ublk_drv: cleanup and bugfix
Subsystems (2)
v2 cover ublk_drv: cleanup and bugfix
The following 3 patches are cleanup and bugfix. Patch 1 and 2 simply inline a function and update comments for ublk_drv's aborting machemism.
Patch 3 fix a null-deref bug reported by myself. Ming gives out a patch and I integrate it with more comments on this bug.
From V1:
- reserve comment on __ublk_fail_req() to notify that aborted requests may be issued again.
- only reference ublk_io in branch of not using task_work_add()
ZiyangZhang (3):
ublk_drv: check ubq_daemon_is_dying() in __ublk_rq_task_work()
ublk_drv: update comment for __ublk_fail_req()
ublk_drv: do not add a re-issued request aborted previously to
ioucmd's task_work
drivers/block/ublk_drv.c | 28 ++++++++++++++++++++++------ 1 file changed, 22 insertions(+), 6 deletions(-)
Signature · 1 line
2.27.0
PATCH v2 1/3 ublk_drv: check ubq_daemon_is_dying() in __ublk_rq_task_work()
Replace direct check on PF_EXITING in __ublk_rq_task_work() by the existing wrapper. Also inline ubq_daemon_is_dying().
--- drivers/block/ublk_drv.c | 5 ++--- 1 file changed, 2 insertions(+), 3 deletions(-)
drivers/block/ublk_drv.c +2 -3
@@ -555,7 +555,7 @@ static inline struct ublk_uring_cmd_pdu *ublk_get_uring_cmd_pdu(return (struct ublk_uring_cmd_pdu *)&ioucmd->pdu; } -static bool ubq_daemon_is_dying (struct ublk_queue *ubq) +static inline bool ubq_daemon_is_dying (struct ublk_queue *ubq) {return ubq->ubq_daemon->flags &PF_EXITING ; }
@@ -644,8 +644,7 @@ static inline void __ublk_rq_task_work(struct request *req)struct ublk_device *ub = ubq->dev;int tag = req->tag;struct ublk_io *io = &ubq->ios[tag]; -bool task_exiting = current != ubq->ubq_daemon || - (current->flags &PF_EXITING ); +bool task_exiting = current != ubq->ubq_daemon ||ubq_daemon_is_dying (ubq);unsigned int mapped_bytes; pr_devel("%s: complete: op %d, qid %d tag %d io_flags %x addr %llx\n",
Signature · 1 line
2.27.0
PATCH v2 2/3 ublk_drv: update comment for __ublk_fail_req()
In reply to ZiyangZhangSince __ublk_rq_task_work always fails requests immediately during exiting, __ublk_fail_req() is only called from abort context during exiting. So lock is unnecessary.
--- drivers/block/ublk_drv.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)
drivers/block/ublk_drv.c +3 -2
@@ -605,8 +605,9 @@ static void ublk_complete_rq(struct request *req) }/* -* __ublk_fail_req() may be called from abort context or ->ubq_daemon -* context during exiting, so lock is required. +* Since __ublk_rq_task_work always fails requests immediately during +* exiting, __ublk_fail_req() is only called from abort context during +* exiting. So lock is unnecessary. * * Also aborting may not be started yet, keep in mind that one failed * request may be issued by block layer again.
Signature · 1 line
2.27.0
Re: [PATCH V2 2/3] ublk_drv: update comment for __ublk_fail_req()
Quoted 28 lines · ZiyangZhang
Since
__ublk_rq_task_workalways fails requests immediately during exiting,__ublk_fail_req()is only called from abort context during exiting. So lock is unnecessary.Signed-off-by ZiyangZhang <[email protected]>--- drivers/block/ublk_drv.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)drivers/block/ublk_drv.c +3
-2@@ -605,8 +605,9 @@ static void ublk_complete_rq(struct request *req) }/* -* __ublk_fail_req() may be called from abort context or ->ubq_daemon -* context during exiting, so lock is required. +* Since __ublk_rq_task_work always fails requests immediately during +* exiting, __ublk_fail_req() is only called from abort context during +* exiting. So lock is unnecessary. * * Also aborting may not be started yet, keep in mind that one failed * request may be issued by block layer again. Signature · 1 line
2.27.0
Signature · 1 line
Ming
PATCH v2 3/3 ublk_drv: do not add a re-issued request aborted previously to ioucmd's task_work
In reply to ZiyangZhangIn ublk_queue_rq(), Assume current request is a re-issued request aborted previously in monitor_work because the ubq_daemon(ioucmd's task) is PF_EXITING. For this request, we cannot call io_uring_cmd_complete_in_task() anymore because at that moment io_uring context may be freed in case that no inflight ioucmd exists. Otherwise, we may cause null-deref in ctx->fallback_work.
