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 static char *envp[] = {
83 "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
87 char **argv = kmalloc(sizeof(char *[5]), GFP_KERNEL);
91 module_name = kstrdup(module_name, GFP_KERNEL);
95 argv[0] = modprobe_path;
98 argv[3] = module_name; /* check free_modprobe_argv() */
101 return call_usermodehelper_fns(modprobe_path, argv, envp,
102 wait | UMH_KILLABLE, NULL, free_modprobe_argv, NULL);
110 * __request_module - try to load a kernel module
111 * @wait: wait (or not) for the operation to complete
112 * @fmt: printf style format string for the name of the module
113 * @...: arguments as specified in the format string
115 * Load a module using the user mode module loader. The function returns
116 * zero on success or a negative errno code on failure. Note that a
117 * successful module load does not mean the module did not then unload
118 * and exit on an error of its own. Callers must check that the service
119 * they requested is now available not blindly invoke it.
121 * If module auto-loading support is disabled then this function
122 * becomes a no-operation.
124 int __request_module(bool wait, const char *fmt, ...)
127 char module_name[MODULE_NAME_LEN];
128 unsigned int max_modprobes;
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;
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.
140 WARN_ON_ONCE(wait && current_is_async());
143 ret = vsnprintf(module_name, MODULE_NAME_LEN, fmt, args);
145 if (ret >= MODULE_NAME_LEN)
146 return -ENAMETOOLONG;
148 ret = security_kernel_module_request(module_name);
152 /* If modprobe needs a service that is in a module, we get a recursive
153 * loop. Limit the number of running kmod threads to max_threads/2 or
154 * MAX_KMOD_CONCURRENT, whichever is the smaller. A cleaner method
155 * would be to run the parents of this process, counting how many times
156 * kmod was invoked. That would mean accessing the internals of the
157 * process tables to get the command line, proc_pid_cmdline is static
158 * and it is not worth changing the proc code just to handle this case.
161 * "trace the ppid" is simple, but will fail if someone's
162 * parent exits. I think this is as good as it gets. --RR
164 max_modprobes = min(max_threads/2, MAX_KMOD_CONCURRENT);
165 atomic_inc(&kmod_concurrent);
166 if (atomic_read(&kmod_concurrent) > max_modprobes) {
167 /* We may be blaming an innocent here, but unlikely */
168 if (kmod_loop_msg < 5) {
170 "request_module: runaway loop modprobe %s\n",
174 atomic_dec(&kmod_concurrent);
178 trace_module_request(module_name, wait, _RET_IP_);
180 ret = call_modprobe(module_name, wait ? UMH_WAIT_PROC : UMH_WAIT_EXEC);
182 atomic_dec(&kmod_concurrent);
185 EXPORT_SYMBOL(__request_module);
186 #endif /* CONFIG_MODULES */
189 * This is the task which runs the usermode application
191 static int ____call_usermodehelper(void *data)
193 struct subprocess_info *sub_info = data;
197 spin_lock_irq(¤t->sighand->siglock);
198 flush_signal_handlers(current, 1);
199 spin_unlock_irq(¤t->sighand->siglock);
201 /* We can run anywhere, unlike our parent keventd(). */
202 set_cpus_allowed_ptr(current, cpu_all_mask);
205 * Our parent is keventd, which runs with elevated scheduling priority.
206 * Avoid propagating that into the userspace child.
208 set_user_nice(current, 0);
211 new = prepare_kernel_cred(current);
215 spin_lock(&umh_sysctl_lock);
216 new->cap_bset = cap_intersect(usermodehelper_bset, new->cap_bset);
217 new->cap_inheritable = cap_intersect(usermodehelper_inheritable,
218 new->cap_inheritable);
219 spin_unlock(&umh_sysctl_lock);
221 if (sub_info->init) {
222 retval = sub_info->init(sub_info, new);
231 retval = do_execve(sub_info->path,
232 (const char __user *const __user *)sub_info->argv,
233 (const char __user *const __user *)sub_info->envp);
239 sub_info->retval = retval;
243 static int call_helper(void *data)
245 /* Worker thread started blocking khelper thread. */
246 kmod_thread_locker = current;
247 return ____call_usermodehelper(data);
250 static void call_usermodehelper_freeinfo(struct subprocess_info *info)
253 (*info->cleanup)(info);
257 static void umh_complete(struct subprocess_info *sub_info)
259 struct completion *comp = xchg(&sub_info->complete, NULL);
261 * See call_usermodehelper_exec(). If xchg() returns NULL
262 * we own sub_info, the UMH_KILLABLE caller has gone away.
