Currently we only post a cqe if we get an error OUTSIDE of submission.
For submission, we return the error directly through io_uring_enter().
This is a bit awkward for applications, and it makes more sense to
always post a cqe with an error, if the error happens on behalf of an
sqe.
This changes submission behavior a bit. io_uring_enter() returns -ERROR
for an error, and > 0 for number of sqes submitted. Before this change,
if you wanted to submit 8 entries and had an error on the 5th entry,
io_uring_enter() would return 4 (for number of entries successfully
submitted) and rewind the sqring. The application would then have to
peek at the sqring and figure out what was wrong with the head sqe, and
then skip it itself. With this change, we'll return 5 since we did
consume 5 sqes, and the last sqe (with the error) will result in a cqe
being posted with the error.
This makes the logic easier to handle in the application, and it cleans
up the submission part.
Suggested-by: Stefan Bühler <source@stbuehler.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
There is no operation to order with afterwards, and removing the flag is
not critical in any way.
There will always be a "race condition" where the application will
trigger IORING_ENTER_SQ_WAKEUP when it isn't actually needed.
Signed-off-by: Stefan Bühler <source@stbuehler.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
smp_store_release in io_commit_sqring already orders the store to
dropped before the update to SQ head.
Signed-off-by: Stefan Bühler <source@stbuehler.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The memory operations before reading cq head are unrelated and we
don't care about their order.
Document that the control dependency in combination with READ_ONCE and
WRITE_ONCE forms a barrier we need.
Signed-off-by: Stefan Bühler <source@stbuehler.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
wq_has_sleeper has a full barrier internally. The smp_rmb barrier in
io_uring_poll synchronizes with it.
Signed-off-by: Stefan Bühler <source@stbuehler.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The application reading the CQ ring needs a barrier to pair with the
smp_store_release in io_commit_cqring, not the barrier after it.
Also a write barrier *after* writing something (but not *before*
writing anything interesting) doesn't order anything, so an smp_wmb()
after writing SQ tail is not needed.
Additionally consider reading SQ head and writing CQ tail in the notes.
Also add some clarifications how the various other fields in the ring
buffers are used.
Signed-off-by: Stefan Bühler <source@stbuehler.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Not all request types set REQ_F_FORCE_NONBLOCK when they needed async
punting; reverse logic instead and set REQ_F_NOWAIT if request mustn't
be punted.
Signed-off-by: Stefan Bühler <source@stbuehler.de>
Merged with my previous patch for this.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Since commit 09bb839434 we don't use the state argument for any sort
of on-stack caching in the io read and write path. Remove the stale
and unused argument from them, and bubble it up to __io_submit_sqe()
and down to io_prep_rw().
Signed-off-by: Jens Axboe <axboe@kernel.dk>
io_uring_poll shouldn't signal EPOLLOUT | EPOLLWRNORM if the queue is
full; the old check would always signal EPOLLOUT | EPOLLWRNORM (unless
there were U32_MAX - 1 entries in the SQ queue).
Signed-off-by: Stefan Bühler <source@stbuehler.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Reading the SQ tail needs to come after setting IORING_SQ_NEED_WAKEUP in
flags; there is no cheap barrier for ordering a store before a load, a
full memory barrier is required.
Userspace needs a full memory barrier between updating SQ tail and
checking for the IORING_SQ_NEED_WAKEUP too.
Signed-off-by: Stefan Bühler <source@stbuehler.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
A read memory barrier is required between reading SQ tail and reading
the actual data belonging to the SQ entry.
Userspace needs a matching write barrier between writing SQ entries and
updating SQ tail (using smp_store_release to update tail will do).
Signed-off-by: Stefan Bühler <source@stbuehler.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
If we have multiple threads doing io_uring_register(2) on an io_uring
fd, then we can potentially try and kill the percpu reference while
someone else has already killed it.
Prevent this race by failing io_uring_register(2) if the ref is marked
dying. This is safe since we're inside the io_uring mutex.
Fixes: b19062a567 ("io_uring: fix possible deadlock between io_uring_{enter,register}")
Reported-by: syzbot <syzbot+10d25e23199614b7721f@syzkaller.appspotmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
This is a leftover from when the rings initially were not free flowing,
and hence a test for tail + 1 == head would indicate full. Since we now
let them wrap instead of mask them with the size, we need to check if
they drift more than the ring size from each other.
