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/binfmts.h>
25 #include <linux/syscalls.h>
26 #include <linux/unistd.h>
27 #include <linux/kmod.h>
28 #include <linux/slab.h>
29 #include <linux/completion.h>
30 #include <linux/cred.h>
31 #include <linux/file.h>
32 #include <linux/fdtable.h>
33 #include <linux/workqueue.h>
34 #include <linux/security.h>
35 #include <linux/mount.h>
36 #include <linux/kernel.h>
37 #include <linux/init.h>
38 #include <linux/resource.h>
39 #include <linux/notifier.h>
40 #include <linux/suspend.h>
41 #include <linux/rwsem.h>
42 #include <linux/ptrace.h>
43 #include <linux/async.h>
44 #include <linux/uaccess.h>
46 #include <trace/events/module.h>
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
);
60 * threads = div64_u64((u64) totalram_pages * (u64) PAGE_SIZE,
61 * (u64) THREAD_SIZE * 8UL);
63 * If you need less than 50 threads would mean we're dealing with systems
64 * smaller than 3200 pages. This assuems you are capable of having ~13M memory,
65 * and this would only be an be an upper limit, after which the OOM killer
66 * would take effect. Systems like these are very unlikely if modules are
69 #define MAX_KMOD_CONCURRENT 50
70 static atomic_t kmod_concurrent_max
= ATOMIC_INIT(MAX_KMOD_CONCURRENT
);
73 modprobe_path is set via /proc/sys.
75 char modprobe_path
[KMOD_PATH_LEN
] = "/sbin/modprobe";
77 static void free_modprobe_argv(struct subprocess_info
*info
)
79 kfree(info
->argv
[3]); /* check call_modprobe() */
83 static int call_modprobe(char *module_name
, int wait
)
85 struct subprocess_info
*info
;
86 static char *envp
[] = {
89 "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
93 char **argv
= kmalloc(sizeof(char *[5]), GFP_KERNEL
);
97 module_name
= kstrdup(module_name
, GFP_KERNEL
);
101 argv
[0] = modprobe_path
;
104 argv
[3] = module_name
; /* check free_modprobe_argv() */
107 info
= call_usermodehelper_setup(modprobe_path
, argv
, envp
, GFP_KERNEL
,
108 NULL
, free_modprobe_argv
, NULL
);
110 goto free_module_name
;
112 return call_usermodehelper_exec(info
, wait
| UMH_KILLABLE
);
123 * __request_module - try to load a kernel module
124 * @wait: wait (or not) for the operation to complete
125 * @fmt: printf style format string for the name of the module
126 * @...: arguments as specified in the format string
128 * Load a module using the user mode module loader. The function returns
129 * zero on success or a negative errno code or positive exit code from
130 * "modprobe" on failure. Note that a successful module load does not mean
131 * the module did not then unload and exit on an error of its own. Callers
132 * must check that the service they requested is now available not blindly
135 * If module auto-loading support is disabled then this function
136 * becomes a no-operation.
138 int __request_module(bool wait
, const char *fmt
, ...)
141 char module_name
[MODULE_NAME_LEN
];
143 static int kmod_loop_msg
;
146 * We don't allow synchronous module loading from async. Module
147 * init may invoke async_synchronize_full() which will end up
148 * waiting for this task which already is waiting for the module
149 * loading to complete, leading to a deadlock.
