cfg80211: ignore spurious deauth

commit 643f82e32f14faf0d0944c804203a6681b6b0a1e upstream.

Ever since mac80211/drivers are no longer
fully in charge of keeping track of the
auth status, trying to make them do so will
fail. Instead of warning and reporting the
deauthentication to userspace, cfg80211 must
simply ignore it so that spurious
deauthentications, e.g. before starting
authentication, aren't seen by userspace as
actual deauthentications.

Reported-by: Paul Stewart <pstew@google.com>
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

iwlwifi: fix scan abort

commit d28232b461b8d54b09e59325dbac8b0913ce2049 upstream.

Fix possible double priv->mutex lock introduced by commit
a69b03e941abae00380fc6bc1877fb797a1b31e6
"iwlwifi: cancel scan watchdog in iwl_bg_abort_scan" .
We can not call cancel_delayed_work_sync(&priv->scan_check) with
priv->mutex locked because workqueue function iwl_bg_scan_check()
take that lock internally.

We do not need to synchronize when canceling priv->scan_check work.
We can avoid races (sending double abort command or send no
command at all) using STATUS_SCAN_ABORT bit. Moreover
current iwl_bg_scan_check() code seems to be broken, as
we should not send abort commands when currently aborting.

Signed-off-by: Stanislaw Gruszka <sgruszka@redhat.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

ath9k: fix yet another buffer leak in the tx aggregation code

commit 4cee78614cfa046a26c4fbf313d5bbacb3ad8efc upstream.

When an aggregation session is being cleaned up, while the tx status
for some frames is being processed, the TID is flushed and its buffers
are sent out.

Unfortunately that left the pending un-acked frames unprocessed, thus
leaking buffers. Fix this by reordering the code so that those frames
are processed first, before the TID is flushed.

Signed-off-by: Felix Fietkau <nbd@openwrt.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

ath9k: fix TSF after reset on AR913x

commit f860d526eb2939a1c37128900b5af2b6f3ff7f20 upstream.

When issuing a reset, the TSF value is lost in the hardware because of
the 913x specific cold reset. As with some AR9280 cards, the TSF needs
to be preserved in software here.

Additionally, there's an issue that frequently prevents a successful
TSF write directly after the chip reset. In this case, repeating the
TSF write after the initval-writes usually works.

This patch detects failed TSF writes and recovers from them, taking
into account the delay caused by the initval writes.

Signed-off-by: Felix Fietkau <nbd@openwrt.org>
Reported-by: Björn Smedman <bjorn.smedman@venatech.se>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

ath9k_hw: fix an off-by-one error in the PDADC boundaries calculation

commit 03b4776c408d2f4bf3a5d204e223724d154716d1 upstream.

PDADC values were only generated for values surrounding the target
index, however not for the target index itself, leading to a minor
error in the generated curve.

Signed-off-by: Felix Fietkau <nbd@openwrt.org>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

ath9k: enable serialize_regmode for non-PCIE AR9160

commit 4c85ab11ca56da1aa59b58c80cc6a356515cc645 upstream.

https://bugzilla.kernel.org/show_bug.cgi?id=16476

Signed-off-by: John W. Linville <linville@tuxdriver.com>
Acked-by: Luis R. Rodriguez <lrodriguez@atheros.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

9p: strlen() doesn't count the terminator

commit 5c4bfa17f3ec46becec4b23d12323f7605ebd696 upstream.

This is an off by one bug because strlen() doesn't count the NULL
terminator.  We strcpy() addr into a fixed length array of size
UNIX_PATH_MAX later on.

The addr variable is the name of the device being mounted.

Signed-off-by: Dan Carpenter <error27@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

e1000e: 82577/82578 PHY register access issues

commit ff847ac2d3e90edd94674c28bade25ae1e6a2e49 upstream.

The MAC-PHY interconnect on 82577/82578 uses a power management feature
(called K1) which must be disabled when in 1Gbps due to a hardware issue on
these parts.  The #define bit setting used to enable/disable K1 is
incorrect and can cause PHY register accesses to stop working altogether
until the next device reset.  This patch sets the register correctly.

This issue is present in kernels since 2.6.32.

Signed-off-by: Bruce Allan <bruce.w.allan@intel.com>
Tested-by: Jeff Pieper <jeffrey.e.pieper@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

e1000e: don't inadvertently re-set INTX_DISABLE

commit 36f2407fe52c55566221f8c68c8fb808abffd2f5 upstream.

Should e1000_test_msi() fail to see an msi interrupt, it attempts to
fallback to legacy INTx interrupts. But an error in the code may prevent
this from happening correctly.

Before calling e1000_test_msi_interrupt(), e1000_test_msi() disables SERR
by clearing the SERR bit from the just read PCI_COMMAND bits as it writes
them back out.

Upon return from calling e1000_test_msi_interrupt(), it re-enables SERR
by writing out the version of PCI_COMMAND it had previously read.

The problem with this is that e1000_test_msi_interrupt() calls
pci_disable_msi(), which eventually ends up in pci_intx(). And because
pci_intx() was called with enable set to 1, the INTX_DISABLE bit gets
cleared from PCI_COMMAND, which is what we want. But when we get back to
e1000_test_msi(), the INTX_DISABLE bit gets inadvertently re-set because
of the attempt by e1000_test_msi() to re-enable SERR.

The solution is to have e1000_test_msi() re-read the PCI_COMMAND bits as
part of its attempt to re-enable SERR.

During debugging/testing of this issue I found that not all the systems
I ran on had the SERR bit set to begin with. And on some of the systems
the same could be said for the INTX_DISABLE bit. Needless to say these
latter systems didn't have a problem falling back to legacy INTx
interrupts with the code as is.

Signed-off-by: Dean Nelson <dnelson@redhat.com>
Tested-by: Emil Tantilov <emil.s.tantilov@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

amd64_edac: Correct scrub rate setting

commit bc57117856cf1e581135810b37d3b75f9d1749f5 upstream.

Exit early when setting scrub rate on unknown/unsupported families.

Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
Acked-by: Doug Thompson <dougthompson@xmission.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

amd64_edac: Fix DCT base address selector

commit 9975a5f22a4fcc8d08035c65439900a983f891ad upstream.

The correct check is to verify whether in high range we're below 4GB
and not to extract the DctSelBaseAddr again. See "2.8.5 Routing DRAM
Requests" in the F10h BKDG.

Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
Acked-by: Doug Thompson <dougthompson@xmission.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

parisc: pass through '\t' to early (iodc) console

commit d9b68e5e88248bb24fd4e455588bea1d56108fd6 upstream.

The firmware handles '\t' internally, so stop trying to emulate it
(which, incidentally, had a bug in it.)

Fixes a really weird hang at bootup in rcu_bootup_announce, which,
as far as I can tell, is the first printk in the core kernel to use
a tab as the first character.

Signed-off-by: Kyle McMartin <kyle@parisc-linux.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

arm/imx/gpio: add spinlock protection

commit 14cb0deb66fcfca8fdbef75da8c84b5405a8c767 upstream.

The GPIO registers need protection from concurrent access for operations that
are not atomic.

Cc: Juergen Beisert <j.beisert@pengutronix.de>
Cc: Daniel Mack <daniel@caiaq.de>
Reported-by: rpkamiak@rockwellcollins.com
Signed-off-by: Baruch Siach <baruch@tkos.co.il>
Signed-off-by: Sascha Hauer <s.hauer@pengutronix.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

PARISC: led.c - fix potential stack overflow in led_proc_write()

commit 4b4fd27c0b5ec638a1f06ced9226fd95229dbbf0 upstream.

avoid potential stack overflow by correctly checking count parameter

Reported-by: Ilja <ilja@netric.org>
Signed-off-by: Helge Deller <deller@gmx.de>
Acked-by: Kyle McMartin <kyle@mcmartin.ca>
Cc: James E.J. Bottomley <jejb@parisc-linux.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

NFS: kswapd must not block in nfs_release_page

commit b608b283a962caaa280756bc8563016a71712acf upstream

See https://bugzilla.kernel.org/show_bug.cgi?id=16056

If other processes are blocked waiting for kswapd to free up some memory so
that they can make progress, then we cannot allow kswapd to block on those
processes.

Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

comedi: Uncripple 8255-based DIO subdevices

(Note: upstream comedi configuration has been overhauled, so this patch
does not apply there.)

