[4/8] Fix aio_poll() races
diff mbox series

Message ID 20190310070822.11564-4-viro@ZenIV.linux.org.uk
State New
Headers show
Series
  • [1/8] pin iocb through aio.
Related show

Commit Message

Al Viro March 10, 2019, 7:08 a.m. UTC
From: Al Viro <viro@zeniv.linux.org.uk>

aio_poll() has to cope with several unpleasant problems:
	* requests that might stay around indefinitely need to
be made visible for io_cancel(2); that must not be done to
a request already completed, though.
	* in cases when ->poll() has placed us on a waitqueue,
wakeup might have happened (and request completed) before ->poll()
returns.
	* worse, in some early wakeup cases request might end
up re-added into the queue later - we can't treat "woken up and
currently not in the queue" as "it's not going to stick around
indefinitely"
	* ... moreover, ->poll() might have decided not to
put it on any queues to start with, and that needs to be distinguished
from the previous case
	* ->poll() might have tried to put us on more than one queue.
Only the first will succeed for aio poll, so we might end up missing
wakeups.  OTOH, we might very well notice that only after the
wakeup hits and request gets completed (all before ->poll() gets
around to the second poll_wait()).  In that case it's too late to
decide that we have an error.

req->woken was an attempt to deal with that.  Unfortunately, it was
broken.  What we need to keep track of is not that wakeup has happened -
the thing might come back after that.  It's that async reference is
already gone and won't come back, so we can't (and needn't) put the
request on the list of cancellables.

The easiest case is "request hadn't been put on any waitqueues"; we
can tell by seeing NULL apt.head, and in that case there won't be
anything async.  We should either complete the request ourselves
(if vfs_poll() reports anything of interest) or return an error.

In all other cases we get exclusion with wakeups by grabbing the
queue lock.

If request is currently on queue and we have something interesting
from vfs_poll(), we can steal it and complete the request ourselves.

If it's on queue and vfs_poll() has not reported anything interesting,
we either put it on the cancellable list, or, if we know that it
hadn't been put on all queues ->poll() wanted it on, we steal it and
return an error.

If it's _not_ on queue, it's either been already dealt with (in which
case we do nothing), or there's aio_poll_complete_work() about to be
executed.  In that case we either put it on the cancellable list,
or, if we know it hadn't been put on all queues ->poll() wanted it on,
simulate what cancel would've done.

It's a lot more convoluted than I'd like it to be.  Single-consumer APIs
suck, and unfortunately aio is not an exception...

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
---
 fs/aio.c | 71 +++++++++++++++++++++++++++++-----------------------------------
 1 file changed, 32 insertions(+), 39 deletions(-)

Comments

hch@lst.de March 11, 2019, 7:58 p.m. UTC | #1
Where do we put the second iocb reference in case we return from
vfs_poll without ever being woken?

Also it seems like the complete code would still benefit from a little
helper, something like:

diff --git a/fs/aio.c b/fs/aio.c
index b2a5c7b3a1fe..8415e5e484ce 100644
--- a/fs/aio.c
+++ b/fs/aio.c
@@ -1611,6 +1611,13 @@ static int aio_fsync(struct fsync_iocb *req, const struct iocb *iocb,
 	return 0;
 }
 
+static void aio_poll_finish(struct aio_kiocb *iocb, __poll_t mask)
+{
+	list_del_init(&iocb->ki_list);
+	iocb->ki_res.res = mangle_poll(mask);
+	iocb->poll.done = true;
+}
+
 static void aio_poll_complete_work(struct work_struct *work)
 {
 	struct poll_iocb *req = container_of(work, struct poll_iocb, work);
@@ -1635,9 +1642,7 @@ static void aio_poll_complete_work(struct work_struct *work)
 		spin_unlock_irq(&ctx->ctx_lock);
 		return;
 	}
-	list_del_init(&iocb->ki_list);
-	iocb->ki_res.res = mangle_poll(mask);
-	req->done = true;
+	aio_poll_finish(iocb, mask);
 	spin_unlock_irq(&ctx->ctx_lock);
 
 	iocb_put(iocb);
@@ -1674,24 +1679,20 @@ static int aio_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync,
 
 	list_del_init(&req->wait.entry);
 
