ftrace: Update the kconfig for DYNAMIC_FTRACE
[linux-2.6.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 #define CAP_BSET        (void *)1
49 #define CAP_PI          (void *)2
50
51 static kernel_cap_t usermodehelper_bset = CAP_FULL_SET;
52 static kernel_cap_t usermodehelper_inheritable = CAP_FULL_SET;
53 static DEFINE_SPINLOCK(umh_sysctl_lock);
54 static DECLARE_RWSEM(umhelper_sem);
55
56 #ifdef CONFIG_MODULES
57
58 /*
59         modprobe_path is set via /proc/sys.
60 */
61 char modprobe_path[KMOD_PATH_LEN] = "/sbin/modprobe";
62
63 static void free_modprobe_argv(struct subprocess_info *info)
64 {
65         kfree(info->argv[3]); /* check call_modprobe() */
66         kfree(info->argv);
67 }
68
69 static int call_modprobe(char *module_name, int wait)
70 {
71         static char *envp[] = {
72                 "HOME=/",
73                 "TERM=linux",
74                 "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
75                 NULL
76         };
77
78         char **argv = kmalloc(sizeof(char *[5]), GFP_KERNEL);
79         if (!argv)
80                 goto out;
81
82         module_name = kstrdup(module_name, GFP_KERNEL);
83         if (!module_name)
84                 goto free_argv;
85
86         argv[0] = modprobe_path;
87         argv[1] = "-q";
88         argv[2] = "--";
89         argv[3] = module_name;  /* check free_modprobe_argv() */
90         argv[4] = NULL;
91
92         return call_usermodehelper_fns(modprobe_path, argv, envp,
93                 wait | UMH_KILLABLE, NULL, free_modprobe_argv, NULL);
94 free_argv:
95         kfree(argv);
96 out:
97         return -ENOMEM;
98 }
99
100 /**
101  * __request_module - try to load a kernel module
102  * @wait: wait (or not) for the operation to complete
103  * @fmt: printf style format string for the name of the module
104  * @...: arguments as specified in the format string
105  *
106  * Load a module using the user mode module loader. The function returns
107  * zero on success or a negative errno code on failure. Note that a
108  * successful module load does not mean the module did not then unload
109  * and exit on an error of its own. Callers must check that the service
110  * they requested is now available not blindly invoke it.
111  *
112  * If module auto-loading support is disabled then this function
113  * becomes a no-operation.
114  */
115 int __request_module(bool wait, const char *fmt, ...)
116 {
117         va_list args;
118         char module_name[MODULE_NAME_LEN];
119         unsigned int max_modprobes;
120         int ret;
121         static atomic_t kmod_concurrent = ATOMIC_INIT(0);
122 #define MAX_KMOD_CONCURRENT 50  /* Completely arbitrary value - KAO */
123         static int kmod_loop_msg;
124
125         va_start(args, fmt);
126         ret = vsnprintf(module_name, MODULE_NAME_LEN, fmt, args);
127         va_end(args);
128         if (ret >= MODULE_NAME_LEN)
129                 return -ENAMETOOLONG;
130
131         ret = security_kernel_module_request(module_name);
132         if (ret)
133                 return ret;
134
135         /* If modprobe needs a service that is in a module, we get a recursive
136          * loop.  Limit the number of running kmod threads to max_threads/2 or
137          * MAX_KMOD_CONCURRENT, whichever is the smaller.  A cleaner method
138          * would be to run the parents of this process, counting how many times
139          * kmod was invoked.  That would mean accessing the internals of the
140          * process tables to get the command line, proc_pid_cmdline is static
141          * and it is not worth changing the proc code just to handle this case. 
142          * KAO.
