[LU-216] Fix fiemap locking issue in 2.6.32 kernels

• Type: Bug
• Resolution: Fixed
• Priority: Critical
• Fix Version/s: Lustre 2.2.0, Lustre 1.8.6
• Affects Version/s: Lustre 2.1.0, Lustre 1.8.6
• Severity: 3
• Bugzilla ID: 23,780
• Assignee: Zhenyu Xu
• Reporter: Cory Spitz

Description

The patch from Lustre Bugzilla 23780 attachment 31890 was added for RHEL6 support with git commit e550abd05cf7ffddaedbef996be1baaae0912b4a, but a better long term fix is needed.

Johann L. writes in Lustre Bugzilla 23780 comment #28:
> The drawback of this patch is a potential race with fiemap (i did not revert
> the git kernel commit fab3a549e204172236779f502eccb4f9bf0dc87d totally).

and follows up in Lustre Bugzilla 23780 comment #34:
> Please note that this patch is a short term solution.
> Girish, it would be great if you could come up with another patch which takes
> i_data_sem write reference where it is needed in ext4_walk_space().

b1_8 also still requires the fix for SLES 11 SP1.

---

Info required for matching: sanity-benchmark test_fsx fsx

Comments

[Andrew Perepechko]

Zhenyu Xu, I was referring to your patch.

> static int ext3_ext_new_extent_cb()
>...
> if (bp->create == 0)

{ >... > return EXT_CONTINUE; > }

>
> down_write((&EXT4_I(inode)->i_data_sem));
>
> /* validate extent, make sure the extent tree does not changed */
> tmppath = ext4_ext_find_extent(inode, cex->ec_block, NULL);
> if (IS_ERR(tmppath))

{ >... > }
> tmpex = tmppath[depth].p_ext;
> if (tmpex != ex) {>...> }

This patch seems to be expecting that the extent map may have changed during write (which I called allocation race earlier) . However, obdfilter takes i_mutex on writes and truncates and serializes extent map updates with it.

Which case/race is expected to be handled with your code that I quoted above?

Thanks.

[Zhenyu Xu]

Panda,

I don't quite get you, i_mutex is always hold during the map block process, it is i_data_sem that be used to protect extent block usage. I don't know what "allocation race" you are referring to, would you mind elaborating it?

[Andrew Perepechko]

Zhenyu Xu, why isn't a hypothetical allocation race in ext3_ext_new_extent_cb() addressed by your patch avoided by means of i_mutex which is taken whenever Lustre calls fsfilt_map_inode_pages(create=1)?

Thanks.

[Alex Zhuravlev]

I tend to think ldiskfs_ext_next_allocated_block() should be called under i_data_sem together with ldiskfs_ext_find_extent(), otherwise ldiskfs_ext_next_allocated_block() is working on data being modified

[Build Master (Inactive)]

Integrated in lustre-master » x86_64,server,el5,ofa #146
LU-216 Protect extent tree during fsfilt_ldiskfs_ext_walk_space()

Oleg Drokin : dea0c1f27c0eb53d4738a70241ff2bcc47201862
Files :

• ldiskfs/kernel_patches/patches/ext4-misc-sles11.patch
• lustre/lvfs/fsfilt_ext3.c
• ldiskfs/kernel_patches/patches/ext4-misc-rhel6.patch
• ldiskfs/kernel_patches/patches/ext4-misc-rhel5.patch

[Build Master (Inactive)]

Integrated in lustre-master » i686,server,el6,inkernel #146
LU-216 Protect extent tree during fsfilt_ldiskfs_ext_walk_space()

Oleg Drokin : dea0c1f27c0eb53d4738a70241ff2bcc47201862
Files :

• ldiskfs/kernel_patches/patches/ext4-misc-sles11.patch
• lustre/lvfs/fsfilt_ext3.c
• ldiskfs/kernel_patches/patches/ext4-misc-rhel5.patch
• ldiskfs/kernel_patches/patches/ext4-misc-rhel6.patch

[Build Master (Inactive)]

Integrated in lustre-master » x86_64,client,el5,ofa #146
LU-216 Protect extent tree during fsfilt_ldiskfs_ext_walk_space()

Oleg Drokin : dea0c1f27c0eb53d4738a70241ff2bcc47201862
Files :

• lustre/lvfs/fsfilt_ext3.c
• ldiskfs/kernel_patches/patches/ext4-misc-sles11.patch
• ldiskfs/kernel_patches/patches/ext4-misc-rhel6.patch
• ldiskfs/kernel_patches/patches/ext4-misc-rhel5.patch

[Build Master (Inactive)]

Integrated in lustre-master » x86_64,server,el6,inkernel #146
LU-216 Protect extent tree during fsfilt_ldiskfs_ext_walk_space()

Oleg Drokin : dea0c1f27c0eb53d4738a70241ff2bcc47201862
Files :

• ldiskfs/kernel_patches/patches/ext4-misc-sles11.patch
• lustre/lvfs/fsfilt_ext3.c
• ldiskfs/kernel_patches/patches/ext4-misc-rhel5.patch
• ldiskfs/kernel_patches/patches/ext4-misc-rhel6.patch

[Build Master (Inactive)]

Integrated in lustre-master » x86_64,server,el5,inkernel #146
LU-216 Protect extent tree during fsfilt_ldiskfs_ext_walk_space()

Oleg Drokin : dea0c1f27c0eb53d4738a70241ff2bcc47201862
Files :

• ldiskfs/kernel_patches/patches/ext4-misc-sles11.patch
• ldiskfs/kernel_patches/patches/ext4-misc-rhel6.patch
• ldiskfs/kernel_patches/patches/ext4-misc-rhel5.patch
• lustre/lvfs/fsfilt_ext3.c

[Build Master (Inactive)]

Integrated in lustre-master » i686,server,el5,ofa #146
LU-216 Protect extent tree during fsfilt_ldiskfs_ext_walk_space()

Oleg Drokin : dea0c1f27c0eb53d4738a70241ff2bcc47201862
Files :

• ldiskfs/kernel_patches/patches/ext4-misc-sles11.patch
• lustre/lvfs/fsfilt_ext3.c
• ldiskfs/kernel_patches/patches/ext4-misc-rhel5.patch
• ldiskfs/kernel_patches/patches/ext4-misc-rhel6.patch

[Build Master (Inactive)]

Integrated in lustre-master » x86_64,client,ubuntu1004,ofa #146
LU-216 Protect extent tree during fsfilt_ldiskfs_ext_walk_space()

Oleg Drokin : dea0c1f27c0eb53d4738a70241ff2bcc47201862
Files :

• ldiskfs/kernel_patches/patches/ext4-misc-sles11.patch
• lustre/lvfs/fsfilt_ext3.c
• ldiskfs/kernel_patches/patches/ext4-misc-rhel5.patch
• ldiskfs/kernel_patches/patches/ext4-misc-rhel6.patch