sqpoll fixes/cleanups
cover sqpoll fixes/cleanups
All can go 5.12, but 1/3 is must have, 2/3 is much preferred to land 5.12. 3/3 is meh.
Pavel Begunkov (3):
io_uring: fix invalid ctx->sq_thread_idle
io_uring: remove indirect ctx into sqo injection
io_uring: simplify io_sqd_update_thread_idle()
fs/io_uring.c | 41 ++++++++--------------------------------- 1 file changed, 8 insertions(+), 33 deletions(-)
Signature · 1 line
2.24.0
PATCH 1/3 io_uring: fix invalid ctx->sq_thread_idle
We have to set ctx->sq_thread_idle before adding a ring to an SQ task, otherwise sqd races for seeing zero and accounting it as such.
--- fs/io_uring.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-)
fs/io_uring.c +4 -4
@@ -7827,14 +7827,14 @@ static int io_sq_offload_create(struct io_ring_ctx *ctx, ctx->sq_creds =get_current_cred (); ctx->sq_data = sqd; -io_sq_thread_park (sqd); -list_add (&ctx->sqd_list, &sqd->ctx_new_list); -io_sq_thread_unpark (sqd); - ctx->sq_thread_idle =msecs_to_jiffies (p->sq_thread_idle);if (!ctx->sq_thread_idle) ctx->sq_thread_idle =HZ ; +io_sq_thread_park (sqd); +list_add (&ctx->sqd_list, &sqd->ctx_new_list); +io_sq_thread_unpark (sqd); +if (sqd->thread)return 0 ;
Signature · 1 line
2.24.0
PATCH 2/3 io_uring: remove indirect ctx into sqo injection
In reply to Pavel BegunkovWe use ->ctx_new_list to notify sqo about new ctx pending, then sqo should stop and splice it to its sqd->ctx_list, paired with ->sq_thread_comp.
The last one is broken because nobody reinitialises it, and trying to fix it would only add more complexity and bugs. And the first isn't really needed as is done under park(), that protects from races well. Add ctx into sqd->ctx_list directly (under park()), it's much simpler and allows to kill both, ctx_new_list and sq_thread_comp.
note: apparently there is no real problem at the moment, because sq_thread_comp is used only by io_sq_thread_finish() followed by parking, where list_del(&ctx->sqd_list) removes it well regardless whether it's in the new or the active list.
--- fs/io_uring.c | 28 +++------------------------- 1 file changed, 3 insertions(+), 25 deletions(-)
fs/io_uring.c +3 -25
@@ -262,7 +262,6 @@ struct io_sq_data {/* ctx's that are using this sqd */ struct list_head ctx_list; -struct list_head ctx_new_list;struct task_struct *thread;struct wait_queue_head wait;
@@ -398,7 +397,6 @@ struct io_ring_ctx {struct user_struct *user;struct completion ref_comp; -struct completion sq_thread_comp;#if defined(CONFIG_UNIX )struct socket *ring_sock;
@@ -1136,7 +1134,6 @@ static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p)init_waitqueue_head (&ctx->cq_wait);INIT_LIST_HEAD (&ctx->cq_overflow_list);init_completion (&ctx->ref_comp); -init_completion (&ctx->sq_thread_comp);idr_init (&ctx->io_buffer_idr);xa_init_flags (&ctx->personalities,XA_FLAGS_ALLOC1 );mutex_init (&ctx->uring_lock);
@@ -6632,19 +6629,6 @@ static void io_sqd_update_thread_idle(struct io_sq_data *sqd) sqd->sq_thread_idle = sq_thread_idle; } -static void io_sqd_init_new (struct io_sq_data *sqd) -{ -struct io_ring_ctx *ctx; - -while (!list_empty (&sqd->ctx_new_list)) { - ctx = list_first_entry(&sqd->ctx_new_list, struct io_ring_ctx, sqd_list); -list_move_tail (&ctx->sqd_list, &sqd->ctx_list); -complete (&ctx->sq_thread_comp); - } - -io_sqd_update_thread_idle (sqd); -} -static int io_sq_thread (void *data) {struct io_sq_data *sqd = data;
@@ -6675,11 +6659,8 @@ static int io_sq_thread(void *data)up_read (&sqd->rw_lock);cond_resched ();down_read (&sqd->rw_lock); -continue ; - } -if (unlikely (!list_empty (&sqd->ctx_new_list))) { -io_sqd_init_new (sqd); timeout = jiffies + sqd->sq_thread_idle; +continue ; }if (fatal_signal_pending (current)) {set_bit (IO_SQ_THREAD_SHOULD_STOP , &sqd->state);
@@ -7097,9 +7078,6 @@ static void io_sq_thread_finish(struct io_ring_ctx *ctx)if (sqd) {complete (&sqd->startup); -if (sqd->thread) -wait_for_completion (&ctx->sq_thread_comp); -io_sq_thread_park (sqd);list_del (&ctx->sqd_list);io_sqd_update_thread_idle (sqd);
@@ -7151,7 +7129,6 @@ static struct io_sq_data *io_get_sq_data(struct io_uring_params *p)refcount_set (&sqd->refs,1 );INIT_LIST_HEAD (&sqd->ctx_list); -INIT_LIST_HEAD (&sqd->ctx_new_list);init_rwsem (&sqd->rw_lock);init_waitqueue_head (&sqd->wait);init_completion (&sqd->startup);
@@ -7832,7 +7809,8 @@ static int io_sq_offload_create(struct io_ring_ctx *ctx, ctx->sq_thread_idle =HZ ;io_sq_thread_park (sqd); -list_add (&ctx->sqd_list, &sqd->ctx_new_list); +list_add (&ctx->sqd_list, &sqd->ctx_list); +io_sqd_update_thread_idle (sqd);io_sq_thread_unpark (sqd);if (sqd->thread)
Signature · 1 line
2.24.0
PATCH 3/3 io_uring: simplify io_sqd_update_thread_idle()
In reply to Pavel BegunkovUse a more comprehensible() max instead of hand coding it with ifs in io_sqd_update_thread_idle().
