io_uring: complete request via task work in case of DEFER_TASKRUN
PATCH 1/1 io_uring: complete request via task work in case of DEFER_TASKRUN
So far io_req_complete_post() only covers DEFER_TASKRUN by completing request via task work when the request is completed from IOWQ.
However, uring command could be completed from any context, and if io uring is setup with DEFER_TASKRUN, the command is required to be completed from current context, otherwise wait on IORING_ENTER_GETEVENTS can't be wakeup, and may hang forever.
The issue can be observed on removing ublk device, but turns out it is one generic issue for uring command & DEFER_TASKRUN, so solve it in io_uring core code.
--- io_uring/io_uring.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
io_uring/io_uring.c +1 -1
@@ -1012,7 +1012,7 @@ static void __io_req_complete_post(struct io_kiocb *req, unsigned issue_flags)void io_req_complete_post (struct io_kiocb *req,unsigned issue_flags) { -if (req->ctx->task_complete && (issue_flags &IO_URING_F_IOWQ )) { +if (req->ctx->task_complete && req->ctx->submitter_task != current) { req->io_task_work.func = io_req_task_complete;io_req_task_work_add (req); } else if (!(issue_flags & IO_URING_F_UNLOCKED) ||
Signature · 1 line
2.38.1
Quoted 11 lines · Ming Lei
So far
io_req_complete_post()only coversDEFER_TASKRUNby completing request via task work when the request is completed from IOWQ.However, uring command could be completed from any context, and if io uring is setup with
DEFER_TASKRUN, the command is required to be completed from current context, otherwise wait onIORING_ENTER_GETEVENTScan't be wakeup, and may hang forever.The issue can be observed on removing ublk device, but turns out it is one generic issue for uring command &
DEFER_TASKRUN, so solve it inio_uringcore code.
Thanks for sharing, this has been fine for nvme-passthrough side though. We usually test with DEFER_TASKRUN option, as both fio and t/io_uring set the option.
Quoted 13 lines
Reported-by Jens Axboe <[email protected]>Cc Kanchan Joshi <[email protected]>Signed-off-by Ming Lei <[email protected]>---
io_uring/io_uring.c| 2 +-1 file changed, 1 insertion(+), 1 deletion(-)io_uring/io_uring.c +0
-0
io_uring/io_uring.c static void __io_req_complete_post(struct io_kiocb *req, unsigned issue_flags) · line 1017
Quoted 7 lines
io_uring/io_uring.c +1
-1@@ -1012,7 +1012,7 @@ static void __io_req_complete_post(struct io_kiocb *req, unsigned issue_flags)void io_req_complete_post (struct io_kiocb *req,unsigned issue_flags) { -if (req->ctx->task_complete && (issue_flags &IO_URING_F_IOWQ )) { +if (req->ctx->task_complete && req->ctx->submitter_task != current) { req->io_task_work.func = io_req_task_complete;io_req_task_work_add (req);
In nvme-side, we always complete in task context, so this seems bit hard to produce.
But this patch ensures that task-work is setup if it is needed, and caller/driver did not get to set that explicitly. So looks fine to me. FWIW, I do not see regression in nvme tests.
Quoted 9 lines · Ming Lei
So far
io_req_complete_post()only coversDEFER_TASKRUNby completing request via task work when the request is completed from IOWQ.However, uring command could be completed from any context, and if io uring is setup with
DEFER_TASKRUN, the command is required to be completed from current context, otherwise wait onIORING_ENTER_GETEVENTScan't be wakeup, and may hang forever.[...]
Applied, thanks!
[1/1] io_uring: complete request via task work in case of DEFER_TASKRUN
commit: 860e1c7f8b0b43fbf91b4d689adfaa13adb89452
Best regards,
Signature · 1 line
Jens Axboe
Quoted 7 lines · Ming Lei
So far
io_req_complete_post()only coversDEFER_TASKRUNby completing request via task work when the request is completed from IOWQ.However, uring command could be completed from any context, and if io uring is setup with
DEFER_TASKRUN, the command is required to be completed from current context, otherwise wait onIORING_ENTER_GETEVENTScan't be wakeup, and may hang forever.
fwiw, there is one legit exception, when the task is half dead task_work will be executed by a kthread. It should be fine as it locks the ctx down, but I can't help but wonder whether it's only ublk_cancel_queue() affected or there are more places in ublk?
