Skip to content
GitLab
  1. Mar 08, 2021
  3. Mar 06, 2021
  5. Mar 05, 2021
    • Jens Axboe's avatar
      io_uring: make SQPOLL thread parking saner · 86e0d676
      Jens Axboe authored 
      

We have this weird true/false return from parking, and then some of the
callers decide to look at that. It can lead to unbalanced parks and
sqd locking. Have the callers check the thread status once it's parked.
We know we have the lock at that point, so it's either valid or it's NULL.

Fix race with parking on thread exit. We need to be careful here with
ordering of the sdq->lock and the IO_SQ_THREAD_SHOULD_PARK bit.

Rename sqd->completion to sqd->parked to reflect that this is the only
thing this completion event doesn.

Signed-off-by: default avatarJens Axboe <axboe@kernel.dk>
      86e0d676
    • Jens Axboe's avatar
      io-wq: kill hashed waitqueue before manager exits · 09ca6c40
      Jens Axboe authored 
      If we race with shutting down the io-wq context and someone queueing
a hashed entry, then we can exit the manager with it armed. If it then
triggers after the manager has exited, we can have a use-after-free where
io_wqe_hash_wake() attempts to wake a now gone manager process.

Move the killing of the hashed write queue into the manager itself, so
that we know we've killed it before the task exits.

Fixes: e941894e

 ("io-wq: make buffered file write hashed work map per-ctx")
Signed-off-by: default avatarJens Axboe <axboe@kernel.dk>
      09ca6c40
    • Jens Axboe's avatar
      io_uring: clear IOCB_WAITQ for non -EIOCBQUEUED return · b5b0ecb7
      Jens Axboe authored 
      

The callback can only be armed, if we get -EIOCBQUEUED returned. It's
important that we clear the WAITQ bit for other cases, otherwise we can
queue for async retry and filemap will assume that we're armed and
return -EAGAIN instead of just blocking for the IO.

Cc: stable@vger.kernel.org # 5.9+
Signed-off-by: default avatarJens Axboe <axboe@kernel.dk>
      b5b0ecb7
    • Jens Axboe's avatar
      io_uring: don't keep looping for more events if we can't flush overflow · ca0a2651
      Jens Axboe authored 
      

It doesn't make sense to wait for more events to come in, if we can't
even flush the overflow we already have to the ring. Return -EBUSY for
that condition, just like we do for attempts to submit with overflow
pending.

Cc: stable@vger.kernel.org # 5.11
Signed-off-by: default avatarJens Axboe <axboe@kernel.dk>
      ca0a2651
    • Jens Axboe's avatar
      io_uring: move to using create_io_thread() · 46fe18b1
      Jens Axboe authored 
      

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: default avatarJens Axboe <axboe@kernel.dk>
      46fe18b1
    • Rafael J. Wysocki's avatar
      Merge branch 'powercap' · 7bff4c26
      Rafael J. Wysocki authored 
      * powercap:
  powercap/drivers/dtpm: Add the experimental label to the option description
  powercap/drivers/dtpm: Fix root node initialization
      7bff4c26
    • Max Gurtovoy's avatar
      nvmet: model_number must be immutable once set · d9f273b7
      Max Gurtovoy authored 
      

In case we have already established connection to nvmf target, it
shouldn't be allowed to change the model_number. E.g. if someone will
identify ctrl and get model_number of "my_model" later on will change
the model_numbel via configfs to "my_new_model" this will break the NVMe
specification for "Get Log Page – Persistent Event Log" that refers to
Model Number as: "This field contains the same value as reported in the
Model Number field of the Identify Controller data structure, bytes
63:24."

Although it doesn't mentioned explicitly that this field can't be
changed, we can assume it.

So allow setting this field only once: using configfs or in the first
identify ctrl operation.

Signed-off-by: default avatarMax Gurtovoy <mgurtovoy@nvidia.com>
Signed-off-by: default avatarChristoph Hellwig <hch@lst.de>
      d9f273b7
    • Martin George's avatar
      nvme-fabrics: fix kato initialization · 32feb6de
      Martin George authored 
      

Currently kato is initialized to NVME_DEFAULT_KATO for both
discovery & i/o controllers. This is a problem specifically
for non-persistent discovery controllers since it always ends
up with a non-zero kato value. Fix this by initializing kato
to zero instead, and ensuring various controllers are assigned
appropriate kato values as follows:

non-persistent controllers  - kato set to zero
persistent controllers      - kato set to NVMF_DEV_DISC_TMO
                              (or any positive int via nvme-cli)
i/o controllers             - kato set to NVME_DEFAULT_KATO
                              (or any positive int via nvme-cli)

Signed-off-by: default avatarMartin George <marting@netapp.com>
Signed-off-by: default avatarChristoph Hellwig <hch@lst.de>
      32feb6de