* [PATCH v6 0/9] ceph: size handling for the fscrypt
@ 2021-11-04 5:52 xiubli
2021-11-04 5:52 ` [PATCH v6 1/9] libceph: add CEPH_OSD_OP_ASSERT_VER support xiubli
` (9 more replies)
0 siblings, 10 replies; 17+ messages in thread
From: xiubli @ 2021-11-04 5:52 UTC (permalink / raw)
To: jlayton; +Cc: idryomov, vshankar, pdonnell, khiremat, ceph-devel, Xiubo Li
From: Xiubo Li <xiubli@redhat.com>
This patch series is based on the "wip-fscrypt-fnames" branch in
repo https://github.com/ceph/ceph-client.git.
And I have picked up 5 patches from the "ceph-fscrypt-size-experimental"
branch in repo
https://git.kernel.org/pub/scm/linux/kernel/git/jlayton/linux.git.
====
This approach is based on the discussion from V1 and V2, which will
pass the encrypted last block contents to MDS along with the truncate
request.
This will send the encrypted last block contents to MDS along with
the truncate request when truncating to a smaller size and at the
same time new size does not align to BLOCK SIZE.
The MDS side patch is raised in PR
https://github.com/ceph/ceph/pull/43588, which is also based Jeff's
previous great work in PR https://github.com/ceph/ceph/pull/41284.
The MDS will use the filer.write_trunc(), which could update and
truncate the file in one shot, instead of filer.truncate().
This just assume kclient won't support the inline data feature, which
will be remove soon, more detail please see:
https://tracker.ceph.com/issues/52916
Changed in V6:
- Fixed the file hole bug, also have updated the MDS side PR.
- Add add object version support for sync read in #8.
Changed in V5:
- Rebase to "wip-fscrypt-fnames" branch in ceph-client.git repo.
- Pick up 5 patches from Jeff's "ceph-fscrypt-size-experimental" branch
in linux.git repo.
- Add "i_truncate_pagecache_size" member support in ceph_inode_info
struct, this will be used to truncate the pagecache only in kclient
side, because the "i_truncate_size" will always be aligned to BLOCK
SIZE. In fscrypt case we need to use the real size to truncate the
pagecache.
Changed in V4:
- Retry the truncate request by 20 times before fail it with -EAGAIN.
- Remove the "fill_last_block" label and move the code to else branch.
- Remove the #3 patch, which has already been sent out separately, in
V3 series.
- Improve some comments in the code.
Changed in V3:
- Fix possibly corrupting the file just before the MDS acquires the
xlock for FILE lock, another client has updated it.
- Flush the pagecache buffer before reading the last block for the
when filling the truncate request.
- Some other minore fixes.
Jeff Layton (5):
libceph: add CEPH_OSD_OP_ASSERT_VER support
ceph: size handling for encrypted inodes in cap updates
ceph: fscrypt_file field handling in MClientRequest messages
ceph: get file size from fscrypt_file when present in inode traces
ceph: handle fscrypt fields in cap messages from MDS
Xiubo Li (4):
ceph: add __ceph_get_caps helper support
ceph: add __ceph_sync_read helper support
ceph: add object version support for sync read
ceph: add truncate size handling support for fscrypt
fs/ceph/caps.c | 136 ++++++++++++++----
fs/ceph/crypto.h | 4 +
fs/ceph/dir.c | 3 +
fs/ceph/file.c | 76 ++++++++--
fs/ceph/inode.c | 243 +++++++++++++++++++++++++++++---
fs/ceph/mds_client.c | 9 +-
fs/ceph/mds_client.h | 2 +
fs/ceph/super.h | 25 ++++
include/linux/ceph/crypto.h | 28 ++++
include/linux/ceph/osd_client.h | 6 +-
include/linux/ceph/rados.h | 4 +
net/ceph/osd_client.c | 5 +
12 files changed, 482 insertions(+), 59 deletions(-)
create mode 100644 include/linux/ceph/crypto.h
--
2.27.0
^ permalink raw reply [flat|nested] 17+ messages in thread
* [PATCH v6 1/9] libceph: add CEPH_OSD_OP_ASSERT_VER support
2021-11-04 5:52 [PATCH v6 0/9] ceph: size handling for the fscrypt xiubli
@ 2021-11-04 5:52 ` xiubli
2021-11-04 5:52 ` [PATCH v6 2/9] ceph: size handling for encrypted inodes in cap updates xiubli
` (8 subsequent siblings)
9 siblings, 0 replies; 17+ messages in thread
From: xiubli @ 2021-11-04 5:52 UTC (permalink / raw)
To: jlayton; +Cc: idryomov, vshankar, pdonnell, khiremat, ceph-devel
From: Jeff Layton <jlayton@kernel.org>
...and record the user_version in the reply in a new field in
ceph_osd_request, so we can populate the assert_ver appropriately.
Shuffle the fields a bit too so that the new field fits in an
existing hole on x86_64.
Signed-off-by: Jeff Layton <jlayton@kernel.org>
---
include/linux/ceph/osd_client.h | 6 +++++-
include/linux/ceph/rados.h | 4 ++++
net/ceph/osd_client.c | 5 +++++
3 files changed, 14 insertions(+), 1 deletion(-)
diff --git a/include/linux/ceph/osd_client.h b/include/linux/ceph/osd_client.h
index 83fa08a06507..7ee1684d3edc 100644
--- a/include/linux/ceph/osd_client.h
+++ b/include/linux/ceph/osd_client.h
@@ -145,6 +145,9 @@ struct ceph_osd_req_op {
u32 src_fadvise_flags;
struct ceph_osd_data osd_data;
} copy_from;
+ struct {
+ u64 ver;
+ } assert_ver;
};
};
@@ -199,6 +202,7 @@ struct ceph_osd_request {
struct ceph_osd_client *r_osdc;
struct kref r_kref;
bool r_mempool;
+ bool r_linger; /* don't resend on failure */
struct completion r_completion; /* private to osd_client.c */
ceph_osdc_callback_t r_callback;
@@ -211,9 +215,9 @@ struct ceph_osd_request {
struct ceph_snap_context *r_snapc; /* for writes */
struct timespec64 r_mtime; /* ditto */
u64 r_data_offset; /* ditto */
- bool r_linger; /* don't resend on failure */
/* internal */
+ u64 r_version; /* data version sent in reply */
unsigned long r_stamp; /* jiffies, send or check time */
unsigned long r_start_stamp; /* jiffies */
ktime_t r_start_latency; /* ktime_t */
diff --git a/include/linux/ceph/rados.h b/include/linux/ceph/rados.h
index 43a7a1573b51..73c3efbec36c 100644
--- a/include/linux/ceph/rados.h
+++ b/include/linux/ceph/rados.h
@@ -523,6 +523,10 @@ struct ceph_osd_op {
struct {
__le64 cookie;
} __attribute__ ((packed)) notify;
+ struct {
+ __le64 unused;
+ __le64 ver;
+ } __attribute__ ((packed)) assert_ver;
struct {
__le64 offset, length;
__le64 src_offset;
diff --git a/net/ceph/osd_client.c b/net/ceph/osd_client.c
index ff8624a7c964..f3a9af012123 100644
--- a/net/ceph/osd_client.c
+++ b/net/ceph/osd_client.c
@@ -1038,6 +1038,10 @@ static u32 osd_req_encode_op(struct ceph_osd_op *dst,
dst->copy_from.src_fadvise_flags =
cpu_to_le32(src->copy_from.src_fadvise_flags);
break;
+ case CEPH_OSD_OP_ASSERT_VER:
+ dst->assert_ver.unused = cpu_to_le64(0);
+ dst->assert_ver.ver = cpu_to_le64(src->assert_ver.ver);
+ break;
default:
pr_err("unsupported osd opcode %s\n",
ceph_osd_op_name(src->op));
@@ -3763,6 +3767,7 @@ static void handle_reply(struct ceph_osd *osd, struct ceph_msg *msg)
* one (type of) reply back.
*/
WARN_ON(!(m.flags & CEPH_OSD_FLAG_ONDISK));
+ req->r_version = m.user_version;
req->r_result = m.result ?: data_len;
finish_request(req);
mutex_unlock(&osd->lock);
--
2.27.0
^ permalink raw reply related [flat|nested] 17+ messages in thread
* [PATCH v6 2/9] ceph: size handling for encrypted inodes in cap updates
2021-11-04 5:52 [PATCH v6 0/9] ceph: size handling for the fscrypt xiubli
2021-11-04 5:52 ` [PATCH v6 1/9] libceph: add CEPH_OSD_OP_ASSERT_VER support xiubli
@ 2021-11-04 5:52 ` xiubli
2021-11-04 5:52 ` [PATCH v6 3/9] ceph: fscrypt_file field handling in MClientRequest messages xiubli
` (7 subsequent siblings)
9 siblings, 0 replies; 17+ messages in thread
From: xiubli @ 2021-11-04 5:52 UTC (permalink / raw)
To: jlayton; +Cc: idryomov, vshankar, pdonnell, khiremat, ceph-devel
From: Jeff Layton <jlayton@kernel.org>
Transmit the rounded-up size as the normal size, and fill out the
fscrypt_file field with the real file size.
Signed-off-by: Jeff Layton <jlayton@kernel.org>
---
fs/ceph/caps.c | 43 +++++++++++++++++++++++++------------------
fs/ceph/crypto.h | 4 ++++
2 files changed, 29 insertions(+), 18 deletions(-)
diff --git a/fs/ceph/caps.c b/fs/ceph/caps.c
index 80f521dd7254..fc367f42536a 100644
--- a/fs/ceph/caps.c
+++ b/fs/ceph/caps.c
@@ -1215,10 +1215,9 @@ struct cap_msg_args {
umode_t mode;
bool inline_data;
bool wake;
+ bool encrypted;
u32 fscrypt_auth_len;
- u32 fscrypt_file_len;
u8 fscrypt_auth[sizeof(struct ceph_fscrypt_auth)]; // for context
- u8 fscrypt_file[sizeof(u64)]; // for size
};
/* Marshal up the cap msg to the MDS */
@@ -1253,7 +1252,12 @@ static void encode_cap_msg(struct ceph_msg *msg, struct cap_msg_args *arg)
fc->ino = cpu_to_le64(arg->ino);
fc->snap_follows = cpu_to_le64(arg->follows);
- fc->size = cpu_to_le64(arg->size);
+#if IS_ENABLED(CONFIG_FS_ENCRYPTION)
+ if (arg->encrypted)
+ fc->size = cpu_to_le64(round_up(arg->size, CEPH_FSCRYPT_BLOCK_SIZE));
+ else
+#endif
+ fc->size = cpu_to_le64(arg->size);
fc->max_size = cpu_to_le64(arg->max_size);
ceph_encode_timespec64(&fc->mtime, &arg->mtime);
ceph_encode_timespec64(&fc->atime, &arg->atime);
@@ -1313,11 +1317,17 @@ static void encode_cap_msg(struct ceph_msg *msg, struct cap_msg_args *arg)
ceph_encode_64(&p, 0);
#if IS_ENABLED(CONFIG_FS_ENCRYPTION)
- /* fscrypt_auth and fscrypt_file (version 12) */
+ /*
+ * fscrypt_auth and fscrypt_file (version 12)
+ *
+ * fscrypt_auth holds the crypto context (if any). fscrypt_file
+ * tracks the real i_size as an __le64 field (and we use a rounded-up
+ * i_size in * the traditional size field).
+ */
ceph_encode_32(&p, arg->fscrypt_auth_len);
ceph_encode_copy(&p, arg->fscrypt_auth, arg->fscrypt_auth_len);
- ceph_encode_32(&p, arg->fscrypt_file_len);
- ceph_encode_copy(&p, arg->fscrypt_file, arg->fscrypt_file_len);
+ ceph_encode_32(&p, sizeof(__le64));
+ ceph_encode_64(&p, arg->size);
#else /* CONFIG_FS_ENCRYPTION */
ceph_encode_32(&p, 0);
ceph_encode_32(&p, 0);
@@ -1389,7 +1399,6 @@ static void __prep_cap(struct cap_msg_args *arg, struct ceph_cap *cap,
arg->follows = flushing ? ci->i_head_snapc->seq : 0;
arg->flush_tid = flush_tid;
arg->oldest_flush_tid = oldest_flush_tid;
-
arg->size = i_size_read(inode);
ci->i_reported_size = arg->size;
arg->max_size = ci->i_wanted_max_size;
@@ -1443,6 +1452,7 @@ static void __prep_cap(struct cap_msg_args *arg, struct ceph_cap *cap,
}
}
arg->flags = flags;
+ arg->encrypted = IS_ENCRYPTED(inode);
#if IS_ENABLED(CONFIG_FS_ENCRYPTION)
if (ci->fscrypt_auth_len &&
WARN_ON_ONCE(ci->fscrypt_auth_len != sizeof(struct ceph_fscrypt_auth))) {
@@ -1453,21 +1463,21 @@ static void __prep_cap(struct cap_msg_args *arg, struct ceph_cap *cap,
memcpy(arg->fscrypt_auth, ci->fscrypt_auth,
min_t(size_t, ci->fscrypt_auth_len, sizeof(arg->fscrypt_auth)));
}
- /* FIXME: use this to track "real" size */
- arg->fscrypt_file_len = 0;
#endif /* CONFIG_FS_ENCRYPTION */
}
+#if IS_ENABLED(CONFIG_FS_ENCRYPTION)
#define CAP_MSG_FIXED_FIELDS (sizeof(struct ceph_mds_caps) + \
- 4 + 8 + 4 + 4 + 8 + 4 + 4 + 4 + 8 + 8 + 4 + 8 + 8 + 4 + 4)
+ 4 + 8 + 4 + 4 + 8 + 4 + 4 + 4 + 8 + 8 + 4 + 8 + 8 + 4 + 4 + 8)
-#if IS_ENABLED(CONFIG_FS_ENCRYPTION)
static inline int cap_msg_size(struct cap_msg_args *arg)
{
- return CAP_MSG_FIXED_FIELDS + arg->fscrypt_auth_len +
- arg->fscrypt_file_len;
+ return CAP_MSG_FIXED_FIELDS + arg->fscrypt_auth_len;
}
#else
+#define CAP_MSG_FIXED_FIELDS (sizeof(struct ceph_mds_caps) + \
+ 4 + 8 + 4 + 4 + 8 + 4 + 4 + 4 + 8 + 8 + 4 + 8 + 8 + 4 + 4)
+
static inline int cap_msg_size(struct cap_msg_args *arg)
{
return CAP_MSG_FIXED_FIELDS;
@@ -1546,13 +1556,10 @@ static inline int __send_flush_snap(struct inode *inode,
arg.inline_data = capsnap->inline_data;
arg.flags = 0;
arg.wake = false;
+ arg.encrypted = IS_ENCRYPTED(inode);
- /*
- * No fscrypt_auth changes from a capsnap. It will need
- * to update fscrypt_file on size changes (TODO).
- */
+ /* No fscrypt_auth changes from a capsnap.*/
arg.fscrypt_auth_len = 0;
- arg.fscrypt_file_len = 0;
msg = ceph_msg_new(CEPH_MSG_CLIENT_CAPS, cap_msg_size(&arg),
GFP_NOFS, false);
diff --git a/fs/ceph/crypto.h b/fs/ceph/crypto.h
index c2e0cbb5667b..ab27a7ed62c3 100644
--- a/fs/ceph/crypto.h
+++ b/fs/ceph/crypto.h
@@ -9,6 +9,10 @@
#include <crypto/sha2.h>
#include <linux/fscrypt.h>
+#define CEPH_FSCRYPT_BLOCK_SHIFT 12
+#define CEPH_FSCRYPT_BLOCK_SIZE (_AC(1,UL) << CEPH_FSCRYPT_BLOCK_SHIFT)
+#define CEPH_FSCRYPT_BLOCK_MASK (~(CEPH_FSCRYPT_BLOCK_SIZE-1))
+
struct ceph_fs_client;
struct ceph_acl_sec_ctx;
struct ceph_mds_request;
--
2.27.0
^ permalink raw reply related [flat|nested] 17+ messages in thread
* [PATCH v6 3/9] ceph: fscrypt_file field handling in MClientRequest messages
2021-11-04 5:52 [PATCH v6 0/9] ceph: size handling for the fscrypt xiubli
2021-11-04 5:52 ` [PATCH v6 1/9] libceph: add CEPH_OSD_OP_ASSERT_VER support xiubli
2021-11-04 5:52 ` [PATCH v6 2/9] ceph: size handling for encrypted inodes in cap updates xiubli
@ 2021-11-04 5:52 ` xiubli
2021-11-04 5:52 ` [PATCH v6 4/9] ceph: get file size from fscrypt_file when present in inode traces xiubli
` (6 subsequent siblings)
9 siblings, 0 replies; 17+ messages in thread
From: xiubli @ 2021-11-04 5:52 UTC (permalink / raw)
To: jlayton; +Cc: idryomov, vshankar, pdonnell, khiremat, ceph-devel
From: Jeff Layton <jlayton@kernel.org>
For encrypted inodes, transmit a rounded-up size to the MDS as the
normal file size and send the real inode size in fscrypt_file field.
Also, fix up creates and truncates to also transmit fscrypt_file.
