sched/loadavg: Fix loadavg artifacts on fully idle and on fully loaded systems
[pandora-kernel.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 <asm/uaccess.h>
40
41 #include <trace/events/module.h>
42
43 extern int max_threads;
44
45 static struct workqueue_struct *khelper_wq;
46
47 #define CAP_BSET        (void *)1
48 #define CAP_PI          (void *)2
49
50 static kernel_cap_t usermodehelper_bset = CAP_FULL_SET;
51 static kernel_cap_t usermodehelper_inheritable = CAP_FULL_SET;
52 static DEFINE_SPINLOCK(umh_sysctl_lock);
53
54 #ifdef CONFIG_MODULES
55
56 /*
57         modprobe_path is set via /proc/sys.
58 */
59 char modprobe_path[KMOD_PATH_LEN] = "/sbin/modprobe";
60
61 static void free_modprobe_argv(struct subprocess_info *info)
62 {
63         kfree(info->argv[3]); /* check call_modprobe() */
64         kfree(info->argv);
65 }
66
67 static int call_modprobe(char *module_name, int wait)
68 {
69         static char *envp[] = {
70                 "HOME=/",
71                 "TERM=linux",
72                 "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
73                 NULL
74         };
75
76         char **argv = kmalloc(sizeof(char *[5]), GFP_KERNEL);
77         if (!argv)
78                 goto out;
79
80         module_name = kstrdup(module_name, GFP_KERNEL);
81         if (!module_name)
82                 goto free_argv;
83
84         argv[0] = modprobe_path;
85         argv[1] = "-q";
86         argv[2] = "--";
87         argv[3] = module_name;  /* check free_modprobe_argv() */
88         argv[4] = NULL;
89
90         return call_usermodehelper_fns(modprobe_path, argv, envp,
91                 wait | UMH_KILLABLE, NULL, free_modprobe_argv, NULL);
92 free_argv:
93         kfree(argv);
94 out:
95         return -ENOMEM;
96 }
97
98 /**
99  * __request_module - try to load a kernel module
100  * @wait: wait (or not) for the operation to complete
101  * @fmt: printf style format string for the name of the module
102  * @...: arguments as specified in the format string
103  *
104  * Load a module using the user mode module loader. The function returns
105  * zero on success or a negative errno code on failure. Note that a
106  * successful module load does not mean the module did not then unload
107  * and exit on an error of its own. Callers must check that the service
108  * they requested is now available not blindly invoke it.
109  *
110  * If module auto-loading support is disabled then this function
111  * becomes a no-operation.
112  */
113 int __request_module(bool wait, const char *fmt, ...)
114 {
115         va_list args;
116         char module_name[MODULE_NAME_LEN];
117         unsigned int max_modprobes;
118         int ret;
119         static atomic_t kmod_concurrent = ATOMIC_INIT(0);
120 #define MAX_KMOD_CONCURRENT 50  /* Completely arbitrary value - KAO */
121         static int kmod_loop_msg;
122
123         va_start(args, fmt);
124         ret = vsnprintf(module_name, MODULE_NAME_LEN, fmt, args);
125         va_end(args);
126         if (ret >= MODULE_NAME_LEN)
127                 return -ENAMETOOLONG;
128
129         ret = security_kernel_module_request(module_name);
130         if (ret)
131                 return ret;
132
133         /* If modprobe needs a service that is in a module, we get a recursive
134          * loop.  Limit the number of running kmod threads to max_threads/2 or
135          * MAX_KMOD_CONCURRENT, whichever is the smaller.  A cleaner method
136          * would be to run the parents of this process, counting how many times
137          * kmod was invoked.  That would mean accessing the internals of the
138          * process tables to get the command line, proc_pid_cmdline is static
139          * and it is not worth changing the proc code just to handle this case. 
140          * KAO.
