v3: use parent inode information more widely when we have it
v2: use r_locked_dir in __choose_mds if it's available
handle NULL d_inode_rcu(parent) properly
In several places, the kcephfs client accesses dentry->d_parent without
holding appropriate locking. This is not safe, as you can race with
renames that can morph the tree such that is changes and dentries found
previously can end up freed. This patchset fixes these places to access
the parent(s) safely, mostly using the rcu_read_lock().
Note that I'm not aware of any specific bug reports in this area --
these were just discovered by inspection (mostly by Zheng). This passes
xfstests at least up to test generic/095, where we hit an unrelated
softlockup in the splice write code (http://tracker.ceph.com/issues/18130).
Tracker bug link: http://tracker.ceph.com/issues/18148
Jeff Layton (5):
ceph: clean up unsafe d_parent access in __choose_mds
ceph: clean up unsafe d_parent accesses in build_dentry_path
ceph: pass parent dir ino info to build_dentry_path
ceph: fix unsafe dcache access in ceph_encode_dentry_release
ceph: pass parent inode info to ceph_encode_dentry_release if we have
it
fs/ceph/caps.c | 8 ++++-
fs/ceph/mds_client.c | 94 +++++++++++++++++++++++++++++++++++-----------------
fs/ceph/super.h | 1 +
3 files changed, 71 insertions(+), 32 deletions(-)
--
2.7.4
--
To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
