firefly-linux-kernel-4.4.55.git
13 years agonet: Fix a memmove bug in dev_gro_receive()
Jarek Poplawski [Wed, 11 Aug 2010 02:02:10 +0000 (02:02 +0000)]
net: Fix a memmove bug in dev_gro_receive()

[ Upstream commit e5093aec2e6b60c3df2420057ffab9ed4a6d2792 ]

>Xin Xiaohui wrote:
> I looked into the code dev_gro_receive(), found the code here:
> if the frags[0] is pulled to 0, then the page will be released,
> and memmove() frags left.
> Is that right? I'm not sure if memmove do right or not, but
> frags[0].size is never set after memove at least. what I think
> a simple way is not to do anything if we found frags[0].size == 0.
> The patch is as followed.
...

This version of the patch fixes the bug directly in memmove.

Reported-by: "Xin, Xiaohui" <xiaohui.xin@intel.com>
Signed-off-by: Jarek Poplawski <jarkao2@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agosparc64: Fix atomic64_t routine return values.
David S. Miller [Wed, 18 Aug 2010 21:47:23 +0000 (14:47 -0700)]
sparc64: Fix atomic64_t routine return values.

[ Upstream commits 86fa04b8742ac681d470786f55e2403ada0075b2
  and b10f997bb0f4e5b34d447f498fb85834a40d3acb ]

Should return 'long' instead of 'int'.

Thanks to Dimitris Michailidis and Tony Luck.

Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agosparc64: Fix rwsem constant bug leading to hangs.
David S. Miller [Wed, 18 Aug 2010 00:09:53 +0000 (17:09 -0700)]
sparc64: Fix rwsem constant bug leading to hangs.

[ Upstream commit ef201bebe5afc91a2b99b45dacc8c6dd88ca9e58 ]

As noticed by Linus, it is critical that some of the
rwsem constants be signed.  Yet, hex constants are
unsigned unless explicitly casted or negated.

The most critical one is RWSEM_WAITING_BIAS.

This bug was exacerbated by commit
424acaaeb3a3932d64a9b4bd59df6cf72c22d8f3 ("rwsem: wake queued readers
when writer blocks on active read lock")

Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agosparc64: Add missing ID to parport probing code.
David S. Miller [Thu, 5 Aug 2010 00:30:21 +0000 (17:30 -0700)]
sparc64: Add missing ID to parport probing code.

[ Upstream commit bf8253bf5e7cfe17dd53e3f6340a45b11d9fb51c ]

SunBlade-2500 has 'parallel' device node with compatible
property "pnpALI,1533,3" so add that to the ID table.

Reported-by: Mikael Pettersson <mikpe@it.uu.se>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agosunxvr500: Ignore secondary output PCI devices.
David S. Miller [Sun, 4 Apr 2010 08:12:50 +0000 (01:12 -0700)]
sunxvr500: Ignore secondary output PCI devices.

[ Upstream commit bdd32ce95f79fb5cc964cd789d7ae4500bba7c6f ]

These just represent the secondary and further heads attached to the
card, and they have different sets of PCI bar registers to map.

So don't try to drive them in the main driver.

Reported-by: Frans van Berckel <fberckel@xs4all.nl>
Tested-by: Frans van Berckel <fberckel@xs4all.nl>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoslab: fix object alignment
Carsten Otte [Fri, 6 Aug 2010 16:19:22 +0000 (18:19 +0200)]
slab: fix object alignment

commit 1ab335d8f85792e3b107ff8237d53cf64db714df upstream.

This patch fixes alignment of slab objects in case CONFIG_DEBUG_PAGEALLOC is
active.
Before this spot in kmem_cache_create, we have this situation:
- align contains the required alignment of the object
- cachep->obj_offset is 0 or equals align in case of CONFIG_DEBUG_SLAB
- size equals the size of the object, or object plus trailing redzone in case
  of CONFIG_DEBUG_SLAB

This spot tries to fill one page per object if the object is in certain size
limits, however setting obj_offset to PAGE_SIZE - size does break the object
alignment since size may not be aligned with the required alignment.
This patch simply adds an ALIGN(size, align) to the equation and fixes the
object size detection accordingly.

This code in drivers/s390/cio/qdio_setup_init has lead to incorrectly aligned
slab objects (sizeof(struct qdio_q) equals 1792):
qdio_q_cache = kmem_cache_create("qdio_q", sizeof(struct qdio_q),
 256, 0, NULL);

Acked-by: Christoph Lameter <cl@linux.com>
Signed-off-by: Carsten Otte <cotte@de.ibm.com>
Signed-off-by: Pekka Enberg <penberg@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agodrm/i915: add 'reclaimable' to i915 self-reclaimable page allocations
Linus Torvalds [Sun, 18 Jul 2010 16:44:37 +0000 (09:44 -0700)]
drm/i915: add 'reclaimable' to i915 self-reclaimable page allocations

commit cd9f040df6ce46573760a507cb88192d05d27d86 upstream.

The hibernate issues that got fixed in commit 985b823b9192 ("drm/i915:
fix hibernation since i915 self-reclaim fixes") turn out to have been
incomplete.  Vefa Bicakci tested lots of hibernate cycles, and without
the __GFP_RECLAIMABLE flag the system eventually fails to resume.

With the flag added, Vefa can apparently hibernate forever (or until he
gets bored running his automated scripts, whichever comes first).

The reclaimable flag was there originally, and was one of the flags that
were dropped (unintentionally) by commit 4bdadb978569 ("drm/i915:
Selectively enable self-reclaim") that introduced all these problems,
but I didn't want to just blindly add back all the flags in commit
985b823b9192, and it looked like __GFP_RECLAIM wasn't necessary.  It
clearly was.

I still suspect that there is some subtle reason we're missing that
causes the problems, but __GFP_RECLAIMABLE is certainly not wrong to use
in this context, and is what the code historically used.  And we have no
idea what the causes the corruption without it.

Reported-and-tested-by: M. Vefa Bicakci <bicave@superonline.com>
Cc: Dave Airlie <airlied@gmail.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agodrm/i915: fix hibernation since i915 self-reclaim fixes
Linus Torvalds [Fri, 2 Jul 2010 00:04:42 +0000 (10:04 +1000)]
drm/i915: fix hibernation since i915 self-reclaim fixes

commit 985b823b919273fe1327d56d2196b4f92e5d0fae upstream.

Since commit 4bdadb9785696439c6e2b3efe34aa76df1149c83 ("drm/i915:
Selectively enable self-reclaim"), we've been passing GFP_MOVABLE to the
i915 page allocator where we weren't before due to some over-eager
removal of the page mapping gfp_flags games the code used to play.

This caused hibernate on Intel hardware to result in a lot of memory
corruptions on resume.  See for example

  http://bugzilla.kernel.org/show_bug.cgi?id=13811

Reported-by: Evengi Golov (in bugzilla)
Signed-off-by: Dave Airlie <airlied@redhat.com>
Tested-by: M. Vefa Bicakci <bicave@superonline.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agomm: make stack guard page logic use vm_prev pointer
Linus Torvalds [Fri, 20 Aug 2010 23:49:40 +0000 (16:49 -0700)]
mm: make stack guard page logic use vm_prev pointer

commit 0e8e50e20c837eeec8323bba7dcd25fe5479194c upstream.

Like the mlock() change previously, this makes the stack guard check
code use vma->vm_prev to see what the mapping below the current stack
is, rather than have to look it up with find_vma().

Also, accept an abutting stack segment, since that happens naturally if
you split the stack with mlock or mprotect.

Tested-by: Ian Campbell <ijc@hellion.org.uk>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agomm: make the mlock() stack guard page checks stricter
Linus Torvalds [Fri, 20 Aug 2010 23:39:25 +0000 (16:39 -0700)]
mm: make the mlock() stack guard page checks stricter

commit 7798330ac8114c731cfab83e634c6ecedaa233d7 upstream.

If we've split the stack vma, only the lowest one has the guard page.
Now that we have a doubly linked list of vma's, checking this is trivial.

Tested-by: Ian Campbell <ijc@hellion.org.uk>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agomm: make the vma list be doubly linked
Linus Torvalds [Fri, 20 Aug 2010 23:24:55 +0000 (16:24 -0700)]
mm: make the vma list be doubly linked

commit 297c5eee372478fc32fec5fe8eed711eedb13f3d upstream.

It's a really simple list, and several of the users want to go backwards
in it to find the previous vma.  So rather than have to look up the
previous entry with 'find_vma_prev()' or something similar, just make it
doubly linked instead.

Tested-by: Ian Campbell <ijc@hellion.org.uk>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agodm ioctl: release _hash_lock between devices in remove_all
Kiyoshi Ueda [Thu, 12 Aug 2010 03:13:55 +0000 (04:13 +0100)]
dm ioctl: release _hash_lock between devices in remove_all

commit 98f332855effef02aeb738e4d62e9a5b903c52fd upstream.

This patch changes dm_hash_remove_all() to release _hash_lock when
removing a device.  After removing the device, dm_hash_remove_all()
takes _hash_lock and searches the hash from scratch again.

This patch is a preparation for the next patch, which changes device
deletion code to wait for md reference to be 0.  Without this patch,
the wait in the next patch may cause AB-BA deadlock:
  CPU0                                CPU1
  -----------------------------------------------------------------------
  dm_hash_remove_all()
    down_write(_hash_lock)
                                      table_status()
                                        md = find_device()
                                               dm_get(md)
                                                 <increment md->holders>
                                        dm_get_live_or_inactive_table()
                                          dm_get_inactive_table()
                                            down_write(_hash_lock)
    <in the md deletion code>
      <wait for md->holders to be 0>

Signed-off-by: Kiyoshi Ueda <k-ueda@ct.jp.nec.com>
Signed-off-by: Jun'ichi Nomura <j-nomura@ce.jp.nec.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agodm mpath: fix NULL pointer dereference when path parameters missing
Alasdair G Kergon [Thu, 12 Aug 2010 03:13:49 +0000 (04:13 +0100)]
dm mpath: fix NULL pointer dereference when path parameters missing

commit 6bbf79a14080a0c61212f53b4b87dc1a99fedf9c upstream.

multipath_ctr() forgets to return an error after detecting
missing path parameters.  Fix this.

Signed-off-by: Patrick LoPresti <lopresti@gmail.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agodrm/i915/edp: Flush the write before waiting for PLLs
Chris Wilson [Sat, 7 Aug 2010 10:01:36 +0000 (11:01 +0100)]
drm/i915/edp: Flush the write before waiting for PLLs

commit 5ddb954b9ee50824977d2931e0ff58b3050b337d upstream.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agopxa3xx: fix ns2cycle equation
Axel Lin [Mon, 16 Aug 2010 08:09:09 +0000 (16:09 +0800)]
pxa3xx: fix ns2cycle equation

commit 93b352fce679945845664b56b0c3afbd655a7a12 upstream.

Test on a PXA310 platform with Samsung K9F2G08X0B NAND flash,
with tCH=5 and clk is 156MHz, ns2cycle(5, 156000000) returns -1.

ns2cycle returns negtive value will break NDTR0_tXX macros.

