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/sched/task.h>
24 #include <linux/syscalls.h>
25 #include <linux/unistd.h>
26 #include <linux/kmod.h>
27 #include <linux/slab.h>
28 #include <linux/completion.h>
29 #include <linux/cred.h>
30 #include <linux/file.h>
31 #include <linux/fdtable.h>
32 #include <linux/workqueue.h>
33 #include <linux/security.h>
34 #include <linux/mount.h>
35 #include <linux/kernel.h>
36 #include <linux/init.h>
37 #include <linux/resource.h>
38 #include <linux/notifier.h>
39 #include <linux/suspend.h>
40 #include <linux/rwsem.h>
41 #include <linux/ptrace.h>
42 #include <linux/async.h>
43 #include <linux/uaccess.h>
45 #include <trace/events/module.h>
47 extern int max_threads;
49 #define CAP_BSET (void *)1
50 #define CAP_PI (void *)2
52 static kernel_cap_t usermodehelper_bset = CAP_FULL_SET;
53 static kernel_cap_t usermodehelper_inheritable = CAP_FULL_SET;
54 static DEFINE_SPINLOCK(umh_sysctl_lock);
55 static DECLARE_RWSEM(umhelper_sem);
60 modprobe_path is set via /proc/sys.
62 char modprobe_path[KMOD_PATH_LEN] = "/sbin/modprobe";
64 static void free_modprobe_argv(struct subprocess_info *info)
66 kfree(info->argv[3]); /* check call_modprobe() */
70 static int call_modprobe(char *module_name, int wait)
72 struct subprocess_info *info;
73 static char *envp[] = {
76 "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
80 char **argv = kmalloc(sizeof(char *[5]), GFP_KERNEL);
84 module_name = kstrdup(module_name, GFP_KERNEL);
88 argv[0] = modprobe_path;
91 argv[3] = module_name; /* check free_modprobe_argv() */
94 info = call_usermodehelper_setup(modprobe_path, argv, envp, GFP_KERNEL,
95 NULL, free_modprobe_argv, NULL);
97 goto free_module_name;
99 return call_usermodehelper_exec(info, wait | UMH_KILLABLE);
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 or positive exit code from
117 * "modprobe" on failure. Note that a successful module load does not mean
118 * the module did not then unload and exit on an error of its own. Callers
119 * must check that the service they requested is now available not blindly
122 * If module auto-loading support is disabled then this function
123 * becomes a no-operation.
125 int __request_module(bool wait, const char *fmt, ...)
128 char module_name[MODULE_NAME_LEN];
129 unsigned int max_modprobes;
131 static atomic_t kmod_concurrent = ATOMIC_INIT(0);
132 #define MAX_KMOD_CONCURRENT 50 /* Completely arbitrary value - KAO */
133 static int kmod_loop_msg;
136 * We don't allow synchronous module loading from async. Module
137 * init may invoke async_synchronize_full() which will end up
138 * waiting for this task which already is waiting for the module
139 * loading to complete, leading to a deadlock.
141 WARN_ON_ONCE(wait && current_is_async());
143 if (!modprobe_path[0])
147 ret = vsnprintf(module_name, MODULE_NAME_LEN, fmt, args);
149 if (ret >= MODULE_NAME_LEN)
150 return -ENAMETOOLONG;
152 ret = security_kernel_module_request(module_name);
156 /* If modprobe needs a service that is in a module, we get a recursive
157 * loop. Limit the number of running kmod threads to max_threads/2 or
158 * MAX_KMOD_CONCURRENT, whichever is the smaller. A cleaner method
159 * would be to run the parents of this process, counting how many times
160 * kmod was invoked. That would mean accessing the internals of the
161 * process tables to get the command line, proc_pid_cmdline is static
162 * and it is not worth changing the proc code just to handle this case.
