Processing links, io_submit_sqe() prepares requests, drops sqes, and
passes them with sqe=NULL to io_queue_sqe(). There IOSQE_DRAIN and/or
IOSQE_ASYNC requests will go through the same prep, which doesn't expect
sqe=NULL and fail with NULL pointer deference.
Always do full prepare including io_alloc_async_ctx() for linked
requests, and then it can skip the second preparation.
Signed-off-by: Pavel Begunkov <[email protected]>
---
fs/io_uring.c | 8 ++++++++
1 file changed, 8 insertions(+)
diff --git a/fs/io_uring.c b/fs/io_uring.c
index 55afae6f0cf4..9d43efbec960 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -4813,6 +4813,9 @@ static int io_req_defer_prep(struct io_kiocb *req,
{
ssize_t ret = 0;
+ if (!sqe)
+ return 0;
+
if (io_op_defs[req->opcode].file_table) {
ret = io_grab_files(req);
if (unlikely(ret))
@@ -5655,6 +5658,11 @@ static bool io_submit_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe,
if (sqe_flags & (IOSQE_IO_LINK|IOSQE_IO_HARDLINK)) {
req->flags |= REQ_F_LINK;
INIT_LIST_HEAD(&req->link_list);
+
+ if (io_alloc_async_ctx(req)) {
+ ret = -EAGAIN;
+ goto err_req;
+ }
ret = io_req_defer_prep(req, sqe);
if (ret)
req->flags |= REQ_F_FAIL_LINK;
--
2.24.0
On 13/03/2020 22:29, Pavel Begunkov wrote:
> Processing links, io_submit_sqe() prepares requests, drops sqes, and
> passes them with sqe=NULL to io_queue_sqe(). There IOSQE_DRAIN and/or
> IOSQE_ASYNC requests will go through the same prep, which doesn't expect
> sqe=NULL and fail with NULL pointer deference.
>
> Always do full prepare including io_alloc_async_ctx() for linked
> requests, and then it can skip the second preparation.
BTW, linked_timeout test fails for a good reason. The test passes NULL buffer to
writev and expects it to -EFAULT in io_req_defer_prep(). However,
io_submit_sqe() catches this case (see head of a link case), sets
REQ_F_FAIL_LINK and allows it to fail with -ECANCELED in io_queue_link_head().
> Signed-off-by: Pavel Begunkov <[email protected]>
> ---
> fs/io_uring.c | 8 ++++++++
> 1 file changed, 8 insertions(+)
>
> diff --git a/fs/io_uring.c b/fs/io_uring.c
> index 55afae6f0cf4..9d43efbec960 100644
> --- a/fs/io_uring.c
> +++ b/fs/io_uring.c
> @@ -4813,6 +4813,9 @@ static int io_req_defer_prep(struct io_kiocb *req,
> {
> ssize_t ret = 0;
>
> + if (!sqe)
> + return 0;
> +
> if (io_op_defs[req->opcode].file_table) {
> ret = io_grab_files(req);
> if (unlikely(ret))
> @@ -5655,6 +5658,11 @@ static bool io_submit_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe,
> if (sqe_flags & (IOSQE_IO_LINK|IOSQE_IO_HARDLINK)) {
> req->flags |= REQ_F_LINK;
> INIT_LIST_HEAD(&req->link_list);
> +
> + if (io_alloc_async_ctx(req)) {
> + ret = -EAGAIN;
> + goto err_req;
> + }
> ret = io_req_defer_prep(req, sqe);
> if (ret)
> req->flags |= REQ_F_FAIL_LINK;
>
--
Pavel Begunkov
On 13/03/2020 22:29, Pavel Begunkov wrote:
> Processing links, io_submit_sqe() prepares requests, drops sqes, and
> passes them with sqe=NULL to io_queue_sqe(). There IOSQE_DRAIN and/or
> IOSQE_ASYNC requests will go through the same prep, which doesn't expect
> sqe=NULL and fail with NULL pointer deference.
>
> Always do full prepare including io_alloc_async_ctx() for linked
> requests, and then it can skip the second preparation.