After checking the commit log, I found the problem is introduced by
commit 5b0d4d7c8a67c5ba3d35e6ceb0c5530cc6846db7
"[MTD] [NAND] pxa3xx: convert from ns to clock ticks more accurately"

To get num of clock cycles, we use below equation:
num of clock cycles = time (ns) / one clock cycle (ns) + 1
We need to add 1 cycle here because integer division will truncate the result.
It is possible the developers set the Min values in SPEC for timing settings.
Thus the truncate may cause problem, and it is safe to add an extra cycle here.

The various fields in NDTR{01} are in units of clock ticks minus one,
thus we should subtract 1 cycle then.

Thus the correct equation should be:
num of clock cycles = time (ns) / one clock cycle (ns) + 1 - 1
                    = time (ns) / one clock cycle (ns)

Signed-off-by: Axel Lin <axel.lin@gmail.com>
Signed-off-by: Lei Wen <leiwen@marvell.com>
Acked-by: Eric Miao <eric.y.miao@gmail.com>
Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoath5k: disable ASPM L0s for all cards
Maxim Levitsky [Fri, 13 Aug 2010 15:27:28 +0000 (11:27 -0400)]
ath5k: disable ASPM L0s for all cards

commit 6ccf15a1a76d2ff915cdef6ae4d12d0170087118 upstream.

Atheros PCIe wireless cards handled by ath5k do require L0s disabled.
For distributions shipping with CONFIG_PCIEASPM (this will be enabled
by default in the future in 2.6.36) this will also mean both L1 and L0s
will be disabled when a pre 1.1 PCIe device is detected. We do know L1
works correctly even for all ath5k pre 1.1 PCIe devices though but cannot
currently undue the effect of a blacklist, for details you can read
pcie_aspm_sanity_check() and see how it adjusts the device link
capability.

It may be possible in the future to implement some PCI API to allow
drivers to override blacklists for pre 1.1 PCIe but for now it is
best to accept that both L0s and L1 will be disabled completely for
distributions shipping with CONFIG_PCIEASPM rather than having this
issue present. Motivation for adding this new API will be to help
with power consumption for some of these devices.

Example of issues you'd see:

  - On the Acer Aspire One (AOA150, Atheros Communications Inc. AR5001
    Wireless Network Adapter [168c:001c] (rev 01)) doesn't work well
    with ASPM enabled, the card will eventually stall on heavy traffic
    with often 'unsupported jumbo' warnings appearing. Disabling
    ASPM L0s in ath5k fixes these problems.

  - On the same card you would see a storm of RXORN interrupts
    even though medium is idle.

Credit for root causing and fixing the bug goes to Jussi Kivilinna.

Cc: David Quan <David.Quan@atheros.com>
Cc: Matthew Garrett <mjg59@srcf.ucam.org>
Cc: Tim Gardner <tim.gardner@canonical.com>
Cc: Jussi Kivilinna <jussi.kivilinna@mbnet.fi>
Signed-off-by: Luis R. Rodriguez <lrodriguez@atheros.com>
Signed-off-by: Maxim Levitsky <maximlevitsky@gmail.com>
Signed-off-by: John W. Linville <linville@tuxdriver.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agonfs: Add "lookupcache" to displayed mount options
Patrick J. LoPresti [Tue, 10 Aug 2010 21:28:01 +0000 (17:28 -0400)]
nfs: Add "lookupcache" to displayed mount options

commit 9b00c64318cc337846a7a08a5678f5f19aeff188 upstream.

Running "cat /proc/mounts" fails to display the "lookupcache" option.
This oversight cost me a bunch of wasted time recently.

The following simple patch fixes it.

Signed-off-by: Patrick LoPresti <lopresti@gmail.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoARM: Tighten check for allowable CPSR values
Russell King [Fri, 13 Aug 2010 22:33:46 +0000 (23:33 +0100)]
ARM: Tighten check for allowable CPSR values

commit 41e2e8fd34fff909a0e40129f6ac4233ecfa67a9 upstream.

Reviewed-by: Arve Hjønnevåg <arve@android.com>
Acked-by: Dima Zavin <dima@android.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agohwmon: (pc87360) Fix device resource declaration
Jean Delvare [Sat, 14 Aug 2010 19:08:48 +0000 (21:08 +0200)]
hwmon: (pc87360) Fix device resource declaration

commit b9783dcebe952bf73449fe70a19ee4814adc81a0 upstream.

It's not OK to call platform_device_add_resources() multiple times
in a row. Despite its name, this functions sets the resources, it
doesn't add them. So we have to prepare an array with all the
resources, and then call platform_device_add_resources() once.

Before this fix, only the last I/O resource would be actually
registered. The other I/O resources were leaked.

Signed-off-by: Jean Delvare <khali@linux-fr.org>
Cc: Jim Cromie <jim.cromie@gmail.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoFix the nested PR lock calling issue in ACL
Jiaju Zhang [Wed, 28 Jul 2010 05:21:06 +0000 (13:21 +0800)]
Fix the nested PR lock calling issue in ACL

commit 845b6cf34150100deb5f58c8a37a372b111f2918 upstream.

Hi,

Thanks a lot for all the review and comments so far;) I'd like to send
the improved (V4) version of this patch.

This patch fixes a deadlock in OCFS2 ACL. We found this bug in OCFS2
and Samba integration using scenario, the symptom is several smbd
processes will be hung under heavy workload. Finally we found out it
is the nested PR lock calling that leads to this deadlock:

 node1        node2
              gr PR
                |
                V
 PR(EX)---> BAST:OCFS2_LOCK_BLOCKED
                |
                V
              rq PR
                |
                V
              wait=1

After requesting the 2nd PR lock, the process "smbd" went into D
state. It can only be woken up when the 1st PR lock's RO holder equals
zero. There should be an ocfs2_inode_unlock in the calling path later
on, which can decrement the RO holder. But since it has been in
uninterruptible sleep, the unlock function has no chance to be called.

The related stack trace is:
smbd          D ffff8800013d0600     0  9522   5608 0x00000000
 ffff88002ca7fb18 0000000000000282 ffff88002f964500 ffff88002ca7fa98
 ffff8800013d0600 ffff88002ca7fae0 ffff88002f964340 ffff88002f964340
 ffff88002ca7ffd8 ffff88002ca7ffd8 ffff88002f964340 ffff88002f964340
Call Trace:
[<ffffffff80350425>] schedule_timeout+0x175/0x210
[<ffffffff8034f580>] wait_for_common+0xf0/0x210
[<ffffffffa03e12b9>] __ocfs2_cluster_lock+0x3b9/0xa90 [ocfs2]
[<ffffffffa03e7665>] ocfs2_inode_lock_full_nested+0x255/0xdb0 [ocfs2]
[<ffffffffa0446019>] ocfs2_get_acl+0x69/0x120 [ocfs2]
[<ffffffffa0446368>] ocfs2_check_acl+0x28/0x80 [ocfs2]
[<ffffffff800e3507>] acl_permission_check+0x57/0xb0
[<ffffffff800e357d>] generic_permission+0x1d/0xc0
[<ffffffffa03eecea>] ocfs2_permission+0x10a/0x1d0 [ocfs2]
[<ffffffff800e3f65>] inode_permission+0x45/0x100
[<ffffffff800d86b3>] sys_chdir+0x53/0x90
[<ffffffff80007458>] system_call_fastpath+0x16/0x1b
[<00007f34a4ef6927>] 0x7f34a4ef6927

For details, please see:
https://bugzilla.novell.com/show_bug.cgi?id=614332 and
http://oss.oracle.com/bugzilla/show_bug.cgi?id=1278

Signed-off-by: Jiaju Zhang <jjzhang@suse.de>
Acked-by: Mark Fasheh <mfasheh@suse.com>
Signed-off-by: Joel Becker <joel.becker@oracle.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agox86, apic: Fix apic=debug boot crash
Daniel Kiper [Thu, 19 Aug 2010 22:46:16 +0000 (00:46 +0200)]
x86, apic: Fix apic=debug boot crash

commit 05e407603e527f9d808dd3866d3a17c2ce4dfcc5 upstream.

Fix a boot crash when apic=debug is used and the APIC is
not properly initialized.

This issue appears during Xen Dom0 kernel boot but the
fix is generic and the crash could occur on real hardware
as well.

Signed-off-by: Daniel Kiper <dkiper@net-space.pl>
Cc: xen-devel@lists.xensource.com
Cc: konrad.wilk@oracle.com
Cc: jeremy@goop.org
LKML-Reference: <20100819224616.GB9967@router-fw-old.local.net-space.pl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agox86, hotplug: Serialize CPU hotplug to avoid bringup concurrency issues
Borislav Petkov [Thu, 19 Aug 2010 18:10:29 +0000 (20:10 +0200)]
x86, hotplug: Serialize CPU hotplug to avoid bringup concurrency issues

commit d7c53c9e822a4fefa13a0cae76f3190bfd0d5c11 upstream.

When testing cpu hotplug code on 32-bit we kept hitting the "CPU%d:
Stuck ??" message due to multiple cores concurrently accessing the
cpu_callin_mask, among others.

Since these codepaths are not protected from concurrent access due to
the fact that there's no sane reason for making an already complex
code unnecessarily more complex - we hit the issue only when insanely
switching cores off- and online - serialize hotplugging cores on the
sysfs level and be done with it.

[ v2.1: fix !HOTPLUG_CPU build ]

Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
LKML-Reference: <20100819181029.GC17171@aftab>
Signed-off-by: H. Peter Anvin <hpa@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoocfs2/dlm: remove potential deadlock -V3
Wengang Wang [Fri, 30 Jul 2010 15:18:00 +0000 (23:18 +0800)]
ocfs2/dlm: remove potential deadlock -V3

commit b11f1f1ab73fd358b1b734a9427744802202ba68 upstream.

When we need to take both dlm_domain_lock and dlm->spinlock, we should take
them in order of: dlm_domain_lock then dlm->spinlock.

There is pathes disobey this order. That is calling dlm_lockres_put() with
dlm->spinlock held in dlm_run_purge_list. dlm_lockres_put() calls dlm_put() at
the ref and dlm_put() locks on dlm_domain_lock.

Fix:
Don't grab/put the dlm when the initialising/releasing lockres.
That grab is not required because we don't call dlm_unregister_domain()
based on refcount.

Signed-off-by: Wengang Wang <wen.gang.wang@oracle.com>
Signed-off-by: Joel Becker <joel.becker@oracle.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoocfs2/dlm: avoid incorrect bit set in refmap on recovery master
Wengang Wang [Fri, 30 Jul 2010 08:14:44 +0000 (16:14 +0800)]
ocfs2/dlm: avoid incorrect bit set in refmap on recovery master

commit a524812b7eaa7783d7811198921100f079034e61 upstream.

In the following situation, there remains an incorrect bit in refmap on the
recovery master. Finally the recovery master will fail at purging the lockres
due to the incorrect bit in refmap.

1) node A has no interest on lockres A any longer, so it is purging it.
2) the owner of lockres A is node B, so node A is sending de-ref message
to node B.
3) at this time, node B crashed. node C becomes the recovery master. it recovers
lockres A(because the master is the dead node B).
4) node A migrated lockres A to node C with a refbit there.
5) node A failed to send de-ref message to node B because it crashed. The failure
is ignored. no other action is done for lockres A any more.

For mormal, re-send the deref message to it to recovery master can fix it. Well,
ignoring the failure of deref to the original master and not recovering the lockres
to recovery master has the same effect. And the later is simpler.

