| Commit message (Collapse) | Author | Age | Files | Lines |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
call_usermodehelper_exec_sync() does fork() + wait() with "unignored"
SIGCHLD. What we have missed is that this worker thread can have other
children previously forked by call_usermodehelper_exec_work() without
UMH_WAIT_PROC. If such a child exits in between it becomes a zombie
because auto-reaping only works if SIGCHLD is ignored, and nobody can
reap it (unless/until this worker thread exits too).
Change the !UMH_WAIT_PROC case to use CLONE_PARENT.
Note: this is only first step. All PF_KTHREAD tasks, even created by
kernel_thread() should have ->parent == kthreadd by default.
Fixes: bb304a5c6fc63d8506c ("kmod: handle UMH_WAIT_PROC from system unbound workqueue")
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The UMH_WAIT_PROC handler runs in its own thread in order to make sure
that waiting for the exec kernel thread completion won't block other
usermodehelper queued jobs.
On older workqueue implementations, worklets couldn't sleep without
blocking the rest of the queue. But now the workqueue subsystem handles
that. Khelper still had the older limitation due to its singlethread
properties but we replaced it to system unbound workqueues.
Those are affine to the current node and can block up to some number of
instances.
They are a good candidate to handle UMH_WAIT_PROC assuming that we have
enough system unbound workers to handle lots of parallel usermodehelper
jobs.
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Reviewed-by: Oleg Nesterov <oleg@redhat.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
We need to launch the usermodehelper kernel threads with the widest
affinity and this is partly why we use khelper. This workqueue has
unbound properties and thus a wide affinity inherited by all its children.
Now khelper also has special properties that we aren't much interested in:
ordered and singlethread. There is really no need about ordering as all
we do is creating kernel threads. This can be done concurrently. And
singlethread is a useless limitation as well.
The workqueue engine already proposes generic unbound workqueues that
don't share these useless properties and handle well parallel jobs.
The only worrysome specific is their affinity to the node of the current
CPU. It's fine for creating the usermodehelper kernel threads but those
inherit this affinity for longer jobs such as requesting modules.
This patch proposes to use these node affine unbound workqueues assuming
that a node is sufficient to handle several parallel usermodehelper
requests.
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Reviewed-by: Oleg Nesterov <oleg@redhat.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
There seem to be quite some confusions on the comments, likely due to
changes that came after them.
Now since it's very non obvious why we have 3 levels of asynchronous code
to implement usermodehelpers, it's important to comment in detail the
reason of this layout.
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Reviewed-by: Oleg Nesterov <oleg@redhat.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Khelper is affine to all CPUs. Now since it creates the
call_usermodehelper_exec_[a]sync() kernel threads, those inherit the wide
affinity.
As such explicitly forcing a wide affinity from those kernel threads
is like a no-op.
Just remove it. It's needless and it breaks CPU isolation users who
rely on workqueue affinity tuning.
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Reviewed-by: Oleg Nesterov <oleg@redhat.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This patchset does a bunch of cleanups and converts khelper to use system
unbound workqueues. The 3 first patches should be uncontroversial. The
last 2 patches are debatable.
Kmod creates kernel threads that perform userspace jobs and we want those
to have a large affinity in order not to contend busy CPUs. This is
(partly) why we use khelper which has a wide affinity that the kernel
threads it create can inherit from. Now khelper is a dedicated workqueue
that has singlethread properties which we aren't interested in.
Hence those two debatable changes:
_ We would like to use generic workqueues. System unbound workqueues are
a very good candidate but they are not wide affine, only node affine.
Now probably a node is enough to perform many parallel kmod jobs.
_ We would like to remove the wait_for_helper kernel thread (UMH_WAIT_PROC
handler) to use the workqueue. It means that if the workqueue blocks,
and no other worker can take pending kmod request, we can be screwed.
Now if we have 512 threads, this should be enough.
This patch (of 5):
Underscores on function names aren't much verbose to explain the purpose
of a function. And kmod has interesting such flavours.