151 WARN_ON_ONCE(wait
&& current_is_async());
153 if (!modprobe_path
[0])
157 ret
= vsnprintf(module_name
, MODULE_NAME_LEN
, fmt
, args
);
159 if (ret
>= MODULE_NAME_LEN
)
160 return -ENAMETOOLONG
;
162 ret
= security_kernel_module_request(module_name
);
166 if (atomic_dec_if_positive(&kmod_concurrent_max
) < 0) {
167 /* We may be blaming an innocent here, but unlikely */
168 if (kmod_loop_msg
< 5) {
170 "request_module: runaway loop modprobe %s\n",
177 trace_module_request(module_name
, wait
, _RET_IP_
);
179 ret
= call_modprobe(module_name
, wait
? UMH_WAIT_PROC
: UMH_WAIT_EXEC
);
181 atomic_inc(&kmod_concurrent_max
);
185 EXPORT_SYMBOL(__request_module
);
187 #endif /* CONFIG_MODULES */
189 static void call_usermodehelper_freeinfo(struct subprocess_info
*info
)
192 (*info
->cleanup
)(info
);
196 static void umh_complete(struct subprocess_info
*sub_info
)
198 struct completion
*comp
= xchg(&sub_info
->complete
, NULL
);
200 * See call_usermodehelper_exec(). If xchg() returns NULL
201 * we own sub_info, the UMH_KILLABLE caller has gone away
202 * or the caller used UMH_NO_WAIT.
207 call_usermodehelper_freeinfo(sub_info
);
211 * This is the task which runs the usermode application
213 static int call_usermodehelper_exec_async(void *data
)
215 struct subprocess_info
*sub_info
= data
;
219 spin_lock_irq(¤t
->sighand
->siglock
);
220 flush_signal_handlers(current
, 1);
221 spin_unlock_irq(¤t
->sighand
->siglock
);
224 * Our parent (unbound workqueue) runs with elevated scheduling
225 * priority. Avoid propagating that into the userspace child.
227 set_user_nice(current
, 0);
230 new = prepare_kernel_cred(current
);
234 spin_lock(&umh_sysctl_lock
);
235 new->cap_bset
= cap_intersect(usermodehelper_bset
, new->cap_bset
);
236 new->cap_inheritable
= cap_intersect(usermodehelper_inheritable
,
237 new->cap_inheritable
);
238 spin_unlock(&umh_sysctl_lock
);
240 if (sub_info
->init
) {
241 retval
= sub_info
->init(sub_info
, new);
250 retval
= do_execve(getname_kernel(sub_info
->path
),
251 (const char __user
*const __user
*)sub_info
->argv
,
252 (const char __user
*const __user
*)sub_info
->envp
);
254 sub_info
->retval
= retval
;
256 * call_usermodehelper_exec_sync() will call umh_complete
259 if (!(sub_info
->wait
& UMH_WAIT_PROC
))
260 umh_complete(sub_info
);
266 /* Handles UMH_WAIT_PROC. */
267 static void call_usermodehelper_exec_sync(struct subprocess_info
*sub_info
)
271 /* If SIGCLD is ignored sys_wait4 won't populate the status. */
272 kernel_sigaction(SIGCHLD
, SIG_DFL
);
273 pid
= kernel_thread(call_usermodehelper_exec_async
, sub_info
, SIGCHLD
);
275 sub_info
->retval
= pid
;
279 * Normally it is bogus to call wait4() from in-kernel because
280 * wait4() wants to write the exit code to a userspace address.
281 * But call_usermodehelper_exec_sync() always runs as kernel
282 * thread (workqueue) and put_user() to a kernel address works
283 * OK for kernel threads, due to their having an mm_segment_t
284 * which spans the entire address space.
286 * Thus the __user pointer cast is valid here.
288 sys_wait4(pid
, (int __user
*)&ret
, 0, NULL
);
291 * If ret is 0, either call_usermodehelper_exec_async failed and
292 * the real error code is already in sub_info->retval or
293 * sub_info->retval is 0 anyway, so don't mess with it then.
296 sub_info
->retval
= ret
;
299 /* Restore default kernel sig handler */
300 kernel_sigaction(SIGCHLD
, SIG_IGN
);
302 umh_complete(sub_info
);
306 * We need to create the usermodehelper kernel thread from a task that is affine
307 * to an optimized set of CPUs (or nohz housekeeping ones) such that they
308 * inherit a widest affinity irrespective of call_usermodehelper() callers with
309 * possibly reduced affinity (eg: per-cpu workqueues). We don't want
310 * usermodehelper targets to contend a busy CPU.