Signed-off-by: Wengang Wang <wen.gang.wang@oracle.com>
Acked-by: Srinivas Eeda <srinivas.eeda@oracle.com>
Signed-off-by: Joel Becker <joel.becker@oracle.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoocfs2: Count more refcount records in file system fragmentation.
Tao Ma [Thu, 22 Jul 2010 05:56:45 +0000 (13:56 +0800)]
ocfs2: Count more refcount records in file system fragmentation.

commit 8a2e70c40ff58f82dde67770e6623ca45f0cb0c8 upstream.

The refcount record calculation in ocfs2_calc_refcount_meta_credits
is too optimistic that we can always allocate contiguous clusters
and handle an already existed refcount rec as a whole. Actually
because of file system fragmentation, we may have the chance to split
a refcount record into 3 parts during the transaction. So consider
the worst case in record calculation.

Signed-off-by: Tao Ma <tao.ma@oracle.com>
Signed-off-by: Joel Becker <joel.becker@oracle.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoocfs2 fix o2dlm dlm run purgelist (rev 3)
Srinivas Eeda [Mon, 19 Jul 2010 23:04:12 +0000 (16:04 -0700)]
ocfs2 fix o2dlm dlm run purgelist (rev 3)

commit 7beaf243787f85a2ef9213ccf13ab4a243283fde upstream.

This patch fixes two problems in dlm_run_purgelist

1. If a lockres is found to be in use, dlm_run_purgelist keeps trying to purge
the same lockres instead of trying the next lockres.

2. When a lockres is found unused, dlm_run_purgelist releases lockres spinlock
before setting DLM_LOCK_RES_DROPPING_REF and calls dlm_purge_lockres.
spinlock is reacquired but in this window lockres can get reused. This leads
to BUG.

This patch modifies dlm_run_purgelist to skip lockres if it's in use and purge
 next lockres. It also sets DLM_LOCK_RES_DROPPING_REF before releasing the
lockres spinlock protecting it from getting reused.

Signed-off-by: Srinivas Eeda <srinivas.eeda@oracle.com>
Acked-by: Sunil Mushran <sunil.mushran@oracle.com>
Signed-off-by: Joel Becker <joel.becker@oracle.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoocfs2/dlm: fix a dead lock
Wengang Wang [Fri, 16 Jul 2010 15:13:33 +0000 (23:13 +0800)]
ocfs2/dlm: fix a dead lock

commit 6d98c3ccb52f692f1a60339dde7c700686a5568b upstream.

When we have to take both dlm->master_lock and lockres->spinlock,
take them in order

lockres->spinlock and then dlm->master_lock.

The patch fixes a violation of the rule.
We can simply move taking dlm->master_lock to where we have dropped res->spinlock
since when we access res->state and free mle memory we don't need master_lock's
protection.

Signed-off-by: Wengang Wang <wen.gang.wang@oracle.com>
Signed-off-by: Joel Becker <joel.becker@oracle.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoocfs2: do not overwrite error codes in ocfs2_init_acl
Tiger Yang [Fri, 16 Jul 2010 03:21:23 +0000 (11:21 +0800)]
ocfs2: do not overwrite error codes in ocfs2_init_acl

commit 6eda3dd33f8a0ce58ee56a11351758643a698db4 upstream.

Setting the acl while creating a new inode depends on
the error codes of posix_acl_create_masq. This patch fix
a issue of overwriting the error codes of it.

Reported-by: Pawel Zawora <pzawora@gmail.com>
Signed-off-by: Tiger Yang <tiger.yang@oracle.com>
Signed-off-by: Joel Becker <joel.becker@oracle.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoALSA: hda - Fix missing stream for second ADC on Realtek ALC260 HDA codec
Jonathan Woithe [Sat, 7 Aug 2010 14:47:05 +0000 (00:17 +0930)]
ALSA: hda - Fix missing stream for second ADC on Realtek ALC260 HDA codec

commit 53bacfbbb2ddd981287b58a511c8b8f5df179886 upstream.

I discovered tonight that ALSA no longer sets up a stream for the second ADC
provided by the Realtek ALC260 HDA codec.  At some point alc_build_pcms()
started using stream_analog_alt_capture when constructing the second ADC
stream, but patch_alc260() was never updated accordingly.  I have no idea
when this regression occurred.  The trivial patch to patch_alc260() given
below fixes the problem as far as I can tell.  The patch is against 2.6.35.

Signed-off-by: Jonathan Woithe <jwoithe@physics.adelaide.edu.au>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoALSA: emu10k1 - delay the PCM interrupts (add pcm_irq_delay parameter)
Jaroslav Kysela [Wed, 18 Aug 2010 12:08:17 +0000 (14:08 +0200)]
ALSA: emu10k1 - delay the PCM interrupts (add pcm_irq_delay parameter)

commit 56385a12d9bb9e173751f74b6c430742018cafc0 upstream.

With some hardware combinations, the PCM interrupts are acknowledged
before the period boundary from the emu10k1 chip. The midlevel PCM code
gets confused and the playback stream is interrupted.

It seems that the interrupt processing shift by 2 samples is enough
to fix this issue. This default value does not harm other,
non-affected hardware.

More information: Kernel bugzilla bug#16300

[A copmile warning fixed by tiwai]

Signed-off-by: Jaroslav Kysela <perex@perex.cz>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoALSA: riptide - Fix detection / load of firmware files
Takashi Iwai [Mon, 16 Aug 2010 06:08:48 +0000 (08:08 +0200)]
ALSA: riptide - Fix detection / load of firmware files

commit a5ba6beb839cfa288960c92cd2668a2601c24dda upstream.

The detection and loading of firmeware on riptide driver has been broken
due to rewrite of some codes, checking the presense wrongly.
This patch fixes the logic again.

Reference: kernel bug 16596
https://bugzilla.kernel.org/show_bug.cgi?id=16596

Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoASoC: Remove DSP mode support for WM8776
Mark Brown [Mon, 16 Aug 2010 10:46:57 +0000 (11:46 +0100)]
ASoC: Remove DSP mode support for WM8776

commit b2c1e07b81a126e5846dfc3d36f559d861df59f4 upstream.

This is not supported by current hardware revisions.

Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
Acked-by: Liam Girdwood <lrg@slimlogic.co.uk>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoASoC: Fix inverted mute controls for WM8580
Mark Brown [Fri, 6 Aug 2010 16:57:12 +0000 (17:57 +0100)]
ASoC: Fix inverted mute controls for WM8580

commit 4f0ed9a51bc8ef16c2589112fdb110479e4b0df1 upstream.

Signed-off-by: Mark Brown <broonie@opensource.wolfsonmicro.com>
Acked-by: Liam Girdwood <lrg@slimlogic.co.uk>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agomemstick: fix hangs on unexpected device removal in mspro_blk
Maxim Levitsky [Wed, 11 Aug 2010 21:17:52 +0000 (14:17 -0700)]
memstick: fix hangs on unexpected device removal in mspro_blk

commit d862b13bc8cbab9692fbe0ef44c40d0488b81af1 upstream.

mspro_block_remove() is called from detect thread that first calls the
mspro_block_stop(), which stops the request queue.  If we call
del_gendisk() with the queue stopped we get a deadlock.

Signed-off-by: Maxim Levitsky <maximlevitsky@gmail.com>
Cc: Alex Dubov <oakad@yahoo.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoLinux 2.6.32.20
Greg Kroah-Hartman [Fri, 20 Aug 2010 18:34:38 +0000 (11:34 -0700)]
Linux 2.6.32.20

13 years agomm: fix up some user-visible effects of the stack guard page
Linus Torvalds [Sun, 15 Aug 2010 18:35:52 +0000 (11:35 -0700)]
mm: fix up some user-visible effects of the stack guard page

commit d7824370e26325c881b665350ce64fb0a4fde24a upstream.

This commit makes the stack guard page somewhat less visible to user
space. It does this by:

 - not showing the guard page in /proc/<pid>/maps

   It looks like lvm-tools will actually read /proc/self/maps to figure
   out where all its mappings are, and effectively do a specialized
   "mlockall()" in user space.  By not showing the guard page as part of
   the mapping (by just adding PAGE_SIZE to the start for grows-up
   pages), lvm-tools ends up not being aware of it.

 - by also teaching the _real_ mlock() functionality not to try to lock
   the guard page.

   That would just expand the mapping down to create a new guard page,
   so there really is no point in trying to lock it in place.

It would perhaps be nice to show the guard page specially in
/proc/<pid>/maps (or at least mark grow-down segments some way), but
let's not open ourselves up to more breakage by user space from programs
that depends on the exact deails of the 'maps' file.

Special thanks to Henrique de Moraes Holschuh for diving into lvm-tools
source code to see what was going on with the whole new warning.

Reported-and-tested-by: François Valenduc <francois.valenduc@tvcablenet.be
Reported-by: Henrique de Moraes Holschuh <hmh@hmh.eng.br>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agomm: fix page table unmap for stack guard page properly
Linus Torvalds [Sat, 14 Aug 2010 18:44:56 +0000 (11:44 -0700)]
mm: fix page table unmap for stack guard page properly

commit 11ac552477e32835cb6970bf0a70c210807f5673 upstream.

We do in fact need to unmap the page table _before_ doing the whole
stack guard page logic, because if it is needed (mainly 32-bit x86 with
PAE and CONFIG_HIGHPTE, but other architectures may use it too) then it
will do a kmap_atomic/kunmap_atomic.

And those kmaps will create an atomic region that we cannot do
allocations in.  However, the whole stack expand code will need to do
anon_vma_prepare() and vma_lock_anon_vma() and they cannot do that in an
atomic region.

Now, a better model might actually be to do the anon_vma_prepare() when
_creating_ a VM_GROWSDOWN segment, and not have to worry about any of
this at page fault time.  But in the meantime, this is the
straightforward fix for the issue.

See https://bugzilla.kernel.org/show_bug.cgi?id=16588 for details.

Reported-by: Wylda <wylda@volny.cz>
Reported-by: Sedat Dilek <sedat.dilek@gmail.com>
Reported-by: Mike Pagano <mpagano@gentoo.org>
Reported-by: François Valenduc <francois.valenduc@tvcablenet.be>
Tested-by: Ed Tomlinson <edt@aei.ca>
Cc: Pekka Enberg <penberg@kernel.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoLinux 2.6.32.19
Greg Kroah-Hartman [Fri, 13 Aug 2010 20:24:37 +0000 (13:24 -0700)]
Linux 2.6.32.19

13 years agox86: don't send SIGBUS for kernel page faults
Linus Torvalds [Fri, 13 Aug 2010 16:49:20 +0000 (09:49 -0700)]
x86: don't send SIGBUS for kernel page faults

commit 96054569190bdec375fe824e48ca1f4e3b53dd36 upstream.

It's wrong for several reasons, but the most direct one is that the
fault may be for the stack accesses to set up a previous SIGBUS.  When
we have a kernel exception, the kernel exception handler does all the
fixups, not some user-level signal handler.

Even apart from the nested SIGBUS issue, it's also wrong to give out
kernel fault addresses in the signal handler info block, or to send a
SIGBUS when a system call already returns EFAULT.

Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agomm: fix missing page table unmap for stack guard page failure case
Linus Torvalds [Fri, 13 Aug 2010 16:24:04 +0000 (09:24 -0700)]
mm: fix missing page table unmap for stack guard page failure case

commit 5528f9132cf65d4d892bcbc5684c61e7822b21e9 upstream.