Several comedi drivers call subdev_8255_init() (declared in
"drivers/staging/comedi/drivers/8255.h") to set up one or more DIO
subdevices.  This should be provided by the 8255.ko module, but unless
the CONFIG_COMEDI_8255 or CONFIG_COMEDI_8255_MODULE macro is defined,
the 8255.h header uses a dummy inline version of the function instead.
This means the comedi devices end up with an "unused" subdevice with 0
channels instead of a "DIO" subdevice with 24 channels!

This patch provides a non-interactive COMEDI_8255 option and selects it
whenever the COMEDI_PCI_DRIVERS or COMEDI_PCMCIA_DRIVERS options are
selected.

Signed-off-by: Ian Abbott <abbotti@mev.co.uk>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

acl trouble after upgrading ubuntu

commit d327cf7449e6fd5cbac784c641770e9366faa386 upstream.

Subject: nfs: fix acl decoding

Commit 28f566942c6b1d929f5e240e69e7081b77b238d3 "NFS: use dynamically
computed compound_hdr.replen for xdr_inline_pages offset" accidentally
changed the amount of space to allow for the acl reply, resulting in an
IO error on attempts to get an acl.

Reported-by: Paul Rudin <paul@rudin.co.uk>
Cc: Benny Halevy <bhalevy@panasas.com>
Signed-off-by: J. Bruce Fields <bfields@citi.umich.edu>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Cc: Jeremy Kerr <jeremy.kerr@canonical.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

mm: fix ia64 crash when gcore reads gate area

commit de51257aa301652876ab6e8f13ea4eadbe4a3846 upstream.

Debian's ia64 autobuilders have been seeing kernel freeze or reboot
when running the gdb testsuite (Debian bug 588574): dannf bisected to
2.6.32 62eede62dafb4a6633eae7ffbeb34c60dba5e7b1 "mm: ZERO_PAGE without
PTE_SPECIAL"; and reproduced it with gdb's gcore on a simple target.

I'd missed updating the gate_vma handling in __get_user_pages(): that
happens to use vm_normal_page() (nowadays failing on the zero page),
yet reported success even when it failed to get a page - boom when
access_process_vm() tried to copy that to its intermediate buffer.

Fix this, resisting cleanups: in particular, leave it for now reporting
success when not asked to get any pages - very probably safe to change,
but let's not risk it without testing exposure.

Why did ia64 crash with 16kB pages, but succeed with 64kB pages?
Because setup_gate() pads each 64kB of its gate area with zero pages.

Reported-by: Andreas Barth <aba@not.so.argh.org>
Bisected-by: dann frazier <dannf@debian.org>
Signed-off-by: Hugh Dickins <hughd@google.com>
Tested-by: dann frazier <dannf@dannf.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

sched: cgroup: Implement different treatment for idle shares

commit cd8ad40de36c2fe75f3b731bd70198b385895246 upstream.

When setting the weight for a per-cpu task-group, we have to put in a
phantom weight when there is no work on that cpu, otherwise we'll not
service that cpu when new work gets placed there until we again update
the per-cpu weights.

We used to add these phantom weights to the total, so that the idle
per-cpu shares don't get inflated, this however causes the non-idle
parts to get deflated, causing unexpected weight distibutions.

Reverse this, so that the non-idle shares are correct but the idle
shares are inflated.

Reported-by: Yasunori Goto <y-goto@jp.fujitsu.com>
Tested-by: Yasunori Goto <y-goto@jp.fujitsu.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <1257934048.23203.76.camel@twins>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

Linux 2.6.32.17

V4L/DVB (13830): uvcvideo: add another YUYV format GUID for iSight cameras

commit 68f194e027ecfbbc8d5515bc40787e542eed59e9 upstream.

For some unknown reason, on a MacBookPro5,3 the iSight sometimes report
a different video format GUID. This patch add the other (wrong) GUID to
the format table, making the iSight work always w/o other problems.

What it should report: 32595559-0000-0010-8000-00aa00389b71
What it often reports: 32595559-0000-0010-8000-000000389b71

Signed-off-by: Daniel Ritz <daniel.ritz@gmx.ch>
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
Cc: Leann Ogasawara <leann.ogasawara@canonical.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

Input: RX51 keymap - fix recent compile breakage

commit 2e65a2075cc740b485ab203430bdf3459d5551b6 upstream.

Commit 3fea60261e73 ("Input: twl40300-keypad - fix handling of "all
ground" rows") broke compilation as I managed to use non-existent
keycodes.

Reported-by: Arjan van de Ven <arjan@infradead.org>
Signed-off-by: Dmitry Torokhov <dtor@mail.ru>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

ath5k: initialize ah->ah_current_channel

commit b6855772f4a22c4fbdd4fcaceff5c8a527035123 upstream.

ath5k assumes ah_current_channel is always a valid pointer in
several places, but a newly created interface may not have a
channel.  To avoid null pointer dereferences, set it up to point
to the first available channel until later reconfigured.

This fixes the following oops:
$ rmmod ath5k
$ insmod ath5k
$ iw phy0 set distance 11000

BUG: unable to handle kernel NULL pointer dereference at 00000006
IP: [<d0a1ff24>] ath5k_hw_set_coverage_class+0x74/0x1b0 [ath5k]
*pde = 00000000
Oops: 0000 [#1]
last sysfs file: /sys/devices/pci0000:00/0000:00:0e.0/ieee80211/phy0/index
Modules linked in: usbhid option usb_storage usbserial usblp evdev lm90
scx200_acb i2c_algo_bit i2c_dev i2c_core via_rhine ohci_hcd ne2k_pci
8390 leds_alix2 xt_IMQ imq nf_nat_tftp nf_conntrack_tftp nf_nat_irc nf_cc

Pid: 1597, comm: iw Not tainted (2.6.32.14 #8)
EIP: 0060:[<d0a1ff24>] EFLAGS: 00010296 CPU: 0
EIP is at ath5k_hw_set_coverage_class+0x74/0x1b0 [ath5k]
EAX: 000000c2 EBX: 00000000 ECX: ffffffff EDX: c12d2080
ESI: 00000019 EDI: cf8c0000 EBP: d0a30edc ESP: cfa09bf4
  DS: 007b ES: 007b FS: 0000 GS: 0000 SS: 0068
Process iw (pid: 1597, ti=cfa09000 task=cf88a000 task.ti=cfa09000)
Stack:
  d0a34f35 d0a353f8 d0a30edc 000000fe cf8c0000 00000000 1900063d cfa8c9e0
<0> cfa8c9e8 cfa8c0c0 cfa8c000 d0a27f0c 199d84b4 cfa8c200 00000010 d09bfdc7
<0> 00000000 00000000 ffffffff d08e0d28 cf9263c0 00000001 cfa09cc4 00000000
Call Trace:
  [<d0a27f0c>] ? ath5k_hw_attach+0xc8c/0x3c10 [ath5k]
  [<d09bfdc7>] ? __ieee80211_request_smps+0x1347/0x1580 [mac80211]
  [<d08e0d28>] ? nl80211_send_scan_start+0x7b8/0x4520 [cfg80211]
  [<c10f5db9>] ? nla_parse+0x59/0xc0
  [<c11ca8d9>] ? genl_rcv_msg+0x169/0x1a0
  [<c11ca770>] ? genl_rcv_msg+0x0/0x1a0
  [<c11c7e68>] ? netlink_rcv_skb+0x38/0x90
  [<c11c9649>] ? genl_rcv+0x19/0x30
  [<c11c7c03>] ? netlink_unicast+0x1b3/0x220
  [<c11c893e>] ? netlink_sendmsg+0x26e/0x290
  [<c11a409e>] ? sock_sendmsg+0xbe/0xf0
  [<c1032780>] ? autoremove_wake_function+0x0/0x50
  [<c104d846>] ? __alloc_pages_nodemask+0x106/0x530
  [<c1074933>] ? do_lookup+0x53/0x1b0
  [<c10766f9>] ? __link_path_walk+0x9b9/0x9e0
  [<c11acab0>] ? verify_iovec+0x50/0x90
  [<c11a42b1>] ? sys_sendmsg+0x1e1/0x270
  [<c1048e50>] ? find_get_page+0x10/0x50
  [<c104a96f>] ? filemap_fault+0x5f/0x370
  [<c1059159>] ? __do_fault+0x319/0x370
  [<c11a55b4>] ? sys_socketcall+0x244/0x290
  [<c101962c>] ? do_page_fault+0x1ec/0x270
  [<c1019440>] ? do_page_fault+0x0/0x270
  [<c1002ae5>] ? syscall_call+0x7/0xb
Code: 00 b8 fe 00 00 00 b9 f8 53 a3 d0 89 5c 24 14 89 7c 24 10 89 44 24
0c 89 6c 24 08 89 4c 24 04 c7 04 24 35 4f a3 d0 e8 7c 30 60 f0 <0f> b7
43 06 ba 06 00 00 00 a8 10 75 0e 83 e0 20 83 f8 01 19 d2
EIP: [<d0a1ff24>] ath5k_hw_set_coverage_class+0x74/0x1b0 [ath5k] SS:ESP
0068:cfa09bf4
CR2: 0000000000000006
---[ end trace 54f73d6b10ceb87b ]---

Reported-by: Steve Brown <sbrown@cortland.com>
Signed-off-by: Bob Copeland <me@bobcopeland.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

ecryptfs: Bugfix for error related to ecryptfs_hash_buckets

commit a6f80fb7b5986fda663d94079d3bba0937a6b6ff upstream.

