diff mbox

[2/2] NFS: Send SIGIO on lost locks

Message ID 1310407110-20600-2-git-send-email-bjschuma@netapp.com (mailing list archive)
State New, archived
Headers show

Commit Message

Bryan Schumaker July 11, 2011, 5:58 p.m. UTC
From: Bryan Schumaker <bjschuma@netapp.com>

If the client loses a lock, we send SIGIO to the application to notify
it.  The application can then handle the error from there.

Signed-off-by: Bryan Schumaker <bjschuma@netapp.com>
---
 Documentation/kernel-parameters.txt |    7 +++++++
 fs/nfs/nfs4_fs.h                    |    2 ++
 fs/nfs/nfs4proc.c                   |   13 +++++++++++++
 fs/nfs/nfs4state.c                  |   14 ++++++++++++++
 fs/nfs/write.c                      |    7 +++++++
 5 files changed, 43 insertions(+), 0 deletions(-)

Comments

Jeff Layton July 11, 2011, 8:08 p.m. UTC | #1
On Mon, 11 Jul 2011 13:58:30 -0400
bjschuma@netapp.com wrote:

> From: Bryan Schumaker <bjschuma@netapp.com>
> 
> If the client loses a lock, we send SIGIO to the application to notify
> it.  The application can then handle the error from there.
> 
> Signed-off-by: Bryan Schumaker <bjschuma@netapp.com>

Would SIGLOST be a better choice? Linux hasn't supported that
historically, but we could add it.

