io-wq defaults to per-node masks for IO workers. This works fine by
default, but isn't particularly handy for workloads that prefer more
specific affinities, for either performance or isolation reasons.
This adds IORING_REGISTER_IOWQ_AFF that allows the user to pass in a CPU
mask that is then applied to IO thread workers, and an
IORING_UNREGISTER_IOWQ_AFF that simply resets the masks back to the
default of per-node.
Note that no care is given to existing IO threads, they will need to go
through a reschedule before the affinity is correct if they are already
running or sleeping.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The req pointer uniquely identify a specific request.
Having it in traces can provide valuable insights that is not possible
to have if the calling process is reusing the same user_data value.
Reviewed-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Olivier Langlois <olivier@trillion01.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
In most cases io_commit_cqring() is just an smp_store_release(), and
it's hot enough, especially for IRQ rw, to want it to save on a function
call. Mark it inline and extract a non-inlined slow path doing drain
and timeout flushing. The inlined part is pretty slim to not cause
binary bloating.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/7350f8b6b92caa50a48a80be39909f0d83eddd93.1623772051.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Place all drain_next logic into io_drain_req(), so it's never executed
if there was no drained requests before. The only thing we need is to
set ->drain_active if we see a request with IOSQE_IO_DRAIN, do that in
io_init_req() where flags are definitely in registers.
Also, all drain-related code is encapsulated in io_drain_req(), makes it
cleaner.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/68bf4f7395ddaafbf1a26bd97b57d57d45a9f900.1623772051.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
->drain_used is one way, which is not optimal if users use DRAIN but
very rarely. However, we can just clear it in io_drain_req() when all
drained before requests are gone. Also rename the flag to reflect the
change and be more clear about it.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/7f37a240857546a94df6348507edddacab150460.1623772051.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
fs/io_uring.c: In function 'io_alloc_page_table':
include/linux/minmax.h:20:28: warning: comparison of distinct pointer
types lacks a cast
Cast everything to size_t using min_t.
Reported-by: Stephen Rothwell <sfr@canb.auug.org.au>
Fixes: 9123c8ffce ("io_uring: add helpers for 2 level table alloc")
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/50f420a956bca070a43810d4a805293ed54f39d8.1623759527.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The sqe_ptr argument has been gone since 709b302fad (io_uring:
simplify io_get_sqring, 2020-04-08), made the return value of the
function. Update the comment accordingly.
Signed-off-by: Fam Zheng <fam.zheng@bytedance.com>
Reviewed-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/20210604164256.12242-1-fam.zheng@bytedance.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Replace drain checks with one-way flag set upon seeing the first
IOSQE_IO_DRAIN request. There are several places where it cuts cycles
well:
1) It's much faster than the fast check with two
conditions in io_drain_req() including pretty complex
list_empty_careful().
2) We can mark io_queue_sqe() inline now, that's a huge win.
3) It replaces timeout and drain checks in io_commit_cqring() with a
single flags test. Also great not touching ->defer_list there without a
reason so limiting cache bouncing.
It adds a small amount of overhead to drain path, but it's negligible.
The main nuisance is that once it meets any DRAIN request in io_uring
instance lifetime it will _always_ go through a slower path, so
drain-less and offset-mode timeout less applications are preferable.
The overhead in that case would be not big, but it's worth to bear in
mind.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/98d2fff8c4da5144bb0d08499f591d4768128ea3.1623709150.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
->uring_lock is prevalently used for submission, even though it protects
many other things like iopoll, registeration, selected bufs, and more.
And it's placed together with ->cq_wait poked on completion and CQ
waiting sides. Move them apart, ->uring_lock goes to the submission
data, and cq_wait to completion related chunk. The last one requires
some reshuffling so everything needed by io_cqring_ev_posted*() is in
one cacheline.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/dea5e845caee4c98aa0922b46d713154d81f7bd8.1623709150.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
We use several wait_queue_head's for different purposes, but namings are
confusing. First rename ctx->cq_wait into ctx->poll_wait, because this
one is used for polling an io_uring instance. Then rename ctx->wait into
ctx->cq_wait, which is responsible for CQE waiting.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/47b97a097780c86c67b20b6ccc4e077523dce682.1623709150.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
There are no users of ->sq_check_overflow, only ->cq_check_overflow is
used. Combine it and move out of completion related part of struct
io_ring_ctx.
