[LU-14267] osd_ldiskfs_write_record(): do not update in-bh inode every time

• Type: Bug
• Resolution: Fixed
• Priority: Critical
• Fix Version/s: Lustre 2.14.0
• Affects Version/s: Lustre 2.14.0, Lustre 2.12.6
• Labels: LTS12
• Severity: 3
• Assignee: Alex Zhuravlev
• Reporter: Alex Zhuravlev

Description

there is no point to update in-bh inode every call to osd_ldiskfs_write_record(), but
the following construction makes it so:

        dirty_inode = test_and_set_bit(LDISKFS_INODE_JOURNAL_DATA,
                                       &ei->i_flags);
....
        if (dirty_inode)
                osd_dirty_inode(inode, I_DIRTY_DATASYNC);

there are two consequences:
1) transaction can exceed declared credits as osd_declare_write() doesn't reserve inode for overwrite case (then current operation aborts with -ENOSPC and filesystem turns read-only)
2) overwrites copy inode into buffer cache impacting performance, should be especially visible with changelogs enabled, where every record causes at least two separate writes

Issue Links

• is related to: LU-13416 Data corruption during IOR testing with DoM files and hard failover