--- fs/io_uring.c | 7 ++----- 1 file changed, 2 insertions(+), 5 deletions(-)
fs/io_uring.c +2 -5
@@ -6621,11 +6621,8 @@ static void io_sqd_update_thread_idle(struct io_sq_data *sqd)struct io_ring_ctx *ctx;unsigned sq_thread_idle =0 ; -list_for_each_entry (ctx, &sqd->ctx_list, sqd_list) { -if (sq_thread_idle < ctx->sq_thread_idle) - sq_thread_idle = ctx->sq_thread_idle; - } - +list_for_each_entry (ctx, &sqd->ctx_list, sqd_list) + sq_thread_idle =max (sq_thread_idle, ctx->sq_thread_idle); sqd->sq_thread_idle = sq_thread_idle; }
Signature · 1 line
2.24.0
Re: [PATCH 1/3] io_uring: fix invalid ctx->sq_thread_idle
In reply to Pavel BegunkovHi Pavel,
Quoted 17 lines
We have to set
ctx->sq_thread_idlebefore adding a ring to an SQ task, otherwise sqd races for seeing zero and accounting it as such.Signed-off-by Pavel Begunkov <[email protected]>--- fs/io_uring.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-)fs/io_uring.c +0
-1@@ -7827,14 +7827,14 @@ static int io_sq_offload_create(struct io_ring_ctx *ctx, ctx->sq_creds =get_current_cred (); ctx->sq_data = sqd; -io_sq_thread_park (sqd);
fs/io_uring.c static int io_sq_offload_create(struct io_ring_ctx *ctx, · line 7836
Quoted 10 lines
fs/io_uring.c +3
-3-list_add (&ctx->sqd_list, &sqd->ctx_new_list); -io_sq_thread_unpark (sqd); - ctx->sq_thread_idle =msecs_to_jiffies (p->sq_thread_idle);if (!ctx->sq_thread_idle) ctx->sq_thread_idle =HZ ; +io_sq_thread_park (sqd); +list_add (&ctx->sqd_list, &sqd->ctx_new_list); +io_sq_thread_unpark (sqd);
I wondered about the exact same change this morning, while researching the IORING_SETUP_ATTACH_WQ behavior :-)
It still seems to me that IORING_SETUP_ATTACH_WQ changed over time. As you introduced that flag, can you summaries it's behavior (and changes) over time (over the releases).
I'm wondering if ctx->sq_creds is really the only thing we need to take care of.
Do we know about existing users of IORING_SETUP_ATTACH_WQ and their use case? As mm, files and other things may differ now between sqe producer and the sq_thread.
metze
Jens Axboe
maintainerQuoted 10 lines · Pavel Begunkov
All can go 5.12, but 1/3 is must have, 2/3 is much preferred to land 5.12. 3/3 is meh.
Pavel Begunkov (3):
io_uring: fix invalidctx->sq_thread_idle
io_uring: remove indirect ctx into sqo injection
io_uring: simplifyio_sqd_update_thread_idle()fs/io_uring.c | 41 ++++++++--------------------------------- 1 file changed, 8 insertions(+), 33 deletions(-)
Thanks, added 1+2 for 5.12, and 3 for 5.13.
Signature · 1 line
Jens Axboe
Re: [PATCH 1/3] io_uring: fix invalid ctx->sq_thread_idle
In reply to Stefan MetzmacherQuoted 11 lines
I wondered about the exact same change this morning, while researching the
IORING_SETUP_ATTACH_WQbehavior :-)It still seems to me that
IORING_SETUP_ATTACH_WQchanged over time. As you introduced that flag, can you summaries it's behavior (and changes) over time (over the releases).I'm wondering if
ctx->sq_credsis really the only thing we need to take care of.Do we know about existing users of
IORING_SETUP_ATTACH_WQand their use case? As mm, files and other things may differ now between sqe producer and thesq_thread.
I'll start a new thread for this...
metze
Re: [PATCH 1/3] io_uring: fix invalid ctx->sq_thread_idle
In reply to Stefan MetzmacherQuoted 9 lines · Stefan Metzmacher
Hi Pavel,
I wondered about the exact same change this morning, while researching the
IORING_SETUP_ATTACH_WQbehavior :-)It still seems to me that
IORING_SETUP_ATTACH_WQchanged over time. As you introduced that flag, can you summaries it's behavior (and changes) over time (over the releases).
Not sure I remember the story in details, but from the beginning it was for io-wq sharing only, then it had expanded to SQPOLL as well. Now it's only about SQPOLL sharing, because of the recent io-wq changes that made it per-task and shared by default.
In all cases it should be checking the passed in file, that should retain the old behaviour of failing setup if the flag is set but wq_fd is not valid.
Quoted 2 lines
I'm wondering if
ctx->sq_credsis really the only thing we need to take care of.
io-wq is not affected by IORING_SETUP_ATTACH_WQ. It's per-task and mimics all the resources of the creator (on the moment of io-wq creation). Off ATTACH_WQ topic, but that's almost matches what it has been before, and with dropped unshare bit, should be totally same.
Regarding SQPOLL, it was always using resources of the first task, so those are just reaped of from it, and not only some particular like mm/files but all of them, like fork does, so should be safer.
Creds are just a special case because of that personality stuff, at least if we add back iowq unshare handling.
Quoted 2 lines
Do we know about existing users of
IORING_SETUP_ATTACH_WQand their use case?
Have no clue.
Quoted 1 line
As mm, files and other things may differ now between sqe producer and the
sq_thread.
It was always using mm/files of the ctx creator's task, aka ctx->sqo_task, but right, for the sharing case those may be different b/w ctx, so looks like a regression to me
Signature · 1 line
Pavel Begunkov
IORING_SETUP_ATTACH_WQ (was Re: [PATCH 1/3] io_uring: fix invalid ctx->sq_thread_idle)
Am 11.03.21 um 12:18 schrieb Pavel Begunkov:
Quoted 11 lines · Stefan Metzmacher
Quoted 9 lines
Hi Pavel,
I wondered about the exact same change this morning, while researching the
IORING_SETUP_ATTACH_WQbehavior :-)It still seems to me that
IORING_SETUP_ATTACH_WQchanged over time. As you introduced that flag, can you summaries it's behavior (and changes) over time (over the releases).
Quoted 7 lines
Not sure I remember the story in details, but from the beginning it was for io-wq sharing only, then it had expanded to SQPOLL as well. Now it's only about SQPOLL sharing, because of the recent io-wq changes that made it per-task and shared by default.
In all cases it should be checking the passed in file, that should retain the old behaviour of failing setup if the flag is set but
wq_fdis not valid.
Thanks, that's what I also found so far, see below for more findings.
Quoted 18 lines
Quoted 2 lines
I'm wondering if
ctx->sq_credsis really the only thing we need to take care of.io-wq is not affected by
IORING_SETUP_ATTACH_WQ. It's per-task and mimics all the resources of the creator (on the moment of io-wq creation). OffATTACH_WQtopic, but that's almost matches what it has been before, and with dropped unshare bit, should be totally same.Regarding SQPOLL, it was always using resources of the first task, so those are just reaped of from it, and not only some particular like mm/files but all of them, like fork does, so should be safer.
Creds are just a special case because of that personality stuff, at least if we add back iowq unshare handling.
Quoted 2 lines
Do we know about existing users of
IORING_SETUP_ATTACH_WQand their use case?
Quoted 7 lines
Have no clue.
Quoted 1 line
As mm, files and other things may differ now between sqe producer and the
sq_thread.It was always using mm/files of the ctx creator's task, aka
ctx->sqo_task, but right, for the sharing case those may be different b/w ctx, so looks like a regression to me
Good. I'll try to explore a possible way out below.
