[PATCH 5.10 0/5] aio poll fixes for 5.10

[PATCH 5.10 0/5] aio poll fixes for 5.10

[Eric Biggers]

Backport the aio poll fixes to 5.10.  This resolves a conflict in
aio_poll_wake() in patch 4.  It's a "trivial" conflict, but I'm sending
this to make sure it doesn't get dropped.

Eric Biggers (5):
  wait: add wake_up_pollfree()
  binder: use wake_up_pollfree()
  signalfd: use wake_up_pollfree()
  aio: keep poll requests on waitqueue until completed
  aio: fix use-after-free due to missing POLLFREE handling

 drivers/android/binder.c        |  21 ++--
 fs/aio.c                        | 184 ++++++++++++++++++++++++++------
 fs/signalfd.c                   |  12 +--
 include/linux/wait.h            |  26 +++++
 include/uapi/asm-generic/poll.h |   2 +-
 kernel/sched/wait.c             |   7 ++
 6 files changed, 195 insertions(+), 57 deletions(-)

-- 
2.34.1

[PATCH 5.10 1/5] wait: add wake_up_pollfree()

[Eric Biggers]

From: Eric Biggers <ebiggers@google.com>

commit 42288cb44c4b5fff7653bc392b583a2b8bd6a8c0 upstream.

Several ->poll() implementations are special in that they use a
waitqueue whose lifetime is the current task, rather than the struct
file as is normally the case.  This is okay for blocking polls, since a
blocking poll occurs within one task; however, non-blocking polls
require another solution.  This solution is for the queue to be cleared
before it is freed, using 'wake_up_poll(wq, EPOLLHUP | POLLFREE);'.

However, that has a bug: wake_up_poll() calls __wake_up() with
nr_exclusive=1.  Therefore, if there are multiple "exclusive" waiters,
and the wakeup function for the first one returns a positive value, only
that one will be called.  That's *not* what's needed for POLLFREE;
POLLFREE is special in that it really needs to wake up everyone.

Considering the three non-blocking poll systems:

- io_uring poll doesn't handle POLLFREE at all, so it is broken anyway.

- aio poll is unaffected, since it doesn't support exclusive waits.
  However, that's fragile, as someone could add this feature later.

- epoll doesn't appear to be broken by this, since its wakeup function
  returns 0 when it sees POLLFREE.  But this is fragile.

Although there is a workaround (see epoll), it's better to define a
function which always sends POLLFREE to all waiters.  Add such a
function.  Also make it verify that the queue really becomes empty after
all waiters have been woken up.

Reported-by: Linus Torvalds <torvalds@linux-foundation.org>
Cc: stable@vger.kernel.org
Link: https://lore.kernel.org/r/20211209010455.42744-2-ebiggers@kernel.org
Signed-off-by: Eric Biggers <ebiggers@google.com>
---
 include/linux/wait.h | 26 ++++++++++++++++++++++++++
 kernel/sched/wait.c  |  7 +++++++
 2 files changed, 33 insertions(+)

diff --git a/include/linux/wait.h b/include/linux/wait.h
index f8b0704968a1e..9b8b0833100a0 100644
--- a/include/linux/wait.h
+++ b/include/linux/wait.h
@@ -207,6 +207,7 @@ void __wake_up_sync_key(struct wait_queue_head *wq_head, unsigned int mode, void
 void __wake_up_locked_sync_key(struct wait_queue_head *wq_head, unsigned int mode, void *key);
 void __wake_up_locked(struct wait_queue_head *wq_head, unsigned int mode, int nr);
 void __wake_up_sync(struct wait_queue_head *wq_head, unsigned int mode);
+void __wake_up_pollfree(struct wait_queue_head *wq_head);
 
 #define wake_up(x)			__wake_up(x, TASK_NORMAL, 1, NULL)
 #define wake_up_nr(x, nr)		__wake_up(x, TASK_NORMAL, nr, NULL)
@@ -235,6 +236,31 @@ void __wake_up_sync(struct wait_queue_head *wq_head, unsigned int mode);
 #define wake_up_interruptible_sync_poll_locked(x, m)				\
 	__wake_up_locked_sync_key((x), TASK_INTERRUPTIBLE, poll_to_key(m))
 
