]> err.no Git - linux-2.6/log
linux-2.6
19 years ago[PATCH] dmapool: Fix "nocast type" warnings
Victor Fusco [Sat, 10 Sep 2005 07:26:49 +0000 (00:26 -0700)]
[PATCH] dmapool: Fix "nocast type" warnings

Fix the sparse warning "implicit cast to nocast type"

Signed-off-by: Victor Fusco <victor@cetuc.puc-rio.br>
Signed-off-by: Domen Puncer <domen@coderock.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] lib/radix-tree: Fix "nocast type" warnings
Victor Fusco [Sat, 10 Sep 2005 07:26:48 +0000 (00:26 -0700)]
[PATCH] lib/radix-tree: Fix "nocast type" warnings

Fix the sparse warning "implicit cast to nocast type"

Signed-off-by: Victor Fusco <victor@cetuc.puc-rio.br>
Signed-off-by: Domen Puncer <domen@coderock.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] Spelling fixes for Documentation/
Tobias Klauser [Sat, 10 Sep 2005 07:26:46 +0000 (00:26 -0700)]
[PATCH] Spelling fixes for Documentation/

The attached patch fixes the following spelling errors in Documentation/
        - double "the"
        - Several misspellings of function/functionality
        - infomation
        - memeory
        - Recieved
        - wether
and possibly others which I forgot ;-)
Trailing whitespaces on the same line as the typo are also deleted.

Signed-off-by: Tobias Klauser <tklauser@nuerscht.ch>
Signed-off-by: Domen Puncer <domen@coderock.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] mm/swap_state: Fix "nocast type" warnings
Victor Fusco [Sat, 10 Sep 2005 07:26:46 +0000 (00:26 -0700)]
[PATCH] mm/swap_state: Fix "nocast type" warnings

Fix the sparse warning "implicit cast to nocast type"

Signed-off-by: Victor Fusco <victor@cetuc.puc-rio.br>
Signed-off-by: Domen Puncer <domen@coderock.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] Add kerneldoc reference to CodingStyle
Pekka J Enberg [Sat, 10 Sep 2005 07:26:44 +0000 (00:26 -0700)]
[PATCH] Add kerneldoc reference to CodingStyle

Signed-off-by: Pekka Enberg <penberg@cs.helsinki.fi>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] SH64: C99 initializers for hw_interrupt_type structures
Thomas Gleixner [Sat, 10 Sep 2005 07:26:44 +0000 (00:26 -0700)]
[PATCH] SH64: C99 initializers for hw_interrupt_type structures

Convert the initializers of hw_interrupt_type structures to C99 initializers.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] V850: C99 initializers for hw_interrupt_type structures
Thomas Gleixner [Sat, 10 Sep 2005 07:26:43 +0000 (00:26 -0700)]
[PATCH] V850: C99 initializers for hw_interrupt_type structures

Convert the initializers of hw_interrupt_type structures to C99 initializers.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] SH: C99 initializers for hw_interrupt_type structures
Thomas Gleixner [Sat, 10 Sep 2005 07:26:42 +0000 (00:26 -0700)]
[PATCH] SH: C99 initializers for hw_interrupt_type structures

Convert the initializers of hw_interrupt_type structures to C99 initializers.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] PPC: C99 initializers for hw_interrupt_type structures
Thomas Gleixner [Sat, 10 Sep 2005 07:26:40 +0000 (00:26 -0700)]
[PATCH] PPC: C99 initializers for hw_interrupt_type structures

Convert the initializers of hw_interrupt_type structures to C99 initializers.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] kernel/acct: add kerneldoc
Randy Dunlap [Sat, 10 Sep 2005 07:26:39 +0000 (00:26 -0700)]
[PATCH] kernel/acct: add kerneldoc

for kernel/acct.c:
- fix typos
- add kerneldoc for non-static functions

Signed-off-by: Randy Dunlap <rdunlap@xenotime.net>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] char/n_tty: fix sparse warnings (__nocast type)
Victor Fusco [Sat, 10 Sep 2005 07:26:38 +0000 (00:26 -0700)]
[PATCH] char/n_tty: fix sparse warnings (__nocast type)

Fix the sparse warning "implicit cast to nocast type"

Signed-off-by: Victor Fusco <victor@cetuc.puc-rio.br>
Signed-off-by: Domen Puncer <domen@coderock.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] mm/slab: fix sparse warnings
Victor Fusco [Sat, 10 Sep 2005 07:26:36 +0000 (00:26 -0700)]
[PATCH] mm/slab: fix sparse warnings

Fix the sparse warning "implicit cast to nocast type"

Signed-off-by: Victor Fusco <victor@cetuc.puc-rio.br>
Signed-off-by: Domen Puncer <domen@coderock.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] sb16_csp: untypedef
Alexey Dobriyan [Sat, 10 Sep 2005 07:26:35 +0000 (00:26 -0700)]
[PATCH] sb16_csp: untypedef

Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Signed-off-by: Domen Puncer <domen@coderock.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] sb16_csp: remove home-grown le??_to_cpu macros
Alexey Dobriyan [Sat, 10 Sep 2005 07:26:34 +0000 (00:26 -0700)]
[PATCH] sb16_csp: remove home-grown le??_to_cpu macros

Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Signed-off-by: Domen Puncer <domen@coderock.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] hdpu_cpustate.c: misc_register() can fail
Christophe Lucas [Sat, 10 Sep 2005 07:26:33 +0000 (00:26 -0700)]
[PATCH] hdpu_cpustate.c: misc_register() can fail

Signed-off-by: Christophe Lucas <clucas@rotomalug.org>
Signed-off-by: Domen Puncer <domen@coderock.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] drivers/char/lcd.c: misc_register() can fail
Christophe Lucas [Sat, 10 Sep 2005 07:26:32 +0000 (00:26 -0700)]
[PATCH] drivers/char/lcd.c: misc_register() can fail

Signed-off-by: Christophe Lucas <clucas@rotomalug.org>
Signed-off-by: Domen Puncer <domen@coderock.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] applicom: fix error handling
Christophe Lucas [Sat, 10 Sep 2005 07:26:29 +0000 (00:26 -0700)]
[PATCH] applicom: fix error handling

misc_register() can fail.

Signed-off-by: Christophe Lucas <clucas@rotomalug.org>
Signed-off-by: Domen Puncer <domen@coderock.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] mm/filemap.c: make two functions static
Adrian Bunk [Sat, 10 Sep 2005 07:26:28 +0000 (00:26 -0700)]
[PATCH] mm/filemap.c: make two functions static

With Nick Piggin <npiggin@suse.de>

Give some things static scope.

Signed-off-by: Adrian Bunk <bunk@stusta.de>
Signed-off-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] Yet another RCU documentation update
Paul E. McKenney [Sat, 10 Sep 2005 07:26:24 +0000 (00:26 -0700)]
[PATCH] Yet another RCU documentation update

Update RCU documentation based on discussions and review of RCU-based tree
patches.  Add an introductory whatisRCU.txt file.

Signed-off-by: <paulmck@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] Remove even more stale references to Documentation/smp.tex
Arthur Othieno [Sat, 10 Sep 2005 07:26:22 +0000 (00:26 -0700)]
[PATCH] Remove even more stale references to Documentation/smp.tex

Randy cleaned out the bulk of these stale references to the now long gone
Documentation/smp.tex back in 2004.  I followed this up with a few more
sweeps.  Somehow, these have managed to sneak back in since.