165 * "trace the ppid" is simple, but will fail if someone's
166 * parent exits. I think this is as good as it gets. --RR
168 max_modprobes = min(max_threads/2, MAX_KMOD_CONCURRENT);
169 atomic_inc(&kmod_concurrent);
170 if (atomic_read(&kmod_concurrent) > max_modprobes) {
171 /* We may be blaming an innocent here, but unlikely */
172 if (kmod_loop_msg < 5) {
174 "request_module: runaway loop modprobe %s\n",
178 atomic_dec(&kmod_concurrent);
182 trace_module_request(module_name, wait, _RET_IP_);
184 ret = call_modprobe(module_name, wait ? UMH_WAIT_PROC : UMH_WAIT_EXEC);
186 atomic_dec(&kmod_concurrent);
189 EXPORT_SYMBOL(__request_module);
190 #endif /* CONFIG_MODULES */
192 static void call_usermodehelper_freeinfo(struct subprocess_info *info)
195 (*info->cleanup)(info);
199 static void umh_complete(struct subprocess_info *sub_info)
201 struct completion *comp = xchg(&sub_info->complete, NULL);
203 * See call_usermodehelper_exec(). If xchg() returns NULL
204 * we own sub_info, the UMH_KILLABLE caller has gone away
205 * or the caller used UMH_NO_WAIT.
210 call_usermodehelper_freeinfo(sub_info);
214 * This is the task which runs the usermode application
216 static int call_usermodehelper_exec_async(void *data)
218 struct subprocess_info *sub_info = data;
222 spin_lock_irq(¤t->sighand->siglock);
223 flush_signal_handlers(current, 1);
224 spin_unlock_irq(¤t->sighand->siglock);
227 * Our parent (unbound workqueue) runs with elevated scheduling
228 * priority. Avoid propagating that into the userspace child.
230 set_user_nice(current, 0);
233 new = prepare_kernel_cred(current);
237 spin_lock(&umh_sysctl_lock);
238 new->cap_bset = cap_intersect(usermodehelper_bset, new->cap_bset);
239 new->cap_inheritable = cap_intersect(usermodehelper_inheritable,
240 new->cap_inheritable);
241 spin_unlock(&umh_sysctl_lock);
243 if (sub_info->init) {
244 retval = sub_info->init(sub_info, new);
253 retval = do_execve(getname_kernel(sub_info->path),
254 (const char __user *const __user *)sub_info->argv,
255 (const char __user *const __user *)sub_info->envp);
257 sub_info->retval = retval;
259 * call_usermodehelper_exec_sync() will call umh_complete
262 if (!(sub_info->wait & UMH_WAIT_PROC))
263 umh_complete(sub_info);
269 /* Handles UMH_WAIT_PROC. */
270 static void call_usermodehelper_exec_sync(struct subprocess_info *sub_info)
274 /* If SIGCLD is ignored sys_wait4 won't populate the status. */
275 kernel_sigaction(SIGCHLD, SIG_DFL);
276 pid = kernel_thread(call_usermodehelper_exec_async, sub_info, SIGCHLD);
278 sub_info->retval = pid;
282 * Normally it is bogus to call wait4() from in-kernel because
283 * wait4() wants to write the exit code to a userspace address.
284 * But call_usermodehelper_exec_sync() always runs as kernel
285 * thread (workqueue) and put_user() to a kernel address works
286 * OK for kernel threads, due to their having an mm_segment_t
287 * which spans the entire address space.
289 * Thus the __user pointer cast is valid here.
291 sys_wait4(pid, (int __user *)&ret, 0, NULL);
294 * If ret is 0, either call_usermodehelper_exec_async failed and
295 * the real error code is already in sub_info->retval or
296 * sub_info->retval is 0 anyway, so don't mess with it then.
299 sub_info->retval = ret;
302 /* Restore default kernel sig handler */
303 kernel_sigaction(SIGCHLD, SIG_IGN);
305 umh_complete(sub_info);
309 * We need to create the usermodehelper kernel thread from a task that is affine
310 * to an optimized set of CPUs (or nohz housekeeping ones) such that they
311 * inherit a widest affinity irrespective of call_usermodehelper() callers with
312 * possibly reduced affinity (eg: per-cpu workqueues). We don't want
313 * usermodehelper targets to contend a busy CPU.