Lets rename the following functions:
* __call_usermodehelper -> call_usermodehelper_exec_work
* ____call_usermodehelper -> call_usermodehelper_exec_async
* wait_for_helper -> call_usermodehelper_exec_sync
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Reviewed-by: Oleg Nesterov <oleg@redhat.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
If request_module() successfully runs modprobe, but modprobe exits with a
non-zero status, then the return value from request_module() will be that
(positive) error status. So the return from request_module can be:
negative errno
zero for success
positive exit code.
Signed-off-by: NeilBrown <neilb@suse.com>
Cc: Goldwyn Rodrigues <rgoldwyn@suse.de>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Now that we do not call kernel_thread(CLONE_VFORK) from the worker
thread we can not deadlock if do_execve() in turn triggers another
call_usermodehelper(), we can remove the kmod_thread_locker code.
Note: we should probably kill khelper_wq and simply use one of the
global workqueues, say, system_unbound_wq, this special wq for umh buys
nothing nowadays.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
After "kernel/kmod: fix use-after-free of the sub_infostructure"
CLONE_VFORK in __call_usermodehelper() buys nothing, we rely on on
umh_complete() in ____call_usermodehelper() anyway.
Remove it. This also eliminates the unnecessary sleep/wakeup in the
likely case, and this allows the next change.
While at it, kill the "int wait" locals in ____call_usermodehelper() and
__call_usermodehelper(), they can safely use sub_info->wait.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Found this in the message log on a s390 system:
BUG kmalloc-192 (Not tainted): Poison overwritten
Disabling lock debugging due to kernel taint
INFO: 0x00000000684761f4-0x00000000684761f7. First byte 0xff instead of 0x6b
INFO: Allocated in call_usermodehelper_setup+0x70/0x128 age=71 cpu=2 pid=648
__slab_alloc.isra.47.constprop.56+0x5f6/0x658
kmem_cache_alloc_trace+0x106/0x408
call_usermodehelper_setup+0x70/0x128
call_usermodehelper+0x62/0x90
cgroup_release_agent+0x178/0x1c0
process_one_work+0x36e/0x680
worker_thread+0x2f0/0x4f8
kthread+0x10a/0x120
kernel_thread_starter+0x6/0xc
kernel_thread_starter+0x0/0xc
INFO: Freed in call_usermodehelper_exec+0x110/0x1b8 age=71 cpu=2 pid=648
__slab_free+0x94/0x560
kfree+0x364/0x3e0
call_usermodehelper_exec+0x110/0x1b8
cgroup_release_agent+0x178/0x1c0
process_one_work+0x36e/0x680
worker_thread+0x2f0/0x4f8
kthread+0x10a/0x120
kernel_thread_starter+0x6/0xc
kernel_thread_starter+0x0/0xc
There is a use-after-free bug on the subprocess_info structure allocated
by the user mode helper. In case do_execve() returns with an error
____call_usermodehelper() stores the error code to sub_info->retval, but
sub_info can already have been freed.
Regarding UMH_NO_WAIT, the sub_info structure can be freed by
__call_usermodehelper() before the worker thread returns from
do_execve(), allowing memory corruption when do_execve() failed after
exec_mmap() is called.
Regarding UMH_WAIT_EXEC, the call to umh_complete() allows
call_usermodehelper_exec() to continue which then frees sub_info.
To fix this race the code needs to make sure that the call to
call_usermodehelper_freeinfo() is always done after the last store to
sub_info->retval.
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Reviewed-by: Oleg Nesterov <oleg@redhat.com>
Cc: Tetsuo Handa <penguin-kernel@i-love.sakura.ne.jp>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Now that we have kernel_sigaction() we can change wait_for_helper() to
use it and cleans up the code a bit.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Al Viro <viro@ZenIV.linux.org.uk>
Cc: David Woodhouse <dwmw2@infradead.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Cc: Richard Weinberger <richard@nod.at>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Tejun Heo <tj@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
|
|
|
|
|
|
|
|
|
| |
Mostly scripted conversion of the smp_mb__* barriers.
Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Acked-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Link: http://lkml.kernel.org/n/tip-55dhyhocezdw1dg7u19hmh1u@git.kernel.org
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: linux-arch@vger.kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This changes 'do_execve()' to get the executable name as a 'struct
filename', and to free it when it is done. This is what the normal
users want, and it simplifies and streamlines their error handling.