Signed-off-by: Jeff Layton <jlayton@kernel.org>
---
fs/ceph/dir.c | 3 +++
fs/ceph/file.c | 2 ++
fs/ceph/inode.c | 18 ++++++++++++++++--
fs/ceph/mds_client.c | 9 ++++++++-
fs/ceph/mds_client.h | 2 ++
5 files changed, 31 insertions(+), 3 deletions(-)
diff --git a/fs/ceph/dir.c b/fs/ceph/dir.c
index 37c9c589ee27..987c1579614c 100644
--- a/fs/ceph/dir.c
+++ b/fs/ceph/dir.c
@@ -916,6 +916,9 @@ static int ceph_mknod(struct user_namespace *mnt_userns, struct inode *dir,
goto out_req;
}
+ if (S_ISREG(mode) && IS_ENCRYPTED(dir))
+ set_bit(CEPH_MDS_R_FSCRYPT_FILE, &req->r_req_flags);
+
req->r_dentry = dget(dentry);
req->r_num_caps = 2;
req->r_parent = dir;
diff --git a/fs/ceph/file.c b/fs/ceph/file.c
index 126d2d80686c..8c0b9ed7f48b 100644
--- a/fs/ceph/file.c
+++ b/fs/ceph/file.c
@@ -715,6 +715,8 @@ int ceph_atomic_open(struct inode *dir, struct dentry *dentry,
req->r_args.open.mask = cpu_to_le32(mask);
req->r_parent = dir;
ihold(dir);
+ if (IS_ENCRYPTED(dir))
+ set_bit(CEPH_MDS_R_FSCRYPT_FILE, &req->r_req_flags);
if (flags & O_CREAT) {
struct ceph_file_layout lo;
diff --git a/fs/ceph/inode.c b/fs/ceph/inode.c
index d24d42c94d43..4a7b2b0d88f7 100644
--- a/fs/ceph/inode.c
+++ b/fs/ceph/inode.c
@@ -2383,11 +2383,25 @@ int __ceph_setattr(struct inode *inode, struct iattr *attr, struct ceph_iattr *c
}
} else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
attr->ia_size != isize) {
- req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
- req->r_args.setattr.old_size = cpu_to_le64(isize);
mask |= CEPH_SETATTR_SIZE;
release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_EXCL |
CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
+ if (IS_ENCRYPTED(inode)) {
+ set_bit(CEPH_MDS_R_FSCRYPT_FILE, &req->r_req_flags);
+ mask |= CEPH_SETATTR_FSCRYPT_FILE;
+ req->r_args.setattr.size =
+ cpu_to_le64(round_up(attr->ia_size,
+ CEPH_FSCRYPT_BLOCK_SIZE));
+ req->r_args.setattr.old_size =
+ cpu_to_le64(round_up(isize,
+ CEPH_FSCRYPT_BLOCK_SIZE));
+ req->r_fscrypt_file = attr->ia_size;
+ /* FIXME: client must zero out any partial blocks! */
+ } else {
+ req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
+ req->r_args.setattr.old_size = cpu_to_le64(isize);
+ req->r_fscrypt_file = 0;
+ }
}
}
if (ia_valid & ATTR_MTIME) {
diff --git a/fs/ceph/mds_client.c b/fs/ceph/mds_client.c
index 69caea1d2444..e2d1b98c61fc 100644
--- a/fs/ceph/mds_client.c
+++ b/fs/ceph/mds_client.c
@@ -2653,7 +2653,12 @@ static void encode_mclientrequest_tail(void **p, const struct ceph_mds_request *
} else {
ceph_encode_32(p, 0);
}
- ceph_encode_32(p, 0); // fscrypt_file for now
+ if (test_bit(CEPH_MDS_R_FSCRYPT_FILE, &req->r_req_flags)) {
+ ceph_encode_32(p, sizeof(__le64));
+ ceph_encode_64(p, req->r_fscrypt_file);
+ } else {
+ ceph_encode_32(p, 0);
+ }
}
/*
@@ -2739,6 +2744,8 @@ static struct ceph_msg *create_request_message(struct ceph_mds_session *session,
/* fscrypt_file */
len += sizeof(u32);
+ if (test_bit(CEPH_MDS_R_FSCRYPT_FILE, &req->r_req_flags))
+ len += sizeof(__le64);
msg = ceph_msg_new2(CEPH_MSG_CLIENT_REQUEST, len, 1, GFP_NOFS, false);
if (!msg) {
diff --git a/fs/ceph/mds_client.h b/fs/ceph/mds_client.h
index 6a2ac489e06e..d64ff1bd2f5d 100644
--- a/fs/ceph/mds_client.h
+++ b/fs/ceph/mds_client.h
@@ -276,6 +276,7 @@ struct ceph_mds_request {
#define CEPH_MDS_R_DID_PREPOPULATE (6) /* prepopulated readdir */
#define CEPH_MDS_R_PARENT_LOCKED (7) /* is r_parent->i_rwsem wlocked? */
#define CEPH_MDS_R_ASYNC (8) /* async request */
+#define CEPH_MDS_R_FSCRYPT_FILE (9) /* must marshal fscrypt_file field */
unsigned long r_req_flags;
struct mutex r_fill_mutex;
@@ -283,6 +284,7 @@ struct ceph_mds_request {
union ceph_mds_request_args r_args;
struct ceph_fscrypt_auth *r_fscrypt_auth;
+ __le64 r_fscrypt_file;
u8 *r_altname; /* fscrypt binary crypttext for long filenames */
u32 r_altname_len; /* length of r_altname */
--
2.27.0
^ permalink raw reply related [flat|nested] 17+ messages in thread
* [PATCH v6 4/9] ceph: get file size from fscrypt_file when present in inode traces
2021-11-04 5:52 [PATCH v6 0/9] ceph: size handling for the fscrypt xiubli
` (2 preceding siblings ...)
2021-11-04 5:52 ` [PATCH v6 3/9] ceph: fscrypt_file field handling in MClientRequest messages xiubli
@ 2021-11-04 5:52 ` xiubli
2021-11-04 5:52 ` [PATCH v6 5/9] ceph: handle fscrypt fields in cap messages from MDS xiubli
` (5 subsequent siblings)
9 siblings, 0 replies; 17+ messages in thread
From: xiubli @ 2021-11-04 5:52 UTC (permalink / raw)
To: jlayton; +Cc: idryomov, vshankar, pdonnell, khiremat, ceph-devel
From: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Jeff Layton <jlayton@kernel.org>
---
fs/ceph/inode.c | 30 +++++++++++++++++++-----------
1 file changed, 19 insertions(+), 11 deletions(-)
diff --git a/fs/ceph/inode.c b/fs/ceph/inode.c
index 4a7b2b0d88f7..15c2fb1e2c8a 100644
--- a/fs/ceph/inode.c
+++ b/fs/ceph/inode.c
@@ -978,6 +978,16 @@ int ceph_fill_inode(struct inode *inode, struct page *locked_page,
from_kgid(&init_user_ns, inode->i_gid));
ceph_decode_timespec64(&ci->i_btime, &iinfo->btime);
ceph_decode_timespec64(&ci->i_snap_btime, &iinfo->snap_btime);
+
+#ifdef CONFIG_FS_ENCRYPTION
+ if (iinfo->fscrypt_auth_len && !ci->fscrypt_auth) {
+ ci->fscrypt_auth_len = iinfo->fscrypt_auth_len;
+ ci->fscrypt_auth = iinfo->fscrypt_auth;
+ iinfo->fscrypt_auth = NULL;
+ iinfo->fscrypt_auth_len = 0;
+ inode_set_flags(inode, S_ENCRYPTED, S_ENCRYPTED);
+ }
+#endif
}
if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) &&
@@ -1001,6 +1011,7 @@ int ceph_fill_inode(struct inode *inode, struct page *locked_page,
if (new_version ||
(new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) {
+ u64 size = info->size;
s64 old_pool = ci->i_layout.pool_id;
struct ceph_string *old_ns;
@@ -1014,10 +1025,17 @@ int ceph_fill_inode(struct inode *inode, struct page *locked_page,
pool_ns = old_ns;
+ if (IS_ENCRYPTED(inode) && size &&
+ (iinfo->fscrypt_file_len == sizeof(__le64))) {
+ size = __le64_to_cpu(*(__le64 *)iinfo->fscrypt_file);
+ if (info->size != round_up(size, CEPH_FSCRYPT_BLOCK_SIZE))
+ pr_warn("size=%llu fscrypt_file=%llu\n", info->size, size);
+ }
+
queue_trunc = ceph_fill_file_size(inode, issued,
le32_to_cpu(info->truncate_seq),
le64_to_cpu(info->truncate_size),
- le64_to_cpu(info->size));
+ le64_to_cpu(size));
/* only update max_size on auth cap */
if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
ci->i_max_size != le64_to_cpu(info->max_size)) {
@@ -1057,16 +1075,6 @@ int ceph_fill_inode(struct inode *inode, struct page *locked_page,
xattr_blob = NULL;
}
-#ifdef CONFIG_FS_ENCRYPTION
- if (iinfo->fscrypt_auth_len && !ci->fscrypt_auth) {
- ci->fscrypt_auth_len = iinfo->fscrypt_auth_len;
- ci->fscrypt_auth = iinfo->fscrypt_auth;
- iinfo->fscrypt_auth = NULL;
- iinfo->fscrypt_auth_len = 0;
- inode_set_flags(inode, S_ENCRYPTED, S_ENCRYPTED);
- }
-#endif
-
/* finally update i_version */
if (le64_to_cpu(info->version) > ci->i_version)
ci->i_version = le64_to_cpu(info->version);
--
2.27.0
^ permalink raw reply related [flat|nested] 17+ messages in thread
* [PATCH v6 5/9] ceph: handle fscrypt fields in cap messages from MDS
2021-11-04 5:52 [PATCH v6 0/9] ceph: size handling for the fscrypt xiubli
` (3 preceding siblings ...)
2021-11-04 5:52 ` [PATCH v6 4/9] ceph: get file size from fscrypt_file when present in inode traces xiubli
@ 2021-11-04 5:52 ` xiubli
2021-11-04 5:52 ` [PATCH v6 6/9] ceph: add __ceph_get_caps helper support xiubli
` (4 subsequent siblings)
9 siblings, 0 replies; 17+ messages in thread
From: xiubli @ 2021-11-04 5:52 UTC (permalink / raw)
To: jlayton; +Cc: idryomov, vshankar, pdonnell, khiremat, ceph-devel
From: Jeff Layton <jlayton@kernel.org>
Signed-off-by: Jeff Layton <jlayton@kernel.org>
---
fs/ceph/caps.c | 74 ++++++++++++++++++++++++++++++++++++++++++++++++--
1 file changed, 72 insertions(+), 2 deletions(-)
diff --git a/fs/ceph/caps.c b/fs/ceph/caps.c
index fc367f42536a..c9f1ac3ad2f3 100644
--- a/fs/ceph/caps.c
+++ b/fs/ceph/caps.c
@@ -3329,6 +3329,9 @@ struct cap_extra_info {
/* currently issued */
int issued;
struct timespec64 btime;
+ u8 *fscrypt_auth;
+ u32 fscrypt_auth_len;
+ u64 fscrypt_file_size;
};
/*
@@ -3361,6 +3364,14 @@ static void handle_cap_grant(struct inode *inode,
bool deleted_inode = false;
bool fill_inline = false;
+ /*
+ * If there is at least one crypto block then we'll trust fscrypt_file_size.
+ * If the real length of the file is 0, then ignore it (it has probably been
+ * truncated down to 0 by the MDS).
+ */
+ if (IS_ENCRYPTED(inode) && size)
+ size = extra_info->fscrypt_file_size;
+
dout("handle_cap_grant inode %p cap %p mds%d seq %d %s\n",
inode, cap, session->s_mds, seq, ceph_cap_string(newcaps));
dout(" size %llu max_size %llu, i_size %llu\n", size, max_size,
@@ -3839,7 +3850,8 @@ static void handle_cap_flushsnap_ack(struct inode *inode, u64 flush_tid,
*/
static bool handle_cap_trunc(struct inode *inode,
struct ceph_mds_caps *trunc,
- struct ceph_mds_session *session)
+ struct ceph_mds_session *session,
+ struct cap_extra_info *extra_info)
{
struct ceph_inode_info *ci = ceph_inode(inode);
int mds = session->s_mds;
@@ -3856,6 +3868,14 @@ static bool handle_cap_trunc(struct inode *inode,
issued |= implemented | dirty;
+ /*
+ * If there is at least one crypto block then we'll trust fscrypt_file_size.
+ * If the real length of the file is 0, then ignore it (it has probably been
+ * truncated down to 0 by the MDS).
+ */
+ if (IS_ENCRYPTED(inode) && size)
+ size = extra_info->fscrypt_file_size;
+
dout("handle_cap_trunc inode %p mds%d seq %d to %lld seq %d\n",
inode, mds, seq, truncate_size, truncate_seq);
queue_trunc = ceph_fill_file_size(inode, issued,
@@ -4074,6 +4094,48 @@ static void handle_cap_import(struct ceph_mds_client *mdsc,
*target_cap = cap;
}
+#ifdef CONFIG_FS_ENCRYPTION
+static int parse_fscrypt_fields(void **p, void *end, struct cap_extra_info *extra)
+{
+ u32 len;
+
+ ceph_decode_32_safe(p, end, extra->fscrypt_auth_len, bad);
+ if (extra->fscrypt_auth_len) {
+ ceph_decode_need(p, end, extra->fscrypt_auth_len, bad);
+ extra->fscrypt_auth = kmalloc(extra->fscrypt_auth_len, GFP_KERNEL);
+ if (!extra->fscrypt_auth)
+ return -ENOMEM;
+ ceph_decode_copy_safe(p, end, extra->fscrypt_auth,
+ extra->fscrypt_auth_len, bad);
+ }
+
+ ceph_decode_32_safe(p, end, len, bad);
+ if (len == sizeof(u64))
+ ceph_decode_64_safe(p, end, extra->fscrypt_file_size, bad);
+ else
+ ceph_decode_skip_n(p, end, len, bad);
+ return 0;
+bad:
+ return -EIO;
+}
+#else
+static int parse_fscrypt_fields(void **p, void *end, struct cap_extra_info *extra)
+{
+ u32 len;
+
+ /* Don't care about these fields unless we're encryption-capable */
+ ceph_decode_32_safe(p, end, len, bad);
+ if (len)
+ ceph_decode_skip_n(p, end, len, bad);
+ ceph_decode_32_safe(p, end, len, bad);
+ if (len)
+ ceph_decode_skip_n(p, end, len, bad);
+ return 0;
+bad:
+ return -EIO;
+}
+#endif
+
/*
* Handle a caps message from the MDS.
*
@@ -4192,6 +4254,12 @@ void ceph_handle_caps(struct ceph_mds_session *session,
ceph_decode_64_safe(&p, end, extra_info.nsubdirs, bad);
}
+ if (msg_version >= 12) {
+ int ret = parse_fscrypt_fields(&p, end, &extra_info);
+ if (ret)
+ goto bad;
+ }
+
/* lookup ino */
inode = ceph_find_inode(mdsc->fsc->sb, vino);
ci = ceph_inode(inode);
@@ -4288,7 +4356,8 @@ void ceph_handle_caps(struct ceph_mds_session *session,
break;
case CEPH_CAP_OP_TRUNC:
- queue_trunc = handle_cap_trunc(inode, h, session);
+ queue_trunc = handle_cap_trunc(inode, h, session,
+ &extra_info);
spin_unlock(&ci->i_ceph_lock);
if (queue_trunc)
ceph_queue_vmtruncate(inode);
@@ -4306,6 +4375,7 @@ void ceph_handle_caps(struct ceph_mds_session *session,
iput(inode);
out:
ceph_put_string(extra_info.pool_ns);
+ kfree(extra_info.fscrypt_auth);
return;
flush_cap_releases:
--
2.27.0
^ permalink raw reply related [flat|nested] 17+ messages in thread
* [PATCH v6 6/9] ceph: add __ceph_get_caps helper support
2021-11-04 5:52 [PATCH v6 0/9] ceph: size handling for the fscrypt xiubli
` (4 preceding siblings ...)
2021-11-04 5:52 ` [PATCH v6 5/9] ceph: handle fscrypt fields in cap messages from MDS xiubli
@ 2021-11-04 5:52 ` xiubli
2021-11-04 5:52 ` [PATCH v6 7/9] ceph: add __ceph_sync_read " xiubli
` (3 subsequent siblings)
9 siblings, 0 replies; 17+ messages in thread
From: xiubli @ 2021-11-04 5:52 UTC (permalink / raw)
To: jlayton; +Cc: idryomov, vshankar, pdonnell, khiremat, ceph-devel, Xiubo Li
From: Xiubo Li <xiubli@redhat.com>
Signed-off-by: Xiubo Li <xiubli@redhat.com>
---
fs/ceph/caps.c | 19 +++++++++++++------
fs/ceph/super.h | 2 ++
2 files changed, 15 insertions(+), 6 deletions(-)
diff --git a/fs/ceph/caps.c b/fs/ceph/caps.c
index c9f1ac3ad2f3..c15c5dd36747 100644
--- a/fs/ceph/caps.c
+++ b/fs/ceph/caps.c
@@ -2911,10 +2911,9 @@ int ceph_try_get_caps(struct inode *inode, int need, int want,
* due to a small max_size, make sure we check_max_size (and possibly
* ask the mds) so we don't get hung up indefinitely.
