This change is meant to add a control for disabling early socket demux.
The main motivation behind this patch is to provide an option to disable
the feature as it adds an additional cost to routing that reduces overall
throughput by up to 5%. For example one of my systems went from 12.1Mpps
to 11.6 after the early socket demux was added. It looks like the reason
for the regression is that we are now having to perform two lookups, first
the one for an established socket, and then the one for the routing table.
By adding this patch and toggling the value for ip_early_demux to 0 I am
able to get back to the 12.1Mpps I was previously seeing.
[ Move local variables in ip_rcv_finish() down into the basic
block in which they are actually used. -DaveM ]
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Pull perf updates from Ingo Molnar.
* 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
ftrace: Make all inline tags also include notrace
perf: Use css_tryget() to avoid propping up css refcount
perf tools: Fix synthesizing tracepoint names from the perf.data headers
perf stat: Fix default output file
perf tools: Fix endianity swapping for adds_features bitmask
Pull two cgroup fixes from Tejun Heo:
"This containes two patches fixing a refcnt race bug during css_put().
Decrementing and checking the value weren't atomic and two tasks could
think that they both pushed the counter to zero."
* 'for-3.5-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/cgroup:
cgroups: Account for CSS_DEACT_BIAS in __css_put
cgroup: make sure that decisions in __css_put are atomic
Pull driver core and printk fixes from Greg Kroah-Hartman:
"Here are some fixes for 3.5-rc4 that resolve the kmsg problems that
people have reported showing up after the printk and kmsg changes went
into 3.5-rc1. There are also a smattering of other tiny fixes for the
extcon and hyper-v drivers that people have reported.
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>"
* tag 'driver-core-3.5-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/driver-core:
extcon: max8997: Add missing kfree for info->edev in max8997_muic_remove()
extcon: Set platform drvdata in gpio_extcon_probe() and fix irq leak
extcon: Fix wrong index in max8997_extcon_cable[]
kmsg - kmsg_dump() fix CONFIG_PRINTK=n compilation
printk: return -EINVAL if the message len is bigger than the buf size
printk: use mutex lock to stop syslog_seq from going wild
kmsg - kmsg_dump() use iterator to receive log buffer content
vme: change maintainer e-mail address
Extcon: Don't try to create duplicate link names
driver core: fixup reversed deferred probe order
printk: Fix alignment of buf causing crash on ARM EABI
Tools: hv: verify origin of netlink connector message
find_new_reaper() changes pid_ns->child_reaper, see add0d4df ("pid_ns:
zap_pid_ns_processes: fix the ->child_reaper changing").
The original reason has gone away after the previous patch, ->children
list must be empty after zap_pid_ns_processes().
However now we can not switch to init_pid_ns.child_reaper.
__unhash_process() relies on the "->child_reaper == parent" check, but
this check does not work if the last exiting task is also the child
reaper.
As Eric sugested, we can change __unhash_process() to use the parent's
pid_ns and remove this code.
Also, with this change we can move detach_pid(PIDTYPE_PID) back, where it
was before the previous fix.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Louis Rilling <louis.rilling@kerlabs.com>
Cc: Mike Galbraith <efault@gmx.de>
Acked-by: Pavel Emelyanov <xemul@parallels.com>
Tested-by: Andrew Wagin <avagin@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Today we have a twofold bug. Sometimes release_task on pid == 1 in a pid
namespace can run before other processes in a pid namespace have had
release task called. With the result that pid_ns_release_proc can be
called before the last proc_flus_task() is done using upid->ns->proc_mnt,
resulting in the use of a stale pointer. This same set of circumstances
can lead to waitpid(...) returning for a processes started with
clone(CLONE_NEWPID) before the every process in the pid namespace has
actually exited.
To fix this modify zap_pid_ns_processess wait until all other processes in
the pid namespace have exited, even EXIT_DEAD zombies.
The delay_group_leader and related tests ensure that the thread gruop
leader will be the last thread of a process group to be reaped, or to
become EXIT_DEAD and self reap. With the change to zap_pid_ns_processes
we get the guarantee that pid == 1 in a pid namespace will be the last
task that release_task is called on.
With pid == 1 being the last task to pass through release_task
pid_ns_release_proc can no longer be called too early nor can wait return
before all of the EXIT_DEAD tasks in a pid namespace have exited.
