2 * umh - the kernel usermode helper
4 #include <linux/module.h>
5 #include <linux/sched.h>
6 #include <linux/sched/task.h>
7 #include <linux/binfmts.h>
8 #include <linux/syscalls.h>
9 #include <linux/unistd.h>
10 #include <linux/kmod.h>
11 #include <linux/slab.h>
12 #include <linux/completion.h>
13 #include <linux/cred.h>
14 #include <linux/file.h>
15 #include <linux/fdtable.h>
16 #include <linux/workqueue.h>
17 #include <linux/security.h>
18 #include <linux/mount.h>
19 #include <linux/kernel.h>
20 #include <linux/init.h>
21 #include <linux/resource.h>
22 #include <linux/notifier.h>
23 #include <linux/suspend.h>
24 #include <linux/rwsem.h>
25 #include <linux/ptrace.h>
26 #include <linux/async.h>
27 #include <linux/uaccess.h>
28 #include <linux/shmem_fs.h>
29 #include <linux/pipe_fs_i.h>
31 #include <trace/events/module.h>
33 #define CAP_BSET (void *)1
34 #define CAP_PI (void *)2
36 static kernel_cap_t usermodehelper_bset
= CAP_FULL_SET
;
37 static kernel_cap_t usermodehelper_inheritable
= CAP_FULL_SET
;
38 static DEFINE_SPINLOCK(umh_sysctl_lock
);
39 static DECLARE_RWSEM(umhelper_sem
);
40 static LIST_HEAD(umh_list
);
41 static DEFINE_MUTEX(umh_list_lock
);
43 static void call_usermodehelper_freeinfo(struct subprocess_info
*info
)
46 (*info
->cleanup
)(info
);
50 static void umh_complete(struct subprocess_info
*sub_info
)
52 struct completion
*comp
= xchg(&sub_info
->complete
, NULL
);
54 * See call_usermodehelper_exec(). If xchg() returns NULL
55 * we own sub_info, the UMH_KILLABLE caller has gone away
56 * or the caller used UMH_NO_WAIT.
61 call_usermodehelper_freeinfo(sub_info
);
65 * This is the task which runs the usermode application
67 static int call_usermodehelper_exec_async(void *data
)
69 struct subprocess_info
*sub_info
= data
;
73 spin_lock_irq(¤t
->sighand
->siglock
);
74 flush_signal_handlers(current
, 1);
75 spin_unlock_irq(¤t
->sighand
->siglock
);
78 * Our parent (unbound workqueue) runs with elevated scheduling
79 * priority. Avoid propagating that into the userspace child.
81 set_user_nice(current
, 0);
84 new = prepare_kernel_cred(current
);
88 spin_lock(&umh_sysctl_lock
);
89 new->cap_bset
= cap_intersect(usermodehelper_bset
, new->cap_bset
);
90 new->cap_inheritable
= cap_intersect(usermodehelper_inheritable
,
91 new->cap_inheritable
);
92 spin_unlock(&umh_sysctl_lock
);
95 retval
= sub_info
->init(sub_info
, new);
104 sub_info
->pid
= task_pid_nr(current
);
105 if (sub_info
->file
) {
106 retval
= do_execve_file(sub_info
->file
,
107 sub_info
->argv
, sub_info
->envp
);
109 current
->flags
|= PF_UMH
;
111 retval
= do_execve(getname_kernel(sub_info
->path
),
112 (const char __user
*const __user
*)sub_info
->argv
,
113 (const char __user
*const __user
*)sub_info
->envp
);
115 sub_info
->retval
= retval
;
117 * call_usermodehelper_exec_sync() will call umh_complete
120 if (!(sub_info
->wait
& UMH_WAIT_PROC
))
121 umh_complete(sub_info
);
127 /* Handles UMH_WAIT_PROC. */
128 static void call_usermodehelper_exec_sync(struct subprocess_info
*sub_info
)
132 /* If SIGCLD is ignored kernel_wait4 won't populate the status. */
133 kernel_sigaction(SIGCHLD
, SIG_DFL
);
134 pid
= kernel_thread(call_usermodehelper_exec_async
, sub_info
, SIGCHLD
);
136 sub_info
->retval
= pid
;
140 * Normally it is bogus to call wait4() from in-kernel because
141 * wait4() wants to write the exit code to a userspace address.