I can't seem to figure out a contact point for M32R (no one listed in
MAINTAINERS!), but, these patches are only but trivial.

Signed-off-by: Arthur Othieno <a.othieno@bluewin.ch>
Acked-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] sched: allow the load to grow upto its cpu_power
Siddha, Suresh B [Sat, 10 Sep 2005 07:26:21 +0000 (00:26 -0700)]
[PATCH] sched: allow the load to grow upto its cpu_power

Don't pull tasks from a group if that would cause the group's total load to
drop below its total cpu_power (ie.  cause the group to start going idle).

Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Signed-off-by: Nick Piggin <npiggin@suse.de>
Acked-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] sched: don't kick ALB in the presence of pinned task
Siddha, Suresh B [Sat, 10 Sep 2005 07:26:21 +0000 (00:26 -0700)]
[PATCH] sched: don't kick ALB in the presence of pinned task

Jack Steiner brought this issue at my OLS talk.

Take a scenario where two tasks are pinned to two HT threads in a physical
package.  Idle packages in the system will keep kicking migration_thread on
the busy package with out any success.

We will run into similar scenarios in the presence of CMP/NUMA.

Signed-off-by: Suresh Siddha <suresh.b.siddha@intel.com>
Acked-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] sched: use cached variable in sys_sched_yield()
Renaud Lienhart [Sat, 10 Sep 2005 07:26:20 +0000 (00:26 -0700)]
[PATCH] sched: use cached variable in sys_sched_yield()

In sys_sched_yield(), we cache current->array in the "array" variable, thus
there's no need to dereference "current" again later.

Signed-Off-By: Renaud Lienhart <renaud.lienhart@free.fr>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] sched: HT optimisation
Nick Piggin [Sat, 10 Sep 2005 07:26:19 +0000 (00:26 -0700)]
[PATCH] sched: HT optimisation

If an idle sibling of an HT queue encounters a busy sibling, then make
higher level load balancing of the non-idle variety.

Performance of multiprocessor HT systems with low numbers of tasks
(generally < number of virtual CPUs) can be significantly worse than the
exact same workloads when running in non-HT mode.  The reason is largely
due to poor scheduling behaviour.

This patch improves the situation, making the performance gap far less
significant on one problematic test case (tbench).

Signed-off-by: Nick Piggin <npiggin@suse.de>
Acked-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] sched: less locking
Nick Piggin [Sat, 10 Sep 2005 07:26:18 +0000 (00:26 -0700)]
[PATCH] sched: less locking

During periodic load balancing, don't hold this runqueue's lock while
scanning remote runqueues, which can take a non trivial amount of time
especially on very large systems.

Holding the runqueue lock will only help to stabilise ->nr_running, however
this doesn't do much to help because tasks being woken will simply get held
up on the runqueue lock, so ->nr_running would not provide a really
accurate picture of runqueue load in that case anyway.

What's more, ->nr_running (and possibly the cpu_load averages) of remote
runqueues won't be stable anyway, so load balancing is always an inexact
operation.

Signed-off-by: Nick Piggin <npiggin@suse.de>
Acked-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] sched: less newidle locking
Nick Piggin [Sat, 10 Sep 2005 07:26:16 +0000 (00:26 -0700)]
[PATCH] sched: less newidle locking

Similarly to the earlier change in load_balance, only lock the runqueue in
load_balance_newidle if the busiest queue found has a nr_running > 1.  This
will reduce frequency of expensive remote runqueue lock aquisitions in the
schedule() path on some workloads.

Signed-off-by: Nick Piggin <npiggin@suse.de>
Acked-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] sched: fix SMT scheduler latency bug
Ingo Molnar [Sat, 10 Sep 2005 07:26:16 +0000 (00:26 -0700)]
[PATCH] sched: fix SMT scheduler latency bug

William Weston reported unusually high scheduling latencies on his x86 HT
box, on the -RT kernel.  I managed to reproduce it on my HT box and the
latency tracer shows the incident in action:

                 _------=> CPU#
                / _-----=> irqs-off
               | / _----=> need-resched
               || / _---=> hardirq/softirq
               ||| / _--=> preempt-depth
               |||| /
               |||||     delay
   cmd     pid ||||| time  |   caller
      \   /    |||||   \   |   /
      du-2803  3Dnh2    0us : __trace_start_sched_wakeup (try_to_wake_up)
        ..............................................................
        ... we are running on CPU#3, PID 2778 gets woken to CPU#1: ...
        ..............................................................
      du-2803  3Dnh2    0us : __trace_start_sched_wakeup <<...>-2778> (73 1)
      du-2803  3Dnh2    0us : _raw_spin_unlock (try_to_wake_up)
        ................................................
        ... still on CPU#3, we send an IPI to CPU#1: ...
        ................................................
      du-2803  3Dnh1    0us : resched_task (try_to_wake_up)
      du-2803  3Dnh1    1us : smp_send_reschedule (try_to_wake_up)
      du-2803  3Dnh1    1us : send_IPI_mask_bitmask (smp_send_reschedule)
      du-2803  3Dnh1    2us : _raw_spin_unlock_irqrestore (try_to_wake_up)
        ...............................................
        ... 1 usec later, the IPI arrives on CPU#1: ...
        ...............................................
  <idle>-0     1Dnh.    2us : smp_reschedule_interrupt (c0100c5a 0 0)

So far so good, this is the normal wakeup/preemption mechanism.  But here
comes the scheduler anomaly on CPU#1:

  <idle>-0     1Dnh.    2us : preempt_schedule_irq (need_resched)
  <idle>-0     1Dnh.    2us : preempt_schedule_irq (need_resched)
  <idle>-0     1Dnh.    3us : __schedule (preempt_schedule_irq)
  <idle>-0     1Dnh.    3us : profile_hit (__schedule)
  <idle>-0     1Dnh1    3us : sched_clock (__schedule)
  <idle>-0     1Dnh1    4us : _raw_spin_lock_irq (__schedule)
  <idle>-0     1Dnh1    4us : _raw_spin_lock_irqsave (__schedule)
  <idle>-0     1Dnh2    5us : _raw_spin_unlock (__schedule)
  <idle>-0     1Dnh1    5us : preempt_schedule (__schedule)
  <idle>-0     1Dnh1    6us : _raw_spin_lock (__schedule)
  <idle>-0     1Dnh2    6us : find_next_bit (__schedule)
  <idle>-0     1Dnh2    6us : _raw_spin_lock (__schedule)
  <idle>-0     1Dnh3    7us : find_next_bit (__schedule)
  <idle>-0     1Dnh3    7us : find_next_bit (__schedule)
  <idle>-0     1Dnh3    8us : _raw_spin_unlock (__schedule)
  <idle>-0     1Dnh2    8us : preempt_schedule (__schedule)
  <idle>-0     1Dnh2    8us : find_next_bit (__schedule)
  <idle>-0     1Dnh2    9us : trace_stop_sched_switched (__schedule)
  <idle>-0     1Dnh2    9us : _raw_spin_lock (trace_stop_sched_switched)
  <idle>-0     1Dnh3   10us : trace_stop_sched_switched <<...>-2778> (73 8c)
  <idle>-0     1Dnh3   10us : _raw_spin_unlock (trace_stop_sched_switched)
  <idle>-0     1Dnh1   10us : _raw_spin_unlock (__schedule)
  <idle>-0     1Dnh.   11us : local_irq_enable_noresched (preempt_schedule_irq)
  <idle>-0     1Dnh.   11us < (0)

we didnt pick up pid 2778! It only gets scheduled much later:

   <...>-2778  1Dnh2  412us : __switch_to (__schedule)
   <...>-2778  1Dnh2  413us : __schedule <<idle>-0> (8c 73)
   <...>-2778  1Dnh2  413us : _raw_spin_unlock (__schedule)
   <...>-2778  1Dnh1  413us : trace_stop_sched_switched (__schedule)
   <...>-2778  1Dnh1  414us : _raw_spin_lock (trace_stop_sched_switched)
   <...>-2778  1Dnh2  414us : trace_stop_sched_switched <<...>-2778> (73 1)
   <...>-2778  1Dnh2  414us : _raw_spin_unlock (trace_stop_sched_switched)
   <...>-2778  1Dnh1  415us : trace_stop_sched_switched (__schedule)

the reason for this anomaly is the following code in dependent_sleeper():

                /*
                 * If a user task with lower static priority than the
                 * running task on the SMT sibling is trying to schedule,
                 * delay it till there is proportionately less timeslice
                 * left of the sibling task to prevent a lower priority
                 * task from using an unfair proportion of the
                 * physical cpu's resources. -ck
                 */
[...]
                        if (((smt_curr->time_slice * (100 - sd->per_cpu_gain) /
                                100) > task_timeslice(p)))
                                        ret = 1;

Note that in contrast to the comment above, we dont actually do the check
based on static priority, we do the check based on timeslices.  But
timeslices go up and down, and even highprio tasks can randomly have very
low timeslices (just before their next refill) and can thus be judged as
'lowprio' by the above piece of code.  This condition is clearly buggy.
The correct test is to check for static_prio _and_ to check for the
preemption priority.  Even on different static priority levels, a
higher-prio interactive task should not be delayed due to a
higher-static-prio CPU hog.

There is a symmetric bug in the 'kick SMT sibling' code of this function as
well, which can be solved in a similar way.

The patch below (against the current scheduler queue in -mm) fixes both
bugs.  I have build and boot-tested this on x86 SMT, and nice +20 tasks
still get properly throttled - so the dependent-sleeper logic is still in
action.

btw., these bugs pessimised the SMT scheduler because the 'delay wakeup'
property was applied too liberally, so this fix is likely a throughput
improvement as well.

I separated out a smt_slice() function to make the code easier to read.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] sched: TASK_NONINTERACTIVE
Ingo Molnar [Sat, 10 Sep 2005 07:26:12 +0000 (00:26 -0700)]
[PATCH] sched: TASK_NONINTERACTIVE

This patch implements a task state bit (TASK_NONINTERACTIVE), which can be
used by blocking points to mark the task's wait as "non-interactive".  This
does not mean the task will be considered a CPU-hog - the wait will simply
not have an effect on the waiting task's priority - positive or negative
alike.  Right now only pipe_wait() will make use of it, because it's a
common source of not-so-interactive waits (kernel compilation jobs, etc.).

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] sched cleanups
Ingo Molnar [Sat, 10 Sep 2005 07:26:11 +0000 (00:26 -0700)]
[PATCH] sched cleanups

whitespace cleanups.

Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] sched: make idlest_group/cpu cpus_allowed-aware
M.Baris Demiray [Sat, 10 Sep 2005 07:26:09 +0000 (00:26 -0700)]
[PATCH] sched: make idlest_group/cpu cpus_allowed-aware

Add relevant checks into find_idlest_group() and find_idlest_cpu() to make
them return only the groups that have allowed CPUs and allowed CPUs
respectively.

Signed-off-by: M.Baris Demiray <baris@labristeknoloji.com>
Signed-off-by: Nick Piggin <nickpiggin@yahoo.com.au>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] sched: run SCHED_NORMAL tasks with real time tasks on SMT siblings
Con Kolivas [Sat, 10 Sep 2005 07:26:08 +0000 (00:26 -0700)]
[PATCH] sched: run SCHED_NORMAL tasks with real time tasks on SMT siblings

The hyperthread aware nice handling currently puts to sleep any non real
time task when a real time task is running on its sibling cpu.  This can
lead to prolonged starvation by having the non real time task pegged to the
cpu with load balancing not pulling that task away.

Currently we force lower priority hyperthread tasks to run a percentage of
time difference based on timeslice differences which is meaningless when
comparing real time tasks to SCHED_NORMAL tasks.  We can allow non real
time tasks to run with real time tasks on the sibling up to per_cpu_gain%
if we use jiffies as a counter.

Cleanups and micro-optimisations to the relevant code section should make
it more understandable as well.

Signed-off-by: Con Kolivas <kernel@kolivas.org>
Acked-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] synclink_cs add statistics clear
Paul Fulghum [Sat, 10 Sep 2005 07:26:07 +0000 (00:26 -0700)]
[PATCH] synclink_cs add statistics clear

Add ability to clear statistics.

Signed-off-by: Paul Fulghum <paulkf@microgate.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] cpuset semaphore depth check deadlock fix
Paul Jackson [Sat, 10 Sep 2005 07:26:06 +0000 (00:26 -0700)]
[PATCH] cpuset semaphore depth check deadlock fix

The cpusets-formalize-intermediate-gfp_kernel-containment patch
has a deadlock problem.

This patch was part of a set of four patches to make more
extensive use of the cpuset 'mem_exclusive' attribute to
manage kernel GFP_KERNEL memory allocations and to constrain
the out-of-memory (oom) killer.

A task that is changing cpusets in particular ways on a system
when it is very short of free memory could double trip over
the global cpuset_sem semaphore (get the lock and then deadlock
trying to get it again).

The second attempt to get cpuset_sem would be in the routine
cpuset_zone_allowed().  This was discovered by code inspection.
I can not reproduce the problem except with an artifically
hacked kernel and a specialized stress test.

In real life you cannot hit this unless you are manipulating
cpusets, and are very unlikely to hit it unless you are rapidly
modifying cpusets on a memory tight system.  Even then it would
be a rare occurence.

If you did hit it, the task double tripping over cpuset_sem
would deadlock in the kernel, and any other task also trying
to manipulate cpusets would deadlock there too, on cpuset_sem.
Your batch manager would be wedged solid (if it was cpuset
savvy), but classic Unix shells and utilities would work well
enough to reboot the system.

The unusual condition that led to this bug is that unlike most
semaphores, cpuset_sem _can_ be acquired while in the page
allocation code, when __alloc_pages() calls cpuset_zone_allowed.
So it easy to mistakenly perform the following sequence:
  1) task makes system call to alter a cpuset
  2) take cpuset_sem
  3) try to allocate memory
  4) memory allocator, via cpuset_zone_allowed, trys to take cpuset_sem
  5) deadlock

The reason that this is not a serious bug for most users
is that almost all calls to allocate memory don't require
taking cpuset_sem.  Only some code paths off the beaten
track require taking cpuset_sem -- which is good.  Taking
a global semaphore on the main code path for allocating
memory would not scale well.

This patch fixes this deadlock by wrapping the up() and down()
calls on cpuset_sem in kernel/cpuset.c with code that tracks
the nesting depth of the current task on that semaphore, and
only does the real down() if the task doesn't hold the lock
already, and only does the real up() if the nesting depth
(number of unmatched downs) is exactly one.