+/**
+ * wake_up_pollfree - signal that a polled waitqueue is going away
+ * @wq_head: the wait queue head
+ *
+ * In the very rare cases where a ->poll() implementation uses a waitqueue whose
+ * lifetime is tied to a task rather than to the 'struct file' being polled,
+ * this function must be called before the waitqueue is freed so that
+ * non-blocking polls (e.g. epoll) are notified that the queue is going away.
+ *
+ * The caller must also RCU-delay the freeing of the wait_queue_head, e.g. via
+ * an explicit synchronize_rcu() or call_rcu(), or via SLAB_TYPESAFE_BY_RCU.
+ */
+static inline void wake_up_pollfree(struct wait_queue_head *wq_head)
+{
+	/*
+	 * For performance reasons, we don't always take the queue lock here.
+	 * Therefore, we might race with someone removing the last entry from
+	 * the queue, and proceed while they still hold the queue lock.
+	 * However, rcu_read_lock() is required to be held in such cases, so we
+	 * can safely proceed with an RCU-delayed free.
+	 */
+	if (waitqueue_active(wq_head))
+		__wake_up_pollfree(wq_head);
+}
+
 #define ___wait_cond_timeout(condition)						\
 ({										\
 	bool __cond = (condition);						\
diff --git a/kernel/sched/wait.c b/kernel/sched/wait.c
index 21005b980a6b7..a55642aa3f68b 100644
--- a/kernel/sched/wait.c
+++ b/kernel/sched/wait.c
@@ -223,6 +223,13 @@ void __wake_up_sync(struct wait_queue_head *wq_head, unsigned int mode)
 }
 EXPORT_SYMBOL_GPL(__wake_up_sync);	/* For internal use only */
 
+void __wake_up_pollfree(struct wait_queue_head *wq_head)
+{
+	__wake_up(wq_head, TASK_NORMAL, 0, poll_to_key(EPOLLHUP | POLLFREE));
+	/* POLLFREE must have cleared the queue. */
+	WARN_ON_ONCE(waitqueue_active(wq_head));
+}
+
 /*
  * Note: we use "set_current_state()" _after_ the wait-queue add,
  * because we need a memory barrier there on SMP, so that any
-- 
2.34.1

[PATCH 5.10 2/5] binder: use wake_up_pollfree()

[Eric Biggers]

From: Eric Biggers <ebiggers@google.com>

commit a880b28a71e39013e357fd3adccd1d8a31bc69a8 upstream.

wake_up_poll() uses nr_exclusive=1, so it's not guaranteed to wake up
all exclusive waiters.  Yet, POLLFREE *must* wake up all waiters.  epoll
and aio poll are fortunately not affected by this, but it's very
fragile.  Thus, the new function wake_up_pollfree() has been introduced.

Convert binder to use wake_up_pollfree().

Reported-by: Linus Torvalds <torvalds@linux-foundation.org>
Fixes: f5cb779ba163 ("ANDROID: binder: remove waitqueue when thread exits.")
Cc: stable@vger.kernel.org
Link: https://lore.kernel.org/r/20211209010455.42744-3-ebiggers@kernel.org
Signed-off-by: Eric Biggers <ebiggers@google.com>
---
 drivers/android/binder.c | 21 +++++++++------------
 1 file changed, 9 insertions(+), 12 deletions(-)

diff --git a/drivers/android/binder.c b/drivers/android/binder.c
index a1255971e50ce..80e2bbb36422e 100644
--- a/drivers/android/binder.c
+++ b/drivers/android/binder.c
@@ -4784,23 +4784,20 @@ static int binder_thread_release(struct binder_proc *proc,
 	__release(&t->lock);
 
 	/*
-	 * If this thread used poll, make sure we remove the waitqueue
-	 * from any epoll data structures holding it with POLLFREE.
-	 * waitqueue_active() is safe to use here because we're holding
-	 * the inner lock.
+	 * If this thread used poll, make sure we remove the waitqueue from any
+	 * poll data structures holding it.
 	 */
-	if ((thread->looper & BINDER_LOOPER_STATE_POLL) &&
-	    waitqueue_active(&thread->wait)) {
-		wake_up_poll(&thread->wait, EPOLLHUP | POLLFREE);
-	}
+	if (thread->looper & BINDER_LOOPER_STATE_POLL)
+		wake_up_pollfree(&thread->wait);
 
 	binder_inner_proc_unlock(thread->proc);
 
 	/*
-	 * This is needed to avoid races between wake_up_poll() above and
-	 * and ep_remove_waitqueue() called for other reasons (eg the epoll file
-	 * descriptor being closed); ep_remove_waitqueue() holds an RCU read
-	 * lock, so we can be sure it's done after calling synchronize_rcu().
+	 * This is needed to avoid races between wake_up_pollfree() above and
+	 * someone else removing the last entry from the queue for other reasons
+	 * (e.g. ep_remove_wait_queue() being called due to an epoll file
+	 * descriptor being closed).  Such other users hold an RCU read lock, so
+	 * we can be sure they're done after we call synchronize_rcu().
 	 */
 	if (thread->looper & BINDER_LOOPER_STATE_POLL)
 		synchronize_rcu();
-- 
2.34.1

[PATCH 5.10 3/5] signalfd: use wake_up_pollfree()

[Eric Biggers]

From: Eric Biggers <ebiggers@google.com>

commit 9537bae0da1f8d1e2361ab6d0479e8af7824e160 upstream.

wake_up_poll() uses nr_exclusive=1, so it's not guaranteed to wake up
all exclusive waiters.  Yet, POLLFREE *must* wake up all waiters.  epoll
and aio poll are fortunately not affected by this, but it's very
fragile.  Thus, the new function wake_up_pollfree() has been introduced.

