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 <asm/uaccess.h>
  40
  41 #include <trace/events/module.h>
  42
  43 extern int max_threads;
  44
  45 static struct workqueue_struct *khelper_wq;
  46
  47 #define CAP_BSET        (void *)1
  48 #define CAP_PI          (void *)2
  49
  50 static kernel_cap_t usermodehelper_bset = CAP_FULL_SET;
  51 static kernel_cap_t usermodehelper_inheritable = CAP_FULL_SET;
  52 static DEFINE_SPINLOCK(umh_sysctl_lock);
  53
  54 #ifdef CONFIG_MODULES
  55
  56 /*
  57         modprobe_path is set via /proc/sys.
  58 */
  59 char modprobe_path[KMOD_PATH_LEN] = "/sbin/modprobe";
  60
  61 /**
  62  * __request_module - try to load a kernel module
  63  * @wait: wait (or not) for the operation to complete
  64  * @fmt: printf style format string for the name of the module
  65  * @...: arguments as specified in the format string
  66  *
  67  * Load a module using the user mode module loader. The function returns
  68  * zero on success or a negative errno code on failure. Note that a
  69  * successful module load does not mean the module did not then unload
  70  * and exit on an error of its own. Callers must check that the service
  71  * they requested is now available not blindly invoke it.
  72  *
  73  * If module auto-loading support is disabled then this function
  74  * becomes a no-operation.
  75  */
  76 int __request_module(bool wait, const char *fmt, ...)
  77 {
  78         va_list args;
  79         char module_name[MODULE_NAME_LEN];
  80         unsigned int max_modprobes;
  81         int ret;
  82         char *argv[] = { modprobe_path, "-q", "--", module_name, NULL };
  83         static char *envp[] = { "HOME=/",
  84                                 "TERM=linux",
  85                                 "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
  86                                 NULL };
  87         static atomic_t kmod_concurrent = ATOMIC_INIT(0);
  88 #define MAX_KMOD_CONCURRENT 50  /* Completely arbitrary value - KAO */
  89         static int kmod_loop_msg;
  90
  91         va_start(args, fmt);
  92         ret = vsnprintf(module_name, MODULE_NAME_LEN, fmt, args);
  93         va_end(args);
  94         if (ret >= MODULE_NAME_LEN)
  95                 return -ENAMETOOLONG;
  96
  97         ret = security_kernel_module_request(module_name);
  98         if (ret)
  99                 return ret;
 100
 101         /* If modprobe needs a service that is in a module, we get a recursive
 102          * loop.  Limit the number of running kmod threads to max_threads/2 or
 103          * MAX_KMOD_CONCURRENT, whichever is the smaller.  A cleaner method
 104          * would be to run the parents of this process, counting how many times
 105          * kmod was invoked.  That would mean accessing the internals of the
 106          * process tables to get the command line, proc_pid_cmdline is static
 107          * and it is not worth changing the proc code just to handle this case.
 108          * KAO.
 109          *
 110          * "trace the ppid" is simple, but will fail if someone's
 111          * parent exits.  I think this is as good as it gets. --RR
 112          */
 113         max_modprobes = min(max_threads/2, MAX_KMOD_CONCURRENT);
 114         atomic_inc(&kmod_concurrent);
 115         if (atomic_read(&kmod_concurrent) > max_modprobes) {
 116                 /* We may be blaming an innocent here, but unlikely */
 117                 if (kmod_loop_msg++ < 5)
 118                         printk(KERN_ERR
 119                                "request_module: runaway loop modprobe %s\n",
 120                                module_name);
 121                 atomic_dec(&kmod_concurrent);
 122                 return -ENOMEM;
 123         }
 124
 125         trace_module_request(module_name, wait, _RET_IP_);
 126
 127         ret = call_usermodehelper_fns(modprobe_path, argv, envp,
 128                         wait ? UMH_WAIT_PROC : UMH_WAIT_EXEC,
 129                         NULL, NULL, NULL);
 130
 131         atomic_dec(&kmod_concurrent);
 132         return ret;
 133 }
 134 EXPORT_SYMBOL(__request_module);
 135 #endif /* CONFIG_MODULES */
 136
 137 /*
 138  * This is the task which runs the usermode application
 139  */
 140 static int ____call_usermodehelper(void *data)
 141 {
 142         struct subprocess_info *sub_info = data;
 143         struct cred *new;
 144         int retval;
 145
 146         spin_lock_irq(&current->sighand->siglock);
 147         flush_signal_handlers(current, 1);
 148         spin_unlock_irq(&current->sighand->siglock);
 149
 150         /* We can run anywhere, unlike our parent keventd(). */
 151         set_cpus_allowed_ptr(current, cpu_all_mask);
 152
 153         /*
 154          * Our parent is keventd, which runs with elevated scheduling priority.