A not so obvious benefit of it is fitting all completion side fields
into a single cacheline. It was taking 2 lines before with 56B padding,
and io_cqring_ev_posted*() were still touching both of them.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/25927394964df31d113e3c729416af573afff5f5.1623709150.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
submit_state.link is used only to assemble a link and not used for
actual submission, so clear it before io_queue_sqe() in io_submit_sqe(),
awhile it's hot and in caches and queueing doesn't spoil it. May also
potentially help compiler with spilling or to do other optimisations.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/1579939426f3ad6b55af3005b1389bbbed7d780d.1623709150.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
io_commit_cqring() might be very hot and we definitely don't want to
touch ->timeout_list there, because 1) it's shared with the submission
side so might lead to cache bouncing and 2) may need to load an extra
cache line, especially for IRQ completions.
We're interested in it at the completion side only when there are
offset-mode timeouts, which are not so popular. Replace
list_empty(->timeout_list) hot path check with a new one-way flag, which
is set when we prepare the first offset-mode timeout.
note: the flag sits in the same line as briefly used after ->rings
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/e4892ec68b71a69f92ffbea4a1499be3ec0d463b.1623709150.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Kill ->cached_sq_dropped and wire DRAIN sequence number correction via
->cq_extra, which is there exactly for that purpose. User visible
dropped counter will be populated by incrementing it instead of keeping
a copy, similarly as it was done not so long ago with cq_overflow.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/088aceb2707a534d531e2770267c4498e0507cc1.1623709150.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Since moving locked_free_* out of struct io_submit_state
ctx->submit_state is accessed on submission side only, so move it into
the submission section. Same goes for rsrc table pointers/nodes/etc.,
they must be taken and checked during submission because sync'ed by
uring_lock, so move them there as well.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/8a5899a50afc6ccca63249e716f580b246f3dec6.1623709150.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
ctx->flags are heavily used by both, completion and submission sides, so
move it out from the ctx fields related to submissions. Instead, place
it together with ctx->refs, because it's already cacheline-aligned and
so pads lots of space, and both almost never change. Also, in most
occasions they are accessed together as refs are taken at submission
time and put back during completion.
Do same with ctx->rings, where the pointer itself is never modified
apart from ring init/free.
Note: in percpu mode, struct percpu_ref doesn't modify the struct itself
but takes indirection with ref->percpu_count_ptr.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/4c48c173e63d35591383ba2b87e8b8e8dfdbd23d.1623709150.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
sq_array and sq_sqes are always used together, however they are in
different cachelines, where the borderline is right before
cq_overflow_list is rather rarely touched. Move the fields together so
it loads only one cacheline.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/3ef2411a94874da06492506a8897eff679244f49.1623709150.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Static analysis is warning that the sizeof being used is should be
of *data->tags[i] and not data->tags[i]. Although these are the same
size on 64 bit systems it is not a portable assumption to assume
this is true for all cases. Fix this by using a temporary pointer
tag_slot to make the code a clearer.
Addresses-Coverity: ("Sizeof not portable")
Fixes: d878c81610 ("io_uring: hide rsrc tag copy into generic helpers")
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Reviewed-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/20210615130011.57387-1-colin.king@canonical.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
tctx in submission part is always synchronised because is executed from
the task's context, so we can batch allocate tctx/task references and
store them across syscall boundaries. It avoids enough of operations,
including an atomic for getting task ref and a percpu_counter_add()
function call, which still fallback to spinlock for large batching
cases (around >=32). Should be good for SQPOLL submitting in small
portions and coming at some moment bpf submissions.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/14b327b973410a3eec1f702ecf650e100513aca9.1623634181.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
There is a complaint against sys_io_uring_enter() blocking if it submits
stdin reads. The problem is in __io_file_supports_async(), which
sees that it's a cdev and allows it to be processed inline.
