Commit Graph

85 Commits

Author SHA1 Message Date
Jens Axboe
414d0f45c3 io_uring/alloc_cache: switch to array based caching
Currently lists are being used to manage this, but best practice is
usually to have these in an array instead as that it cheaper to manage.

Outside of that detail, games are also played with KASAN as the list
is inside the cached entry itself.

Finally, all users of this need a struct io_cache_entry embedded in
their struct, which is union'ized with something else in there that
isn't used across the free -> realloc cycle.

Get rid of all of that, and simply have it be an array. This will not
change the memory used, as we're just trading an 8-byte member entry
for the per-elem array size.

This reduces the overhead of the recycled allocations, and it reduces
the amount of code code needed to support recycling to about half of
what it currently is.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
2024-04-15 08:10:25 -06:00
Pavel Begunkov
e5c12945be io_uring: refactor io_fill_cqe_req_aux
The restriction on multishot execution context disallowing io-wq is
driven by rules of io_fill_cqe_req_aux(), it should only be called in
the master task context, either from the syscall path or in task_work.
Since task_work now always takes the ctx lock implying
IO_URING_F_COMPLETE_DEFER, we can just assume that the function is
always called with its defer argument set to true.

Kill the argument. Also rename the function for more consistency as
"fill" in CQE related functions was usually meant for raw interfaces
only copying data into the CQ without any locking, waking the user
and other accounting "post" functions take care of.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Tested-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/93423d106c33116c7d06bf277f651aa68b427328.1710799188.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2024-04-15 08:10:24 -06:00
Pavel Begunkov
8e5b3b89ec io_uring: remove struct io_tw_state::locked
ctx is always locked for task_work now, so get rid of struct
io_tw_state::locked. Note I'm stopping one step before removing
io_tw_state altogether, which is not empty, because it still serves the
purpose of indicating which function is a tw callback and forcing users
not to invoke them carelessly out of a wrong context. The removal can
always be done later.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Tested-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/e95e1ea116d0bfa54b656076e6a977bc221392a4.1710799188.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2024-04-15 08:10:24 -06:00
Pavel Begunkov
5e3afe580a io_uring: fix poll_remove stalled req completion
Taking the ctx lock is not enough to use the deferred request completion
infrastructure, it'll get queued into the list but no one would expect
it there, so it will sit there until next io_submit_flush_completions().
It's hard to care about the cancellation path, so complete it via tw.

Fixes: ef7dfac51d ("io_uring/poll: serialize poll linked timer start with poll removal")
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/c446740bc16858f8a2a8dcdce899812f21d15f23.1710514702.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2024-03-15 09:36:56 -06:00
Pavel Begunkov
1a8ec63b2b io_uring: fix io_queue_proc modifying req->flags
With multiple poll entries __io_queue_proc() might be running in
parallel with poll handlers and possibly task_work, we should not be
carelessly modifying req->flags there. io_poll_double_prepare() handles
a similar case with locking but it's much easier to move it into
__io_arm_poll_handler().

Cc: stable@vger.kernel.org
Fixes: 595e52284d ("io_uring/poll: don't enable lazy wake for POLLEXCLUSIVE")
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/455cc49e38cf32026fa1b49670be8c162c2cb583.1709834755.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2024-03-07 11:10:28 -07:00
Stefan Roesch
8d0c12a80c io-uring: add napi busy poll support
This adds the napi busy polling support in io_uring.c. It adds a new
napi_list to the io_ring_ctx structure. This list contains the list of
napi_id's that are currently enabled for busy polling. The list is
synchronized by the new napi_lock spin lock. The current default napi
busy polling time is stored in napi_busy_poll_to. If napi busy polling
is not enabled, the value is 0.

In addition there is also a hash table. The hash table store the napi
id and the pointer to the above list nodes. The hash table is used to
speed up the lookup to the list elements. The hash table is synchronized
with rcu.

The NAPI_TIMEOUT is stored as a timeout to make sure that the time a
napi entry is stored in the napi list is limited.

The busy poll timeout is also stored as part of the io_wait_queue. This
is necessary as for sq polling the poll interval needs to be adjusted
and the napi callback allows only to pass in one value.

This has been tested with two simple programs from the liburing library
repository: the napi client and the napi server program. The client
sends a request, which has a timestamp in its payload and the server
replies with the same payload. The client calculates the roundtrip time
and stores it to calculate the results.