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Cc: Louis Rilling <louis.rilling@kerlabs.com>
Cc: Mike Galbraith <efault@gmx.de>
Acked-by: Pavel Emelyanov <xemul@parallels.com>
Tested-by: Andrew Wagin <avagin@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
When we fixed the race between atomic_dec and css_refcnt, we missed
the fact that css_refcnt internally subtracts CSS_DEACT_BIAS to get
the actual reference count. This can potentially cause a refcount leak
if __css_put races with cgroup_clear_css_refs.
Signed-off-by: Salman Qazi <sqazi@google.com>
Acked-by: Li Zefan <lizefan@huawei.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
An rmdir pushes css's ref count to zero. However, if the associated
directory is open at the time, the dentry ref count is non-zero. If
the fd for this directory is then passed into perf_event_open, it
does a css_get(). This bounces the ref count back up from zero. This
is a problem by itself. But what makes it turn into a crash is the
fact that we end up doing an extra dput, since we perform a dput
when css_put sees the ref count go down to zero.
css_tryget() does not fall into that trap. So, we use that instead.
Reproduction test-case for the bug:
#include <unistd.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <linux/unistd.h>
#include <linux/perf_event.h>
#include <string.h>
#include <errno.h>
#include <stdio.h>
#define PERF_FLAG_PID_CGROUP (1U << 2)
int perf_event_open(struct perf_event_attr *hw_event_uptr,
pid_t pid, int cpu, int group_fd, unsigned long flags) {
return syscall(__NR_perf_event_open,hw_event_uptr, pid, cpu,
group_fd, flags);
}
/*
* Directly poke at the perf_event bug, since it's proving hard to repro
* depending on where in the kernel tree. what moved?
*/
int main(int argc, char **argv)
{
int fd;
struct perf_event_attr attr;
memset(&attr, 0, sizeof(attr));
attr.exclude_kernel = 1;
attr.size = sizeof(attr);
mkdir("/dev/cgroup/perf_event/blah", 0777);
fd = open("/dev/cgroup/perf_event/blah", O_RDONLY);
perror("open");
rmdir("/dev/cgroup/perf_event/blah");
sleep(2);
perf_event_open(&attr, fd, 0, -1, PERF_FLAG_PID_CGROUP);
perror("perf_event_open");
close(fd);
return 0;
}
Signed-off-by: Salman Qazi <sqazi@google.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Tejun Heo <tj@kernel.org>
Link: http://lkml.kernel.org/r/20120614223108.1025.2503.stgit@dungbeetle.mtv.corp.google.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Just like what devkmsg_read() does, return -EINVAL if the message len is
bigger than the buf size, or it will trigger a segfault error.
Acked-by: Kay Sievers <kay@vrfy.org>
Acked-by: Fengguang Wu <wfg@linux.intel.com>
Signed-off-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Although syslog_seq and log_next_seq stuff are protected by logbuf_lock
spin log, it's not enough. Say we have two processes A and B, and let
syslog_seq = N, while log_next_seq = N + 1, and the two processes both
come to syslog_print at almost the same time. And No matter which
process get the spin lock first, it will increase syslog_seq by one,
then release spin lock; thus later, another process increase syslog_seq
by one again. In this case, syslog_seq is bigger than syslog_next_seq.
And latter, it would make:
wait_event_interruptiable(log_wait, syslog != log_next_seq)
don't wait any more even there is no new write comes. Thus it introduce
a infinite loop reading.
I can easily see this kind of issue by the following steps:
# cat /proc/kmsg # at meantime, I don't kill rsyslog
# So they are the two processes.
# xinit # I added drm.debug=6 in the kernel parameter line,
# so that it will produce lots of message and let that
# issue happen
It's 100% reproducable on my side. And my disk will be filled up by
/var/log/messages in a quite short time.
So, introduce a mutex_lock to stop syslog_seq from going wild just like
what devkmsg_read() does. It does fix this issue as expected.
v2: use mutex_lock_interruptiable() instead (comments from Kay)
Signed-off-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
Reviewed-by: Fengguang Wu <fengguang.wu@intel.com>
Acked-By: Kay Sievers <kay@vrfy.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
