Add a pollin/pollout field to the request table, and have commands that
we can safely poll for properly marked.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
For poll requests, it's not uncommon to link a read (or write) after
the poll to execute immediately after the file is marked as ready.
Since the poll completion is called inside the waitqueue wake up handler,
we have to punt that linked request to async context. This slows down
the processing, and actually means it's faster to not use a link for this
use case.
We also run into problems if the completion_lock is contended, as we're
doing a different lock ordering than the issue side is. Hence we have
to do trylock for completion, and if that fails, go async. Poll removal
needs to go async as well, for the same reason.
eventfd notification needs special case as well, to avoid stack blowing
recursion or deadlocks.
These are all deficiencies that were inherited from the aio poll
implementation, but I think we can do better. When a poll completes,
simply queue it up in the task poll list. When the task completes the
list, we can run dependent links inline as well. This means we never
have to go async, and we can remove a bunch of code associated with
that, and optimizations to try and make that run faster. The diffstat
speaks for itself.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Store the io_kiocb in the private field instead of the poll entry, this
is in preparation for allowing multiple waitqueues.
No functional changes in this patch.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
IO_WQ_WORK_CB is used only for linked timeouts, which will be armed
before the work setup (i.e. mm, override creds, etc). The setup
shouldn't take long, so it's ok to arm it a bit later and get rid
of IO_WQ_WORK_CB.
Make io-wq call work->func() only once, callbacks will handle the rest.
i.e. the linked timeout handler will do the actual issue. And as a
bonus, it removes an extra indirect call.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Deduplicate call to io_cqring_fill_event(), plain and easy
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Add support for splice(2).
- output file is specified as sqe->fd, so it's handled by generic code
- hash_reg_file handled by generic code as well
- len is 32bit, but should be fine
- the fd_in is registered file, when SPLICE_F_FD_IN_FIXED is set, which
is a splice flag (i.e. sqe->splice_flags).
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Preparation without functional changes. Adds io_get_file(), that allows
to grab files not only into req->file.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
req->in_async is not really needed, it only prevents propagation of
@nxt for fast not-blocked submissions. Remove it.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
io_prep_async_worker() called io_wq_assign_next() do many useless checks:
io_req_work_grab_env() was already called during prep, and @do_hashed
is not ever used. Add io_prep_next_work() -- simplified version, that
can be called io-wq.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Many operations define custom work.func before getting into an io-wq.
There are several points against:
- it calls io_wq_assign_next() from outside io-wq, that may be confusing
- sync context would go unnecessary through io_req_cancelled()
- prototypes are quite different, so work!=old_work looks strange
- makes async/sync responsibilities fuzzy
- adds extra overhead
Don't call generic path and io-wq handlers from each other, but use
helpers instead
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Don't drop an early reference, hang on to it and let the caller drop
it. This makes it behave more like "regular" requests.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
If the -EAGAIN happens because of a static condition, then a poll
or later retry won't fix it. We must call it again from blocking
condition. Play it safe and ensure that any -EAGAIN condition from read
or write must retry from async context.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
We must set MSG_CMSG_COMPAT if we're in compatability mode, otherwise
the iovec import for these commands will not do the right thing and fail
the command with -EINVAL.
Found by running the test suite compiled as 32-bit.
Cc: stable@vger.kernel.org
Fixes: aa1fa28fc7 ("io_uring: add support for recvmsg()")
Fixes: 0fa03c624d ("io_uring: add support for sendmsg()")
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Follow the pattern used with other *_show_fdinfo functions and only
define and use io_uring_show_fdinfo and its helper functions if
CONFIG_PROC_FS is set.
Fixes: 87ce955b24 ("io_uring: add ->show_fdinfo() for the io_uring file descriptor")
Signed-off-by: Tobias Klauser <tklauser@distanz.ch>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Unlike the other core import helpers, import_single_range() returns 0 on
success, not the length imported. This means that links that depend on
the result of non-vec based IORING_OP_{READ,WRITE} that were added for
5.5 get errored when they should not be.
Fixes: 3a6820f2bb ("io_uring: add non-vectored read/write commands")
Signed-off-by: Jens Axboe <axboe@kernel.dk>
If work completes inline, then we should pick up a dependent link item
in __io_queue_sqe() as well. If we don't do so, we're forced to go async
with that item, which is suboptimal.
This also fixes an issue with io_put_req_find_next(), which always looks
up the next work item. That should only be done if we're dropping the
last reference to the request, to prevent multiple lookups of the same
work item.
Outside of being a fix, this also enables a good cleanup series for 5.7,
where we never have to pass 'nxt' around or into the work handlers.