143          *
144          * "trace the ppid" is simple, but will fail if someone's
145          * parent exits.  I think this is as good as it gets. --RR
146          */
147         max_modprobes = min(max_threads/2, MAX_KMOD_CONCURRENT);
148         atomic_inc(&kmod_concurrent);
149         if (atomic_read(&kmod_concurrent) > max_modprobes) {
150                 /* We may be blaming an innocent here, but unlikely */
151                 if (kmod_loop_msg < 5) {
152                         printk(KERN_ERR
153                                "request_module: runaway loop modprobe %s\n",
154                                module_name);
155                         kmod_loop_msg++;
156                 }
157                 atomic_dec(&kmod_concurrent);
158                 return -ENOMEM;
159         }
160
161         trace_module_request(module_name, wait, _RET_IP_);
162
163         ret = call_modprobe(module_name, wait ? UMH_WAIT_PROC : UMH_WAIT_EXEC);
164
165         atomic_dec(&kmod_concurrent);
166         return ret;
167 }
168 EXPORT_SYMBOL(__request_module);
169 #endif /* CONFIG_MODULES */
170
171 /*
172  * This is the task which runs the usermode application
173  */
174 static int ____call_usermodehelper(void *data)
175 {
176         struct subprocess_info *sub_info = data;
177         struct cred *new;
178         int retval;
179
180         spin_lock_irq(&current->sighand->siglock);
181         flush_signal_handlers(current, 1);
182         spin_unlock_irq(&current->sighand->siglock);
183
184         /* We can run anywhere, unlike our parent keventd(). */
185         set_cpus_allowed_ptr(current, cpu_all_mask);
186
187         /*
188          * Our parent is keventd, which runs with elevated scheduling priority.
189          * Avoid propagating that into the userspace child.
190          */
191         set_user_nice(current, 0);
192
193         retval = -ENOMEM;
194         new = prepare_kernel_cred(current);
195         if (!new)
196                 goto fail;
197
198         spin_lock(&umh_sysctl_lock);
199         new->cap_bset = cap_intersect(usermodehelper_bset, new->cap_bset);
200         new->cap_inheritable = cap_intersect(usermodehelper_inheritable,
201                                              new->cap_inheritable);
202         spin_unlock(&umh_sysctl_lock);
203
204         if (sub_info->init) {
205                 retval = sub_info->init(sub_info, new);
206                 if (retval) {
207                         abort_creds(new);
208                         goto fail;
209                 }
210         }
211
212         commit_creds(new);
213
214         retval = kernel_execve(sub_info->path,
215                                (const char *const *)sub_info->argv,
216                                (const char *const *)sub_info->envp);
217
218         /* Exec failed? */
219 fail:
220         sub_info->retval = retval;
221         return 0;
222 }
223
224 void call_usermodehelper_freeinfo(struct subprocess_info *info)
225 {
226         if (info->cleanup)
227                 (*info->cleanup)(info);
228         kfree(info);
229 }
230 EXPORT_SYMBOL(call_usermodehelper_freeinfo);
231
232 static void umh_complete(struct subprocess_info *sub_info)
233 {
234         struct completion *comp = xchg(&sub_info->complete, NULL);
235         /*
236          * See call_usermodehelper_exec(). If xchg() returns NULL
237          * we own sub_info, the UMH_KILLABLE caller has gone away.
238          */
239         if (comp)
240                 complete(comp);
241         else
242                 call_usermodehelper_freeinfo(sub_info);
243 }
244
245 /* Keventd can't block, but this (a child) can. */
246 static int wait_for_helper(void *data)
247 {
248         struct subprocess_info *sub_info = data;
249         pid_t pid;
250
251         /* If SIGCLD is ignored sys_wait4 won't populate the status. */
252         spin_lock_irq(&current->sighand->siglock);
253         current->sighand->action[SIGCHLD-1].sa.sa_handler = SIG_DFL;
254         spin_unlock_irq(&current->sighand->siglock);
255
256         pid = kernel_thread(____call_usermodehelper, sub_info, SIGCHLD);
257         if (pid < 0) {
258                 sub_info->retval = pid;
259         } else {
260                 int ret = -ECHILD;
261                 /*
262                  * Normally it is bogus to call wait4() from in-kernel because
263                  * wait4() wants to write the exit code to a userspace address.
264                  * But wait_for_helper() always runs as keventd, and put_user()
265                  * to a kernel address works OK for kernel threads, due to their
266                  * having an mm_segment_t which spans the entire address space.
267                  *
268                  * Thus the __user pointer cast is valid here.
269                  */
270                 sys_wait4(pid, (int __user *)&ret, 0, NULL);
271
272                 /*
273                  * If ret is 0, either ____call_usermodehelper failed and the
274                  * real error code is already in sub_info->retval or
275                  * sub_info->retval is 0 anyway, so don't mess with it then.