The previous required use of refresh_mems(), anytime that
the cpuset_sem semaphore was acquired and the code executed
while holding that semaphore might try to allocate memory, is
no longer required.  Two refresh_mems() calls were removed
thanks to this.  This is a good change, as failing to get
all the necessary refresh_mems() calls placed was a primary
source of bugs in this cpuset code.  The only remaining call
to refresh_mems() is made while doing a memory allocation,
if certain task memory placement data needs to be updated
from its cpuset, due to the cpuset having been changed behind
the tasks back.

Signed-off-by: Paul Jackson <pj@sgi.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] spinlock consolidation
Ingo Molnar [Sat, 10 Sep 2005 07:25:56 +0000 (00:25 -0700)]
[PATCH] spinlock consolidation

This patch (written by me and also containing many suggestions of Arjan van
de Ven) does a major cleanup of the spinlock code.  It does the following
things:

 - consolidates and enhances the spinlock/rwlock debugging code

 - simplifies the asm/spinlock.h files

 - encapsulates the raw spinlock type and moves generic spinlock
   features (such as ->break_lock) into the generic code.

 - cleans up the spinlock code hierarchy to get rid of the spaghetti.

Most notably there's now only a single variant of the debugging code,
located in lib/spinlock_debug.c.  (previously we had one SMP debugging
variant per architecture, plus a separate generic one for UP builds)

Also, i've enhanced the rwlock debugging facility, it will now track
write-owners.  There is new spinlock-owner/CPU-tracking on SMP builds too.
All locks have lockup detection now, which will work for both soft and hard
spin/rwlock lockups.

The arch-level include files now only contain the minimally necessary
subset of the spinlock code - all the rest that can be generalized now
lives in the generic headers:

 include/asm-i386/spinlock_types.h       |   16
 include/asm-x86_64/spinlock_types.h     |   16

I have also split up the various spinlock variants into separate files,
making it easier to see which does what. The new layout is:

   SMP                         |  UP
   ----------------------------|-----------------------------------
   asm/spinlock_types_smp.h    |  linux/spinlock_types_up.h
   linux/spinlock_types.h      |  linux/spinlock_types.h
   asm/spinlock_smp.h          |  linux/spinlock_up.h
   linux/spinlock_api_smp.h    |  linux/spinlock_api_up.h
   linux/spinlock.h            |  linux/spinlock.h

/*
 * here's the role of the various spinlock/rwlock related include files:
 *
 * on SMP builds:
 *
 *  asm/spinlock_types.h: contains the raw_spinlock_t/raw_rwlock_t and the
 *                        initializers
 *
 *  linux/spinlock_types.h:
 *                        defines the generic type and initializers
 *
 *  asm/spinlock.h:       contains the __raw_spin_*()/etc. lowlevel
 *                        implementations, mostly inline assembly code
 *
 *   (also included on UP-debug builds:)
 *
 *  linux/spinlock_api_smp.h:
 *                        contains the prototypes for the _spin_*() APIs.
 *
 *  linux/spinlock.h:     builds the final spin_*() APIs.
 *
 * on UP builds:
 *
 *  linux/spinlock_type_up.h:
 *                        contains the generic, simplified UP spinlock type.
 *                        (which is an empty structure on non-debug builds)
 *
 *  linux/spinlock_types.h:
 *                        defines the generic type and initializers
 *
 *  linux/spinlock_up.h:
 *                        contains the __raw_spin_*()/etc. version of UP
 *                        builds. (which are NOPs on non-debug, non-preempt
 *                        builds)
 *
 *   (included on UP-non-debug builds:)
 *
 *  linux/spinlock_api_up.h:
 *                        builds the _spin_*() APIs.
 *
 *  linux/spinlock.h:     builds the final spin_*() APIs.
 */

All SMP and UP architectures are converted by this patch.

arm, i386, ia64, ppc, ppc64, s390/s390x, x64 was build-tested via
crosscompilers.  m32r, mips, sh, sparc, have not been tested yet, but should
be mostly fine.

From: Grant Grundler <grundler@parisc-linux.org>

  Booted and lightly tested on a500-44 (64-bit, SMP kernel, dual CPU).
  Builds 32-bit SMP kernel (not booted or tested).  I did not try to build
  non-SMP kernels.  That should be trivial to fix up later if necessary.

  I converted bit ops atomic_hash lock to raw_spinlock_t.  Doing so avoids
  some ugly nesting of linux/*.h and asm/*.h files.  Those particular locks
  are well tested and contained entirely inside arch specific code.  I do NOT
  expect any new issues to arise with them.

 If someone does ever need to use debug/metrics with them, then they will
  need to unravel this hairball between spinlocks, atomic ops, and bit ops
  that exist only because parisc has exactly one atomic instruction: LDCW
  (load and clear word).

From: "Luck, Tony" <tony.luck@intel.com>

   ia64 fix

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Arjan van de Ven <arjanv@infradead.org>
Signed-off-by: Grant Grundler <grundler@parisc-linux.org>
Cc: Matthew Wilcox <willy@debian.org>
Signed-off-by: Hirokazu Takata <takata@linux-m32r.org>
Signed-off-by: Mikael Pettersson <mikpe@csd.uu.se>
Signed-off-by: Benoit Boissinot <benoit.boissinot@ens-lyon.org>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] Subject: PATCH: fix numa caused compile warnings
Alan Cox [Sat, 10 Sep 2005 07:25:49 +0000 (00:25 -0700)]
[PATCH] Subject: PATCH: fix numa caused compile warnings

pcibus_to_cpumask expands into more than just an initialiser so gcc
moans about code before variable declarations.

Signed-off-by: Alan Cox <alan@redhat.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] ntfs build fix
Andrew Morton [Sat, 10 Sep 2005 07:25:47 +0000 (00:25 -0700)]
[PATCH] ntfs build fix

*** Warning: "bit_spin_lock" [fs/ntfs/ntfs.ko] undefined!
*** Warning: "bit_spin_unlock" [fs/ntfs/ntfs.ko] undefined!

Cc: Anton Altaparmakov <aia21@cantab.net>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] md: fix BUG when raid10 rebuilds without enough drives
NeilBrown [Fri, 9 Sep 2005 23:24:04 +0000 (16:24 -0700)]
[PATCH] md: fix BUG when raid10 rebuilds without enough drives

This shouldn't be a BUG.  We should cope.

Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] md: fix raid10 assembly when too many devices are missing
NeilBrown [Fri, 9 Sep 2005 23:24:03 +0000 (16:24 -0700)]
[PATCH] md: fix raid10 assembly when too many devices are missing

If you try to assemble an array with too many missing devices, raid10 will now
reject the attempt, instead of allowing it.

Also check when hot-adding a drive and refuse the hot-add if the array is
beyond hope.

Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] md: really get sb_size setting right in all cases
NeilBrown [Fri, 9 Sep 2005 23:24:02 +0000 (16:24 -0700)]
[PATCH] md: really get sb_size setting right in all cases

There was another case where sb_size wasn't being set, so instead do the
sensible thing and set if when filling in the content of a superblock.  That
ensures that whenever we write a superblock, the sb_size MUST be set.

Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] md: make sure the new 'sb_size' is set properly device added without pre...
NeilBrown [Fri, 9 Sep 2005 23:24:01 +0000 (16:24 -0700)]
[PATCH] md: make sure the new 'sb_size' is set properly device added without pre-existing superblock.

There are two ways to add devices to an md/raid array.

  It can have superblock written to it, and then given to the md driver,
  which will read the superblock (the new way)

or

  md can be told (through SET_ARRAY_INFO) the shape of the array, and
  the told about individual drives, and md will create the required
  superblock (the old way).

