They contain 64-bit instructions so wouldn't work on 32-bit kernels or
32-bit hardware. Since there are no users, blow them away. They
probably were only ever created because there are atomic_sub_if_positive
and atomic_dec_if_positive which exist only for sake of semaphores.
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
A "cleanup" almost two years ago deleted the old definition from
<asm/fcntl.h>, so asm-generic/fcntl.h defaulted it to the the same
value as FASYNC ... which happened to be the wrong thing.
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
The comment being removed by this patch is incorrect and misleading.
In the following situation:
1. load ...
2. store 1 -> X
3. wmb
4. rmb
5. load a <- Y
6. store ...
4 will only ensure ordering of 1 with 5.
3 will only ensure ordering of 2 with 6.
Further, a CPU with strictly in-order stores will still only provide that
2 and 6 are ordered (effectively, it is the same as a weakly ordered CPU
with wmb after every store).
In all cases, 5 may still be executed before 2 is visible to other CPUs!
The additional piece of the puzzle that mb() provides is the store/load
ordering, which fundamentally cannot be achieved with any combination of
rmb()s and wmb()s.
This can be an unexpected result if one expected any sort of global ordering
guarantee to barriers (eg. that the barriers themselves are sequentially
consistent with other types of barriers). However sfence or lfence barriers
need only provide an ordering partial ordering of memory operations -- Consider
that wmb may be implemented as nothing more than inserting a special barrier
entry in the store queue, or, in the case of x86, it can be a noop as the store
queue is in order. And an rmb may be implemented as a directive to prevent
subsequent loads only so long as their are no previous outstanding loads (while
there could be stores still in store queues).
I can actually see the occasional load/store being reordered around lfence on
my core2. That doesn't prove my above assertions, but it does show the comment
is wrong (unless my program is -- can send it out by request).
So:
mb() and smp_mb() always have and always will require a full mfence
or lock prefixed instruction on x86. And we should remove this comment.
Signed-off-by: Nick Piggin <npiggin@suse.de>
Cc: Paul McKenney <paulmck@us.ibm.com>
Cc: David Howells <dhowells@redhat.com>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Based upon a report and initial patch by Peter Lieven.
tcp4_md5sig_key and tcp6_md5sig_key need to start with
the exact same members as tcp_md5sig_key. Because they
are both cast to that type by tcp_v{4,6}_md5_do_lookup().
Unfortunately tcp{4,6}_md5sig_key use a u16 for the key
length instead of a u8, which is what tcp_md5sig_key
uses. This just so happens to work by accident on
little-endian, but on big-endian it doesn't.
Instead of casting, just place tcp_md5sig_key as the first member of
the address-family specific structures, adjust the access sites, and
kill off the ugly casts.
Signed-off-by: David S. Miller <davem@davemloft.net>
The __pa() for those did assume that all symbols have XKPHYS values and
the math fails for any other address range.
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
The __pa() for those did assume that all symbols have XKPHYS values and
the math fails for any other address range.
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
This reverts commit 184c44d204.
As noted by Dave Jones:
"Linus, please revert the above cset. It doesn't seem to be
necessary (it was added to fix a miscompile in 'make allnoconfig'
which doesn't seem to be repeatable with it reverted) and actively
breaks the ARM SA1100 framebuffer driver."
Requested-by: Dave Jones <davej@redhat.com>
Cc: Russell King <rmk+lkml@arm.linux.org.uk>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This reverts commit e66485d747, since
Rafael Wysocki noticed that the change only works for his in -mm, not in
mainline (and that both "noapictimer" _and_ "apicmaintimer" are broken
on his hardware, but that's apparently not a regression, just a symptom
of the same issue that causes the automatic apic timer disable to not
work).
It turns out that it really doesn't work correctly on x86-64, since
x86-64 doesn't use the generic clock events for timers yet.
