Root/kernel/kmod.c

1/*
2    kmod, the new module loader (replaces kerneld)
3    Kirk Petersen
4
5    Reorganized not to be a daemon by Adam Richter, with guidance
6    from Greg Zornetzer.
7
8    Modified to avoid chroot and file sharing problems.
9    Mikael Pettersson
10
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
14
15    Unblock all signals when we exec a usermode process.
16    Shuu Yamaguchi <shuu@wondernetworkresources.com> December 2000
17
18    call_usermodehelper wait flag, and remove exec_usermodehelper.
19    Rusty Russell <rusty@rustcorp.com.au> Jan 2003
20*/
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>
43
44#include <trace/events/module.h>
45
46extern int max_threads;
47
48static struct workqueue_struct *khelper_wq;
49
50/*
51 * kmod_thread_locker is used for deadlock avoidance. There is no explicit
52 * locking to protect this global - it is private to the singleton khelper
53 * thread and should only ever be modified by that thread.
54 */
55static const struct task_struct *kmod_thread_locker;
56
57#define CAP_BSET (void *)1
58#define CAP_PI (void *)2
59
60static kernel_cap_t usermodehelper_bset = CAP_FULL_SET;
61static kernel_cap_t usermodehelper_inheritable = CAP_FULL_SET;
62static DEFINE_SPINLOCK(umh_sysctl_lock);
63static DECLARE_RWSEM(umhelper_sem);
64
65#ifdef CONFIG_MODULES
66
67/*
68    modprobe_path is set via /proc/sys.
69*/
70char modprobe_path[KMOD_PATH_LEN] = "/sbin/modprobe";
71
72static void free_modprobe_argv(struct subprocess_info *info)
73{
74    kfree(info->argv[3]); /* check call_modprobe() */
75    kfree(info->argv);
76}
77
78static int call_modprobe(char *module_name, int wait)
79{
80    struct subprocess_info *info;
81    static char *envp[] = {
82        "HOME=/",
83        "TERM=linux",
84        "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
85        NULL
86    };
87
88    char **argv = kmalloc(sizeof(char *[5]), GFP_KERNEL);
89    if (!argv)
90        goto out;
91
92    module_name = kstrdup(module_name, GFP_KERNEL);
93    if (!module_name)
94        goto free_argv;
95
96    argv[0] = modprobe_path;
97    argv[1] = "-q";
98    argv[2] = "--";
99    argv[3] = module_name; /* check free_modprobe_argv() */
100    argv[4] = NULL;
101
102    info = call_usermodehelper_setup(modprobe_path, argv, envp, GFP_KERNEL,
103                     NULL, free_modprobe_argv, NULL);
104    if (!info)
105        goto free_module_name;
106
107    return call_usermodehelper_exec(info, wait | UMH_KILLABLE);
108
109free_module_name:
110    kfree(module_name);
111free_argv:
112    kfree(argv);
113out:
114    return -ENOMEM;
115}
116
117/**
118 * __request_module - try to load a kernel module
119 * @wait: wait (or not) for the operation to complete
120 * @fmt: printf style format string for the name of the module
121 * @...: arguments as specified in the format string
122 *
123 * Load a module using the user mode module loader. The function returns
124 * zero on success or a negative errno code on failure. Note that a
125 * successful module load does not mean the module did not then unload
126 * and exit on an error of its own. Callers must check that the service
127 * they requested is now available not blindly invoke it.
128 *
129 * If module auto-loading support is disabled then this function
130 * becomes a no-operation.
131 */
132int __request_module(bool wait, const char *fmt, ...)
133{
134    va_list args;
135    char module_name[MODULE_NAME_LEN];
136    unsigned int max_modprobes;
137    int ret;
138    static atomic_t kmod_concurrent = ATOMIC_INIT(0);
139#define MAX_KMOD_CONCURRENT 50 /* Completely arbitrary value - KAO */
140    static int kmod_loop_msg;
141
142    /*
143     * We don't allow synchronous module loading from async. Module
144     * init may invoke async_synchronize_full() which will end up
145     * waiting for this task which already is waiting for the module
146     * loading to complete, leading to a deadlock.
