linux/kernel/kmod.c
<<
>>
Prefs
   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 <asm/uaccess.h>
  41
  42#include <trace/events/module.h>
  43
  44extern int max_threads;
  45
  46static struct workqueue_struct *khelper_wq;
  47
  48#define CAP_BSET        (void *)1
  49#define CAP_PI          (void *)2
  50
  51static kernel_cap_t usermodehelper_bset = CAP_FULL_SET;
  52static kernel_cap_t usermodehelper_inheritable = CAP_FULL_SET;
  53static DEFINE_SPINLOCK(umh_sysctl_lock);
  54static DECLARE_RWSEM(umhelper_sem);
  55
  56#ifdef CONFIG_MODULES
  57
  58/*
  59        modprobe_path is set via /proc/sys.
  60*/
  61char modprobe_path[KMOD_PATH_LEN] = "/sbin/modprobe";
  62
  63static void free_modprobe_argv(struct subprocess_info *info)
  64{
  65        kfree(info->argv[3]); /* check call_modprobe() */
  66        kfree(info->argv);
  67}
  68
  69static int call_modprobe(char *module_name, int wait)
  70{
  71        static char *envp[] = {
  72                "HOME=/",
  73                "TERM=linux",
  74                "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
  75                NULL
  76        };
  77
  78        char **argv = kmalloc(sizeof(char *[5]), GFP_KERNEL);
  79        if (!argv)
  80                goto out;
  81
  82        module_name = kstrdup(module_name, GFP_KERNEL);
  83        if (!module_name)
  84                goto free_argv;
  85
  86        argv[0] = modprobe_path;
  87        argv[1] = "-q";
  88        argv[2] = "--";
  89        argv[3] = module_name;  /* check free_modprobe_argv() */
  90        argv[4] = NULL;
  91
  92        return call_usermodehelper_fns(modprobe_path, argv, envp,
  93                wait | UMH_KILLABLE, NULL, free_modprobe_argv, NULL);
  94free_argv:
  95        kfree(argv);
  96out:
  97        return -ENOMEM;
  98}
  99
 100/**
 101 * __request_module - try to load a kernel module
 102 * @wait: wait (or not) for the operation to complete
 103 * @fmt: printf style format string for the name of the module
 104 * @...: arguments as specified in the format string
 105 *
 106 * Load a module using the user mode module loader. The function returns
 107 * zero on success or a negative errno code on failure. Note that a
 108 * successful module load does not mean the module did not then unload
 109 * and exit on an error of its own. Callers must check that the service
 110 * they requested is now available not blindly invoke it.
 111 *
 112 * If module auto-loading support is disabled then this function
 113 * becomes a no-operation.
 114 */
 115int __request_module(bool wait, const char *fmt, ...)
 116{
 117        va_list args;
 118        char module_name[MODULE_NAME_LEN];
 119        unsigned int max_modprobes;
 120        int ret;
 121        static atomic_t kmod_concurrent = ATOMIC_INIT(0);
 122#define MAX_KMOD_CONCURRENT 50  /* Completely arbitrary value - KAO */
 123        static int kmod_loop_msg;
 124
 125        va_start(args, fmt);
 126        ret = vsnprintf(module_name, MODULE_NAME_LEN, fmt, args);
 127        va_end(args);
 128        if (ret >= MODULE_NAME_LEN)
 129                return -ENAMETOOLONG;
 130
 131        ret = security_kernel_module_request(module_name);
 132        if (ret)
 133                return ret;
 134
 135        /* If modprobe needs a service that is in a module, we get a recursive
 136         * loop.  Limit the number of running kmod threads to max_threads/2 or
 137         * MAX_KMOD_CONCURRENT, whichever is the smaller.  A cleaner method
 138         * would be to run the parents of this process, counting how many times
 139         * kmod was invoked.  That would mean accessing the internals of the
 140         * process tables to get the command line, proc_pid_cmdline is static
 141         * and it is not worth changing the proc code just to handle this case. 
 142         * KAO.
