Skip to contentSkip to conversation

aio: Do not return ERESTARTSYS as a result of AIO (v2)

PATCH In review Jan Kara 2 participants 4 messages Sep 9, 2010 – Sep 10, 2010

Subsystems (2)
lore
Share & tools
lore

Jan Kara

author reviewer
#0 lore raw

PATCH 1/1 aio: Do not return ERESTARTSYS as a result of AIO (v2)

OCFS2 can return ERESTARTSYS from its write function when the process is signalled while waiting for a cluster lock (and the filesystem is mounted with intr mount option). Generally, it seems reasonable to allow filesystems to return this error code from its IO functions. As we must not leak ERESTARTSYS (and similar error codes) to userspace as a result of an AIO operation, we have to properly convert it to EINTR inside AIO code (restarting the syscall isn't really an option because other AIO could have been already submitted by the same io_submit syscall).

---
 fs/aio.c |   10 +++++++++-
 1 files changed, 9 insertions(+), 1 deletions(-)

Andrew, I'm not sure what happened with the old version of the patch (I had
a feeling you carried it in -mm but I cannot see it there now). Anyway, here's a new version with better changelog as several people raised questions about the previous one. Please consider merging it. Thanks.

fs/aio.c +9 -1
index 3006b5b..b73c9b5 100644
@@ -712,8 +712,16 @@ static ssize_t aio_run_iocb(struct kiocb *iocb)
 	 */
 	ret = retry(iocb);
 
-	if (ret != -EIOCBRETRY && ret != -EIOCBQUEUED)
+	if (ret != -EIOCBRETRY && ret != -EIOCBQUEUED) {
+		/*
+		 * There's no easy way to restart the syscall since other AIO's
+		 * may be already running. Just fail this IO with EINTR.
+		 */
+		if (unlikely(ret == -ERESTARTSYS || ret == -ERESTARTNOINTR ||
+			     ret == -ERESTARTNOHAND || ret == -ERESTART_RESTARTBLOCK))
+			ret = -EINTR;
 		aio_complete(iocb, ret, 0);
+	}
 out:
 	spin_lock_irq(&ctx->ctx_lock);
 
Signature · 1 line
1.6.4.2
Reviewed-by on v1 1/1
#1 lore raw
Quoted 8 lines · Jan Kara

OCFS2 can return ERESTARTSYS from its write function when the process is signalled while waiting for a cluster lock (and the filesystem is mounted with intr mount option). Generally, it seems reasonable to allow filesystems to return this error code from its IO functions. As we must not leak ERESTARTSYS (and similar error codes) to userspace as a result of an AIO operation, we have to properly convert it to EINTR inside AIO code (restarting the syscall isn't really an option because other AIO could have been already submitted by the same io_submit syscall).

I had responded to your last posting with:

This is a confusing problem description. First, I'm assuming you actually experienced this, is that right? Care to share the details of that?

The above paragraph I think you addressed. I'm still curious about this, though. The submission path should be as non-blocking as possible. I guess you can't really get around locking, though. Was the problem witnessed when doing O_DIRECT or buffered AIO?

Next, assuming we can get ERSTARTSYS and friends, it will be the return code of a single iocb (reaped via io_getevents), not the return code of the io_submit system call. I'm not saying this is right, I'm just saying that your description of the problem is misleading.

That objection stands, but just warrants correcting the problem description.

Anyway, the code itself looks fine. ;-)

author reviewer
#2 lore raw
Quoted 22 lines · Jeff Moyer
Quoted 8 lines · Jan Kara

OCFS2 can return ERESTARTSYS from its write function when the process is signalled while waiting for a cluster lock (and the filesystem is mounted with intr mount option). Generally, it seems reasonable to allow filesystems to return this error code from its IO functions. As we must not leak ERESTARTSYS (and similar error codes) to userspace as a result of an AIO operation, we have to properly convert it to EINTR inside AIO code (restarting the syscall isn't really an option because other AIO could have been already submitted by the same io_submit syscall).

I had responded to your last posting with:

This is a confusing problem description. First, I'm assuming you actually experienced this, is that right? Care to share the details of that?

The above paragraph I think you addressed. I'm still curious about this, though. The submission path should be as non-blocking as possible. I guess you can't really get around locking, though. Was the problem witnessed when doing O_DIRECT or buffered AIO?

Quoted 6 lines

Next, assuming we can get ERSTARTSYS and friends, it will be the return code of a single iocb (reaped via io_getevents), not the return code of the io_submit system call. I'm not saying this is right, I'm just saying that your description of the problem is misleading.

That objection stands, but just warrants correcting the problem description.

Hmm, I tried to address that by saying:

As we must not leak ERESTARTSYS (and similar error codes) to userspace as a result of an AIO operation

     ^^^^^^^^^^^^^^^^^^^^^^^^^^

by which I meant the result value in the iocb structure. But apparently it's not explicit enough. So would you be happier with something like "result received via io_getevents() syscall"?

Quoted 3 lines

Anyway, the code itself looks fine. ;-)

Thanks.

Honza

Signature · 2 lines
Jan Kara <[email protected]>
SUSE Labs, CR
#3 lore raw
Quoted 9 lines · Jan Kara
Quoted 8 lines · Jeff Moyer

Jan Kara <[email protected]> writes:

Next, assuming we can get ERSTARTSYS and friends, it will be the return code of a single iocb (reaped via io_getevents), not the return code of the io_submit system call. I'm not saying this is right, I'm just saying that your description of the problem is misleading.

That objection stands, but just warrants correcting the problem description.

Quoted 8 lines

Hmm, I tried to address that by saying:

As we must not leak ERESTARTSYS (and similar error codes) to userspace as a result of an AIO operation

     ^^^^^^^^^^^^^^^^^^^^^^^^^^

by which I meant the result value in the iocb structure. But apparently it's not explicit enough. So would you be happier with something like "result received via io_getevents() syscall"?

That part looks fine, no need to change it. What tripped me up was this:

(restarting the syscall isn't really an option because other AIO could have been already submitted by the same io_submit syscall)

To me, that sounded like the result would be seen by io_submit instead of io_getevents.

Cheers,
Jeff