kmalloc'ing struct io_notif is too expensive when done frequently, cache
them as many other resources in io_uring. Keep two list, the first one
is from where we're getting notifiers, it's protected by ->uring_lock.
The second is protected by ->completion_lock, to which we queue released
notifiers. Then we splice one list into another when needed.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/9dec18f7fcbab9f4bd40b96e5ae158b119945230.1657643355.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Add internal part of send zerocopy notifications. There are two main
structures, the first one is struct io_notif, which carries inside
struct ubuf_info and maps 1:1 to it. io_uring will be binding a number
of zerocopy send requests to it and ask to complete (aka flush) it. When
flushed and all attached requests and skbs complete, it'll generate one
and only one CQE. There are intended to be passed into the network layer
as struct msghdr::msg_ubuf.
The second concept is notification slots. The userspace will be able to
register an array of slots and subsequently addressing them by the index
in the array. Slots are independent of each other. Each slot can have
only one notifier at a time (called active notifier) but many notifiers
during the lifetime. When active, a notifier not going to post any
completion but the userspace can attach requests to it by specifying
the corresponding slot while issueing send zc requests. Eventually, the
userspace will want to "flush" the notifier losing any way to attach
new requests to it, however it can use the next atomatically added
notifier of this slot or of any other slot.
When the network layer is done with all enqueued skbs attached to a
notifier and doesn't need the specified in them user data, the flushed
notifier will post a CQE.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/3ecf54c31a85762bf679b0a432c9f43ecf7e61cc.1657643355.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
If we're offloading requests directly to io-wq because IOSQE_ASYNC was
set in the sqe, we can miss hashing writes appropriately because we
haven't set REQ_F_ISREG yet. This can cause a performance regression
with buffered writes, as io-wq then no longer correctly serializes writes
to that file.
Ensure that we set the flags in io_prep_async_work(), which will cause
the io-wq work item to be hashed appropriately.
Fixes: 584b0180f0 ("io_uring: move read/write file prep state into actual opcode handler")
Link: https://lore.kernel.org/io-uring/20220608080054.GB22428@xsang-OptiPlex-9020/
Reported-by: kernel test robot <oliver.sang@intel.com>
Tested-by: Yin Fengwei <fengwei.yin@intel.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The commit 8bb649ee1d ("io_uring: remove ring quiesce for
io_uring_register") removed the worklow relying on reinit/resurrection
of the percpu_ref, hence, initialization with that requested is a relic.
This is based on code review, this causes no real bug (and theoretically
can't). Technically it's a revert of commit 214828962d ("io_uring:
initialize percpu refcounters using PERCU_REF_ALLOW_REINIT") but since
the flag omission is now justified, I'm not making this a revert.
Fixes: 8bb649ee1d ("io_uring: remove ring quiesce for io_uring_register")
Signed-off-by: Michal Koutný <mkoutny@suse.com>
Acked-by: Roman Gushchin <roman.gushchin@linux.dev>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
For recvmsg/sendmsg, if they don't complete inline, we currently need
to allocate a struct io_async_msghdr for each request. This is a
somewhat large struct.
Hook up sendmsg/recvmsg to use the io_alloc_cache. This reduces the
alloc + free overhead considerably, yielding 4-5% of extra performance
running netbench.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Caches like this tend to grow to the peak size, and then never get any
smaller. Impose a max limit on the size, to prevent it from growing too
big.
A somewhat randomly chosen 512 is the max size we'll allow the cache
to get. If a batch of frees come in and would bring it over that, we
simply start kfree'ing the surplus.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
In preparation for adding limits, and one more user, abstract out the
core bits of the allocation+free cache.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
In overflow we see a duplcate line in the trace, and in some cases 3
lines (if initial io_post_aux_cqe fails).
