[PATCH v5] aio: Add support for the POLLFREE

[Ramji Jiyani]

Add support for the POLLFREE flag to force complete iocb inline in
aio_poll_wake(). A thread may use it to signal it's exit and/or request
to cleanup while pending poll request. In this case, aio_poll_wake()
needs to make sure it doesn't keep any reference to the queue entry
before returning from wake to avoid possible use after free via
poll_cancel() path.

UAF issue was found during binder and aio interactions in certain
sequence of events [1].

The POLLFREE flag is no more exclusive to the epoll and is being
shared with the aio. Remove comment from poll.h to avoid confusion.

[1] https://lore.kernel.org/r/CAKUd0B_TCXRY4h1hTztfwWbNSFQqsudDLn2S_28csgWZmZAG3Q@xxxxxxxxxxxxxx/

Fixes: af5c72b1fc7a ("Fix aio_poll() races")
Signed-off-by: Ramji Jiyani <ramjiyani@xxxxxxxxxx>
Reviewed-by: Jeff Moyer <jmoyer@xxxxxxxxxx>
Reviewed-by: Christoph Hellwig <hch@xxxxxx>
Cc: stable@xxxxxxxxxxxxxxx # 4.19+
---
Changes since v1:
- Removed parenthesis around POLLFREE macro definition as per review.
- Updated description to refer UAF issue discussion this patch fixes.
- Updated description to remove reference to parenthesis change.
- Added Reviewed-by from Jeff Moyer

Changes since v2:
- Added Fixes tag.
- Added stable tag for backporting on 4.19+ LTS releases

Changes since v3:
- Updated patch description
- Updated Fixes tag to issue manifestation origin

Changes since v4:
- Added Reviewed-by from Christoph Hellwig
---
 fs/aio.c                        | 45 ++++++++++++++++++---------------
 include/uapi/asm-generic/poll.h |  2 +-
 2 files changed, 26 insertions(+), 21 deletions(-)

diff --git a/fs/aio.c b/fs/aio.c
index 51b08ab01dff..5d539c05df42 100644
--- a/fs/aio.c
+++ b/fs/aio.c
@@ -1674,6 +1674,7 @@ static int aio_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync,
 {
 	struct poll_iocb *req = container_of(wait, struct poll_iocb, wait);
 	struct aio_kiocb *iocb = container_of(req, struct aio_kiocb, poll);
+	struct kioctx *ctx = iocb->ki_ctx;
 	__poll_t mask = key_to_poll(key);
 	unsigned long flags;
 
@@ -1683,29 +1684,33 @@ static int aio_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync,
 
 	list_del_init(&req->wait.entry);
 
-	if (mask && spin_trylock_irqsave(&iocb->ki_ctx->ctx_lock, flags)) {
-		struct kioctx *ctx = iocb->ki_ctx;
+	/*
+	 * 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 (mask & POLLFREE) {
+		/* Force complete iocb inline to remove refs to deleted entry */
+		spin_lock_irqsave(&ctx->ctx_lock, flags);
+	} else if (!(mask && spin_trylock_irqsave(&ctx->ctx_lock, flags))) {
+		/* Can't complete iocb inline; schedule for later */
+		schedule_work(&req->work);
+		return 1;
+	}
 
-		/*
-		 * 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(&iocb->ki_list);
-		iocb->ki_res.res = mangle_poll(mask);
-		req->done = true;
-		if (iocb->ki_eventfd && eventfd_signal_allowed()) {
-			iocb = NULL;
-			INIT_WORK(&req->work, aio_poll_put_work);
-			schedule_work(&req->work);
-		}
-		spin_unlock_irqrestore(&ctx->ctx_lock, flags);
-		if (iocb)
-			iocb_put(iocb);
-	} else {
+	/* complete iocb inline */
+	list_del(&iocb->ki_list);
+	iocb->ki_res.res = mangle_poll(mask);
+	req->done = true;
+	if (iocb->ki_eventfd && eventfd_signal_allowed()) {
+		iocb = NULL;
+		INIT_WORK(&req->work, aio_poll_put_work);
 		schedule_work(&req->work);
 	}
+	spin_unlock_irqrestore(&ctx->ctx_lock, flags);
+	if (iocb)
+		iocb_put(iocb);
+
 	return 1;
 }
 
diff --git a/include/uapi/asm-generic/poll.h b/include/uapi/asm-generic/poll.h
index 41b509f410bf..f9c520ce4bf4 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
 
-- 
2.33.0.1079.g6e70778dc9-goog