This fixes a case where we'd overwrite CQ ring entries, if the user
failed to reap completions. Both cases would ultimately result in lost
completions as the application violated the depth it asked for. The only
difference is that before this fix we'd return invalid entries for the
overflowed completions, instead of properly flagging it in the
cq_ring->overflow variable.
Reported-by: Stefan Bühler <source@stbuehler.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
If we have multiple threads, one doing io_uring_enter() while the other
is doing io_uring_register(), we can run into a deadlock between the
two. io_uring_register() must wait for existing users of the io_uring
instance to exit. But it does so while holding the io_uring mutex.
Callers of io_uring_enter() may need this mutex to make progress (and
eventually exit). If we wait for users to exit in io_uring_register(),
we can't do so with the io_uring mutex held without potentially risking
a deadlock.
Drop the io_uring mutex while waiting for existing callers to exit. This
is safe and guaranteed to make forward progress, since we already killed
the percpu ref before doing so. Hence later callers of io_uring_enter()
will be rejected.
Reported-by: syzbot+16dc03452dee970a0c3e@syzkaller.appspotmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Since the fget/fput handling was reworked in commit 09bb839434, we
never call io_file_put() with state == NULL (and hence file != NULL)
anymore. Remove that case.
Reported-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
We currently call cpu_possible() even if we don't use the CPU. Move the
test under the SQ_AFF branch, which is the only place where we'll use
the value. Do the cpu_possible() test AFTER we've limited it to a max
of NR_CPUS. This avoids triggering the following warning:
WARNING: CPU: 1 PID: 7600 at include/linux/cpumask.h:121 cpu_max_bits_warn
if CONFIG_DEBUG_PER_CPU_MAPS is enabled.
While in there, also move the SQ thread idle period assignment inside
SETUP_SQPOLL, as we don't use it otherwise either.
Reported-by: syzbot+cd714a07c6de2bc34293@syzkaller.appspotmail.com
Fixes: 6c271ce2f1 ("io_uring: add submission polling")
Signed-off-by: Jens Axboe <axboe@kernel.dk>
This options spawns a kernel side thread that will poll for submissions
(and completions, if IORING_SETUP_IOPOLL is set). As this allows a user
to potentially use more cycles outside of the normal hierarchy,
restrict the use of this feature to root.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Will Deacon reported the following KASAN complaint:
[ 149.890370] ==================================================================
[ 149.891266] BUG: KASAN: double-free or invalid-free in io_sqe_files_unregister+0xa8/0x140
[ 149.892218]
[ 149.892411] CPU: 113 PID: 3974 Comm: io_uring_regist Tainted: G B 5.1.0-rc3-00012-g40b114779944 #3
[ 149.893623] Hardware name: linux,dummy-virt (DT)
[ 149.894169] Call trace:
[ 149.894539] dump_backtrace+0x0/0x228
[ 149.895172] show_stack+0x14/0x20
[ 149.895747] dump_stack+0xe8/0x124
[ 149.896335] print_address_description+0x60/0x258
[ 149.897148] kasan_report_invalid_free+0x78/0xb8
[ 149.897936] __kasan_slab_free+0x1fc/0x228
[ 149.898641] kasan_slab_free+0x10/0x18
[ 149.899283] kfree+0x70/0x1f8
[ 149.899798] io_sqe_files_unregister+0xa8/0x140
[ 149.900574] io_ring_ctx_wait_and_kill+0x190/0x3c0
[ 149.901402] io_uring_release+0x2c/0x48
[ 149.902068] __fput+0x18c/0x510
[ 149.902612] ____fput+0xc/0x18
[ 149.903146] task_work_run+0xf0/0x148
[ 149.903778] do_notify_resume+0x554/0x748
[ 149.904467] work_pending+0x8/0x10
[ 149.905060]
[ 149.905331] Allocated by task 3974:
[ 149.905934] __kasan_kmalloc.isra.0.part.1+0x48/0xf8
[ 149.906786] __kasan_kmalloc.isra.0+0xb8/0xd8
[ 149.907531] kasan_kmalloc+0xc/0x18
[ 149.908134] __kmalloc+0x168/0x248
[ 149.908724] __arm64_sys_io_uring_register+0x2b8/0x15a8
[ 149.909622] el0_svc_common+0x100/0x258
[ 149.910281] el0_svc_handler+0x48/0xc0
[ 149.910928] el0_svc+0x8/0xc
[ 149.911425]
[ 149.911696] Freed by task 3974:
[ 149.912242] __kasan_slab_free+0x114/0x228
[ 149.912955] kasan_slab_free+0x10/0x18
[ 149.913602] kfree+0x70/0x1f8
[ 149.914118] __arm64_sys_io_uring_register+0xc2c/0x15a8
[ 149.915009] el0_svc_common+0x100/0x258
[ 149.915670] el0_svc_handler+0x48/0xc0
[ 149.916317] el0_svc+0x8/0xc
