We used to need rather convoluted ordering trickery to guarantee
that dput() of ex-mountpoints happens before the final mntput()
of the same. Since we don't need that anymore, there's no point
playing with fs_pin for that.
Signed-off-by: Al Viro <[email protected]>
Lift getting the original mount (dentry is actually not needed at all)
of the mountpoint into the callers - to do_move_mount() and pivot_root()
level. That simplifies the cleanup in those and allows to get saner
arguments for attach_mnt_recursive().
Signed-off-by: Al Viro <[email protected]>
Using dput_to_list() to shift the contributing reference from ->mnt_mountpoint
to ->mnt_mp->m_dentry. Dentries are dropped (with dput_to_list()) as soon
as struct mountpoint is destroyed; in cases where we are under namespace_sem
we use the global list, shrinking it in namespace_unlock(). In case of
detaching stuck MNT_LOCKed children at final mntput_no_expire() we use a local
list and shrink it ourselves. ->mnt_ex_mountpoint crap is gone.
Signed-off-by: Al Viro <[email protected]>
Currently, running into a shrink list that contains dentries from different
filesystems can cause several unpleasant things for shrink_dcache_parent()
and for umount(2).
The first problem is that there's a window during shrink_dentry_list() between
__dentry_kill() takes a victim out and dropping reference to its parent. During
that window the parent looks like a genuine busy dentry. shrink_dcache_parent()
(or, worse yet, shrink_dcache_for_umount()) coming at that time will see no
eviction candidates and no indication that it needs to wait for some
shrink_dentry_list() to proceed further.
That applies for any shrink list that might intersect with the subtree we are
trying to shrink; the only reason it does not blow on umount(2) in the mainline
is that we unregister the memory shrinker before hitting shrink_dcache_for_umount().
Another problem happens if something in a mixed-filesystem shrink list gets
be stuck in e.g. iput(), getting umount of unrelated fs to spin waiting for
the stuck shrinker to get around to our dentries.
Solution:
1) have shrink_dentry_list() decrement the parent's refcount and
make sure it's on a shrink list (ours unless it already had been on some
other) before calling __dentry_kill(). That eliminates the window when
shrink_dcache_parent() would've blown past the entire subtree without
noticing anything with zero refcount not on shrink lists.
2) when shrink_dcache_parent() has found no eviction candidates,
but some dentries are still sitting on shrink lists, rather than
repeating the scan in hope that shrinkers have progressed, scan looking
for something on shrink lists with zero refcount. If such a thing is
found, grab rcu_read_lock() and stop the scan, with caller locking
it for eviction, dropping out of RCU and doing __dentry_kill(), with
the same treatment for parent as shrink_dentry_list() would do.
Note that right now mixed-filesystem shrink lists do not occur, so this
is not a mainline bug. Howevere, there's a bunch of uses for such
beasts (e.g. the "try and evict everything we can out of given page"
patches; there are potential uses in mount-related code, considerably
simplifying the life in fs/namespace.c, etc.)
Signed-off-by: Al Viro <[email protected]>
socket->wq is assign-once, set when we are initializing both
struct socket it's in and struct socket_wq it points to. As the
matter of fact, the only reason for separate allocation was the
ability to RCU-delay freeing of socket_wq. RCU-delaying the
freeing of socket itself gets rid of that need, so we can just
fold struct socket_wq into the end of struct socket and simplify
the life both for sock_alloc_inode() (one allocation instead of
two) and for tun/tap oddballs, where we used to embed struct socket
and struct socket_wq into the same structure (now - embedding just
the struct socket).
Note that reference to struct socket_wq in struct sock does remain
a reference - that's unchanged.
Signed-off-by: Al Viro <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
we do have an RCU-delayed part there already (freeing the wq),
so it's not like the pipe situation; moreover, it might be
worth considering coallocating wq with the rest of struct sock_alloc.
->sk_wq in struct sock would remain a pointer as it is, but
the object it normally points to would be coallocated with
struct socket...
Signed-off-by: Al Viro <[email protected]>
Signed-off-by: David S. Miller <[email protected]>
No point having two call sites (earlier in init_rootfs() from
mnt_init() in case we are going to use shmem-style rootfs,
later from do_basic_setup() unconditionally), along with the
logics in shmem_init() itself to make the second call a no-op...