The function ecryptfs_uid_hash wrongly assumes that the
second parameter to hash_long() is the number of hash
buckets instead of the number of hash bits.
This patch fixes that and renames the variable
ecryptfs_hash_buckets to ecryptfs_hash_bits to make it
clearer.

Fixes: CVE-2010-2492
Signed-off-by: Andre Osterhues <aosterhues@escrypt.com>
Signed-off-by: Tyler Hicks <tyhicks@linux.vnet.ibm.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

Fix spinaphore down_spin()

commit b70f4e85bfc4d7000036355b714a92d5c574f1be upstream.

Typo in down_spin() meant it only read the low 32 bits of the
"serve" value, instead of the full 64 bits. This results in the
system hanging when the values in ticket/serve get larger than
32-bits. A big enough system running the right test can hit this
in a just a few hours.

Broken since 883a3acf5b0d4782ac35981227a0d094e8b44850
    [IA64] Re-implement spinaphores using ticket lock concepts

Reported via IRC by Bjorn Helgaas

Signed-off-by: Tony Luck <tony.luck@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

eeepc-laptop: check wireless hotplug events

commit bc9d24a3aeb1532fc3e234907a8b6d671f7ed68f upstream.

Before we mark the wireless device as unplugged, check PCI config space
to see whether the wireless device is really disabled (and vice versa).
This works around newer models which don't want the hotplug code, where
we end up disabling the wired network device.

My old 701 still works correctly with this.  I can also simulate an
afflicted model by changing the hardcoded PCI bus/slot number in the
driver, and it seems to work nicely (although it is a bit noisy).

In future this type of hotplug support will be implemented by the PCI
core.  The existing blacklist and the new warning message will be
removed at that point.

Signed-off-by: Alan Jenkins <alan-jenkins@tuffmail.co.uk>
Signed-off-by: Corentin Chary <corentincj@iksaif.net>
Cc: Tim Gardner <tim.gardner@canonical.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

kbuild: Fix modpost segfault

commit 1c938663d58b5b2965976a6f54cc51b5d6f691aa upstream.

Alan <alan@clueserver.org> writes:

> program: /home/alan/GitTrees/linux-2.6-mid-ref/scripts/mod/modpost -o
> Module.symvers -S vmlinux.o
>
> Program received signal SIGSEGV, Segmentation fault.

It just hit me.
It's the offset calculation in reloc_location() which overflows:
        return (void *)elf->hdr + sechdrs[section].sh_offset +
               (r->r_offset - sechdrs[section].sh_addr);

E.g. for the first rodata r entry:
r->r_offset < sechdrs[section].sh_addr
and the expression in the parenthesis produces 0xFFFFFFE0 or something
equally wise.

Reported-by: Alan <alan@clueserver.org>
Signed-off-by: Krzysztof Hałasa <khc@pm.waw.pl>
Tested-by: Alan <alan@clueserver.org>
Signed-off-by: Michal Marek <mmarek@suse.cz>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

iwlagn: verify flow id in compressed BA packet

commit b561e8274f75831ee87e4ea378cbb1f9f050a51a upstream.

The flow id (scd_flow) in a compressed BA packet should match the txq_id
of the queue from which the aggregated packets were sent. However, in
some hardware like the 1000 series, sometimes the flow id is 0 for the
txq_id (10 to 19). This can cause the annoying message:
[ 2213.306191] iwlagn 0000:01:00.0: Received BA when not expected
[ 2213.310178] iwlagn 0000:01:00.0: Read index for DMA queue txq id (0),
index 5, is out of range [0-256] 7 7.

And even worse, if agg->wait_for_ba is true when the bad BA is arriving,
this can cause system hang due to NULL pointer dereference because the
code is operating in a wrong tx queue!

Signed-off-by: Shanyu Zhao <shanyu.zhao@intel.com>
Signed-off-by: Pradeep Kulkarni <pradeepx.kulkarni@intel.com>
Signed-off-by: Reinette Chatre <reinette.chatre@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

V4L/DVB: uvcvideo: Add support for V4L2_PIX_FMT_Y16

commit 61421206833a4085d9bdf35b2b84cd9a67dfdfac upstream.

The Miricle 307K (17dc:0202) camera reports a 16-bit greyscale format,
support it in the driver.

Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

V4L/DVB: uvcvideo: Add support for Packard Bell EasyNote MX52 integrated webcam

commit f129b03ba272c86c42ad476684caa0d6109cb383 upstream.

The camera requires the STREAM_NO_FID quirk. Add a corresponding entry
in the device IDs list.

Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

V4L/DVB: uvcvideo: Add support for unbranded Arkmicro 18ec:3290 webcams

commit 1e4d05bc95a0fe2972c5c91ed45466587d07cd2c upstream.

The camera requires the PROBE_DEF quirk. Add a corresponding entry in
the device IDs list.

Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

mac80211: fix supported rates IE if AP doesn't give us it's rates

commit 76f273640134f3eb8257179cd5b3bc6ba5fe4a96 upstream.

If AP do not provide us supported rates before assiociation, send
all rates we are supporting instead of empty information element.

v1 -> v2: Add comment.

Signed-off-by: Stanislaw Gruszka <sgruszka@redhat.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

cxgb3: fix linkup issue

commit 9441cad99b4b09d6b627351c2d282833868c116c upstream.

I encountered an issue that not to link up on cxgb3 fabric.
I bisected and found that this regression was introduced by
0f07c4ee8c800923ae7918c231532a9256233eed.

Correct to pass phy_addr to cphy_init() at t3_xaui_direct_phy_prep().

Signed-off-by: Hiroshi Shimamoto <h-shimamoto@ct.jp.nec.com>
Acked-by: Divy Le Ray <divy@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

V4L/DVB: budget: Select correct frontends

commit d46b36e7f927772bb72524dc9f1e384e3cb4a975 upstream.

Update the Kconfig selections to match the code.

Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

MIPS FPU emulator: allow Cause bits of FCSR to be writeable by ctc1

commit 95e8f634d7a3ea5af40ec3fa42c8a152fd3a0624 upstream.

    In the FPU emulator code of the MIPS, the Cause bits of the FCSR register
    are not currently writeable by the ctc1 instruction.  In odd corner cases,
    this can cause problems.  For example, a case existed where a divide-by-zero
    exception was generated by the FPU, and the signal handler attempted to
    restore the FPU registers to their state before the exception occurred.  In
    this particular setup, writing the old value to the FCSR register would
    cause another divide-by-zero exception to occur immediately.  The solution
    is to change the ctc1 instruction emulator code to allow the Cause bits of
    the FCSR register to be writeable.  This is the behaviour of the hardware
    that the code is emulating.

    This problem was found by Shane McDonald, but the credit for the fix goes
    to Kevin Kissell.  In Kevin's words:

    I submit that the bug is indeed in that ctc_op:  case of the emulator.  The
    Cause bits (17:12) are supposed to be writable by that instruction, but the
    CTC1 emulation won't let them be updated by the instruction.  I think that
    actually if you just completely removed lines 387-388 [...] things would
    work a good deal better.  At least, it would be a more accurate emulation of
    the architecturally defined FPU.  If I wanted to be really, really pedantic
    (which I sometimes do), I'd also protect the reserved bits that aren't
    necessarily writable.

