pandora-kernel.git
11 years agocpuset: fix cpuset_print_task_mems_allowed() vs rename() race
Li Zefan [Fri, 25 Jan 2013 08:08:01 +0000 (16:08 +0800)]
cpuset: fix cpuset_print_task_mems_allowed() vs rename() race

rename() will change dentry->d_name. The result of this race can
be worse than seeing partially rewritten name, but we might access
a stale pointer because rename() will re-allocate memory to hold
a longer name.

It's safe in the protection of dentry->d_lock.

v2: check NULL dentry before acquiring dentry lock.

Signed-off-by: Li Zefan <lizefan@huawei.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: stable@vger.kernel.org
11 years agocgroup: fix exit() vs rmdir() race
Li Zefan [Thu, 24 Jan 2013 06:43:28 +0000 (14:43 +0800)]
cgroup: fix exit() vs rmdir() race

In cgroup_exit() put_css_set_taskexit() is called without any lock,
which might lead to accessing a freed cgroup:

thread1                           thread2
---------------------------------------------
exit()
  cgroup_exit()
    put_css_set_taskexit()
      atomic_dec(cgrp->count);
                                   rmdir();
      /* not safe !! */
      check_for_release(cgrp);

rcu_read_lock() can be used to make sure the cgroup is alive.

Signed-off-by: Li Zefan <lizefan@huawei.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: stable@vger.kernel.org
11 years agocgroup: remove bogus comments in cgroup_diput()
Li Zefan [Thu, 24 Jan 2013 06:32:02 +0000 (14:32 +0800)]
cgroup: remove bogus comments in cgroup_diput()

Since commit 48ddbe194623ae089cc0576e60363f2d2e85662a
("cgroup: make css->refcnt clearing on cgroup removal optional"),
each css holds a ref on cgroup's dentry, so cgroup_diput() won't be
called until all css' refs go down to 0, which invalids the comments.

Signed-off-by: Li Zefan <lizefan@huawei.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
11 years agocgroup: remove synchronize_rcu() from cgroup_diput()
Li Zefan [Thu, 24 Jan 2013 06:31:42 +0000 (14:31 +0800)]
cgroup: remove synchronize_rcu() from cgroup_diput()

Free cgroup via call_rcu(). The actual work is done through
workqueue.

Signed-off-by: Li Zefan <lizefan@huawei.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
11 years agocgroup: remove duplicate RCU free on struct cgroup
Li Zefan [Thu, 24 Jan 2013 06:31:27 +0000 (14:31 +0800)]
cgroup: remove duplicate RCU free on struct cgroup

When destroying a cgroup, though in cgroup_diput() we've called
synchronize_rcu(), we then still have to free it via call_rcu().

The story is, long ago to fix a race between reading /proc/sched_debug
and freeing cgroup, the code was changed to utilize call_rcu(). See
commit a47295e6bc42ad35f9c15ac66f598aa24debd4e2 ("cgroups: make
cgroup_path() RCU-safe")

As we've fixed cpu cgroup that cpu_cgroup_offline_css() is used
to unregister a task_group so there won't be concurrent access
to this task_group after synchronize_rcu() in diput(). Now we can
just kfree(cgrp).

Signed-off-by: Li Zefan <lizefan@huawei.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
11 years agosched: remove redundant NULL cgroup check in task_group_path()
Li Zefan [Thu, 24 Jan 2013 06:31:11 +0000 (14:31 +0800)]
sched: remove redundant NULL cgroup check in task_group_path()

A task_group won't be online (thus no one can see it) until
cpu_cgroup_css_online(), and at that time tg->css.cgroup has
been initialized, so this NULL check is redundant.

Signed-off-by: Li Zefan <lizefan@huawei.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
11 years agosched: split out css_online/css_offline from tg creation/destruction
Li Zefan [Thu, 24 Jan 2013 06:30:48 +0000 (14:30 +0800)]
sched: split out css_online/css_offline from tg creation/destruction

This is a preparaton for later patches.

- What do we gain from cpu_cgroup_css_online():

After ss->css_alloc() and before ss->css_online(), there's a small
window that tg->css.cgroup is NULL. With this change, tg won't be seen
before ss->css_online(), where it's added to the global list, so we're
guaranteed we'll never see NULL tg->css.cgroup.

- What do we gain from cpu_cgroup_css_offline():

tg is freed via RCU, so is cgroup. Without this change, This is how
synchronization works:

cgroup_rmdir()
  no ss->css_offline()
diput()
  syncornize_rcu()
  ss->css_free()       <-- unregister tg, and free it via call_rcu()
  kfree_rcu(cgroup)    <-- wait possible refs to cgroup, and free cgroup

We can't just kfree(cgroup), because tg might access tg->css.cgroup.

With this change:

cgroup_rmdir()
  ss->css_offline()    <-- unregister tg
diput()
  synchronize_rcu()    <-- wait possible refs to tg and cgroup
  ss->css_free()       <-- free tg
  kfree_rcu(cgroup)    <-- free cgroup

As you see, kfree_rcu() is redundant now.

Signed-off-by: Li Zefan <lizefan@huawei.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Ingo Molnar <mingo@kernel.org>
11 years agocgroup: initialize cgrp->dentry before css_alloc()
Li Zefan [Thu, 24 Jan 2013 06:30:22 +0000 (14:30 +0800)]
cgroup: initialize cgrp->dentry before css_alloc()

With this change, we're guaranteed that cgroup_path() won't see NULL
cgrp->dentry, and thus we can remove the NULL check in it.

(Well, it's not strictly true, because dummptop.dentry is always NULL
 but we already handle that separately.)

Signed-off-by: Li Zefan <lizefan@huawei.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
11 years agocgroup: remove a NULL check in cgroup_exit()
Li Zefan [Thu, 24 Jan 2013 06:43:51 +0000 (14:43 +0800)]
cgroup: remove a NULL check in cgroup_exit()

init_task.cgroups is initialized at boot phase, and whenver a ask
is forked, it's cgroups pointer is inherited from its parent, and
it's never set to NULL afterwards.

Signed-off-by: Li Zefan <lizefan@huawei.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
11 years agocgroup: fix bogus kernel warnings when cgroup_create() failed
Li Zefan [Mon, 21 Jan 2013 10:18:33 +0000 (18:18 +0800)]
cgroup: fix bogus kernel warnings when cgroup_create() failed

If cgroup_create() failed and cgroup_destroy_locked() is called to
do cleanup, we'll see a bunch of warnings:

cgroup_addrm_files: failed to remove 2MB.limit_in_bytes, err=-2
cgroup_addrm_files: failed to remove 2MB.usage_in_bytes, err=-2
cgroup_addrm_files: failed to remove 2MB.max_usage_in_bytes, err=-2
cgroup_addrm_files: failed to remove 2MB.failcnt, err=-2
cgroup_addrm_files: failed to remove prioidx, err=-2
cgroup_addrm_files: failed to remove ifpriomap, err=-2
...

We failed to remove those files, because cgroup_create() has failed
before creating those cgroup files.

To fix this, we simply don't warn if cgroup_rm_file() can't find the
cft entry.

Signed-off-by: Li Zefan <lizefan@huawei.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
11 years agocgroup: remove synchronize_rcu() from rebind_subsystems()
Li Zefan [Mon, 14 Jan 2013 09:24:18 +0000 (17:24 +0800)]
cgroup: remove synchronize_rcu() from rebind_subsystems()

Nothing's protected by RCU in rebind_subsystems(), and I can't think
of a reason why it is needed.

Signed-off-by: Li Zefan <lizefan@huawei.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
11 years agocgroup: remove synchronize_rcu() from cgroup_attach_{task|proc}()
Li Zefan [Mon, 14 Jan 2013 09:23:26 +0000 (17:23 +0800)]
cgroup: remove synchronize_rcu() from cgroup_attach_{task|proc}()

These 2 syncronize_rcu()s make attaching a task to a cgroup
quite slow, and it can't be ignored in some situations.

A real case from Colin Cross: Android uses cgroups heavily to
manage thread priorities, putting threads in a background group
with reduced cpu.shares when they are not visible to the user,
and in a foreground group when they are. Some RPCs from foreground
threads to background threads will temporarily move the background
thread into the foreground group for the duration of the RPC.
This results in many calls to cgroup_attach_task.

In cgroup_attach_task() it's task->cgroups that is protected by RCU,
and put_css_set() calls kfree_rcu() to free it.

If we remove this synchronize_rcu(), there can be threads in RCU-read
sections accessing their old cgroup via current->cgroups with
concurrent rmdir operation, but this is safe.

 # time for ((i=0; i<50; i++)) { echo $$ > /mnt/sub/tasks; echo $$ > /mnt/tasks; }

real    0m2.524s
user    0m0.008s
sys     0m0.004s

With this patch:

real    0m0.004s
user    0m0.004s
sys     0m0.000s

tj: These synchronize_rcu()s are utterly confused.  synchornize_rcu()
    necessarily has to come between two operations to guarantee that
    the changes made by the former operation are visible to all rcu
    readers before proceeding to the latter operation.  Here,
    synchornize_rcu() are at the end of attach operations with nothing
    beyond it.  Its only effect would be delaying completion of
    write(2) to sysfs tasks/procs files until all rcu readers see the
    change, which doesn't mean anything.

Signed-off-by: Li Zefan <lizefan@huawei.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
Reported-by: Colin Cross <ccross@google.com>
11 years agocgroup: use new hashtable implementation
Li Zefan [Thu, 10 Jan 2013 03:49:27 +0000 (11:49 +0800)]
cgroup: use new hashtable implementation

Switch cgroup to use the new hashtable implementation. No functional changes.

Signed-off-by: Li Zefan <lizefan@huawei.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
11 years agocgroups: fix cgroup_event_listener error handling
Greg Thelen [Mon, 7 Jan 2013 19:50:17 +0000 (11:50 -0800)]
cgroups: fix cgroup_event_listener error handling

The error handling in cgroup_event_listener.c did not correctly deal
with either an error opening either <control_file> or
cgroup.event_control.  Due to an uninitialized variable the program
exit code was undefined if either of these opens failed.