142 * But call_usermodehelper_exec_sync() always runs as kernel
143 * thread (workqueue) and put_user() to a kernel address works
144 * OK for kernel threads, due to their having an mm_segment_t
145 * which spans the entire address space.
147 * Thus the __user pointer cast is valid here.
149 kernel_wait4(pid
, (int __user
*)&ret
, 0, NULL
);
152 * If ret is 0, either call_usermodehelper_exec_async failed and
153 * the real error code is already in sub_info->retval or
154 * sub_info->retval is 0 anyway, so don't mess with it then.
157 sub_info
->retval
= ret
;
160 /* Restore default kernel sig handler */
161 kernel_sigaction(SIGCHLD
, SIG_IGN
);
163 umh_complete(sub_info
);
167 * We need to create the usermodehelper kernel thread from a task that is affine
168 * to an optimized set of CPUs (or nohz housekeeping ones) such that they
169 * inherit a widest affinity irrespective of call_usermodehelper() callers with
170 * possibly reduced affinity (eg: per-cpu workqueues). We don't want
171 * usermodehelper targets to contend a busy CPU.
173 * Unbound workqueues provide such wide affinity and allow to block on
174 * UMH_WAIT_PROC requests without blocking pending request (up to some limit).
176 * Besides, workqueues provide the privilege level that caller might not have
177 * to perform the usermodehelper request.
180 static void call_usermodehelper_exec_work(struct work_struct
*work
)
182 struct subprocess_info
*sub_info
=
183 container_of(work
, struct subprocess_info
, work
);
185 if (sub_info
->wait
& UMH_WAIT_PROC
) {
186 call_usermodehelper_exec_sync(sub_info
);
190 * Use CLONE_PARENT to reparent it to kthreadd; we do not
191 * want to pollute current->children, and we need a parent
192 * that always ignores SIGCHLD to ensure auto-reaping.
194 pid
= kernel_thread(call_usermodehelper_exec_async
, sub_info
,
195 CLONE_PARENT
| SIGCHLD
);
197 sub_info
->retval
= pid
;
198 umh_complete(sub_info
);
204 * If set, call_usermodehelper_exec() will exit immediately returning -EBUSY
205 * (used for preventing user land processes from being created after the user
206 * land has been frozen during a system-wide hibernation or suspend operation).
207 * Should always be manipulated under umhelper_sem acquired for write.
209 static enum umh_disable_depth usermodehelper_disabled
= UMH_DISABLED
;
211 /* Number of helpers running */
212 static atomic_t running_helpers
= ATOMIC_INIT(0);
215 * Wait queue head used by usermodehelper_disable() to wait for all running
218 static DECLARE_WAIT_QUEUE_HEAD(running_helpers_waitq
);
221 * Used by usermodehelper_read_lock_wait() to wait for usermodehelper_disabled
224 static DECLARE_WAIT_QUEUE_HEAD(usermodehelper_disabled_waitq
);
227 * Time to wait for running_helpers to become zero before the setting of
228 * usermodehelper_disabled in usermodehelper_disable() fails
230 #define RUNNING_HELPERS_TIMEOUT (5 * HZ)
232 int usermodehelper_read_trylock(void)
237 down_read(&umhelper_sem
);
239 prepare_to_wait(&usermodehelper_disabled_waitq
, &wait
,
241 if (!usermodehelper_disabled
)
244 if (usermodehelper_disabled
== UMH_DISABLED
)
247 up_read(&umhelper_sem
);
255 down_read(&umhelper_sem
);
257 finish_wait(&usermodehelper_disabled_waitq
, &wait
);
260 EXPORT_SYMBOL_GPL(usermodehelper_read_trylock
);
262 long usermodehelper_read_lock_wait(long timeout
)
269 down_read(&umhelper_sem
);
271 prepare_to_wait(&usermodehelper_disabled_waitq
, &wait
,
272 TASK_UNINTERRUPTIBLE
);
273 if (!usermodehelper_disabled
)
276 up_read(&umhelper_sem
);
278 timeout
= schedule_timeout(timeout
);
282 down_read(&umhelper_sem
);
284 finish_wait(&usermodehelper_disabled_waitq
, &wait
);
287 EXPORT_SYMBOL_GPL(usermodehelper_read_lock_wait
);
289 void usermodehelper_read_unlock(void)
291 up_read(&umhelper_sem
);
293 EXPORT_SYMBOL_GPL(usermodehelper_read_unlock
);
296 * __usermodehelper_set_disable_depth - Modify usermodehelper_disabled.
