There are two reasons why this isn't the best idea:
- It's an odd area to grab a bit of storage space, hence it's an odd area
to grab storage from.
- It puts the 3rd io_kiocb cacheline into the hot path, where normal hot
path just needs the first two.
Use 'cflags' for joint fd/cflags storage. We only need fd until we
successfully issue, and we only need cflags once a request is done and is
completed.
Fixes: 6bf9c47a39 ("io_uring: defer file assignment")
Signed-off-by: Jens Axboe <axboe@kernel.dk>
If an application uses direct open or accept, it knows in advance what
direct descriptor value it will get as it picks it itself. This allows
combined requests such as:
sqe = io_uring_get_sqe(ring);
io_uring_prep_openat_direct(sqe, ..., file_slot);
sqe->flags |= IOSQE_IO_LINK | IOSQE_CQE_SKIP_SUCCESS;
sqe = io_uring_get_sqe(ring);
io_uring_prep_read(sqe,file_slot, buf, buf_size, 0);
sqe->flags |= IOSQE_FIXED_FILE;
io_uring_submit(ring);
where we prepare both a file open and read, and only get a completion
event for the read when both have completed successfully.
Currently links are fully prepared before the head is issued, but that
fails if the dependent link needs a file assigned that isn't valid until
the head has completed.
Conversely, if the same chain is performed but the fixed file slot is
already valid, then we would be unexpectedly returning data from the
old file slot rather than the newly opened one. Make sure we're
consistent here.
Allow deferral of file setup, which makes this documented case work.
Cc: stable@vger.kernel.org # v5.15+
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Pull signal/exit/ptrace updates from Eric Biederman:
"This set of changes deletes some dead code, makes a lot of cleanups
which hopefully make the code easier to follow, and fixes bugs found
along the way.
The end-game which I have not yet reached yet is for fatal signals
that generate coredumps to be short-circuit deliverable from
complete_signal, for force_siginfo_to_task not to require changing
userspace configured signal delivery state, and for the ptrace stops
to always happen in locations where we can guarantee on all
architectures that the all of the registers are saved and available on
the stack.
Removal of profile_task_ext, profile_munmap, and profile_handoff_task
are the big successes for dead code removal this round.
A bunch of small bug fixes are included, as most of the issues
reported were small enough that they would not affect bisection so I
simply added the fixes and did not fold the fixes into the changes
they were fixing.
There was a bug that broke coredumps piped to systemd-coredump. I
dropped the change that caused that bug and replaced it entirely with
something much more restrained. Unfortunately that required some
rebasing.
Some successes after this set of changes: There are few enough calls
to do_exit to audit in a reasonable amount of time. The lifetime of
struct kthread now matches the lifetime of struct task, and the
pointer to struct kthread is no longer stored in set_child_tid. The
flag SIGNAL_GROUP_COREDUMP is removed. The field group_exit_task is
removed. Issues where task->exit_code was examined with
signal->group_exit_code should been examined were fixed.
There are several loosely related changes included because I am
cleaning up and if I don't include them they will probably get lost.
The original postings of these changes can be found at:
https://lkml.kernel.org/r/87a6ha4zsd.fsf@email.froward.int.ebiederm.orghttps://lkml.kernel.org/r/87bl1kunjj.fsf@email.froward.int.ebiederm.orghttps://lkml.kernel.org/r/87r19opkx1.fsf_-_@email.froward.int.ebiederm.org
I trimmed back the last set of changes to only the obviously correct
once. Simply because there was less time for review than I had hoped"
* 'signal-for-v5.17' of git://git.kernel.org/pub/scm/linux/kernel/git/ebiederm/user-namespace: (44 commits)
ptrace/m68k: Stop open coding ptrace_report_syscall
ptrace: Remove unused regs argument from ptrace_report_syscall
ptrace: Remove second setting of PT_SEIZED in ptrace_attach
taskstats: Cleanup the use of task->exit_code
exit: Use the correct exit_code in /proc/<pid>/stat
exit: Fix the exit_code for wait_task_zombie
exit: Coredumps reach do_group_exit
exit: Remove profile_handoff_task
exit: Remove profile_task_exit & profile_munmap
signal: clean up kernel-doc comments
signal: Remove the helper signal_group_exit
signal: Rename group_exit_task group_exec_task
coredump: Stop setting signal->group_exit_task
signal: Remove SIGNAL_GROUP_COREDUMP
signal: During coredumps set SIGNAL_GROUP_EXIT in zap_process
signal: Make coredump handling explicit in complete_signal
signal: Have prepare_signal detect coredumps using signal->core_state
signal: Have the oom killer detect coredumps using signal->core_state
exit: Move force_uaccess back into do_exit
exit: Guarantee make_task_dead leaks the tsk when calling do_task_exit
...