312 * Unbound workqueues provide such wide affinity and allow to block on
313 * UMH_WAIT_PROC requests without blocking pending request (up to some limit).
315 * Besides, workqueues provide the privilege level that caller might not have
316 * to perform the usermodehelper request.
319 static void call_usermodehelper_exec_work(struct work_struct
*work
)
321 struct subprocess_info
*sub_info
=
322 container_of(work
, struct subprocess_info
, work
);
324 if (sub_info
->wait
& UMH_WAIT_PROC
) {
325 call_usermodehelper_exec_sync(sub_info
);
329 * Use CLONE_PARENT to reparent it to kthreadd; we do not
330 * want to pollute current->children, and we need a parent
331 * that always ignores SIGCHLD to ensure auto-reaping.
333 pid
= kernel_thread(call_usermodehelper_exec_async
, sub_info
,
334 CLONE_PARENT
| SIGCHLD
);
336 sub_info
->retval
= pid
;
337 umh_complete(sub_info
);
343 * If set, call_usermodehelper_exec() will exit immediately returning -EBUSY
344 * (used for preventing user land processes from being created after the user
345 * land has been frozen during a system-wide hibernation or suspend operation).
346 * Should always be manipulated under umhelper_sem acquired for write.
348 static enum umh_disable_depth usermodehelper_disabled
= UMH_DISABLED
;
350 /* Number of helpers running */
351 static atomic_t running_helpers
= ATOMIC_INIT(0);
354 * Wait queue head used by usermodehelper_disable() to wait for all running
357 static DECLARE_WAIT_QUEUE_HEAD(running_helpers_waitq
);
360 * Used by usermodehelper_read_lock_wait() to wait for usermodehelper_disabled
363 static DECLARE_WAIT_QUEUE_HEAD(usermodehelper_disabled_waitq
);
366 * Time to wait for running_helpers to become zero before the setting of
367 * usermodehelper_disabled in usermodehelper_disable() fails
369 #define RUNNING_HELPERS_TIMEOUT (5 * HZ)
371 int usermodehelper_read_trylock(void)
376 down_read(&umhelper_sem
);
378 prepare_to_wait(&usermodehelper_disabled_waitq
, &wait
,
380 if (!usermodehelper_disabled
)
383 if (usermodehelper_disabled
== UMH_DISABLED
)
386 up_read(&umhelper_sem
);
394 down_read(&umhelper_sem
);
396 finish_wait(&usermodehelper_disabled_waitq
, &wait
);
399 EXPORT_SYMBOL_GPL(usermodehelper_read_trylock
);
401 long usermodehelper_read_lock_wait(long timeout
)
408 down_read(&umhelper_sem
);
410 prepare_to_wait(&usermodehelper_disabled_waitq
, &wait
,
411 TASK_UNINTERRUPTIBLE
);
412 if (!usermodehelper_disabled
)
415 up_read(&umhelper_sem
);
417 timeout
= schedule_timeout(timeout
);
421 down_read(&umhelper_sem
);
423 finish_wait(&usermodehelper_disabled_waitq
, &wait
);
426 EXPORT_SYMBOL_GPL(usermodehelper_read_lock_wait
);
428 void usermodehelper_read_unlock(void)
430 up_read(&umhelper_sem
);
432 EXPORT_SYMBOL_GPL(usermodehelper_read_unlock
);
435 * __usermodehelper_set_disable_depth - Modify usermodehelper_disabled.
436 * @depth: New value to assign to usermodehelper_disabled.
438 * Change the value of usermodehelper_disabled (under umhelper_sem locked for
439 * writing) and wakeup tasks waiting for it to change.
441 void __usermodehelper_set_disable_depth(enum umh_disable_depth depth
)
443 down_write(&umhelper_sem
);
444 usermodehelper_disabled
= depth
;
445 wake_up(&usermodehelper_disabled_waitq
);
446 up_write(&umhelper_sem
);
450 * __usermodehelper_disable - Prevent new helpers from being started.
