kmod: split out umh code into its own file
[sfrench/cifs-2.6.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
29 #include <trace/events/module.h>
30
31 #define CAP_BSET        (void *)1
32 #define CAP_PI          (void *)2
33
34 static kernel_cap_t usermodehelper_bset = CAP_FULL_SET;
35 static kernel_cap_t usermodehelper_inheritable = CAP_FULL_SET;
36 static DEFINE_SPINLOCK(umh_sysctl_lock);
37 static DECLARE_RWSEM(umhelper_sem);
38
39 static void call_usermodehelper_freeinfo(struct subprocess_info *info)
40 {
41         if (info->cleanup)
42                 (*info->cleanup)(info);
43         kfree(info);
44 }
45
46 static void umh_complete(struct subprocess_info *sub_info)
47 {
48         struct completion *comp = xchg(&sub_info->complete, NULL);
49         /*
50          * See call_usermodehelper_exec(). If xchg() returns NULL
51          * we own sub_info, the UMH_KILLABLE caller has gone away
52          * or the caller used UMH_NO_WAIT.
53          */
54         if (comp)
55                 complete(comp);
56         else
57                 call_usermodehelper_freeinfo(sub_info);
58 }
59
60 /*
61  * This is the task which runs the usermode application
62  */
63 static int call_usermodehelper_exec_async(void *data)
64 {
65         struct subprocess_info *sub_info = data;
66         struct cred *new;
67         int retval;
68
69         spin_lock_irq(&current->sighand->siglock);
70         flush_signal_handlers(current, 1);
71         spin_unlock_irq(&current->sighand->siglock);
72
73         /*
74          * Our parent (unbound workqueue) runs with elevated scheduling
75          * priority. Avoid propagating that into the userspace child.
76          */
77         set_user_nice(current, 0);
78
79         retval = -ENOMEM;
80         new = prepare_kernel_cred(current);
81         if (!new)
82                 goto out;
83
84         spin_lock(&umh_sysctl_lock);
85         new->cap_bset = cap_intersect(usermodehelper_bset, new->cap_bset);
86         new->cap_inheritable = cap_intersect(usermodehelper_inheritable,
87                                              new->cap_inheritable);
88         spin_unlock(&umh_sysctl_lock);
89
90         if (sub_info->init) {
91                 retval = sub_info->init(sub_info, new);
92                 if (retval) {
93                         abort_creds(new);
94                         goto out;
95                 }
96         }
97
98         commit_creds(new);
99
100         retval = do_execve(getname_kernel(sub_info->path),
101                            (const char __user *const __user *)sub_info->argv,
102                            (const char __user *const __user *)sub_info->envp);
103 out:
104         sub_info->retval = retval;
105         /*
106          * call_usermodehelper_exec_sync() will call umh_complete
107          * if UHM_WAIT_PROC.
108          */
109         if (!(sub_info->wait & UMH_WAIT_PROC))
110                 umh_complete(sub_info);
111         if (!retval)
112                 return 0;
113         do_exit(0);
114 }
115
116 /* Handles UMH_WAIT_PROC.  */
117 static void call_usermodehelper_exec_sync(struct subprocess_info *sub_info)
118 {
119         pid_t pid;
120
121         /* If SIGCLD is ignored sys_wait4 won't populate the status. */
122         kernel_sigaction(SIGCHLD, SIG_DFL);
123         pid = kernel_thread(call_usermodehelper_exec_async, sub_info, SIGCHLD);
124         if (pid < 0) {
125                 sub_info->retval = pid;
126         } else {
127                 int ret = -ECHILD;
128                 /*
129                  * Normally it is bogus to call wait4() from in-kernel because
130                  * wait4() wants to write the exit code to a userspace address.
131                  * But call_usermodehelper_exec_sync() always runs as kernel
132                  * thread (workqueue) and put_user() to a kernel address works
133                  * OK for kernel threads, due to their having an mm_segment_t
134                  * which spans the entire address space.
