In chasing a performance issue between using IORING_OP_RECVMSG and
IORING_OP_READV on sockets, tracing showed that we always punt the
socket reads to async offload. This is due to io_file_supports_async()
not checking for S_ISSOCK on the inode. Since sockets supports the
O_NONBLOCK (or MSG_DONTWAIT) flag just fine, add sockets to the list
of file types that we can do a non-blocking issue to.
Signed-off-by: Jens Axboe <axboe@kernel.dk>

The socket read/write helpers only look at the file O_NONBLOCK. not
the iocb IOCB_NOWAIT flag. This breaks users like preadv2/pwritev2
and io_uring that rely on not having the file itself marked nonblocking,
but rather the iocb itself.

Cc: netdev@vger.kernel.org
Acked-by: David Miller <davem@davemloft.net>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

We hash regular files to avoid having multiple threads hammer on the
inode mutex, but it should not be needed on other types of files
(like sockets).
Signed-off-by: Jens Axboe <axboe@kernel.dk>

One major use case of linked commands is the ability to run the next
link inline, if at all possible. This is done correctly for async
offload, but somewhere along the line we lost the ability to do so when
we were able to complete a request without having to punt it. Ensure
that we do so correctly.
Signed-off-by: Jens Axboe <axboe@kernel.dk>

This essentially reverts commit e944475e. For high poll ops
workloads, like TAO, the dynamic allocation of the wait_queue
entry for IORING_OP_POLL_ADD adds considerable extra overhead.
Go back to embedding the wait_queue_entry, but keep the usage of
wait->private for the pointer stashing.
Signed-off-by: Jens Axboe <axboe@kernel.dk>

Don't just assign it from the main call path, that can miss the case
when we're called from issue deferral.
Signed-off-by: Jens Axboe <axboe@kernel.dk>

We use the mutex to guard against registered file updates, for instance.
Ensure we're safe in accessing that state against concurrent updates.
Signed-off-by: Jens Axboe <axboe@kernel.dk>

To avoid going to sleep only to get woken shortly thereafter, spin
briefly for new work upon completion of work.
Signed-off-by: Jens Axboe <axboe@kernel.dk>

We only have one cases of using the waitqueue to wake the worker, the
rest are using wake_up_process(). Since we can save some cycles not
fiddling with the waitqueue io_wqe_worker(), switch the work activation
to task wakeup and get rid of the now unused wait_queue_head_t in
struct io_worker.
Signed-off-by: Jens Axboe <axboe@kernel.dk>

Some commands will invariably end in a failure in the sense that the
completion result will be less than zero. One such example is timeouts
that don't have a completion count set, they will always complete with
-ETIME unless cancelled.

For linked commands, we sever links and fail the rest of the chain if
the result is less than zero. Since we have commands where we know that
will happen, add IOSQE_IO_HARDLINK as a stronger link that doesn't sever
regardless of the completion result. Note that the link will still sever
if we fail submitting the parent request, hard links are only resilient
in the presence of completion results for requests that did submit
correctly.

Cc: stable@vger.kernel.org # v5.4
Reviewed-by: Pavel Begunkov <asml.silence@gmail.com>
Reported-by: 李通洲 <carter.li@eoitek.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

thatn -> than.
Signed-off-by: Liming Wu <19092205@suning.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

Links are created by chaining requests through req->list with an
exception that head uses req->link_list. (e.g. link_list->list->list)
Because of that, io_req_link_next() needs complex splicing to advance.

Link them all through list_list. Also, it seems to be simpler and more
consistent IMHO.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

In case of an error io_submit_sqe() drops a request and continues
without it, even if the request was a part of a link. Not only it
doesn't cancel links, but also may execute wrong sequence of actions.

