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 <asm/uaccess.h>
42 #include <trace/events/module.h>
44 extern int max_threads;
46 static struct workqueue_struct *khelper_wq;
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 int call_modprobe(char *module_name, int wait)
65 static char *envp[] = {
68 "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
72 char *argv[] = { modprobe_path, "-q", "--", module_name, NULL };
74 return call_usermodehelper_fns(modprobe_path, argv, envp,
75 wait, NULL, NULL, NULL);
79 * __request_module - try to load a kernel module
80 * @wait: wait (or not) for the operation to complete
81 * @fmt: printf style format string for the name of the module
82 * @...: arguments as specified in the format string
84 * Load a module using the user mode module loader. The function returns
85 * zero on success or a negative errno code on failure. Note that a
86 * successful module load does not mean the module did not then unload
87 * and exit on an error of its own. Callers must check that the service
88 * they requested is now available not blindly invoke it.
90 * If module auto-loading support is disabled then this function
91 * becomes a no-operation.
93 int __request_module(bool wait, const char *fmt, ...)
96 char module_name[MODULE_NAME_LEN];
97 unsigned int max_modprobes;
99 static atomic_t kmod_concurrent = ATOMIC_INIT(0);
100 #define MAX_KMOD_CONCURRENT 50 /* Completely arbitrary value - KAO */
101 static int kmod_loop_msg;
104 ret = vsnprintf(module_name, MODULE_NAME_LEN, fmt, args);
106 if (ret >= MODULE_NAME_LEN)
107 return -ENAMETOOLONG;
109 ret = security_kernel_module_request(module_name);
113 /* If modprobe needs a service that is in a module, we get a recursive
114 * loop. Limit the number of running kmod threads to max_threads/2 or
115 * MAX_KMOD_CONCURRENT, whichever is the smaller. A cleaner method
116 * would be to run the parents of this process, counting how many times
117 * kmod was invoked. That would mean accessing the internals of the
118 * process tables to get the command line, proc_pid_cmdline is static
119 * and it is not worth changing the proc code just to handle this case.
122 * "trace the ppid" is simple, but will fail if someone's
123 * parent exits. I think this is as good as it gets. --RR
125 max_modprobes = min(max_threads/2, MAX_KMOD_CONCURRENT);
126 atomic_inc(&kmod_concurrent);
127 if (atomic_read(&kmod_concurrent) > max_modprobes) {
128 /* We may be blaming an innocent here, but unlikely */
129 if (kmod_loop_msg < 5) {
131 "request_module: runaway loop modprobe %s\n",
135 atomic_dec(&kmod_concurrent);
139 trace_module_request(module_name, wait, _RET_IP_);
141 ret = call_modprobe(module_name, wait ? UMH_WAIT_PROC : UMH_WAIT_EXEC);
143 atomic_dec(&kmod_concurrent);
146 EXPORT_SYMBOL(__request_module);
147 #endif /* CONFIG_MODULES */
150 * This is the task which runs the usermode application
152 static int ____call_usermodehelper(void *data)
154 struct subprocess_info *sub_info = data;
158 spin_lock_irq(¤t->sighand->siglock);
159 flush_signal_handlers(current, 1);
160 spin_unlock_irq(¤t->sighand->siglock);
162 /* We can run anywhere, unlike our parent keventd(). */
163 set_cpus_allowed_ptr(current, cpu_all_mask);
166 * Our parent is keventd, which runs with elevated scheduling priority.
167 * Avoid propagating that into the userspace child.
169 set_user_nice(current, 0);
172 new = prepare_kernel_cred(current);
176 spin_lock(&umh_sysctl_lock);
177 new->cap_bset = cap_intersect(usermodehelper_bset, new->cap_bset);
178 new->cap_inheritable = cap_intersect(usermodehelper_inheritable,
179 new->cap_inheritable);
180 spin_unlock(&umh_sysctl_lock);
182 if (sub_info->init) {
183 retval = sub_info->init(sub_info, new);
192 retval = kernel_execve(sub_info->path,
193 (const char *const *)sub_info->argv,
194 (const char *const *)sub_info->envp);
198 sub_info->retval = retval;
202 void call_usermodehelper_freeinfo(struct subprocess_info *info)
205 (*info->cleanup)(info);
208 EXPORT_SYMBOL(call_usermodehelper_freeinfo);
210 static void umh_complete(struct subprocess_info *sub_info)
212 struct completion *comp = xchg(&sub_info->complete, NULL);
214 * See call_usermodehelper_exec(). If xchg() returns NULL
215 * we own sub_info, the UMH_KILLABLE caller has gone away.