 155          * Avoid propagating that into the userspace child.
 156          */
 157         set_user_nice(current, 0);
 158
 159         retval = -ENOMEM;
 160         new = prepare_kernel_cred(current);
 161         if (!new)
 162                 goto fail;
 163
 164         spin_lock(&umh_sysctl_lock);
 165         new->cap_bset = cap_intersect(usermodehelper_bset, new->cap_bset);
 166         new->cap_inheritable = cap_intersect(usermodehelper_inheritable,
 167                                              new->cap_inheritable);
 168         spin_unlock(&umh_sysctl_lock);
 169
 170         if (sub_info->init) {
 171                 retval = sub_info->init(sub_info, new);
 172                 if (retval) {
 173                         abort_creds(new);
 174                         goto fail;
 175                 }
 176         }
 177
 178         commit_creds(new);
 179
 180         retval = kernel_execve(sub_info->path,
 181                                (const char *const *)sub_info->argv,
 182                                (const char *const *)sub_info->envp);
 183
 184         /* Exec failed? */
 185 fail:
 186         sub_info->retval = retval;
 187         do_exit(0);
 188 }
 189
 190 void call_usermodehelper_freeinfo(struct subprocess_info *info)
 191 {
 192         if (info->cleanup)
 193                 (*info->cleanup)(info);
 194         kfree(info);
 195 }
 196 EXPORT_SYMBOL(call_usermodehelper_freeinfo);
 197
 198 /* Keventd can't block, but this (a child) can. */
 199 static int wait_for_helper(void *data)
 200 {
 201         struct subprocess_info *sub_info = data;
 202         pid_t pid;
 203
 204         /* If SIGCLD is ignored sys_wait4 won't populate the status. */
 205         spin_lock_irq(&current->sighand->siglock);
 206         current->sighand->action[SIGCHLD-1].sa.sa_handler = SIG_DFL;
 207         spin_unlock_irq(&current->sighand->siglock);
 208
 209         pid = kernel_thread(____call_usermodehelper, sub_info, SIGCHLD);
 210         if (pid < 0) {
 211                 sub_info->retval = pid;
 212         } else {
 213                 int ret = -ECHILD;
 214                 /*
 215                  * Normally it is bogus to call wait4() from in-kernel because
 216                  * wait4() wants to write the exit code to a userspace address.
 217                  * But wait_for_helper() always runs as keventd, and put_user()
 218                  * to a kernel address works OK for kernel threads, due to their
 219                  * having an mm_segment_t which spans the entire address space.
 220                  *
 221                  * Thus the __user pointer cast is valid here.
 222                  */
 223                 sys_wait4(pid, (int __user *)&ret, 0, NULL);
 224
 225                 /*
 226                  * If ret is 0, either ____call_usermodehelper failed and the
 227                  * real error code is already in sub_info->retval or
 228                  * sub_info->retval is 0 anyway, so don't mess with it then.
 229                  */
 230                 if (ret)
 231                         sub_info->retval = ret;
 232         }
 233
 234         complete(sub_info->complete);
 235         return 0;
 236 }
 237
 238 /* This is run by khelper thread  */
 239 static void __call_usermodehelper(struct work_struct *work)
 240 {
 241         struct subprocess_info *sub_info =
 242                 container_of(work, struct subprocess_info, work);
 243         enum umh_wait wait = sub_info->wait;
 244         pid_t pid;
 245
 246         /* CLONE_VFORK: wait until the usermode helper has execve'd
 247          * successfully We need the data structures to stay around
 248          * until that is done.  */
 249         if (wait == UMH_WAIT_PROC)
 250                 pid = kernel_thread(wait_for_helper, sub_info,
 251                                     CLONE_FS | CLONE_FILES | SIGCHLD);
 252         else
 253                 pid = kernel_thread(____call_usermodehelper, sub_info,
 254                                     CLONE_VFORK | SIGCHLD);
 255
 256         switch (wait) {
 257         case UMH_NO_WAIT:
 258                 call_usermodehelper_freeinfo(sub_info);
 259                 break;
 260
 261         case UMH_WAIT_PROC:
 262                 if (pid > 0)
 263                         break;
 264                 /* FALLTHROUGH */
 265         case UMH_WAIT_EXEC:
 266                 if (pid < 0)
 267                         sub_info->retval = pid;
 268                 complete(sub_info->complete);
 269         }
 270 }
 271
 272 /*
 273  * If set, call_usermodehelper_exec() will exit immediately returning -EBUSY
 274  * (used for preventing user land processes from being created after the user
 275  * land has been frozen during a system-wide hibernation or suspend operation).