297 * @depth: New value to assign to usermodehelper_disabled.
299 * Change the value of usermodehelper_disabled (under umhelper_sem locked for
300 * writing) and wakeup tasks waiting for it to change.
302 void __usermodehelper_set_disable_depth(enum umh_disable_depth depth
)
304 down_write(&umhelper_sem
);
305 usermodehelper_disabled
= depth
;
306 wake_up(&usermodehelper_disabled_waitq
);
307 up_write(&umhelper_sem
);
311 * __usermodehelper_disable - Prevent new helpers from being started.
312 * @depth: New value to assign to usermodehelper_disabled.
314 * Set usermodehelper_disabled to @depth and wait for running helpers to exit.
316 int __usermodehelper_disable(enum umh_disable_depth depth
)
323 down_write(&umhelper_sem
);
324 usermodehelper_disabled
= depth
;
325 up_write(&umhelper_sem
);
328 * From now on call_usermodehelper_exec() won't start any new
329 * helpers, so it is sufficient if running_helpers turns out to
330 * be zero at one point (it may be increased later, but that
333 retval
= wait_event_timeout(running_helpers_waitq
,
334 atomic_read(&running_helpers
) == 0,
335 RUNNING_HELPERS_TIMEOUT
);
339 __usermodehelper_set_disable_depth(UMH_ENABLED
);
343 static void helper_lock(void)
345 atomic_inc(&running_helpers
);
346 smp_mb__after_atomic();
349 static void helper_unlock(void)
351 if (atomic_dec_and_test(&running_helpers
))
352 wake_up(&running_helpers_waitq
);
356 * call_usermodehelper_setup - prepare to call a usermode helper
357 * @path: path to usermode executable
358 * @argv: arg vector for process
359 * @envp: environment for process
360 * @gfp_mask: gfp mask for memory allocation
361 * @cleanup: a cleanup function
362 * @init: an init function
363 * @data: arbitrary context sensitive data
365 * Returns either %NULL on allocation failure, or a subprocess_info
366 * structure. This should be passed to call_usermodehelper_exec to
367 * exec the process and free the structure.
369 * The init function is used to customize the helper process prior to
370 * exec. A non-zero return code causes the process to error out, exit,
371 * and return the failure to the calling process
373 * The cleanup function is just before ethe subprocess_info is about to
374 * be freed. This can be used for freeing the argv and envp. The
375 * Function must be runnable in either a process context or the
376 * context in which call_usermodehelper_exec is called.
378 struct subprocess_info
*call_usermodehelper_setup(const char *path
, char **argv
,
379 char **envp
, gfp_t gfp_mask
,
380 int (*init
)(struct subprocess_info
*info
, struct cred
*new),
381 void (*cleanup
)(struct subprocess_info
*info
),
384 struct subprocess_info
*sub_info
;
385 sub_info
= kzalloc(sizeof(struct subprocess_info
), gfp_mask
);
389 INIT_WORK(&sub_info
->work
, call_usermodehelper_exec_work
);
391 #ifdef CONFIG_STATIC_USERMODEHELPER
392 sub_info
->path
= CONFIG_STATIC_USERMODEHELPER_PATH
;
394 sub_info
->path
= path
;
396 sub_info
->argv
= argv
;
397 sub_info
->envp
= envp
;
399 sub_info
->cleanup
= cleanup
;
400 sub_info
->init
= init
;
401 sub_info
->data
= data
;
405 EXPORT_SYMBOL(call_usermodehelper_setup
);
407 struct subprocess_info
*call_usermodehelper_setup_file(struct file
*file
,
408 int (*init
)(struct subprocess_info
*info
, struct cred
*new),
409 void (*cleanup
)(struct subprocess_info
*info
), void *data
)
411 struct subprocess_info
*sub_info
;
412 struct umh_info
*info
= data
;
413 const char *cmdline
= (info
->cmdline
) ? info
->cmdline
: "usermodehelper";
415 sub_info
= kzalloc(sizeof(struct subprocess_info
), GFP_KERNEL
);
419 sub_info
->argv