This patch simplifies and corrects cgroup_event_listener.c error
handling by:
1. using err*() rather than printf(),exit()
2. depending on process exit to close open files

With this patch failures always return non-zero error.

Signed-off-by: Greg Thelen <gthelen@google.com>
Acked-by: Li Zefan <lizefan@huawei.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
11 years agocgroups: move cgroup_event_listener.c to tools/cgroup
Greg Thelen [Fri, 4 Jan 2013 21:05:17 +0000 (13:05 -0800)]
cgroups: move cgroup_event_listener.c to tools/cgroup

Move the cgroup_event_listener.c tool from Documentation into the new
tools/cgroup directory.

This change involves wiring cgroup_event_listener.c into the tools/
make system so that is can be built with:
  $ make tools/cgroup

Signed-off-by: Greg Thelen <gthelen@google.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
11 years agocgroup: implement cgroup_rightmost_descendant()
Tejun Heo [Mon, 7 Jan 2013 16:49:33 +0000 (08:49 -0800)]
cgroup: implement cgroup_rightmost_descendant()

Implement cgroup_rightmost_descendant() which returns the right most
descendant of the specified cgroup.  This can be used to skip the
cgroup's subtree while iterating with
cgroup_for_each_descendant_pre().

Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Michal Hocko <mhocko@suse.cz>
Acked-by: Li Zefan <lizefan@huawei.com>
11 years agocgroup: remove unused dummy cgroup_fork_callbacks()
Tejun Heo [Fri, 28 Dec 2012 21:18:28 +0000 (13:18 -0800)]
cgroup: remove unused dummy cgroup_fork_callbacks()

5edee61ede ("cgroup: cgroup_subsys->fork() should be called after the
task is added to css_set") removed cgroup_fork_callbacks() but forgot
to remove its dummy version for !CONFIG_CGROUPS.  Remove it.

Signed-off-by: Tejun Heo <tj@kernel.org>
Reported-by: Herton Ronaldo Krzesinski <herton.krzesinski@canonical.com>
11 years agoLinux 3.8-rc2 v3.8-rc2
Linus Torvalds [Thu, 3 Jan 2013 02:13:21 +0000 (18:13 -0800)]
Linux 3.8-rc2

11 years agoMerge branch 'fixes-for-3.8' of git://git.kernel.org/pub/scm/linux/kernel/git/coolone...
Linus Torvalds [Thu, 3 Jan 2013 02:12:35 +0000 (18:12 -0800)]
Merge branch 'fixes-for-3.8' of git://git./linux/kernel/git/cooloney/linux-leds

Pull LED fix from Bryan Wu.

* 'fixes-for-3.8' of git://git.kernel.org/pub/scm/linux/kernel/git/cooloney/linux-leds:
  leds: leds-gpio: set devm_gpio_request_one() flags param correctly

11 years agoleds: leds-gpio: set devm_gpio_request_one() flags param correctly
Javier Martinez Canillas [Thu, 20 Dec 2012 12:56:59 +0000 (04:56 -0800)]
leds: leds-gpio: set devm_gpio_request_one() flags param correctly

commit a99d76f leds: leds-gpio: use gpio_request_one

changed the leds-gpio driver to use gpio_request_one() instead
of gpio_request() + gpio_direction_output()

Unfortunately, it also made a semantic change that breaks the
leds-gpio driver.

The gpio_request_one() flags parameter was set to:

GPIOF_DIR_OUT | (led_dat->active_low ^ state)

Since GPIOF_DIR_OUT is 0, the final flags value will just be the
XOR'ed value of led_dat->active_low and state.

This value were used to distinguish between HIGH/LOW output initial
level and call gpio_direction_output() accordingly.

With this new semantic gpio_request_one() will take the flags value
of 1 as a configuration of input direction (GPIOF_DIR_IN) and will
call gpio_direction_input() instead of gpio_direction_output().

int gpio_request_one(unsigned gpio, unsigned long flags, const char *label)
{
..
if (flags & GPIOF_DIR_IN)
err = gpio_direction_input(gpio);
else
err = gpio_direction_output(gpio,
(flags & GPIOF_INIT_HIGH) ? 1 : 0);
..
}

The right semantic is to evaluate led_dat->active_low ^ state and
set the output initial level explicitly.

Signed-off-by: Javier Martinez Canillas <javier.martinez@collabora.co.uk>
Reported-by: Arnaud Patard <arnaud.patard@rtp-net.org>
Tested-by: Ezequiel Garcia <ezequiel.garcia@free-electrons.com>
Signed-off-by: Bryan Wu <cooloney@gmail.com>
11 years agoMerge git://www.linux-watchdog.org/linux-watchdog
Linus Torvalds [Thu, 3 Jan 2013 01:46:14 +0000 (17:46 -0800)]
Merge git://www.linux-watchdog.org/linux-watchdog

Pull watchdog fixes from Wim Van Sebroeck:
 "This fixes some small errors in the new da9055 driver, eliminates a
  compiler warning and adds DT support for the twl4030_wdt driver (so
  that we can have multiple watchdogs with DT on the omap platforms)."

* git://www.linux-watchdog.org/linux-watchdog:
  watchdog: twl4030_wdt: add DT support
  watchdog: omap_wdt: eliminate unused variable and a compiler warning
  watchdog: da9055: Don't update wdt_dev->timeout in da9055_wdt_set_timeout error path
  watchdog: da9055: Fix invalid free of devm_ allocated data

11 years agoMerge tag '3.8-pci-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/helgaas/pci
Linus Torvalds [Thu, 3 Jan 2013 01:44:29 +0000 (17:44 -0800)]
Merge tag '3.8-pci-fixes' of git://git./linux/kernel/git/helgaas/pci

Pull PCI updates from Bjorn Helgaas:
 "Some fixes for v3.8.  They include a fix for the new SR-IOV sysfs
  management support, an expanded quirk for Ricoh SD card readers, a
  Stratus DMI quirk fix, and a PME polling fix."

* tag '3.8-pci-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/helgaas/pci:
  PCI: Reduce Ricoh 0xe822 SD card reader base clock frequency to 50MHz
  PCI/PM: Do not suspend port if any subordinate device needs PME polling
  PCI: Add PCIe Link Capability link speed and width names
  PCI: Work around Stratus ftServer broken PCIe hierarchy (fix DMI check)
  PCI: Remove spurious error for sriov_numvfs store and simplify flow

11 years agoUAPI: Strip _UAPI prefix on header install no matter the whitespace
David Howells [Wed, 2 Jan 2013 15:13:02 +0000 (15:13 +0000)]
UAPI: Strip _UAPI prefix on header install no matter the whitespace

Commit 56c176c9cac9 ("UAPI: strip the _UAPI prefix from header guards
during header installation") strips the _UAPI prefix from header guards,
but only if there's a single space between the cpp directive and the
label.

Make it more flexible and able to handle tabs and multiple white space
characters.

Signed-off-by: David Howells <dhowell@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoUAPI: Remove empty Kbuild files
David Howells [Wed, 2 Jan 2013 15:12:55 +0000 (15:12 +0000)]
UAPI: Remove empty Kbuild files

Empty files can get deleted by the patch program, so remove empty Kbuild
files and their links from the parent Kbuilds.

Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoMerge tag 'ecryptfs-3.8-rc2-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Thu, 3 Jan 2013 01:33:50 +0000 (17:33 -0800)]
Merge tag 'ecryptfs-3.8-rc2-fixes' of git://git./linux/kernel/git/tyhicks/ecryptfs

Pull ecryptfs fixes from Tyler Hicks:
 "Two self-explanatory fixes and a third patch which improves
  performance: when overwriting a full page in the eCryptfs page cache,
  skip reading in and decrypting the corresponding lower page."

* tag 'ecryptfs-3.8-rc2-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/tyhicks/ecryptfs:
  fs/ecryptfs/crypto.c: make ecryptfs_encode_for_filename() static
  eCryptfs: fix to use list_for_each_entry_safe() when delete items
  eCryptfs: Avoid unnecessary disk read and data decryption during writing

11 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/sage/ceph...
Linus Torvalds [Thu, 3 Jan 2013 01:32:49 +0000 (17:32 -0800)]
Merge branch 'for-linus' of git://git./linux/kernel/git/sage/ceph-client

Pull Ceph fixes from Sage Weil:
 "Two of Alex's patches deal with a race when reseting server
  connections for open RBD images, one demotes some non-fatal BUGs to
  WARNs, and my patch fixes a protocol feature bit failure path."

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/sage/ceph-client:
  libceph: fix protocol feature mismatch failure path
  libceph: WARN, don't BUG on unexpected connection states
  libceph: always reset osds when kicking
  libceph: move linger requests sooner in kick_requests()

11 years agomm: mempolicy: Convert shared_policy mutex to spinlock
Mel Gorman [Fri, 21 Dec 2012 23:10:25 +0000 (23:10 +0000)]
mm: mempolicy: Convert shared_policy mutex to spinlock

Sasha was fuzzing with trinity and reported the following problem:

  BUG: sleeping function called from invalid context at kernel/mutex.c:269
  in_atomic(): 1, irqs_disabled(): 0, pid: 6361, name: trinity-main
  2 locks held by trinity-main/6361:
   #0:  (&mm->mmap_sem){++++++}, at: [<ffffffff810aa314>] __do_page_fault+0x1e4/0x4f0
   #1:  (&(&mm->page_table_lock)->rlock){+.+...}, at: [<ffffffff8122f017>] handle_pte_fault+0x3f7/0x6a0
  Pid: 6361, comm: trinity-main Tainted: G        W
  3.7.0-rc2-next-20121024-sasha-00001-gd95ef01-dirty #74
  Call Trace:
    __might_sleep+0x1c3/0x1e0
    mutex_lock_nested+0x29/0x50
    mpol_shared_policy_lookup+0x2e/0x90
    shmem_get_policy+0x2e/0x30
    get_vma_policy+0x5a/0xa0
    mpol_misplaced+0x41/0x1d0
    handle_pte_fault+0x465/0x6a0

This was triggered by a different version of automatic NUMA balancing
but in theory the current version is vunerable to the same problem.

do_numa_page
  -> numa_migrate_prep
    -> mpol_misplaced
      -> get_vma_policy
        -> shmem_get_policy

It's very unlikely this will happen as shared pages are not marked
pte_numa -- see the page_mapcount() check in change_pte_range() -- but
it is possible.

