c449858946afc23706e06ab1e893ba0cef4c4f99
[muen/linux.git] / kernel / umh.c
1 /*
2  * umh - the kernel usermode helper
3  */
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>
30
31 #include <trace/events/module.h>
32
33 #define CAP_BSET        (void *)1
34 #define CAP_PI          (void *)2
35
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
41 static void call_usermodehelper_freeinfo(struct subprocess_info *info)
42 {
43         if (info->cleanup)
44                 (*info->cleanup)(info);
45         kfree(info);
46 }
47
48 static void umh_complete(struct subprocess_info *sub_info)
49 {
50         struct completion *comp = xchg(&sub_info->complete, NULL);
51         /*
52          * See call_usermodehelper_exec(). If xchg() returns NULL
53          * we own sub_info, the UMH_KILLABLE caller has gone away
54          * or the caller used UMH_NO_WAIT.
55          */
56         if (comp)
57                 complete(comp);
58         else
59                 call_usermodehelper_freeinfo(sub_info);
60 }
61
62 /*
63  * This is the task which runs the usermode application
64  */
65 static int call_usermodehelper_exec_async(void *data)
66 {
67         struct subprocess_info *sub_info = data;
68         struct cred *new;
69         int retval;
70
71         spin_lock_irq(&current->sighand->siglock);
72         flush_signal_handlers(current, 1);
73         spin_unlock_irq(&current->sighand->siglock);
74
75         /*
76          * Our parent (unbound workqueue) runs with elevated scheduling
77          * priority. Avoid propagating that into the userspace child.
78          */
79         set_user_nice(current, 0);
80
81         retval = -ENOMEM;
82         new = prepare_kernel_cred(current);
83         if (!new)
84                 goto out;
85
86         spin_lock(&umh_sysctl_lock);
87         new->cap_bset = cap_intersect(usermodehelper_bset, new->cap_bset);
88         new->cap_inheritable = cap_intersect(usermodehelper_inheritable,
89                                              new->cap_inheritable);
90         spin_unlock(&umh_sysctl_lock);
91
92         if (sub_info->init) {
93                 retval = sub_info->init(sub_info, new);
94                 if (retval) {
95                         abort_creds(new);
96                         goto out;
97                 }
98         }
99
100         commit_creds(new);
101
102         sub_info->pid = task_pid_nr(current);
103         if (sub_info->file)
104                 retval = do_execve_file(sub_info->file,
105                                         sub_info->argv, sub_info->envp);
106         else
107                 retval = do_execve(getname_kernel(sub_info->path),
108                                    (const char __user *const __user *)sub_info->argv,
109                                    (const char __user *const __user *)sub_info->envp);
110 out:
111         sub_info->retval = retval;
112         /*
113          * call_usermodehelper_exec_sync() will call umh_complete
114          * if UHM_WAIT_PROC.
115          */
116         if (!(sub_info->wait & UMH_WAIT_PROC))
117                 umh_complete(sub_info);
118         if (!retval)
119                 return 0;
120         do_exit(0);
121 }
122
123 /* Handles UMH_WAIT_PROC.  */
124 static void call_usermodehelper_exec_sync(struct subprocess_info *sub_info)
125 {
126         pid_t pid;
127
128         /* If SIGCLD is ignored kernel_wait4 won't populate the status. */
129         kernel_sigaction(SIGCHLD, SIG_DFL);
130         pid = kernel_thread(call_usermodehelper_exec_async, sub_info, SIGCHLD);
131         if (pid < 0) {
132                 sub_info->retval = pid;
133         } else {
134                 int ret = -ECHILD;
135                 /*
136                  * Normally it is bogus to call wait4() from in-kernel because
137                  * wait4() wants to write the exit code to a userspace address.
138                  * But call_usermodehelper_exec_sync() always runs as kernel
139                  * thread (workqueue) and put_user() to a kernel address works
140                  * OK for kernel threads, due to their having an mm_segment_t
141                  * which spans the entire address space.
