Skip to content

Commit

Permalink
ceph: handle 'session get evicted while there are file locks'
Browse files Browse the repository at this point in the history 
When session get evicted, all file locks associated with the session
get released remotely by mds. File locks tracked by kernel become
stale. In this situation, set an error flag on inode. The flag makes
further file locks return -EIO.

Another option to handle this situation is cleanup file locks tracked
kernel. I do not choose it because it is inconvenient to notify user
program about the error.

Signed-off-by: "Yan, Zheng" <[email protected]>
Acked-by: Jeff Layton <[email protected]>
Signed-off-by: Ilya Dryomov <[email protected]>
  • Loading branch information
@ukernel @idryomov
ukernel authored and idryomov committed Nov 13, 2017
1 parent 4deb14a commit b3f8d68
Show file tree
Hide file tree
Showing 3 changed files with 58 additions and 17 deletions.
52 changes: 40 additions & 12 deletions fs/ceph/locks.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -39,7 +39,13 @@ static void ceph_fl_copy_lock(struct file_lock *dst, struct file_lock *src)
static void ceph_fl_release_lock(struct file_lock *fl)
{
struct inode *inode = file_inode(fl->fl_file);
atomic_dec(&ceph_inode(inode)->i_filelock_ref);
struct ceph_inode_info *ci = ceph_inode(inode);
if (atomic_dec_and_test(&ci->i_filelock_ref)) {
/* clear error when all locks are released */
spin_lock(&ci->i_ceph_lock);
ci->i_ceph_flags &= ~CEPH_I_ERROR_FILELOCK;
spin_unlock(&ci->i_ceph_lock);
}
}

static const struct file_lock_operations ceph_fl_lock_ops = {
Expand Down Expand Up @@ -208,10 +214,11 @@ static int ceph_lock_wait_for_completion(struct ceph_mds_client *mdsc,
int ceph_lock(struct file *file, int cmd, struct file_lock *fl)
{
struct inode *inode = file_inode(file);
int err;
struct ceph_inode_info *ci = ceph_inode(inode);
int err = 0;
u16 op = CEPH_MDS_OP_SETFILELOCK;
u8 lock_cmd;
u8 wait = 0;
u8 lock_cmd;

if (!(fl->fl_flags & FL_POSIX))
return -ENOLCK;
Expand All @@ -227,15 +234,24 @@ int ceph_lock(struct file *file, int cmd, struct file_lock *fl)
else if (IS_SETLKW(cmd))
wait = 1;

if (op == CEPH_MDS_OP_SETFILELOCK) {
spin_lock(&ci->i_ceph_lock);
if (ci->i_ceph_flags & CEPH_I_ERROR_FILELOCK) {
err = -EIO;
} else if (op == CEPH_MDS_OP_SETFILELOCK) {
/*
* increasing i_filelock_ref closes race window between
* handling request reply and adding file_lock struct to
* inode. Otherwise, i_auth_cap may get trimmed in the
* window. Caller function will decrease the counter.
*/
fl->fl_ops = &ceph_fl_lock_ops;
atomic_inc(&ceph_inode(inode)->i_filelock_ref);
atomic_inc(&ci->i_filelock_ref);
}
spin_unlock(&ci->i_ceph_lock);
if (err < 0) {
if (op == CEPH_MDS_OP_SETFILELOCK && F_UNLCK == fl->fl_type)
posix_lock_file(file, fl, NULL);
return err;
}

if (F_RDLCK == fl->fl_type)
Expand All @@ -247,10 +263,10 @@ int ceph_lock(struct file *file, int cmd, struct file_lock *fl)

err = ceph_lock_message(CEPH_LOCK_FCNTL, op, inode, lock_cmd, wait, fl);
if (!err) {
if (op != CEPH_MDS_OP_GETFILELOCK) {
if (op == CEPH_MDS_OP_SETFILELOCK) {
dout("mds locked, locking locally");
err = posix_lock_file(file, fl, NULL);
if (err && (CEPH_MDS_OP_SETFILELOCK == op)) {
if (err) {
/* undo! This should only happen if
* the kernel detects local
* deadlock. */
Expand All @@ -267,9 +283,10 @@ int ceph_lock(struct file *file, int cmd, struct file_lock *fl)
int ceph_flock(struct file *file, int cmd, struct file_lock *fl)
{
struct inode *inode = file_inode(file);
int err;
u8 lock_cmd;
struct ceph_inode_info *ci = ceph_inode(inode);
int err = 0;
u8 wait = 0;
u8 lock_cmd;

if (!(fl->fl_flags & FL_FLOCK))
return -ENOLCK;
Expand All @@ -279,9 +296,20 @@ int ceph_flock(struct file *file, int cmd, struct file_lock *fl)

dout("ceph_flock, fl_file: %p", fl->fl_file);