 276  */
 277 static int usermodehelper_disabled;
 278
 279 /* Number of helpers running */
 280 static atomic_t running_helpers = ATOMIC_INIT(0);
 281
 282 /*
 283  * Wait queue head used by usermodehelper_pm_callback() to wait for all running
 284  * helpers to finish.
 285  */
 286 static DECLARE_WAIT_QUEUE_HEAD(running_helpers_waitq);
 287
 288 /*
 289  * Time to wait for running_helpers to become zero before the setting of
 290  * usermodehelper_disabled in usermodehelper_pm_callback() fails
 291  */
 292 #define RUNNING_HELPERS_TIMEOUT (5 * HZ)
 293
 294 /**
 295  * usermodehelper_disable - prevent new helpers from being started
 296  */
 297 int usermodehelper_disable(void)
 298 {
 299         long retval;
 300
 301         usermodehelper_disabled = 1;
 302         smp_mb();
 303         /*
 304          * From now on call_usermodehelper_exec() won't start any new
 305          * helpers, so it is sufficient if running_helpers turns out to
 306          * be zero at one point (it may be increased later, but that
 307          * doesn't matter).
 308          */
 309         retval = wait_event_timeout(running_helpers_waitq,
 310                                         atomic_read(&running_helpers) == 0,
 311                                         RUNNING_HELPERS_TIMEOUT);
 312         if (retval)
 313                 return 0;
 314
 315         usermodehelper_disabled = 0;
 316         return -EAGAIN;
 317 }
 318
 319 /**
 320  * usermodehelper_enable - allow new helpers to be started again
 321  */
 322 void usermodehelper_enable(void)
 323 {
 324         usermodehelper_disabled = 0;
 325 }
 326
 327 /**
 328  * usermodehelper_is_disabled - check if new helpers are allowed to be started
 329  */
 330 bool usermodehelper_is_disabled(void)
 331 {
 332         return usermodehelper_disabled;
 333 }
 334 EXPORT_SYMBOL_GPL(usermodehelper_is_disabled);
 335
 336 static void helper_lock(void)
 337 {
 338         atomic_inc(&running_helpers);
 339         smp_mb__after_atomic_inc();
 340 }
 341
 342 static void helper_unlock(void)
 343 {
 344         if (atomic_dec_and_test(&running_helpers))
 345                 wake_up(&running_helpers_waitq);
 346 }
 347
 348 /**
 349  * call_usermodehelper_setup - prepare to call a usermode helper
 350  * @path: path to usermode executable
 351  * @argv: arg vector for process
 352  * @envp: environment for process
 353  * @gfp_mask: gfp mask for memory allocation
 354  *
 355  * Returns either %NULL on allocation failure, or a subprocess_info
 356  * structure.  This should be passed to call_usermodehelper_exec to
 357  * exec the process and free the structure.
 358  */
 359 struct subprocess_info *call_usermodehelper_setup(char *path, char **argv,
 360                                                   char **envp, gfp_t gfp_mask)
 361 {
 362         struct subprocess_info *sub_info;
 363         sub_info = kzalloc(sizeof(struct subprocess_info), gfp_mask);
 364         if (!sub_info)
 365                 goto out;
 366
 367         INIT_WORK(&sub_info->work, __call_usermodehelper);
 368         sub_info->path = path;
 369         sub_info->argv = argv;
 370         sub_info->envp = envp;
 371   out:
 372         return sub_info;
 373 }
 374 EXPORT_SYMBOL(call_usermodehelper_setup);
 375
 376 /**
 377  * call_usermodehelper_setfns - set a cleanup/init function
 378  * @info: a subprocess_info returned by call_usermodehelper_setup
 379  * @cleanup: a cleanup function
 380  * @init: an init function
 381  * @data: arbitrary context sensitive data
 382  *
 383  * The init function is used to customize the helper process prior to
 384  * exec.  A non-zero return code causes the process to error out, exit,
 385  * and return the failure to the calling process
 386  *
 387  * The cleanup function is just before ethe subprocess_info is about to
 388  * be freed.  This can be used for freeing the argv and envp.  The
 389  * Function must be runnable in either a process context or the
 390  * context in which call_usermodehelper_exec is called.