*/
-int ceph_get_caps(struct file *filp, int need, int want, loff_t endoff, int *got)
+int __ceph_get_caps(struct inode *inode, struct ceph_file_info *fi, int need,
+ int want, loff_t endoff, int *got)
{
- struct ceph_file_info *fi = filp->private_data;
- struct inode *inode = file_inode(filp);
struct ceph_inode_info *ci = ceph_inode(inode);
struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
int ret, _got, flags;
@@ -2923,7 +2922,7 @@ int ceph_get_caps(struct file *filp, int need, int want, loff_t endoff, int *got
if (ret < 0)
return ret;
- if ((fi->fmode & CEPH_FILE_MODE_WR) &&
+ if (fi && (fi->fmode & CEPH_FILE_MODE_WR) &&
fi->filp_gen != READ_ONCE(fsc->filp_gen))
return -EBADF;
@@ -2931,7 +2930,7 @@ int ceph_get_caps(struct file *filp, int need, int want, loff_t endoff, int *got
while (true) {
flags &= CEPH_FILE_MODE_MASK;
- if (atomic_read(&fi->num_locks))
+ if (fi && atomic_read(&fi->num_locks))
flags |= CHECK_FILELOCK;
_got = 0;
ret = try_get_cap_refs(inode, need, want, endoff,
@@ -2976,7 +2975,7 @@ int ceph_get_caps(struct file *filp, int need, int want, loff_t endoff, int *got
continue;
}
- if ((fi->fmode & CEPH_FILE_MODE_WR) &&
+ if (fi && (fi->fmode & CEPH_FILE_MODE_WR) &&
fi->filp_gen != READ_ONCE(fsc->filp_gen)) {
if (ret >= 0 && _got)
ceph_put_cap_refs(ci, _got);
@@ -3039,6 +3038,14 @@ int ceph_get_caps(struct file *filp, int need, int want, loff_t endoff, int *got
return 0;
}
+int ceph_get_caps(struct file *filp, int need, int want, loff_t endoff, int *got)
+{
+ struct ceph_file_info *fi = filp->private_data;
+ struct inode *inode = file_inode(filp);
+
+ return __ceph_get_caps(inode, fi, need, want, endoff, got);
+}
+
/*
* Take cap refs. Caller must already know we hold at least one ref
* on the caps in question or we don't know this is safe.
diff --git a/fs/ceph/super.h b/fs/ceph/super.h
index ea95c958202f..403918a4cdb3 100644
--- a/fs/ceph/super.h
+++ b/fs/ceph/super.h
@@ -1225,6 +1225,8 @@ extern int ceph_encode_dentry_release(void **p, struct dentry *dn,
struct inode *dir,
int mds, int drop, int unless);
+extern int __ceph_get_caps(struct inode *inode, struct ceph_file_info *fi,
+ int need, int want, loff_t endoff, int *got);
extern int ceph_get_caps(struct file *filp, int need, int want,
loff_t endoff, int *got);
extern int ceph_try_get_caps(struct inode *inode,
--
2.27.0
^ permalink raw reply related [flat|nested] 17+ messages in thread
* [PATCH v6 7/9] ceph: add __ceph_sync_read helper support
2021-11-04 5:52 [PATCH v6 0/9] ceph: size handling for the fscrypt xiubli
` (5 preceding siblings ...)
2021-11-04 5:52 ` [PATCH v6 6/9] ceph: add __ceph_get_caps helper support xiubli
@ 2021-11-04 5:52 ` xiubli
2021-11-04 5:52 ` [PATCH v6 8/9] ceph: add object version support for sync read xiubli
` (2 subsequent siblings)
9 siblings, 0 replies; 17+ messages in thread
From: xiubli @ 2021-11-04 5:52 UTC (permalink / raw)
To: jlayton; +Cc: idryomov, vshankar, pdonnell, khiremat, ceph-devel, Xiubo Li
From: Xiubo Li <xiubli@redhat.com>
Signed-off-by: Xiubo Li <xiubli@redhat.com>
---
fs/ceph/file.c | 34 ++++++++++++++++++++++------------
fs/ceph/super.h | 2 ++
2 files changed, 24 insertions(+), 12 deletions(-)
diff --git a/fs/ceph/file.c b/fs/ceph/file.c
index 8c0b9ed7f48b..129f6a642f8e 100644
--- a/fs/ceph/file.c
+++ b/fs/ceph/file.c
@@ -870,21 +870,18 @@ enum {
* If we get a short result from the OSD, check against i_size; we need to
* only return a short read to the caller if we hit EOF.
*/
-static ssize_t ceph_sync_read(struct kiocb *iocb, struct iov_iter *to,
- int *retry_op)
+ssize_t __ceph_sync_read(struct inode *inode, loff_t *ki_pos,
+ struct iov_iter *to, int *retry_op)
{
- struct file *file = iocb->ki_filp;
- struct inode *inode = file_inode(file);
struct ceph_inode_info *ci = ceph_inode(inode);
struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
struct ceph_osd_client *osdc = &fsc->client->osdc;
ssize_t ret;
- u64 off = iocb->ki_pos;
+ u64 off = *ki_pos;
u64 len = iov_iter_count(to);
u64 i_size;
- dout("sync_read on file %p %llu~%u %s\n", file, off, (unsigned)len,
- (file->f_flags & O_DIRECT) ? "O_DIRECT" : "");
+ dout("sync_read on inode %p %llu~%u\n", inode, *ki_pos, (unsigned)len);
if (!len)
return 0;
@@ -986,14 +983,14 @@ static ssize_t ceph_sync_read(struct kiocb *iocb, struct iov_iter *to,
break;
}
- if (off > iocb->ki_pos) {
+ if (off > *ki_pos) {
if (off >= i_size) {
*retry_op = CHECK_EOF;
- ret = i_size - iocb->ki_pos;
- iocb->ki_pos = i_size;
+ ret = i_size - *ki_pos;
+ *ki_pos = i_size;
} else {
- ret = off - iocb->ki_pos;
- iocb->ki_pos = off;
+ ret = off - *ki_pos;
+ *ki_pos = off;
}
}
@@ -1001,6 +998,19 @@ static ssize_t ceph_sync_read(struct kiocb *iocb, struct iov_iter *to,
return ret;
}
+static ssize_t ceph_sync_read(struct kiocb *iocb, struct iov_iter *to,
+ int *retry_op)
+{
+ struct file *file = iocb->ki_filp;
+ struct inode *inode = file_inode(file);
+
+ dout("sync_read on file %p %llu~%u %s\n", file, iocb->ki_pos,
+ (unsigned)iov_iter_count(to),
+ (file->f_flags & O_DIRECT) ? "O_DIRECT" : "");
+
+ return __ceph_sync_read(inode, &iocb->ki_pos, to, retry_op);
+}
+
struct ceph_aio_request {
struct kiocb *iocb;
size_t total_len;
diff --git a/fs/ceph/super.h b/fs/ceph/super.h
index 403918a4cdb3..2362d758af97 100644
--- a/fs/ceph/super.h
+++ b/fs/ceph/super.h
@@ -1253,6 +1253,8 @@ extern int ceph_renew_caps(struct inode *inode, int fmode);
extern int ceph_open(struct inode *inode, struct file *file);
extern int ceph_atomic_open(struct inode *dir, struct dentry *dentry,
struct file *file, unsigned flags, umode_t mode);
+extern ssize_t __ceph_sync_read(struct inode *inode, loff_t *ki_pos,
+ struct iov_iter *to, int *retry_op);
extern int ceph_release(struct inode *inode, struct file *filp);
extern void ceph_fill_inline_data(struct inode *inode, struct page *locked_page,
char *data, size_t len);
--
2.27.0
^ permalink raw reply related [flat|nested] 17+ messages in thread
* [PATCH v6 8/9] ceph: add object version support for sync read
2021-11-04 5:52 [PATCH v6 0/9] ceph: size handling for the fscrypt xiubli
` (6 preceding siblings ...)
2021-11-04 5:52 ` [PATCH v6 7/9] ceph: add __ceph_sync_read " xiubli
@ 2021-11-04 5:52 ` xiubli
2021-12-07 17:31 ` Jeff Layton
2021-11-04 5:52 ` [PATCH v6 9/9] ceph: add truncate size handling support for fscrypt xiubli
2021-11-04 6:00 ` [PATCH v6 0/9] ceph: size handling for the fscrypt Xiubo Li
9 siblings, 1 reply; 17+ messages in thread
From: xiubli @ 2021-11-04 5:52 UTC (permalink / raw)
To: jlayton; +Cc: idryomov, vshankar, pdonnell, khiremat, ceph-devel, Xiubo Li
From: Xiubo Li <xiubli@redhat.com>
The sync read may split the read into several osdc requests, so
for each it may in different Rados objects.
Signed-off-by: Xiubo Li <xiubli@redhat.com>
---
fs/ceph/file.c | 44 ++++++++++++++++++++++++++++++++++++++++++--
fs/ceph/super.h | 18 +++++++++++++++++-
2 files changed, 59 insertions(+), 3 deletions(-)
diff --git a/fs/ceph/file.c b/fs/ceph/file.c
index 129f6a642f8e..cedd86a6058d 100644
--- a/fs/ceph/file.c
+++ b/fs/ceph/file.c
@@ -871,7 +871,8 @@ enum {
* only return a short read to the caller if we hit EOF.
*/
ssize_t __ceph_sync_read(struct inode *inode, loff_t *ki_pos,
- struct iov_iter *to, int *retry_op)
+ struct iov_iter *to, int *retry_op,
+ struct ceph_object_vers *objvers)
{
struct ceph_inode_info *ci = ceph_inode(inode);
struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
@@ -880,6 +881,7 @@ ssize_t __ceph_sync_read(struct inode *inode, loff_t *ki_pos,
u64 off = *ki_pos;
u64 len = iov_iter_count(to);
u64 i_size;
+ u32 object_count = 8;
dout("sync_read on inode %p %llu~%u\n", inode, *ki_pos, (unsigned)len);
@@ -896,6 +898,15 @@ ssize_t __ceph_sync_read(struct inode *inode, loff_t *ki_pos,
if (ret < 0)
return ret;
+ if (objvers) {
+ objvers->count = 0;
+ objvers->objvers = kcalloc(object_count,
+ sizeof(struct ceph_object_ver),
+ GFP_KERNEL);
+ if (!objvers->objvers)
+ return -ENOMEM;
+ }
+
ret = 0;
while ((len = iov_iter_count(to)) > 0) {
struct ceph_osd_request *req;
@@ -938,6 +949,30 @@ ssize_t __ceph_sync_read(struct inode *inode, loff_t *ki_pos,
req->r_end_latency,
len, ret);
+ if (objvers) {
+ u32 ind = objvers->count;
+
+ if (objvers->count >= object_count) {
+ int ov_size;
+
+ object_count *= 2;
+ ov_size = sizeof(struct ceph_object_ver);
+ objvers->objvers = krealloc_array(objvers,
+ object_count,
+ ov_size,
+ GFP_KERNEL);
+ if (!objvers->objvers) {
+ objvers->count = 0;
+ ret = -ENOMEM;
+ break;
+ }
+ }
+
+ objvers->objvers[ind].offset = off;
+ objvers->objvers[ind].length = len;
+ objvers->objvers[ind].objver = req->r_version;
+ objvers->count++;
+ }
ceph_osdc_put_request(req);
i_size = i_size_read(inode);
@@ -995,6 +1030,11 @@ ssize_t __ceph_sync_read(struct inode *inode, loff_t *ki_pos,
}
dout("sync_read result %zd retry_op %d\n", ret, *retry_op);
+ if (ret < 0 && objvers) {
+ objvers->count = 0;
+ kfree(objvers->objvers);
+ objvers->objvers = NULL;
+ }
return ret;
}
@@ -1008,7 +1048,7 @@ static ssize_t ceph_sync_read(struct kiocb *iocb, struct iov_iter *to,
(unsigned)iov_iter_count(to),
(file->f_flags & O_DIRECT) ? "O_DIRECT" : "");
- return __ceph_sync_read(inode, &iocb->ki_pos, to, retry_op);
+ return __ceph_sync_read(inode, &iocb->ki_pos, to, retry_op, NULL);
}
struct ceph_aio_request {
diff --git a/fs/ceph/super.h b/fs/ceph/super.h
index 2362d758af97..b347b12e86a9 100644
--- a/fs/ceph/super.h
+++ b/fs/ceph/super.h
@@ -451,6 +451,21 @@ struct ceph_inode_info {
struct inode vfs_inode; /* at end */
};
+/*
+ * The version of an object which contains the
+ * file range of [offset, offset + length).
+ */
+struct ceph_object_ver {
+ u64 offset;
+ u64 length;
+ u64 objver;
+};
+
+struct ceph_object_vers {
+ u32 count;
+ struct ceph_object_ver *objvers;
+};
+
static inline struct ceph_inode_info *
ceph_inode(const struct inode *inode)
{
@@ -1254,7 +1269,8 @@ extern int ceph_open(struct inode *inode, struct file *file);
extern int ceph_atomic_open(struct inode *dir, struct dentry *dentry,
struct file *file, unsigned flags, umode_t mode);
extern ssize_t __ceph_sync_read(struct inode *inode, loff_t *ki_pos,
- struct iov_iter *to, int *retry_op);
+ struct iov_iter *to, int *retry_op,
+ struct ceph_object_vers *objvers);
extern int ceph_release(struct inode *inode, struct file *filp);
extern void ceph_fill_inline_data(struct inode *inode, struct page *locked_page,
char *data, size_t len);
--
2.27.0
^ permalink raw reply related [flat|nested] 17+ messages in thread
* [PATCH v6 9/9] ceph: add truncate size handling support for fscrypt
2021-11-04 5:52 [PATCH v6 0/9] ceph: size handling for the fscrypt xiubli
` (7 preceding siblings ...)
2021-11-04 5:52 ` [PATCH v6 8/9] ceph: add object version support for sync read xiubli
@ 2021-11-04 5:52 ` xiubli
2021-11-05 13:36 ` Jeff Layton
2021-11-04 6:00 ` [PATCH v6 0/9] ceph: size handling for the fscrypt Xiubo Li
9 siblings, 1 reply; 17+ messages in thread
From: xiubli @ 2021-11-04 5:52 UTC (permalink / raw)
To: jlayton; +Cc: idryomov, vshankar, pdonnell, khiremat, ceph-devel, Xiubo Li
From: Xiubo Li <xiubli@redhat.com>
This will transfer the encrypted last block contents to the MDS
along with the truncate request only when the new size is smaller
and not aligned to the fscrypt BLOCK size. When the last block is
located in the file hole, the truncate request will only contain
the header.
The MDS could fail to do the truncate if there has another client
or process has already updated the Rados object which contains
the last block, and will return -EAGAIN, then the kclient needs
to retry it. The RMW will take around 50ms, and will let it retry
20 times for now.
Signed-off-by: Xiubo Li <xiubli@redhat.com>
---
fs/ceph/inode.c | 205 ++++++++++++++++++++++++++++++++++--
fs/ceph/super.h | 5 +
include/linux/ceph/crypto.h | 28 +++++
3 files changed, 227 insertions(+), 11 deletions(-)
create mode 100644 include/linux/ceph/crypto.h
diff --git a/fs/ceph/inode.c b/fs/ceph/inode.c
index 15c2fb1e2c8a..5817685ea9a5 100644
--- a/fs/ceph/inode.c
+++ b/fs/ceph/inode.c
@@ -21,6 +21,7 @@
#include "cache.h"
#include "crypto.h"
#include <linux/ceph/decode.h>
+#include <linux/ceph/crypto.h>
/*
* Ceph inode operations
@@ -586,6 +587,7 @@ struct inode *ceph_alloc_inode(struct super_block *sb)
ci->i_truncate_seq = 0;
ci->i_truncate_size = 0;
ci->i_truncate_pending = 0;
+ ci->i_truncate_pagecache_size = 0;
ci->i_max_size = 0;
ci->i_reported_size = 0;
@@ -751,6 +753,10 @@ int ceph_fill_file_size(struct inode *inode, int issued,
dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
truncate_size);
ci->i_truncate_size = truncate_size;
+ if (IS_ENCRYPTED(inode))
+ ci->i_truncate_pagecache_size = size;
+ else
+ ci->i_truncate_pagecache_size = truncate_size;
}
if (queue_trunc)
@@ -1026,10 +1032,14 @@ int ceph_fill_inode(struct inode *inode, struct page *locked_page,
pool_ns = old_ns;
if (IS_ENCRYPTED(inode) && size &&
- (iinfo->fscrypt_file_len == sizeof(__le64))) {
- size = __le64_to_cpu(*(__le64 *)iinfo->fscrypt_file);
- if (info->size != round_up(size, CEPH_FSCRYPT_BLOCK_SIZE))
- pr_warn("size=%llu fscrypt_file=%llu\n", info->size, size);
+ (iinfo->fscrypt_file_len >= sizeof(__le64))) {
+ u64 fsize = __le64_to_cpu(*(__le64 *)iinfo->fscrypt_file);
+ if (fsize) {
+ size = fsize;
+ if (info->size != round_up(size, CEPH_FSCRYPT_BLOCK_SIZE))
+ pr_warn("size=%llu fscrypt_file=%llu\n",
+ info->size, size);
+ }
}
queue_trunc = ceph_fill_file_size(inode, issued,
@@ -2142,7 +2152,7 @@ void __ceph_do_pending_vmtruncate(struct inode *inode)
/* there should be no reader or writer */
WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
- to = ci->i_truncate_size;
+ to = ci->i_truncate_pagecache_size;
wrbuffer_refs = ci->i_wrbuffer_ref;
dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
ci->i_truncate_pending, to);
@@ -2151,7 +2161,7 @@ void __ceph_do_pending_vmtruncate(struct inode *inode)
truncate_pagecache(inode, to);
spin_lock(&ci->i_ceph_lock);
- if (to == ci->i_truncate_size) {
+ if (to == ci->i_truncate_pagecache_size) {
ci->i_truncate_pending = 0;
finish = 1;
}
@@ -2232,6 +2242,141 @@ static const struct inode_operations ceph_encrypted_symlink_iops = {
.listxattr = ceph_listxattr,
};
+/*
+ * Transfer the encrypted last block to the MDS and the MDS
+ * will help update it when truncating a smaller size.