Ok, I'm continuing the thread here (just pasting the mail I already started to write :-)
I did some more research regarding IORING_SETUP_ATTACH_WQ in 5.12.
The current logic in io_sq_offload_create() is this:
+ /* Retain compatibility with failing for an invalid attach attempt */
+ if ((ctx->flags & (IORING_SETUP_ATTACH_WQ | IORING_SETUP_SQPOLL)) ==
+ IORING_SETUP_ATTACH_WQ) {
+ struct fd f;
+
+ f = fdget(p->wq_fd);
+ if (!f.file)
+ return -ENXIO;
+ if (f.file->f_op != &io_uring_fops) {
+ fdput(f);
+ return -EINVAL;
+ }
+ fdput(f);
+ }
That means that IORING_SETUP_ATTACH_WQ (without IORING_SETUP_SQPOLL) is completely ignored (except that we still simulate the -ENXIO and -EINVAL cases), correct? (You already agreed on that above :-)
The reason for this is that io_wq is no longer maintained per io_ring_ctx, but instead it is now global per io_uring_task.
Which means each userspace thread (or the sq_thread) has its own io_uring_task and thus its own io_wq.
Regarding the IORING_SETUP_SQPOLL|IORING_SETUP_ATTACH_WQ case we still allow attaching to the sq_thread of a different io_ring_ctx. The sq_thread runs in the context of the io_uring_setup() syscall that created it. We used to switch current->mm, current->files and other things before calling __io_sq_thread() before, but we no longer do that. And this seems to be security problem to me, as it's now possible for the attached io_ring_ctx to start sqe's copying the whole address space of the donator into a registered fixed file of the attached process.
As we already ignore IORING_SETUP_ATTACH_WQ without IORING_SETUP_SQPOLL, what about ignoring it as well if the attaching task uses different ->mm, ->files, ...
So IORING_SETUP_ATTACH_WQ would only have any effect if the task calling io_uring_setup() runs in the same context (except of the creds) as the existing sq_thread, which means it would work if multiple userspace threads of the same userspace process want to share the sq_thread and its io_wq. Everything else would be stupid (similar to the unshare() cases).
But as this has worked before, we just silently ignore IORING_SETUP_ATTACH_WQ is
we find a context mismatch and let io_uring_setup() silently create a new sq_thread.
What do you think?
metze
Re: IORING_SETUP_ATTACH_WQ (was Re: [PATCH 1/3] io_uring: fix invalid ctx->sq_thread_idle)
Am 11.03.21 um 12:46 schrieb Stefan Metzmacher:
Quoted 89 lines
Am 11.03.21 um 12:18 schrieb Pavel Begunkov:
Quoted 11 lines · Stefan Metzmacher
Quoted 9 lines
Hi Pavel,
I wondered about the exact same change this morning, while researching the
IORING_SETUP_ATTACH_WQbehavior :-)It still seems to me that
IORING_SETUP_ATTACH_WQchanged over time. As you introduced that flag, can you summaries it's behavior (and changes) over time (over the releases).Quoted 7 lines
Not sure I remember the story in details, but from the beginning it was for io-wq sharing only, then it had expanded to SQPOLL as well. Now it's only about SQPOLL sharing, because of the recent io-wq changes that made it per-task and shared by default.
In all cases it should be checking the passed in file, that should retain the old behaviour of failing setup if the flag is set but
wq_fdis not valid.Thanks, that's what I also found so far, see below for more findings.
Quoted 18 lines
Quoted 2 lines
I'm wondering if
ctx->sq_credsis really the only thing we need to take care of.io-wq is not affected by
IORING_SETUP_ATTACH_WQ. It's per-task and mimics all the resources of the creator (on the moment of io-wq creation). OffATTACH_WQtopic, but that's almost matches what it has been before, and with dropped unshare bit, should be totally same.Regarding SQPOLL, it was always using resources of the first task, so those are just reaped of from it, and not only some particular like mm/files but all of them, like fork does, so should be safer.
Creds are just a special case because of that personality stuff, at least if we add back iowq unshare handling.
Quoted 2 lines
Do we know about existing users of
IORING_SETUP_ATTACH_WQand their use case?Quoted 7 lines
Have no clue.
Quoted 1 line
As mm, files and other things may differ now between sqe producer and the
sq_thread.It was always using mm/files of the ctx creator's task, aka
ctx->sqo_task, but right, for the sharing case those may be different b/w ctx, so looks like a regression to meGood. I'll try to explore a possible way out below.
Ok, I'm continuing the thread here (just pasting the mail I already started to write :-)
I did some more research regarding
IORING_SETUP_ATTACH_WQin 5.12.The current logic in
io_sq_offload_create()is this:(quoted diff) +14
-0+ /* Retain compatibility with failing for an invalid attach attempt */ + if ((ctx->flags & (IORING_SETUP_ATTACH_WQ | IORING_SETUP_SQPOLL)) == + IORING_SETUP_ATTACH_WQ) { + struct fd f; + + f = fdget(p->wq_fd); + if (!f.file) + return -ENXIO; + if (f.file->f_op != &io_uring_fops) { + fdput(f); + return -EINVAL; + } + fdput(f); + }That means that
IORING_SETUP_ATTACH_WQ(withoutIORING_SETUP_SQPOLL) is completely ignored (except that we still simulate the-ENXIOand-EINVALcases), correct? (You already agreed on that above :-)The reason for this is that
io_wqis no longer maintained perio_ring_ctx, but instead it is now global perio_uring_task.
Which means each userspace thread (or thesq_thread) has its ownio_uring_taskand thus its ownio_wq.Regarding the IORING_SETUP_SQPOLL|IORING_SETUP_ATTACH_WQ case we still allow attaching to the
sq_threadof a differentio_ring_ctx. Thesq_threadruns in the context of theio_uring_setup()syscall that created it. We used to switchcurrent->mm,current->filesand other things before calling__io_sq_thread()before, but we no longer do that. And this seems to be security problem to me, as it's now possible for the attachedio_ring_ctxto start sqe's copying the whole address space of the donator into a registered fixed file of the attached process.
Quoted 8 lines
As we already ignore
IORING_SETUP_ATTACH_WQwithoutIORING_SETUP_SQPOLL, what about ignoring it as well if the attaching task uses different ->mm, ->files, ...
SoIORING_SETUP_ATTACH_WQwould only have any effect if the task callingio_uring_setup()runs in the same context (except of the creds) as the existingsq_thread, which means it would work if multiple userspace threads of the same userspace process want to share thesq_threadand itsio_wq. Everything else would be stupid (similar to theunshare()cases).
But as this has worked before, we just silently ignoreIORING_SETUP_ATTACH_WQis
we find a context mismatch and letio_uring_setup()silently create a newsq_thread.
Or we completely ignore IORING_SETUP_ATTACH_WQ (execpt the error cases).