Stop consuming sqes, and let the user handle errors.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

We recently changed this from a single list to an rbtree, but for some
real life workloads, the rbtree slows down the submission/insertion
case enough so that it's the top cycle consumer on the io_uring side.
In testing, using a hash table is a more well rounded compromise. It
is fast for insertion, and as long as it's sized appropriately, it
works well for the cancellation case as well. Running TAO with a lot
of network sockets, this removes io_poll_req_insert() from spending
2% of the CPU cycles.
Reported-by: Dan Melnic <dmm@fb.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

If someone removes a node from a list, and then later adds it back to
a list, we can have invalid data in ->next. This can cause all sorts
of issues. One such use case is the IORING_OP_POLL_ADD command, which
will do just that if we race and get woken twice without any pending
events. This is a pretty rare case, but can happen under extreme loads.
Dan reports that he saw the following crash:

BUG: kernel NULL pointer dereference, address: 0000000000000000
PGD d283ce067 P4D d283ce067 PUD e5ca04067 PMD 0
Oops: 0002 [#1] SMP
CPU: 17 PID: 10726 Comm: tao:fast-fiber Kdump: loaded Not tainted 5.2.9-02851-gac7bc042d2d1 #116
Hardware name: Quanta Twin Lakes MP/Twin Lakes Passive MP, BIOS F09_3A17 05/03/2019
RIP: 0010:io_wqe_enqueue+0x3e/0xd0
Code: 34 24 74 55 8b 47 58 48 8d 6f 50 85 c0 74 50 48 89 df e8 35 7c 75 00 48 83 7b 08 00 48 8b 14 24 0f 84 84 00 00 00 48 8b 4b 10 <48> 89 11 48 89 53 10 83 63 20 fe 48 89 c6 48 89 df e8 0c 7a 75 00
RSP: 0000:ffffc90006858a08 EFLAGS: 00010082
RAX: 0000000000000002 RBX: ffff889037492fc0 RCX: 0000000000000000
RDX: ffff888e40cc11a8 RSI: ffff888e40cc11a8 RDI: ffff889037492fc0
RBP: ffff889037493010 R08: 00000000000000c3 R09: ffffc90006858ab8
R10: 0000000000000000 R11: 0000000000000000 R12: ffff888e40cc11a8
R13: 0000000000000000 R14: 00000000000000c3 R15: ffff888e40cc1100
FS:  00007fcddc9db700(0000) GS:ffff88903fa40000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 0000000000000000 CR3: 0000000e479f5003 CR4: 00000000007606e0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
PKRU: 55555554
Call Trace:
 <IRQ>
 io_poll_wake+0x12f/0x2a0
 __wake_up_common+0x86/0x120
 __wake_up_common_lock+0x7a/0xc0
 sock_def_readable+0x3c/0x70
 tcp_rcv_established+0x557/0x630
 tcp_v6_do_rcv+0x118/0x3c0
 tcp_v6_rcv+0x97e/0x9d0
 ip6_protocol_deliver_rcu+0xe3/0x440
 ip6_input+0x3d/0xc0
 ? ip6_protocol_deliver_rcu+0x440/0x440
 ipv6_rcv+0x56/0xd0
 ? ip6_rcv_finish_core.isra.18+0x80/0x80
 __netif_receive_skb_one_core+0x50/0x70
 netif_receive_skb_internal+0x2f/0xa0
 napi_gro_receive+0x125/0x150
 mlx5e_handle_rx_cqe+0x1d9/0x5a0
 ? mlx5e_poll_tx_cq+0x305/0x560
 mlx5e_poll_rx_cq+0x49f/0x9c5
 mlx5e_napi_poll+0xee/0x640
 ? smp_reschedule_interrupt+0x16/0xd0
 ? reschedule_interrupt+0xf/0x20
 net_rx_action+0x286/0x3d0
 __do_softirq+0xca/0x297
 irq_exit+0x96/0xa0
 do_IRQ+0x54/0xe0
 common_interrupt+0xf/0xf
 </IRQ>
RIP: 0033:0x7fdc627a2e3a
Code: 31 c0 85 d2 0f 88 f6 00 00 00 55 48 89 e5 41 57 41 56 4c 63 f2 41 55 41 54 53 48 83 ec 18 48 85 ff 0f 84 c7 00 00 00 48 8b 07 <41> 89 d4 49 89 f5 48 89 fb 48 85 c0 0f 84 64 01 00 00 48 83 78 10

when running a networked workload with about 5000 sockets being polled
for. Fix this by clearing node->next when the node is being removed from
the list.