Convert signalfd to use wake_up_pollfree().

Reported-by: Linus Torvalds <torvalds@linux-foundation.org>
Fixes: d80e731ecab4 ("epoll: introduce POLLFREE to flush ->signalfd_wqh before kfree()")
Cc: stable@vger.kernel.org
Link: https://lore.kernel.org/r/20211209010455.42744-4-ebiggers@kernel.org
Signed-off-by: Eric Biggers <ebiggers@google.com>
---
 fs/signalfd.c | 12 +-----------
 1 file changed, 1 insertion(+), 11 deletions(-)

diff --git a/fs/signalfd.c b/fs/signalfd.c
index 456046e158737..b94fb5f81797a 100644
--- a/fs/signalfd.c
+++ b/fs/signalfd.c
@@ -35,17 +35,7 @@
 
 void signalfd_cleanup(struct sighand_struct *sighand)
 {
-	wait_queue_head_t *wqh = &sighand->signalfd_wqh;
-	/*
-	 * The lockless check can race with remove_wait_queue() in progress,
-	 * but in this case its caller should run under rcu_read_lock() and
-	 * sighand_cachep is SLAB_TYPESAFE_BY_RCU, we can safely return.
-	 */
-	if (likely(!waitqueue_active(wqh)))
-		return;
-
-	/* wait_queue_entry_t->func(POLLFREE) should do remove_wait_queue() */
-	wake_up_poll(wqh, EPOLLHUP | POLLFREE);
+	wake_up_pollfree(&sighand->signalfd_wqh);
 }
 
 struct signalfd_ctx {
-- 
2.34.1

[PATCH 5.10 4/5] aio: keep poll requests on waitqueue until completed

[Eric Biggers]

From: Eric Biggers <ebiggers@google.com>

commit 363bee27e25804d8981dd1c025b4ad49dc39c530 upstream.

Currently, aio_poll_wake() will always remove the poll request from the
waitqueue.  Then, if aio_poll_complete_work() sees that none of the
polled events are ready and the request isn't cancelled, it re-adds the
request to the waitqueue.  (This can easily happen when polling a file
that doesn't pass an event mask when waking up its waitqueue.)

This is fundamentally broken for two reasons:

  1. If a wakeup occurs between vfs_poll() and the request being
     re-added to the waitqueue, it will be missed because the request
     wasn't on the waitqueue at the time.  Therefore, IOCB_CMD_POLL
     might never complete even if the polled file is ready.

  2. When the request isn't on the waitqueue, there is no way to be
     notified that the waitqueue is being freed (which happens when its
     lifetime is shorter than the struct file's).  This is supposed to
     happen via the waitqueue entries being woken up with POLLFREE.

Therefore, leave the requests on the waitqueue until they are actually
completed (or cancelled).  To keep track of when aio_poll_complete_work
needs to be scheduled, use new fields in struct poll_iocb.  Remove the
'done' field which is now redundant.

Note that this is consistent with how sys_poll() and eventpoll work;
their wakeup functions do *not* remove the waitqueue entries.

Fixes: 2c14fa838cbe ("aio: implement IOCB_CMD_POLL")
Cc: <stable@vger.kernel.org> # v4.18+
Link: https://lore.kernel.org/r/20211209010455.42744-5-ebiggers@kernel.org
Signed-off-by: Eric Biggers <ebiggers@google.com>
---
 fs/aio.c | 83 ++++++++++++++++++++++++++++++++++++++++++--------------
 1 file changed, 63 insertions(+), 20 deletions(-)

