We currently have a race where we recycle the selected buffer if poll
returns IO_APOLL_OK. But that's too late, as the poll could already be
triggering or have triggered. If that race happens, then we're putting a
buffer that's already being used.

Fix this by recycling before we arm poll. This does mean that we'll
sometimes almost instantly re-select the buffer, but it's rare enough in
testing that it should not pose a performance issue.

Fixes: b1c62645 ("io_uring: recycle provided buffers if request goes async")
Signed-off-by: Jens Axboe <axboe@kernel.dk>

The fix for not advancing the iterator if we're using fixed buffers is
broken in that it can hit a condition where we don't terminate the loop.
This results in io-wq looping forever, asking to read (or write) 0 bytes
for every subsequent loop.
Reported-by: Joel Jaeschke <joel.jaeschke@gmail.com>
Link: https://github.com/axboe/liburing/issues/549
Fixes: 16c8d2df ("io_uring: ensure symmetry in handling iter types in loop_rw_iter()")
Signed-off-by: Jens Axboe <axboe@kernel.dk>

Looks like a victim of too much copy/paste, we should not be looking
at req->open.how in accept. The point is to check CLOEXEC and error
out, which we don't invalid direct descriptors on exec. Hence any
attempt to get a direct descriptor with CLOEXEC is invalid.

No harm is done here, as req->open.how.flags overlaps with
req->accept.flags, but it's very confusing and might change if either of
those command structs are modified.

Fixes: aaa4db12 ("io_uring: accept directly into fixed file table")
Signed-off-by: Jens Axboe <axboe@kernel.dk>

Workloads using provided buffers benefit from using and returning buffers
in the right order, and so does TLBs for that matter. Manage the internal
buffer list in a straight list, rather than use the head buffer as the
insertion node. Use a hashed list for the buffer group IDs instead of
xarray, the overhead is much lower this way. xarray provides internal
locking and other trickery that is handy for some uses cases, but
io_uring already locks internally for the buffer manipulation and needs
none of that.

This is good for about a 2% reduction in overhead, combination of the
improved management and the fact that the workload has an easier time
bundling back provided buffers.
Signed-off-by: Jens Axboe <axboe@kernel.dk>

Add ->has_evfd flag, which is true IFF there is an eventfd attached, and
use it to hide io_eventfd_signal() into __io_commit_cqring_flush() and
combine fast checks in a single if. Also, gcc 11.2 wasn't inlining
io_cqring_ev_posted() without this change, so helps with that as well.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/f6168471997decded475a063f92915787975a30b.1647481208.git.asml.silence@gmail.comSigned-off-by: Jens Axboe <axboe@kernel.dk>

io_commit_cqring() is currently always under spinlock section, so it's
always better to keep it as slim as possible. Move
__io_commit_cqring_flush() out of it into ev_posted*(). If fast checks
do fail and this post-processing is required, we'll reacquire
->completion_lock, which is fine as we don't care about performance of
draining and offset timeouts.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/ec4e81fd720d3bc7bca8cb9152e080dad1a052f1.1647481208.git.asml.silence@gmail.comSigned-off-by: Jens Axboe <axboe@kernel.dk>

A preparation patch, which moves a fast ->io_ev_fd check out of
io_eventfd_signal() into ev_posted*(). Compilers are smart enough for it
to not change anything, but will need it later.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/ec4091ac76d43912b73917e8db651c2dac4b7b01.1647481208.git.asml.silence@gmail.comSigned-off-by: Jens Axboe <axboe@kernel.dk>

smp_mb() in io_cqring_ev_posted_iopoll() is only there because of
waitqueue_active(). However, non-SQPOLL IOPOLL ring doesn't wake the CQ
and so the barrier there is useless. Kill it, it's usually pretty
expensive.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/d72e8ef6f7a3f6a72e18fad8409f7d47afc8da7d.1647481208.git.asml.silence@gmail.comSigned-off-by: Jens Axboe <axboe@kernel.dk>

