linux.git/io_uring, branch v6.3-rc2

io_uring: silence variable ‘prev’ set but not used warning

2023-03-09T17:10:58+00:00

If io_uring.o is built with W=1, it triggers a warning:

io_uring/io_uring.c: In function ‘__io_submit_flush_completions’:
io_uring/io_uring.c:1502:40: warning: variable ‘prev’ set but not used [-Wunused-but-set-variable]
 1502 |         struct io_wq_work_node *node, *prev;
      |                                        ^~~~

which is due to the wq_list_for_each() iterator always keeping a 'prev'
variable. Most users need this to remove an entry from a list, for
example, but __io_submit_flush_completions() never does that.

Add a basic helper that doesn't track prev instead, and use that in
that function.

Reported-by: Vincenzo Palazzo 
Reviewed-by: Vincenzo Palazzo 
Signed-off-by: Jens Axboe

io_uring/uring_cmd: ensure that device supports IOPOLL

2023-03-09T16:23:59+00:00

It's possible for a file type to support uring commands, but not
pollable ones. Hence before issuing one of those, we should check
that it is supported and error out upfront if it isn't.

Cc: stable@vger.kernel.org
Fixes: 5756a3a7e713 ("io_uring: add iopoll infrastructure for io_uring_cmd")
Link: https://github.com/axboe/liburing/issues/816
Reviewed-by: Kanchan Joshi 
Signed-off-by: Jens Axboe

io_uring/io-wq: stop setting PF_NO_SETAFFINITY on io-wq workers

2023-03-08T15:48:13+00:00

Every now and then reports come in that are puzzled on why changing
affinity on the io-wq workers fails with EINVAL. This happens because they
set PF_NO_SETAFFINITY as part of their creation, as io-wq organizes
workers into groups based on what CPU they are running on.

However, this is purely an optimization and not a functional requirement.
We can allow setting affinity, and just lazily update our worker to wqe
mappings. If a given io-wq thread times out, it normally exits if there's
no more work to do. The exception is if it's the last worker available.
For the timeout case, check the affinity of the worker against group mask
and exit even if it's the last worker. New workers should be created with
the right mask and in the right location.

Reported-by:Daniel Dao 
Link: https://lore.kernel.org/io-uring/CA+wXwBQwgxB3_UphSny-yAP5b26meeOu1W4TwYVcD_+5gOhvPw@mail.gmail.com/
Signed-off-by: Jens Axboe

Merge tag 'io_uring-6.3-2023-03-03' of git://git.kernel.dk/linux

2023-03-03T18:25:29+00:00

Pull more io_uring updates from Jens Axboe:
 "Here's a set of fixes/changes that didn't make the first cut, either
  because they got queued before I sent the early merge request, or
  fixes that came in afterwards. In detail:

   - Don't set MSG_NOSIGNAL on recv/recvmsg opcodes, as AF_PACKET will
     error out (David)

   - Fix for spurious poll wakeups (me)

   - Fix for a file leak for buffered reads in certain conditions
     (Joseph)

   - Don't allow registered buffers of mixed types (Pavel)

   - Improve handling of huge pages for registered buffers (Pavel)

   - Provided buffer ring size calculation fix (Wojciech)

   - Minor cleanups (me)"

* tag 'io_uring-6.3-2023-03-03' of git://git.kernel.dk/linux:
  io_uring/poll: don't pass in wake func to io_init_poll_iocb()
  io_uring: fix fget leak when fs don't support nowait buffered read
  io_uring/poll: allow some retries for poll triggering spuriously
  io_uring: remove MSG_NOSIGNAL from recvmsg
  io_uring/rsrc: always initialize 'folio' to NULL
  io_uring/rsrc: optimise registered huge pages
  io_uring/rsrc: optimise single entry advance
  io_uring/rsrc: disallow multi-source reg buffers
  io_uring: remove unused wq_list_merge
  io_uring: fix size calculation when registering buf ring
  io_uring/rsrc: fix a comment in io_import_fixed()
  io_uring: rename 'in_idle' to 'in_cancel'
  io_uring: consolidate the put_ref-and-return section of adding work

capability: just use a 'u64' instead of a 'u32[2]' array

2023-03-01T18:01:22+00:00

Back in 2008 we extended the capability bits from 32 to 64, and we did
it by extending the single 32-bit capability word from one word to an
array of two words.  It was then obfuscated by hiding the "2" behind two
macro expansions, with the reasoning being that maybe it gets extended
further some day.