267 call_usermodehelper_freeinfo(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 spin_lock_irq(¤t->sighand->siglock);
278 current->sighand->action[SIGCHLD-1].sa.sa_handler = SIG_DFL;
279 spin_unlock_irq(¤t->sighand->siglock);
281 pid = kernel_thread(____call_usermodehelper, sub_info, SIGCHLD);
283 sub_info->retval = pid;
287 * Normally it is bogus to call wait4() from in-kernel because
288 * wait4() wants to write the exit code to a userspace address.
289 * But wait_for_helper() always runs as keventd, and put_user()
290 * to a kernel address works OK for kernel threads, due to their
291 * having an mm_segment_t which spans the entire address space.
293 * Thus the __user pointer cast is valid here.
295 sys_wait4(pid, (int __user *)&ret, 0, NULL);
298 * If ret is 0, either ____call_usermodehelper failed and the
299 * real error code is already in sub_info->retval or
300 * sub_info->retval is 0 anyway, so don't mess with it then.
303 sub_info->retval = ret;
306 umh_complete(sub_info);
310 /* This is run by khelper thread */
311 static void __call_usermodehelper(struct work_struct *work)
313 struct subprocess_info *sub_info =
314 container_of(work, struct subprocess_info, work);
315 int wait = sub_info->wait & ~UMH_KILLABLE;
318 /* CLONE_VFORK: wait until the usermode helper has execve'd
319 * successfully We need the data structures to stay around
320 * until that is done. */
321 if (wait == UMH_WAIT_PROC)
322 pid = kernel_thread(wait_for_helper, sub_info,
323 CLONE_FS | CLONE_FILES | SIGCHLD);
325 pid = kernel_thread(call_helper, sub_info,
326 CLONE_VFORK | SIGCHLD);
327 /* Worker thread stopped blocking khelper thread. */
328 kmod_thread_locker = NULL;
333 call_usermodehelper_freeinfo(sub_info);
342 sub_info->retval = pid;
343 umh_complete(sub_info);
348 * If set, call_usermodehelper_exec() will exit immediately returning -EBUSY
349 * (used for preventing user land processes from being created after the user
350 * land has been frozen during a system-wide hibernation or suspend operation).
351 * Should always be manipulated under umhelper_sem acquired for write.
353 static enum umh_disable_depth usermodehelper_disabled = UMH_DISABLED;
355 /* Number of helpers running */
356 static atomic_t running_helpers = ATOMIC_INIT(0);
359 * Wait queue head used by usermodehelper_disable() to wait for all running
362 static DECLARE_WAIT_QUEUE_HEAD(running_helpers_waitq);
365 * Used by usermodehelper_read_lock_wait() to wait for usermodehelper_disabled
368 static DECLARE_WAIT_QUEUE_HEAD(usermodehelper_disabled_waitq);
371 * Time to wait for running_helpers to become zero before the setting of
372 * usermodehelper_disabled in usermodehelper_disable() fails
374 #define RUNNING_HELPERS_TIMEOUT (5 * HZ)
376 int usermodehelper_read_trylock(void)
381 down_read(&umhelper_sem);
383 prepare_to_wait(&usermodehelper_disabled_waitq, &wait,
385 if (!usermodehelper_disabled)
388 if (usermodehelper_disabled == UMH_DISABLED)
391 up_read(&umhelper_sem);
399 down_read(&umhelper_sem);
401 finish_wait(&usermodehelper_disabled_waitq, &wait);
404 EXPORT_SYMBOL_GPL(usermodehelper_read_trylock);
406 long usermodehelper_read_lock_wait(long timeout)
413 down_read(&umhelper_sem);
415 prepare_to_wait(&usermodehelper_disabled_waitq, &wait,
416 TASK_UNINTERRUPTIBLE);
417 if (!usermodehelper_disabled)
420 up_read(&umhelper_sem);
422 timeout = schedule_timeout(timeout);
426 down_read(&umhelper_sem);
428 finish_wait(&usermodehelper_disabled_waitq, &wait);
431 EXPORT_SYMBOL_GPL(usermodehelper_read_lock_wait);
433 void usermodehelper_read_unlock(void)
435 up_read(&umhelper_sem);
437 EXPORT_SYMBOL_GPL(usermodehelper_read_unlock);
440 * __usermodehelper_set_disable_depth - Modify usermodehelper_disabled.