147     */
148    WARN_ON_ONCE(wait && current_is_async());
149
150    if (!modprobe_path[0])
151        return 0;
152
153    va_start(args, fmt);
154    ret = vsnprintf(module_name, MODULE_NAME_LEN, fmt, args);
155    va_end(args);
156    if (ret >= MODULE_NAME_LEN)
157        return -ENAMETOOLONG;
158
159    ret = security_kernel_module_request(module_name);
160    if (ret)
161        return ret;
162
163    /* If modprobe needs a service that is in a module, we get a recursive
164     * loop. Limit the number of running kmod threads to max_threads/2 or
165     * MAX_KMOD_CONCURRENT, whichever is the smaller. A cleaner method
166     * would be to run the parents of this process, counting how many times
167     * kmod was invoked. That would mean accessing the internals of the
168     * process tables to get the command line, proc_pid_cmdline is static
169     * and it is not worth changing the proc code just to handle this case.
170     * KAO.
171     *
172     * "trace the ppid" is simple, but will fail if someone's
173     * parent exits. I think this is as good as it gets. --RR
174     */
175    max_modprobes = min(max_threads/2, MAX_KMOD_CONCURRENT);
176    atomic_inc(&kmod_concurrent);
177    if (atomic_read(&kmod_concurrent) > max_modprobes) {
178        /* We may be blaming an innocent here, but unlikely */
179        if (kmod_loop_msg < 5) {
180            printk(KERN_ERR
181                   "request_module: runaway loop modprobe %s\n",
182                   module_name);
183            kmod_loop_msg++;
184        }
185        atomic_dec(&kmod_concurrent);
186        return -ENOMEM;
187    }
188
189    trace_module_request(module_name, wait, _RET_IP_);
190
191    ret = call_modprobe(module_name, wait ? UMH_WAIT_PROC : UMH_WAIT_EXEC);
192
193    atomic_dec(&kmod_concurrent);
194    return ret;
195}
196EXPORT_SYMBOL(__request_module);
197#endif /* CONFIG_MODULES */
198
199/*
200 * This is the task which runs the usermode application
201 */
202static int ____call_usermodehelper(void *data)
203{
204    struct subprocess_info *sub_info = data;
205    struct cred *new;
206    int retval;
207
208    spin_lock_irq(&current->sighand->siglock);
209    flush_signal_handlers(current, 1);
210    spin_unlock_irq(&current->sighand->siglock);
211
212    /* We can run anywhere, unlike our parent keventd(). */
213    set_cpus_allowed_ptr(current, cpu_all_mask);
214
215    /*
216     * Our parent is keventd, which runs with elevated scheduling priority.
217     * Avoid propagating that into the userspace child.
218     */
219    set_user_nice(current, 0);
220
221    retval = -ENOMEM;
222    new = prepare_kernel_cred(current);
223    if (!new)
224        goto fail;
225
226    spin_lock(&umh_sysctl_lock);
227    new->cap_bset = cap_intersect(usermodehelper_bset, new->cap_bset);
228    new->cap_inheritable = cap_intersect(usermodehelper_inheritable,
229                         new->cap_inheritable);
230    spin_unlock(&umh_sysctl_lock);
231
232    if (sub_info->init) {
233        retval = sub_info->init(sub_info, new);
234        if (retval) {
235            abort_creds(new);
236            goto fail;
237        }
238    }
239
240    commit_creds(new);
241
242    retval = do_execve(getname_kernel(sub_info->path),
243               (const char __user *const __user *)sub_info->argv,
244               (const char __user *const __user *)sub_info->envp);
245    if (!retval)
246        return 0;
247
248    /* Exec failed? */
249fail:
250    sub_info->retval = retval;
251    do_exit(0);
252}
253
254static int call_helper(void *data)
255{
256    /* Worker thread started blocking khelper thread. */
257    kmod_thread_locker = current;
258    return ____call_usermodehelper(data);
259}
260
261static void call_usermodehelper_freeinfo(struct subprocess_info *info)
262{
263    if (info->cleanup)
264        (*info->cleanup)(info);
265    kfree(info);
266}
267
268static void umh_complete(struct subprocess_info *sub_info)
269{
270    struct completion *comp = xchg(&sub_info->complete, NULL);
271    /*
272     * See call_usermodehelper_exec(). If xchg() returns NULL
273     * we own sub_info, the UMH_KILLABLE caller has gone away.
