Skipped unlocks
Subsystems (4)
cover Skipped unlocks
Well over a year ago I made other attempts to fix the problem of NFS failing to send an unlock when signalled. Those attempts were terrible.
Here's another, smaller version that keeps two simple fixes and splits out the behavior for waiting on I/O completion to be different for the v3 and v4 case. For v4, we can easily ensure the unlock is always sent by having the rpc_task perform the wait. For v3, that approach would require significant changes to the way NLM works, so a simpler approach of skipping the wait is used as was suggested by Trond.
I think this is probably a little late for 4.11. Comments and review are welcomed.
Benjamin Coddington (4):
NFS4: remove a redundant lock range check
NFS: Move the flock open mode check into nfs_flock()
locks: Set FL_CLOSE when removing flock locks on close()
NFS: Always wait for I/O completion before unlock
fs/locks.c | 2 +- fs/nfs/file.c | 33 +++++++++++++++++---------------- fs/nfs/nfs3proc.c | 13 +++++++++++++ fs/nfs/nfs4proc.c | 24 +++++++----------------- fs/nfs/pagelist.c | 1 + 5 files changed, 39 insertions(+), 34 deletions(-)
Signature · 1 line
2.9.3
PATCH 1/4 NFS4: remove a redundant lock range check
flock64_to_posix_lock() is already doing this check
--- fs/nfs/nfs4proc.c | 3 --- 1 file changed, 3 deletions(-)
fs/nfs/nfs4proc.c +0 -3
@@ -6570,9 +6570,6 @@ nfs4_proc_lock(struct file *filp, int cmd, struct file_lock *request) ctx =nfs_file_open_context (filp); state = ctx->state; -if (request->fl_start <0 || request->fl_end <0 ) -return -EINVAL ; -if (IS_GETLK (cmd)) {if (state !=NULL )return nfs4_proc_getlk (state,F_GETLK , request);
Signature · 1 line
2.9.3
PATCH 2/4 NFS: Move the flock open mode check into nfs_flock()
In reply to Benjamin CoddingtonWe only need to check lock exclusive/shared types against open mode when flock() is used on NFS, so move it into the flock-specific path instead of checking it for all locks.
--- fs/nfs/file.c | 20 +++++++++++++++++--- fs/nfs/nfs4proc.c | 14 -------------- 2 files changed, 17 insertions(+), 17 deletions(-)
fs/nfs/file.c +17 -3
@@ -820,9 +820,23 @@ int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)if (NFS_SERVER (inode)->flags &NFS_MOUNT_LOCAL_FLOCK ) is_local =1 ; -/* We're simulating flock() locks using posix locks on the server */ -if (fl->fl_type ==F_UNLCK ) -return do_unlk (filp, cmd, fl, is_local); +/* +* VFS doesn't require the open mode to match a flock() lock's type. +* NFS, however, may simulate flock() locking with posix locking which +* requires the open mode to match the lock type. +*/ +switch (fl->fl_type) { +case F_UNLCK : +return do_unlk (filp, cmd, fl, is_local); +case F_RDLCK : +if (!(filp->f_mode &FMODE_READ )) +return -EBADF ; +break ; +case F_WRLCK : +if (!(filp->f_mode &FMODE_WRITE )) +return -EBADF ; + } +return do_setlk (filp, cmd, fl, is_local); }EXPORT_SYMBOL_GPL (nfs_flock);
fs/nfs/nfs4proc.c +0 -14
@@ -6592,20 +6592,6 @@ nfs4_proc_lock(struct file *filp, int cmd, struct file_lock *request) !test_bit(NFS_STATE_POSIX_LOCKS, &state->flags))return -ENOLCK ; -/* -* Don't rely on the VFS having checked the file open mode, -* since it won't do this for flock() locks. -*/ -switch (request->fl_type) { -case F_RDLCK : -if (!(filp->f_mode &FMODE_READ )) -return -EBADF ; -break ; -case F_WRLCK : -if (!(filp->f_mode &FMODE_WRITE )) -return -EBADF ; - } - status =nfs4_set_lock_state (state, request);if (status !=0 )return status;
Signature · 1 line
2.9.3
PATCH 3/4 locks: Set FL_CLOSE when removing flock locks on close()
In reply to Benjamin CoddingtonSet FL_CLOSE in fl_flags as in locks_remove_posix() when clearing locks. NFS will check for this flag to ensure an unlock is sent in a following patch.
