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";
64 * __request_module - try to load a kernel module
65 * @wait: wait (or not) for the operation to complete
66 * @fmt: printf style format string for the name of the module
67 * @...: arguments as specified in the format string
69 * Load a module using the user mode module loader. The function returns
70 * zero on success or a negative errno code on failure. Note that a
71 * successful module load does not mean the module did not then unload
72 * and exit on an error of its own. Callers must check that the service
73 * they requested is now available not blindly invoke it.
75 * If module auto-loading support is disabled then this function
76 * becomes a no-operation.
78 int __request_module(bool wait, const char *fmt, ...)
81 char module_name[MODULE_NAME_LEN];
82 unsigned int max_modprobes;
84 char *argv[] = { modprobe_path, "-q", "--", module_name, NULL };
85 static char *envp[] = { "HOME=/",
87 "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
89 static atomic_t kmod_concurrent = ATOMIC_INIT(0);
90 #define MAX_KMOD_CONCURRENT 50 /* Completely arbitrary value - KAO */
91 static int kmod_loop_msg;
94 ret = vsnprintf(module_name, MODULE_NAME_LEN, fmt, args);
96 if (ret >= MODULE_NAME_LEN)
99 ret = security_kernel_module_request(module_name);
103 /* If modprobe needs a service that is in a module, we get a recursive
104 * loop. Limit the number of running kmod threads to max_threads/2 or
105 * MAX_KMOD_CONCURRENT, whichever is the smaller. A cleaner method
106 * would be to run the parents of this process, counting how many times
107 * kmod was invoked. That would mean accessing the internals of the
108 * process tables to get the command line, proc_pid_cmdline is static
109 * and it is not worth changing the proc code just to handle this case.
112 * "trace the ppid" is simple, but will fail if someone's
113 * parent exits. I think this is as good as it gets. --RR
115 max_modprobes = min(max_threads/2, MAX_KMOD_CONCURRENT);
116 atomic_inc(&kmod_concurrent);
117 if (atomic_read(&kmod_concurrent) > max_modprobes) {
118 /* We may be blaming an innocent here, but unlikely */
119 if (kmod_loop_msg < 5) {
121 "request_module: runaway loop modprobe %s\n",
125 atomic_dec(&kmod_concurrent);
129 trace_module_request(module_name, wait, _RET_IP_);
131 ret = call_usermodehelper_fns(modprobe_path, argv, envp,
132 wait ? UMH_WAIT_PROC : UMH_WAIT_EXEC,
135 atomic_dec(&kmod_concurrent);
138 EXPORT_SYMBOL(__request_module);
139 #endif /* CONFIG_MODULES */
142 * This is the task which runs the usermode application
144 static int ____call_usermodehelper(void *data)
146 struct subprocess_info *sub_info = data;
150 spin_lock_irq(¤t->sighand->siglock);
151 flush_signal_handlers(current, 1);
152 spin_unlock_irq(¤t->sighand->siglock);
154 /* We can run anywhere, unlike our parent keventd(). */
155 set_cpus_allowed_ptr(current, cpu_all_mask);
158 * Our parent is keventd, which runs with elevated scheduling priority.
159 * Avoid propagating that into the userspace child.
161 set_user_nice(current, 0);
164 new = prepare_kernel_cred(current);
168 spin_lock(&umh_sysctl_lock);
169 new->cap_bset = cap_intersect(usermodehelper_bset, new->cap_bset);
170 new->cap_inheritable = cap_intersect(usermodehelper_inheritable,
171 new->cap_inheritable);
172 spin_unlock(&umh_sysctl_lock);
174 if (sub_info->init) {
175 retval = sub_info->init(sub_info, new);
184 retval = kernel_execve(sub_info->path,
185 (const char *const *)sub_info->argv,
186 (const char *const *)sub_info->envp);
190 sub_info->retval = retval;
194 void call_usermodehelper_freeinfo(struct subprocess_info *info)
197 (*info->cleanup)(info);
200 EXPORT_SYMBOL(call_usermodehelper_freeinfo);
202 static void umh_complete(struct subprocess_info *sub_info)
204 complete(sub_info->complete);
207 /* Keventd can't block, but this (a child) can. */
208 static int wait_for_helper(void *data)
210 struct subprocess_info *sub_info = data;
213 /* If SIGCLD is ignored sys_wait4 won't populate the status. */
214 spin_lock_irq(¤t->sighand->siglock);
215 current->sighand->action[SIGCHLD-1].sa.sa_handler = SIG_DFL;
216 spin_unlock_irq(¤t->sighand->siglock);
218 pid = kernel_thread(____call_usermodehelper, sub_info, SIGCHLD);
220 sub_info->retval = pid;
224 * Normally it is bogus to call wait4() from in-kernel because
225 * wait4() wants to write the exit code to a userspace address.
226 * But wait_for_helper() always runs as keventd, and put_user()
227 * to a kernel address works OK for kernel threads, due to their
228 * having an mm_segment_t which spans the entire address space.
230 * Thus the __user pointer cast is valid here.
232 sys_wait4(pid, (int __user *)&ret, 0, NULL);
235 * If ret is 0, either ____call_usermodehelper failed and the
236 * real error code is already in sub_info->retval or
237 * sub_info->retval is 0 anyway, so don't mess with it then.
240 sub_info->retval = ret;
243 umh_complete(sub_info);
247 /* This is run by khelper thread */
248 static void __call_usermodehelper(struct work_struct *work)
250 struct subprocess_info *sub_info =
251 container_of(work, struct subprocess_info, work);
252 enum umh_wait wait = sub_info->wait;
255 /* CLONE_VFORK: wait until the usermode helper has execve'd
256 * successfully We need the data structures to stay around
257 * until that is done. */
258 if (wait == UMH_WAIT_PROC)
259 pid = kernel_thread(wait_for_helper, sub_info,
260 CLONE_FS | CLONE_FILES | SIGCHLD);
262 pid = kernel_thread(____call_usermodehelper, sub_info,
263 CLONE_VFORK | SIGCHLD);
267 call_usermodehelper_freeinfo(sub_info);
276 sub_info->retval = pid;
277 umh_complete(sub_info);
282 * If set, call_usermodehelper_exec() will exit immediately returning -EBUSY
283 * (used for preventing user land processes from being created after the user
284 * land has been frozen during a system-wide hibernation or suspend operation).
285 * Should always be manipulated under umhelper_sem acquired for write.
287 static int usermodehelper_disabled = 1;
289 /* Number of helpers running */
290 static atomic_t running_helpers = ATOMIC_INIT(0);
293 * Wait queue head used by usermodehelper_disable() to wait for all running
296 static DECLARE_WAIT_QUEUE_HEAD(running_helpers_waitq);
299 * Time to wait for running_helpers to become zero before the setting of
300 * usermodehelper_disabled in usermodehelper_disable() fails
302 #define RUNNING_HELPERS_TIMEOUT (5 * HZ)
304 void read_lock_usermodehelper(void)
306 down_read(&umhelper_sem);
308 EXPORT_SYMBOL_GPL(read_lock_usermodehelper);
310 void read_unlock_usermodehelper(void)
312 up_read(&umhelper_sem);
314 EXPORT_SYMBOL_GPL(read_unlock_usermodehelper);
317 * usermodehelper_disable - prevent new helpers from being started
319 int usermodehelper_disable(void)
323 down_write(&umhelper_sem);
324 usermodehelper_disabled = 1;
325 up_write(&umhelper_sem);
328 * From now on call_usermodehelper_exec() won't start any new
329 * helpers, so it is sufficient if running_helpers turns out to
330 * be zero at one point (it may be increased later, but that
333 retval = wait_event_timeout(running_helpers_waitq,
334 atomic_read(&running_helpers) == 0,
335 RUNNING_HELPERS_TIMEOUT);
339 down_write(&umhelper_sem);
340 usermodehelper_disabled = 0;
341 up_write(&umhelper_sem);
346 * usermodehelper_enable - allow new helpers to be started again
348 void usermodehelper_enable(void)
350 down_write(&umhelper_sem);
351 usermodehelper_disabled = 0;
352 up_write(&umhelper_sem);
356 * usermodehelper_is_disabled - check if new helpers are allowed to be started
358 bool usermodehelper_is_disabled(void)
360 return usermodehelper_disabled;
362 EXPORT_SYMBOL_GPL(usermodehelper_is_disabled);
364 static void helper_lock(void)
366 atomic_inc(&running_helpers);
367 smp_mb__after_atomic_inc();
370 static void helper_unlock(void)
372 if (atomic_dec_and_test(&running_helpers))
373 wake_up(&running_helpers_waitq);
377 * call_usermodehelper_setup - prepare to call a usermode helper
378 * @path: path to usermode executable
379 * @argv: arg vector for process
380 * @envp: environment for process
381 * @gfp_mask: gfp mask for memory allocation
383 * Returns either %NULL on allocation failure, or a subprocess_info
384 * structure. This should be passed to call_usermodehelper_exec to
385 * exec the process and free the structure.
387 struct subprocess_info *call_usermodehelper_setup(char *path, char **argv,
388 char **envp, gfp_t gfp_mask)
390 struct subprocess_info *sub_info;
391 sub_info = kzalloc(sizeof(struct subprocess_info), gfp_mask);
395 INIT_WORK(&sub_info->work, __call_usermodehelper);
396 sub_info->path = path;
397 sub_info->argv = argv;
398 sub_info->envp = envp;
402 EXPORT_SYMBOL(call_usermodehelper_setup);
405 * call_usermodehelper_setfns - set a cleanup/init function
406 * @info: a subprocess_info returned by call_usermodehelper_setup
407 * @cleanup: a cleanup function
408 * @init: an init function
409 * @data: arbitrary context sensitive data
411 * The init function is used to customize the helper process prior to
412 * exec. A non-zero return code causes the process to error out, exit,
413 * and return the failure to the calling process
415 * The cleanup function is just before ethe subprocess_info is about to
416 * be freed. This can be used for freeing the argv and envp. The
417 * Function must be runnable in either a process context or the
418 * context in which call_usermodehelper_exec is called.
420 void call_usermodehelper_setfns(struct subprocess_info *info,
421 int (*init)(struct subprocess_info *info, struct cred *new),
422 void (*cleanup)(struct subprocess_info *info),
425 info->cleanup = cleanup;
429 EXPORT_SYMBOL(call_usermodehelper_setfns);
432 * call_usermodehelper_exec - start a usermode application
433 * @sub_info: information about the subprocessa
434 * @wait: wait for the application to finish and return status.
435 * when -1 don't wait at all, but you get no useful error back when
436 * the program couldn't be exec'ed. This makes it safe to call
437 * from interrupt context.
439 * Runs a user-space application. The application is started
440 * asynchronously if wait is not set, and runs as a child of keventd.
441 * (ie. it runs with full root capabilities).
443 int call_usermodehelper_exec(struct subprocess_info *sub_info,
446 DECLARE_COMPLETION_ONSTACK(done);
450 if (sub_info->path[0] == '\0')
453 if (!khelper_wq || usermodehelper_disabled) {
458 sub_info->complete = &done;
459 sub_info->wait = wait;
461 queue_work(khelper_wq, &sub_info->work);
462 if (wait == UMH_NO_WAIT) /* task has freed sub_info */
464 wait_for_completion(&done);
465 retval = sub_info->retval;
468 call_usermodehelper_freeinfo(sub_info);
473 EXPORT_SYMBOL(call_usermodehelper_exec);
475 static int proc_cap_handler(struct ctl_table *table, int write,
476 void __user *buffer, size_t *lenp, loff_t *ppos)
479 unsigned long cap_array[_KERNEL_CAPABILITY_U32S];
480 kernel_cap_t new_cap;
483 if (write && (!capable(CAP_SETPCAP) ||
484 !capable(CAP_SYS_MODULE)))
488 * convert from the global kernel_cap_t to the ulong array to print to
489 * userspace if this is a read.
491 spin_lock(&umh_sysctl_lock);
492 for (i = 0; i < _KERNEL_CAPABILITY_U32S; i++) {
493 if (table->data == CAP_BSET)
494 cap_array[i] = usermodehelper_bset.cap[i];
495 else if (table->data == CAP_PI)
496 cap_array[i] = usermodehelper_inheritable.cap[i];
500 spin_unlock(&umh_sysctl_lock);
506 * actually read or write and array of ulongs from userspace. Remember
507 * these are least significant 32 bits first
509 err = proc_doulongvec_minmax(&t, write, buffer, lenp, ppos);
514 * convert from the sysctl array of ulongs to the kernel_cap_t
515 * internal representation
517 for (i = 0; i < _KERNEL_CAPABILITY_U32S; i++)
518 new_cap.cap[i] = cap_array[i];
521 * Drop everything not in the new_cap (but don't add things)
523 spin_lock(&umh_sysctl_lock);
525 if (table->data == CAP_BSET)
526 usermodehelper_bset = cap_intersect(usermodehelper_bset, new_cap);
527 if (table->data == CAP_PI)
528 usermodehelper_inheritable = cap_intersect(usermodehelper_inheritable, new_cap);
530 spin_unlock(&umh_sysctl_lock);
535 struct ctl_table usermodehelper_table[] = {
539 .maxlen = _KERNEL_CAPABILITY_U32S * sizeof(unsigned long),
541 .proc_handler = proc_cap_handler,
544 .procname = "inheritable",
546 .maxlen = _KERNEL_CAPABILITY_U32S * sizeof(unsigned long),
548 .proc_handler = proc_cap_handler,
553 void __init usermodehelper_init(void)
555 khelper_wq = create_singlethread_workqueue("khelper");