> ---
>  Documentation/kernel-parameters.txt |    7 +++++++
>  fs/nfs/nfs4_fs.h                    |    2 ++
>  fs/nfs/nfs4proc.c                   |   13 +++++++++++++
>  fs/nfs/nfs4state.c                  |   14 ++++++++++++++
>  fs/nfs/write.c                      |    7 +++++++
>  5 files changed, 43 insertions(+), 0 deletions(-)
> 
> diff --git a/Documentation/kernel-parameters.txt b/Documentation/kernel-parameters.txt
> index fd248a31..988ebdc 100644
> --- a/Documentation/kernel-parameters.txt
> +++ b/Documentation/kernel-parameters.txt
> @@ -1593,6 +1593,13 @@ bytes respectively. Such letter suffixes can also be entirely omitted.
>  			of returning the full 64-bit number.
>  			The default is to return 64-bit inode numbers.
>  
> +	nfs.enable_sigio=
> +			[NFSv4.1] enable sending SIGIO to applications.
> +			If true, the NFS client will send SIGIO to applications
> +			when a lost file lock is detected.  If false, the client
> +			will attempt to recover the lock.
> +			The default is to send SIGIO.
> +
>  	nfs.nfs4_disable_idmapping=
>  			[NFSv4] When set, this option disables the NFSv4
>  			idmapper on the client, but only if the mount
> diff --git a/fs/nfs/nfs4_fs.h b/fs/nfs/nfs4_fs.h
> index c30aed2..8fa02af 100644
> --- a/fs/nfs/nfs4_fs.h
> +++ b/fs/nfs/nfs4_fs.h
> @@ -163,6 +163,7 @@ struct nfs4_lock_state {
>  	struct list_head	ls_locks;	/* Other lock stateids */
>  	struct nfs4_state *	ls_state;	/* Pointer to open state */
>  #define NFS_LOCK_INITIALIZED 1
> +#define NFS_LOCK_INVALID     2
>  	int			ls_flags;
>  	struct nfs_seqid_counter	ls_seqid;
>  	struct rpc_sequence	ls_sequence;
> @@ -346,6 +347,7 @@ extern struct nfs4_state * nfs4_get_open_state(struct inode *, struct nfs4_state
>  extern void nfs4_put_open_state(struct nfs4_state *);
>  extern void nfs4_close_state(struct path *, struct nfs4_state *, fmode_t);
>  extern void nfs4_close_sync(struct path *, struct nfs4_state *, fmode_t);
> +extern int nfs4_validate_lock_stateid(struct nfs4_state *, struct nfs_lock_context *);
>  extern void nfs4_state_set_mode_locked(struct nfs4_state *, fmode_t);
>  extern void nfs4_schedule_lease_recovery(struct nfs_client *);
>  extern void nfs4_schedule_state_manager(struct nfs_client *);
> diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c
> index f37218b..8757dbd 100644
> --- a/fs/nfs/nfs4proc.c
> +++ b/fs/nfs/nfs4proc.c
> @@ -4462,6 +4462,8 @@ out:
>  }
>  
>  #if defined(CONFIG_NFS_V4_1)
> +bool enable_sigio = true;
> +
>  static int nfs41_lock_expired(struct nfs4_state *state, struct file_lock *request)
>  {
>  	int status;
> @@ -4472,10 +4474,21 @@ static int nfs41_lock_expired(struct nfs4_state *state, struct file_lock *reques
>  	status = nfs41_test_stateid(server, stateid);
>  	if (status == NFS_OK)
>  		return 0;
> +
>  	nfs41_free_stateid(server, stateid);
>  	lock_state->ls_seqid.flags &= !NFS_SEQID_CONFIRMED;
> +
> +	if (enable_sigio) {
> +		lock_state->ls_flags |= NFS_LOCK_INVALID;
> +		kill_pid(request->fl_nspid, SIGIO, 1);
> +		return 0;
> +	}
>  	return nfs4_lock_expired(state, request);
>  }
> +
> +module_param(enable_sigio, bool, 0644);
> +MODULE_PARM_DESC(enable_sigio, "Send SIGIO to an application when a lost lock"
> +			"is detected instead of attempting recovery");
>  #endif
>  
>  static int _nfs4_proc_setlk(struct nfs4_state *state, int cmd, struct file_lock *request)
> diff --git a/fs/nfs/nfs4state.c b/fs/nfs/nfs4state.c
> index 5d744a5..101d9b4 100644
> --- a/fs/nfs/nfs4state.c
> +++ b/fs/nfs/nfs4state.c
> @@ -725,6 +725,20 @@ __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner, pid_t fl_p
>  	return NULL;
>  }
>  
> +int nfs4_validate_lock_stateid(struct nfs4_state *state, struct nfs_lock_context *lock_ctx)
> +{
> +	int valid = 1;
> +	struct nfs4_lock_state *lock_state;
> +
> +	spin_lock(&state->state_lock);
> +	lock_state = __nfs4_find_lock_state(state, lock_ctx->lockowner, lock_ctx->pid, NFS4_ANY_LOCK_TYPE);
> +	if (lock_state != NULL)
> +		valid = (lock_state->ls_flags & NFS_LOCK_INVALID) == 0;
> +	spin_unlock(&state->state_lock);
> +	nfs4_put_lock_state(lock_state);
> +	return valid;
> +}
> +
>  /*
>   * Return a compatible lock_state. If no initialized lock_state structure
>   * exists, return an uninitialized one.
> diff --git a/fs/nfs/write.c b/fs/nfs/write.c
> index 7f8732e..49ca975 100644
> --- a/fs/nfs/write.c
> +++ b/fs/nfs/write.c
> @@ -1114,7 +1114,14 @@ out:
>  void nfs_write_prepare(struct rpc_task *task, void *calldata)
>  {
>  	struct nfs_write_data *data = calldata;
> +	struct nfs4_state *state = data->args.context->state;
> +	struct nfs_lock_context *lock_context = data->args.lock_context;
>  
> +	if (!nfs4_validate_lock_stateid(state, lock_context)) {
> +		task->tk_status = -EIO;
> +		task->tk_action = NULL;
> +		return;
> +	}
>  	if (nfs4_setup_sequence(NFS_SERVER(data->inode),
>  				&data->args.seq_args,
>  				&data->res.seq_res, 1, task))
Trond Myklebust July 11, 2011, 8:12 p.m. UTC | #2
On Mon, 2011-07-11 at 16:08 -0400, Jeff Layton wrote: 
> On Mon, 11 Jul 2011 13:58:30 -0400
> bjschuma@netapp.com wrote:
> 
> > From: Bryan Schumaker <bjschuma@netapp.com>
> > 
> > If the client loses a lock, we send SIGIO to the application to notify
> > it.  The application can then handle the error from there.
> > 
> > Signed-off-by: Bryan Schumaker <bjschuma@netapp.com>
> 
> Would SIGLOST be a better choice? Linux hasn't supported that
> historically, but we could add it.

SIGLOST is 'defined' in the kernel as follows:

#define SIGIO           29
#define SIGPOLL         SIGIO
/*
#define SIGLOST         29
*/

IOW: it is synonymous with SIGPOLL and SIGIO. This explains Bryan's
choice.