.. which didn't show up in my tests because it's a no-op on x86-64 and
most other architectures.  But we enter the function with the last-level
page table mapped, and should unmap it at exit.

Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agomm: keep a guard page below a grow-down stack segment
Linus Torvalds [Fri, 13 Aug 2010 00:54:33 +0000 (17:54 -0700)]
mm: keep a guard page below a grow-down stack segment

commit 320b2b8de12698082609ebbc1a17165727f4c893 upstream.

This is a rather minimally invasive patch to solve the problem of the
user stack growing into a memory mapped area below it.  Whenever we fill
the first page of the stack segment, expand the segment down by one
page.

Now, admittedly some odd application might _want_ the stack to grow down
into the preceding memory mapping, and so we may at some point need to
make this a process tunable (some people might also want to have more
than a single page of guarding), but let's try the minimal approach
first.

Tested with trivial application that maps a single page just below the
stack, and then starts recursing.  Without this, we will get a SIGSEGV
_after_ the stack has smashed the mapping.  With this patch, we'll get a
nice SIGBUS just as the stack touches the page just above the mapping.

Requested-by: Keith Packard <keithp@keithp.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agomm: fix corruption of hibernation caused by reusing swap during image saving
KAMEZAWA Hiroyuki [Wed, 11 Aug 2010 22:32:15 +0000 (15:32 -0700)]
mm: fix corruption of hibernation caused by reusing swap during image saving

commit 966cca029f739716fbcc8068b8c6dfe381f86fc3 upstream.

Since 2.6.31, swap_map[]'s refcounting was changed to show that a used
swap entry is just for swap-cache, can be reused.  Then, while scanning
free entry in swap_map[], a swap entry may be able to be reclaimed and
reused.  It was caused by commit c9e444103b5e7a5 ("mm: reuse unused swap
entry if necessary").

But this caused deta corruption at resume. The scenario is

- Assume a clean-swap cache, but mapped.

- at hibernation_snapshot[], clean-swap-cache is saved as
  clean-swap-cache and swap_map[] is marked as SWAP_HAS_CACHE.

- then, save_image() is called.  And reuse SWAP_HAS_CACHE entry to save
  image, and break the contents.

After resume:

- the memory reclaim runs and finds clean-not-referenced-swap-cache and
  discards it because it's marked as clean.  But here, the contents on
  disk and swap-cache is inconsistent.

Hance memory is corrupted.

This patch avoids the bug by not reclaiming swap-entry during hibernation.
This is a quick fix for backporting.

Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Rafael J. Wysocki <rjw@sisk.pl>
Reported-by: Ondreg Zary <linux@rainbow-software.org>
Tested-by: Ondreg Zary <linux@rainbow-software.org>
Tested-by: Andrea Gelmini <andrea.gelmini@gmail.com>
Signed-off-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agomd/raid1: delay reads that could overtake behind-writes.
NeilBrown [Wed, 31 Mar 2010 00:21:44 +0000 (11:21 +1100)]
md/raid1: delay reads that could overtake behind-writes.

commit e555190d82c0f58e825e3cbd9e6ebe2e7ac713bd upstream.

When a raid1 array is configured to support write-behind
on some devices, it normally only reads from other devices.
If all devices are write-behind (because the rest have failed)
it is possible for a read request to be serviced before a
behind-write request, which would appear as data corruption.

So when forced to read from a WriteMostly device, wait for any
write-behind to complete, and don't start any more behind-writes.

Signed-off-by: NeilBrown <neilb@suse.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoibmvfc: Reduce error recovery timeout
Brian King [Tue, 20 Apr 2010 19:21:35 +0000 (14:21 -0500)]
ibmvfc: Reduce error recovery timeout

commit daa142d1773dd3a986f02a8a4da929608d24daaa upstream.

If a command times out resulting in EH getting invoked, we wait for the
aborted commands to come back after sending the abort. Shorten
the amount of time we wait for these responses, to ensure we don't
get stuck in EH for several minutes.

Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoibmvfc: Fix command completion handling
Brian King [Tue, 20 Apr 2010 19:21:33 +0000 (14:21 -0500)]
ibmvfc: Fix command completion handling

commit f5832fa2f8dc39adcf3ae348d2d6383163235e79 upstream.

Commands which are completed by the VIOS are placed on a CRQ
in kernel memory for the ibmvfc driver to process. Each CRQ
entry is 16 bytes. The ibmvfc driver reads the first 8 bytes
to check if the entry is valid, then reads the next 8 bytes to get
the handle, which is a pointer the completed command. This fixes
an issue seen on Power 7 where the processor reordered the
loads from memory, resulting in processing command completion
with a stale handle. This could result in command timeouts,
and also early completion of commands.

Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoaic79xx: check for non-NULL scb in ahd_handle_nonpkt_busfree
Hannes Reinecke [Fri, 15 Jan 2010 12:07:34 +0000 (13:07 +0100)]
aic79xx: check for non-NULL scb in ahd_handle_nonpkt_busfree

commit 534ef056db8a8fb6b9d50188d88ed5d1fbc66673 upstream.

When removing several devices aic79xx will occasionally Oops
in ahd_handle_nonpkt_busfree during rescan. Looking at the
code I found that we're indeed not checking if the scb in
question is NULL. So check for it before accessing it.

Signed-off-by: Hannes Reinecke <hare@suse.de>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoloop: Update mtime when writing using aops
Nikanth Karthikesan [Thu, 8 Apr 2010 19:39:31 +0000 (21:39 +0200)]
loop: Update mtime when writing using aops

commit 02246c41171097ceab3246f6dc251ac89de6004b upstream.

Update mtime when writing to backing filesystem using the address space
operations write_begin and write_end.

Signed-off-by: Nikanth Karthikesan <knikanth@suse.de>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoSkip check for mandatory locks when unlocking
Sachin Prabhu [Wed, 10 Mar 2010 15:28:40 +0000 (10:28 -0500)]
Skip check for mandatory locks when unlocking

commit ee860b6a650360c91f5d5f9a94262aad9be90015 upstream.

ocfs2_lock() will skip locks on file which has mode set to 02666. This
is a problem in cases where the mode of the file is changed after a
process has obtained a lock on the file.

ocfs2_lock() should skip the check for mandatory locks when unlocking a
file.

Signed-off-by: Sachin Prabhu <sprabhu@redhat.com>
Signed-off-by: Joel Becker <joel.becker@oracle.com>
Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoocfs2: Set MS_POSIXACL on remount
Jan Kara [Thu, 15 Oct 2009 12:54:05 +0000 (14:54 +0200)]
ocfs2: Set MS_POSIXACL on remount

commit 57b09bb5e492c37c1e4273fe4e435ffd1d2ddbe0 upstream.

We have to set MS_POSIXACL on remount as well. Otherwise VFS
would not know we started supporting ACLs after remount and
thus ACLs would not work.

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Joel Becker <joel.becker@oracle.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoocfs2: Find proper end cpos for a leaf refcount block.
Tao Ma [Mon, 30 Nov 2009 06:32:19 +0000 (14:32 +0800)]
ocfs2: Find proper end cpos for a leaf refcount block.

commit 38a04e432768ec0b016f3c687b4de31ac111ae59 upstream.

ocfs2 refcount tree is stored as an extent tree while
the leaf ocfs2_refcount_rec points to a refcount block.

The following step can trip a kernel panic.
mkfs.ocfs2 -b 512 -C 1M --fs-features=refcount $DEVICE
mount -t ocfs2 $DEVICE $MNT_DIR
FILE_NAME=$RANDOM
FILE_NAME_1=$RANDOM
FILE_REF="${FILE_NAME}_ref"
FILE_REF_1="${FILE_NAME}_ref_1"
for((i=0;i<305;i++))
do
# /mnt/1048576 is a file with 1048576 sizes.
cat /mnt/1048576 >> $MNT_DIR/$FILE_NAME
cat /mnt/1048576 >> $MNT_DIR/$FILE_NAME_1
done
for((i=0;i<3;i++))
do
cat /mnt/1048576 >> $MNT_DIR/$FILE_NAME
done

for((i=0;i<2;i++))
do
cat /mnt/1048576 >> $MNT_DIR/$FILE_NAME
cat /mnt/1048576 >> $MNT_DIR/$FILE_NAME_1
done

cat /mnt/1048576 >> $MNT_DIR/$FILE_NAME

for((i=0;i<11;i++))
do
cat /mnt/1048576 >> $MNT_DIR/$FILE_NAME
cat /mnt/1048576 >> $MNT_DIR/$FILE_NAME_1
done
reflink $MNT_DIR/$FILE_NAME $MNT_DIR/$FILE_REF
# write_f is a program which will write some bytes to a file at offset.
# write_f -f file_name -l offset -w write_bytes.
./write_f -f $MNT_DIR/$FILE_REF -l $[310*1048576] -w 4096
./write_f -f $MNT_DIR/$FILE_REF -l $[306*1048576] -w 4096
./write_f -f $MNT_DIR/$FILE_REF -l $[311*1048576] -w 4096
./write_f -f $MNT_DIR/$FILE_NAME -l $[310*1048576] -w 4096
./write_f -f $MNT_DIR/$FILE_NAME -l $[311*1048576] -w 4096
reflink $MNT_DIR/$FILE_NAME $MNT_DIR/$FILE_REF_1
./write_f -f $MNT_DIR/$FILE_NAME -l $[311*1048576] -w 4096
#kernel panic here.

The reason is that if the ocfs2_extent_rec is the last record
in a leaf extent block, the old solution fails to find the
suitable end cpos. So this patch try to walk through the b-tree,
find the next sub root and get the c_pos the next sub-tree starts
from.

btw, I have runned tristan's test case against the patched kernel
for several days and this type of kernel panic never happens again.

Signed-off-by: Tao Ma <tao.ma@oracle.com>
Signed-off-by: Joel Becker <joel.becker@oracle.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agodlm: send reply before bast
David Teigland [Wed, 24 Feb 2010 17:59:23 +0000 (11:59 -0600)]
dlm: send reply before bast

commit cf6620acc0f6fac57968aafef79ab372bdcf6157 upstream.

When the lock master processes a successful operation (request,
convert, cancel, or unlock), it will process the effects of the
change before sending the reply for the operation.  The "effects"
of the operation are:

- blocking callbacks (basts) for any newly granted locks
- waiting or converting locks that can now be granted

The cast is queued on the local node when the reply from the lock
master is received.  This means that a lock holder can receive a
bast for a lock mode that is doesn't yet know has been granted.

Signed-off-by: David Teigland <teigland@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agodlm: fix ordering of bast and cast
David Teigland [Wed, 24 Feb 2010 17:08:18 +0000 (11:08 -0600)]
dlm: fix ordering of bast and cast

commit 7fe2b3190b8b299409f13cf3a6f85c2bd371f8bb upstream.

When both blocking and completion callbacks are queued for lock,
the dlm would always deliver the completion callback (cast) first.
In some cases the blocking callback (bast) is queued before the
cast, though, and should be delivered first.  This patch keeps
track of the order in which they were queued and delivers them
in that order.

