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;
50 #define CAP_BSET (void *)1
51 #define CAP_PI (void *)2
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);
61 modprobe_path is set via /proc/sys.
63 char modprobe_path[KMOD_PATH_LEN] = "/sbin/modprobe";
65 static void free_modprobe_argv(struct subprocess_info *info)
67 kfree(info->argv[3]); /* check call_modprobe() */
71 static int call_modprobe(char *module_name, int wait)
73 struct subprocess_info *info;
74 static char *envp[] = {
77 "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
81 char **argv = kmalloc(sizeof(char *[5]), GFP_KERNEL);
85 module_name = kstrdup(module_name, GFP_KERNEL);
89 argv[0] = modprobe_path;
92 argv[3] = module_name; /* check free_modprobe_argv() */
95 info = call_usermodehelper_setup(modprobe_path, argv, envp, GFP_KERNEL,
96 NULL, free_modprobe_argv, NULL);
98 goto free_module_name;
100 return call_usermodehelper_exec(info, wait | UMH_KILLABLE);
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
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
123 * If module auto-loading support is disabled then this function
124 * becomes a no-operation.
126 int __request_module(bool wait, const char *fmt, ...)
129 char module_name[MODULE_NAME_LEN];
130 unsigned int max_modprobes;
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;
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.
142 WARN_ON_ONCE(wait && current_is_async());
144 if (!modprobe_path[0])
148 ret = vsnprintf(module_name, MODULE_NAME_LEN, fmt, args);
150 if (ret >= MODULE_NAME_LEN)
151 return -ENAMETOOLONG;
153 ret = security_kernel_module_request(module_name);
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.
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
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) {
175 "request_module: runaway loop modprobe %s\n",
179 atomic_dec(&kmod_concurrent);
183 trace_module_request(module_name, wait, _RET_IP_);
185 ret = call_modprobe(module_name, wait ? UMH_WAIT_PROC : UMH_WAIT_EXEC);
187 atomic_dec(&kmod_concurrent);
190 EXPORT_SYMBOL(__request_module);
191 #endif /* CONFIG_MODULES */
193 static void call_usermodehelper_freeinfo(struct subprocess_info *info)
196 (*info->cleanup)(info);
200 static void umh_complete(struct subprocess_info *sub_info)
202 struct completion *comp = xchg(&sub_info->complete, NULL);
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.
211 call_usermodehelper_freeinfo(sub_info);
215 * This is the task which runs the usermode application
217 static int ____call_usermodehelper(void *data)
219 struct subprocess_info *sub_info = data;
223 spin_lock_irq(¤t->sighand->siglock);
224 flush_signal_handlers(current, 1);
225 spin_unlock_irq(¤t->sighand->siglock);
227 /* We can run anywhere, unlike our parent keventd(). */
228 set_cpus_allowed_ptr(current, cpu_all_mask);
231 * Our parent is keventd, which runs with elevated scheduling priority.
232 * Avoid propagating that into the userspace child.
234 set_user_nice(current, 0);
237 new = prepare_kernel_cred(current);
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);
247 if (sub_info->init) {
248 retval = sub_info->init(sub_info, new);
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);
261 sub_info->retval = retval;
262 /* wait_for_helper() will call umh_complete if UHM_WAIT_PROC. */
263 if (!(sub_info->wait & UMH_WAIT_PROC))
264 umh_complete(sub_info);
270 /* Keventd can't block, but this (a child) can. */
271 static int wait_for_helper(void *data)
273 struct subprocess_info *sub_info = data;
276 /* If SIGCLD is ignored sys_wait4 won't populate the status. */
277 kernel_sigaction(SIGCHLD, SIG_DFL);
278 pid = kernel_thread(____call_usermodehelper, sub_info, SIGCHLD);
280 sub_info->retval = pid;
284 * Normally it is bogus to call wait4() from in-kernel because
285 * wait4() wants to write the exit code to a userspace address.
286 * But wait_for_helper() always runs as keventd, and put_user()
287 * to a kernel address works OK for kernel threads, due to their
288 * having an mm_segment_t which spans the entire address space.
290 * Thus the __user pointer cast is valid here.
292 sys_wait4(pid, (int __user *)&ret, 0, NULL);
295 * If ret is 0, either ____call_usermodehelper failed and the
296 * real error code is already in sub_info->retval or
297 * sub_info->retval is 0 anyway, so don't mess with it then.
300 sub_info->retval = ret;
303 umh_complete(sub_info);
307 /* This is run by khelper thread */
308 static void __call_usermodehelper(struct work_struct *work)
310 struct subprocess_info *sub_info =
311 container_of(work, struct subprocess_info, work);
314 if (sub_info->wait & UMH_WAIT_PROC)
315 pid = kernel_thread(wait_for_helper, sub_info,
316 CLONE_FS | CLONE_FILES | SIGCHLD);
318 pid = kernel_thread(____call_usermodehelper, sub_info,
322 sub_info->retval = pid;
323 umh_complete(sub_info);
328 * If set, call_usermodehelper_exec() will exit immediately returning -EBUSY
329 * (used for preventing user land processes from being created after the user
330 * land has been frozen during a system-wide hibernation or suspend operation).
