vfs: count unlinked inodes

Add a new counter to the superblock that keeps track of unlinked but
not yet deleted inodes.

Do not WARN_ON if set_nlink is called with zero count, just do a
ratelimited printk.  This happens on xfs and probably other
filesystems after an unclean shutdown when the filesystem reads inodes
which already have zero i_nlink.  Reported by Christoph Hellwig.

Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Tested-by: Toshiyuki Okajima <toshi.okajima@jp.fujitsu.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

vfs: protect remounting superblock read-only

Currently remouting superblock read-only is racy in a major way.

With the per mount read-only infrastructure it is now possible to
prevent most races, which this patch attempts.

Before starting the remount read-only, iterate through all mounts
belonging to the superblock and if none of them have any pending
writes, set sb->s_readonly_remount.  This indicates that remount is in
progress and no further write requests are allowed.  If the remount
succeeds set MS_RDONLY and reset s_readonly_remount.

If the remounting is unsuccessful just reset s_readonly_remount.
This can result in transient EROFS errors, despite the fact the
remount failed.  Unfortunately hodling off writes is difficult as
remount itself may touch the filesystem (e.g. through load_nls())
which would deadlock.

A later patch deals with delayed writes due to nlink going to zero.

Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Tested-by: Toshiyuki Okajima <toshi.okajima@jp.fujitsu.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

vfs: keep list of mounts for each superblock

Keep track of vfsmounts belonging to a superblock.  List is protected
by vfsmount_lock.

Signed-off-by: Miklos Szeredi <mszeredi@suse.cz>
Tested-by: Toshiyuki Okajima <toshi.okajima@jp.fujitsu.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

vfs: switch ->show_options() to struct dentry *

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

vfs: switch ->show_path() to struct dentry *

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

vfs: switch ->show_devname() to struct dentry *

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

vfs: switch ->show_stats to struct dentry *

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

switch security_path_chmod() to struct path *

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

vfs: prefer ->dentry->d_sb to ->mnt->mnt_sb

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

Merge branches 'vfsmount-guts', 'umode_t' and 'partitions' into Z

vfs: trim includes a bit

[folded fix for missing magic.h from Tetsuo Handa]

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

switch mnt_namespace ->root to struct mount

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

vfs: take /proc/*/mounts and friends to fs/proc_namespace.c

rationale: that stuff is far tighter bound to fs/namespace.c than to
the guts of procfs proper.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

vfs: opencode mntget() mnt_set_mountpoint()

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

vfs: spread struct mount - remaining argument of next_mnt()

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

vfs: move fsnotify junk to struct mount

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

vfs: move mnt_devname

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

vfs: move mnt_list to struct mount

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

vfs: switch pnode.h macros to struct mount *

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

vfs: move the rest of int fields to struct mount

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

vfs: mnt_id/mnt_group_id moved

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

vfs: mnt_ns moved to struct mount

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

vfs: spread struct mount - mntput_no_expire

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

vfs: spread struct mount - do_add_mount and graft_tree

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

vfs: take mnt_share/mnt_slave/mnt_slave_list and mnt_expire to struct mount

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

vfs: and now we can make ->mnt_master point to struct mount

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

vfs: take mnt_master to struct mount

make IS_MNT_SLAVE take struct mount * at the same time

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

vfs: spread struct mount - remaining argument of mnt_set_mountpoint()

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

vfs: spread struct mount - propagate_mnt()

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

vfs: spread struct mount - shared subtree iterators

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

vfs: spread struct mount - get_dominating_id / do_make_slave

next pile of horrors, similar to mnt_parent one; this time it's
mnt_master.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

vfs: take mnt_child/mnt_mounts to struct mount

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

vfs: all counters taken to struct mount

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

vfs: spread struct mount - work with counters

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

vfs: move mnt_mountpoint to struct mount

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

vfs: now it can be done - make mnt_parent point to struct mount

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

vfs: mnt_parent moved to struct mount

the second victim...

