Hi folks,
This is a followup to the original RFC patchset posted here:
http://oss.sgi.com/archives/xfs/2016-01/msg00366.html
This patchset actually works, and passes through xfstests without
any regressions at all. So feel free to test and comment on this
version.
-Dave.
Version 2
- timestamps are signed, so need casting to int before being stored
in a 64 bit variable when read from disk. Fixes generic/258
failure (patch 2).
- nlink wasn't being accounted correctly for O_TMPFILE files and
rename whiteouts. Fixes generic/078 and others. (patch 5)
- VFS does not initialise i_mode, ever, so we need to initialise it
directly after slab allocation to ensure that we correctly detect
new file creates and hence don't add the inode the sb list twice,
corrupting it. (busy inodes after unmount, self destruct in 5
seconds, have a nice day!) (patch 8)
- when reallocating a reclaimable inode, we need to save the
variables that hold on-disk values across the inode_init_always()
call, otherwise we lose them. This results in corruption of these
fields. (patch 8)
|