The client is running on host1 and the server is running on host 2 (in
the same rack). The measured times below are roundtrip times. They are
average times over 5 runs each. Each run measures 1 million roundtrips.

                   no rx coal          rx coal: frames=88,usecs=33
Default              57us                    56us

client_poll=100us    47us                    46us

server_poll=100us    51us                    46us

client_poll=100us+   40us                    40us
server_poll=100us

client_poll=100us+   41us                    39us
server_poll=100us+
prefer napi busy poll on client

client_poll=100us+   41us                    39us
server_poll=100us+
prefer napi busy poll on server

client_poll=100us+   41us                    39us
server_poll=100us+
prefer napi busy poll on client + server

Signed-off-by: Stefan Roesch <shr@devkernel.io>
Suggested-by: Olivier Langlois <olivier@trillion01.com>
Acked-by: Jakub Kicinski <kuba@kernel.org>
Link: https://lore.kernel.org/r/20230608163839.2891748-5-shr@devkernel.io
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2024-02-09 11:54:19 -07:00
Jens Axboe
3cdc4be114 io_uring/poll: improve readability of poll reference decrementing
This overly long line is hard to read. Break it up by AND'ing the
ref mask first, then perform the atomic_sub_return() with the value
itself.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
2024-02-08 13:27:06 -07:00
Jens Axboe
95041b93e9 io_uring: add io_file_can_poll() helper
This adds a flag to avoid dipping dereferencing file and then f_op to
figure out if the file has a poll handler defined or not. We generally
call this at least twice for networked workloads, and if using ring
provided buffers, we do it on every buffer selection. Particularly the
latter is troublesome, as it's otherwise a very fast operation.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
2024-02-08 13:27:06 -07:00
Jens Axboe
521223d7c2 io_uring/cancel: don't default to setting req->work.cancel_seq
Just leave it unset by default, avoiding dipping into the last
cacheline (which is otherwise untouched) for the fast path of using
poll to drive networked traffic. Add a flag that tells us if the
sequence is valid or not, and then we can defer actually assigning
the flag and sequence until someone runs cancelations.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
2024-02-08 13:27:06 -07:00
Jens Axboe
704ea888d6 io_uring/poll: add requeue return code from poll multishot handling
Since our poll handling is edge triggered, multishot handlers retry
internally until they know that no more data is available. In
preparation for limiting these retries, add an internal return code,
IOU_REQUEUE, which can be used to inform the poll backend about the
handler wanting to retry, but that this should happen through a normal
task_work requeue rather than keep hammering on the issue side for this
one request.

No functional changes in this patch, nobody is using this return code
just yet.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
2024-01-29 13:19:47 -07:00
Jens Axboe
e84b01a880 io_uring/poll: move poll execution helpers higher up
In preparation for calling __io_poll_execute() higher up, move the
functions to avoid forward declarations.

No functional changes in this patch.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
2024-01-29 13:19:17 -07:00
Jens Axboe
595e52284d io_uring/poll: don't enable lazy wake for POLLEXCLUSIVE
There are a few quirks around using lazy wake for poll unconditionally,
and one of them is related the EPOLLEXCLUSIVE. Those may trigger
exclusive wakeups, which wake a limited number of entries in the wait
queue. If that wake number is less than the number of entries someone is
waiting for (and that someone is also using DEFER_TASKRUN), then we can
get stuck waiting for more entries while we should be processing the ones
we already got.

If we're doing exclusive poll waits, flag the request as not being
compatible with lazy wakeups.

Reported-by: Pavel Begunkov <asml.silence@gmail.com>
Fixes: 6ce4a93dbb ("io_uring/poll: use IOU_F_TWQ_LAZY_WAKE for wakeups")
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-12-13 08:58:15 -07:00
Jens Axboe
6ce4a93dbb io_uring/poll: use IOU_F_TWQ_LAZY_WAKE for wakeups
With poll triggered retries, each event trigger will cause a task_work
item to be added for processing. If the ring is setup with
IORING_SETUP_DEFER_TASKRUN and a task is waiting on multiple events to
complete, any task_work addition will wake the task for processing these
items. This can cause more context switches than we would like, if the
application is deliberately waiting on multiple items to increase
efficiency.

For example, if an application has receive multishot armed for sockets
and wants to wait for N to complete within M usec of time, we should not
be waking up and processing these items until we have all the events we
asked for. By switching the poll trigger to lazy wake, we'll process
them when they are all ready, in one swoop, rather than wake multiple
times only to process one and then go back to sleep.

At some point we probably want to look at just making the lazy wake
the default, but for now, let's just selectively enable it where it
makes sense.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-10-19 06:42:29 -06:00
Pavel Begunkov
b6b2bb58a7 io_uring: never overflow io_aux_cqe
Now all callers of io_aux_cqe() set allow_overflow to false, remove the
parameter and not allow overflowing auxilary multishot cqes.

When CQ is full the function callers and all multishot requests in
general are expected to complete the request. That prevents indefinite
in-background grows of the overflow list and let's the userspace to
handle the backlog at its own pace.