Instead just trace once for each CQE
Signed-off-by: Dylan Yudaken <dylany@fb.com>
Link: https://lore.kernel.org/r/20220630091231.1456789-13-dylany@fb.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Some use cases of io_post_aux_cqe would not want to overflow as is, but
might want to change the flags/result. For example multishot receive
requires in order CQE, and so if there is an overflow it would need to
stop receiving until the overflow is taken care of.
Signed-off-by: Dylan Yudaken <dylany@fb.com>
Link: https://lore.kernel.org/r/20220630091231.1456789-8-dylany@fb.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
From recently io_uring provides an option to allocate a file index for
operation registering fixed files. However, it's utterly unusable with
mixed approaches when for a part of files the userspace knows better
where to place it, as it may race and users don't have any sane way to
pick a slot and hoping it will not be taken.
Let the userspace to register a range of fixed file slots in which the
auto-allocation happens. The use case is splittting the fixed table in
two parts, where on of them is used for auto-allocation and another for
slot-specified operations.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/66ab0394e436f38437cf7c44676e1920d09687ad.1656154403.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
io_uring_enter() takes ctx->refs, which was previously preventing racing
with register quiesce. However, as register now doesn't touch the refs,
we can freely kill extra ctx pinning and rely on the fact that we're
holding a file reference preventing the ring from being destroyed.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/a11c57ad33a1be53541fce90669c1b79cf4d8940.1656153286.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The io_uring cancelation API is async, like any other API that we expose
there. For the case of finding a request to cancel, or not finding one,
it is fully sync in that when submission returns, the CQE for both the
cancelation request and the targeted request have been posted to the
CQ ring.
However, if the targeted work is being executed by io-wq, the API can
only start the act of canceling it. This makes it difficult to use in
some circumstances, as the caller then has to wait for the CQEs to come
in and match on the same cancelation data there.
Provide a IORING_REGISTER_SYNC_CANCEL command for io_uring_register()
that does sync cancelations, always. For the io-wq case, it'll wait
for the cancelation to come in before returning. The only expected
returns from this API is:
0 Request found and canceled fine.
> 0 Requests found and canceled. Only happens if asked to
cancel multiple requests, and if the work wasn't in
progress.
-ENOENT Request not found.
-ETIME A timeout on the operation was requested, but the timeout
expired before we could cancel.
and we won't get -EALREADY via this API.
If the timeout value passed in is -1 (tv_sec and tv_nsec), then that
means that no timeout is requested. Otherwise, the timespec passed in
is the amount of time the sync cancel will wait for a successful
cancelation.
Link: https://github.com/axboe/liburing/discussions/608
Signed-off-by: Jens Axboe <axboe@kernel.dk>
trace task_work_run to help provide stats on how often task work is run
and what batch sizes are coming through.
Signed-off-by: Dylan Yudaken <dylany@fb.com>
Link: https://lore.kernel.org/r/20220622134028.2013417-9-dylany@fb.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Batching task work up is an important performance optimisation, as
task_work_add is expensive.
In order to keep the semantics replace the task_list with a fake node
while processing the old list, and then do a cmpxchg at the end to see if
there is more work.
Signed-off-by: Dylan Yudaken <dylany@fb.com>
Link: https://lore.kernel.org/r/20220622134028.2013417-6-dylany@fb.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
With networking use cases we see contention on the spinlock used to
protect the task_list when multiple threads try and add completions at once.
Instead we can use a lockless list, and assume that the first caller to
add to the list is responsible for kicking off task work.
Signed-off-by: Dylan Yudaken <dylany@fb.com>
Link: https://lore.kernel.org/r/20220622134028.2013417-4-dylany@fb.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
This optimisation has some built in assumptions that make it easy to
introduce bugs. It also does not have clear wins that make it worth keeping.
Signed-off-by: Dylan Yudaken <dylany@fb.com>
Link: https://lore.kernel.org/r/20220622134028.2013417-2-dylany@fb.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
spin_lock(&ctx->completion_lock);
/* post CQEs */
io_commit_cqring(ctx);
spin_unlock(&ctx->completion_lock);
io_cqring_ev_posted(ctx);
We have many places repeating this sequence, and the three function
unlock section is not perfect from the maintainance perspective and also
makes it harder to add new locking/sync trick.
