Merge git://git.kernel.org/pub/scm/linux/kernel/git/pablo/nf-next
[linux-2.6-microblaze.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         if (sub_info->file)
103                 retval = do_execve_file(sub_info->file,
104                                         sub_info->argv, sub_info->envp);
105         else
106                 retval = do_execve(getname_kernel(sub_info->path),
107                                    (const char __user *const __user *)sub_info->argv,
108                                    (const char __user *const __user *)sub_info->envp);
109 out:
110         sub_info->retval = retval;
111         /*
112          * call_usermodehelper_exec_sync() will call umh_complete
113          * if UHM_WAIT_PROC.
114          */
115         if (!(sub_info->wait & UMH_WAIT_PROC))
116                 umh_complete(sub_info);
117         if (!retval)
118                 return 0;
119         do_exit(0);
120 }
121
122 /* Handles UMH_WAIT_PROC.  */
123 static void call_usermodehelper_exec_sync(struct subprocess_info *sub_info)
124 {
125         pid_t pid;
126
127         /* If SIGCLD is ignored kernel_wait4 won't populate the status. */
128         kernel_sigaction(SIGCHLD, SIG_DFL);
129         pid = kernel_thread(call_usermodehelper_exec_async, sub_info, SIGCHLD);
130         if (pid < 0) {
131                 sub_info->retval = pid;
132         } else {
133                 int ret = -ECHILD;
134                 /*
135                  * Normally it is bogus to call wait4() from in-kernel because
136                  * wait4() wants to write the exit code to a userspace address.
137                  * But call_usermodehelper_exec_sync() always runs as kernel
138                  * thread (workqueue) and put_user() to a kernel address works
139                  * OK for kernel threads, due to their having an mm_segment_t
140                  * which spans the entire address space.
141                  *
142                  * Thus the __user pointer cast is valid here.
143                  */
144                 kernel_wait4(pid, (int __user *)&ret, 0, NULL);
145
146                 /*
147                  * If ret is 0, either call_usermodehelper_exec_async failed and
148                  * the real error code is already in sub_info->retval or
149                  * sub_info->retval is 0 anyway, so don't mess with it then.
150                  */
151                 if (ret)
152                         sub_info->retval = ret;
153         }
154
155         /* Restore default kernel sig handler */
156         kernel_sigaction(SIGCHLD, SIG_IGN);
157
158         umh_complete(sub_info);
159 }
160
161 /*
162  * We need to create the usermodehelper kernel thread from a task that is affine
163  * to an optimized set of CPUs (or nohz housekeeping ones) such that they
164  * inherit a widest affinity irrespective of call_usermodehelper() callers with
165  * possibly reduced affinity (eg: per-cpu workqueues). We don't want
166  * usermodehelper targets to contend a busy CPU.
167  *
168  * Unbound workqueues provide such wide affinity and allow to block on
169  * UMH_WAIT_PROC requests without blocking pending request (up to some limit).
170  *
171  * Besides, workqueues provide the privilege level that caller might not have
172  * to perform the usermodehelper request.
173  *
174  */
175 static void call_usermodehelper_exec_work(struct work_struct *work)
176 {
177         struct subprocess_info *sub_info =
178                 container_of(work, struct subprocess_info, work);
179
180         if (sub_info->wait & UMH_WAIT_PROC) {
181                 call_usermodehelper_exec_sync(sub_info);
182         } else {
183                 pid_t pid;
184                 /*
185                  * Use CLONE_PARENT to reparent it to kthreadd; we do not
186                  * want to pollute current->children, and we need a parent
187                  * that always ignores SIGCHLD to ensure auto-reaping.
188                  */
189                 pid = kernel_thread(call_usermodehelper_exec_async, sub_info,
190                                     CLONE_PARENT | SIGCHLD);
191                 if (pid < 0) {
192                         sub_info->retval = pid;
193                         umh_complete(sub_info);
194                 } else {
195                         sub_info->pid = pid;
196                 }
197         }
198 }
199
200 /*
201  * If set, call_usermodehelper_exec() will exit immediately returning -EBUSY
202  * (used for preventing user land processes from being created after the user
203  * land has been frozen during a system-wide hibernation or suspend operation).
204  * Should always be manipulated under umhelper_sem acquired for write.
