Re: [PATCH V5 0/4] btrf_show_devname related fixes

[Su Yue <l@damenly.su>]

On Tue 31 Aug 2021 at 06:41, Anand Jain <anand.jain@oracle.com> 
wrote:

> Ping.
>
In the past week, I spent some time on testing this patchset and
the patch 'btrfs: fix lockdep warning while mounting sprout fs'.

Same as you, I wonder if there is any race condition. Just wrote
some scripts including device removal, replace and sprout while
busy looping showing mount info. It runned well and passed 
xfstests
in my two VMs. I'm not a pundit in btrfs device field so I can't 
give my
RBs even these patches looks almost okay to me.

--
Su

> This patchset is fixing a bug reported by btrfs/238.
>
> Here I am proposing to remove traversing the device list in
> btrfs_show_devname(), instead use device->name as stored in
> fs_devices::latest_dev.
>
> Kindly review/comment.
>
> Thx, Anand
>
>
>
> On 24/08/2021 13:05, Anand Jain wrote:
>> v5: Patches reorged.
>>   Patch (btrfs: consolidate device_list_mutex in prepare_sprout 
>>   to its parent)
>>   moved into a new set as it has a dependency on an older patch 
>>   in the ML.
>>   Change log updated.
>> v4: Fix unrelated changes
>> v3: Add missing rcu_lock in show_devname
>> v2: Use latest_dev so that device path is also shown
>> Su Yue reported [1] warn() as a result of a race between the 
>> following two
>> threads,
>>    Thread-A: function stack leading to btrfs_prepare_sprout()
>> and
>>    Thread-B: function stack leading to btrfs_show_devname()
>> [1]
>> https://patchwork.kernel.org/project/linux-btrfs/patch/20210818041944.5793-1-l@damenly.su/
>> While btrfs_prepare_sprout() moves the fs_devices::devices into
>> fs_devices::seed_list, the btrfs_show_devname searched for the 
>> devices
>> and found none, leading to the warning as in [1] (above).
>> The btrfs_prepare_sprout() uses device_list_mutex however
>> btrfs_show_devname() don't and, the device_list_mutex in
>> btrfs_show_devname() was removed by the patch 88c14590cdd6
>> (btrfs: use RCU in btrfs_show_devname for device list 
>> traversal)
>> for the perforamcne reasons.
>> This series does not intend to reintroduce the 
>> device_list_mutex in
>> btrfs_prepare_sprout() but instead saves the pointer to 
>> btrfs_devices
>> in the fs_devices::latest_dev so that btrfs_show_devname() can 
>> use it.
>> patch 1 converts fs_devices::latest_bdev type from struct 
>> block_device to
>>          struct btrfs_device and renames it to latest_dev
>> patch 2 btrfs_show_devname() uses the 
>> fs_devices::latest_dev::name to show
>>          the device path in the /proc/self/mounts
>> patch 3 fixes a stale latest_dev pointer after the sprout 
>> operation
>> patch 4 fixes an old comment about the function 
>> btrfs_show_devname()
>> Anand Jain (4):
>>    btrfs: convert latest_bdev type to struct btrfs_device and 
>>    rename
>>    btrfs: use latest_dev in btrfs_show_devname
>>    btrfs: update latest_dev when we sprout
>>    btrfs: fix comment about the btrfs_show_devname
>>   fs/btrfs/disk-io.c   |  6 +++---
>>   fs/btrfs/extent_io.c |  2 +-
>>   fs/btrfs/inode.c     |  2 +-
>>   fs/btrfs/super.c     | 26 ++++----------------------
>>   fs/btrfs/volumes.c   | 17 ++++++++---------
>>   fs/btrfs/volumes.h   |  2 +-
>>   6 files changed, 18 insertions(+), 37 deletions(-)
>>