Introduce two helpers. io_cq_lock(), which is simple and only grabs
->completion_lock, and io_cq_unlock_post() encapsulating the three call
section.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/fe0c682bf7f7b55d9be55b0d034be9c1949277dc.1655684496.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Some io_uring-eventfd users assume that there won't be spurious wakeups.
That assumption has to be honoured by all io_cqring_ev_posted() callers,
which is inconvenient and from time to time leads to problems but should
be maintained to not break the userspace.
Instead of making the callers track whether a CQE was posted or not, hide
it inside io_eventfd_signal(). It saves ->cached_cq_tail it saw last time
and triggers the eventfd only when ->cached_cq_tail changed since then.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/0ffc66bae37a2513080b601e4370e147faaa72c5.1655684496.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
io_uring_try_cancel_requests() loops until there is nothing left to do
with the ring, however there might be several rings and they might have
dependencies between them, e.g. via poll requests.
Instead of cancelling rings one by one, try to cancel them all and only
then loop over if we still potenially some work to do.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/8d491fe02d8ac4c77ff38061cf86b9a827e8845c.1655684496.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
It's not clear how widely used IOSQE_CQE_SKIP_SUCCESS is, and how often
->flush_cqes flag prevents from completion being flushed. Sometimes it's
high level of concurrency that enables it at least for one CQE, but
sometimes it doesn't save much because nobody waiting on the CQ.
Remove ->flush_cqes flag and the optimisation, it should benefit the
normal use case. Note, that there is no spurious eventfd problem with
that as checks for spuriousness were incorporated into
io_eventfd_signal().
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/692e81eeddccc096f449a7960365fa7b4a18f8e6.1655637157.git.asml.silence@gmail.com
[axboe: remove now dead state->flush_cqes variable]
Signed-off-by: Jens Axboe <axboe@kernel.dk>
We have lots of trace events accepting an io_uring request and wanting
to print some of its fields like user_data, opcode, flags and so on.
However, as trace points were unaware of io_uring structures, we had to
pass all the fields as arguments. Teach trace/events/io_uring.h about
struct io_kiocb and stop the misery of passing a horde of arguments to
trace helpers.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/40ff72f92798114e56d400f2b003beb6cde6ef53.1655384063.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
__io_fill_cqe_req() is hot and inlined, we want it to be as small as
possible. Add io_req_cqe_overflow() accepting only a request and doing
all overflow accounting, and replace with it two calls to 6 argument
io_cqring_event_overflow().
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/048b9fbcce56814d77a1a540409c98c3d383edcb.1655455613.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Currently we do two extra spin lock/unlock pairs to add a poll/apoll
request to the cancellation hash table and remove it from there.
On the submission side we often already hold ->uring_lock and tw
completion is likely to hold it as well. Add a second cancellation hash
table protected by ->uring_lock. In concerns for latency because of a
need to have the mutex locked on the completion side, use the new table
only in following cases:
1) IORING_SETUP_SINGLE_ISSUER: only one task grabs uring_lock, so there
is little to no contention and so the main tw hander will almost
always end up grabbing it before calling callbacks.
2) IORING_SETUP_SQPOLL: same as with single issuer, only one task is
a major user of ->uring_lock.