diff --git a/fs/aio.c b/fs/aio.c
index 6a21d8919409c..37c5f450d1596 100644
--- a/fs/aio.c
+++ b/fs/aio.c
@@ -182,8 +182,9 @@ struct poll_iocb {
 	struct file		*file;
 	struct wait_queue_head	*head;
 	__poll_t		events;
-	bool			done;
 	bool			cancelled;
+	bool			work_scheduled;
+	bool			work_need_resched;
 	struct wait_queue_entry	wait;
 	struct work_struct	work;
 };
@@ -1640,14 +1641,26 @@ static void aio_poll_complete_work(struct work_struct *work)
 	 * avoid further branches in the fast path.
 	 */
 	spin_lock_irq(&ctx->ctx_lock);
+	spin_lock(&req->head->lock);
 	if (!mask && !READ_ONCE(req->cancelled)) {
-		add_wait_queue(req->head, &req->wait);
+		/*
+		 * The request isn't actually ready to be completed yet.
+		 * Reschedule completion if another wakeup came in.
+		 */
+		if (req->work_need_resched) {
+			schedule_work(&req->work);
+			req->work_need_resched = false;
+		} else {
+			req->work_scheduled = false;
+		}
+		spin_unlock(&req->head->lock);
 		spin_unlock_irq(&ctx->ctx_lock);
 		return;
 	}
+	list_del_init(&req->wait.entry);
+	spin_unlock(&req->head->lock);
 	list_del_init(&iocb->ki_list);
 	iocb->ki_res.res = mangle_poll(mask);
-	req->done = true;
 	spin_unlock_irq(&ctx->ctx_lock);
 
 	iocb_put(iocb);
@@ -1661,9 +1674,9 @@ static int aio_poll_cancel(struct kiocb *iocb)
 
 	spin_lock(&req->head->lock);
 	WRITE_ONCE(req->cancelled, true);
-	if (!list_empty(&req->wait.entry)) {
-		list_del_init(&req->wait.entry);
+	if (!req->work_scheduled) {
 		schedule_work(&aiocb->poll.work);
+		req->work_scheduled = true;
 	}
 	spin_unlock(&req->head->lock);
 
@@ -1682,20 +1695,26 @@ static int aio_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync,
 	if (mask && !(mask & req->events))
 		return 0;
 
-	list_del_init(&req->wait.entry);
-
-	if (mask && spin_trylock_irqsave(&iocb->ki_ctx->ctx_lock, flags)) {
+	/*
+	 * Complete the request inline if possible.  This requires that three
+	 * conditions be met:
+	 *   1. An event mask must have been passed.  If a plain wakeup was done
+	 *	instead, then mask == 0 and we have to call vfs_poll() to get
+	 *	the events, so inline completion isn't possible.
+	 *   2. The completion work must not have already been scheduled.
+	 *   3. ctx_lock must not be busy.  We have to use trylock because we
+	 *	already hold the waitqueue lock, so this inverts the normal
+	 *	locking order.  Use irqsave/irqrestore because not all
+	 *	filesystems (e.g. fuse) call this function with IRQs disabled,
+	 *	yet IRQs have to be disabled before ctx_lock is obtained.
+	 */
+	if (mask && !req->work_scheduled &&
+	    spin_trylock_irqsave(&iocb->ki_ctx->ctx_lock, flags)) {
 		struct kioctx *ctx = iocb->ki_ctx;
 
-		/*
-		 * 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.
-		 */
+		list_del_init(&req->wait.entry);
 		list_del(&iocb->ki_list);
 		iocb->ki_res.res = mangle_poll(mask);
-		req->done = true;
 		if (iocb->ki_eventfd && eventfd_signal_count()) {
 			iocb = NULL;
 			INIT_WORK(&req->work, aio_poll_put_work);
@@ -1705,7 +1724,20 @@ static int aio_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync,
 		if (iocb)
 			iocb_put(iocb);
 	} else {
-		schedule_work(&req->work);
+		/*
+		 * Schedule the completion work if needed.  If it was already
+		 * scheduled, record that another wakeup came in.
+		 *
+		 * Don't remove the request from the waitqueue here, as it might
+		 * not actually be complete yet (we won't know until vfs_poll()
+		 * is called), and we must not miss any wakeups.
+		 */
+		if (req->work_scheduled) {
+			req->work_need_resched = true;
+		} else {
+			schedule_work(&req->work);
+			req->work_scheduled = true;
+		}
 	}
 	return 1;
 }
@@ -1752,8 +1784,9 @@ static int aio_poll(struct aio_kiocb *aiocb, const struct iocb *iocb)
 	req->events = demangle_poll(iocb->aio_buf) | EPOLLERR | EPOLLHUP;
 
 	req->head = NULL;
-	req->done = false;
 	req->cancelled = false;
+	req->work_scheduled = false;
+	req->work_need_resched = false;
 