That reasoning may have been valid at the time, but the last thing we
want to do is to extend the capability set any more.  And the array of
values not only causes source code oddities (with loops to deal with
it), but also results in worse code generation.  It's a lose-lose
situation.

So just change the 'u32[2]' into a 'u64' and be done with it.

We still have to deal with the fact that the user space interface is
designed around an array of these 32-bit values, but that was the case
before too, since the array layouts were different (ie user space
doesn't use an array of 32-bit values for individual capability masks,
but an array of 32-bit slices of multiple masks).

So that marshalling of data is actually simplified too, even if it does
remain somewhat obscure and odd.

This was all triggered by my reaction to the new "cap_isidentical()"
introduced recently.  By just using a saner data structure, it went from

	unsigned __capi;
	CAP_FOR_EACH_U32(__capi) {
		if (a.cap[__capi] != b.cap[__capi])
			return false;
	}
	return true;

to just being

	return a.val == b.val;

instead.  Which is rather more obvious both to humans and to compilers.

Cc: Mateusz Guzik 
Cc: Casey Schaufler 
Cc: Serge Hallyn 
Cc: Al Viro 
Cc: Paul Moore 
Signed-off-by: Linus Torvalds

io_uring/poll: don't pass in wake func to io_init_poll_iocb()

2023-03-01T17:06:53+00:00

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

io_uring: fix fget leak when fs don't support nowait buffered read

2023-02-28T12:58:42+00:00

Heming reported a BUG when using io_uring doing link-cp on ocfs2. [1]

Do the following steps can reproduce this BUG:
mount -t ocfs2 /dev/vdc /mnt/ocfs2
cp testfile /mnt/ocfs2/
./link-cp /mnt/ocfs2/testfile /mnt/ocfs2/testfile.1
umount /mnt/ocfs2

Then umount will fail, and it outputs:
umount: /mnt/ocfs2: target is busy.

While tracing umount, it blames mnt_get_count() not return as expected.
Do a deep investigation for fget()/fput() on related code flow, I've
finally found that fget() leaks since ocfs2 doesn't support nowait
buffered read.

io_issue_sqe
|-io_assign_file  // do fget() first
  |-io_read
  |-io_iter_do_read
    |-ocfs2_file_read_iter  // return -EOPNOTSUPP
  |-kiocb_done
    |-io_rw_done
      |-__io_complete_rw_common  // set REQ_F_REISSUE
    |-io_resubmit_prep
      |-io_req_prep_async  // override req->file, leak happens

This was introduced by commit a196c78b5443 in v5.18. Fix it by don't
re-assign req->file if it has already been assigned.

[1] https://lore.kernel.org/ocfs2-devel/ab580a75-91c8-d68a-3455-40361be1bfa8@linux.alibaba.com/T/#t

Fixes: a196c78b5443 ("io_uring: assign non-fixed early for async work")
Cc: 
Reported-by: Heming Zhao 
Signed-off-by: Joseph Qi 
Cc: Xiaoguang Wang 
Link: https://lore.kernel.org/r/20230228045459.13524-1-joseph.qi@linux.alibaba.com
Signed-off-by: Jens Axboe

io_uring/poll: allow some retries for poll triggering spuriously

2023-02-26T03:10:13+00:00

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

io_uring: remove MSG_NOSIGNAL from recvmsg

2023-02-24T19:59:02+00:00

MSG_NOSIGNAL is not applicable for the receiving side, SIGPIPE is
generated when trying to write to a "broken pipe".  AF_PACKET's
packet_recvmsg() does enforce this, giving back EINVAL when MSG_NOSIGNAL
is set - making it unuseable in io_uring's recvmsg.

Remove MSG_NOSIGNAL from io_recvmsg_prep().

Cc: stable@vger.kernel.org # v5.10+
Signed-off-by: David Lamparter 
Cc: Eric Dumazet 
Cc: Jens Axboe 
Reviewed-by: Eric Dumazet 
Link: https://lore.kernel.org/r/20230224150123.128346-1-equinox@diac24.net
Signed-off-by: Jens Axboe

io_uring/rsrc: always initialize 'folio' to NULL

2023-02-24T19:58:31+00:00

Smatch complains that:

smatch warnings:
io_uring/rsrc.c:1262 io_sqe_buffer_register() error: uninitialized symbol 'folio'.

'folio' may be used uninitialized, which can happen if we end up with a
single page mapped. Ensure that we clear folio to NULL at the top so
it's always set.

Reported-by: kernel test robot 
Reported-by: Dan Carpenter 
Link: https://lore.kernel.org/r/202302241432.YML1CD5C-lkp@intel.com/
Signed-off-by: Jens Axboe