Merge tag 'please-pull-naveen' of git://git.kernel.org/pub/scm/linux/kernel/git/ras...
[linux-3.10.git] / kernel / kmod.c
1 /*
2         kmod, the new module loader (replaces kerneld)
3         Kirk Petersen
4
5         Reorganized not to be a daemon by Adam Richter, with guidance
6         from Greg Zornetzer.
7
8         Modified to avoid chroot and file sharing problems.
9         Mikael Pettersson
10
11         Limit the concurrent number of kmod modprobes to catch loops from
12         "modprobe needs a service that is in a module".
13         Keith Owens <kaos@ocs.com.au> December 1999
14
15         Unblock all signals when we exec a usermode process.
16         Shuu Yamaguchi <shuu@wondernetworkresources.com> December 2000
17
18         call_usermodehelper wait flag, and remove exec_usermodehelper.
19         Rusty Russell <rusty@rustcorp.com.au>  Jan 2003
20 */
21 #include <linux/module.h>
22 #include <linux/sched.h>
23 #include <linux/syscalls.h>
24 #include <linux/unistd.h>
25 #include <linux/kmod.h>
26 #include <linux/slab.h>
27 #include <linux/completion.h>
28 #include <linux/cred.h>
29 #include <linux/file.h>
30 #include <linux/fdtable.h>
31 #include <linux/workqueue.h>
32 #include <linux/security.h>
33 #include <linux/mount.h>
34 #include <linux/kernel.h>
35 #include <linux/init.h>
36 #include <linux/resource.h>
37 #include <linux/notifier.h>
38 #include <linux/suspend.h>
39 #include <linux/rwsem.h>
40 #include <asm/uaccess.h>
41
42 #include <trace/events/module.h>
43
44 extern int max_threads;
45
46 static struct workqueue_struct *khelper_wq;
47
48 /*
49  * kmod_thread_locker is used for deadlock avoidance.  There is no explicit
50  * locking to protect this global - it is private to the singleton khelper
51  * thread and should only ever be modified by that thread.
52  */
53 static const struct task_struct *kmod_thread_locker;
54
55 #define CAP_BSET        (void *)1
56 #define CAP_PI          (void *)2
57
58 static kernel_cap_t usermodehelper_bset = CAP_FULL_SET;
59 static kernel_cap_t usermodehelper_inheritable = CAP_FULL_SET;
60 static DEFINE_SPINLOCK(umh_sysctl_lock);
61 static DECLARE_RWSEM(umhelper_sem);
62
63 #ifdef CONFIG_MODULES
64
65 /*
66         modprobe_path is set via /proc/sys.
67 */
68 char modprobe_path[KMOD_PATH_LEN] = "/sbin/modprobe";
69
70 static void free_modprobe_argv(struct subprocess_info *info)
71 {
72         kfree(info->argv[3]); /* check call_modprobe() */
73         kfree(info->argv);
74 }
75
76 static int call_modprobe(char *module_name, int wait)
77 {
78         static char *envp[] = {
79                 "HOME=/",
80                 "TERM=linux",
81                 "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
82                 NULL
83         };
84
85         char **argv = kmalloc(sizeof(char *[5]), GFP_KERNEL);
86         if (!argv)
87                 goto out;
88
89         module_name = kstrdup(module_name, GFP_KERNEL);
90         if (!module_name)
91                 goto free_argv;
92
93         argv[0] = modprobe_path;
94         argv[1] = "-q";
95         argv[2] = "--";
96         argv[3] = module_name;  /* check free_modprobe_argv() */
97         argv[4] = NULL;
98
99         return call_usermodehelper_fns(modprobe_path, argv, envp,
100                 wait | UMH_KILLABLE, NULL, free_modprobe_argv, NULL);
101 free_argv:
102         kfree(argv);
103 out:
104         return -ENOMEM;
105 }
106
107 /**
108  * __request_module - try to load a kernel module
109  * @wait: wait (or not) for the operation to complete
110  * @fmt: printf style format string for the name of the module
111  * @...: arguments as specified in the format string
112  *
113  * Load a module using the user mode module loader. The function returns
114  * zero on success or a negative errno code on failure. Note that a
115  * successful module load does not mean the module did not then unload
116  * and exit on an error of its own. Callers must check that the service
117  * they requested is now available not blindly invoke it.