315 * Unbound workqueues provide such wide affinity and allow to block on
316 * UMH_WAIT_PROC requests without blocking pending request (up to some limit).
318 * Besides, workqueues provide the privilege level that caller might not have
319 * to perform the usermodehelper request.
322 static void call_usermodehelper_exec_work(struct work_struct *work)
324 struct subprocess_info *sub_info =
325 container_of(work, struct subprocess_info, work);
327 if (sub_info->wait & UMH_WAIT_PROC) {
328 call_usermodehelper_exec_sync(sub_info);
332 * Use CLONE_PARENT to reparent it to kthreadd; we do not
333 * want to pollute current->children, and we need a parent
334 * that always ignores SIGCHLD to ensure auto-reaping.
336 pid = kernel_thread(call_usermodehelper_exec_async, sub_info,
337 CLONE_PARENT | SIGCHLD);
339 sub_info->retval = pid;
340 umh_complete(sub_info);
346 * If set, call_usermodehelper_exec() will exit immediately returning -EBUSY
347 * (used for preventing user land processes from being created after the user
348 * land has been frozen during a system-wide hibernation or suspend operation).
349 * Should always be manipulated under umhelper_sem acquired for write.
351 static enum umh_disable_depth usermodehelper_disabled = UMH_DISABLED;
353 /* Number of helpers running */
354 static atomic_t running_helpers = ATOMIC_INIT(0);
357 * Wait queue head used by usermodehelper_disable() to wait for all running
360 static DECLARE_WAIT_QUEUE_HEAD(running_helpers_waitq);
363 * Used by usermodehelper_read_lock_wait() to wait for usermodehelper_disabled
366 static DECLARE_WAIT_QUEUE_HEAD(usermodehelper_disabled_waitq);
369 * Time to wait for running_helpers to become zero before the setting of
370 * usermodehelper_disabled in usermodehelper_disable() fails
372 #define RUNNING_HELPERS_TIMEOUT (5 * HZ)
374 int usermodehelper_read_trylock(void)
379 down_read(&umhelper_sem);
381 prepare_to_wait(&usermodehelper_disabled_waitq, &wait,
383 if (!usermodehelper_disabled)
386 if (usermodehelper_disabled == UMH_DISABLED)
389 up_read(&umhelper_sem);
397 down_read(&umhelper_sem);
399 finish_wait(&usermodehelper_disabled_waitq, &wait);
402 EXPORT_SYMBOL_GPL(usermodehelper_read_trylock);
404 long usermodehelper_read_lock_wait(long timeout)
411 down_read(&umhelper_sem);
413 prepare_to_wait(&usermodehelper_disabled_waitq, &wait,
414 TASK_UNINTERRUPTIBLE);
415 if (!usermodehelper_disabled)
418 up_read(&umhelper_sem);
420 timeout = schedule_timeout(timeout);
424 down_read(&umhelper_sem);
426 finish_wait(&usermodehelper_disabled_waitq, &wait);
429 EXPORT_SYMBOL_GPL(usermodehelper_read_lock_wait);
431 void usermodehelper_read_unlock(void)
433 up_read(&umhelper_sem);
435 EXPORT_SYMBOL_GPL(usermodehelper_read_unlock);
438 * __usermodehelper_set_disable_depth - Modify usermodehelper_disabled.
439 * @depth: New value to assign to usermodehelper_disabled.
441 * Change the value of usermodehelper_disabled (under umhelper_sem locked for
442 * writing) and wakeup tasks waiting for it to change.
444 void __usermodehelper_set_disable_depth(enum umh_disable_depth depth)
446 down_write(&umhelper_sem);
447 usermodehelper_disabled = depth;
448 wake_up(&usermodehelper_disabled_waitq);
449 up_write(&umhelper_sem);
453 * __usermodehelper_disable - Prevent new helpers from being started.