 143         *
 144         * "trace the ppid" is simple, but will fail if someone's
 145         * parent exits.  I think this is as good as it gets. --RR
 146         */
 147        max_modprobes = min(max_threads/2, MAX_KMOD_CONCURRENT);
 148        atomic_inc(&kmod_concurrent);
 149        if (atomic_read(&kmod_concurrent) > max_modprobes) {
 150                /* We may be blaming an innocent here, but unlikely */
 151                if (kmod_loop_msg < 5) {
 152                        printk(KERN_ERR
 153                               "request_module: runaway loop modprobe %s\n",
 154                               module_name);
 155                        kmod_loop_msg++;
 156                }
 157                atomic_dec(&kmod_concurrent);
 158                return -ENOMEM;
 159        }
 160
 161        trace_module_request(module_name, wait, _RET_IP_);
 162
 163        ret = call_modprobe(module_name, wait ? UMH_WAIT_PROC : UMH_WAIT_EXEC);
 164
 165        atomic_dec(&kmod_concurrent);
 166        return ret;
 167}
 168EXPORT_SYMBOL(__request_module);
 169#endif /* CONFIG_MODULES */
 170
 171/*
 172 * This is the task which runs the usermode application
 173 */
 174static int ____call_usermodehelper(void *data)
 175{
 176        struct subprocess_info *sub_info = data;
 177        struct cred *new;
 178        int retval;
 179
 180        spin_lock_irq(&current->sighand->siglock);
 181        flush_signal_handlers(current, 1);
 182        spin_unlock_irq(&current->sighand->siglock);
 183
 184        /* We can run anywhere, unlike our parent keventd(). */
 185        set_cpus_allowed_ptr(current, cpu_all_mask);
 186
 187        /*
 188         * Our parent is keventd, which runs with elevated scheduling priority.
 189         * Avoid propagating that into the userspace child.
 190         */
 191        set_user_nice(current, 0);
 192
 193        retval = -ENOMEM;
 194        new = prepare_kernel_cred(current);
 195        if (!new)
 196                goto fail;
 197
 198        spin_lock(&umh_sysctl_lock);
 199        new->cap_bset = cap_intersect(usermodehelper_bset, new->cap_bset);
 200        new->cap_inheritable = cap_intersect(usermodehelper_inheritable,
 201                                             new->cap_inheritable);
 202        spin_unlock(&umh_sysctl_lock);
 203
 204        if (sub_info->init) {
 205                retval = sub_info->init(sub_info, new);
 206                if (retval) {
 207                        abort_creds(new);
 208                        goto fail;
 209                }
 210        }
 211
 212        commit_creds(new);
 213
 214        retval = kernel_execve(sub_info->path,
 215                               (const char *const *)sub_info->argv,
 216                               (const char *const *)sub_info->envp);
 217
 218        /* Exec failed? */
 219fail:
 220        sub_info->retval = retval;
 221        return 0;
 222}
 223
 224static void call_usermodehelper_freeinfo(struct subprocess_info *info)
 225{
 226        if (info->cleanup)
 227                (*info->cleanup)(info);
 228        kfree(info);
 229}
 230
 231static void umh_complete(struct subprocess_info *sub_info)
 232{
 233        struct completion *comp = xchg(&sub_info->complete, NULL);
 234        /*
 235         * See call_usermodehelper_exec(). If xchg() returns NULL
 236         * we own sub_info, the UMH_KILLABLE caller has gone away.
 237         */
 238        if (comp)
 239                complete(comp);
 240        else
 241                call_usermodehelper_freeinfo(sub_info);
 242}
 243
 244/* Keventd can't block, but this (a child) can. */
 245static int wait_for_helper(void *data)
 246{
 247        struct subprocess_info *sub_info = data;
 248        pid_t pid;
 249
 250        /* If SIGCLD is ignored sys_wait4 won't populate the status. */
 251        spin_lock_irq(&current->sighand->siglock);
 252        current->sighand->action[SIGCHLD-1].sa.sa_handler = SIG_DFL;
 253        spin_unlock_irq(&current->sighand->siglock);
 254
 255        pid = kernel_thread(____call_usermodehelper, sub_info, SIGCHLD);
 256        if (pid < 0) {
 257                sub_info->retval = pid;
 258        } else {
 259                int ret = -ECHILD;
 260                /*
 261                 * Normally it is bogus to call wait4() from in-kernel because
 262                 * wait4() wants to write the exit code to a userspace address.
