Re: [PATCH] ext4: Do not destroy ext4_groupinfo_caches if ext4_mb_init() fails

[Andrey Tsyvarev <tsyvarev@ispras.ru>]

12.05.2014 19:08, Lukáš Czerner пишет:
> On Mon, 12 May 2014, Andrey Tsyvarev wrote:
>
>> Date: Mon, 12 May 2014 12:23:59 +0400
>> From: Andrey Tsyvarev<tsyvarev@ispras.ru>
>> To: Theodore Ts'o<tytso@mit.edu>
>> Cc: Andrey Tsyvarev<tsyvarev@ispras.ru>,
>>      Andreas Dilger<adilger.kernel@dilger.ca>,linux-ext4@vger.kernel.org,
>>      linux-kernel@vger.kernel.org, Alexey Khoroshilov<khoroshilov@ispras.ru>
>> Subject: [PATCH] ext4: Do not destroy ext4_groupinfo_caches if ext4_mb_init()
>>      fails
>>
>> Caches from 'ext4_groupinfo_caches' may be in use by other mounts, which have already existed.
>> So, it is incorrect to destroy them when newly requested mount fails.
>>
>> Found by Linux File System Verification project (linuxtesting.org).
> Makes sense, thanks! Can you please share the test case which
> triggered this ? It might be worth including in xfstests.

Actually it was triggered by xfstests themselves but run with fault simulation.
The method of fault simulation is under development/evaluation now, we expect to publish a paper describing it in the near future.

BUG_ON() in get_groupinfo_cache() was firstly triggered by test generic/003, but actually it could be any other test, which uses a scratch device: xftests itself requires test device(TEST_DEV) mounted, so a fault simulated while mount scratch device causes the problem described.


> Reviewed-by: Lukas Czerner<lczerner@redhat.com>
>
>> Signed-off-by: Andrey Tsyvarev<tsyvarev@ispras.ru>
>> ---
>>   fs/ext4/mballoc.c | 4 +---
>>   1 file changed, 1 insertion(+), 3 deletions(-)
>>
>> diff --git a/fs/ext4/mballoc.c b/fs/ext4/mballoc.c
>> index 04a5c75..becea1d 100644
>> --- a/fs/ext4/mballoc.c
>> +++ b/fs/ext4/mballoc.c
>> @@ -2607,7 +2607,7 @@ int ext4_mb_init(struct super_block *sb)
>>   	sbi->s_locality_groups = alloc_percpu(struct ext4_locality_group);
>>   	if (sbi->s_locality_groups == NULL) {
>>   		ret = -ENOMEM;
>> -		goto out_free_groupinfo_slab;
>> +		goto out;
>>   	}
>>   	for_each_possible_cpu(i) {
>>   		struct ext4_locality_group *lg;
>> @@ -2632,8 +2632,6 @@ int ext4_mb_init(struct super_block *sb)
>>   out_free_locality_groups:
>>   	free_percpu(sbi->s_locality_groups);
>>   	sbi->s_locality_groups = NULL;
>> -out_free_groupinfo_slab:
>> -	ext4_groupinfo_destroy_slabs();
>>   out:
>>   	kfree(sbi->s_mb_offsets);
>>   	sbi->s_mb_offsets = NULL;
>>

--
Andrey Tsyvarev
Linux Verification Center, ISPRAS