 	apt.pt._qproc = aio_poll_queue_proc;
 	apt.pt._key = req->events;
@@ -1768,17 +1801,27 @@ static int aio_poll(struct aio_kiocb *aiocb, const struct iocb *iocb)
 	spin_lock_irq(&ctx->ctx_lock);
 	if (likely(req->head)) {
 		spin_lock(&req->head->lock);
-		if (unlikely(list_empty(&req->wait.entry))) {
-			if (apt.error)
+		if (list_empty(&req->wait.entry) || req->work_scheduled) {
+			/*
+			 * aio_poll_wake() already either scheduled the async
+			 * completion work, or completed the request inline.
+			 */
+			if (apt.error) /* unsupported case: multiple queues */
 				cancel = true;
 			apt.error = 0;
 			mask = 0;
 		}
 		if (mask || apt.error) {
+			/* Steal to complete synchronously. */
 			list_del_init(&req->wait.entry);
 		} else if (cancel) {
+			/* Cancel if possible (may be too late though). */
 			WRITE_ONCE(req->cancelled, true);
-		} else if (!req->done) { /* actually waiting for an event */
+		} else if (!list_empty(&req->wait.entry)) {
+			/*
+			 * Actually waiting for an event, so add the request to
+			 * active_reqs so that it can be cancelled if needed.
+			 */
 			list_add_tail(&aiocb->ki_list, &ctx->active_reqs);
 			aiocb->ki_cancel = aio_poll_cancel;
 		}
-- 
2.34.1

[PATCH 5.10 5/5] aio: fix use-after-free due to missing POLLFREE handling

[Eric Biggers]

From: Eric Biggers <ebiggers@google.com>

commit 50252e4b5e989ce64555c7aef7516bdefc2fea72 upstream.

signalfd_poll() and binder_poll() are special in that they use a
waitqueue whose lifetime is the current task, rather than the struct
file as is normally the case.  This is okay for blocking polls, since a
blocking poll occurs within one task; however, non-blocking polls
require another solution.  This solution is for the queue to be cleared
before it is freed, by sending a POLLFREE notification to all waiters.

Unfortunately, only eventpoll handles POLLFREE.  A second type of
non-blocking poll, aio poll, was added in kernel v4.18, and it doesn't
handle POLLFREE.  This allows a use-after-free to occur if a signalfd or
binder fd is polled with aio poll, and the waitqueue gets freed.

Fix this by making aio poll handle POLLFREE.

A patch by Ramji Jiyani <ramjiyani@google.com>
(https://lore.kernel.org/r/20211027011834.2497484-1-ramjiyani@google.com)
tried to do this by making aio_poll_wake() always complete the request
inline if POLLFREE is seen.  However, that solution had two bugs.
First, it introduced a deadlock, as it unconditionally locked the aio
context while holding the waitqueue lock, which inverts the normal
locking order.  Second, it didn't consider that POLLFREE notifications
are missed while the request has been temporarily de-queued.

The second problem was solved by my previous patch.  This patch then
properly fixes the use-after-free by handling POLLFREE in a
deadlock-free way.  It does this by taking advantage of the fact that
freeing of the waitqueue is RCU-delayed, similar to what eventpoll does.

Fixes: 2c14fa838cbe ("aio: implement IOCB_CMD_POLL")
Cc: <stable@vger.kernel.org> # v4.18+
Link: https://lore.kernel.org/r/20211209010455.42744-6-ebiggers@kernel.org
Signed-off-by: Eric Biggers <ebiggers@google.com>
---
 fs/aio.c                        | 137 ++++++++++++++++++++++++--------
 include/uapi/asm-generic/poll.h |   2 +-
 2 files changed, 107 insertions(+), 32 deletions(-)

diff --git a/fs/aio.c b/fs/aio.c
index 37c5f450d1596..2a9dfa58ec3ab 100644
--- a/fs/aio.c
+++ b/fs/aio.c
@@ -1622,6 +1622,51 @@ static void aio_poll_put_work(struct work_struct *work)
 	iocb_put(iocb);
 }
 