141          *
142          * "trace the ppid" is simple, but will fail if someone's
143          * parent exits.  I think this is as good as it gets. --RR
144          */
145         max_modprobes = min(max_threads/2, MAX_KMOD_CONCURRENT);
146         atomic_inc(&kmod_concurrent);
147         if (atomic_read(&kmod_concurrent) > max_modprobes) {
148                 /* We may be blaming an innocent here, but unlikely */
149                 if (kmod_loop_msg < 5) {
150                         printk(KERN_ERR
151                                "request_module: runaway loop modprobe %s\n",
152                                module_name);
153                         kmod_loop_msg++;
154                 }
155                 atomic_dec(&kmod_concurrent);
156                 return -ENOMEM;
157         }
158
159         trace_module_request(module_name, wait, _RET_IP_);
160
161         ret = call_modprobe(module_name, wait ? UMH_WAIT_PROC : UMH_WAIT_EXEC);
162
163         atomic_dec(&kmod_concurrent);
164         return ret;
165 }
166 EXPORT_SYMBOL(__request_module);
167 #endif /* CONFIG_MODULES */
168
169 /*
170  * This is the task which runs the usermode application
171  */
172 static int ____call_usermodehelper(void *data)
173 {
174         struct subprocess_info *sub_info = data;
175         struct cred *new;
176         int retval;
177
178         spin_lock_irq(&current->sighand->siglock);
179         flush_signal_handlers(current, 1);
180         spin_unlock_irq(&current->sighand->siglock);
181
182         /* We can run anywhere, unlike our parent keventd(). */
183         set_cpus_allowed_ptr(current, cpu_all_mask);
184
185         /*
186          * Our parent is keventd, which runs with elevated scheduling priority.
187          * Avoid propagating that into the userspace child.
188          */
189         set_user_nice(current, 0);
190
191         retval = -ENOMEM;
192         new = prepare_kernel_cred(current);
193         if (!new)
194                 goto fail;
195
196         spin_lock(&umh_sysctl_lock);
197         new->cap_bset = cap_intersect(usermodehelper_bset, new->cap_bset);
198         new->cap_inheritable = cap_intersect(usermodehelper_inheritable,
199                                              new->cap_inheritable);
200         spin_unlock(&umh_sysctl_lock);
201
202         if (sub_info->init) {
203                 retval = sub_info->init(sub_info, new);
204                 if (retval) {
205                         abort_creds(new);
206                         goto fail;
207                 }
208         }
209
210         commit_creds(new);
211
212         retval = kernel_execve(sub_info->path,
213                                (const char *const *)sub_info->argv,
214                                (const char *const *)sub_info->envp);
215
216         /* Exec failed? */
217 fail:
218         sub_info->retval = retval;
219         return 0;
220 }
221
222 void call_usermodehelper_freeinfo(struct subprocess_info *info)
223 {
224         if (info->cleanup)
225                 (*info->cleanup)(info);
226         kfree(info);
227 }
228 EXPORT_SYMBOL(call_usermodehelper_freeinfo);
229
230 static void umh_complete(struct subprocess_info *sub_info)
231 {
232         struct completion *comp = xchg(&sub_info->complete, NULL);
233         /*
234          * See call_usermodehelper_exec(). If xchg() returns NULL
235          * we own sub_info, the UMH_KILLABLE caller has gone away.
236          */
237         if (comp)
238                 complete(comp);
239         else
240                 call_usermodehelper_freeinfo(sub_info);
241 }
242
243 /* Keventd can't block, but this (a child) can. */
244 static int wait_for_helper(void *data)
245 {
246         struct subprocess_info *sub_info = data;
247         pid_t pid;
248
249         /* If SIGCLD is ignored sys_wait4 won't populate the status. */
250         spin_lock_irq(&current->sighand->siglock);
251         current->sighand->action[SIGCHLD-1].sa.sa_handler = SIG_DFL;
252         spin_unlock_irq(&current->sighand->siglock);
253
254         pid = kernel_thread(____call_usermodehelper, sub_info, SIGCHLD);
255         if (pid < 0) {
256                 sub_info->retval = pid;
257         } else {
258                 int ret = -ECHILD;
259                 /*
260                  * Normally it is bogus to call wait4() from in-kernel because
261                  * wait4() wants to write the exit code to a userspace address.