One more thing, cmds should not be setting issue_flags but only forwarding what the core io_uring code passed, it'll get tons of bugs in no time otherwise.
static void ublk_cancel_queue(struct ublk_queue *ubq)
{
...
io_uring_cmd_done(io->cmd, UBLK_IO_RES_ABORT, 0,
IO_URING_F_UNLOCKED);
}Can we replace it with task_work? It should be cold, and I assume ublk_cancel_queue() doesn't assume that all requests will put down by the end of the function as io_uring_cmd_done() can offload it in any case.
Signature · 1 line
Pavel Begunkov
Quoted 9 lines · Pavel Begunkov
Quoted 7 lines · Ming Lei
So far
io_req_complete_post()only coversDEFER_TASKRUNby completing request via task work when the request is completed from IOWQ.However, uring command could be completed from any context, and if io uring is setup with
DEFER_TASKRUN, the command is required to be completed from current context, otherwise wait onIORING_ENTER_GETEVENTScan't be wakeup, and may hang forever.
Quoted 4 lines
fwiw, there is one legit exception, when the task is half dead
task_workwill be executed by a kthread. It should be fine as it locks the ctx down, but I can't help but wonder whether it's onlyublk_cancel_queue()affected or there are more places in ublk?
No, it isn't.
It isn't triggered on nvme-pt just because command is always done in task context.
And we know more uring command cases are coming.
Quoted 4 lines
One more thing, cmds should not be setting
issue_flagsbut only forwarding what the coreio_uringcode passed, it'll get tons of bugs in no time otherwise.
Here io_uring_cmd_done() is changed to this way recently, and it could be another topic.
Quoted 8 lines
static void
ublk_cancel_queue(struct ublk_queue *ubq){ ... io_uring_cmd_done(io->cmd, UBLK_IO_RES_ABORT, 0, IO_URING_F_UNLOCKED); }
Quoted 4 lines
Can we replace it with
task_work? It should be cold, and I assumeublk_cancel_queue()doesn't assume that all requests will put down by the end of the function asio_uring_cmd_done()can offload it in any case.
But it isn't specific for ublk, any caller of io_uring_cmd_done() has such issue since io_uring_cmd_done() is one generic API.
thanks,
Ming
Quoted 15 lines · Ming Lei
Quoted 9 lines · Pavel Begunkov
Quoted 7 lines · Ming Lei
So far
io_req_complete_post()only coversDEFER_TASKRUNby completing request via task work when the request is completed from IOWQ.However, uring command could be completed from any context, and if io uring is setup with
DEFER_TASKRUN, the command is required to be completed from current context, otherwise wait onIORING_ENTER_GETEVENTScan't be wakeup, and may hang forever.Quoted 4 lines
fwiw, there is one legit exception, when the task is half dead
task_workwill be executed by a kthread. It should be fine as it locks the ctx down, but I can't help but wonder whether it's onlyublk_cancel_queue()affected or there are more places in ublk?
Quoted 6 lines
No, it isn't.
It isn't triggered on nvme-pt just because command is always done in task context.
And we know more uring command cases are coming.
FWIW, the model I had in mind (in initial days) was this -
1) io_uring_cmd_done is a simple API, it just posts one/two results into reuglar/big SQE
2) for anything complex completion (that requires task-work), it will use another API io_uring_cmd_complete_in_task with the provider-specific callback (that will call above simple API eventually)
Quoted 21 lines · Kanchan Joshi
Quoted 20 lines · Ming Lei
Quoted 9 lines · Pavel Begunkov
Quoted 7 lines · Ming Lei
So far
io_req_complete_post()only coversDEFER_TASKRUNby completing request via task work when the request is completed from IOWQ.However, uring command could be completed from any context, and if io uring is setup with
DEFER_TASKRUN, the command is required to be completed from current context, otherwise wait onIORING_ENTER_GETEVENTScan't be wakeup, and may hang forever.Quoted 4 lines
fwiw, there is one legit exception, when the task is half dead
task_workwill be executed by a kthread. It should be fine as it locks the ctx down, but I can't help but wonder whether it's onlyublk_cancel_queue()affected or there are more places in ublk?No, it isn't.
It isn't triggered on nvme-pt just because command is always done in task context.
Quoted 8 lines
Quoted 1 line
And we know more uring command cases are coming.
FWIW, the model I had in mind (in initial days) was this -
1)io_uring_cmd_doneis a simple API, it just posts one/two results into reuglar/big SQE
2) for anything complex completion (that requires task-work), it will use another APIio_uring_cmd_complete_in_taskwith the provider-specific callback (that will call above simple API eventually)
IMO, the current two APIs type are fine, from ublk viewpoint at least.
io_uring setup/setting is transparent/invisible to driver, and it is reasonable for the two interfaces to hide any io_uring implementation details.