205  */
206 static enum umh_disable_depth usermodehelper_disabled = UMH_DISABLED;
207
208 /* Number of helpers running */
209 static atomic_t running_helpers = ATOMIC_INIT(0);
210
211 /*
212  * Wait queue head used by usermodehelper_disable() to wait for all running
213  * helpers to finish.
214  */
215 static DECLARE_WAIT_QUEUE_HEAD(running_helpers_waitq);
216
217 /*
218  * Used by usermodehelper_read_lock_wait() to wait for usermodehelper_disabled
219  * to become 'false'.
220  */
221 static DECLARE_WAIT_QUEUE_HEAD(usermodehelper_disabled_waitq);
222
223 /*
224  * Time to wait for running_helpers to become zero before the setting of
225  * usermodehelper_disabled in usermodehelper_disable() fails
226  */
227 #define RUNNING_HELPERS_TIMEOUT (5 * HZ)
228
229 int usermodehelper_read_trylock(void)
230 {
231         DEFINE_WAIT(wait);
232         int ret = 0;
233
234         down_read(&umhelper_sem);
235         for (;;) {
236                 prepare_to_wait(&usermodehelper_disabled_waitq, &wait,
237                                 TASK_INTERRUPTIBLE);
238                 if (!usermodehelper_disabled)
239                         break;
240
241                 if (usermodehelper_disabled == UMH_DISABLED)
242                         ret = -EAGAIN;
243
244                 up_read(&umhelper_sem);
245
246                 if (ret)
247                         break;
248
249                 schedule();
250                 try_to_freeze();
251
252                 down_read(&umhelper_sem);
253         }
254         finish_wait(&usermodehelper_disabled_waitq, &wait);
255         return ret;
256 }
257 EXPORT_SYMBOL_GPL(usermodehelper_read_trylock);
258
259 long usermodehelper_read_lock_wait(long timeout)
260 {
261         DEFINE_WAIT(wait);
262
263         if (timeout < 0)
264                 return -EINVAL;
265
266         down_read(&umhelper_sem);
267         for (;;) {
268                 prepare_to_wait(&usermodehelper_disabled_waitq, &wait,
269                                 TASK_UNINTERRUPTIBLE);
270                 if (!usermodehelper_disabled)
271                         break;
272
273                 up_read(&umhelper_sem);
274
275                 timeout = schedule_timeout(timeout);
276                 if (!timeout)
277                         break;
278
279                 down_read(&umhelper_sem);
280         }
281         finish_wait(&usermodehelper_disabled_waitq, &wait);
282         return timeout;
283 }
284 EXPORT_SYMBOL_GPL(usermodehelper_read_lock_wait);
285
286 void usermodehelper_read_unlock(void)
287 {
288         up_read(&umhelper_sem);
289 }
290 EXPORT_SYMBOL_GPL(usermodehelper_read_unlock);
291
292 /**
293  * __usermodehelper_set_disable_depth - Modify usermodehelper_disabled.
294  * @depth: New value to assign to usermodehelper_disabled.
295  *
296  * Change the value of usermodehelper_disabled (under umhelper_sem locked for
297  * writing) and wakeup tasks waiting for it to change.
298  */
299 void __usermodehelper_set_disable_depth(enum umh_disable_depth depth)
300 {
301         down_write(&umhelper_sem);
302         usermodehelper_disabled = depth;
303         wake_up(&usermodehelper_disabled_waitq);
304         up_write(&umhelper_sem);
305 }
306
307 /**
308  * __usermodehelper_disable - Prevent new helpers from being started.
309  * @depth: New value to assign to usermodehelper_disabled.
310  *
311  * Set usermodehelper_disabled to @depth and wait for running helpers to exit.
312  */
313 int __usermodehelper_disable(enum umh_disable_depth depth)
314 {
315         long retval;
316
317         if (!depth)
318                 return -EINVAL;
319
320         down_write(&umhelper_sem);
321         usermodehelper_disabled = depth;
322         up_write(&umhelper_sem);
323
324         /*
325          * From now on call_usermodehelper_exec() won't start any new
326          * helpers, so it is sufficient if running_helpers turns out to
327          * be zero at one point (it may be increased later, but that
328          * doesn't matter).