142                  *
143                  * Thus the __user pointer cast is valid here.
144                  */
145                 kernel_wait4(pid, (int __user *)&ret, 0, NULL);
146
147                 /*
148                  * If ret is 0, either call_usermodehelper_exec_async failed and
149                  * the real error code is already in sub_info->retval or
150                  * sub_info->retval is 0 anyway, so don't mess with it then.
151                  */
152                 if (ret)
153                         sub_info->retval = ret;
154         }
155
156         /* Restore default kernel sig handler */
157         kernel_sigaction(SIGCHLD, SIG_IGN);
158
159         umh_complete(sub_info);
160 }
161
162 /*
163  * We need to create the usermodehelper kernel thread from a task that is affine
164  * to an optimized set of CPUs (or nohz housekeeping ones) such that they
165  * inherit a widest affinity irrespective of call_usermodehelper() callers with
166  * possibly reduced affinity (eg: per-cpu workqueues). We don't want
167  * usermodehelper targets to contend a busy CPU.
168  *
169  * Unbound workqueues provide such wide affinity and allow to block on
170  * UMH_WAIT_PROC requests without blocking pending request (up to some limit).
171  *
172  * Besides, workqueues provide the privilege level that caller might not have
173  * to perform the usermodehelper request.
174  *
175  */
176 static void call_usermodehelper_exec_work(struct work_struct *work)
177 {
178         struct subprocess_info *sub_info =
179                 container_of(work, struct subprocess_info, work);
180
181         if (sub_info->wait & UMH_WAIT_PROC) {
182                 call_usermodehelper_exec_sync(sub_info);
183         } else {
184                 pid_t pid;
185                 /*
186                  * Use CLONE_PARENT to reparent it to kthreadd; we do not
187                  * want to pollute current->children, and we need a parent
188                  * that always ignores SIGCHLD to ensure auto-reaping.
189                  */
190                 pid = kernel_thread(call_usermodehelper_exec_async, sub_info,
191                                     CLONE_PARENT | SIGCHLD);
192                 if (pid < 0) {
193                         sub_info->retval = pid;
194                         umh_complete(sub_info);
195                 }
196         }
197 }
198
199 /*
200  * If set, call_usermodehelper_exec() will exit immediately returning -EBUSY
201  * (used for preventing user land processes from being created after the user
202  * land has been frozen during a system-wide hibernation or suspend operation).
203  * Should always be manipulated under umhelper_sem acquired for write.
204  */
205 static enum umh_disable_depth usermodehelper_disabled = UMH_DISABLED;
206
207 /* Number of helpers running */
208 static atomic_t running_helpers = ATOMIC_INIT(0);
209
210 /*
211  * Wait queue head used by usermodehelper_disable() to wait for all running
212  * helpers to finish.
213  */
214 static DECLARE_WAIT_QUEUE_HEAD(running_helpers_waitq);
215
216 /*
217  * Used by usermodehelper_read_lock_wait() to wait for usermodehelper_disabled
218  * to become 'false'.
