1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * umh - the kernel usermode helper
4 */
5#include <linux/module.h>
6#include <linux/sched.h>
7#include <linux/sched/task.h>
8#include <linux/binfmts.h>
9#include <linux/syscalls.h>
10#include <linux/unistd.h>
11#include <linux/kmod.h>
12#include <linux/slab.h>
13#include <linux/completion.h>
14#include <linux/cred.h>
15#include <linux/file.h>
16#include <linux/fdtable.h>
17#include <linux/fs_struct.h>
18#include <linux/workqueue.h>
19#include <linux/security.h>
20#include <linux/mount.h>
21#include <linux/kernel.h>
22#include <linux/init.h>
23#include <linux/resource.h>
24#include <linux/notifier.h>
25#include <linux/suspend.h>
26#include <linux/rwsem.h>
27#include <linux/ptrace.h>
28#include <linux/async.h>
29#include <linux/uaccess.h>
30#include <linux/initrd.h>
31#include <linux/freezer.h>
32
33#include <trace/events/module.h>
34
35static kernel_cap_t usermodehelper_bset = CAP_FULL_SET;
36static kernel_cap_t usermodehelper_inheritable = CAP_FULL_SET;
37static DEFINE_SPINLOCK(umh_sysctl_lock);
38static DECLARE_RWSEM(umhelper_sem);
39
40static void call_usermodehelper_freeinfo(struct subprocess_info *info)
41{
42	if (info->cleanup)
43		(*info->cleanup)(info);
44	kfree(info);
45}
46
47static void umh_complete(struct subprocess_info *sub_info)
48{
49	struct completion *comp = xchg(&sub_info->complete, NULL);
50	/*
51	 * See call_usermodehelper_exec(). If xchg() returns NULL
52	 * we own sub_info, the UMH_KILLABLE caller has gone away
53	 * or the caller used UMH_NO_WAIT.
54	 */
55	if (comp)
56		complete(comp);
57	else
58		call_usermodehelper_freeinfo(sub_info);
59}
60
61/*
62 * This is the task which runs the usermode application
63 */
64static int call_usermodehelper_exec_async(void *data)
65{
66	struct subprocess_info *sub_info = data;
67	struct cred *new;
68	int retval;
69
70	spin_lock_irq(&current->sighand->siglock);
71	flush_signal_handlers(current, 1);
72	spin_unlock_irq(&current->sighand->siglock);
73
74	/*
75	 * Initial kernel threads share ther FS with init, in order to
76	 * get the init root directory. But we've now created a new
77	 * thread that is going to execve a user process and has its own
78	 * 'struct fs_struct'. Reset umask to the default.
79	 */
80	current->fs->umask = 0022;
81
82	/*
83	 * Our parent (unbound workqueue) runs with elevated scheduling
84	 * priority. Avoid propagating that into the userspace child.
85	 */
86	set_user_nice(current, 0);
87
88	retval = -ENOMEM;
89	new = prepare_kernel_cred(current);
90	if (!new)
91		goto out;
92
93	spin_lock(&umh_sysctl_lock);
94	new->cap_bset = cap_intersect(usermodehelper_bset, new->cap_bset);
95	new->cap_inheritable = cap_intersect(usermodehelper_inheritable,
96					     new->cap_inheritable);
97	spin_unlock(&umh_sysctl_lock);
98
99	if (sub_info->init) {
100		retval = sub_info->init(sub_info, new);
101		if (retval) {
102			abort_creds(new);
103			goto out;
104		}
105	}
106
107	commit_creds(new);
108
109	wait_for_initramfs();
110	retval = kernel_execve(sub_info->path,
111			       (const char *const *)sub_info->argv,
112			       (const char *const *)sub_info->envp);
113out:
114	sub_info->retval = retval;
115	/*
116	 * call_usermodehelper_exec_sync() will call umh_complete
117	 * if UHM_WAIT_PROC.