To address this, this patch restores sp->lock as originally implemented
by Kosaki Motohiro.  In the path where get_vma_policy() is called, it
should not be calling sp_alloc() so it is not necessary to treat the PTL
specially.

Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Tested-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Mel Gorman <mgorman@suse.de>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoMerge tag 'ext4_for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso...
Linus Torvalds [Wed, 2 Jan 2013 17:57:34 +0000 (09:57 -0800)]
Merge tag 'ext4_for_linus' of git://git./linux/kernel/git/tytso/ext4

Pull ext4 bug fixes from Ted Ts'o:
 "Various bug fixes for ext4.  Perhaps the most serious bug fixed is one
  which could cause file system corruptions when performing file punch
  operations."

* tag 'ext4_for_linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tytso/ext4:
  ext4: avoid hang when mounting non-journal filesystems with orphan list
  ext4: lock i_mutex when truncating orphan inodes
  ext4: do not try to write superblock on ro remount w/o journal
  ext4: include journal blocks in df overhead calcs
  ext4: remove unaligned AIO warning printk
  ext4: fix an incorrect comment about i_mutex
  ext4: fix deadlock in journal_unmap_buffer()
  ext4: split off ext4_journalled_invalidatepage()
  jbd2: fix assertion failure in jbd2_journal_flush()
  ext4: check dioread_nolock on remount
  ext4: fix extent tree corruption caused by hole punch

11 years agomempolicy: remove arg from mpol_parse_str, mpol_to_str
Hugh Dickins [Wed, 2 Jan 2013 10:04:23 +0000 (02:04 -0800)]
mempolicy: remove arg from mpol_parse_str, mpol_to_str

Remove the unused argument (formerly no_context) from mpol_parse_str()
and from mpol_to_str().

Signed-off-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agotmpfs mempolicy: fix /proc/mounts corrupting memory
Hugh Dickins [Wed, 2 Jan 2013 10:01:33 +0000 (02:01 -0800)]
tmpfs mempolicy: fix /proc/mounts corrupting memory

Recently I suggested using "mount -o remount,mpol=local /tmp" in NUMA
mempolicy testing.  Very nasty.  Reading /proc/mounts, /proc/pid/mounts
or /proc/pid/mountinfo may then corrupt one bit of kernel memory, often
in a page table (causing "Bad swap" or "Bad page map" warning or "Bad
pagetable" oops), sometimes in a vm_area_struct or rbnode or somewhere
worse.  "mpol=prefer" and "mpol=prefer:Node" are equally toxic.

Recent NUMA enhancements are not to blame: this dates back to 2.6.35,
when commit e17f74af351c "mempolicy: don't call mpol_set_nodemask() when
no_context" skipped mpol_parse_str()'s call to mpol_set_nodemask(),
which used to initialize v.preferred_node, or set MPOL_F_LOCAL in flags.
With slab poisoning, you can then rely on mpol_to_str() to set the bit
for node 0x6b6b, probably in the next page above the caller's stack.

mpol_parse_str() is only called from shmem_parse_options(): no_context
is always true, so call it unused for now, and remove !no_context code.
Set v.nodes or v.preferred_node or MPOL_F_LOCAL as mpol_to_str() might
expect.  Then mpol_to_str() can ignore its no_context argument also,
the mpol being appropriately initialized whether contextualized or not.
Rename its no_context unused too, and let subsequent patch remove them
(that's not needed for stable backporting, which would involve rejects).

I don't understand why MPOL_LOCAL is described as a pseudo-policy:
it's a reasonable policy which suffers from a confusing implementation
in terms of MPOL_PREFERRED with MPOL_F_LOCAL.  I believe this would be
much more robust if MPOL_LOCAL were recognized in switch statements
throughout, MPOL_F_LOCAL deleted, and MPOL_PREFERRED use the (possibly
empty) nodes mask like everyone else, instead of its preferred_node
variant (I presume an optimization from the days before MPOL_LOCAL).
But that would take me too long to get right and fully tested.

Signed-off-by: Hugh Dickins <hughd@google.com>
Cc: stable@vger.kernel.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoepoll: prevent missed events on EPOLL_CTL_MOD
Eric Wong [Tue, 1 Jan 2013 21:20:27 +0000 (21:20 +0000)]
epoll: prevent missed events on EPOLL_CTL_MOD

EPOLL_CTL_MOD sets the interest mask before calling f_op->poll() to
ensure events are not missed.  Since the modifications to the interest
mask are not protected by the same lock as ep_poll_callback, we need to
ensure the change is visible to other CPUs calling ep_poll_callback.

We also need to ensure f_op->poll() has an up-to-date view of past
events which occured before we modified the interest mask.  So this
barrier also pairs with the barrier in wq_has_sleeper().

This should guarantee either ep_poll_callback or f_op->poll() (or both)
will notice the readiness of a recently-ready/modified item.

This issue was encountered by Andreas Voellmy and Junchang(Jason) Wang in:
http://thread.gmane.org/gmane.linux.kernel/1408782/

Signed-off-by: Eric Wong <normalperson@yhbt.net>
Cc: Hans Verkuil <hans.verkuil@cisco.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Davide Libenzi <davidel@xmailserver.org>
Cc: Hans de Goede <hdegoede@redhat.com>
Cc: Mauro Carvalho Chehab <mchehab@infradead.org>
Cc: David Miller <davem@davemloft.net>
Cc: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Andreas Voellmy <andreas.voellmy@yale.edu>
Tested-by: "Junchang(Jason) Wang" <junchang.wang@yale.edu>
Cc: netdev@vger.kernel.org
Cc: linux-fsdevel@vger.kernel.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agowatchdog: twl4030_wdt: add DT support
Aaro Koskinen [Sun, 23 Dec 2012 20:03:37 +0000 (22:03 +0200)]
watchdog: twl4030_wdt: add DT support

Add DT support for twl4030_wdt. This is needed to get twl4030_wdt to
probe when booting with DT.

Signed-off-by: Aaro Koskinen <aaro.koskinen@iki.fi>
Signed-off-by: Wim Van Sebroeck <wim@iguana.be>
11 years agowatchdog: omap_wdt: eliminate unused variable and a compiler warning
Aaro Koskinen [Sun, 23 Dec 2012 20:03:36 +0000 (22:03 +0200)]
watchdog: omap_wdt: eliminate unused variable and a compiler warning

We forgot to delete this in the commit 4f4753d9 (watchdog: omap_wdt:
convert to devm_ functions), and as a result the following compilation
warning was introduced:

drivers/watchdog/omap_wdt.c: In function 'omap_wdt_remove':
drivers/watchdog/omap_wdt.c:299:19: warning: unused variable 'res' [-Wunused-variable]

Signed-off-by: Aaro Koskinen <aaro.koskinen@iki.fi>
Reviewed-by: Paul Walmsley <paul@pwsan.com>
Signed-off-by: Wim Van Sebroeck <wim@iguana.be>
11 years agowatchdog: da9055: Don't update wdt_dev->timeout in da9055_wdt_set_timeout error path
Axel Lin [Sat, 22 Dec 2012 03:07:01 +0000 (11:07 +0800)]
watchdog: da9055: Don't update wdt_dev->timeout in da9055_wdt_set_timeout error path

Otherwise, WDIOC_GETTIMEOUT returns wrong value if set_timeout fails.
This patch also removes unnecessary ret variable in da9055_wdt_ping function.

Signed-off-by: Axel Lin <axel.lin@ingics.com>
Signed-off-by: Wim Van Sebroeck <wim@iguana.be>
11 years agowatchdog: da9055: Fix invalid free of devm_ allocated data
Axel Lin [Fri, 21 Dec 2012 13:09:06 +0000 (21:09 +0800)]
watchdog: da9055: Fix invalid free of devm_ allocated data

It is not required to free devm_ allocated data. Since kref_put
needs a valid release function, da9055_wdt_release_resources()
is not deleted.

Signed-off-by: Axel Lin <axel.lin@ingics.com>
Signed-off-by: Wim Van Sebroeck <wim@iguana.be>
11 years agoMerge branch 'drm-next' of git://people.freedesktop.org/~airlied/linux
Linus Torvalds [Sun, 30 Dec 2012 18:00:37 +0000 (10:00 -0800)]
Merge branch 'drm-next' of git://people.freedesktop.org/~airlied/linux

Pull DRM update from Dave Airlie:
 "This is a bit larger due to me not bothering to do anything since
  before Xmas, and other people working too hard after I had clearly
  given up.

  It's got the 3 main x86 driver fixes pulls, and a bunch of tegra
  fixes, doesn't fix the Ironlake bug yet, but that does seem to be
  getting closer.

   - radeon: gpu reset fixes and userspace packet support
   - i915: watermark fixes, workarounds, i830/845 fix,
   - nouveau: nvd9/kepler microcode fixes, accel is now enabled and
     working, gk106 support
   - tegra: misc fixes."

* 'drm-next' of git://people.freedesktop.org/~airlied/linux: (34 commits)
  Revert "drm: tegra: protect DC register access with mutex"
  drm: tegra: program only one window during modeset
  drm: tegra: clean out old gem prototypes
  drm: tegra: remove redundant tegra2_tmds_config entry
  drm: tegra: protect DC register access with mutex
  drm: tegra: don't leave clients host1x member uninitialized
  drm: tegra: fix front_porch <-> back_porch mixup
  drm/nve0/graph: fix fuc, and enable acceleration on all known chipsets
  drm/nvc0/graph: fix fuc, and enable acceleration on GF119
  drm/nouveau/bios: cache ramcfg strap on later chipsets
  drm/nouveau/mxm: silence output if no bios data
  drm/nouveau/bios: parse/display extra version component
  drm/nouveau/bios: implement opcode 0xa9
  drm/nouveau/bios: update gpio parsing apis to match current design
  drm/nouveau: initial support for GK106
  drm/radeon: add WAIT_UNTIL to evergreen VM safe reg list
  drm/i915: disable shrinker lock stealing for create_mmap_offset
  drm/i915: optionally disable shrinker lock stealing
  drm/i915: fix flags in dma buf exporting
  drm/radeon: add support for MEM_WRITE packet
  ...