118  *
119  * If module auto-loading support is disabled then this function
120  * becomes a no-operation.
121  */
122 int __request_module(bool wait, const char *fmt, ...)
123 {
124         va_list args;
125         char module_name[MODULE_NAME_LEN];
126         unsigned int max_modprobes;
127         int ret;
128         static atomic_t kmod_concurrent = ATOMIC_INIT(0);
129 #define MAX_KMOD_CONCURRENT 50  /* Completely arbitrary value - KAO */
130         static int kmod_loop_msg;
131
132         va_start(args, fmt);
133         ret = vsnprintf(module_name, MODULE_NAME_LEN, fmt, args);
134         va_end(args);
135         if (ret >= MODULE_NAME_LEN)
136                 return -ENAMETOOLONG;
137
138         ret = security_kernel_module_request(module_name);
139         if (ret)
140                 return ret;
141
142         /* If modprobe needs a service that is in a module, we get a recursive
143          * loop.  Limit the number of running kmod threads to max_threads/2 or
144          * MAX_KMOD_CONCURRENT, whichever is the smaller.  A cleaner method
145          * would be to run the parents of this process, counting how many times
146          * kmod was invoked.  That would mean accessing the internals of the
147          * process tables to get the command line, proc_pid_cmdline is static
148          * and it is not worth changing the proc code just to handle this case. 
149          * KAO.
150          *
151          * "trace the ppid" is simple, but will fail if someone's
152          * parent exits.  I think this is as good as it gets. --RR
153          */
154         max_modprobes = min(max_threads/2, MAX_KMOD_CONCURRENT);
155         atomic_inc(&kmod_concurrent);
156         if (atomic_read(&kmod_concurrent) > max_modprobes) {
157                 /* We may be blaming an innocent here, but unlikely */
158                 if (kmod_loop_msg < 5) {
159                         printk(KERN_ERR
160                                "request_module: runaway loop modprobe %s\n",
161                                module_name);
162                         kmod_loop_msg++;
163                 }
164                 atomic_dec(&kmod_concurrent);
165                 return -ENOMEM;
166         }
167
168         trace_module_request(module_name, wait, _RET_IP_);
169
170         ret = call_modprobe(module_name, wait ? UMH_WAIT_PROC : UMH_WAIT_EXEC);
171
172         atomic_dec(&kmod_concurrent);
173         return ret;
174 }
175 EXPORT_SYMBOL(__request_module);
176 #endif /* CONFIG_MODULES */
177
178 /*
179  * This is the task which runs the usermode application
180  */
181 static int ____call_usermodehelper(void *data)
182 {
183         struct subprocess_info *sub_info = data;
184         struct cred *new;
185         int retval;
186
187         spin_lock_irq(&current->sighand->siglock);
188         flush_signal_handlers(current, 1);
189         spin_unlock_irq(&current->sighand->siglock);
190
191         /* We can run anywhere, unlike our parent keventd(). */
192         set_cpus_allowed_ptr(current, cpu_all_mask);
193
194         /*
195          * Our parent is keventd, which runs with elevated scheduling priority.
196          * Avoid propagating that into the userspace child.
197          */
198         set_user_nice(current, 0);
199
200         retval = -ENOMEM;
201         new = prepare_kernel_cred(current);
202         if (!new)
203                 goto fail;
204
205         spin_lock(&umh_sysctl_lock);
206         new->cap_bset = cap_intersect(usermodehelper_bset, new->cap_bset);
207         new->cap_inheritable = cap_intersect(usermodehelper_inheritable,
208                                              new->cap_inheritable);
209         spin_unlock(&umh_sysctl_lock);
210
211         if (sub_info->init) {
212                 retval = sub_info->init(sub_info, new);
213                 if (retval) {
214                         abort_creds(new);
215                         goto fail;
216                 }
217         }
218
219         commit_creds(new);
220
221         retval = kernel_execve(sub_info->path,
222                                (const char *const *)sub_info->argv,
223                                (const char *const *)sub_info->envp);
224
225         /* Exec failed? */
226 fail:
227         sub_info->retval = retval;
228         return 0;
229 }
230
231 static int call_helper(void *data)
232 {
233         /* Worker thread started blocking khelper thread. */
234         kmod_thread_locker = current;
235         return ____call_usermodehelper(data);
236 }
237
238 static void call_usermodehelper_freeinfo(struct subprocess_info *info)
239 {
240         if (info->cleanup)
241                 (*info->cleanup)(info);
242         kfree(info);
243 }
244
245 static void umh_complete(struct subprocess_info *sub_info)
246 {
247         struct completion *comp = xchg(&sub_info->complete, NULL);
248         /*
249          * See call_usermodehelper_exec(). If xchg() returns NULL
250          * we own sub_info, the UMH_KILLABLE caller has gone away.
