scsi_proc: make proc_scsi_host_open() preallocate a bigger buffer

Some of the ->show_info() instances really spew a lot; it's not a problem
wrt correctness (seq_read() will grow buffer and call the sucker again),
but in this case it makes sense to start with a somewhat bigger one -
they often do exceed one page worth of output.

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

new helper: single_open_size()

Same as single_open(), but preallocates the buffer of given size.
Doesn't make any sense for sizes up to PAGE_SIZE and doesn't make
sense if output of show() exceeds PAGE_SIZE only rarely - seq_read()
will take care of growing the buffer and redoing show().  If you
_know_ that it will be large, it might make more sense to look into
saner iterator, rather than go with single-shot one.  If that's
impossible, single_open_size() might be for you.

Again, don't use that without a good reason; occasionally that's really
the best way to go, but very often there are better solutions.

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

scsi: bury ->proc_info()

all users converted to ->show_info()/->write_info()

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

atari_scsi: switch to ->show_info()

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

aix7xxx_old: switch to ->show_info()

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

sata_svw: switch to ->show_info()

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

ips: fix format bugs exposed by previous commit

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

ips: switch to ->show_info()

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

acornscsi: switch to ->show_info()

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

fas216: switch to ->show_info()

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

aic7xxx: switch to ->show_info()

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

aic79xx: switch to ->show_info()

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

in2000: switch to ->show_info()

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

atp870u: switch to ->show_info()

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

dc395x: switch to ->show_info()

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

dpt_i2o: switch to ->show_info()

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

eata_pio: switch to ->show_info()

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

aha1740: switch to ->show_info()

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

aha152x: switch to ->show_info()

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

advansys: fix buggered formats - unsigned long != u32 on 64bit

asc_prt_line() had been hiding several places where formats had not
matched the argument types.  The previous commit has finally made them
visible...

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

advansys: switch to ->show_info()

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

nsp_cs: switch to ->show_info()

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

nsp32: switch to ->show_info()

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

keucr: switch to ->show_info()

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

rts5139: switch to ->show_info()

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

tcm: switch to ->show_info()

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

cciss: switch to ->show_info()

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

BusLogic: switch to ->show_info()

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

scsi_debug: switch to ->show_info()

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

fusion: switch to ->show_info()

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

usb-storage: switch to ->show_info()

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

NCR5830: switch to ->show_info()

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

ppa: switch to ->show_info()

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

imm: switch to ->show_info()

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

gdth: switch to ->show_info()

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

switch wd33c93 to ->show_info()

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

wd7000: switch to ->show_info()

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

sym53c8xx_2: switch to ->show_info()

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

scsi: saner replacements for ->proc_info()

It's still an obsolete interface; don't introduce those in new drivers.
However, it's saner than the ->proc_info() and commits after this one
will convert the existing ->proc_info() users to it.

The read side is ->show_info(seq_file *, struct Scsi_Host *); use
seq_... for generating contents.

The write side is ->write_info(struct Scsi_Host *, char *, int).

Again, this is driven by procfs needs; we are going to kill ->write_proc()
and ->read_proc() and this is the main obstacle to burying that piece of
shit.

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

procfs: don't allow to use proc_create, create_proc_entry, etc. for directories

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

i2o: use proc_remove_subtree()

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

silicom: bury bp_proc.c

It's a seriously rotten copy of parts of bp_mod.c; had been
ifdefed out all along, lacks a bunch of declarations that would
be needed if ifdef had been removed, all stuff in it is duplicated
in bp_mod.c anyway...

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

silicom: switch to create_proc_data(), clean procfs side of things up

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

reiserfs: use proc_remove_subtree()

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

silicom: untangle module_init

Helper functions are often useful.  So are local variables...

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

silicom: get_bypass_slave_pfs() open-codes lookup_port()

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

silicom: unobfuscate get_{status,master}_port_fn()

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

silicom: helper functions are often useful...

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

get rid of a bunch of open-coded create_proc_read_entry()

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

isp1362-hcd: don't reimplement proc_create_data()

... especially in a racy way

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

rtl8192u: don't play with reassigning ->proc_fops, just use proc_create()

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

dgrp procfs fixes, part 6: just use proc_create{,_data}

->permission() of its own is a rudiment of sysctl imitation;
normal procfs logics will do just fine here, no need to
mess with ->proc_iops at all.