276                  */
277                 if (ret)
278                         sub_info->retval = ret;
279         }
280
281         umh_complete(sub_info);
282         return 0;
283 }
284
285 /* This is run by khelper thread  */
286 static void __call_usermodehelper(struct work_struct *work)
287 {
288         struct subprocess_info *sub_info =
289                 container_of(work, struct subprocess_info, work);
290         int wait = sub_info->wait & ~UMH_KILLABLE;
291         pid_t pid;
292
293         /* CLONE_VFORK: wait until the usermode helper has execve'd
294          * successfully We need the data structures to stay around
295          * until that is done.  */
296         if (wait == UMH_WAIT_PROC)
297                 pid = kernel_thread(wait_for_helper, sub_info,
298                                     CLONE_FS | CLONE_FILES | SIGCHLD);
299         else
300                 pid = kernel_thread(____call_usermodehelper, sub_info,
301                                     CLONE_VFORK | SIGCHLD);
302
303         switch (wait) {
304         case UMH_NO_WAIT:
305                 call_usermodehelper_freeinfo(sub_info);
306                 break;
307
308         case UMH_WAIT_PROC:
309                 if (pid > 0)
310                         break;
311                 /* FALLTHROUGH */
312         case UMH_WAIT_EXEC:
313                 if (pid < 0)
314                         sub_info->retval = pid;
315                 umh_complete(sub_info);
316         }
317 }
318
319 /*
320  * If set, call_usermodehelper_exec() will exit immediately returning -EBUSY
321  * (used for preventing user land processes from being created after the user
322  * land has been frozen during a system-wide hibernation or suspend operation).
323  * Should always be manipulated under umhelper_sem acquired for write.
324  */
325 static enum umh_disable_depth usermodehelper_disabled = UMH_DISABLED;
326
327 /* Number of helpers running */
328 static atomic_t running_helpers = ATOMIC_INIT(0);
329
330 /*
331  * Wait queue head used by usermodehelper_disable() to wait for all running
332  * helpers to finish.
333  */
334 static DECLARE_WAIT_QUEUE_HEAD(running_helpers_waitq);
335
336 /*
337  * Used by usermodehelper_read_lock_wait() to wait for usermodehelper_disabled
338  * to become 'false'.
339  */
340 static DECLARE_WAIT_QUEUE_HEAD(usermodehelper_disabled_waitq);
341
342 /*
343  * Time to wait for running_helpers to become zero before the setting of
344  * usermodehelper_disabled in usermodehelper_disable() fails
345  */
346 #define RUNNING_HELPERS_TIMEOUT (5 * HZ)
347
348 int usermodehelper_read_trylock(void)
349 {
350         DEFINE_WAIT(wait);
351         int ret = 0;
352
353         down_read(&umhelper_sem);
354         for (;;) {
355                 prepare_to_wait(&usermodehelper_disabled_waitq, &wait,
356                                 TASK_INTERRUPTIBLE);
357                 if (!usermodehelper_disabled)
358                         break;
359
360                 if (usermodehelper_disabled == UMH_DISABLED)
361                         ret = -EAGAIN;
362
363                 up_read(&umhelper_sem);
364
365                 if (ret)
366                         break;
367
368                 schedule();
369                 try_to_freeze();
370
371                 down_read(&umhelper_sem);
372         }
373         finish_wait(&usermodehelper_disabled_waitq, &wait);
374         return ret;
375 }
376 EXPORT_SYMBOL_GPL(usermodehelper_read_trylock);
377
378 long usermodehelper_read_lock_wait(long timeout)
379 {
380         DEFINE_WAIT(wait);
381
382         if (timeout < 0)
383                 return -EINVAL;
384
385         down_read(&umhelper_sem);
386         for (;;) {
387                 prepare_to_wait(&usermodehelper_disabled_waitq, &wait,
388                                 TASK_UNINTERRUPTIBLE);
389                 if (!usermodehelper_disabled)
390                         break;
391
392                 up_read(&umhelper_sem);
393
394                 timeout = schedule_timeout(timeout);
395                 if (!timeout)
396                         break;
397
398                 down_read(&umhelper_sem);
399         }
400         finish_wait(&usermodehelper_disabled_waitq, &wait);
401         return timeout;
402 }
403 EXPORT_SYMBOL_GPL(usermodehelper_read_lock_wait);
404
405 void usermodehelper_read_unlock(void)
406 {
407         up_read(&umhelper_sem);
408 }
409 EXPORT_SYMBOL_GPL(usermodehelper_read_unlock);
410
411 /**
412  * __usermodehelper_set_disable_depth - Modify usermodehelper_disabled.
413  * depth: New value to assign to usermodehelper_disabled.