Punt char devices using generic rules of io_file_supports_async(),
including checking for presence of *_iter() versions of rw callbacks.
Apparently, it will affect most of cdevs with some exceptions like
null and zero devices.
Cc: stable@vger.kernel.org
Reported-by: Birk Hirdman <lonjil@gmail.com>
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/d60270856b8a4560a639ef5f76e55eb563633599.1623236455.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Relax buffer registration restictions, which filters out file backed
memory, and allow shmem/memfd as they have normal anonymous pages
underneath.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
struct io_submit_state contains struct io_comp_state and so
locked_free_*, that renders cachelines around ->locked_free* being
invalidated on most non-inline completions, that may terrorise caches if
submissions and completions are done by different tasks.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/290cb5412b76892e8631978ee8ab9db0c6290dd5.1621201931.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
There are two copies of cq_overflow, shared with userspace and internal
cached one. It was needed for DRAIN accounting, but now we have yet
another knob to tune the accounting, i.e. cq_extra, and we can throw
away the internal counter and just increment the one in the shared ring.
If user modifies it as so never gets the right overflow value ever
again, it's its problem, even though before we would have restored it
back by next overflow.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/8427965f5175dd051febc63804909861109ce859.1621201931.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
ring has two types of resource-related fields: used for request
submission, and field needed for update/registration. Reshuffle them
into these two groups for better locality and readability. The second
group is not in the hot path, so it's natural to place them somewhere in
the end. Also update an outdated comment.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/05b34795bb4440f4ec4510f08abd5a31830f8ca0.1621201931.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The main difference is in req_set_fail_links() renamed into
req_set_fail(), which now sets REQ_F_FAIL_LINK/REQ_F_FAIL flag
unconditional on whether it has been a link or not. It only matters in
io_disarm_next(), which already handles it well, and all calls to it
have a fast path checking REQ_F_LINK/HARDLINK.
It looks cleaner, and sheds binary size
text data bss dec hex filename
84235 12390 8 96633 17979 ./fs/io_uring.o
84151 12414 8 96573 1793d ./fs/io_uring.o
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/e2224154dd6e53b665ac835d29436b177872fa10.1621201931.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Going through submission in __io_sq_thread() and still having a full SQ
is rather unexpected, so remove a check for SQ fullness and just wake up
whoever wait on sqo_sq_wait. Also skip if it doesn't do submission in
the first place, likely may to happen for SQPOLL sharing and/or IOPOLL.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/e2e91751e87b1a39f8d63ef884aaff578123f61e.1621201931.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Add IORING_FEAT_RSRC_TAGS indicating that io_uring supports a bunch of
new IORING_REGISTER operations, in particular
IORING_REGISTER_[FILES[,UPDATE]2,BUFFERS[2,UPDATE]] that support rsrc
tagging, and also indicating implemented dynamic fixed buffer updates.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/9b995d4045b6c6b4ab7510ca124fd25ac2203af7.1623339162.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
There are ABI moments about recently added rsrc registration/update and
tagging that might become a nuisance in the future. First,
IORING_REGISTER_RSRC[_UPD] hide different types of resources under it,
so breaks fine control over them by restrictions. It works for now, but
once those are wanted under restrictions it would require a rework.
It was also inconvenient trying to fit a new resource not supporting
all the features (e.g. dynamic update) into the interface, so better
to return to IORING_REGISTER_* top level dispatching.
Second, register/update were considered to accept a type of resource,
however that's not a good idea because there might be several ways of
registration of a single resource type, e.g. we may want to add
non-contig buffers or anything more exquisite as dma mapped memory.
So, remove IORING_RSRC_[FILE,BUFFER] out of the ABI, and place them
internally for now to limit changes.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/9b554897a7c17ad6e3becc48dfed2f7af9f423d5.1623339162.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
As Andres reports "... io_sqe_buffer_register() doesn't initialize imu.
io_buffer_account_pin() does imu->acct_pages++, before calling
io_account_mem(ctx, imu->acct_pages).", leading to evevntual -ENOMEM.