251          */
252         if (comp)
253                 complete(comp);
254         else
255                 call_usermodehelper_freeinfo(sub_info);
256 }
257
258 /* Keventd can't block, but this (a child) can. */
259 static int wait_for_helper(void *data)
260 {
261         struct subprocess_info *sub_info = data;
262         pid_t pid;
263
264         /* If SIGCLD is ignored sys_wait4 won't populate the status. */
265         spin_lock_irq(&current->sighand->siglock);
266         current->sighand->action[SIGCHLD-1].sa.sa_handler = SIG_DFL;
267         spin_unlock_irq(&current->sighand->siglock);
268
269         pid = kernel_thread(____call_usermodehelper, sub_info, SIGCHLD);
270         if (pid < 0) {
271                 sub_info->retval = pid;
272         } else {
273                 int ret = -ECHILD;
274                 /*
275                  * Normally it is bogus to call wait4() from in-kernel because
276                  * wait4() wants to write the exit code to a userspace address.
277                  * But wait_for_helper() always runs as keventd, and put_user()
278                  * to a kernel address works OK for kernel threads, due to their
279                  * having an mm_segment_t which spans the entire address space.
280                  *
281                  * Thus the __user pointer cast is valid here.
282                  */
283                 sys_wait4(pid, (int __user *)&ret, 0, NULL);
284
285                 /*
286                  * If ret is 0, either ____call_usermodehelper failed and the
287                  * real error code is already in sub_info->retval or
288                  * sub_info->retval is 0 anyway, so don't mess with it then.
289                  */
290                 if (ret)
291                         sub_info->retval = ret;
292         }
293
294         umh_complete(sub_info);
295         return 0;
296 }
297
298 /* This is run by khelper thread  */
299 static void __call_usermodehelper(struct work_struct *work)
300 {
301         struct subprocess_info *sub_info =
302                 container_of(work, struct subprocess_info, work);
303         int wait = sub_info->wait & ~UMH_KILLABLE;
304         pid_t pid;
305
306         /* CLONE_VFORK: wait until the usermode helper has execve'd
307          * successfully We need the data structures to stay around
308          * until that is done.  */
309         if (wait == UMH_WAIT_PROC)
310                 pid = kernel_thread(wait_for_helper, sub_info,
311                                     CLONE_FS | CLONE_FILES | SIGCHLD);
312         else {
313                 pid = kernel_thread(call_helper, sub_info,
314                                     CLONE_VFORK | SIGCHLD);
315                 /* Worker thread stopped blocking khelper thread. */
316                 kmod_thread_locker = NULL;
317         }
318
319         switch (wait) {
320         case UMH_NO_WAIT:
321                 call_usermodehelper_freeinfo(sub_info);
322                 break;
323
324         case UMH_WAIT_PROC:
325                 if (pid > 0)
326                         break;
327                 /* FALLTHROUGH */
328         case UMH_WAIT_EXEC:
329                 if (pid < 0)
330                         sub_info->retval = pid;
331                 umh_complete(sub_info);
332         }
333 }
334
335 /*
336  * If set, call_usermodehelper_exec() will exit immediately returning -EBUSY
337  * (used for preventing user land processes from being created after the user
338  * land has been frozen during a system-wide hibernation or suspend operation).
339  * Should always be manipulated under umhelper_sem acquired for write.
340  */
341 static enum umh_disable_depth usermodehelper_disabled = UMH_DISABLED;
342
343 /* Number of helpers running */
344 static atomic_t running_helpers = ATOMIC_INIT(0);
345
346 /*
347  * Wait queue head used by usermodehelper_disable() to wait for all running
348  * helpers to finish.