11 years agoMerge tag 'omap-late-cleanups' of git://git.kernel.org/pub/scm/linux/kernel/git/arm...
Linus Torvalds [Sun, 30 Dec 2012 17:59:21 +0000 (09:59 -0800)]
Merge tag 'omap-late-cleanups' of git://git./linux/kernel/git/arm/arm-soc

Pull late ARM cleanups for omap from Olof Johansson:
 "From Tony Lindgren:

  Here are few more patches to finish the omap changes for multiplatform
  conversion that are not strictly fixes, but were too complex to do
  with the dependencies during the merge window.  Those are to move of
  serial-omap.h to platform_data, and the removal of remaining
  cpu_is_omap macro usage outside mach-omap2.

  Then there are several trivial fixes for typos and few minimal
  omap2plus_defconfig updates."

* tag 'omap-late-cleanups' of git://git.kernel.org/pub/scm/linux/kernel/git/arm/arm-soc:
  arch/arm/mach-omap2/dpll3xxx.c: drop if around WARN_ON
  OMAP2: Fix a typo - replace regist with register.
  ARM/omap: use module_platform_driver macro
  ARM: OMAP2+: PMU: Remove unused header
  ARM: OMAP4: remove duplicated include from omap_hwmod_44xx_data.c
  ARM: OMAP2+: omap2plus_defconfig: enable twl4030 SoC audio
  ARM: OMAP2+: omap2plus_defconfig: Add tps65217 support
  ARM: OMAP2+: enable devtmpfs and devtmpfs automount
  ARM: OMAP2+: omap_twl: Change TWL4030_MODULE_PM_RECEIVER to TWL_MODULE_PM_RECEIVER
  ARM: OMAP2+: Drop plat/cpu.h for omap2plus
  ARM: OMAP: Split fb.c to remove last remaining cpu_is_omap usage
  MAINTAINERS: Add an entry for omap related .dts files

11 years agoMerge tag 'fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/arm/arm-soc
Linus Torvalds [Sun, 30 Dec 2012 17:58:36 +0000 (09:58 -0800)]
Merge tag 'fixes' of git://git./linux/kernel/git/arm/arm-soc

Pull ARM SoC fixes from Olof Johansson:
 "It's been quiet over the holidays, but we have had a couple of trivial
  fixes coming in for the newly introduced sunxi platform; one to add it
  to the multiplatform defconfig for build coverage, and one fixup for
  device tree strings."

* tag 'fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/arm/arm-soc:
  sunxi: Change the machine compatible string.
  ARM: multi_v7_defconfig: Add ARCH_SUNXI

11 years agoRevert "drm: tegra: protect DC register access with mutex"
Dave Airlie [Sun, 30 Dec 2012 11:58:20 +0000 (21:58 +1000)]
Revert "drm: tegra: protect DC register access with mutex"

This reverts commit 83c0bcb694be31dcd6c04bdd935b96a95a0af548.

Lucas pointed out this was a mistake, and I missed the discussion,
so just revert it out to save a rebase.

Signed-off-by: Dave Airlie <airlied@redhat.com>
11 years agodrm: tegra: program only one window during modeset
Lucas Stach [Wed, 19 Dec 2012 21:38:57 +0000 (21:38 +0000)]
drm: tegra: program only one window during modeset

The intention is to program exactly WIN_A, not WIN_A and possibly
others.

Signed-off-by: Lucas Stach <dev@lynxeye.de>
Signed-off-by: Dave Airlie <airlied@redhat.com>
11 years agodrm: tegra: clean out old gem prototypes
Lucas Stach [Wed, 19 Dec 2012 21:38:56 +0000 (21:38 +0000)]
drm: tegra: clean out old gem prototypes

There is no gem.c anymore, those functions are implemented by the
drm_cma_helpers now.

Signed-off-by: Lucas Stach <dev@lynxeye.de>
Signed-off-by: Dave Airlie <airlied@redhat.com>
11 years agodrm: tegra: remove redundant tegra2_tmds_config entry
Lucas Stach [Wed, 19 Dec 2012 21:38:55 +0000 (21:38 +0000)]
drm: tegra: remove redundant tegra2_tmds_config entry

The 720p and 1080p entries are completely redundant, as we are matching
the table entries against <=pclk.
Also generalize the comment, as we are using those table entries even
when driving other modes than the standard TV ones.

Signed-off-by: Lucas Stach <dev@lynxeye.de>
Signed-off-by: Dave Airlie <airlied@redhat.com>
11 years agodrm: tegra: protect DC register access with mutex
Lucas Stach [Wed, 19 Dec 2012 21:38:54 +0000 (21:38 +0000)]
drm: tegra: protect DC register access with mutex

Window properties are programmed through a shared aperture and have to
happen atomically. Also we do the read-update-write dance on some of the
shared regs.
To make sure that different functions don't stumble over each other
protect the register access with a mutex.

Signed-off-by: Lucas Stach <dev@lynxeye.de>
Signed-off-by: Dave Airlie <airlied@redhat.com>
11 years agodrm: tegra: don't leave clients host1x member uninitialized
Lucas Stach [Wed, 19 Dec 2012 21:38:53 +0000 (21:38 +0000)]
drm: tegra: don't leave clients host1x member uninitialized

No real problem for now, as nothing is using this, but leaving it
unitialized is asking for trouble later on.

Signed-off-by: Lucas Stach <dev@lynxeye.de>
Signed-off-by: Dave Airlie <airlied@redhat.com>
11 years agodrm: tegra: fix front_porch <-> back_porch mixup
Lucas Stach [Wed, 19 Dec 2012 21:38:52 +0000 (21:38 +0000)]
drm: tegra: fix front_porch <-> back_porch mixup

Fixes wrong picture offset observed when using HDMI output with a
Technisat HD TV.

Signed-off-by: Lucas Stach <dev@lynxeye.de>
Acked-by: Mark Zhang <markz@nvidia.com>
Tested-by: Mark Zhang <markz@nvidia.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
11 years agoMerge branch 'drm-intel-fixes' of git://people.freedesktop.org/~danvet/drm-intel...
Dave Airlie [Sun, 30 Dec 2012 03:54:12 +0000 (13:54 +1000)]
Merge branch 'drm-intel-fixes' of git://people.freedesktop.org/~danvet/drm-intel into drm-next

Some fixes for 3.8:
- Watermark fixups from Chris Wilson (4 pieces).
- 2 snb workarounds, seem to be recently added to our internal DB.
- workaround for the infamous i830/i845 hang, seems now finally solid!
  Based on Chris' fix for SNA, now also for UXA/mesa&old SNA.
- Some more fixlets for shrinker-pulls-the-rug issues (Chris&me).
- Fix dma-buf flags when exporting (you).
- Disable the VGA plane if it's enabled on lid open - similar fix in
  spirit to the one I've sent you last weeek, BIOS' really like to mess
  with the display when closing the lid (awesome debug work from Krzysztof
  Mazur).

* 'drm-intel-fixes' of git://people.freedesktop.org/~danvet/drm-intel:
  drm/i915: disable shrinker lock stealing for create_mmap_offset
  drm/i915: optionally disable shrinker lock stealing
  drm/i915: fix flags in dma buf exporting
  i915: ensure that VGA plane is disabled
  drm/i915: Preallocate the drm_mm_node prior to manipulating the GTT drm_mm manager
  drm: Export routines for inserting preallocated nodes into the mm manager
  drm/i915: don't disable disconnected outputs
  drm/i915: Implement workaround for broken CS tlb on i830/845
  drm/i915: Implement WaSetupGtModeTdRowDispatch
  drm/i915: Implement WaDisableHiZPlanesWhenMSAAEnabled
  drm/i915: Prefer CRTC 'active' rather than 'enabled' during WM computations
  drm/i915: Clear self-refresh watermarks when disabled
  drm/i915: Double the cursor self-refresh latency on Valleyview
  drm/i915: Fixup cursor latency used for IVB lp3 watermarks

11 years agoMerge branch 'drm-fixes-3.8' of git://people.freedesktop.org/~agd5f/linux into drm...
Dave Airlie [Sun, 30 Dec 2012 03:02:48 +0000 (13:02 +1000)]
Merge branch 'drm-fixes-3.8' of git://people.freedesktop.org/~agd5f/linux into drm-next

Misc fixes for reset and new packets for userspace usage.

* 'drm-fixes-3.8' of git://people.freedesktop.org/~agd5f/linux:
  drm/radeon: add WAIT_UNTIL to evergreen VM safe reg list
  drm/radeon: add support for MEM_WRITE packet
  drm/radeon: restore modeset late in GPU reset path
  drm/radeon: avoid deadlock in pm path when waiting for fence
  drm/radeon: don't leave fence blocked process on failed GPU reset

11 years agoMerge branch 'drm-nouveau-fixes-3.8' of git://anongit.freedesktop.org/git/nouveau...
Dave Airlie [Sun, 30 Dec 2012 03:01:52 +0000 (13:01 +1000)]
Merge branch 'drm-nouveau-fixes-3.8' of git://anongit.freedesktop.org/git/nouveau/linux-2.6 into drm-next

Fixes the accel support for nvd9 + kepler chipsets, also fixes GK106 support.

* 'drm-nouveau-fixes-3.8' of git://anongit.freedesktop.org/git/nouveau/linux-2.6:
  drm/nve0/graph: fix fuc, and enable acceleration on all known chipsets
  drm/nvc0/graph: fix fuc, and enable acceleration on GF119
  drm/nouveau/bios: cache ramcfg strap on later chipsets
  drm/nouveau/mxm: silence output if no bios data
  drm/nouveau/bios: parse/display extra version component
  drm/nouveau/bios: implement opcode 0xa9
  drm/nouveau/bios: update gpio parsing apis to match current design
  drm/nouveau: initial support for GK106

11 years agomm: fix null pointer dereference in wait_iff_congested()
Zlatko Calusic [Fri, 28 Dec 2012 02:16:38 +0000 (03:16 +0100)]
mm: fix null pointer dereference in wait_iff_congested()

An unintended consequence of commit 4ae0a48b5efc ("mm: modify
pgdat_balanced() so that it also handles order-0") is that
wait_iff_congested() can now be called with NULL 'struct zone *'
producing kernel oops like this:

  BUG: unable to handle kernel NULL pointer dereference
  IP: [<ffffffff811542d9>] wait_iff_congested+0x59/0x140

This trivial patch fixes it.