Reviewed-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
After making ext4 support iopoll method:
let ext4_file_operations's iopoll method be iomap_dio_iopoll(),
we found fio can easily hang in fio_ioring_getevents() with below fio
job:
rm -f testfile; sync;
sudo fio -name=fiotest -filename=testfile -iodepth=128 -thread
-rw=write -ioengine=io_uring -hipri=1 -sqthread_poll=1 -direct=1
-bs=4k -size=10G -numjobs=8 -runtime=2000 -group_reporting
with IORING_SETUP_SQPOLL and IORING_SETUP_IOPOLL enabled.
There are two issues that results in this hang, one reason is that
when IORING_SETUP_SQPOLL and IORING_SETUP_IOPOLL are enabled, fio
does not use io_uring_enter to get completed events, it relies on
kernel io_sq_thread to poll for completed events.
Another reason is that there is a race: when io_submit_sqes() in
io_sq_thread() submits a batch of sqes, variable 'inflight' will
record the number of submitted reqs, then io_sq_thread will poll for
reqs which have been added to poll_list. But note, if some previous
reqs have been punted to io worker, these reqs will won't be in
poll_list timely. io_sq_thread() will only poll for a part of previous
submitted reqs, and then find poll_list is empty, reset variable
'inflight' to be zero. If app just waits these deferred reqs and does
not wake up io_sq_thread again, then hang happens.
For app that entirely relies on io_sq_thread to poll completed requests,
let io_iopoll_req_issued() wake up io_sq_thread properly when adding new
element to poll_list, and when io_sq_thread prepares to sleep, check
whether poll_list is empty again, if not empty, continue to poll.
Signed-off-by: Xiaoguang Wang <xiaoguang.wang@linux.alibaba.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
We somehow never free the idr, even though we init it for every ctx.
Free it when the rest of the ring data is freed.
Fixes: 071698e13a ("io_uring: allow registering credentials")
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
If we have a chain of requests and they don't all use the same
credentials, then the head of the chain will be issued with the
credentails of the tail of the chain.
Ensure __io_queue_sqe() overrides the credentials, if they are different.
Once we do that, we can clean up the creds handling as well, by only
having io_submit_sqe() do the lookup of a personality. It doesn't need
to assign it, since __io_queue_sqe() now always does the right thing.
Fixes: 75c6a03904 ("io_uring: support using a registered personality for commands")
Reported-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Since commit a3a0e43fd7 ("io_uring: don't enter poll loop if we have
CQEs pending"), if we already events pending, we won't enter poll loop.
In case SETUP_IOPOLL and SETUP_SQPOLL are both enabled, if app has
been terminated and don't reap pending events which are already in cq
ring, and there are some reqs in poll_list, io_sq_thread will enter
__io_iopoll_check(), and find pending events, then return, this loop
will never have a chance to exit.
I have seen this issue in fio stress tests, to fix this issue, let
io_sq_thread call io_iopoll_getevents() with argument 'min' being zero,
and remove __io_iopoll_check().
Fixes: a3a0e43fd7 ("io_uring: don't enter poll loop if we have CQEs pending")
Signed-off-by: Xiaoguang Wang <xiaoguang.wang@linux.alibaba.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
This patch drops 'cur_mm' before calling cond_resched(), to prevent
the sq_thread from spinning even when the user process is finished.
Before this patch, if the user process ended without closing the
io_uring fd, the sq_thread continues to spin until the
'sq_thread_idle' timeout ends.
In the worst case where the 'sq_thread_idle' parameter is bigger than
INT_MAX, the sq_thread will spin forever.
Fixes: 6c271ce2f1 ("io_uring: add submission polling")
Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
io_cleanup_req() should be called before req->io is freed, and so
shouldn't be after __io_free_req() -> __io_req_aux_free(). Also,
it will be ignored for in io_free_req_many(), which use
__io_req_aux_free().
Place cleanup_req() into __io_req_aux_free().
Fixes: 99bc4c3853 ("io_uring: fix iovec leaks")
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The "kmsg" pointer can't be NULL and we have already dereferenced it so
a check here would be useless.
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
fallocate_finish() is missing cancellation check. Add it.
It's safe to do that, as only flags setup and sqe fields copy are done
before it gets into __io_fallocate().
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Carter reported an issue where he could produce a stall on ring exit,
when we're cleaning up requests that match the given file table. For
this particular test case, a combination of a few things caused the
issue:
- The cq ring was overflown
- The request being canceled was in the overflow list
The combination of the above means that the cq overflow list holds a
reference to the request. The request is canceled correctly, but since
the overflow list holds a reference to it, the final put won't happen.