[ 149.916817]
[ 149.917101] The buggy address belongs to the object at ffff8004ce07ed00
[ 149.917101] which belongs to the cache kmalloc-128 of size 128
[ 149.919197] The buggy address is located 0 bytes inside of
[ 149.919197] 128-byte region [ffff8004ce07ed00, ffff8004ce07ed80)
[ 149.921142] The buggy address belongs to the page:
[ 149.921953] page:ffff7e0013381f00 count:1 mapcount:0 mapping:ffff800503417c00 index:0x0 compound_mapcount: 0
[ 149.923595] flags: 0x1ffff00000010200(slab|head)
[ 149.924388] raw: 1ffff00000010200 dead000000000100 dead000000000200 ffff800503417c00
[ 149.925706] raw: 0000000000000000 0000000080400040 00000001ffffffff 0000000000000000
[ 149.927011] page dumped because: kasan: bad access detected
[ 149.927956]
[ 149.928224] Memory state around the buggy address:
[ 149.929054] ffff8004ce07ec00: 00 00 00 00 00 00 00 00 fc fc fc fc fc fc fc fc
[ 149.930274] ffff8004ce07ec80: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc
[ 149.931494] >ffff8004ce07ed00: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
[ 149.932712] ^
[ 149.933281] ffff8004ce07ed80: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc
[ 149.934508] ffff8004ce07ee00: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc
[ 149.935725] ==================================================================
which is due to a failure in registrering a fileset. This frees the
ctx->user_files pointer, but doesn't clear it. When the io_uring
instance is later freed through the normal channels, we free this
pointer again. At this point it's invalid.
Ensure we clear the pointer when we free it for the error case.
Reported-by: Will Deacon <will.deacon@arm.com>
Tested-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
In case of direct write -EAGAIN will be returned if page cache was
previously populated. To avoid immediate completion of a request
with -EAGAIN error write has to be offloaded to the async worker,
like io_read() does.
Signed-off-by: Roman Penyaev <rpenyaev@suse.de>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: linux-block@vger.kernel.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
On big-endian architectures, the signal masks are differnet
between 32-bit and 64-bit tasks, so we have to use a different
function for reading them from user space.
io_cqring_wait() initially got this wrong, and always interprets
this as a native structure. This is ok on x86 and most arm64,
but not on s390, ppc64be, mips64be, sparc64 and parisc.
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
-----BEGIN PGP SIGNATURE-----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=l4lx
-----END PGP SIGNATURE-----
Merge tag 'io_uring-20190323' of git://git.kernel.dk/linux-block
Pull io_uring fixes and improvements from Jens Axboe:
"The first five in this series are heavily inspired by the work Al did
on the aio side to fix the races there.
The last two re-introduce a feature that was in io_uring before it got
merged, but which I pulled since we didn't have a good way to have
BVEC iters that already have a stable reference. These aren't
necessarily related to block, it's just how io_uring pins fixed
buffers"
* tag 'io_uring-20190323' of git://git.kernel.dk/linux-block:
block: add BIO_NO_PAGE_REF flag
iov_iter: add ITER_BVEC_FLAG_NO_REF flag
io_uring: mark me as the maintainer
io_uring: retry bulk slab allocs as single allocs
io_uring: fix poll races
io_uring: fix fget/fput handling
io_uring: add prepped flag
io_uring: make io_read/write return an integer
io_uring: use regular request ref counts
For ITER_BVEC, if we're holding on to kernel pages, the caller
doesn't need to grab a reference to the bvec pages, and drop that
same reference on IO completion. This is essentially safe for any
ITER_BVEC, but some use cases end up reusing pages and uncondtionally
dropping a page reference on completion. And example of that is
sendfile(2), that ends up being a splice_in + splice_out on the
pipe pages.
Add a flag that tells us it's fine to not grab a page reference
to the bvec pages, since that caller knows not to drop a reference
when it's done with the pages.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
I've seen cases where bulk alloc fails, since the bulk alloc API
is all-or-nothing - either we get the number we ask for, or it
returns 0 as number of entries.