The controlled lifetime of the executable name also fixes a
use-after-free problem with the trace_sched_process_exec tracepoint: the
lifetime of the passed-in string for kernel users was not at all
obvious, and the user-mode helper code used UMH_WAIT_EXEC to serialize
the pathname allocation lifetime with the execve() having finished,
which in turn meant that the trace point that happened after
mm_release() of the old process VM ended up using already free'd memory.
To solve the kernel string lifetime issue, this simply introduces
"getname_kernel()" that works like the normal user-space getname()
function, except with the source coming from kernel memory.
As Oleg points out, this also means that we could drop the tcomm[] array
from 'struct linux_binprm', since the pathname lifetime now covers
setup_new_exec(). That would be a separate cleanup.
Reported-by: Igor Zhbanov <i.zhbanov@samsung.com>
Tested-by: Steven Rostedt <rostedt@goodmis.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
If /proc/sys/kernel/core_pattern contains only "|", a NULL pointer
dereference happens upon core dump because argv_split("") returns
argv[0] == NULL.
This bug was once fixed by commit 264b83c07a84 ("usermodehelper: check
subprocess_info->path != NULL") but was by error reintroduced by commit
7f57cfa4e2aa ("usermodehelper: kill the sub_info->path[0] check").
This bug seems to exist since 2.6.19 (the version which core dump to
pipe was added). Depending on kernel version and config, some side
effect might happen immediately after this oops (e.g. kernel panic with
2.6.32-358.18.1.el6).
Signed-off-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Acked-by: Oleg Nesterov <oleg@redhat.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
call_usermodehelper_exec() does nothing but returns success if path[0] ==
0. The only user which needs this strange feature is request_module(), it
can check modprobe_path[0] itself like other users do if they want to
detect the "disabled by admin" case.
Kill it. Not only it looks strange, it can confuse other callers. And
this allows us to revert 264b83c0 ("usermodehelper: check
subprocess_info->path != NULL"), do_execve(NULL) is safe.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Rusty Russell <rusty@rustcorp.com.au>
Cc: Lucas De Marchi <lucas.de.marchi@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
argv_split(empty_or_all_spaces) happily succeeds, it simply returns
argc == 0 and argv[0] == NULL. Change call_usermodehelper_exec() to
check sub_info->path != NULL to avoid the crash.
This is the minimal fix, todo:
- perhaps we should change argv_split() to return NULL or change the
callers.
- kill or justify ->path[0] check
- narrow the scope of helper_lock()
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-By: Lucas De Marchi <lucas.demarchi@intel.com>
Cc: stable@vger.kernel.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This function suffers from not being able to determine if the cleanup is
called in case it returns -ENOMEM. Nobody is using it anymore, so let's
remove it.
Signed-off-by: Lucas De Marchi <lucas.demarchi@profusion.mobi>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: David Howells <dhowells@redhat.com>
Cc: James Morris <james.l.morris@oracle.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Tejun Heo <tj@kernel.org>
Cc: "Rafael J. Wysocki" <rjw@sisk.pl>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Use call_usermodehelper_setup() + call_usermodehelper_exec() instead of
calling call_usermodehelper_fns(). In case the latter returns -ENOMEM the
cleanup function may had not been called - in this case we would not free
argv and module_name.
Signed-off-by: Lucas De Marchi <lucas.demarchi@profusion.mobi>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: David Howells <dhowells@redhat.com>
Cc: James Morris <james.l.morris@oracle.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Tejun Heo <tj@kernel.org>
Cc: "Rafael J. Wysocki" <rjw@sisk.pl>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
call_usermodehelper_setup()
call_usermodehelper_setup() + call_usermodehelper_exec() need to be
called instead of call_usermodehelper_fns() when the cleanup function
needs to be called even when an ENOMEM error occurs. In this case using
call_usermodehelper_fns() the user can't distinguish if the cleanup
function was called or not.