Signed-off-by: Shane McDonald <mcdonald.shane@gmail.com>
To: anemo@mba.ocn.ne.jp
To: kevink@paralogos.com
To: sshtylyov@mvista.com
Patchwork: http://patchwork.linux-mips.org/patch/1205/
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Cc: Ben Hutchings <ben@decadent.org.uk>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

staging: rtl8192su: add Support for Belkin F5D8053 v6

commit d615da093eb0f691a73a754589e2a4a24a6f1ca7 upstream.

Please find attached a patch which adds the device ID for the Belkin
F5D8053 v6 to the rtl8192su driver. I've tested this in 2.6.34-rc3
(Ubuntu 9.10 amd64) and the network adapter is working flawlessly.

Signed-off-by: Richard Airlie <richard@backtrace.co.uk>
Cc: Ben Hutchings <ben@decadent.org.uk>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

forcedeth: fix tx limit2 flag check

commit 5c659322a904a7cc0265e7b183372b9bdebec6db upstream.

This is a fix for bug 572201 @ bugs.debian.org

This patch fixes the TX_LIMIT feature flag. The previous logic check
for TX_LIMIT2 also took into account a device that only had TX_LIMIT
set.

Reported-by: Stephen Mulcahu <stephen.mulcahy@deri.org>
Reported-by: Ben Huchings <ben@decadent.org.uk>
Signed-off-by: Ayaz Abdulla <aabdulla@nvidia.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Cc: Ben Hutchings <ben@decadent.org.uk>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

V4L/DVB: FusionHDTV: Use quick reads for I2C IR device probing

commit 806b07c29b711aaf90c81d2a19711607769f8246 upstream.

IR support on FusionHDTV cards is broken since kernel 2.6.31. One side
effect of the switch to the standard binding model for IR I2C devices
was to let i2c-core do the probing instead of the ir-kbd-i2c driver.
There is a slight difference between the two probe methods: i2c-core
uses 0-byte writes, while the ir-kbd-i2c was using 0-byte reads. As
some IR I2C devices only support reads, the new probe method fails to
detect them.

For now, revert to letting the driver do the probe, using 0-byte
reads. In the future, i2c-core will be extended to let callers of
i2c_new_probed_device() provide a custom probing function.

Signed-off-by: Jean Delvare <khali@linux-fr.org>
Tested-by: "Timothy D. Lenz" <tlenz@vorgon.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

V4L/DVB: dvb-core: Fix ULE decapsulation bug

commit 5c331fc8c19e181bffab46e9d18e1637cdc47170 upstream.

Fix ULE decapsulation bug when less than 4 bytes of ULE SNDU is packed
into the remaining bytes of a MPEG2-TS frame

ULE (Unidirectional Lightweight Encapsulation RFC 4326) decapsulation
code has a bug that incorrectly treats ULE SNDU packed into the
remaining 2 or 3 bytes of a MPEG2-TS frame as having invalid pointer
field on the subsequent MPEG2-TS frame.

Signed-off-by: Ang Way Chuang <wcang@nav6.org>
Acked-by: Jarod Wilson <jarod@redhat.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

ALSA: Echoaudio, fix Guru Meditation #00000005.48454C50

commit b721e68bdc5b39c51bf6a1469f8d3663fbe03243 upstream.

This patch fixes a division by zero error in the irq handler.

There is a small window between the hw_params() callback and when
runtime->frame_bits is set by ALSA middle layer. When another substream is
already running, if an interrupt is delivered during that window the irq
handler calls pcm_pointer() which does a division by zero. The patch below
makes the irq handler skip substreams that are initialized but not started
yet. Cc to Clemens Ladisch because he proposed an alternate fix.

For more information, please read the original thread in the linux-kernel
mailing list: http://lkml.org/lkml/2010/2/2/187

Signed-off-by: Giuliano Pochini <pochini@shiny.it>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

revert "[CPUFREQ] remove rwsem lock from CPUFREQ_GOV_STOP call (second call site)"

commit accd846698439ba18250e8fd5681af280446b853 upstream.

395913d0b1db37092ea3d9d69b832183b1dd84c5 ("[CPUFREQ] remove rwsem lock
from CPUFREQ_GOV_STOP call (second call site)") is not needed, because
there is no rwsem lock in cpufreq_ondemand and cpufreq_conservative
anymore.  Lock should not be released until the work done.

Addresses https://bugzilla.kernel.org/show_bug.cgi?id=1594

Signed-off-by: Andrej Gelenberg <andrej.gelenberg@udo.edu>
Cc: Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>
Cc: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: Dave Jones <davej@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

firmware_class: fix memory leak - free allocated pages

commit dd336c554d8926c3348a2d5f2a5ef5597f6d1a06 upstream.

fix memory leak introduced by the patch 6e03a201bbe:
firmware: speed up request_firmware()

1. vfree won't release pages there were allocated explicitly and mapped
using vmap. The memory has to be vunmap-ed and the pages needs
to be freed explicitly

2. page array is moved into the 'struct
firmware' so that we can free it from release_firmware()
and not only in fw_dev_release()

The fix doesn't break the firmware load speed.

Cc: Johannes Berg <johannes@sipsolutions.net>
Cc: Ming Lei <tom.leiming@gmail.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Singed-off-by: Kay Sievers <kay.sievers@vrfy.org>
Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>
Signed-off-by: Tomas Winkler <tomas.winkler@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

dm9000: fix "BUG: spinlock recursion"

commit 380fefb2ddabd4cd5f14dbe090481f0544e65078 upstream.

dm9000_set_rx_csum and dm9000_hash_table are called from atomic context (in
dm9000_init_dm9000), and from non-atomic context (via ethtool_ops and
net_device_ops respectively). This causes a spinlock recursion BUG. Fix this by
renaming these functions to *_unlocked for the atomic context, and make the
original functions locking wrappers for use in the non-atomic context.

Signed-off-by: Baruch Siach <baruch@tkos.co.il>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

libertas/sdio: 8686: set ECSI bit for 1-bit transfers

commit 8a64c0f6b7ec7f758c4ef445e49f479e27fa2236 upstream.

When operating in 1-bit mode, SDAT1 is used as dedicated interrupt line.
However, the 8686 will only drive this line when the ECSI bit is set in
the CCCR_IF register.

Thanks to Alagu Sankar for pointing me in the right direction.

Signed-off-by: Daniel Mack <daniel@caiaq.de>
Cc: Alagu Sankar <alagusankar@embwise.com>
Cc: Volker Ernst <volker.ernst@txtr.com>
Cc: Dan Williams <dcbw@redhat.com>
Cc: John W. Linville <linville@tuxdriver.com>
Cc: Holger Schurig <hs4233@mail.mn-solutions.de>
Cc: Bing Zhao <bzhao@marvell.com>
Cc: libertas-dev@lists.infradead.org
Cc: linux-wireless@vger.kernel.org
Cc: linux-mmc@vger.kernel.org
Acked-by: Dan Williams <dcbw@redhat.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

ACPI: Unconditionally set SCI_EN on resume

commit b6dacf63e9fb2e7a1369843d6cef332f76fca6a3 upstream.

The ACPI spec tells us that the firmware will reenable SCI_EN on resume.
Reality disagrees in some cases. The ACPI spec tells us that the only way
to set SCI_EN is via an SMM call.
https://bugzilla.kernel.org/show_bug.cgi?id=13745 shows us that doing so
may break machines. Tracing the ACPI calls made by Windows shows that it
unconditionally sets SCI_EN on resume with a direct register write, and
therefore the overwhelming probability is that everything is fine with
this behaviour.

Signed-off-by: Matthew Garrett <mjg@redhat.com>
Tested-by: Rafael J. Wysocki <rjw@sisk.pl>
Signed-off-by: Len Brown <len.brown@intel.com>
Cc: Kamal Mostafa <kamal@canonical.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

ACPI: skip checking BM_STS if the BIOS doesn't ask for it

commit 718be4aaf3613cf7c2d097f925abc3d3553c0605 upstream.

It turns out that there is a bit in the _CST for Intel FFH C3
that tells the OS if we should be checking BM_STS or not.