The point of using set_child_tid to hold the kthread pointer was that
it already did what is necessary. There are now restrictions on when
set_child_tid can be initialized and when set_child_tid can be used in
schedule_tail. Which indicates that continuing to use set_child_tid
to hold the kthread pointer is a bad idea.
Instead of continuing to use the set_child_tid field of task_struct
generalize the pf_io_worker field of task_struct and use it to hold
the kthread pointer.
Rename pf_io_worker (which is a void * pointer) to worker_private so
it can be used to store kthreads struct kthread pointer. Update the
kthread code to store the kthread pointer in the worker_private field.
Remove the places where set_child_tid had to be dealt with carefully
because kthreads also used it.
Link: https://lkml.kernel.org/r/CAHk-=wgtFAA9SbVYg0gR1tqPMC17-NYcs0GQkaYg1bGhh1uJQQ@mail.gmail.com
Link: https://lkml.kernel.org/r/87a6grvqy8.fsf_-_@email.froward.int.ebiederm.org
Suggested-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
add a helper to merge two wq_lists, it will be useful in the next
patches.
Reviewed-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Hao Xu <haoxu@linux.alibaba.com>
Link: https://lore.kernel.org/r/20211207093951.247840-2-haoxu@linux.alibaba.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The invariant of io_wq_work_list is that it's empty IFF ->first is NULL,
so no need to initially set ->last. With now having more users of the
list it may play a role, i.e. used in each tw iteration and on every
completion flushing.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/c464ab5cab6e46a858c6d39c107e92b3b5291f13.1633373302.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Apart from just using lists (i.e. io_wq_work_list), we also want to have
stacks, which are a bit faster, and have some interoperability between
them. Add a stack implementation based on io_wq_work_node and some
helpers.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/5d3a412a5ac0d47e0f0499d70d2207d70a68925e.1632516769.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
io-wq divides work into two categories:
1) Work that completes in a bounded time, like reading from a regular file
or a block device. This type of work is limited based on the size of
the SQ ring.
2) Work that may never complete, we call this unbounded work. The amount
of workers here is just limited by RLIMIT_NPROC.
For various uses cases, it's handy to have the kernel limit the maximum
amount of pending workers for both categories. Provide a way to do with
with a new IORING_REGISTER_IOWQ_MAX_WORKERS operation.
IORING_REGISTER_IOWQ_MAX_WORKERS takes an array of two integers and sets
the max worker count to what is being passed in for each category. The
old values are returned into that same array. If 0 is being passed in for
either category, it simply returns the current value.
The value is capped at RLIMIT_NPROC. This actually isn't that important
as it's more of a hint, if we're exceeding the value then our attempt
to fork a new worker will fail. This happens naturally already if more
than one node is in the system, as these values are per-node internally
for io-wq.
Reported-by: Johannes Lundberg <johalun0@gmail.com>
Link: https://github.com/axboe/liburing/issues/420
Signed-off-by: Jens Axboe <axboe@kernel.dk>
io-wq defaults to per-node masks for IO workers. This works fine by
default, but isn't particularly handy for workloads that prefer more
specific affinities, for either performance or isolation reasons.
This adds IORING_REGISTER_IOWQ_AFF that allows the user to pass in a CPU
mask that is then applied to IO thread workers, and an
IORING_UNREGISTER_IOWQ_AFF that simply resets the masks back to the
default of per-node.