Meantime driver should be free to choose either of the two.
Thanks,
Ming
Quoted 15 lines · Ming Lei
Quoted 9 lines · Pavel Begunkov
Quoted 7 lines · Ming Lei
So far
io_req_complete_post()only coversDEFER_TASKRUNby completing request via task work when the request is completed from IOWQ.However, uring command could be completed from any context, and if io uring is setup with
DEFER_TASKRUN, the command is required to be completed from current context, otherwise wait onIORING_ENTER_GETEVENTScan't be wakeup, and may hang forever.Quoted 4 lines
fwiw, there is one legit exception, when the task is half dead
task_workwill be executed by a kthread. It should be fine as it locks the ctx down, but I can't help but wonder whether it's onlyublk_cancel_queue()affected or there are more places in ublk?
Quoted 6 lines
No, it isn't.
It isn't triggered on nvme-pt just because command is always done in task context.
And we know more uring command cases are coming.
Because all requests and cmds but ublk complete it from another task, ublk is special in this regard.
I have several more not so related questions:
1) Can requests be submitted by some other task than ->ubq_daemon? Looking at
static int ublk_ch_uring_cmd(struct io_uring_cmd *cmd, unsigned int issue_flags)
{
...
if (ubq->ubq_daemon && ubq->ubq_daemon != current)
goto out;
}ublk_queue_cmd() avoiding io_uring way of delivery and using raw task_work doesn't seem great. Especially with TWA_SIGNAL_NO_IPI.
2) What the purpose of the two lines below? I see how UBLK_F_URING_CMD_COMP_IN_TASK is used, but don't understand why it changes depending on whether it's a module or not.
3) The long comment in ublk_queue_cmd() seems quite scary. If you have a cmd / io_uring request it hold a ctx reference and is always allowed to use io_uring's task_work infra like io_uring_cmd_complete_in_task(). Why it's different for ublk?
Quoted 7 lines
Quoted 4 lines
One more thing, cmds should not be setting
issue_flagsbut only forwarding what the coreio_uringcode passed, it'll get tons of bugs in no time otherwise.Here
io_uring_cmd_done()is changed to this way recently, and it could be another topic.
And it's abused, but as you said, not particularly related to this patch.
Quoted 7 lines
Quoted 7 lines
static void
ublk_cancel_queue(struct ublk_queue *ubq){ ... io_uring_cmd_done(io->cmd, UBLK_IO_RES_ABORT, 0, IO_URING_F_UNLOCKED); }
Quoted 7 lines
Quoted 4 lines
Can we replace it with
task_work? It should be cold, and I assumeublk_cancel_queue()doesn't assume that all requests will put down by the end of the function asio_uring_cmd_done()can offload it in any case.But it isn't specific for ublk, any caller of
io_uring_cmd_done()has such issue sinceio_uring_cmd_done()is one generic API.
Well, fair enough, considering that IO_URING_F_UNLOCKED was just added (*still naively hoping it'll be clean up*)
Signature · 1 line
Pavel Begunkov
Quoted 18 lines · Pavel Begunkov
Quoted 17 lines · Ming Lei
Quoted 9 lines · Pavel Begunkov
Quoted 7 lines · Ming Lei
So far
io_req_complete_post()only coversDEFER_TASKRUNby completing request via task work when the request is completed from IOWQ.However, uring command could be completed from any context, and if io uring is setup with
DEFER_TASKRUN, the command is required to be completed from current context, otherwise wait onIORING_ENTER_GETEVENTScan't be wakeup, and may hang forever.Quoted 4 lines
fwiw, there is one legit exception, when the task is half dead
task_workwill be executed by a kthread. It should be fine as it locks the ctx down, but I can't help but wonder whether it's onlyublk_cancel_queue()affected or there are more places in ublk?No, it isn't.
Quoted 7 lines
Quoted 4 lines
It isn't triggered on nvme-pt just because command is always done in task context.
And we know more uring command cases are coming.
Because all requests and cmds but ublk complete it from another task, ublk is special in this regard.
Not sure it is true, cause it is allowed to call io_uring_cmd_done from other task technically. And it could be more friendly for driver to not limit its caller in the task context. Especially we have another API of io_uring_cmd_complete_in_task().
Quoted 4 lines
I have several more not so related questions:
1) Can requests be submitted by some other task than ->ubq_daemon?