+/*
+ * Safely lock the waitqueue which the request is on, synchronizing with the
+ * case where the ->poll() provider decides to free its waitqueue early.
+ *
+ * Returns true on success, meaning that req->head->lock was locked, req->wait
+ * is on req->head, and an RCU read lock was taken.  Returns false if the
+ * request was already removed from its waitqueue (which might no longer exist).
+ */
+static bool poll_iocb_lock_wq(struct poll_iocb *req)
+{
+	wait_queue_head_t *head;
+
+	/*
+	 * While we hold the waitqueue lock and the waitqueue is nonempty,
+	 * wake_up_pollfree() will wait for us.  However, taking the waitqueue
+	 * lock in the first place can race with the waitqueue being freed.
+	 *
+	 * We solve this as eventpoll does: by taking advantage of the fact that
+	 * all users of wake_up_pollfree() will RCU-delay the actual free.  If
+	 * we enter rcu_read_lock() and see that the pointer to the queue is
+	 * non-NULL, we can then lock it without the memory being freed out from
+	 * under us, then check whether the request is still on the queue.
+	 *
+	 * Keep holding rcu_read_lock() as long as we hold the queue lock, in
+	 * case the caller deletes the entry from the queue, leaving it empty.
+	 * In that case, only RCU prevents the queue memory from being freed.
+	 */
+	rcu_read_lock();
+	head = smp_load_acquire(&req->head);
+	if (head) {
+		spin_lock(&head->lock);
+		if (!list_empty(&req->wait.entry))
+			return true;
+		spin_unlock(&head->lock);
+	}
+	rcu_read_unlock();
+	return false;
+}
+
+static void poll_iocb_unlock_wq(struct poll_iocb *req)
+{
+	spin_unlock(&req->head->lock);
+	rcu_read_unlock();
+}
+
 static void aio_poll_complete_work(struct work_struct *work)
 {
 	struct poll_iocb *req = container_of(work, struct poll_iocb, work);
@@ -1641,24 +1686,25 @@ static void aio_poll_complete_work(struct work_struct *work)
 	 * avoid further branches in the fast path.
 	 */
 	spin_lock_irq(&ctx->ctx_lock);
-	spin_lock(&req->head->lock);
-	if (!mask && !READ_ONCE(req->cancelled)) {
-		/*
-		 * The request isn't actually ready to be completed yet.
-		 * Reschedule completion if another wakeup came in.
-		 */
-		if (req->work_need_resched) {
-			schedule_work(&req->work);
-			req->work_need_resched = false;
-		} else {
-			req->work_scheduled = false;
+	if (poll_iocb_lock_wq(req)) {
+		if (!mask && !READ_ONCE(req->cancelled)) {
+			/*
+			 * The request isn't actually ready to be completed yet.
+			 * Reschedule completion if another wakeup came in.
+			 */
+			if (req->work_need_resched) {
+				schedule_work(&req->work);
+				req->work_need_resched = false;
+			} else {
+				req->work_scheduled = false;
+			}
+			poll_iocb_unlock_wq(req);
+			spin_unlock_irq(&ctx->ctx_lock);
+			return;
 		}
-		spin_unlock(&req->head->lock);
-		spin_unlock_irq(&ctx->ctx_lock);
-		return;
-	}
-	list_del_init(&req->wait.entry);
-	spin_unlock(&req->head->lock);
+		list_del_init(&req->wait.entry);
+		poll_iocb_unlock_wq(req);
+	} /* else, POLLFREE has freed the waitqueue, so we must complete */
 	list_del_init(&iocb->ki_list);
 	iocb->ki_res.res = mangle_poll(mask);
 	spin_unlock_irq(&ctx->ctx_lock);
@@ -1672,13 +1718,14 @@ static int aio_poll_cancel(struct kiocb *iocb)
 	struct aio_kiocb *aiocb = container_of(iocb, struct aio_kiocb, rw);
 	struct poll_iocb *req = &aiocb->poll;
 
-	spin_lock(&req->head->lock);
-	WRITE_ONCE(req->cancelled, true);
-	if (!req->work_scheduled) {
-		schedule_work(&aiocb->poll.work);
-		req->work_scheduled = true;
-	}
-	spin_unlock(&req->head->lock);
+	if (poll_iocb_lock_wq(req)) {
+		WRITE_ONCE(req->cancelled, true);
+		if (!req->work_scheduled) {
+			schedule_work(&aiocb->poll.work);
+			req->work_scheduled = true;
+		}
+		poll_iocb_unlock_wq(req);
+	} /* else, the request was force-cancelled by POLLFREE already */
 
 	return 0;
 }
@@ -1730,7 +1777,8 @@ static int aio_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync,
 		 *
 		 * Don't remove the request from the waitqueue here, as it might
 		 * not actually be complete yet (we won't know until vfs_poll()
-		 * is called), and we must not miss any wakeups.
+		 * is called), and we must not miss any wakeups.  POLLFREE is an
+		 * exception to this; see below.
 		 */
 		if (req->work_scheduled) {
 			req->work_need_resched = true;
@@ -1738,6 +1786,28 @@ static int aio_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync,
 			schedule_work(&req->work);
 			req->work_scheduled = true;
 		}
+
+		/*
+		 * If the waitqueue is being freed early but we can't complete
+		 * the request inline, we have to tear down the request as best
+		 * we can.  That means immediately removing the request from its
+		 * waitqueue and preventing all further accesses to the
+		 * waitqueue via the request.  We also need to schedule the
+		 * completion work (done above).  Also mark the request as
+		 * cancelled, to potentially skip an unneeded call to ->poll().
+		 */
+		if (mask & POLLFREE) {
+			WRITE_ONCE(req->cancelled, true);
+			list_del_init(&req->wait.entry);
+
+			/*
+			 * Careful: this *must* be the last step, since as soon
+			 * as req->head is NULL'ed out, the request can be
+			 * completed and freed, since aio_poll_complete_work()
+			 * will no longer need to take the waitqueue lock.
+			 */
+			smp_store_release(&req->head, NULL);
+		}
 	}
 	return 1;
 }
@@ -1745,6 +1815,7 @@ static int aio_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync,
 struct aio_poll_table {
 	struct poll_table_struct	pt;
 	struct aio_kiocb		*iocb;
+	bool				queued;
 	int				error;
 };
 
