tun: fasync BKL pushdown

Signed-off-by: Jonathan Corbet <corbet@lwn.net>

i2o: fasync BKL pushdown

This driver appears to really need the BKL to protect open_files.

Signed-off-by: Jonathan Corbet <corbet@lwn.net>

mpt: fasync BKL pushdown

It looks like this driver really needs the BKL here.

Signed-off-by: Jonathan Corbet <corbet@lwn.net>

Remove BKL from remote_llseek v2

- Replace remote_llseek with generic_file_llseek_unlocked (to force compilation
failures in all users)
- Change all users to either use generic_file_llseek_unlocked directly or
take the BKL around. I changed the file systems who don't use the BKL
for anything (CIFS, GFS) to call it directly. NCPFS and SMBFS and NFS
take the BKL, but explicitely in their own source now.

I moved them all over in a single patch to avoid unbisectable sections.

Open problem: 32bit kernels can corrupt fpos because its modification
is not atomic, but they can do that anyways because there's other paths who
modify it without BKL.

Do we need a special lock for the pos/f_version = 0 checks?

Trond says the NFS BKL is likely not needed, but keep it for now
until his full audit.

v2: Use generic_file_llseek_unlocked instead of remote_llseek_unlocked
    and factor duplicated code (suggested by hch)

Cc: Trond.Myklebust@netapp.com
Cc: swhiteho@redhat.com
Cc: sfrench@samba.org
Cc: vandrove@vc.cvut.cz
Signed-off-by: Andi Kleen <ak@suse.de>
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: Jonathan Corbet <corbet@lwn.net>

Make FAT users happier by not deadlocking

The FAT BKL removal patch can cause deadlocks.  It turns out that the new
lock_super() calls are unneeded, remove them (as directed by Linus).

Reported-by: "Tony Luck" <tony.luck@intel.com>
Signed-off-by: Jonathan Corbet <corbet@lwn.net>

x86-mce: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

vmwatchdog: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

vmcp: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

via-pmu: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

uml-random: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

uml-mmapper: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

uml-harddog: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

uinput: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

uctrl: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

tpm-tpm: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

sparc64-rtc: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

sparc-apc: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

sonypi: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

sony-laptop: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

serio: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

scsi-tgt: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

sbus-rtc: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

rtc-rtc-m41t80: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

rtc: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

riowatchdog: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

pm_qos_params: BKL pushdown

[jmc: added <linux/smp_lock.h>]

Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Jonathan Corbet <corbet@lwn.net>

parisc-kernel-perf: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

parisc-eisa_eeprom: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

openprom: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

ocfs2-stack_user: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

nvram: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

net-tun: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

mwave-mwavedd: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

mvme16x-rtc: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

mousedev: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

monwriter: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

monreader: BKL pushdown

[jmc: added <linux/smp_lock.h>]

Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Jonathan Corbet <corbet@lwn.net>

microcode: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

message-i2o-i2o_config: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

megaraid: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

media-radio-miropcm20-rds: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

macintosh-smu: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

lcd: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

jsflash: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

irnet_ppp: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

ipmi-watchdog: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

ip27-rtc: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

infiniband-ucma: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

hw-random: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

hpet: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

hp_sdc_rtc: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

hdpu_cpustate: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

genrtc: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

flash: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

envctrl: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

efirtc: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

ds1620: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

ds1286: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

dlm-user: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

display7seg: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

crypto-zcrypt_api: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

crypto-prng: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

cpwatchdog: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

bvme6000-rtc: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

briq_panel: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

bluetooth-vhci: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

block-dasd_eer: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

bf561-coreb: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

basler-excite: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

apm_32: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

apm-emulation: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

ans-lcd: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

agp-frontend: BKL pushdown

Signed-off-by: Arnd Bergmann <arnd@arndb.de>

Replace BKL with superblock lock in fat/msdos/vfat

This replaces the use of the BKL in the FAT family of filesystems with the
existing superblock lock instead.

The code already appears to do mostly proper locking with its own private
spinlocks (and mutexes), but while the BKL could possibly have been
dropped entirely, converting it to use the superblock lock (which is just
a regular mutex) is the conservative thing to do.

As a per-filesystem mutex, it not only won't have any of the possible
latency issues related to the BKL, but the lock is obviously private to
the particular filesystem instance and will thus not cause problems for
entirely unrelated users like the BKL can.

Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Cc: OGAWA Hirofumi <hirofumi@mail.parknet.co.jp>
Signed-off-by: Jonathan Corbet <corbet@lwn.net>

bkl-removal viotape fixup

Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Jonathan Corbet <corbet@lwn.net>

Remove the lock_kernel() call from chrdev_open()

All in-kernel char device open() functions now either have their own
lock_kernel() calls or clearly do not need one.