274     */
275    if (comp)
276        complete(comp);
277    else
278        call_usermodehelper_freeinfo(sub_info);
279}
280
281/* Keventd can't block, but this (a child) can. */
282static int wait_for_helper(void *data)
283{
284    struct subprocess_info *sub_info = data;
285    pid_t pid;
286
287    /* If SIGCLD is ignored sys_wait4 won't populate the status. */
288    spin_lock_irq(&current->sighand->siglock);
289    current->sighand->action[SIGCHLD-1].sa.sa_handler = SIG_DFL;
290    spin_unlock_irq(&current->sighand->siglock);
291
292    pid = kernel_thread(____call_usermodehelper, sub_info, SIGCHLD);
293    if (pid < 0) {
294        sub_info->retval = pid;
295    } else {
296        int ret = -ECHILD;
297        /*
298         * Normally it is bogus to call wait4() from in-kernel because
299         * wait4() wants to write the exit code to a userspace address.
300         * But wait_for_helper() always runs as keventd, and put_user()
301         * to a kernel address works OK for kernel threads, due to their
302         * having an mm_segment_t which spans the entire address space.
303         *
304         * Thus the __user pointer cast is valid here.
305         */
306        sys_wait4(pid, (int __user *)&ret, 0, NULL);
307
308        /*
309         * If ret is 0, either ____call_usermodehelper failed and the
310         * real error code is already in sub_info->retval or
311         * sub_info->retval is 0 anyway, so don't mess with it then.
312         */
313        if (ret)
314            sub_info->retval = ret;
315    }
316
317    umh_complete(sub_info);
318    do_exit(0);
319}
320
321/* This is run by khelper thread */
322static void __call_usermodehelper(struct work_struct *work)
323{
324    struct subprocess_info *sub_info =
325        container_of(work, struct subprocess_info, work);
326    int wait = sub_info->wait & ~UMH_KILLABLE;
327    pid_t pid;
328
329    /* CLONE_VFORK: wait until the usermode helper has execve'd
330     * successfully We need the data structures to stay around
331     * until that is done. */
332    if (wait == UMH_WAIT_PROC)
333        pid = kernel_thread(wait_for_helper, sub_info,
334                    CLONE_FS | CLONE_FILES | SIGCHLD);
335    else {
336        pid = kernel_thread(call_helper, sub_info,
337                    CLONE_VFORK | SIGCHLD);
338        /* Worker thread stopped blocking khelper thread. */
339        kmod_thread_locker = NULL;
340    }
341
342    switch (wait) {
343    case UMH_NO_WAIT:
344        call_usermodehelper_freeinfo(sub_info);
345        break;
346
347    case UMH_WAIT_PROC:
348        if (pid > 0)
349            break;
350        /* FALLTHROUGH */
351    case UMH_WAIT_EXEC:
352        if (pid < 0)
353            sub_info->retval = pid;
354        umh_complete(sub_info);
355    }
356}
357
358/*
359 * If set, call_usermodehelper_exec() will exit immediately returning -EBUSY
360 * (used for preventing user land processes from being created after the user
361 * land has been frozen during a system-wide hibernation or suspend operation).
362 * Should always be manipulated under umhelper_sem acquired for write.
363 */
364static enum umh_disable_depth usermodehelper_disabled = UMH_DISABLED;
365
366/* Number of helpers running */
367static atomic_t running_helpers = ATOMIC_INIT(0);
368
369/*
370 * Wait queue head used by usermodehelper_disable() to wait for all running
371 * helpers to finish.
372 */
373static DECLARE_WAIT_QUEUE_HEAD(running_helpers_waitq);
374
375/*
376 * Used by usermodehelper_read_lock_wait() to wait for usermodehelper_disabled
377 * to become 'false'.
