[PATCH 1/1] io_uring: sanitise nr_pages for SQ/CQ

[PATCH 1/1] io_uring: sanitise nr_pages for SQ/CQ

[Pavel Begunkov]

WARNING: CPU: 0 PID: 5834 at io_uring/memmap.c:144 io_pin_pages+0x149/0x180 io_uring/memmap.c:144
CPU: 0 UID: 0 PID: 5834 Comm: syz-executor825 Not tainted 6.12.0-next-20241118-syzkaller #0
Call Trace:
 <TASK>
 __io_uaddr_map+0xfb/0x2d0 io_uring/memmap.c:183
 io_rings_map io_uring/io_uring.c:2611 [inline]
 io_allocate_scq_urings+0x1c0/0x650 io_uring/io_uring.c:3470
 io_uring_create+0x5b5/0xc00 io_uring/io_uring.c:3692
 io_uring_setup io_uring/io_uring.c:3781 [inline]
 ...
 </TASK>

Apparently there is a way to request a large enough CQ/SQ so that the
number of pages used doesn't fit into int. Even worse, then it's
truncated further to ushort. Limit them to the type size for now, but
it needs a better follow up.

Cc: stable@vger.kernel.org
Reported-by: syzbot+2159cbb522b02847c053@syzkaller.appspotmail.com
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
---
 io_uring/io_uring.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/io_uring/io_uring.c b/io_uring/io_uring.c
index bfa93888f862..82d217cfdebc 100644
--- a/io_uring/io_uring.c
+++ b/io_uring/io_uring.c
@@ -3482,7 +3482,7 @@ static __cold int io_allocate_scq_urings(struct io_ring_ctx *ctx,
 
 	size = rings_size(ctx->flags, p->sq_entries, p->cq_entries,
 			  &sq_array_offset);
-	if (size == SIZE_MAX)
+	if (size == SIZE_MAX || (PAGE_ALIGN(size) >> PAGE_SHIFT) > USHRT_MAX)
 		return -EOVERFLOW;
 
 	if (!(ctx->flags & IORING_SETUP_NO_MMAP))
@@ -3505,7 +3505,7 @@ static __cold int io_allocate_scq_urings(struct io_ring_ctx *ctx,
 		size = array_size(2 * sizeof(struct io_uring_sqe), p->sq_entries);
 	else
 		size = array_size(sizeof(struct io_uring_sqe), p->sq_entries);
-	if (size == SIZE_MAX) {
+	if (size == SIZE_MAX || (PAGE_ALIGN(size) >> PAGE_SHIFT) > USHRT_MAX) {
 		io_rings_free(ctx);
 		return -EOVERFLOW;
 	}
-- 
2.47.1

Re: [PATCH 1/1] io_uring: sanitise nr_pages for SQ/CQ

[Pavel Begunkov]

On 11/25/24 23:22, Pavel Begunkov wrote:
> WARNING: CPU: 0 PID: 5834 at io_uring/memmap.c:144 io_pin_pages+0x149/0x180 io_uring/memmap.c:144
> CPU: 0 UID: 0 PID: 5834 Comm: syz-executor825 Not tainted 6.12.0-next-20241118-syzkaller #0
> Call Trace:
>   <TASK>
>   __io_uaddr_map+0xfb/0x2d0 io_uring/memmap.c:183
>   io_rings_map io_uring/io_uring.c:2611 [inline]
>   io_allocate_scq_urings+0x1c0/0x650 io_uring/io_uring.c:3470
>   io_uring_create+0x5b5/0xc00 io_uring/io_uring.c:3692
>   io_uring_setup io_uring/io_uring.c:3781 [inline]
>   ...
>   </TASK>
> 
> Apparently there is a way to request a large enough CQ/SQ so that the
> number of pages used doesn't fit into int. Even worse, then it's
> truncated further to ushort. Limit them to the type size for now, but
> it needs a better follow up.

Nevermind, syz says it doesn't fix it.

-- 
Pavel Begunkov