Reported-by: Zhouping Liu <zliu@redhat.com>
Reported-and-tested-by: Sedat Dilek <sedat.dilek@gmail.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Hugh Dickins <hughd@google.com>
Signed-off-by: Zlatko Calusic <zlatko.calusic@iskon.hr>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoMerge tag 'sunxi-fixes-for-3.8-rc2' of git://github.com/mripard/linux into fixes
Olof Johansson [Fri, 28 Dec 2012 07:53:01 +0000 (08:53 +0100)]
Merge tag 'sunxi-fixes-for-3.8-rc2' of git://github.com/mripard/linux into fixes

From Maxime Ripard:
Fixes for the sunxi core to be merged in 3.8-rc2

* tag 'sunxi-fixes-for-3.8-rc2' of git://github.com/mripard/linux:
  sunxi: Change the machine compatible string.
  ARM: multi_v7_defconfig: Add ARCH_SUNXI

11 years agolibceph: fix protocol feature mismatch failure path
Sage Weil [Fri, 28 Dec 2012 02:27:04 +0000 (20:27 -0600)]
libceph: fix protocol feature mismatch failure path

We should not set con->state to CLOSED here; that happens in
ceph_fault() in the caller, where it first asserts that the state
is not yet CLOSED.  Avoids a BUG when the features don't match.

Since the fail_protocol() has become a trivial wrapper, replace
calls to it with direct calls to reset_connection().

Signed-off-by: Sage Weil <sage@inktank.com>
Reviewed-by: Alex Elder <elder@inktank.com>
11 years agolibceph: WARN, don't BUG on unexpected connection states
Alex Elder [Wed, 26 Dec 2012 16:43:57 +0000 (10:43 -0600)]
libceph: WARN, don't BUG on unexpected connection states

A number of assertions in the ceph messenger are implemented with
BUG_ON(), killing the system if connection's state doesn't match
what's expected.  At this point our state model is (evidently) not
well understood enough for these assertions to trigger a BUG().
Convert all BUG_ON(con->state...) calls to be WARN_ON(con->state...)
so we learn about these issues without killing the machine.

We now recognize that a connection fault can occur due to a socket
closure at any time, regardless of the state of the connection.  So
there is really nothing we can assert about the state of the
connection at that point so eliminate that assertion.

Reported-by: Ugis <ugis22@gmail.com>
Tested-by: Ugis <ugis22@gmail.com>
Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
11 years agolibceph: always reset osds when kicking
Alex Elder [Wed, 26 Dec 2012 20:31:40 +0000 (14:31 -0600)]
libceph: always reset osds when kicking

When ceph_osdc_handle_map() is called to process a new osd map,
kick_requests() is called to ensure all affected requests are
updated if necessary to reflect changes in the osd map.  This
happens in two cases:  whenever an incremental map update is
processed; and when a full map update (or the last one if there is
more than one) gets processed.

In the former case, the kick_requests() call is followed immediately
by a call to reset_changed_osds() to ensure any connections to osds
affected by the map change are reset.  But for full map updates
this isn't done.

Both cases should be doing this osd reset.

Rather than duplicating the reset_changed_osds() call, move it into
the end of kick_requests().

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
11 years agolibceph: move linger requests sooner in kick_requests()
Alex Elder [Wed, 19 Dec 2012 21:52:36 +0000 (15:52 -0600)]
libceph: move linger requests sooner in kick_requests()

The kick_requests() function is called by ceph_osdc_handle_map()
when an osd map change has been indicated.  Its purpose is to
re-queue any request whose target osd is different from what it
was when it was originally sent.

It is structured as two loops, one for incomplete but registered
requests, and a second for handling completed linger requests.
As a special case, in the first loop if a request marked to linger
has not yet completed, it is moved from the request list to the
linger list.  This is as a quick and dirty way to have the second
loop handle sending the request along with all the other linger
requests.

Because of the way it's done now, however, this quick and dirty
solution can result in these incomplete linger requests never
getting re-sent as desired.  The problem lies in the fact that
the second loop only arranges for a linger request to be sent
if it appears its target osd has changed.  This is the proper
handling for *completed* linger requests (it avoids issuing
the same linger request twice to the same osd).

But although the linger requests added to the list in the first loop
may have been sent, they have not yet completed, so they need to be
re-sent regardless of whether their target osd has changed.

The first required fix is we need to avoid calling __map_request()
on any incomplete linger request.  Otherwise the subsequent
__map_request() call in the second loop will find the target osd
has not changed and will therefore not re-send the request.

Second, we need to be sure that a sent but incomplete linger request
gets re-sent.  If the target osd is the same with the new osd map as
it was when the request was originally sent, this won't happen.
This can be fixed through careful handling when we move these
requests from the request list to the linger list, by unregistering
the request *before* it is registered as a linger request.  This
works because a side-effect of unregistering the request is to make
the request's r_osd pointer be NULL, and *that* will ensure the
second loop actually re-sends the linger request.

Processing of such a request is done at that point, so continue with
the next one once it's been moved.

Signed-off-by: Alex Elder <elder@inktank.com>
Reviewed-by: Sage Weil <sage@inktank.com>
11 years agoMerge tag 'hwmon-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/groeck...
Linus Torvalds [Thu, 27 Dec 2012 18:46:47 +0000 (10:46 -0800)]
Merge tag 'hwmon-for-linus' of git://git./linux/kernel/git/groeck/linux-staging

Pull hwmon fixes from Guenter Roeck:

 - Report i2c errors to userspace in lm73 driver

 - Fix problem with DIV_ROUND_CLOSEST and unsigned divisors in emc6w201
   driver

* tag 'hwmon-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/groeck/linux-staging:
  hwmon: (emc6w201) Fix DIV_ROUND_CLOSEST problem with unsigned divisors
  hwmon: (lm73} Detect and report i2c bus errors

11 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ebiederm...
Linus Torvalds [Thu, 27 Dec 2012 18:42:46 +0000 (10:42 -0800)]
Merge branch 'for-linus' of git://git./linux/kernel/git/ebiederm/user-namespace

Pull namespace fixes from Eric Biederman:
 "This tree includes two bug fixes for problems Oleg spotted on his
  review of the recent pid namespace work.  A small fix to not enable
  bottom halves with irqs disabled, and a trivial build fix for f2fs
  with user namespaces enabled."

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ebiederm/user-namespace:
  f2fs: Don't assign e_id in f2fs_acl_from_disk
  proc: Allow proc_free_inum to be called from any context
  pidns: Stop pid allocation when init dies
  pidns: Outlaw thread creation after unshare(CLONE_NEWPID)

11 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net
Linus Torvalds [Thu, 27 Dec 2012 18:40:30 +0000 (10:40 -0800)]
Merge git://git./linux/kernel/git/davem/net

Pull networking fixes from David Miller:

1) GRE tunnel drivers don't set the transport header properly, they also
   blindly deref the inner protocol ipv4 and needs some checks.  Fixes
   from Isaku Yamahata.

2) Fix sleeps while atomic in netdevice rename code, from Eric Dumazet.

3) Fix double-spinlock in solos-pci driver, from Dan Carpenter.

4) More ARP bug fixes.  Fix lockdep splat in arp_solicit() and then the
   bug accidentally added by that fix.  From Eric Dumazet and Cong Wang.

5) Remove some __dev* annotations that slipped back in, as well as all
   HOTPLUG references.  From Greg KH

6) RDS protocol uses wrong interfaces to access scatter-gather elements,
   causing a regression.  From Mike Marciniszyn.

7) Fix build error in cpts driver, from Richard Cochran.

8) Fix arithmetic in packet scheduler, from Stefan Hasko.

9) Similarly, fix association during calculation of random backoff in
   batman-adv.  From Akinobu Mita.

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net: (21 commits)
  ipv6/ip6_gre: set transport header correctly
  ipv4/ip_gre: set transport header correctly to gre header
  IB/rds: suppress incompatible protocol when version is known
  IB/rds: Correct ib_api use with gs_dma_address/sg_dma_len
  net/vxlan: Use the underlying device index when joining/leaving multicast groups
  tcp: should drop incoming frames without ACK flag set
  netprio_cgroup: define sk_cgrp_prioidx only if NETPRIO_CGROUP is enabled
  cpts: fix a run time warn_on.
  cpts: fix build error by removing useless code.
  batman-adv: fix random jitter calculation
  arp: fix a regression in arp_solicit()
  net: sched: integer overflow fix
  CONFIG_HOTPLUG removal from networking core
  Drivers: network: more __dev* removal
  bridge: call br_netpoll_disable in br_add_if
  ipv4: arp: fix a lockdep splat in arp_solicit()
  tuntap: dont use a private kmem_cache
  net: devnet_rename_seq should be a seqcount
  ip_gre: fix possible use after free
  ip_gre: make ipgre_tunnel_xmit() not parse network header as IP unconditionally
  ...

11 years agoext4: avoid hang when mounting non-journal filesystems with orphan list
Theodore Ts'o [Thu, 27 Dec 2012 06:42:50 +0000 (01:42 -0500)]
ext4: avoid hang when mounting non-journal filesystems with orphan list

When trying to mount a file system which does not contain a journal,
but which does have a orphan list containing an inode which needs to
be truncated, the mount call with hang forever in
ext4_orphan_cleanup() because ext4_orphan_del() will return
immediately without removing the inode from the orphan list, leading
to an uninterruptible loop in kernel code which will busy out one of
the CPU's on the system.

This can be trivially reproduced by trying to mount the file system
found in tests/f_orphan_extents_inode/image.gz from the e2fsprogs
source tree.  If a malicious user were to put this on a USB stick, and
mount it on a Linux desktop which has automatic mounts enabled, this
could be considered a potential denial of service attack.  (Not a big
deal in practice, but professional paranoids worry about such things,
and have even been known to allocate CVE numbers for such problems.)

Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Reviewed-by: Zheng Liu <wenqing.lz@taobao.com>
Cc: stable@vger.kernel.org
11 years agoext4: lock i_mutex when truncating orphan inodes
Theodore Ts'o [Thu, 27 Dec 2012 06:42:48 +0000 (01:42 -0500)]
ext4: lock i_mutex when truncating orphan inodes

Commit c278531d39 added a warning when ext4_flush_unwritten_io() is
called without i_mutex being taken.  It had previously not been taken
during orphan cleanup since races weren't possible at that point in
the mount process, but as a result of this c278531d39, we will now see
a kernel WARN_ON in this case.  Take the i_mutex in
ext4_orphan_cleanup() to suppress this warning.

Reported-by: Alexander Beregalov <a.beregalov@gmail.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Reviewed-by: Zheng Liu <wenqing.lz@taobao.com>
Cc: stable@vger.kernel.org
11 years agoipv6/ip6_gre: set transport header correctly
Isaku Yamahata [Mon, 24 Dec 2012 16:51:04 +0000 (16:51 +0000)]
ipv6/ip6_gre: set transport header correctly

ip6gre_xmit2() incorrectly sets transport header to inner payload
instead of GRE header. It seems copy-and-pasted from ipip.c.
Set transport header to gre header.
(In ipip case the transport header is the inner ip header, so that's
correct.)

Found by inspection. In practice the incorrect transport header
doesn't matter because the skb usually is sent to another net_device
or socket, so the transport header isn't referenced.

Signed-off-by: Isaku Yamahata <yamahata@valinux.co.jp>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoipv4/ip_gre: set transport header correctly to gre header
Isaku Yamahata [Mon, 24 Dec 2012 16:51:03 +0000 (16:51 +0000)]
ipv4/ip_gre: set transport header correctly to gre header

ipgre_tunnel_xmit() incorrectly sets transport header to inner payload
instead of GRE header. It seems copy-and-pasted from ipip.c.
So set transport header to gre header.
(In ipip case the transport header is the inner ip header, so that's
correct.)

Found by inspection. In practice the incorrect transport header
doesn't matter because the skb usually is sent to another net_device
or socket, so the transport header isn't referenced.

Signed-off-by: Isaku Yamahata <yamahata@valinux.co.jp>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoIB/rds: suppress incompatible protocol when version is known
Marciniszyn, Mike [Fri, 21 Dec 2012 08:01:54 +0000 (08:01 +0000)]
IB/rds: suppress incompatible protocol when version is known

Add an else to only print the incompatible protocol message
when version hasn't been established.

Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoIB/rds: Correct ib_api use with gs_dma_address/sg_dma_len
Marciniszyn, Mike [Fri, 21 Dec 2012 08:01:49 +0000 (08:01 +0000)]
IB/rds: Correct ib_api use with gs_dma_address/sg_dma_len

0b088e00 ("RDS: Use page_remainder_alloc() for recv bufs")
added uses of sg_dma_len() and sg_dma_address(). This makes
RDS DOA with the qib driver.

IB ulps should use ib_sg_dma_len() and ib_sg_dma_address
respectively since some HCAs overload ib_sg_dma* operations.

Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agonet/vxlan: Use the underlying device index when joining/leaving multicast groups
Yan Burman [Thu, 20 Dec 2012 03:36:08 +0000 (03:36 +0000)]
net/vxlan: Use the underlying device index when joining/leaving multicast groups

The socket calls from vxlan to join/leave multicast group aren't
using the index of the underlying device, as a result the stack uses
the first interface that is up. This results in vxlan being non functional
over a device which isn't the 1st to be up.
Fix this by providing the iflink field to the vxlan instance
to the multicast calls.

Signed-off-by: Yan Burman <yanb@mellanox.com>
Acked-by: Stephen Hemminger <shemminger@vyatta.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agotcp: should drop incoming frames without ACK flag set
Eric Dumazet [Wed, 26 Dec 2012 12:44:34 +0000 (12:44 +0000)]
tcp: should drop incoming frames without ACK flag set

In commit 96e0bf4b5193d (tcp: Discard segments that ack data not yet
sent) John Dykstra enforced a check against ack sequences.

In commit 354e4aa391ed5 (tcp: RFC 5961 5.2 Blind Data Injection Attack
Mitigation) I added more safety tests.

But we missed fact that these tests are not performed if ACK bit is
not set.

RFC 793 3.9 mandates TCP should drop a frame without ACK flag set.

" fifth check the ACK field,
      if the ACK bit is off drop the segment and return"

Not doing so permits an attacker to only guess an acceptable sequence
number, evading stronger checks.

Many thanks to Zhiyun Qian for bringing this issue to our attention.

See :
http://web.eecs.umich.edu/~zhiyunq/pub/ccs12_TCP_sequence_number_inference.pdf

Reported-by: Zhiyun Qian <zhiyunq@umich.edu>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Nandita Dukkipati <nanditad@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Cc: John Dykstra <john.dykstra1@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agomm: Fix PageHead when !CONFIG_PAGEFLAGS_EXTENDED
Christoffer Dall [Fri, 21 Dec 2012 18:03:50 +0000 (13:03 -0500)]
mm: Fix PageHead when !CONFIG_PAGEFLAGS_EXTENDED

Unfortunately with !CONFIG_PAGEFLAGS_EXTENDED, (!PageHead) is false, and
(PageHead) is true, for tail pages.  If this is indeed the intended
behavior, which I doubt because it breaks cache cleaning on some ARM
systems, then the nomenclature is highly problematic.

This patch makes sure PageHead is only true for head pages and PageTail
is only true for tail pages, and neither is true for non-compound pages.

[ This buglet seems ancient - seems to have been introduced back in Apr
  2008 in commit 6a1e7f777f61: "pageflags: convert to the use of new
  macros".  And the reason nobody noticed is because the PageHead()
  tests are almost all about just sanity-checking, and only used on
  pages that are actual page heads.  The fact that the old code returned
  true for tail pages too was thus not really noticeable.   - Linus ]

Signed-off-by: Christoffer Dall <cdall@cs.columbia.edu>
Acked-by: Andrea Arcangeli <aarcange@redhat.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Will Deacon <Will.Deacon@arm.com>
Cc: Steve Capper <Steve.Capper@arm.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: stable@kernel.org # 2.6.26+
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agonetprio_cgroup: define sk_cgrp_prioidx only if NETPRIO_CGROUP is enabled
Li Zefan [Tue, 25 Dec 2012 20:48:24 +0000 (20:48 +0000)]
netprio_cgroup: define sk_cgrp_prioidx only if NETPRIO_CGROUP is enabled

sock->sk_cgrp_prioidx won't be used at all if CONFIG_NETPRIO_CGROUP=n.

Signed-off-by: Li Zefan <lizefan@huawei.com>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agocpts: fix a run time warn_on.
Richard Cochran [Sun, 23 Dec 2012 21:19:10 +0000 (21:19 +0000)]
cpts: fix a run time warn_on.

This patch fixes a warning in clk_enable by calling clk_prepare_enable
instead.

Signed-off-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agocpts: fix build error by removing useless code.
Richard Cochran [Sun, 23 Dec 2012 21:19:09 +0000 (21:19 +0000)]
cpts: fix build error by removing useless code.

The cpts driver tries to obtain the input clock frequency by calling the
clock's internal 'recalc' method. Since <plat/clock.h> has been removed,
this code can no longer compile.

However, the driver never makes use of the frequency value, so this patch
fixes the issue by removing the offending code altogether.

Signed-off-by: Richard Cochran <richardcochran@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agobatman-adv: fix random jitter calculation
Akinobu Mita [Wed, 26 Dec 2012 02:32:10 +0000 (02:32 +0000)]
batman-adv: fix random jitter calculation

batadv_iv_ogm_emit_send_time() attempts to calculates a random integer
in the range of 'orig_interval +- BATADV_JITTER' by the below lines.

        msecs = atomic_read(&bat_priv->orig_interval) - BATADV_JITTER;
        msecs += (random32() % 2 * BATADV_JITTER);

But it actually gets 'orig_interval' or 'orig_interval - BATADV_JITTER'
because '%' and '*' have same precedence and associativity is
left-to-right.

This adds the parentheses at the appropriate position so that it matches
original intension.

Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Acked-by: Antonio Quartulli <ordex@autistici.org>
Cc: Marek Lindner <lindner_marek@yahoo.de>
Cc: Simon Wunderlich <siwu@hrz.tu-chemnitz.de>
Cc: Antonio Quartulli <ordex@autistici.org>
Cc: b.a.t.m.a.n@lists.open-mesh.org
Cc: "David S. Miller" <davem@davemloft.net>
Cc: netdev@vger.kernel.org
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoPCI: Reduce Ricoh 0xe822 SD card reader base clock frequency to 50MHz
Andy Lutomirski [Sat, 1 Dec 2012 20:37:20 +0000 (12:37 -0800)]
PCI: Reduce Ricoh 0xe822 SD card reader base clock frequency to 50MHz

Otherwise it fails like this on cards like the Transcend 16GB SDHC card:

    mmc0: new SDHC card at address b368
    mmcblk0: mmc0:b368 SDC   15.0 GiB
    mmcblk0: error -110 sending status command, retrying
    mmcblk0: error -84 transferring data, sector 0, nr 8, cmd response 0x900, card status 0xb0

Tested on my Lenovo x200 laptop.

[bhelgaas: changelog]
Signed-off-by: Andy Lutomirski <luto@amacapital.net>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Acked-by: Chris Ball <cjb@laptop.org>
CC: Manoj Iyer <manoj.iyer@canonical.com>
CC: stable@vger.kernel.org
11 years agoPCI/PM: Do not suspend port if any subordinate device needs PME polling
Huang Ying [Wed, 26 Dec 2012 17:39:23 +0000 (10:39 -0700)]
PCI/PM: Do not suspend port if any subordinate device needs PME polling

Ulrich reported that his USB3 cardreader does not work reliably when
connected to the USB3 port.  It turns out that USB3 controller failed to
awaken when plugging in the USB3 cardreader.  Further experiments found
that the USB3 host controller can only be awakened via polling, not via PME
interrupt.  But if the PCIe port to which the USB3 host controller is
connected is suspended, we cannot poll the controller because its config
space is not accessible when the PCIe port is in a low power state.