Then we can implement a new IORING_SETUP_ATTACH_SQ with new semantics, that the existing sq_thread will be used as it and both sides now what it means to them. We also add a new IORING_REGISTER_RESTRICTIONS/IORING_RESTRICTION_ALLOW_SQ_ATTACHMENTS which prepares the first io_ring_ctx to allow others to attach.
Would that make sense?
metze
Re: IORING_SETUP_ATTACH_WQ (was Re: [PATCH 1/3] io_uring: fix invalid ctx->sq_thread_idle)
In reply to Stefan MetzmacherQuoted 71 lines · Stefan Metzmacher
Am 11.03.21 um 12:18 schrieb Pavel Begunkov:
Quoted 11 lines · Stefan Metzmacher
Quoted 9 lines
Hi Pavel,
I wondered about the exact same change this morning, while researching the
IORING_SETUP_ATTACH_WQbehavior :-)It still seems to me that
IORING_SETUP_ATTACH_WQchanged over time. As you introduced that flag, can you summaries it's behavior (and changes) over time (over the releases).Quoted 7 lines
Not sure I remember the story in details, but from the beginning it was for io-wq sharing only, then it had expanded to SQPOLL as well. Now it's only about SQPOLL sharing, because of the recent io-wq changes that made it per-task and shared by default.
In all cases it should be checking the passed in file, that should retain the old behaviour of failing setup if the flag is set but
wq_fdis not valid.Thanks, that's what I also found so far, see below for more findings.
Quoted 18 lines
Quoted 2 lines
I'm wondering if
ctx->sq_credsis really the only thing we need to take care of.io-wq is not affected by
IORING_SETUP_ATTACH_WQ. It's per-task and mimics all the resources of the creator (on the moment of io-wq creation). OffATTACH_WQtopic, but that's almost matches what it has been before, and with dropped unshare bit, should be totally same.Regarding SQPOLL, it was always using resources of the first task, so those are just reaped of from it, and not only some particular like mm/files but all of them, like fork does, so should be safer.
Creds are just a special case because of that personality stuff, at least if we add back iowq unshare handling.
Quoted 2 lines
Do we know about existing users of
IORING_SETUP_ATTACH_WQand their use case?Quoted 7 lines
Have no clue.
Quoted 1 line
As mm, files and other things may differ now between sqe producer and the
sq_thread.It was always using mm/files of the ctx creator's task, aka
ctx->sqo_task, but right, for the sharing case those may be different b/w ctx, so looks like a regression to meGood. I'll try to explore a possible way out below.
Ok, I'm continuing the thread here (just pasting the mail I already started to write :-)
I did some more research regarding
IORING_SETUP_ATTACH_WQin 5.12.The current logic in
io_sq_offload_create()is this:(quoted diff) +14
-0+ /* Retain compatibility with failing for an invalid attach attempt */ + if ((ctx->flags & (IORING_SETUP_ATTACH_WQ | IORING_SETUP_SQPOLL)) == + IORING_SETUP_ATTACH_WQ) { + struct fd f; + + f = fdget(p->wq_fd); + if (!f.file) + return -ENXIO; + if (f.file->f_op != &io_uring_fops) { + fdput(f); + return -EINVAL; + } + fdput(f); + }
Quoted 3 lines
That means that
IORING_SETUP_ATTACH_WQ(withoutIORING_SETUP_SQPOLL) is completely ignored (except that we still simulate the-ENXIOand-EINVALcases), correct? (You already agreed on that above :-)
Yep, and we do these -ENXIO and -EINVAL for SQPOLL as well.
Quoted 4 lines
The reason for this is that
io_wqis no longer maintained perio_ring_ctx, but instead it is now global perio_uring_task.
Which means each userspace thread (or thesq_thread) has its ownio_uring_taskand thus its ownio_wq.
Just for anyone out of context, it's per process/thread/struct task/etc. struct io_uring_task is just a bit of a context attached to a task ever submitted io_uring requests, and its' not some special kind of a task.
Quoted 7 lines
Regarding the IORING_SETUP_SQPOLL|IORING_SETUP_ATTACH_WQ case we still allow attaching to the
sq_threadof a differentio_ring_ctx. Thesq_threadruns in the context of theio_uring_setup()syscall that created it. We used to switchcurrent->mm,current->filesand other things before calling__io_sq_thread()before, but we no longer do that. And this seems to be security problem to me, as it's now possible for the attachedio_ring_ctxto start sqe's copying the whole address space of the donator into a registered fixed file of the attached process.
It's not as bad, because 1) you voluntarily passes fd and 2) requires privileges, but it's a change of behaviour, which, well, can be exploited as you said.
Quoted 8 lines
As we already ignore
IORING_SETUP_ATTACH_WQwithoutIORING_SETUP_SQPOLL, what about ignoring it as well if the attaching task uses different ->mm, ->files, ...
SoIORING_SETUP_ATTACH_WQwould only have any effect if the task callingio_uring_setup()runs in the same context (except of the creds) as the existingsq_thread, which means it would work if multiple userspace threads of the same userspace process want to share thesq_threadand itsio_wq. Everything else would be stupid (similar to theunshare()cases).
But as this has worked before, we just silently ignoreIORING_SETUP_ATTACH_WQis
we find a context mismatch and letio_uring_setup()silently create a newsq_thread.
options:
1. Return back all that acquire_mm_files. Not great, not as safe as new io-wq, etc.
2. Completely ignore SQPOLL sharing. Performance regressions...
3. Do selected sharing. Maybe if thread group or so matches, should be safer than just mm/files check (or any subset of possibly long list). And there may be differences when the creator task do unshare/etc., but can be patched up (from where the unshare hook came in the first place).
I like 3) but 2) may do as well. The only performance problem I see is for those who wanted to use it out of threads. E.g. there even was a proposal to have per-CPU SQPOLL tasks and keep them per whole system.
Signature · 1 line
Pavel Begunkov
Re: IORING_SETUP_ATTACH_WQ (was Re: [PATCH 1/3] io_uring: fix invalid ctx->sq_thread_idle)
Am 11.03.21 um 13:27 schrieb Pavel Begunkov:
Quoted 95 lines · Stefan Metzmacher
Quoted 71 lines
Am 11.03.21 um 12:18 schrieb Pavel Begunkov:
Quoted 11 lines · Stefan Metzmacher
Quoted 9 lines
Hi Pavel,
I wondered about the exact same change this morning, while researching the
IORING_SETUP_ATTACH_WQbehavior :-)It still seems to me that
IORING_SETUP_ATTACH_WQchanged over time. As you introduced that flag, can you summaries it's behavior (and changes) over time (over the releases).Quoted 7 lines
Not sure I remember the story in details, but from the beginning it was for io-wq sharing only, then it had expanded to SQPOLL as well. Now it's only about SQPOLL sharing, because of the recent io-wq changes that made it per-task and shared by default.
In all cases it should be checking the passed in file, that should retain the old behaviour of failing setup if the flag is set but
wq_fdis not valid.Thanks, that's what I also found so far, see below for more findings.