-	if (mask) {
+	if (mask && spin_trylock_irqsave(&iocb->ki_ctx->ctx_lock, flags)) {
 		/*
 		 * Try to complete the iocb inline if we can. Use
 		 * irqsave/irqrestore because not all filesystems (e.g. fuse)
 		 * call this function with IRQs disabled and because IRQs
 		 * have to be disabled before ctx_lock is obtained.
 		 */
-		if (spin_trylock_irqsave(&iocb->ki_ctx->ctx_lock, flags)) {
-			list_del(&iocb->ki_list);
-			iocb->ki_res.res = mangle_poll(mask);
-			req->done = true;
-			spin_unlock_irqrestore(&iocb->ki_ctx->ctx_lock, flags);
-			iocb_put(iocb);
-			return 1;
-		}
+		aio_poll_finish(iocb, mask);
+		spin_unlock_irqrestore(&iocb->ki_ctx->ctx_lock, flags);
+		iocb_put(iocb);
+	} else {
+		schedule_work(&req->work);
 	}
 
-	schedule_work(&req->work);
 	return 1;
 }
Al Viro March 11, 2019, 9:06 p.m. UTC | #2
On Mon, Mar 11, 2019 at 08:58:31PM +0100, Christoph Hellwig wrote:
> Where do we put the second iocb reference in case we return from
> vfs_poll without ever being woken?

Depends.  If mask is non-zero (i.e. vfs_poll() has returned something
we care about) and it has never been woken, we steal it and drop the
reference ourselves.  If it is zero and we see that ->poll() has tried
to put it on two queues, we steal it (again, assuming it's not on
waitqueue and _can_ be stolen) and return -EINVAL.  In that case
__io_submit_one() (or, by the end of the series, io_submit_one())
will call iocb_destroy().  And in the normal waiting case (nothing
interesting reported and no errors) it will end up on the list of
cancellables.  Then it either will get completed by later wakeup, which
will drop the reference, or it will get eventually cancelled, which will
hit the same aio_poll_complete_work() and drop the reference...

> Also it seems like the complete code would still benefit from a little
> helper, something like:

Umm...  Not sure I like the name (something like aio_poll_done() seems
to be better), but other than that - no problem.
hch@lst.de March 12, 2019, 7:18 p.m. UTC | #3
On Mon, Mar 11, 2019 at 09:06:18PM +0000, Al Viro wrote:
> On Mon, Mar 11, 2019 at 08:58:31PM +0100, Christoph Hellwig wrote:
> > Where do we put the second iocb reference in case we return from
> > vfs_poll without ever being woken?
> 
> Depends.  If mask is non-zero (i.e. vfs_poll() has returned something
> we care about) and it has never been woken, we steal it and drop the
> reference ourselves.  If it is zero and we see that ->poll() has tried
> to put it on two queues, we steal it (again, assuming it's not on
> waitqueue and _can_ be stolen) and return -EINVAL.  In that case
> __io_submit_one() (or, by the end of the series, io_submit_one())
> will call iocb_destroy().  And in the normal waiting case (nothing
> interesting reported and no errors) it will end up on the list of
> cancellables.  Then it either will get completed by later wakeup, which
> will drop the reference, or it will get eventually cancelled, which will
> hit the same aio_poll_complete_work() and drop the reference...

Ok, seems like the logic is sane.  I was missing how the actual
mask logic worked in aio_poll().

> > Also it seems like the complete code would still benefit from a little
> > helper, something like:
> 
> Umm...  Not sure I like the name (something like aio_poll_done() seems
> to be better), but other than that - no problem.

I don't care about the name.  Feel free to change it to whatever suits
you.

Patch
diff mbox series

diff --git a/fs/aio.c b/fs/aio.c
index b9c4c1894020..f47a29f7f201 100644
--- a/fs/aio.c
+++ b/fs/aio.c
@@ -181,7 +181,7 @@  struct poll_iocb {
 	struct file		*file;
 	struct wait_queue_head	*head;
 	__poll_t		events;
-	bool			woken;
+	bool			done;
 	bool			cancelled;
 	struct wait_queue_entry	wait;
 	struct work_struct	work;
@@ -1606,12 +1606,6 @@  static int aio_fsync(struct fsync_iocb *req, const struct iocb *iocb,
 	return 0;
 }
 
-static inline void aio_poll_complete(struct aio_kiocb *iocb, __poll_t mask)
-{
-	iocb->ki_res.res = mangle_poll(mask);
-	iocb_put(iocb);
-}
-
 static void aio_poll_complete_work(struct work_struct *work)
 {
 	struct poll_iocb *req = container_of(work, struct poll_iocb, work);
@@ -1637,9 +1631,11 @@  static void aio_poll_complete_work(struct work_struct *work)
 		return;
 	}
 	list_del_init(&iocb->ki_list);
+	iocb->ki_res.res = mangle_poll(mask);
+	req->done = true;
 	spin_unlock_irq(&ctx->ctx_lock);
 