Add a check on UBLK_IO_FLAG_ABORTED to prevent the above situation. This check is safe and makes sense.
Note: monitor_work sets UBLK_IO_FLAG_ABORTED and ends this request (releasing the tag). Then the request is restarted(allocating the tag) and we are here. Since releasing/allocating a tag implies smp_mb(), finding UBLK_IO_FLAG_ABORTED guarantees that here is a re-issued request aborted previously.
--- drivers/block/ublk_drv.c | 18 +++++++++++++++++- 1 file changed, 17 insertions(+), 1 deletion(-)
drivers/block/ublk_drv.c +17 -1
@@ -756,9 +756,25 @@ static blk_status_t ublk_queue_rq(struct blk_mq_hw_ctx *hctx,if (task_work_add (ubq->ubq_daemon, &data->work, notify_mode)) goto fail; }else { -struct io_uring_cmd *cmd = ubq->ios[rq->tag].cmd; +struct ublk_io *io = &ubq->ios[rq->tag]; +struct io_uring_cmd *cmd = io->cmd;struct ublk_uring_cmd_pdu *pdu =ublk_get_uring_cmd_pdu (cmd); +/* +* If the check pass, we know that this is a re-issued request aborted +* previously in monitor_work because the ubq_daemon(cmd's task) is +* PF_EXITING. We cannot call io_uring_cmd_complete_in_task() anymore +* because this ioucmd's io_uring context may be freed now if no inflight +* ioucmd exists. Otherwise we may cause null-deref in ctx->fallback_work. +* +* Note: monitor_work sets UBLK_IO_FLAG_ABORTED and ends this request(releasing +* the tag). Then the request is re-started(allocating the tag) and we are here. +* Since releasing/allocating a tag implies smp_mb(), finding UBLK_IO_FLAG_ABORTED +* guarantees that here is a re-issued request aborted previously. +*/ +if ((io->flags &UBLK_IO_FLAG_ABORTED )) + goto fail; + pdu->req = rq;io_uring_cmd_complete_in_task (cmd, ublk_rq_task_work_cb); }
Signature · 1 line
2.27.0
Re: [PATCH V2 3/3] ublk_drv: do not add a re-issued request aborted previously to ioucmd's task_work
Quoted 10 lines · ZiyangZhang
In
ublk_queue_rq(), Assume current request is a re-issued request aborted previously inmonitor_workbecause theubq_daemon(ioucmd's task)isPF_EXITING. For this request, we cannot callio_uring_cmd_complete_in_task()anymore because at that momentio_uringcontext may be freed in case that no inflight ioucmd exists. Otherwise, we may cause null-deref inctx->fallback_work.Add a check on
UBLK_IO_FLAG_ABORTEDto prevent the above situation. This check is safe and makes sense.
Quoted 8 lines
Note:
monitor_worksetsUBLK_IO_FLAG_ABORTEDand ends this request (releasing the tag). Then the request is restarted(allocating the tag) and we are here. Since releasing/allocating a tag impliessmp_mb(), findingUBLK_IO_FLAG_ABORTEDguarantees that here is a re-issued request aborted previously.Suggested-by Ming Lei <[email protected]>Signed-off-by ZiyangZhang <[email protected]>
Thanks,
Ming
Quoted 8 lines · ZiyangZhang
The following 3 patches are cleanup and bugfix. Patch 1 and 2 simply inline a function and update comments for ublk_drv's aborting machemism.
Patch 3 fix a null-deref bug reported by myself. Ming gives out a patch and I integrate it with more comments on this bug.
[...]
Applied, thanks!
[1/3] ublk_drv: check ubq_daemon_is_dying() in __ublk_rq_task_work()
commit: 966120b51a245c9ff5857c5b169310c248e0ae87
[2/3] ublk_drv: update comment for __ublk_fail_req()
commit: bb24174754afc5a7d185ca5406dcfbc608cdf157
[3/3] ublk_drv: do not add a re-issued request aborted previously to ioucmd's task_work
commit: e6190dd0031d335c22586d34ef898301ed20f230Best regards,
Signature · 1 line
Jens Axboe