Note that no care is given to existing IO threads, they will need to go
through a reschedule before the affinity is correct if they are already
running or sleeping.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
There is an old problem with io-wq cancellation where requests should be
killed and are in io-wq but are not discoverable, e.g. in @next_hashed
or @linked vars of io_worker_handle_work(). It adds some unreliability
to individual request canellation, but also may potentially get
__io_uring_cancel() stuck. For instance:
1) An __io_uring_cancel()'s cancellation round have not found any
request but there are some as desribed.
2) __io_uring_cancel() goes to sleep
3) Then workers wake up and try to execute those hidden requests
that happen to be unbound.
As we already cancel all requests of io-wq there, set IO_WQ_BIT_EXIT
in advance, so preventing 3) from executing unbound requests. The
workers will initially break looping because of getting a signal as they
are threads of the dying/exec()'ing user task.
Cc: stable@vger.kernel.org
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/abfcf8c54cb9e8f7bfbad7e9a0cc5433cc70bdc2.1621781238.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
io-wq relies on a manager thread to create/fork new workers, as needed.
But there's really no strong need for it anymore. We have the following
cases that fork a new worker:
1) Work queue. This is done from the task itself always, and it's trivial
to create a worker off that path, if needed.
2) All workers have gone to sleep, and we have more work. This is called
off the sched out path. For this case, use a task_work items to queue
a fork-worker operation.
3) Hashed work completion. Don't think we need to do anything off this
case. If need be, it could just use approach 2 as well.
Part of this change is incrementing the running worker count before the
fork, to avoid cases where we observe we need a worker and then queue
creation of one. Then new work comes in, we fork a new one. That last
queue operation should have waited for the previous worker to come up,
it's quite possible we don't even need it. Hence move the worker running
from before we fork it off to more efficiently handle that case.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
If we go async with a request, grab the creds that the task currently has
assigned and make sure that the async side switches to them. This is
handled in the same way that we do for registered personalities.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
This allows us to do task creation and setup without needing to use
completions to try and synchronize with the starting thread. Get rid of
the old io_wq_fork_thread() wrapper, and the 'wq' and 'worker' startup
completion events - we can now do setup before the task is running.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
If we move it in there, then we no longer have to care about it in io-wq.
This means we can drop the cred handling in io-wq, and we can drop the
REQ_F_WORK_INITIALIZED flag and async init functions as that was the last
user of it since we moved to the new workers. Then we can also drop
io_wq_work->creds, and just hold the personality u16 in there instead.
Suggested-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
If we put the io-wq from io_uring, we really want it to exit. Provide
a helper that does that for us. Couple that with not having the manager
hold a reference to the 'wq' and the normal SQPOLL exit will tear down
the io-wq context appropriate.
On the io-wq side, our wq context is per task, so only the task itself
is manipulating ->manager and hence it's safe to check and clear without
any extra locking. We just need to ensure that the manager task stays
around, in case it exits.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
exec will cancel any threads, including the ones that io-wq is using. This
isn't a problem, in fact we'd prefer it to be that way since it means we
know that any async work cancels naturally without having to handle it
proactively.
But it does mean that we need to setup a new manager, as the manager and
workers are gone. Handle this at queue time, and cancel work if we fail.
Since the manager can go away without us noticing, ensure that the manager
itself holds a reference to the 'wq' as well. Rename io_wq_destroy() to
io_wq_put() to reflect that.
In the future we can now simplify exec cancelation handling, for now just
leave it the same.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Before the io-wq thread change, we maintained a hash work map and lock
per-node per-ring. That wasn't ideal, as we really wanted it to be per
ring. But now that we have per-task workers, the hash map ends up being
just per-task. That'll work just fine for the normal case of having
one task use a ring, but if you share the ring between tasks, then it's
considerably worse than it was before.
Make the hash map per ctx instead, which provides full per-ctx buffered
write serialization on hashed writes.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
We're now just using fork like we would from userspace, so there's no
need to try and impose extra restrictions or accounting on the user
side of things. That's already being done for us. That also means we
don't have to pass in the user_struct anymore, that's correctly inherited
through ->creds on fork.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
We want to use this in io_uring proper as well, for the SQPOLL thread.