329          */
330         retval = wait_event_timeout(running_helpers_waitq,
331                                         atomic_read(&running_helpers) == 0,
332                                         RUNNING_HELPERS_TIMEOUT);
333         if (retval)
334                 return 0;
335
336         __usermodehelper_set_disable_depth(UMH_ENABLED);
337         return -EAGAIN;
338 }
339
340 static void helper_lock(void)
341 {
342         atomic_inc(&running_helpers);
343         smp_mb__after_atomic();
344 }
345
346 static void helper_unlock(void)
347 {
348         if (atomic_dec_and_test(&running_helpers))
349                 wake_up(&running_helpers_waitq);
350 }
351
352 /**
353  * call_usermodehelper_setup - prepare to call a usermode helper
354  * @path: path to usermode executable
355  * @argv: arg vector for process
356  * @envp: environment for process
357  * @gfp_mask: gfp mask for memory allocation
358  * @cleanup: a cleanup function
359  * @init: an init function
360  * @data: arbitrary context sensitive data
361  *
362  * Returns either %NULL on allocation failure, or a subprocess_info
363  * structure.  This should be passed to call_usermodehelper_exec to
364  * exec the process and free the structure.
365  *
366  * The init function is used to customize the helper process prior to
367  * exec.  A non-zero return code causes the process to error out, exit,
368  * and return the failure to the calling process
369  *
370  * The cleanup function is just before ethe subprocess_info is about to
371  * be freed.  This can be used for freeing the argv and envp.  The
372  * Function must be runnable in either a process context or the
373  * context in which call_usermodehelper_exec is called.
374  */
375 struct subprocess_info *call_usermodehelper_setup(const char *path, char **argv,
376                 char **envp, gfp_t gfp_mask,
377                 int (*init)(struct subprocess_info *info, struct cred *new),
378                 void (*cleanup)(struct subprocess_info *info),
379                 void *data)
380 {
381         struct subprocess_info *sub_info;
382         sub_info = kzalloc(sizeof(struct subprocess_info), gfp_mask);
383         if (!sub_info)
384                 goto out;
385
386         INIT_WORK(&sub_info->work, call_usermodehelper_exec_work);
387
388 #ifdef CONFIG_STATIC_USERMODEHELPER
389         sub_info->path = CONFIG_STATIC_USERMODEHELPER_PATH;
390 #else
391         sub_info->path = path;
392 #endif
393         sub_info->argv = argv;
394         sub_info->envp = envp;
395
396         sub_info->cleanup = cleanup;
397         sub_info->init = init;
398         sub_info->data = data;
399   out:
400         return sub_info;
401 }
402 EXPORT_SYMBOL(call_usermodehelper_setup);
403
404 struct subprocess_info *call_usermodehelper_setup_file(struct file *file,
405                 int (*init)(struct subprocess_info *info, struct cred *new),
406                 void (*cleanup)(struct subprocess_info *info), void *data)
407 {
408         struct subprocess_info *sub_info;
409
410         sub_info = kzalloc(sizeof(struct subprocess_info), GFP_KERNEL);
411         if (!sub_info)
412                 return NULL;
413
414         INIT_WORK(&sub_info->work, call_usermodehelper_exec_work);
415         sub_info->path = "none";
416         sub_info->file = file;
417         sub_info->init = init;
418         sub_info->cleanup = cleanup;
419         sub_info->data = data;
420         return sub_info;
421 }
422
423 static int umh_pipe_setup(struct subprocess_info *info, struct cred *new)
424 {
425         struct umh_info *umh_info = info->data;
426         struct file *from_umh[2];
427         struct file *to_umh[2];
428         int err;
429
430         /* create pipe to send data to umh */
431         err = create_pipe_files(to_umh, 0);
432         if (err)
433                 return err;
434         err = replace_fd(0, to_umh[0], 0);
435         fput(to_umh[0]);
436         if (err < 0) {
437                 fput(to_umh[1]);
438                 return err;
439         }
440
441         /* create pipe to receive data from umh */
442         err = create_pipe_files(from_umh, 0);
443         if (err) {
444                 fput(to_umh[1]);
445                 replace_fd(0, NULL, 0);
446                 return err;
447         }
448         err = replace_fd(1, from_umh[1], 0);
449         fput(from_umh[1]);
450         if (err < 0) {
451                 fput(to_umh[1]);
452                 replace_fd(0, NULL, 0);
453                 fput(from_umh[0]);
454                 return err;
455         }
456
457         umh_info->pipe_to_umh = to_umh[1];
458         umh_info->pipe_from_umh = from_umh[0];
459         return 0;
460 }
461
462 static void umh_save_pid(struct subprocess_info *info)
463 {
464         struct umh_info *umh_info = info->data;
465
466         umh_info->pid = info->pid;
467 }
468
469 /**
470  * fork_usermode_blob - fork a blob of bytes as a usermode process
471  * @data: a blob of bytes that can be do_execv-ed as a file
472  * @len: length of the blob
473  * @info: information about usermode process (shouldn't be NULL)
474  *
475  * Returns either negative error or zero which indicates success
476  * in executing a blob of bytes as a usermode process. In such
477  * case 'struct umh_info *info' is populated with two pipes
478  * and a pid of the process. The caller is responsible for health
479  * check of the user process, killing it via pid, and closing the
480  * pipes when user process is no longer needed.
