[LU-8265] one possible ldiskfs deadlock for rhel7

Thu, 23 Feb 2017 07:35:51 +0000

This bug was originally motivated by porting project quota support to ldiskfs:

https://testing.hpdd.intel.com/test_logs/5de868b0-2377-11e6-a8f9-5254006e85c2/show_text
ext4_setattr
 down_read(&EXT4_I(inode)->i_data_sem);
 ext4_mark_inode_dirty
  ext4_expand_extra_isize
   ext4_expand_extra_isize_ea
    ext4_xattr_block_set
     down_read(&EXT4_I(inode)->i_data_sem);--->deadlock here.
     ext4_new_meta_blocks

This problem happen when we trying to upgrade ldiskfs to support project
quota which will expand ext4 inode size.

I think this problem might also happen when we try to upgrade Lustre from
rhel6 to rhel7 in the future, so we better fix it.

I also opened a redhat ticket for this.

Whamcloud Community JIRA

[LU-8265] one possible ldiskfs deadlock for rhel7