Use io_req_complete_failed() in kiocb_done(). This cleans up the code,
but also ensures that a provided buffers is correctly freed on failure.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/a4880106fcf199d5810707fe2d17126fcdf18bc4.1647481208.git.asml.silence@gmail.com
[axboe: split from previous patch]
Signed-off-by: Jens Axboe <axboe@kernel.dk>

It's never a good idea to put provided buffers without notifying the
userspace, it'll lead to userspace leaks, so add io_put_kbuf() in
io_req_complete_failed(). The fail helper is called by all sorts of
requests, but it's still safe to do as io_put_kbuf() will return 0 in
for all requests that don't support and so don't expect provided buffers.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/a4880106fcf199d5810707fe2d17126fcdf18bc4.1647481208.git.asml.silence@gmail.comSigned-off-by: Jens Axboe <axboe@kernel.dk>

io_fill_cqe*() is not always the best way to post CQEs just because
there is enough of infrastructure on top. Replace a raw call to a
variant of it inside of io_timeout_cancel(), which also saves us some
bloating and might help with batching later.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/46113ec4345764b4aef3b384ce38cceabaeedcbb.1647481208.git.asml.silence@gmail.comSigned-off-by: Jens Axboe <axboe@kernel.dk>

Restore consistency in __io_fill_cqe* like helpers, always honouring
"io_" prefix and adding "req" when we're passing in a request.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/bd016ff5c1a4f74687828069d2619d8a65e0c6d7.1647481208.git.asml.silence@gmail.comSigned-off-by: Jens Axboe <axboe@kernel.dk>

With commit "io_uring: cache req->apoll->events in req->cflags" applied,
we now have just io_poll_remove_entries() dipping into req->apoll when
it isn't strictly necessary.

Mark poll and double-poll with a flag, so we know if we need to look
at apoll->double_poll. This avoids pulling in those cachelines if we
don't need them. The common case is that the poll wake handler already
removed these entries while hot off the completion path.
Signed-off-by: Jens Axboe <axboe@kernel.dk>

When we arm poll on behalf of a different type of request, like a network
receive, then we allocate req->apoll as our poll entry. Running network
workloads shows io_poll_check_events() as the most expensive part of
io_uring, and it's all due to having to pull in req->apoll instead of
just the request which we have hot already.

Cache poll->events in req->cflags, which isn't used until the request
completes anyway. This isn't strictly needed for regular poll, where
req->poll.events is used and thus already hot, but for the sake of
unification we do it all around.

This saves 3-4% of overhead in certain request workloads.
Signed-off-by: Jens Axboe <axboe@kernel.dk>

This serves two purposes:

- We now have the last cacheline mostly unused for generic workloads,
  instead of having to pull in the poll refs explicitly for workloads
  that rely on poll arming.

- It shrinks the io_kiocb from 232 to 224 bytes.
Signed-off-by: Jens Axboe <axboe@kernel.dk>

Make the tracing formatting for user_data and flags consistent.

Having consistent formatting allows one for example to grep for a specific
user_data/flags and be able to trace a single sqe through easily.

Change user_data to 0x%llx and flags to 0x%x everywhere. The '0x' is
useful to disambiguate for example "user_data 100".

Additionally remove the '=' for flags in io_uring_req_failed, again for consistency.
Signed-off-by: Dylan Yudaken <dylany@fb.com>
Link: https://lore.kernel.org/r/20220316095204.2191498-1-dylany@fb.comSigned-off-by: Jens Axboe <axboe@kernel.dk>

Particularly for networked workloads, io_uring intensively uses its
poll based backend to get a notification when data/space is available.
Profiling workloads, we see 3-4% of alloc+free that is directly attributed
to just the apoll allocation and free (and the rest being skb alloc+free).

For the fast path, we have ctx->uring_lock held already for both issue
and the inline completions, and we can utilize that to avoid any extra
locking needed to have a basic recycling cache for the apoll entries on
both the alloc and free side.

Double poll still requires an allocation. But those are rare and not
a fast path item.