Quoted 18 lines
Quoted 2 lines
I'm wondering if
ctx->sq_credsis really the only thing we need to take care of.io-wq is not affected by
IORING_SETUP_ATTACH_WQ. It's per-task and mimics all the resources of the creator (on the moment of io-wq creation). OffATTACH_WQtopic, but that's almost matches what it has been before, and with dropped unshare bit, should be totally same.Regarding SQPOLL, it was always using resources of the first task, so those are just reaped of from it, and not only some particular like mm/files but all of them, like fork does, so should be safer.
Creds are just a special case because of that personality stuff, at least if we add back iowq unshare handling.
Quoted 2 lines
Do we know about existing users of
IORING_SETUP_ATTACH_WQand their use case?Quoted 7 lines
Have no clue.
Quoted 1 line
As mm, files and other things may differ now between sqe producer and the
sq_thread.It was always using mm/files of the ctx creator's task, aka
ctx->sqo_task, but right, for the sharing case those may be different b/w ctx, so looks like a regression to meGood. I'll try to explore a possible way out below.
Ok, I'm continuing the thread here (just pasting the mail I already started to write :-)
I did some more research regarding
IORING_SETUP_ATTACH_WQin 5.12.The current logic in
io_sq_offload_create()is this:(quoted diff) +14
-0+ /* Retain compatibility with failing for an invalid attach attempt */ + if ((ctx->flags & (IORING_SETUP_ATTACH_WQ | IORING_SETUP_SQPOLL)) == + IORING_SETUP_ATTACH_WQ) { + struct fd f; + + f = fdget(p->wq_fd); + if (!f.file) + return -ENXIO; + if (f.file->f_op != &io_uring_fops) { + fdput(f); + return -EINVAL; + } + fdput(f); + }Quoted 3 lines
That means that
IORING_SETUP_ATTACH_WQ(withoutIORING_SETUP_SQPOLL) is completely ignored (except that we still simulate the-ENXIOand-EINVALcases), correct? (You already agreed on that above :-)Yep, and we do these
-ENXIOand-EINVALfor SQPOLL as well.Quoted 4 lines
The reason for this is that
io_wqis no longer maintained perio_ring_ctx, but instead it is now global perio_uring_task.
Which means each userspace thread (or thesq_thread) has its ownio_uring_taskand thus its ownio_wq.Just for anyone out of context, it's per process/thread/struct task/etc. struct
io_uring_taskis just a bit of a context attached to a task ever submittedio_uringrequests, and its' not some special kind of a task.Quoted 7 lines
Regarding the IORING_SETUP_SQPOLL|IORING_SETUP_ATTACH_WQ case we still allow attaching to the
sq_threadof a differentio_ring_ctx. Thesq_threadruns in the context of theio_uring_setup()syscall that created it. We used to switchcurrent->mm,current->filesand other things before calling__io_sq_thread()before, but we no longer do that. And this seems to be security problem to me, as it's now possible for the attachedio_ring_ctxto start sqe's copying the whole address space of the donator into a registered fixed file of the attached process.
Quoted 2 lines
It's not as bad, because 1) you voluntarily passes fd and 2) requires privileges, but it's a change of behaviour, which, well, can be exploited as you said.
Yes, but pointers and other things may have a different meaning now, as they were against the thread that produced the sqe's and now it's relativ to the unchanged sq_thread. So unmodified application may corrupt/leak there data.
Quoted 21 lines
Quoted 8 lines
As we already ignore
IORING_SETUP_ATTACH_WQwithoutIORING_SETUP_SQPOLL, what about ignoring it as well if the attaching task uses different ->mm, ->files, ...
SoIORING_SETUP_ATTACH_WQwould only have any effect if the task callingio_uring_setup()runs in the same context (except of the creds) as the existingsq_thread, which means it would work if multiple userspace threads of the same userspace process want to share thesq_threadand itsio_wq. Everything else would be stupid (similar to theunshare()cases).
But as this has worked before, we just silently ignoreIORING_SETUP_ATTACH_WQis
we find a context mismatch and letio_uring_setup()silently create a newsq_thread.options:
1. Return back all thatacquire_mm_files. Not great, not as safe as new io-wq, etc.2. Completely ignore SQPOLL sharing. Performance regressions...
3. Do selected sharing. Maybe if thread group or so matches, should be safer than just mm/files check (or any subset of possibly long list). And there may be differences when the creator task do unshare/etc., but can be patched up (from where the unshare hook came in the first place).
Quoted 4 lines
I like 3) but 2) may do as well. The only performance problem I see is for those who wanted to use it out of threads. E.g. there even was a proposal to have per-CPU SQPOLL tasks and keep them per whole system.
Yes 2. with having a new IORING_SETUP_ATTACH_SQ (see my other mail)
Or 3. and I guess the thread group might be ok.
But somehow 2 feels safer and we could start with fresh ideas from there.
metze
Jens Axboe
maintainerRe: IORING_SETUP_ATTACH_WQ (was Re: [PATCH 1/3] io_uring: fix invalid ctx->sq_thread_idle)
In reply to Stefan MetzmacherQuoted 73 lines · Stefan Metzmacher
Am 11.03.21 um 12:18 schrieb Pavel Begunkov:
Quoted 11 lines · Stefan Metzmacher
Quoted 9 lines
Hi Pavel,
I wondered about the exact same change this morning, while researching the
IORING_SETUP_ATTACH_WQbehavior :-)It still seems to me that
IORING_SETUP_ATTACH_WQchanged over time. As you introduced that flag, can you summaries it's behavior (and changes) over time (over the releases).Quoted 7 lines
Not sure I remember the story in details, but from the beginning it was for io-wq sharing only, then it had expanded to SQPOLL as well. Now it's only about SQPOLL sharing, because of the recent io-wq changes that made it per-task and shared by default.
In all cases it should be checking the passed in file, that should retain the old behaviour of failing setup if the flag is set but
wq_fdis not valid.Thanks, that's what I also found so far, see below for more findings.
Quoted 18 lines
Quoted 2 lines
I'm wondering if
ctx->sq_credsis really the only thing we need to take care of.io-wq is not affected by
IORING_SETUP_ATTACH_WQ. It's per-task and mimics all the resources of the creator (on the moment of io-wq creation). OffATTACH_WQtopic, but that's almost matches what it has been before, and with dropped unshare bit, should be totally same.Regarding SQPOLL, it was always using resources of the first task, so those are just reaped of from it, and not only some particular like mm/files but all of them, like fork does, so should be safer.
Creds are just a special case because of that personality stuff, at least if we add back iowq unshare handling.
Quoted 2 lines
Do we know about existing users of
IORING_SETUP_ATTACH_WQand their use case?Quoted 7 lines
Have no clue.
