]> git.baikalelectronics.ru Git - kernel.git/commit
io_uring: fix potential req refcount underflow
authorHao Xu <haoxu@linux.alibaba.com>
Wed, 22 Sep 2021 10:12:38 +0000 (18:12 +0800)
committerJens Axboe <axboe@kernel.dk>
Fri, 24 Sep 2021 16:24:34 +0000 (10:24 -0600)
commit5b7aa38d86f348847a48f71e9ac7715406de900e
treef564a01423b873890b12f8c170858e7f2020334d
parenta62682f92eedb41c1cd8290fa875a4b85624fb9a
io_uring: fix potential req refcount underflow

For multishot mode, there may be cases like:

iowq                                 original context
io_poll_add
  _arm_poll()
  mask = vfs_poll() is not 0
  if mask
(2)  io_poll_complete()
  compl_unlock
   (interruption happens
    tw queued to original
    context)
                                     io_poll_task_func()
                                     compl_lock
                                 (3) done = io_poll_complete() is true
                                     compl_unlock
                                     put req ref
(1) if (poll->flags & EPOLLONESHOT)
      put req ref

EPOLLONESHOT flag in (1) may be from (2) or (3), so there are multiple
combinations that can cause ref underfow.
Let's address it by:
- check the return value in (2) as done
- change (1) to if (done)
    in this way, we only do ref put in (1) if 'oneshot flag' is from
    (2)
- do poll.done check in io_poll_task_func(), so that we won't put ref
  for the second time.

Signed-off-by: Hao Xu <haoxu@linux.alibaba.com>
Link: https://lore.kernel.org/r/20210922101238.7177-4-haoxu@linux.alibaba.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
fs/io_uring.c