2 kmod, the new module loader (replaces kerneld)
5 Reorganized not to be a daemon by Adam Richter, with guidance
8 Modified to avoid chroot and file sharing problems.
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
15 Unblock all signals when we exec a usermode process.
16 Shuu Yamaguchi <shuu@wondernetworkresources.com> December 2000
18 call_usermodehelper wait flag, and remove exec_usermodehelper.
19 Rusty Russell <rusty@rustcorp.com.au> Jan 2003
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>
44 #include <trace/events/module.h>
46 extern int max_threads;
48 static struct workqueue_struct *khelper_wq;
51 * kmod_thread_locker is used for deadlock avoidance. There is no explicit
52 * locking to protect this global - it is private to the singleton khelper
53 * thread and should only ever be modified by that thread.
55 static const struct task_struct *kmod_thread_locker;
57 #define CAP_BSET (void *)1
58 #define CAP_PI (void *)2
60 static kernel_cap_t usermodehelper_bset = CAP_FULL_SET;
61 static kernel_cap_t usermodehelper_inheritable = CAP_FULL_SET;
62 static DEFINE_SPINLOCK(umh_sysctl_lock);
63 static DECLARE_RWSEM(umhelper_sem);
68 modprobe_path is set via /proc/sys.
70 char modprobe_path[KMOD_PATH_LEN] = "/sbin/modprobe";
72 static void free_modprobe_argv(struct subprocess_info *info)
74 kfree(info->argv[3]); /* check call_modprobe() */
78 static int call_modprobe(char *module_name, int wait)
80 struct subprocess_info *info;
81 static char *envp[] = {
84 "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
88 char **argv = kmalloc(sizeof(char *[5]), GFP_KERNEL);
92 module_name = kstrdup(module_name, GFP_KERNEL);
96 argv[0] = modprobe_path;
99 argv[3] = module_name; /* check free_modprobe_argv() */
102 info = call_usermodehelper_setup(modprobe_path, argv, envp, GFP_KERNEL,
103 NULL, free_modprobe_argv, NULL);
105 goto free_module_name;
107 return call_usermodehelper_exec(info, wait | UMH_KILLABLE);
118 * __request_module - try to load a kernel module
119 * @wait: wait (or not) for the operation to complete
120 * @fmt: printf style format string for the name of the module
121 * @...: arguments as specified in the format string
123 * Load a module using the user mode module loader. The function returns
124 * zero on success or a negative errno code on failure. Note that a
125 * successful module load does not mean the module did not then unload
126 * and exit on an error of its own. Callers must check that the service
127 * they requested is now available not blindly invoke it.
129 * If module auto-loading support is disabled then this function
130 * becomes a no-operation.
132 int __request_module(bool wait, const char *fmt, ...)
135 char module_name[MODULE_NAME_LEN];
136 unsigned int max_modprobes;
138 static atomic_t kmod_concurrent = ATOMIC_INIT(0);
139 #define MAX_KMOD_CONCURRENT 50 /* Completely arbitrary value - KAO */
140 static int kmod_loop_msg;
143 * We don't allow synchronous module loading from async. Module
144 * init may invoke async_synchronize_full() which will end up
145 * waiting for this task which already is waiting for the module
146 * loading to complete, leading to a deadlock.
148 WARN_ON_ONCE(wait && current_is_async());
150 if (!modprobe_path[0])
154 ret = vsnprintf(module_name, MODULE_NAME_LEN, fmt, args);
156 if (ret >= MODULE_NAME_LEN)
157 return -ENAMETOOLONG;
159 ret = security_kernel_module_request(module_name);
163 /* If modprobe needs a service that is in a module, we get a recursive
164 * loop. Limit the number of running kmod threads to max_threads/2 or
165 * MAX_KMOD_CONCURRENT, whichever is the smaller. A cleaner method
166 * would be to run the parents of this process, counting how many times
167 * kmod was invoked. That would mean accessing the internals of the
168 * process tables to get the command line, proc_pid_cmdline is static
169 * and it is not worth changing the proc code just to handle this case.
172 * "trace the ppid" is simple, but will fail if someone's
173 * parent exits. I think this is as good as it gets. --RR
175 max_modprobes = min(max_threads/2, MAX_KMOD_CONCURRENT);
176 atomic_inc(&kmod_concurrent);
177 if (atomic_read(&kmod_concurrent) > max_modprobes) {
178 /* We may be blaming an innocent here, but unlikely */
179 if (kmod_loop_msg < 5) {
181 "request_module: runaway loop modprobe %s\n",
185 atomic_dec(&kmod_concurrent);
189 trace_module_request(module_name, wait, _RET_IP_);
191 ret = call_modprobe(module_name, wait ? UMH_WAIT_PROC : UMH_WAIT_EXEC);
193 atomic_dec(&kmod_concurrent);
196 EXPORT_SYMBOL(__request_module);
197 #endif /* CONFIG_MODULES */
200 * This is the task which runs the usermode application
202 static int ____call_usermodehelper(void *data)
204 struct subprocess_info *sub_info = data;
208 spin_lock_irq(¤t->sighand->siglock);
209 flush_signal_handlers(current, 1);
210 spin_unlock_irq(¤t->sighand->siglock);
212 /* We can run anywhere, unlike our parent keventd(). */
213 set_cpus_allowed_ptr(current, cpu_all_mask);
216 * Our parent is keventd, which runs with elevated scheduling priority.
