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