--- fs/locks.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
fs/locks.c +1 -1
@@ -2504,7 +2504,7 @@ locks_remove_flock(struct file *filp, struct file_lock_context *flctx) .fl_owner = filp, .fl_pid = current->tgid, .fl_file = filp, - .fl_flags = FL_FLOCK, + .fl_flags = FL_FLOCK | FL_CLOSE, .fl_type = F_UNLCK, .fl_end = OFFSET_MAX, };
Signature · 1 line
2.9.3
PATCH 4/4 NFS: Always wait for I/O completion before unlock
In reply to Benjamin CoddingtonNFS attempts to wait for read and write completion before unlocking in order to ensure that the data returned was protected by the lock. When this waiting is interrupted by a signal, the unlock may never be sent, and messages similar to the following are seen in the kernel ring buffer:
[20.167876] Leaked locks on dev=0x0:0x2b ino=0x8dd4c3: [20.168286] POSIX: fl_owner=ffff880078b06940 fl_flags=0x1 fl_type=0x0 fl_pid=20183 [20.168727] POSIX: fl_owner=ffff880078b06680 fl_flags=0x1 fl_type=0x0 fl_pid=20185
For NFSv3, the missing unlock will cause the server to refuse conflicting locks indefinitely. For NFSv4, the leftover lock will be removed by the server after the lease timeout.
This patch fixes this for NFSv3 by skipping the wait in order to immediately send the unlock if the FL_CLOSE flag is set when signaled. For NFSv4, this approach may cause the server to see the I/O as arriving with an old stateid, so, for the v4 case the fix is different: the wait on I/O completion is moved into nfs4_locku_ops' rpc_call_prepare(). This will cause the sleep to happen in rpciod context, and a signal to the originally waiting process will not cause the unlock to be skipped.
--- fs/nfs/file.c | 13 ------------- fs/nfs/nfs3proc.c | 13 +++++++++++++ fs/nfs/nfs4proc.c | 7 +++++++ fs/nfs/pagelist.c | 1 + 4 files changed, 21 insertions(+), 13 deletions(-)
fs/nfs/file.c +0 -13
@@ -684,7 +684,6 @@ static int do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local) { struct inode *inode = filp->f_mapping->host; -struct nfs_lock_context *l_ctx;int status;/*
@@ -693,18 +692,6 @@ do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)*/ vfs_fsync (filp,0 ); - l_ctx =nfs_get_lock_context (nfs_file_open_context (filp)); -if (!IS_ERR (l_ctx)) { - status =nfs_iocounter_wait (l_ctx); -nfs_put_lock_context (l_ctx); -if (status <0 ) -return status; - } - -/* NOTE: special case -* If we're signalled while cleaning up locks on process exit, we -* still need to complete the unlock. -*/ /* * Use local locking if mounted with "-onolock" or with appropriate * "-olocal_lock="
fs/nfs/nfs3proc.c +13 -0
@@ -869,6 +869,19 @@ static int nfs3_proc_lock(struct file *filp, int cmd, struct file_lock *fl) { struct inode *inode = file_inode(filp); +int status; +struct nfs_lock_context *l_ctx; + +/* For v3, always send the unlock on FL_CLOSE */ +if (fl->fl_type ==F_UNLCK ) { + l_ctx =nfs_get_lock_context (nfs_file_open_context (filp)); +if (!IS_ERR (l_ctx)) { + status =nfs_iocounter_wait (l_ctx); +nfs_put_lock_context (l_ctx); +if (status <0 && !(fl->fl_flags &FL_CLOSE )) +return status; + } + }return nlmclnt_proc (NFS_SERVER (inode)->nlm_host, cmd, fl); }
fs/nfs/nfs4proc.c +7 -0
@@ -5906,6 +5906,7 @@ struct nfs4_unlockdata {struct nfs_locku_res res;struct nfs4_lock_state *lsp;struct nfs_open_context *ctx; +struct nfs_lock_context *l_ctx;struct file_lock fl;struct nfs_server *server;unsigned long timestamp;
@@ -5930,6 +5931,7 @@ static struct nfs4_unlockdata *nfs4_alloc_unlockdata(struct file_lock *fl,atomic_inc (&lsp->ls_count);/* Ensure we don't close file until we're done freeing locks! */ p->ctx =get_nfs_open_context (ctx); + p->l_ctx =nfs_get_lock_context (ctx);memcpy (&p->fl, fl,sizeof (p->fl)); p->server =NFS_SERVER (inode);return p;
@@ -5988,6 +5990,11 @@ static void nfs4_locku_prepare(struct rpc_task *task, void *data)/* Note: exit _without_ running nfs4_locku_done */ goto out_no_action; } + +if (!IS_ERR (calldata->l_ctx)) { +nfs_iocounter_wait (calldata->l_ctx); +nfs_put_lock_context (calldata->l_ctx); + } calldata->timestamp = jiffies; if (nfs4_setup_sequence(calldata->server, &calldata->arg.seq_args,
fs/nfs/pagelist.c +1 -0
@@ -114,6 +114,7 @@ nfs_iocounter_wait(struct nfs_lock_context *l_ctx)return wait_on_atomic_t (&l_ctx->io_count, nfs_wait_atomic_killable,TASK_KILLABLE ); } +EXPORT_SYMBOL (nfs_iocounter_wait);/* * nfs_page_group_lock - lock the head of the page group
Signature · 1 line
2.9.3