349  */
350 static DECLARE_WAIT_QUEUE_HEAD(running_helpers_waitq);
351
352 /*
353  * Used by usermodehelper_read_lock_wait() to wait for usermodehelper_disabled
354  * to become 'false'.
355  */
356 static DECLARE_WAIT_QUEUE_HEAD(usermodehelper_disabled_waitq);
357
358 /*
359  * Time to wait for running_helpers to become zero before the setting of
360  * usermodehelper_disabled in usermodehelper_disable() fails
361  */
362 #define RUNNING_HELPERS_TIMEOUT (5 * HZ)
363
364 int usermodehelper_read_trylock(void)
365 {
366         DEFINE_WAIT(wait);
367         int ret = 0;
368
369         down_read(&umhelper_sem);
370         for (;;) {
371                 prepare_to_wait(&usermodehelper_disabled_waitq, &wait,
372                                 TASK_INTERRUPTIBLE);
373                 if (!usermodehelper_disabled)
374                         break;
375
376                 if (usermodehelper_disabled == UMH_DISABLED)
377                         ret = -EAGAIN;
378
379                 up_read(&umhelper_sem);
380
381                 if (ret)
382                         break;
383
384                 schedule();
385                 try_to_freeze();
386
387                 down_read(&umhelper_sem);
388         }
389         finish_wait(&usermodehelper_disabled_waitq, &wait);
390         return ret;
391 }
392 EXPORT_SYMBOL_GPL(usermodehelper_read_trylock);
393
394 long usermodehelper_read_lock_wait(long timeout)
395 {
396         DEFINE_WAIT(wait);
397
398         if (timeout < 0)
399                 return -EINVAL;
400
401         down_read(&umhelper_sem);
402         for (;;) {
403                 prepare_to_wait(&usermodehelper_disabled_waitq, &wait,
404                                 TASK_UNINTERRUPTIBLE);
405                 if (!usermodehelper_disabled)
406                         break;
407
408                 up_read(&umhelper_sem);
409
410                 timeout = schedule_timeout(timeout);
411                 if (!timeout)
412                         break;
413
414                 down_read(&umhelper_sem);
415         }
416         finish_wait(&usermodehelper_disabled_waitq, &wait);
417         return timeout;
418 }
419 EXPORT_SYMBOL_GPL(usermodehelper_read_lock_wait);
420
421 void usermodehelper_read_unlock(void)
422 {
423         up_read(&umhelper_sem);
424 }
425 EXPORT_SYMBOL_GPL(usermodehelper_read_unlock);
426
427 /**
428  * __usermodehelper_set_disable_depth - Modify usermodehelper_disabled.
429  * @depth: New value to assign to usermodehelper_disabled.
430  *
431  * Change the value of usermodehelper_disabled (under umhelper_sem locked for
432  * writing) and wakeup tasks waiting for it to change.
433  */
434 void __usermodehelper_set_disable_depth(enum umh_disable_depth depth)
435 {
436         down_write(&umhelper_sem);
437         usermodehelper_disabled = depth;
438         wake_up(&usermodehelper_disabled_waitq);
439         up_write(&umhelper_sem);
440 }
441
442 /**
443  * __usermodehelper_disable - Prevent new helpers from being started.
444  * @depth: New value to assign to usermodehelper_disabled.
445  *
446  * Set usermodehelper_disabled to @depth and wait for running helpers to exit.
447  */
448 int __usermodehelper_disable(enum umh_disable_depth depth)
449 {
450         long retval;
451
452         if (!depth)
453                 return -EINVAL;
454
455         down_write(&umhelper_sem);
456         usermodehelper_disabled = depth;
457         up_write(&umhelper_sem);
458
459         /*
460          * From now on call_usermodehelper_exec() won't start any new
461          * helpers, so it is sufficient if running_helpers turns out to
462          * be zero at one point (it may be increased later, but that
463          * doesn't matter).