[akpm@linux-foundation.org: export call_usermodehelper_setup() to modules]
Signed-off-by: Lucas De Marchi <lucas.demarchi@profusion.mobi>
Reviewed-by: Oleg Nesterov <oleg@redhat.com>
Cc: David Howells <dhowells@redhat.com>
Cc: James Morris <james.l.morris@oracle.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Tejun Heo <tj@kernel.org>
Cc: "Rafael J. Wysocki" <rjw@sisk.pl>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|\
| |
| |
| |
| |
| |
| | |
To receive f56c3196f251012de9b3ebaff55732a9074fdaae ("async: fix
__lowest_in_progress()").
Signed-off-by: Tejun Heo <tj@kernel.org>
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
All architectures have
CONFIG_GENERIC_KERNEL_THREAD
CONFIG_GENERIC_KERNEL_EXECVE
__ARCH_WANT_SYS_EXECVE
None of them have __ARCH_WANT_KERNEL_EXECVE and there are only two callers
of kernel_execve() (which is a trivial wrapper for do_execve() now) left.
Kill the conditionals and make both callers use do_execve().
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
|
|/
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Synchronous requet_module() from an async worker can lead to deadlock
because module init path may invoke async_synchronize_full(). The
async worker waits for request_module() to complete and the module
loading waits for the async task to finish. This bug happened in the
block layer because of default elevator auto-loading.
Block layer has been updated not to do default elevator auto-loading
and it has been decided to disallow synchronous request_module() from
async workers.
Trigger WARN_ON_ONCE() on synchronous request_module() from async
workers.
For more details, please refer to the following thread.
http://thread.gmane.org/gmane.linux.kernel/1420814
Signed-off-by: Tejun Heo <tj@kernel.org>
Reported-by: Alex Riesen <raa.lkml@gmail.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Arjan van de Ven <arjan@linux.intel.com>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
* allow kernel_execve() leave the actual return to userland to
caller (selected by CONFIG_GENERIC_KERNEL_EXECVE). Callers
updated accordingly.
* architecture that does select GENERIC_KERNEL_EXECVE in its
Kconfig should have its ret_from_kernel_thread() do this:
call schedule_tail
call the callback left for it by copy_thread(); if it ever
returns, that's because it has just done successful kernel_execve()
jump to return from syscall
IOW, its only difference from ret_from_fork() is that it does call the
callback.
* such an architecture should also get rid of ret_from_kernel_execve()
and __ARCH_WANT_KERNEL_EXECVE
This is the last part of infrastructure patches in that area - from
that point on work on different architectures can live independently.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Most of them never returned anyway - only two functions had to be
changed. That allows to simplify their callers a whole lot.
Note that this does *not* apply to kthread_run() callbacks - all of
those had been called from the same kernel_thread() callback, which
did do_exit() already. This is strictly about very few low-level
kernel_thread() callbacks (there are only 6 of those, mostly as part
of kthread.h and kmod.h exported mechanisms, plus kernel_init()
itself).
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The system deadlocks (at least since 2.6.10) when
call_usermodehelper(UMH_WAIT_EXEC) request triggers
call_usermodehelper(UMH_WAIT_PROC) request.
This is because "khelper thread is waiting for the worker thread at
wait_for_completion() in do_fork() since the worker thread was created
with CLONE_VFORK flag" and "the worker thread cannot call complete()
because do_execve() is blocked at UMH_WAIT_PROC request" and "the khelper
thread cannot start processing UMH_WAIT_PROC request because the khelper
thread is waiting for the worker thread at wait_for_completion() in
do_fork()".
The easiest example to observe this deadlock is to use a corrupted
/sbin/hotplug binary (like shown below).
# : > /tmp/dummy
# chmod 755 /tmp/dummy
# echo /tmp/dummy > /proc/sys/kernel/hotplug
# modprobe whatever
call_usermodehelper("/tmp/dummy", UMH_WAIT_EXEC) is called from
kobject_uevent_env() in lib/kobject_uevent.c upon loading/unloading a
module. do_execve("/tmp/dummy") triggers a call to
request_module("binfmt-0000") from search_binary_handler() which in turn
calls call_usermodehelper(UMH_WAIT_PROC).
