ocfs2: give an obvious tip for dismatch cluster names
diff mbox

Message ID 1495089336-19312-1-git-send-email-ghe@suse.com
State New
Headers show

Commit Message

Gang He May 18, 2017, 6:35 a.m. UTC
This patch is used to add an obvious error message, due to
dismatch cluster names between on-disk and in the current cluster.
We can meet this case during OCFS2 cluster migration, if we can
give the user an obvious tip for why they can not mount the file
system after migration, they can quickly fix this dismatch problem.
Second, also move printing ocfs2_fill_super() errno to the front
of ocfs2_dismount_volume() function, since ocfs2_dismount_volume()
will also print it's own message.

Signed-off-by: Gang He <ghe@suse.com>
---
 fs/ocfs2/super.c | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)

Comments

Joseph Qi May 18, 2017, 9:42 a.m. UTC | #1
Hi Gang,

How can we confirm EBADR is only because cluster name mismatch?
Since the cluster stack may be o2cb(o2dlm) or user(fsdlm).

Thanks,
Joseph

On 17/5/18 14:35, Gang He wrote:
> This patch is used to add an obvious error message, due to
> dismatch cluster names between on-disk and in the current cluster.
> We can meet this case during OCFS2 cluster migration, if we can
> give the user an obvious tip for why they can not mount the file
> system after migration, they can quickly fix this dismatch problem.
> Second, also move printing ocfs2_fill_super() errno to the front
> of ocfs2_dismount_volume() function, since ocfs2_dismount_volume()
> will also print it's own message.
> 
> Signed-off-by: Gang He <ghe@suse.com>
> ---
>  fs/ocfs2/super.c | 8 ++++++--
>  1 file changed, 6 insertions(+), 2 deletions(-)
> 
> diff --git a/fs/ocfs2/super.c b/fs/ocfs2/super.c
> index ca1646f..5575918 100644
> --- a/fs/ocfs2/super.c
> +++ b/fs/ocfs2/super.c
> @@ -1208,14 +1208,15 @@ static int ocfs2_fill_super(struct super_block *sb, void *data, int silent)
>  read_super_error:
>  	brelse(bh);
>  
> +	if (status)
> +		mlog_errno(status);
> +
>  	if (osb) {
>  		atomic_set(&osb->vol_state, VOLUME_DISABLED);
>  		wake_up(&osb->osb_mount_event);
>  		ocfs2_dismount_volume(sb, 1);
>  	}
>  
> -	if (status)
> -		mlog_errno(status);
>  	return status;
>  }
>  
> @@ -1843,6 +1844,9 @@ static int ocfs2_mount_volume(struct super_block *sb)
>  	status = ocfs2_dlm_init(osb);
>  	if (status < 0) {
>  		mlog_errno(status);
> +		if (status == -EBADR)
> +			mlog(ML_ERROR, "couldn't mount because cluster name on"
> +			" disk does not match the running cluster name.\n");
>  		goto leave;
>  	}
>  
>
Gang He May 18, 2017, 10:43 a.m. UTC | #2
Hi Joseph,


>>> 
> Hi Gang,
> 
> How can we confirm EBADR is only because cluster name mismatch?
> Since the cluster stack may be o2cb(o2dlm) or user(fsdlm).
I looked through all the code of OCFS2 (include o2cb), there is not any place which returns this error.
In fact, the function calling patch ocfs2_fill_super -> ocfs2_mount_volume -> ocfs2_dlm_init -> dlm_new_lockspace
is very specific path, we can use this errorno to give the uses a more clear tip, 
since this case looks like a little common during cluster migration, but the customer can quickly
get the failure cause if there is a error printing.
Also, I think there is not possible to add this errorno in o2cb path during ocfs2_dlm_init, since o2cb code has been stable for 
a long time.   