Fixes: 6206f0e1 ("io-wq: shrink io_wq_work a bit")
Reported-by: Dan Melnic <dmm@fb.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

If we defer a timeout, we should ensure that we copy the timespec
when we have consumed the sqe. This is similar to commit f67676d1
for read/write requests. We already did this correctly for timeouts
deferred as links, but do it generally and use the infrastructure added
by commit 1a6b74fc instead of having the timeout deferral use its
own.
Signed-off-by: Jens Axboe <axboe@kernel.dk>

There's really no reason why we forbid things like link/drain etc on
regular timeout commands. Enable the usual SQE flags on timeouts.
Reported-by: 李通洲 <carter.li@eoitek.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

If BLK_DEV_ZONED isn't set, 'ret' isn't used. This makes gcc complain,
rightfully. Move ret where it is used.

Fixes: 979d5447 ("null_blk: cleanup null_gendisk_register")
Signed-off-by: Jens Axboe <axboe@kernel.dk>

Queue dma alignment limit requires users(fs, target, ...) of block layer
to pass aligned buffer.

So far brd doesn't support un-aligned buffer, even though it is easy
to support it.

However, given brd is often used for debug purpose, and there are other
drivers which can't support un-aligned buffer too.

So add warning so that brd users know what to fix.
Reported-by: Stephen Rust <srust@blockbridge.com>
Cc: Stephen Rust <srust@blockbridge.com>
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

Now we depend on blk_queue_split() to respect most of queue limit
(the only one exception could be dma alignment), however
blk_queue_split() isn't used for brd, so this limit isn't respected
since v4.3.

Also max_hw_sectors limit doesn't play a big role for brd, which is
added since brd is added to tree for unknown reason.

So remove it.
Signed-off-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

For each I/O request, blkback first maps the foreign pages for the
request to its local pages.  If an allocation of a local page for the
mapping fails, it should unmap every mapping already made for the
request.

However, blkback's handling mechanism for the allocation failure does
not mark the remaining foreign pages as unmapped.  Therefore, the unmap
function merely tries to unmap every valid grant page for the request,
including the pages not mapped due to the allocation failure.  On a
system that fails the allocation frequently, this problem leads to
following kernel crash.

  [  372.012538] BUG: unable to handle kernel NULL pointer dereference at 0000000000000001
  [  372.012546] IP: [<ffffffff814071ac>] gnttab_unmap_refs.part.7+0x1c/0x40
  [  372.012557] PGD 16f3e9067 PUD 16426e067 PMD 0
  [  372.012562] Oops: 0002 [#1] SMP
  [  372.012566] Modules linked in: act_police sch_ingress cls_u32
  ...
  [  372.012746] Call Trace:
  [  372.012752]  [<ffffffff81407204>] gnttab_unmap_refs+0x34/0x40
  [  372.012759]  [<ffffffffa0335ae3>] xen_blkbk_unmap+0x83/0x150 [xen_blkback]
  ...
  [  372.012802]  [<ffffffffa0336c50>] dispatch_rw_block_io+0x970/0x980 [xen_blkback]
  ...
  Decompressing Linux... Parsing ELF... done.
  Booting the kernel.
  [    0.000000] Initializing cgroup subsys cpuset

This commit fixes this problem by marking the grant pages of the given
request that didn't mapped due to the allocation failure as invalid.

Fixes: c6cc142d ("xen-blkback: use balloon pages for all mappings")
Reviewed-by: David Woodhouse <dwmw@amazon.de>
Reviewed-by: Maximilian Heyne <mheyne@amazon.de>
Reviewed-by: Paul Durrant <pdurrant@amazon.co.uk>
Reviewed-by: Roger Pau Monné <roger.pau@citrix.com>
Signed-off-by: SeongJae Park <sjpark@amazon.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

Right now we return it to userspace, which means the application has
to poll for the socket to be writeable. Let's just treat it like
-EAGAIN and have io_uring handle it internally, this makes it much
easier to use.
Signed-off-by: Jens Axboe <axboe@kernel.dk>

The current zone revalidation code has a major problem in that it
doesn't update the zone size and q->nr_zones atomically, leading
to a short window where an out of bounds access to the zone arrays
is possible.

To fix this move the setting of the zone size into the crticial
sections blk_revalidate_disk_zones so that it gets updated together
with the zone bitmaps and q->nr_zones.  This also slightly simplifies
the caller as it deducts the zone size from the report_zones.