441 * @depth: New value to assign to usermodehelper_disabled.
443 * Change the value of usermodehelper_disabled (under umhelper_sem locked for
444 * writing) and wakeup tasks waiting for it to change.
446 void __usermodehelper_set_disable_depth(enum umh_disable_depth depth)
448 down_write(&umhelper_sem);
449 usermodehelper_disabled = depth;
450 wake_up(&usermodehelper_disabled_waitq);
451 up_write(&umhelper_sem);
455 * __usermodehelper_disable - Prevent new helpers from being started.
456 * @depth: New value to assign to usermodehelper_disabled.
458 * Set usermodehelper_disabled to @depth and wait for running helpers to exit.
460 int __usermodehelper_disable(enum umh_disable_depth depth)
467 down_write(&umhelper_sem);
468 usermodehelper_disabled = depth;
469 up_write(&umhelper_sem);
472 * From now on call_usermodehelper_exec() won't start any new
473 * helpers, so it is sufficient if running_helpers turns out to
474 * be zero at one point (it may be increased later, but that
477 retval = wait_event_timeout(running_helpers_waitq,
478 atomic_read(&running_helpers) == 0,
479 RUNNING_HELPERS_TIMEOUT);
483 __usermodehelper_set_disable_depth(UMH_ENABLED);
487 static void helper_lock(void)
489 atomic_inc(&running_helpers);
490 smp_mb__after_atomic_inc();
493 static void helper_unlock(void)
495 if (atomic_dec_and_test(&running_helpers))
496 wake_up(&running_helpers_waitq);
500 * call_usermodehelper_setup - prepare to call a usermode helper
501 * @path: path to usermode executable
502 * @argv: arg vector for process
503 * @envp: environment for process
504 * @gfp_mask: gfp mask for memory allocation
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.
511 struct subprocess_info *call_usermodehelper_setup(char *path, char **argv,
512 char **envp, gfp_t gfp_mask)
514 struct subprocess_info *sub_info;
515 sub_info = kzalloc(sizeof(struct subprocess_info), gfp_mask);
519 INIT_WORK(&sub_info->work, __call_usermodehelper);
520 sub_info->path = path;
521 sub_info->argv = argv;
522 sub_info->envp = envp;
528 * call_usermodehelper_setfns - set a cleanup/init function
529 * @info: a subprocess_info returned by call_usermodehelper_setup
530 * @cleanup: a cleanup function
531 * @init: an init function
532 * @data: arbitrary context sensitive data
534 * The init function is used to customize the helper process prior to
535 * exec. A non-zero return code causes the process to error out, exit,
536 * and return the failure to the calling process
538 * The cleanup function is just before ethe subprocess_info is about to
539 * be freed. This can be used for freeing the argv and envp. The
540 * Function must be runnable in either a process context or the
541 * context in which call_usermodehelper_exec is called.
544 void call_usermodehelper_setfns(struct subprocess_info *info,
545 int (*init)(struct subprocess_info *info, struct cred *new),
546 void (*cleanup)(struct subprocess_info *info),
549 info->cleanup = cleanup;
555 * call_usermodehelper_exec - start a usermode application
556 * @sub_info: information about the subprocessa
557 * @wait: wait for the application to finish and return status.
558 * when -1 don't wait at all, but you get no useful error back when
559 * the program couldn't be exec'ed. This makes it safe to call
560 * from interrupt context.
562 * Runs a user-space application. The application is started
563 * asynchronously if wait is not set, and runs as a child of keventd.
564 * (ie. it runs with full root capabilities).
