Merge tag 'for-5.16/block-2021-11-09' of git://git.kernel.dk/linux-block
[linux-2.6-microblaze.git] / kernel / umh.c
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
32 #include <trace/events/module.h>
33
34 #define CAP_BSET        (void *)1
35 #define CAP_PI          (void *)2
36
37 static kernel_cap_t usermodehelper_bset = CAP_FULL_SET;
38 static kernel_cap_t usermodehelper_inheritable = CAP_FULL_SET;
39 static DEFINE_SPINLOCK(umh_sysctl_lock);
40 static DECLARE_RWSEM(umhelper_sem);
41
42 static void call_usermodehelper_freeinfo(struct subprocess_info *info)
43 {
44         if (info->cleanup)
45                 (*info->cleanup)(info);
46         kfree(info);
47 }
48
49 static void umh_complete(struct subprocess_info *sub_info)
50 {
51         struct completion *comp = xchg(&sub_info->complete, NULL);
52         /*
53          * See call_usermodehelper_exec(). If xchg() returns NULL
54          * we own sub_info, the UMH_KILLABLE caller has gone away
55          * or the caller used UMH_NO_WAIT.
56          */
57         if (comp)
58                 complete(comp);
59         else
60                 call_usermodehelper_freeinfo(sub_info);
61 }
62
63 /*
64  * This is the task which runs the usermode application
65  */
66 static int call_usermodehelper_exec_async(void *data)
67 {
68         struct subprocess_info *sub_info = data;
69         struct cred *new;
70         int retval;
71
72         spin_lock_irq(&current->sighand->siglock);
73         flush_signal_handlers(current, 1);
74         spin_unlock_irq(&current->sighand->siglock);
75
76         /*
77          * Initial kernel threads share ther FS with init, in order to
78          * get the init root directory. But we've now created a new
79          * thread that is going to execve a user process and has its own
80          * 'struct fs_struct'. Reset umask to the default.
81          */
82         current->fs->umask = 0022;
83
84         /*
85          * Our parent (unbound workqueue) runs with elevated scheduling
86          * priority. Avoid propagating that into the userspace child.
87          */
88         set_user_nice(current, 0);
89
90         retval = -ENOMEM;
91         new = prepare_kernel_cred(current);
92         if (!new)
93                 goto out;
94
95         spin_lock(&umh_sysctl_lock);
96         new->cap_bset = cap_intersect(usermodehelper_bset, new->cap_bset);
97         new->cap_inheritable = cap_intersect(usermodehelper_inheritable,
98                                              new->cap_inheritable);
99         spin_unlock(&umh_sysctl_lock);
100
101         if (sub_info->init) {
102                 retval = sub_info->init(sub_info, new);
103                 if (retval) {
104                         abort_creds(new);
105                         goto out;
106                 }
107         }
108
109         commit_creds(new);
110
111         wait_for_initramfs();
112         retval = kernel_execve(sub_info->path,
113                                (const char *const *)sub_info->argv,
114                                (const char *const *)sub_info->envp);
115 out:
116         sub_info->retval = retval;
117         /*
118          * call_usermodehelper_exec_sync() will call umh_complete
119          * if UHM_WAIT_PROC.
120          */
121         if (!(sub_info->wait & UMH_WAIT_PROC))
122                 umh_complete(sub_info);
123         if (!retval)
124                 return 0;
125         do_exit(0);
126 }
127
128 /* Handles UMH_WAIT_PROC.  */
129 static void call_usermodehelper_exec_sync(struct subprocess_info *sub_info)
130 {
131         pid_t pid;
132
133         /* If SIGCLD is ignored do_wait won't populate the status. */
134         kernel_sigaction(SIGCHLD, SIG_DFL);
135         pid = kernel_thread(call_usermodehelper_exec_async, sub_info, SIGCHLD);
136         if (pid < 0)
137                 sub_info->retval = pid;
138         else
139                 kernel_wait(pid, &sub_info->retval);
140
141         /* Restore default kernel sig handler */
142         kernel_sigaction(SIGCHLD, SIG_IGN);
143         umh_complete(sub_info);
144 }
145
146 /*
147  * We need to create the usermodehelper kernel thread from a task that is affine
148  * to an optimized set of CPUs (or nohz housekeeping ones) such that they
149  * inherit a widest affinity irrespective of call_usermodehelper() callers with
150  * possibly reduced affinity (eg: per-cpu workqueues). We don't want
151  * usermodehelper targets to contend a busy CPU.
