Hi,
This patch set fix a memory corruption problem caused by read stale
extent block from disk in ext4_ext_split()->memset(). The root cause is
we do not clear buffer's verified bit before read metadata block from
disk again if it has been failed to write out to disk, if the block is
mew allocated, we may propably get stale data from disk and lead to
out-of-bounds access when we use this stale data.
The first patch is the same to my v1 iteration, just clear buffer
verified bit before read, this patch can fix this problem.
The remaining patches remove all open codes that read metadata blocks
in ext4 and introduce common read helpers as Jan suggested. I have test
them on my xfstests and there is no degeneration.
Thanks,
Yi.
zhangyi (F) (7):
ext4: clear buffer verified flag if read meta block from disk
ext4: introduce new metadata buffer read helpers
ext4: use common helpers in all places reading metadata buffers
ext4: use ext4_buffer_uptodate() in __ext4_get_inode_loc()
ext4: introduce ext4_sb_breadahead_unmovable() to replace
sb_breadahead_unmovable()
ext4: use ext4_sb_bread() instead of sb_bread()
ext4: introduce ext4_sb_bread_unmovable() to replace
sb_bread_unmovable()
fs/ext4/balloc.c | 7 +--
fs/ext4/ext4.h | 8 +++
fs/ext4/extents.c | 2 +-
fs/ext4/ialloc.c | 5 +-
fs/ext4/indirect.c | 8 +--
fs/ext4/inode.c | 43 +++++---------
fs/ext4/mmp.c | 10 +---
fs/ext4/move_extent.c | 2 +-
fs/ext4/resize.c | 10 ++--
fs/ext4/super.c | 131 ++++++++++++++++++++++++++++++++++++------
10 files changed, 153 insertions(+), 73 deletions(-)
--
2.25.4