Thanks to Rafal for testing, and here's the ugly details on x86-64 as
per Thomas:
"I just looked into the code and the logic vs. noapictimer on SMP is
completely broken.
On i386 the noapictimer option not only disables the local APIC
timer, it also registers the CPUs for broadcasting via IPI on SMP
systems.
The x86-64 code uses the broadcast only when the local apic timer is
active, i.e. "noapictimer" is not on the command line. This defeats
the whole purpose of "noapictimer". It should be there to make boxen
work, where the local APIC timer actually has a hardware problem,
e.g. the nx6325.
The current implementation of x86_64 only fixes the ACPI c-states
related problem where the APIC timer stops in C3(2), nothing else.
On nx6325 and other AMD X2 equipped systems which have the C1E
enabled we run into the following:
PIT keeps jiffies (and the system) running, but the local APIC timer
interrupts can get out of sync due to this C1E effect.
I don't think this is a critical problem, but it is wrong
nevertheless.
I think it's safe to revert the C1E patch and postpone the fix to the
clock events conversion."
On further reflection, Thomas noted:
"It's even worse than I thought on the first check:
"noapictimer" on the command line of an SMP box prevents _ONLY_ the
boot CPU apic timer from being used. But the secondary CPU is still
unconditionally setting up the APIC timer and uses the non
calibrated variable calibration_result, which is of course 0, to
setup the APIC timer. Wreckage guaranteed."
so we'll just have to wait for the x86 merge to hopefully fix this up
for x86-64.
Tested-and-requested-by: Rafael J. Wysocki <rjw@sisk.pl>
Acked-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
commit 3556ddfa92 titled
[PATCH] x86-64: Disable local APIC timer use on AMD systems with C1E
solves a problem with AMD dual core laptops e.g. HP nx6325 (Turion 64
X2) with C1E enabled:
When both cores go into idle at the same time, then the system switches
into C1E state, which is basically the same as C3. This stops the local
apic timer.
This was debugged right after the dyntick merge on i386 and despite the
patch title it fixes only the 32 bit path.
x86_64 is still missing this fix. It seems that mainline is not really
affected by this issue, as the PIT is running and keeps jiffies
incrementing, but that's just waiting for trouble.
-mm suffers from this problem due to the x86_64 high resolution timer
patches.
This is a quick and dirty port of the i386 code to x86_64.
I spent quite a time with Rafael to debug the -mm / hrt wreckage until
someone pointed us to this. I really had forgotten that we debugged this
half a year ago already.
Sigh, is it just me or is there something yelling arch/x86 into my ear?
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Rafael J. Wysocki <rjw@sisk.pl>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
It gets pointer to fastcall function, expects a pointer to normal
one and calls the sucker.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
If ADDIP is enabled, when an ASCONF chunk is received with ASCONF
paramter length set to zero, this will cause infinite loop.
By the way, if an malformed ASCONF chunk is received, will cause
processing to access memory without verifying.
This is because of not check the validity of parameters in ASCONF chunk.
This patch fixed this.
Signed-off-by: Wei Yongjun <yjwei@cn.fujitsu.com>
Signed-off-by: Vlad Yasevich <vladislav.yasevich@hp.com>
While processing OOTB chunks as well as chunks with an invalid
length of 0, it was possible to SCTP to get wedged inside an
infinite loop because we didn't catch the condition correctly,
or didn't mark the packet for discard correctly.
This work is based on original findings and work by
Wei Yongjun <yjwei@cn.fujitsu.com>
Signed-off-by: Vlad Yasevich <vladislav.yasevich@hp.com>
device_suspend() calls ACPI suspend functions, which seems to have undesired
side effects on lower idle C-states. It took me some time to realize that
especially the VAIO BIOSes (both Andrews jinxed UP and my elfstruck SMP one)
show this effect. I'm quite sure that other bug reports against suspend/resume
about turning the system into a brick have the same root cause.