567 int call_usermodehelper_exec(struct subprocess_info *sub_info, int wait)
569 DECLARE_COMPLETION_ONSTACK(done);
573 if (sub_info->path[0] == '\0')
576 if (!khelper_wq || usermodehelper_disabled) {
581 * Worker thread must not wait for khelper thread at below
582 * wait_for_completion() if the thread was created with CLONE_VFORK
583 * flag, for khelper thread is already waiting for the thread at
584 * wait_for_completion() in do_fork().
586 if (wait != UMH_NO_WAIT && current == kmod_thread_locker) {
591 sub_info->complete = &done;
592 sub_info->wait = wait;
594 queue_work(khelper_wq, &sub_info->work);
595 if (wait == UMH_NO_WAIT) /* task has freed sub_info */
598 if (wait & UMH_KILLABLE) {
599 retval = wait_for_completion_killable(&done);
603 /* umh_complete() will see NULL and free sub_info */
604 if (xchg(&sub_info->complete, NULL))
606 /* fallthrough, umh_complete() was already called */
609 wait_for_completion(&done);
611 retval = sub_info->retval;
613 call_usermodehelper_freeinfo(sub_info);
620 * call_usermodehelper_fns() will not run the caller-provided cleanup function
621 * if a memory allocation failure is experienced. So the caller might need to
622 * check the call_usermodehelper_fns() return value: if it is -ENOMEM, perform
623 * the necessaary cleanup within the caller.
625 int call_usermodehelper_fns(
626 char *path, char **argv, char **envp, int wait,
627 int (*init)(struct subprocess_info *info, struct cred *new),
628 void (*cleanup)(struct subprocess_info *), void *data)
630 struct subprocess_info *info;
631 gfp_t gfp_mask = (wait == UMH_NO_WAIT) ? GFP_ATOMIC : GFP_KERNEL;
633 info = call_usermodehelper_setup(path, argv, envp, gfp_mask);
638 call_usermodehelper_setfns(info, init, cleanup, data);
640 return call_usermodehelper_exec(info, wait);
642 EXPORT_SYMBOL(call_usermodehelper_fns);
644 static int proc_cap_handler(struct ctl_table *table, int write,
645 void __user *buffer, size_t *lenp, loff_t *ppos)
648 unsigned long cap_array[_KERNEL_CAPABILITY_U32S];
649 kernel_cap_t new_cap;
652 if (write && (!capable(CAP_SETPCAP) ||
653 !capable(CAP_SYS_MODULE)))
657 * convert from the global kernel_cap_t to the ulong array to print to
658 * userspace if this is a read.
660 spin_lock(&umh_sysctl_lock);
661 for (i = 0; i < _KERNEL_CAPABILITY_U32S; i++) {
662 if (table->data == CAP_BSET)
663 cap_array[i] = usermodehelper_bset.cap[i];
664 else if (table->data == CAP_PI)
665 cap_array[i] = usermodehelper_inheritable.cap[i];
669 spin_unlock(&umh_sysctl_lock);
675 * actually read or write and array of ulongs from userspace. Remember
676 * these are least significant 32 bits first
678 err = proc_doulongvec_minmax(&t, write, buffer, lenp, ppos);
683 * convert from the sysctl array of ulongs to the kernel_cap_t
684 * internal representation
686 for (i = 0; i < _KERNEL_CAPABILITY_U32S; i++)
687 new_cap.cap[i] = cap_array[i];
690 * Drop everything not in the new_cap (but don't add things)
692 spin_lock(&umh_sysctl_lock);
694 if (table->data == CAP_BSET)
695 usermodehelper_bset = cap_intersect(usermodehelper_bset, new_cap);
696 if (table->data == CAP_PI)
697 usermodehelper_inheritable = cap_intersect(usermodehelper_inheritable, new_cap);
699 spin_unlock(&umh_sysctl_lock);
704 struct ctl_table usermodehelper_table[] = {
708 .maxlen = _KERNEL_CAPABILITY_U32S * sizeof(unsigned long),
710 .proc_handler = proc_cap_handler,
713 .procname = "inheritable",
715 .maxlen = _KERNEL_CAPABILITY_U32S * sizeof(unsigned long),
717 .proc_handler = proc_cap_handler,
722 void __init usermodehelper_init(void)
724 khelper_wq = create_singlethread_workqueue("khelper");