This change also allows to check for a power of two zone size in generic
code.
Reported-by: Hans Holmberg <hans@owltronix.com>
Reviewed-by: Javier González <javier@javigon.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

bio based drivers only need to update q->nr_zones.  Do that manually
instead of overloading blk_revalidate_disk_zones to keep that function
simpler for the next round of changes that will rely even more on the
request based functionality.
Reviewed-by: Javier González <javier@javigon.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

Allocate the conventional zone bitmap and the sequential zone locking
bitmap only when we find a zone of the respective type.  This avoids
wasting memory on the conventional zone bitmap for devices that only
have sequential zones, and will also prepare for other future changes.
Reviewed-by: Javier González <javier@javigon.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

Invert the meaning of seq_zones_bitmap by keeping a bitmap of
conventional zones.  This allows not having a bitmap for devices
that do not have conventional zones.
Reviewed-by: Javier González <javier@javigon.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

Simplify the arguments to blkdev_nr_zones by passing a gendisk instead
of the block_device and capacity.  This also removes the need for
__blkdev_nr_zones as all callers are outside the fast path and can
deal with the additional branch.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

Use a saner size calculation, and do a trivial cleanup on the zone
revalidation to prepare to future changes.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

For zoned=1 mode, the zone size must be a power of 2. Check this not
only when the zone size is specified during modprobe, but also when
creating a zoned null_blk device using configfs.
Signed-off-by: Damien Le Moal <damien.lemoal@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

bvec_iter_advance() is quite popular, but compilers fail to do proper
alias analysis and optimise it good enough. The assembly is checked
for gcc 9.2, x86-64.

- remove @iter->bi_size from min(...), as it's always less than @bytes.
Modify at the beginning and forget about it.

- the compiler isn't able to collapse memory dependencies and remove
writes in the loop. Help it by explicitely using local vars.
Signed-off-by: Arvind Sankar <nivedita@alum.mit.edu>
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

Since commit b18fdf71 ("io_uring: simplify io_req_link_next()"),
the io_wq_current_is_worker function is no longer needed, clean it
up.
Signed-off-by: Jackie Liu <liuyun01@kylinos.cn>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

Parameter ctx we have never used, clean it up.
Signed-off-by: Jackie Liu <liuyun01@kylinos.cn>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

If this flag is set, applications can be certain that any data for
async offload has been consumed when the kernel has consumed the
SQE.
Signed-off-by: Jens Axboe <axboe@kernel.dk>

Just like commit f67676d1 for read/write requests, this one ensures
that the sockaddr data has been copied for IORING_OP_CONNECT if we need
to punt the request to async context.
Signed-off-by: Jens Axboe <axboe@kernel.dk>

Just like commit f67676d1 for read/write requests, this one ensures
that the msghdr data is fully copied if we need to punt a recvmsg or
sendmsg system call to async context.
Signed-off-by: Jens Axboe <axboe@kernel.dk>

Currently we don't copy the iovecs when we punt to async context. This
can be problematic for applications that store the iovec on the stack,
as they often assume that it's safe to let the iovec go out of scope
as soon as IO submission has been called. This isn't always safe, as we
will re-copy the iovec once we're in async context.

Make this 100% safe by copying the iovec just once. With this change,
applications may safely store the iovec on the stack for all cases.
Reported-by: 李通洲 <carter.li@eoitek.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

Right now we just copy the sqe for async offload, but we want to store
more context across an async punt. In preparation for doing so, put the
sqe copy inside a structure that we can expand. With this pointer added,
we can get rid of REQ_F_FREE_SQE, as that is now indicated by whether
req->io is NULL or not.

No functional changes in this patch.
Signed-off-by: Jens Axboe <axboe@kernel.dk>

Commit 6917d068 ("block: merge invalidate_partitions into
rescan_partitions") caused a regression where systemd-udevd spins
forever using max CPU starting at boot time.

It's caused by a behavior change where a KOBJ_CHANGE uevent is now sent
in a case where previously it wasn't.

Restore the old behavior.

Fixes: 6917d068 ("block: merge invalidate_partitions into rescan_partitions")
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

We should never return -ERESTARTSYS to userspace, transform it into
-EINTR.

Cc: stable@vger.kernel.org # v5.3+
Signed-off-by: Jens Axboe <axboe@kernel.dk>