After fishing in the dark for quite some time, I realized that removing the ACPI
processor module before suspend (this removes the lower C-state functionality)
made the problem disappear. Interestingly enough the propability of having a
bricked box is influenced by various factors (interrupts, size of the ram image,
...). Even adding a bunch of printks in the wrong places made the problem go
away. The previous periodic tick implementation simply pampered over the
problem, which explains why the dyntick / clockevents changes made this more
prominent.
We avoid complex functionality during the boot process and we have to do the
same during suspend/resume. It is a similar scenario and equaly fragile.
Add suspend / resume functions to the ACPI processor code and disable the lower
idle C-states across suspend/resume. Fall back to the default idle
implementation (halt) instead.
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Len Brown <lenb@kernel.org>
Cc: Venkatesh Pallipadi <venkatesh.pallipadi@intel.com>
Cc: Rafael J. Wysocki <rjw@sisk.pl>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
When compiling the Blackfin kernel, checksyscalls.pl will report lots of missing syscalls warnings.
This patch will add some missing syscalls which make sense on Blackfin arch
After appling this patch, toolchain should be rebuilt. Then recompiling the kernel with the new
toolchain.
Signed-off-by: Bryan Wu <bryan.wu@analog.com>
Add minimum support for the Blackfin relocations, since we don't have
enough space in each reloc. The idea is to store a value with one
relocation so that subsequent ones can access it.
Actually, this patch is required for Blackfin. Currently if BINFMT_FLAT is
enabled, git-tree kernel will fail to compile.
Signed-off-by: Bernd Schmidt <bernd.schmidt@analog.com>
Signed-off-by: Bryan Wu <bryan.wu@analog.com>
Cc: David McCullough <davidm@snapgear.com>
Cc: Greg Ungerer <gerg@snapgear.com>
Cc: Miles Bader <miles.bader@necel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
This reverts commit 34feb2c83b.
Suresh Siddha points out that this one breaks the fundamental
requirement that you cannot free page table pages before the TLB caches
are flushed. The quicklists do not give the same kinds of guarantees
that the mmu_gather structure does, at least not in NUMA configurations.
Requested-by: Suresh Siddha <suresh.b.siddha@intel.com>
Acked-by: Andi Kleen <ak@suse.de>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Christoph Lameter <clameter@sgi.com>
Cc: Asit Mallick <asit.k.mallick@intel.com>
Cc: Tony Luck <tony.luck@intel.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This simplifies signalfd code, by avoiding it to remain attached to the
sighand during its lifetime.
In this way, the signalfd remain attached to the sighand only during
poll(2) (and select and epoll) and read(2). This also allows to remove
all the custom "tsk == current" checks in kernel/signal.c, since
dequeue_signal() will only be called by "current".
I think this is also what Ben was suggesting time ago.
The external effect of this, is that a thread can extract only its own
private signals and the group ones. I think this is an acceptable
behaviour, in that those are the signals the thread would be able to
fetch w/out signalfd.
Signed-off-by: Davide Libenzi <davidel@xmailserver.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
add /proc/sys/kernel/sched_compat_yield to make sys_sched_yield()
more agressive, by moving the yielding task to the last position
in the rbtree.
with sched_compat_yield=0:
PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
2539 mingo 20 0 1576 252 204 R 50 0.0 0:02.03 loop_yield
2541 mingo 20 0 1576 244 196 R 50 0.0 0:02.05 loop
with sched_compat_yield=1:
PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
2584 mingo 20 0 1576 248 196 R 99 0.0 0:52.45 loop
2582 mingo 20 0 1576 256 204 R 0 0.0 0:00.00 loop_yield
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Add a workaround to address warnings generated on the "n" constraint by
GCC 3.3 and below.
Signed-off-by: Maciej W. Rozycki <macro@linux-mips.org>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
This patch proposes fixes to the reference counting of memory policy in the
page allocation paths and in show_numa_map(). Extracted from my "Memory
Policy Cleanups and Enhancements" series as stand-alone.