262                  * But wait_for_helper() always runs as keventd, and put_user()
263                  * to a kernel address works OK for kernel threads, due to their
264                  * having an mm_segment_t which spans the entire address space.
265                  *
266                  * Thus the __user pointer cast is valid here.
267                  */
268                 sys_wait4(pid, (int __user *)&ret, 0, NULL);
269
270                 /*
271                  * If ret is 0, either ____call_usermodehelper failed and the
272                  * real error code is already in sub_info->retval or
273                  * sub_info->retval is 0 anyway, so don't mess with it then.
274                  */
275                 if (ret)
276                         sub_info->retval = ret;
277         }
278
279         umh_complete(sub_info);
280         return 0;
281 }
282
283 /* This is run by khelper thread  */
284 static void __call_usermodehelper(struct work_struct *work)
285 {
286         struct subprocess_info *sub_info =
287                 container_of(work, struct subprocess_info, work);
288         enum umh_wait wait = sub_info->wait;
289         pid_t pid;
290
291         if (wait != UMH_NO_WAIT)
292                 wait &= ~UMH_KILLABLE;
293
294         /* CLONE_VFORK: wait until the usermode helper has execve'd
295          * successfully We need the data structures to stay around
296          * until that is done.  */
297         if (wait == UMH_WAIT_PROC)
298                 pid = kernel_thread(wait_for_helper, sub_info,
299                                     CLONE_FS | CLONE_FILES | SIGCHLD);
300         else
301                 pid = kernel_thread(____call_usermodehelper, sub_info,
302                                     CLONE_VFORK | SIGCHLD);
303
304         switch (wait) {
305         case UMH_NO_WAIT:
306                 call_usermodehelper_freeinfo(sub_info);
307                 break;
308
309         case UMH_WAIT_PROC:
310                 if (pid > 0)
311                         break;
312                 /* FALLTHROUGH */
313         case UMH_WAIT_EXEC:
314                 if (pid < 0)
315                         sub_info->retval = pid;
316                 umh_complete(sub_info);
317         }
318 }
319
320 /*
321  * If set, call_usermodehelper_exec() will exit immediately returning -EBUSY
322  * (used for preventing user land processes from being created after the user
323  * land has been frozen during a system-wide hibernation or suspend operation).
324  */
325 static int usermodehelper_disabled = 1;
326
327 /* Number of helpers running */
328 static atomic_t running_helpers = ATOMIC_INIT(0);
329
330 /*
331  * Wait queue head used by usermodehelper_pm_callback() to wait for all running
332  * helpers to finish.
333  */
334 static DECLARE_WAIT_QUEUE_HEAD(running_helpers_waitq);
335
336 /*
337  * Time to wait for running_helpers to become zero before the setting of
338  * usermodehelper_disabled in usermodehelper_pm_callback() fails
339  */
340 #define RUNNING_HELPERS_TIMEOUT (5 * HZ)
341
342 /**
343  * usermodehelper_disable - prevent new helpers from being started
344  */
345 int usermodehelper_disable(void)
346 {
347         long retval;
348
349         usermodehelper_disabled = 1;
350         smp_mb();
351         /*
352          * From now on call_usermodehelper_exec() won't start any new
353          * helpers, so it is sufficient if running_helpers turns out to
354          * be zero at one point (it may be increased later, but that
355          * doesn't matter).