464          */
465         retval = wait_event_timeout(running_helpers_waitq,
466                                         atomic_read(&running_helpers) == 0,
467                                         RUNNING_HELPERS_TIMEOUT);
468         if (retval)
469                 return 0;
470
471         __usermodehelper_set_disable_depth(UMH_ENABLED);
472         return -EAGAIN;
473 }
474
475 static void helper_lock(void)
476 {
477         atomic_inc(&running_helpers);
478         smp_mb__after_atomic_inc();
479 }
480
481 static void helper_unlock(void)
482 {
483         if (atomic_dec_and_test(&running_helpers))
484                 wake_up(&running_helpers_waitq);
485 }
486
487 /**
488  * call_usermodehelper_setup - prepare to call a usermode helper
489  * @path: path to usermode executable
490  * @argv: arg vector for process
491  * @envp: environment for process
492  * @gfp_mask: gfp mask for memory allocation
493  *
494  * Returns either %NULL on allocation failure, or a subprocess_info
495  * structure.  This should be passed to call_usermodehelper_exec to
496  * exec the process and free the structure.
497  */
498 static
499 struct subprocess_info *call_usermodehelper_setup(char *path, char **argv,
500                                                   char **envp, gfp_t gfp_mask)
501 {
502         struct subprocess_info *sub_info;
503         sub_info = kzalloc(sizeof(struct subprocess_info), gfp_mask);
504         if (!sub_info)
505                 goto out;
506
507         INIT_WORK(&sub_info->work, __call_usermodehelper);
508         sub_info->path = path;
509         sub_info->argv = argv;
510         sub_info->envp = envp;
511   out:
512         return sub_info;
513 }
514
515 /**
516  * call_usermodehelper_setfns - set a cleanup/init function
517  * @info: a subprocess_info returned by call_usermodehelper_setup
518  * @cleanup: a cleanup function
519  * @init: an init function
520  * @data: arbitrary context sensitive data
521  *
522  * The init function is used to customize the helper process prior to
523  * exec.  A non-zero return code causes the process to error out, exit,
524  * and return the failure to the calling process
525  *
526  * The cleanup function is just before ethe subprocess_info is about to
527  * be freed.  This can be used for freeing the argv and envp.  The
528  * Function must be runnable in either a process context or the
529  * context in which call_usermodehelper_exec is called.
530  */
531 static
532 void call_usermodehelper_setfns(struct subprocess_info *info,
533                     int (*init)(struct subprocess_info *info, struct cred *new),
534                     void (*cleanup)(struct subprocess_info *info),
535                     void *data)
536 {
537         info->cleanup = cleanup;
538         info->init = init;
539         info->data = data;
540 }
541
542 /**
543  * call_usermodehelper_exec - start a usermode application
544  * @sub_info: information about the subprocessa
545  * @wait: wait for the application to finish and return status.
546  *        when -1 don't wait at all, but you get no useful error back when
547  *        the program couldn't be exec'ed. This makes it safe to call
548  *        from interrupt context.
549  *
550  * Runs a user-space application.  The application is started
551  * asynchronously if wait is not set, and runs as a child of keventd.
552  * (ie. it runs with full root capabilities).
553  */
554 static
555 int call_usermodehelper_exec(struct subprocess_info *sub_info, int wait)
556 {
557         DECLARE_COMPLETION_ONSTACK(done);
558         int retval = 0;
559
560         helper_lock();
561         if (sub_info->path[0] == '\0')
562                 goto out;
563
564         if (!khelper_wq || usermodehelper_disabled) {
565                 retval = -EBUSY;
566                 goto out;
567         }
568         /*
569          * Worker thread must not wait for khelper thread at below
570          * wait_for_completion() if the thread was created with CLONE_VFORK
571          * flag, for khelper thread is already waiting for the thread at
572          * wait_for_completion() in do_fork().
573          */
574         if (wait != UMH_NO_WAIT && current == kmod_thread_locker) {
575                 retval = -EBUSY;
576                 goto out;
577         }
578
579         sub_info->complete = &done;
580         sub_info->wait = wait;
581
582         queue_work(khelper_wq, &sub_info->work);
583         if (wait == UMH_NO_WAIT)        /* task has freed sub_info */
584                 goto unlock;
585
586         if (wait & UMH_KILLABLE) {
587                 retval = wait_for_completion_killable(&done);
588                 if (!retval)
589                         goto wait_done;
590
591                 /* umh_complete() will see NULL and free sub_info */
592                 if (xchg(&sub_info->complete, NULL))
593                         goto unlock;
594                 /* fallthrough, umh_complete() was already called */
595         }
596
597         wait_for_completion(&done);
598 wait_done:
599         retval = sub_info->retval;
600 out:
601         call_usermodehelper_freeinfo(sub_info);
602 unlock:
603         helper_unlock();
604         return retval;
605 }
606
607 /*
608  * call_usermodehelper_fns() will not run the caller-provided cleanup function
609  * if a memory allocation failure is experienced.  So the caller might need to
610  * check the call_usermodehelper_fns() return value: if it is -ENOMEM, perform
611  * the necessaary cleanup within the caller.