+ *
+ * We don't support a PAGE_SIZE that is smaller than the
+ * CEPH_FSCRYPT_BLOCK_SIZE.
+ */
+static int fill_fscrypt_truncate(struct inode *inode,
+ struct ceph_mds_request *req,
+ struct iattr *attr)
+{
+ struct ceph_inode_info *ci = ceph_inode(inode);
+ int boff = attr->ia_size % CEPH_FSCRYPT_BLOCK_SIZE;
+ loff_t pos, orig_pos = round_down(attr->ia_size, CEPH_FSCRYPT_BLOCK_SIZE);
+#if 0
+ u64 block = orig_pos >> CEPH_FSCRYPT_BLOCK_SHIFT;
+#endif
+ struct ceph_pagelist *pagelist = NULL;
+ struct kvec iov;
+ struct iov_iter iter;
+ struct page *page = NULL;
+ struct ceph_fscrypt_truncate_size_header header;
+ int retry_op = 0;
+ int len = CEPH_FSCRYPT_BLOCK_SIZE;
+ loff_t i_size = i_size_read(inode);
+ struct ceph_object_vers objvers = {0, NULL};
+ int got, ret, issued;
+
+ ret = __ceph_get_caps(inode, NULL, CEPH_CAP_FILE_RD, 0, -1, &got);
+ if (ret < 0)
+ return ret;
+
+ issued = __ceph_caps_issued(ci, NULL);
+
+ dout("%s size %lld -> %lld got cap refs on %s, issued %s\n", __func__,
+ i_size, attr->ia_size, ceph_cap_string(got),
+ ceph_cap_string(issued));
+
+ /* Try to writeback the dirty pagecaches */
+ if (issued & (CEPH_CAP_FILE_BUFFER))
+ filemap_fdatawrite(&inode->i_data);
+
+ page = __page_cache_alloc(GFP_KERNEL);
+ if (page == NULL) {
+ ret = -ENOMEM;
+ goto out;
+ }
+
+ pagelist = ceph_pagelist_alloc(GFP_KERNEL);
+ if (!pagelist) {
+ ret = -ENOMEM;
+ goto out;
+ }
+
+ iov.iov_base = kmap_local_page(page);
+ iov.iov_len = len;
+ iov_iter_kvec(&iter, READ, &iov, 1, len);
+
+ pos = orig_pos;
+ ret = __ceph_sync_read(inode, &pos, &iter, &retry_op, &objvers);
+ ceph_put_cap_refs(ci, got);
+ if (ret < 0)
+ goto out;
+
+ WARN_ON_ONCE(objvers.count != 1);
+
+ /* Insert the header first */
+ header.ver = 1;
+ header.compat = 1;
+
+ /*
+ * If we hit a hole here, we should just skip filling
+ * the fscrypt for the request, because once the fscrypt
+ * is enabled, the file will be split into many blocks
+ * with the size of CEPH_FSCRYPT_BLOCK_SIZE, if there
+ * has a hole, the hole size should be multiple of block
+ * size.
+ *
+ * If the Rados object doesn't exist, it will be set 0.
+ */
+ if (!objvers.objvers[0].objver) {
+ dout("%s hit hole, ppos %lld < size %lld\n", __func__,
+ pos, i_size);
+
+ header.data_len = cpu_to_le32(8 + 8 + 4);
+ header.assert_ver = cpu_to_le64(0);
+ header.file_offset = cpu_to_le64(0);
+ header.block_size = cpu_to_le64(0);
+ ret = 0;
+ } else {
+ header.data_len = cpu_to_le32(8 + 8 + 4 + CEPH_FSCRYPT_BLOCK_SIZE);
+ header.assert_ver = objvers.objvers[0].objver;
+ header.file_offset = cpu_to_le64(orig_pos);
+ header.block_size = cpu_to_le64(CEPH_FSCRYPT_BLOCK_SIZE);
+
+ /* truncate and zero out the extra contents for the last block */
+ memset(iov.iov_base + boff, 0, PAGE_SIZE - boff);
+
+#if 0 // Uncomment this when the fscrypt is enabled globally in kceph
+
+ /* encrypt the last block */
+ ret = fscrypt_encrypt_block_inplace(inode, page,
+ CEPH_FSCRYPT_BLOCK_SIZE,
+ 0, block,
+ GFP_KERNEL);
+ if (ret)
+ goto out;
+#endif
+ }
+
+ /* Insert the header */
+ ret = ceph_pagelist_append(pagelist, &header, sizeof(header));
+ if (ret)
+ goto out;
+
+ if (header.block_size) {
+ /* Append the last block contents to pagelist */
+ ret = ceph_pagelist_append(pagelist, iov.iov_base,
+ CEPH_FSCRYPT_BLOCK_SIZE);
+ if (ret)
+ goto out;
+ }
+ req->r_pagelist = pagelist;
+out:
+ dout("%s %p size dropping cap refs on %s\n", __func__,
+ inode, ceph_cap_string(got));
+ kunmap_local(iov.iov_base);
+ if (page)
+ __free_pages(page, 0);
+ if (ret && pagelist)
+ ceph_pagelist_release(pagelist);
+ kfree(objvers.objvers);
+ return ret;
+}
+
int __ceph_setattr(struct inode *inode, struct iattr *attr, struct ceph_iattr *cia)
{
struct ceph_inode_info *ci = ceph_inode(inode);
@@ -2239,12 +2384,15 @@ int __ceph_setattr(struct inode *inode, struct iattr *attr, struct ceph_iattr *c
struct ceph_mds_request *req;
struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
struct ceph_cap_flush *prealloc_cf;
+ loff_t isize = i_size_read(inode);
int issued;
int release = 0, dirtied = 0;
int mask = 0;
int err = 0;
int inode_dirty_flags = 0;
bool lock_snap_rwsem = false;
+ bool fill_fscrypt;
+ int truncate_retry = 20; /* The RMW will take around 50ms */
prealloc_cf = ceph_alloc_cap_flush();
if (!prealloc_cf)
@@ -2257,6 +2405,8 @@ int __ceph_setattr(struct inode *inode, struct iattr *attr, struct ceph_iattr *c
return PTR_ERR(req);
}
+retry:
+ fill_fscrypt = false;
spin_lock(&ci->i_ceph_lock);
issued = __ceph_caps_issued(ci, NULL);
@@ -2378,10 +2528,27 @@ int __ceph_setattr(struct inode *inode, struct iattr *attr, struct ceph_iattr *c
}
}
if (ia_valid & ATTR_SIZE) {
- loff_t isize = i_size_read(inode);
-
dout("setattr %p size %lld -> %lld\n", inode, isize, attr->ia_size);
- if ((issued & CEPH_CAP_FILE_EXCL) && attr->ia_size >= isize) {
+ /*
+ * Only when the new size is smaller and not aligned to
+ * CEPH_FSCRYPT_BLOCK_SIZE will the RMW is needed.
+ */
+ if (IS_ENCRYPTED(inode) && attr->ia_size < isize &&
+ (attr->ia_size % CEPH_FSCRYPT_BLOCK_SIZE)) {
+ mask |= CEPH_SETATTR_SIZE;
+ release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_EXCL |
+ CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
+ set_bit(CEPH_MDS_R_FSCRYPT_FILE, &req->r_req_flags);
+ mask |= CEPH_SETATTR_FSCRYPT_FILE;
+ req->r_args.setattr.size =
+ cpu_to_le64(round_up(attr->ia_size,
+ CEPH_FSCRYPT_BLOCK_SIZE));
+ req->r_args.setattr.old_size =
+ cpu_to_le64(round_up(isize,
+ CEPH_FSCRYPT_BLOCK_SIZE));
+ req->r_fscrypt_file = attr->ia_size;
+ fill_fscrypt = true;
+ } else if ((issued & CEPH_CAP_FILE_EXCL) && attr->ia_size >= isize) {
if (attr->ia_size > isize) {
i_size_write(inode, attr->ia_size);
inode->i_blocks = calc_inode_blocks(attr->ia_size);
@@ -2404,7 +2571,6 @@ int __ceph_setattr(struct inode *inode, struct iattr *attr, struct ceph_iattr *c
cpu_to_le64(round_up(isize,
CEPH_FSCRYPT_BLOCK_SIZE));
req->r_fscrypt_file = attr->ia_size;
- /* FIXME: client must zero out any partial blocks! */
} else {
req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
req->r_args.setattr.old_size = cpu_to_le64(isize);
@@ -2476,7 +2642,6 @@ int __ceph_setattr(struct inode *inode, struct iattr *attr, struct ceph_iattr *c
if (inode_dirty_flags)
__mark_inode_dirty(inode, inode_dirty_flags);
-
if (mask) {
req->r_inode = inode;
ihold(inode);
@@ -2484,7 +2649,25 @@ int __ceph_setattr(struct inode *inode, struct iattr *attr, struct ceph_iattr *c
req->r_args.setattr.mask = cpu_to_le32(mask);
req->r_num_caps = 1;
req->r_stamp = attr->ia_ctime;
+ if (fill_fscrypt) {
+ err = fill_fscrypt_truncate(inode, req, attr);
+ if (err)
+ goto out;
+ }
+
+ /*
+ * The truncate request will return -EAGAIN when the
+ * last block has been updated just before the MDS
+ * successfully gets the xlock for the FILE lock. To
+ * avoid corrupting the file contents we need to retry
+ * it.
+ */
err = ceph_mdsc_do_request(mdsc, NULL, req);
+ if (err == -EAGAIN && truncate_retry--) {
+ dout("setattr %p result=%d (%s locally, %d remote), retry it!\n",
+ inode, err, ceph_cap_string(dirtied), mask);
+ goto retry;
+ }
}
out:
dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
diff --git a/fs/ceph/super.h b/fs/ceph/super.h
index b347b12e86a9..071857bb59d8 100644
--- a/fs/ceph/super.h
+++ b/fs/ceph/super.h
@@ -408,6 +408,11 @@ struct ceph_inode_info {
u32 i_truncate_seq; /* last truncate to smaller size */
u64 i_truncate_size; /* and the size we last truncated down to */
int i_truncate_pending; /* still need to call vmtruncate */
+ /*
+ * For none fscrypt case it equals to i_truncate_size or it will
+ * equals to fscrypt_file_size
+ */
+ u64 i_truncate_pagecache_size;
u64 i_max_size; /* max file size authorized by mds */
u64 i_reported_size; /* (max_)size reported to or requested of mds */
diff --git a/include/linux/ceph/crypto.h b/include/linux/ceph/crypto.h
new file mode 100644
index 000000000000..2b0961902887
--- /dev/null
+++ b/include/linux/ceph/crypto.h
@@ -0,0 +1,28 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef _FS_CEPH_CRYPTO_H
+#define _FS_CEPH_CRYPTO_H
+
+#include <linux/types.h>
+
+/*
+ * Header for the crypted file when truncating the size, this
+ * will be sent to MDS, and the MDS will update the encrypted
+ * last block and then truncate the size.
+ */
+struct ceph_fscrypt_truncate_size_header {
+ __u8 ver;
+ __u8 compat;
+
+ /*
+ * It will be sizeof(assert_ver + file_offset + block_size)
+ * if the last block is empty when it's located in a file
+ * hole. Or the data_len will plus CEPH_FSCRYPT_BLOCK_SIZE.
+ */
+ __le32 data_len;
+
+ __le64 assert_ver;
+ __le64 file_offset;
+ __le32 block_size;
+} __packed;
+
+#endif
--
2.27.0
^ permalink raw reply related [flat|nested] 17+ messages in thread
* Re: [PATCH v6 0/9] ceph: size handling for the fscrypt
2021-11-04 5:52 [PATCH v6 0/9] ceph: size handling for the fscrypt xiubli
` (8 preceding siblings ...)
2021-11-04 5:52 ` [PATCH v6 9/9] ceph: add truncate size handling support for fscrypt xiubli
@ 2021-11-04 6:00 ` Xiubo Li
2021-11-04 10:20 ` Jeff Layton
9 siblings, 1 reply; 17+ messages in thread
From: Xiubo Li @ 2021-11-04 6:00 UTC (permalink / raw)
To: jlayton; +Cc: idryomov, vshankar, pdonnell, khiremat, ceph-devel
The xfstests generic/014 test passed:
[root@ceph1 xfstests]# pwd
/mnt/kcephfs/xfstests
[root@ceph1 xfstests]# cat ./local.config
export FSTYP=ceph
export TEST_DEV=10.72.49.127:40084:/test
export TEST_DIR=/mnt/kcephfs/test/_brpcfnn
export TEST_FS_MOUNT_OPTS="-o
test_dummy_encryption,name=admin,secret=AQDS3IFhEtxvORAAxn1d4FVN2bRUsc/TZMpQvQ=="
[root@ceph1 xfstests]# ./check generic/014
FSTYP -- ceph
PLATFORM -- Linux/x86_64 ceph1 5.15.0-rc6+
generic/014 533s ... 558s
Ran: generic/014
Passed all 1 tests
On 11/4/21 1:52 PM, xiubli@redhat.com wrote:
> From: Xiubo Li <xiubli@redhat.com>
>
> This patch series is based on the "wip-fscrypt-fnames" branch in
> repo https://github.com/ceph/ceph-client.git.
>
> And I have picked up 5 patches from the "ceph-fscrypt-size-experimental"
> branch in repo
> https://git.kernel.org/pub/scm/linux/kernel/git/jlayton/linux.git.
>
> ====
>
> This approach is based on the discussion from V1 and V2, which will
> pass the encrypted last block contents to MDS along with the truncate
> request.
>
> This will send the encrypted last block contents to MDS along with
> the truncate request when truncating to a smaller size and at the
> same time new size does not align to BLOCK SIZE.
>
> The MDS side patch is raised in PR
> https://github.com/ceph/ceph/pull/43588, which is also based Jeff's
> previous great work in PR https://github.com/ceph/ceph/pull/41284.
>
> The MDS will use the filer.write_trunc(), which could update and
> truncate the file in one shot, instead of filer.truncate().
>
> This just assume kclient won't support the inline data feature, which
> will be remove soon, more detail please see:
> https://tracker.ceph.com/issues/52916
>
>
> Changed in V6:
> - Fixed the file hole bug, also have updated the MDS side PR.
> - Add add object version support for sync read in #8.
>
>
> Changed in V5:
> - Rebase to "wip-fscrypt-fnames" branch in ceph-client.git repo.
> - Pick up 5 patches from Jeff's "ceph-fscrypt-size-experimental" branch
> in linux.git repo.
> - Add "i_truncate_pagecache_size" member support in ceph_inode_info
> struct, this will be used to truncate the pagecache only in kclient
> side, because the "i_truncate_size" will always be aligned to BLOCK
> SIZE. In fscrypt case we need to use the real size to truncate the
> pagecache.
>
>
> Changed in V4:
> - Retry the truncate request by 20 times before fail it with -EAGAIN.
> - Remove the "fill_last_block" label and move the code to else branch.
> - Remove the #3 patch, which has already been sent out separately, in
> V3 series.
> - Improve some comments in the code.
>
>
> Changed in V3:
> - Fix possibly corrupting the file just before the MDS acquires the
> xlock for FILE lock, another client has updated it.
> - Flush the pagecache buffer before reading the last block for the
> when filling the truncate request.
> - Some other minore fixes.
>
>
>
> Jeff Layton (5):
> libceph: add CEPH_OSD_OP_ASSERT_VER support
> ceph: size handling for encrypted inodes in cap updates
> ceph: fscrypt_file field handling in MClientRequest messages
> ceph: get file size from fscrypt_file when present in inode traces
> ceph: handle fscrypt fields in cap messages from MDS
>
> Xiubo Li (4):
> ceph: add __ceph_get_caps helper support
> ceph: add __ceph_sync_read helper support
> ceph: add object version support for sync read
> ceph: add truncate size handling support for fscrypt
>
> fs/ceph/caps.c | 136 ++++++++++++++----
> fs/ceph/crypto.h | 4 +
> fs/ceph/dir.c | 3 +
> fs/ceph/file.c | 76 ++++++++--
> fs/ceph/inode.c | 243 +++++++++++++++++++++++++++++---
> fs/ceph/mds_client.c | 9 +-
> fs/ceph/mds_client.h | 2 +
> fs/ceph/super.h | 25 ++++
> include/linux/ceph/crypto.h | 28 ++++
> include/linux/ceph/osd_client.h | 6 +-
> include/linux/ceph/rados.h | 4 +
> net/ceph/osd_client.c | 5 +
> 12 files changed, 482 insertions(+), 59 deletions(-)
> create mode 100644 include/linux/ceph/crypto.h
>
^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: [PATCH v6 0/9] ceph: size handling for the fscrypt
2021-11-04 6:00 ` [PATCH v6 0/9] ceph: size handling for the fscrypt Xiubo Li
@ 2021-11-04 10:20 ` Jeff Layton
2021-11-04 10:45 ` Xiubo Li
0 siblings, 1 reply; 17+ messages in thread
From: Jeff Layton @ 2021-11-04 10:20 UTC (permalink / raw)
To: Xiubo Li; +Cc: idryomov, vshankar, pdonnell, khiremat, ceph-devel
Thanks. I'm now seeing a MDS crash when I run that test, but my MDS
branch isn't quite up to date with yours. I'll rebuild that and re-test.
