Re: [PATCH v3] btrfs: handle dynamically reappearing missing device

[Anand Jain <anand.jain@oracle.com>]

On 12/18/2017 08:01 PM, Nikolay Borisov wrote:
> 
> 
> On 17.12.2017 05:04, Anand Jain wrote:
>> If the device is not present at the time of (-o degrade) mount,
>> the mount context will create a dummy missing struct btrfs_device.
>> Later this device may reappear after the FS is mounted and
>> then device is included in the device list but it missed the
>> open_device part. So this patch handles that case by going
>> through the open_device steps which this device missed and finally
>> adds to the device alloc list.
>>
>> So now with this patch, to bring back the missing device user can run,
>>
>>     btrfs dev scan <path-of-missing-device>
>>
>> Without this kernel patch, even though 'btrfs fi show' and 'btrfs
>> dev ready' would tell you that missing device has reappeared
>> successfully but actually in kernel FS layer it didn't.
>>
>> Signed-off-by: Anand Jain <anand.jain@oracle.com>
>> ---
>> This patch needs:
>>   [PATCH 0/4]  factor __btrfs_open_devices()
>> v3:
>> The check for missing in the device_list_add() is now a another
>> patch as its not related.
>>   btrfs: fix inconsistency during missing device rejoin
>>
>> v2:
>> Add more comments.
>> Add more change log.
>> Add to check if device missing is set, to handle the case
>> dev open fail and user will rerun the dev scan.
>>
>>   fs/btrfs/volumes.c | 57 ++++++++++++++++++++++++++++++++++++++++++++++++++++--
>>   1 file changed, 55 insertions(+), 2 deletions(-)
>>
>> diff --git a/fs/btrfs/volumes.c b/fs/btrfs/volumes.c
>> index 93d65c72b731..5c3190c65f81 100644
>> --- a/fs/btrfs/volumes.c
>> +++ b/fs/btrfs/volumes.c
>> @@ -812,8 +812,61 @@ static noinline int device_list_add(const char *path,
>>   		rcu_string_free(device->name);
>>   		rcu_assign_pointer(device->name, name);
>>   		if (test_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state)) {
>> -			fs_devices->missing_devices--;
>> -			clear_bit(BTRFS_DEV_STATE_MISSING, &device->dev_state);
>> +			int ret;
>> +			struct btrfs_fs_info *fs_info = fs_devices->fs_info;
>> +			fmode_t fmode = FMODE_READ | FMODE_WRITE | FMODE_EXCL;
>> +
>> +			if (btrfs_super_flags(disk_super) &
>> +					BTRFS_SUPER_FLAG_SEEDING)
>> +				fmode &= ~FMODE_WRITE;
>> +
>> +			/*
>> +			 * Missing can be set only when FS is mounted.
>> +			 * So here its always fs_devices->opened > 0 and most
>> +			 * of the struct device members are already updated by
>> +			 * the mount process even if this device was missing, so
>> +			 * now follow the normal open device procedure for this
>> +			 * device. The scrub will take care of filling the
> 
> So how is scrub supposed to be initiated - automatically or by the user
> since it's assumed he will have seen the message that a device has been
> added? Reading the comment I'd expect scrub is kicked automatically.

  Oh. Right. it isn't a good idea to reset degraded option (below) in
  this thread, instead let scrub do it. Then user can notice too. In fact
  even if all the devices are present but if chunks does not match to
  what user has configured for, then its a degraded volume.

  When I was resetting the degraded option below. I had that question,
  but now the thought process and reasoning is satisfying. Will update.
  Thanks.


>> +			 * missing stripes for raid56 and balance for raid1 and
>> +			 * raid10.
>> +			 */
>> +			ASSERT(fs_devices->opened);
>> +			mutex_lock(&fs_devices->device_list_mutex);
>> +			mutex_lock(&fs_info->chunk_mutex);
>> +			/*
>> +			 * As of now do not fail the dev scan thread for the
>> +			 * reason that btrfs_open_one_device() fails and keep
>> +			 * the legacy dev scan requisites as it is.
>> +			 * And reset missing only if open is successful, as
>> +			 * user can rerun dev scan after fixing the device
>> +			 * for which the device open (below) failed.
>> +			 */
>> +			ret = btrfs_open_one_device(fs_devices, device, fmode,
>> +							fs_info->bdev_holder);
>> +			if (!ret) {
>> +				fs_devices->missing_devices--;
>> +				clear_bit(BTRFS_DEV_STATE_MISSING,
>> +							&device->dev_state);
>> +				btrfs_clear_opt(fs_info->mount_opt, DEGRADED);  <-- 


>> +				btrfs_warn(fs_info,
>> +					"BTRFS: device %s devid %llu joined\n",
>> +					path, devid);
> 
> why do we have to warn, this is considered to be a "good" thing so
> perhaps btrfs_info?

  you are right, it should be btrfs_info. Will change.

>> +			}
>> +
>> +			if (test_bit(BTRFS_DEV_STATE_WRITEABLE,
>> +				     &device->dev_state) &&
> 
> Can a device that is missing really be writable? So we have 2 cases
> where we add a missing device:
> 
> 1. Is via add_missing_dev which sets dev_state_missing so writable will
> be false.
> 
> 2. In read_one_dev when we have successfully found the device from
> btrfs_find_device but it doesn't have a ->bdev member, in which case we
> don't really clear the writable fact (but perhaps we should) ?

  Writeable flag is more about seed device which isn't a writeable
  device. btrfs_open_one_device() does reset writeable flag.

> Overall the lifecycle of these flags is not very clear ;\

  Yep. Its bit overlapping and needs cleanup.

Thanks, Anand


>> +			    !test_bit(BTRFS_DEV_STATE_REPLACE_TGT,
>> +				      &device->dev_state)) {
>> +				fs_devices->total_rw_bytes +=
>> +							device->total_bytes;
>> +				atomic64_add(device->total_bytes -
>> +						device->bytes_used,
>> +						&fs_info->free_chunk_space);
>> +			}
>> +			set_bit(BTRFS_DEV_STATE_IN_FS_METADATA,
>> +							&device->dev_state);
>> +			mutex_unlock(&fs_info->chunk_mutex);
>> +			mutex_unlock(&fs_devices->device_list_mutex);
>>   		}
>>   	}
>>   
>>
> --
> 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
>