481  */
482 int fork_usermode_blob(void *data, size_t len, struct umh_info *info)
483 {
484         struct subprocess_info *sub_info;
485         struct file *file;
486         ssize_t written;
487         loff_t pos = 0;
488         int err;
489
490         file = shmem_kernel_file_setup("", len, 0);
491         if (IS_ERR(file))
492                 return PTR_ERR(file);
493
494         written = kernel_write(file, data, len, &pos);
495         if (written != len) {
496                 err = written;
497                 if (err >= 0)
498                         err = -ENOMEM;
499                 goto out;
500         }
501
502         err = -ENOMEM;
503         sub_info = call_usermodehelper_setup_file(file, umh_pipe_setup,
504                                                   umh_save_pid, info);
505         if (!sub_info)
506                 goto out;
507
508         err = call_usermodehelper_exec(sub_info, UMH_WAIT_EXEC);
509 out:
510         fput(file);
511         return err;
512 }
513 EXPORT_SYMBOL_GPL(fork_usermode_blob);
514
515 /**
516  * call_usermodehelper_exec - start a usermode application
517  * @sub_info: information about the subprocessa
518  * @wait: wait for the application to finish and return status.
519  *        when UMH_NO_WAIT don't wait at all, but you get no useful error back
520  *        when the program couldn't be exec'ed. This makes it safe to call
521  *        from interrupt context.
522  *
523  * Runs a user-space application.  The application is started
524  * asynchronously if wait is not set, and runs as a child of system workqueues.
525  * (ie. it runs with full root capabilities and optimized affinity).
526  */
527 int call_usermodehelper_exec(struct subprocess_info *sub_info, int wait)
528 {
529         DECLARE_COMPLETION_ONSTACK(done);
530         int retval = 0;
531
532         if (!sub_info->path) {
533                 call_usermodehelper_freeinfo(sub_info);
534                 return -EINVAL;
535         }
536         helper_lock();
537         if (usermodehelper_disabled) {
538                 retval = -EBUSY;
539                 goto out;
540         }
541
542         /*
543          * If there is no binary for us to call, then just return and get out of
544          * here.  This allows us to set STATIC_USERMODEHELPER_PATH to "" and
545          * disable all call_usermodehelper() calls.
546          */
547         if (strlen(sub_info->path) == 0)
548                 goto out;
549
550         /*
551          * Set the completion pointer only if there is a waiter.
552          * This makes it possible to use umh_complete to free
553          * the data structure in case of UMH_NO_WAIT.
554          */
555         sub_info->complete = (wait == UMH_NO_WAIT) ? NULL : &done;
556         sub_info->wait = wait;
557
558         queue_work(system_unbound_wq, &sub_info->work);
559         if (wait == UMH_NO_WAIT)        /* task has freed sub_info */
560                 goto unlock;
561
562         if (wait & UMH_KILLABLE) {
563                 retval = wait_for_completion_killable(&done);
564                 if (!retval)
565                         goto wait_done;
566
567                 /* umh_complete() will see NULL and free sub_info */
568                 if (xchg(&sub_info->complete, NULL))
569                         goto unlock;
570                 /* fallthrough, umh_complete() was already called */
571         }
572
573         wait_for_completion(&done);
574 wait_done:
575         retval = sub_info->retval;
576 out:
577         call_usermodehelper_freeinfo(sub_info);
578 unlock:
579         helper_unlock();
580         return retval;
581 }
582 EXPORT_SYMBOL(call_usermodehelper_exec);
583
584 /**
585  * call_usermodehelper() - prepare and start a usermode application
586  * @path: path to usermode executable
587  * @argv: arg vector for process
588  * @envp: environment for process
589  * @wait: wait for the application to finish and return status.