Thanks
Gang

> 
> Thanks,
> Joseph
> 
> On 17/5/18 14:35, Gang He wrote:
>> This patch is used to add an obvious error message, due to
>> dismatch cluster names between on-disk and in the current cluster.
>> We can meet this case during OCFS2 cluster migration, if we can
>> give the user an obvious tip for why they can not mount the file
>> system after migration, they can quickly fix this dismatch problem.
>> Second, also move printing ocfs2_fill_super() errno to the front
>> of ocfs2_dismount_volume() function, since ocfs2_dismount_volume()
>> will also print it's own message.
>> 
>> Signed-off-by: Gang He <ghe@suse.com>
>> ---
>>  fs/ocfs2/super.c | 8 ++++++--
>>  1 file changed, 6 insertions(+), 2 deletions(-)
>> 
>> diff --git a/fs/ocfs2/super.c b/fs/ocfs2/super.c
>> index ca1646f..5575918 100644
>> --- a/fs/ocfs2/super.c
>> +++ b/fs/ocfs2/super.c
>> @@ -1208,14 +1208,15 @@ static int ocfs2_fill_super(struct super_block *sb, 
> void *data, int silent)
>>  read_super_error:
>>  	brelse(bh);
>>  
>> +	if (status)
>> +		mlog_errno(status);
>> +
>>  	if (osb) {
>>  		atomic_set(&osb->vol_state, VOLUME_DISABLED);
>>  		wake_up(&osb->osb_mount_event);
>>  		ocfs2_dismount_volume(sb, 1);
>>  	}
>>  
>> -	if (status)
>> -		mlog_errno(status);
>>  	return status;
>>  }
>>  
>> @@ -1843,6 +1844,9 @@ static int ocfs2_mount_volume(struct super_block *sb)
>>  	status = ocfs2_dlm_init(osb);
>>  	if (status < 0) {
>>  		mlog_errno(status);
>> +		if (status == -EBADR)
>> +			mlog(ML_ERROR, "couldn't mount because cluster name on"
>> +			" disk does not match the running cluster name.\n");
>>  		goto leave;
>>  	}
>>  
>>
Joseph Qi May 19, 2017, 12:46 a.m. UTC | #3
Hi Gang,

As you described, only fsdlm will return this error and fsdlm has
already print the same message. So why should we add it outside again?

Thanks,
Joseph

On 17/5/18 18:43, Gang He wrote:
> Hi Joseph,
> 
> 
>>>>
>> Hi Gang,
>>
>> How can we confirm EBADR is only because cluster name mismatch?
>> Since the cluster stack may be o2cb(o2dlm) or user(fsdlm).
> I looked through all the code of OCFS2 (include o2cb), there is not any place which returns this error.
> In fact, the function calling patch ocfs2_fill_super -> ocfs2_mount_volume -> ocfs2_dlm_init -> dlm_new_lockspace
> is very specific path, we can use this errorno to give the uses a more clear tip, 
> since this case looks like a little common during cluster migration, but the customer can quickly
> get the failure cause if there is a error printing.
> Also, I think there is not possible to add this errorno in o2cb path during ocfs2_dlm_init, since o2cb code has been stable for 
> a long time.   
> 
> Thanks
> Gang
> 
>>
>> Thanks,
>> Joseph
>>
>> On 17/5/18 14:35, Gang He wrote:
>>> This patch is used to add an obvious error message, due to
>>> dismatch cluster names between on-disk and in the current cluster.
>>> We can meet this case during OCFS2 cluster migration, if we can
>>> give the user an obvious tip for why they can not mount the file
>>> system after migration, they can quickly fix this dismatch problem.
>>> Second, also move printing ocfs2_fill_super() errno to the front
>>> of ocfs2_dismount_volume() function, since ocfs2_dismount_volume()
>>> will also print it's own message.
>>>
>>> Signed-off-by: Gang He <ghe@suse.com>
>>> ---
>>>  fs/ocfs2/super.c | 8 ++++++--
>>>  1 file changed, 6 insertions(+), 2 deletions(-)
>>>
>>> diff --git a/fs/ocfs2/super.c b/fs/ocfs2/super.c
>>> index ca1646f..5575918 100644
>>> --- a/fs/ocfs2/super.c
>>> +++ b/fs/ocfs2/super.c
>>> @@ -1208,14 +1208,15 @@ static int ocfs2_fill_super(struct super_block *sb, 
>> void *data, int silent)
>>>  read_super_error:
>>>  	brelse(bh);
>>>  
>>> +	if (status)
>>> +		mlog_errno(status);
>>> +
>>>  	if (osb) {
>>>  		atomic_set(&osb->vol_state, VOLUME_DISABLED);
>>>  		wake_up(&osb->osb_mount_event);
>>>  		ocfs2_dismount_volume(sb, 1);
>>>  	}
>>>  
>>> -	if (status)
>>> -		mlog_errno(status);
>>>  	return status;
>>>  }
>>>  
>>> @@ -1843,6 +1844,9 @@ static int ocfs2_mount_volume(struct super_block *sb)
>>>  	status = ocfs2_dlm_init(osb);
>>>  	if (status < 0) {
>>>  		mlog_errno(status);
>>> +		if (status == -EBADR)
>>> +			mlog(ML_ERROR, "couldn't mount because cluster name on"
>>> +			" disk does not match the running cluster name.\n");
>>>  		goto leave;
>>>  	}
>>>  
>>>
>
Gang He May 19, 2017, 2:15 a.m. UTC | #4
Hello Joseph,


>>> 
> Hi Gang,
> 
> As you described, only fsdlm will return this error and fsdlm has
> already print the same message. So why should we add it outside again?
Yes, DLM kernel module has printed a message in this case, likes "dlm: dlm cluster name XXX mismatch YYY",
then return error to ocfs2 layer, but the user can not understand the error meanings from this message, usually ignore this tip.
then, I submitted a new printing in DLM kernel module likes "dlm cluster name '%s' does not match the application cluster name '%s'",
it maybe can help the user a little, but the user still can not find the real cause in file system layer easily, 
since DLM kernel module can be used by any upper applications (not only for ocfs2).
So, that is why I want to add a obvious error message in file system layer, 
in this layer we can tell the user the reason is there is a cluster name dismatch between on disk and on the running cluster environment.