217 * Avoid propagating that into the userspace child.
219 set_user_nice(current, 0);
222 new = prepare_kernel_cred(current);
226 spin_lock(&umh_sysctl_lock);
227 new->cap_bset = cap_intersect(usermodehelper_bset, new->cap_bset);
228 new->cap_inheritable = cap_intersect(usermodehelper_inheritable,
229 new->cap_inheritable);
230 spin_unlock(&umh_sysctl_lock);
232 if (sub_info->init) {
233 retval = sub_info->init(sub_info, new);
242 retval = do_execve(sub_info->path,
243 (const char __user *const __user *)sub_info->argv,
244 (const char __user *const __user *)sub_info->envp);
250 sub_info->retval = retval;
254 static int call_helper(void *data)
256 /* Worker thread started blocking khelper thread. */
257 kmod_thread_locker = current;
258 return ____call_usermodehelper(data);
261 static void call_usermodehelper_freeinfo(struct subprocess_info *info)
264 (*info->cleanup)(info);
268 static void umh_complete(struct subprocess_info *sub_info)
270 struct completion *comp = xchg(&sub_info->complete, NULL);
272 * See call_usermodehelper_exec(). If xchg() returns NULL
273 * we own sub_info, the UMH_KILLABLE caller has gone away.
278 call_usermodehelper_freeinfo(sub_info);
281 /* Keventd can't block, but this (a child) can. */
282 static int wait_for_helper(void *data)
284 struct subprocess_info *sub_info = data;
287 /* If SIGCLD is ignored sys_wait4 won't populate the status. */
288 spin_lock_irq(¤t->sighand->siglock);
289 current->sighand->action[SIGCHLD-1].sa.sa_handler = SIG_DFL;
290 spin_unlock_irq(¤t->sighand->siglock);
292 pid = kernel_thread(____call_usermodehelper, sub_info, SIGCHLD);
294 sub_info->retval = pid;
298 * Normally it is bogus to call wait4() from in-kernel because
299 * wait4() wants to write the exit code to a userspace address.
300 * But wait_for_helper() always runs as keventd, and put_user()
301 * to a kernel address works OK for kernel threads, due to their
302 * having an mm_segment_t which spans the entire address space.
304 * Thus the __user pointer cast is valid here.
306 sys_wait4(pid, (int __user *)&ret, 0, NULL);
309 * If ret is 0, either ____call_usermodehelper failed and the
310 * real error code is already in sub_info->retval or
311 * sub_info->retval is 0 anyway, so don't mess with it then.
314 sub_info->retval = ret;
317 umh_complete(sub_info);
321 /* This is run by khelper thread */
322 static void __call_usermodehelper(struct work_struct *work)
324 struct subprocess_info *sub_info =
325 container_of(work, struct subprocess_info, work);
326 int wait = sub_info->wait & ~UMH_KILLABLE;
329 /* CLONE_VFORK: wait until the usermode helper has execve'd
330 * successfully We need the data structures to stay around
331 * until that is done. */
332 if (wait == UMH_WAIT_PROC)
333 pid = kernel_thread(wait_for_helper, sub_info,
334 CLONE_FS | CLONE_FILES | SIGCHLD);
336 pid = kernel_thread(call_helper, sub_info,
337 CLONE_VFORK | SIGCHLD);
338 /* Worker thread stopped blocking khelper thread. */
339 kmod_thread_locker = NULL;
344 call_usermodehelper_freeinfo(sub_info);
353 sub_info->retval = pid;
354 umh_complete(sub_info);
359 * If set, call_usermodehelper_exec() will exit immediately returning -EBUSY
360 * (used for preventing user land processes from being created after the user
361 * land has been frozen during a system-wide hibernation or suspend operation).
362 * Should always be manipulated under umhelper_sem acquired for write.