3) apoll: we normally grab the lock on the completion side anyway to
execute the request, so it's free.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/1bbad9c78c454b7b92f100bbf46730a37df7194f.1655371007.git.asml.silence@gmail.com
Reviewed-by: Hao Xu <howeyxu@tencent.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Add a new IORING_SETUP_SINGLE_ISSUER flag and the userspace visible part
of it, i.e. put limitations of submitters. Also, don't allow it together
with IOPOLL as we're not going to put it to good use.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/4bcc41ee467fdf04c8aab8baf6ce3ba21858c3d4.1655371007.git.asml.silence@gmail.com
Reviewed-by: Hao Xu <howeyxu@tencent.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Don't allocate to many hash/cancellation buckets, there might be too
many, clamp it to 8 bits, or 256 * 64B = 16KB. We don't usually have too
many requests, and 256 buckets should be enough, especially since we
do hash search only in the cancellation path.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/b9620c8072ba61a2d50eba894b89bd93a94a9abd.1655371007.git.asml.silence@gmail.com
Reviewed-by: Hao Xu <howeyxu@tencent.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Add a new io_hash_bucket structure so that each bucket in cancel_hash
has separate spinlock. Use per entry lock for cancel_hash, this removes
some completion lock invocation and remove contension between different
cancel_hash entries.
Signed-off-by: Hao Xu <howeyxu@tencent.com>
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/05d1e135b0c8bce9d1441e6346776589e5783e26.1655371007.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
REQ_F_COMPLETE_INLINE is only needed to delay queueing into the
completion list to io_queue_sqe() as __io_req_complete() is inlined and
we don't want to bloat the kernel.
As now we complete in a more centralised fashion in io_issue_sqe() we
can get rid of the flag and queue to the list directly.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/600ba20a9338b8a39b249b23d3d177803613dde4.1655371007.git.asml.silence@gmail.com
Reviewed-by: Hao Xu <howeyxu@tencent.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Luckily, nnobody completes multi-apoll requests outside the polling
functions, but don't set IO_URING_F_COMPLETE_DEFER in any case as
there is nobody who is catching REQ_F_COMPLETE_INLINE, and so will leak
requests if used.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/a65ed3f5effd9321ee06e6edea294a03be3e15a0.1655310733.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
We already have the declarations in opdef.h, move the rest into its own
file rather than in the main io_uring.c file.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
This also helps cleanup the io_uring.h cancel parts, as we can make
things static in the cancel.c file, mostly.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Add a io_poll_issue() rather than export the general task_work locking
and io_issue_sqe(), and put the io_op_defs definition and structure into
a separate header file so that poll can use it.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Remove some dead headers we no longer need, and get rid of the
io_ring_ctx and io_uring_fops forward declarations.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
This also means moving a bit more of the fixed file handling to the
filetable side, which makes sense separately too.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Would be nice to sort out Kconfig for this and don't even compile
epoll.c if we don't have epoll configured.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Normally request handlers complete requests themselves, if they don't
return an error. For the latter case, the core will complete it for
them.
This is unhandy for pushing opcode handlers further out, as we don't
want a bunch of inline completion code and we don't want to make the
completion path slower than it is now.
Let the core handle any completion, unless the handler explicitly
asks us not to.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
This can move request type specific cleanup into a private handler,
removing the need for the core io_uring parts to know what types
they are dealing with.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
There's a special case for recvmsg with MSG_ERRQUEUE set. This is
problematic as it means the core needs to know about this special
request type.
For now, just add a generic flag for it.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Remove struct io_poll_update from io_kiocb, and convert the poll path to
use the io_cmd_type approach instead.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Remove struct io_poll_iocb from io_kiocb, and convert the poll path to
use the io_cmd_type approach instead.
While at it, rename io_poll_iocb to io_poll which is consistent with the
other request type private structures.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Remove struct io_rw from io_kiocb, and convert the read/write path to
use the io_cmd_type approach instead.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Each opcode generally has a command structure in io_kiocb which it can
use to store data associated with that request.
In preparation for having the core layer not know about what's inside
these fields, add a generic io_cmd_data type and put in the union as
well.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Define an io_op_def->prep_async() handler and push the async preparation
to there. Since we now have that, we can drop ->needs_async_setup, as
they mean the same thing.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
In preparation for splitting io_uring up a bit, move it into its own
top level directory. It didn't really belong in fs/ anyway, as it's
not a file system only API.
This adds io_uring/ and moves the core files in there, and updates the
MAINTAINERS file for the new location.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