612  */
613 int call_usermodehelper_fns(
614         char *path, char **argv, char **envp, int wait,
615         int (*init)(struct subprocess_info *info, struct cred *new),
616         void (*cleanup)(struct subprocess_info *), void *data)
617 {
618         struct subprocess_info *info;
619         gfp_t gfp_mask = (wait == UMH_NO_WAIT) ? GFP_ATOMIC : GFP_KERNEL;
620
621         info = call_usermodehelper_setup(path, argv, envp, gfp_mask);
622
623         if (info == NULL)
624                 return -ENOMEM;
625
626         call_usermodehelper_setfns(info, init, cleanup, data);
627
628         return call_usermodehelper_exec(info, wait);
629 }
630 EXPORT_SYMBOL(call_usermodehelper_fns);
631
632 static int proc_cap_handler(struct ctl_table *table, int write,
633                          void __user *buffer, size_t *lenp, loff_t *ppos)
634 {
635         struct ctl_table t;
636         unsigned long cap_array[_KERNEL_CAPABILITY_U32S];
637         kernel_cap_t new_cap;
638         int err, i;
639
640         if (write && (!capable(CAP_SETPCAP) ||
641                       !capable(CAP_SYS_MODULE)))
642                 return -EPERM;
643
644         /*
645          * convert from the global kernel_cap_t to the ulong array to print to
646          * userspace if this is a read.
647          */
648         spin_lock(&umh_sysctl_lock);
649         for (i = 0; i < _KERNEL_CAPABILITY_U32S; i++)  {
650                 if (table->data == CAP_BSET)
651                         cap_array[i] = usermodehelper_bset.cap[i];
652                 else if (table->data == CAP_PI)
653                         cap_array[i] = usermodehelper_inheritable.cap[i];
654                 else
655                         BUG();
656         }
657         spin_unlock(&umh_sysctl_lock);
658
659         t = *table;
660         t.data = &cap_array;
661
662         /*
663          * actually read or write and array of ulongs from userspace.  Remember
664          * these are least significant 32 bits first
665          */
666         err = proc_doulongvec_minmax(&t, write, buffer, lenp, ppos);
667         if (err < 0)
668                 return err;
669
670         /*
671          * convert from the sysctl array of ulongs to the kernel_cap_t
672          * internal representation
673          */
674         for (i = 0; i < _KERNEL_CAPABILITY_U32S; i++)
675                 new_cap.cap[i] = cap_array[i];
676
677         /*
678          * Drop everything not in the new_cap (but don't add things)
679          */
680         spin_lock(&umh_sysctl_lock);
681         if (write) {
682                 if (table->data == CAP_BSET)
683                         usermodehelper_bset = cap_intersect(usermodehelper_bset, new_cap);
684                 if (table->data == CAP_PI)
685                         usermodehelper_inheritable = cap_intersect(usermodehelper_inheritable, new_cap);
686         }
687         spin_unlock(&umh_sysctl_lock);
688
689         return 0;
690 }
691
692 struct ctl_table usermodehelper_table[] = {
693         {
694                 .procname       = "bset",
695                 .data           = CAP_BSET,
696                 .maxlen         = _KERNEL_CAPABILITY_U32S * sizeof(unsigned long),
697                 .mode           = 0600,
698                 .proc_handler   = proc_cap_handler,
699         },
700         {
701                 .procname       = "inheritable",
702                 .data           = CAP_PI,
703                 .maxlen         = _KERNEL_CAPABILITY_U32S * sizeof(unsigned long),
704                 .mode           = 0600,
705                 .proc_handler   = proc_cap_handler,
706         },
707         { }
708 };
709
710 void __init usermodehelper_init(void)
711 {
712         khelper_wq = create_singlethread_workqueue("khelper");
713         BUG_ON(!khelper_wq);
714 }