@@ -1755,11 +1826,12 @@ aio_poll_queue_proc(struct file *file, struct wait_queue_head *head,
 	struct aio_poll_table *pt = container_of(p, struct aio_poll_table, pt);
 
 	/* multiple wait queues per file are not supported */
-	if (unlikely(pt->iocb->poll.head)) {
+	if (unlikely(pt->queued)) {
 		pt->error = -EINVAL;
 		return;
 	}
 
+	pt->queued = true;
 	pt->error = 0;
 	pt->iocb->poll.head = head;
 	add_wait_queue(head, &pt->iocb->poll.wait);
@@ -1791,6 +1863,7 @@ static int aio_poll(struct aio_kiocb *aiocb, const struct iocb *iocb)
 	apt.pt._qproc = aio_poll_queue_proc;
 	apt.pt._key = req->events;
 	apt.iocb = aiocb;
+	apt.queued = false;
 	apt.error = -EINVAL; /* same as no support for IOCB_CMD_POLL */
 
 	/* initialized the list so that we can do list_empty checks */
@@ -1799,9 +1872,10 @@ static int aio_poll(struct aio_kiocb *aiocb, const struct iocb *iocb)
 
 	mask = vfs_poll(req->file, &apt.pt) & req->events;
 	spin_lock_irq(&ctx->ctx_lock);
-	if (likely(req->head)) {
-		spin_lock(&req->head->lock);
-		if (list_empty(&req->wait.entry) || req->work_scheduled) {
+	if (likely(apt.queued)) {
+		bool on_queue = poll_iocb_lock_wq(req);
+
+		if (!on_queue || req->work_scheduled) {
 			/*
 			 * aio_poll_wake() already either scheduled the async
 			 * completion work, or completed the request inline.
@@ -1817,7 +1891,7 @@ static int aio_poll(struct aio_kiocb *aiocb, const struct iocb *iocb)
 		} else if (cancel) {
 			/* Cancel if possible (may be too late though). */
 			WRITE_ONCE(req->cancelled, true);
-		} else if (!list_empty(&req->wait.entry)) {
+		} else if (on_queue) {
 			/*
 			 * Actually waiting for an event, so add the request to
 			 * active_reqs so that it can be cancelled if needed.
@@ -1825,7 +1899,8 @@ static int aio_poll(struct aio_kiocb *aiocb, const struct iocb *iocb)
 			list_add_tail(&aiocb->ki_list, &ctx->active_reqs);
 			aiocb->ki_cancel = aio_poll_cancel;
 		}
-		spin_unlock(&req->head->lock);
+		if (on_queue)
+			poll_iocb_unlock_wq(req);
 	}
 	if (mask) { /* no async, we'd stolen it */
 		aiocb->ki_res.res = mangle_poll(mask);
diff --git a/include/uapi/asm-generic/poll.h b/include/uapi/asm-generic/poll.h
index 41b509f410bf9..f9c520ce4bf4e 100644
--- a/include/uapi/asm-generic/poll.h
+++ b/include/uapi/asm-generic/poll.h
@@ -29,7 +29,7 @@
 #define POLLRDHUP       0x2000
 #endif
 
-#define POLLFREE	(__force __poll_t)0x4000	/* currently only for epoll */
+#define POLLFREE	(__force __poll_t)0x4000
 
 #define POLL_BUSY_LOOP	(__force __poll_t)0x8000
 
-- 
2.34.1

Re: [PATCH 5.10 0/5] aio poll fixes for 5.10

[Greg KH]