Cheers
  Trond
Jeff Layton July 12, 2011, 12:47 a.m. UTC | #3
On Mon, 11 Jul 2011 16:12:24 -0400
Trond Myklebust <Trond.Myklebust@netapp.com> wrote:

> On Mon, 2011-07-11 at 16:08 -0400, Jeff Layton wrote: 
> > On Mon, 11 Jul 2011 13:58:30 -0400
> > bjschuma@netapp.com wrote:
> > 
> > > From: Bryan Schumaker <bjschuma@netapp.com>
> > > 
> > > If the client loses a lock, we send SIGIO to the application to notify
> > > it.  The application can then handle the error from there.
> > > 
> > > Signed-off-by: Bryan Schumaker <bjschuma@netapp.com>
> > 
> > Would SIGLOST be a better choice? Linux hasn't supported that
> > historically, but we could add it.
> 
> SIGLOST is 'defined' in the kernel as follows:
> 
> #define SIGIO           29
> #define SIGPOLL         SIGIO
> /*
> #define SIGLOST         29
> */
> 
> IOW: it is synonymous with SIGPOLL and SIGIO. This explains Bryan's
> choice.
> 
> Cheers
>   Trond
> 

Right. I just wonder whether we'd be better off making this a distinct
signal with its own number. It seems like there would be value in being
able to distinguish between SIGLOST and SIGIO.
Trond Myklebust July 12, 2011, 2:31 a.m. UTC | #4
On Mon, 2011-07-11 at 20:47 -0400, Jeff Layton wrote: 
> On Mon, 11 Jul 2011 16:12:24 -0400
> Trond Myklebust <Trond.Myklebust@netapp.com> wrote:
> 
> > On Mon, 2011-07-11 at 16:08 -0400, Jeff Layton wrote: 
> > > On Mon, 11 Jul 2011 13:58:30 -0400
> > > bjschuma@netapp.com wrote:
> > > 
> > > > From: Bryan Schumaker <bjschuma@netapp.com>
> > > > 
> > > > If the client loses a lock, we send SIGIO to the application to notify
> > > > it.  The application can then handle the error from there.
> > > > 
> > > > Signed-off-by: Bryan Schumaker <bjschuma@netapp.com>
> > > 
> > > Would SIGLOST be a better choice? Linux hasn't supported that
> > > historically, but we could add it.
> > 
> > SIGLOST is 'defined' in the kernel as follows:
> > 
> > #define SIGIO           29
> > #define SIGPOLL         SIGIO
> > /*
> > #define SIGLOST         29
> > */
> > 
> > IOW: it is synonymous with SIGPOLL and SIGIO. This explains Bryan's
> > choice.
> > 
> > Cheers
> >   Trond
> > 
> 
> Right. I just wonder whether we'd be better off making this a distinct
> signal with its own number. It seems like there would be value in being
> able to distinguish between SIGLOST and SIGIO.

Actually, I think there is value in distinguishing between SIGIO and
SIGPOLL too, but it would appear that it is a bit late to do that.

Note that we can (and probably should) use a siginfo structure to pass a
bit more info to the application. That would mean filling out a siginfo
and then replacing kill_pid() with kill_pid_info(), but that should be
easily doable and can be made to resolve the above ambiguities.

Cheers
  Trond
diff mbox

Patch

diff --git a/Documentation/kernel-parameters.txt b/Documentation/kernel-parameters.txt
index fd248a31..988ebdc 100644
--- a/Documentation/kernel-parameters.txt
+++ b/Documentation/kernel-parameters.txt
@@ -1593,6 +1593,13 @@  bytes respectively. Such letter suffixes can also be entirely omitted.
 			of returning the full 64-bit number.
 			The default is to return 64-bit inode numbers.
 