135                  *
136                  * Thus the __user pointer cast is valid here.
137                  */
138                 sys_wait4(pid, (int __user *)&ret, 0, NULL);
139
140                 /*
141                  * If ret is 0, either call_usermodehelper_exec_async failed and
142                  * the real error code is already in sub_info->retval or
143                  * sub_info->retval is 0 anyway, so don't mess with it then.
144                  */
145                 if (ret)
146                         sub_info->retval = ret;
147         }
148
149         /* Restore default kernel sig handler */
150         kernel_sigaction(SIGCHLD, SIG_IGN);
151
152         umh_complete(sub_info);
153 }
154
155 /*
156  * We need to create the usermodehelper kernel thread from a task that is affine
157  * to an optimized set of CPUs (or nohz housekeeping ones) such that they
158  * inherit a widest affinity irrespective of call_usermodehelper() callers with
159  * possibly reduced affinity (eg: per-cpu workqueues). We don't want
160  * usermodehelper targets to contend a busy CPU.
161  *
162  * Unbound workqueues provide such wide affinity and allow to block on
163  * UMH_WAIT_PROC requests without blocking pending request (up to some limit).
164  *
165  * Besides, workqueues provide the privilege level that caller might not have
166  * to perform the usermodehelper request.
167  *
168  */
169 static void call_usermodehelper_exec_work(struct work_struct *work)
170 {
171         struct subprocess_info *sub_info =
172                 container_of(work, struct subprocess_info, work);
173
174         if (sub_info->wait & UMH_WAIT_PROC) {
175                 call_usermodehelper_exec_sync(sub_info);
176         } else {
177                 pid_t pid;
178                 /*
179                  * Use CLONE_PARENT to reparent it to kthreadd; we do not
180                  * want to pollute current->children, and we need a parent
181                  * that always ignores SIGCHLD to ensure auto-reaping.
182                  */
183                 pid = kernel_thread(call_usermodehelper_exec_async, sub_info,
184                                     CLONE_PARENT | SIGCHLD);
185                 if (pid < 0) {
186                         sub_info->retval = pid;
187                         umh_complete(sub_info);
188                 }
189         }
190 }
191
192 /*
193  * If set, call_usermodehelper_exec() will exit immediately returning -EBUSY
194  * (used for preventing user land processes from being created after the user
195  * land has been frozen during a system-wide hibernation or suspend operation).
196  * Should always be manipulated under umhelper_sem acquired for write.
197  */
198 static enum umh_disable_depth usermodehelper_disabled = UMH_DISABLED;
199
200 /* Number of helpers running */
201 static atomic_t running_helpers = ATOMIC_INIT(0);
202
203 /*
204  * Wait queue head used by usermodehelper_disable() to wait for all running
205  * helpers to finish.
206  */
207 static DECLARE_WAIT_QUEUE_HEAD(running_helpers_waitq);
208
209 /*
210  * Used by usermodehelper_read_lock_wait() to wait for usermodehelper_disabled
211  * to become 'false'.
