On 08/01/2018 10:29 PM, David Sterba wrote:

On Thu, Jul 26, 2018 at 02:53:32PM +0800, Anand Jain wrote:

quoted

From: Anand Jain <redacted>

%fs_devices can be free-ed by btrfs_free_stale_devices() when the
close_fs_devices() drops fs_devices::opened to zero, but close_fs_devices
tries to access the %fs_devices again without the device_list_mutex.

Fix this by bringing the %fs_devices access with in the device_list_mutex.

AFAICS this cannot happen anymore because the two calls are serialized
by the uuid_mutex. But this was not the case when syzbot reported the
problem where your patch would apply.

The parallell access to opened and device list cannot happen when:

* btrfs_scan_one_device that wants to call btrfs_free_stale_devices
* btrfs_close_devices calls close_fs_devices

Fixed by the series:

btrfs: lift uuid_mutex to callers of btrfs_scan_one_device
btrfs: lift uuid_mutex to callers of btrfs_open_devices
btrfs: lift uuid_mutex to callers of btrfs_parse_early_options
btrfs: reorder initialization before the mount locks uuid_mutex
btrfs: fix mount and ioctl device scan ioctl race

If there's a race I don't see, please describe in more detail.

  Right. There is no race with the uuid_mutex patches as above.

  And I just found this- can we make close be consistent with its
  open part.
  btrfs_open_devices() hold device_list_mutex before the update to
  fs_devices::opened. So close_fs_device() could do the same, and be
  theoretically correct.

Thanks, Anand

--
To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html