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 #define CAP_BSET (void *)1
49 #define CAP_PI (void *)2
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);
59 modprobe_path is set via /proc/sys.
61 char modprobe_path[KMOD_PATH_LEN] = "/sbin/modprobe";
63 static void free_modprobe_argv(struct subprocess_info *info)
65 kfree(info->argv[3]); /* check call_modprobe() */
69 static int call_modprobe(char *module_name, int wait)
71 struct subprocess_info *info;
72 static char *envp[] = {
75 "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
79 char **argv = kmalloc(sizeof(char *[5]), GFP_KERNEL);
83 module_name = kstrdup(module_name, GFP_KERNEL);
87 argv[0] = modprobe_path;
90 argv[3] = module_name; /* check free_modprobe_argv() */
93 info = call_usermodehelper_setup(modprobe_path, argv, envp, GFP_KERNEL,
94 NULL, free_modprobe_argv, NULL);
96 goto free_module_name;
98 return call_usermodehelper_exec(info, wait | UMH_KILLABLE);
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
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
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());
142 if (!modprobe_path[0])
146 ret = vsnprintf(module_name, MODULE_NAME_LEN, fmt, args);
148 if (ret >= MODULE_NAME_LEN)
149 return -ENAMETOOLONG;
151 ret = security_kernel_module_request(module_name);
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.
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
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) {
173 "request_module: runaway loop modprobe %s\n",
177 atomic_dec(&kmod_concurrent);
181 trace_module_request(module_name, wait, _RET_IP_);
183 ret = call_modprobe(module_name, wait ? UMH_WAIT_PROC : UMH_WAIT_EXEC);
185 atomic_dec(&kmod_concurrent);
188 EXPORT_SYMBOL(__request_module);
189 #endif /* CONFIG_MODULES */
191 static void call_usermodehelper_freeinfo(struct subprocess_info *info)
194 (*info->cleanup)(info);
198 static void umh_complete(struct subprocess_info *sub_info)
200 struct completion *comp = xchg(&sub_info->complete, NULL);
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.
209 call_usermodehelper_freeinfo(sub_info);
213 * This is the task which runs the usermode application
215 static int call_usermodehelper_exec_async(void *data)
217 struct subprocess_info *sub_info = data;
221 spin_lock_irq(¤t->sighand->siglock);
222 flush_signal_handlers(current, 1);
223 spin_unlock_irq(¤t->sighand->siglock);
226 * Our parent (unbound workqueue) runs with elevated scheduling
227 * priority. Avoid propagating that into the userspace child.
229 set_user_nice(current, 0);
232 new = prepare_kernel_cred(current);
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);
242 if (sub_info->init) {
243 retval = sub_info->init(sub_info, new);
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);
256 sub_info->retval = retval;
258 * call_usermodehelper_exec_sync() will call umh_complete
261 if (!(sub_info->wait & UMH_WAIT_PROC))
262 umh_complete(sub_info);
268 /* Handles UMH_WAIT_PROC. */
269 static void call_usermodehelper_exec_sync(struct subprocess_info *sub_info)
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);
277 sub_info->retval = pid;
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.
288 * Thus the __user pointer cast is valid here.
290 sys_wait4(pid, (int __user *)&ret, 0, NULL);
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.
298 sub_info->retval = ret;
301 /* Restore default kernel sig handler */
302 kernel_sigaction(SIGCHLD, SIG_IGN);
304 umh_complete(sub_info);
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.
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).
317 * Besides, workqueues provide the privilege level that caller might not have
318 * to perform the usermodehelper request.
321 static void call_usermodehelper_exec_work(struct work_struct *work)
323 struct subprocess_info *sub_info =
324 container_of(work, struct subprocess_info, work);
326 if (sub_info->wait & UMH_WAIT_PROC) {
327 call_usermodehelper_exec_sync(sub_info);
331 pid = kernel_thread(call_usermodehelper_exec_async, sub_info,
334 sub_info->retval = pid;
335 umh_complete(sub_info);
341 * If set, call_usermodehelper_exec() will exit immediately returning -EBUSY
342 * (used for preventing user land processes from being created after the user
343 * land has been frozen during a system-wide hibernation or suspend operation).
344 * Should always be manipulated under umhelper_sem acquired for write.