378 */
379static DECLARE_WAIT_QUEUE_HEAD(usermodehelper_disabled_waitq);
380
381/*
382 * Time to wait for running_helpers to become zero before the setting of
383 * usermodehelper_disabled in usermodehelper_disable() fails
384 */
385#define RUNNING_HELPERS_TIMEOUT (5 * HZ)
386
387int usermodehelper_read_trylock(void)
388{
389    DEFINE_WAIT(wait);
390    int ret = 0;
391
392    down_read(&umhelper_sem);
393    for (;;) {
394        prepare_to_wait(&usermodehelper_disabled_waitq, &wait,
395                TASK_INTERRUPTIBLE);
396        if (!usermodehelper_disabled)
397            break;
398
399        if (usermodehelper_disabled == UMH_DISABLED)
400            ret = -EAGAIN;
401
402        up_read(&umhelper_sem);
403
404        if (ret)
405            break;
406
407        schedule();
408        try_to_freeze();
409
410        down_read(&umhelper_sem);
411    }
412    finish_wait(&usermodehelper_disabled_waitq, &wait);
413    return ret;
414}
415EXPORT_SYMBOL_GPL(usermodehelper_read_trylock);
416
417long usermodehelper_read_lock_wait(long timeout)
418{
419    DEFINE_WAIT(wait);
420
421    if (timeout < 0)
422        return -EINVAL;
423
424    down_read(&umhelper_sem);
425    for (;;) {
426        prepare_to_wait(&usermodehelper_disabled_waitq, &wait,
427                TASK_UNINTERRUPTIBLE);
428        if (!usermodehelper_disabled)
429            break;
430
431        up_read(&umhelper_sem);
432
433        timeout = schedule_timeout(timeout);
434        if (!timeout)
435            break;
436
437        down_read(&umhelper_sem);
438    }
439    finish_wait(&usermodehelper_disabled_waitq, &wait);
440    return timeout;
441}
442EXPORT_SYMBOL_GPL(usermodehelper_read_lock_wait);
443
444void usermodehelper_read_unlock(void)
445{
446    up_read(&umhelper_sem);
447}
448EXPORT_SYMBOL_GPL(usermodehelper_read_unlock);
449
450/**
451 * __usermodehelper_set_disable_depth - Modify usermodehelper_disabled.
452 * @depth: New value to assign to usermodehelper_disabled.
453 *
454 * Change the value of usermodehelper_disabled (under umhelper_sem locked for
455 * writing) and wakeup tasks waiting for it to change.
456 */
457void __usermodehelper_set_disable_depth(enum umh_disable_depth depth)
458{
459    down_write(&umhelper_sem);
460    usermodehelper_disabled = depth;
461    wake_up(&usermodehelper_disabled_waitq);
462    up_write(&umhelper_sem);
463}
464
465/**
466 * __usermodehelper_disable - Prevent new helpers from being started.
467 * @depth: New value to assign to usermodehelper_disabled.
468 *
469 * Set usermodehelper_disabled to @depth and wait for running helpers to exit.
470 */
471int __usermodehelper_disable(enum umh_disable_depth depth)
472{
473    long retval;
474
475    if (!depth)
476        return -EINVAL;
477
478    down_write(&umhelper_sem);
479    usermodehelper_disabled = depth;
480    up_write(&umhelper_sem);
481
482    /*
483     * From now on call_usermodehelper_exec() won't start any new
484     * helpers, so it is sufficient if running_helpers turns out to
485     * be zero at one point (it may be increased later, but that
486     * doesn't matter).