This patch also keeps track of the granted mode in the last cast
and eliminates the following bast if the bast mode is compatible
with the preceding cast mode.  This happens when a remotely mastered
lock is demoted, e.g. EX->NL, in which case the local node queues
a cast immediately after sending the demote message.  In this way
a cast can be queued for a mode, e.g. NL, that makes an in-transit
bast extraneous.

Signed-off-by: David Teigland <teigland@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agodlm: always use GFP_NOFS
David Teigland [Mon, 30 Nov 2009 22:34:43 +0000 (16:34 -0600)]
dlm: always use GFP_NOFS

commit 573c24c4af6664ffcd9aa7ba617a35fde2b95534 upstream.

Replace all GFP_KERNEL and ls_allocation with GFP_NOFS.
ls_allocation would be GFP_KERNEL for userland lockspaces
and GFP_NOFS for file system lockspaces.

It was discovered that any lockspaces on the system can
affect all others by triggering memory reclaim in the
file system which could in turn call back into the dlm
to acquire locks, deadlocking dlm threads that were
shared by all lockspaces, like dlm_recv.

Signed-off-by: David Teigland <teigland@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoreiserfs: fix oops while creating privroot with selinux enabled
Jeff Mahoney [Tue, 23 Mar 2010 20:35:38 +0000 (13:35 -0700)]
reiserfs: fix oops while creating privroot with selinux enabled

commit 6cb4aff0a77cc0e6bae9475d62205319e3ebbf3f upstream.

Commit 57fe60df ("reiserfs: add atomic addition of selinux attributes
during inode creation") contains a bug that will cause it to oops when
mounting a file system that didn't previously contain extended attributes
on a system using security.* xattrs.

The issue is that while creating the privroot during mount
reiserfs_security_init calls reiserfs_xattr_jcreate_nblocks which
dereferences the xattr root.  The xattr root doesn't exist, so we get an
oops.

Addresses http://bugzilla.kernel.org/show_bug.cgi?id=15309

Signed-off-by: Jeff Mahoney <jeffm@suse.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoreiserfs: properly honor read-only devices
Jeff Mahoney [Tue, 23 Mar 2010 20:35:39 +0000 (13:35 -0700)]
reiserfs: properly honor read-only devices

commit 3f8b5ee33293d43ca360771b535dfae8c57259dc upstream.

The reiserfs journal behaves inconsistently when determining whether to
allow a mount of a read-only device.

This is due to the use of the continue_replay variable to short circuit
the journal scanning.  If it's set, it's assumed that there are
transactions to replay, but there may not be.  If it's unset, it's assumed
that there aren't any, and that may not be the case either.

I've observed two failure cases:
1) Where a clean file system on a read-only device refuses to mount
2) Where a clean file system on a read-only device passes the
   optimization and then tries writing the journal header to update
   the latest mount id.

The former is easily observable by using a freshly created file system on
a read-only loopback device.

This patch moves the check into journal_read_transaction, where it can
bail out before it's about to replay a transaction.  That way it can go
through and skip transactions where appropriate, yet still refuse to mount
a file system with outstanding transactions.

Signed-off-by: Jeff Mahoney <jeffm@suse.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoext4: Fix optional-arg mount options
Eric Sandeen [Tue, 16 Feb 2010 01:17:55 +0000 (20:17 -0500)]
ext4: Fix optional-arg mount options

commit 15121c18a22ae483279f76dc9e554334b800d0f7 upstream.

We have 2 mount options, "barrier" and "auto_da_alloc" which may or
may not take a 1/0 argument.  This causes the ext4 superblock mount
code to subtract uninitialized pointers and pass the result to
kmalloc, which results in very noisy failures.

Per Ted's suggestion, initialize the args struct so that
we know whether match_token() found an argument for the
option, and skip match_int() if not.

Also, return error (0) from parse_options if we thought
we found an argument, but match_int() Fails.

Reported-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Acked-by: Jeff Mahoney <jeffm@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoext4: Make sure the MOVE_EXT ioctl can't overwrite append-only files
Theodore Ts'o [Thu, 3 Jun 2010 02:04:39 +0000 (22:04 -0400)]
ext4: Make sure the MOVE_EXT ioctl can't overwrite append-only files

commit 1f5a81e41f8b1a782c68d3843e9ec1bfaadf7d72 upstream.

Dan Roseberg has reported a problem with the MOVE_EXT ioctl.  If the
donor file is an append-only file, we should not allow the operation
to proceed, lest we end up overwriting the contents of an append-only
file.

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Cc: Dan Rosenberg <dan.j.rosenberg@gmail.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoACPI: Fix regression where _PPC is not read at boot even when ignore_ppc=0
Darrick J. Wong [Thu, 18 Feb 2010 18:28:20 +0000 (10:28 -0800)]
ACPI: Fix regression where _PPC is not read at boot even when ignore_ppc=0

commit 455c0d71d46e86b0b7ff2c9dcfc19bc162302ee9 upstream.

Earlier, Ingo Molnar posted a patch to make it so that the kernel would avoid
reading _PPC on his broken T60.  Unfortunately, it seems that with Thomas
Renninger's patch last July to eliminate _PPC evaluations when the processor
driver loads, the kernel never actually reads _PPC at all!  This is problematic
if you happen to boot your non-T60 computer in a state where the BIOS _wants_
_PPC to be something other than zero.

So, put the _PPC evaluation back into acpi_processor_get_performance_info if
ignore_ppc isn't 1.

Signed-off-by: Darrick J. Wong <djwong@us.ibm.com>
Signed-off-by: Len Brown <len.brown@intel.com>
Acked-by: Jeff Mahoney <jeffm@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agopowerpc/eeh: Fix a bug when pci structure is null
Breno Leitao [Wed, 3 Feb 2010 05:56:41 +0000 (05:56 +0000)]
powerpc/eeh: Fix a bug when pci structure is null

commit 8d3d50bf1913561ef3b1f5b53115c5a481ba9b1e upstream.

During a EEH recover, the pci_dev structure can be null, mainly if an
eeh event is detected during cpi config operation. In this case, the
pci_dev will not be known (and will be null) the kernel will crash
with the following message:

Unable to handle kernel paging request for data at address 0x000000a0
Faulting instruction address: 0xc00000000006b8b4
Oops: Kernel access of bad area, sig: 11 [#1]

NIP [c00000000006b8b4] .eeh_event_handler+0x10c/0x1a0
LR [c00000000006b8a8] .eeh_event_handler+0x100/0x1a0
Call Trace:
[c0000003a80dff00] [c00000000006b8a8] .eeh_event_handler+0x100/0x1a0
[c0000003a80dff90] [c000000000031f1c] .kernel_thread+0x54/0x70

The bug occurs because pci_name() tries to access a null pointer.
This patch just guarantee that pci_name() is not called on Null pointers.

Signed-off-by: Breno Leitao <leitao@linux.vnet.ibm.com>
Signed-off-by: Linas Vepstas <linasvepstas@gmail.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Acked-by: Jeff Mahoney <jeffm@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoHWPOISON: abort on failed unmap
Wu Fengguang [Wed, 16 Dec 2009 11:19:58 +0000 (12:19 +0100)]
HWPOISON: abort on failed unmap

commit 1668bfd5be9d8a52536c4865000fbbe065a3613b upstream.

Don't try to isolate a still mapped page. Otherwise we will hit the
BUG_ON(page_mapped(page)) in __remove_from_page_cache().

Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: Thomas Renninger <trenn@suse.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoHWPOISON: remove the anonymous entry
Wu Fengguang [Wed, 16 Dec 2009 11:19:57 +0000 (12:19 +0100)]
HWPOISON: remove the anonymous entry

commit 9b9a29ecd75e310f75a9243e1c3538ad34598fcb upstream.

(PG_swapbacked && !PG_lru) pages should not happen.
Better to treat them as unknown pages.

Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
Signed-off-by: Andi Kleen <ak@linux.intel.com>
Signed-off-by: Thomas Renninger <trenn@suse.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agox86: Fix out of order of gsi
Eric W. Biederman [Sun, 28 Feb 2010 09:06:34 +0000 (01:06 -0800)]
x86: Fix out of order of gsi

commit fad539956c9e69749a03f7817d22d1bab87657bf upstream.

Iranna D Ankad reported that IBM x3950 systems have boot
problems after this commit:

 |
 | commit b9c61b70075c87a8612624736faf4a2de5b1ed30
 |
 |    x86/pci: update pirq_enable_irq() to setup io apic routing
 |

The problem is that with the patch, the machine freezes when
console=ttyS0,... kernel serial parameter is passed.

It seem to freeze at DVD initialization and the whole problem
seem to be DVD/pata related, but somehow exposed through the
serial parameter.

Such apic problems can expose really weird behavior:

  ACPI: IOAPIC (id[0x10] address[0xfecff000] gsi_base[0])
  IOAPIC[0]: apic_id 16, version 0, address 0xfecff000, GSI 0-2
  ACPI: IOAPIC (id[0x0f] address[0xfec00000] gsi_base[3])
  IOAPIC[1]: apic_id 15, version 0, address 0xfec00000, GSI 3-38
  ACPI: IOAPIC (id[0x0e] address[0xfec01000] gsi_base[39])
  IOAPIC[2]: apic_id 14, version 0, address 0xfec01000, GSI 39-74
  ACPI: INT_SRC_OVR (bus 0 bus_irq 1 global_irq 4 dfl dfl)
  ACPI: INT_SRC_OVR (bus 0 bus_irq 0 global_irq 5 dfl dfl)
  ACPI: INT_SRC_OVR (bus 0 bus_irq 3 global_irq 6 dfl dfl)
  ACPI: INT_SRC_OVR (bus 0 bus_irq 4 global_irq 7 dfl dfl)
  ACPI: INT_SRC_OVR (bus 0 bus_irq 6 global_irq 9 dfl dfl)
  ACPI: INT_SRC_OVR (bus 0 bus_irq 7 global_irq 10 dfl dfl)
  ACPI: INT_SRC_OVR (bus 0 bus_irq 8 global_irq 11 low edge)
  ACPI: INT_SRC_OVR (bus 0 bus_irq 9 global_irq 12 dfl dfl)
  ACPI: INT_SRC_OVR (bus 0 bus_irq 12 global_irq 15 dfl dfl)
  ACPI: INT_SRC_OVR (bus 0 bus_irq 13 global_irq 16 dfl dfl)
  ACPI: INT_SRC_OVR (bus 0 bus_irq 14 global_irq 17 low edge)
  ACPI: INT_SRC_OVR (bus 0 bus_irq 15 global_irq 18 dfl dfl)

It turns out that the system has three io apic controllers, but
boot ioapic routing is in the second one, and that gsi_base is
not 0 - it is using a bunch of INT_SRC_OVR...

So these recent changes:

 1. one set routing for first io apic controller
 2. assume irq = gsi

... will break that system.

So try to remap those gsis, need to seperate boot_ioapic_idx
detection out of enable_IO_APIC() and call them early.

So introduce boot_ioapic_idx, and remap_ioapic_gsi()...

 -v2: shift gsi with delta instead of gsi_base of boot_ioapic_idx

 -v3: double check with find_isa_irq_apic(0, mp_INT) to get right
      boot_ioapic_idx

 -v4: nr_legacy_irqs

 -v5: add print out for boot_ioapic_idx, and also make it could be
      applied for current kernel and previous kernel

 -v6: add bus_irq, in acpi_sci_ioapic_setup, so can get overwride
      for sci right mapping...