With the simple cache in place, we see a 3-4% reduction in overhead for
the workload.
Signed-off-by: Jens Axboe <axboe@kernel.dk>

Julia and the kernel test robot report that the prep handling for this
command inadvertently checks one field twice:

fs/io_uring.c:4338:42-56: duplicated argument to && or ||

Get rid of it.
Reported-by: kernel test robot <lkp@intel.com>
Reported-by: Julia Lawall <julia.lawall@lip6.fr>
Fixes: 4f57f06c ("io_uring: add support for IORING_OP_MSG_RING command")
Signed-off-by: Jens Axboe <axboe@kernel.dk>

By default, io_uring will stop submitting a batch of requests if we run
into an error submitting a request. This isn't strictly necessary, as
the error result is passed out-of-band via a CQE anyway. And it can be
a bit confusing for some applications.

Provide a way to setup a ring that will continue submitting on error,
when the error CQE has been posted.

There's still one case that will break out of submission. If we fail
allocating a request, then we'll still return -ENOMEM. We could in theory
post a CQE for that condition too even if we never got a request. Leave
that for a potential followup.
Reported-by: Dylan Yudaken <dylany@fb.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

If we are using provided buffers, it's less than useful to have a buffer
selected and pinned if a request needs to go async or arms poll for
notification trigger on when we can process it.

Recycle the buffer in those events, so we don't pin it for the duration
of the request.
Signed-off-by: Jens Axboe <axboe@kernel.dk>

If we drop buffers between scheduling a retry, then we need to re-import
when we start the request again.
Signed-off-by: Jens Axboe <axboe@kernel.dk>

Most of the logic in io_read() deals with regular files, and in some ways
it would make sense to split the handling into S_IFREG and others. But
at least for retry, we don't need to bother setting up a bunch of state
just to abort in the loop later. In particular, don't bother forcing
setup of async data for a normal non-vectored read when we don't need it.
Signed-off-by: Jens Axboe <axboe@kernel.dk>

The sqpoll thread can be used for performing the napi busy poll in a
similar way that it does io polling for file systems supporting direct
access bypassing the page cache.

The other way that io_uring can be used for napi busy poll is by
calling io_uring_enter() to get events.

If the user specify a timeout value, it is distributed between polling
and sleeping by using the systemwide setting
/proc/sys/net/core/busy_poll.

The changes have been tested with this program:
https://github.com/lano1106/io_uring_udp_ping

and the result is:
Without sqpoll:
NAPI busy loop disabled:
rtt min/avg/max/mdev = 40.631/42.050/58.667/1.547 us
NAPI busy loop enabled:
rtt min/avg/max/mdev = 30.619/31.753/61.433/1.456 us

With sqpoll:
NAPI busy loop disabled:
rtt min/avg/max/mdev = 42.087/44.438/59.508/1.533 us
NAPI busy loop enabled:
rtt min/avg/max/mdev = 35.779/37.347/52.201/0.924 us
Co-developed-by: Hao Xu <haoxu@linux.alibaba.com>
Signed-off-by: Hao Xu <haoxu@linux.alibaba.com>
Signed-off-by: Olivier Langlois <olivier@trillion01.com>
Link: https://lore.kernel.org/r/810bd9408ffc510ff08269e78dca9df4af0b9e4e.1646777484.git.olivier@trillion01.comSigned-off-by: Jens Axboe <axboe@kernel.dk>

Move up the block manipulating the sig variable to execute code
that may encounter an error and exit first before continuing
executing the rest of the function and avoid useless computations
Signed-off-by: Olivier Langlois <olivier@trillion01.com>
Link: https://lore.kernel.org/r/84513f7cc1b1fb31d8f4cb910aee033391d036b4.1646777484.git.olivier@trillion01.comSigned-off-by: Jens Axboe <axboe@kernel.dk>

This adds support for IORING_OP_MSG_RING, which allows an SQE to signal
another ring. That allows either waking up someone waiting on the ring,
or even passing a 64-bit value via the user_data field in the CQE.

sqe->fd must contain the fd of a ring that should receive the CQE.
sqe->off will be propagated to the cqe->user_data on the target ring,
and sqe->len will be propagated to cqe->res. The results CQE will have
IORING_CQE_F_MSG set in its flags, to indicate that this CQE was generated
from a messaging request rather than a SQE issued locally on that ring.
This effectively allows passing a 64-bit and a 32-bit quantify between
the two rings.

