Hi all, Sorry for the late. This patch series tries to fix up some regressions after we try to use extent status tree as a cache, and give it some improvements. The first commit tries to fix a e4defrag regression which is reported by Dmitry. As Dmitry reported after applied this patch there still has a failure in Dmitry's xfstests #300. The second commit improves ext4_es_can_be_merged function to avoid a potential overflow. The third commit adds a self-testing infrastructure for extent status tree to do a sanity check. After as a extent cache we need to make sure that extent status tree caches right things. After applied this patch, we will get a lot of messages when we run xfstests, e.g. #83, #91, ... Please define ES_AGGRESSIVE_TEST to enable it. The fourth commit lets ext4_ext_handle_uninitialized_extents return right length of converted extent. After applied this patch, some messages that reports the return value is not equal to map->m_len disappears. The fifth commit fixes a bug that will causes an initialized extent is marked as uninitialized when we try to split an extent. This patch is very important for the later one because we will not eliminate all error messages if this patch is missing. The last commit updates extent status tree after we zeroout an extent. Meanwhile we will not update it in ext4_map_blocks. Otherwise we will track an written extent as unwritten. It might cause a potential bug. After applied this patch, we will never get any messages from self- testing infrastructure except that xfstest #223 when dioread_nolock enables. We don't need to worry about it because when we lookup a block mapping from status tree i_data_sem hasn't been taken. So there is a race window that an unwritten extent is converted by end_io when we compare the result between extent tree and status tree. I have tried to take i_data_sem locking and run #223 again, and the message doesn't be printed. After applied the patch series, I run xfstests with the following configurations because it is easy to trigger a unwritten->written conversion. * EXT4 4k Block * EXT4 4k Block with dioread_nolock * EXT4 4k Block with bigalloc And I don't get any message from self-testing infrastructure except #223 as said before. I will go on testing other configurations. Hope I don't screw up something. The patch series also needs to do more tests, though. I decide to send it out now to let others review because we still have a pending regression. I want to know whether this patch series can it or not. Dmitry, I still couldn't reproduce the failure of #300 in my desktop. I change another machine with a HDD to try to reproduce it, but I failed. Could you please test this patch series to confirm this regression can be fixed or not. I really appreicate for your help. If you have any update please let me know. This patch series is against the latest dev branch of ext4. As always, any feedback is appreicated. Thanks!!! - Zheng Dmitry Monakhov (3): ext4: invalidate exntent-status-tree during extent_migration ext4: add self-testing infrastructure to do a sanity check ext4: ext4_split_extent shoult take care about extent zeroout v3 Zheng Liu (3): ext4: improve ext4_es_can_be_merged() to avoid a potential overflow ext4: fix wrong m_len value after unwritten extent conversion ext4: update extent status tree after an extent is zeroed out fs/ext4/extents.c | 53 ++++++++++++++++++++---- fs/ext4/extents_status.c | 104 +++++++++++++++++++++++++++++++++++++++++++--- fs/ext4/extents_status.h | 9 ++++ fs/ext4/inode.c | 105 +++++++++++++++++++++++++++++++++++++++++++++++ fs/ext4/move_extent.c | 8 ++++ 5 files changed, 266 insertions(+), 13 deletions(-) -- 1.7.12.rc2.18.g61b472e -- To unsubscribe from this list: send the line "unsubscribe linux-ext4" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