219  */
220 static DECLARE_WAIT_QUEUE_HEAD(usermodehelper_disabled_waitq);
221
222 /*
223  * Time to wait for running_helpers to become zero before the setting of
224  * usermodehelper_disabled in usermodehelper_disable() fails
225  */
226 #define RUNNING_HELPERS_TIMEOUT (5 * HZ)
227
228 int usermodehelper_read_trylock(void)
229 {
230         DEFINE_WAIT(wait);
231         int ret = 0;
232
233         down_read(&umhelper_sem);
234         for (;;) {
235                 prepare_to_wait(&usermodehelper_disabled_waitq, &wait,
236                                 TASK_INTERRUPTIBLE);
237                 if (!usermodehelper_disabled)
238                         break;
239
240                 if (usermodehelper_disabled == UMH_DISABLED)
241                         ret = -EAGAIN;
242
243                 up_read(&umhelper_sem);
244
245                 if (ret)
246                         break;
247
248                 schedule();
249                 try_to_freeze();
250
251                 down_read(&umhelper_sem);
252         }
253         finish_wait(&usermodehelper_disabled_waitq, &wait);
254         return ret;
255 }
256 EXPORT_SYMBOL_GPL(usermodehelper_read_trylock);
257
258 long usermodehelper_read_lock_wait(long timeout)
259 {
260         DEFINE_WAIT(wait);
261
262         if (timeout < 0)
263                 return -EINVAL;
264
265         down_read(&umhelper_sem);
266         for (;;) {
267                 prepare_to_wait(&usermodehelper_disabled_waitq, &wait,
268                                 TASK_UNINTERRUPTIBLE);
269                 if (!usermodehelper_disabled)
270                         break;
271
272                 up_read(&umhelper_sem);
273
274                 timeout = schedule_timeout(timeout);
275                 if (!timeout)
276                         break;
277
278                 down_read(&umhelper_sem);
279         }
280         finish_wait(&usermodehelper_disabled_waitq, &wait);
281         return timeout;
282 }
283 EXPORT_SYMBOL_GPL(usermodehelper_read_lock_wait);
284
285 void usermodehelper_read_unlock(void)
286 {
287         up_read(&umhelper_sem);
288 }
289 EXPORT_SYMBOL_GPL(usermodehelper_read_unlock);
290
291 /**
292  * __usermodehelper_set_disable_depth - Modify usermodehelper_disabled.
293  * @depth: New value to assign to usermodehelper_disabled.
294  *
295  * Change the value of usermodehelper_disabled (under umhelper_sem locked for
296  * writing) and wakeup tasks waiting for it to change.
297  */
298 void __usermodehelper_set_disable_depth(enum umh_disable_depth depth)
299 {
300         down_write(&umhelper_sem);
301         usermodehelper_disabled = depth;
302         wake_up(&usermodehelper_disabled_waitq);
303         up_write(&umhelper_sem);
304 }
305
306 /**
307  * __usermodehelper_disable - Prevent new helpers from being started.
308  * @depth: New value to assign to usermodehelper_disabled.
309  *
310  * Set usermodehelper_disabled to @depth and wait for running helpers to exit.
311  */
312 int __usermodehelper_disable(enum umh_disable_depth depth)
313 {
314         long retval;
315
316         if (!depth)
317                 return -EINVAL;
318
319         down_write(&umhelper_sem);
320         usermodehelper_disabled = depth;
321         up_write(&umhelper_sem);
322
323         /*
324          * From now on call_usermodehelper_exec() won't start any new
325          * helpers, so it is sufficient if running_helpers turns out to
326          * be zero at one point (it may be increased later, but that
327          * doesn't matter).
328          */
329         retval = wait_event_timeout(running_helpers_waitq,
330                                         atomic_read(&running_helpers) == 0,
331                                         RUNNING_HELPERS_TIMEOUT);
332         if (retval)
333                 return 0;
334
335         __usermodehelper_set_disable_depth(UMH_ENABLED);
336         return -EAGAIN;
337 }
338
339 static void helper_lock(void)
340 {
341         atomic_inc(&running_helpers);
342         smp_mb__after_atomic();
343 }
344
345 static void helper_unlock(void)
346 {
347         if (atomic_dec_and_test(&running_helpers))
348                 wake_up(&running_helpers_waitq);
349 }
350
351 /**
352  * call_usermodehelper_setup - prepare to call a usermode helper
353  * @path: path to usermode executable
354  * @argv: arg vector for process
355  * @envp: environment for process
356  * @gfp_mask: gfp mask for memory allocation
357  * @cleanup: a cleanup function
358  * @init: an init function
359  * @data: arbitrary context sensitive data
360  *
361  * Returns either %NULL on allocation failure, or a subprocess_info
362  * structure.  This should be passed to call_usermodehelper_exec to
363  * exec the process and free the structure.