In order to avoid deadlock, as a for-now and easy-to-backport solution, do
not try to call wait_for_completion() in call_usermodehelper_exec() if the
worker thread was created by khelper thread with CLONE_VFORK flag. Future
and fundamental solution might be replacing singleton khelper thread with
some workqueue so that recursive calls up to max_active dependency loop
can be handled without deadlock.
[akpm@linux-foundation.org: add comment to kmod_thread_locker]
Signed-off-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Cc: Arjan van de Ven <arjan@linux.intel.com>
Acked-by: Rusty Russell <rusty@rustcorp.com.au>
Cc: Tejun Heo <tj@kernel.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This function's interface is, uh, subtle. Attempt to apologise for it.
Cc: WANG Cong <xiyou.wangcong@gmail.com>
Cc: Cyrill Gorcunov <gorcunov@openvz.org>
Cc: Kees Cook <keescook@chromium.org>
Cc: Serge Hallyn <serge.hallyn@canonical.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Alan Cox <alan@lxorguk.ukuu.org.uk>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
|
|
|
|
|
|
| |
Warning(kernel/kmod.c:419): No description found for parameter 'depth'
Signed-off-by: Randy Dunlap <rdunlap@xenotime.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
If we move call_usermodehelper_fns() to kmod.c file and EXPORT_SYMBOL it
we can avoid exporting all it's helper functions:
call_usermodehelper_setup
call_usermodehelper_setfns
call_usermodehelper_exec
And make all of them static to kmod.c
Since the optimizer will see all these as a single call site it will
inline them inside call_usermodehelper_fns(). So we loose the call to
_fns but gain 3 calls to the helpers. (Not that it matters)
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
call_usermodehelper_freeinfo() is not used outside of kmod.c. So unexport
it, and make it static to kmod.c
Signed-off-by: Boaz Harrosh <bharrosh@panasas.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
There is a race condition between the freezer and request_firmware()
such that if request_firmware() is run on one CPU and
freeze_processes() is run on another CPU and usermodehelper_disable()
called by it succeeds to grab umhelper_sem for writing before
usermodehelper_read_trylock() called from request_firmware()
acquires it for reading, the request_firmware() will fail and
trigger a WARN_ON() complaining that it was called at a wrong time.
However, in fact, it wasn't called at a wrong time and
freeze_processes() simply happened to be executed simultaneously.
To avoid this race, at least in some cases, modify
usermodehelper_read_trylock() so that it doesn't fail if the
freezing of tasks has just started and hasn't been completed yet.
Instead, during the freezing of tasks, it will try to freeze the
task that has called it so that it can wait until user space is
thawed without triggering the scary warning.
For this purpose, change usermodehelper_disabled so that it can
take three different values, UMH_ENABLED (0), UMH_FREEZING and
UMH_DISABLED. The first one means that usermode helpers are
enabled, the last one means "hard disable" (i.e. the system is not
ready for usermode helpers to be used) and the second one
is reserved for the freezer. Namely, when freeze_processes() is
started, it sets usermodehelper_disabled to UMH_FREEZING which
tells usermodehelper_read_trylock() that it shouldn't fail just
yet and should call try_to_freeze() if woken up and cannot
return immediately. This way all freezable tasks that happen
to call request_firmware() right before freeze_processes() is
started and lose the race for umhelper_sem with it will be
frozen and will sleep until thaw_processes() unsets
usermodehelper_disabled. [For the non-freezable callers of
request_firmware() the race for umhelper_sem against
freeze_processes() is unfortunately unavoidable.]
Reported-by: Stephen Boyd <sboyd@codeaurora.org>
Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Acked-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: stable@vger.kernel.org
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
If firmware is requested asynchronously, by calling
request_firmware_nowait(), there is no reason to fail the request
(and warn the user) when the system is (presumably temporarily)
unready to handle it (because user space is not available yet or
frozen). For this reason, introduce an alternative routine for
read-locking umhelper_sem, usermodehelper_read_lock_wait(), that
will wait for usermodehelper_disabled to be unset (possibly with
a timeout) and make request_firmware_work_func() use it instead of
usermodehelper_read_trylock().
Accordingly, modify request_firmware() so that it uses
usermodehelper_read_trylock() to acquire umhelper_sem and remove
the code related to that lock from _request_firmware().
Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Acked-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: stable@vger.kernel.org
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Instead of two functions, read_lock_usermodehelper() and
usermodehelper_is_disabled(), used in combination, introduce
usermodehelper_read_trylock() that will only return with umhelper_sem
held if usermodehelper_disabled is unset (and will return -EAGAIN
otherwise) and make _request_firmware() use it.
Rename read_unlock_usermodehelper() to
usermodehelper_read_unlock() to follow the naming convention of the
new function.
Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Acked-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: stable@vger.kernel.org
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
As Tetsuo Handa pointed out, request_module() can stress the system
while the oom-killed caller sleeps in TASK_UNINTERRUPTIBLE.
The task T uses "almost all" memory, then it does something which
triggers request_module(). Say, it can simply call sys_socket(). This
in turn needs more memory and leads to OOM. oom-killer correctly
chooses T and kills it, but this can't help because it sleeps in
TASK_UNINTERRUPTIBLE and after that oom-killer becomes "disabled" by the
TIF_MEMDIE task T.
Make __request_module() killable. The only necessary change is that
call_modprobe() should kmalloc argv and module_name, they can't live in
the stack if we use UMH_KILLABLE. This memory is freed via
call_usermodehelper_freeinfo()->cleanup.
Reported-by: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Tejun Heo <tj@kernel.org>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
No functional changes. Move the call_usermodehelper code from
__request_module() into the new simple helper, call_modprobe().
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Tejun Heo <tj@kernel.org>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Minor cleanup. ____call_usermodehelper() can simply return, no need to
call do_exit() explicitely.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Tejun Heo <tj@kernel.org>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
No functional changes. It is not sane to use UMH_KILLABLE with enum
umh_wait, but obviously we do not want another argument in
call_usermodehelper_* helpers. Kill this enum, use the plain int.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Tejun Heo <tj@kernel.org>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Implement UMH_KILLABLE, should be used along with UMH_WAIT_EXEC/PROC.
The caller must ensure that subprocess_info->path/etc can not go away
until call_usermodehelper_freeinfo().
call_usermodehelper_exec(UMH_KILLABLE) does
wait_for_completion_killable. If it fails, it uses
xchg(&sub_info->complete, NULL) to serialize with umh_complete() which
does the same xhcg() to access sub_info->complete.
If call_usermodehelper_exec wins, it can safely return. umh_complete()
should get NULL and call call_usermodehelper_freeinfo().
Otherwise we know that umh_complete() was already called, in this case
call_usermodehelper_exec() falls back to wait_for_completion() which
should succeed "very soon".
Note: UMH_NO_WAIT == -1 but it obviously should not be used with
UMH_KILLABLE. We delay the neccessary cleanup to simplify the back
porting.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Tejun Heo <tj@kernel.org>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Preparation. Add the new trivial helper, umh_complete(). Currently it
simply does complete(sub_info->complete).
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Tejun Heo <tj@kernel.org>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|\
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
* pm-sleep: (51 commits)
PM: Drop generic_subsys_pm_ops
PM / Sleep: Remove forward-only callbacks from AMBA bus type
PM / Sleep: Remove forward-only callbacks from platform bus type
PM: Run the driver callback directly if the subsystem one is not there
PM / Sleep: Make pm_op() and pm_noirq_op() return callback pointers
PM / Sleep: Merge internal functions in generic_ops.c
PM / Sleep: Simplify generic system suspend callbacks
PM / Hibernate: Remove deprecated hibernation snapshot ioctls
PM / Sleep: Fix freezer failures due to racy usermodehelper_is_disabled()
PM / Sleep: Recommend [un]lock_system_sleep() over using pm_mutex directly
PM / Sleep: Replace mutex_[un]lock(&pm_mutex) with [un]lock_system_sleep()
PM / Sleep: Make [un]lock_system_sleep() generic
PM / Sleep: Use the freezer_count() functions in [un]lock_system_sleep() APIs
PM / Freezer: Remove the "userspace only" constraint from freezer[_do_not]_count()
PM / Hibernate: Replace unintuitive 'if' condition in kernel/power/user.c with 'else'
Freezer / sunrpc / NFS: don't allow TASK_KILLABLE sleeps to block the freezer
PM / Sleep: Unify diagnostic messages from device suspend/resume
ACPI / PM: Do not save/restore NVS on Asus K54C/K54HR
PM / Hibernate: Remove deprecated hibernation test modes
PM / Hibernate: Thaw processes in SNAPSHOT_CREATE_IMAGE ioctl test path
...