212  */
213 static DECLARE_WAIT_QUEUE_HEAD(usermodehelper_disabled_waitq);
214
215 /*
216  * Time to wait for running_helpers to become zero before the setting of
217  * usermodehelper_disabled in usermodehelper_disable() fails
218  */
219 #define RUNNING_HELPERS_TIMEOUT (5 * HZ)
220
221 int usermodehelper_read_trylock(void)
222 {
223         DEFINE_WAIT(wait);
224         int ret = 0;
225
226         down_read(&umhelper_sem);
227         for (;;) {
228                 prepare_to_wait(&usermodehelper_disabled_waitq, &wait,
229                                 TASK_INTERRUPTIBLE);
230                 if (!usermodehelper_disabled)
231                         break;
232
233                 if (usermodehelper_disabled == UMH_DISABLED)
234                         ret = -EAGAIN;
235
236                 up_read(&umhelper_sem);
237
238                 if (ret)
239                         break;
240
241                 schedule();
242                 try_to_freeze();
243
244                 down_read(&umhelper_sem);
245         }
246         finish_wait(&usermodehelper_disabled_waitq, &wait);
247         return ret;
248 }
249 EXPORT_SYMBOL_GPL(usermodehelper_read_trylock);
250
251 long usermodehelper_read_lock_wait(long timeout)
252 {
253         DEFINE_WAIT(wait);
254
255         if (timeout < 0)
256                 return -EINVAL;
257
258         down_read(&umhelper_sem);
259         for (;;) {
260                 prepare_to_wait(&usermodehelper_disabled_waitq, &wait,
261                                 TASK_UNINTERRUPTIBLE);
262                 if (!usermodehelper_disabled)
263                         break;
264
265                 up_read(&umhelper_sem);
266
267                 timeout = schedule_timeout(timeout);
268                 if (!timeout)
269                         break;
270
271                 down_read(&umhelper_sem);
272         }
273         finish_wait(&usermodehelper_disabled_waitq, &wait);
274         return timeout;
275 }
276 EXPORT_SYMBOL_GPL(usermodehelper_read_lock_wait);
277
278 void usermodehelper_read_unlock(void)
279 {
280         up_read(&umhelper_sem);
281 }
282 EXPORT_SYMBOL_GPL(usermodehelper_read_unlock);
283
284 /**
285  * __usermodehelper_set_disable_depth - Modify usermodehelper_disabled.
286  * @depth: New value to assign to usermodehelper_disabled.
287  *
288  * Change the value of usermodehelper_disabled (under umhelper_sem locked for
289  * writing) and wakeup tasks waiting for it to change.
290  */
291 void __usermodehelper_set_disable_depth(enum umh_disable_depth depth)
292 {
293         down_write(&umhelper_sem);
294         usermodehelper_disabled = depth;
295         wake_up(&usermodehelper_disabled_waitq);
296         up_write(&umhelper_sem);
297 }
298
299 /**
300  * __usermodehelper_disable - Prevent new helpers from being started.
301  * @depth: New value to assign to usermodehelper_disabled.
302  *
303  * Set usermodehelper_disabled to @depth and wait for running helpers to exit.
304  */
305 int __usermodehelper_disable(enum umh_disable_depth depth)
306 {
307         long retval;
308
309         if (!depth)
310                 return -EINVAL;
311
312         down_write(&umhelper_sem);
313         usermodehelper_disabled = depth;
314         up_write(&umhelper_sem);
315
316         /*
317          * From now on call_usermodehelper_exec() won't start any new
318          * helpers, so it is sufficient if running_helpers turns out to
319          * be zero at one point (it may be increased later, but that
320          * doesn't matter).
321          */
322         retval = wait_event_timeout(running_helpers_waitq,
323                                         atomic_read(&running_helpers) == 0,
324                                         RUNNING_HELPERS_TIMEOUT);
325         if (retval)
326                 return 0;
327
328         __usermodehelper_set_disable_depth(UMH_ENABLED);
329         return -EAGAIN;
330 }
331
332 static void helper_lock(void)
333 {
334         atomic_inc(&running_helpers);
335         smp_mb__after_atomic();
336 }
337
338 static void helper_unlock(void)
339 {
340         if (atomic_dec_and_test(&running_helpers))
341                 wake_up(&running_helpers_waitq);
342 }
343
344 /**
345  * call_usermodehelper_setup - prepare to call a usermode helper
346  * @path: path to usermode executable
347  * @argv: arg vector for process
348  * @envp: environment for process
349  * @gfp_mask: gfp mask for memory allocation
350  * @cleanup: a cleanup function
351  * @init: an init function
352  * @data: arbitrary context sensitive data
353  *
354  * Returns either %NULL on allocation failure, or a subprocess_info
355  * structure.  This should be passed to call_usermodehelper_exec to
356  * exec the process and free the structure.