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

dgrp procfs fixes, part 5: per-node files

this "hooks" scheme is pointless - just make file_operations non-static
and consolidate initialiazation bits.

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

dgrp procfs fixes, part 4: get rid of sysctl-like machinery

racy and very overblown...

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

dgrp procfs fixes, part 3: kill dead code

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

dgrp procfs fixes, part 2

All table entries either have non-NULL ->proc_file_fops or
non-NULL child.

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

dgrp procfs fixes, part 1

proc_create() has shat upon fops argument when mode is S_IFDIR.
Good thing, too, since fops passed to it is completely useless
for any directory.  Just use proc_mkdir(), damnit.

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

The rtl8192e procfs-based debug interface seems very broken

The procfs debug code in rtl_debug.c is, ironically, very buggy: it lacks proper locking.
Since the most useful part of the code (the stats) are available through more
standard APIs, I think it is best to just delete the whole mess.

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

rtl8192e: switch to proc_create()

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

rtl8192e: don't use create_proc_entry() for directories

proc_mkdir() is there for purpose...

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

procfs: switch /proc/self away from proc_dir_entry

Just have it pinned in dcache all along and let procfs ->kill_sb()
drop it before kill_anon_super().

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

mode_t, whack-a-mole at 11...

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

snd_info_register: switch to proc_create_data/proc_mkdir_mode

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

hysdn: stash pointer to card into proc_dir_entry->data

no need to search later - we know the card when we are
creating procfs entries

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

atags_proc: switch to proc_create_data()

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

gadgetfs: don't bother with fops->owner

filesystem module as whole is pinned down by its superblock, no need
to have opened files on it to add anything to that.

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

ccg: don't bother with fops->owner

filesystem module as whole is pinned down by its superblock, no need
to have opened files on it to add anything to that.

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

spufs: don't bother with fops->owner

filesystem module as whole is pinned down by its superblock, no need
to have opened files on it to add anything to that.

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

get rid of the last free_pipe_info() callers

and rename __free_pipe_info() to free_pipe_info()

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

get rid of alloc_pipe_info() argument

not used anymore

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

get rid of pipe->inode

it's used only as a flag to distinguish normal pipes/FIFOs from the
internal per-task one used by file-to-file splice.  And pipe->files
would work just as well for that purpose...

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

introduce variants of pipe_lock/pipe_unlock for real pipes/FIFOs

fs/pipe.c file_operations methods *know* that pipe is not an internal one;
no need to check pipe->inode for those callers.

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

pipe: set file->private_data to ->i_pipe

simplify get_pipe_info(), while we are at it

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

pipe: don't use ->i_mutex

now it can be done - put mutex into pipe_inode_info, use it instead
of ->i_mutex

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

pipe: take allocation and freeing of pipe_inode_info out of ->i_mutex

* new field - pipe->files; number of struct file over that pipe (all
  sharing the same inode, of course); protected by inode->i_lock.
* pipe_release() decrements pipe->files, clears inode->i_pipe when
  if the counter has reached 0 (all under ->i_lock) and, in that case,
  frees pipe after having done pipe_unlock()
* fifo_open() starts with grabbing ->i_lock, and either bumps pipe->files
  if ->i_pipe was non-NULL or allocates a new pipe (dropping and regaining
  ->i_lock) and rechecks ->i_pipe; if it's still NULL, inserts new pipe
  there, otherwise bumps ->i_pipe->files and frees the one we'd allocated.
  At that point we know that ->i_pipe is non-NULL and won't go away, so
  we can do pipe_lock() on it and proceed as we used to.  If we end up
  failing, decrement pipe->files and if it reaches 0 clear ->i_pipe and
  free the sucker after pipe_unlock().