The newly introduced sb_size was only set for drives being added the
new way, not the old ways.  Oops :-(

Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] md: report spare drives in /proc/mdstat
NeilBrown [Fri, 9 Sep 2005 23:24:00 +0000 (16:24 -0700)]
[PATCH] md: report spare drives in /proc/mdstat

Just like failed drives have (F), so spare drives now have (S).

Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] md: add information about superblock version to /proc/mdstat
NeilBrown [Fri, 9 Sep 2005 23:24:00 +0000 (16:24 -0700)]
[PATCH] md: add information about superblock version to /proc/mdstat

Leave it unchanged if the original (0.90) is used, incase it might be a
compatability problem.

Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] md: use queue_hardsect_size instead of block_size for md superblock size...
NeilBrown [Fri, 9 Sep 2005 23:23:59 +0000 (16:23 -0700)]
[PATCH] md: use queue_hardsect_size instead of block_size for md superblock size calc.

Doh.  I want the physical hard-sector-size, not the current block size...

Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] md: choose better default offset for bitmap.
NeilBrown [Fri, 9 Sep 2005 23:23:58 +0000 (16:23 -0700)]
[PATCH] md: choose better default offset for bitmap.

On reflection, a better default location for hot-adding bitmaps with version-1
superblocks is immediately after the superblock.  There might not be much room
there, but there is usually atleast 3k, and that is a good start.

Signed-off-by: Neil Brown <neilb@suse.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] md: tidy up daemon stop/start code in md/bitmap.c
NeilBrown [Fri, 9 Sep 2005 23:23:58 +0000 (16:23 -0700)]
[PATCH] md: tidy up daemon stop/start code in md/bitmap.c

The bitmap code used to have two daemons, so there is some 'common' start/stop
code.  But now there is only one, so the common code is just noise.

This patch tidies this up somewhat.

Signed-off-by: Neil Brown <neilb@cse.unsw.edu.au>
Signed-off-by: Adrian Bunk <bunk@stusta.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] md: ensure bitmap_writeback_daemon handles shutdown properly.
NeilBrown [Fri, 9 Sep 2005 23:23:57 +0000 (16:23 -0700)]
[PATCH] md: ensure bitmap_writeback_daemon handles shutdown properly.

mddev->bitmap gets clearred before the writeback daemon is stopped.  So the
write_back daemon needs to be careful not to dereference the 'bitmap' if it is
NULL.

Signed-off-by: Neil Brown <neilb@cse.unsw.edu.au>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] md: use kthread infrastructure in md
NeilBrown [Fri, 9 Sep 2005 23:23:56 +0000 (16:23 -0700)]
[PATCH] md: use kthread infrastructure in md

Switch MD to use the kthread infrastructure, to simplify the code and get rid
of tasklist_lock abuse in md_unregister_thread.

Also don't flush signals in md_thread, as the called thread will always do
that.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Neil Brown <neilb@cse.unsw.edu.au>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] md: write-intent bitmap support for raid6
NeilBrown [Fri, 9 Sep 2005 23:23:55 +0000 (16:23 -0700)]
[PATCH] md: write-intent bitmap support for raid6

This is a direct port of the raid5 patch.

Signed-off-by: Neil Brown <neilb@cse.unsw.edu.au>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] md: add write-intent-bitmap support to raid5
NeilBrown [Fri, 9 Sep 2005 23:23:54 +0000 (16:23 -0700)]
[PATCH] md: add write-intent-bitmap support to raid5

Most awkward part of this is delaying write requests until bitmap updates have
been flushed.

To achieve this, we have a sequence number (seq_flush) which is incremented
each time the raid5 is unplugged.

If the raid thread notices that this has changed, it flushes bitmap changes,
and assigned the value of seq_flush to seq_write.

When a write request arrives, it is given the number from seq_write, and that
write request may not complete until seq_flush is larger than the saved seq
number.

We have a new queue for storing stripes which are waiting for a bitmap flush
and an extra flag for stripes to record if the write was 'degraded' and so
should not clear the a bit in the bitmap.

Signed-off-by: Neil Brown <neilb@cse.unsw.edu.au>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] md: limit size of sb read/written to appropriate amount
NeilBrown [Fri, 9 Sep 2005 23:23:53 +0000 (16:23 -0700)]
[PATCH] md: limit size of sb read/written to appropriate amount

version-1 superblocks are not (normally) 4K long, and can be of variable size.
 Writing the full 4K can cause corruption (but only in non-default
configurations).

With this patch the super-block-flavour can choose a size to read, and set a
size to write based on what it finds.

Signed-off-by: Neil Brown <neilb@cse.unsw.edu.au>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] md: remove old cruft from md_k.h header file
NeilBrown [Fri, 9 Sep 2005 23:23:53 +0000 (16:23 -0700)]
[PATCH] md: remove old cruft from md_k.h header file

These inlines haven't been used for ages, they should go.

Signed-off-by: Neil Brown <neilb@cse.unsw.edu.au>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] md: fix bitmap/read_sb_page so that it handles errors properly.
NeilBrown [Fri, 9 Sep 2005 23:23:52 +0000 (16:23 -0700)]
[PATCH] md: fix bitmap/read_sb_page so that it handles errors properly.

read_sb_page() assumed that if sync_page_io fails, the device would be marked
faultly.  However it isn't.  So in the face of error, read_sb_page would loop
forever.

Redo the logic so that this cannot happen.

Signed-off-by: Neil Brown <neilb@cse.unsw.edu.au>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] md: allow md to load a superblock with feature-bit '1' set
NeilBrown [Fri, 9 Sep 2005 23:23:51 +0000 (16:23 -0700)]
[PATCH] md: allow md to load a superblock with feature-bit '1' set

As this is used to flag an internal bitmap.

Also, introduce symbolic names for feature bits.

Signed-off-by: Neil Brown <neilb@cse.unsw.edu.au>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] md: allow hot-adding devices to arrays with non-persistant superblocks.
NeilBrown [Fri, 9 Sep 2005 23:23:50 +0000 (16:23 -0700)]
[PATCH] md: allow hot-adding devices to arrays with non-persistant superblocks.

It is possibly (and occasionally useful) to have a raid1 without persistent
superblocks.  The code in add_new_disk for adding a device to such an array
always tries to read a superblock.

This will obviously fail.

So do the appropriate test and call md_import_device with
appropriate args.

Signed-off-by: Neil Brown <neilb@cse.unsw.edu.au>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] md: do not set mddev->bitmap until bitmap is fully initialised
NeilBrown [Fri, 9 Sep 2005 23:23:50 +0000 (16:23 -0700)]
[PATCH] md: do not set mddev->bitmap until bitmap is fully initialised

When hot-adding a bitmap, bitmap_daemon_work could get called while the bitmap
is being created, so don't set mddev->bitmap until the bitmap is ready.

This requires freeing the bitmap inside bitmap_create if creation failed
part-way through.

Signed-off-by: Neil Brown <neilb@cse.unsw.edu.au>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] md: make sure bitmap_daemon_work actually does work.
NeilBrown [Fri, 9 Sep 2005 23:23:49 +0000 (16:23 -0700)]
[PATCH] md: make sure bitmap_daemon_work actually does work.

The 'lastrun' time wasn't being initialised, so it could be half a
jiffie-cycle before it seemed to be time to do work again.