 263                 * But wait_for_helper() always runs as keventd, and put_user()
 264                 * to a kernel address works OK for kernel threads, due to their
 265                 * having an mm_segment_t which spans the entire address space.
 266                 *
 267                 * Thus the __user pointer cast is valid here.
 268                 */
 269                sys_wait4(pid, (int __user *)&ret, 0, NULL);
 270
 271                /*
 272                 * If ret is 0, either ____call_usermodehelper failed and the
 273                 * real error code is already in sub_info->retval or
 274                 * sub_info->retval is 0 anyway, so don't mess with it then.
 275                 */
 276                if (ret)
 277                        sub_info->retval = ret;
 278        }
 279
 280        umh_complete(sub_info);
 281        return 0;
 282}
 283
 284/* This is run by khelper thread  */
 285static void __call_usermodehelper(struct work_struct *work)
 286{
 287        struct subprocess_info *sub_info =
 288                container_of(work, struct subprocess_info, work);
 289        int wait = sub_info->wait & ~UMH_KILLABLE;
 290        pid_t pid;
 291
 292        /* CLONE_VFORK: wait until the usermode helper has execve'd
 293         * successfully We need the data structures to stay around
 294         * until that is done.  */
 295        if (wait == UMH_WAIT_PROC)
 296                pid = kernel_thread(wait_for_helper, sub_info,
 297                                    CLONE_FS | CLONE_FILES | SIGCHLD);
 298        else
 299                pid = kernel_thread(____call_usermodehelper, sub_info,
 300                                    CLONE_VFORK | SIGCHLD);
 301
 302        switch (wait) {
 303        case UMH_NO_WAIT:
 304                call_usermodehelper_freeinfo(sub_info);
 305                break;
 306
 307        case UMH_WAIT_PROC:
 308                if (pid > 0)
 309                        break;
 310                /* FALLTHROUGH */
 311        case UMH_WAIT_EXEC:
 312                if (pid < 0)
 313                        sub_info->retval = pid;
 314                umh_complete(sub_info);
 315        }
 316}
 317
 318/*
 319 * If set, call_usermodehelper_exec() will exit immediately returning -EBUSY
 320 * (used for preventing user land processes from being created after the user
 321 * land has been frozen during a system-wide hibernation or suspend operation).
 322 * Should always be manipulated under umhelper_sem acquired for write.
 323 */
 324static enum umh_disable_depth usermodehelper_disabled = UMH_DISABLED;
 325
 326/* Number of helpers running */
 327static atomic_t running_helpers = ATOMIC_INIT(0);
 328
 329/*
 330 * Wait queue head used by usermodehelper_disable() to wait for all running
 331 * helpers to finish.
 332 */
 333static DECLARE_WAIT_QUEUE_HEAD(running_helpers_waitq);
 334
 335/*
 336 * Used by usermodehelper_read_lock_wait() to wait for usermodehelper_disabled
 337 * to become 'false'.