Thanks
Gang

> 
> Thanks,
> Joseph
> 
> On 17/5/18 18:43, Gang He wrote:
>> Hi Joseph,
>> 
>> 
>>>>>
>>> Hi Gang,
>>>
>>> How can we confirm EBADR is only because cluster name mismatch?
>>> Since the cluster stack may be o2cb(o2dlm) or user(fsdlm).
>> I looked through all the code of OCFS2 (include o2cb), there is not any 
> place which returns this error.
>> In fact, the function calling patch ocfs2_fill_super -> ocfs2_mount_volume -> 
> ocfs2_dlm_init -> dlm_new_lockspace
>> is very specific path, we can use this errorno to give the uses a more clear 
> tip, 
>> since this case looks like a little common during cluster migration, but the 
> customer can quickly
>> get the failure cause if there is a error printing.
>> Also, I think there is not possible to add this errorno in o2cb path during 
> ocfs2_dlm_init, since o2cb code has been stable for 
>> a long time.   
>> 
>> Thanks
>> Gang
>> 
>>>
>>> Thanks,
>>> Joseph
>>>
>>> On 17/5/18 14:35, Gang He wrote:
>>>> This patch is used to add an obvious error message, due to
>>>> dismatch cluster names between on-disk and in the current cluster.
>>>> We can meet this case during OCFS2 cluster migration, if we can
>>>> give the user an obvious tip for why they can not mount the file
>>>> system after migration, they can quickly fix this dismatch problem.
>>>> Second, also move printing ocfs2_fill_super() errno to the front
>>>> of ocfs2_dismount_volume() function, since ocfs2_dismount_volume()
>>>> will also print it's own message.
>>>>
>>>> Signed-off-by: Gang He <ghe@suse.com>
>>>> ---
>>>>  fs/ocfs2/super.c | 8 ++++++--
>>>>  1 file changed, 6 insertions(+), 2 deletions(-)
>>>>
>>>> diff --git a/fs/ocfs2/super.c b/fs/ocfs2/super.c
>>>> index ca1646f..5575918 100644
>>>> --- a/fs/ocfs2/super.c
>>>> +++ b/fs/ocfs2/super.c
>>>> @@ -1208,14 +1208,15 @@ static int ocfs2_fill_super(struct super_block *sb, 
>>> void *data, int silent)
>>>>  read_super_error:
>>>>  	brelse(bh);
>>>>  
>>>> +	if (status)
>>>> +		mlog_errno(status);
>>>> +
>>>>  	if (osb) {
>>>>  		atomic_set(&osb->vol_state, VOLUME_DISABLED);
>>>>  		wake_up(&osb->osb_mount_event);
>>>>  		ocfs2_dismount_volume(sb, 1);
>>>>  	}
>>>>  
>>>> -	if (status)
>>>> -		mlog_errno(status);
>>>>  	return status;
>>>>  }
>>>>  
>>>> @@ -1843,6 +1844,9 @@ static int ocfs2_mount_volume(struct super_block *sb)
>>>>  	status = ocfs2_dlm_init(osb);
>>>>  	if (status < 0) {
>>>>  		mlog_errno(status);
>>>> +		if (status == -EBADR)
>>>> +			mlog(ML_ERROR, "couldn't mount because cluster name on"
>>>> +			" disk does not match the running cluster name.\n");
>>>>  		goto leave;
>>>>  	}
>>>>  
>>>>
>>

Patch
diff mbox

diff --git a/fs/ocfs2/super.c b/fs/ocfs2/super.c
index ca1646f..5575918 100644
--- a/fs/ocfs2/super.c
+++ b/fs/ocfs2/super.c
@@ -1208,14 +1208,15 @@  static int ocfs2_fill_super(struct super_block *sb, void *data, int silent)
 read_super_error:
 	brelse(bh);
 
+	if (status)
+		mlog_errno(status);
+
 	if (osb) {
 		atomic_set(&osb->vol_state, VOLUME_DISABLED);
 		wake_up(&osb->osb_mount_event);
 		ocfs2_dismount_volume(sb, 1);
 	}
 
-	if (status)
-		mlog_errno(status);
 	return status;
 }
 
@@ -1843,6 +1844,9 @@  static int ocfs2_mount_volume(struct super_block *sb)
 	status = ocfs2_dlm_init(osb);
 	if (status < 0) {
 		mlog_errno(status);
+		if (status == -EBADR)
+			mlog(ML_ERROR, "couldn't mount because cluster name on"
+			" disk does not match the running cluster name.\n");
 		goto leave;
 	}