Shared policy lookup [shmem] has always added a reference to the policy,
but this was never unrefed after page allocation or after formatting the
numa map data.
Default system policy should not require additional ref counting, nor
should the current task's task policy. However, show_numa_map() calls
get_vma_policy() to examine what may be [likely is] another task's policy.
The latter case needs protection against freeing of the policy.
This patch adds a reference count to a mempolicy returned by
get_vma_policy() when the policy is a vma policy or another task's
mempolicy. Again, shared policy is already reference counted on lookup. A
matching "unref" [__mpol_free()] is performed in alloc_page_vma() for
shared and vma policies, and in show_numa_map() for shared and another
task's mempolicy. We can call __mpol_free() directly, saving an admittedly
inexpensive inline NULL test, because we know we have a non-NULL policy.
Handling policy ref counts for hugepages is a bit trickier.
huge_zonelist() returns a zone list that might come from a shared or vma
'BIND policy. In this case, we should hold the reference until after the
huge page allocation in dequeue_hugepage(). The patch modifies
huge_zonelist() to return a pointer to the mempolicy if it needs to be
unref'd after allocation.
Kernel Build [16cpu, 32GB, ia64] - average of 10 runs:
w/o patch w/ refcount patch
Avg Std Devn Avg Std Devn
Real: 100.59 0.38 100.63 0.43
User: 1209.60 0.37 1209.91 0.31
System: 81.52 0.42 81.64 0.34
Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Acked-by: Andi Kleen <ak@suse.de>
Cc: Christoph Lameter <clameter@sgi.com>
Acked-by: Mel Gorman <mel@csn.ul.ie>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
It turned out, that the user namespace is released during the do_exit() in
exit_task_namespaces(), but the struct user_struct is released only during the
put_task_struct(), i.e. MUCH later.
On debug kernels with poisoned slabs this will cause the oops in
uid_hash_remove() because the head of the chain, which resides inside the
struct user_namespace, will be already freed and poisoned.
Since the uid hash itself is required only when someone can search it, i.e.
when the namespace is alive, we can safely unhash all the user_struct-s from
it during the namespace exiting. The subsequent free_uid() will complete the
user_struct destruction.
For example simple program
#include <sched.h>
char stack[2 * 1024 * 1024];
int f(void *foo)
{
return 0;
}
int main(void)
{
clone(f, stack + 1 * 1024 * 1024, 0x10000000, 0);
return 0;
}
run on kernel with CONFIG_USER_NS turned on will oops the
kernel immediately.
This was spotted during OpenVZ kernel testing.
Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Alexey Dobriyan <adobriyan@openvz.org>
Acked-by: "Serge E. Hallyn" <serue@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Surprisingly, but (spotted by Alexey Dobriyan) the uid hash still uses
list_heads, thus occupying twice as much place as it could. Convert it to
hlist_heads.
Signed-off-by: Pavel Emelyanov <xemul@openvz.org>
Signed-off-by: Alexey Dobriyan <adobriyan@openvz.org>
Acked-by: Serge Hallyn <serue@us.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Recent changes to the timekeeping code broke support for the PowerPC 601
processor which doesn't have the usual timebase facility but a slightly
different thing called (yuck) the RTC.
This fixes it, boot tested on an old 601 based PowerMac 7200.
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>
When CONFIG_ISA is disabled, the isa_driver support will not be compiled
in. Define stubs so that we don't get link-time errors.
Signed-off-by: Matthew Wilcox <matthew@wil.cx>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This patch adds an optimised version of skb_cow that avoids the copy if
the header can be modified even if the rest of the payload is cloned.
This can be used in encapsulating paths where we only need to modify the
header. As it is, this can be used in PPPOE and bridging.
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
Since the sctp_sockaddr_entry is now RCU enabled as part of
the patch to synchronize sctp_localaddr_list, it makes sense to
change all handling of these entries to RCU. This includes the
sctp_bind_addrs structure and it's list of bound addresses.