454 * @depth: New value to assign to usermodehelper_disabled.
456 * Set usermodehelper_disabled to @depth and wait for running helpers to exit.
458 int __usermodehelper_disable(enum umh_disable_depth depth)
465 down_write(&umhelper_sem);
466 usermodehelper_disabled = depth;
467 up_write(&umhelper_sem);
470 * From now on call_usermodehelper_exec() won't start any new
471 * helpers, so it is sufficient if running_helpers turns out to
472 * be zero at one point (it may be increased later, but that
475 retval = wait_event_timeout(running_helpers_waitq,
476 atomic_read(&running_helpers) == 0,
477 RUNNING_HELPERS_TIMEOUT);
481 __usermodehelper_set_disable_depth(UMH_ENABLED);
485 static void helper_lock(void)
487 atomic_inc(&running_helpers);
488 smp_mb__after_atomic();
491 static void helper_unlock(void)
493 if (atomic_dec_and_test(&running_helpers))
494 wake_up(&running_helpers_waitq);
498 * call_usermodehelper_setup - prepare to call a usermode helper
499 * @path: path to usermode executable
500 * @argv: arg vector for process
501 * @envp: environment for process
502 * @gfp_mask: gfp mask for memory allocation
503 * @cleanup: a cleanup function
504 * @init: an init function
505 * @data: arbitrary context sensitive data
507 * Returns either %NULL on allocation failure, or a subprocess_info
508 * structure. This should be passed to call_usermodehelper_exec to
509 * exec the process and free the structure.
511 * The init function is used to customize the helper process prior to
512 * exec. A non-zero return code causes the process to error out, exit,
513 * and return the failure to the calling process
515 * The cleanup function is just before ethe subprocess_info is about to
516 * be freed. This can be used for freeing the argv and envp. The
517 * Function must be runnable in either a process context or the
518 * context in which call_usermodehelper_exec is called.
520 struct subprocess_info *call_usermodehelper_setup(const char *path, char **argv,
521 char **envp, gfp_t gfp_mask,
522 int (*init)(struct subprocess_info *info, struct cred *new),
523 void (*cleanup)(struct subprocess_info *info),
526 struct subprocess_info *sub_info;
527 sub_info = kzalloc(sizeof(struct subprocess_info), gfp_mask);
531 INIT_WORK(&sub_info->work, call_usermodehelper_exec_work);
533 #ifdef CONFIG_STATIC_USERMODEHELPER
534 sub_info->path = CONFIG_STATIC_USERMODEHELPER_PATH;
536 sub_info->path = path;
538 sub_info->argv = argv;
539 sub_info->envp = envp;
541 sub_info->cleanup = cleanup;
542 sub_info->init = init;
543 sub_info->data = data;
547 EXPORT_SYMBOL(call_usermodehelper_setup);
550 * call_usermodehelper_exec - start a usermode application
551 * @sub_info: information about the subprocessa
552 * @wait: wait for the application to finish and return status.
553 * when UMH_NO_WAIT don't wait at all, but you get no useful error back
554 * when the program couldn't be exec'ed. This makes it safe to call
555 * from interrupt context.
557 * Runs a user-space application. The application is started
558 * asynchronously if wait is not set, and runs as a child of system workqueues.
559 * (ie. it runs with full root capabilities and optimized affinity).
561 int call_usermodehelper_exec(struct subprocess_info *sub_info, int wait)
563 DECLARE_COMPLETION_ONSTACK(done);
566 if (!sub_info->path) {
567 call_usermodehelper_freeinfo(sub_info);
571 if (usermodehelper_disabled) {
577 * If there is no binary for us to call, then just return and get out of
578 * here. This allows us to set STATIC_USERMODEHELPER_PATH to "" and
579 * disable all call_usermodehelper() calls.
581 if (strlen(sub_info->path) == 0)
585 * Set the completion pointer only if there is a waiter.