= argv_split(GFP_KERNEL
, cmdline
, NULL
);
420 if (!sub_info
->argv
) {
425 INIT_WORK(&sub_info
->work
, call_usermodehelper_exec_work
);
426 sub_info
->path
= "none";
427 sub_info
->file
= file
;
428 sub_info
->init
= init
;
429 sub_info
->cleanup
= cleanup
;
430 sub_info
->data
= data
;
434 static int umh_pipe_setup(struct subprocess_info
*info
, struct cred
*new)
436 struct umh_info
*umh_info
= info
->data
;
437 struct file
*from_umh
[2];
438 struct file
*to_umh
[2];
441 /* create pipe to send data to umh */
442 err
= create_pipe_files(to_umh
, 0);
445 err
= replace_fd(0, to_umh
[0], 0);
452 /* create pipe to receive data from umh */
453 err
= create_pipe_files(from_umh
, 0);
456 replace_fd(0, NULL
, 0);
459 err
= replace_fd(1, from_umh
[1], 0);
463 replace_fd(0, NULL
, 0);
468 umh_info
->pipe_to_umh
= to_umh
[1];
469 umh_info
->pipe_from_umh
= from_umh
[0];
473 static void umh_clean_and_save_pid(struct subprocess_info
*info
)
475 struct umh_info
*umh_info
= info
->data
;
477 argv_free(info
->argv
);
478 umh_info
->pid
= info
->pid
;
482 * fork_usermode_blob - fork a blob of bytes as a usermode process
483 * @data: a blob of bytes that can be do_execv-ed as a file
484 * @len: length of the blob
485 * @info: information about usermode process (shouldn't be NULL)
487 * If info->cmdline is set it will be used as command line for the
488 * user process, else "usermodehelper" is used.
490 * Returns either negative error or zero which indicates success
491 * in executing a blob of bytes as a usermode process. In such
492 * case 'struct umh_info *info' is populated with two pipes
493 * and a pid of the process. The caller is responsible for health
494 * check of the user process, killing it via pid, and closing the
495 * pipes when user process is no longer needed.
497 int fork_usermode_blob(void *data
, size_t len
, struct umh_info
*info
)
499 struct subprocess_info
*sub_info
;
505 file
= shmem_kernel_file_setup("", len
, 0);
507 return PTR_ERR(file
);
509 written
= kernel_write(file
, data
, len
, &pos
);
510 if (written
!= len
) {
518 sub_info
= call_usermodehelper_setup_file(file
, umh_pipe_setup
,
519 umh_clean_and_save_pid
, info
);
523 err
= call_usermodehelper_exec(sub_info
, UMH_WAIT_EXEC
);
525 mutex_lock(&umh_list_lock
);
526 list_add(&info
->list
, &umh_list
);
527 mutex_unlock(&umh_list_lock
);
533 EXPORT_SYMBOL_GPL(fork_usermode_blob
);
536 * call_usermodehelper_exec - start a usermode application
537 * @sub_info: information about the subprocessa
538 * @wait: wait for the application to finish and return status.
539 * when UMH_NO_WAIT don't wait at all, but you get no useful error back
540 * when the program couldn't be exec'ed. This makes it safe to call
541 * from interrupt context.
543 * Runs a user-space application. The application is started
544 * asynchronously if wait is not set, and runs as a child of system workqueues.
545 * (ie. it runs with full root capabilities and optimized affinity).
547 int call_usermodehelper_exec(struct subprocess_info
*sub_info
, int wait
)
549 DECLARE_COMPLETION_ONSTACK(done
);
552 if (!sub_info
->path
) {
553 call_usermodehelper_freeinfo(sub_info
);
557 if (usermodehelper_disabled
) {
563 * If there is no binary for us to call, then just return and get out of
564 * here. This allows us to set STATIC_USERMODEHELPER_PATH to "" and
565 * disable all call_usermodehelper() calls.
567 if (strlen(sub_info
->path
) == 0)
571 * Set the completion pointer only if there is a waiter.
572 * This makes it possible to use umh_complete to free
573 * the data structure in case of UMH_NO_WAIT.