346 static enum umh_disable_depth usermodehelper_disabled = UMH_DISABLED;
348 /* Number of helpers running */
349 static atomic_t running_helpers = ATOMIC_INIT(0);
352 * Wait queue head used by usermodehelper_disable() to wait for all running
355 static DECLARE_WAIT_QUEUE_HEAD(running_helpers_waitq);
358 * Used by usermodehelper_read_lock_wait() to wait for usermodehelper_disabled
361 static DECLARE_WAIT_QUEUE_HEAD(usermodehelper_disabled_waitq);
364 * Time to wait for running_helpers to become zero before the setting of
365 * usermodehelper_disabled in usermodehelper_disable() fails
367 #define RUNNING_HELPERS_TIMEOUT (5 * HZ)
369 int usermodehelper_read_trylock(void)
374 down_read(&umhelper_sem);
376 prepare_to_wait(&usermodehelper_disabled_waitq, &wait,
378 if (!usermodehelper_disabled)
381 if (usermodehelper_disabled == UMH_DISABLED)
384 up_read(&umhelper_sem);
392 down_read(&umhelper_sem);
394 finish_wait(&usermodehelper_disabled_waitq, &wait);
397 EXPORT_SYMBOL_GPL(usermodehelper_read_trylock);
399 long usermodehelper_read_lock_wait(long timeout)
406 down_read(&umhelper_sem);
408 prepare_to_wait(&usermodehelper_disabled_waitq, &wait,
409 TASK_UNINTERRUPTIBLE);
410 if (!usermodehelper_disabled)
413 up_read(&umhelper_sem);
415 timeout = schedule_timeout(timeout);
419 down_read(&umhelper_sem);
421 finish_wait(&usermodehelper_disabled_waitq, &wait);
424 EXPORT_SYMBOL_GPL(usermodehelper_read_lock_wait);
426 void usermodehelper_read_unlock(void)
428 up_read(&umhelper_sem);
430 EXPORT_SYMBOL_GPL(usermodehelper_read_unlock);
433 * __usermodehelper_set_disable_depth - Modify usermodehelper_disabled.
434 * @depth: New value to assign to usermodehelper_disabled.
436 * Change the value of usermodehelper_disabled (under umhelper_sem locked for
437 * writing) and wakeup tasks waiting for it to change.
439 void __usermodehelper_set_disable_depth(enum umh_disable_depth depth)
441 down_write(&umhelper_sem);
442 usermodehelper_disabled = depth;
443 wake_up(&usermodehelper_disabled_waitq);
444 up_write(&umhelper_sem);
448 * __usermodehelper_disable - Prevent new helpers from being started.
449 * @depth: New value to assign to usermodehelper_disabled.
451 * Set usermodehelper_disabled to @depth and wait for running helpers to exit.
453 int __usermodehelper_disable(enum umh_disable_depth depth)
460 down_write(&umhelper_sem);
461 usermodehelper_disabled = depth;
462 up_write(&umhelper_sem);
465 * From now on call_usermodehelper_exec() won't start any new
466 * helpers, so it is sufficient if running_helpers turns out to
467 * be zero at one point (it may be increased later, but that
470 retval = wait_event_timeout(running_helpers_waitq,
471 atomic_read(&running_helpers) == 0,
472 RUNNING_HELPERS_TIMEOUT);
476 __usermodehelper_set_disable_depth(UMH_ENABLED);
480 static void helper_lock(void)
482 atomic_inc(&running_helpers);
483 smp_mb__after_atomic();
486 static void helper_unlock(void)
488 if (atomic_dec_and_test(&running_helpers))
489 wake_up(&running_helpers_waitq);
493 * call_usermodehelper_setup - prepare to call a usermode helper
494 * @path: path to usermode executable
495 * @argv: arg vector for process
496 * @envp: environment for process
497 * @gfp_mask: gfp mask for memory allocation
498 * @cleanup: a cleanup function
499 * @init: an init function
500 * @data: arbitrary context sensitive data
502 * Returns either %NULL on allocation failure, or a subprocess_info
503 * structure. This should be passed to call_usermodehelper_exec to
504 * exec the process and free the structure.
506 * The init function is used to customize the helper process prior to
507 * exec. A non-zero return code causes the process to error out, exit,
508 * and return the failure to the calling process
510 * The cleanup function is just before ethe subprocess_info is about to
511 * be freed. This can be used for freeing the argv and envp. The
512 * Function must be runnable in either a process context or the
513 * context in which call_usermodehelper_exec is called.
515 struct subprocess_info *call_usermodehelper_setup(char *path, char **argv,
516 char **envp, gfp_t gfp_mask,
517 int (*init)(struct subprocess_info *info, struct cred *new),
518 void (*cleanup)(struct subprocess_info *info),
521 struct subprocess_info *sub_info;
522 sub_info = kzalloc(sizeof(struct subprocess_info), gfp_mask);
526 INIT_WORK(&sub_info->work, call_usermodehelper_exec_work);
527 sub_info->path = path;
528 sub_info->argv = argv;
529 sub_info->envp = envp;
531 sub_info->cleanup = cleanup;
532 sub_info->init = init;
533 sub_info->data = data;
537 EXPORT_SYMBOL(call_usermodehelper_setup);
540 * call_usermodehelper_exec - start a usermode application
541 * @sub_info: information about the subprocessa
542 * @wait: wait for the application to finish and return status.