586 * This makes it possible to use umh_complete to free
587 * the data structure in case of UMH_NO_WAIT.
589 sub_info->complete = (wait == UMH_NO_WAIT) ? NULL : &done;
590 sub_info->wait = wait;
592 queue_work(system_unbound_wq, &sub_info->work);
593 if (wait == UMH_NO_WAIT) /* task has freed sub_info */
596 if (wait & UMH_KILLABLE) {
597 retval = wait_for_completion_killable(&done);
601 /* umh_complete() will see NULL and free sub_info */
602 if (xchg(&sub_info->complete, NULL))
604 /* fallthrough, umh_complete() was already called */
607 wait_for_completion(&done);
609 retval = sub_info->retval;
611 call_usermodehelper_freeinfo(sub_info);
616 EXPORT_SYMBOL(call_usermodehelper_exec);
619 * call_usermodehelper() - prepare and start a usermode application
620 * @path: path to usermode executable
621 * @argv: arg vector for process
622 * @envp: environment for process
623 * @wait: wait for the application to finish and return status.
624 * when UMH_NO_WAIT don't wait at all, but you get no useful error back
625 * when the program couldn't be exec'ed. This makes it safe to call
626 * from interrupt context.
628 * This function is the equivalent to use call_usermodehelper_setup() and
629 * call_usermodehelper_exec().
631 int call_usermodehelper(const char *path, char **argv, char **envp, int wait)
633 struct subprocess_info *info;
634 gfp_t gfp_mask = (wait == UMH_NO_WAIT) ? GFP_ATOMIC : GFP_KERNEL;
636 info = call_usermodehelper_setup(path, argv, envp, gfp_mask,
641 return call_usermodehelper_exec(info, wait);
643 EXPORT_SYMBOL(call_usermodehelper);
645 static int proc_cap_handler(struct ctl_table *table, int write,
646 void __user *buffer, size_t *lenp, loff_t *ppos)
649 unsigned long cap_array[_KERNEL_CAPABILITY_U32S];
650 kernel_cap_t new_cap;
653 if (write && (!capable(CAP_SETPCAP) ||
654 !capable(CAP_SYS_MODULE)))
658 * convert from the global kernel_cap_t to the ulong array to print to
659 * userspace if this is a read.
661 spin_lock(&umh_sysctl_lock);
662 for (i = 0; i < _KERNEL_CAPABILITY_U32S; i++) {
663 if (table->data == CAP_BSET)
664 cap_array[i] = usermodehelper_bset.cap[i];
665 else if (table->data == CAP_PI)
666 cap_array[i] = usermodehelper_inheritable.cap[i];
670 spin_unlock(&umh_sysctl_lock);
676 * actually read or write and array of ulongs from userspace. Remember
677 * these are least significant 32 bits first
679 err = proc_doulongvec_minmax(&t, write, buffer, lenp, ppos);
684 * convert from the sysctl array of ulongs to the kernel_cap_t
685 * internal representation
687 for (i = 0; i < _KERNEL_CAPABILITY_U32S; i++)
688 new_cap.cap[i] = cap_array[i];
691 * Drop everything not in the new_cap (but don't add things)
693 spin_lock(&umh_sysctl_lock);
695 if (table->data == CAP_BSET)
696 usermodehelper_bset = cap_intersect(usermodehelper_bset, new_cap);
697 if (table->data == CAP_PI)
698 usermodehelper_inheritable = cap_intersect(usermodehelper_inheritable, new_cap);
700 spin_unlock(&umh_sysctl_lock);
705 struct ctl_table usermodehelper_table[] = {
709 .maxlen = _KERNEL_CAPABILITY_U32S * sizeof(unsigned long),
711 .proc_handler = proc_cap_handler,
714 .procname = "inheritable",
716 .maxlen = _KERNEL_CAPABILITY_U32S * sizeof(unsigned long),
718 .proc_handler = proc_cap_handler,