Hi all,
This is the second part of correcting XFS to reload the incore unlinked
inode list from the ondisk contents. Whereas part one tackled failures
from regular filesystem calls, this part takes on the problem of needing
to reload the entire incore unlinked inode list on account of somebody
loading an inode that's in the /middle/ of an unlinked list. This
happens during quotacheck, bulkstat, or even opening a file by handle.
In this case we don't know the length of the list that we're reloading,
so we don't want to create a new unbounded memory load while holding
resources locked. Instead, we'll target UNTRUSTED iget calls to reload
the entire bucket.
Note that this changes the definition of the incore unlinked inode list
slightly -- i_prev_unlinked == 0 now means "not on the incore list".
v2: rebase to for-next, resend without changes
v3: add necessary prerequisites
v4: fix accidental commit to wrong patch
If you're going to start using this code, I strongly recommend pulling
from my git trees, which are linked below.
This has been running on the djcloud for months with no problems. Enjoy!
Comments and questions are, as always, welcome.
--D
kernel git tree:
https://git.kernel.org/cgit/linux/kernel/git/djwong/xfs-linux.git/log/?h=fix-iunlink-list
fstests git tree:
https://git.kernel.org/cgit/linux/kernel/git/djwong/xfstests-dev.git/log/?h=fix-iunlink-list
---
common/fuzzy | 4 -
common/rc | 36 ++++++++-
tests/xfs/1872 | 111 +++++++++++++++++++++++++++
tests/xfs/1872.out | 5 +
tests/xfs/1873 | 215 ++++++++++++++++++++++++++++++++++++++++++++++++++++
tests/xfs/1873.out | 6 +
tests/xfs/329 | 4 -
tests/xfs/434 | 2
tests/xfs/435 | 2
tests/xfs/436 | 2
tests/xfs/444 | 2
tests/xfs/516 | 2
12 files changed, 381 insertions(+), 10 deletions(-)
create mode 100755 tests/xfs/1872
create mode 100644 tests/xfs/1872.out
create mode 100755 tests/xfs/1873
create mode 100644 tests/xfs/1873.out