If we fail a batch bulk alloc, retry a "normal" kmem_cache_alloc()
and just use that instead of failing with -EAGAIN.
While in there, ensure we use GFP_KERNEL. That was an oversight in
the original code, when we switched away from GFP_ATOMIC.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
This is a straight port of Al's fix for the aio poll implementation,
since the io_uring version is heavily based on that. The below
description is almost straight from that patch, just modified to
fit the io_uring situation.
io_poll() has to cope with several unpleasant problems:
* requests that might stay around indefinitely need to
be made visible for io_cancel(2); that must not be done to
a request already completed, though.
* in cases when ->poll() has placed us on a waitqueue,
wakeup might have happened (and request completed) before ->poll()
returns.
* worse, in some early wakeup cases request might end
up re-added into the queue later - we can't treat "woken up and
currently not in the queue" as "it's not going to stick around
indefinitely"
* ... moreover, ->poll() might have decided not to
put it on any queues to start with, and that needs to be distinguished
from the previous case
* ->poll() might have tried to put us on more than one queue.
Only the first will succeed for io poll, so we might end up missing
wakeups. OTOH, we might very well notice that only after the
wakeup hits and request gets completed (all before ->poll() gets
around to the second poll_wait()). In that case it's too late to
decide that we have an error.
req->woken was an attempt to deal with that. Unfortunately, it was
broken. What we need to keep track of is not that wakeup has happened -
the thing might come back after that. It's that async reference is
already gone and won't come back, so we can't (and needn't) put the
request on the list of cancellables.
The easiest case is "request hadn't been put on any waitqueues"; we
can tell by seeing NULL apt.head, and in that case there won't be
anything async. We should either complete the request ourselves
(if vfs_poll() reports anything of interest) or return an error.
In all other cases we get exclusion with wakeups by grabbing the
queue lock.
If request is currently on queue and we have something interesting
from vfs_poll(), we can steal it and complete the request ourselves.
If it's on queue and vfs_poll() has not reported anything interesting,
we either put it on the cancellable list, or, if we know that it
hadn't been put on all queues ->poll() wanted it on, we steal it and
return an error.
If it's _not_ on queue, it's either been already dealt with (in which
case we do nothing), or there's io_poll_complete_work() about to be
executed. In that case we either put it on the cancellable list,
or, if we know it hadn't been put on all queues ->poll() wanted it on,
simulate what cancel would've done.
Fixes: 221c5eb233 ("io_uring: add support for IORING_OP_POLL")
Signed-off-by: Jens Axboe <axboe@kernel.dk>
This isn't a straight port of commit 84c4e1f89f for aio.c, since
io_uring doesn't use files in exactly the same way. But it's pretty
close. See the commit message for that commit.
This essentially fixes a use-after-free with the poll command
handling, but it takes cue from Linus's approach to just simplifying
the file handling. We move the setup of the file into a higher level
location, so the individual commands don't have to deal with it. And
then we release the reference when we free the associated io_kiocb.
Fixes: 221c5eb233 ("io_uring: add support for IORING_OP_POLL")
Signed-off-by: Jens Axboe <axboe@kernel.dk>
We currently use the fact that if ->ki_filp is already set, then we've
done the prep. In preparation for moving the file assignment earlier,
use a separate flag to tell whether the request has been prepped for
IO or not.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Get rid of the special casing of "normal" requests not having
any references to the io_kiocb. We initialize the ref count to 2,
one for the submission side, and one or the completion side.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
All users of VM_MAX_READAHEAD actually convert it to kbytes and then to
pages. Define the macro explicitly as (SZ_128K / PAGE_SIZE). This
simplifies the expression in every filesystem. Also rename the macro to
VM_READAHEAD_PAGES to properly convey its meaning. Finally remove unused
VM_MIN_READAHEAD
[akpm@linux-foundation.org: fix fs/io_uring.c, per Stephen]
Link: http://lkml.kernel.org/r/20181221144053.24318-1-nborisov@suse.com
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: Matthew Wilcox <willy@infradead.org>
Reviewed-by: David Hildenbrand <david@redhat.com>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Eric Van Hensbergen <ericvh@gmail.com>
Cc: Latchesar Ionkov <lucho@ionkov.net>
Cc: Dominique Martinet <asmadeus@codewreck.org>
Cc: David Howells <dhowells@redhat.com>
Cc: Chris Mason <clm@fb.com>
Cc: Josef Bacik <josef@toxicpanda.com>
Cc: David Sterba <dsterba@suse.com>
Cc: Miklos Szeredi <miklos@szeredi.hu>
Cc: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Right now we punt any buffered request that ends up triggering an
-EAGAIN to an async workqueue. This works fine in terms of providing
async execution of them, but it also can create quite a lot of work
queue items. For sequentially buffered IO, it's advantageous to
serialize the issue of them. For reads, the first one will trigger a
read-ahead, and subsequent request merely end up waiting on later pages
to complete. For writes, devices usually respond better to streamed
sequential writes.