+	nfs.enable_sigio=
+			[NFSv4.1] enable sending SIGIO to applications.
+			If true, the NFS client will send SIGIO to applications
+			when a lost file lock is detected.  If false, the client
+			will attempt to recover the lock.
+			The default is to send SIGIO.
+
 	nfs.nfs4_disable_idmapping=
 			[NFSv4] When set, this option disables the NFSv4
 			idmapper on the client, but only if the mount
diff --git a/fs/nfs/nfs4_fs.h b/fs/nfs/nfs4_fs.h
index c30aed2..8fa02af 100644
--- a/fs/nfs/nfs4_fs.h
+++ b/fs/nfs/nfs4_fs.h
@@ -163,6 +163,7 @@  struct nfs4_lock_state {
 	struct list_head	ls_locks;	/* Other lock stateids */
 	struct nfs4_state *	ls_state;	/* Pointer to open state */
 #define NFS_LOCK_INITIALIZED 1
+#define NFS_LOCK_INVALID     2
 	int			ls_flags;
 	struct nfs_seqid_counter	ls_seqid;
 	struct rpc_sequence	ls_sequence;
@@ -346,6 +347,7 @@  extern struct nfs4_state * nfs4_get_open_state(struct inode *, struct nfs4_state
 extern void nfs4_put_open_state(struct nfs4_state *);
 extern void nfs4_close_state(struct path *, struct nfs4_state *, fmode_t);
 extern void nfs4_close_sync(struct path *, struct nfs4_state *, fmode_t);
+extern int nfs4_validate_lock_stateid(struct nfs4_state *, struct nfs_lock_context *);
 extern void nfs4_state_set_mode_locked(struct nfs4_state *, fmode_t);
 extern void nfs4_schedule_lease_recovery(struct nfs_client *);
 extern void nfs4_schedule_state_manager(struct nfs_client *);
diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c
index f37218b..8757dbd 100644
--- a/fs/nfs/nfs4proc.c
+++ b/fs/nfs/nfs4proc.c
@@ -4462,6 +4462,8 @@  out:
 }
 
 #if defined(CONFIG_NFS_V4_1)
+bool enable_sigio = true;
+
 static int nfs41_lock_expired(struct nfs4_state *state, struct file_lock *request)
 {
 	int status;
@@ -4472,10 +4474,21 @@  static int nfs41_lock_expired(struct nfs4_state *state, struct file_lock *reques
 	status = nfs41_test_stateid(server, stateid);
 	if (status == NFS_OK)
 		return 0;
+
 	nfs41_free_stateid(server, stateid);
 	lock_state->ls_seqid.flags &= !NFS_SEQID_CONFIRMED;
+
+	if (enable_sigio) {
+		lock_state->ls_flags |= NFS_LOCK_INVALID;
+		kill_pid(request->fl_nspid, SIGIO, 1);
+		return 0;
+	}
 	return nfs4_lock_expired(state, request);
 }
+
+module_param(enable_sigio, bool, 0644);
+MODULE_PARM_DESC(enable_sigio, "Send SIGIO to an application when a lost lock"
+			"is detected instead of attempting recovery");
 #endif
 
 static int _nfs4_proc_setlk(struct nfs4_state *state, int cmd, struct file_lock *request)
diff --git a/fs/nfs/nfs4state.c b/fs/nfs/nfs4state.c
index 5d744a5..101d9b4 100644
--- a/fs/nfs/nfs4state.c
+++ b/fs/nfs/nfs4state.c
@@ -725,6 +725,20 @@  __nfs4_find_lock_state(struct nfs4_state *state, fl_owner_t fl_owner, pid_t fl_p
 	return NULL;
 }
 
+int nfs4_validate_lock_stateid(struct nfs4_state *state, struct nfs_lock_context *lock_ctx)
+{
+	int valid = 1;
+	struct nfs4_lock_state *lock_state;
+
+	spin_lock(&state->state_lock);
+	lock_state = __nfs4_find_lock_state(state, lock_ctx->lockowner, lock_ctx->pid, NFS4_ANY_LOCK_TYPE);
+	if (lock_state != NULL)
+		valid = (lock_state->ls_flags & NFS_LOCK_INVALID) == 0;
+	spin_unlock(&state->state_lock);
+	nfs4_put_lock_state(lock_state);
+	return valid;
+}
+
 /*
  * Return a compatible lock_state. If no initialized lock_state structure
  * exists, return an uninitialized one.
diff --git a/fs/nfs/write.c b/fs/nfs/write.c
index 7f8732e..49ca975 100644
--- a/fs/nfs/write.c
+++ b/fs/nfs/write.c
@@ -1114,7 +1114,14 @@  out:
 void nfs_write_prepare(struct rpc_task *task, void *calldata)
 {
 	struct nfs_write_data *data = calldata;
+	struct nfs4_state *state = data->args.context->state;
+	struct nfs_lock_context *lock_context = data->args.lock_context;
 
+	if (!nfs4_validate_lock_stateid(state, lock_context)) {
+		task->tk_status = -EIO;
+		task->tk_action = NULL;
+		return;
+	}
 	if (nfs4_setup_sequence(NFS_SERVER(data->inode),
 				&data->args.seq_args,
 				&data->res.seq_res, 1, task))