On 2020/3/19 上午4:47, Marcos Paulo de Souza wrote:
> On Wed, Mar 18, 2020 at 06:56:51PM +0800, Qu Wenruo wrote:
>>
>>
>> On 2020/3/8 上午6:45, Marcos Paulo de Souza wrote:
>>> From: Marcos Paulo de Souza <mpdesouza@suse.com>
>>>
>>> By using the -a flag on replace makes btrfs issue a resize ioctl after
>>> the replace finishes. This argument is a shortcut for
>>>
>>> btrfs replace start -f 3 /dev/sdf BTRFS/
>>> btrfs fi resize 3:max BTRFS/
>>>
>>> The -a stands for "automatically resize"
>>>
>>> Fixes: #21
>>>
>>> Signed-off-by: Marcos Paulo de Souza <mpdesouza@suse.com>
>>> ---
>>>  Documentation/btrfs-replace.asciidoc |  4 +++-
>>>  cmds/replace.c                       | 19 +++++++++++++++++--
>>>  2 files changed, 20 insertions(+), 3 deletions(-)
>>>
>>> diff --git a/Documentation/btrfs-replace.asciidoc b/Documentation/btrfs-replace.asciidoc
>>> index b73bf1b3..e0b30066 100644
>>> --- a/Documentation/btrfs-replace.asciidoc
>>> +++ b/Documentation/btrfs-replace.asciidoc
>>> @@ -18,7 +18,7 @@ SUBCOMMAND
>>>  *cancel* <mount_point>::
>>>  Cancel a running device replace operation.
>>>  
>>> -*start* [-Bfr] <srcdev>|<devid> <targetdev> <path>::
>>> +*start* [-aBfr] <srcdev>|<devid> <targetdev> <path>::
>>>  Replace device of a btrfs filesystem.
>>>  +
>>>  On a live filesystem, duplicate the data to the target device which
>>> @@ -53,6 +53,8 @@ never allowed to be used as the <targetdev>.
>>>  +
>>>  -B::::
>>>  no background replace.
>>> ++a::::
>>> +automatically resizes the filesystem if the <targetdev> is bigger than <srcdev>.
>>
>> Resizes "to its max size".
>>
>>>  
>>>  *status* [-1] <mount_point>::
>>>  Print status and progress information of a running device replace operation.
>>> diff --git a/cmds/replace.c b/cmds/replace.c
>>> index 2321aa15..48f470cd 100644
>>> --- a/cmds/replace.c
>>> +++ b/cmds/replace.c
>>> @@ -91,7 +91,7 @@ static int dev_replace_handle_sigint(int fd)
>>>  }
>>>  
>>>  static const char *const cmd_replace_start_usage[] = {
>>> -	"btrfs replace start [-Bfr] <srcdev>|<devid> <targetdev> <mount_point>",
>>> +	"btrfs replace start [-aBfr] <srcdev>|<devid> <targetdev> <mount_point>",
>>>  	"Replace device of a btrfs filesystem.",
>>>  	"On a live filesystem, duplicate the data to the target device which",
>>>  	"is currently stored on the source device. If the source device is not",
>>> @@ -104,6 +104,8 @@ static const char *const cmd_replace_start_usage[] = {
>>>  	"from the system, you have to use the <devid> parameter format.",
>>>  	"The <targetdev> needs to be same size or larger than the <srcdev>.",
>>>  	"",
>>> +	"-a     automatically resize the filesystem if the <targetdev> is bigger",
>>> +	"       than <srcdev>",
>>
>> Same here, resize to its max size.
>>
>>>  	"-r     only read from <srcdev> if no other zero-defect mirror exists",
>>>  	"       (enable this if your drive has lots of read errors, the access",
>>>  	"       would be very slow)",
>>> @@ -129,6 +131,7 @@ static int cmd_replace_start(const struct cmd_struct *cmd,
>>>  	char *path;
>>>  	char *srcdev;
>>>  	char *dstdev = NULL;
>>> +	bool auto_resize = false;
>>>  	int avoid_reading_from_srcdev = 0;
>>>  	int force_using_targetdev = 0;
>>>  	u64 dstdev_block_count;
>>> @@ -138,8 +141,11 @@ static int cmd_replace_start(const struct cmd_struct *cmd,
>>>  	u64 dstdev_size;
>>>  
>>>  	optind = 0;
>>> -	while ((c = getopt(argc, argv, "Brf")) != -1) {
>>> +	while ((c = getopt(argc, argv, "aBrf")) != -1) {
>>>  		switch (c) {
>>> +		case 'a':
>>> +			auto_resize = true;
>>> +			break;
>>>  		case 'B':
>>>  			do_not_background = 1;
>>>  			break;
>>> @@ -309,6 +315,15 @@ static int cmd_replace_start(const struct cmd_struct *cmd,
>>>  			goto leave_with_error;
>>>  		}
>>>  	}
>>> +
>>> +	if (ret == 0 && auto_resize && dstdev_size > srcdev_size) {
>>> +		char amount[BTRFS_PATH_NAME_MAX + 1];
>>> +		snprintf(amount, BTRFS_PATH_NAME_MAX, "%s:max", srcdev);
>>> +
>>> +		if (resize_filesystem(amount, path))
>>> +			goto leave_with_error;
>>> +	}
>>> +
>>
>> I'm pretty happy since it's all done in user space.
>>
>> But this is a different error, here we succeeded in replace, but only
>> failed in resize (which should be pretty rare to hit though).
>>
>> We should provide some better error message for it other than just error
>> out.
> 
> Function resize_filesystem already checks for errors, and it even print's a
> resize message before calling the resize ioctl. Or do you mean another message
> specifying that "replace finished, but resize failed" along the messages already
> being provided by resize_filesystem function?

Yes, that's what I mean.

Thanks,
Qu
> 
> Thanks,
>   Marcos
> 
>>
>> Thanks,
>> Qu
>>
>>>  	close_file_or_dir(fdmnt, dirstream);
>>>  	return 0;
>>>  
>>>
>>
> 
> 
>