On Fri, Dec 10, 2021 at 03:48:00PM -0800, Eric Biggers wrote:
> Backport the aio poll fixes to 5.10.  This resolves a conflict in
> aio_poll_wake() in patch 4.  It's a "trivial" conflict, but I'm sending
> this to make sure it doesn't get dropped.
> 
> Eric Biggers (5):
>   wait: add wake_up_pollfree()
>   binder: use wake_up_pollfree()
>   signalfd: use wake_up_pollfree()
>   aio: keep poll requests on waitqueue until completed
>   aio: fix use-after-free due to missing POLLFREE handling
> 
>  drivers/android/binder.c        |  21 ++--
>  fs/aio.c                        | 184 ++++++++++++++++++++++++++------
>  fs/signalfd.c                   |  12 +--
>  include/linux/wait.h            |  26 +++++
>  include/uapi/asm-generic/poll.h |   2 +-
>  kernel/sched/wait.c             |   7 ++
>  6 files changed, 195 insertions(+), 57 deletions(-)
> 
> -- 
> 2.34.1
> 

Thanks for all of the backports, much appreciated and now queued up.

greg k-h

Re: [PATCH 5.10 0/5] aio poll fixes for 5.10

[Eric Biggers]

On Sun, Dec 12, 2021 at 02:54:45PM +0100, Greg KH wrote:
> On Fri, Dec 10, 2021 at 03:48:00PM -0800, Eric Biggers wrote:
> > Backport the aio poll fixes to 5.10.  This resolves a conflict in
> > aio_poll_wake() in patch 4.  It's a "trivial" conflict, but I'm sending
> > this to make sure it doesn't get dropped.
> > 
> > Eric Biggers (5):
> >   wait: add wake_up_pollfree()
> >   binder: use wake_up_pollfree()
> >   signalfd: use wake_up_pollfree()
> >   aio: keep poll requests on waitqueue until completed
> >   aio: fix use-after-free due to missing POLLFREE handling
> > 
> >  drivers/android/binder.c        |  21 ++--
> >  fs/aio.c                        | 184 ++++++++++++++++++++++++++------
> >  fs/signalfd.c                   |  12 +--
> >  include/linux/wait.h            |  26 +++++
> >  include/uapi/asm-generic/poll.h |   2 +-
> >  kernel/sched/wait.c             |   7 ++
> >  6 files changed, 195 insertions(+), 57 deletions(-)
> > 
> > -- 
> > 2.34.1
> > 
> 
> Thanks for all of the backports, much appreciated and now queued up.
> 
> greg k-h

Thanks!  Can you apply the following commit to 5.15-stable too?  I missed that
it's needed in 5.15:

	commit 4b3749865374899e115aa8c48681709b086fe6d3
	Author: Xie Yongji <xieyongji@bytedance.com>
	Date:   Mon Sep 13 19:19:28 2021 +0800

	    aio: Fix incorrect usage of eventfd_signal_allowed()

Re: [PATCH 5.10 0/5] aio poll fixes for 5.10

[Greg KH]

On Sun, Dec 12, 2021 at 11:46:03AM -0800, Eric Biggers wrote:
> On Sun, Dec 12, 2021 at 02:54:45PM +0100, Greg KH wrote:
> > On Fri, Dec 10, 2021 at 03:48:00PM -0800, Eric Biggers wrote:
> > > Backport the aio poll fixes to 5.10.  This resolves a conflict in
> > > aio_poll_wake() in patch 4.  It's a "trivial" conflict, but I'm sending
> > > this to make sure it doesn't get dropped.
> > > 
> > > Eric Biggers (5):
> > >   wait: add wake_up_pollfree()
> > >   binder: use wake_up_pollfree()
> > >   signalfd: use wake_up_pollfree()
> > >   aio: keep poll requests on waitqueue until completed
> > >   aio: fix use-after-free due to missing POLLFREE handling
> > > 
> > >  drivers/android/binder.c        |  21 ++--
> > >  fs/aio.c                        | 184 ++++++++++++++++++++++++++------
> > >  fs/signalfd.c                   |  12 +--
> > >  include/linux/wait.h            |  26 +++++
> > >  include/uapi/asm-generic/poll.h |   2 +-
> > >  kernel/sched/wait.c             |   7 ++
> > >  6 files changed, 195 insertions(+), 57 deletions(-)
> > > 
> > > -- 
> > > 2.34.1
> > > 
> > 
> > Thanks for all of the backports, much appreciated and now queued up.
> > 
> > greg k-h
> 
> Thanks!  Can you apply the following commit to 5.15-stable too?  I missed that
> it's needed in 5.15:
> 
> 	commit 4b3749865374899e115aa8c48681709b086fe6d3
> 	Author: Xie Yongji <xieyongji@bytedance.com>
> 	Date:   Mon Sep 13 19:19:28 2021 +0800
> 
> 	    aio: Fix incorrect usage of eventfd_signal_allowed()

Now queued up, thanks.

greg k-h