Signed-off-by: Neil Brown <neilb@cse.unsw.edu.au>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] md: raid1_quiesce is back to front, fix it.
NeilBrown [Fri, 9 Sep 2005 23:23:48 +0000 (16:23 -0700)]
[PATCH] md: raid1_quiesce is back to front, fix it.

A state of 0 mean 'not quiesced'
A state of 1 means 'is quiesced'

The original code got this wrong.

Signed-off-by: Neil Brown <neilb@cse.unsw.edu.au>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] md: support md/linear array with components greater than 2 terabytes.
NeilBrown [Fri, 9 Sep 2005 23:23:47 +0000 (16:23 -0700)]
[PATCH] md: support md/linear array with components greater than 2 terabytes.

linear currently uses division by the size of the smallest componenet device
to find which device a request goes to.  If that smallest device is larger
than 2 terabytes, then the division will not work on some systems.

So we introduce a pre-shift, and take care not to make the hash table too
large, much like the code in raid0.

Also get rid of conf->nr_zones, which is not needed.

Signed-off-by: Neil Brown <neilb@cse.unsw.edu.au>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] md: add write-behind support for md/raid1
NeilBrown [Fri, 9 Sep 2005 23:23:47 +0000 (16:23 -0700)]
[PATCH] md: add write-behind support for md/raid1

If a device is flagged 'WriteMostly' and the array has a bitmap, and the
bitmap superblock indicates that write_behind is allowed, then write_behind is
enabled for WriteMostly devices.

Write requests will be acknowledges as complete to the caller (via b_end_io)
when all non-WriteMostly devices have completed the write, but will not be
cleared from the bitmap until all devices complete.

This requires memory allocation to make a local copy of the data being
written.  If there is insufficient memory, then we fall-back on normal write
semantics.

Signed-Off-By: Paul Clements <paul.clements@steeleye.com>
Signed-off-by: Neil Brown <neilb@cse.unsw.edu.au>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] md: support write-mostly device in raid1
NeilBrown [Fri, 9 Sep 2005 23:23:45 +0000 (16:23 -0700)]
[PATCH] md: support write-mostly device in raid1

This allows a device in a raid1 to be marked as "write mostly".  Read requests
will only be sent if there is no other option.

Signed-off-by: Neil Brown <neilb@cse.unsw.edu.au>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] md: all hot-add and hot-remove of md intent logging bitmaps
NeilBrown [Fri, 9 Sep 2005 23:23:45 +0000 (16:23 -0700)]
[PATCH] md: all hot-add and hot-remove of md intent logging bitmaps

Both file-bitmaps and superblock bitmaps are supported.

If you add a bitmap file on the array device, you lose.

This introduces a 'default_bitmap_offset' field in mddev, as the ioctl used
for adding a superblock bitmap doesn't have room for giving an offset.  Later,
this value will be setable via sysfs.

Signed-off-by: Neil Brown <neilb@cse.unsw.edu.au>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] md: improve handling of bitmap initialisation.
NeilBrown [Fri, 9 Sep 2005 23:23:44 +0000 (16:23 -0700)]
[PATCH] md: improve handling of bitmap initialisation.

When we find a 'stale' bitmap, possibly because it is new, we should just
assume every bit needs to be set, but rather base the setting of bits on the
current state of the array (degraded and recovery_cp).

Signed-off-by: Neil Brown <neilb@cse.unsw.edu.au>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] md: don't allow new md/bitmap file to be set if one already exists
NeilBrown [Fri, 9 Sep 2005 23:23:43 +0000 (16:23 -0700)]
[PATCH] md: don't allow new md/bitmap file to be set if one already exists

... otherwise we loose a reference and can never free the file.

Signed-off-by: Neil Brown <neilb@cse.unsw.edu.au>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] dm: fix rh_dec()/rh_inc() race in dm-raid1.c
Jun'ichi Nomura [Fri, 9 Sep 2005 23:23:42 +0000 (16:23 -0700)]
[PATCH] dm: fix rh_dec()/rh_inc() race in dm-raid1.c

Fix another bug in dm-raid1.c that the dirty region may stay in or be moved
to clean list and freed while in use.

It happens as follows:

   CPU0                                   CPU1
   ------------------------------------------------------------------------------
   rh_dec()
     if (atomic_dec_and_test(pending))
        <the region is still marked dirty>
                                          rh_inc()
                                            if the region is clean
                                               mark the region dirty
                                               and remove from clean list
        mark the region clean
        and move to clean list
                                                  atomic_inc(pending)

At this stage, the region is in clean list and will be mistakenly reclaimed
by rh_update_states() later.

Signed-off-by: Jun'ichi Nomura <j-nomura@ce.jp.nec.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] md: fail IO request to md that require a barrier.
NeilBrown [Fri, 9 Sep 2005 23:23:41 +0000 (16:23 -0700)]
[PATCH] md: fail IO request to md that require a barrier.

md does not yet support BIO_RW_BARRIER, so be honest about it and fail
(-EOPNOTSUPP) any such requests.

Signed-off-by: Neil Brown <neilb@cse.unsw.edu.au>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] md: fix minor error in raid10 read-balancing calculation.
NeilBrown [Fri, 9 Sep 2005 23:23:40 +0000 (16:23 -0700)]
[PATCH] md: fix minor error in raid10 read-balancing calculation.

'this_sector' is a virtual (array) address while 'head_position' is a physical
(device) address, so substraction doesn't make any sense.  devs[slot].addr
should be used instead of this_sector.

However, this patch doesn't make much practical different to the read
balancing due to the effects of later code.

Signed-off-by: Neil Brown <neilb@cse.unsw.edu.au>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years agoMerge master.kernel.org:/pub/scm/linux/kernel/git/sam/kbuild
Linus Torvalds [Fri, 9 Sep 2005 22:46:49 +0000 (15:46 -0700)]
Merge master.kernel.org:/pub/scm/linux/kernel/git/sam/kbuild

19 years agoRemove "must_check" attributes in PCI-land
Linus Torvalds [Fri, 9 Sep 2005 22:43:46 +0000 (15:43 -0700)]
Remove "must_check" attributes in PCI-land

Don't just irritate all other kernel developers.  Fix the users first,
then you can re-introduce the must-check infrastructure to avoid new
cases creeping in.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years agoPreempt-safe RCU file usage
Linus Torvalds [Fri, 9 Sep 2005 22:42:34 +0000 (15:42 -0700)]
Preempt-safe RCU file usage

Fix up fs/compat.c fixes.

19 years ago[PATCH] Update PCI IOMEM allocation start
Daniel Ritz [Thu, 8 Sep 2005 22:57:14 +0000 (00:57 +0200)]
[PATCH] Update PCI IOMEM allocation start

This fixes the problem with "Averatec 6240 pcmcia_socket0: unable to
apply power", which was due to the CardBus IOMEM register region being
allocated at an address that was actually inside the RAM window that had
been reserved for video frame-buffers in an UMA setup.

The BIOS _should_ have marked that region reserved in the e820 memory
descriptor tables, but did not.

It is fixed by rounding up the default starting address of PCI memory
allocations, so that we leave a bigger gap after the final known memory
location.  The amount of rounding depends on how big the unused memory
gap is that we can allocate IOMEM from.

Based on example code by Linus.