356          */
357         retval = wait_event_timeout(running_helpers_waitq,
358                                         atomic_read(&running_helpers) == 0,
359                                         RUNNING_HELPERS_TIMEOUT);
360         if (retval)
361                 return 0;
362
363         usermodehelper_disabled = 0;
364         return -EAGAIN;
365 }
366
367 /**
368  * usermodehelper_enable - allow new helpers to be started again
369  */
370 void usermodehelper_enable(void)
371 {
372         usermodehelper_disabled = 0;
373 }
374
375 /**
376  * usermodehelper_is_disabled - check if new helpers are allowed to be started
377  */
378 bool usermodehelper_is_disabled(void)
379 {
380         return usermodehelper_disabled;
381 }
382 EXPORT_SYMBOL_GPL(usermodehelper_is_disabled);
383
384 static void helper_lock(void)
385 {
386         atomic_inc(&running_helpers);
387         smp_mb__after_atomic_inc();
388 }
389
390 static void helper_unlock(void)
391 {
392         if (atomic_dec_and_test(&running_helpers))
393                 wake_up(&running_helpers_waitq);
394 }
395
396 /**
397  * call_usermodehelper_setup - prepare to call a usermode helper
398  * @path: path to usermode executable
399  * @argv: arg vector for process
400  * @envp: environment for process
401  * @gfp_mask: gfp mask for memory allocation
402  *
403  * Returns either %NULL on allocation failure, or a subprocess_info
404  * structure.  This should be passed to call_usermodehelper_exec to
405  * exec the process and free the structure.
406  */
407 struct subprocess_info *call_usermodehelper_setup(char *path, char **argv,
408                                                   char **envp, gfp_t gfp_mask)
409 {
410         struct subprocess_info *sub_info;
411         sub_info = kzalloc(sizeof(struct subprocess_info), gfp_mask);
412         if (!sub_info)
413                 goto out;
414
415         INIT_WORK(&sub_info->work, __call_usermodehelper);
416         sub_info->path = path;
417         sub_info->argv = argv;
418         sub_info->envp = envp;
419   out:
420         return sub_info;
421 }
422 EXPORT_SYMBOL(call_usermodehelper_setup);
423
424 /**
425  * call_usermodehelper_setfns - set a cleanup/init function
426  * @info: a subprocess_info returned by call_usermodehelper_setup
427  * @cleanup: a cleanup function
428  * @init: an init function
429  * @data: arbitrary context sensitive data
430  *
431  * The init function is used to customize the helper process prior to
432  * exec.  A non-zero return code causes the process to error out, exit,
433  * and return the failure to the calling process
434  *
435  * The cleanup function is just before ethe subprocess_info is about to
436  * be freed.  This can be used for freeing the argv and envp.  The
437  * Function must be runnable in either a process context or the
438  * context in which call_usermodehelper_exec is called.
439  */
440 void call_usermodehelper_setfns(struct subprocess_info *info,
441                     int (*init)(struct subprocess_info *info, struct cred *new),
442                     void (*cleanup)(struct subprocess_info *info),
443                     void *data)
444 {
445         info->cleanup = cleanup;
446         info->init = init;
447         info->data = data;
448 }
449 EXPORT_SYMBOL(call_usermodehelper_setfns);
450
451 /**
452  * call_usermodehelper_exec - start a usermode application
453  * @sub_info: information about the subprocessa
454  * @wait: wait for the application to finish and return status.
455  *        when -1 don't wait at all, but you get no useful error back when
456  *        the program couldn't be exec'ed. This makes it safe to call
457  *        from interrupt context.
458  *
459  * Runs a user-space application.  The application is started
460  * asynchronously if wait is not set, and runs as a child of keventd.
461  * (ie. it runs with full root capabilities).