Thanks,
Jeff
On Thu, 2021-11-04 at 14:00 +0800, Xiubo Li wrote:
> The xfstests generic/014 test passed:
>
> [root@ceph1 xfstests]# pwd
> /mnt/kcephfs/xfstests
> [root@ceph1 xfstests]# cat ./local.config
> export FSTYP=ceph
> export TEST_DEV=10.72.49.127:40084:/test
> export TEST_DIR=/mnt/kcephfs/test/_brpcfnn
> export TEST_FS_MOUNT_OPTS="-o
> test_dummy_encryption,name=admin,secret=AQDS3IFhEtxvORAAxn1d4FVN2bRUsc/TZMpQvQ=="
>
> [root@ceph1 xfstests]# ./check generic/014
>
> FSTYP -- ceph
> PLATFORM -- Linux/x86_64 ceph1 5.15.0-rc6+
>
> generic/014 533s ... 558s
> Ran: generic/014
> Passed all 1 tests
>
>
>
>
> On 11/4/21 1:52 PM, xiubli@redhat.com wrote:
> > From: Xiubo Li <xiubli@redhat.com>
> >
> > This patch series is based on the "wip-fscrypt-fnames" branch in
> > repo https://github.com/ceph/ceph-client.git.
> >
> > And I have picked up 5 patches from the "ceph-fscrypt-size-experimental"
> > branch in repo
> > https://git.kernel.org/pub/scm/linux/kernel/git/jlayton/linux.git.
> >
> > ====
> >
> > This approach is based on the discussion from V1 and V2, which will
> > pass the encrypted last block contents to MDS along with the truncate
> > request.
> >
> > This will send the encrypted last block contents to MDS along with
> > the truncate request when truncating to a smaller size and at the
> > same time new size does not align to BLOCK SIZE.
> >
> > The MDS side patch is raised in PR
> > https://github.com/ceph/ceph/pull/43588, which is also based Jeff's
> > previous great work in PR https://github.com/ceph/ceph/pull/41284.
> >
> > The MDS will use the filer.write_trunc(), which could update and
> > truncate the file in one shot, instead of filer.truncate().
> >
> > This just assume kclient won't support the inline data feature, which
> > will be remove soon, more detail please see:
> > https://tracker.ceph.com/issues/52916
> >
> >
> > Changed in V6:
> > - Fixed the file hole bug, also have updated the MDS side PR.
> > - Add add object version support for sync read in #8.
> >
> >
> > Changed in V5:
> > - Rebase to "wip-fscrypt-fnames" branch in ceph-client.git repo.
> > - Pick up 5 patches from Jeff's "ceph-fscrypt-size-experimental" branch
> > in linux.git repo.
> > - Add "i_truncate_pagecache_size" member support in ceph_inode_info
> > struct, this will be used to truncate the pagecache only in kclient
> > side, because the "i_truncate_size" will always be aligned to BLOCK
> > SIZE. In fscrypt case we need to use the real size to truncate the
> > pagecache.
> >
> >
> > Changed in V4:
> > - Retry the truncate request by 20 times before fail it with -EAGAIN.
> > - Remove the "fill_last_block" label and move the code to else branch.
> > - Remove the #3 patch, which has already been sent out separately, in
> > V3 series.
> > - Improve some comments in the code.
> >
> >
> > Changed in V3:
> > - Fix possibly corrupting the file just before the MDS acquires the
> > xlock for FILE lock, another client has updated it.
> > - Flush the pagecache buffer before reading the last block for the
> > when filling the truncate request.
> > - Some other minore fixes.
> >
> >
> >
> > Jeff Layton (5):
> > libceph: add CEPH_OSD_OP_ASSERT_VER support
> > ceph: size handling for encrypted inodes in cap updates
> > ceph: fscrypt_file field handling in MClientRequest messages
> > ceph: get file size from fscrypt_file when present in inode traces
> > ceph: handle fscrypt fields in cap messages from MDS
> >
> > Xiubo Li (4):
> > ceph: add __ceph_get_caps helper support
> > ceph: add __ceph_sync_read helper support
> > ceph: add object version support for sync read
> > ceph: add truncate size handling support for fscrypt
> >
> > fs/ceph/caps.c | 136 ++++++++++++++----
> > fs/ceph/crypto.h | 4 +
> > fs/ceph/dir.c | 3 +
> > fs/ceph/file.c | 76 ++++++++--
> > fs/ceph/inode.c | 243 +++++++++++++++++++++++++++++---
> > fs/ceph/mds_client.c | 9 +-
> > fs/ceph/mds_client.h | 2 +
> > fs/ceph/super.h | 25 ++++
> > include/linux/ceph/crypto.h | 28 ++++
> > include/linux/ceph/osd_client.h | 6 +-
> > include/linux/ceph/rados.h | 4 +
> > net/ceph/osd_client.c | 5 +
> > 12 files changed, 482 insertions(+), 59 deletions(-)
> > create mode 100644 include/linux/ceph/crypto.h
> >
>
--
Jeff Layton <jlayton@kernel.org>
^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: [PATCH v6 0/9] ceph: size handling for the fscrypt
2021-11-04 10:20 ` Jeff Layton
@ 2021-11-04 10:45 ` Xiubo Li
0 siblings, 0 replies; 17+ messages in thread
From: Xiubo Li @ 2021-11-04 10:45 UTC (permalink / raw)
To: Jeff Layton; +Cc: idryomov, vshankar, pdonnell, khiremat, ceph-devel
On 11/4/21 6:20 PM, Jeff Layton wrote:
> Thanks. I'm now seeing a MDS crash when I run that test, but my MDS
> branch isn't quite up to date with yours. I'll rebuild that and re-test.
I didn't hit the crash today, you need to update that branch.
Let me know if you still can see it after that.
Thanks.
> Thanks,
> Jeff
>
> On Thu, 2021-11-04 at 14:00 +0800, Xiubo Li wrote:
>> The xfstests generic/014 test passed:
>>
>> [root@ceph1 xfstests]# pwd
>> /mnt/kcephfs/xfstests
>> [root@ceph1 xfstests]# cat ./local.config
>> export FSTYP=ceph
>> export TEST_DEV=10.72.49.127:40084:/test
>> export TEST_DIR=/mnt/kcephfs/test/_brpcfnn
>> export TEST_FS_MOUNT_OPTS="-o
>> test_dummy_encryption,name=admin,secret=AQDS3IFhEtxvORAAxn1d4FVN2bRUsc/TZMpQvQ=="
>>
>> [root@ceph1 xfstests]# ./check generic/014
>>
>> FSTYP -- ceph
>> PLATFORM -- Linux/x86_64 ceph1 5.15.0-rc6+
>>
>> generic/014 533s ... 558s
>> Ran: generic/014
>> Passed all 1 tests
>>
>>
>>
>>
>> On 11/4/21 1:52 PM, xiubli@redhat.com wrote:
>>> From: Xiubo Li <xiubli@redhat.com>
>>>
>>> This patch series is based on the "wip-fscrypt-fnames" branch in
>>> repo https://github.com/ceph/ceph-client.git.
>>>
>>> And I have picked up 5 patches from the "ceph-fscrypt-size-experimental"
>>> branch in repo
>>> https://git.kernel.org/pub/scm/linux/kernel/git/jlayton/linux.git.
>>>
>>> ====
>>>
>>> This approach is based on the discussion from V1 and V2, which will
>>> pass the encrypted last block contents to MDS along with the truncate
>>> request.
>>>
>>> This will send the encrypted last block contents to MDS along with
>>> the truncate request when truncating to a smaller size and at the
>>> same time new size does not align to BLOCK SIZE.
>>>
>>> The MDS side patch is raised in PR
>>> https://github.com/ceph/ceph/pull/43588, which is also based Jeff's
>>> previous great work in PR https://github.com/ceph/ceph/pull/41284.
>>>
>>> The MDS will use the filer.write_trunc(), which could update and
>>> truncate the file in one shot, instead of filer.truncate().
>>>
>>> This just assume kclient won't support the inline data feature, which
>>> will be remove soon, more detail please see:
>>> https://tracker.ceph.com/issues/52916
>>>
>>>
>>> Changed in V6:
>>> - Fixed the file hole bug, also have updated the MDS side PR.
>>> - Add add object version support for sync read in #8.
>>>
>>>
>>> Changed in V5:
>>> - Rebase to "wip-fscrypt-fnames" branch in ceph-client.git repo.
>>> - Pick up 5 patches from Jeff's "ceph-fscrypt-size-experimental" branch
>>> in linux.git repo.
>>> - Add "i_truncate_pagecache_size" member support in ceph_inode_info
>>> struct, this will be used to truncate the pagecache only in kclient
>>> side, because the "i_truncate_size" will always be aligned to BLOCK
>>> SIZE. In fscrypt case we need to use the real size to truncate the
>>> pagecache.
>>>
>>>
>>> Changed in V4:
>>> - Retry the truncate request by 20 times before fail it with -EAGAIN.
>>> - Remove the "fill_last_block" label and move the code to else branch.
>>> - Remove the #3 patch, which has already been sent out separately, in
>>> V3 series.
>>> - Improve some comments in the code.
>>>
>>>
>>> Changed in V3:
>>> - Fix possibly corrupting the file just before the MDS acquires the
>>> xlock for FILE lock, another client has updated it.
>>> - Flush the pagecache buffer before reading the last block for the
>>> when filling the truncate request.
>>> - Some other minore fixes.
>>>
>>>
>>>
>>> Jeff Layton (5):
>>> libceph: add CEPH_OSD_OP_ASSERT_VER support
>>> ceph: size handling for encrypted inodes in cap updates
>>> ceph: fscrypt_file field handling in MClientRequest messages
>>> ceph: get file size from fscrypt_file when present in inode traces
>>> ceph: handle fscrypt fields in cap messages from MDS
>>>
>>> Xiubo Li (4):
>>> ceph: add __ceph_get_caps helper support
>>> ceph: add __ceph_sync_read helper support
>>> ceph: add object version support for sync read
>>> ceph: add truncate size handling support for fscrypt
>>>
>>> fs/ceph/caps.c | 136 ++++++++++++++----
>>> fs/ceph/crypto.h | 4 +
>>> fs/ceph/dir.c | 3 +
>>> fs/ceph/file.c | 76 ++++++++--
>>> fs/ceph/inode.c | 243 +++++++++++++++++++++++++++++---
>>> fs/ceph/mds_client.c | 9 +-
>>> fs/ceph/mds_client.h | 2 +
>>> fs/ceph/super.h | 25 ++++
>>> include/linux/ceph/crypto.h | 28 ++++
>>> include/linux/ceph/osd_client.h | 6 +-
>>> include/linux/ceph/rados.h | 4 +
>>> net/ceph/osd_client.c | 5 +
>>> 12 files changed, 482 insertions(+), 59 deletions(-)
>>> create mode 100644 include/linux/ceph/crypto.h
>>>
^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: [PATCH v6 9/9] ceph: add truncate size handling support for fscrypt
2021-11-04 5:52 ` [PATCH v6 9/9] ceph: add truncate size handling support for fscrypt xiubli
@ 2021-11-05 13:36 ` Jeff Layton
2021-11-05 14:11 ` Xiubo Li
0 siblings, 1 reply; 17+ messages in thread
From: Jeff Layton @ 2021-11-05 13:36 UTC (permalink / raw)
To: xiubli; +Cc: idryomov, vshankar, pdonnell, khiremat, ceph-devel
[-- Attachment #1: Type: text/plain, Size: 15716 bytes --]
On Thu, 2021-11-04 at 13:52 +0800, xiubli@redhat.com wrote:
> From: Xiubo Li <xiubli@redhat.com>
>
> This will transfer the encrypted last block contents to the MDS
> along with the truncate request only when the new size is smaller
> and not aligned to the fscrypt BLOCK size. When the last block is
> located in the file hole, the truncate request will only contain
> the header.
>
> The MDS could fail to do the truncate if there has another client
> or process has already updated the Rados object which contains
> the last block, and will return -EAGAIN, then the kclient needs
> to retry it. The RMW will take around 50ms, and will let it retry
> 20 times for now.
>
> Signed-off-by: Xiubo Li <xiubli@redhat.com>
> ---
> fs/ceph/inode.c | 205 ++++++++++++++++++++++++++++++++++--
> fs/ceph/super.h | 5 +
> include/linux/ceph/crypto.h | 28 +++++
> 3 files changed, 227 insertions(+), 11 deletions(-)
> create mode 100644 include/linux/ceph/crypto.h
>
> diff --git a/fs/ceph/inode.c b/fs/ceph/inode.c
> index 15c2fb1e2c8a..5817685ea9a5 100644
> --- a/fs/ceph/inode.c
> +++ b/fs/ceph/inode.c
> @@ -21,6 +21,7 @@
> #include "cache.h"
> #include "crypto.h"
> #include <linux/ceph/decode.h>
> +#include <linux/ceph/crypto.h>
>
> /*
> * Ceph inode operations
> @@ -586,6 +587,7 @@ struct inode *ceph_alloc_inode(struct super_block *sb)
> ci->i_truncate_seq = 0;
> ci->i_truncate_size = 0;
> ci->i_truncate_pending = 0;
> + ci->i_truncate_pagecache_size = 0;
>
> ci->i_max_size = 0;
> ci->i_reported_size = 0;
> @@ -751,6 +753,10 @@ int ceph_fill_file_size(struct inode *inode, int issued,
> dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
> truncate_size);
> ci->i_truncate_size = truncate_size;
> + if (IS_ENCRYPTED(inode))
> + ci->i_truncate_pagecache_size = size;
> + else
> + ci->i_truncate_pagecache_size = truncate_size;
> }
>
> if (queue_trunc)
> @@ -1026,10 +1032,14 @@ int ceph_fill_inode(struct inode *inode, struct page *locked_page,
> pool_ns = old_ns;
>
> if (IS_ENCRYPTED(inode) && size &&
> - (iinfo->fscrypt_file_len == sizeof(__le64))) {
> - size = __le64_to_cpu(*(__le64 *)iinfo->fscrypt_file);
> - if (info->size != round_up(size, CEPH_FSCRYPT_BLOCK_SIZE))
> - pr_warn("size=%llu fscrypt_file=%llu\n", info->size, size);
> + (iinfo->fscrypt_file_len >= sizeof(__le64))) {
> + u64 fsize = __le64_to_cpu(*(__le64 *)iinfo->fscrypt_file);
> + if (fsize) {
> + size = fsize;
> + if (info->size != round_up(size, CEPH_FSCRYPT_BLOCK_SIZE))
> + pr_warn("size=%llu fscrypt_file=%llu\n",
> + info->size, size);
> + }
> }
>
> queue_trunc = ceph_fill_file_size(inode, issued,
> @@ -2142,7 +2152,7 @@ void __ceph_do_pending_vmtruncate(struct inode *inode)
> /* there should be no reader or writer */
> WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
>
> - to = ci->i_truncate_size;
> + to = ci->i_truncate_pagecache_size;
> wrbuffer_refs = ci->i_wrbuffer_ref;
> dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
> ci->i_truncate_pending, to);
> @@ -2151,7 +2161,7 @@ void __ceph_do_pending_vmtruncate(struct inode *inode)
> truncate_pagecache(inode, to);
>
> spin_lock(&ci->i_ceph_lock);
> - if (to == ci->i_truncate_size) {
> + if (to == ci->i_truncate_pagecache_size) {
> ci->i_truncate_pending = 0;
> finish = 1;
> }
> @@ -2232,6 +2242,141 @@ static const struct inode_operations ceph_encrypted_symlink_iops = {
> .listxattr = ceph_listxattr,
> };
>
> +/*
> + * Transfer the encrypted last block to the MDS and the MDS
> + * will help update it when truncating a smaller size.
> + *
> + * We don't support a PAGE_SIZE that is smaller than the
> + * CEPH_FSCRYPT_BLOCK_SIZE.