Linux has been unconditionally checking BM_STS.
If the chip-set is configured to enable BM_STS,
it can retard or completely prevent entry into
deep C-states -- as illustrated by turbostat:

http://userweb.kernel.org/~lenb/acpi/utils/pmtools/turbostat/

ref: Intel Processor Vendor-Specific ACPI Interface Specification
table 4 "_CST FFH GAS Field Encoding"
Bit 1: Set to 1 if OSPM should use Bus Master avoidance for this C-state

https://bugzilla.kernel.org/show_bug.cgi?id=15886

Signed-off-by: Len Brown <len.brown@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

PM / x86: Save/restore MISC_ENABLE register

commit 85a0e7539781dad4bfcffd98e72fa9f130f4e40d upstream.

Save/restore MISC_ENABLE register on suspend/resume.
This fixes OOPS (invalid opcode) on resume from STR on Asus P4P800-VM,
which wakes up with MWAIT disabled.

Fixes https://bugzilla.kernel.org/show_bug.cgi?id=15385

Signed-off-by: Ondrej Zary <linux@rainbow-software.org>
Tested-by: Alan Stern <stern@rowland.harvard.edu>
Acked-by: H. Peter Anvin <hpa@linux.intel.com>
Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

GFS2: Fix up system xattrs

commit 2646a1f61a3b5525914757f10fa12b5b94713648 upstream.

This code has been shamelessly stolen from XFS at the suggestion
of Christoph Hellwig. I've not added support for cached ACLs so
far... watch for that in a later patch, although this is designed
in such a way that they should be easy to add.

Signed-off-by: Steven Whitehouse <swhiteho@redhat.com>
Cc: Christoph Hellwig <hch@infradead.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

futex: futex_find_get_task remove credentails check

commit 7a0ea09ad5352efce8fe79ed853150449903b9f5 upstream.

futex_find_get_task is currently used (through lookup_pi_state) from two
contexts, futex_requeue and futex_lock_pi_atomic.  None of the paths
looks it needs the credentials check, though.  Different (e)uids
shouldn't matter at all because the only thing that is important for
shared futex is the accessibility of the shared memory.

The credentail check results in glibc assert failure or process hang (if
glibc is compiled without assert support) for shared robust pthread
mutex with priority inheritance if a process tries to lock already held
lock owned by a process with a different euid:

pthread_mutex_lock.c:312: __pthread_mutex_lock_full: Assertion `(-(e)) != 3 || !robust' failed.

The problem is that futex_lock_pi_atomic which is called when we try to
lock already held lock checks the current holder (tid is stored in the
futex value) to get the PI state.  It uses lookup_pi_state which in turn
gets task struct from futex_find_get_task.  ESRCH is returned either
when the task is not found or if credentials check fails.

futex_lock_pi_atomic simply returns if it gets ESRCH.  glibc code,
however, doesn't expect that robust lock returns with ESRCH because it
should get either success or owner died.

Signed-off-by: Michal Hocko <mhocko@suse.cz>
Acked-by: Darren Hart <dvhltc@us.ibm.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
Cc: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

KVM: MMU: invalidate and flush on spte small->large page size change

Always invalidate spte and flush TLBs when changing page size, to make
sure different sized translations for the same address are never cached
in a CPU's TLB.

Currently the only case where this occurs is when a non-leaf spte pointer is
overwritten by a leaf, large spte entry. This can happen after dirty
logging is disabled on a memslot, for example.

Noticed by Andrea.

KVM-Stable-Tag
Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
(cherry picked from commit 3be2264be3c00865116f997dc53ebcc90fe7fc4b)

KVM: SVM: Implement workaround for Erratum 383

This patch implements a workaround for AMD erratum 383 into
KVM. Without this erratum fix it is possible for a guest to
kill the host machine. This patch implements the suggested
workaround for hypervisors which will be published by the
next revision guide update.

[jan: fix overflow warning on i386]
[xiao: fix unused variable warning]

Cc: stable@kernel.org
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>
Signed-off-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
(cherry picked from commit 67ec66077799f2fef84b21a643912b179c422281)

KVM: SVM: Handle MCEs early in the vmexit process

This patch moves handling of the MC vmexits to an earlier
point in the vmexit. The handle_exit function is too late
because the vcpu might alreadry have changed its physical
cpu.

Cc: stable@kernel.org
Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
(cherry picked from commit fe5913e4e1700cbfc337f4b1da9ddb26f6a55586)

KVM: MMU: Remove user access when allowing kernel access to gpte.w=0 page

If cr0.wp=0, we have to allow the guest kernel access to a page with pte.w=0.
We do that by setting spte.w=1, since the host cr0.wp must remain set so the
host can write protect pages.  Once we allow write access, we must remove
user access otherwise we mistakenly allow the user to write the page.

Reviewed-by: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Avi Kivity <avi@redhat.com>
(cherry picked from commit 69325a122580d3a7b26589e8efdd6663001c3297)

ext4: Make fsync sync new parent directories in no-journal mode

commit 14ece1028b3ed53ffec1b1213ffc6acaf79ad77c upstream (as of v2.6.34-git13)

Add a new ext4 state to tell us when a file has been newly created; use
that state in ext4_sync_file in no-journal mode to tell us when we need
to sync the parent directory as well as the inode and data itself.  This
fixes a problem in which a panic or power failure may lose the entire
file even when using fsync, since the parent directory entry is lost.

Addresses-Google-Bug: #2480057

Signed-off-by: Frank Mayhar <fmayhar@google.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

ext4: Fix compat EXT4_IOC_ADD_GROUP

commit 4d92dc0f00a775dc2e1267b0e00befb783902fe7 upstream (as of v2.6.34-git13)

struct ext4_new_group_input needs to be converted because u64 has
only 32-bit alignment on some 32-bit architectures, notably i386.

Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

ext4: Conditionally define compat ioctl numbers

commit 899ad0cea6ad7ff4ba24b16318edbc3cbbe03fad upstream (as of v2.6.34-git13)

It is unnecessary, and in general impossible, to define the compat
ioctl numbers except when building the filesystem with CONFIG_COMPAT
defined.

Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

ext4: restart ext4_ext_remove_space() after transaction restart

commit 0617b83fa239db9743a18ce6cc0e556f4d0fd567 upstream (as of v2.6.34-git13)

If i_data_sem was internally dropped due to transaction restart, it is
necessary to restart path look-up because extents tree was possibly
modified by ext4_get_block().

https://bugzilla.kernel.org/show_bug.cgi?id=15827

Signed-off-by: Dmitry Monakhov <dmonakhov@openvz.org>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Acked-by: Jan Kara <jack@suse.cz>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

ext4: Clear the EXT4_EOFBLOCKS_FL flag only when warranted

commit 786ec7915e530936b9eb2e3d12274145cab7aa7d upstream (as of v2.6.34-git13)

Dimitry Monakhov discovered an edge case where it was possible for the
EXT4_EOFBLOCKS_FL flag could get cleared unnecessarily.  This is true;
I have a test case that can be exercised via downloading and
decompressing the file:

wget ftp://ftp.kernel.org/pub/linux/kernel/people/tytso/ext4-testcases/eofblocks-fl-test-case.img.bz2
bunzip2 eofblocks-fl-test-case.img
dd if=/dev/zero of=eofblocks-fl-test-case.img bs=1k seek=17925 bs=1k count=1 conv=notrunc

However, triggering it in real life is highly unlikely since it
requires an extremely fragmented sparse file with a hole in exactly
the right place in the extent tree.  (It actually took quite a bit of
work to generate this test case.)  Still, it's nice to get even
extreme corner cases to be correct, so this patch makes sure that we
don't clear the EXT4_EOFBLOCKS_FL incorrectly even in this corner
case.

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

ext4: Avoid crashing on NULL ptr dereference on a filesystem error

commit f70f362b4a6fe47c239dbfb3efc0cc2c10e4f09c upstream (as of v2.6.34-git13)

If the EOFBLOCK_FL flag is set when it should not be and the inode is
zero length, then eh_entries is zero, and ex is NULL, so dereferencing
ex to print ex->ee_block causes a kernel OOPS in
ext4_ext_map_blocks().

On top of that, the error message which is printed isn't very helpful.
So we fix this by printing something more explanatory which doesn't
involve trying to print ex->ee_block.

