2d83511e96108ffac0f1430d4e3d39188a1536ed
[sfrench/cifs-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 <linux/ptrace.h>
41 #include <linux/async.h>
42 #include <asm/uaccess.h>
43
44 #include <trace/events/module.h>
45
46 extern int max_threads;
47
48 static struct workqueue_struct *khelper_wq;
49
50 #define CAP_BSET        (void *)1
51 #define CAP_PI          (void *)2
52
53 static kernel_cap_t usermodehelper_bset = CAP_FULL_SET;
54 static kernel_cap_t usermodehelper_inheritable = CAP_FULL_SET;
55 static DEFINE_SPINLOCK(umh_sysctl_lock);
56 static DECLARE_RWSEM(umhelper_sem);
57
58 #ifdef CONFIG_MODULES
59
60 /*
61         modprobe_path is set via /proc/sys.
62 */
63 char modprobe_path[KMOD_PATH_LEN] = "/sbin/modprobe";
64
65 static void free_modprobe_argv(struct subprocess_info *info)
66 {
67         kfree(info->argv[3]); /* check call_modprobe() */
68         kfree(info->argv);
69 }
70
71 static int call_modprobe(char *module_name, int wait)
72 {
73         struct subprocess_info *info;
74         static char *envp[] = {
75                 "HOME=/",
76                 "TERM=linux",
77                 "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
78                 NULL
79         };
80
81         char **argv = kmalloc(sizeof(char *[5]), GFP_KERNEL);
82         if (!argv)
83                 goto out;
84
85         module_name = kstrdup(module_name, GFP_KERNEL);
86         if (!module_name)
87                 goto free_argv;
88
89         argv[0] = modprobe_path;
90         argv[1] = "-q";
91         argv[2] = "--";
92         argv[3] = module_name;  /* check free_modprobe_argv() */
93         argv[4] = NULL;
94
95         info = call_usermodehelper_setup(modprobe_path, argv, envp, GFP_KERNEL,
96                                          NULL, free_modprobe_argv, NULL);
97         if (!info)
98                 goto free_module_name;
99
100         return call_usermodehelper_exec(info, wait | UMH_KILLABLE);
101
102 free_module_name:
103         kfree(module_name);
104 free_argv:
105         kfree(argv);
106 out:
107         return -ENOMEM;
108 }
109
110 /**
111  * __request_module - try to load a kernel module
112  * @wait: wait (or not) for the operation to complete
113  * @fmt: printf style format string for the name of the module
114  * @...: arguments as specified in the format string
115  *
116  * Load a module using the user mode module loader. The function returns
117  * zero on success or a negative errno code or positive exit code from
118  * "modprobe" on failure. Note that a successful module load does not mean
119  * the module did not then unload and exit on an error of its own. Callers
120  * must check that the service they requested is now available not blindly
121  * invoke it.
122  *
123  * If module auto-loading support is disabled then this function
124  * becomes a no-operation.
125  */
126 int __request_module(bool wait, const char *fmt, ...)
127 {
128         va_list args;
129         char module_name[MODULE_NAME_LEN];
130         unsigned int max_modprobes;
131         int ret;
132         static atomic_t kmod_concurrent = ATOMIC_INIT(0);
133 #define MAX_KMOD_CONCURRENT 50  /* Completely arbitrary value - KAO */
134         static int kmod_loop_msg;
135
136         /*
137          * We don't allow synchronous module loading from async.  Module
138          * init may invoke async_synchronize_full() which will end up
139          * waiting for this task which already is waiting for the module
140          * loading to complete, leading to a deadlock.