Resubmitting a request should also be faster than accounting a bunch of
overflows, so it should be better for perf when it happens, but a well
behaving userspace should be trying to avoid overflows in any case.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/bb20d14d708ea174721e58bb53786b0521e4dd6d.1691757663.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-08-11 10:42:57 -06:00
Jens Axboe
d7b8b079a8 io_uring/cancel: support opcode based lookup and cancelation
Add IORING_ASYNC_CANCEL_OP flag for cancelation, which allows the
application to target cancelation based on the opcode of the original
request.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-07-17 10:05:48 -06:00
Jens Axboe
a30badf66d io_uring: use cancelation match helper for poll and timeout requests
Get rid of the request vs io_cancel_data checking and just use the
exported helper for this.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-07-17 10:05:48 -06:00
Jens Axboe
ad711c5d11 io_uring/poll: always set 'ctx' in io_cancel_data
This isn't strictly necessary for this callsite, as it uses it's
internal lookup for this cancelation purpose. But let's be consistent
with how it's used in general and set ctx as well.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-07-17 10:05:48 -06:00
Linus Torvalds
0aa69d53ac for-6.5/io_uring-2023-06-23
-----BEGIN PGP SIGNATURE-----
 
 iQJEBAABCAAuFiEEwPw5LcreJtl1+l5K99NY+ylx4KYFAmSV8cEQHGF4Ym9lQGtl
 cm5lbC5kawAKCRD301j7KXHgpnvZD/0QWstFCe1CSLWaycdC9fhWftFt3hyEIST5
 CYEL56UZrDWNkv9xTLe855xvMavjd4sdHlUa8NUPghRQeJyKYgRxHBLXRWmy0uNN
 l47Zjiwsolmbr3Nt6qViLdCDYmG39ZGNwWo8b6p3ybWYLtzxeOblocOBTPzoCtkS
 hjo7Z0eMONvsvLX+l0o9IDdWtZIQ2fGo4VYkMIVb6CyxRPpuUuPKbE25qaTx+uBg
 Fy6Qa3SlaTwzqcg3dttggjIP792L/eETUCWGndg5pJNbrkj/fI4Vm4bEljID76eS
 HODl+pWHmyM6avVkypr7N3Tp5HKF0OTUa4vJTLIZo1QiRu6zlXphtuvGn+McEmgV
 hbYmQMYWzqJ22k2iEpCR58pdhmZJC9uB8r4Rwgr/t9GKqt4E+15EzmqkG9cUVMGV
 rfbBwVLwBUd5+0WHwQ8RzdtaUPt17vSIW/8WhU5zoMGVotqVBHO/H+5BtmKPWWpq
 fx1etQ8XJVPIxziJvgsEitb1s6KZzJspcONDlLEitmZkflv3gGdVm99KNbXwJpcp
 m6+FcYQ5d5FivfLPGgpx8go+4M2QuoW2yRGwZHu54buCnpxgNjIk898OjrUrdXCg
 3/0m99GXmOWQQl0VrrTr+Fv99nVsQ2hMQzOFJGMYRtHEEc5xiTcJiZmoxmF7T7/C
 TipyW3czsw==
 =5Me8
 -----END PGP SIGNATURE-----

Merge tag 'for-6.5/io_uring-2023-06-23' of git://git.kernel.dk/linux

Pull io_uring updates from Jens Axboe:
 "Nothing major in this release, just a bunch of cleanups and some
  optimizations around networking mostly.

   - clean up file request flags handling (Christoph)

   - clean up request freeing and CQ locking (Pavel)

   - support for using pre-registering the io_uring fd at setup time
     (Josh)

   - Add support for user allocated ring memory, rather than having the
     kernel allocate it. Mostly for packing rings into a huge page (me)

   - avoid an unnecessary double retry on receive (me)

   - maintain ordering for task_work, which also improves performance
     (me)

   - misc cleanups/fixes (Pavel, me)"

* tag 'for-6.5/io_uring-2023-06-23' of git://git.kernel.dk/linux: (39 commits)
  io_uring: merge conditional unlock flush helpers
  io_uring: make io_cq_unlock_post static
  io_uring: inline __io_cq_unlock
  io_uring: fix acquire/release annotations
  io_uring: kill io_cq_unlock()
  io_uring: remove IOU_F_TWQ_FORCE_NORMAL
  io_uring: don't batch task put on reqs free
  io_uring: move io_clean_op()
  io_uring: inline io_dismantle_req()
  io_uring: remove io_free_req_tw
  io_uring: open code io_put_req_find_next
  io_uring: add helpers to decode the fixed file file_ptr
  io_uring: use io_file_from_index in io_msg_grab_file
  io_uring: use io_file_from_index in __io_sync_cancel
  io_uring: return REQ_F_ flags from io_file_get_flags
  io_uring: remove io_req_ffs_set
  io_uring: remove a confusing comment above io_file_get_flags
  io_uring: remove the mode variable in io_file_get_flags
  io_uring: remove __io_file_supports_nowait
  io_uring: wait interruptibly for request completions on exit
  ...
2023-06-26 12:30:26 -07:00
Jens Axboe
ef7dfac51d io_uring/poll: serialize poll linked timer start with poll removal
We selectively grab the ctx->uring_lock for poll update/removal, but
we really should grab it from the start to fully synchronize with
linked timeouts. Normally this is indeed the case, but if requests
are forced async by the application, we don't fully cover removal
and timer disarm within the uring_lock.