Addresses-Google-Bug: #2655740

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

ext4: Use bitops to read/modify i_flags in struct ext4_inode_info

commit 12e9b892002d9af057655d35b44db8ee9243b0dc upstream (as of v2.6.34-git13)

At several places we modify EXT4_I(inode)->i_flags without holding
i_mutex (ext4_do_update_inode, ...). These modifications are racy and
we can lose updates to i_flags. So convert handling of i_flags to use
bitops which are atomic.

https://bugzilla.kernel.org/show_bug.cgi?id=15792

Signed-off-by: Dmitry Monakhov <dmonakhov@openvz.org>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

ext4: Show journal_checksum option

commit 39a4bade8c1826b658316d66ee81c09b0a4d7d42 upstream (as of v2.6.34-git13)

We failed to show journal_checksum option in /proc/mounts. Fix it.

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

ext4: check for a good block group before loading buddy pages

commit 8a57d9d61a6e361c7bb159dda797672c1df1a691 upstream (as of v2.6.34-git13)

This adds a new field in ext4_group_info to cache the largest available
block range in a block group; and don't load the buddy pages until *after*
we've done a sanity check on the block group.

With large allocation requests (e.g., fallocate(), 8MiB) and relatively full
partitions, it's easy to have no block groups with a block extent large
enough to satisfy the input request length.  This currently causes the loop
during cr == 0 in ext4_mb_regular_allocator() to load the buddy bitmap pages
for EVERY block group.  That can be a lot of pages.  The patch below allows
us to call ext4_mb_good_group() BEFORE we load the buddy pages (although we
have check again after we lock the block group).

Addresses-Google-Bug: #2578108
Addresses-Google-Bug: #2704453

Signed-off-by: Curt Wohlgemuth <curtw@google.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

ext4: Prevent creation of files larger than RLIMIT_FSIZE using fallocate

commit 6d19c42b7cf81c39632b6d4dbc514e8449bcd346 upstream (as of v2.6.34-git13)

Currently using posix_fallocate one can bypass an RLIMIT_FSIZE limit
and create a file larger than the limit. Add a check for that.

Signed-off-by: Nikanth Karthikesan <knikanth@suse.de>
Signed-off-by: Amit Arora <aarora@in.ibm.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

ext4: Remove extraneous newlines in ext4_msg() calls

commit fbe845ddf368f77f86aa7500f8fd2690f54c66a8 upstream (as of v2.6.34-git13)

Addresses-Google-Bug: #2562325

Signed-off-by: Curt Wohlgemuth <curtw@google.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

ext4: init statistics after journal recovery

commit 84061e07c5fbbbf9dc8aef8fb750fc3a2dfc31f3 upstream (as of v2.6.34-git13)

Currently block/inode/dir counters initialized before journal was
recovered. In fact after journal recovery this info will probably
change. And freeblocks it critical for correct delalloc mode
accounting.

https://bugzilla.kernel.org/show_bug.cgi?id=15768

Signed-off-by: Dmitry Monakhov <dmonakhov@openvz.org>
Acked-by: Jan Kara <jack@suse.cz>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

ext4: clean up inode bitmaps manipulation in ext4_free_inode

commit d17413c08cd2b1dd2bf2cfdbb0f7b736b2b2b15c upstrea (as of v2..34-git13)

- Reorganize locking scheme to batch two atomic operation in to one.
  This also allow us to state what healthy group must obey following rule
  ext4_free_inodes_count(sb, gdp) == ext4_count_free(inode_bitmap, NUM);
- Fix possible undefined pointer dereference.
- Even if group descriptor stats aren't accessible we have to update
  inode bitmaps.
- Move non-group members update out of group_lock.

Note: this commit has been observed to fix fs corruption problems
under heavy fs load

Signed-off-by: Dmitry Monakhov <dmonakhov@openvz.org>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

ext4: Do not zero out uninitialized extents beyond i_size

commit 21ca087a3891efab4d45488db8febee474d26c68 upstream (as of v2.6.34-git13)

The extents code will sometimes zero out blocks and mark them as
initialized instead of splitting an extent into several smaller ones.
This optimization however, causes problems if the extent is beyond
i_size because fsck will complain if there are uninitialized blocks
after i_size as this can not be distinguished from an inode that has
an incorrect i_size field.

https://bugzilla.kernel.org/show_bug.cgi?id=15742

Signed-off-by: Dmitry Monakhov <dmonakhov@openvz.org>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

ext4: don't scan/accumulate more pages than mballoc will allocate

commit c445e3e0a5c2804524dec6e55f66d63f6bc5bc3e upstream (as of v2.6.34-git13)

There was a bug reported on RHEL5 that a 10G dd on a 12G box
had a very, very slow sync after that.

At issue was the loop in write_cache_pages scanning all the way
to the end of the 10G file, even though the subsequent call
to mpage_da_submit_io would only actually write a smallish amt; then
we went back to the write_cache_pages loop ... wasting tons of time
in calling __mpage_da_writepage for thousands of pages we would
just revisit (many times) later.

Upstream it's not such a big issue for sys_sync because we get
to the loop with a much smaller nr_to_write, which limits the loop.

However, talking with Aneesh he realized that fsync upstream still
gets here with a very large nr_to_write and we face the same problem.

This patch makes mpage_add_bh_to_extent stop the loop after we've
accumulated 2048 pages, by setting mpd->io_done = 1; which ultimately
causes the write_cache_pages loop to break.

Repeating the test with a dirty_ratio of 80 (to leave something for
fsync to do), I don't see huge IO performance gains, but the reduction
in cpu usage is striking: 80% usage with stock, and 2% with the
below patch.  Instrumenting the loop in write_cache_pages clearly
shows that we are wasting time here.

Eventually we need to change mpage_da_map_pages() also submit its I/O
to the block layer, subsuming mpage_da_submit_io(), and then change it
call ext4_get_blocks() multiple times.

Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

ext4: stop issuing discards if not supported by device

commit a30eec2a8650a77f754e84b2e15f062fe652baa7 upstream (as of v2.6.34-git13)

Turn off issuance of discard requests if the device does
not support it - similar to the action we take for barriers.
This will save a little computation time if a non-discardable
device is mounted with -o discard, and also makes it obvious
that it's not doing what was asked at mount time ...

Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

ext4: don't return to userspace after freezing the fs with a mutex held

commit 6b0310fbf087ad6e9e3b8392adca97cd77184084 upstream (as of v2.6.34-git13)

ext4_freeze() used jbd2_journal_lock_updates() which takes
the j_barrier mutex, and then returns to userspace.  The
kernel does not like this:

================================================
[ BUG: lock held when returning to user space! ]
------------------------------------------------
lvcreate/1075 is leaving the kernel with locks still held!
1 lock held by lvcreate/1075:
 #0:  (&journal->j_barrier){+.+...}, at: [<ffffffff811c6214>]
jbd2_journal_lock_updates+0xe1/0xf0

Use vfs_check_frozen() added to ext4_journal_start_sb() and
ext4_force_commit() instead.

Addresses-Red-Hat-Bugzilla: #568503

Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

ext4: check s_log_groups_per_flex in online resize code

commit 42007efd569f1cf3bfb9a61da60ef6c2179508ca upstream (as of v2.6.34-git13)

If groups_per_flex < 2, sbi->s_flex_groups[] doesn't get filled out,
and every other access to this first tests s_log_groups_per_flex;
same thing needs to happen in resize or we'll wander off into
a null pointer when doing an online resize of the file system.

Thanks to Christoph Biedl, who came up with the trivial testcase:

# truncate --size 128M fsfile
# mkfs.ext3 -F fsfile
# tune2fs -O extents,uninit_bg,dir_index,flex_bg,huge_file,dir_nlink,extra_isize fsfile
# e2fsck -yDf -C0 fsfile
# truncate --size 132M fsfile
# losetup /dev/loop0 fsfile
# mount /dev/loop0 mnt
# resize2fs -p /dev/loop0

https://bugzilla.kernel.org/show_bug.cgi?id=13549

Reported-by: Alessandro Polverini <alex@nibbles.it>
Test-case-by: Christoph Biedl <bugzilla.kernel.bpeb@manchmal.in-ulm.de>
Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

ext4: fix quota accounting in case of fallocate

commit 35121c9860316d7799cea0fbc359a9186e7c2747 upstream (as of v2.6.34-git13)

allocated_meta_data is already included in 'used' variable.