 -v7: looks like pnpacpi get irq instead of gsi, so need to revert
      them back...

 -v8: split into two patches

 -v9: according to Eric, use fixed 16 for shifting instead of remap

 -v10: still need to touch rsparser.c

 -v11: just revert back to way Eric suggest...
      anyway the ioapic in first ioapic is blocked by second...

 -v12: two patches, this one will add more loop but check apic_id and irq > 16

Reported-by: Iranna D Ankad <iranna.ankad@in.ibm.com>
Bisected-by: Iranna D Ankad <iranna.ankad@in.ibm.com>
Tested-by: Gary Hade <garyhade@us.ibm.com>
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Cc: Eric W. Biederman <ebiederm@xmission.com>
Cc: Thomas Renninger <trenn@suse.de>
Cc: Eric W. Biederman <ebiederm@xmission.com>
Cc: Suresh Siddha <suresh.b.siddha@intel.com>
Cc: len.brown@intel.com
LKML-Reference: <4B8A321A.1000008@kernel.org>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agomemory hotplug: fix a bug on /dev/mem for 64-bit kernels
Shaohui Zheng [Tue, 2 Feb 2010 21:44:16 +0000 (13:44 -0800)]
memory hotplug: fix a bug on /dev/mem for 64-bit kernels

commit ea0854170c95245a258b386c7a9314399c949fe0 upstream.

Newly added memory can not be accessed via /dev/mem, because we do not
update the variables high_memory, max_pfn and max_low_pfn.

Add a function update_end_of_memory_vars() to update these variables for
64-bit kernels.

[akpm@linux-foundation.org: simplify comment]
Signed-off-by: Shaohui Zheng <shaohui.zheng@intel.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Li Haicheng <haicheng.li@intel.com>
Reviewed-by: Wu Fengguang <fengguang.wu@intel.com>
Reviewed-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agocrypto: testmgr - Fix complain about lack test for internal used algorithm
Song Youquan [Wed, 23 Dec 2009 11:45:20 +0000 (19:45 +0800)]
crypto: testmgr - Fix complain about lack test for internal used algorithm

commit 863b557a88f8c033f7419fabafef4712a5055f85 upstream.

When load aesni-intel and ghash_clmulni-intel driver,kernel will complain no
 test for some internal used algorithm.
The strange information as following:

alg: No test for __aes-aesni (__driver-aes-aesni)
alg: No test for __ecb-aes-aesni (__driver-ecb-aes-aesni)
alg: No test for __cbc-aes-aesni (__driver-cbc-aes-aesni)
alg: No test for __ecb-aes-aesni (cryptd(__driver-ecb-aes-aesni)
alg: No test for __ghash (__ghash-pclmulqdqni)
alg: No test for __ghash (cryptd(__ghash-pclmulqdqni))

This patch add NULL test entries for these algorithm and driver.

Signed-off-by: Song Youquan <youquan.song@intel.com>
Signed-off-by: Hang Ying <ying.huang@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Acked-by: Jiri Kosina <jkosina@suse.cz>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agofix SBA IOMMU to handle allocation failure properly
FUJITA Tomonori [Tue, 17 Nov 2009 22:44:35 +0000 (14:44 -0800)]
fix SBA IOMMU to handle allocation failure properly

commit e2a465675dc089e9a56ba2fa2a5fbd9bd8844d18 upstream.

It's possible that SBA IOMMU might fail to find I/O space under heavy
I/Os.  SBA IOMMU panics on allocation failure but it shouldn't; drivers
can handle the failure.  The majority of other IOMMU drivers don't panic
on allocation failure.

This patch fixes SBA IOMMU path to handle allocation failure properly.

Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Cc: Fenghua Yu <fenghua.yu@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Tony Luck <tony.luck@intel.com>
Acked-by: Leonardo Chiquitto <lchiquitto@novell.com>
Acked-by: Jeff Mahoney <jeffm@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agomutex: Don't spin when the owner CPU is offline or other weird cases
Benjamin Herrenschmidt [Fri, 16 Apr 2010 21:20:00 +0000 (23:20 +0200)]
mutex: Don't spin when the owner CPU is offline or other weird cases

commit 4b402210486c6414fe5fbfd85934a0a22da56b04 upstream.

Due to recent load-balancer changes that delay the task migration to
the next wakeup, the adaptive mutex spinning ends up in a live lock
when the owner's CPU gets offlined because the cpu_online() check
lives before the owner running check.

This patch changes mutex_spin_on_owner() to return 0 (don't spin) in
any case where we aren't sure about the owner struct validity or CPU
number, and if the said CPU is offline. There is no point going back &
re-evaluate spinning in corner cases like that, let's just go to
sleep.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <1271212509.13059.135.camel@pasglop>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agosched, cputime: Introduce thread_group_times()
Hidetoshi Seto [Wed, 2 Dec 2009 08:28:07 +0000 (17:28 +0900)]
sched, cputime: Introduce thread_group_times()

commit 0cf55e1ec08bb5a22e068309e2d8ba1180ab4239 upstream.

This is a real fix for problem of utime/stime values decreasing
described in the thread:

   http://lkml.org/lkml/2009/11/3/522

Now cputime is accounted in the following way:

 - {u,s}time in task_struct are increased every time when the thread
   is interrupted by a tick (timer interrupt).

 - When a thread exits, its {u,s}time are added to signal->{u,s}time,
   after adjusted by task_times().

 - When all threads in a thread_group exits, accumulated {u,s}time
   (and also c{u,s}time) in signal struct are added to c{u,s}time
   in signal struct of the group's parent.

So {u,s}time in task struct are "raw" tick count, while
{u,s}time and c{u,s}time in signal struct are "adjusted" values.

And accounted values are used by:

 - task_times(), to get cputime of a thread:
   This function returns adjusted values that originates from raw
   {u,s}time and scaled by sum_exec_runtime that accounted by CFS.

 - thread_group_cputime(), to get cputime of a thread group:
   This function returns sum of all {u,s}time of living threads in
   the group, plus {u,s}time in the signal struct that is sum of
   adjusted cputimes of all exited threads belonged to the group.

The problem is the return value of thread_group_cputime(),
because it is mixed sum of "raw" value and "adjusted" value:

  group's {u,s}time = foreach(thread){{u,s}time} + exited({u,s}time)

This misbehavior can break {u,s}time monotonicity.
Assume that if there is a thread that have raw values greater
than adjusted values (e.g. interrupted by 1000Hz ticks 50 times
but only runs 45ms) and if it exits, cputime will decrease (e.g.
-5ms).

To fix this, we could do:

  group's {u,s}time = foreach(t){task_times(t)} + exited({u,s}time)

But task_times() contains hard divisions, so applying it for
every thread should be avoided.

This patch fixes the above problem in the following way:

 - Modify thread's exit (= __exit_signal()) not to use task_times().
   It means {u,s}time in signal struct accumulates raw values instead
   of adjusted values.  As the result it makes thread_group_cputime()
   to return pure sum of "raw" values.

 - Introduce a new function thread_group_times(*task, *utime, *stime)
   that converts "raw" values of thread_group_cputime() to "adjusted"
   values, in same calculation procedure as task_times().

 - Modify group's exit (= wait_task_zombie()) to use this introduced
   thread_group_times().  It make c{u,s}time in signal struct to
   have adjusted values like before this patch.

 - Replace some thread_group_cputime() by thread_group_times().
   This replacements are only applied where conveys the "adjusted"
   cputime to users, and where already uses task_times() near by it.
   (i.e. sys_times(), getrusage(), and /proc/<PID>/stat.)

This patch have a positive side effect:

 - Before this patch, if a group contains many short-life threads
   (e.g. runs 0.9ms and not interrupted by ticks), the group's
   cputime could be invisible since thread's cputime was accumulated
   after adjusted: imagine adjustment function as adj(ticks, runtime),
     {adj(0, 0.9) + adj(0, 0.9) + ....} = {0 + 0 + ....} = 0.
   After this patch it will not happen because the adjustment is
   applied after accumulated.

v2:
 - remove if()s, put new variables into signal_struct.

Signed-off-by: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Acked-by: Peter Zijlstra <peterz@infradead.org>
Cc: Spencer Candland <spencer@bluehost.com>
Cc: Americo Wang <xiyou.wangcong@gmail.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Balbir Singh <balbir@in.ibm.com>
Cc: Stanislaw Gruszka <sgruszka@redhat.com>
LKML-Reference: <4B162517.8040909@jp.fujitsu.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agosched: Fix granularity of task_u/stime()
Hidetoshi Seto [Thu, 12 Nov 2009 04:33:45 +0000 (13:33 +0900)]
sched: Fix granularity of task_u/stime()

commit 761b1d26df542fd5eb348837351e4d2f3bc7bffe upstream.

Originally task_s/utime() were designed to return clock_t but
later changed to return cputime_t by following commit:

  commit efe567fc8281661524ffa75477a7c4ca9b466c63
  Author: Christian Borntraeger <borntraeger@de.ibm.com>
  Date:   Thu Aug 23 15:18:02 2007 +0200

It only changed the type of return value, but not the
implementation. As the result the granularity of task_s/utime()
is still that of clock_t, not that of cputime_t.

So using task_s/utime() in __exit_signal() makes values
accumulated to the signal struct to be rounded and coarse
grained.

This patch removes casts to clock_t in task_u/stime(), to keep
granularity of cputime_t over the calculation.

v2:
  Use div_u64() to avoid error "undefined reference to `__udivdi3`"
  on some 32bit systems.

Signed-off-by: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Acked-by: Peter Zijlstra <peterz@infradead.org>
Cc: xiyou.wangcong@gmail.com
Cc: Spencer Candland <spencer@bluehost.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Stanislaw Gruszka <sgruszka@redhat.com>
LKML-Reference: <4AFB9029.9000208@jp.fujitsu.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agotimekeeping: Fix clock_gettime vsyscall time warp
Lin Ming [Tue, 17 Nov 2009 05:49:50 +0000 (13:49 +0800)]
timekeeping: Fix clock_gettime vsyscall time warp

commit 0696b711e4be45fa104c12329f617beb29c03f78 upstream.

Since commit 0a544198 "timekeeping: Move NTP adjusted clock multiplier
to struct timekeeper" the clock multiplier of vsyscall is updated with
the unmodified clock multiplier of the clock source and not with the
NTP adjusted multiplier of the timekeeper.

This causes user space observerable time warps:
new CLOCK-warp maximum: 120 nsecs,  00000025c337c537 -> 00000025c337c4bf

Add a new argument "mult" to update_vsyscall() and hand in the
timekeeping internal NTP adjusted multiplier.

Signed-off-by: Lin Ming <ming.m.lin@intel.com>
Cc: "Zhang Yanmin" <yanmin_zhang@linux.intel.com>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Tony Luck <tony.luck@intel.com>
LKML-Reference: <1258436990.17765.83.camel@minggr.sh.intel.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Kurt Garloff <garloff@suse.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agonohz: Reuse ktime in sub-functions of tick_check_idle.
Martin Schwidefsky [Tue, 29 Sep 2009 12:25:15 +0000 (14:25 +0200)]
nohz: Reuse ktime in sub-functions of tick_check_idle.

commit eed3b9cf3fe3fcc7a50238dfcab63a63914e8f42 upstream.