Yeah, requests can be submitted by other task, but ublk driver doesn't allow it because ublk driver has not knowledge when the io_uring context goes away, so has to limit requests submitted from ->ubq_daemon only, then use this task's information for checking if the io_uring context is going to exit. When the io_uring context is dying, we need to abort these uring commands(may never complete), see ublk_cancel_queue().
The only difference is that the uring command may never complete, because one uring cmd is only completed when the associated block request is coming. The situation could be improved by adding API/callback for notifying io_uring exit.
Quoted 11 lines
Looking at
static int
ublk_ch_uring_cmd(struct io_uring_cmd *cmd, unsigned int issue_flags){ ... if (ubq->ubq_daemon && ubq->ubq_daemon != current) goto out; }
ublk_queue_cmd()avoidingio_uringway of delivery and using rawtask_workdoesn't seem great. Especially withTWA_SIGNAL_NO_IPI.
Yeah, it has been in my todo list to kill task work. In ublk early time, task work just performs better than io_uring_cmd_complete_in_task(), but the gap becomes pretty small or even not visible now.
Quoted 4 lines
2) What the purpose of the two lines below? I see how
UBLK_F_URING_CMD_COMP_IN_TASKis used, but don't understand why it changes depending on whether it's a module or not.
task work isn't available in case of building ublk as module.
Quoted 5 lines
3) The long comment in
ublk_queue_cmd()seems quite scary. If you have a cmd /io_uringrequest it hold a ctx reference and is always allowed to use io_uring'stask_workinfra likeio_uring_cmd_complete_in_task(). Why it's different for ublk?
The thing is that we don't know if there is io_uring request for the coming blk request. UBLK_IO_FLAG_ABORTED just means that the io_uring context is dead, and we can't use io_uring_cmd_complete_in_task() any more.
Quoted 25 lines
Quoted 8 lines
Quoted 5 lines
One more thing, cmds should not be setting
issue_flagsbut only forwarding what the coreio_uringcode passed, it'll get tons of bugs in no time otherwise.Here
io_uring_cmd_done()is changed to this way recently, and it could be another topic.And it's abused, but as you said, not particularly related to this patch.
Quoted 12 lines
Quoted 11 lines
static void
ublk_cancel_queue(struct ublk_queue *ubq){ ... io_uring_cmd_done(io->cmd, UBLK_IO_RES_ABORT, 0, IO_URING_F_UNLOCKED); }Can we replace it with
task_work? It should be cold, and I assumeublk_cancel_queue()doesn't assume that all requests will put down by the end of the function asio_uring_cmd_done()can offload it in any case.
Quoted 5 lines
Quoted 2 lines
But it isn't specific for ublk, any caller of
io_uring_cmd_done()has such issue sinceio_uring_cmd_done()is one generic API.Well, fair enough, considering that
IO_URING_F_UNLOCKEDwas just added (*still naively hoping it'll be clean up*)
IMO, it is reasonable for io_uring_cmd_done to hide any io_uring implementation details, even the task context concept, but not sure if it is doable.
Thanks,
Ming
Quoted 24 lines · Ming Lei
Quoted 23 lines · Pavel Begunkov
Quoted 21 lines · Ming Lei
Quoted 9 lines · Pavel Begunkov
Quoted 7 lines · Ming Lei
So far
io_req_complete_post()only coversDEFER_TASKRUNby completing request via task work when the request is completed from IOWQ.However, uring command could be completed from any context, and if io uring is setup with
DEFER_TASKRUN, the command is required to be completed from current context, otherwise wait onIORING_ENTER_GETEVENTScan't be wakeup, and may hang forever.Quoted 4 lines
fwiw, there is one legit exception, when the task is half dead
task_workwill be executed by a kthread. It should be fine as it locks the ctx down, but I can't help but wonder whether it's onlyublk_cancel_queue()affected or there are more places in ublk?No, it isn't.
It isn't triggered on nvme-pt just because command is always done in task context.
And we know more uring command cases are coming.
Quoted 7 lines
Quoted 2 lines
Because all requests and cmds but ublk complete it from another task, ublk is special in this regard.
Not sure it is true, cause it is allowed to call
io_uring_cmd_donefrom other task technically. And it could be more friendly for driver to not limit its caller in the task context. Especially we have another API ofio_uring_cmd_complete_in_task().
I agree that the cmd io_uring API can do better.
Quoted 11 lines
Quoted 3 lines
I have several more not so related questions:
1) Can requests be submitted by some other task than ->ubq_daemon?