Signed-off-by: Dmitry Monakhov <dmonakhov@openvz.org>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

ext4: allow defrag (EXT4_IOC_MOVE_EXT) in 32bit compat mode

commit b684b2ee9409f2890a8b3aea98525bbe5f84e276 upstream (as of v2.6.34-git13)

I have an x86_64 kernel with i386 userspace. e4defrag fails on the
EXT4_IOC_MOVE_EXT ioctl because it is not wired up for the compat
case. It seems that struct move_extent is compat save, only types
with fixed widths are used:
{
        __u32 reserved;         /* should be zero */
        __u32 donor_fd;         /* donor file descriptor */
        __u64 orig_start;       /* logical start offset in block for orig */
        __u64 donor_start;      /* logical start offset in block for donor */
        __u64 len;              /* block length to be moved */
        __u64 moved_len;        /* moved block length */
};

Lets just wire up EXT4_IOC_MOVE_EXT for the compat case.

Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Reviewed-by: Eric Sandeen <sandeen@redhat.com>
CC: Akira Fujita <a-fujita@rs.jp.nec.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

ext4: rename ext4_mb_release_desc() to ext4_mb_unload_buddy()

commit e39e07fdfd98be8650385f12a7b81d6adc547510 upstream (as of v2.6.34-git13)

This function cleans up after ext4_mb_load_buddy(), so the renaming
makes the code clearer.

Signed-off-by: Jing Zhang <zj.barak@gmail.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

ext4: Remove unnecessary call to ext4_get_group_desc() in mballoc

commit 62e823a2cba18509ee826d775270e8ef9071b5bc upstream (as of v2.6.34-git13)

Signed-off-by: Jing Zhang <zj.barak@gmail.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

ext4: fix memory leaks in error path handling of ext4_ext_zeroout()

commit b720303df7352d4a7a1f61e467e0a124913c0d41 upstream (as of v2.6.34-git13)

When EIO occurs after bio is submitted, there is no memory free
operation for bio, which results in memory leakage. And there is also
no check against bio_alloc() for bio.

Acked-by: Dave Kleikamp <shaggy@linux.vnet.ibm.com>
Signed-off-by: Jing Zhang <zj.barak@gmail.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

ext4: check missed return value in ext4_sync_file()

commit 0671e704658b9f26f85e78d51176daa861f955c7 upstream (as of v2.6.34-git13)

Signed-off-by: Dmitry Monakhov <dmonakhov@openvz.org>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

ext4: Issue the discard operation *before* releasing the blocks to be reused

commit b90f687018e6d6c77d981b09203780f7001407e5 upstream (as of v2.6.34-rc6)

Otherwise, we can end up having data corruption because the blocks
could get reused and then discarded!

https://bugzilla.kernel.org/show_bug.cgi?id=15579

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

ext4: Fix buffer head leaks after calls to ext4_get_inode_loc()

commit fd2dd9fbaf9e498ec63eef298921e36556f7214c upstream (as of v2.6.34-rc6)

Calls to ext4_get_inode_loc() returns with a reference to a buffer
head in iloc->bh.  The callers of this function in ext4_write_inode()
when in no journal mode and in ext4_xattr_fiemap() don't release the
buffer head after using it.

Addresses-Google-Bug: #2548165

Signed-off-by: Curt Wohlgemuth <curtw@google.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

ext4: Fix possible lost inode write in no journal mode

commit 8b472d739b2ddd8ab7fb278874f696cd95b25a5e upstream (as of v2.6.34-rc6)

In the no-journal case, ext4_write_inode() will fetch the bh and call
sync_dirty_buffer() on it.  However, if the bh has already been
written and the bh reclaimed for some other purpose, AND if the inode
is the only one in the inode table block in use, then
ext4_get_inode_loc() will not read the inode table block from disk,
but as an optimization, fill the block with zero's assuming that its
caller will copy in the on-disk version of the inode.  This is not
done by ext4_write_inode(), so the contents of the inode can simply
get lost.  The fix is to use __ext4_get_inode_loc() with in_mem set to
0, instead of ext4_get_inode_loc().  Long term the API needs to be
fixed so it's obvious why latter is not safe.

Addresses-Google-Bug: #2526446

Signed-off-by: Curt Wohlgemuth <curtw@google.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

ext4: Fixed inode allocator to correctly track a flex_bg's used_dirs

commit c4caae25187ff3f5e837c6f04eb1acc2723c72d3 upstream (as of v2.6.34-rc3)

When used_dirs was introduced for the flex_groups struct, it looks
like the accounting was not put into place properly, in some places
manipulating free_inodes rather than used_dirs.

Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

ext4: Fix estimate of # of blocks needed to write indirect-mapped files

commit d330a5befb88875a9b3d2db62f9b74dadf660b13 upstream (as of v2.6.34-rc3)

http://bugzilla.kernel.org/show_bug.cgi?id=15420

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

ext4: Code cleanup for EXT4_IOC_MOVE_EXT ioctl

commit c437b2733520599a2c6e0dbcdeae611319f84707 upstream (as of v2.6.33-git11)

a) Fix sparse warning in ext4_ioctl()
b) Remove unneeded variable in mext_leaf_block()
c) Fix spelling typo in mext_check_arguments()

Signed-off-by: Akira Fujita <a-fujita@rs.jp.nec.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

ext4: Fix the NULL reference in double_down_write_data_sem()

commit 7247c0caa23d94a1cb6b307edba9dc45fb0798d4 upstream (as of v2.6.33-git11)

If EXT4_IOC_MOVE_EXT ioctl is called with NULL donor_fd, fget() in
ext4_ioctl() gets inappropriate file structure for donor; so we need
to do this check earlier, before calling double_down_write_data_sem().

Signed-off-by: Akira Fujita <a-fujita@rs.jp.nec.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

ext4: Fix insertion point of extent in mext_insert_across_blocks()

commit 5fd5249aa36fad98c9fd5edced352939e54f9324 upstream (as of v2.6.33-git11)

If the leaf node has 2 extent space or fewer and EXT4_IOC_MOVE_EXT
ioctl is called with the file offset where after the 2nd extent
covers, mext_insert_across_blocks() always tries to insert extent into
the first extent.  As a result, the file gets corrupted because of
wrong extent order.  The patch fixes this problem.

Signed-off-by: Akira Fujita <a-fujita@rs.jp.nec.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

ext4: make "offset" consistent in ext4_check_dir_entry()

commit b8b8afe236e97b6359d46d3a3f8c46455e192271 upstream (as of v2.6.33-git11)

The callers of ext4_check_dir_entry() usually pass in the "file
offset" (ext4_readdir, htree_dirblock_to_tree, search_dirblock,
ext4_dx_find_entry, empty_dir), but a few callers (add_dirent_to_buf,
ext4_delete_entry) only pass in the buffer offset.

To accomodate those last two (which would be hard to fix otherwise),
this patch changes ext4_check_dir_entry() to print the physical block
number and the relative offset as well as the passed-in offset.

Signed-off-by: Toshiyuki Okajima <toshi.okajima@jp.fujitsu.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

ext4: Handle non empty on-disk orphan link

commit 6e3617e579e070d3655a93ee9ed7149113e795e0 upstream (as of v2.6.33-git11)

In case of truncate errors we explicitly remove inode from in-core
orphan list via orphan_del(NULL, inode) without modifying the on-disk list.

But later on, the same inode may be inserted in the orphan list again
which will result the on-disk linked list getting corrupted.  If inode
i_dtime contains valid value, then skip on-disk list modification.

Signed-off-by: Dmitry Monakhov <dmonakhov@openvz.org>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

ext4: explicitly remove inode from orphan list after failed direct io

commit da1dafca84413145f5ac59998b4cdd06fb89f721 upstream (as of v2.6.33-git11)

Otherwise non-empty orphan list will be triggered on umount.

Signed-off-by: Dmitry Monakhov <dmonakhov@openvz.org>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

ext4: fix error handling in migrate

commit f39490bcd1691d65dc33689222a12e1fc13dd824 upstream (as of v2.6.33-git11)

Set i_nlink to zero for temporary inode from very beginning.
otherwise we may fail to start new journal handle and this
inode will be unreferenced but with i_nlink == 1
Since we hold inode reference it can not be pruned.

Also add missed journal_start retval check.