220 call_usermodehelper_freeinfo(sub_info);
223 /* Keventd can't block, but this (a child) can. */
224 static int wait_for_helper(void *data)
226 struct subprocess_info *sub_info = data;
229 /* If SIGCLD is ignored sys_wait4 won't populate the status. */
230 spin_lock_irq(¤t->sighand->siglock);
231 current->sighand->action[SIGCHLD-1].sa.sa_handler = SIG_DFL;
232 spin_unlock_irq(¤t->sighand->siglock);
234 pid = kernel_thread(____call_usermodehelper, sub_info, SIGCHLD);
236 sub_info->retval = pid;
240 * Normally it is bogus to call wait4() from in-kernel because
241 * wait4() wants to write the exit code to a userspace address.
242 * But wait_for_helper() always runs as keventd, and put_user()
243 * to a kernel address works OK for kernel threads, due to their
244 * having an mm_segment_t which spans the entire address space.
246 * Thus the __user pointer cast is valid here.
248 sys_wait4(pid, (int __user *)&ret, 0, NULL);
251 * If ret is 0, either ____call_usermodehelper failed and the
252 * real error code is already in sub_info->retval or
253 * sub_info->retval is 0 anyway, so don't mess with it then.
256 sub_info->retval = ret;
259 umh_complete(sub_info);
263 /* This is run by khelper thread */
264 static void __call_usermodehelper(struct work_struct *work)
266 struct subprocess_info *sub_info =
267 container_of(work, struct subprocess_info, work);
268 int wait = sub_info->wait & ~UMH_KILLABLE;
271 /* CLONE_VFORK: wait until the usermode helper has execve'd
272 * successfully We need the data structures to stay around
273 * until that is done. */
274 if (wait == UMH_WAIT_PROC)
275 pid = kernel_thread(wait_for_helper, sub_info,
276 CLONE_FS | CLONE_FILES | SIGCHLD);
278 pid = kernel_thread(____call_usermodehelper, sub_info,
279 CLONE_VFORK | SIGCHLD);
283 call_usermodehelper_freeinfo(sub_info);
292 sub_info->retval = pid;
293 umh_complete(sub_info);
298 * If set, call_usermodehelper_exec() will exit immediately returning -EBUSY
299 * (used for preventing user land processes from being created after the user
300 * land has been frozen during a system-wide hibernation or suspend operation).
301 * Should always be manipulated under umhelper_sem acquired for write.
303 static int usermodehelper_disabled = 1;
305 /* Number of helpers running */
306 static atomic_t running_helpers = ATOMIC_INIT(0);
309 * Wait queue head used by usermodehelper_disable() to wait for all running
312 static DECLARE_WAIT_QUEUE_HEAD(running_helpers_waitq);
315 * Time to wait for running_helpers to become zero before the setting of
316 * usermodehelper_disabled in usermodehelper_disable() fails
318 #define RUNNING_HELPERS_TIMEOUT (5 * HZ)
320 void read_lock_usermodehelper(void)
322 down_read(&umhelper_sem);
324 EXPORT_SYMBOL_GPL(read_lock_usermodehelper);
326 void read_unlock_usermodehelper(void)
328 up_read(&umhelper_sem);
330 EXPORT_SYMBOL_GPL(read_unlock_usermodehelper);
333 * usermodehelper_disable - prevent new helpers from being started
335 int usermodehelper_disable(void)
339 down_write(&umhelper_sem);
340 usermodehelper_disabled = 1;
341 up_write(&umhelper_sem);
344 * From now on call_usermodehelper_exec() won't start any new
345 * helpers, so it is sufficient if running_helpers turns out to
346 * be zero at one point (it may be increased later, but that
349 retval = wait_event_timeout(running_helpers_waitq,
350 atomic_read(&running_helpers) == 0,
351 RUNNING_HELPERS_TIMEOUT);
355 down_write(&umhelper_sem);
356 usermodehelper_disabled = 0;
357 up_write(&umhelper_sem);
362 * usermodehelper_enable - allow new helpers to be started again
364 void usermodehelper_enable(void)
366 down_write(&umhelper_sem);
367 usermodehelper_disabled = 0;
368 up_write(&umhelper_sem);
372 * usermodehelper_is_disabled - check if new helpers are allowed to be started
374 bool usermodehelper_is_disabled(void)
376 return usermodehelper_disabled;
378 EXPORT_SYMBOL_GPL(usermodehelper_is_disabled);
380 static void helper_lock(void)
382 atomic_inc(&running_helpers);
383 smp_mb__after_atomic_inc();
386 static void helper_unlock(void)
388 if (atomic_dec_and_test(&running_helpers))
389 wake_up(&running_helpers_waitq);
393 * call_usermodehelper_setup - prepare to call a usermode helper
394 * @path: path to usermode executable
395 * @argv: arg vector for process
396 * @envp: environment for process
397 * @gfp_mask: gfp mask for memory allocation
399 * Returns either %NULL on allocation failure, or a subprocess_info
400 * structure. This should be passed to call_usermodehelper_exec to
401 * exec the process and free the structure.
403 struct subprocess_info *call_usermodehelper_setup(char *path, char **argv,
404 char **envp, gfp_t gfp_mask)
406 struct subprocess_info *sub_info;
407 sub_info = kzalloc(sizeof(struct subprocess_info), gfp_mask);
411 INIT_WORK(&sub_info->work, __call_usermodehelper);
412 sub_info->path = path;
413 sub_info->argv = argv;
414 sub_info->envp = envp;
418 EXPORT_SYMBOL(call_usermodehelper_setup);
421 * call_usermodehelper_setfns - set a cleanup/init function
422 * @info: a subprocess_info returned by call_usermodehelper_setup
423 * @cleanup: a cleanup function
424 * @init: an init function
425 * @data: arbitrary context sensitive data
427 * The init function is used to customize the helper process prior to
428 * exec. A non-zero return code causes the process to error out, exit,
429 * and return the failure to the calling process
431 * The cleanup function is just before ethe subprocess_info is about to
432 * be freed. This can be used for freeing the argv and envp. The
433 * Function must be runnable in either a process context or the
434 * context in which call_usermodehelper_exec is called.
436 void call_usermodehelper_setfns(struct subprocess_info *info,
437 int (*init)(struct subprocess_info *info, struct cred *new),
438 void (*cleanup)(struct subprocess_info *info),
441 info->cleanup = cleanup;
445 EXPORT_SYMBOL(call_usermodehelper_setfns);
448 * call_usermodehelper_exec - start a usermode application
449 * @sub_info: information about the subprocessa
450 * @wait: wait for the application to finish and return status.
451 * when -1 don't wait at all, but you get no useful error back when
452 * the program couldn't be exec'ed. This makes it safe to call
453 * from interrupt context.
455 * Runs a user-space application. The application is started
456 * asynchronously if wait is not set, and runs as a child of keventd.
457 * (ie. it runs with full root capabilities).
459 int call_usermodehelper_exec(struct subprocess_info *sub_info, int wait)
461 DECLARE_COMPLETION_ONSTACK(done);
465 if (sub_info->path[0] == '\0')
468 if (!khelper_wq || usermodehelper_disabled) {
473 sub_info->complete = &done;
474 sub_info->wait = wait;
476 queue_work(khelper_wq, &sub_info->work);
477 if (wait == UMH_NO_WAIT) /* task has freed sub_info */
480 if (wait & UMH_KILLABLE) {
481 retval = wait_for_completion_killable(&done);
485 /* umh_complete() will see NULL and free sub_info */
486 if (xchg(&sub_info->complete, NULL))
488 /* fallthrough, umh_complete() was already called */
491 wait_for_completion(&done);
493 retval = sub_info->retval;
495 call_usermodehelper_freeinfo(sub_info);
500 EXPORT_SYMBOL(call_usermodehelper_exec);
502 static int proc_cap_handler(struct ctl_table *table, int write,
503 void __user *buffer, size_t *lenp, loff_t *ppos)
506 unsigned long cap_array[_KERNEL_CAPABILITY_U32S];
507 kernel_cap_t new_cap;
510 if (write && (!capable(CAP_SETPCAP) ||
511 !capable(CAP_SYS_MODULE)))
515 * convert from the global kernel_cap_t to the ulong array to print to
516 * userspace if this is a read.
518 spin_lock(&umh_sysctl_lock);
519 for (i = 0; i < _KERNEL_CAPABILITY_U32S; i++) {
520 if (table->data == CAP_BSET)
521 cap_array[i] = usermodehelper_bset.cap[i];
522 else if (table->data == CAP_PI)
523 cap_array[i] = usermodehelper_inheritable.cap[i];
527 spin_unlock(&umh_sysctl_lock);
533 * actually read or write and array of ulongs from userspace. Remember
534 * these are least significant 32 bits first
536 err = proc_doulongvec_minmax(&t, write, buffer, lenp, ppos);
541 * convert from the sysctl array of ulongs to the kernel_cap_t
542 * internal representation
544 for (i = 0; i < _KERNEL_CAPABILITY_U32S; i++)
545 new_cap.cap[i] = cap_array[i];
548 * Drop everything not in the new_cap (but don't add things)
550 spin_lock(&umh_sysctl_lock);
552 if (table->data == CAP_BSET)
553 usermodehelper_bset = cap_intersect(usermodehelper_bset, new_cap);
554 if (table->data == CAP_PI)
555 usermodehelper_inheritable = cap_intersect(usermodehelper_inheritable, new_cap);
557 spin_unlock(&umh_sysctl_lock);
562 struct ctl_table usermodehelper_table[] = {
566 .maxlen = _KERNEL_CAPABILITY_U32S * sizeof(unsigned long),
568 .proc_handler = proc_cap_handler,
571 .procname = "inheritable",
573 .maxlen = _KERNEL_CAPABILITY_U32S * sizeof(unsigned long),
575 .proc_handler = proc_cap_handler,
580 void __init usermodehelper_init(void)
582 khelper_wq = create_singlethread_workqueue("khelper");