575 sub_info
->complete
= (wait
== UMH_NO_WAIT
) ? NULL
: &done
;
576 sub_info
->wait
= wait
;
578 queue_work(system_unbound_wq
, &sub_info
->work
);
579 if (wait
== UMH_NO_WAIT
) /* task has freed sub_info */
582 if (wait
& UMH_KILLABLE
) {
583 retval
= wait_for_completion_killable(&done
);
587 /* umh_complete() will see NULL and free sub_info */
588 if (xchg(&sub_info
->complete
, NULL
))
590 /* fallthrough, umh_complete() was already called */
593 wait_for_completion(&done
);
595 retval
= sub_info
->retval
;
597 call_usermodehelper_freeinfo(sub_info
);
602 EXPORT_SYMBOL(call_usermodehelper_exec
);
605 * call_usermodehelper() - prepare and start a usermode application
606 * @path: path to usermode executable
607 * @argv: arg vector for process
608 * @envp: environment for process
609 * @wait: wait for the application to finish and return status.
610 * when UMH_NO_WAIT don't wait at all, but you get no useful error back
611 * when the program couldn't be exec'ed. This makes it safe to call
612 * from interrupt context.
614 * This function is the equivalent to use call_usermodehelper_setup() and
615 * call_usermodehelper_exec().
617 int call_usermodehelper(const char *path
, char **argv
, char **envp
, int wait
)
619 struct subprocess_info
*info
;
620 gfp_t gfp_mask
= (wait
== UMH_NO_WAIT
) ? GFP_ATOMIC
: GFP_KERNEL
;
622 info
= call_usermodehelper_setup(path
, argv
, envp
, gfp_mask
,
627 return call_usermodehelper_exec(info
, wait
);
629 EXPORT_SYMBOL(call_usermodehelper
);
631 static int proc_cap_handler(struct ctl_table
*table
, int write
,
632 void __user
*buffer
, size_t *lenp
, loff_t
*ppos
)
635 unsigned long cap_array
[_KERNEL_CAPABILITY_U32S
];
636 kernel_cap_t new_cap
;
639 if (write
&& (!capable(CAP_SETPCAP
) ||
640 !capable(CAP_SYS_MODULE
)))
644 * convert from the global kernel_cap_t to the ulong array to print to
645 * userspace if this is a read.
647 spin_lock(&umh_sysctl_lock
);
648 for (i
= 0; i
< _KERNEL_CAPABILITY_U32S
; i
++) {
649 if (table
->data
== CAP_BSET
)
650 cap_array
[i
] = usermodehelper_bset
.cap
[i
];
651 else if (table
->data
== CAP_PI
)
652 cap_array
[i
] = usermodehelper_inheritable
.cap
[i
];
656 spin_unlock(&umh_sysctl_lock
);
662 * actually read or write and array of ulongs from userspace. Remember
663 * these are least significant 32 bits first
665 err
= proc_doulongvec_minmax(&t
, write
, buffer
, lenp
, ppos
);
670 * convert from the sysctl array of ulongs to the kernel_cap_t
671 * internal representation
673 for (i
= 0; i
< _KERNEL_CAPABILITY_U32S
; i
++)
674 new_cap
.cap
[i
] = cap_array
[i
];
677 * Drop everything not in the new_cap (but don't add things)
680 spin_lock(&umh_sysctl_lock
);
681 if (table
->data
== CAP_BSET
)
682 usermodehelper_bset
= cap_intersect(usermodehelper_bset
, new_cap
);
683 if (table
->data
== CAP_PI
)
684 usermodehelper_inheritable
= cap_intersect(usermodehelper_inheritable
, new_cap
);
685 spin_unlock(&umh_sysctl_lock
);
691 void __exit_umh(struct task_struct
*tsk
)
693 struct umh_info
*info
;
694 pid_t pid
= tsk
->pid
;
696 mutex_lock(&umh_list_lock
);
697 list_for_each_entry(info
, &umh_list
, list
) {
698 if (info
->pid
== pid
) {
699 list_del(&info
->list
);
700 mutex_unlock(&umh_list_lock
);
704 mutex_unlock(&umh_list_lock
);
711 struct ctl_table usermodehelper_table
[] = {
715 .maxlen
= _KERNEL_CAPABILITY_U32S
* sizeof(unsigned long),
717 .proc_handler
= proc_cap_handler
,
720 .procname
= "inheritable",
722 .maxlen
= _KERNEL_CAPABILITY_U32S
* sizeof(unsigned long),
724 .proc_handler
= proc_cap_handler
,