462  */
463 int call_usermodehelper_exec(struct subprocess_info *sub_info,
464                              enum umh_wait wait)
465 {
466         DECLARE_COMPLETION_ONSTACK(done);
467         int retval = 0;
468
469         helper_lock();
470         if (!sub_info->path) {
471                 retval = -EINVAL;
472                 goto out;
473         }
474
475         if (sub_info->path[0] == '\0')
476                 goto out;
477
478         if (!khelper_wq || usermodehelper_disabled) {
479                 retval = -EBUSY;
480                 goto out;
481         }
482
483         sub_info->complete = &done;
484         sub_info->wait = wait;
485
486         queue_work(khelper_wq, &sub_info->work);
487         if (wait == UMH_NO_WAIT)        /* task has freed sub_info */
488                 goto unlock;
489
490         if (wait & UMH_KILLABLE) {
491                 retval = wait_for_completion_killable(&done);
492                 if (!retval)
493                         goto wait_done;
494
495                 /* umh_complete() will see NULL and free sub_info */
496                 if (xchg(&sub_info->complete, NULL))
497                         goto unlock;
498                 /* fallthrough, umh_complete() was already called */
499         }
500
501         wait_for_completion(&done);
502 wait_done:
503         retval = sub_info->retval;
504 out:
505         call_usermodehelper_freeinfo(sub_info);
506 unlock:
507         helper_unlock();
508         return retval;
509 }
510 EXPORT_SYMBOL(call_usermodehelper_exec);
511
512 static int proc_cap_handler(struct ctl_table *table, int write,
513                          void __user *buffer, size_t *lenp, loff_t *ppos)
514 {
515         struct ctl_table t;
516         unsigned long cap_array[_KERNEL_CAPABILITY_U32S];
517         kernel_cap_t new_cap;
518         int err, i;
519
520         if (write && (!capable(CAP_SETPCAP) ||
521                       !capable(CAP_SYS_MODULE)))
522                 return -EPERM;
523
524         /*
525          * convert from the global kernel_cap_t to the ulong array to print to
526          * userspace if this is a read.
527          */
528         spin_lock(&umh_sysctl_lock);
529         for (i = 0; i < _KERNEL_CAPABILITY_U32S; i++)  {
530                 if (table->data == CAP_BSET)
531                         cap_array[i] = usermodehelper_bset.cap[i];
532                 else if (table->data == CAP_PI)
533                         cap_array[i] = usermodehelper_inheritable.cap[i];
534                 else
535                         BUG();
536         }
537         spin_unlock(&umh_sysctl_lock);
538
539         t = *table;
540         t.data = &cap_array;
541
542         /*
543          * actually read or write and array of ulongs from userspace.  Remember
544          * these are least significant 32 bits first
545          */
546         err = proc_doulongvec_minmax(&t, write, buffer, lenp, ppos);
547         if (err < 0)
548                 return err;
549
550         /*
551          * convert from the sysctl array of ulongs to the kernel_cap_t
552          * internal representation
553          */
554         for (i = 0; i < _KERNEL_CAPABILITY_U32S; i++)
555                 new_cap.cap[i] = cap_array[i];
556
557         /*
558          * Drop everything not in the new_cap (but don't add things)
559          */
560         spin_lock(&umh_sysctl_lock);
561         if (write) {
562                 if (table->data == CAP_BSET)
563                         usermodehelper_bset = cap_intersect(usermodehelper_bset, new_cap);
564                 if (table->data == CAP_PI)
565                         usermodehelper_inheritable = cap_intersect(usermodehelper_inheritable, new_cap);
566         }
567         spin_unlock(&umh_sysctl_lock);
568
569         return 0;
570 }
571
572 struct ctl_table usermodehelper_table[] = {
573         {
574                 .procname       = "bset",
575                 .data           = CAP_BSET,
576                 .maxlen         = _KERNEL_CAPABILITY_U32S * sizeof(unsigned long),
577                 .mode           = 0600,
578                 .proc_handler   = proc_cap_handler,
579         },
580         {
581                 .procname       = "inheritable",
582                 .data           = CAP_PI,
583                 .maxlen         = _KERNEL_CAPABILITY_U32S * sizeof(unsigned long),
584                 .mode           = 0600,
585                 .proc_handler   = proc_cap_handler,
586         },
587         { }
588 };
589
590 void __init usermodehelper_init(void)
591 {
592         khelper_wq = create_singlethread_workqueue("khelper");
593         BUG_ON(!khelper_wq);
594 }