Rename it from fork_thread() to io_wq_fork_thread(), and make it
available through the io-wq.h header.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The async workers are siblings of the task itself, so by definition we
have all the state that we need. Remove any of the state grabbing that
we have, and requests flagging what they need.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Instead of using regular kthread kernel threads, create kernel threads
that are like a real thread that the task would create. This ensures that
we get all the context that we need, without having to carry that state
around. This greatly reduces the code complexity, and the risk of missing
state for a given request type.
With the move away from kthread, we can also dump everything related to
assigned state to the new threads.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
We don't support attach anymore, so doesn't make sense to carry the
use_refs reference count. Get rid of it.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
We hit this case when the task is exiting, and we need somewhere to
do background cleanup of requests. Instead of relying on the io-wq
task manager to do this work for us, just stuff it somewhere where
we can safely run it ourselves directly.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
task_work is a LIFO list, due to how it's implemented as a lockless
list. For long chains of task_work, this can be problematic as the
first entry added is the last one processed. Similarly, we'd waste
a lot of CPU cycles reversing this list.
Wrap the task_work so we have a single task_work entry per task per
ctx, and use that to run it in the right order.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Saving one lock/unlock for io-wq is not super important, but adds some
ugliness in the code. More important, atomic decs not turning it to zero
for some archs won't give the right ordering/barriers so the
io_steal_work() may pretty easily get subtly and completely broken.
Return back 2-step io-wq work exchange and clean it up.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
io_uring no longer issues full cancelations on the io-wq, so remove any
remnants of this code and the IO_WQ_BIT_CANCEL flag.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
For the first time a req punted to io-wq, we'll initialize io_wq_work's
list to be NULL, then insert req to io_wqe->work_list. If this req is not
inserted into tail of io_wqe->work_list, this req's io_wq_work list will
point to another req's io_wq_work. For splitted bio case, this req maybe
inserted to io_wqe->work_list repeatedly, once we insert it to tail of
io_wqe->work_list for the second time, now io_wq_work->list->next will be
invalid pointer, which then result in many strang error, panic, kernel
soft-lockup, rcu stall, etc.
In my vm, kernel doest not have commit cc29e1bf0d ("block: disable
iopoll for split bio"), below fio job can reproduce this bug steadily:
[global]
name=iouring-sqpoll-iopoll-1
ioengine=io_uring
iodepth=128
numjobs=1
thread
rw=randread
direct=1
registerfiles=1
hipri=1
bs=4m
size=100M
runtime=120
time_based
group_reporting
randrepeat=0
[device]
directory=/home/feiman.wxg/mntpoint/ # an ext4 mount point
If we have commit cc29e1bf0d ("block: disable iopoll for split bio"),
there will no splitted bio case for polled io, but I think we still to need
to fix this list corruption, it also should maybe go to stable branchs.
To fix this corruption, if a req is inserted into tail of io_wqe->work_list,
initialize req->io_wq_work->list->next to bu NULL.
Cc: stable@vger.kernel.org
Signed-off-by: Xiaoguang Wang <xiaoguang.wang@linux.alibaba.com>
Reviewed-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Instead of iterating over each request and cancelling it individually in
io_uring_cancel_files(), try to cancel all matching requests and use
->inflight_list only to check if there anything left.
In many cases it should be faster, and we can reuse a lot of code from
task cancellation.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
This one was missed in the earlier conversion, should be included like
any of the other IO identity flags. Make sure we restore to RLIM_INIFITY
when dropping the personality again.
Fixes: 98447d65b4 ("io_uring: move io identity items into separate struct")
Signed-off-by: Jens Axboe <axboe@kernel.dk>
io-wq contains a pointer to the identity, which we just hold in io_kiocb
for now. This is in preparation for putting this outside io_kiocb. The
only exception is struct files_struct, which we'll need different rules
for to avoid a circular dependency.
No functional changes in this patch.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
We have a number of bits that decide what context to inherit. Set up
io-wq flags for these instead. This is in preparation for always having
the various members set, but not always needing them for all requests.
No intended functional changes in this patch.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
There are a few operations that are offloaded to the worker threads. In
this case, we lose process context and end up in kthread context. This
results in ios to be not accounted to the issuing cgroup and
consequently end up as issued by root. Just like others, adopt the
personality of the blkcg too when issuing via the workqueues.