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

pipe: preparation to new locking rules

* use the fact that file_inode(file)->i_pipe doesn't change
  while the file is opened - no locks needed to access that.
* switch to pipe_lock/pipe_unlock where it's easy to do

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

pipe: switch wait_for_partner() and wake_up_partner() to pipe_inode_info

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

pipe: fold file_operations instances in one

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

fold fifo.c into pipe.c

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

lift sb_start_write out of ->splice_write()

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

lift sb_start_write into default_file_splice_write()

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

lift sb_start_write() out of ->write()

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

switch compat readv/writev variants to COMPAT_SYSCALL_DEFINE

... and take to fs/read_write.c

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

f2fs: use mnt_want_write_file() in ioctl

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

lift sb_start_write/sb_end_write out of ->aio_write()

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

hpfs: move setting hpfs-private i_dirty to ->write_end()

... so that writev(2) doesn't miss it.  Get rid of hpfs_file_write().

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

reiserfs: don't wank with EFBIG before calling do_sync_write()

look for file_capable() in there...

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

fold release_mounts() into namespace_unlock()

... and provide namespace_lock() as a trivial wrapper;
switch to those two consistently.

Result is patterned after rtnl_lock/rtnl_unlock pair.

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

switch unlock_mount() to namespace_unlock(), convert all umount_tree() callers

which allows to kill the last argument of umount_tree() and make release_mounts()
static.

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

more conversions to namespace_unlock()

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

get rid of the second argument of shrink_submounts()

... it's always &unmounted.

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

saner umount_tree()/release_mounts(), part 1

global list of release_mounts() fodder, protected by namespace_sem;
eventually, all umount_tree() callers will use it as kill list.
Helper picking the contents of that list, releasing namespace_sem
and doing release_mounts() on what it got.

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

get rid of full-hash scan on detaching vfsmounts

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

mnt: release locks on error path in do_loopback

do_loopback calls lock_mount(path) and forget to unlock_mount
if clone_mnt or copy_mnt fails.

[   77.661566] ================================================
[   77.662939] [ BUG: lock held when returning to user space! ]
[   77.664104] 3.9.0-rc5+ #17 Not tainted
[   77.664982] ------------------------------------------------
[   77.666488] mount/514 is leaving the kernel with locks still held!
[   77.668027] 2 locks held by mount/514:
[   77.668817]  #0:  (&sb->s_type->i_mutex_key#7){+.+.+.}, at: [<ffffffff811cca22>] lock_mount+0x32/0xe0
[   77.671755]  #1:  (&namespace_sem){+++++.}, at: [<ffffffff811cca3a>] lock_mount+0x4a/0xe0

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

palinfo fixes

* check for proc_mkdir() failures
* fix buffer overrun - sizeof(format string) is *not* enough to
hold sprintf() result.
* use proc_remove_subtree(); life's much easier with it

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

procfs: add proc_remove_subtree()

just what it sounds like; do that only to procfs subtrees you've
created - doing that to something shared with another driver is
not only antisocial, but might cause interesting races with
proc_create() and its ilk.

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

ecryptfs: close rmmod race

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

Merge branch 'for-linus' of git://git./linux/kernel/git/viro/vfs

Pull vfs fixes from Al Viro:
 "stable fodder; assorted deadlock fixes"

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs:
  vt: synchronize_rcu() under spinlock is not nice...
  Nest rename_lock inside vfsmount_lock
  Don't bother with redoing rw_verify_area() from default_file_splice_from()

vt: synchronize_rcu() under spinlock is not nice...

vcs_poll_data_free() calls unregister_vt_notifier(), which calls
atomic_notifier_chain_unregister(), which calls synchronize_rcu().
Do it *after* we'd dropped ->f_lock.

Cc: stable@vger.kernel.org (all kernels since 2.6.37)
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>

Nest rename_lock inside vfsmount_lock

... lest we get livelocks between path_is_under() and d_path() and friends.

The thing is, wrt fairness lglocks are more similar to rwsems than to rwlocks;
it is possible to have thread B spin on attempt to take lock shared while thread
A is already holding it shared, if B is on lower-numbered CPU than A and there's
a thread C spinning on attempt to take the same lock exclusive.

As the result, we need consistent ordering between vfsmount_lock (lglock) and
rename_lock (seq_lock), even though everything that takes both is going to take
vfsmount_lock only shared.

Spotted-by: Brad Spengler <spender@grsecurity.net>
Cc: stable@vger.kernel.org
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>