414  *
415  * Change the value of usermodehelper_disabled (under umhelper_sem locked for
416  * writing) and wakeup tasks waiting for it to change.
417  */
418 void __usermodehelper_set_disable_depth(enum umh_disable_depth depth)
419 {
420         down_write(&umhelper_sem);
421         usermodehelper_disabled = depth;
422         wake_up(&usermodehelper_disabled_waitq);
423         up_write(&umhelper_sem);
424 }
425
426 /**
427  * __usermodehelper_disable - Prevent new helpers from being started.
428  * @depth: New value to assign to usermodehelper_disabled.
429  *
430  * Set usermodehelper_disabled to @depth and wait for running helpers to exit.
431  */
432 int __usermodehelper_disable(enum umh_disable_depth depth)
433 {
434         long retval;
435
436         if (!depth)
437                 return -EINVAL;
438
439         down_write(&umhelper_sem);
440         usermodehelper_disabled = depth;
441         up_write(&umhelper_sem);
442
443         /*
444          * From now on call_usermodehelper_exec() won't start any new
445          * helpers, so it is sufficient if running_helpers turns out to
446          * be zero at one point (it may be increased later, but that
447          * doesn't matter).
448          */
449         retval = wait_event_timeout(running_helpers_waitq,
450                                         atomic_read(&running_helpers) == 0,
451                                         RUNNING_HELPERS_TIMEOUT);
452         if (retval)
453                 return 0;
454
455         __usermodehelper_set_disable_depth(UMH_ENABLED);
456         return -EAGAIN;
457 }
458
459 static void helper_lock(void)
460 {
461         atomic_inc(&running_helpers);
462         smp_mb__after_atomic_inc();
463 }
464
465 static void helper_unlock(void)
466 {
467         if (atomic_dec_and_test(&running_helpers))
468                 wake_up(&running_helpers_waitq);
469 }
470
471 /**
472  * call_usermodehelper_setup - prepare to call a usermode helper
473  * @path: path to usermode executable
474  * @argv: arg vector for process
475  * @envp: environment for process
476  * @gfp_mask: gfp mask for memory allocation
477  *
478  * Returns either %NULL on allocation failure, or a subprocess_info
479  * structure.  This should be passed to call_usermodehelper_exec to
480  * exec the process and free the structure.
481  */
482 struct subprocess_info *call_usermodehelper_setup(char *path, char **argv,
483                                                   char **envp, gfp_t gfp_mask)
484 {
485         struct subprocess_info *sub_info;
486         sub_info = kzalloc(sizeof(struct subprocess_info), gfp_mask);
487         if (!sub_info)
488                 goto out;
489
490         INIT_WORK(&sub_info->work, __call_usermodehelper);
491         sub_info->path = path;
492         sub_info->argv = argv;
493         sub_info->envp = envp;
494   out:
495         return sub_info;
496 }
497 EXPORT_SYMBOL(call_usermodehelper_setup);
498
499 /**
500  * call_usermodehelper_setfns - set a cleanup/init function
501  * @info: a subprocess_info returned by call_usermodehelper_setup
502  * @cleanup: a cleanup function
503  * @init: an init function
504  * @data: arbitrary context sensitive data
505  *
506  * The init function is used to customize the helper process prior to
507  * exec.  A non-zero return code causes the process to error out, exit,
508  * and return the failure to the calling process
509  *
510  * The cleanup function is just before ethe subprocess_info is about to
511  * be freed.  This can be used for freeing the argv and envp.  The
512  * Function must be runnable in either a process context or the
513  * context in which call_usermodehelper_exec is called.
514  */
515 void call_usermodehelper_setfns(struct subprocess_info *info,
516                     int (*init)(struct subprocess_info *info, struct cred *new),
517                     void (*cleanup)(struct subprocess_info *info),
518                     void *data)
519 {
520         info->cleanup = cleanup;
521         info->init = init;
522         info->data = data;
523 }
524 EXPORT_SYMBOL(call_usermodehelper_setfns);
525
526 /**
527  * call_usermodehelper_exec - start a usermode application
528  * @sub_info: information about the subprocessa
529  * @wait: wait for the application to finish and return status.
530  *        when -1 don't wait at all, but you get no useful error back when
531  *        the program couldn't be exec'ed. This makes it safe to call
532  *        from interrupt context.
533  *
534  * Runs a user-space application.  The application is started
535  * asynchronously if wait is not set, and runs as a child of keventd.