/* see comment in ceph_lock */
fl->fl_ops = &ceph_fl_lock_ops;
atomic_inc(&ceph_inode(inode)->i_filelock_ref);
spin_lock(&ci->i_ceph_lock);
if (ci->i_ceph_flags & CEPH_I_ERROR_FILELOCK) {
err = -EIO;
} else {
/* see comment in ceph_lock */
fl->fl_ops = &ceph_fl_lock_ops;
atomic_inc(&ci->i_filelock_ref);
}
spin_unlock(&ci->i_ceph_lock);
if (err < 0) {
if (F_UNLCK == fl->fl_type)
locks_lock_file_wait(file, fl);
return err;
}

if (IS_SETLKW(cmd))
wait = 1;
Expand Down
21 changes: 16 additions & 5 deletions fs/ceph/mds_client.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -1215,6 +1215,13 @@ static int remove_session_caps_cb(struct inode *inode, struct ceph_cap *cap,
}
spin_unlock(&mdsc->cap_dirty_lock);

if (atomic_read(&ci->i_filelock_ref) > 0) {
/* make further file lock syscall return -EIO */
ci->i_ceph_flags |= CEPH_I_ERROR_FILELOCK;
pr_warn_ratelimited(" dropping file locks for %p %lld\n",
inode, ceph_ino(inode));
}

if (!ci->i_dirty_caps && ci->i_prealloc_cap_flush) {
list_add(&ci->i_prealloc_cap_flush->i_list, &to_remove);
ci->i_prealloc_cap_flush = NULL;
Expand Down Expand Up @@ -2832,7 +2839,7 @@ static int encode_caps_cb(struct inode *inode, struct ceph_cap *cap,
struct ceph_mds_cap_reconnect v2;
struct ceph_mds_cap_reconnect_v1 v1;
} rec;
struct ceph_inode_info *ci;
struct ceph_inode_info *ci = cap->ci;
struct ceph_reconnect_state *recon_state = arg;
struct ceph_pagelist *pagelist = recon_state->pagelist;
char *path;
Expand All @@ -2841,8 +2848,6 @@ static int encode_caps_cb(struct inode *inode, struct ceph_cap *cap,
u64 snap_follows;
struct dentry *dentry;

ci = cap->ci;

dout(" adding %p ino %llx.%llx cap %p %lld %s\n",
inode, ceph_vinop(inode), cap, cap->cap_id,
ceph_cap_string(cap->issued));
Expand Down Expand Up @@ -2875,7 +2880,8 @@ static int encode_caps_cb(struct inode *inode, struct ceph_cap *cap,
rec.v2.issued = cpu_to_le32(cap->issued);
rec.v2.snaprealm = cpu_to_le64(ci->i_snap_realm->ino);
rec.v2.pathbase = cpu_to_le64(pathbase);
rec.v2.flock_len = 0;
rec.v2.flock_len =
(ci->i_ceph_flags & CEPH_I_ERROR_FILELOCK) ? 0 : 1;
} else {
rec.v1.cap_id = cpu_to_le64(cap->cap_id);
rec.v1.wanted = cpu_to_le32(__ceph_caps_wanted(ci));
Expand Down Expand Up @@ -2904,7 +2910,12 @@ static int encode_caps_cb(struct inode *inode, struct ceph_cap *cap,
u8 struct_v = 0;

encode_again:
ceph_count_locks(inode, &num_fcntl_locks, &num_flock_locks);
if (rec.v2.flock_len) {
ceph_count_locks(inode, &num_fcntl_locks, &num_flock_locks);
} else {
num_fcntl_locks = 0;
num_flock_locks = 0;
}
if (num_fcntl_locks + num_flock_locks > 0) {
flocks = kmalloc((num_fcntl_locks + num_flock_locks) *
sizeof(struct ceph_filelock), GFP_NOFS);
Expand Down
2 changes: 2 additions & 0 deletions fs/ceph/super.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -488,6 +488,8 @@ static inline struct inode *ceph_find_inode(struct super_block *sb,
#define CEPH_I_KICK_FLUSH (1 << 9) /* kick flushing caps */
#define CEPH_I_FLUSH_SNAPS (1 << 10) /* need flush snapss */
#define CEPH_I_ERROR_WRITE (1 << 11) /* have seen write errors */
#define CEPH_I_ERROR_FILELOCK (1 << 12) /* have seen file lock errors */


/*
* We set the ERROR_WRITE bit when we start seeing write errors on an inode
Expand Down

0 comments on commit b3f8d68

Please sign in to comment.