Quoted 1 line
As mm, files and other things may differ now between sqe producer and the
sq_thread.It was always using mm/files of the ctx creator's task, aka
ctx->sqo_task, but right, for the sharing case those may be different b/w ctx, so looks like a regression to meGood. I'll try to explore a possible way out below.
Ok, I'm continuing the thread here (just pasting the mail I already started to write :-)
I did some more research regarding
IORING_SETUP_ATTACH_WQin 5.12.The current logic in
io_sq_offload_create()is this:(quoted diff) +14
-0+ /* Retain compatibility with failing for an invalid attach attempt */ + if ((ctx->flags & (IORING_SETUP_ATTACH_WQ | IORING_SETUP_SQPOLL)) == + IORING_SETUP_ATTACH_WQ) { + struct fd f; + + f = fdget(p->wq_fd); + if (!f.file) + return -ENXIO; + if (f.file->f_op != &io_uring_fops) { + fdput(f); + return -EINVAL; + } + fdput(f); + }
Quoted 3 lines
That means that
IORING_SETUP_ATTACH_WQ(withoutIORING_SETUP_SQPOLL) is completely ignored (except that we still simulate the-ENXIOand-EINVALcases), correct? (You already agreed on that above :-)
Correct, there's no need to share anything there anymore. So we just pretend that we do, since the application is none the wiser on that front anyway.
Quoted 4 lines
The reason for this is that
io_wqis no longer maintained perio_ring_ctx, but instead it is now global perio_uring_task. Which means each userspace thread (or thesq_thread) has its ownio_uring_taskand thus its ownio_wq.
Right
Quoted 9 lines
Regarding the IORING_SETUP_SQPOLL|IORING_SETUP_ATTACH_WQ case we still allow attaching to the
sq_threadof a differentio_ring_ctx. Thesq_threadruns in the context of theio_uring_setup()syscall that created it. We used to switchcurrent->mm,current->filesand other things before calling__io_sq_thread()before, but we no longer do that. And this seems to be security problem to me, as it's now possible for the attachedio_ring_ctxto start sqe's copying the whole address space of the donator into a registered fixed file of the attached process.
As Pavel commented, this isn't really a security issue, as you're willingly passing the fd for someone to attach to. That's why I made it an fd based attachment in the first place. That said, it can cause breakage now where if you rely on your ring creator mm (etc) to be used, then you'll have a bad time since things will fail. That's obviously not ideal, and certainly needs fixing.
Quoted 12 lines
As we already ignore
IORING_SETUP_ATTACH_WQwithoutIORING_SETUP_SQPOLL, what about ignoring it as well if the attaching task uses different ->mm, ->files, ... SoIORING_SETUP_ATTACH_WQwould only have any effect if the task callingio_uring_setup()runs in the same context (except of the creds) as the existingsq_thread, which means it would work if multiple userspace threads of the same userspace process want to share thesq_threadand itsio_wq. Everything else would be stupid (similar to theunshare()cases). But as this has worked before, we just silently ignoreIORING_SETUP_ATTACH_WQis we find a context mismatch and letio_uring_setup()silently create a newsq_thread.
Quoted 1 line
What do you think?
I think that's a very reasonable first approach, and would avoid a direct regression in terms of having the thread now uses different resources. It might very well cause a CPU regression since you now have eg 2 threads instead of 1, but that's something we can tackle down the line.
Signature · 1 line
Jens Axboe
Jens Axboe
maintainerRe: IORING_SETUP_ATTACH_WQ (was Re: [PATCH 1/3] io_uring: fix invalid ctx->sq_thread_idle)
In reply to Stefan MetzmacherQuoted 10 lines · Stefan Metzmacher
Or we completely ignore
IORING_SETUP_ATTACH_WQ(execpt the error cases).Then we can implement a new
IORING_SETUP_ATTACH_SQwith new semantics, that the existingsq_threadwill be used as it and both sides now what it means to them. We also add a new IORING_REGISTER_RESTRICTIONS/IORING_RESTRICTION_ALLOW_SQ_ATTACHMENTS which prepares the firstio_ring_ctxto allow others to attach.Would that make sense?
I think we should retain ATTACH_WQ semantics with SQPOLL for the cases that are possible. That would not exclude doing a more specific ATTACH_SQ for the new case. And maybe those two cases can then be folded down the line.
Signature · 1 line
Jens Axboe
Jens Axboe
maintainerRe: IORING_SETUP_ATTACH_WQ (was Re: [PATCH 1/3] io_uring: fix invalid ctx->sq_thread_idle)
In reply to Stefan MetzmacherQuoted 130 lines · Stefan Metzmacher
Am 11.03.21 um 13:27 schrieb Pavel Begunkov:
Quoted 95 lines · Stefan Metzmacher
Quoted 71 lines
Am 11.03.21 um 12:18 schrieb Pavel Begunkov:
Quoted 11 lines · Stefan Metzmacher
Quoted 9 lines
Hi Pavel,
I wondered about the exact same change this morning, while researching the
IORING_SETUP_ATTACH_WQbehavior :-)It still seems to me that
IORING_SETUP_ATTACH_WQchanged over time. As you introduced that flag, can you summaries it's behavior (and changes) over time (over the releases).Quoted 7 lines
Not sure I remember the story in details, but from the beginning it was for io-wq sharing only, then it had expanded to SQPOLL as well. Now it's only about SQPOLL sharing, because of the recent io-wq changes that made it per-task and shared by default.
In all cases it should be checking the passed in file, that should retain the old behaviour of failing setup if the flag is set but
wq_fdis not valid.Thanks, that's what I also found so far, see below for more findings.
Quoted 18 lines
Quoted 2 lines
I'm wondering if
ctx->sq_credsis really the only thing we need to take care of.io-wq is not affected by
IORING_SETUP_ATTACH_WQ. It's per-task and mimics all the resources of the creator (on the moment of io-wq creation). OffATTACH_WQtopic, but that's almost matches what it has been before, and with dropped unshare bit, should be totally same.Regarding SQPOLL, it was always using resources of the first task, so those are just reaped of from it, and not only some particular like mm/files but all of them, like fork does, so should be safer.
Creds are just a special case because of that personality stuff, at least if we add back iowq unshare handling.
Quoted 2 lines
Do we know about existing users of
IORING_SETUP_ATTACH_WQand their use case?Quoted 7 lines
Have no clue.
Quoted 1 line
As mm, files and other things may differ now between sqe producer and the
sq_thread.It was always using mm/files of the ctx creator's task, aka
ctx->sqo_task, but right, for the sharing case those may be different b/w ctx, so looks like a regression to meGood. I'll try to explore a possible way out below.