This list is currently protected by an external rw_lock and that
looks like an overkill. There are only 2 writers to the list:
bind()/bindx() calls, and BH processing of ASCONF-ACK chunks.
These are already seriealized via the socket lock, so they will
not step on each other. These are also relatively rare, so we
should be good with RCU.
The readers are varied and they are easily converted to RCU.
Signed-off-by: Vlad Yasevich <vladislav.yasevich@hp.com>
Acked-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Acked-by: Sridhar Samdurala <sri@us.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
sctp_localaddr_list is modified dynamically via NETDEV_UP
and NETDEV_DOWN events, but there is not synchronization
between writer (even handler) and readers. As a result,
the readers can access an entry that has been freed and
crash the sytem.
Signed-off-by: Vlad Yasevich <vladislav.yasevich@hp.com>
Acked-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Acked-by: Sridhar Samdurala <sri@us.ibm.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
As noted by Al Viro, when we try to call prom_set_trap_table()
in the SMP trampoline code we try to take the PROM call spinlock
which doesn't work because the current thread pointer isn't
valid yet and lockdep depends upon that being correct.
Furthermore, we cannot set the current thread pointer register
because it can't be properly dereferenced until we return from
prom_set_trap_table(). Kernel TLB misses only work after that
call.
So do the PROM call to set the trap table directly instead of
going through the OBP library C code, and thus avoid the lock
altogether.
These calls are guarenteed to be serialized fully.
Since there are now no calls to the prom_set_trap_table{_sun4v}()
library functions, they can be deleted.
Signed-off-by: David S. Miller <davem@davemloft.net>
Revert b543858209 and add
no_pci_devices() check to avoid crash due to early calling of
pci_get_class().
Signed-off-by: Atsushi Nemoto <anemo@mba.ocn.ne.jp>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Commit f629307c85 introduced uses of
kernel_termios_to_user_termios_1 and user_termios_to_kernel_termios_1
on all architectures. However, powerpc, s390, avr32 and frv don't
currently define those functions since their termios struct didn't
need to be changed when the arbitrary baud rate stuff was added, and
thus the kernel won't currently build on those architectures.
This adds definitions of kernel_termios_to_user_termios_1 and
user_termios_to_kernel_termios_1 to include/asm-generic/termios.h
which are identical to kernel_termios_to_user_termios and
user_termios_to_kernel_termios respectively. The definitions are the
same because the "old" termios and "new" termios are in fact the same
on these architectures (which are the same ones that use
asm-generic/termios.h).
Signed-off-by: Paul Mackerras <paulus@samba.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Alan Cox <alan@redhat.com>
Cc: David Miller <davem@davemloft.net>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Taneli Vähäkangas <vahakang@cs.helsinki.fi> reported that commit
786d7e1612 aka "Fix rmmod/read/write races
in /proc entries" broke SBCL + SLIME combo.
The old code in do_select() used DEFAULT_POLLMASK, if couldn't find
->poll handler. The new code makes ->poll always there and returns 0 by
default, which is not correct. Return DEFAULT_POLLMASK instead.
Steps to reproduce:
install emacs, SBCL, SLIME
emacs
M-x slime in *inferior-lisp* buffer
[watch it doing "Connecting to Swank on port X.."]
Please, apply before 2.6.23.
P.S.: why SBCL can't just read(2) /proc/cpuinfo is a mystery.
Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Cc: T Taneli Vahakangas <vahakang@cs.helsinki.fi>
Cc: Oleg Nesterov <oleg@tv-sign.ru>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The AdvanSys driver wants to align some pointers, and the ALIGN macro
doesn't work for pointers. Rather than try to make it work, add a new
PTR_ALIGN macro which is typesafe.
Signed-off-by: Matthew Wilcox <matthew@wil.cx>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This patch has added #include <linux/spinlock.h> to include/linux/leds.h
for rwlock_t.