Add state to track the last buffered request we punted to a work queue,
and if the next one is sequential to the previous, attempt to get the
previous work item to handle it. We limit the number of sequential
add-ons to the a multiple (8) of the max read-ahead size of the file.
This should be a good number for both reads and wries, as it defines the
max IO size the device can do directly.
This drastically cuts down on the number of context switches we need to
handle buffered sequential IO, and a basic test case of copying a big
file with io_uring sees a 5x speedup.
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
This is basically a direct port of bfe4037e72, which implements a
one-shot poll command through aio. Description below is based on that
commit as well. However, instead of adding a POLL command and relying
on io_cancel(2) to remove it, we mimic the epoll(2) interface of
having a command to add a poll notification, IORING_OP_POLL_ADD,
and one to remove it again, IORING_OP_POLL_REMOVE.
To poll for a file descriptor the application should submit an sqe of
type IORING_OP_POLL. It will poll the fd for the events specified in the
poll_events field.
Unlike poll or epoll without EPOLLONESHOT this interface always works in
one shot mode, that is once the sqe is completed, it will have to be
resubmitted.
Reviewed-by: Hannes Reinecke <hare@suse.com>
Based-on-code-from: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
We'll use this for the POLL implementation. Regular requests will
NOT be using references, so initialize it to 0. Any real use of
the io_kiocb ref will initialize it to at least 2.
Reviewed-by: Hannes Reinecke <hare@suse.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
This enables an application to do IO, without ever entering the kernel.
By using the SQ ring to fill in new sqes and watching for completions
on the CQ ring, we can submit and reap IOs without doing a single system
call. The kernel side thread will poll for new submissions, and in case
of HIPRI/polled IO, it'll also poll for completions.
By default, we allow 1 second of active spinning. This can by changed
by passing in a different grace period at io_uring_register(2) time.
If the thread exceeds this idle time without having any work to do, it
will set:
sq_ring->flags |= IORING_SQ_NEED_WAKEUP.
The application will have to call io_uring_enter() to start things back
up again. If IO is kept busy, that will never be needed. Basically an
application that has this feature enabled will guard it's
io_uring_enter(2) call with:
read_barrier();
if (*sq_ring->flags & IORING_SQ_NEED_WAKEUP)
io_uring_enter(fd, 0, 0, IORING_ENTER_SQ_WAKEUP);
instead of calling it unconditionally.
It's mandatory to use fixed files with this feature. Failure to do so
will result in the application getting an -EBADF CQ entry when
submitting IO.
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
We normally have to fget/fput for each IO we do on a file. Even with
the batching we do, the cost of the atomic inc/dec of the file usage
count adds up.
This adds IORING_REGISTER_FILES, and IORING_UNREGISTER_FILES opcodes
for the io_uring_register(2) system call. The arguments passed in must
be an array of __s32 holding file descriptors, and nr_args should hold
the number of file descriptors the application wishes to pin for the
duration of the io_uring instance (or until IORING_UNREGISTER_FILES is
called).
When used, the application must set IOSQE_FIXED_FILE in the sqe->flags
member. Then, instead of setting sqe->fd to the real fd, it sets sqe->fd
to the index in the array passed in to IORING_REGISTER_FILES.
Files are automatically unregistered when the io_uring instance is torn
down. An application need only unregister if it wishes to register a new
set of fds.
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
If we have fixed user buffers, we can map them into the kernel when we
setup the io_uring. That avoids the need to do get_user_pages() for
each and every IO.
To utilize this feature, the application must call io_uring_register()
after having setup an io_uring instance, passing in
IORING_REGISTER_BUFFERS as the opcode. The argument must be a pointer to
an iovec array, and the nr_args should contain how many iovecs the
application wishes to map.