Make this simpler by having consistent locking state for poll removal.

Cc: stable@vger.kernel.org # 6.1+
Reported-by: Querijn Voet <querijnqyn@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-06-17 20:21:52 -06:00
Jens Axboe
d86eaed185 io_uring: cleanup io_aux_cqe() API
Everybody is passing in the request, so get rid of the io_ring_ctx and
explicit user_data pass-in. Both the ctx and user_data can be deduced
from the request at hand.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-06-07 14:59:22 -06:00
Jens Axboe
c92fcfc2ba io_uring: avoid indirect function calls for the hottest task_work
We use task_work for a variety of reasons, but doing completions or
triggering rety after poll are by far the hottest two. Use the indirect
funtion call wrappers to avoid the indirect function call if
CONFIG_RETPOLINE is set.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-06-02 08:55:37 -06:00
Pavel Begunkov
a282967c84 io_uring: encapsulate task_work state
For task works we're passing around a bool pointer for whether the
current ring is locked or not, let's wrap it in a structure, that
will make it more opaque preventing abuse and will also help us
to pass more info in the future if needed.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/1ecec9483d58696e248d1bfd52cf62b04442df1d.1679931367.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-04-03 07:16:15 -06:00
Jens Axboe
005308f7bd io_uring/poll: clear single/double poll flags on poll arming
Unless we have at least one entry queued, then don't call into
io_poll_remove_entries(). Normally this isn't possible, but if we
retry poll then we can have ->nr_entries cleared again as we're
setting it up. If this happens for a poll retry, then we'll still have
at least REQ_F_SINGLE_POLL set. io_poll_remove_entries() then thinks
it has entries to remove.

Clear REQ_F_SINGLE_POLL and REQ_F_DOUBLE_POLL unconditionally when
arming a poll request.

Fixes: c16bda3759 ("io_uring/poll: allow some retries for poll triggering spuriously")
Cc: stable@vger.kernel.org
Reported-by: Pengfei Xu <pengfei.xu@intel.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-03-28 07:09:01 -06:00
Jens Axboe
1947ddf9b3 io_uring/poll: don't pass in wake func to io_init_poll_iocb()
We only use one, and it's io_poll_wake(). Hardwire that in the initial
init, as well as in __io_queue_proc() if we're setting up for double
poll.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-03-01 10:06:53 -07:00
Jens Axboe
c16bda3759 io_uring/poll: allow some retries for poll triggering spuriously
If we get woken spuriously when polling and fail the operation with
-EAGAIN again, then we generally only allow polling again if data
had been transferred at some point. This is indicated with
REQ_F_PARTIAL_IO. However, if the spurious poll triggers when the socket
was originally empty, then we haven't transferred data yet and we will
fail the poll re-arm. This either punts the socket to io-wq if it's
blocking, or it fails the request with -EAGAIN if not. Neither condition
is desirable, as the former will slow things down, while the latter
will make the application confused.

We want to ensure that a repeated poll trigger doesn't lead to infinite
work making no progress, that's what the REQ_F_PARTIAL_IO check was
for. But it doesn't protect against a loop post the first receive, and
it's unnecessarily strict if we started out with an empty socket.

Add a somewhat random retry count, just to put an upper limit on the
potential number of retries that will be done. This should be high enough
that we won't really hit it in practice, unless something needs to be
aborted anyway.

Cc: stable@vger.kernel.org # v5.10+
Link: https://github.com/axboe/liburing/issues/364
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-02-25 20:10:13 -07:00
Breno Leitao
a7dd27828b io_uring: Rename struct io_op_def
The current io_op_def struct is becoming huge and the name is a bit
generic.

The goal of this patch is to rename this struct to `io_issue_def`. This
struct will contain the hot functions associated with the issue code
path.

For now, this patch only renames the structure, and an upcoming patch
will break up the structure in two, moving the non-issue fields to a
secondary struct.

Signed-off-by: Breno Leitao <leitao@debian.org>
Reviewed-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/20230112144411.2624698-1-leitao@debian.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-01-29 15:17:41 -07:00
Jens Axboe
8caa03f10b io_uring/poll: don't reissue in case of poll race on multishot request
A previous commit fixed a poll race that can occur, but it's only
applicable for multishot requests. For a multishot request, we can safely
ignore a spurious wakeup, as we never leave the waitqueue to begin with.

A blunt reissue of a multishot armed request can cause us to leak a
buffer, if they are ring provided. While this seems like a bug in itself,
it's not really defined behavior to reissue a multishot request directly.
It's less efficient to do so as well, and not required to rearm anything
like it is for singleshot poll requests.

Cc: stable@vger.kernel.org
Fixes: 6e5aedb932 ("io_uring/poll: attempt request issue after racy poll wakeup")
Reported-and-tested-by: Olivier Langlois <olivier@trillion01.com>
Link: https://github.com/axboe/liburing/issues/778
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-01-20 15:11:54 -07:00
Jens Axboe
6e5aedb932 io_uring/poll: attempt request issue after racy poll wakeup
If we have multiple requests waiting on the same target poll waitqueue,
then it's quite possible to get a request triggered and get disappointed
in not being able to make any progress with it. If we race in doing so,
we'll potentially leave the poll request on the internal tables, but
removed from the waitqueue. That means that any subsequent trigger of
the poll waitqueue will not kick that request into action, causing an
application to potentially wait for completion of a request that will
never happen.

Fix this by adding a new poll return state, IOU_POLL_REISSUE. Rather
than have complicated logic for how to re-arm a given type of request,
just punt it for a reissue.

While in there, move the 'ret' variable to the only section where it
gets used. This avoids confusion the scope of it.

Cc: stable@vger.kernel.org
Fixes: eb0089d629 ("io_uring: single shot poll removal optimisation")
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-01-12 10:35:51 -07:00
Jens Axboe
febb985c06 io_uring/poll: add hash if ready poll request can't complete inline
If we don't, then we may lose access to it completely, leading to a
request leak. This will eventually stall the ring exit process as
well.

Cc: stable@vger.kernel.org
Fixes: 49f1c68e04 ("io_uring: optimise submission side poll_refs")
Reported-and-tested-by: syzbot+6c95df01470a47fc3af4@syzkaller.appspotmail.com
Link: https://lore.kernel.org/io-uring/0000000000009f829805f1ce87b2@google.com/
Suggested-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2023-01-09 15:46:57 -07:00
Pavel Begunkov
443e575506 io_uring: combine poll tw handlers
Merge apoll and regular poll tw handlers, it will help with inlining.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/482e59edb9fc81bd275fdbf486837330fb27120a.1669821213.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-11-30 10:27:34 -07:00
Pavel Begunkov
c3bfb57ea7 io_uring: improve poll warning handling
Don't try to complete requests if their refs are broken and we've got
a warning, it's much better to drop them and potentially leaking than
double freeing.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/31edf9f96f05d03ab62c114508a231a2dce434cb.1669821213.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-11-30 10:27:22 -07:00
Pavel Begunkov
047b6aef09 io_uring: remove ctx variable in io_poll_check_events
ctx is only used by io_poll_check_events() for multishot poll CQE
posting, don't save it on stack in advance.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/552c1771f8a0e7688afdb4f538ead245f53e80e7.1669821213.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-11-30 10:26:57 -07:00
Pavel Begunkov
9805fa2d94 io_uring: carve io_poll_check_events fast path
The fast path in io_poll_check_events() is when we have only one
(i.e. master) reference. Move all verification, cancellations
checks, edge case handling and so on under a common if.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/8c21c5d5e027e32dc553705e88796dec79ff6f93.1669821213.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-11-30 10:26:57 -07:00
Jens Axboe
b2cf789f6c Merge branch 'for-6.2/io_uring' into for-6.2/io_uring-next
* for-6.2/io_uring: (41 commits)
  io_uring: keep unlock_post inlined in hot path
  io_uring: don't use complete_post in kbuf
  io_uring: spelling fix
  io_uring: remove io_req_complete_post_tw
  io_uring: allow multishot polled reqs to defer completion
  io_uring: remove overflow param from io_post_aux_cqe
  io_uring: add lockdep assertion in io_fill_cqe_aux
  io_uring: make io_fill_cqe_aux static
  io_uring: add io_aux_cqe which allows deferred completion
  io_uring: allow defer completion for aux posted cqes
  io_uring: defer all io_req_complete_failed
  io_uring: always lock in io_apoll_task_func
  io_uring: remove iopoll spinlock
  io_uring: iopoll protect complete_post
  io_uring: inline __io_req_complete_put()
  io_uring: remove io_req_tw_post_queue
  io_uring: use io_req_task_complete() in timeout
  io_uring: hold locks for io_req_complete_failed
  io_uring: add completion locking for iopoll
  io_uring: kill io_cqring_ev_posted() and __io_cq_unlock_post()
  ...
2022-11-29 12:08:37 -07:00
Lin Ma
12ad3d2d6c io_uring/poll: fix poll_refs race with cancelation
There is an interesting race condition of poll_refs which could result
in a NULL pointer dereference. The crash trace is like:

