Skip to contentSkip to conversation

vfs.git - dcache and mountpoint stuff

GIT PULL Merged Al Viro 2 participants 2 messages Jul 20, 2019 VFS

lore
Share & tools
lore

Al Viro

author maintainer
#0 lore raw

Saner handling of refcounts to mountpoints.
Transfer the counting reference from struct mount ->mnt_mountpoint over to struct mountpoint ->m_dentry. That allows to get rid of the convoluted games with ordering of mount shutdowns. The cost is in teaching shrink_dcache_{parent,for_umount} to cope with mixed-filesystem shrink lists, which we'll also need for the Slab Movable Objects patchset.

	I'm not sure what's the best way to do that pull request;

there are two branches with identical (modulo one space in comment) contents. The former (#work.dcache) sat in -next; the latter (#work.dcache2) has the last commit of the former split and folded. Said last commit is basically what you'd asked to change - comment for locking of ex_mountpoints, separating the default variant of put_namespace() from "save to this list" one, having the former explicitly pass &ex_mountpoints to the latter, comment updates.

Conservative approach would be to pull #work.dcache, but...
consider e.g. put_mountpoint() changes in work.dcache: in the middle of the series we have
-static void put_mountpoint(struct mountpoint *mp)
+static void put_mountpoint(struct mountpoint *mp, struct list_head *list)
...
+ if (!list)
+ list = &ex_mountpoints;
+ dput_to_list(dentry, list);
with corresponding callers' updates (all but one passing NULL). In this last commit it becomes
static void __put_mountpoint(struct mountpoint *mp, struct list_head *list) with if (!list) part gone and
+static void put_mountpoint(struct mountpoint *mp)
+{
+ __put_mountpoint(mp, &ex_mountpoints);
+}
with callers reverted to the original state. IOW, pointless noise. And
-static LIST_HEAD(ex_mountpoints);
+static LIST_HEAD(ex_mountpoints); /* protected by namespace_sem */ is better folded into the into the commit introducing the list.

IOW, if you are OK with pulling #work.dcache2, it would, IMO, be
better. Below is the pull request for it; the variant for #work.dcache would differ in having 9 commits instead of 8. Up to you...

The following changes since commit 570d7a98e7d6d5d8706d94ffd2d40adeaa318332:

vfs: move_mount: reject moving kernel internal mounts (2019-07-01 10:46:36 -0400)

are available in the git repository at:

git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs.git work.dcache2

for you to fetch changes up to 56cbb429d911991170fe867b4bba14f0efed5829:

switch the remnants of releasing the mountpoint away from fs_pin (2019-07-16 22:52:37 -0400)

----------------------------------------------------------------
Al Viro (8):
ceph: don't open-code the check for dead lockref
nfs: dget_parent() never returns NULL
__detach_mounts(): lookup_mountpoint() can't return ERR_PTR() anymore fs/namespace.c: shift put_mountpoint() to callers of unhash_mnt() Teach shrink_dcache_parent() to cope with mixed-filesystem shrink lists make struct mountpoint bear the dentry reference to mountpoint, not struct mount get rid of detach_mnt()
switch the remnants of releasing the mountpoint away from fs_pin

 fs/ceph/dir.c          |   2 +-
 fs/dcache.c            | 100 +++++++++++++++++++++++++------
 fs/fs_pin.c            |  10 +---
 fs/internal.h          |   2 +
 fs/mount.h             |   8 ++-
 fs/namespace.c         | 159 ++++++++++++++++++++++++-------------------------
 fs/nfs/super.c         |   6 +-
 include/linux/fs_pin.h |   1 -
 8 files changed, 172 insertions(+), 116 deletions(-)