141          */
142         WARN_ON_ONCE(wait && current_is_async());
143
144         if (!modprobe_path[0])
145                 return 0;
146
147         va_start(args, fmt);
148         ret = vsnprintf(module_name, MODULE_NAME_LEN, fmt, args);
149         va_end(args);
150         if (ret >= MODULE_NAME_LEN)
151                 return -ENAMETOOLONG;
152
153         ret = security_kernel_module_request(module_name);
154         if (ret)
155                 return ret;
156
157         /* If modprobe needs a service that is in a module, we get a recursive
158          * loop.  Limit the number of running kmod threads to max_threads/2 or
159          * MAX_KMOD_CONCURRENT, whichever is the smaller.  A cleaner method
160          * would be to run the parents of this process, counting how many times
161          * kmod was invoked.  That would mean accessing the internals of the
162          * process tables to get the command line, proc_pid_cmdline is static
163          * and it is not worth changing the proc code just to handle this case. 
164          * KAO.
165          *
166          * "trace the ppid" is simple, but will fail if someone's
167          * parent exits.  I think this is as good as it gets. --RR
168          */
169         max_modprobes = min(max_threads/2, MAX_KMOD_CONCURRENT);
170         atomic_inc(&kmod_concurrent);
171         if (atomic_read(&kmod_concurrent) > max_modprobes) {
172                 /* We may be blaming an innocent here, but unlikely */
173                 if (kmod_loop_msg < 5) {
174                         printk(KERN_ERR
175                                "request_module: runaway loop modprobe %s\n",
176                                module_name);
177                         kmod_loop_msg++;
178                 }
179                 atomic_dec(&kmod_concurrent);
180                 return -ENOMEM;
181         }
182
183         trace_module_request(module_name, wait, _RET_IP_);
184
185         ret = call_modprobe(module_name, wait ? UMH_WAIT_PROC : UMH_WAIT_EXEC);
186
187         atomic_dec(&kmod_concurrent);
188         return ret;
189 }
190 EXPORT_SYMBOL(__request_module);
191 #endif /* CONFIG_MODULES */
192
193 static void call_usermodehelper_freeinfo(struct subprocess_info *info)
194 {
195         if (info->cleanup)
196                 (*info->cleanup)(info);
197         kfree(info);
198 }
199
200 static void umh_complete(struct subprocess_info *sub_info)
201 {
202         struct completion *comp = xchg(&sub_info->complete, NULL);
203         /*
204          * See call_usermodehelper_exec(). If xchg() returns NULL
205          * we own sub_info, the UMH_KILLABLE caller has gone away
206          * or the caller used UMH_NO_WAIT.
207          */
208         if (comp)
209                 complete(comp);
210         else
211                 call_usermodehelper_freeinfo(sub_info);
212 }
213
214 /*
215  * This is the task which runs the usermode application
216  */
217 static int call_usermodehelper_exec_async(void *data)
218 {
219         struct subprocess_info *sub_info = data;
220         struct cred *new;
221         int retval;
222
223         spin_lock_irq(&current->sighand->siglock);
224         flush_signal_handlers(current, 1);
225         spin_unlock_irq(&current->sighand->siglock);
226
227         /* We can run anywhere, unlike our parent keventd(). */
228         set_cpus_allowed_ptr(current, cpu_all_mask);
229
230         /*
231          * Our parent is keventd, which runs with elevated scheduling priority.
232          * Avoid propagating that into the userspace child.
233          */
234         set_user_nice(current, 0);
235
236         retval = -ENOMEM;
237         new = prepare_kernel_cred(current);
238         if (!new)
239                 goto out;
240
241         spin_lock(&umh_sysctl_lock);
242         new->cap_bset = cap_intersect(usermodehelper_bset, new->cap_bset);
243         new->cap_inheritable = cap_intersect(usermodehelper_inheritable,
244                                              new->cap_inheritable);
245         spin_unlock(&umh_sysctl_lock);
246
247         if (sub_info->init) {
248                 retval = sub_info->init(sub_info, new);
249                 if (retval) {
250                         abort_creds(new);
251                         goto out;
252                 }
253         }
254
255         commit_creds(new);
256
257         retval = do_execve(getname_kernel(sub_info->path),
258                            (const char __user *const __user *)sub_info->argv,
259                            (const char __user *const __user *)sub_info->envp);
260 out:
261         sub_info->retval = retval;
262         /*
263          * call_usermodehelper_exec_sync() will call umh_complete
264          * if UHM_WAIT_PROC.
265          */
266         if (!(sub_info->wait & UMH_WAIT_PROC))
267                 umh_complete(sub_info);
268         if (!retval)
269                 return 0;
270         do_exit(0);
271 }
272
273 /* Keventd can't block, but this (a child) can. */
274 static int call_usermodehelper_exec_sync(void *data)
275 {
276         struct subprocess_info *sub_info = data;
277         pid_t pid;
278
279         /* If SIGCLD is ignored sys_wait4 won't populate the status. */
280         kernel_sigaction(SIGCHLD, SIG_DFL);
281         pid = kernel_thread(call_usermodehelper_exec_async, sub_info, SIGCHLD);
282         if (pid < 0) {
283                 sub_info->retval = pid;
284         } else {
285                 int ret = -ECHILD;
286                 /*
287                  * Normally it is bogus to call wait4() from in-kernel because
288                  * wait4() wants to write the exit code to a userspace address.
289                  * But call_usermodehelper_exec_sync() always runs as keventd,
290                  * and put_user() to a kernel address works OK for kernel
291                  * threads, due to their having an mm_segment_t which spans the
292                  * entire address space.
293                  *
294                  * Thus the __user pointer cast is valid here.
295                  */
296                 sys_wait4(pid, (int __user *)&ret, 0, NULL);
297
298                 /*
299                  * If ret is 0, either call_usermodehelper_exec_async failed and
300                  * the real error code is already in sub_info->retval or
301                  * sub_info->retval is 0 anyway, so don't mess with it then.
302                  */
303                 if (ret)
304                         sub_info->retval = ret;
305         }
306
307         umh_complete(sub_info);
308         do_exit(0);
309 }
310
311 /* This is run by khelper thread  */
312 static void call_usermodehelper_exec_work(struct work_struct *work)
313 {
314         struct subprocess_info *sub_info =
315                 container_of(work, struct subprocess_info, work);
316         pid_t pid;
317
318         if (sub_info->wait & UMH_WAIT_PROC)
319                 pid = kernel_thread(call_usermodehelper_exec_sync, sub_info,
320                                     CLONE_FS | CLONE_FILES | SIGCHLD);
321         else
322                 pid = kernel_thread(call_usermodehelper_exec_async, sub_info,
323                                     SIGCHLD);
324
325         if (pid < 0) {
326                 sub_info->retval = pid;
327                 umh_complete(sub_info);
328         }
329 }
330
331 /*
332  * If set, call_usermodehelper_exec() will exit immediately returning -EBUSY
333  * (used for preventing user land processes from being created after the user
334  * land has been frozen during a system-wide hibernation or suspend operation).
335  * Should always be manipulated under umhelper_sem acquired for write.
336  */
337 static enum umh_disable_depth usermodehelper_disabled = UMH_DISABLED;
338
339 /* Number of helpers running */
340 static atomic_t running_helpers = ATOMIC_INIT(0);
341
342 /*
343  * Wait queue head used by usermodehelper_disable() to wait for all running
344  * helpers to finish.
345  */
346 static DECLARE_WAIT_QUEUE_HEAD(running_helpers_waitq);
347
348 /*
349  * Used by usermodehelper_read_lock_wait() to wait for usermodehelper_disabled
350  * to become 'false'.
351  */
352 static DECLARE_WAIT_QUEUE_HEAD(usermodehelper_disabled_waitq);
353
354 /*
355  * Time to wait for running_helpers to become zero before the setting of
356  * usermodehelper_disabled in usermodehelper_disable() fails
357  */
358 #define RUNNING_HELPERS_TIMEOUT (5 * HZ)
359
360 int usermodehelper_read_trylock(void)
361 {
362         DEFINE_WAIT(wait);
363         int ret = 0;
364
365         down_read(&umhelper_sem);
366         for (;;) {
367                 prepare_to_wait(&usermodehelper_disabled_waitq, &wait,
368                                 TASK_INTERRUPTIBLE);
369                 if (!usermodehelper_disabled)
370                         break;
371
372                 if (usermodehelper_disabled == UMH_DISABLED)
373                         ret = -EAGAIN;
374
375                 up_read(&umhelper_sem);
376
377                 if (ret)
378                         break;
379
380                 schedule();
381                 try_to_freeze();
382
383                 down_read(&umhelper_sem);
384         }
385         finish_wait(&usermodehelper_disabled_waitq, &wait);
386         return ret;
387 }
388 EXPORT_SYMBOL_GPL(usermodehelper_read_trylock);
389
390 long usermodehelper_read_lock_wait(long timeout)
391 {
392         DEFINE_WAIT(wait);
393
394         if (timeout < 0)
395                 return -EINVAL;
396
397         down_read(&umhelper_sem);
398         for (;;) {
399                 prepare_to_wait(&usermodehelper_disabled_waitq, &wait,
400                                 TASK_UNINTERRUPTIBLE);
401                 if (!usermodehelper_disabled)
402                         break;
403
404                 up_read(&umhelper_sem);
405
406                 timeout = schedule_timeout(timeout);
407                 if (!timeout)
408                         break;
409
410                 down_read(&umhelper_sem);
411         }
412         finish_wait(&usermodehelper_disabled_waitq, &wait);
413         return timeout;
414 }
415 EXPORT_SYMBOL_GPL(usermodehelper_read_lock_wait);
416
417 void usermodehelper_read_unlock(void)
418 {
419         up_read(&umhelper_sem);
420 }
421 EXPORT_SYMBOL_GPL(usermodehelper_read_unlock);
422
423 /**
424  * __usermodehelper_set_disable_depth - Modify usermodehelper_disabled.
425  * @depth: New value to assign to usermodehelper_disabled.
426  *
427  * Change the value of usermodehelper_disabled (under umhelper_sem locked for
428  * writing) and wakeup tasks waiting for it to change.
429  */
430 void __usermodehelper_set_disable_depth(enum umh_disable_depth depth)
431 {
432         down_write(&umhelper_sem);
433         usermodehelper_disabled = depth;
434         wake_up(&usermodehelper_disabled_waitq);
435         up_write(&umhelper_sem);
436 }
437
438 /**
439  * __usermodehelper_disable - Prevent new helpers from being started.
440  * @depth: New value to assign to usermodehelper_disabled.
441  *
442  * Set usermodehelper_disabled to @depth and wait for running helpers to exit.
443  */
444 int __usermodehelper_disable(enum umh_disable_depth depth)
445 {
446         long retval;
447
448         if (!depth)
449                 return -EINVAL;
450
451         down_write(&umhelper_sem);
452         usermodehelper_disabled = depth;
453         up_write(&umhelper_sem);
454
455         /*
456          * From now on call_usermodehelper_exec() won't start any new
457          * helpers, so it is sufficient if running_helpers turns out to
458          * be zero at one point (it may be increased later, but that
459          * doesn't matter).
460          */
461         retval = wait_event_timeout(running_helpers_waitq,
462                                         atomic_read(&running_helpers) == 0,
463                                         RUNNING_HELPERS_TIMEOUT);
464         if (retval)
465                 return 0;
466
467         __usermodehelper_set_disable_depth(UMH_ENABLED);
468         return -EAGAIN;
469 }
470
471 static void helper_lock(void)
472 {
473         atomic_inc(&running_helpers);
474         smp_mb__after_atomic();
475 }
476
477 static void helper_unlock(void)
478 {
479         if (atomic_dec_and_test(&running_helpers))
480                 wake_up(&running_helpers_waitq);
481 }
482
483 /**
484  * call_usermodehelper_setup - prepare to call a usermode helper
485  * @path: path to usermode executable
486  * @argv: arg vector for process
487  * @envp: environment for process
488  * @gfp_mask: gfp mask for memory allocation
489  * @cleanup: a cleanup function
490  * @init: an init function
491  * @data: arbitrary context sensitive data
492  *
493  * Returns either %NULL on allocation failure, or a subprocess_info
494  * structure.  This should be passed to call_usermodehelper_exec to
495  * exec the process and free the structure.
496  *
497  * The init function is used to customize the helper process prior to
498  * exec.  A non-zero return code causes the process to error out, exit,
499  * and return the failure to the calling process
500  *
501  * The cleanup function is just before ethe subprocess_info is about to
502  * be freed.  This can be used for freeing the argv and envp.  The
503  * Function must be runnable in either a process context or the
504  * context in which call_usermodehelper_exec is called.
505  */
506 struct subprocess_info *call_usermodehelper_setup(char *path, char **argv,
507                 char **envp, gfp_t gfp_mask,
508                 int (*init)(struct subprocess_info *info, struct cred *new),
509                 void (*cleanup)(struct subprocess_info *info),
510                 void *data)
511 {
512         struct subprocess_info *sub_info;
513         sub_info = kzalloc(sizeof(struct subprocess_info), gfp_mask);
514         if (!sub_info)
515                 goto out;
516
517         INIT_WORK(&sub_info->work, call_usermodehelper_exec_work);
518         sub_info->path = path;
519         sub_info->argv = argv;
520         sub_info->envp = envp;
521
522         sub_info->cleanup = cleanup;
523         sub_info->init = init;
524         sub_info->data = data;
525   out:
526         return sub_info;
527 }
528 EXPORT_SYMBOL(call_usermodehelper_setup);
529
530 /**
531  * call_usermodehelper_exec - start a usermode application
532  * @sub_info: information about the subprocessa
533  * @wait: wait for the application to finish and return status.
534  *        when UMH_NO_WAIT don't wait at all, but you get no useful error back
535  *        when the program couldn't be exec'ed. This makes it safe to call
536  *        from interrupt context.
537  *
538  * Runs a user-space application.  The application is started
539  * asynchronously if wait is not set, and runs as a child of keventd.
540  * (ie. it runs with full root capabilities).
541  */
542 int call_usermodehelper_exec(struct subprocess_info *sub_info, int wait)
543 {
544         DECLARE_COMPLETION_ONSTACK(done);
545         int retval = 0;
546
547         if (!sub_info->path) {
548                 call_usermodehelper_freeinfo(sub_info);
549                 return -EINVAL;
550         }
551         helper_lock();
552         if (!khelper_wq || usermodehelper_disabled) {
553                 retval = -EBUSY;
554                 goto out;
555         }
556         /*
557          * Set the completion pointer only if there is a waiter.
558          * This makes it possible to use umh_complete to free
559          * the data structure in case of UMH_NO_WAIT.
560          */
561         sub_info->complete = (wait == UMH_NO_WAIT) ? NULL : &done;
562         sub_info->wait = wait;
563
564         queue_work(khelper_wq, &sub_info->work);
565         if (wait == UMH_NO_WAIT)        /* task has freed sub_info */
566                 goto unlock;
567
568         if (wait & UMH_KILLABLE) {
569                 retval = wait_for_completion_killable(&done);
570                 if (!retval)
571                         goto wait_done;
572
573                 /* umh_complete() will see NULL and free sub_info */
574                 if (xchg(&sub_info->complete, NULL))
575                         goto unlock;
576                 /* fallthrough, umh_complete() was already called */
577         }
578
579         wait_for_completion(&done);
580 wait_done:
581         retval = sub_info->retval;
582 out:
583         call_usermodehelper_freeinfo(sub_info);
584 unlock:
585         helper_unlock();
586         return retval;
587 }
588 EXPORT_SYMBOL(call_usermodehelper_exec);
589
590 /**
591  * call_usermodehelper() - prepare and start a usermode application
592  * @path: path to usermode executable
593  * @argv: arg vector for process
594  * @envp: environment for process
595  * @wait: wait for the application to finish and return status.
596  *        when UMH_NO_WAIT don't wait at all, but you get no useful error back
597  *        when the program couldn't be exec'ed. This makes it safe to call
598  *        from interrupt context.
599  *
600  * This function is the equivalent to use call_usermodehelper_setup() and
601  * call_usermodehelper_exec().
602  */
603 int call_usermodehelper(char *path, char **argv, char **envp, int wait)
604 {
605         struct subprocess_info *info;
606         gfp_t gfp_mask = (wait == UMH_NO_WAIT) ? GFP_ATOMIC : GFP_KERNEL;
607
608         info = call_usermodehelper_setup(path, argv, envp, gfp_mask,
609                                          NULL, NULL, NULL);
610         if (info == NULL)
611                 return -ENOMEM;
612
613         return call_usermodehelper_exec(info, wait);
614 }
615 EXPORT_SYMBOL(call_usermodehelper);
616
617 static int proc_cap_handler(struct ctl_table *table, int write,
618                          void __user *buffer, size_t *lenp, loff_t *ppos)
619 {
620         struct ctl_table t;
621         unsigned long cap_array[_KERNEL_CAPABILITY_U32S];
622         kernel_cap_t new_cap;
623         int err, i;
624
625         if (write && (!capable(CAP_SETPCAP) ||
626                       !capable(CAP_SYS_MODULE)))
627                 return -EPERM;
628
629         /*
630          * convert from the global kernel_cap_t to the ulong array to print to
631          * userspace if this is a read.
632          */
633         spin_lock(&umh_sysctl_lock);
634         for (i = 0; i < _KERNEL_CAPABILITY_U32S; i++)  {
635                 if (table->data == CAP_BSET)
636                         cap_array[i] = usermodehelper_bset.cap[i];
637                 else if (table->data == CAP_PI)
638                         cap_array[i] = usermodehelper_inheritable.cap[i];
639                 else
640                         BUG();
641         }
642         spin_unlock(&umh_sysctl_lock);
643
644         t = *table;
645         t.data = &cap_array;
646
647         /*
648          * actually read or write and array of ulongs from userspace.  Remember
649          * these are least significant 32 bits first
650          */
651         err = proc_doulongvec_minmax(&t, write, buffer, lenp, ppos);
652         if (err < 0)
653                 return err;
654
655         /*
656          * convert from the sysctl array of ulongs to the kernel_cap_t
657          * internal representation
658          */
659         for (i = 0; i < _KERNEL_CAPABILITY_U32S; i++)
660                 new_cap.cap[i] = cap_array[i];
661
662         /*
663          * Drop everything not in the new_cap (but don't add things)
664          */
665         spin_lock(&umh_sysctl_lock);
666         if (write) {
667                 if (table->data == CAP_BSET)
668                         usermodehelper_bset = cap_intersect(usermodehelper_bset, new_cap);
669                 if (table->data == CAP_PI)
670                         usermodehelper_inheritable = cap_intersect(usermodehelper_inheritable, new_cap);
671         }
672         spin_unlock(&umh_sysctl_lock);
673
674         return 0;
675 }
676
677 struct ctl_table usermodehelper_table[] = {
678         {
679                 .procname       = "bset",
680                 .data           = CAP_BSET,
681                 .maxlen         = _KERNEL_CAPABILITY_U32S * sizeof(unsigned long),
682                 .mode           = 0600,
683                 .proc_handler   = proc_cap_handler,
684         },
685         {
686                 .procname       = "inheritable",
687                 .data           = CAP_PI,
688                 .maxlen         = _KERNEL_CAPABILITY_U32S * sizeof(unsigned long),
689                 .mode           = 0600,
690                 .proc_handler   = proc_cap_handler,
691         },
692         { }
693 };
694
695 void __init usermodehelper_init(void)
696 {
697         khelper_wq = create_singlethread_workqueue("khelper");
698         BUG_ON(!khelper_wq);
699 }