364  *
365  * The init function is used to customize the helper process prior to
366  * exec.  A non-zero return code causes the process to error out, exit,
367  * and return the failure to the calling process
368  *
369  * The cleanup function is just before ethe subprocess_info is about to
370  * be freed.  This can be used for freeing the argv and envp.  The
371  * Function must be runnable in either a process context or the
372  * context in which call_usermodehelper_exec is called.
373  */
374 struct subprocess_info *call_usermodehelper_setup(const char *path, char **argv,
375                 char **envp, gfp_t gfp_mask,
376                 int (*init)(struct subprocess_info *info, struct cred *new),
377                 void (*cleanup)(struct subprocess_info *info),
378                 void *data)
379 {
380         struct subprocess_info *sub_info;
381         sub_info = kzalloc(sizeof(struct subprocess_info), gfp_mask);
382         if (!sub_info)
383                 goto out;
384
385         INIT_WORK(&sub_info->work, call_usermodehelper_exec_work);
386
387 #ifdef CONFIG_STATIC_USERMODEHELPER
388         sub_info->path = CONFIG_STATIC_USERMODEHELPER_PATH;
389 #else
390         sub_info->path = path;
391 #endif
392         sub_info->argv = argv;
393         sub_info->envp = envp;
394
395         sub_info->cleanup = cleanup;
396         sub_info->init = init;
397         sub_info->data = data;
398   out:
399         return sub_info;
400 }
401 EXPORT_SYMBOL(call_usermodehelper_setup);
402
403 struct subprocess_info *call_usermodehelper_setup_file(struct file *file,
404                 int (*init)(struct subprocess_info *info, struct cred *new),
405                 void (*cleanup)(struct subprocess_info *info), void *data)
406 {
407         struct subprocess_info *sub_info;
408
409         sub_info = kzalloc(sizeof(struct subprocess_info), GFP_KERNEL);
410         if (!sub_info)
411                 return NULL;
412
413         INIT_WORK(&sub_info->work, call_usermodehelper_exec_work);
414         sub_info->path = "none";
415         sub_info->file = file;
416         sub_info->init = init;
417         sub_info->cleanup = cleanup;
418         sub_info->data = data;
419         return sub_info;
420 }
421
422 static int umh_pipe_setup(struct subprocess_info *info, struct cred *new)
423 {
424         struct umh_info *umh_info = info->data;
425         struct file *from_umh[2];
426         struct file *to_umh[2];
427         int err;
428
429         /* create pipe to send data to umh */
430         err = create_pipe_files(to_umh, 0);
431         if (err)
432                 return err;
433         err = replace_fd(0, to_umh[0], 0);
434         fput(to_umh[0]);
435         if (err < 0) {
436                 fput(to_umh[1]);
437                 return err;
438         }
439
440         /* create pipe to receive data from umh */
441         err = create_pipe_files(from_umh, 0);
442         if (err) {
443                 fput(to_umh[1]);
444                 replace_fd(0, NULL, 0);
445                 return err;
446         }
447         err = replace_fd(1, from_umh[1], 0);
448         fput(from_umh[1]);
449         if (err < 0) {
450                 fput(to_umh[1]);
451                 replace_fd(0, NULL, 0);
452                 fput(from_umh[0]);
453                 return err;
454         }
455
456         umh_info->pipe_to_umh = to_umh[1];
457         umh_info->pipe_from_umh = from_umh[0];
458         return 0;
459 }
460
461 static void umh_save_pid(struct subprocess_info *info)
462 {
463         struct umh_info *umh_info = info->data;
464
465         umh_info->pid = info->pid;
466 }
467
468 /**
469  * fork_usermode_blob - fork a blob of bytes as a usermode process
470  * @data: a blob of bytes that can be do_execv-ed as a file
471  * @len: length of the blob
472  * @info: information about usermode process (shouldn't be NULL)
473  *
474  * Returns either negative error or zero which indicates success
475  * in executing a blob of bytes as a usermode process. In such
476  * case 'struct umh_info *info' is populated with two pipes
477  * and a pid of the process. The caller is responsible for health
478  * check of the user process, killing it via pid, and closing the
479  * pipes when user process is no longer needed.
