
@@ -146,6 +147,7 @@ xfs_free_perag(
 		spin_unlock(&mp->m_perag_lock);
 		ASSERT(pag);
 		ASSERT(atomic_read(&pag->pag_ref) == 0);
+		cancel_delayed_work_sync(&pag->pag_blockgc_work);
 		xfs_iunlink_destroy(pag);
 		xfs_buf_hash_destroy(pag);
 		call_rcu(&pag->rcu_head, __xfs_free_perag);
@@ -201,6 +203,7 @@ xfs_initialize_perag(
 		pag->pag_agno = index;
 		pag->pag_mount = mp;
 		spin_lock_init(&pag->pag_ici_lock);
+		INIT_DELAYED_WORK(&pag->pag_blockgc_work, xfs_blockgc_worker);
 		INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC);
 
 		error = xfs_buf_hash_init(pag);
diff --git a/fs/xfs/xfs_mount.h b/fs/xfs/xfs_mount.h
index 316e0d79cc40..659ad95fe3e0 100644
--- a/fs/xfs/xfs_mount.h
+++ b/fs/xfs/xfs_mount.h
@@ -177,8 +177,6 @@ typedef struct xfs_mount {
 	uint64_t		m_resblks_avail;/* available reserved blocks */
 	uint64_t		m_resblks_save;	/* reserved blks @ remount,ro */
 	struct delayed_work	m_reclaim_work;	/* background inode reclaim */
-	struct delayed_work	m_blockgc_work; /* background prealloc blocks
-						     trimming */
 	struct xfs_kobj		m_kobj;
 	struct xfs_kobj		m_error_kobj;
 	struct xfs_kobj		m_error_meta_kobj;
@@ -367,6 +365,9 @@ typedef struct xfs_perag {
 	/* Blocks reserved for the reverse mapping btree. */
 	struct xfs_ag_resv	pag_rmapbt_resv;
 
+	/* background prealloc block trimming */
+	struct delayed_work	pag_blockgc_work;
+
 	/* reference count */
 	uint8_t			pagf_refcount_level;
 
diff --git a/fs/xfs/xfs_super.c b/fs/xfs/xfs_super.c
index ea942089d074..2b04818627e9 100644
--- a/fs/xfs/xfs_super.c
+++ b/fs/xfs/xfs_super.c
@@ -35,6 +35,7 @@
 #include "xfs_refcount_item.h"
 #include "xfs_bmap_item.h"
 #include "xfs_reflink.h"
+#include "xfs_pwork.h"
 
 #include <linux/magic.h>
 #include <linux/fs_context.h>
@@ -519,7 +520,7 @@ xfs_init_mount_workqueues(
 		goto out_destroy_cil;
 
 	mp->m_blockgc_workqueue = alloc_workqueue("xfs-blockgc/%s",
-			XFS_WQFLAGS(WQ_FREEZABLE | WQ_MEM_RECLAIM),
+			XFS_WQFLAGS(WQ_UNBOUND | WQ_FREEZABLE | WQ_MEM_RECLAIM),
 			0, mp->m_super->s_id);
 	if (!mp->m_blockgc_workqueue)
 		goto out_destroy_reclaim;
@@ -1842,7 +1843,6 @@ static int xfs_init_fs_context(
 	mutex_init(&mp->m_growlock);
 	INIT_WORK(&mp->m_flush_inodes_work, xfs_flush_inodes_worker);
 	INIT_DELAYED_WORK(&mp->m_reclaim_work, xfs_reclaim_worker);
-	INIT_DELAYED_WORK(&mp->m_blockgc_work, xfs_blockgc_worker);
 	mp->m_kobj.kobject.kset = xfs_kset;
 	/*
 	 * We don't create the finobt per-ag space reservation until after log

[PATCH 10/11] xfs: expose the blockgc workqueue knobs publicly

[Darrick J. Wong]

From: Darrick J. Wong <djwong@kernel.org>

Expose the workqueue sysfs knobs for the speculative preallocation gc
workers on all kernels, and update the sysadmin information.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
---
 Documentation/admin-guide/xfs.rst |    3 +++
 fs/xfs/xfs_super.c                |    2 +-
 2 files changed, 4 insertions(+), 1 deletion(-)


diff --git a/Documentation/admin-guide/xfs.rst b/Documentation/admin-guide/xfs.rst
index b00b1eece9de..d2064a52811b 100644
--- a/Documentation/admin-guide/xfs.rst
+++ b/Documentation/admin-guide/xfs.rst
@@ -518,6 +518,9 @@ and the short name of the data device.  They all can be found in:
 ================  ===========
   xfs_iwalk-$pid  Inode scans of the entire filesystem. Currently limited to
                   mount time quotacheck.
+  xfs-blockgc     Background garbage collection of disk space that have been
+                  speculatively allocated beyond EOF or for staging copy on
+                  write operations.
 ================  ===========
 
 For example, the knobs for the quotacheck workqueue for /dev/nvme0n1 would be
diff --git a/fs/xfs/xfs_super.c b/fs/xfs/xfs_super.c
index 2b04818627e9..21b1d034aca3 100644
--- a/fs/xfs/xfs_super.c
+++ b/fs/xfs/xfs_super.c
@@ -520,7 +520,7 @@ xfs_init_mount_workqueues(
 		goto out_destroy_cil;
 
 	mp->m_blockgc_workqueue = alloc_workqueue("xfs-blockgc/%s",
-			XFS_WQFLAGS(WQ_UNBOUND | WQ_FREEZABLE | WQ_MEM_RECLAIM),
+			WQ_SYSFS | WQ_UNBOUND | WQ_FREEZABLE | WQ_MEM_RECLAIM,
 			0, mp->m_super->s_id);
 	if (!mp->m_blockgc_workqueue)
 		goto out_destroy_reclaim;