118	 */
119	if (!(sub_info->wait & UMH_WAIT_PROC))
120		umh_complete(sub_info);
121	if (!retval)
122		return 0;
123	do_exit(0);
124}
125
126/* Handles UMH_WAIT_PROC.  */
127static void call_usermodehelper_exec_sync(struct subprocess_info *sub_info)
128{
129	pid_t pid;
130
131	/* If SIGCLD is ignored do_wait won't populate the status. */
132	kernel_sigaction(SIGCHLD, SIG_DFL);
133	pid = user_mode_thread(call_usermodehelper_exec_async, sub_info, SIGCHLD);
134	if (pid < 0)
135		sub_info->retval = pid;
136	else
137		kernel_wait(pid, &sub_info->retval);
138
139	/* Restore default kernel sig handler */
140	kernel_sigaction(SIGCHLD, SIG_IGN);
141	umh_complete(sub_info);
142}
143
144/*
145 * We need to create the usermodehelper kernel thread from a task that is affine
146 * to an optimized set of CPUs (or nohz housekeeping ones) such that they
147 * inherit a widest affinity irrespective of call_usermodehelper() callers with
148 * possibly reduced affinity (eg: per-cpu workqueues). We don't want
149 * usermodehelper targets to contend a busy CPU.
150 *
151 * Unbound workqueues provide such wide affinity and allow to block on
152 * UMH_WAIT_PROC requests without blocking pending request (up to some limit).
153 *
154 * Besides, workqueues provide the privilege level that caller might not have
155 * to perform the usermodehelper request.
156 *
157 */
158static void call_usermodehelper_exec_work(struct work_struct *work)
159{
160	struct subprocess_info *sub_info =
161		container_of(work, struct subprocess_info, work);
162
163	if (sub_info->wait & UMH_WAIT_PROC) {
164		call_usermodehelper_exec_sync(sub_info);
165	} else {
166		pid_t pid;
167		/*
168		 * Use CLONE_PARENT to reparent it to kthreadd; we do not
169		 * want to pollute current->children, and we need a parent
170		 * that always ignores SIGCHLD to ensure auto-reaping.
171		 */
172		pid = user_mode_thread(call_usermodehelper_exec_async, sub_info,
173				       CLONE_PARENT | SIGCHLD);
174		if (pid < 0) {
175			sub_info->retval = pid;
176			umh_complete(sub_info);
177		}
178	}
179}
180
181/*
182 * If set, call_usermodehelper_exec() will exit immediately returning -EBUSY
183 * (used for preventing user land processes from being created after the user
184 * land has been frozen during a system-wide hibernation or suspend operation).
185 * Should always be manipulated under umhelper_sem acquired for write.
186 */
187static enum umh_disable_depth usermodehelper_disabled = UMH_DISABLED;
188
189/* Number of helpers running */
190static atomic_t running_helpers = ATOMIC_INIT(0);
191
192/*
193 * Wait queue head used by usermodehelper_disable() to wait for all running
194 * helpers to finish.
195 */
196static DECLARE_WAIT_QUEUE_HEAD(running_helpers_waitq);
197
198/*
199 * Used by usermodehelper_read_lock_wait() to wait for usermodehelper_disabled
200 * to become 'false'.
