Skip to contentSkip to conversation

io_uring: REQ_F_PREPPED race condition with punting to workers

PATCH In review Stefan Bühler 2 participants 4 messages May 3, 2019 – May 13, 2019

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

Hi,

if the initial operation returns EAGAIN (and REQ_F_NOWAIT) is not set, io_submit_sqe copies the SQE for processing in a worker.

The worker will then read from the SQE copy to determine (some) parameters for operations, but not all of those parameters will be validated again, as the initial operation sets REQ_F_PREPPED.

So between the initial operation and the memcpy is a race in which the application could change the SQE: for example it could change from IORING_OP_FSYNC to IORING_OP_READV, which would result in broken kiocb data afaict.

The only way around that I can see right now is copying the SQE in io_submit_sqe (moving the call to io_cqring_add_event to io_submit_sqe should simplify this afaict): does that sound acceptable?

cheers,
Stefan

maintainer
#1 lore raw
Quoted 9 lines · Stefan Bühler

Hi,

if the initial operation returns EAGAIN (and REQ_F_NOWAIT) is not set, io_submit_sqe copies the SQE for processing in a worker.

The worker will then read from the SQE copy to determine (some) parameters for operations, but not all of those parameters will be validated again, as the initial operation sets REQ_F_PREPPED.

Quoted 8 lines

So between the initial operation and the memcpy is a race in which the application could change the SQE: for example it could change from IORING_OP_FSYNC to IORING_OP_READV, which would result in broken kiocb data afaict.

The only way around that I can see right now is copying the SQE in io_submit_sqe (moving the call to io_cqring_add_event to io_submit_sqe should simplify this afaict): does that sound acceptable?

I'd be inclined to just fold the prep into the regular handling. The only prep routine that does any significant work is the read/write one, and if we're punting to async anyway, it's not a huge hit.

If we do that, then we can get rid of the PREPPED flag and the separate need to call io_prep_xxx() for the command type.

Signature · 1 line
Jens Axboe
#2 lore raw

PATCH 1/1 io_uring: fix race condition reading SQE data

When punting to workers the SQE gets copied after the initial try. There is a race condition between reading SQE data for the initial try and copying it for punting it to the workers.

For example io_rw_done calls kiocb->ki_complete even if it was prepared for IORING_OP_FSYNC (and would be NULL).

The easiest solution for now is to alway prepare again in the worker.

req->file is safe to prepare though as long as it is checked before use.

---
 fs/io_uring.c | 17 ++---------------
 1 file changed, 2 insertions(+), 15 deletions(-)
fs/io_uring.c +2 -15
index 48ea3977012a..576d9c652b4c 100644
@@ -329,9 +329,8 @@ struct io_kiocb {
 #define REQ_F_IOPOLL_COMPLETED	2	/* polled IO has completed */
 #define REQ_F_FIXED_FILE	4	/* ctx owns file */
 #define REQ_F_SEQ_PREV		8	/* sequential with previous */
-#define REQ_F_PREPPED		16	/* prep already done */
-#define REQ_F_IO_DRAIN		32	/* drain existing IO first */
-#define REQ_F_IO_DRAINED	64	/* drain done */
+#define REQ_F_IO_DRAIN		16	/* drain existing IO first */
+#define REQ_F_IO_DRAINED	32	/* drain done */
 	u64			user_data;
 	u32			error;	/* iopoll result from callback */
 	u32			sequence;
@@ -896,9 +895,6 @@ static int io_prep_rw(struct io_kiocb *req, const struct sqe_submit *s,
 
 	if (!req->file)
 		return -EBADF;
-	/* For -EAGAIN retry, everything is already prepped */
-	if (req->flags & REQ_F_PREPPED)
-		return 0;
 
 	if (force_nonblock && !io_file_supports_async(req->file))
 		force_nonblock = false;
@@ -941,7 +937,6 @@ static int io_prep_rw(struct io_kiocb *req, const struct sqe_submit *s,
 			return -EINVAL;
 		kiocb->ki_complete = io_complete_rw;
 	}
-	req->flags |= REQ_F_PREPPED;
 	return 0;
 }
 
@@ -1227,16 +1222,12 @@ static int io_prep_fsync(struct io_kiocb *req, const struct io_uring_sqe *sqe)
 
 	if (!req->file)
 		return -EBADF;
-	/* Prep already done (EAGAIN retry) */
-	if (req->flags & REQ_F_PREPPED)
-		return 0;
 
 	if (unlikely(ctx->flags & IORING_SETUP_IOPOLL))
 		return -EINVAL;
 	if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index))
 		return -EINVAL;
 
-	req->flags |= REQ_F_PREPPED;
 	return 0;
 }
 
@@ -1277,16 +1268,12 @@ static int io_prep_sfr(struct io_kiocb *req, const struct io_uring_sqe *sqe)
 
 	if (!req->file)
 		return -EBADF;
-	/* Prep already done (EAGAIN retry) */
-	if (req->flags & REQ_F_PREPPED)
-		return 0;
 
 	if (unlikely(ctx->flags & IORING_SETUP_IOPOLL))
 		return -EINVAL;
 	if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index))
 		return -EINVAL;
 
-	req->flags |= REQ_F_PREPPED;
 	return ret;
 }
 
Signature · 1 line
2.20.1
maintainer
#3 lore raw

Re: [PATCH 1/1] io_uring: fix race condition reading SQE data

Quoted 10 lines · Stefan Bühler

When punting to workers the SQE gets copied after the initial try. There is a race condition between reading SQE data for the initial try and copying it for punting it to the workers.

For example io_rw_done calls kiocb->ki_complete even if it was prepared for IORING_OP_FSYNC (and would be NULL).

The easiest solution for now is to alway prepare again in the worker.

req->file is safe to prepare though as long as it is checked before use.

Looks good, as we discussed a week or so ago. Applied, thanks.

Signature · 1 line
Jens Axboe