451 * @depth: New value to assign to usermodehelper_disabled.
453 * Set usermodehelper_disabled to @depth and wait for running helpers to exit.
455 int __usermodehelper_disable(enum umh_disable_depth depth
)
462 down_write(&umhelper_sem
);
463 usermodehelper_disabled
= depth
;
464 up_write(&umhelper_sem
);
467 * From now on call_usermodehelper_exec() won't start any new
468 * helpers, so it is sufficient if running_helpers turns out to
469 * be zero at one point (it may be increased later, but that
472 retval
= wait_event_timeout(running_helpers_waitq
,
473 atomic_read(&running_helpers
) == 0,
474 RUNNING_HELPERS_TIMEOUT
);
478 __usermodehelper_set_disable_depth(UMH_ENABLED
);
482 static void helper_lock(void)
484 atomic_inc(&running_helpers
);
485 smp_mb__after_atomic();
488 static void helper_unlock(void)
490 if (atomic_dec_and_test(&running_helpers
))
491 wake_up(&running_helpers_waitq
);
495 * call_usermodehelper_setup - prepare to call a usermode helper
496 * @path: path to usermode executable
497 * @argv: arg vector for process
498 * @envp: environment for process
499 * @gfp_mask: gfp mask for memory allocation
500 * @cleanup: a cleanup function
501 * @init: an init function
502 * @data: arbitrary context sensitive data
504 * Returns either %NULL on allocation failure, or a subprocess_info
505 * structure. This should be passed to call_usermodehelper_exec to
506 * exec the process and free the structure.
508 * The init function is used to customize the helper process prior to
509 * exec. A non-zero return code causes the process to error out, exit,
510 * and return the failure to the calling process
512 * The cleanup function is just before ethe subprocess_info is about to
513 * be freed. This can be used for freeing the argv and envp. The
514 * Function must be runnable in either a process context or the
515 * context in which call_usermodehelper_exec is called.
517 struct subprocess_info
*call_usermodehelper_setup(const char *path
, char **argv
,
518 char **envp
, gfp_t gfp_mask
,
519 int (*init
)(struct subprocess_info
*info
, struct cred
*new),
520 void (*cleanup
)(struct subprocess_info
*info
),
523 struct subprocess_info
*sub_info
;
524 sub_info
= kzalloc(sizeof(struct subprocess_info
), gfp_mask
);
528 INIT_WORK(&sub_info
->work
, call_usermodehelper_exec_work
);
530 #ifdef CONFIG_STATIC_USERMODEHELPER
531 sub_info
->path
= CONFIG_STATIC_USERMODEHELPER_PATH
;
533 sub_info
->path
= path
;
535 sub_info
->argv
= argv
;
536 sub_info
->envp
= envp
;
538 sub_info
->cleanup
= cleanup
;
539 sub_info
->init
= init
;
540 sub_info
->data
= data
;
544 EXPORT_SYMBOL(call_usermodehelper_setup
);
547 * call_usermodehelper_exec - start a usermode application
548 * @sub_info: information about the subprocessa
549 * @wait: wait for the application to finish and return status.
550 * when UMH_NO_WAIT don't wait at all, but you get no useful error back
551 * when the program couldn't be exec'ed. This makes it safe to call
552 * from interrupt context.
554 * Runs a user-space application. The application is started
555 * asynchronously if wait is not set, and runs as a child of system workqueues.
556 * (ie. it runs with full root capabilities and optimized affinity).
558 int call_usermodehelper_exec(struct subprocess_info
*sub_info
, int wait
)
560 DECLARE_COMPLETION_ONSTACK(done
);
563 if (!sub_info
->path
) {
564 call_usermodehelper_freeinfo(sub_info
);
568 if (usermodehelper_disabled
) {
574 * If there is no binary for us to call, then just return and get out of
575 * here. This allows us to set STATIC_USERMODEHELPER_PATH to "" and
576 * disable all call_usermodehelper() calls.
578 if (strlen(sub_info
->path
) == 0)
582 * Set the completion pointer only if there is a waiter.