480  */
481 int fork_usermode_blob(void *data, size_t len, struct umh_info *info)
482 {
483         struct subprocess_info *sub_info;
484         struct file *file;
485         ssize_t written;
486         loff_t pos = 0;
487         int err;
488
489         file = shmem_kernel_file_setup("", len, 0);
490         if (IS_ERR(file))
491                 return PTR_ERR(file);
492
493         written = kernel_write(file, data, len, &pos);
494         if (written != len) {
495                 err = written;
496                 if (err >= 0)
497                         err = -ENOMEM;
498                 goto out;
499         }
500
501         err = -ENOMEM;
502         sub_info = call_usermodehelper_setup_file(file, umh_pipe_setup,
503                                                   umh_save_pid, info);
504         if (!sub_info)
505                 goto out;
506
507         err = call_usermodehelper_exec(sub_info, UMH_WAIT_EXEC);
508 out:
509         fput(file);
510         return err;
511 }
512 EXPORT_SYMBOL_GPL(fork_usermode_blob);
513
514 /**
515  * call_usermodehelper_exec - start a usermode application
516  * @sub_info: information about the subprocessa
517  * @wait: wait for the application to finish and return status.
518  *        when UMH_NO_WAIT don't wait at all, but you get no useful error back
519  *        when the program couldn't be exec'ed. This makes it safe to call
520  *        from interrupt context.
521  *
522  * Runs a user-space application.  The application is started
523  * asynchronously if wait is not set, and runs as a child of system workqueues.
524  * (ie. it runs with full root capabilities and optimized affinity).
525  */
526 int call_usermodehelper_exec(struct subprocess_info *sub_info, int wait)
527 {
528         DECLARE_COMPLETION_ONSTACK(done);
529         int retval = 0;
530
531         if (!sub_info->path) {
532                 call_usermodehelper_freeinfo(sub_info);
533                 return -EINVAL;
534         }
535         helper_lock();
536         if (usermodehelper_disabled) {
537                 retval = -EBUSY;
538                 goto out;
539         }
540
541         /*
542          * If there is no binary for us to call, then just return and get out of
543          * here.  This allows us to set STATIC_USERMODEHELPER_PATH to "" and
544          * disable all call_usermodehelper() calls.
545          */
546         if (strlen(sub_info->path) == 0)
547                 goto out;
548
549         /*
550          * Set the completion pointer only if there is a waiter.
551          * This makes it possible to use umh_complete to free
552          * the data structure in case of UMH_NO_WAIT.
553          */
554         sub_info->complete = (wait == UMH_NO_WAIT) ? NULL : &done;
555         sub_info->wait = wait;
556
557         queue_work(system_unbound_wq, &sub_info->work);
558         if (wait == UMH_NO_WAIT)        /* task has freed sub_info */
559                 goto unlock;
560
561         if (wait & UMH_KILLABLE) {
562                 retval = wait_for_completion_killable(&done);
563                 if (!retval)
564                         goto wait_done;
565
566                 /* umh_complete() will see NULL and free sub_info */
567                 if (xchg(&sub_info->complete, NULL))
568                         goto unlock;
569                 /* fallthrough, umh_complete() was already called */
570         }
571
572         wait_for_completion(&done);
573 wait_done:
574         retval = sub_info->retval;
575 out:
576         call_usermodehelper_freeinfo(sub_info);
577 unlock:
578         helper_unlock();
579         return retval;
580 }
581 EXPORT_SYMBOL(call_usermodehelper_exec);
582
583 /**
584  * call_usermodehelper() - prepare and start a usermode application
585  * @path: path to usermode executable
586  * @argv: arg vector for process
587  * @envp: environment for process
588  * @wait: wait for the application to finish and return status.