Acked-by: Greg KH <greg@kroah.com>
Acked-by: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years agoMerge master.kernel.org:/pub/scm/linux/kernel/git/gregkh/pci-2.6
Linus Torvalds [Fri, 9 Sep 2005 22:17:00 +0000 (15:17 -0700)]
Merge master.kernel.org:/pub/scm/linux/kernel/git/gregkh/pci-2.6

19 years agoFix up lost patch in compat_sys_select() for new RCU files world order
Linus Torvalds [Fri, 9 Sep 2005 22:10:52 +0000 (15:10 -0700)]
Fix up lost patch in compat_sys_select() for new RCU files world order

Andrew lost this in patch reject resolution, and never noticed, since
the compat code isn't in use on x86.

Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years agoMerge master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6
Linus Torvalds [Fri, 9 Sep 2005 21:25:22 +0000 (14:25 -0700)]
Merge master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6

19 years ago[PATCH] Lost sockfd_put() in routing_ioctl()
Kirill Korotaev [Fri, 9 Sep 2005 09:59:48 +0000 (13:59 +0400)]
[PATCH] Lost sockfd_put() in routing_ioctl()

This patch adds lost sockfd_put() in 32bit compat rounting_ioctl() on
64bit platforms

Signed-Off-By: Kirill Korotaev <dev@sw.ru>
Signed-Off-By: Maxim Giryaev <gem@sw.ru>
Signed-off-By: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] lost fput in 32bit ioctl on x86-64
Kirill Korotaev [Fri, 9 Sep 2005 20:05:53 +0000 (13:05 -0700)]
[PATCH] lost fput in 32bit ioctl on x86-64

This adds a lost fput in 32bit tiocgdev ioctl on x86-64

[ chrisw: Updated to use fget_light/fput_light ]

Signed-Off-By: Kirill Korotaev <dev@sw.ru>
Signed-Off-By: Maxim Giryaev <gem@sw.ru>
Signed-off-by: Chris Wright <chrisw@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years agokbuild: um fix so it compile with generic asm-offsets.h support
Sam Ravnborg [Fri, 9 Sep 2005 21:10:54 +0000 (23:10 +0200)]
kbuild: um fix so it compile with generic asm-offsets.h support

um has it own set of files for asm-offsets. So for now the
gen-asm-offset macro is just duplicated in the um Makefile.

This may well be the final solution since um is a bit special compared
to other architectures - time will tell.

Also added a dummy arch/um/kernel/asm-offsets.h file to keep kbuild happy.

Signed-off-by: Sam Ravnborg <sam@ravnborg.org>
19 years ago[PATCH] trivial iomem annotations in qla2xxx/qla_dbg.c
viro@ZenIV.linux.org.uk [Fri, 9 Sep 2005 20:42:25 +0000 (21:42 +0100)]
[PATCH] trivial iomem annotations in qla2xxx/qla_dbg.c

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] trivial __user annotations (md)
viro@ZenIV.linux.org.uk [Fri, 9 Sep 2005 19:36:43 +0000 (20:36 +0100)]
[PATCH] trivial __user annotations (md)

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] trivial __user annotations (evdev)
viro@ZenIV.linux.org.uk [Fri, 9 Sep 2005 19:29:12 +0000 (20:29 +0100)]
[PATCH] trivial __user annotations (evdev)

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] trivial __user cleanup (video1394)
viro@ZenIV.linux.org.uk [Fri, 9 Sep 2005 19:26:01 +0000 (20:26 +0100)]
[PATCH] trivial __user cleanup (video1394)

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] trivial __user annotations (ipmi)
viro@ZenIV.linux.org.uk [Fri, 9 Sep 2005 19:14:05 +0000 (20:14 +0100)]
[PATCH] trivial __user annotations (ipmi)

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] more SPIN_LOCK_UNLOCKED -> DEFINE_SPINLOCK conversions
Ingo Molnar [Fri, 9 Sep 2005 20:10:41 +0000 (13:10 -0700)]
[PATCH] more SPIN_LOCK_UNLOCKED -> DEFINE_SPINLOCK conversions

This converts the final 20 DEFINE_SPINLOCK holdouts.  (another 580 places
are already using DEFINE_SPINLOCK).  Build tested on x86.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] timer initialization cleanup: DEFINE_TIMER
Ingo Molnar [Fri, 9 Sep 2005 20:10:40 +0000 (13:10 -0700)]
[PATCH] timer initialization cleanup: DEFINE_TIMER

Clean up timer initialization by introducing DEFINE_TIMER a'la
DEFINE_SPINLOCK.  Build and boot-tested on x86.  A similar patch has been
been in the -RT tree for some time.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] FUSE: don't allow restarting of system calls
Miklos Szeredi [Fri, 9 Sep 2005 20:10:39 +0000 (13:10 -0700)]
[PATCH] FUSE: don't allow restarting of system calls

This patch removes ability to interrupt and restart operations while there
hasn't been any side-effect.

The reason: applications.  There are some apps it seems that generate
signals at a fast rate.  This means, that if the operation cannot make
enough progress between two signals, it will be restarted for ever.  This
bug actually manifested itself with 'krusader' trying to open a file for
writing under sshfs.  Thanks to Eduard Czimbalmos for the report.

The problem can be solved just by making open() uninterruptible, because in
this case it was the truncate operation that slowed down the progress.  But
it's better to solve this by simply not allowing interrupts at all (except
SIGKILL), because applications don't expect file operations to be
interruptible anyway.  As an added bonus the code is simplified somewhat.

Signed-off-by: Miklos Szeredi <miklos@szeredi.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] FUSE: add fsync operation for directories
Miklos Szeredi [Fri, 9 Sep 2005 20:10:38 +0000 (13:10 -0700)]
[PATCH] FUSE: add fsync operation for directories

This patch adds a new FSYNCDIR request, which is sent when fsync is called
on directories.  This operation is available in libfuse 2.3-pre1 or
greater.

Signed-off-by: Miklos Szeredi <miklos@szeredi.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] fuse: don't update file times
Miklos Szeredi [Fri, 9 Sep 2005 20:10:38 +0000 (13:10 -0700)]
[PATCH] fuse: don't update file times

Don't change mtime/ctime/atime to local time on read/write.  Rather invalidate
file attributes, so next stat() will force a GETATTR call.  Bug reported by
Ben Grimm.

Signed-off-by: Miklos Szeredi <miklos@szeredi.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] fuse: more flexible caching
Miklos Szeredi [Fri, 9 Sep 2005 20:10:37 +0000 (13:10 -0700)]
[PATCH] fuse: more flexible caching

Make data caching behavior selectable on a per-open basis instead of
per-mount.  Compatibility for the old mount options 'kernel_cache' and
'direct_io' is retained in the userspace library (version 2.4.0-pre1 or
later).

Signed-off-by: Miklos Szeredi <miklos@szeredi.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] fuse: transfer readdir data through device
Miklos Szeredi [Fri, 9 Sep 2005 20:10:36 +0000 (13:10 -0700)]
[PATCH] fuse: transfer readdir data through device

This patch removes a long lasting "hack" in FUSE, which used a separate
channel (a file descriptor refering to a disk-file) to transfer directory
contents from userspace to the kernel.

The patch adds three new operations (OPENDIR, READDIR, RELEASEDIR), which
have semantics and implementation exactly maching the respective file
operations (OPEN, READ, RELEASE).

This simplifies the directory reading code.  Also disk space is not
necessary, which can be important in embedded systems.

Signed-off-by: Miklos Szeredi <miklos@szeredi.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] FUSE - direct I/O
Miklos Szeredi [Fri, 9 Sep 2005 20:10:35 +0000 (13:10 -0700)]
[PATCH] FUSE - direct I/O

This patch adds support for the "direct_io" mount option of FUSE.