Yeah, requests can be submitted by other task, but ublk driver doesn't allow it because ublk driver has not knowledge when the
io_uringcontext goes away, so has to limit requests submitted from ->ubq_daemon only, then use this task's information for checking if theio_uringcontext is going to exit. When theio_uringcontext is dying, we need to abort these uring commands(may never complete), seeublk_cancel_queue().
Quoted 4 lines
The only difference is that the uring command may never complete, because one uring cmd is only completed when the associated block request is coming. The situation could be improved by adding API/callback for notifying
io_uringexit.
Got it. And it sounds like you can use IORING_SETUP_SINGLE_ISSUER and possibly IORING_SETUP_DEFER_TASKRUN, if not already.
Quoted 9 lines
Quoted 9 lines
Looking at
static int
ublk_ch_uring_cmd(struct io_uring_cmd *cmd, unsigned int issue_flags){ ... if (ubq->ubq_daemon && ubq->ubq_daemon != current) goto out; }
Quoted 4 lines
Quoted 2 lines
ublk_queue_cmd()avoidingio_uringway of delivery and using rawtask_workdoesn't seem great. Especially withTWA_SIGNAL_NO_IPI.Yeah, it has been in my todo list to kill task work. In ublk early time,
I see
Quoted 2 lines
task work just performs better than
io_uring_cmd_complete_in_task(), but the gap becomes pretty small or even not visible now.
It seems a bit strange, non DEFER_TASKRUN tw is almost identical to what you do, see __io_req_task_work_add(). Maybe it's extra callbacks on the execution side.
Did you try DEFER_TASKRUN? Not sure it suits your case as there are limitations, but the queueing side of it, as well as execution and waiting are well optimised and should do better.
Quoted 5 lines
Quoted 3 lines
2) What the purpose of the two lines below? I see how
UBLK_F_URING_CMD_COMP_IN_TASKis used, but don't understand why it changes depending on whether it's a module or not.task work isn't available in case of building ublk as module.
Ah, makes sense now, thanks
Quoted 9 lines
Quoted 4 lines
3) The long comment in
ublk_queue_cmd()seems quite scary. If you have a cmd /io_uringrequest it hold a ctx reference and is always allowed to use io_uring'stask_workinfra likeio_uring_cmd_complete_in_task(). Why it's different for ublk?The thing is that we don't know if there is
io_uringrequest for the coming blk request.UBLK_IO_FLAG_ABORTEDjust means that theio_uringcontext is dead, and we can't useio_uring_cmd_complete_in_task()any more.
Roughly got it, IIUC, there might not be a (valid) io_uring request backing this block request in the first place because of this aborting thing.
Quoted 26 lines
Quoted 26 lines
Quoted 6 lines
Quoted 3 lines
One more thing, cmds should not be setting
issue_flagsbut only forwarding what the coreio_uringcode passed, it'll get tons of bugs in no time otherwise.Here
io_uring_cmd_done()is changed to this way recently, and it could be another topic.And it's abused, but as you said, not particularly related to this patch.
Quoted 14 lines
Quoted 11 lines
static void
ublk_cancel_queue(struct ublk_queue *ubq){ ... io_uring_cmd_done(io->cmd, UBLK_IO_RES_ABORT, 0, IO_URING_F_UNLOCKED); }Can we replace it with
task_work? It should be cold, and I assumeublk_cancel_queue()doesn't assume that all requests will put down by the end of the function asio_uring_cmd_done()can offload it in any case.But it isn't specific for ublk, any caller of
io_uring_cmd_done()has such issue sinceio_uring_cmd_done()is one generic API.
Quoted 6 lines
Quoted 2 lines
Well, fair enough, considering that
IO_URING_F_UNLOCKEDwas just added (*still naively hoping it'll be clean up*)IMO, it is reasonable for
io_uring_cmd_doneto hide anyio_uringimplementation details, even the task context concept, but not sure if it is doable.
I agree that there should be a function doing the right thing without extra flags, i.e. completing via tw, and there should also be a helper for more advanced performant cases like when we know the context.