487     */
488    retval = wait_event_timeout(running_helpers_waitq,
489                    atomic_read(&running_helpers) == 0,
490                    RUNNING_HELPERS_TIMEOUT);
491    if (retval)
492        return 0;
493
494    __usermodehelper_set_disable_depth(UMH_ENABLED);
495    return -EAGAIN;
496}
497
498static void helper_lock(void)
499{
500    atomic_inc(&running_helpers);
501    smp_mb__after_atomic_inc();
502}
503
504static void helper_unlock(void)
505{
506    if (atomic_dec_and_test(&running_helpers))
507        wake_up(&running_helpers_waitq);
508}
509
510/**
511 * call_usermodehelper_setup - prepare to call a usermode helper
512 * @path: path to usermode executable
513 * @argv: arg vector for process
514 * @envp: environment for process
515 * @gfp_mask: gfp mask for memory allocation
516 * @cleanup: a cleanup function
517 * @init: an init function
518 * @data: arbitrary context sensitive data
519 *
520 * Returns either %NULL on allocation failure, or a subprocess_info
521 * structure. This should be passed to call_usermodehelper_exec to
522 * exec the process and free the structure.
523 *
524 * The init function is used to customize the helper process prior to
525 * exec. A non-zero return code causes the process to error out, exit,
526 * and return the failure to the calling process
527 *
528 * The cleanup function is just before ethe subprocess_info is about to
529 * be freed. This can be used for freeing the argv and envp. The
530 * Function must be runnable in either a process context or the
531 * context in which call_usermodehelper_exec is called.
532 */
533struct subprocess_info *call_usermodehelper_setup(char *path, char **argv,
534        char **envp, gfp_t gfp_mask,
535        int (*init)(struct subprocess_info *info, struct cred *new),
536        void (*cleanup)(struct subprocess_info *info),
537        void *data)
538{
539    struct subprocess_info *sub_info;
540    sub_info = kzalloc(sizeof(struct subprocess_info), gfp_mask);
541    if (!sub_info)
542        goto out;
543
544    INIT_WORK(&sub_info->work, __call_usermodehelper);
545    sub_info->path = path;
546    sub_info->argv = argv;
547    sub_info->envp = envp;
548
549    sub_info->cleanup = cleanup;
550    sub_info->init = init;
551    sub_info->data = data;
552  out:
553    return sub_info;
554}
555EXPORT_SYMBOL(call_usermodehelper_setup);
556
557/**
558 * call_usermodehelper_exec - start a usermode application
559 * @sub_info: information about the subprocessa
560 * @wait: wait for the application to finish and return status.
561 * when UMH_NO_WAIT don't wait at all, but you get no useful error back
562 * when the program couldn't be exec'ed. This makes it safe to call
563 * from interrupt context.
564 *
565 * Runs a user-space application. The application is started
566 * asynchronously if wait is not set, and runs as a child of keventd.
567 * (ie. it runs with full root capabilities).
568 */
569int call_usermodehelper_exec(struct subprocess_info *sub_info, int wait)
570{
571    DECLARE_COMPLETION_ONSTACK(done);
572    int retval = 0;
573
574    if (!sub_info->path) {
575        call_usermodehelper_freeinfo(sub_info);
576        return -EINVAL;
577    }
578    helper_lock();
579    if (!khelper_wq || usermodehelper_disabled) {
580        retval = -EBUSY;
581        goto out;
582    }
583    /*
584     * Worker thread must not wait for khelper thread at below
585     * wait_for_completion() if the thread was created with CLONE_VFORK
586     * flag, for khelper thread is already waiting for the thread at
587     * wait_for_completion() in do_fork().
588     */
589    if (wait != UMH_NO_WAIT && current == kmod_thread_locker) {
590        retval = -EBUSY;
591        goto out;
592    }
593
594    sub_info->complete = &done;
595    sub_info->wait = wait;
596
597    queue_work(khelper_wq, &sub_info->work);
598    if (wait == UMH_NO_WAIT) /* task has freed sub_info */
599        goto unlock;
600
601    if (wait & UMH_KILLABLE) {
602        retval = wait_for_completion_killable(&done);
603        if (!retval)
604            goto wait_done;
605
606        /* umh_complete() will see NULL and free sub_info */
607        if (xchg(&sub_info->complete, NULL))
608            goto unlock;
609        /* fallthrough, umh_complete() was already called */
610    }
611
612    wait_for_completion(&done);
613wait_done:
614    retval = sub_info->retval;
615out:
616    call_usermodehelper_freeinfo(sub_info);
617unlock:
618    helper_unlock();
619    return retval;
620}
621EXPORT_SYMBOL(call_usermodehelper_exec);
622
623/**
624 * call_usermodehelper() - prepare and start a usermode application
625 * @path: path to usermode executable
626 * @argv: arg vector for process
627 * @envp: environment for process
628 * @wait: wait for the application to finish and return status.