583 * This makes it possible to use umh_complete to free
584 * the data structure in case of UMH_NO_WAIT.
586 sub_info
->complete
= (wait
== UMH_NO_WAIT
) ? NULL
: &done
;
587 sub_info
->wait
= wait
;
589 queue_work(system_unbound_wq
, &sub_info
->work
);
590 if (wait
== UMH_NO_WAIT
) /* task has freed sub_info */
593 if (wait
& UMH_KILLABLE
) {
594 retval
= wait_for_completion_killable(&done
);
598 /* umh_complete() will see NULL and free sub_info */
599 if (xchg(&sub_info
->complete
, NULL
))
601 /* fallthrough, umh_complete() was already called */
604 wait_for_completion(&done
);
606 retval
= sub_info
->retval
;
608 call_usermodehelper_freeinfo(sub_info
);
613 EXPORT_SYMBOL(call_usermodehelper_exec
);
616 * call_usermodehelper() - prepare and start a usermode application
617 * @path: path to usermode executable
618 * @argv: arg vector for process
619 * @envp: environment for process
620 * @wait: wait for the application to finish and return status.
621 * when UMH_NO_WAIT don't wait at all, but you get no useful error back
622 * when the program couldn't be exec'ed. This makes it safe to call
623 * from interrupt context.
625 * This function is the equivalent to use call_usermodehelper_setup() and
626 * call_usermodehelper_exec().
628 int call_usermodehelper(const char *path
, char **argv
, char **envp
, int wait
)
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 return call_usermodehelper_exec(info
, wait
);
640 EXPORT_SYMBOL(call_usermodehelper
);
642 static int proc_cap_handler(struct ctl_table
*table
, int write
,
643 void __user
*buffer
, size_t *lenp
, loff_t
*ppos
)
646 unsigned long cap_array
[_KERNEL_CAPABILITY_U32S
];
647 kernel_cap_t new_cap
;
650 if (write
&& (!capable(CAP_SETPCAP
) ||
651 !capable(CAP_SYS_MODULE
)))
655 * convert from the global kernel_cap_t to the ulong array to print to
656 * userspace if this is a read.
658 spin_lock(&umh_sysctl_lock
);
659 for (i
= 0; i
< _KERNEL_CAPABILITY_U32S
; i
++) {
660 if (table
->data
== CAP_BSET
)
661 cap_array
[i
] = usermodehelper_bset
.cap
[i
];
662 else if (table
->data
== CAP_PI
)
663 cap_array
[i
] = usermodehelper_inheritable
.cap
[i
];
667 spin_unlock(&umh_sysctl_lock
);
673 * actually read or write and array of ulongs from userspace. Remember
674 * these are least significant 32 bits first
676 err
= proc_doulongvec_minmax(&t
, write
, buffer
, lenp
, ppos
);
681 * convert from the sysctl array of ulongs to the kernel_cap_t
682 * internal representation
684 for (i
= 0; i
< _KERNEL_CAPABILITY_U32S
; i
++)
685 new_cap
.cap
[i
] = cap_array
[i
];
688 * Drop everything not in the new_cap (but don't add things)
690 spin_lock(&umh_sysctl_lock
);
692 if (table
->data
== CAP_BSET
)
693 usermodehelper_bset
= cap_intersect(usermodehelper_bset
, new_cap
);
694 if (table
->data
== CAP_PI
)
695 usermodehelper_inheritable
= cap_intersect(usermodehelper_inheritable
, new_cap
);
697 spin_unlock(&umh_sysctl_lock
);
702 struct ctl_table usermodehelper_table
[] = {
706 .maxlen
= _KERNEL_CAPABILITY_U32S
* sizeof(unsigned long),
708 .proc_handler
= proc_cap_handler
,
711 .procname
= "inheritable",
713 .maxlen
= _KERNEL_CAPABILITY_U32S
* sizeof(unsigned long),
715 .proc_handler
= proc_cap_handler
,