No upstream commit, this is a fix to a stable 5.4 specific backport.
The intention of backport commit cac68d12c531 ("io_uring: grab ->fs as part
of async offload") as found in the stable 5.4 tree was to make
io_sq_wq_submit_work() to switch the workqueue task's ->fs over to the
submitting task's one during the IO operation.
However, due to a small logic error, this change turned out to not have any
actual effect. From a high level, the relevant code in
io_sq_wq_submit_work() looks like
old_fs_struct = current->fs;
do {
...
if (req->fs != current->fs && current->fs != old_fs_struct) {
task_lock(current);
if (req->fs)
current->fs = req->fs;
else
current->fs = old_fs_struct;
task_unlock(current);
}
...
} while (req);
The if condition is supposed to cover the case that current->fs doesn't
match what's needed for processing the request, but observe how it fails
to ever evaluate to true due to the second clause:
current->fs != old_fs_struct will be false in the first iteration as per
the initialization of old_fs_struct and because this prevents current->fs
from getting replaced, the same follows inductively for all subsequent
iterations.
Fix said if condition such that
- if req->fs is set and doesn't match current->fs, the latter will be
switched to the former
- or if req->fs is unset, the switch back to the initial old_fs_struct
will be made, if necessary.
While at it, also correct the condition for the ->fs related cleanup right
before the return of io_sq_wq_submit_work(): currently, old_fs_struct is
restored only if it's non-NULL. It is always non-NULL though and thus, the
if-condition is rendundant. Supposedly, the motivation had been to optimize
and avoid switching current->fs back to the initial old_fs_struct in case
it is found to have the desired value already. Make it so.
Cc: stable@vger.kernel.org # v5.4
Fixes: cac68d12c531 ("io_uring: grab ->fs as part of async offload")
Reviewed-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Nicolai Stange <nstange@suse.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
If we queue work in io_poll_wake(), it will leads to list double
add. So we should add the list when the callback func is the
io_sq_wq_submit_work.
The following oops was seen:
list_add double add: new=ffff9ca6a8f1b0e0, prev=ffff9ca62001cee8,
next=ffff9ca6a8f1b0e0.
------------[ cut here ]------------
kernel BUG at lib/list_debug.c:31!
Call Trace:
<IRQ>
io_poll_wake+0xf3/0x230
__wake_up_common+0x91/0x170
__wake_up_common_lock+0x7a/0xc0
io_commit_cqring+0xea/0x280
? blkcg_iolatency_done_bio+0x2b/0x610
io_cqring_add_event+0x3e/0x60
io_complete_rw+0x58/0x80
dio_complete+0x106/0x250
blk_update_request+0xa0/0x3b0
blk_mq_end_request+0x1a/0x110
blk_mq_complete_request+0xd0/0xe0
nvme_irq+0x129/0x270 [nvme]
__handle_irq_event_percpu+0x7b/0x190
handle_irq_event_percpu+0x30/0x80
handle_irq_event+0x3c/0x60
handle_edge_irq+0x91/0x1e0
do_IRQ+0x4d/0xd0
common_interrupt+0xf/0xf
Fixes: 1c4404efcf2c ("io_uring: make sure async workqueue is canceled on exit")
Reported-by: Jiachen Zhang <zhangjiachen.jaycee@bytedance.com>
Signed-off-by: Muchun Song <songmuchun@bytedance.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
If the process 0 has been initialized io_uring is complete, and
then fork process 1. If process 1 exits and it leads to delete
all reqs from the task_list. If we kill process 0. We will not
send SIGINT signal to the kworker. So we can not remove the req
from the task_list. The io_sq_wq_submit_work() can do that for
us.
Fixes: 1c4404efcf2c ("io_uring: make sure async workqueue is canceled on exit")
Signed-off-by: Muchun Song <songmuchun@bytedance.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The store to req->flags and load req->work_task should not be
reordering in io_cancel_async_work(). We should make sure that
either we store REQ_F_CANCE flag to req->flags or we see the
req->work_task setted in io_sq_wq_submit_work().
Fixes: 1c4404efcf2c ("io_uring: make sure async workqueue is canceled on exit")
Signed-off-by: Muchun Song <songmuchun@bytedance.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The commit
1c4404efcf2c0> ("<io_uring: make sure async workqueue is canceled on exit>")
doesn't solve the resource leak problem totally! When kworker is doing a
io task for the io_uring, The process which submitted the io task has
received a SIGKILL signal from the user. Then the io_cancel_async_work
function could have sent a SIGINT signal to the kworker, but the judging
condition is wrong. So it doesn't send a SIGINT signal to the kworker,
then caused the resource leaking problem.