If successful, these buffers are now mapped into the kernel, eligible
for IO. To use these fixed buffers, the application must use the
IORING_OP_READ_FIXED and IORING_OP_WRITE_FIXED opcodes, and then
set sqe->index to the desired buffer index. sqe->addr..sqe->addr+seq->len
must point to somewhere inside the indexed buffer.
The application may register buffers throughout the lifetime of the
io_uring instance. It can call io_uring_register() with
IORING_UNREGISTER_BUFFERS as the opcode to unregister the current set of
buffers, and then register a new set. The application need not
unregister buffers explicitly before shutting down the io_uring
instance.
It's perfectly valid to setup a larger buffer, and then sometimes only
use parts of it for an IO. As long as the range is within the originally
mapped region, it will work just fine.
For now, buffers must not be file backed. If file backed buffers are
passed in, the registration will fail with -1/EOPNOTSUPP. This
restriction may be relaxed in the future.
RLIMIT_MEMLOCK is used to check how much memory we can pin. A somewhat
arbitrary 1G per buffer size is also imposed.
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Similarly to how we use the state->ios_left to know how many references
to get to a file, we can use it to allocate the io_kiocb's we need in
bulk.
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Add a separate io_submit_state structure, to cache some of the things
we need for IO submission.
One such example is file reference batching. io_submit_state. We get as
many references as the number of sqes we are submitting, and drop
unused ones if we end up switching files. The assumption here is that
we're usually only dealing with one fd, and if there are multiple,
hopefuly they are at least somewhat ordered. Could trivially be extended
to cover multiple fds, if needed.
On the completion side we do the same thing, except this is trivially
done just locally in io_iopoll_reap().
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Add support for a polled io_uring instance. When a read or write is
submitted to a polled io_uring, the application must poll for
completions on the CQ ring through io_uring_enter(2). Polled IO may not
generate IRQ completions, hence they need to be actively found by the
application itself.
To use polling, io_uring_setup() must be used with the
IORING_SETUP_IOPOLL flag being set. It is illegal to mix and match
polled and non-polled IO on an io_uring.
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Add a new fsync opcode, which either syncs a range if one is passed,
or the whole file if the offset and length fields are both cleared
to zero. A flag is provided to use fdatasync semantics, that is only
force out metadata which is required to retrieve the file data, but
not others like metadata.
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The submission queue (SQ) and completion queue (CQ) rings are shared
between the application and the kernel. This eliminates the need to
copy data back and forth to submit and complete IO.
IO submissions use the io_uring_sqe data structure, and completions
are generated in the form of io_uring_cqe data structures. The SQ
ring is an index into the io_uring_sqe array, which makes it possible
to submit a batch of IOs without them being contiguous in the ring.
The CQ ring is always contiguous, as completion events are inherently
unordered, and hence any io_uring_cqe entry can point back to an
arbitrary submission.
Two new system calls are added for this:
io_uring_setup(entries, params)
Sets up an io_uring instance for doing async IO. On success,
returns a file descriptor that the application can mmap to
gain access to the SQ ring, CQ ring, and io_uring_sqes.
io_uring_enter(fd, to_submit, min_complete, flags, sigset, sigsetsize)
Initiates IO against the rings mapped to this fd, or waits for
them to complete, or both. The behavior is controlled by the
parameters passed in. If 'to_submit' is non-zero, then we'll
try and submit new IO. If IORING_ENTER_GETEVENTS is set, the
kernel will wait for 'min_complete' events, if they aren't
already available. It's valid to set IORING_ENTER_GETEVENTS
and 'min_complete' == 0 at the same time, this allows the
kernel to return already completed events without waiting
for them. This is useful only for polling, as for IRQ
driven IO, the application can just check the CQ ring
without entering the kernel.
With this setup, it's possible to do async IO with a single system
call. Future developments will enable polled IO with this interface,
and polled submission as well. The latter will enable an application
to do IO without doing ANY system calls at all.
For IRQ driven IO, an application only needs to enter the kernel for
completions if it wants to wait for them to occur.
Each io_uring is backed by a workqueue, to support buffered async IO
as well. We will only punt to an async context if the command would
need to wait for IO on the device side. Any data that can be accessed
directly in the page cache is done inline. This avoids the slowness
issue of usual threadpools, since cached data is accessed as quickly
as a sync interface.
Sample application: http://git.kernel.dk/cgit/fio/plain/t/io_uring.c
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Jens Axboe