331 * Should always be manipulated under umhelper_sem acquired for write.
333 static enum umh_disable_depth usermodehelper_disabled = UMH_DISABLED;
335 /* Number of helpers running */
336 static atomic_t running_helpers = ATOMIC_INIT(0);
339 * Wait queue head used by usermodehelper_disable() to wait for all running
342 static DECLARE_WAIT_QUEUE_HEAD(running_helpers_waitq);
345 * Used by usermodehelper_read_lock_wait() to wait for usermodehelper_disabled
348 static DECLARE_WAIT_QUEUE_HEAD(usermodehelper_disabled_waitq);
351 * Time to wait for running_helpers to become zero before the setting of
352 * usermodehelper_disabled in usermodehelper_disable() fails
354 #define RUNNING_HELPERS_TIMEOUT (5 * HZ)
356 int usermodehelper_read_trylock(void)
361 down_read(&umhelper_sem);
363 prepare_to_wait(&usermodehelper_disabled_waitq, &wait,
365 if (!usermodehelper_disabled)
368 if (usermodehelper_disabled == UMH_DISABLED)
371 up_read(&umhelper_sem);
379 down_read(&umhelper_sem);
381 finish_wait(&usermodehelper_disabled_waitq, &wait);
384 EXPORT_SYMBOL_GPL(usermodehelper_read_trylock);
386 long usermodehelper_read_lock_wait(long timeout)
393 down_read(&umhelper_sem);
395 prepare_to_wait(&usermodehelper_disabled_waitq, &wait,
396 TASK_UNINTERRUPTIBLE);
397 if (!usermodehelper_disabled)
400 up_read(&umhelper_sem);
402 timeout = schedule_timeout(timeout);
406 down_read(&umhelper_sem);
408 finish_wait(&usermodehelper_disabled_waitq, &wait);
411 EXPORT_SYMBOL_GPL(usermodehelper_read_lock_wait);
413 void usermodehelper_read_unlock(void)
415 up_read(&umhelper_sem);
417 EXPORT_SYMBOL_GPL(usermodehelper_read_unlock);
420 * __usermodehelper_set_disable_depth - Modify usermodehelper_disabled.
421 * @depth: New value to assign to usermodehelper_disabled.
423 * Change the value of usermodehelper_disabled (under umhelper_sem locked for
424 * writing) and wakeup tasks waiting for it to change.
426 void __usermodehelper_set_disable_depth(enum umh_disable_depth depth)
428 down_write(&umhelper_sem);
429 usermodehelper_disabled = depth;
430 wake_up(&usermodehelper_disabled_waitq);
431 up_write(&umhelper_sem);
435 * __usermodehelper_disable - Prevent new helpers from being started.
436 * @depth: New value to assign to usermodehelper_disabled.
438 * Set usermodehelper_disabled to @depth and wait for running helpers to exit.
440 int __usermodehelper_disable(enum umh_disable_depth depth)
447 down_write(&umhelper_sem);
448 usermodehelper_disabled = depth;
449 up_write(&umhelper_sem);
452 * From now on call_usermodehelper_exec() won't start any new
453 * helpers, so it is sufficient if running_helpers turns out to
454 * be zero at one point (it may be increased later, but that
457 retval = wait_event_timeout(running_helpers_waitq,
458 atomic_read(&running_helpers) == 0,
459 RUNNING_HELPERS_TIMEOUT);
463 __usermodehelper_set_disable_depth(UMH_ENABLED);
467 static void helper_lock(void)
469 atomic_inc(&running_helpers);
470 smp_mb__after_atomic();
473 static void helper_unlock(void)
475 if (atomic_dec_and_test(&running_helpers))
476 wake_up(&running_helpers_waitq);
480 * call_usermodehelper_setup - prepare to call a usermode helper
481 * @path: path to usermode executable
482 * @argv: arg vector for process
483 * @envp: environment for process
484 * @gfp_mask: gfp mask for memory allocation
485 * @cleanup: a cleanup function
486 * @init: an init function
487 * @data: arbitrary context sensitive data
489 * Returns either %NULL on allocation failure, or a subprocess_info
490 * structure. This should be passed to call_usermodehelper_exec to
491 * exec the process and free the structure.
493 * The init function is used to customize the helper process prior to
494 * exec. A non-zero return code causes the process to error out, exit,
495 * and return the failure to the calling process
497 * The cleanup function is just before ethe subprocess_info is about to
498 * be freed. This can be used for freeing the argv and envp. The
499 * Function must be runnable in either a process context or the
500 * context in which call_usermodehelper_exec is called.
502 struct subprocess_info *call_usermodehelper_setup(char *path, char **argv,
503 char **envp, gfp_t gfp_mask,
504 int (*init)(struct subprocess_info *info, struct cred *new),
505 void (*cleanup)(struct subprocess_info *info),
508 struct subprocess_info *sub_info;
509 sub_info = kzalloc(sizeof(struct subprocess_info), gfp_mask);
513 INIT_WORK(&sub_info->work, __call_usermodehelper);
514 sub_info->path = path;
515 sub_info->argv = argv;
516 sub_info->envp = envp;
518 sub_info->cleanup = cleanup;
519 sub_info->init = init;
520 sub_info->data = data;
524 EXPORT_SYMBOL(call_usermodehelper_setup);
527 * call_usermodehelper_exec - start a usermode application
528 * @sub_info: information about the subprocessa
529 * @wait: wait for the application to finish and return status.