364 static enum umh_disable_depth usermodehelper_disabled = UMH_DISABLED;
366 /* Number of helpers running */
367 static atomic_t running_helpers = ATOMIC_INIT(0);
370 * Wait queue head used by usermodehelper_disable() to wait for all running
373 static DECLARE_WAIT_QUEUE_HEAD(running_helpers_waitq);
376 * Used by usermodehelper_read_lock_wait() to wait for usermodehelper_disabled
379 static DECLARE_WAIT_QUEUE_HEAD(usermodehelper_disabled_waitq);
382 * Time to wait for running_helpers to become zero before the setting of
383 * usermodehelper_disabled in usermodehelper_disable() fails
385 #define RUNNING_HELPERS_TIMEOUT (5 * HZ)
387 int usermodehelper_read_trylock(void)
392 down_read(&umhelper_sem);
394 prepare_to_wait(&usermodehelper_disabled_waitq, &wait,
396 if (!usermodehelper_disabled)
399 if (usermodehelper_disabled == UMH_DISABLED)
402 up_read(&umhelper_sem);
410 down_read(&umhelper_sem);
412 finish_wait(&usermodehelper_disabled_waitq, &wait);
415 EXPORT_SYMBOL_GPL(usermodehelper_read_trylock);
417 long usermodehelper_read_lock_wait(long timeout)
424 down_read(&umhelper_sem);
426 prepare_to_wait(&usermodehelper_disabled_waitq, &wait,
427 TASK_UNINTERRUPTIBLE);
428 if (!usermodehelper_disabled)
431 up_read(&umhelper_sem);
433 timeout = schedule_timeout(timeout);
437 down_read(&umhelper_sem);
439 finish_wait(&usermodehelper_disabled_waitq, &wait);
442 EXPORT_SYMBOL_GPL(usermodehelper_read_lock_wait);
444 void usermodehelper_read_unlock(void)
446 up_read(&umhelper_sem);
448 EXPORT_SYMBOL_GPL(usermodehelper_read_unlock);
451 * __usermodehelper_set_disable_depth - Modify usermodehelper_disabled.
452 * @depth: New value to assign to usermodehelper_disabled.
454 * Change the value of usermodehelper_disabled (under umhelper_sem locked for
455 * writing) and wakeup tasks waiting for it to change.
457 void __usermodehelper_set_disable_depth(enum umh_disable_depth depth)
459 down_write(&umhelper_sem);
460 usermodehelper_disabled = depth;
461 wake_up(&usermodehelper_disabled_waitq);
462 up_write(&umhelper_sem);
466 * __usermodehelper_disable - Prevent new helpers from being started.
467 * @depth: New value to assign to usermodehelper_disabled.
469 * Set usermodehelper_disabled to @depth and wait for running helpers to exit.
471 int __usermodehelper_disable(enum umh_disable_depth depth)
478 down_write(&umhelper_sem);
479 usermodehelper_disabled = depth;
480 up_write(&umhelper_sem);
483 * From now on call_usermodehelper_exec() won't start any new
484 * helpers, so it is sufficient if running_helpers turns out to
485 * be zero at one point (it may be increased later, but that
488 retval = wait_event_timeout(running_helpers_waitq,
489 atomic_read(&running_helpers) == 0,
490 RUNNING_HELPERS_TIMEOUT);
494 __usermodehelper_set_disable_depth(UMH_ENABLED);
498 static void helper_lock(void)
500 atomic_inc(&running_helpers);
501 smp_mb__after_atomic_inc();
504 static void helper_unlock(void)
506 if (atomic_dec_and_test(&running_helpers))
507 wake_up(&running_helpers_waitq);
511 * call_usermodehelper_setup - prepare to call a usermode helper
512 * @path: path to usermode executable
513 * @argv: arg vector for process
514 * @envp: environment for process
515 * @gfp_mask: gfp mask for memory allocation
516 * @cleanup: a cleanup function
517 * @init: an init function
518 * @data: arbitrary context sensitive data
520 * Returns either %NULL on allocation failure, or a subprocess_info
521 * structure. This should be passed to call_usermodehelper_exec to
522 * exec the process and free the structure.
524 * The init function is used to customize the helper process prior to
525 * exec. A non-zero return code causes the process to error out, exit,
526 * and return the failure to the calling process
528 * The cleanup function is just before ethe subprocess_info is about to
529 * be freed. This can be used for freeing the argv and envp. The
530 * Function must be runnable in either a process context or the
531 * context in which call_usermodehelper_exec is called.
533 struct subprocess_info *call_usermodehelper_setup(char *path, char **argv,
534 char **envp, gfp_t gfp_mask,
535 int (*init)(struct subprocess_info *info, struct cred *new),
536 void (*cleanup)(struct subprocess_info *info),
539 struct subprocess_info *sub_info;
540 sub_info = kzalloc(sizeof(struct subprocess_info), gfp_mask);
544 INIT_WORK(&sub_info->work, __call_usermodehelper);
545 sub_info->path = path;
546 sub_info->argv = argv;
547 sub_info->envp = envp;
549 sub_info->cleanup = cleanup;
550 sub_info->init = init;
551 sub_info->data = data;
555 EXPORT_SYMBOL(call_usermodehelper_setup);
558 * call_usermodehelper_exec - start a usermode application
559 * @sub_info: information about the subprocessa
560 * @wait: wait for the application to finish and return status.