KASAN: null-ptr-deref in range [0x0000000000000008-0x000000000000000f]
CPU: 0 PID: 30781 Comm: syz-executor.2 Not tainted 6.0.0-g493ffd6605b2 #1
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS
1.13.0-1ubuntu1.1 04/01/2014
RIP: 0010:io_poll_remove_entry io_uring/poll.c:154 [inline]
RIP: 0010:io_poll_remove_entries+0x171/0x5b4 io_uring/poll.c:190
Code: ...
RSP: 0018:ffff88810dfefba0 EFLAGS: 00010202
RAX: 0000000000000001 RBX: 0000000000000000 RCX: 0000000000040000
RDX: ffffc900030c4000 RSI: 000000000003ffff RDI: 0000000000040000
RBP: 0000000000000008 R08: ffffffff9764d3dd R09: fffffbfff3836781
R10: fffffbfff3836781 R11: 0000000000000000 R12: 1ffff11003422d60
R13: ffff88801a116b04 R14: ffff88801a116ac0 R15: dffffc0000000000
FS:  00007f9c07497700(0000) GS:ffff88811a600000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 00007ffb5c00ea98 CR3: 0000000105680005 CR4: 0000000000770ef0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
PKRU: 55555554
Call Trace:
 <TASK>
 io_apoll_task_func+0x3f/0xa0 io_uring/poll.c:299
 handle_tw_list io_uring/io_uring.c:1037 [inline]
 tctx_task_work+0x37e/0x4f0 io_uring/io_uring.c:1090
 task_work_run+0x13a/0x1b0 kernel/task_work.c:177
 get_signal+0x2402/0x25a0 kernel/signal.c:2635
 arch_do_signal_or_restart+0x3b/0x660 arch/x86/kernel/signal.c:869
 exit_to_user_mode_loop kernel/entry/common.c:166 [inline]
 exit_to_user_mode_prepare+0xc2/0x160 kernel/entry/common.c:201
 __syscall_exit_to_user_mode_work kernel/entry/common.c:283 [inline]
 syscall_exit_to_user_mode+0x58/0x160 kernel/entry/common.c:294
 entry_SYSCALL_64_after_hwframe+0x63/0xcd

The root cause for this is a tiny overlooking in
io_poll_check_events() when cocurrently run with poll cancel routine
io_poll_cancel_req().

The interleaving to trigger use-after-free:

CPU0                                       |  CPU1
                                           |
io_apoll_task_func()                       |  io_poll_cancel_req()
 io_poll_check_events()                    |
  // do while first loop                   |
  v = atomic_read(...)                     |
  // v = poll_refs = 1                     |
  ...                                      |  io_poll_mark_cancelled()
                                           |   atomic_or()
                                           |   // poll_refs =
IO_POLL_CANCEL_FLAG | 1
                                           |
  atomic_sub_return(...)                   |
  // poll_refs = IO_POLL_CANCEL_FLAG       |
  // loop continue                         |
                                           |
                                           |  io_poll_execute()
                                           |   io_poll_get_ownership()
                                           |   // poll_refs =
IO_POLL_CANCEL_FLAG | 1
                                           |   // gets the ownership
  v = atomic_read(...)                     |
  // poll_refs not change                  |
                                           |
  if (v & IO_POLL_CANCEL_FLAG)             |
   return -ECANCELED;                      |
  // io_poll_check_events return           |
  // will go into                          |
  // io_req_complete_failed() free req     |
                                           |
                                           |  io_apoll_task_func()
                                           |  // also go into
io_req_complete_failed()

And the interleaving to trigger the kernel WARNING:

CPU0                                       |  CPU1
                                           |
io_apoll_task_func()                       |  io_poll_cancel_req()
 io_poll_check_events()                    |
  // do while first loop                   |
  v = atomic_read(...)                     |
  // v = poll_refs = 1                     |
  ...                                      |  io_poll_mark_cancelled()
                                           |   atomic_or()
                                           |   // poll_refs =
IO_POLL_CANCEL_FLAG | 1
                                           |
  atomic_sub_return(...)                   |
  // poll_refs = IO_POLL_CANCEL_FLAG       |
  // loop continue                         |
                                           |
  v = atomic_read(...)                     |
  // v = IO_POLL_CANCEL_FLAG               |
                                           |  io_poll_execute()
                                           |   io_poll_get_ownership()
                                           |   // poll_refs =
IO_POLL_CANCEL_FLAG | 1
                                           |   // gets the ownership
                                           |
  WARN_ON_ONCE(!(v & IO_POLL_REF_MASK)))   |
  // v & IO_POLL_REF_MASK = 0 WARN         |
                                           |
                                           |  io_apoll_task_func()
                                           |  // also go into
io_req_complete_failed()

By looking up the source code and communicating with Pavel, the
implementation of this atomic poll refs should continue the loop of
io_poll_check_events() just to avoid somewhere else to grab the
ownership. Therefore, this patch simply adds another AND operation to
make sure the loop will stop if it finds the poll_refs is exactly equal
to IO_POLL_CANCEL_FLAG. Since io_poll_cancel_req() grabs ownership and
will finally make its way to io_req_complete_failed(), the req will
be reclaimed as expected.

Fixes: aa43477b04 ("io_uring: poll rework")
Signed-off-by: Lin Ma <linma@zju.edu.cn>
Reviewed-by: Pavel Begunkov <asml.silence@gmail.com>
[axboe: tweak description and code style]
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-11-25 07:17:33 -07:00
Pavel Begunkov
a26a35e901 io_uring: make poll refs more robust
poll_refs carry two functions, the first is ownership over the request.
The second is notifying the io_poll_check_events() that there was an
event but wake up couldn't grab the ownership, so io_poll_check_events()
should retry.

We want to make poll_refs more robust against overflows. Instead of
always incrementing it, which covers two purposes with one atomic, check
if poll_refs is elevated enough and if so set a retry flag without
attempts to grab ownership. The gap between the bias check and following
atomics may seem racy, but we don't need it to be strict. Moreover there
might only be maximum 4 parallel updates: by the first and the second
poll entries, __io_arm_poll_handler() and cancellation. From those four,
only poll wake ups may be executed multiple times, but they're protected
by a spin.

Cc: stable@vger.kernel.org
Reported-by: Lin Ma <linma@zju.edu.cn>
Fixes: aa43477b04 ("io_uring: poll rework")
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/c762bc31f8683b3270f3587691348a7119ef9c9d.1668963050.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-11-25 06:54:46 -07:00
Pavel Begunkov
2f3893437a io_uring: cmpxchg for poll arm refs release
Replace atomically substracting the ownership reference at the end of
arming a poll with a cmpxchg. We try to release ownership by setting 0
assuming that poll_refs didn't change while we were arming. If it did
change, we keep the ownership and use it to queue a tw, which is fully
capable to process all events and (even tolerates spurious wake ups).

It's a bit more elegant as we reduce races b/w setting the cancellation
flag and getting refs with this release, and with that we don't have to
worry about any kinds of underflows. It's not the fastest path for
polling. The performance difference b/w cmpxchg and atomic dec is
usually negligible and it's not the fastest path.

Cc: stable@vger.kernel.org
Fixes: aa43477b04 ("io_uring: poll rework")
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/0c95251624397ea6def568ff040cad2d7926fd51.1668963050.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-11-25 06:54:16 -07:00
Dylan Yudaken
9b8c54755a io_uring: add io_aux_cqe which allows deferred completion
Use the just introduced deferred post cqe completion state when possible
in io_aux_cqe. If not possible fallback to io_post_aux_cqe.

This introduces a complication because of allow_overflow. For deferred
completions we cannot know without locking the completion_lock if it will
overflow (and even if we locked it, another post could sneak in and cause
this cqe to be in overflow).
However since overflow protection is mostly a best effort defence in depth
to prevent infinite loops of CQEs for poll, just checking the overflow bit
is going to be good enough and will result in at most 16 (array size of
deferred cqes) overflows.

Suggested-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Dylan Yudaken <dylany@meta.com>
Link: https://lore.kernel.org/r/20221124093559.3780686-6-dylany@meta.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-11-25 06:10:04 -07:00
Dylan Yudaken
973fc83f3a io_uring: defer all io_req_complete_failed
All failures happen under lock now, and can be deferred. To be consistent
when the failure has happened after some multishot cqe has been
deferred (and keep ordering), always defer failures.

To make this obvious at the caller (and to help prevent a future bug)
rename io_req_complete_failed to io_req_defer_failed.

Signed-off-by: Dylan Yudaken <dylany@meta.com>
Link: https://lore.kernel.org/r/20221124093559.3780686-4-dylany@meta.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-11-25 06:10:04 -07:00
Dylan Yudaken
c06c6c5d27 io_uring: always lock in io_apoll_task_func
This is required for the failure case (io_req_complete_failed) and is
missing.

The alternative would be to only lock in the failure path, however all of
the non-error paths in io_poll_check_events that do not do not return
IOU_POLL_NO_ACTION end up locking anyway. The only extraneous lock would
be for the multishot poll overflowing the CQE ring, however multishot poll
would probably benefit from being locked as it will allow completions to
be batched.

So it seems reasonable to lock always.

Signed-off-by: Dylan Yudaken <dylany@meta.com>
Link: https://lore.kernel.org/r/20221124093559.3780686-3-dylany@meta.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-11-25 06:10:04 -07:00
Pavel Begunkov
1bec951c38 io_uring: iopoll protect complete_post
io_req_complete_post() may be used by iopoll enabled rings, grab locks
in this case. That requires to pass issue_flags to propagate the locking
state.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/cc6d854065c57c838ca8e8806f707a226b70fd2d.1669203009.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-11-23 10:45:31 -07:00
Jens Axboe
4061f0ef73 Revert "io_uring: disallow self-propelled ring polling"
This reverts commit 7fdbc5f014.

This patch dealt with a subset of the real problem, which is a potential
circular dependency on the wakup path for io_uring itself. Outside of
io_uring, eventfd can also trigger this (see details in 03e02acda8)
and so can epoll (see details in caf1aeaffc). Now that we have a
generic solution to this problem, get rid of the io_uring specific
work-around.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-11-22 06:08:31 -07:00
Jens Axboe
4464853277 io_uring: pass in EPOLL_URING_WAKE for eventfd signaling and wakeups
Pass in EPOLL_URING_WAKE when signaling eventfd or doing poll related
wakups, so that we can check for a circular event dependency between
eventfd and epoll. If this flag is set when our wakeup handlers are
called, then we know we have a dependency that needs to terminate
multishot requests.

eventfd and epoll are the only such possible dependencies.