Signed-off-by: Yoichi Yuasa <yoichi_yuasa@tripeaks.co.jp>
Signed-off-by: Richard Purdie <rpurdie@rpsys.net>
Make the SATA drive detection code from eighty_ninty_three() into inline
ide_dev_is_sata() helper fixing it along the way to be more strict while
checking word 80 for the reserved values...
Signed-off-by: Sergei Shtylyov <sshtylyov@ru.mvista.com>
Signed-off-by: Bartlomiej Zolnierkiewicz <bzolnier@gmail.com>
Following a strict interpretation the empty definition of irq_enable_hazard
has always been a bug - but an intentional one because it didn't bite.
This has now changed, for uniprocessor kernels mm/slab.c:do_drain()
[...]
on_each_cpu(do_drain, cachep, 1, 1);
check_irq_on();
[...]
may be compiled into a mtc0 c0_status; mfc0 c0_status sequence resulting
in a back-to-back hazard.
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
Same as all the others, just put in the constants for the existing kernel
code and termios2 structure
Signed-off-by: Alan Cox <alan@redhat.com>
Acked-by: Ralf Baechle <ralf@linux-mips.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Ralf Baechle <ralf@linux-mips.org>
This patch adds the Intel Tolapai LPC and SMBus Controller DID's.
Signed-off-by: Jason Gaston <jason.d.gaston@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Fix warnings when CONFIG_PCIEAER=n:
drivers/pci/pcie/portdrv_pci.c:105: warning: statement with no effect
drivers/pci/pcie/portdrv_pci.c:226: warning: statement with no effect
drivers/scsi/arcmsr/arcmsr_hba.c:352: warning: statement with no effect
Signed-off-by: Randy Dunlap <randy.dunlap@oracle.com>
Acked-by: Linas Vepstas <linas@austin.ibm.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
So I've had a deadlock reported to me. I've found that the sequence of
events goes like this:
1) process A (modprobe) runs to remove ip_tables.ko
2) process B (iptables-restore) runs and calls setsockopt on a netfilter socket,
increasing the ip_tables socket_ops use count
3) process A acquires a file lock on the file ip_tables.ko, calls remove_module
in the kernel, which in turn executes the ip_tables module cleanup routine,
which calls nf_unregister_sockopt
4) nf_unregister_sockopt, seeing that the use count is non-zero, puts the
calling process into uninterruptible sleep, expecting the process using the
socket option code to wake it up when it exits the kernel
4) the user of the socket option code (process B) in do_ipt_get_ctl, calls
ipt_find_table_lock, which in this case calls request_module to load
ip_tables_nat.ko
5) request_module forks a copy of modprobe (process C) to load the module and
blocks until modprobe exits.
6) Process C. forked by request_module process the dependencies of
ip_tables_nat.ko, of which ip_tables.ko is one.
7) Process C attempts to lock the request module and all its dependencies, it
blocks when it attempts to lock ip_tables.ko (which was previously locked in
step 3)
Theres not really any great permanent solution to this that I can see, but I've
developed a two part solution that corrects the problem
Part 1) Modifies the nf_sockopt registration code so that, instead of using a
use counter internal to the nf_sockopt_ops structure, we instead use a pointer
to the registering modules owner to do module reference counting when nf_sockopt
calls a modules set/get routine. This prevents the deadlock by preventing set 4
from happening.
Part 2) Enhances the modprobe utilty so that by default it preforms non-blocking
remove operations (the same way rmmod does), and add an option to explicity
request blocking operation. So if you select blocking operation in modprobe you
can still cause the above deadlock, but only if you explicity try (and since
root can do any old stupid thing it would like.... :) ).
Signed-off-by: Neil Horman <nhorman@tuxdriver.com>
Signed-off-by: Patrick McHardy <kaber@trash.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
Michael Hennerich