561 * when UMH_NO_WAIT don't wait at all, but you get no useful error back
562 * when the program couldn't be exec'ed. This makes it safe to call
563 * from interrupt context.
565 * Runs a user-space application. The application is started
566 * asynchronously if wait is not set, and runs as a child of keventd.
567 * (ie. it runs with full root capabilities).
569 int call_usermodehelper_exec(struct subprocess_info *sub_info, int wait)
571 DECLARE_COMPLETION_ONSTACK(done);
575 if (!khelper_wq || usermodehelper_disabled) {
580 * Worker thread must not wait for khelper thread at below
581 * wait_for_completion() if the thread was created with CLONE_VFORK
582 * flag, for khelper thread is already waiting for the thread at
583 * wait_for_completion() in do_fork().
585 if (wait != UMH_NO_WAIT && current == kmod_thread_locker) {
590 sub_info->complete = &done;
591 sub_info->wait = wait;
593 queue_work(khelper_wq, &sub_info->work);
594 if (wait == UMH_NO_WAIT) /* task has freed sub_info */
597 if (wait & UMH_KILLABLE) {
598 retval = wait_for_completion_killable(&done);
602 /* umh_complete() will see NULL and free sub_info */
603 if (xchg(&sub_info->complete, NULL))
605 /* fallthrough, umh_complete() was already called */
608 wait_for_completion(&done);
610 retval = sub_info->retval;
612 call_usermodehelper_freeinfo(sub_info);
617 EXPORT_SYMBOL(call_usermodehelper_exec);
620 * call_usermodehelper() - prepare and start a usermode application
621 * @path: path to usermode executable
622 * @argv: arg vector for process
623 * @envp: environment for process
624 * @wait: wait for the application to finish and return status.
625 * when UMH_NO_WAIT don't wait at all, but you get no useful error back
626 * when the program couldn't be exec'ed. This makes it safe to call
627 * from interrupt context.
629 * This function is the equivalent to use call_usermodehelper_setup() and
630 * call_usermodehelper_exec().
632 int call_usermodehelper(char *path, char **argv, char **envp, int wait)
634 struct subprocess_info *info;
635 gfp_t gfp_mask = (wait == UMH_NO_WAIT) ? GFP_ATOMIC : GFP_KERNEL;
637 info = call_usermodehelper_setup(path, argv, envp, gfp_mask,
642 return call_usermodehelper_exec(info, wait);
644 EXPORT_SYMBOL(call_usermodehelper);
646 static int proc_cap_handler(struct ctl_table *table, int write,
647 void __user *buffer, size_t *lenp, loff_t *ppos)
650 unsigned long cap_array[_KERNEL_CAPABILITY_U32S];
651 kernel_cap_t new_cap;
654 if (write && (!capable(CAP_SETPCAP) ||
655 !capable(CAP_SYS_MODULE)))
659 * convert from the global kernel_cap_t to the ulong array to print to
660 * userspace if this is a read.
662 spin_lock(&umh_sysctl_lock);
663 for (i = 0; i < _KERNEL_CAPABILITY_U32S; i++) {
664 if (table->data == CAP_BSET)
665 cap_array[i] = usermodehelper_bset.cap[i];
666 else if (table->data == CAP_PI)
667 cap_array[i] = usermodehelper_inheritable.cap[i];
671 spin_unlock(&umh_sysctl_lock);
677 * actually read or write and array of ulongs from userspace. Remember
678 * these are least significant 32 bits first
680 err = proc_doulongvec_minmax(&t, write, buffer, lenp, ppos);
685 * convert from the sysctl array of ulongs to the kernel_cap_t
686 * internal representation
688 for (i = 0; i < _KERNEL_CAPABILITY_U32S; i++)
689 new_cap.cap[i] = cap_array[i];
692 * Drop everything not in the new_cap (but don't add things)
694 spin_lock(&umh_sysctl_lock);
696 if (table->data == CAP_BSET)
697 usermodehelper_bset = cap_intersect(usermodehelper_bset, new_cap);
698 if (table->data == CAP_PI)
699 usermodehelper_inheritable = cap_intersect(usermodehelper_inheritable, new_cap);
701 spin_unlock(&umh_sysctl_lock);
706 struct ctl_table usermodehelper_table[] = {
710 .maxlen = _KERNEL_CAPABILITY_U32S * sizeof(unsigned long),
712 .proc_handler = proc_cap_handler,
715 .procname = "inheritable",
717 .maxlen = _KERNEL_CAPABILITY_U32S * sizeof(unsigned long),
719 .proc_handler = proc_cap_handler,
724 void __init usermodehelper_init(void)
726 khelper_wq = create_singlethread_workqueue("khelper");