-	aio_poll_complete(iocb, mask);
+	iocb_put(iocb);
 }
 
 /* assumes we are called with irqs disabled */
@@ -1671,7 +1667,7 @@  static int aio_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync,
 	if (mask && !(mask & req->events))
 		return 0;
 
-	req->woken = true;
+	list_del_init(&req->wait.entry);
 
 	if (mask) {
 		/*
@@ -1682,15 +1678,14 @@  static int aio_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync,
 		 */
 		if (spin_trylock_irqsave(&iocb->ki_ctx->ctx_lock, flags)) {
 			list_del(&iocb->ki_list);
+			iocb->ki_res.res = mangle_poll(mask);
+			req->done = true;
 			spin_unlock_irqrestore(&iocb->ki_ctx->ctx_lock, flags);
-
-			list_del_init(&req->wait.entry);
-			aio_poll_complete(iocb, mask);
+			iocb_put(iocb);
 			return 1;
 		}
 	}
 
-	list_del_init(&req->wait.entry);
 	schedule_work(&req->work);
 	return 1;
 }
@@ -1723,6 +1718,7 @@  static ssize_t aio_poll(struct aio_kiocb *aiocb, const struct iocb *iocb)
 	struct kioctx *ctx = aiocb->ki_ctx;
 	struct poll_iocb *req = &aiocb->poll;
 	struct aio_poll_table apt;
+	bool cancel = false;
 	__poll_t mask;
 
 	/* reject any unknown events outside the normal event mask. */
@@ -1736,7 +1732,7 @@  static ssize_t aio_poll(struct aio_kiocb *aiocb, const struct iocb *iocb)
 	req->events = demangle_poll(iocb->aio_buf) | EPOLLERR | EPOLLHUP;
 
 	req->head = NULL;
-	req->woken = false;
+	req->done = false;
 	req->cancelled = false;
 
 	apt.pt._qproc = aio_poll_queue_proc;
@@ -1749,36 +1745,33 @@  static ssize_t aio_poll(struct aio_kiocb *aiocb, const struct iocb *iocb)
 	init_waitqueue_func_entry(&req->wait, aio_poll_wake);
 
 	mask = vfs_poll(req->file, &apt.pt) & req->events;
-	if (unlikely(!req->head)) {
-		/* we did not manage to set up a waitqueue, done */
-		goto out;
-	}
-
 	spin_lock_irq(&ctx->ctx_lock);
-	spin_lock(&req->head->lock);
-	if (req->woken) {
-		/* wake_up context handles the rest */
-		mask = 0;
+	if (likely(req->head)) {
+		spin_lock(&req->head->lock);
+		if (unlikely(list_empty(&req->wait.entry))) {
+			if (apt.error)
+				cancel = true;
+			apt.error = 0;
+			mask = 0;
+		}
+		if (mask || apt.error) {
+			list_del_init(&req->wait.entry);
+		} else if (cancel) {
+			WRITE_ONCE(req->cancelled, true);
+		} else if (!req->done) { /* actually waiting for an event */
+			list_add_tail(&aiocb->ki_list, &ctx->active_reqs);
+			aiocb->ki_cancel = aio_poll_cancel;
+		}
+		spin_unlock(&req->head->lock);
+	}
+	if (mask) { /* no async, we'd stolen it */
+		aiocb->ki_res.res = mangle_poll(mask);
 		apt.error = 0;
-	} else if (mask || apt.error) {
-		/* if we get an error or a mask we are done */
-		WARN_ON_ONCE(list_empty(&req->wait.entry));
-		list_del_init(&req->wait.entry);
-	} else {
-		/* actually waiting for an event */
-		list_add_tail(&aiocb->ki_list, &ctx->active_reqs);
-		aiocb->ki_cancel = aio_poll_cancel;
 	}
-	spin_unlock(&req->head->lock);
 	spin_unlock_irq(&ctx->ctx_lock);
-
-out:
-	if (unlikely(apt.error))
-		return apt.error;
-
 	if (mask)
-		aio_poll_complete(aiocb, mask);
-	return 0;
+		iocb_put(aiocb);
+	return apt.error;
 }
 
 static int __io_submit_one(struct kioctx *ctx, const struct iocb *iocb,