Initialise the field.
Reported-by: Andres Freund <andres@anarazel.de>
Fixes: 41edf1a5ec ("io_uring: keep table of pointers to ubufs")
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/438a6f46739ae5e05d9c75a0c8fa235320ff367c.1622285901.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Commit ba5ef6dc8a ("io_uring: fortify tctx/io_wq cleanup") introduced
setting tctx->io_wq to NULL a bit earlier. This has caused KCSAN to
detect a data race between accesses to tctx->io_wq:
write to 0xffff88811d8df330 of 8 bytes by task 3709 on cpu 1:
io_uring_clean_tctx fs/io_uring.c:9042 [inline]
__io_uring_cancel fs/io_uring.c:9136
io_uring_files_cancel include/linux/io_uring.h:16 [inline]
do_exit kernel/exit.c:781
do_group_exit kernel/exit.c:923
get_signal kernel/signal.c:2835
arch_do_signal_or_restart arch/x86/kernel/signal.c:789
handle_signal_work kernel/entry/common.c:147 [inline]
exit_to_user_mode_loop kernel/entry/common.c:171 [inline]
...
read to 0xffff88811d8df330 of 8 bytes by task 6412 on cpu 0:
io_uring_try_cancel_iowq fs/io_uring.c:8911 [inline]
io_uring_try_cancel_requests fs/io_uring.c:8933
io_ring_exit_work fs/io_uring.c:8736
process_one_work kernel/workqueue.c:2276
...
With the config used, KCSAN only reports data races with value changes:
this implies that in the case here we also know that tctx->io_wq was
non-NULL. Therefore, depending on interleaving, we may end up with:
[CPU 0] | [CPU 1]
io_uring_try_cancel_iowq() | io_uring_clean_tctx()
if (!tctx->io_wq) // false | ...
... | tctx->io_wq = NULL
io_wq_cancel_cb(tctx->io_wq, ...) | ...
-> NULL-deref |
Note: It is likely that thus far we've gotten lucky and the compiler
optimizes the double-read into a single read into a register -- but this
is never guaranteed, and can easily change with a different config!
Fix the data race by restoring the previous behaviour, where both
setting io_wq to NULL and put of the wq are _serialized_ after
concurrent io_uring_try_cancel_iowq() via acquisition of the uring_lock
and removal of the node in io_uring_del_task_file().
Fixes: ba5ef6dc8a ("io_uring: fortify tctx/io_wq cleanup")
Suggested-by: Pavel Begunkov <asml.silence@gmail.com>
Reported-by: syzbot+bf2b3d0435b9b728946c@syzkaller.appspotmail.com
Signed-off-by: Marco Elver <elver@google.com>
Cc: Jens Axboe <axboe@kernel.dk>
Link: https://lore.kernel.org/r/20210527092547.2656514-1-elver@google.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
There is an old problem with io-wq cancellation where requests should be
killed and are in io-wq but are not discoverable, e.g. in @next_hashed
or @linked vars of io_worker_handle_work(). It adds some unreliability
to individual request canellation, but also may potentially get
__io_uring_cancel() stuck. For instance:
1) An __io_uring_cancel()'s cancellation round have not found any
request but there are some as desribed.
2) __io_uring_cancel() goes to sleep
3) Then workers wake up and try to execute those hidden requests
that happen to be unbound.
As we already cancel all requests of io-wq there, set IO_WQ_BIT_EXIT
in advance, so preventing 3) from executing unbound requests. The
workers will initially break looping because of getting a signal as they
are threads of the dying/exec()'ing user task.
Cc: stable@vger.kernel.org
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/abfcf8c54cb9e8f7bfbad7e9a0cc5433cc70bdc2.1621781238.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
We don't want anyone poking into tctx->io_wq awhile it's being destroyed
by io_wq_put_and_exit(), and even though it shouldn't even happen, if
buggy would be preferable to get a NULL-deref instead of subtle delayed
failure or UAF.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/827b021de17926fd807610b3e53a5a5fa8530856.1621513214.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