[PATCH 11/11] xfs: don't bounce the iolock between free_{eof,cow}blocks

[Darrick J. Wong]

From: Darrick J. Wong <djwong@kernel.org>

Since xfs_inode_free_eofblocks and xfs_inode_free_cowblocks are now
internal static functions, we can save ourselves a cycling of the iolock
by passing the lock state out to xfs_blockgc_scan_inode and letting it
do all the unlocking.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
---
 fs/xfs/xfs_icache.c |   43 +++++++++++++++++++++----------------------
 1 file changed, 21 insertions(+), 22 deletions(-)


diff --git a/fs/xfs/xfs_icache.c b/fs/xfs/xfs_icache.c
index ad881e92d6cd..2e14ee4bddae 100644
--- a/fs/xfs/xfs_icache.c
+++ b/fs/xfs/xfs_icache.c
@@ -1283,11 +1283,11 @@ xfs_reclaim_worker(
 STATIC int
 xfs_inode_free_eofblocks(
 	struct xfs_inode	*ip,
-	void			*args)
+	void			*args,
+	unsigned int		*unlockflags)
 {
 	struct xfs_eofblocks	*eofb = args;
 	bool			wait;
-	int			ret;
 
 	wait = eofb && (eofb->eof_flags & XFS_EOF_FLAGS_SYNC);
 
@@ -1320,11 +1320,9 @@ xfs_inode_free_eofblocks(
 			return -EAGAIN;
 		return 0;
 	}
+	*unlockflags |= XFS_IOLOCK_EXCL;
 
-	ret = xfs_free_eofblocks(ip);
-	xfs_iunlock(ip, XFS_IOLOCK_EXCL);
-
-	return ret;
+	return xfs_free_eofblocks(ip);
 }
 
 /*
@@ -1493,7 +1491,8 @@ xfs_prep_free_cowblocks(
 STATIC int
 xfs_inode_free_cowblocks(
 	struct xfs_inode	*ip,
-	void			*args)
+	void			*args,
+	unsigned int		*unlockflags)
 {
 	struct xfs_eofblocks	*eofb = args;
 	bool			wait;
@@ -1514,16 +1513,20 @@ xfs_inode_free_cowblocks(
 	 * If the caller is waiting, return -EAGAIN to keep the background
 	 * scanner moving and revisit the inode in a subsequent pass.
 	 */
-	if (!xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) {
+	if (!(*unlockflags & XFS_IOLOCK_EXCL) &&
+	    !xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) {
 		if (wait)
 			return -EAGAIN;
 		return 0;
 	}
+	*unlockflags |= XFS_IOLOCK_EXCL;
+
 	if (!xfs_ilock_nowait(ip, XFS_MMAPLOCK_EXCL)) {
 		if (wait)
-			ret = -EAGAIN;
-		goto out_iolock;
+			return -EAGAIN;
+		return 0;
 	}
+	*unlockflags |= XFS_MMAPLOCK_EXCL;
 
 	/*
 	 * Check again, nobody else should be able to dirty blocks or change
@@ -1531,11 +1534,6 @@ xfs_inode_free_cowblocks(
 	 */
 	if (xfs_prep_free_cowblocks(ip))
 		ret = xfs_reflink_cancel_cow_range(ip, 0, NULLFILEOFF, false);
-
-	xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
-out_iolock:
-	xfs_iunlock(ip, XFS_IOLOCK_EXCL);
-
 	return ret;
 }
 
@@ -1593,17 +1591,18 @@ xfs_blockgc_scan_inode(
 	struct xfs_inode	*ip,
 	void			*args)
 {
+	unsigned int		unlockflags = 0;
 	int			error;
 
-	error = xfs_inode_free_eofblocks(ip, args);
+	error = xfs_inode_free_eofblocks(ip, args, &unlockflags);
 	if (error)
-		return error;
+		goto unlock;
 
-	error = xfs_inode_free_cowblocks(ip, args);
-	if (error)
-		return error;
-
-	return 0;
+	error = xfs_inode_free_cowblocks(ip, args, &unlockflags);
+unlock:
+	if (unlockflags)
+		xfs_iunlock(ip, unlockflags);
+	return error;
 }
 
 /* Background worker that trims preallocated space. */

Re: [PATCH 11/11] xfs: don't bounce the iolock between free_{eof,cow}blocks

[Christoph Hellwig]

On Wed, Jan 27, 2021 at 10:04:41PM -0800, Darrick J. Wong wrote:
> From: Darrick J. Wong <djwong@kernel.org>
> 
> Since xfs_inode_free_eofblocks and xfs_inode_free_cowblocks are now
> internal static functions, we can save ourselves a cycling of the iolock
> by passing the lock state out to xfs_blockgc_scan_inode and letting it
> do all the unlocking.

Looks good, although normally we use lock_flags and not unlockflags
for this kind of variable, and maybe we should stick to that here.

Reviewed-by: Christoph Hellwig <hch@lst.de>