201 */
202static DECLARE_WAIT_QUEUE_HEAD(usermodehelper_disabled_waitq);
203
204/*
205 * Time to wait for running_helpers to become zero before the setting of
206 * usermodehelper_disabled in usermodehelper_disable() fails
207 */
208#define RUNNING_HELPERS_TIMEOUT	(5 * HZ)
209
210int usermodehelper_read_trylock(void)
211{
212	DEFINE_WAIT(wait);
213	int ret = 0;
214
215	down_read(&umhelper_sem);
216	for (;;) {
217		prepare_to_wait(&usermodehelper_disabled_waitq, &wait,
218				TASK_INTERRUPTIBLE);
219		if (!usermodehelper_disabled)
220			break;
221
222		if (usermodehelper_disabled == UMH_DISABLED)
223			ret = -EAGAIN;
224
225		up_read(&umhelper_sem);
226
227		if (ret)
228			break;
229
230		schedule();
231		try_to_freeze();
232
233		down_read(&umhelper_sem);
234	}
235	finish_wait(&usermodehelper_disabled_waitq, &wait);
236	return ret;
237}
238EXPORT_SYMBOL_GPL(usermodehelper_read_trylock);
239
240long usermodehelper_read_lock_wait(long timeout)
241{
242	DEFINE_WAIT(wait);
243
244	if (timeout < 0)
245		return -EINVAL;
246
247	down_read(&umhelper_sem);
248	for (;;) {
249		prepare_to_wait(&usermodehelper_disabled_waitq, &wait,
250				TASK_UNINTERRUPTIBLE);
251		if (!usermodehelper_disabled)
252			break;
253
254		up_read(&umhelper_sem);
255
256		timeout = schedule_timeout(timeout);
257		if (!timeout)
258			break;
259
260		down_read(&umhelper_sem);
261	}
262	finish_wait(&usermodehelper_disabled_waitq, &wait);
263	return timeout;
264}
265EXPORT_SYMBOL_GPL(usermodehelper_read_lock_wait);
266
267void usermodehelper_read_unlock(void)
268{
269	up_read(&umhelper_sem);
270}
271EXPORT_SYMBOL_GPL(usermodehelper_read_unlock);
272
273/**
274 * __usermodehelper_set_disable_depth - Modify usermodehelper_disabled.
275 * @depth: New value to assign to usermodehelper_disabled.
276 *
277 * Change the value of usermodehelper_disabled (under umhelper_sem locked for
278 * writing) and wakeup tasks waiting for it to change.
279 */
280void __usermodehelper_set_disable_depth(enum umh_disable_depth depth)
281{
282	down_write(&umhelper_sem);
283	usermodehelper_disabled = depth;
284	wake_up(&usermodehelper_disabled_waitq);
285	up_write(&umhelper_sem);
286}
287
288/**
289 * __usermodehelper_disable - Prevent new helpers from being started.
290 * @depth: New value to assign to usermodehelper_disabled.
291 *
292 * Set usermodehelper_disabled to @depth and wait for running helpers to exit.
293 */
294int __usermodehelper_disable(enum umh_disable_depth depth)
295{
296	long retval;
297
298	if (!depth)
299		return -EINVAL;
300
301	down_write(&umhelper_sem);
302	usermodehelper_disabled = depth;
303	up_write(&umhelper_sem);
304
305	/*
306	 * From now on call_usermodehelper_exec() won't start any new
307	 * helpers, so it is sufficient if running_helpers turns out to
308	 * be zero at one point (it may be increased later, but that
309	 * doesn't matter).
310	 */
311	retval = wait_event_timeout(running_helpers_waitq,
312					atomic_read(&running_helpers) == 0,
313					RUNNING_HELPERS_TIMEOUT);
314	if (retval)
315		return 0;
316
317	__usermodehelper_set_disable_depth(UMH_ENABLED);
318	return -EAGAIN;
319}
320
321static void helper_lock(void)
322{
323	atomic_inc(&running_helpers);
324	smp_mb__after_atomic();
325}
326
327static void helper_unlock(void)
328{
329	if (atomic_dec_and_test(&running_helpers))
330		wake_up(&running_helpers_waitq);
331}
332
333/**
334 * call_usermodehelper_setup - prepare to call a usermode helper
335 * @path: path to usermode executable
336 * @argv: arg vector for process
337 * @envp: environment for process
338 * @gfp_mask: gfp mask for memory allocation
339 * @init: an init function
340 * @cleanup: a cleanup function
341 * @data: arbitrary context sensitive data
342 *
343 * Returns either %NULL on allocation failure, or a subprocess_info
344 * structure.  This should be passed to call_usermodehelper_exec to
345 * exec the process and free the structure.
346 *
347 * The init function is used to customize the helper process prior to
348 * exec.  A non-zero return code causes the process to error out, exit,
349 * and return the failure to the calling process
350 *
351 * The cleanup function is just before the subprocess_info is about to
352 * be freed.  This can be used for freeing the argv and envp.  The
353 * Function must be runnable in either a process context or the
354 * context in which call_usermodehelper_exec is called.