Why the juding condition is wrong? The process is a multi-threaded process,
we call the thread of the process which has submitted the io task Thread1.
So the req->task is the current macro of the Thread1. when all the threads
of the process have done exit procedure, the last thread will call the
io_cancel_async_work, but the last thread may not the Thread1, so the task
is not equal and doesn't send the SIGINT signal. To fix this bug, we alter
the task attribute of the req with struct files_struct. And check the files
instead.
Fixes: 1c4404efcf2c0 ("io_uring: make sure async workqueue is canceled on exit")
Signed-off-by: Yinyin Zhu <zhuyinyin@bytedance.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit bd74048108c179cea0ff52979506164c80f29da7 upstream.
If we hit an earlier error path in io_uring_create(), then we will have
accounted memory, but not set ctx->{sq,cq}_entries yet. Then when the
ring is torn down in error, we use those values to unaccount the memory.
Ensure we set the ctx entries before we're able to hit a potential error
path.
Cc: stable@vger.kernel.org
Reported-by: Tomáš Chaloupka <chalucha@gmail.com>
Tested-by: Tomáš Chaloupka <chalucha@gmail.com>
Reviewed-by: Stefano Garzarella <sgarzare@redhat.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
[ Upstream commit b36200f543ff07a1cb346aa582349141df2c8068 ]
rings_size() sets sq_offset to the total size of the rings (the returned
value which is used for memory allocation). This is wrong: sq array should
be located within the rings, not after them. Set sq_offset to where it
should be.
Fixes: 75b28affdd6a ("io_uring: allocate the two rings together")
Signed-off-by: Dmitry Vyukov <dvyukov@google.com>
Acked-by: Hristo Venev <hristo@venev.name>
Cc: io-uring@vger.kernel.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Sasha Levin <sashal@kernel.org>
the commit <a4d61e66ee4a> ("<io_uring: prevent re-read of sqe->opcode>")
caused another vulnerability. After io_get_req(), the sqe_submit struct
in req is not initialized, but the following code defaults that
req->submit.opcode is available.
Signed-off-by: Liu Yong <pkfxxxing@gmail.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
when ctx->sqo_mm is zero, io_sq_wq_submit_work() frees 'req'
without deleting it from 'task_list'. After that, 'req' is
accessed in io_ring_ctx_wait_and_kill() which lead to
a use-after-free.
Signed-off-by: Guoyu Huang <hgy5945@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Liu reports that he can trigger a NULL pointer dereference with
IORING_OP_SENDMSG, by changing the sqe->opcode after we've validated
that the previous opcode didn't need a file and didn't assign one.
Ensure we validate and read the opcode only once.
Reported-by: Liu Yong <pkfxxxing@gmail.com>
Tested-by: Liu Yong <pkfxxxing@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Track async work items that we queue, so we can safely cancel them
if the ring is closed or the process exits. Newer kernels handle
this automatically with io-wq, but the old workqueue based setup needs
a bit of special help to get there.
There's no upstream variant of this, as that would require backporting
all the io-wq changes from 5.5 and on. Hence I made a one-off that
ensures that we don't leak memory if we have async work items that
need active cancelation (like socket IO).
Reported-by: Agarwal, Anchal <anchalag@amazon.com>
Tested-by: Agarwal, Anchal <anchalag@amazon.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Sasha Levin <sashal@kernel.org>
commit a8c73c1a614f6da6c0b04c393f87447e28cb6de4 upstream.
Use kvfree() to free the pages and vmas, since they are allocated by
kvmalloc_array() in a loop.
Fixes: d4ef647510b1 ("io_uring: avoid page allocation warnings")
Signed-off-by: Denis Efremov <efremov@linux.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Cc: stable@vger.kernel.org
Link: https://lore.kernel.org/r/20200605093203.40087-1-efremov@linux.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit 4ed734b0d0913e566a9d871e15d24eb240f269f7 upstream.
With the previous fixes for number of files open checking, I added some
debug code to see if we had other spots where we're checking rlimit()
against the async io-wq workers. The only one I found was file size
checking, which we should also honor.
During write and fallocate prep, store the max file size and override
that for the current ask if we're in io-wq worker context.
Cc: stable@vger.kernel.org # 5.1+
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
commit c336e992cb1cb1db9ee608dfb30342ae781057ab upstream.
We already checked this limit when the file was opened, and we keep it
open in the file table. Hence when we added unit_inflight to the count
we want to register, we're doubly accounting these files. This results
in -EMFILE for file registration, if we're at half the limit.
Cc: stable@vger.kernel.org # v5.1+
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Nicolai Stange