This request type has the following request specific error cases:

- -EBADFD. Set if the sqe->fd doesn't point to a file descriptor that is
  of the io_uring type.
- -EOVERFLOW. Set if we were not able to deliver a request to the target
  ring.
Signed-off-by: Jens Axboe <axboe@kernel.dk>

In testing high frequency workloads with provided buffers, we spend a
lot of time in allocating and freeing the buffer units themselves.
Rather than repeatedly free and alloc them, add a recycling cache
instead. There are two caches:

- ctx->io_buffers_cache. This is the one we grab from in the submission
  path, and it's protected by ctx->uring_lock. For inline completions,
  we can recycle straight back to this cache and not need any extra
  locking.

- ctx->io_buffers_comp. If we're not under uring_lock, then we use this
  list to recycle buffers. It's protected by the completion_lock.

On adding a new buffer, check io_buffers_cache. If it's empty, check if
we can splice entries from the io_buffers_comp_cache.

This reduces about 5-10% of overhead from provided buffers, bringing it
pretty close to the non-provided path.
Signed-off-by: Jens Axboe <axboe@kernel.dk>

Lots of workloads use multiple threads, in which case the file table is
shared between them. This makes getting and putting the ring file
descriptor for each io_uring_enter(2) system call more expensive, as it
involves an atomic get and put for each call.

Similarly to how we allow registering normal file descriptors to avoid
this overhead, add support for an io_uring_register(2) API that allows
to register the ring fds themselves:

1) IORING_REGISTER_RING_FDS - takes an array of io_uring_rsrc_update
   structs, and registers them with the task.
2) IORING_UNREGISTER_RING_FDS - takes an array of io_uring_src_update
   structs, and unregisters them.

When a ring fd is registered, it is internally represented by an offset.
This offset is returned to the application, and the application then
uses this offset and sets IORING_ENTER_REGISTERED_RING for the
io_uring_enter(2) system call. This works just like using a registered
file descriptor, rather than a real one, in an SQE, where
IOSQE_FIXED_FILE gets set to tell io_uring that we're using an internal
offset/descriptor rather than a real file descriptor.

In initial testing, this provides a nice bump in performance for
threaded applications in real world cases where the batch count (eg
number of requests submitted per io_uring_enter(2) invocation) is low.
In a microbenchmark, submitting NOP requests, we see the following
increases in performance:

Requests per syscall	Baseline	Registered	Increase
----------------------------------------------------------------
1			 ~7030K		 ~8080K		+15%
2			~13120K		~14800K		+13%
4			~22740K		~25300K		+11%
Co-developed-by: Xiaoguang Wang <xiaoguang.wang@linux.alibaba.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

Fix incorrect name reference in comment. ki_filp does not exist in the
struct, but file does.
Signed-off-by: Dylan Yudaken <dylany@fb.com>
Link: https://lore.kernel.org/r/20220224105157.1332353-1-dylany@fb.comSigned-off-by: Jens Axboe <axboe@kernel.dk>

There is a slight optimisation to be had by calculating the correct pos
pointer inside io_kiocb_update_pos and then using that later.

It seems code size drops by a bit:
000000000000a1b0 0000000000000400 t io_read
000000000000a5b0 0000000000000319 t io_write

vs
000000000000a1b0 00000000000003f6 t io_read
000000000000a5b0 0000000000000310 t io_write
Signed-off-by: Dylan Yudaken <dylany@fb.com>
Reviewed-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

Update kiocb->ki_pos at execution time rather than in io_prep_rw().
io_prep_rw() happens before the job is enqueued to a worker and so the
offset might be read multiple times before being executed once.