152  *
153  * Unbound workqueues provide such wide affinity and allow to block on
154  * UMH_WAIT_PROC requests without blocking pending request (up to some limit).
155  *
156  * Besides, workqueues provide the privilege level that caller might not have
157  * to perform the usermodehelper request.
158  *
159  */
160 static void call_usermodehelper_exec_work(struct work_struct *work)
161 {
162         struct subprocess_info *sub_info =
163                 container_of(work, struct subprocess_info, work);
164
165         if (sub_info->wait & UMH_WAIT_PROC) {
166                 call_usermodehelper_exec_sync(sub_info);
167         } else {
168                 pid_t pid;
169                 /*
170                  * Use CLONE_PARENT to reparent it to kthreadd; we do not
171                  * want to pollute current->children, and we need a parent
172                  * that always ignores SIGCHLD to ensure auto-reaping.
173                  */
174                 pid = kernel_thread(call_usermodehelper_exec_async, sub_info,
175                                     CLONE_PARENT | SIGCHLD);
176                 if (pid < 0) {
177                         sub_info->retval = pid;
178                         umh_complete(sub_info);
179                 }
180         }
181 }
182
183 /*
184  * If set, call_usermodehelper_exec() will exit immediately returning -EBUSY
185  * (used for preventing user land processes from being created after the user
186  * land has been frozen during a system-wide hibernation or suspend operation).
187  * Should always be manipulated under umhelper_sem acquired for write.
188  */
189 static enum umh_disable_depth usermodehelper_disabled = UMH_DISABLED;
190
191 /* Number of helpers running */
192 static atomic_t running_helpers = ATOMIC_INIT(0);
193
194 /*
195  * Wait queue head used by usermodehelper_disable() to wait for all running
196  * helpers to finish.
197  */
198 static DECLARE_WAIT_QUEUE_HEAD(running_helpers_waitq);
199
200 /*
201  * Used by usermodehelper_read_lock_wait() to wait for usermodehelper_disabled
202  * to become 'false'.
203  */
204 static DECLARE_WAIT_QUEUE_HEAD(usermodehelper_disabled_waitq);
205
206 /*
207  * Time to wait for running_helpers to become zero before the setting of
208  * usermodehelper_disabled in usermodehelper_disable() fails
209  */
210 #define RUNNING_HELPERS_TIMEOUT (5 * HZ)
211
212 int usermodehelper_read_trylock(void)
213 {
214         DEFINE_WAIT(wait);
215         int ret = 0;
216
217         down_read(&umhelper_sem);
218         for (;;) {
219                 prepare_to_wait(&usermodehelper_disabled_waitq, &wait,
220                                 TASK_INTERRUPTIBLE);
221                 if (!usermodehelper_disabled)
222                         break;
223
224                 if (usermodehelper_disabled == UMH_DISABLED)
225                         ret = -EAGAIN;
226
227                 up_read(&umhelper_sem);
228
229                 if (ret)
230                         break;
231
232                 schedule();
233                 try_to_freeze();
234
235                 down_read(&umhelper_sem);
236         }
237         finish_wait(&usermodehelper_disabled_waitq, &wait);
238         return ret;
239 }
240 EXPORT_SYMBOL_GPL(usermodehelper_read_trylock);
241
242 long usermodehelper_read_lock_wait(long timeout)
243 {
244         DEFINE_WAIT(wait);
245
246         if (timeout < 0)
247                 return -EINVAL;
248
249         down_read(&umhelper_sem);
250         for (;;) {
251                 prepare_to_wait(&usermodehelper_disabled_waitq, &wait,
252                                 TASK_UNINTERRUPTIBLE);
253                 if (!usermodehelper_disabled)
254                         break;
255
256                 up_read(&umhelper_sem);
257
258                 timeout = schedule_timeout(timeout);
259                 if (!timeout)
260                         break;
261
262                 down_read(&umhelper_sem);
263         }
264         finish_wait(&usermodehelper_disabled_waitq, &wait);
265         return timeout;
266 }
267 EXPORT_SYMBOL_GPL(usermodehelper_read_lock_wait);
268
269 void usermodehelper_read_unlock(void)
270 {
271         up_read(&umhelper_sem);
272 }
273 EXPORT_SYMBOL_GPL(usermodehelper_read_unlock);
274
275 /**
276  * __usermodehelper_set_disable_depth - Modify usermodehelper_disabled.