Hmm, found unreliably failing the across-fork test. I don't know whether it's
this patch specific, but need to take a look there first.
>
> Signed-off-by: Pavel Begunkov <[email protected]>
> ---
> fs/io_uring.c | 8 ++++++++
> 1 file changed, 8 insertions(+)
>
> diff --git a/fs/io_uring.c b/fs/io_uring.c
> index 55afae6f0cf4..9d43efbec960 100644
> --- a/fs/io_uring.c
> +++ b/fs/io_uring.c
> @@ -4813,6 +4813,9 @@ static int io_req_defer_prep(struct io_kiocb *req,
> {
> ssize_t ret = 0;
>
> + if (!sqe)
> + return 0;
> +
> if (io_op_defs[req->opcode].file_table) {
> ret = io_grab_files(req);
> if (unlikely(ret))
> @@ -5655,6 +5658,11 @@ static bool io_submit_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe,
> if (sqe_flags & (IOSQE_IO_LINK|IOSQE_IO_HARDLINK)) {
> req->flags |= REQ_F_LINK;
> INIT_LIST_HEAD(&req->link_list);
> +
> + if (io_alloc_async_ctx(req)) {
> + ret = -EAGAIN;
> + goto err_req;
> + }
> ret = io_req_defer_prep(req, sqe);
> if (ret)
> req->flags |= REQ_F_FAIL_LINK;
>
--
Pavel Begunkov
On 13/03/2020 23:28, Pavel Begunkov wrote:
> Hmm, found unreliably failing the across-fork test. I don't know whether it's
> this patch specific, but need to take a look there first.
It's good to go, just used outdated tests.
The reproducer is attached.
>
>>
>> Signed-off-by: Pavel Begunkov <[email protected]>
>> ---
>> fs/io_uring.c | 8 ++++++++
>> 1 file changed, 8 insertions(+)
>>
>> diff --git a/fs/io_uring.c b/fs/io_uring.c
>> index 55afae6f0cf4..9d43efbec960 100644
>> --- a/fs/io_uring.c
>> +++ b/fs/io_uring.c
>> @@ -4813,6 +4813,9 @@ static int io_req_defer_prep(struct io_kiocb *req,
>> {
>> ssize_t ret = 0;
>>
>> + if (!sqe)
>> + return 0;
>> +
>> if (io_op_defs[req->opcode].file_table) {
>> ret = io_grab_files(req);
>> if (unlikely(ret))
>> @@ -5655,6 +5658,11 @@ static bool io_submit_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe,
>> if (sqe_flags & (IOSQE_IO_LINK|IOSQE_IO_HARDLINK)) {
>> req->flags |= REQ_F_LINK;
>> INIT_LIST_HEAD(&req->link_list);
>> +
>> + if (io_alloc_async_ctx(req)) {
>> + ret = -EAGAIN;
>> + goto err_req;
>> + }
>> ret = io_req_defer_prep(req, sqe);
>> if (ret)
>> req->flags |= REQ_F_FAIL_LINK;
>>
>
--
Pavel Begunkov
On 3/13/20 3:29 PM, Pavel Begunkov wrote:
> On 13/03/2020 23:28, Pavel Begunkov wrote:
>> Hmm, found unreliably failing the across-fork test. I don't know whether it's
>> this patch specific, but need to take a look there first.
>
> It's good to go, just used outdated tests.
> The reproducer is attached.
I integrated this into the existing read-write in liburing, thanks.
--
Jens Axboe
On 3/13/20 1:29 PM, Pavel Begunkov wrote:
> Processing links, io_submit_sqe() prepares requests, drops sqes, and
> passes them with sqe=NULL to io_queue_sqe(). There IOSQE_DRAIN and/or
> IOSQE_ASYNC requests will go through the same prep, which doesn't expect
> sqe=NULL and fail with NULL pointer deference.
>
> Always do full prepare including io_alloc_async_ctx() for linked
> requests, and then it can skip the second preparation.
Thanks, applied.
--
Jens Axboe