357  *
358  * The init function is used to customize the helper process prior to
359  * exec.  A non-zero return code causes the process to error out, exit,
360  * and return the failure to the calling process
361  *
362  * The cleanup function is just before ethe subprocess_info is about to
363  * be freed.  This can be used for freeing the argv and envp.  The
364  * Function must be runnable in either a process context or the
365  * context in which call_usermodehelper_exec is called.
366  */
367 struct subprocess_info *call_usermodehelper_setup(const char *path, char **argv,
368                 char **envp, gfp_t gfp_mask,
369                 int (*init)(struct subprocess_info *info, struct cred *new),
370                 void (*cleanup)(struct subprocess_info *info),
371                 void *data)
372 {
373         struct subprocess_info *sub_info;
374         sub_info = kzalloc(sizeof(struct subprocess_info), gfp_mask);
375         if (!sub_info)
376                 goto out;
377
378         INIT_WORK(&sub_info->work, call_usermodehelper_exec_work);
379
380 #ifdef CONFIG_STATIC_USERMODEHELPER
381         sub_info->path = CONFIG_STATIC_USERMODEHELPER_PATH;
382 #else
383         sub_info->path = path;
384 #endif
385         sub_info->argv = argv;
386         sub_info->envp = envp;
387
388         sub_info->cleanup = cleanup;
389         sub_info->init = init;
390         sub_info->data = data;
391   out:
392         return sub_info;
393 }
394 EXPORT_SYMBOL(call_usermodehelper_setup);
395
396 /**
397  * call_usermodehelper_exec - start a usermode application
398  * @sub_info: information about the subprocessa
399  * @wait: wait for the application to finish and return status.
400  *        when UMH_NO_WAIT don't wait at all, but you get no useful error back
401  *        when the program couldn't be exec'ed. This makes it safe to call
402  *        from interrupt context.
403  *
404  * Runs a user-space application.  The application is started
405  * asynchronously if wait is not set, and runs as a child of system workqueues.
406  * (ie. it runs with full root capabilities and optimized affinity).
407  */
408 int call_usermodehelper_exec(struct subprocess_info *sub_info, int wait)
409 {
410         DECLARE_COMPLETION_ONSTACK(done);
411         int retval = 0;
412
413         if (!sub_info->path) {
414                 call_usermodehelper_freeinfo(sub_info);
415                 return -EINVAL;
416         }
417         helper_lock();
418         if (usermodehelper_disabled) {
419                 retval = -EBUSY;
420                 goto out;
421         }
422
423         /*
424          * If there is no binary for us to call, then just return and get out of
425          * here.  This allows us to set STATIC_USERMODEHELPER_PATH to "" and
426          * disable all call_usermodehelper() calls.
427          */
428         if (strlen(sub_info->path) == 0)
429                 goto out;
430
431         /*
432          * Set the completion pointer only if there is a waiter.
433          * This makes it possible to use umh_complete to free
434          * the data structure in case of UMH_NO_WAIT.
435          */
436         sub_info->complete = (wait == UMH_NO_WAIT) ? NULL : &done;
437         sub_info->wait = wait;
438
439         queue_work(system_unbound_wq, &sub_info->work);
440         if (wait == UMH_NO_WAIT)        /* task has freed sub_info */
441                 goto unlock;
442
443         if (wait & UMH_KILLABLE) {
444                 retval = wait_for_completion_killable(&done);
445                 if (!retval)
446                         goto wait_done;
447
448                 /* umh_complete() will see NULL and free sub_info */
449                 if (xchg(&sub_info->complete, NULL))
450                         goto unlock;
451                 /* fallthrough, umh_complete() was already called */
452         }
453
454         wait_for_completion(&done);
455 wait_done:
456         retval = sub_info->retval;
457 out:
458         call_usermodehelper_freeinfo(sub_info);
459 unlock:
460         helper_unlock();
461         return retval;
462 }
463 EXPORT_SYMBOL(call_usermodehelper_exec);
464
465 /**
466  * call_usermodehelper() - prepare and start a usermode application
467  * @path: path to usermode executable
468  * @argv: arg vector for process
469  * @envp: environment for process
470  * @wait: wait for the application to finish and return status.