Conflicts:
kernel/kmod.c
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
Commit a144c6a (PM: Print a warning if firmware is requested when tasks
are frozen) introduced usermodehelper_is_disabled() to warn and exit
immediately if firmware is requested when usermodehelpers are disabled.
However, it is racy. Consider the following scenario, currently used in
drivers/base/firmware_class.c:
...
if (usermodehelper_is_disabled())
goto out;
/* Do actual work */
...
out:
return err;
Nothing prevents someone from disabling usermodehelpers just after the check
in the 'if' condition, which means that it is quite possible to try doing the
"actual work" with usermodehelpers disabled, leading to undesirable
consequences.
In particular, this race condition in _request_firmware() causes task freezing
failures whenever suspend/hibernation is in progress because, it wrongly waits
to get the firmware/microcode image from userspace when actually the
usermodehelpers are disabled or userspace has been frozen.
Some of the example scenarios that cause freezing failures due to this race
are those that depend on userspace via request_firmware(), such as x86
microcode module initialization and microcode image reload.
Previous discussions about this issue can be found at:
http://thread.gmane.org/gmane.linux.kernel/1198291/focus=1200591
This patch adds proper synchronization to fix this issue.
It is to be noted that this patchset fixes the freezing failures but doesn't
remove the warnings. IOW, it does not attempt to add explicit synchronization
to x86 microcode driver to avoid requesting microcode image at inopportune
moments. Because, the warnings were introduced to highlight such cases, in the
first place. And we need not silence the warnings, since we take care of the
*real* problem (freezing failure) and hence, after that, the warnings are
pretty harmless anyway.
Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
|
|/
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
usermodehelper_pm_callback() no longer exists in the kernel. There are 2
comments in kernel/kmod.c that still refer to it.
Also, the patch that introduced usermodehelper_pm_callback(), #included
two header files: <linux/notifier.h> and <linux/suspend.h>. But these are
no longer necessary.
This patch updates the comments as appropriate and removes the unnecessary
header file inclusions.
Signed-off-by: Srivatsa S. Bhat <srivatsa.bhat@linux.vnet.ibm.com>
Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Due to post-increment in condition of kmod_loop_msg in __request_module(),
the system log can be spammed by much more than 5 instances of the 'runaway
loop' message if the number of events triggering it makes the kmod_loop_msg
to overflow.
Fix that by making sure we never increment it past the threshold.
Signed-off-by: Jiri Kosina <jkosina@suse.cz>
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
CC: stable@kernel.org
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The core device layer sends tons of uevent notifications for each device
it finds, and if the kernel has been built with a non-empty
CONFIG_UEVENT_HELPER_PATH that will make us try to execute the usermode
helper binary for all these events very early in the boot.
Not only won't the root filesystem even be mounted at that point, we
literally won't have necessarily even initialized all the process
handling data structures at that point, which causes no end of silly
problems even when the usermode helper doesn't actually succeed in
executing.
So just use our existing infrastructure to disable the usermodehelpers
to make the kernel start out with them disabled. We enable them when
we've at least initialized stuff a bit.
Problems related to an uninitialized
init_ipc_ns.ids[IPC_SHM_IDS].rw_mutex
reported by various people.
Reported-by: Manuel Lauss <manuel.lauss@googlemail.com>
Reported-by: Richard Weinberger <richard@nod.at>
Reported-by: Marc Zyngier <maz@misterjones.org>
Acked-by: Kay Sievers <kay.sievers@vrfy.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Vasiliy Kulikov <segoon@openwall.com>
Cc: Greg KH <greg@kroah.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
____call_usermodehelper() now erases any credentials set by the
subprocess_inf::init() function. The problem is that commit
17f60a7da150 ("capabilites: allow the application of capability limits
to usermode helpers") creates and commits new credentials with
prepare_kernel_cred() after the call to the init() function. This wipes
all keyrings after umh_keys_init() is called.