> + */
> +static int fill_fscrypt_truncate(struct inode *inode,
> + struct ceph_mds_request *req,
> + struct iattr *attr)
> +{
> + struct ceph_inode_info *ci = ceph_inode(inode);
> + int boff = attr->ia_size % CEPH_FSCRYPT_BLOCK_SIZE;
> + loff_t pos, orig_pos = round_down(attr->ia_size, CEPH_FSCRYPT_BLOCK_SIZE);
> +#if 0
> + u64 block = orig_pos >> CEPH_FSCRYPT_BLOCK_SHIFT;
> +#endif
> + struct ceph_pagelist *pagelist = NULL;
> + struct kvec iov;
> + struct iov_iter iter;
> + struct page *page = NULL;
> + struct ceph_fscrypt_truncate_size_header header;
> + int retry_op = 0;
> + int len = CEPH_FSCRYPT_BLOCK_SIZE;
> + loff_t i_size = i_size_read(inode);
> + struct ceph_object_vers objvers = {0, NULL};
> + int got, ret, issued;
> +
> + ret = __ceph_get_caps(inode, NULL, CEPH_CAP_FILE_RD, 0, -1, &got);
> + if (ret < 0)
> + return ret;
> +
> + issued = __ceph_caps_issued(ci, NULL);
> +
> + dout("%s size %lld -> %lld got cap refs on %s, issued %s\n", __func__,
> + i_size, attr->ia_size, ceph_cap_string(got),
> + ceph_cap_string(issued));
> +
> + /* Try to writeback the dirty pagecaches */
> + if (issued & (CEPH_CAP_FILE_BUFFER))
> + filemap_fdatawrite(&inode->i_data);
> +
> + page = __page_cache_alloc(GFP_KERNEL);
> + if (page == NULL) {
> + ret = -ENOMEM;
> + goto out;
> + }
> +
> + pagelist = ceph_pagelist_alloc(GFP_KERNEL);
> + if (!pagelist) {
> + ret = -ENOMEM;
> + goto out;
> + }
> +
> + iov.iov_base = kmap_local_page(page);
> + iov.iov_len = len;
> + iov_iter_kvec(&iter, READ, &iov, 1, len);
> +
> + pos = orig_pos;
> + ret = __ceph_sync_read(inode, &pos, &iter, &retry_op, &objvers);
> + ceph_put_cap_refs(ci, got);
> + if (ret < 0)
> + goto out;
> +
> + WARN_ON_ONCE(objvers.count != 1);
> +
> + /* Insert the header first */
> + header.ver = 1;
> + header.compat = 1;
> +
> + /*
> + * If we hit a hole here, we should just skip filling
> + * the fscrypt for the request, because once the fscrypt
> + * is enabled, the file will be split into many blocks
> + * with the size of CEPH_FSCRYPT_BLOCK_SIZE, if there
> + * has a hole, the hole size should be multiple of block
> + * size.
> + *
> + * If the Rados object doesn't exist, it will be set 0.
> + */
> + if (!objvers.objvers[0].objver) {
> + dout("%s hit hole, ppos %lld < size %lld\n", __func__,
> + pos, i_size);
> +
> + header.data_len = cpu_to_le32(8 + 8 + 4);
> + header.assert_ver = cpu_to_le64(0);
> + header.file_offset = cpu_to_le64(0);
> + header.block_size = cpu_to_le64(0);
Note that 0 is a special case, and nothing will complain if you don't
endian-convert it before assigning it.
> + ret = 0;
> + } else {
> + header.data_len = cpu_to_le32(8 + 8 + 4 + CEPH_FSCRYPT_BLOCK_SIZE);
> + header.assert_ver = objvers.objvers[0].objver;
> + header.file_offset = cpu_to_le64(orig_pos);
> + header.block_size = cpu_to_le64(CEPH_FSCRYPT_BLOCK_SIZE);
> +
> + /* truncate and zero out the extra contents for the last block */
> + memset(iov.iov_base + boff, 0, PAGE_SIZE - boff);
>
>
sparse throws some warnings here:
CHECK fs/ceph/inode.c
fs/ceph/inode.c:1020:32: warning: incorrect type in initializer (different base types)
fs/ceph/inode.c:1020:32: expected unsigned long long [usertype] size
fs/ceph/inode.c:1020:32: got restricted __le64 [usertype] size
fs/ceph/inode.c:1039:41: warning: restricted __le64 degrades to integer
fs/ceph/inode.c:1048:41: warning: cast to restricted __le64
fs/ceph/inode.c:2332:35: warning: incorrect type in assignment (different base types)
fs/ceph/inode.c:2332:35: expected restricted __le32 [assigned] [usertype] block_size
fs/ceph/inode.c:2332:35: got restricted __le64 [usertype]
fs/ceph/inode.c:2336:35: warning: incorrect type in assignment (different base types)
fs/ceph/inode.c:2336:35: expected restricted __le64 [assigned] [usertype] assert_ver
fs/ceph/inode.c:2336:35: got unsigned long long [usertype] objver
fs/ceph/inode.c:2338:35: warning: incorrect type in assignment (different base types)
fs/ceph/inode.c:2338:35: expected restricted __le32 [assigned] [usertype] block_size
fs/ceph/inode.c:2338:35: got restricted __le64 [usertype]
fs/ceph/inode.c:2549:45: warning: incorrect type in assignment (different base types)
fs/ceph/inode.c:2549:45: expected restricted __le64 [usertype] r_fscrypt_file
fs/ceph/inode.c:2549:45: got long long [usertype] ia_size
fs/ceph/inode.c:2573:53: warning: incorrect type in assignment (different base types)
fs/ceph/inode.c:2573:53: expected restricted __le64 [usertype] r_fscrypt_file
fs/ceph/inode.c:2573:53: got long long [usertype] ia_size
I've attached a patch to this email. Can you fold those deltas into the
appropriate patches in your series and resend?
FWIW, esp. when dealing with this sort of endianness-converting code,
it's often a good idea to install sparse and build the kmod with C=1,
which will catch these sorts of warnings.
Here's what I usually use this to build the module:
$ make -j16 M=fs/ceph W=1 C=1
There are some persistent bogus warnings about lock imbalances that I
sitll need to fix one of these days, but I've just been ignoring those
for now.
> +
> +#if 0 // Uncomment this when the fscrypt is enabled globally in kceph
> +
> + /* encrypt the last block */
> + ret = fscrypt_encrypt_block_inplace(inode, page,
> + CEPH_FSCRYPT_BLOCK_SIZE,
> + 0, block,
> + GFP_KERNEL);
> + if (ret)
> + goto out;
> +#endif
> + }
> +
> + /* Insert the header */
> + ret = ceph_pagelist_append(pagelist, &header, sizeof(header));
> + if (ret)
> + goto out;
> +
> + if (header.block_size) {
> + /* Append the last block contents to pagelist */
> + ret = ceph_pagelist_append(pagelist, iov.iov_base,
> + CEPH_FSCRYPT_BLOCK_SIZE);
> + if (ret)
> + goto out;
> + }
> + req->r_pagelist = pagelist;
> +out:
> + dout("%s %p size dropping cap refs on %s\n", __func__,
> + inode, ceph_cap_string(got));
> + kunmap_local(iov.iov_base);
> + if (page)
> + __free_pages(page, 0);
> + if (ret && pagelist)
> + ceph_pagelist_release(pagelist);
> + kfree(objvers.objvers);
> + return ret;
> +}
> +
> int __ceph_setattr(struct inode *inode, struct iattr *attr, struct ceph_iattr *cia)
> {
> struct ceph_inode_info *ci = ceph_inode(inode);
> @@ -2239,12 +2384,15 @@ int __ceph_setattr(struct inode *inode, struct iattr *attr, struct ceph_iattr *c
> struct ceph_mds_request *req;
> struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
> struct ceph_cap_flush *prealloc_cf;
> + loff_t isize = i_size_read(inode);
> int issued;
> int release = 0, dirtied = 0;
> int mask = 0;
> int err = 0;
> int inode_dirty_flags = 0;
> bool lock_snap_rwsem = false;
> + bool fill_fscrypt;
> + int truncate_retry = 20; /* The RMW will take around 50ms */
>
> prealloc_cf = ceph_alloc_cap_flush();
> if (!prealloc_cf)
> @@ -2257,6 +2405,8 @@ int __ceph_setattr(struct inode *inode, struct iattr *attr, struct ceph_iattr *c
> return PTR_ERR(req);
> }
>
> +retry:
> + fill_fscrypt = false;
> spin_lock(&ci->i_ceph_lock);
> issued = __ceph_caps_issued(ci, NULL);
>
> @@ -2378,10 +2528,27 @@ int __ceph_setattr(struct inode *inode, struct iattr *attr, struct ceph_iattr *c
> }
> }
> if (ia_valid & ATTR_SIZE) {
> - loff_t isize = i_size_read(inode);
> -
> dout("setattr %p size %lld -> %lld\n", inode, isize, attr->ia_size);
> - if ((issued & CEPH_CAP_FILE_EXCL) && attr->ia_size >= isize) {
> + /*
> + * Only when the new size is smaller and not aligned to
> + * CEPH_FSCRYPT_BLOCK_SIZE will the RMW is needed.
> + */
> + if (IS_ENCRYPTED(inode) && attr->ia_size < isize &&
> + (attr->ia_size % CEPH_FSCRYPT_BLOCK_SIZE)) {
> + mask |= CEPH_SETATTR_SIZE;
> + release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_EXCL |
> + CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
> + set_bit(CEPH_MDS_R_FSCRYPT_FILE, &req->r_req_flags);
> + mask |= CEPH_SETATTR_FSCRYPT_FILE;
> + req->r_args.setattr.size =
> + cpu_to_le64(round_up(attr->ia_size,
> + CEPH_FSCRYPT_BLOCK_SIZE));
> + req->r_args.setattr.old_size =
> + cpu_to_le64(round_up(isize,
> + CEPH_FSCRYPT_BLOCK_SIZE));
> + req->r_fscrypt_file = attr->ia_size;
> + fill_fscrypt = true;
> + } else if ((issued & CEPH_CAP_FILE_EXCL) && attr->ia_size >= isize) {
> if (attr->ia_size > isize) {
> i_size_write(inode, attr->ia_size);
> inode->i_blocks = calc_inode_blocks(attr->ia_size);
> @@ -2404,7 +2571,6 @@ int __ceph_setattr(struct inode *inode, struct iattr *attr, struct ceph_iattr *c
> cpu_to_le64(round_up(isize,
> CEPH_FSCRYPT_BLOCK_SIZE));
> req->r_fscrypt_file = attr->ia_size;
> - /* FIXME: client must zero out any partial blocks! */
> } else {
> req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
> req->r_args.setattr.old_size = cpu_to_le64(isize);
> @@ -2476,7 +2642,6 @@ int __ceph_setattr(struct inode *inode, struct iattr *attr, struct ceph_iattr *c
> if (inode_dirty_flags)
> __mark_inode_dirty(inode, inode_dirty_flags);
>
> -
> if (mask) {
> req->r_inode = inode;
> ihold(inode);
> @@ -2484,7 +2649,25 @@ int __ceph_setattr(struct inode *inode, struct iattr *attr, struct ceph_iattr *c
> req->r_args.setattr.mask = cpu_to_le32(mask);
> req->r_num_caps = 1;
> req->r_stamp = attr->ia_ctime;
> + if (fill_fscrypt) {
> + err = fill_fscrypt_truncate(inode, req, attr);
> + if (err)
> + goto out;
> + }
> +
> + /*
> + * The truncate request will return -EAGAIN when the
> + * last block has been updated just before the MDS
> + * successfully gets the xlock for the FILE lock. To
> + * avoid corrupting the file contents we need to retry
> + * it.
> + */
> err = ceph_mdsc_do_request(mdsc, NULL, req);
> + if (err == -EAGAIN && truncate_retry--) {
> + dout("setattr %p result=%d (%s locally, %d remote), retry it!\n",
> + inode, err, ceph_cap_string(dirtied), mask);
> + goto retry;
> + }
> }
> out:
> dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
> diff --git a/fs/ceph/super.h b/fs/ceph/super.h
> index b347b12e86a9..071857bb59d8 100644
> --- a/fs/ceph/super.h
> +++ b/fs/ceph/super.h
> @@ -408,6 +408,11 @@ struct ceph_inode_info {
> u32 i_truncate_seq; /* last truncate to smaller size */
> u64 i_truncate_size; /* and the size we last truncated down to */
> int i_truncate_pending; /* still need to call vmtruncate */
> + /*
> + * For none fscrypt case it equals to i_truncate_size or it will
> + * equals to fscrypt_file_size
> + */
> + u64 i_truncate_pagecache_size;
>
> u64 i_max_size; /* max file size authorized by mds */
> u64 i_reported_size; /* (max_)size reported to or requested of mds */
> diff --git a/include/linux/ceph/crypto.h b/include/linux/ceph/crypto.h
> new file mode 100644
> index 000000000000..2b0961902887
> --- /dev/null
> +++ b/include/linux/ceph/crypto.h
> @@ -0,0 +1,28 @@
> +/* SPDX-License-Identifier: GPL-2.0 */
> +#ifndef _FS_CEPH_CRYPTO_H
> +#define _FS_CEPH_CRYPTO_H
> +
> +#include <linux/types.h>
> +
> +/*
> + * Header for the crypted file when truncating the size, this
> + * will be sent to MDS, and the MDS will update the encrypted
> + * last block and then truncate the size.
> + */
> +struct ceph_fscrypt_truncate_size_header {
> + __u8 ver;
> + __u8 compat;
> +
> + /*
> + * It will be sizeof(assert_ver + file_offset + block_size)
> + * if the last block is empty when it's located in a file
> + * hole. Or the data_len will plus CEPH_FSCRYPT_BLOCK_SIZE.
> + */
> + __le32 data_len;
> +
> + __le64 assert_ver;
> + __le64 file_offset;
> + __le32 block_size;
> +} __packed;
> +
> +#endif
When I said to move this to crypto.h, I meant fs/ceph/crypto.h. Let's
not add a new header file for this. Can you move this definition into
there?
Thanks,
--
Jeff Layton <jlayton@kernel.org>
[-- Attachment #2: 0001-SQUASH-fix-some-endianness-issues-noticed-by-sparse.patch --]
[-- Type: text/x-patch, Size: 3018 bytes --]
From 651d7b1539b4e7b0273fb5ccf9b865ef31f50d60 Mon Sep 17 00:00:00 2001
From: Jeff Layton <jlayton@kernel.org>
Date: Fri, 5 Nov 2021 09:24:02 -0400
Subject: [PATCH] SQUASH: fix some endianness issues noticed by sparse
Signed-off-by: Jeff Layton <jlayton@kernel.org>
---
fs/ceph/inode.c | 18 +++++++++---------
fs/ceph/mds_client.h | 2 +-
2 files changed, 10 insertions(+), 10 deletions(-)
diff --git a/fs/ceph/inode.c b/fs/ceph/inode.c
index 5817685ea9a5..8693007a6f2e 100644
--- a/fs/ceph/inode.c
+++ b/fs/ceph/inode.c
@@ -1017,7 +1017,7 @@ int ceph_fill_inode(struct inode *inode, struct page *locked_page,
if (new_version ||
(new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) {
- u64 size = info->size;
+ u64 size = le64_to_cpu(info->size);
s64 old_pool = ci->i_layout.pool_id;
struct ceph_string *old_ns;
@@ -1036,7 +1036,8 @@ int ceph_fill_inode(struct inode *inode, struct page *locked_page,
u64 fsize = __le64_to_cpu(*(__le64 *)iinfo->fscrypt_file);
if (fsize) {
size = fsize;
- if (info->size != round_up(size, CEPH_FSCRYPT_BLOCK_SIZE))
+ if (le64_to_cpu(info->size) !=
+ round_up(size, CEPH_FSCRYPT_BLOCK_SIZE))
pr_warn("size=%llu fscrypt_file=%llu\n",
info->size, size);
}
@@ -1044,8 +1045,7 @@ int ceph_fill_inode(struct inode *inode, struct page *locked_page,
queue_trunc = ceph_fill_file_size(inode, issued,
le32_to_cpu(info->truncate_seq),
- le64_to_cpu(info->truncate_size),
- le64_to_cpu(size));
+ le64_to_cpu(info->truncate_size), size);
/* only update max_size on auth cap */
if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
ci->i_max_size != le64_to_cpu(info->max_size)) {
@@ -2327,15 +2327,15 @@ static int fill_fscrypt_truncate(struct inode *inode,
pos, i_size);
header.data_len = cpu_to_le32(8 + 8 + 4);
- header.assert_ver = cpu_to_le64(0);
- header.file_offset = cpu_to_le64(0);
- header.block_size = cpu_to_le64(0);
+ header.assert_ver = 0;
+ header.file_offset = 0;
+ header.block_size = 0;
ret = 0;
} else {
header.data_len = cpu_to_le32(8 + 8 + 4 + CEPH_FSCRYPT_BLOCK_SIZE);
- header.assert_ver = objvers.objvers[0].objver;
+ header.assert_ver = cpu_to_le64(objvers.objvers[0].objver);
header.file_offset = cpu_to_le64(orig_pos);
- header.block_size = cpu_to_le64(CEPH_FSCRYPT_BLOCK_SIZE);
+ header.block_size = cpu_to_le32(CEPH_FSCRYPT_BLOCK_SIZE);
/* truncate and zero out the extra contents for the last block */
memset(iov.iov_base + boff, 0, PAGE_SIZE - boff);
diff --git a/fs/ceph/mds_client.h b/fs/ceph/mds_client.h
index d64ff1bd2f5d..149a3a828472 100644
--- a/fs/ceph/mds_client.h
+++ b/fs/ceph/mds_client.h
@@ -284,7 +284,7 @@ struct ceph_mds_request {
union ceph_mds_request_args r_args;
struct ceph_fscrypt_auth *r_fscrypt_auth;
- __le64 r_fscrypt_file;
+ u64 r_fscrypt_file;
u8 *r_altname; /* fscrypt binary crypttext for long filenames */
u32 r_altname_len; /* length of r_altname */
--
2.33.1
^ permalink raw reply related [flat|nested] 17+ messages in thread
* Re: [PATCH v6 9/9] ceph: add truncate size handling support for fscrypt
2021-11-05 13:36 ` Jeff Layton
@ 2021-11-05 14:11 ` Xiubo Li
0 siblings, 0 replies; 17+ messages in thread
From: Xiubo Li @ 2021-11-05 14:11 UTC (permalink / raw)
To: Jeff Layton; +Cc: idryomov, vshankar, pdonnell, khiremat, ceph-devel
On 11/5/21 9:36 PM, Jeff Layton wrote:
> On Thu, 2021-11-04 at 13:52 +0800, xiubli@redhat.com wrote:
>> From: Xiubo Li <xiubli@redhat.com>
>>
>> This will transfer the encrypted last block contents to the MDS
>> along with the truncate request only when the new size is smaller
>> and not aligned to the fscrypt BLOCK size. When the last block is
>> located in the file hole, the truncate request will only contain
>> the header.