To solve the issue, the PCIe port will not be suspended if any subordinate
device needs PME polling.

[bhelgaas: use bool consistently rather than mixing int/bool]
Reference: http://lkml.kernel.org/r/50841CCC.9030809@uli-eckhardt.de
Reported-by: Ulrich Eckhardt <usb@uli-eckhardt.de>
Tested-by: Sarah Sharp <sarah.a.sharp@linux.intel.com>
Signed-off-by: Huang Ying <ying.huang@intel.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Acked-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
CC: stable@vger.kernel.org # v3.6+
11 years agoPCI: Add PCIe Link Capability link speed and width names
Bjorn Helgaas [Wed, 26 Dec 2012 17:39:23 +0000 (10:39 -0700)]
PCI: Add PCIe Link Capability link speed and width names

Add standard #defines for the Supported Link Speeds field in the PCIe
Link Capabilities register.

Note that prior to PCIe spec r3.0, these encodings were defined:

    0001b  2.5GT/s Link speed supported
    0010b  5.0GT/s and 2.5GT/s Link speed supported

Starting with spec r3.0, these encodings refer to bits 0 and 1 in the
Supported Link Speeds Vector in the Link Capabilities 2 register, and bits
0 and 1 there mean 2.5 GT/s and 5.0 GT/s, respectively.  Therefore, code
that followed r2.0 and interpreted 0x1 as 2.5GT/s and 0x2 as 5.0GT/s will
continue to work, and we can identify a device using the new encodings
because it will have a non-zero Link Capabilities 2 register.

Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
11 years agoPCI: Work around Stratus ftServer broken PCIe hierarchy (fix DMI check)
Myron Stowe [Wed, 26 Dec 2012 17:39:23 +0000 (10:39 -0700)]
PCI: Work around Stratus ftServer broken PCIe hierarchy (fix DMI check)

Commit 284f5f9 was intended to disable the "only_one_child()" optimization
on Stratus ftServer systems, but its DMI check is wrong.  It looks for
DMI_SYS_VENDOR that contains "ftServer", when it should look for
DMI_SYS_VENDOR containing "Stratus" and DMI_PRODUCT_NAME containing
"ftServer".

Tested on Stratus ftServer 6400.

Reported-by: Fadeeva Marina <astarta@rat.ru>
Reference: https://bugzilla.kernel.org/show_bug.cgi?id=51331
Signed-off-by: Myron Stowe <myron.stowe@redhat.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
CC: stable@vger.kernel.org # v3.5+
11 years agoPCI: Remove spurious error for sriov_numvfs store and simplify flow
Bjorn Helgaas [Wed, 26 Dec 2012 17:39:22 +0000 (10:39 -0700)]
PCI: Remove spurious error for sriov_numvfs store and simplify flow

If we request "num_vfs" and the driver's sriov_configure() method enables
exactly that number ("num_vfs_enabled"), we complain "Invalid value for
number of VFs to enable" and return an error.  We should silently return
success instead.

Also, use kstrtou16() since numVFs is defined to be a 16-bit field and
rework to simplify control flow.

Reported-by: Greg Rose <gregory.v.rose@intel.com>
Reference: http://lkml.kernel.org/r/20121214101911.00002f59@unknown
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Tested-by: Donald Dutile <ddutile@redhat.com>
11 years agof2fs: Don't assign e_id in f2fs_acl_from_disk
Eric W. Biederman [Sat, 22 Dec 2012 09:52:39 +0000 (01:52 -0800)]
f2fs: Don't assign e_id in f2fs_acl_from_disk

With user namespaces enabled building f2fs fails with:

 CC      fs/f2fs/acl.o
fs/f2fs/acl.c: In function ‘f2fs_acl_from_disk’:
fs/f2fs/acl.c:85:21: error: ‘struct posix_acl_entry’ has no member named ‘e_id’
make[2]: *** [fs/f2fs/acl.o] Error 1
make[2]: Target `__build' not remade because of errors.

e_id is a backwards compatibility field only used for file systems
that haven't been converted to use kuids and kgids.  When the posix
acl tag field is neither ACL_USER nor ACL_GROUP assigning e_id is
unnecessary.  Remove the assignment so f2fs will build with user
namespaces enabled.

Cc: Namjae Jeon <namjae.jeon@samsung.com>
Cc: Amit Sahrawat <a.sahrawat@samsung.com>
Acked-by: Jaegeuk Kim <jaegeuk.kim@samsung.com>
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
11 years agoproc: Allow proc_free_inum to be called from any context
Eric W. Biederman [Sat, 22 Dec 2012 04:38:00 +0000 (20:38 -0800)]
proc: Allow proc_free_inum to be called from any context

While testing the pid namespace code I hit this nasty warning.

[  176.262617] ------------[ cut here ]------------
[  176.263388] WARNING: at /home/eric/projects/linux/linux-userns-devel/kernel/softirq.c:160 local_bh_enable_ip+0x7a/0xa0()
[  176.265145] Hardware name: Bochs
[  176.265677] Modules linked in:
[  176.266341] Pid: 742, comm: bash Not tainted 3.7.0userns+ #18
[  176.266564] Call Trace:
[  176.266564]  [<ffffffff810a539f>] warn_slowpath_common+0x7f/0xc0
[  176.266564]  [<ffffffff810a53fa>] warn_slowpath_null+0x1a/0x20
[  176.266564]  [<ffffffff810ad9ea>] local_bh_enable_ip+0x7a/0xa0
[  176.266564]  [<ffffffff819308c9>] _raw_spin_unlock_bh+0x19/0x20
[  176.266564]  [<ffffffff8123dbda>] proc_free_inum+0x3a/0x50
[  176.266564]  [<ffffffff8111d0dc>] free_pid_ns+0x1c/0x80
[  176.266564]  [<ffffffff8111d195>] put_pid_ns+0x35/0x50
[  176.266564]  [<ffffffff810c608a>] put_pid+0x4a/0x60
[  176.266564]  [<ffffffff8146b177>] tty_ioctl+0x717/0xc10
[  176.266564]  [<ffffffff810aa4d5>] ? wait_consider_task+0x855/0xb90
[  176.266564]  [<ffffffff81086bf9>] ? default_spin_lock_flags+0x9/0x10
[  176.266564]  [<ffffffff810cab0a>] ? remove_wait_queue+0x5a/0x70
[  176.266564]  [<ffffffff811e37e8>] do_vfs_ioctl+0x98/0x550
[  176.266564]  [<ffffffff810b8a0f>] ? recalc_sigpending+0x1f/0x60
[  176.266564]  [<ffffffff810b9127>] ? __set_task_blocked+0x37/0x80
[  176.266564]  [<ffffffff810ab95b>] ? sys_wait4+0xab/0xf0
[  176.266564]  [<ffffffff811e3d31>] sys_ioctl+0x91/0xb0
[  176.266564]  [<ffffffff810a95f0>] ? task_stopped_code+0x50/0x50
[  176.266564]  [<ffffffff81939199>] system_call_fastpath+0x16/0x1b
[  176.266564] ---[ end trace 387af88219ad6143 ]---

It turns out that spin_unlock_bh(proc_inum_lock) is not safe when
put_pid is called with another spinlock held and irqs disabled.

For now take the easy path and use spin_lock_irqsave(proc_inum_lock)
in proc_free_inum and spin_loc_irq in proc_alloc_inum(proc_inum_lock).

Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
11 years agopidns: Stop pid allocation when init dies
Eric W. Biederman [Sat, 22 Dec 2012 04:27:12 +0000 (20:27 -0800)]
pidns: Stop pid allocation when init dies

Oleg pointed out that in a pid namespace the sequence.
- pid 1 becomes a zombie
- setns(thepidns), fork,...
- reaping pid 1.
- The injected processes exiting.

Can lead to processes attempting access their child reaper and
instead following a stale pointer.

That waitpid for init can return before all of the processes in
the pid namespace have exited is also unfortunate.

Avoid these problems by disabling the allocation of new pids in a pid
namespace when init dies, instead of when the last process in a pid
namespace is reaped.

Pointed-out-by: Oleg Nesterov <oleg@redhat.com>
Reviewed-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
11 years agoext4: do not try to write superblock on ro remount w/o journal
Michael Tokarev [Tue, 25 Dec 2012 19:08:16 +0000 (14:08 -0500)]
ext4: do not try to write superblock on ro remount w/o journal

When a journal-less ext4 filesystem is mounted on a read-only block
device (blockdev --setro will do), each remount (for other, unrelated,
flags, like suid=>nosuid etc) results in a series of scary messages
from kernel telling about I/O errors on the device.

This is becauese of the following code ext4_remount():

       if (sbi->s_journal == NULL)
                ext4_commit_super(sb, 1);

at the end of remount procedure, which forces writing (flushing) of
a superblock regardless whenever it is dirty or not, if the filesystem
is readonly or not, and whenever the device itself is readonly or not.

We only need call ext4_commit_super when the file system had been
previously mounted read/write.

Thanks to Eric Sandeen for help in diagnosing this issue.

Signed-off-By: Michael Tokarev <mjt@tls.msk.ru>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Cc: stable@vger.kernel.org
11 years agoext4: include journal blocks in df overhead calcs
Eric Sandeen [Tue, 25 Dec 2012 18:56:01 +0000 (13:56 -0500)]
ext4: include journal blocks in df overhead calcs

To more accurately calculate overhead for "bsd" style
df reporting, we should count the journal blocks as
overhead as well.

Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
Tested-by: Eric Whitney <enwlinux@gmail.com>
11 years agoext4: remove unaligned AIO warning printk
Eric Sandeen [Tue, 25 Dec 2012 18:33:13 +0000 (13:33 -0500)]
ext4: remove unaligned AIO warning printk

Although I put this in, I now think it was a bad decision.  For most
users, there is very little to be done in this case.  They get the
message, once per day, with no real context or proposed action.  TBH,
it generates support calls when it probably does not need to; the
message sounds more dire than the situation really is.