Ok, I'm continuing the thread here (just pasting the mail I already started to write :-)
I did some more research regarding
IORING_SETUP_ATTACH_WQin 5.12.The current logic in
io_sq_offload_create()is this:(quoted diff) +14
-0+ /* Retain compatibility with failing for an invalid attach attempt */ + if ((ctx->flags & (IORING_SETUP_ATTACH_WQ | IORING_SETUP_SQPOLL)) == + IORING_SETUP_ATTACH_WQ) { + struct fd f; + + f = fdget(p->wq_fd); + if (!f.file) + return -ENXIO; + if (f.file->f_op != &io_uring_fops) { + fdput(f); + return -EINVAL; + } + fdput(f); + }Quoted 3 lines
That means that
IORING_SETUP_ATTACH_WQ(withoutIORING_SETUP_SQPOLL) is completely ignored (except that we still simulate the-ENXIOand-EINVALcases), correct? (You already agreed on that above :-)Yep, and we do these
-ENXIOand-EINVALfor SQPOLL as well.Quoted 4 lines
The reason for this is that
io_wqis no longer maintained perio_ring_ctx, but instead it is now global perio_uring_task.
Which means each userspace thread (or thesq_thread) has its ownio_uring_taskand thus its ownio_wq.Just for anyone out of context, it's per process/thread/struct task/etc. struct
io_uring_taskis just a bit of a context attached to a task ever submittedio_uringrequests, and its' not some special kind of a task.Quoted 7 lines
Regarding the IORING_SETUP_SQPOLL|IORING_SETUP_ATTACH_WQ case we still allow attaching to the
sq_threadof a differentio_ring_ctx. Thesq_threadruns in the context of theio_uring_setup()syscall that created it. We used to switchcurrent->mm,current->filesand other things before calling__io_sq_thread()before, but we no longer do that. And this seems to be security problem to me, as it's now possible for the attachedio_ring_ctxto start sqe's copying the whole address space of the donator into a registered fixed file of the attached process.Quoted 2 lines
It's not as bad, because 1) you voluntarily passes fd and 2) requires privileges, but it's a change of behaviour, which, well, can be exploited as you said.
Yes, but pointers and other things may have a different meaning now, as they were against the thread that produced the sqe's and now it's relativ to the unchanged
sq_thread. So unmodified application may corrupt/leak there data.Quoted 21 lines
Quoted 8 lines
As we already ignore
IORING_SETUP_ATTACH_WQwithoutIORING_SETUP_SQPOLL, what about ignoring it as well if the attaching task uses different ->mm, ->files, ...
SoIORING_SETUP_ATTACH_WQwould only have any effect if the task callingio_uring_setup()runs in the same context (except of the creds) as the existingsq_thread, which means it would work if multiple userspace threads of the same userspace process want to share thesq_threadand itsio_wq. Everything else would be stupid (similar to theunshare()cases).
But as this has worked before, we just silently ignoreIORING_SETUP_ATTACH_WQis
we find a context mismatch and letio_uring_setup()silently create a newsq_thread.options:
1. Return back all thatacquire_mm_files. Not great, not as safe as new io-wq, etc.2. Completely ignore SQPOLL sharing. Performance regressions...
3. Do selected sharing. Maybe if thread group or so matches, should be safer than just mm/files check (or any subset of possibly long list). And there may be differences when the creator task do unshare/etc., but can be patched up (from where the unshare hook came in the first place).
Quoted 4 lines
I like 3) but 2) may do as well. The only performance problem I see is for those who wanted to use it out of threads. E.g. there even was a proposal to have per-CPU SQPOLL tasks and keep them per whole system.
Quoted 4 lines
Yes 2. with having a new
IORING_SETUP_ATTACH_SQ(see my other mail)Or 3. and I guess the thread group might be ok.
But somehow 2 feels safer and we could start with fresh ideas from there.
3 should be perfectly safe, outside of someone doing unshare(). And I think we already agreed that this case is in the realm of "garbage in, garbage out" and we don't need to specifically cater to it. If we match at attach time, we should be good to go.
Signature · 1 line
Jens Axboe
Jens Axboe
maintainerRe: IORING_SETUP_ATTACH_WQ (was Re: [PATCH 1/3] io_uring: fix invalid ctx->sq_thread_idle)
Quoted 133 lines · Jens Axboe
Quoted 132 lines · Stefan Metzmacher
Am 11.03.21 um 13:27 schrieb Pavel Begunkov:
Quoted 95 lines · Stefan Metzmacher
Quoted 71 lines
Am 11.03.21 um 12:18 schrieb Pavel Begunkov:
Quoted 11 lines · Stefan Metzmacher
Quoted 9 lines
Hi Pavel,
I wondered about the exact same change this morning, while researching the
IORING_SETUP_ATTACH_WQbehavior :-)It still seems to me that
IORING_SETUP_ATTACH_WQchanged over time. As you introduced that flag, can you summaries it's behavior (and changes) over time (over the releases).Quoted 7 lines
Not sure I remember the story in details, but from the beginning it was for io-wq sharing only, then it had expanded to SQPOLL as well. Now it's only about SQPOLL sharing, because of the recent io-wq changes that made it per-task and shared by default.
In all cases it should be checking the passed in file, that should retain the old behaviour of failing setup if the flag is set but
wq_fdis not valid.Thanks, that's what I also found so far, see below for more findings.
Quoted 18 lines
Quoted 2 lines
I'm wondering if
ctx->sq_credsis really the only thing we need to take care of.io-wq is not affected by
IORING_SETUP_ATTACH_WQ. It's per-task and mimics all the resources of the creator (on the moment of io-wq creation). OffATTACH_WQtopic, but that's almost matches what it has been before, and with dropped unshare bit, should be totally same.Regarding SQPOLL, it was always using resources of the first task, so those are just reaped of from it, and not only some particular like mm/files but all of them, like fork does, so should be safer.
Creds are just a special case because of that personality stuff, at least if we add back iowq unshare handling.
Quoted 2 lines
Do we know about existing users of
IORING_SETUP_ATTACH_WQand their use case?Quoted 7 lines
Have no clue.
Quoted 1 line
As mm, files and other things may differ now between sqe producer and the
sq_thread.It was always using mm/files of the ctx creator's task, aka
ctx->sqo_task, but right, for the sharing case those may be different b/w ctx, so looks like a regression to meGood. I'll try to explore a possible way out below.
Ok, I'm continuing the thread here (just pasting the mail I already started to write :-)
I did some more research regarding
IORING_SETUP_ATTACH_WQin 5.12.The current logic in
io_sq_offload_create()is this:(quoted diff) +14
-0+ /* Retain compatibility with failing for an invalid attach attempt */ + if ((ctx->flags & (IORING_SETUP_ATTACH_WQ | IORING_SETUP_SQPOLL)) == + IORING_SETUP_ATTACH_WQ) { + struct fd f; + + f = fdget(p->wq_fd); + if (!f.file) + return -ENXIO; + if (f.file->f_op != &io_uring_fops) { + fdput(f); + return -EINVAL; + } + fdput(f); + }Quoted 3 lines
That means that
IORING_SETUP_ATTACH_WQ(withoutIORING_SETUP_SQPOLL) is completely ignored (except that we still simulate the-ENXIOand-EINVALcases), correct? (You already agreed on that above :-)Yep, and we do these
-ENXIOand-EINVALfor SQPOLL as well.Quoted 4 lines
The reason for this is that
io_wqis no longer maintained perio_ring_ctx, but instead it is now global perio_uring_task.