>>
>> The MDS could fail to do the truncate if there has another client
>> or process has already updated the Rados object which contains
>> the last block, and will return -EAGAIN, then the kclient needs
>> to retry it. The RMW will take around 50ms, and will let it retry
>> 20 times for now.
>>
>> Signed-off-by: Xiubo Li <xiubli@redhat.com>
>> ---
>> fs/ceph/inode.c | 205 ++++++++++++++++++++++++++++++++++--
>> fs/ceph/super.h | 5 +
>> include/linux/ceph/crypto.h | 28 +++++
>> 3 files changed, 227 insertions(+), 11 deletions(-)
>> create mode 100644 include/linux/ceph/crypto.h
>>
>> diff --git a/fs/ceph/inode.c b/fs/ceph/inode.c
>> index 15c2fb1e2c8a..5817685ea9a5 100644
>> --- a/fs/ceph/inode.c
>> +++ b/fs/ceph/inode.c
>> @@ -21,6 +21,7 @@
>> #include "cache.h"
>> #include "crypto.h"
>> #include <linux/ceph/decode.h>
>> +#include <linux/ceph/crypto.h>
>>
>> /*
>> * Ceph inode operations
>> @@ -586,6 +587,7 @@ struct inode *ceph_alloc_inode(struct super_block *sb)
>> ci->i_truncate_seq = 0;
>> ci->i_truncate_size = 0;
>> ci->i_truncate_pending = 0;
>> + ci->i_truncate_pagecache_size = 0;
>>
>> ci->i_max_size = 0;
>> ci->i_reported_size = 0;
>> @@ -751,6 +753,10 @@ int ceph_fill_file_size(struct inode *inode, int issued,
>> dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
>> truncate_size);
>> ci->i_truncate_size = truncate_size;
>> + if (IS_ENCRYPTED(inode))
>> + ci->i_truncate_pagecache_size = size;
>> + else
>> + ci->i_truncate_pagecache_size = truncate_size;
>> }
>>
>> if (queue_trunc)
>> @@ -1026,10 +1032,14 @@ int ceph_fill_inode(struct inode *inode, struct page *locked_page,
>> pool_ns = old_ns;
>>
>> if (IS_ENCRYPTED(inode) && size &&
>> - (iinfo->fscrypt_file_len == sizeof(__le64))) {
>> - size = __le64_to_cpu(*(__le64 *)iinfo->fscrypt_file);
>> - if (info->size != round_up(size, CEPH_FSCRYPT_BLOCK_SIZE))
>> - pr_warn("size=%llu fscrypt_file=%llu\n", info->size, size);
>> + (iinfo->fscrypt_file_len >= sizeof(__le64))) {
>> + u64 fsize = __le64_to_cpu(*(__le64 *)iinfo->fscrypt_file);
>> + if (fsize) {
>> + size = fsize;
>> + if (info->size != round_up(size, CEPH_FSCRYPT_BLOCK_SIZE))
>> + pr_warn("size=%llu fscrypt_file=%llu\n",
>> + info->size, size);
>> + }
>> }
>>
>> queue_trunc = ceph_fill_file_size(inode, issued,
>> @@ -2142,7 +2152,7 @@ void __ceph_do_pending_vmtruncate(struct inode *inode)
>> /* there should be no reader or writer */
>> WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
>>
>> - to = ci->i_truncate_size;
>> + to = ci->i_truncate_pagecache_size;
>> wrbuffer_refs = ci->i_wrbuffer_ref;
>> dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
>> ci->i_truncate_pending, to);
>> @@ -2151,7 +2161,7 @@ void __ceph_do_pending_vmtruncate(struct inode *inode)
>> truncate_pagecache(inode, to);
>>
>> spin_lock(&ci->i_ceph_lock);
>> - if (to == ci->i_truncate_size) {
>> + if (to == ci->i_truncate_pagecache_size) {
>> ci->i_truncate_pending = 0;
>> finish = 1;
>> }
>> @@ -2232,6 +2242,141 @@ static const struct inode_operations ceph_encrypted_symlink_iops = {
>> .listxattr = ceph_listxattr,
>> };
>>
>> +/*
>> + * Transfer the encrypted last block to the MDS and the MDS
>> + * will help update it when truncating a smaller size.
>> + *
>> + * We don't support a PAGE_SIZE that is smaller than the
>> + * CEPH_FSCRYPT_BLOCK_SIZE.
>> + */
>> +static int fill_fscrypt_truncate(struct inode *inode,
>> + struct ceph_mds_request *req,
>> + struct iattr *attr)
>> +{
>> + struct ceph_inode_info *ci = ceph_inode(inode);
>> + int boff = attr->ia_size % CEPH_FSCRYPT_BLOCK_SIZE;
>> + loff_t pos, orig_pos = round_down(attr->ia_size, CEPH_FSCRYPT_BLOCK_SIZE);
>> +#if 0
>> + u64 block = orig_pos >> CEPH_FSCRYPT_BLOCK_SHIFT;
>> +#endif
>> + struct ceph_pagelist *pagelist = NULL;
>> + struct kvec iov;
>> + struct iov_iter iter;
>> + struct page *page = NULL;
>> + struct ceph_fscrypt_truncate_size_header header;
>> + int retry_op = 0;
>> + int len = CEPH_FSCRYPT_BLOCK_SIZE;
>> + loff_t i_size = i_size_read(inode);
>> + struct ceph_object_vers objvers = {0, NULL};
>> + int got, ret, issued;
>> +
>> + ret = __ceph_get_caps(inode, NULL, CEPH_CAP_FILE_RD, 0, -1, &got);
>> + if (ret < 0)
>> + return ret;
>> +
>> + issued = __ceph_caps_issued(ci, NULL);
>> +
>> + dout("%s size %lld -> %lld got cap refs on %s, issued %s\n", __func__,
>> + i_size, attr->ia_size, ceph_cap_string(got),
>> + ceph_cap_string(issued));
>> +
>> + /* Try to writeback the dirty pagecaches */
>> + if (issued & (CEPH_CAP_FILE_BUFFER))
>> + filemap_fdatawrite(&inode->i_data);
>> +
>> + page = __page_cache_alloc(GFP_KERNEL);
>> + if (page == NULL) {
>> + ret = -ENOMEM;
>> + goto out;
>> + }
>> +
>> + pagelist = ceph_pagelist_alloc(GFP_KERNEL);
>> + if (!pagelist) {
>> + ret = -ENOMEM;
>> + goto out;
>> + }
>> +
>> + iov.iov_base = kmap_local_page(page);
>> + iov.iov_len = len;
>> + iov_iter_kvec(&iter, READ, &iov, 1, len);
>> +
>> + pos = orig_pos;
>> + ret = __ceph_sync_read(inode, &pos, &iter, &retry_op, &objvers);
>> + ceph_put_cap_refs(ci, got);
>> + if (ret < 0)
>> + goto out;
>> +
>> + WARN_ON_ONCE(objvers.count != 1);
>> +
>> + /* Insert the header first */
>> + header.ver = 1;
>> + header.compat = 1;
>> +
>> + /*
>> + * If we hit a hole here, we should just skip filling
>> + * the fscrypt for the request, because once the fscrypt
>> + * is enabled, the file will be split into many blocks
>> + * with the size of CEPH_FSCRYPT_BLOCK_SIZE, if there
>> + * has a hole, the hole size should be multiple of block
>> + * size.
>> + *
>> + * If the Rados object doesn't exist, it will be set 0.
>> + */
>> + if (!objvers.objvers[0].objver) {
>> + dout("%s hit hole, ppos %lld < size %lld\n", __func__,
>> + pos, i_size);
>> +
>> + header.data_len = cpu_to_le32(8 + 8 + 4);
>> + header.assert_ver = cpu_to_le64(0);
>> + header.file_offset = cpu_to_le64(0);
>> + header.block_size = cpu_to_le64(0);
> Note that 0 is a special case, and nothing will complain if you don't
> endian-convert it before assigning it.
Yeah.
>
>> + ret = 0;
>> + } else {
>> + header.data_len = cpu_to_le32(8 + 8 + 4 + CEPH_FSCRYPT_BLOCK_SIZE);
>> + header.assert_ver = objvers.objvers[0].objver;
>> + header.file_offset = cpu_to_le64(orig_pos);
>> + header.block_size = cpu_to_le64(CEPH_FSCRYPT_BLOCK_SIZE);
>> +
>> + /* truncate and zero out the extra contents for the last block */
>> + memset(iov.iov_base + boff, 0, PAGE_SIZE - boff);
>>
>>
> sparse throws some warnings here:
>
> CHECK fs/ceph/inode.c
> fs/ceph/inode.c:1020:32: warning: incorrect type in initializer (different base types)
> fs/ceph/inode.c:1020:32: expected unsigned long long [usertype] size
> fs/ceph/inode.c:1020:32: got restricted __le64 [usertype] size
> fs/ceph/inode.c:1039:41: warning: restricted __le64 degrades to integer
> fs/ceph/inode.c:1048:41: warning: cast to restricted __le64
> fs/ceph/inode.c:2332:35: warning: incorrect type in assignment (different base types)
> fs/ceph/inode.c:2332:35: expected restricted __le32 [assigned] [usertype] block_size
> fs/ceph/inode.c:2332:35: got restricted __le64 [usertype]
> fs/ceph/inode.c:2336:35: warning: incorrect type in assignment (different base types)
> fs/ceph/inode.c:2336:35: expected restricted __le64 [assigned] [usertype] assert_ver
> fs/ceph/inode.c:2336:35: got unsigned long long [usertype] objver
> fs/ceph/inode.c:2338:35: warning: incorrect type in assignment (different base types)
> fs/ceph/inode.c:2338:35: expected restricted __le32 [assigned] [usertype] block_size
> fs/ceph/inode.c:2338:35: got restricted __le64 [usertype]
> fs/ceph/inode.c:2549:45: warning: incorrect type in assignment (different base types)
> fs/ceph/inode.c:2549:45: expected restricted __le64 [usertype] r_fscrypt_file
> fs/ceph/inode.c:2549:45: got long long [usertype] ia_size
> fs/ceph/inode.c:2573:53: warning: incorrect type in assignment (different base types)
> fs/ceph/inode.c:2573:53: expected restricted __le64 [usertype] r_fscrypt_file
> fs/ceph/inode.c:2573:53: got long long [usertype] ia_size
>
> I've attached a patch to this email. Can you fold those deltas into the
> appropriate patches in your series and resend?
>
> FWIW, esp. when dealing with this sort of endianness-converting code,
> it's often a good idea to install sparse and build the kmod with C=1,
> which will catch these sorts of warnings.
>
> Here's what I usually use this to build the module:
>
> $ make -j16 M=fs/ceph W=1 C=1
>
> There are some persistent bogus warnings about lock imbalances that I
> sitll need to fix one of these days, but I've just been ignoring those
> for now.
Will fix this and post the V7 later.
>> +
>> +#if 0 // Uncomment this when the fscrypt is enabled globally in kceph
>> +
>> + /* encrypt the last block */
>> + ret = fscrypt_encrypt_block_inplace(inode, page,
>> + CEPH_FSCRYPT_BLOCK_SIZE,
>> + 0, block,
>> + GFP_KERNEL);
>> + if (ret)
>> + goto out;
>> +#endif
>> + }
>> +
>> + /* Insert the header */
>> + ret = ceph_pagelist_append(pagelist, &header, sizeof(header));
>> + if (ret)
>> + goto out;
>> +
>> + if (header.block_size) {
>> + /* Append the last block contents to pagelist */
>> + ret = ceph_pagelist_append(pagelist, iov.iov_base,
>> + CEPH_FSCRYPT_BLOCK_SIZE);
>> + if (ret)
>> + goto out;
>> + }
>> + req->r_pagelist = pagelist;
>> +out:
>> + dout("%s %p size dropping cap refs on %s\n", __func__,
>> + inode, ceph_cap_string(got));
>> + kunmap_local(iov.iov_base);
>> + if (page)
>> + __free_pages(page, 0);
>> + if (ret && pagelist)
>> + ceph_pagelist_release(pagelist);
>> + kfree(objvers.objvers);
>> + return ret;
>> +}
>> +
>> int __ceph_setattr(struct inode *inode, struct iattr *attr, struct ceph_iattr *cia)
>> {
>> struct ceph_inode_info *ci = ceph_inode(inode);
>> @@ -2239,12 +2384,15 @@ int __ceph_setattr(struct inode *inode, struct iattr *attr, struct ceph_iattr *c
>> struct ceph_mds_request *req;
>> struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
>> struct ceph_cap_flush *prealloc_cf;
>> + loff_t isize = i_size_read(inode);
>> int issued;
>> int release = 0, dirtied = 0;
>> int mask = 0;
>> int err = 0;
>> int inode_dirty_flags = 0;
>> bool lock_snap_rwsem = false;
>> + bool fill_fscrypt;
>> + int truncate_retry = 20; /* The RMW will take around 50ms */
>>
>> prealloc_cf = ceph_alloc_cap_flush();
>> if (!prealloc_cf)
>> @@ -2257,6 +2405,8 @@ int __ceph_setattr(struct inode *inode, struct iattr *attr, struct ceph_iattr *c
>> return PTR_ERR(req);
>> }
>>
>> +retry:
>> + fill_fscrypt = false;
>> spin_lock(&ci->i_ceph_lock);
>> issued = __ceph_caps_issued(ci, NULL);
>>
>> @@ -2378,10 +2528,27 @@ int __ceph_setattr(struct inode *inode, struct iattr *attr, struct ceph_iattr *c
>> }
>> }
>> if (ia_valid & ATTR_SIZE) {
>> - loff_t isize = i_size_read(inode);
>> -
>> dout("setattr %p size %lld -> %lld\n", inode, isize, attr->ia_size);
>> - if ((issued & CEPH_CAP_FILE_EXCL) && attr->ia_size >= isize) {
>> + /*
>> + * Only when the new size is smaller and not aligned to
>> + * CEPH_FSCRYPT_BLOCK_SIZE will the RMW is needed.
>> + */
>> + if (IS_ENCRYPTED(inode) && attr->ia_size < isize &&
>> + (attr->ia_size % CEPH_FSCRYPT_BLOCK_SIZE)) {
>> + mask |= CEPH_SETATTR_SIZE;
>> + release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_EXCL |
>> + CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
>> + set_bit(CEPH_MDS_R_FSCRYPT_FILE, &req->r_req_flags);
>> + mask |= CEPH_SETATTR_FSCRYPT_FILE;
>> + req->r_args.setattr.size =
>> + cpu_to_le64(round_up(attr->ia_size,
>> + CEPH_FSCRYPT_BLOCK_SIZE));
>> + req->r_args.setattr.old_size =
>> + cpu_to_le64(round_up(isize,
>> + CEPH_FSCRYPT_BLOCK_SIZE));
>> + req->r_fscrypt_file = attr->ia_size;
>> + fill_fscrypt = true;
>> + } else if ((issued & CEPH_CAP_FILE_EXCL) && attr->ia_size >= isize) {
>> if (attr->ia_size > isize) {
>> i_size_write(inode, attr->ia_size);
>> inode->i_blocks = calc_inode_blocks(attr->ia_size);
>> @@ -2404,7 +2571,6 @@ int __ceph_setattr(struct inode *inode, struct iattr *attr, struct ceph_iattr *c
>> cpu_to_le64(round_up(isize,
>> CEPH_FSCRYPT_BLOCK_SIZE));
>> req->r_fscrypt_file = attr->ia_size;
>> - /* FIXME: client must zero out any partial blocks! */
>> } else {
>> req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
>> req->r_args.setattr.old_size = cpu_to_le64(isize);
>> @@ -2476,7 +2642,6 @@ int __ceph_setattr(struct inode *inode, struct iattr *attr, struct ceph_iattr *c
>> if (inode_dirty_flags)
>> __mark_inode_dirty(inode, inode_dirty_flags);
>>
>> -
>> if (mask) {
>> req->r_inode = inode;
>> ihold(inode);
>> @@ -2484,7 +2649,25 @@ int __ceph_setattr(struct inode *inode, struct iattr *attr, struct ceph_iattr *c
>> req->r_args.setattr.mask = cpu_to_le32(mask);
>> req->r_num_caps = 1;
>> req->r_stamp = attr->ia_ctime;
>> + if (fill_fscrypt) {
>> + err = fill_fscrypt_truncate(inode, req, attr);
>> + if (err)
>> + goto out;
>> + }
>> +
>> + /*
>> + * The truncate request will return -EAGAIN when the
>> + * last block has been updated just before the MDS
>> + * successfully gets the xlock for the FILE lock. To
>> + * avoid corrupting the file contents we need to retry
>> + * it.