On a system with NOHZ=y tick_check_idle calls tick_nohz_stop_idle and
tick_nohz_update_jiffies. Given the right conditions (ts->idle_active
and/or ts->tick_stopped) both function get a time stamp with ktime_get.
The same time stamp can be reused if both function require one.

On s390 this change has the additional benefit that gcc inlines the
tick_nohz_stop_idle function into tick_check_idle. The number of
instructions to execute tick_check_idle drops from 225 to 144
(without the ktime_get optimization it is 367 vs 215 instructions).

before:

 0)               |  tick_check_idle() {
 0)               |    tick_nohz_stop_idle() {
 0)               |      ktime_get() {
 0)               |        read_tod_clock() {
 0)   0.601 us    |        }
 0)   1.765 us    |      }
 0)   3.047 us    |    }
 0)               |    ktime_get() {
 0)               |      read_tod_clock() {
 0)   0.570 us    |      }
 0)   1.727 us    |    }
 0)               |    tick_do_update_jiffies64() {
 0)   0.609 us    |    }
 0)   8.055 us    |  }

after:

 0)               |  tick_check_idle() {
 0)               |    ktime_get() {
 0)               |      read_tod_clock() {
 0)   0.617 us    |      }
 0)   1.773 us    |    }
 0)               |    tick_do_update_jiffies64() {
 0)   0.593 us    |    }
 0)   4.477 us    |  }

Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: john stultz <johnstul@us.ibm.com>
LKML-Reference: <20090929122533.206589318@de.ibm.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: John Jolly <jjolly@suse.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agonohz: Introduce arch_needs_cpu
Martin Schwidefsky [Tue, 29 Sep 2009 12:25:16 +0000 (14:25 +0200)]
nohz: Introduce arch_needs_cpu

commit 3c5d92a0cfb5103c0d5ab74d4ae6373d3af38148 upstream.

Allow the architecture to request a normal jiffy tick when the system
goes idle and tick_nohz_stop_sched_tick is called . On s390 the hook is
used to prevent the system going fully idle if there has been an
interrupt other than a clock comparator interrupt since the last wakeup.

On s390 the HiperSockets response time for 1 connection ping-pong goes
down from 42 to 34 microseconds. The CPU cost decreases by 27%.

Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
LKML-Reference: <20090929122533.402715150@de.ibm.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: John Jolly <jjolly@suse.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoBtrfs: kfree correct pointer during mount option parsing
Josef Bacik [Thu, 25 Feb 2010 20:38:35 +0000 (20:38 +0000)]
Btrfs: kfree correct pointer during mount option parsing

commit da495ecc0fb096b383754952a1c152147bc95b52 upstream.

We kstrdup the options string, but then strsep screws with the pointer,
so when we kfree() it, we're not giving it the right pointer.

Tested-by: Andy Lutomirski <luto@mit.edu>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Acked-by: Jeff Mahoney <jeffm@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoBtrfs: btrfs_mark_extent_written uses the wrong slot
Shaohua Li [Thu, 11 Feb 2010 07:43:00 +0000 (07:43 +0000)]
Btrfs: btrfs_mark_extent_written uses the wrong slot

commit 3f6fae9559225741c91f1320090b285da1413290 upstream.

My test do: fallocate a big file and do write. The file is 512M, but
after file write is done btrfs-debug-tree shows:
item 6 key (257 EXTENT_DATA 0) itemoff 3516 itemsize 53
                extent data disk byte 1103101952 nr 536870912
                extent data offset 0 nr 399634432 ram 536870912
                extent compression 0
Looks like a regression introducted by
6c7d54ac87f338c479d9729e8392eca3f76e11e1, where we set wrong slot.

Signed-off-by: Shaohua Li <shaohua.li@intel.com>
Acked-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Acked-by: Jeff Mahoney <jeffm@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoBtrfs: apply updated fallocate i_size fix
Aneesh Kumar K.V [Thu, 4 Feb 2010 16:33:03 +0000 (11:33 -0500)]
Btrfs: apply updated fallocate i_size fix

commit 23b5c50945f2294add0137799400329c0ebba290 upstream.

This version of the i_size fix for fallocate makes sure we only update
the i_size when the current fallocate is really operating outside of
i_size.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
Acked-by: Jeff Mahoney <jeffm@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoBtrfs: do not try and lookup the file extent when finishing ordered io
Josef Bacik [Tue, 2 Feb 2010 20:50:10 +0000 (20:50 +0000)]
Btrfs: do not try and lookup the file extent when finishing ordered io

commit efd049fb26a162c3830fd3cb1001fdc09b147f3b upstream.

When running the following fio job

[torrent]
filename=torrent-test
rw=randwrite
size=4g
filesize=4g
bs=4k
ioengine=sync

you would see long stalls where no work was being done.  That is because we were
doing all this extra work to read in the file extent outside of the transaction,
however in the random io case this ends up hurting us because the file extents
are not there to begin with.  So axe this logic, since we end up reading in the
file extent when we go to update it anyway.  This took the fio job from 11 mb/s
with several ~10 second stalls to 24 mb/s to a couple of 1-2 second stalls.

Signed-off-by: Josef Bacik <josef@redhat.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Acked-by: Jeff Mahoney <jeffm@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoBtrfs: Fix oopsen when dropping empty tree.
Yan, Zheng [Mon, 1 Feb 2010 02:41:17 +0000 (02:41 +0000)]
Btrfs: Fix oopsen when dropping empty tree.

commit 7a7965f83e89f0be506a96769938a721e4e5ae50 upstream.

When dropping a empty tree, walk_down_tree() skips checking
extent information for the tree root. This will triggers a
BUG_ON in walk_up_proc().

Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Acked-by: Jeff Mahoney <jeffm@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoBtrfs: remove BUG_ON() due to mounting bad filesystem
Miao Xie [Tue, 2 Feb 2010 08:46:44 +0000 (08:46 +0000)]
Btrfs: remove BUG_ON() due to mounting bad filesystem

commit d7ce5843bb28ada6845ab2ae8510ba3f12d33154 upstream.

Mounting a bad filesystem caused a BUG_ON(). The following is steps to
reproduce it.
 # mkfs.btrfs /dev/sda2
 # mount /dev/sda2 /mnt
 # mkfs.btrfs /dev/sda1 /dev/sda2
 (the program says that /dev/sda2 was mounted, and then exits. )
 # umount /mnt
 # mount /dev/sda1 /mnt

At the third step, mkfs.btrfs exited in the way of make filesystem. So the
initialization of the filesystem didn't finish. So the filesystem was bad, and
it caused BUG_ON() when mounting it. But BUG_ON() should be called by the wrong
code, not user's operation, so I think it is a bug of btrfs.

This patch fixes it.

Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Acked-by: Jeff Mahoney <jeffm@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoBtrfs: make error return negative in btrfs_sync_file()
Roel Kluin [Fri, 29 Jan 2010 10:42:11 +0000 (10:42 +0000)]
Btrfs: make error return negative in btrfs_sync_file()

commit 014e4ac4f7d9c981750491fa40ea35efadc9ed49 upstream.

It appears the error return should be negative

Signed-off-by: Roel Kluin <roel.kluin@gmail.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Acked-by: Jeff Mahoney <jeffm@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoBtrfs: fix race between allocate and release extent buffer.
Yan, Zheng [Thu, 4 Feb 2010 08:46:56 +0000 (08:46 +0000)]
Btrfs: fix race between allocate and release extent buffer.

commit f044ba7835b84e69c68b620ca8fa27e5ef67759d upstream.

Increase extent buffer's reference count while holding the lock.
Otherwise it can race with try_release_extent_buffer.

Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Acked-by: Jeff Mahoney <jeffm@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoBtrfs: check total number of devices when removing missing
Josef Bacik [Wed, 27 Jan 2010 02:09:38 +0000 (02:09 +0000)]
Btrfs: check total number of devices when removing missing

commit 035fe03a7ad56982b30ab3a522b7b08d58feccd0 upstream.

If you have a disk failure in RAID1 and then add a new disk to the
array, and then try to remove the missing volume, it will fail.  The
reason is the sanity check only looks at the total number of rw devices,
which is just 2 because we have 2 good disks and 1 bad one.  Instead
check the total number of devices in the array to make sure we can
actually remove the device.  Tested this with a failed disk setup and
with this test we can now run

btrfs-vol -r missing /mount/point

and it works fine.

Signed-off-by: Josef Bacik <josef@redhat.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Acked-by: Jeff Mahoney <jeffm@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoBtrfs: check return value of open_bdev_exclusive properly
Josef Bacik [Wed, 27 Jan 2010 02:09:00 +0000 (02:09 +0000)]
Btrfs: check return value of open_bdev_exclusive properly

commit 7f59203abeaf18bf3497b308891f95a4489810ad upstream.

Hit this problem while testing RAID1 failure stuff.  open_bdev_exclusive
returns ERR_PTR(), not NULL.  So change the return value properly.  This
is important if you accidently specify a device that doesn't exist when
trying to add a new device to an array, you will panic the box
dereferencing bdev.

Signed-off-by: Josef Bacik <josef@redhat.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Acked-by: Jeff Mahoney <jeffm@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoBtrfs: do not mark the chunk as readonly if in degraded mode
Josef Bacik [Wed, 27 Jan 2010 02:07:59 +0000 (02:07 +0000)]
Btrfs: do not mark the chunk as readonly if in degraded mode

commit f48b90756bd834dda852ff514f2690d3175b1f44 upstream.

If a RAID setup has chunks that span multiple disks, and one of those
disks has failed, btrfs_chunk_readonly will return 1 since one of the
disks in that chunk's stripes is dead and therefore not writeable.  So
instead if we are in degraded mode, return 0 so we can go ahead and
allocate stuff.  Without this patch all of the block groups in a RAID1
setup will end up read-only, which will mean we can't add new disks to
the array since we won't be able to make allocations.

Signed-off-by: Josef Bacik <josef@redhat.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Acked-by: Jeff Mahoney <jeffm@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoBtrfs: run orphan cleanup on default fs root
Josef Bacik [Tue, 26 Jan 2010 14:30:53 +0000 (14:30 +0000)]
Btrfs: run orphan cleanup on default fs root

commit e3acc2a6850efff647f1c5458524eb3a8bcba20a upstream.

This patch revert's commit

6c090a11e1c403b727a6a8eff0b97d5fb9e95cb5

Since it introduces this problem where we can run orphan cleanup on a
volume that can have orphan entries re-added.  Instead of my original
fix, Yan Zheng pointed out that we can just revert my original fix and
then run the orphan cleanup in open_ctree after we look up the fs_root.
I have tested this with all the tests that gave me problems and this
patch fixes both problems.  Thanks,

Signed-off-by: Josef Bacik <josef@redhat.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Acked-by: Jeff Mahoney <jeffm@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoBtrfs: fix a memory leak in btrfs_init_acl
Yang Hongyang [Tue, 26 Jan 2010 00:48:23 +0000 (00:48 +0000)]
Btrfs: fix a memory leak in btrfs_init_acl

commit f858153c367a397235d3e81136741e40e44faf1d upstream.