629 * when UMH_NO_WAIT don't wait at all, but you get no useful error back
630 * when the program couldn't be exec'ed. This makes it safe to call
631 * from interrupt context.
632 *
633 * This function is the equivalent to use call_usermodehelper_setup() and
634 * call_usermodehelper_exec().
635 */
636int call_usermodehelper(char *path, char **argv, char **envp, int wait)
637{
638    struct subprocess_info *info;
639    gfp_t gfp_mask = (wait == UMH_NO_WAIT) ? GFP_ATOMIC : GFP_KERNEL;
640
641    info = call_usermodehelper_setup(path, argv, envp, gfp_mask,
642                     NULL, NULL, NULL);
643    if (info == NULL)
644        return -ENOMEM;
645
646    return call_usermodehelper_exec(info, wait);
647}
648EXPORT_SYMBOL(call_usermodehelper);
649
650static int proc_cap_handler(struct ctl_table *table, int write,
651             void __user *buffer, size_t *lenp, loff_t *ppos)
652{
653    struct ctl_table t;
654    unsigned long cap_array[_KERNEL_CAPABILITY_U32S];
655    kernel_cap_t new_cap;
656    int err, i;
657
658    if (write && (!capable(CAP_SETPCAP) ||
659              !capable(CAP_SYS_MODULE)))
660        return -EPERM;
661
662    /*
663     * convert from the global kernel_cap_t to the ulong array to print to
664     * userspace if this is a read.
665     */
666    spin_lock(&umh_sysctl_lock);
667    for (i = 0; i < _KERNEL_CAPABILITY_U32S; i++) {
668        if (table->data == CAP_BSET)
669            cap_array[i] = usermodehelper_bset.cap[i];
670        else if (table->data == CAP_PI)
671            cap_array[i] = usermodehelper_inheritable.cap[i];
672        else
673            BUG();
674    }
675    spin_unlock(&umh_sysctl_lock);
676
677    t = *table;
678    t.data = &cap_array;
679
680    /*
681     * actually read or write and array of ulongs from userspace. Remember
682     * these are least significant 32 bits first
683     */
684    err = proc_doulongvec_minmax(&t, write, buffer, lenp, ppos);
685    if (err < 0)
686        return err;
687
688    /*
689     * convert from the sysctl array of ulongs to the kernel_cap_t
690     * internal representation
691     */
692    for (i = 0; i < _KERNEL_CAPABILITY_U32S; i++)
693        new_cap.cap[i] = cap_array[i];
694
695    /*
696     * Drop everything not in the new_cap (but don't add things)
697     */
698    spin_lock(&umh_sysctl_lock);
699    if (write) {
700        if (table->data == CAP_BSET)
701            usermodehelper_bset = cap_intersect(usermodehelper_bset, new_cap);
702        if (table->data == CAP_PI)
703            usermodehelper_inheritable = cap_intersect(usermodehelper_inheritable, new_cap);
704    }
705    spin_unlock(&umh_sysctl_lock);
706
707    return 0;
708}
709
710struct ctl_table usermodehelper_table[] = {
711    {
712        .procname = "bset",
713        .data = CAP_BSET,
714        .maxlen = _KERNEL_CAPABILITY_U32S * sizeof(unsigned long),
715        .mode = 0600,
716        .proc_handler = proc_cap_handler,
717    },
718    {
719        .procname = "inheritable",
720        .data = CAP_PI,
721        .maxlen = _KERNEL_CAPABILITY_U32S * sizeof(unsigned long),
722        .mode = 0600,
723        .proc_handler = proc_cap_handler,
724    },
725    { }
726};
727
728void __init usermodehelper_init(void)
729{
730    khelper_wq = create_singlethread_workqueue("khelper");
731    BUG_ON(!khelper_wq);
732}
733

Archive Download this file



interactive