 338 */
 339static DECLARE_WAIT_QUEUE_HEAD(usermodehelper_disabled_waitq);
 340
 341/*
 342 * Time to wait for running_helpers to become zero before the setting of
 343 * usermodehelper_disabled in usermodehelper_disable() fails
 344 */
 345#define RUNNING_HELPERS_TIMEOUT (5 * HZ)
 346
 347int usermodehelper_read_trylock(void)
 348{
 349        DEFINE_WAIT(wait);
 350        int ret = 0;
 351
 352        down_read(&umhelper_sem);
 353        for (;;) {
 354                prepare_to_wait(&usermodehelper_disabled_waitq, &wait,
 355                                TASK_INTERRUPTIBLE);
 356                if (!usermodehelper_disabled)
 357                        break;
 358
 359                if (usermodehelper_disabled == UMH_DISABLED)
 360                        ret = -EAGAIN;
 361
 362                up_read(&umhelper_sem);
 363
 364                if (ret)
 365                        break;
 366
 367                schedule();
 368                try_to_freeze();
 369
 370                down_read(&umhelper_sem);
 371        }
 372        finish_wait(&usermodehelper_disabled_waitq, &wait);
 373        return ret;
 374}
 375EXPORT_SYMBOL_GPL(usermodehelper_read_trylock);
 376
 377long usermodehelper_read_lock_wait(long timeout)
 378{
 379        DEFINE_WAIT(wait);
 380
 381        if (timeout < 0)
 382                return -EINVAL;
 383
 384        down_read(&umhelper_sem);
 385        for (;;) {
 386                prepare_to_wait(&usermodehelper_disabled_waitq, &wait,
 387                                TASK_UNINTERRUPTIBLE);
 388                if (!usermodehelper_disabled)
 389                        break;
 390
 391                up_read(&umhelper_sem);
 392
 393                timeout = schedule_timeout(timeout);
 394                if (!timeout)
 395                        break;
 396
 397                down_read(&umhelper_sem);
 398        }
 399        finish_wait(&usermodehelper_disabled_waitq, &wait);
 400        return timeout;
 401}
 402EXPORT_SYMBOL_GPL(usermodehelper_read_lock_wait);
 403
 404void usermodehelper_read_unlock(void)
 405{
 406        up_read(&umhelper_sem);
 407}
 408EXPORT_SYMBOL_GPL(usermodehelper_read_unlock);
 409
 410/**
 411 * __usermodehelper_set_disable_depth - Modify usermodehelper_disabled.
 412 * @depth: New value to assign to usermodehelper_disabled.
 413 *
 414 * Change the value of usermodehelper_disabled (under umhelper_sem locked for
 415 * writing) and wakeup tasks waiting for it to change.
 416 */
 417void __usermodehelper_set_disable_depth(enum umh_disable_depth depth)
 418{
 419        down_write(&umhelper_sem);
 420        usermodehelper_disabled = depth;
 421        wake_up(&usermodehelper_disabled_waitq);
 422        up_write(&umhelper_sem);
 423}
 424
 425/**
 426 * __usermodehelper_disable - Prevent new helpers from being started.
 427 * @depth: New value to assign to usermodehelper_disabled.
 428 *
 429 * Set usermodehelper_disabled to @depth and wait for running helpers to exit.
 430 */
 431int __usermodehelper_disable(enum umh_disable_depth depth)
 432{
 433        long retval;
 434
 435        if (!depth)
 436                return -EINVAL;
 437
 438        down_write(&umhelper_sem);
 439        usermodehelper_disabled = depth;
 440        up_write(&umhelper_sem);
 441
 442        /*
 443         * From now on call_usermodehelper_exec() won't start any new
 444         * helpers, so it is sufficient if running_helpers turns out to
 445         * be zero at one point (it may be increased later, but that
 446         * doesn't matter).
 447         */
 448        retval = wait_event_timeout(running_helpers_waitq,
 449                                        atomic_read(&running_helpers) == 0,
 450                                        RUNNING_HELPERS_TIMEOUT);
 451        if (retval)
 452                return 0;
 453
 454        __usermodehelper_set_disable_depth(UMH_ENABLED);
 455        return -EAGAIN;
 456}
 457
 458static void helper_lock(void)
 459{
 460        atomic_inc(&running_helpers);
 461        smp_mb__after_atomic_inc();
 462}
 463
 464static void helper_unlock(void)
 465{
 466        if (atomic_dec_and_test(&running_helpers))
 467                wake_up(&running_helpers_waitq);
 468}
 469
 470/**
 471 * call_usermodehelper_setup - prepare to call a usermode helper
 472 * @path: path to usermode executable
 473 * @argv: arg vector for process
 474 * @envp: environment for process
 475 * @gfp_mask: gfp mask for memory allocation
 476 *
 477 * Returns either %NULL on allocation failure, or a subprocess_info
 478 * structure.  This should be passed to call_usermodehelper_exec to
 479 * exec the process and free the structure.
