On 05/16/2018 12:33 AM, David Sterba wrote:
On Thu, Apr 12, 2018 at 10:29:23AM +0800, Anand Jain wrote:
uuid_mutex lock is not a per-fs lock but a global lock. The main aim of
this patch-set is to critically review the usage of this lock, and delete
the unnecessary once. By doing this we improve the concurrency of
device operations across multiple btrfs filesystems is in the system.
patch 1: Was sent before, I am including it here, as its about uuid_mutex.
patch 2-9: Are cleanup and or preparatory patches.
patch 10-14: Drops the uuid_mutex and makes sure there is enough lock,
as discussed in the patch change log.
patch 15: A generic cleanup patch around functions in the same context.
These patches are on top of
https://github.com/kdave/btrfs-devel.git remove-volume-mutex
And it will be a good idea to go along with the kill-volume-mutex patches.
This is tested with xfstests and there are no _new_ regression. And I am
trying to understand the old regressions, and notice that they are
inconsistent.
Anand Jain (15):
btrfs: optimize move uuid_mutex closer to the critical section
btrfs: rename struct btrfs_fs_devices::list
btrfs: cleanup __btrfs_open_devices() drop head pointer
btrfs: rename __btrfs_close_devices to close_fs_devices
btrfs: rename __btrfs_open_devices to open_fs_devices
btrfs: cleanup find_device() drop list_head pointer
btrfs: cleanup btrfs_rm_device() promote fs_devices pointer
btrfs: cleanup btrfs_rm_device() use cur_devices
btrfs: uuid_mutex in read_chunk_tree, add a comment
btrfs: drop uuid_mutex in btrfs_free_extra_devids()
btrfs: drop uuid_mutex in btrfs_open_devices()
btrfs: drop uuid_mutex in close_fs_devices()
btrfs: drop uuid_mutex in btrfs_dev_replace_finishing()
btrfs: drop uuid_mutex in btrfs_destroy_dev_replace_tgtdev()
btrfs: cleanup btrfs_destroy_dev_replace_tgtdev() localize
btrfs_fs_devices
Patches 10 and 12 haven't been merged, the rest is now in misc-next.
Testing hasn't revealed any problems related to the uuid/device locks
but as said before we don't have stress tests.
Our test cases are sequential. We need same device related test cases
in a concurrent manner.
Just for experiment, I ran two instances of xfstests concurrently (on a
separate set of test and scratch devices), they ran fine with these
patches, not a perfect solution though. We need to think of better
ways.
The main challenge with testing concurrency / racing is how to
synchronize racing threads.
Thanks, Anand
--
To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
--
To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html