355 */
356struct subprocess_info *call_usermodehelper_setup(const char *path, char **argv,
357		char **envp, gfp_t gfp_mask,
358		int (*init)(struct subprocess_info *info, struct cred *new),
359		void (*cleanup)(struct subprocess_info *info),
360		void *data)
361{
362	struct subprocess_info *sub_info;
363	sub_info = kzalloc(sizeof(struct subprocess_info), gfp_mask);
364	if (!sub_info)
365		goto out;
366
367	INIT_WORK(&sub_info->work, call_usermodehelper_exec_work);
368
369#ifdef CONFIG_STATIC_USERMODEHELPER
370	sub_info->path = CONFIG_STATIC_USERMODEHELPER_PATH;
371#else
372	sub_info->path = path;
373#endif
374	sub_info->argv = argv;
375	sub_info->envp = envp;
376
377	sub_info->cleanup = cleanup;
378	sub_info->init = init;
379	sub_info->data = data;
380  out:
381	return sub_info;
382}
383EXPORT_SYMBOL(call_usermodehelper_setup);
384
385/**
386 * call_usermodehelper_exec - start a usermode application
387 * @sub_info: information about the subprocess
388 * @wait: wait for the application to finish and return status.
389 *        when UMH_NO_WAIT don't wait at all, but you get no useful error back
390 *        when the program couldn't be exec'ed. This makes it safe to call
391 *        from interrupt context.
392 *
393 * Runs a user-space application.  The application is started
394 * asynchronously if wait is not set, and runs as a child of system workqueues.
395 * (ie. it runs with full root capabilities and optimized affinity).
396 *
397 * Note: successful return value does not guarantee the helper was called at
398 * all. You can't rely on sub_info->{init,cleanup} being called even for
399 * UMH_WAIT_* wait modes as STATIC_USERMODEHELPER_PATH="" turns all helpers
400 * into a successful no-op.
401 */
402int call_usermodehelper_exec(struct subprocess_info *sub_info, int wait)
403{
404	unsigned int state = TASK_UNINTERRUPTIBLE;
405	DECLARE_COMPLETION_ONSTACK(done);
406	int retval = 0;
407
408	if (!sub_info->path) {
409		call_usermodehelper_freeinfo(sub_info);
410		return -EINVAL;
411	}
412	helper_lock();
413	if (usermodehelper_disabled) {
414		retval = -EBUSY;
415		goto out;
416	}
417
418	/*
419	 * If there is no binary for us to call, then just return and get out of
420	 * here.  This allows us to set STATIC_USERMODEHELPER_PATH to "" and
421	 * disable all call_usermodehelper() calls.
422	 */
423	if (strlen(sub_info->path) == 0)
424		goto out;
425
426	/*
427	 * Set the completion pointer only if there is a waiter.
428	 * This makes it possible to use umh_complete to free
429	 * the data structure in case of UMH_NO_WAIT.
430	 */
431	sub_info->complete = (wait == UMH_NO_WAIT) ? NULL : &done;
432	sub_info->wait = wait;
433
434	queue_work(system_unbound_wq, &sub_info->work);
435	if (wait == UMH_NO_WAIT)	/* task has freed sub_info */
436		goto unlock;
437
438	if (wait & UMH_FREEZABLE)
439		state |= TASK_FREEZABLE;
440
441	if (wait & UMH_KILLABLE) {
442		retval = wait_for_completion_state(&done, state | TASK_KILLABLE);
443		if (!retval)
444			goto wait_done;
445
446		/* umh_complete() will see NULL and free sub_info */
447		if (xchg(&sub_info->complete, NULL))
448			goto unlock;
449
450		/*
451		 * fallthrough; in case of -ERESTARTSYS now do uninterruptible
452		 * wait_for_completion_state(). Since umh_complete() shall call
453		 * complete() in a moment if xchg() above returned NULL, this
454		 * uninterruptible wait_for_completion_state() will not block
455		 * SIGKILL'ed processes for long.