In btrfs_init_acl() cloned acl is not released

Signed-off-by: Yang Hongyang <yanghy@cn.fujitsu.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Acked-by: Jeff Mahoney <jeffm@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoBtrfs: Use correct values when updating inode i_size on fallocate
Aneesh Kumar K.V [Wed, 20 Jan 2010 07:28:54 +0000 (07:28 +0000)]
Btrfs: Use correct values when updating inode i_size on fallocate

commit d1ea6a61454e7d7ff0873d0ad1ae27d5807da0d3 upstream.

commit f2bc9dd07e3424c4ec5f3949961fe053d47bc825
Author: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Date:   Wed Jan 20 12:57:53 2010 +0530

    Btrfs: Use correct values when updating inode i_size on fallocate

    Even though we allocate more, we should be updating inode i_size
    as per the arguments passed

Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Acked-by: Jeff Mahoney <jeffm@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoBtrfs: fix possible panic on unmount
Josef Bacik [Fri, 13 Nov 2009 20:12:59 +0000 (20:12 +0000)]
Btrfs: fix possible panic on unmount

commit 11dfe35a0108097f2df1f042c485fa7f758c2cdf upstream.

We can race with the unmount of an fs and the stopping of a kthread where we
will free the block group before we're done using it.  The reason for this is
because we do not hold a reference on the block group while its caching, since
the allocator drops its reference once it exits or moves on to the next block
group.  This patch fixes the problem by taking a reference to the block group
before we start caching and dropping it when we're done to make sure all
accesses to the block group are safe.  Thanks,

Signed-off-by: Josef Bacik <josef@redhat.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Acked-by: Jeff Mahoney <jeffm@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoBtrfs: deal with NULL acl sent to btrfs_set_acl
Chris Mason [Mon, 18 Jan 2010 01:36:18 +0000 (20:36 -0500)]
Btrfs: deal with NULL acl sent to btrfs_set_acl

commit a9cc71a60c29a09174bee2fcef8f924c529fd4b7 upstream.

It is legal for btrfs_set_acl to be sent a NULL acl.  This
makes sure we don't dereference it.  A similar patch was sent by
Johannes Hirte <johannes.hirte@fem.tu-ilmenau.de>

Signed-off-by: Chris Mason <chris.mason@oracle.com>
Acked-by: Jeff Mahoney <jeffm@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoBtrfs: fix regression in orphan cleanup
Josef Bacik [Fri, 15 Jan 2010 20:08:22 +0000 (20:08 +0000)]
Btrfs: fix regression in orphan cleanup

commit 6c090a11e1c403b727a6a8eff0b97d5fb9e95cb5 upstream.

Currently orphan cleanup only ever gets triggered if we cross subvolumes during
a lookup, which means that if we just mount a plain jane fs that has orphans in
it, they will never get cleaned up.  This results in panic's like these

http://www.kerneloops.org/oops.php?number=1109085

where adding an orphan entry results in -EEXIST being returned and we panic.  In
order to fix this, we check to see on lookup if our root has had the orphan
cleanup done, and if not go ahead and do it.  This is easily reproduceable by
running this testcase

#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <string.h>
#include <unistd.h>
#include <stdio.h>

int main(int argc, char **argv)
{
char data[4096];
char newdata[4096];
int fd1, fd2;

memset(data, 'a', 4096);
memset(newdata, 'b', 4096);

while (1) {
int i;

fd1 = creat("file1", 0666);
if (fd1 < 0)
break;

for (i = 0; i < 512; i++)
write(fd1, data, 4096);

fsync(fd1);
close(fd1);

fd2 = creat("file2", 0666);
if (fd2 < 0)
break;

ftruncate(fd2, 4096 * 512);

for (i = 0; i < 512; i++)
write(fd2, newdata, 4096);
close(fd2);

i = rename("file2", "file1");
unlink("file1");
}

return 0;
}

and then pulling the power on the box, and then trying to run that test again
when the box comes back up.  I've tested this locally and it fixes the problem.
Thanks to Tomas Carnecky for helping me track this down initially.

Signed-off-by: Josef Bacik <josef@redhat.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Acked-by: Jeff Mahoney <jeffm@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoBtrfs: Fix race in btrfs_mark_extent_written
Yan, Zheng [Fri, 15 Jan 2010 08:43:09 +0000 (08:43 +0000)]
Btrfs: Fix race in btrfs_mark_extent_written

commit 6c7d54ac87f338c479d9729e8392eca3f76e11e1 upstream.

Fix bug reported by Johannes Hirte. The reason of that bug
is btrfs_del_items is called after btrfs_duplicate_item and
btrfs_del_items triggers tree balance. The fix is check that
case and call btrfs_search_slot when needed.

Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Acked-by: Jeff Mahoney <jeffm@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoBtrfs, fix memory leaks in error paths
Jiri Slaby [Wed, 6 Jan 2010 16:57:22 +0000 (16:57 +0000)]
Btrfs, fix memory leaks in error paths

commit 2423fdfb96e3f9ff3baeb6c4c78d74145547891d upstream.

Stanse found 2 memory leaks in relocate_block_group and
__btrfs_map_block. cluster and multi are not freed/assigned on all
paths. Fix that.

Signed-off-by: Jiri Slaby <jslaby@suse.cz>
Cc: linux-btrfs@vger.kernel.org
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Acked-by: Jeff Mahoney <jeffm@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoBtrfs: align offsets for btrfs_ordered_update_i_size
Yan, Zheng [Mon, 28 Dec 2009 05:01:58 +0000 (05:01 +0000)]
Btrfs: align offsets for btrfs_ordered_update_i_size

commit a038fab0cb873c75d6675e2bcffce8a3935bdce7 upstream.

Some callers of btrfs_ordered_update_i_size can now pass in
a NULL for the ordered extent to update against.  This makes
sure we properly align the offset they pass in when deciding
how much to bump the on disk i_size.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
Acked-by: Jeff Mahoney <jeffm@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agobtrfs: fix missing last-entry in readdir(3)
Jan Engelhardt [Wed, 9 Dec 2009 22:00:38 +0000 (22:00 +0000)]
btrfs: fix missing last-entry in readdir(3)

commit 406266ab9ac8ed8b085c58aacd9e3161480dc5d5 upstream.

parent 49313cdac7b34c9f7ecbb1780cfc648b1c082cd7 (v2.6.32-1-g49313cd)
commit ff48c08e1c05c67e8348ab6f8a24de8034e0e34d
Author: Jan Engelhardt <jengelh@medozas.de>
Date:   Wed Dec 9 22:57:36 2009 +0100

Btrfs: fix missing last-entry in readdir(3)

When one does a 32-bit readdir(3), the last entry of a directory is
missing. This is however not due to passing a large value to filldir,
but it seems to have to do with glibc doing telldir or something
quirky. In any case, this patch fixes it in practice.

Signed-off-by: Jan Engelhardt <jengelh@medozas.de>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Acked-by: Jeff Mahoney <jeffm@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoBtrfs: make sure fallocate properly starts a transaction
Chris Mason [Thu, 17 Dec 2009 20:47:17 +0000 (15:47 -0500)]
Btrfs: make sure fallocate properly starts a transaction

commit 3a1abec9f6880cf406593c392636199ea1c6c917 upstream.

The recent patch to make fallocate enospc friendly would send
down a NULL trans handle to the allocator.  This moves the
transaction start to properly fix things.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
Acked-by: Jeff Mahoney <jeffm@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoBtrfs: make metadata chunks smaller
Josef Bacik [Mon, 7 Dec 2009 21:45:59 +0000 (21:45 +0000)]
Btrfs: make metadata chunks smaller

commit 83d3c9696fed237a3d96fce18299e2fcf112109f upstream.

This patch makes us a bit less zealous about making sure we have enough free
metadata space by pearing down the size of new metadata chunks to 256mb instead
of 1gb.  Also, we used to try an allocate metadata chunks when allocating data,
but that sort of thing is done elsewhere now so we can just remove it.  With my
-ENOSPC test I used to have 3gb reserved for metadata out of 75gb, now I have
1.7gb.  Thanks,

Signed-off-by: Josef Bacik <josef@redhat.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Acked-by: Jeff Mahoney <jeffm@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoBtrfs: Show discard option in /proc/mounts
Matthew Wilcox [Mon, 14 Dec 2009 22:01:12 +0000 (22:01 +0000)]
Btrfs: Show discard option in /proc/mounts

commit 20a5239a5d0f340e29827a6a2d28a138001c44b8 upstream.

Christoph's patch e244a0aeb6a599c19a7c802cda6e2d67c847b154 doesn't display
the discard option in /proc/mounts, leading to some confusion for me.
Here's the missing bit.

Signed-off-by: Matthew Wilcox <willy@linux.intel.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Acked-by: Jeff Mahoney <jeffm@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoBtrfs: deny sys_link across subvolumes.
TARUISI Hiroaki [Thu, 12 Nov 2009 07:14:26 +0000 (07:14 +0000)]
Btrfs: deny sys_link across subvolumes.

commit 4a8be425a8fb8fbb5d881eb55fa6634c3463b9c9 upstream.

I rebased Christian Parpart's patch to deny hard link across
subvolumes. Original patch modifies also btrfs_rename, but
I excluded it because we can move across subvolumes now and
it make no problem.
-----------------

Hard link across subvolumes should not allowed in Btrfs.
btrfs_link checks root of 'to' directory is same as root
of 'from' file. If not same, btrfs_link returns -EPERM.

Signed-off-by: TARUISI Hiroaki <taruishi.hiroak@jp.fujitsu.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Acked-by: Jeff Mahoney <jeffm@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoBtrfs: fail mount on bad mount options
Sage Weil [Sat, 7 Nov 2009 06:19:16 +0000 (06:19 +0000)]
Btrfs: fail mount on bad mount options

commit a7a3f7cadd9bdee569243f7ead9550aa16b60e07 upstream.

We shouldn't silently ignore unrecognized options.

Signed-off-by: Sage Weil <sage@newdream.net>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Acked-by: Jeff Mahoney <jeffm@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoBtrfs: don't add extent 0 to the free space cache v2
Yan, Zheng [Thu, 26 Nov 2009 09:31:11 +0000 (09:31 +0000)]
Btrfs: don't add extent 0 to the free space cache v2

commit 06b2331f8333ec6edf41662757ce8882cc1747d5 upstream.

If block group 0 is completely free, btrfs_read_block_groups will
add extent [0, BTRFS_SUPER_INFO_OFFSET) to the free space cache.

Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Acked-by: Jeff Mahoney <jeffm@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoBtrfs: Fix per root used space accounting
Yan, Zheng [Thu, 12 Nov 2009 09:36:50 +0000 (09:36 +0000)]
Btrfs: Fix per root used space accounting

commit 86b9f2eca5e0984145e3c7698a7cd6dd65c2a93f upstream.

The bytes_used field in root item was originally planned to
trace the amount of used data and tree blocks. But it never
worked right since we can't trace freeing of data accurately.
This patch changes it to only trace the amount of tree blocks.

Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Acked-by: Jeff Mahoney <jeffm@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
13 years agoBtrfs: Fix btrfs_drop_extent_cache for skip pinned case
Yan, Zheng [Thu, 12 Nov 2009 09:36:44 +0000 (09:36 +0000)]
Btrfs: Fix btrfs_drop_extent_cache for skip pinned case

commit 55ef68990029fcd8d04d42fc184aa7fb18cf309e upstream.

The check for skip pinned case is wrong, it may breaks the
while loop too soon.

Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
Acked-by: Jeff Mahoney <jeffm@suse.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>