The best way to deal with this is to put the init() call just prior to
the commit_creds() call, and pass the cred pointer to init(). That
means that umh_keys_init() and suchlike can modify the credentials
_before_ they are published and potentially in use by the rest of the
system.
This prevents request_key() from working as it is prevented from passing
the session keyring it set up with the authorisation token to
/sbin/request-key, and so the latter can't assume the authority to
instantiate the key. This causes the in-kernel DNS resolver to fail
with ENOKEY unconditionally.
Signed-off-by: David Howells <dhowells@redhat.com>
Acked-by: Eric Paris <eparis@redhat.com>
Tested-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|\ |
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
There is no way to limit the capabilities of usermodehelpers. This problem
reared its head recently when someone complained that any user with
cap_net_admin was able to load arbitrary kernel modules, even though the user
didn't have cap_sys_module. The reason is because the actual load is done by
a usermode helper and those always have the full cap set. This patch addes new
sysctls which allow us to bound the permissions of usermode helpers.
/proc/sys/kernel/usermodehelper/bset
/proc/sys/kernel/usermodehelper/inheritable
You must have CAP_SYS_MODULE and CAP_SETPCAP to change these (changes are
&= ONLY). When the kernel launches a usermodehelper it will do so with these
as the bset and pI.
-v2: make globals static
create spinlock to protect globals
-v3: require both CAP_SETPCAP and CAP_SYS_MODULE
-v4: fix the typo s/CAP_SET_PCAP/CAP_SETPCAP/ because I didn't commit
Signed-off-by: Eric Paris <eparis@redhat.com>
No-objection-from: Serge E. Hallyn <serge.hallyn@canonical.com>
Acked-by: David Howells <dhowells@redhat.com>
Acked-by: Serge E. Hallyn <serge.hallyn@canonical.com>
Acked-by: Andrew G. Morgan <morgan@kernel.org>
Signed-off-by: James Morris <jmorris@namei.org>
|
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
We need to prevent kernel-forked processes during system poweroff.
Such processes try to access the filesystem whose disks we are
trying to shutdown at the same time. This causes delays and exceptions
in the storage drivers.
A follow-up patch will add these calls and need usermodehelper_disable()
also on systems without suspend support.
Signed-off-by: Kay Sievers <kay.sievers@vrfy.org>
Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
|
|/
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Some drivers erroneously use request_firmware() from their ->resume()
(or ->thaw(), or ->restore()) callbacks, which is not going to work
unless the firmware has been built in. This causes system resume to
stall until the firmware-loading timeout expires, which makes users
think that the resume has failed and reboot their machines
unnecessarily. For this reason, make _request_firmware() print a
warning and return immediately with error code if it has been called
when tasks are frozen and it's impossible to start any new usermode
helpers.
Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Acked-by: Greg Kroah-Hartman <gregkh@suse.de>
Reviewed-by: Valdis Kletnieks <valdis.kletnieks@vt.edu>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Make do_execve() take a const filename pointer so that kernel_execve() compiles
correctly on ARM:
arch/arm/kernel/sys_arm.c:88: warning: passing argument 1 of 'do_execve' discards qualifiers from pointer target type
This also requires the argv and envp arguments to be consted twice, once for
the pointer array and once for the strings the array points to. This is
because do_execve() passes a pointer to the filename (now const) to
copy_strings_kernel(). A simpler alternative would be to cast the filename
pointer in do_execve() when it's passed to copy_strings_kernel().
do_execve() may not change any of the strings it is passed as part of the argv
or envp lists as they are some of them in .rodata, so marking these strings as
const should be fine.
Further kernel_execve() and sys_execve() need to be changed to match.
This has been test built on x86_64, frv, arm and mips.
Signed-off-by: David Howells <dhowells@redhat.com>
Tested-by: Ralf Baechle <ralf@linux-mips.org>
Acked-by: Russell King <rmk+kernel@arm.linux.org.uk>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|
|
|
|
|
|
|
|
|
| |
UMH_WAIT_EXEC should report the error if kernel_thread() fails, like
UMH_WAIT_PROC does.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
|