Which means each userspace thread (or thesq_thread) has its ownio_uring_taskand thus its ownio_wq.Just for anyone out of context, it's per process/thread/struct task/etc. struct
io_uring_taskis just a bit of a context attached to a task ever submittedio_uringrequests, and its' not some special kind of a task.Quoted 7 lines
Regarding the IORING_SETUP_SQPOLL|IORING_SETUP_ATTACH_WQ case we still allow attaching to the
sq_threadof a differentio_ring_ctx. Thesq_threadruns in the context of theio_uring_setup()syscall that created it. We used to switchcurrent->mm,current->filesand other things before calling__io_sq_thread()before, but we no longer do that. And this seems to be security problem to me, as it's now possible for the attachedio_ring_ctxto start sqe's copying the whole address space of the donator into a registered fixed file of the attached process.Quoted 2 lines
It's not as bad, because 1) you voluntarily passes fd and 2) requires privileges, but it's a change of behaviour, which, well, can be exploited as you said.
Yes, but pointers and other things may have a different meaning now, as they were against the thread that produced the sqe's and now it's relativ to the unchanged
sq_thread. So unmodified application may corrupt/leak there data.Quoted 21 lines
Quoted 8 lines
As we already ignore
IORING_SETUP_ATTACH_WQwithoutIORING_SETUP_SQPOLL, what about ignoring it as well if the attaching task uses different ->mm, ->files, ...
SoIORING_SETUP_ATTACH_WQwould only have any effect if the task callingio_uring_setup()runs in the same context (except of the creds) as the existingsq_thread, which means it would work if multiple userspace threads of the same userspace process want to share thesq_threadand itsio_wq. Everything else would be stupid (similar to theunshare()cases).
But as this has worked before, we just silently ignoreIORING_SETUP_ATTACH_WQis
we find a context mismatch and letio_uring_setup()silently create a newsq_thread.options:
1. Return back all thatacquire_mm_files. Not great, not as safe as new io-wq, etc.2. Completely ignore SQPOLL sharing. Performance regressions...
3. Do selected sharing. Maybe if thread group or so matches, should be safer than just mm/files check (or any subset of possibly long list). And there may be differences when the creator task do unshare/etc., but can be patched up (from where the unshare hook came in the first place).
Quoted 4 lines
I like 3) but 2) may do as well. The only performance problem I see is for those who wanted to use it out of threads. E.g. there even was a proposal to have per-CPU SQPOLL tasks and keep them per whole system.
Yes 2. with having a new
IORING_SETUP_ATTACH_SQ(see my other mail)
Quoted 7 lines
Quoted 2 lines
Or 3. and I guess the thread group might be ok.
But somehow 2 feels safer and we could start with fresh ideas from there.3 should be perfectly safe, outside of someone doing
unshare(). And I think we already agreed that this case is in the realm of "garbage in, garbage out" and we don't need to specifically cater to it. If we match at attach time, we should be good to go.
Just to be clear, something like the below - totally untested, but in principle.
fs/io_uring.c +14 -2
@@ -269,6 +269,7 @@ struct io_sq_data {unsigned sq_thread_idle;int sq_cpu;pid_t task_pid; +pid_t task_tgid;unsigned long state;struct completion startup;
@@ -7081,6 +7082,10 @@ static struct io_sq_data *io_attach_sq_data(struct io_uring_params *p)fdput (f);return ERR_PTR (-EINVAL ); } +if (sqd->task_tgid != current->tgid) { +fdput (f); +return ERR_PTR (-EPERM ); + }refcount_inc (&sqd->refs);fdput (f);
@@ -7091,8 +7096,14 @@ static struct io_sq_data *io_get_sq_data(struct io_uring_params *p) {struct io_sq_data *sqd; -if (p->flags &IORING_SETUP_ATTACH_WQ ) -return io_attach_sq_data (p); +if (p->flags &IORING_SETUP_ATTACH_WQ ) { + sqd =io_attach_sq_data (p); +if (!IS_ERR (sqd)) +return sqd; +/* fall through for EPERM case, setup new sqd/task */ +if (PTR_ERR (sqd) != -EPERM ) +return sqd; + } sqd =kzalloc (sizeof (*sqd),GFP_KERNEL );if (!sqd)
@@ -7793,6 +7804,7 @@ static int io_sq_offload_create(struct io_ring_ctx *ctx, } sqd->task_pid = current->pid; + sqd->task_tgid = current->tgid; tsk =create_io_thread (io_sq_thread, sqd,NUMA_NO_NODE );if (IS_ERR (tsk)) { ret =PTR_ERR (tsk);
Signature · 1 line
Jens Axboe
Re: IORING_SETUP_ATTACH_WQ (was Re: [PATCH 1/3] io_uring: fix invalid ctx->sq_thread_idle)
Hi Jens,
Quoted 42 lines
fs/io_uring.c +13
-2@@ -269,6 +269,7 @@ struct io_sq_data {unsigned sq_thread_idle;int sq_cpu;pid_t task_pid; +pid_t task_tgid;unsigned long state;struct completion startup;@@ -7081,6 +7082,10 @@ static struct io_sq_data *io_attach_sq_data(struct io_uring_params *p)fdput (f);return ERR_PTR (-EINVAL ); } +if (sqd->task_tgid != current->tgid) { +fdput (f); +return ERR_PTR (-EPERM ); + }refcount_inc (&sqd->refs);fdput (f);@@ -7091,8 +7096,14 @@ static struct io_sq_data *io_get_sq_data(struct io_uring_params *p) {struct io_sq_data *sqd; -if (p->flags &IORING_SETUP_ATTACH_WQ ) -return io_attach_sq_data (p); +if (p->flags &IORING_SETUP_ATTACH_WQ ) { + sqd =io_attach_sq_data (p); +if (!IS_ERR (sqd)) +return sqd; +/* fall through for EPERM case, setup new sqd/task */ +if (PTR_ERR (sqd) != -EPERM ) +return sqd; + } sqd =kzalloc (sizeof (*sqd),GFP_KERNEL );if (!sqd)
fs/io_uring.c static int io_sq_offload_create(struct io_ring_ctx *ctx, · line 7810
Quoted 7 lines
fs/io_uring.c +1
-0@@ -7793,6 +7804,7 @@ static int io_sq_offload_create(struct io_ring_ctx *ctx, } sqd->task_pid = current->pid; + sqd->task_tgid = current->tgid; tsk =create_io_thread (io_sq_thread, sqd,NUMA_NO_NODE );if (IS_ERR (tsk)) { ret =PTR_ERR (tsk);
Ok, that looks nice and simple, thanks!
metze