For the SQPOLL thread, it will live and attach in the inited cgroup's
context.
Signed-off-by: Dennis Zhou <dennis@kernel.org>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
If we don't get and assign the namespace for the async work, then certain
paths just don't work properly (like /dev/stdin, /proc/mounts, etc).
Anything that references the current namespace of the given task should
be assigned for async work on behalf of that task.
Cc: stable@vger.kernel.org # v5.5+
Reported-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
RLIMIT_SIZE in needed only for execution from an io-wq context, hence
move all preparations from hot path to io-wq work setup.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
It's easier to return next work from ->do_work() than
having an in-out argument. Looks nicer and easier to compile.
Also, merge io_wq_assign_next() into its only user.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
For an exiting process it tries to cancel all its inflight requests. Use
req->task to match such instead of work.pid. We always have req->task
set, and it will be valid because we're matching only current exiting
task.
Also, remove work.pid and everything related, it's useless now.
Reported-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
If a process is going away, io_uring_flush() will cancel only 1
request with a matching pid. Cancel all of them
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
This adds support for cancelling all io-wq works matching a predicate.
It isn't used yet, so no change in observable behaviour.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
If requests can be submitted and completed inline, we don't need to
initialize whole io_wq_work in io_init_req(), which is an expensive
operation, add a new 'REQ_F_WORK_INITIALIZED' to determine whether
io_wq_work is initialized and add a helper io_req_init_async(), users
must call io_req_init_async() for the first time touching any members
of io_wq_work.
I use /dev/nullb0 to evaluate performance improvement in my physical
machine:
modprobe null_blk nr_devices=1 completion_nsec=0
sudo taskset -c 60 fio -name=fiotest -filename=/dev/nullb0 -iodepth=128
-thread -rw=read -ioengine=io_uring -direct=1 -bs=4k -size=100G -numjobs=1
-time_based -runtime=120
before this patch:
Run status group 0 (all jobs):
READ: bw=724MiB/s (759MB/s), 724MiB/s-724MiB/s (759MB/s-759MB/s),
io=84.8GiB (91.1GB), run=120001-120001msec
With this patch:
Run status group 0 (all jobs):
READ: bw=761MiB/s (798MB/s), 761MiB/s-761MiB/s (798MB/s-798MB/s),
io=89.2GiB (95.8GB), run=120001-120001msec
About 5% improvement.
Signed-off-by: Xiaoguang Wang <xiaoguang.wang@linux.alibaba.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
io_uring is the only user of io-wq, and now it uses only io-wq callback
for all its requests, namely io_wq_submit_work(). Instead of storing
work->runner callback in each instance of io_wq_work, keep it in io-wq
itself.
pros:
- reduces io_wq_work size
- more robust -- ->func won't be invalidated with mem{cpy,set}(req)
- helps other work
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
If the original task is (or has) exited, then the task work will not get
queued properly. Allow for using the io-wq manager task to queue this
work for execution, and ensure that the io-wq manager notices and runs
this work if woken up (or exiting).
Reported-by: Dan Melnic <dmm@fb.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
We always punt async buffered writes to an io-wq helper, as the core
kernel does not have IOCB_NOWAIT support for that. Most buffered async
writes complete very quickly, as it's just a copy operation. This means
that doing multiple locking roundtrips on the shared wqe lock for each
buffered write is wasteful. Additionally, buffered writes are hashed
work items, which means that any buffered write to a given file is
serialized.
Keep identicaly hashed work items contiguously in @wqe->work_list, and
track a tail for each hash bucket. On dequeue of a hashed item, splice
all of the same hash in one go using the tracked tail. Until the batch
is done, the caller doesn't have to synchronize with the wqe or worker
locks again.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
work->data and work->list are shared in union. io_wq_assign_next() sets
->data if a req having a linked_timeout, but then io-wq may want to use
work->list, e.g. to do re-enqueue of a request, so corrupting ->data.
->data is not necessary, just remove it and extract linked_timeout
through @link_list.
Fixes: 60cf46ae60 ("io-wq: hash dependent work")
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
