vfs: export lseek_execute() to modules

For those file systems(btrfs/ext4/ocfs2/tmpfs) that support
SEEK_DATA/SEEK_HOLE functions, we end up handling the similar
matter in lseek_execute() to update the current file offset
to the desired offset if it is valid, ceph also does the
simliar things at ceph_llseek().

To reduce the duplications, this patch make lseek_execute()
public accessible so that we can call it directly from the
underlying file systems.

Thanks Dave Chinner for this suggestion.

[AV: call it vfs_setpos(), don't bring the removed 'inode' argument back]

v2->v1:
- Add kernel-doc comments for lseek_execute()
- Call lseek_execute() in ceph->llseek()

Signed-off-by: Jie Liu <jeff.liu@oracle.com>
Cc: Dave Chinner <dchinner@redhat.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Chris Mason <chris.mason@fusionio.com>
Cc: Josef Bacik <jbacik@fusionio.com>
Cc: Ben Myers <bpm@sgi.com>
Cc: Ted Tso <tytso@mit.edu>
Cc: Hugh Dickins <hughd@google.com>
Cc: Mark Fasheh <mfasheh@suse.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Sage Weil <sage@inktank.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

lseek_execute() doesn't need an inode passed to it

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

block_dev: switch to fixed_size_llseek()

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

cpqphp_sysfs: switch to fixed_size_llseek()

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

tile-srom: switch to fixed_size_llseek()

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

proc_powerpc: switch to fixed_size_llseek()

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

ubi/cdev: switch to fixed_size_llseek()

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

pci/proc: switch to fixed_size_llseek()

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

isapnp: switch to fixed_size_llseek()

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

lpfc: switch to fixed_size_llseek()

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

locks: give the blocked_hash its own spinlock

There's no reason we have to protect the blocked_hash and file_lock_list
with the same spinlock. With the tests I have, breaking it in two gives
a barely measurable performance benefit, but it seems reasonable to make
this locking as granular as possible.

Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

locks: add a new "lm_owner_key" lock operation

Currently, the hashing that the locking code uses to add these values
to the blocked_hash is simply calculated using fl_owner field. That's
valid in most cases except for server-side lockd, which validates the
owner of a lock based on fl_owner and fl_pid.

In the case where you have a small number of NFS clients doing a lot
of locking between different processes, you could end up with all
the blocked requests sitting in a very small number of hash buckets.

Add a new lm_owner_key operation to the lock_manager_operations that
will generate an unsigned long to use as the key in the hashtable.
That function is only implemented for server-side lockd, and simply
XORs the fl_owner and fl_pid.

Signed-off-by: Jeff Layton <jlayton@redhat.com>
Acked-by: J. Bruce Fields <bfields@fieldses.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

locks: turn the blocked_list into a hashtable

Break up the blocked_list into a hashtable, using the fl_owner as a key.
This speeds up searching the hash chains, which is especially significant
for deadlock detection.

Note that the initial implementation assumes that hashing on fl_owner is
sufficient. In most cases it should be, with the notable exception being
server-side lockd, which compares ownership using a tuple of the
nlm_host and the pid sent in the lock request. So, this may degrade to a
single hash bucket when you only have a single NFS client. That will be
addressed in a later patch.

The careful observer may note that this patch leaves the file_lock_list
alone. There's much less of a case for turning the file_lock_list into a
hashtable. The only user of that list is the code that generates
/proc/locks, and it always walks the entire list.

Signed-off-by: Jeff Layton <jlayton@redhat.com>
Acked-by: J. Bruce Fields <bfields@fieldses.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

locks: convert fl_link to a hlist_node

Testing has shown that iterating over the blocked_list for deadlock
detection turns out to be a bottleneck. In order to alleviate that,
begin the process of turning it into a hashtable. We start by turning
the fl_link into a hlist_node and the global lists into hlists. A later
patch will do the conversion of the blocked_list to a hashtable.

Signed-off-by: Jeff Layton <jlayton@redhat.com>
Acked-by: J. Bruce Fields <bfields@fieldses.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

locks: avoid taking global lock if possible when waking up blocked waiters

Since we always hold the i_lock when inserting a new waiter onto the
fl_block list, we can avoid taking the global lock at all if we find
that it's empty when we go to wake up blocked waiters.

Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

locks: protect most of the file_lock handling with i_lock

Having a global lock that protects all of this code is a clear
scalability problem. Instead of doing that, move most of the code to be
protected by the i_lock instead. The exceptions are the global lists
that the ->fl_link sits on, and the ->fl_block list.

->fl_link is what connects these structures to the
global lists, so we must ensure that we hold those locks when iterating
over or updating these lists.

Furthermore, sound deadlock detection requires that we hold the
blocked_list state steady while checking for loops. We also must ensure
that the search and update to the list are atomic.

For the checking and insertion side of the blocked_list, push the
acquisition of the global lock into __posix_lock_file and ensure that
checking and update of the  blocked_list is done without dropping the
lock in between.

On the removal side, when waking up blocked lock waiters, take the
global lock before walking the blocked list and dequeue the waiters from
the global list prior to removal from the fl_block list.

With this, deadlock detection should be race free while we minimize
excessive file_lock_lock thrashing.

Finally, in order to avoid a lock inversion problem when handling
/proc/locks output we must ensure that manipulations of the fl_block
list are also protected by the file_lock_lock.

Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

locks: encapsulate the fl_link list handling

Move the fl_link list handling routines into a separate set of helpers.
Also ensure that locks and requests are always put on global lists
last (after fully initializing them) and are taken off before unintializing
them.

Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

locks: make "added" in __posix_lock_file a bool

Signed-off-by: Jeff Layton <jlayton@redhat.com>
Acked-by: J. Bruce Fields <bfields@fieldses.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

locks: comment cleanups and clarifications

Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

locks: make generic_add_lease and generic_delete_lease static

Signed-off-by: Jeff Layton <jlayton@redhat.com>
Acked-by: J. Bruce Fields <bfields@fieldses.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

cifs: use posix_unblock_lock instead of locks_delete_block

commit 66189be74 (CIFS: Fix VFS lock usage for oplocked files) exported
the locks_delete_block symbol. There's already an exported helper
function that provides this capability however, so make cifs use that
instead and turn locks_delete_block back into a static function.

Note that if fl->fl_next == NULL then this lock has already been through
locks_delete_block(), so we should be OK to ignore an ENOENT error here
and simply not retry the lock.

Cc: Pavel Shilovsky <piastryyy@gmail.com>
Signed-off-by: Jeff Layton <jlayton@redhat.com>
Acked-by: J. Bruce Fields <bfields@fieldses.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

locks: drop the unused filp argument to posix_unblock_lock

Signed-off-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

Don't pass inode to ->d_hash() and ->d_compare()

Instances either don't look at it at all (the majority of cases) or
only want it to find the superblock (which can be had as dentry->d_sb).
A few cases that want more are actually safe with dentry->d_inode -
the only precaution needed is the check that it hadn't been replaced with
NULL by rmdir() or by overwriting rename(), which case should be simply
treated as cache miss.

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

minix: bug widening a binary "not" operation

"chunk_size" is an unsigned int and "pos" is an unsigned long.  The
"& ~(chunk_size-1)" operation clears the high 32 bits unintentionally.

The ALIGN() macro does the correct thing.

Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

splice: lift checks from do_splice_from() into callers

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

constify rw_verify_area()

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

ps3flash: switch to generic_file_llseek_size()

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

wlcore: use *ppos, not file->f_pos

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

bfa: switch to fixed_size_llseek()

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

fnic: switch to fixed_size_llseek()

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

vc: switch to fixed_size_llseek()

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

eisa_eeprom: switch to fixed_size_llseek()

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

bna: switch to fixed_size_llseek()

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

zorro: switch to fixed_size_llseek()

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

mtdchar: switch to fixed_size_llseek()

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

new helper: fixed_size_llseek()

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

ecryptfs: switch ecryptfs_decode_and_decrypt_filename() from dentry to sb

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

fuse: another open-coded file_inode()

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

btrfs: more open-coded file_inode()

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

fanotify: quit wanking with FASYNC in ->release()

... especially since there's no way to get that sucker
on the list fsnotify_fasync() works with - the only thing
adding to it is fsnotify_fasync() itself and it's never
called for fanotify files while they are opened.