Signed-off-by: Al Viro <[email protected]>
init_mount_tree() can get to rootfs_fs_type directly and that simplifies
a lot of things. We don't need to register it, we don't need to look
it up *and* we don't need to bother with preventing subsequent userland
mounts. That's the way we should've done that from the very beginning.
There is a user-visible change, namely the disappearance of "rootfs"
from /proc/filesystems. Note that it's been unmountable all along
and it didn't show up in /proc/mounts; however, it *is* a user-visible
change and theoretically some script might've been using its presence
in /proc/filesystems to tell 2.4.11+ from earlier kernels.
*IF* any complaints about behaviour change do show up, we could fake
it in /proc/filesystems. I very much doubt we'll have to, though.
Signed-off-by: Al Viro <[email protected]>
the only thing done by the latter is making ramfs visible
to mount(2); we don't need it there - rootfs is separate
and, in fact, made visible to mount(2) in the same init_rootfs().
Signed-off-by: Al Viro <[email protected]>
anon_inode_getfd() should be used *ONLY* in situations when we are
guaranteed to be past the last failure point (including copying the
descriptor number to userland, at that). And ksys_close() should
not be used for cleanups at all.
anon_inode_getfile() is there for all nontrivial cases like that.
Just use that...
Fixes: b3e5838252 ("clone: add CLONE_PIDFD")
Signed-off-by: Al Viro <[email protected]>
Reviewed-by: Jann Horn <[email protected]>
Signed-off-by: Christian Brauner <[email protected]>
... since we immediately follow that with check that it *is* an
opened perf file, with O_PATH ones ending with with the same
-EBADF we'd get for descriptor that isn't opened at all.
Signed-off-by: Al Viro <[email protected]>
Call graph of vfs_get_tree():
vfs_fsconfig_locked() # neither kernmount, nor submount
do_new_mount() # neither kernmount, nor submount
fc_mount()
afs_mntpt_do_automount() # submount
mount_one_hugetlbfs() # kernmount
pid_ns_prepare_proc() # kernmount
mq_create_mount() # kernmount
vfs_kern_mount()
simple_pin_fs() # kernmount
vfs_submount() # submount
kern_mount() # kernmount
init_mount_tree()
btrfs_mount()
nfs_do_root_mount()
The first two need the check (unconditionally).
init_mount_tree() is setting rootfs up; any capability
checks make zero sense for that one. And btrfs_mount()/
nfs_do_root_mount() have the checks already done in their
callers.
IOW, we can shift mount_capable() handling into
the two callers - one in the normal case of mount(2),
another - in fsconfig(2) handling of FSCONFIG_CMD_CREATE.
I.e. the syscalls that set a new filesystem up.
Signed-off-by: Al Viro <[email protected]>
sget_fc() is called only from ->get_tree() instances and
the only instance not calling it is legacy_get_tree(),
which calls mount_capable() directly.
In all sget_fc() callers the checks could be moved to the
very beginning of ->get_tree() - ->user_ns is not changed
in between. So lifting the checks to the only caller of
->get_tree() is OK.
Signed-off-by: Al Viro <[email protected]>
... and get rid of the weird dances in ->get_tree() - that logics
can be easily handled in ->init_fs_context().
Signed-off-by: Al Viro <[email protected]>
now both callers of mount_capable() have access to fs_context;
the only difference is that for sget_fc() we have the possibility
of fc->global being true, while for legacy_get_tree() it's guaranteed
to be impossible. Unify to more generic variant...
Signed-off-by: Al Viro <[email protected]>
guaranteed to be equal to current_user_ns() here - it has not
been changed since alloc_fs_context() (nothing in legacy
methods changes it) and since we don't have SB_SUBMOUNT,
that must've been FS_CONTEXT_FOR_MOUNT. And in that case
we have fc->user_ns set to fc->cred->user_ns, i.e.
current_cred()->user_ns, i.e. current_user_ns()
Signed-off-by: Al Viro <[email protected]>
1) all call chains leading to sget_userns() pass through ->mount()
instances.
2) none of ->mount() instances is ever called directly - the only
call site is legacy_get_tree()
3) all remaining ->mount() instances end up calling sget_userns()
IOW, we might as well do the capability checks just before calling
->mount(). As for the arguments passed to mount_capable(),
in case of call chains to sget_userns() going through sget(),
we either don't call mount_capable() at all, or pass current_user_ns()
to it. The call chains going through mount_pseudo_xattr() don't
call mount_capable() at all (SB_KERNMOUNT in flags on those).