277  * @depth: New value to assign to usermodehelper_disabled.
278  *
279  * Change the value of usermodehelper_disabled (under umhelper_sem locked for
280  * writing) and wakeup tasks waiting for it to change.
281  */
282 void __usermodehelper_set_disable_depth(enum umh_disable_depth depth)
283 {
284         down_write(&umhelper_sem);
285         usermodehelper_disabled = depth;
286         wake_up(&usermodehelper_disabled_waitq);
287         up_write(&umhelper_sem);
288 }
289
290 /**
291  * __usermodehelper_disable - Prevent new helpers from being started.
292  * @depth: New value to assign to usermodehelper_disabled.
293  *
294  * Set usermodehelper_disabled to @depth and wait for running helpers to exit.
295  */
296 int __usermodehelper_disable(enum umh_disable_depth depth)
297 {
298         long retval;
299
300         if (!depth)
301                 return -EINVAL;
302
303         down_write(&umhelper_sem);
304         usermodehelper_disabled = depth;
305         up_write(&umhelper_sem);
306
307         /*
308          * From now on call_usermodehelper_exec() won't start any new
309          * helpers, so it is sufficient if running_helpers turns out to
310          * be zero at one point (it may be increased later, but that
311          * doesn't matter).
312          */
313         retval = wait_event_timeout(running_helpers_waitq,
314                                         atomic_read(&running_helpers) == 0,
315                                         RUNNING_HELPERS_TIMEOUT);
316         if (retval)
317                 return 0;
318
319         __usermodehelper_set_disable_depth(UMH_ENABLED);
320         return -EAGAIN;
321 }
322
323 static void helper_lock(void)
324 {
325         atomic_inc(&running_helpers);
326         smp_mb__after_atomic();
327 }
328
329 static void helper_unlock(void)
330 {
331         if (atomic_dec_and_test(&running_helpers))
332                 wake_up(&running_helpers_waitq);
333 }
334
335 /**
336  * call_usermodehelper_setup - prepare to call a usermode helper
337  * @path: path to usermode executable
338  * @argv: arg vector for process
339  * @envp: environment for process
340  * @gfp_mask: gfp mask for memory allocation
341  * @init: an init function
342  * @cleanup: a cleanup function
343  * @data: arbitrary context sensitive data
344  *
345  * Returns either %NULL on allocation failure, or a subprocess_info
346  * structure.  This should be passed to call_usermodehelper_exec to
347  * exec the process and free the structure.
348  *
349  * The init function is used to customize the helper process prior to
350  * exec.  A non-zero return code causes the process to error out, exit,
351  * and return the failure to the calling process
352  *
353  * The cleanup function is just before the subprocess_info is about to
354  * be freed.  This can be used for freeing the argv and envp.  The
355  * Function must be runnable in either a process context or the
356  * context in which call_usermodehelper_exec is called.