456		 */
457	}
458	wait_for_completion_state(&done, state);
459
460wait_done:
461	retval = sub_info->retval;
462out:
463	call_usermodehelper_freeinfo(sub_info);
464unlock:
465	helper_unlock();
466	return retval;
467}
468EXPORT_SYMBOL(call_usermodehelper_exec);
469
470/**
471 * call_usermodehelper() - prepare and start a usermode application
472 * @path: path to usermode executable
473 * @argv: arg vector for process
474 * @envp: environment for process
475 * @wait: wait for the application to finish and return status.
476 *        when UMH_NO_WAIT don't wait at all, but you get no useful error back
477 *        when the program couldn't be exec'ed. This makes it safe to call
478 *        from interrupt context.
479 *
480 * This function is the equivalent to use call_usermodehelper_setup() and
481 * call_usermodehelper_exec().
482 */
483int call_usermodehelper(const char *path, char **argv, char **envp, int wait)
484{
485	struct subprocess_info *info;
486	gfp_t gfp_mask = (wait == UMH_NO_WAIT) ? GFP_ATOMIC : GFP_KERNEL;
487
488	info = call_usermodehelper_setup(path, argv, envp, gfp_mask,
489					 NULL, NULL, NULL);
490	if (info == NULL)
491		return -ENOMEM;
492
493	return call_usermodehelper_exec(info, wait);
494}
495EXPORT_SYMBOL(call_usermodehelper);
496
497#if defined(CONFIG_SYSCTL)
498static int proc_cap_handler(struct ctl_table *table, int write,
499			 void *buffer, size_t *lenp, loff_t *ppos)
500{
501	struct ctl_table t;
502	unsigned long cap_array[2];
503	kernel_cap_t new_cap, *cap;
504	int err;
505
506	if (write && (!capable(CAP_SETPCAP) ||
507		      !capable(CAP_SYS_MODULE)))
508		return -EPERM;
509
510	/*
511	 * convert from the global kernel_cap_t to the ulong array to print to
512	 * userspace if this is a read.
513	 *
514	 * Legacy format: capabilities are exposed as two 32-bit values
515	 */
516	cap = table->data;
517	spin_lock(&umh_sysctl_lock);
518	cap_array[0] = (u32) cap->val;
519	cap_array[1] = cap->val >> 32;
520	spin_unlock(&umh_sysctl_lock);
521
522	t = *table;
523	t.data = &cap_array;
524
525	/*
526	 * actually read or write and array of ulongs from userspace.  Remember
527	 * these are least significant 32 bits first
528	 */
529	err = proc_doulongvec_minmax(&t, write, buffer, lenp, ppos);
530	if (err < 0)
531		return err;
532
533	new_cap.val = (u32)cap_array[0];
534	new_cap.val += (u64)cap_array[1] << 32;
535
536	/*
537	 * Drop everything not in the new_cap (but don't add things)
538	 */
539	if (write) {
540		spin_lock(&umh_sysctl_lock);
541		*cap = cap_intersect(*cap, new_cap);
542		spin_unlock(&umh_sysctl_lock);
543	}
544
545	return 0;
546}
547
548static struct ctl_table usermodehelper_table[] = {
549	{
550		.procname	= "bset",
551		.data		= &usermodehelper_bset,
552		.maxlen		= 2 * sizeof(unsigned long),
553		.mode		= 0600,
554		.proc_handler	= proc_cap_handler,
555	},
556	{
557		.procname	= "inheritable",
558		.data		= &usermodehelper_inheritable,
559		.maxlen		= 2 * sizeof(unsigned long),
560		.mode		= 0600,
561		.proc_handler	= proc_cap_handler,
562	},
563	{ }
564};
565
566static int __init init_umh_sysctls(void)
567{
568	register_sysctl_init("kernel/usermodehelper", usermodehelper_table);
569	return 0;
570}
571early_initcall(init_umh_sysctls);
572#endif /* CONFIG_SYSCTL */
573