Signed-off-by: Dmitry Monakhov <dmonakhov@openvz.org>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

ext4: Fix fencepost error in chosing choosing group vs file preallocation.

commit cc483f102c3f703e853c96f95a654f0106fb2603 upstream (as of v2.6.33-git11)

The ext4 multiblock allocator decides whether to use group or file
preallocation based on the file size.  When the file size reaches
s_mb_stream_request (default is 16 blocks), it changes to use a
file-specific preallocation. This is cool, but it has a tiny problem.

See a simple script:
mkfs.ext4 -b 1024 /dev/sda8 1000000
mount -t ext4 -o nodelalloc /dev/sda8 /mnt/ext4
for((i=0;i<5;i++))
do
cat /mnt/4096>>/mnt/ext4/a #4096 is a file with 4096 characters.
cat /mnt/4096>>/mnt/ext4/b
done
debuge4fs -R 'stat a' /dev/sda8|grep BLOCKS -A 1

And you get
BLOCKS:
(0-14):8705-8719, (15):2356, (16-19):8465-8468

So there are 3 extents, a bit strange for the lonely 15th logical
block.  As we write to the 16 blocks, we choose file preallocation in
ext4_mb_group_or_file, but in ext4_mb_normalize_request, we meet with
the 16*1024 range, so no preallocation will be carried. file b then
reserves the space after '2356', so when when write 16, we start from
another part.

This patch just change the check in ext4_mb_group_or_file, so
that for the lonely 15 we will still use group preallocation.
After the patch, we will get:
debuge4fs -R 'stat a' /dev/sda8|grep BLOCKS -A 1
BLOCKS:
(0-15):8705-8720, (16-19):8465-8468

Looks more sane. Thanks.

Signed-off-by: Tao Ma <tao.ma@oracle.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

ext4: Add flag to files with blocks intentionally past EOF

commit c8d46e41bc744c8fa0092112af3942fcd46c8b18 upstream (as of v2.6.33-git11)

fallocate() may potentially instantiate blocks past EOF, depending
on the flags used when it is called.

e2fsck currently has a test for blocks past i_size, and it
sometimes trips up - noticeably on xfstests 013 which runs fsstress.

This patch from Jiayang does fix it up - it (along with
e2fsprogs updates and other patches recently from Aneesh) has
survived many fsstress runs in a row.

Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: Jiaying Zhang <jiayingz@google.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

ext4: Fix BUG_ON at fs/buffer.c:652 in no journal mode

commit 73b50c1c92666d326b5fa2c945d46509f2f6d91f upstream (as of v2.6.33-git11)

Calls to ext4_handle_dirty_metadata should only pass in an inode
pointer for inode-specific metadata, and not for shared metadata
blocks such as inode table blocks, block group descriptors, the
superblock, etc.

The BUG_ON can get tripped when updating a special device (such as a
block device) that is opened (so that i_mapping is set in
fs/block_dev.c) and the file system is mounted in no journal mode.

Addresses-Google-Bug: #2404870

Signed-off-by: Curt Wohlgemuth <curtw@google.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

ext4: Use bitops to read/modify EXT4_I(inode)->i_state

commit 19f5fb7ad679bb361222c7916086435020c37cce upstream (as of v2.6.33-git11)

At several places we modify EXT4_I(inode)->i_state without holding
i_mutex (ext4_release_file, ext4_bmap, ext4_journalled_writepage,
ext4_do_update_inode, ...). These modifications are racy and we can
lose updates to i_state. So convert handling of i_state to use bitops
which are atomic.

Cc: Jan Kara <jack@suse.cz>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

ext4: Drop EXT4_GET_BLOCKS_UPDATE_RESERVE_SPACE flag

commit 1296cc85c26e94eb865d03f82140f27d598de467 upstream (as of v2.6.33-rc6)

We should update reserve space if it is delalloc buffer
and that is indicated by EXT4_GET_BLOCKS_DELALLOC_RESERVE flag.
So use EXT4_GET_BLOCKS_DELALLOC_RESERVE in place of
EXT4_GET_BLOCKS_UPDATE_RESERVE_SPACE

[ Stable note: This fixes a corruption cuased by the following
  reproduction case:

  rm -f $TEST_FN
  touch $TEST_FN
  fallocate -n -o 656712 -l 858907 $TEST_FN
  dd if=/dev/zero of=$TEST_FN conv=notrunc bs=1 seek=1011020 count=36983
  sync
  dd if=/dev/zero of=$TEST_FN conv=notrunc bs=1 seek=332121 count=24005
  dd if=/dev/zero of=$TEST_FN conv=notrunc bs=1 seek=1040179 count=93319

  If the filesystem is then unmounted and e2fsck run forced, the
  i_blocks field for the file $TEST_FN will be found to be incorrect. ]

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

ext4: Fix quota accounting error with fallocate

commit 5f634d064c709ea02c3cdaa850a08323a4a4bf28 upstream (as of v2.6.33-rc6)

When we fallocate a region of the file which we had recently written,
and which is still in the page cache marked as delayed allocated blocks
we need to make sure we don't do the quota update on writepage path.
This is because the needed quota updated would have already be done
by fallocate.

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

ext4: Handle -EDQUOT error on write

commit 1db913823c0f8360fccbd24ca67eb073966a5ffd upstream (as of v2.6.33-rc6)

We need to release the journal before we do a write_inode.  Otherwise
we could deadlock.

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

ext4: Calculate metadata requirements more accurately

commit 9d0be50230b333005635967f7ecd4897dbfd181b upstream (as of v2.6.33-rc3)

In the past, ext4_calc_metadata_amount(), and its sub-functions
ext4_ext_calc_metadata_amount() and ext4_indirect_calc_metadata_amount()
badly over-estimated the number of metadata blocks that might be
required for delayed allocation blocks.  This didn't matter as much
when functions which managed the reserved metadata blocks were more
aggressive about dropping reserved metadata blocks as delayed
allocation blocks were written, but unfortunately they were too
aggressive.  This was fixed in commit 0637c6f, but as a result the
over-estimation by ext4_calc_metadata_amount() would lead to reserving
2-3 times the number of pending delayed allocation blocks as
potentially required metadata blocks.  So if there are 1 megabytes of
blocks which have been not yet been allocation, up to 3 megabytes of
space would get reserved out of the user's quota and from the file
system free space pool until all of the inode's data blocks have been
allocated.

This commit addresses this problem by much more accurately estimating
the number of metadata blocks that will be required.  It will still
somewhat over-estimate the number of blocks needed, since it must make
a worst case estimate not knowing which physical blocks will be
needed, but it is much more accurate than before.

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

ext4: Fix accounting of reserved metadata blocks

commit ee5f4d9cdf32fd99172d11665c592a288c2b1ff4 upstream (as of v2.6.33-rc3)

Commit 0637c6f had a typo which caused the reserved metadata blocks to
not be released correctly.   Fix this.

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

ext4: Patch up how we claim metadata blocks for quota purposes

commit 0637c6f4135f592f094207c7c21e7c0fc5557834 upstream (as of v2.6.33-rc3)

As reported in Kernel Bugzilla #14936, commit d21cd8f triggered a BUG
in the function ext4_da_update_reserve_space() found in
fs/ext4/inode.c.  The root cause of this BUG() was caused by the fact
that ext4_calc_metadata_amount() can severely over-estimate how many
metadata blocks will be needed, especially when using direct
block-mapped files.

In addition, it can also badly *under* estimate how much space is
needed, since ext4_calc_metadata_amount() assumes that the blocks are
contiguous, and this is not always true.  If the application is
writing blocks to a sparse file, the number of metadata blocks
necessary can be severly underestimated by the functions
ext4_da_reserve_space(), ext4_da_update_reserve_space() and
ext4_da_release_space().  This was the cause of the dq_claim_space
reports found on kerneloops.org.

Unfortunately, doing this right means that we need to massively
over-estimate the amount of free space needed.  So in some cases we
may need to force the inode to be written to disk asynchronously in
to avoid spurious quota failures.

http://bugzilla.kernel.org/show_bug.cgi?id=14936

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

ext4: Ensure zeroout blocks have no dirty metadata

commit 515f41c33a9d44a964264c9511ad2c869af1fac3 upstream (as of v2.6.33-rc3)

This fixes a bug (found by Curt Wohlgemuth) in which new blocks
returned from an extent created with ext4_ext_zeroout() can have dirty
metadata still associated with them.

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Curt Wohlgemuth <curtw@google.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>