Just nuke it.  Normal investigation via blktrace or whatnot can
reveal poor IO patterns if bad performance is encountered.

Signed-off-by: Eric Sandeen <sandeen@redhat.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
11 years agoext4: fix an incorrect comment about i_mutex
Andy Lutomirski [Tue, 25 Dec 2012 18:31:52 +0000 (13:31 -0500)]
ext4: fix an incorrect comment about i_mutex

i_mutex is not held when ->sync_file is called.

Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Andy Lutomirski <luto@amacapital.net>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
11 years agoext4: fix deadlock in journal_unmap_buffer()
Jan Kara [Tue, 25 Dec 2012 18:29:52 +0000 (13:29 -0500)]
ext4: fix deadlock in journal_unmap_buffer()

We cannot wait for transaction commit in journal_unmap_buffer()
because we hold page lock which ranks below transaction start.  We
solve the issue by bailing out of journal_unmap_buffer() and
jbd2_journal_invalidatepage() with -EBUSY.  Caller is then responsible
for waiting for transaction commit to finish and try invalidation
again. Since the issue can happen only for page stradding i_size, it
is simple enough to manually call jbd2_journal_invalidatepage() for
such page from ext4_setattr(), check the return value and wait if
necessary.

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
11 years agoext4: split off ext4_journalled_invalidatepage()
Jan Kara [Tue, 25 Dec 2012 18:28:54 +0000 (13:28 -0500)]
ext4: split off ext4_journalled_invalidatepage()

In data=journal mode we don't need delalloc or DIO handling in invalidatepage
and similarly in other modes we don't need the journal handling. So split
invalidatepage implementations.

Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
11 years agopidns: Outlaw thread creation after unshare(CLONE_NEWPID)
Eric W. Biederman [Fri, 21 Dec 2012 03:26:06 +0000 (19:26 -0800)]
pidns: Outlaw thread creation after unshare(CLONE_NEWPID)

The sequence:
unshare(CLONE_NEWPID)
clone(CLONE_THREAD|CLONE_SIGHAND|CLONE_VM)

Creates a new process in the new pid namespace without setting
pid_ns->child_reaper.  After forking this results in a NULL
pointer dereference.

Avoid this and other nonsense scenarios that can show up after
creating a new pid namespace with unshare by adding a new
check in copy_prodcess.

Pointed-out-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
11 years agoarp: fix a regression in arp_solicit()
Cong Wang [Sun, 23 Dec 2012 15:23:16 +0000 (15:23 +0000)]
arp: fix a regression in arp_solicit()

Sedat reported the following commit caused a regression:

commit 9650388b5c56578fdccc79c57a8c82fb92b8e7f1
Author: Eric Dumazet <edumazet@google.com>
Date:   Fri Dec 21 07:32:10 2012 +0000

    ipv4: arp: fix a lockdep splat in arp_solicit

This is due to the 6th parameter of arp_send() needs to be NULL
for the broadcast case, the above commit changed it to an all-zero
array by mistake.

Reported-by: Sedat Dilek <sedat.dilek@gmail.com>
Tested-by: Sedat Dilek <sedat.dilek@gmail.com>
Cc: Sedat Dilek <sedat.dilek@gmail.com>
Cc: Eric Dumazet <edumazet@google.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Julian Anastasov <ja@ssi.bg>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
11 years agoMerge branch 'i2c-embedded/for-next' of git://git.pengutronix.de/git/wsa/linux
Linus Torvalds [Sun, 23 Dec 2012 17:48:33 +0000 (09:48 -0800)]
Merge branch 'i2c-embedded/for-next' of git://git.pengutronix.de/git/wsa/linux

Pull i2c __dev* attribute removal from Wolfram Sang:
 "The squashed patches from Bill to get rid of the __dev* annotations in
  the i2c subsystem.  I couldn't include it in my previous pull request
  due to some dependency with the mfd subsystem.  I had this patch in
  linux-next for two days before rc1 and nothing popped up."

* 'i2c-embedded/for-next' of git://git.pengutronix.de/git/wsa/linux:
  i2c: remove __dev* attributes from subsystem

11 years agomm: modify pgdat_balanced() so that it also handles order-0
Zlatko Calusic [Sun, 23 Dec 2012 14:12:54 +0000 (15:12 +0100)]
mm: modify pgdat_balanced() so that it also handles order-0

Teach pgdat_balanced() about order-0 allocations so that we can simplify
code in a few places in vmstat.c.

Suggested-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Zlatko Calusic <zlatko.calusic@iskon.hr>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Hugh Dickins <hughd@google.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agoPartly revert "[media] uvcvideo: Set error_idx properly for extended controls API...
Rafael J. Wysocki [Sun, 23 Dec 2012 13:39:32 +0000 (14:39 +0100)]
Partly revert "[media] uvcvideo: Set error_idx properly for extended controls API failures"

Commit f0ed2ce840b3 ("[media] uvcvideo: Set error_idx properly for
extended controls API failures") causes user space to behave incorrectly
on one of my test machines (there is no sound under KDE 4.9.4 using
pulseaudio and there is a knotify4 process occupying one of the CPU
cores 100% of the time).  Reverting that commit entirely fixes the
problem for me.

However, commit f0ed2ce840b3 appears to do more than it follows from its
changelog, because the changelog only says about the changes related to
ctrls->error_idx, while the commit additionally changes error codes
returned by various functions in uvc_ctrl.c and uvc_v4l2.c.  It turns
out that the changes of the returned error codes confuse the user spce,
so it is sufficient to revert the part of commit f0ed2ce840b3 not
mentioned in its changelog to fix the problem.

[ 'ENOENT' is not a valid error return from an ioctl to begin with, and
  I don't understand how anybody ever even thought it would be.  - Linus ]

Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Cc: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Cc: Mauro Carvalho Chehab <mchehab@redhat.com>
Cc: Hans Verkuil <hans.verkuil@cisco.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
11 years agosunxi: Change the machine compatible string.
Maxime Ripard [Tue, 18 Dec 2012 14:17:12 +0000 (15:17 +0100)]
sunxi: Change the machine compatible string.

Commit 68136b10 ("ARM: sunxi: Change device tree naming scheme for
sunxi") changed the naming scheme and the compatible strings used in the
device trees related to the sunXi platform, but forgot to change the
compatible string in the DT machine definition.

This prevents the kernel from booting on these boards.

Signed-off-by: Maxime Ripard <maxime.ripard@free-electrons.com>
11 years agoARM: multi_v7_defconfig: Add ARCH_SUNXI
Maxime Ripard [Wed, 21 Nov 2012 09:57:49 +0000 (10:57 +0100)]
ARM: multi_v7_defconfig: Add ARCH_SUNXI

Signed-off-by: Maxime Ripard <maxime.ripard@free-electrons.com>
11 years agodrm/nve0/graph: fix fuc, and enable acceleration on all known chipsets
Ben Skeggs [Thu, 20 Dec 2012 02:52:06 +0000 (12:52 +1000)]
drm/nve0/graph: fix fuc, and enable acceleration on all known chipsets

Also adds GK106 to chipsets known by ucode.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
11 years agodrm/nvc0/graph: fix fuc, and enable acceleration on GF119
Ben Skeggs [Thu, 20 Dec 2012 02:50:52 +0000 (12:50 +1000)]
drm/nvc0/graph: fix fuc, and enable acceleration on GF119

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
11 years agodrm/nouveau/bios: cache ramcfg strap on later chipsets
Ben Skeggs [Wed, 19 Dec 2012 21:48:51 +0000 (07:48 +1000)]
drm/nouveau/bios: cache ramcfg strap on later chipsets

This fixes suspend/resume on at least Quadro 400.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
11 years agodrm/nouveau/mxm: silence output if no bios data
Ben Skeggs [Tue, 11 Dec 2012 03:32:59 +0000 (13:32 +1000)]
drm/nouveau/mxm: silence output if no bios data

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
11 years agodrm/nouveau/bios: parse/display extra version component
Ben Skeggs [Tue, 11 Dec 2012 02:49:18 +0000 (12:49 +1000)]
drm/nouveau/bios: parse/display extra version component

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
11 years agodrm/nouveau/bios: implement opcode 0xa9
Ben Skeggs [Fri, 7 Dec 2012 03:46:52 +0000 (13:46 +1000)]
drm/nouveau/bios: implement opcode 0xa9

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
11 years agodrm/nouveau/bios: update gpio parsing apis to match current design
Ben Skeggs [Fri, 7 Dec 2012 00:09:17 +0000 (10:09 +1000)]
drm/nouveau/bios: update gpio parsing apis to match current design

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
11 years agodrm/nouveau: initial support for GK106
Ben Skeggs [Thu, 6 Dec 2012 04:45:57 +0000 (14:45 +1000)]
drm/nouveau: initial support for GK106

Modesetting seems to work alright, as does graphics (using binary driver
fuc from nve7...).

Lots to be done no doubt, but this'll get an image on the screen for
people.

Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
11 years agoi2c: remove __dev* attributes from subsystem
Bill Pemberton [Tue, 27 Nov 2012 20:59:38 +0000 (15:59 -0500)]
i2c: remove __dev* attributes from subsystem

CONFIG_HOTPLUG is going away as an option.  As result the __dev*
markings will be going away.

Remove use of __devinit, __devexit_p, __devinitdata, __devinitconst,
and __devexit.

Signed-off-by: Bill Pemberton <wfp5p@virginia.edu>
Acked-by: Peter Korsgaard <peter.korsgaard@barco.com> (for ocores and mux-gpio)
Acked-by: Havard Skinnemoen <hskinnemoen@gmail.com> (for i2c-gpio)
Acked-by: Guan Xuetao <gxt@mprc.pku.edu.cn> (for puf3)
Acked-by: Barry Song <baohua.song@csr.com> (for sirf)
Reviewed-by: Jean Delvare <khali@linux-fr.org>
[wsa: Fixed "foo* bar" flaws while we are here]
Signed-off-by: Wolfram Sang <w.sang@pengutronix.de>