 480 */
 481static
 482struct subprocess_info *call_usermodehelper_setup(char *path, char **argv,
 483                                                  char **envp, gfp_t gfp_mask)
 484{
 485        struct subprocess_info *sub_info;
 486        sub_info = kzalloc(sizeof(struct subprocess_info), gfp_mask);
 487        if (!sub_info)
 488                goto out;
 489
 490        INIT_WORK(&sub_info->work, __call_usermodehelper);
 491        sub_info->path = path;
 492        sub_info->argv = argv;
 493        sub_info->envp = envp;
 494  out:
 495        return sub_info;
 496}
 497
 498/**
 499 * call_usermodehelper_setfns - set a cleanup/init function
 500 * @info: a subprocess_info returned by call_usermodehelper_setup
 501 * @cleanup: a cleanup function
 502 * @init: an init function
 503 * @data: arbitrary context sensitive data
 504 *
 505 * The init function is used to customize the helper process prior to
 506 * exec.  A non-zero return code causes the process to error out, exit,
 507 * and return the failure to the calling process
 508 *
 509 * The cleanup function is just before ethe subprocess_info is about to
 510 * be freed.  This can be used for freeing the argv and envp.  The
 511 * Function must be runnable in either a process context or the
 512 * context in which call_usermodehelper_exec is called.
 513 */
 514static
 515void call_usermodehelper_setfns(struct subprocess_info *info,
 516                    int (*init)(struct subprocess_info *info, struct cred *new),
 517                    void (*cleanup)(struct subprocess_info *info),
 518                    void *data)
 519{
 520        info->cleanup = cleanup;
 521        info->init = init;
 522        info->data = data;
 523}
 524
 525/**
 526 * call_usermodehelper_exec - start a usermode application
 527 * @sub_info: information about the subprocessa
 528 * @wait: wait for the application to finish and return status.
 529 *        when -1 don't wait at all, but you get no useful error back when
 530 *        the program couldn't be exec'ed. This makes it safe to call
 531 *        from interrupt context.
 532 *
 533 * Runs a user-space application.  The application is started
 534 * asynchronously if wait is not set, and runs as a child of keventd.
 535 * (ie. it runs with full root capabilities).
 536 */
 537static
 538int call_usermodehelper_exec(struct subprocess_info *sub_info, int wait)
 539{
 540        DECLARE_COMPLETION_ONSTACK(done);
 541        int retval = 0;
 542
 543        helper_lock();
 544        if (sub_info->path[0] == '\0')
 545                goto out;
 546
 547        if (!khelper_wq || usermodehelper_disabled) {
 548                retval = -EBUSY;
 549                goto out;
 550        }
 551
 552        sub_info->complete = &done;
 553        sub_info->wait = wait;
 554
 555        queue_work(khelper_wq, &sub_info->work);
 556        if (wait == UMH_NO_WAIT)        /* task has freed sub_info */
 557                goto unlock;
 558
 559        if (wait & UMH_KILLABLE) {
 560                retval = wait_for_completion_killable(&done);
 561                if (!retval)
 562                        goto wait_done;
 563
 564                /* umh_complete() will see NULL and free sub_info */
 565                if (xchg(&sub_info->complete, NULL))
 566                        goto unlock;
 567                /* fallthrough, umh_complete() was already called */
 568        }
 569
 570        wait_for_completion(&done);
 571wait_done:
 572        retval = sub_info->retval;
 573out:
 574        call_usermodehelper_freeinfo(sub_info);
 575unlock:
 576        helper_unlock();
 577        return retval;
 578}
 579
 580int call_usermodehelper_fns(
 581        char *path, char **argv, char **envp, int wait,