Cc: stable@vger.kernel.org # 6.0
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-11-22 06:08:31 -07:00
Lin Ma
cd42a53d25 io_uring/poll: remove outdated comments of caching
Previous commit 13a99017ff ("io_uring: remove events caching
atavisms") entirely removes the events caching optimization introduced
by commit 81459350d5 ("io_uring: cache req->apoll->events in
req->cflags"). Hence the related comment should also be removed to avoid
misunderstanding.

Fixes: 13a99017ff ("io_uring: remove events caching atavisms")
Signed-off-by: Lin Ma <linma@zju.edu.cn>
Link: https://lore.kernel.org/r/20221110060313.16303-1-linma@zju.edu.cn
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-11-21 07:44:14 -07:00
Xinghui Li
df730ec21f io_uring: fix two assignments in if conditions
Fixes two errors:

"ERROR: do not use assignment in if condition
130: FILE: io_uring/net.c:130:
+       if (!(issue_flags & IO_URING_F_UNLOCKED) &&

ERROR: do not use assignment in if condition
599: FILE: io_uring/poll.c:599:
+       } else if (!(issue_flags & IO_URING_F_UNLOCKED) &&"
reported by checkpatch.pl in net.c and poll.c .

Signed-off-by: Xinghui Li <korantli@tencent.com>
Reported-by: kernel test robot <lkp@intel.com>
Link: https://lore.kernel.org/r/20221102082503.32236-1-korantwork@gmail.com
[axboe: style tweaks]
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-11-21 07:38:31 -07:00
Pavel Begunkov
7fdbc5f014 io_uring: disallow self-propelled ring polling
When we post a CQE we wake all ring pollers as it normally should be.
However, if a CQE was generated by a multishot poll request targeting
its own ring, it'll wake that request up, which will make it to post
a new CQE, which will wake the request and so on until it exhausts all
CQ entries.

Don't allow multishot polling io_uring files but downgrade them to
oneshots, which was always stated as a correct behaviour that the
userspace should check for.

Cc: stable@vger.kernel.org
Fixes: aa43477b04 ("io_uring: poll rework")
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/3124038c0e7474d427538c2d915335ec28c92d21.1668785722.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-11-18 09:29:31 -07:00
Pavel Begunkov
539bcb57da io_uring: fix tw losing poll events
We may never try to process a poll wake and its mask if there was
multiple wake ups racing for queueing up a tw. Force
io_poll_check_events() to update the mask by vfs_poll().

Cc: stable@vger.kernel.org
Fixes: aa43477b04 ("io_uring: poll rework")
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/00344d60f8b18907171178d7cf598de71d127b0b.1668710222.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-11-17 12:33:33 -07:00
Pavel Begunkov
b98186aee2 io_uring: update res mask in io_poll_check_events
When io_poll_check_events() collides with someone attempting to queue a
task work, it'll spin for one more time. However, it'll continue to use
the mask from the first iteration instead of updating it. For example,
if the first wake up was a EPOLLIN and the second EPOLLOUT, the
userspace will not get EPOLLOUT in time.

Clear the mask for all subsequent iterations to force vfs_poll().

Cc: stable@vger.kernel.org
Fixes: aa43477b04 ("io_uring: poll rework")
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/2dac97e8f691231049cb259c4ae57e79e40b537c.1668710222.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-11-17 12:33:33 -07:00
Pavel Begunkov
5576035f15 io_uring/poll: lockdep annote io_poll_req_insert_locked
Add a lockdep annotation in io_poll_req_insert_locked().

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/8115d8e702733754d0aea119e9b5bb63d1eb8b24.1668184658.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-11-11 09:59:27 -07:00
Pavel Begunkov
30a33669fa io_uring/poll: fix double poll req->flags races
io_poll_double_prepare()            | io_poll_wake()
                                    | poll->head = NULL
smp_load(&poll->head); /* NULL */   |
flags = req->flags;                 |
                                    | req->flags &= ~SINGLE_POLL;
req->flags = flags | DOUBLE_POLL    |

The idea behind io_poll_double_prepare() is to serialise with the
first poll entry by taking the wq lock. However, it's not safe to assume
that io_poll_wake() is not running when we can't grab the lock and so we
may race modifying req->flags.

Skip double poll setup if that happens. It's ok because the first poll
entry will only be removed when it's definitely completing, e.g.
pollfree or oneshot with a valid mask.

Fixes: 49f1c68e04 ("io_uring: optimise submission side poll_refs")
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/b7fab2d502f6121a7d7b199fe4d914a43ca9cdfd.1668184658.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2022-11-11 09:59:27 -07:00