543 * when UMH_NO_WAIT don't wait at all, but you get no useful error back
544 * when the program couldn't be exec'ed. This makes it safe to call
545 * from interrupt context.
547 * Runs a user-space application. The application is started
548 * asynchronously if wait is not set, and runs as a child of system workqueues.
549 * (ie. it runs with full root capabilities and optimized affinity).
551 int call_usermodehelper_exec(struct subprocess_info *sub_info, int wait)
553 DECLARE_COMPLETION_ONSTACK(done);
556 if (!sub_info->path) {
557 call_usermodehelper_freeinfo(sub_info);
561 if (usermodehelper_disabled) {
566 * Set the completion pointer only if there is a waiter.
567 * This makes it possible to use umh_complete to free
568 * the data structure in case of UMH_NO_WAIT.
570 sub_info->complete = (wait == UMH_NO_WAIT) ? NULL : &done;
571 sub_info->wait = wait;
573 queue_work(system_unbound_wq, &sub_info->work);
574 if (wait == UMH_NO_WAIT) /* task has freed sub_info */
577 if (wait & UMH_KILLABLE) {
578 retval = wait_for_completion_killable(&done);
582 /* umh_complete() will see NULL and free sub_info */
583 if (xchg(&sub_info->complete, NULL))
585 /* fallthrough, umh_complete() was already called */
588 wait_for_completion(&done);
590 retval = sub_info->retval;
592 call_usermodehelper_freeinfo(sub_info);
597 EXPORT_SYMBOL(call_usermodehelper_exec);
600 * call_usermodehelper() - prepare and start a usermode application
601 * @path: path to usermode executable
602 * @argv: arg vector for process
603 * @envp: environment for process
604 * @wait: wait for the application to finish and return status.
605 * when UMH_NO_WAIT don't wait at all, but you get no useful error back
606 * when the program couldn't be exec'ed. This makes it safe to call
607 * from interrupt context.
609 * This function is the equivalent to use call_usermodehelper_setup() and
610 * call_usermodehelper_exec().
612 int call_usermodehelper(char *path, char **argv, char **envp, int wait)
614 struct subprocess_info *info;
615 gfp_t gfp_mask = (wait == UMH_NO_WAIT) ? GFP_ATOMIC : GFP_KERNEL;
617 info = call_usermodehelper_setup(path, argv, envp, gfp_mask,
622 return call_usermodehelper_exec(info, wait);
624 EXPORT_SYMBOL(call_usermodehelper);
626 static int proc_cap_handler(struct ctl_table *table, int write,
627 void __user *buffer, size_t *lenp, loff_t *ppos)
630 unsigned long cap_array[_KERNEL_CAPABILITY_U32S];
631 kernel_cap_t new_cap;
634 if (write && (!capable(CAP_SETPCAP) ||
635 !capable(CAP_SYS_MODULE)))
639 * convert from the global kernel_cap_t to the ulong array to print to
640 * userspace if this is a read.
642 spin_lock(&umh_sysctl_lock);
643 for (i = 0; i < _KERNEL_CAPABILITY_U32S; i++) {
644 if (table->data == CAP_BSET)
645 cap_array[i] = usermodehelper_bset.cap[i];
646 else if (table->data == CAP_PI)
647 cap_array[i] = usermodehelper_inheritable.cap[i];
651 spin_unlock(&umh_sysctl_lock);
657 * actually read or write and array of ulongs from userspace. Remember
658 * these are least significant 32 bits first
660 err = proc_doulongvec_minmax(&t, write, buffer, lenp, ppos);
665 * convert from the sysctl array of ulongs to the kernel_cap_t
666 * internal representation
668 for (i = 0; i < _KERNEL_CAPABILITY_U32S; i++)
669 new_cap.cap[i] = cap_array[i];
672 * Drop everything not in the new_cap (but don't add things)
674 spin_lock(&umh_sysctl_lock);
676 if (table->data == CAP_BSET)
677 usermodehelper_bset = cap_intersect(usermodehelper_bset, new_cap);
678 if (table->data == CAP_PI)
679 usermodehelper_inheritable = cap_intersect(usermodehelper_inheritable, new_cap);
681 spin_unlock(&umh_sysctl_lock);
686 struct ctl_table usermodehelper_table[] = {
690 .maxlen = _KERNEL_CAPABILITY_U32S * sizeof(unsigned long),
692 .proc_handler = proc_cap_handler,
695 .procname = "inheritable",
697 .maxlen = _KERNEL_CAPABILITY_U32S * sizeof(unsigned long),
699 .proc_handler = proc_cap_handler,