530 * when UMH_NO_WAIT don't wait at all, but you get no useful error back
531 * when the program couldn't be exec'ed. This makes it safe to call
532 * from interrupt context.
534 * Runs a user-space application. The application is started
535 * asynchronously if wait is not set, and runs as a child of keventd.
536 * (ie. it runs with full root capabilities).
538 int call_usermodehelper_exec(struct subprocess_info *sub_info, int wait)
540 DECLARE_COMPLETION_ONSTACK(done);
543 if (!sub_info->path) {
544 call_usermodehelper_freeinfo(sub_info);
548 if (!khelper_wq || usermodehelper_disabled) {
553 * Set the completion pointer only if there is a waiter.
554 * This makes it possible to use umh_complete to free
555 * the data structure in case of UMH_NO_WAIT.
557 sub_info->complete = (wait == UMH_NO_WAIT) ? NULL : &done;
558 sub_info->wait = wait;
560 queue_work(khelper_wq, &sub_info->work);
561 if (wait == UMH_NO_WAIT) /* task has freed sub_info */
564 if (wait & UMH_KILLABLE) {
565 retval = wait_for_completion_killable(&done);
569 /* umh_complete() will see NULL and free sub_info */
570 if (xchg(&sub_info->complete, NULL))
572 /* fallthrough, umh_complete() was already called */
575 wait_for_completion(&done);
577 retval = sub_info->retval;
579 call_usermodehelper_freeinfo(sub_info);
584 EXPORT_SYMBOL(call_usermodehelper_exec);
587 * call_usermodehelper() - prepare and start a usermode application
588 * @path: path to usermode executable
589 * @argv: arg vector for process
590 * @envp: environment for process
591 * @wait: wait for the application to finish and return status.
592 * when UMH_NO_WAIT don't wait at all, but you get no useful error back
593 * when the program couldn't be exec'ed. This makes it safe to call
594 * from interrupt context.
596 * This function is the equivalent to use call_usermodehelper_setup() and
597 * call_usermodehelper_exec().
599 int call_usermodehelper(char *path, char **argv, char **envp, int wait)
601 struct subprocess_info *info;
602 gfp_t gfp_mask = (wait == UMH_NO_WAIT) ? GFP_ATOMIC : GFP_KERNEL;
604 info = call_usermodehelper_setup(path, argv, envp, gfp_mask,
609 return call_usermodehelper_exec(info, wait);
611 EXPORT_SYMBOL(call_usermodehelper);
613 static int proc_cap_handler(struct ctl_table *table, int write,
614 void __user *buffer, size_t *lenp, loff_t *ppos)
617 unsigned long cap_array[_KERNEL_CAPABILITY_U32S];
618 kernel_cap_t new_cap;
621 if (write && (!capable(CAP_SETPCAP) ||
622 !capable(CAP_SYS_MODULE)))
626 * convert from the global kernel_cap_t to the ulong array to print to
627 * userspace if this is a read.
629 spin_lock(&umh_sysctl_lock);
630 for (i = 0; i < _KERNEL_CAPABILITY_U32S; i++) {
631 if (table->data == CAP_BSET)
632 cap_array[i] = usermodehelper_bset.cap[i];
633 else if (table->data == CAP_PI)
634 cap_array[i] = usermodehelper_inheritable.cap[i];
638 spin_unlock(&umh_sysctl_lock);
644 * actually read or write and array of ulongs from userspace. Remember
645 * these are least significant 32 bits first
647 err = proc_doulongvec_minmax(&t, write, buffer, lenp, ppos);
652 * convert from the sysctl array of ulongs to the kernel_cap_t
653 * internal representation
655 for (i = 0; i < _KERNEL_CAPABILITY_U32S; i++)
656 new_cap.cap[i] = cap_array[i];
659 * Drop everything not in the new_cap (but don't add things)
661 spin_lock(&umh_sysctl_lock);
663 if (table->data == CAP_BSET)
664 usermodehelper_bset = cap_intersect(usermodehelper_bset, new_cap);
665 if (table->data == CAP_PI)
666 usermodehelper_inheritable = cap_intersect(usermodehelper_inheritable, new_cap);
668 spin_unlock(&umh_sysctl_lock);
673 struct ctl_table usermodehelper_table[] = {
677 .maxlen = _KERNEL_CAPABILITY_U32S * sizeof(unsigned long),
679 .proc_handler = proc_cap_handler,
682 .procname = "inheritable",
684 .maxlen = _KERNEL_CAPABILITY_U32S * sizeof(unsigned long),
686 .proc_handler = proc_cap_handler,
691 void __init usermodehelper_init(void)
693 khelper_wq = create_singlethread_workqueue("khelper");