Since the final put doesn't happen, the request remains in the inflight.
Hence we never finish the cancelation flush.
Fix this by removing requests from the overflow list if we're canceling
them.
Cc: stable@vger.kernel.org # 5.5
Reported-by: Carter Li 李通洲 <carter.li@eoitek.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jonas reports that he sometimes sees -97/-22 error returns from
sendmsg, if it gets punted async. This is due to not retaining the
sockaddr_storage between calls. Include that in the state we copy when
going async.
Cc: stable@vger.kernel.org # 5.3+
Reported-by: Jonas Bonn <jonas@norrbonn.se>
Tested-by: Jonas Bonn <jonas@norrbonn.se>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Normally we cancel all work we track, but for untracked work we could
leave the async worker behind until that work completes. This is totally
fine, but does leave resources pending after the task is gone until that
work completes.
Cancel work that this task queued up when it goes away.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
As in the previous patch, make openat*_prep() and statx_prep() handle
double preparation to avoid resource leakage.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Requests may be prepared multiple times with ->io allocated (i.e. async
prepared). Preparation functions don't handle it and forget about
previously allocated resources. This may happen in case of:
- spurious defer_check
- non-head (i.e. async prepared) request executed in sync (via nxt).
Make the handlers check, whether they already allocated resources, which
is true IFF REQ_F_NEED_CLEANUP is set.
Cc: stable@vger.kernel.org # 5.5
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
First, io_close() misses filp_close() and io_cqring_add_event(), when
f_op->flush is defined. That's because in this case it will
io_queue_async_work() itself not grabbing files, so the corresponding
chunk in io_close_finish() won't be executed.
Second, when submitted through io_wq_submit_work(), it will do
filp_close() and *_add_event() twice: first inline in io_close(),
and the second one in call to io_close_finish() from io_close().
The second one will also fire, because it was submitted async through
generic path, and so have grabbed files.
And the last nice thing is to remove this weird pilgrimage with checking
work/old_work and casting it to nxt. Just use a helper instead.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
This passes it in to io-wq, so it assumes the right fs_struct when
executing async work that may need to do lookups.
Cc: stable@vger.kernel.org # 5.3+
Signed-off-by: Jens Axboe <axboe@kernel.dk>
For non-blocking issue, we set IOCB_NOWAIT in the kiocb. However, on a
raw block device, this yields an -EOPNOTSUPP return, as non-blocking
writes aren't supported. Turn this -EOPNOTSUPP into -EAGAIN, so we retry
from blocking context with IOCB_NOWAIT cleared.
Cc: stable@vger.kernel.org # 5.5
Signed-off-by: Jens Axboe <axboe@kernel.dk>
openat() and statx() may have allocated ->open.filename, which should be
be put. Add cleanup handlers for them.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Allocated iovec is freed only in io_{read,write,send,recv)(), and just
leaves it if an error occured. There are plenty of such cases:
- cancellation of non-head requests
- fail grabbing files in __io_queue_sqe()
- set REQ_F_NOWAIT and returning in __io_queue_sqe()
Add REQ_F_NEED_CLEANUP, which will force such requests with custom
allocated resourses go through cleanup handlers on put.
Cc: stable@vger.kernel.org # 5.5
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
In io_uring_poll() we must flush overflowed CQ events before to
check if there are CQ events available, to avoid missing events.
We call the io_cqring_events() that checks and flushes any overflow
and returns the number of CQ events available.
Signed-off-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
All of these opcodes take a directory file descriptor. We can't easily
support fixed files for these operations, and the use case for that
probably isn't all that clear (or sensible) anyway.
Disable IOSQE_FIXED_FILE for these operations.
Reported-by: Stefan Metzmacher <metze@samba.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
After defer, a request will be prepared, that includes allocating iovec
if needed, and then submitted through io_wq_submit_work() but not custom
handler (e.g. io_rw_async()/io_sendrecv_async()). However, it'll leak
iovec, as it's in io-wq and the code goes as follows:
io_read() {
if (!io_wq_current_is_worker())
kfree(iovec);
}
Put all deallocation logic in io_{read,write,send,recv}(), which will
leave the memory, if going async with -EAGAIN.
It also fixes a leak after failed io_alloc_async_ctx() in
io_{recv,send}_msg().
Cc: stable@vger.kernel.org # 5.5
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Make bitfields of size 1 bit be unsigned (since there is no room
for the sign bit).