536  * (ie. it runs with full root capabilities).
537  */
538 int call_usermodehelper_exec(struct subprocess_info *sub_info, int wait)
539 {
540         DECLARE_COMPLETION_ONSTACK(done);
541         int retval = 0;
542
543         helper_lock();
544         if (sub_info->path[0] == '\0')
545                 goto out;
546
547         if (!khelper_wq || usermodehelper_disabled) {
548                 retval = -EBUSY;
549                 goto out;
550         }
551
552         sub_info->complete = &done;
553         sub_info->wait = wait;
554
555         queue_work(khelper_wq, &sub_info->work);
556         if (wait == UMH_NO_WAIT)        /* task has freed sub_info */
557                 goto unlock;
558
559         if (wait & UMH_KILLABLE) {
560                 retval = wait_for_completion_killable(&done);
561                 if (!retval)
562                         goto wait_done;
563
564                 /* umh_complete() will see NULL and free sub_info */
565                 if (xchg(&sub_info->complete, NULL))
566                         goto unlock;
567                 /* fallthrough, umh_complete() was already called */
568         }
569
570         wait_for_completion(&done);
571 wait_done:
572         retval = sub_info->retval;
573 out:
574         call_usermodehelper_freeinfo(sub_info);
575 unlock:
576         helper_unlock();
577         return retval;
578 }
579 EXPORT_SYMBOL(call_usermodehelper_exec);
580
581 static int proc_cap_handler(struct ctl_table *table, int write,
582                          void __user *buffer, size_t *lenp, loff_t *ppos)
583 {
584         struct ctl_table t;
585         unsigned long cap_array[_KERNEL_CAPABILITY_U32S];
586         kernel_cap_t new_cap;
587         int err, i;
588
589         if (write && (!capable(CAP_SETPCAP) ||
590                       !capable(CAP_SYS_MODULE)))
591                 return -EPERM;
592
593         /*
594          * convert from the global kernel_cap_t to the ulong array to print to
595          * userspace if this is a read.
596          */
597         spin_lock(&umh_sysctl_lock);
598         for (i = 0; i < _KERNEL_CAPABILITY_U32S; i++)  {
599                 if (table->data == CAP_BSET)
600                         cap_array[i] = usermodehelper_bset.cap[i];
601                 else if (table->data == CAP_PI)
602                         cap_array[i] = usermodehelper_inheritable.cap[i];
603                 else
604                         BUG();
605         }
606         spin_unlock(&umh_sysctl_lock);
607
608         t = *table;
609         t.data = &cap_array;
610
611         /*
612          * actually read or write and array of ulongs from userspace.  Remember
613          * these are least significant 32 bits first
614          */
615         err = proc_doulongvec_minmax(&t, write, buffer, lenp, ppos);
616         if (err < 0)
617                 return err;
618
619         /*
620          * convert from the sysctl array of ulongs to the kernel_cap_t
621          * internal representation
622          */
623         for (i = 0; i < _KERNEL_CAPABILITY_U32S; i++)
624                 new_cap.cap[i] = cap_array[i];
625
626         /*
627          * Drop everything not in the new_cap (but don't add things)
628          */
629         spin_lock(&umh_sysctl_lock);
630         if (write) {
631                 if (table->data == CAP_BSET)
632                         usermodehelper_bset = cap_intersect(usermodehelper_bset, new_cap);
633                 if (table->data == CAP_PI)
634                         usermodehelper_inheritable = cap_intersect(usermodehelper_inheritable, new_cap);
635         }
636         spin_unlock(&umh_sysctl_lock);
637
638         return 0;
639 }
640
641 struct ctl_table usermodehelper_table[] = {
642         {
643                 .procname       = "bset",
644                 .data           = CAP_BSET,
645                 .maxlen         = _KERNEL_CAPABILITY_U32S * sizeof(unsigned long),
646                 .mode           = 0600,
647                 .proc_handler   = proc_cap_handler,
648         },
649         {
650                 .procname       = "inheritable",
651                 .data           = CAP_PI,
652                 .maxlen         = _KERNEL_CAPABILITY_U32S * sizeof(unsigned long),
653                 .mode           = 0600,
654                 .proc_handler   = proc_cap_handler,
655         },
656         { }
657 };
658
659 void __init usermodehelper_init(void)
660 {
661         khelper_wq = create_singlethread_workqueue("khelper");
662         BUG_ON(!khelper_wq);
663 }