Signature · 1 line
Pavel Begunkov
Quoted 47 lines · Pavel Begunkov
Quoted 24 lines · Ming Lei
Quoted 23 lines · Pavel Begunkov
Quoted 21 lines · Ming Lei
Quoted 9 lines · Pavel Begunkov
Quoted 7 lines · Ming Lei
So far
io_req_complete_post()only coversDEFER_TASKRUNby completing request via task work when the request is completed from IOWQ.However, uring command could be completed from any context, and if io uring is setup with
DEFER_TASKRUN, the command is required to be completed from current context, otherwise wait onIORING_ENTER_GETEVENTScan't be wakeup, and may hang forever.Quoted 4 lines
fwiw, there is one legit exception, when the task is half dead
task_workwill be executed by a kthread. It should be fine as it locks the ctx down, but I can't help but wonder whether it's onlyublk_cancel_queue()affected or there are more places in ublk?No, it isn't.
It isn't triggered on nvme-pt just because command is always done in task context.
And we know more uring command cases are coming.
Quoted 7 lines
Quoted 2 lines
Because all requests and cmds but ublk complete it from another task, ublk is special in this regard.
Not sure it is true, cause it is allowed to call
io_uring_cmd_donefrom other task technically. And it could be more friendly for driver to not limit its caller in the task context. Especially we have another API ofio_uring_cmd_complete_in_task().I agree that the cmd
io_uringAPI can do better.Quoted 11 lines
Quoted 3 lines
I have several more not so related questions:
1) Can requests be submitted by some other task than ->ubq_daemon?
Yeah, requests can be submitted by other task, but ublk driver doesn't allow it because ublk driver has not knowledge when the
io_uringcontext goes away, so has to limit requests submitted from ->ubq_daemon only, then use this task's information for checking if theio_uringcontext is going to exit. When theio_uringcontext is dying, we need to abort these uring commands(may never complete), seeublk_cancel_queue().
Quoted 7 lines
Quoted 4 lines
The only difference is that the uring command may never complete, because one uring cmd is only completed when the associated block request is coming. The situation could be improved by adding API/callback for notifying
io_uringexit.Got it. And it sounds like you can use
IORING_SETUP_SINGLE_ISSUERand possiblyIORING_SETUP_DEFER_TASKRUN, if not already.
ublk driver is simple, but the userspace ublk server can be quite complicated and need flexible setting, and we shouldn't put any limit on userspace in theory.
Quoted 21 lines
Quoted 11 lines
Quoted 11 lines
Looking at
static int
ublk_ch_uring_cmd(struct io_uring_cmd *cmd, unsigned int issue_flags){ ... if (ubq->ubq_daemon && ubq->ubq_daemon != current) goto out; }Quoted 4 lines
Quoted 2 lines
ublk_queue_cmd()avoidingio_uringway of delivery and using rawtask_workdoesn't seem great. Especially withTWA_SIGNAL_NO_IPI.Yeah, it has been in my todo list to kill task work. In ublk early time,
I see
Quoted 2 lines
task work just performs better than
io_uring_cmd_complete_in_task(), but the gap becomes pretty small or even not visible now.
Quoted 7 lines
It seems a bit strange, non
DEFER_TASKRUNtw is almost identical to what you do, see__io_req_task_work_add(). Maybe it's extra callbacks on the execution side.Did you try
DEFER_TASKRUN? Not sure it suits your case as there are limitations, but the queueing side of it, as well as execution and waiting are well optimised and should do better.
I tried DEFER_TASKRUN which need this fix, not see obvious IOPS boost against IORING_SETUP_COOP_TASKRUN, which does make big difference.
Quoted 15 lines
Quoted 7 lines
Quoted 5 lines
2) What the purpose of the two lines below? I see how
UBLK_F_URING_CMD_COMP_IN_TASKis used, but don't understand why it changes depending on whether it's a module or not.task work isn't available in case of building ublk as module.
Ah, makes sense now, thanks
Quoted 5 lines
Quoted 4 lines
3) The long comment in
ublk_queue_cmd()seems quite scary. If you have a cmd /io_uringrequest it hold a ctx reference and is always allowed to use io_uring'stask_workinfra likeio_uring_cmd_complete_in_task(). Why it's different for ublk?
Quoted 8 lines
Quoted 4 lines
The thing is that we don't know if there is
io_uringrequest for the coming blk request.UBLK_IO_FLAG_ABORTEDjust means that theio_uringcontext is dead, and we can't useio_uring_cmd_complete_in_task()any more.Roughly got it, IIUC, there might not be a (valid)
io_uringrequest backing this block request in the first place because of this aborting thing.
I am working on adding notifier cb in io_uring_try_cancel_requests(), and looks it works. With this way, ublk server implementation can become quite flexible and aborting becomes simpler, such as, not need limit of single per-queue submitter any more, and I remember that spdk guys did complain this kind of limit.
Thanks,
Ming