That could've been split into smaller steps (lifting the checks
into sget(), then callers of sget(), then all the way to the
entries of every ->mount() out there, then to the sole caller),
but that would be too much churn for little benefit...
Signed-off-by: Al Viro <[email protected]>
there are 3 remaining callers of sget_userns() - sget(), mount_ns()
and mount_pseudo_xattr(). Extra check in sget() is conditional
upon mount being neither KERNMOUNT nor SUBMOUNT, the identical one
in mount_ns() - upon being not KERNMOUNT; mount_pseudo_xattr()
has no such checks at all.
However, mount_ns() is never used with SUBMOUNT and mount_pseudo_xattr()
is used only for KERNMOUNT, so both would be fine with the same logics
as currently done in sget(), allowing to consolidate the entire thing
in sget_userns() itself.
That's not where these checks will end up in the long run, though -
the whole reason why they'd been done so deep in the bowels of
mount(2) was that there had been no way for a filesystem to specify
which userns to look at until it has entered ->mount().
Now there is a place where filesystem could override the defaults -
->init_fs_context(). Which allows to pull the checks out into
the callers of vfs_get_tree(). That'll take quite a bit of
massage, but that mess is possible to tease apart.
Signed-off-by: Al Viro <[email protected]>
Once upon a time we used to set ->d_name of e.g. pipefs root
so that d_path() on pipes would work. These days it's
completely pointless - dentries of pipes are not even connected
to pipefs root. However, mount_pseudo() had set the root
dentry name (passed as the second argument) and callers
kept inventing names to pass to it. Including those that
didn't *have* any non-root dentries to start with...
All of that had been pointless for about 8 years now; it's
time to get rid of that cargo-culting...
Signed-off-by: Al Viro <[email protected]>
What triggers it is a race between mount --move and umount -l
of the source; we should reject it (the source is parentless *and*
not the root of anon namespace at that), but the check for namespace
being an anon one is broken in that case - is_anon_ns() needs
ns to be non-NULL. Better fixed here than in is_anon_ns(), since
the rest of the callers is guaranteed to get a non-NULL argument...
Reported-by: [email protected]
Signed-off-by: Al Viro <[email protected]>
To choose whether to pick the GID from the old (16bit) or new (32bit)
field, we should check if the old gid field is set to 0xffff. Mainline
checks the old *UID* field instead - cut'n'paste from the corresponding
code in ufs_get_inode_uid().
Fixes: 252e211e90
Signed-off-by: Al Viro <[email protected]>
moving synchronous parts of ->destroy_inode() to ->evict_inode() is
not possible here - they are balancing the stuff done in ->alloc_inode(),
not the things acquired while using it or sanity checks.
Signed-off-by: Al Viro <[email protected]>
fuse_destroy_inode() is gone - sanity checks that need the stack
trace of the caller get moved into ->evict_inode(), the rest joins
the RCU-delayed part which becomes ->free_inode().
While we are at it, don't just pass the address of what happens
to be the first member of structure to kmem_cache_free() -
get_fuse_inode() is there for purpose and it gives the proper
container_of() use. No behaviour change, but verifying correctness
is easier that way.
Signed-off-by: Al Viro <[email protected]>
the rest of this ->destroy_inode() instance could probably be folded
into ext4_evict_inode()
Reviewed-by: Jan Kara <[email protected]>
Signed-off-by: Al Viro <[email protected]>
debugging printks left in ->destroy_inode() and so's the
update of inode count; we could take the latter to RCU-delayed
part (would take only moving the check on module exit past
rcu_barrier() there), but debugging output ought to either
stay where it is or go into ->evict_inode()
Signed-off-by: Al Viro <[email protected]>
we might want to drop ->destroy_inode() there - it's used only for
WARN_ON() now, and AFAICS that could be moved to ->evict_inode()
if we had one...
Reviewed-by: Jan Kara <[email protected]>
Acked-by: Dan Williams <[email protected]>
Signed-off-by: Al Viro <[email protected]>
move the synchronous stuff from ->destroy_inode() to ->evict_inode(),
turn the RCU-delayed part into ->free_inode()
Signed-off-by: Al Viro <[email protected]>
... and use GFS2_I() to get the containing gfs2_inode by inode;
yes, we can feed the address of the first member of structure
to kmem_cache_free(), but let's do it in an obviously safe way.
Signed-off-by: Al Viro <[email protected]>
Al Viro