This clears up the sparse warnings:
CHECK ../fs/io_uring.c
../fs/io_uring.c:207:50: error: dubious one-bit signed bitfield
../fs/io_uring.c:208:55: error: dubious one-bit signed bitfield
../fs/io_uring.c:209:63: error: dubious one-bit signed bitfield
../fs/io_uring.c:210:54: error: dubious one-bit signed bitfield
../fs/io_uring.c:211:57: error: dubious one-bit signed bitfield
Found by sight and then verified with sparse.
Fixes: 69b3e54613 ("io_uring: change io_ring_ctx bool fields into bit fields")
Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: io-uring@vger.kernel.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Fail fast if can't grab mm, so past that requests always have an mm
when required. This allows us to remove req->user altogether.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
-----BEGIN PGP SIGNATURE-----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=RzPL
-----END PGP SIGNATURE-----
Merge tag 'io_uring-5.6-2020-02-05' of git://git.kernel.dk/linux-block
Pull io_uring updates from Jens Axboe:
"Some later fixes for io_uring:
- Small cleanup series from Pavel
- Belt and suspenders build time check of sqe size and layout
(Stefan)
- Addition of ->show_fdinfo() on request of Jann Horn, to aid in
understanding mapped personalities
- eventfd recursion/deadlock fix, for both io_uring and aio
- Fixup for send/recv handling
- Fixup for double deferral of read/write request
- Fix for potential double completion event for close request
- Adjust fadvise advice async/inline behavior
- Fix for shutdown hang with SQPOLL thread
- Fix for potential use-after-free of fixed file table"
* tag 'io_uring-5.6-2020-02-05' of git://git.kernel.dk/linux-block:
io_uring: cleanup fixed file data table references
io_uring: spin for sq thread to idle on shutdown
aio: prevent potential eventfd recursion on poll
io_uring: put the flag changing code in the same spot
io_uring: iterate req cache backwards
io_uring: punt even fadvise() WILLNEED to async context
io_uring: fix sporadic double CQE entry for close
io_uring: remove extra ->file check
io_uring: don't map read/write iovec potentially twice
io_uring: use the proper helpers for io_send/recv
io_uring: prevent potential eventfd recursion on poll
eventfd: track eventfd_signal() recursion depth
io_uring: add BUILD_BUG_ON() to assert the layout of struct io_uring_sqe
io_uring: add ->show_fdinfo() for the io_uring file descriptor
syzbot reports a use-after-free in io_ring_file_ref_switch() when it
tries to switch back to percpu mode. When we put the final reference to
the table by calling percpu_ref_kill_and_confirm(), we don't want the
zero reference to queue async work for flushing the potentially queued
up items. We currently do a few flush_work(), but they merely paper
around the issue, since the work item may not have been queued yet
depending on the when the percpu-ref callback gets run.
Coming into the file unregister, we know we have the ring quiesced.
io_ring_file_ref_switch() can check for whether or not the ref is dying
or not, and not queue anything async at that point. Once the ref has
been confirmed killed, flush any potential items manually.
Reported-by: syzbot+7caeaea49c2c8a591e3d@syzkaller.appspotmail.com
Fixes: 05f3fb3c53 ("io_uring: avoid ring quiesce for fixed file set unregister and update")
Signed-off-by: Jens Axboe <axboe@kernel.dk>
As part of io_uring shutdown, we cancel work that is pending and won't
necessarily complete on its own. That includes requests like poll
commands and timeouts.
If we're using SQPOLL for kernel side submission and we shutdown the
ring immediately after queueing such work, we can race with the sqthread
doing the submission. This means we may miss cancelling some work, which
results in the io_uring shutdown hanging forever.
Cc: stable@vger.kernel.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Both iocb_flags() and kiocb_set_rw_flags() are inline and modify
kiocb->ki_flags. Place them close, so they can be potentially better
optimised.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Grab requests from cache-array from the end, so can get by only
free_reqs.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Andres correctly points out that read-ahead can block, if it needs to
read in meta data (or even just through the page cache page allocations).
Play it safe for now and just ensure WILLNEED is also punted to async
context.
While in there, allow the file settings hints from non-blocking
context. They don't need to start/do IO, and we can safely do them
inline.
Fixes: 4840e418c2 ("io_uring: add IORING_OP_FADVISE")
Reported-by: Andres Freund <andres@anarazel.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
We punt close to async for the final fput(), but we log the completion
even before that even in that case. We rely on the request not having
a files table assigned to detect what the final async close should do.
However, if we punt the async queue to __io_queue_sqe(), we'll get
->files assigned and this makes io_close_finish() think it should both
close the filp again (which does no harm) AND log a new CQE event for
this request. This causes duplicate CQEs.
Queue the request up for async manually so we don't grab files
needlessly and trigger this condition.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe