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