471  *        when UMH_NO_WAIT don't wait at all, but you get no useful error back
472  *        when the program couldn't be exec'ed. This makes it safe to call
473  *        from interrupt context.
474  *
475  * This function is the equivalent to use call_usermodehelper_setup() and
476  * call_usermodehelper_exec().
477  */
478 int call_usermodehelper(const char *path, char **argv, char **envp, int wait)
479 {
480         struct subprocess_info *info;
481         gfp_t gfp_mask = (wait == UMH_NO_WAIT) ? GFP_ATOMIC : GFP_KERNEL;
482
483         info = call_usermodehelper_setup(path, argv, envp, gfp_mask,
484                                          NULL, NULL, NULL);
485         if (info == NULL)
486                 return -ENOMEM;
487
488         return call_usermodehelper_exec(info, wait);
489 }
490 EXPORT_SYMBOL(call_usermodehelper);
491
492 static int proc_cap_handler(struct ctl_table *table, int write,
493                          void __user *buffer, size_t *lenp, loff_t *ppos)
494 {
495         struct ctl_table t;
496         unsigned long cap_array[_KERNEL_CAPABILITY_U32S];
497         kernel_cap_t new_cap;
498         int err, i;
499
500         if (write && (!capable(CAP_SETPCAP) ||
501                       !capable(CAP_SYS_MODULE)))
502                 return -EPERM;
503
504         /*
505          * convert from the global kernel_cap_t to the ulong array to print to
506          * userspace if this is a read.
507          */
508         spin_lock(&umh_sysctl_lock);
509         for (i = 0; i < _KERNEL_CAPABILITY_U32S; i++)  {
510                 if (table->data == CAP_BSET)
511                         cap_array[i] = usermodehelper_bset.cap[i];
512                 else if (table->data == CAP_PI)
513                         cap_array[i] = usermodehelper_inheritable.cap[i];
514                 else
515                         BUG();
516         }
517         spin_unlock(&umh_sysctl_lock);
518
519         t = *table;
520         t.data = &cap_array;
521
522         /*
523          * actually read or write and array of ulongs from userspace.  Remember
524          * these are least significant 32 bits first
525          */
526         err = proc_doulongvec_minmax(&t, write, buffer, lenp, ppos);
527         if (err < 0)
528                 return err;
529
530         /*
531          * convert from the sysctl array of ulongs to the kernel_cap_t
532          * internal representation
533          */
534         for (i = 0; i < _KERNEL_CAPABILITY_U32S; i++)
535                 new_cap.cap[i] = cap_array[i];
536
537         /*
538          * Drop everything not in the new_cap (but don't add things)
539          */
540         spin_lock(&umh_sysctl_lock);
541         if (write) {
542                 if (table->data == CAP_BSET)
543                         usermodehelper_bset = cap_intersect(usermodehelper_bset, new_cap);
544                 if (table->data == CAP_PI)
545                         usermodehelper_inheritable = cap_intersect(usermodehelper_inheritable, new_cap);
546         }
547         spin_unlock(&umh_sysctl_lock);
548
549         return 0;
550 }
551
552 struct ctl_table usermodehelper_table[] = {
553         {
554                 .procname       = "bset",
555                 .data           = CAP_BSET,
556                 .maxlen         = _KERNEL_CAPABILITY_U32S * sizeof(unsigned long),
557                 .mode           = 0600,
558                 .proc_handler   = proc_cap_handler,
559         },
560         {
561                 .procname       = "inheritable",
562                 .data           = CAP_PI,
563                 .maxlen         = _KERNEL_CAPABILITY_U32S * sizeof(unsigned long),
564                 .mode           = 0600,
565                 .proc_handler   = proc_cap_handler,
566         },
567         { }
568 };