>> + */
>> err = ceph_mdsc_do_request(mdsc, NULL, req);
>> + if (err == -EAGAIN && truncate_retry--) {
>> + dout("setattr %p result=%d (%s locally, %d remote), retry it!\n",
>> + inode, err, ceph_cap_string(dirtied), mask);
>> + goto retry;
>> + }
>> }
>> out:
>> dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
>> diff --git a/fs/ceph/super.h b/fs/ceph/super.h
>> index b347b12e86a9..071857bb59d8 100644
>> --- a/fs/ceph/super.h
>> +++ b/fs/ceph/super.h
>> @@ -408,6 +408,11 @@ struct ceph_inode_info {
>> u32 i_truncate_seq; /* last truncate to smaller size */
>> u64 i_truncate_size; /* and the size we last truncated down to */
>> int i_truncate_pending; /* still need to call vmtruncate */
>> + /*
>> + * For none fscrypt case it equals to i_truncate_size or it will
>> + * equals to fscrypt_file_size
>> + */
>> + u64 i_truncate_pagecache_size;
>>
>> u64 i_max_size; /* max file size authorized by mds */
>> u64 i_reported_size; /* (max_)size reported to or requested of mds */
>> diff --git a/include/linux/ceph/crypto.h b/include/linux/ceph/crypto.h
>> new file mode 100644
>> index 000000000000..2b0961902887
>> --- /dev/null
>> +++ b/include/linux/ceph/crypto.h
>> @@ -0,0 +1,28 @@
>> +/* SPDX-License-Identifier: GPL-2.0 */
>> +#ifndef _FS_CEPH_CRYPTO_H
>> +#define _FS_CEPH_CRYPTO_H
>> +
>> +#include <linux/types.h>
>> +
>> +/*
>> + * Header for the crypted file when truncating the size, this
>> + * will be sent to MDS, and the MDS will update the encrypted
>> + * last block and then truncate the size.
>> + */
>> +struct ceph_fscrypt_truncate_size_header {
>> + __u8 ver;
>> + __u8 compat;
>> +
>> + /*
>> + * It will be sizeof(assert_ver + file_offset + block_size)
>> + * if the last block is empty when it's located in a file
>> + * hole. Or the data_len will plus CEPH_FSCRYPT_BLOCK_SIZE.
>> + */
>> + __le32 data_len;
>> +
>> + __le64 assert_ver;
>> + __le64 file_offset;
>> + __le32 block_size;
>> +} __packed;
>> +
>> +#endif
> When I said to move this to crypto.h, I meant fs/ceph/crypto.h. Let's
> not add a new header file for this. Can you move this definition into
> there?
Sure.
BRs
-- Xiubo
>
> Thanks,
^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: [PATCH v6 8/9] ceph: add object version support for sync read
2021-11-04 5:52 ` [PATCH v6 8/9] ceph: add object version support for sync read xiubli
@ 2021-12-07 17:31 ` Jeff Layton
2021-12-08 0:48 ` Xiubo Li
0 siblings, 1 reply; 17+ messages in thread
From: Jeff Layton @ 2021-12-07 17:31 UTC (permalink / raw)
To: xiubli; +Cc: idryomov, vshankar, pdonnell, khiremat, ceph-devel
On Thu, 2021-11-04 at 13:52 +0800, xiubli@redhat.com wrote:
> From: Xiubo Li <xiubli@redhat.com>
>
> The sync read may split the read into several osdc requests, so
> for each it may in different Rados objects.
>
> Signed-off-by: Xiubo Li <xiubli@redhat.com>
> ---
> fs/ceph/file.c | 44 ++++++++++++++++++++++++++++++++++++++++++--
> fs/ceph/super.h | 18 +++++++++++++++++-
> 2 files changed, 59 insertions(+), 3 deletions(-)
>
> diff --git a/fs/ceph/file.c b/fs/ceph/file.c
> index 129f6a642f8e..cedd86a6058d 100644
> --- a/fs/ceph/file.c
> +++ b/fs/ceph/file.c
> @@ -871,7 +871,8 @@ enum {
> * only return a short read to the caller if we hit EOF.
> */
> ssize_t __ceph_sync_read(struct inode *inode, loff_t *ki_pos,
> - struct iov_iter *to, int *retry_op)
> + struct iov_iter *to, int *retry_op,
> + struct ceph_object_vers *objvers)
> {
> struct ceph_inode_info *ci = ceph_inode(inode);
> struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
> @@ -880,6 +881,7 @@ ssize_t __ceph_sync_read(struct inode *inode, loff_t *ki_pos,
> u64 off = *ki_pos;
> u64 len = iov_iter_count(to);
> u64 i_size;
> + u32 object_count = 8;
>
> dout("sync_read on inode %p %llu~%u\n", inode, *ki_pos, (unsigned)len);
>
> @@ -896,6 +898,15 @@ ssize_t __ceph_sync_read(struct inode *inode, loff_t *ki_pos,
> if (ret < 0)
> return ret;
>
> + if (objvers) {
> + objvers->count = 0;
> + objvers->objvers = kcalloc(object_count,
> + sizeof(struct ceph_object_ver),
> + GFP_KERNEL);
> + if (!objvers->objvers)
> + return -ENOMEM;
> + }
> +
> ret = 0;
> while ((len = iov_iter_count(to)) > 0) {
> struct ceph_osd_request *req;
> @@ -938,6 +949,30 @@ ssize_t __ceph_sync_read(struct inode *inode, loff_t *ki_pos,
> req->r_end_latency,
> len, ret);
>
> + if (objvers) {
> + u32 ind = objvers->count;
> +
> + if (objvers->count >= object_count) {
> + int ov_size;
> +
> + object_count *= 2;
> + ov_size = sizeof(struct ceph_object_ver);
> + objvers->objvers = krealloc_array(objvers,
> + object_count,
> + ov_size,
> + GFP_KERNEL);
> + if (!objvers->objvers) {
> + objvers->count = 0;
> + ret = -ENOMEM;
> + break;
> + }
> + }
> +
> + objvers->objvers[ind].offset = off;
> + objvers->objvers[ind].length = len;
> + objvers->objvers[ind].objver = req->r_version;
> + objvers->count++;
> + }
> ceph_osdc_put_request(req);
>
> i_size = i_size_read(inode);
> @@ -995,6 +1030,11 @@ ssize_t __ceph_sync_read(struct inode *inode, loff_t *ki_pos,
> }
>
> dout("sync_read result %zd retry_op %d\n", ret, *retry_op);
> + if (ret < 0 && objvers) {
> + objvers->count = 0;
> + kfree(objvers->objvers);
> + objvers->objvers = NULL;
> + }
> return ret;
> }
>
> @@ -1008,7 +1048,7 @@ static ssize_t ceph_sync_read(struct kiocb *iocb, struct iov_iter *to,
> (unsigned)iov_iter_count(to),
> (file->f_flags & O_DIRECT) ? "O_DIRECT" : "");
>
> - return __ceph_sync_read(inode, &iocb->ki_pos, to, retry_op);
> + return __ceph_sync_read(inode, &iocb->ki_pos, to, retry_op, NULL);
> }
>
> struct ceph_aio_request {
> diff --git a/fs/ceph/super.h b/fs/ceph/super.h
> index 2362d758af97..b347b12e86a9 100644
> --- a/fs/ceph/super.h
> +++ b/fs/ceph/super.h
> @@ -451,6 +451,21 @@ struct ceph_inode_info {
> struct inode vfs_inode; /* at end */
> };
>
> +/*
> + * The version of an object which contains the
> + * file range of [offset, offset + length).
> + */
> +struct ceph_object_ver {
> + u64 offset;
> + u64 length;
> + u64 objver;
> +};
> +
> +struct ceph_object_vers {
> + u32 count;
> + struct ceph_object_ver *objvers;
> +};
> +
> static inline struct ceph_inode_info *
> ceph_inode(const struct inode *inode)
> {
> @@ -1254,7 +1269,8 @@ extern int ceph_open(struct inode *inode, struct file *file);
> extern int ceph_atomic_open(struct inode *dir, struct dentry *dentry,
> struct file *file, unsigned flags, umode_t mode);
> extern ssize_t __ceph_sync_read(struct inode *inode, loff_t *ki_pos,
> - struct iov_iter *to, int *retry_op);
> + struct iov_iter *to, int *retry_op,
> + struct ceph_object_vers *objvers);
>
>
I think this patch is probably overkill. It's not clear to me what you
gain from tracking the offset and length in the object version.
Reads across multiple objects aren't expected to be atomic in cephfs. It
sucks, but RADOS can't reasonably guarantee that sort of consistency. I
don't think you need to track all of the object versions since you only
care about the last object in the read.
Instead, could we just allow __ceph_sync_read to take a pointer to a u64
instead of struct ceph_object_vers * ? Then just have it fill that out
with the version of the last object in the read.
> extern int ceph_release(struct inode *inode, struct file *filp);
> extern void ceph_fill_inline_data(struct inode *inode, struct page *locked_page,
> char *data, size_t len);
--
Jeff Layton <jlayton@kernel.org>
^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: [PATCH v6 8/9] ceph: add object version support for sync read
2021-12-07 17:31 ` Jeff Layton
@ 2021-12-08 0:48 ` Xiubo Li
0 siblings, 0 replies; 17+ messages in thread
From: Xiubo Li @ 2021-12-08 0:48 UTC (permalink / raw)
To: Jeff Layton; +Cc: idryomov, vshankar, pdonnell, khiremat, ceph-devel
On 12/8/21 1:31 AM, Jeff Layton wrote:
> On Thu, 2021-11-04 at 13:52 +0800, xiubli@redhat.com wrote:
>> From: Xiubo Li <xiubli@redhat.com>
>>
>> The sync read may split the read into several osdc requests, so
>> for each it may in different Rados objects.
>>
>> Signed-off-by: Xiubo Li <xiubli@redhat.com>
>> ---
>> fs/ceph/file.c | 44 ++++++++++++++++++++++++++++++++++++++++++--
>> fs/ceph/super.h | 18 +++++++++++++++++-
>> 2 files changed, 59 insertions(+), 3 deletions(-)
>>
>> diff --git a/fs/ceph/file.c b/fs/ceph/file.c
>> index 129f6a642f8e..cedd86a6058d 100644
>> --- a/fs/ceph/file.c
>> +++ b/fs/ceph/file.c
>> @@ -871,7 +871,8 @@ enum {
>> * only return a short read to the caller if we hit EOF.
>> */
>> ssize_t __ceph_sync_read(struct inode *inode, loff_t *ki_pos,
>> - struct iov_iter *to, int *retry_op)
>> + struct iov_iter *to, int *retry_op,
>> + struct ceph_object_vers *objvers)
>> {
>> struct ceph_inode_info *ci = ceph_inode(inode);
>> struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
>> @@ -880,6 +881,7 @@ ssize_t __ceph_sync_read(struct inode *inode, loff_t *ki_pos,
>> u64 off = *ki_pos;
>> u64 len = iov_iter_count(to);
>> u64 i_size;
>> + u32 object_count = 8;
>>
>> dout("sync_read on inode %p %llu~%u\n", inode, *ki_pos, (unsigned)len);
>>
>> @@ -896,6 +898,15 @@ ssize_t __ceph_sync_read(struct inode *inode, loff_t *ki_pos,
>> if (ret < 0)
>> return ret;
>>
>> + if (objvers) {
>> + objvers->count = 0;
>> + objvers->objvers = kcalloc(object_count,
>> + sizeof(struct ceph_object_ver),
>> + GFP_KERNEL);
>> + if (!objvers->objvers)
>> + return -ENOMEM;
>> + }
>> +
>> ret = 0;
>> while ((len = iov_iter_count(to)) > 0) {
>> struct ceph_osd_request *req;
>> @@ -938,6 +949,30 @@ ssize_t __ceph_sync_read(struct inode *inode, loff_t *ki_pos,
>> req->r_end_latency,
>> len, ret);
>>
>> + if (objvers) {
>> + u32 ind = objvers->count;
>> +
>> + if (objvers->count >= object_count) {
>> + int ov_size;
>> +
>> + object_count *= 2;
>> + ov_size = sizeof(struct ceph_object_ver);
>> + objvers->objvers = krealloc_array(objvers,
>> + object_count,
>> + ov_size,
>> + GFP_KERNEL);
>> + if (!objvers->objvers) {
>> + objvers->count = 0;
>> + ret = -ENOMEM;
>> + break;
>> + }
>> + }
>> +
>> + objvers->objvers[ind].offset = off;
>> + objvers->objvers[ind].length = len;
>> + objvers->objvers[ind].objver = req->r_version;
>> + objvers->count++;
>> + }
>> ceph_osdc_put_request(req);
>>
>> i_size = i_size_read(inode);
>> @@ -995,6 +1030,11 @@ ssize_t __ceph_sync_read(struct inode *inode, loff_t *ki_pos,
>> }
>>
>> dout("sync_read result %zd retry_op %d\n", ret, *retry_op);
>> + if (ret < 0 && objvers) {
>> + objvers->count = 0;
>> + kfree(objvers->objvers);
>> + objvers->objvers = NULL;
>> + }
>> return ret;
>> }
>>
>> @@ -1008,7 +1048,7 @@ static ssize_t ceph_sync_read(struct kiocb *iocb, struct iov_iter *to,
>> (unsigned)iov_iter_count(to),
>> (file->f_flags & O_DIRECT) ? "O_DIRECT" : "");
>>
>> - return __ceph_sync_read(inode, &iocb->ki_pos, to, retry_op);
>> + return __ceph_sync_read(inode, &iocb->ki_pos, to, retry_op, NULL);
>> }
>>
>> struct ceph_aio_request {
>> diff --git a/fs/ceph/super.h b/fs/ceph/super.h
>> index 2362d758af97..b347b12e86a9 100644
>> --- a/fs/ceph/super.h
>> +++ b/fs/ceph/super.h
>> @@ -451,6 +451,21 @@ struct ceph_inode_info {
>> struct inode vfs_inode; /* at end */
>> };
>>
>> +/*
>> + * The version of an object which contains the
>> + * file range of [offset, offset + length).
>> + */
>> +struct ceph_object_ver {
>> + u64 offset;
>> + u64 length;
>> + u64 objver;
>> +};
>> +
>> +struct ceph_object_vers {
>> + u32 count;
>> + struct ceph_object_ver *objvers;
>> +};
>> +
>> static inline struct ceph_inode_info *
>> ceph_inode(const struct inode *inode)
>> {
>> @@ -1254,7 +1269,8 @@ extern int ceph_open(struct inode *inode, struct file *file);
>> extern int ceph_atomic_open(struct inode *dir, struct dentry *dentry,
>> struct file *file, unsigned flags, umode_t mode);
>> extern ssize_t __ceph_sync_read(struct inode *inode, loff_t *ki_pos,
>> - struct iov_iter *to, int *retry_op);
>> + struct iov_iter *to, int *retry_op,
>> + struct ceph_object_vers *objvers);
>>
>>
>
> I think this patch is probably overkill. It's not clear to me what you
> gain from tracking the offset and length in the object version.
>
> Reads across multiple objects aren't expected to be atomic in cephfs. It
> sucks, but RADOS can't reasonably guarantee that sort of consistency. I
> don't think you need to track all of the object versions since you only
> care about the last object in the read.
>
> Instead, could we just allow __ceph_sync_read to take a pointer to a u64
> instead of struct ceph_object_vers * ?
The last version was doing like this.
> Then just have it fill that out
> with the version of the last object in the read.
I thought since this will be a general helper and might read more than
one Rados objects and just switched to ceph_object_vers.
I am okay to take a pointer to a u64 to fill the last object's version.
I will send a fixing patch.
BRs
>
>> extern int ceph_release(struct inode *inode, struct file *filp);
>> extern void ceph_fill_inline_data(struct inode *inode, struct page *locked_page,
>> char *data, size_t len);
^ permalink raw reply [flat|nested] 17+ messages in thread
end of thread, other threads:[~2021-12-08 0:48 UTC | newest]
Thread overview: 17+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2021-11-04 5:52 [PATCH v6 0/9] ceph: size handling for the fscrypt xiubli
2021-11-04 5:52 ` [PATCH v6 1/9] libceph: add CEPH_OSD_OP_ASSERT_VER support xiubli
2021-11-04 5:52 ` [PATCH v6 2/9] ceph: size handling for encrypted inodes in cap updates xiubli
2021-11-04 5:52 ` [PATCH v6 3/9] ceph: fscrypt_file field handling in MClientRequest messages xiubli
2021-11-04 5:52 ` [PATCH v6 4/9] ceph: get file size from fscrypt_file when present in inode traces xiubli
2021-11-04 5:52 ` [PATCH v6 5/9] ceph: handle fscrypt fields in cap messages from MDS xiubli
2021-11-04 5:52 ` [PATCH v6 6/9] ceph: add __ceph_get_caps helper support xiubli
2021-11-04 5:52 ` [PATCH v6 7/9] ceph: add __ceph_sync_read " xiubli
2021-11-04 5:52 ` [PATCH v6 8/9] ceph: add object version support for sync read xiubli
2021-12-07 17:31 ` Jeff Layton
2021-12-08 0:48 ` Xiubo Li
2021-11-04 5:52 ` [PATCH v6 9/9] ceph: add truncate size handling support for fscrypt xiubli
2021-11-05 13:36 ` Jeff Layton
2021-11-05 14:11 ` Xiubo Li
2021-11-04 6:00 ` [PATCH v6 0/9] ceph: size handling for the fscrypt Xiubo Li
2021-11-04 10:20 ` Jeff Layton
2021-11-04 10:45 ` Xiubo Li
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.