Ensures that the file position in a set of _linked_ SQEs will be only
obtained after earlier SQEs have completed, and so will include their
incremented file position.
Signed-off-by: Dylan Yudaken <dylany@fb.com>
Reviewed-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

io_kiocb_ppos is called in both branches, and it seems that the compiler
does not fuse this. Fusing removes a few bytes from loop_rw_iter.

Before:
$ nm -S fs/io_uring.o | grep loop_rw_iter
0000000000002430 0000000000000124 t loop_rw_iter

After:
$ nm -S fs/io_uring.o | grep loop_rw_iter
0000000000002430 000000000000010d t loop_rw_iter
Signed-off-by: Dylan Yudaken <dylany@fb.com>
Reviewed-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

Avoid testing TIF_NOTIFY_SIGNAL twice by calling task_sigpending()
directly from io_run_task_work_sig()
Signed-off-by: Olivier Langlois <olivier@trillion01.com>
Link: https://lore.kernel.org/r/bd7c0495f7656e803e5736708591bb665e6eaacd.1645041650.git.olivier@trillion01.comSigned-off-by: Jens Axboe <axboe@kernel.dk>

This makes the io-uring tracepoints consistent. Where it makes sense
the tracepoints start with the following four fields:
- context (ring)
- request
- user_data
- opcode.
Signed-off-by: Stefan Roesch <shr@fb.com>
Link: https://lore.kernel.org/r/20220214180430.70572-3-shr@fb.comSigned-off-by: Jens Axboe <axboe@kernel.dk>

This introduces the __fill_cqe function. This is necessary
to correctly issue the io_uring_complete tracepoint.
Signed-off-by: Stefan Roesch <shr@fb.com>
Link: https://lore.kernel.org/r/20220214180430.70572-2-shr@fb.comSigned-off-by: Jens Axboe <axboe@kernel.dk>

It's better to use the defined enum stuff not the hardcoded number to
define array.
Signed-off-by: Hao Xu <haoxu@linux.alibaba.com>
Link: https://lore.kernel.org/r/20220206095241.121485-4-haoxu@linux.alibaba.comSigned-off-by: Jens Axboe <axboe@kernel.dk>

reduce acct->lock lock and unlock in different functions to make the
code clearer.
Signed-off-by: Hao Xu <haoxu@linux.alibaba.com>
Link: https://lore.kernel.org/r/20220206095241.121485-3-haoxu@linux.alibaba.comSigned-off-by: Jens Axboe <axboe@kernel.dk>

wqe->lock is abused, it now protects acct->work_list, hash stuff,
nr_workers, wqe->free_list and so on. Lets first get the work_list out
of the wqe-lock mess by introduce a specific lock for work list. This
is the first step to solve the huge contension between work insertion
and work consumption.
good thing:
  - split locking for bound and unbound work list
  - reduce contension between work_list visit and (worker's)free_list.

For the hash stuff, since there won't be a work with same file in both
bound and unbound work list, thus they won't visit same hash entry. it
works well to use the new lock to protect hash stuff.

Results:
set max_unbound_worker = 4, test with echo-server:
nice -n -15 ./io_uring_echo_server -p 8081 -f -n 1000 -l 16
(-n connection, -l workload)
before this patch:
Samples: 2M of event 'cycles:ppp', Event count (approx.): 1239982111074
Overhead  Command          Shared Object         Symbol
  28.59%  iou-wrk-10021    [kernel.vmlinux]      [k] native_queued_spin_lock_slowpath
   8.89%  io_uring_echo_s  [kernel.vmlinux]      [k] native_queued_spin_lock_slowpath
   6.20%  iou-wrk-10021    [kernel.vmlinux]      [k] _raw_spin_lock
   2.45%  io_uring_echo_s  [kernel.vmlinux]      [k] io_prep_async_work
   2.36%  iou-wrk-10021    [kernel.vmlinux]      [k] _raw_spin_lock_irqsave
   2.29%  iou-wrk-10021    [kernel.vmlinux]      [k] io_worker_handle_work
   1.29%  io_uring_echo_s  [kernel.vmlinux]      [k] io_wqe_enqueue
   1.06%  iou-wrk-10021    [kernel.vmlinux]      [k] io_wqe_worker
   1.06%  io_uring_echo_s  [kernel.vmlinux]      [k] _raw_spin_lock
   1.03%  iou-wrk-10021    [kernel.vmlinux]      [k] __schedule
   0.99%  iou-wrk-10021    [kernel.vmlinux]      [k] tcp_sendmsg_locked