357  */
358 struct subprocess_info *call_usermodehelper_setup(const char *path, char **argv,
359                 char **envp, gfp_t gfp_mask,
360                 int (*init)(struct subprocess_info *info, struct cred *new),
361                 void (*cleanup)(struct subprocess_info *info),
362                 void *data)
363 {
364         struct subprocess_info *sub_info;
365         sub_info = kzalloc(sizeof(struct subprocess_info), gfp_mask);
366         if (!sub_info)
367                 goto out;
368
369         INIT_WORK(&sub_info->work, call_usermodehelper_exec_work);
370
371 #ifdef CONFIG_STATIC_USERMODEHELPER
372         sub_info->path = CONFIG_STATIC_USERMODEHELPER_PATH;
373 #else
374         sub_info->path = path;
375 #endif
376         sub_info->argv = argv;
377         sub_info->envp = envp;
378
379         sub_info->cleanup = cleanup;
380         sub_info->init = init;
381         sub_info->data = data;
382   out:
383         return sub_info;
384 }
385 EXPORT_SYMBOL(call_usermodehelper_setup);
386
387 /**
388  * call_usermodehelper_exec - start a usermode application
389  * @sub_info: information about the subprocess
390  * @wait: wait for the application to finish and return status.
391  *        when UMH_NO_WAIT don't wait at all, but you get no useful error back
392  *        when the program couldn't be exec'ed. This makes it safe to call
393  *        from interrupt context.
394  *
395  * Runs a user-space application.  The application is started
396  * asynchronously if wait is not set, and runs as a child of system workqueues.
397  * (ie. it runs with full root capabilities and optimized affinity).
398  *
399  * Note: successful return value does not guarantee the helper was called at
400  * all. You can't rely on sub_info->{init,cleanup} being called even for
401  * UMH_WAIT_* wait modes as STATIC_USERMODEHELPER_PATH="" turns all helpers
402  * into a successful no-op.
403  */
404 int call_usermodehelper_exec(struct subprocess_info *sub_info, int wait)
405 {
406         DECLARE_COMPLETION_ONSTACK(done);
407         int retval = 0;
408
409         if (!sub_info->path) {
410                 call_usermodehelper_freeinfo(sub_info);
411                 return -EINVAL;
412         }
413         helper_lock();
414         if (usermodehelper_disabled) {
415                 retval = -EBUSY;
416                 goto out;
417         }
418
419         /*
420          * If there is no binary for us to call, then just return and get out of
421          * here.  This allows us to set STATIC_USERMODEHELPER_PATH to "" and
422          * disable all call_usermodehelper() calls.
423          */
424         if (strlen(sub_info->path) == 0)
425                 goto out;
426
427         /*
428          * Set the completion pointer only if there is a waiter.
429          * This makes it possible to use umh_complete to free
430          * the data structure in case of UMH_NO_WAIT.
431          */
432         sub_info->complete = (wait == UMH_NO_WAIT) ? NULL : &done;
433         sub_info->wait = wait;
434
435         queue_work(system_unbound_wq, &sub_info->work);
436         if (wait == UMH_NO_WAIT)        /* task has freed sub_info */
437                 goto unlock;
438
439         if (wait & UMH_KILLABLE) {
440                 retval = wait_for_completion_killable(&done);
441                 if (!retval)
442                         goto wait_done;
443
444                 /* umh_complete() will see NULL and free sub_info */
445                 if (xchg(&sub_info->complete, NULL))
446                         goto unlock;
447                 /* fallthrough, umh_complete() was already called */
448         }
449
450         wait_for_completion(&done);
451 wait_done:
452         retval = sub_info->retval;
453 out:
454         call_usermodehelper_freeinfo(sub_info);
455 unlock:
456         helper_unlock();
457         return retval;
458 }
459 EXPORT_SYMBOL(call_usermodehelper_exec);
460
461 /**
462  * call_usermodehelper() - prepare and start a usermode application
463  * @path: path to usermode executable
464  * @argv: arg vector for process
465  * @envp: environment for process
466  * @wait: wait for the application to finish and return status.