 582        int (*init)(struct subprocess_info *info, struct cred *new),
 583        void (*cleanup)(struct subprocess_info *), void *data)
 584{
 585        struct subprocess_info *info;
 586        gfp_t gfp_mask = (wait == UMH_NO_WAIT) ? GFP_ATOMIC : GFP_KERNEL;
 587
 588        info = call_usermodehelper_setup(path, argv, envp, gfp_mask);
 589
 590        if (info == NULL)
 591                return -ENOMEM;
 592
 593        call_usermodehelper_setfns(info, init, cleanup, data);
 594
 595        return call_usermodehelper_exec(info, wait);
 596}
 597EXPORT_SYMBOL(call_usermodehelper_fns);
 598
 599static int proc_cap_handler(struct ctl_table *table, int write,
 600                         void __user *buffer, size_t *lenp, loff_t *ppos)
 601{
 602        struct ctl_table t;
 603        unsigned long cap_array[_KERNEL_CAPABILITY_U32S];
 604        kernel_cap_t new_cap;
 605        int err, i;
 606
 607        if (write && (!capable(CAP_SETPCAP) ||
 608                      !capable(CAP_SYS_MODULE)))
 609                return -EPERM;
 610
 611        /*
 612         * convert from the global kernel_cap_t to the ulong array to print to
 613         * userspace if this is a read.
 614         */
 615        spin_lock(&umh_sysctl_lock);
 616        for (i = 0; i < _KERNEL_CAPABILITY_U32S; i++)  {
 617                if (table->data == CAP_BSET)
 618                        cap_array[i] = usermodehelper_bset.cap[i];
 619                else if (table->data == CAP_PI)
 620                        cap_array[i] = usermodehelper_inheritable.cap[i];
 621                else
 622                        BUG();
 623        }
 624        spin_unlock(&umh_sysctl_lock);
 625
 626        t = *table;
 627        t.data = &cap_array;
 628
 629        /*
 630         * actually read or write and array of ulongs from userspace.  Remember
 631         * these are least significant 32 bits first
 632         */
 633        err = proc_doulongvec_minmax(&t, write, buffer, lenp, ppos);
 634        if (err < 0)
 635                return err;
 636
 637        /*
 638         * convert from the sysctl array of ulongs to the kernel_cap_t
 639         * internal representation
 640         */
 641        for (i = 0; i < _KERNEL_CAPABILITY_U32S; i++)
 642                new_cap.cap[i] = cap_array[i];
 643
 644        /*
 645         * Drop everything not in the new_cap (but don't add things)
 646         */
 647        spin_lock(&umh_sysctl_lock);
 648        if (write) {
 649                if (table->data == CAP_BSET)
 650                        usermodehelper_bset = cap_intersect(usermodehelper_bset, new_cap);
 651                if (table->data == CAP_PI)
 652                        usermodehelper_inheritable = cap_intersect(usermodehelper_inheritable, new_cap);
 653        }
 654        spin_unlock(&umh_sysctl_lock);
 655
 656        return 0;
 657}
 658
 659struct ctl_table usermodehelper_table[] = {
 660        {
 661                .procname       = "bset",
 662                .data           = CAP_BSET,
 663                .maxlen         = _KERNEL_CAPABILITY_U32S * sizeof(unsigned long),
 664                .mode           = 0600,
 665                .proc_handler   = proc_cap_handler,
 666        },
 667        {
 668                .procname       = "inheritable",
 669                .data           = CAP_PI,
 670                .maxlen         = _KERNEL_CAPABILITY_U32S * sizeof(unsigned long),
 671                .mode           = 0600,
 672                .proc_handler   = proc_cap_handler,
 673        },
 674        { }
 675};
 676
 677void __init usermodehelper_init(void)
 678{
 679        khelper_wq = create_singlethread_workqueue("khelper");
 680        BUG_ON(!khelper_wq);
 681}
 682
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.