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

comedi: quit wanking with FASYNC in ->release()

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

more open-coded file_inode() calls

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

kill find_inode_number()

the only remaining caller (in ncpfs) is guaranteed to return 0 -
we only hit it if we'd just checked that there's no dentry with
such name.

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

coda: don't bother with find_inode_number()

the fallback it's using for dcache misses is actually the
same value we would've used for inumber anyway.

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

proc_fill_cache(): clean up, get rid of pointless find_inode_number() use

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

proc_fill_cache(): just make instantiate_t return int

all instances always return ERR_PTR(-E...) or NULL, anyway

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

proc_pid_readdir(): stop wanking with proc_fill_cache() for /proc/self

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

proc_fill_cache(): kill pointless check

we'd just checked that child->d_inode is non-NULL, for fuck sake!

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

ncpfs: don't bother with EBUSY on removal of busy directories

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

don't call file_pos_write() if vfs_{read,write}{,v}() fails

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

SELinux: Institute file_path_has_perm()

Create a file_path_has_perm() function that is like path_has_perm() but
instead takes a file struct that is the source of both the path and the
inode (rather than getting the inode from the dentry in the path).  This
is then used where appropriate.

This will be useful for situations like unionmount where it will be
possible to have an apparently-negative dentry (eg. a fallthrough) that is
open with the file struct pointing to an inode on the lower fs.

Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

Replace a bunch of file->dentry->d_inode refs with file_inode()

Replace a bunch of file->dentry->d_inode refs with file_inode().

In __fput(), use file->f_inode instead so as not to be affected by any tricks
that file_inode() might grow.

Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

udf: provide ->tmpfile()

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

ext3 ->tmpfile() support

In this case we do need a bit more than usual, due to orphan
list handling.

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

allow the temp files created by open() to be linked to

O_TMPFILE | O_CREAT => linkat() with AT_SYMLINK_FOLLOW and /proc/self/fd/<n>
as oldpath (i.e. flink()) will create a link
O_TMPFILE | O_CREAT | O_EXCL => ENOENT on attempt to link those guys

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

[O_TMPFILE] it's still short a few helpers, but infrastructure should be OK now...

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

allow build_open_flags() to return an error

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

lift file_*_write out of do_splice_direct()

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

lift file_*_write out of do_splice_from()

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

do_last(): fix missing checks for LAST_BIND case

/proc/self/cwd with O_CREAT should fail with EISDIR.  /proc/self/exe, OTOH,
should fail with ENOTDIR when opened with O_DIRECTORY.

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

pcm_native: switch to fdget()/fdput()

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

[readdir] constify ->actor

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

[readdir] ->readdir() is gone

everything's converted to ->iterate()

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

[readdir] convert ecryptfs

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

[readdir] convert coda

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

[readdir] convert ocfs2

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

[readdir] convert fatfs

... pox upon the idiotic ioctls; life would be much easier without
those.

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

[readdir] convert xfs

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

[readdir] convert btrfs

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

[readdir] convert hostfs

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

[readdir] convert afs

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

[readdir] convert ncpfs

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

[readdir] convert hfsplus

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

[readdir] convert hfs

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

[readdir] convert befs

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

[readdir] convert cifs

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

[readdir] convert freevxfs

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

[readdir] convert fuse

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

[readdir] convert hpfs

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

reiserfs: switch reiserfs_readdir_dentry to inode

... and clean the callers up a bit

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

reiserfs: is_privroot_deh() needs only directory inode, actually

... and that - only to get the superblock.  Privroot is a directory
and we don't allow hardlinks to those...

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

[readdir] convert reiserfs

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

[readdir] convert ntfs

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

[readdir] convert isofs

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

[readdir] convert jffs2

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

[readdir] convert f2fs

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

[readdir] convert 9p

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

[readdir] convert affs

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

[readdir] convert adfs

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

[readdir] convert logfs

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

[readdir] convert jfs

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

[readdir] convert ceph

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

[readdir] convert nfs

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

[readdir] convert ext4

and trim the living hell out bogosities in inline dir case

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

[readdir] convert qnx6

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

[readdir] convert qnx4

... and use strnlen() instead of strlen() - it's done on untrusted data,
after all.

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

[readdir] convert omfs

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

[readdir] convert nilfs2

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

[readdir] convert sysfs

get rid of the kludges in sysfs_readdir()

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

[readdir] convert gfs2

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