467  *        when UMH_NO_WAIT don't wait at all, but you get no useful error back
468  *        when the program couldn't be exec'ed. This makes it safe to call
469  *        from interrupt context.
470  *
471  * This function is the equivalent to use call_usermodehelper_setup() and
472  * call_usermodehelper_exec().
473  */
474 int call_usermodehelper(const char *path, char **argv, char **envp, int wait)
475 {
476         struct subprocess_info *info;
477         gfp_t gfp_mask = (wait == UMH_NO_WAIT) ? GFP_ATOMIC : GFP_KERNEL;
478
479         info = call_usermodehelper_setup(path, argv, envp, gfp_mask,
480                                          NULL, NULL, NULL);
481         if (info == NULL)
482                 return -ENOMEM;
483
484         return call_usermodehelper_exec(info, wait);
485 }
486 EXPORT_SYMBOL(call_usermodehelper);
487
488 static int proc_cap_handler(struct ctl_table *table, int write,
489                          void *buffer, size_t *lenp, loff_t *ppos)
490 {
491         struct ctl_table t;
492         unsigned long cap_array[_KERNEL_CAPABILITY_U32S];
493         kernel_cap_t new_cap;
494         int err, i;
495
496         if (write && (!capable(CAP_SETPCAP) ||
497                       !capable(CAP_SYS_MODULE)))
498                 return -EPERM;
499
500         /*
501          * convert from the global kernel_cap_t to the ulong array to print to
502          * userspace if this is a read.
503          */
504         spin_lock(&umh_sysctl_lock);
505         for (i = 0; i < _KERNEL_CAPABILITY_U32S; i++)  {
506                 if (table->data == CAP_BSET)
507                         cap_array[i] = usermodehelper_bset.cap[i];
508                 else if (table->data == CAP_PI)
509                         cap_array[i] = usermodehelper_inheritable.cap[i];
510                 else
511                         BUG();
512         }
513         spin_unlock(&umh_sysctl_lock);
514
515         t = *table;
516         t.data = &cap_array;
517
518         /*
519          * actually read or write and array of ulongs from userspace.  Remember
520          * these are least significant 32 bits first
521          */
522         err = proc_doulongvec_minmax(&t, write, buffer, lenp, ppos);
523         if (err < 0)
524                 return err;
525
526         /*
527          * convert from the sysctl array of ulongs to the kernel_cap_t
528          * internal representation
529          */
530         for (i = 0; i < _KERNEL_CAPABILITY_U32S; i++)
531                 new_cap.cap[i] = cap_array[i];
532
533         /*
534          * Drop everything not in the new_cap (but don't add things)
535          */
536         if (write) {
537                 spin_lock(&umh_sysctl_lock);
538                 if (table->data == CAP_BSET)
539                         usermodehelper_bset = cap_intersect(usermodehelper_bset, new_cap);
540                 if (table->data == CAP_PI)
541                         usermodehelper_inheritable = cap_intersect(usermodehelper_inheritable, new_cap);
542                 spin_unlock(&umh_sysctl_lock);
543         }
544
545         return 0;
546 }
547
548 struct ctl_table usermodehelper_table[] = {
549         {
550                 .procname       = "bset",
551                 .data           = CAP_BSET,
552                 .maxlen         = _KERNEL_CAPABILITY_U32S * sizeof(unsigned long),
553                 .mode           = 0600,
554                 .proc_handler   = proc_cap_handler,
555         },
556         {
557                 .procname       = "inheritable",
558                 .data           = CAP_PI,
559                 .maxlen         = _KERNEL_CAPABILITY_U32S * sizeof(unsigned long),
560                 .mode           = 0600,
561                 .proc_handler   = proc_cap_handler,
562         },
563         { }
564 };