Hi!
I think I might have caught the issue:
diff --git a/fs/xfs/xfs_super.c b/fs/xfs/xfs_super.c
index e5e0713bebcd8..9231457371100 100644
--- a/fs/xfs/xfs_super.c
+++ b/fs/xfs/xfs_super.c
@@ -1937,17 +1937,17 @@ xfs_init_zones(void)
goto out_destroy_trans_zone;
xfs_efd_zone = kmem_cache_create("xfs_efd_item",
- (sizeof(struct xfs_efd_log_item) +
- (XFS_EFD_MAX_FAST_EXTENTS - 1) *
- sizeof(struct xfs_extent)),
- 0, 0, NULL);
+ struct_size((struct xfs_efd_log_item *)0,
+ efd_format.efd_extents,
+ XFS_EFD_MAX_FAST_EXTENTS),
+ 0, 0, NULL);
if (!xfs_efd_zone)
goto out_destroy_buf_item_zone;
xfs_efi_zone = kmem_cache_create("xfs_efi_item",
- (sizeof(struct xfs_efi_log_item) +
- (XFS_EFI_MAX_FAST_EXTENTS - 1) *
- sizeof(struct xfs_extent)),
+ struct_size((struct xfs_efi_log_item *)0,
+ efi_format.efi_extents,
+ XFS_EFI_MAX_FAST_EXTENTS),
0, 0, NULL);
if (!xfs_efi_zone)
goto out_destroy_efd_zone;
I'm currently testing the patch with the fix above:
https://git.kernel.org/pub/scm/linux/kernel/git/gustavoars/linux.git/commit/?h=testing/xfs-fixed
--
Gustavo
On 3/20/21 16:48, Darrick J. Wong wrote:
> On Sat, Mar 20, 2021 at 03:20:55PM -0500, Gustavo A. R. Silva wrote:
>>
>>
>> On 3/20/21 15:17, Darrick J. Wong wrote:
>>>>>> Below are the results of running xfstests for groups shutdown and log
>>>>>> with the following configuration in local.config:
>>>>>>
>>>>>> export TEST_DEV=/dev/sda3
>>>>>> export TEST_DIR=/mnt/test
>>>>>> export SCRATCH_DEV=/dev/sda4
>>>>>> export SCRATCH_MNT=/mnt/scratch
>>>>>>
>>>>>> The size for both partitions /dev/sda3 and /dev/sda4 is 25GB.
>>>>>
>>>>> Looks good to me, will toss it at my fstests cloud and see if anything
>>>>> shakes out. Thanks for cleaning up this goofy thorn-pile!
>>>>
>>>> Great. It's been fun to work on this. :p
>>>
>>> Did you run the /entire/ fstests suite? With this patch applied to
>>> 5.12-rc2, I keep seeing list corruption assertions about an hour into
>>
>> Nope; I run xfstests 'shutdown' and 'log' groups on 5.11.0, only.
>>
>> How do you run the entire fstests?
>> Could you give me some pointers?
>
> ./check -g all
>
> (instead of "./check -g shutdown")
>
>>> the test run, and usually on some test that heavily exercises allocating
>>> and deleting file extents. I'll try to look at this patch more closely
>>> next week, but I figured I should let you know early, on the off chance
>>> something sticks out to you.
>>
>> OK. I'll go run my tests on 5.12-rc2.
>>
>> Should I run the entire xfstests, too?
>
> Yes, please.
>
> --D
>
>> Thanks
>> --
>> Gustavo