When this mount option is specified, the page cache is bypassed for
read and write operations.  This is useful for example, if the
filesystem doesn't know the size of files before reading them, or when
any kind of caching is harmful.

Signed-off-by: Miklos Szeredi <miklos@szeredi.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] fuse: stricter mount option checking
Miklos Szeredi [Fri, 9 Sep 2005 20:10:34 +0000 (13:10 -0700)]
[PATCH] fuse: stricter mount option checking

Check for the presence of all mandatory mount options.

Signed-off-by: Miklos Szeredi <miklos@szeredi.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] FUSE: tighten check for processes allowed access
Miklos Szeredi [Fri, 9 Sep 2005 20:10:34 +0000 (13:10 -0700)]
[PATCH] FUSE: tighten check for processes allowed access

This patch tightens the check for allowing processes to access non-privileged
mounts.  The rational is that the filesystem implementation can control the
behavior or get otherwise unavailable information of the filesystem user.  If
the filesystem user process has the same uid, gid, and is not suid or sgid
application, then access is safe.  Otherwise access is not allowed unless the
"allow_other" mount option is given (for which policy is controlled by the
userspace mount utility).

Thanks to everyone linux-fsdevel, especially Martin Mares who helped uncover
problems with the previous approach.

Signed-off-by: Miklos Szeredi <miklos@szeredi.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] FUSE - readpages operation
Miklos Szeredi [Fri, 9 Sep 2005 20:10:33 +0000 (13:10 -0700)]
[PATCH] FUSE - readpages operation

This patch adds readpages support to FUSE.

With the help of the readpages() operation multiple reads are bundled
together and sent as a single request to userspace.  This can improve
reading performace.

Signed-off-by: Miklos Szeredi <miklos@szeredi.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] FUSE: add padding
Miklos Szeredi [Fri, 9 Sep 2005 20:10:32 +0000 (13:10 -0700)]
[PATCH] FUSE: add padding

Add padding to structures to make sizes the same on 32bit and 64bit archs.
Initial testing and test machine generously provided by Franco Broi.

Signed-off-by: Miklos Szeredi <miklos@szeredi.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] FUSE - extended attribute operations
Miklos Szeredi [Fri, 9 Sep 2005 20:10:31 +0000 (13:10 -0700)]
[PATCH] FUSE - extended attribute operations

This patch adds the extended attribute operations to FUSE.

The following operations are added:

 o getxattr
 o setxattr
 o listxattr
 o removexattr

Signed-off-by: Miklos Szeredi <miklos@szeredi.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] FUSE - mount options
Miklos Szeredi [Fri, 9 Sep 2005 20:10:31 +0000 (13:10 -0700)]
[PATCH] FUSE - mount options

This patch adds miscellaneous mount options to the FUSE filesystem.

The following mount options are added:

 o default_permissions:  check permissions with generic_permission()
 o allow_other:          allow other users to access files
 o allow_root:           allow root to access files
 o kernel_cache:         don't invalidate page cache on open

Signed-off-by: Miklos Szeredi <miklos@szeredi.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] FUSE - file operations
Miklos Szeredi [Fri, 9 Sep 2005 20:10:30 +0000 (13:10 -0700)]
[PATCH] FUSE - file operations

This patch adds the file operations of FUSE.

The following operations are added:

 o open
 o flush
 o release
 o fsync
 o readpage
 o commit_write

Signed-off-by: Miklos Szeredi <miklos@szeredi.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] FUSE - read-write operations
Miklos Szeredi [Fri, 9 Sep 2005 20:10:29 +0000 (13:10 -0700)]
[PATCH] FUSE - read-write operations

This patch adds the write filesystem operations of FUSE.

The following operations are added:

 o setattr
 o symlink
 o mknod
 o mkdir
 o create
 o unlink
 o rmdir
 o rename
 o link

Signed-off-by: Miklos Szeredi <miklos@szeredi.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] FUSE - read-only operations
Miklos Szeredi [Fri, 9 Sep 2005 20:10:28 +0000 (13:10 -0700)]
[PATCH] FUSE - read-only operations

This patch adds the read-only filesystem operations of FUSE.

This contains the following files:

 o dir.c
    - directory, symlink and file-inode operations

The following operations are added:

 o lookup
 o getattr
 o readlink
 o follow_link
 o directory open
 o readdir
 o directory release
 o permission
 o dentry revalidate
 o statfs

Signed-off-by: Miklos Szeredi <miklos@szeredi.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] FUSE - device functions
Miklos Szeredi [Fri, 9 Sep 2005 20:10:27 +0000 (13:10 -0700)]
[PATCH] FUSE - device functions

This adds the FUSE device handling functions.

This contains the following files:

 o dev.c
    - fuse device operations (read, write, release, poll)
    - registers misc device
    - support for sending requests to userspace

Signed-off-by: Miklos Szeredi <miklos@szeredi.hu>
Signed-off-by: Adrian Bunk <bunk@stusta.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
19 years ago[PATCH] FUSE - core
Miklos Szeredi [Fri, 9 Sep 2005 20:10:26 +0000 (13:10 -0700)]
[PATCH] FUSE - core

This patch adds FUSE core.

This contains the following files:

 o inode.c
    - superblock operations (alloc_inode, destroy_inode, read_inode,
      clear_inode, put_super, show_options)
    - registers FUSE filesystem

 o fuse_i.h
    - private header file

Requirements
============

 The most important difference between orinary filesystems and FUSE is
 the fact, that the filesystem data/metadata is provided by a userspace
 process run with the privileges of the mount "owner" instead of the
 kernel, or some remote entity usually running with elevated
 privileges.

 The security implication of this is that a non-privileged user must
 not be able to use this capability to compromise the system.  Obvious
 requirements arising from this are:

  - mount owner should not be able to get elevated privileges with the
    help of the mounted filesystem

  - mount owner should not be able to induce undesired behavior in
    other users' or the super user's processes

  - mount owner should not get illegitimate access to information from
    other users' and the super user's processes

 These are currently ensured with the following constraints:

  1) mount is only allowed to directory or file which the mount owner
    can modify without limitation (write access + no sticky bit for
    directories)

  2) nosuid,nodev mount options are forced

  3) any process running with fsuid different from the owner is denied
     all access to the filesystem

 1) and 2) are ensured by the "fusermount" mount utility which is a
    setuid root application doing the actual mount operation.

 3) is ensured by a check in the permission() method in kernel

 I started thinking about doing 3) in a different way because Christoph
 H. made a big deal out of it, saying that FUSE is unacceptable into
 mainline in this form.

 The suggested use of private namespaces would be OK, but in their
 current form have many limitations that make their use impractical (as
 discussed in this thread).

 Suggested improvements that would address these limitations:

   - implement shared subtrees

   - allow a process to join an existing namespace (make namespaces
     first-class objects)

   - implement the namespace creation/joining in a PAM module

 With all that in place the check of owner against current->fsuid may
 be removed from the FUSE kernel module, without compromising the
 security requirements.

 Suid programs still interesting questions, since they get access even
 to the private namespace causing some information leak (exact
 order/timing of filesystem operations performed), giving some
 ptrace-like capabilities to unprivileged users.  BTW this problem is
 not strictly limited to the namespace approach, since suid programs
 setting fsuid and accessing users' files will succeed with the current
 approach too.

Signed-off-by: Miklos Szeredi <miklos@szeredi.hu>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>