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

vfs: spread struct mount - is_path_reachable

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

vfs: spread struct mount - mnt_has_parent

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

vfs: spread struct mount - do_umount/propagate_mount_busy

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

vfs: spread struct mount mnt_set_mountpoint child argument

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

vfs: spread struct mount - clone_mnt/copy_tree argument

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

vfs: spread struct mount - shrink_submounts/select_submounts

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

vfs: spread struct mount - umount_tree argument

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

vfs: the first spoils - mnt_hash moved

taken out of struct vfsmount into struct mount

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

vfs: spread struct mount to remaining users of ->mnt_hash

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

vfs: spread struct mount - clone_mnt/copy_tree result

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

vfs: spread struct mount - change_mnt_propagation/set_mnt_shared

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

vfs: spread struct mount - alloc_vfsmnt/free_vfsmnt/mnt_alloc_id/mnt_free_id

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

vfs: spread struct mount - tree_contains_unbindable

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

vfs: spread struct mount - attach_recursive_mnt

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

vfs: spread struct mount - mount group id handling

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

vfs: spread struct mount - commit_tree

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

vfs: spread struct mount - attach_mnt/detach_mnt

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

vfs: spread struct mount - namespace.c internal iterators

next_mnt() return value, first argument
skip_mnt_tree() return value and argument

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

vfs: spread struct mount - __propagate_umount() argument

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

vfs: spread struct mount - __lookup_mnt() result

switch __lookup_mnt() to returning struct mount *; callers adjusted.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

vfs: start hiding vfsmount guts series

Almost all fields of struct vfsmount are used only by core VFS (and
a fairly small part of it, at that).  The plan: embed struct vfsmount
into struct mount, making the latter visible only to core parts of VFS.
Then move fields from vfsmount to mount, eventually leaving only
mnt_root/mnt_sb/mnt_flags in struct vfsmount.  Filesystem code still
gets pointers to struct vfsmount and remains unchanged; all such
pointers go to struct vfsmount embedded into the instances of struct
mount allocated by fs/namespace.c.  When fs/namespace.c et.al. get
a pointer to vfsmount, they turn it into pointer to mount (using
container_of) and work with that.

This is the first part of series; struct mount is introduced,
allocation switched to using it.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

switch open and mkdir syscalls to umode_t

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

switch ->path_mknod() to umode_t

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

switch ->path_mkdir() to umode_t

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

tomoyo_mini_stat: switch to umode_t

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

consolidate a bunch of ipcbuf.h instances

... some still remain weird :-/

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

auditsc: propage umode_t

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

switch kern_ipc_perm to umode_t

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

consolidate umode_t declarations

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

switch spu_create(2) to use of SYSCALL_DEFINE4, make it use umode_t

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

switch mq_open() to umode_t

ceph: propagate umode_t

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

switch hostfs_iattr to explicit unsigned short

It's shared between kernel-compiled hostfs_kern and userland-compiled
hostfs_user (it's uml stuff).  Use explicit type instead of playing
silly buggers with mode_t.  It's not a userland API per se; it interacts
between code compiled with types same as for host kernel and, directly
linked to it, code talking to libc.  Both sides come from the same
kernel source...

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

switch miscdevice to umode_t

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

switch spufs guts to umode_t

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

init/initramfs.c: should use umode_t

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

switch may_mknod() to umode_t

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

should_remove_suid(): inode->i_mode is umode_t

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

switch securityfs_create_file() to umode_t

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

switch security_path_chmod() to umode_t

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

switch sys_chmod()/sys_fchmod()/sys_fchmodat() to umode_t

SYSCALLx magic should take care of things, according to Linus...

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

sysctl: use umode_t for table permissions

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

get rid of open-coded S_ISREG(), etc.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

mqueue: propagate umode_t

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

switch is_sxid() to umode_t

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

switch inode_init_owner() to umode_t

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

fs: propagate umode_t, misc bits

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

autofs4: propagate umode_t

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

hfsplus: propagate umode_t

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

hfs: propagate umode_t

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

cifs: propagate umode_t

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

fat: propagate umode_t

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

ntfs: propagate umode_t

same story as with isofs and udf...

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

isofs: propagate umode_t

situation with mount options is the same as for udf

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

udf: propagate umode_t

note re mount options: fmask and dmask are explicitly truncated to 12bit,
UDF_INVALID_MODE just needs to be guaranteed to differ from any such value.
And umask is used only in &= with umode_t, so we ignore other bits anyway.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

fuse: propagate umode_t

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

pohmelfs: propagate umode_t

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

shmem, ramfs: propagate umode_t, open-coded S_ISREG

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

logfs: propagate umode_t

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

ubifs: propagate umode_t

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

ncpfs: propagate umode_t

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

hugetlbfs: propagate umode_t

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

exofs: propagate umode_t

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>