Hi Amir,
How about this as a set of patches to do what you suggest[1] and hoist the
handler functions for I_OVL_INUSE into common code and rename the flag to
I_EXCL_INUSE. This can then be shared with cachefiles - allowing me to get
rid of S_KERNEL_FILE.
I did split out the functionality for preventing file/dir removal to a
separate flag, I_NO_REMOVE, so that it's not tied to I_EXCL_INUSE in case
overlayfs doesn't want to use it. The downside to that, though is that it
requires a separate locking of i_lock to set/clear it.
I also added four general tracepoints to log successful lock/unlock,
failure to lock and a bad unlock. The lock tracepoints log which driver
asked for the lock and all tracepoints allow the driver to log an arbitrary
reference number (in cachefiles's case this is the object debug ID).
Questions:
(1) Should it be using a flag in i_state or a flag in i_flags? I'm not
sure what the difference is really.
(2) Do we really need to take i_lock when testing I_EXCL_INUSE? Would
READ_ONCE() suffice?
The patches are on a branch here:
https://git.kernel.org/pub/scm/linux/kernel/git/dhowells/linux-fs.git/log/?h=fscache-fixes
David
Link: https://lore.kernel.org/r/CAOQ4uxhRS3MGEnCUDcsB1RL0d1Oy0g0Rzm75hVFAJw2dJ7uKSA@xxxxxxxxxxxxxx/ [1]
---
David Howells (5):
vfs, overlayfs, cachefiles: Turn I_OVL_INUSE into something generic
vfs: Add tracepoints for inode_excl_inuse_trylock/unlock
cachefiles: Split removal-prevention from S_KERNEL_FILE and extend effects
cachefiles: Use I_EXCL_INUSE instead of S_KERNEL_FILE
cachefiles: Remove the now-unused mark-inode-in-use tracepoints
fs/cachefiles/namei.c | 54 +++++++++++++-------------
fs/inode.c | 56 +++++++++++++++++++++++++++
fs/namei.c | 8 ++--
fs/overlayfs/overlayfs.h | 3 --
fs/overlayfs/super.c | 14 ++++---
fs/overlayfs/util.c | 43 ---------------------
include/linux/fs.h | 33 ++++++++++++++--
include/trace/events/cachefiles.h | 63 -------------------------------
8 files changed, 126 insertions(+), 148 deletions(-)