590  *        when UMH_NO_WAIT don't wait at all, but you get no useful error back
591  *        when the program couldn't be exec'ed. This makes it safe to call
592  *        from interrupt context.
593  *
594  * This function is the equivalent to use call_usermodehelper_setup() and
595  * call_usermodehelper_exec().
596  */
597 int call_usermodehelper(const char *path, char **argv, char **envp, int wait)
598 {
599         struct subprocess_info *info;
600         gfp_t gfp_mask = (wait == UMH_NO_WAIT) ? GFP_ATOMIC : GFP_KERNEL;
601
602         info = call_usermodehelper_setup(path, argv, envp, gfp_mask,
603                                          NULL, NULL, NULL);
604         if (info == NULL)
605                 return -ENOMEM;
606
607         return call_usermodehelper_exec(info, wait);
608 }
609 EXPORT_SYMBOL(call_usermodehelper);
610
611 static int proc_cap_handler(struct ctl_table *table, int write,
612                          void __user *buffer, size_t *lenp, loff_t *ppos)
613 {
614         struct ctl_table t;
615         unsigned long cap_array[_KERNEL_CAPABILITY_U32S];
616         kernel_cap_t new_cap;
617         int err, i;
618
619         if (write && (!capable(CAP_SETPCAP) ||
620                       !capable(CAP_SYS_MODULE)))
621                 return -EPERM;
622
623         /*
624          * convert from the global kernel_cap_t to the ulong array to print to
625          * userspace if this is a read.
626          */
627         spin_lock(&umh_sysctl_lock);
628         for (i = 0; i < _KERNEL_CAPABILITY_U32S; i++)  {
629                 if (table->data == CAP_BSET)
630                         cap_array[i] = usermodehelper_bset.cap[i];
631                 else if (table->data == CAP_PI)
632                         cap_array[i] = usermodehelper_inheritable.cap[i];
633                 else
634                         BUG();
635         }
636         spin_unlock(&umh_sysctl_lock);
637
638         t = *table;
639         t.data = &cap_array;
640
641         /*
642          * actually read or write and array of ulongs from userspace.  Remember
643          * these are least significant 32 bits first
644          */
645         err = proc_doulongvec_minmax(&t, write, buffer, lenp, ppos);
646         if (err < 0)
647                 return err;
648
649         /*
650          * convert from the sysctl array of ulongs to the kernel_cap_t
651          * internal representation
652          */
653         for (i = 0; i < _KERNEL_CAPABILITY_U32S; i++)
654                 new_cap.cap[i] = cap_array[i];
655
656         /*
657          * Drop everything not in the new_cap (but don't add things)
658          */
659         if (write) {
660                 spin_lock(&umh_sysctl_lock);
661                 if (table->data == CAP_BSET)
662                         usermodehelper_bset = cap_intersect(usermodehelper_bset, new_cap);
663                 if (table->data == CAP_PI)
664                         usermodehelper_inheritable = cap_intersect(usermodehelper_inheritable, new_cap);
665                 spin_unlock(&umh_sysctl_lock);
666         }
667
668         return 0;
669 }
670
671 struct ctl_table usermodehelper_table[] = {
672         {
673                 .procname       = "bset",
674                 .data           = CAP_BSET,
675                 .maxlen         = _KERNEL_CAPABILITY_U32S * sizeof(unsigned long),
676                 .mode           = 0600,
677                 .proc_handler   = proc_cap_handler,
678         },
679         {
680                 .procname       = "inheritable",
681                 .data           = CAP_PI,
682                 .maxlen         = _KERNEL_CAPABILITY_U32S * sizeof(unsigned long),
683                 .mode           = 0600,
684                 .proc_handler   = proc_cap_handler,
685         },
686         { }
687 };