with this patch:
Samples: 1M of event 'cycles:ppp', Event count (approx.): 708446691943
Overhead  Command          Shared Object         Symbol
  16.86%  iou-wrk-10893    [kernel.vmlinux]      [k] native_queued_spin_lock_slowpat
   9.10%  iou-wrk-10893    [kernel.vmlinux]      [k] _raw_spin_lock
   4.53%  io_uring_echo_s  [kernel.vmlinux]      [k] native_queued_spin_lock_slowpat
   2.87%  iou-wrk-10893    [kernel.vmlinux]      [k] io_worker_handle_work
   2.57%  iou-wrk-10893    [kernel.vmlinux]      [k] _raw_spin_lock_irqsave
   2.56%  io_uring_echo_s  [kernel.vmlinux]      [k] io_prep_async_work
   1.82%  io_uring_echo_s  [kernel.vmlinux]      [k] _raw_spin_lock
   1.33%  iou-wrk-10893    [kernel.vmlinux]      [k] io_wqe_worker
   1.26%  io_uring_echo_s  [kernel.vmlinux]      [k] try_to_wake_up

spin_lock failure from 25.59% + 8.89% =  34.48% to 16.86% + 4.53% = 21.39%
TPS is similar, while cpu usage is from almost 400% to 350%
Signed-off-by: Hao Xu <haoxu@linux.alibaba.com>
Link: https://lore.kernel.org/r/20220206095241.121485-2-haoxu@linux.alibaba.comSigned-off-by: Jens Axboe <axboe@kernel.dk>

Clang warns:

  fs/io_uring.c:9396:9: warning: variable 'ret' is uninitialized when used here [-Wuninitialized]
          return ret;
                 ^~~
  fs/io_uring.c:9373:13: note: initialize the variable 'ret' to silence this warning
          int fd, ret;
                     ^
                      = 0
  1 warning generated.

Just return 0 directly and reduce the scope of ret to the if statement,
as that is the only place that it is used, which is how the function was
before the fixes commit.

Fixes: 1a75fac9a0f9 ("io_uring: avoid ring quiesce while registering/unregistering eventfd")
Link: https://github.com/ClangBuiltLinux/linux/issues/1579Signed-off-by: Nathan Chancellor <nathan@kernel.org>
Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>
Link: https://lore.kernel.org/r/20220207162410.1013466-1-nathan@kernel.orgSigned-off-by: Jens Axboe <axboe@kernel.dk>

None of the opcodes in io_uring_register use ring quiesce anymore. Hence
io_register_op_must_quiesce always returns false and io_ctx_quiesce is
never called.
Signed-off-by: Usama Arif <usama.arif@bytedance.com>
Link: https://lore.kernel.org/r/20220204145117.1186568-6-usama.arif@bytedance.comSigned-off-by: Jens Axboe <axboe@kernel.dk>

IORING_SETUP_R_DISABLED prevents submitting requests and so there will be
no requests until IORING_REGISTER_ENABLE_RINGS is called. And
IORING_REGISTER_RESTRICTIONS works only before
IORING_REGISTER_ENABLE_RINGS is called. Hence ring quiesce is not needed
for these opcodes.
Signed-off-by: Usama Arif <usama.arif@bytedance.com>
Link: https://lore.kernel.org/r/20220204145117.1186568-5-usama.arif@bytedance.comSigned-off-by: Jens Axboe <axboe@kernel.dk>