Signed-off-by: Jonathan Corbet <corbet@lwn.net>

Add a comment in chrdev_open()

I stared at this code for a while and almost deleted it before
understanding crept into my slow brain.  Hopefully this makes life easier
for the next person to happen on it.

Signed-off-by: Jonathan Corbet <corbet@lwn.net>

Add a bunch of cycle_kernel_lock() calls

All of the open() functions which don't need the BKL on their face may
still depend on its acquisition to serialize opens against driver
initialization.  So make those functions acquire then release the BKL to be
on the safe side.

Signed-off-by: Jonathan Corbet <corbet@lwn.net>

Add cycle_kernel_lock()

A number of driver functions are so obviously trivial that they do not need
the big kernel lock - at least not overtly.  It turns out that the
acquisition of the BKL in driver open() functions can perform a sort of
poor-hacker's serialization function, delaying the open operation until the
driver is certain to have completed its initialization.  Add a simple
cycle_kernel_lock() function for these cases to make it clear that there is
no need to *hold* the BKL, just to be sure that we can acquire it.

Signed-off-by: Jonathan Corbet <corbet@lwn.net>

videodev: BKL pushdown

Put explicit lock_kernel() calls into videodev_open().  That function
itself seems OK, but one never knows about all the open() functions
provided by underlying video drivers.

Signed-off-by: Jonathan Corbet <corbet@lwn.net>

usbdev: BKL pushdown

Add explicit lock_kernel() calls to usbdev_open()

Signed-off-by: Jonathan Corbet <corbet@lwn.net>

USB Monitor: BKL pushdown

Add explicit lock_kernel() calls to mon_bin_open()

Signed-off-by: Jonathan Corbet <corbet@lwn.net>

printer gadget: BKL pushdown

Add explicit lock_kernel() calls to printer_open()

Signed-off-by: Jonathan Corbet <corbet@lwn.net>

adb: BKL pushdown

Put explicit lock_kernel() calls in adb_open().  The fact that
adb_release() already has them suggests this is necessary.

Signed-off-by: Jonathan Corbet <corbet@lwn.net>

divamnt: BKL pushdown

Put explicit lock_kernel() calls into maint_open().

Signed-off-by: Jonathan Corbet <corbet@lwn.net>

CAPI: BKL pushdown

Put explicit lock_kernel() calls into capi_open()

Signed-off-by: Jonathan Corbet <corbet@lwn.net>

changer: BKL pushdown

Add lock_kernel() calls to ch_open(), though the existing locking looks
adequate.

Signed-off-by: Jonathan Corbet <corbet@lwn.net>

dpt_i20: BKL pushdown

Add lock_kernel() calls to adpt_open()

Signed-off-by: Jonathan Corbet <corbet@lwn.net>

videopix: BKL pushdown

Add explicit lock_kernel() calls to vfc_open().

Signed-off-by: Jonathan Corbet <corbet@lwn.net>

bpp: bkl pushdown

Put explicit lock_kernel() calls into bpp_open().  It has locking, but I'm
not convinced it won't race with ioctl().

Signed-off-by: Jonathan Corbet <corbet@lwn.net>

phantom: BKL pushdown

Add explicit lock_kernel calls to phantom_open().

Signed-off-by: Jonathan Corbet <corbet@lwn.net>

infiniband: more BKL pushdown

Be extra-cautious and protect the remaining open() functions.

Signed-off-by: Jonathan Corbet <corbet@lwn.net>

dsp56k: BKL pushdown

Put explicit lock_kernel calls into dsp56k_open().

[Stupid missing label error fixed]

Signed-off-by: Jonathan Corbet <corbet@lwn.net>

raw: BKL pushdown

Put explicit lock_kernel() calls into raw_open(), even though the existing
locking looks adequate.

Signed-off-by: Jonathan Corbet <corbet@lwn.net>

tlckl: BKL pushdown

Put explicit lock_kernel calls into tlclk_open()

Signed-off-by: Jonathan Corbet <corbet@lwn.net>

xilinx icap: BKL pushdown

Add explicit lock_kernel() calls to hwicap_open() even though the existing
locking looks adequate.

Signed-off-by: Jonathan Corbet <corbet@lwn.net>

vcs: BKL pushdown

Add explicit BKL to vcs_open().

Signed-off-by: Jonathan Corbet <corbet@lwn.net>

spidev: BKL pushdown

Add the BKL to spidev_open(), even though the existing locking looks
adequate.

Signed-off-by: Jonathan Corbet <corbet@lwn.net>

Add "no BKL needed" comments to several drivers

This documents the fact that somebody looked at the relevant open()
functions and concluded that, due to their trivial nature, no locking was
needed.

Signed-off-by: Jonathan Corbet <corbet@lwn.net>

tty: cdev lock_kernel() pushdown

Parts of the serial code actually BUG() if we don't do this.