589  *        when UMH_NO_WAIT don't wait at all, but you get no useful error back
590  *        when the program couldn't be exec'ed. This makes it safe to call
591  *        from interrupt context.
592  *
593  * This function is the equivalent to use call_usermodehelper_setup() and
594  * call_usermodehelper_exec().
595  */
596 int call_usermodehelper(const char *path, char **argv, char **envp, int wait)
597 {
598         struct subprocess_info *info;
599         gfp_t gfp_mask = (wait == UMH_NO_WAIT) ? GFP_ATOMIC : GFP_KERNEL;
600
601         info = call_usermodehelper_setup(path, argv, envp, gfp_mask,
602                                          NULL, NULL, NULL);
603         if (info == NULL)
604                 return -ENOMEM;
605
606         return call_usermodehelper_exec(info, wait);
607 }
608 EXPORT_SYMBOL(call_usermodehelper);
609
610 static int proc_cap_handler(struct ctl_table *table, int write,
611                          void __user *buffer, size_t *lenp, loff_t *ppos)
612 {
613         struct ctl_table t;
614         unsigned long cap_array[_KERNEL_CAPABILITY_U32S];
615         kernel_cap_t new_cap;
616         int err, i;
617
618         if (write && (!capable(CAP_SETPCAP) ||
619                       !capable(CAP_SYS_MODULE)))
620                 return -EPERM;
621
622         /*
623          * convert from the global kernel_cap_t to the ulong array to print to
624          * userspace if this is a read.
625          */
626         spin_lock(&umh_sysctl_lock);
627         for (i = 0; i < _KERNEL_CAPABILITY_U32S; i++)  {
628                 if (table->data == CAP_BSET)
629                         cap_array[i] = usermodehelper_bset.cap[i];
630                 else if (table->data == CAP_PI)
631                         cap_array[i] = usermodehelper_inheritable.cap[i];
632                 else
633                         BUG();
634         }
635         spin_unlock(&umh_sysctl_lock);
636
637         t = *table;
638         t.data = &cap_array;
639
640         /*
641          * actually read or write and array of ulongs from userspace.  Remember
642          * these are least significant 32 bits first
643          */
644         err = proc_doulongvec_minmax(&t, write, buffer, lenp, ppos);
645         if (err < 0)
646                 return err;
647
648         /*
649          * convert from the sysctl array of ulongs to the kernel_cap_t
650          * internal representation
651          */
652         for (i = 0; i < _KERNEL_CAPABILITY_U32S; i++)
653                 new_cap.cap[i] = cap_array[i];
654
655         /*
656          * Drop everything not in the new_cap (but don't add things)
657          */
658         if (write) {
659                 spin_lock(&umh_sysctl_lock);
660                 if (table->data == CAP_BSET)
661                         usermodehelper_bset = cap_intersect(usermodehelper_bset, new_cap);
662                 if (table->data == CAP_PI)
663                         usermodehelper_inheritable = cap_intersect(usermodehelper_inheritable, new_cap);
664                 spin_unlock(&umh_sysctl_lock);
665         }
666
667         return 0;
668 }
669
670 struct ctl_table usermodehelper_table[] = {
671         {
672                 .procname       = "bset",
673                 .data           = CAP_BSET,
674                 .maxlen         = _KERNEL_CAPABILITY_U32S * sizeof(unsigned long),
675                 .mode           = 0600,
676                 .proc_handler   = proc_cap_handler,
677         },
678         {
679                 .procname       = "inheritable",
680                 .data           = CAP_PI,
681                 .maxlen         = _KERNEL_CAPABILITY_U32S * sizeof(unsigned long),
682                 .mode           = 0600,
683                 .proc_handler   = proc_cap_handler,
684         },
685         { }
686 };