 391  */
 392 void call_usermodehelper_setfns(struct subprocess_info *info,
 393                     int (*init)(struct subprocess_info *info, struct cred *new),
 394                     void (*cleanup)(struct subprocess_info *info),
 395                     void *data)
 396 {
 397         info->cleanup = cleanup;
 398         info->init = init;
 399         info->data = data;
 400 }
 401 EXPORT_SYMBOL(call_usermodehelper_setfns);
 402
 403 /**
 404  * call_usermodehelper_exec - start a usermode application
 405  * @sub_info: information about the subprocessa
 406  * @wait: wait for the application to finish and return status.
 407  *        when -1 don't wait at all, but you get no useful error back when
 408  *        the program couldn't be exec'ed. This makes it safe to call
 409  *        from interrupt context.
 410  *
 411  * Runs a user-space application.  The application is started
 412  * asynchronously if wait is not set, and runs as a child of keventd.
 413  * (ie. it runs with full root capabilities).
 414  */
 415 int call_usermodehelper_exec(struct subprocess_info *sub_info,
 416                              enum umh_wait wait)
 417 {
 418         DECLARE_COMPLETION_ONSTACK(done);
 419         int retval = 0;
 420
 421         helper_lock();
 422         if (sub_info->path[0] == '\0')
 423                 goto out;
 424
 425         if (!khelper_wq || usermodehelper_disabled) {
 426                 retval = -EBUSY;
 427                 goto out;
 428         }
 429
 430         sub_info->complete = &done;
 431         sub_info->wait = wait;
 432
 433         queue_work(khelper_wq, &sub_info->work);
 434         if (wait == UMH_NO_WAIT)        /* task has freed sub_info */
 435                 goto unlock;
 436         wait_for_completion(&done);
 437         retval = sub_info->retval;
 438
 439 out:
 440         call_usermodehelper_freeinfo(sub_info);
 441 unlock:
 442         helper_unlock();
 443         return retval;
 444 }
 445 EXPORT_SYMBOL(call_usermodehelper_exec);
 446
 447 static int proc_cap_handler(struct ctl_table *table, int write,
 448                          void __user *buffer, size_t *lenp, loff_t *ppos)
 449 {
 450         struct ctl_table t;
 451         unsigned long cap_array[_KERNEL_CAPABILITY_U32S];
 452         kernel_cap_t new_cap;
 453         int err, i;
 454
 455         if (write && (!capable(CAP_SETPCAP) ||
 456                       !capable(CAP_SYS_MODULE)))
 457                 return -EPERM;
 458
 459         /*
 460          * convert from the global kernel_cap_t to the ulong array to print to
 461          * userspace if this is a read.
 462          */
 463         spin_lock(&umh_sysctl_lock);
 464         for (i = 0; i < _KERNEL_CAPABILITY_U32S; i++)  {
 465                 if (table->data == CAP_BSET)
 466                         cap_array[i] = usermodehelper_bset.cap[i];
 467                 else if (table->data == CAP_PI)
 468                         cap_array[i] = usermodehelper_inheritable.cap[i];
 469                 else
 470                         BUG();
 471         }
 472         spin_unlock(&umh_sysctl_lock);
 473
 474         t = *table;
 475         t.data = &cap_array;
 476
 477         /*
 478          * actually read or write and array of ulongs from userspace.  Remember
 479          * these are least significant 32 bits first
 480          */
 481         err = proc_doulongvec_minmax(&t, write, buffer, lenp, ppos);
 482         if (err < 0)
 483                 return err;
 484
 485         /*
 486          * convert from the sysctl array of ulongs to the kernel_cap_t
 487          * internal representation
 488          */
 489         for (i = 0; i < _KERNEL_CAPABILITY_U32S; i++)
 490                 new_cap.cap[i] = cap_array[i];
 491
 492         /*
 493          * Drop everything not in the new_cap (but don't add things)
 494          */
 495         spin_lock(&umh_sysctl_lock);
 496         if (write) {
 497                 if (table->data == CAP_BSET)
 498                         usermodehelper_bset = cap_intersect(usermodehelper_bset, new_cap);
 499                 if (table->data == CAP_PI)
 500                         usermodehelper_inheritable = cap_intersect(usermodehelper_inheritable, new_cap);
 501         }
 502         spin_unlock(&umh_sysctl_lock);
 503
 504         return 0;
 505 }
 506
 507 struct ctl_table usermodehelper_table[] = {
 508         {
 509                 .procname       = "bset",
 510                 .data           = CAP_BSET,
 511                 .maxlen         = _KERNEL_CAPABILITY_U32S * sizeof(unsigned long),
 512                 .mode           = 0600,
 513                 .proc_handler   = proc_cap_handler,
 514         },
 515         {
 516                 .procname       = "inheritable",
 517                 .data           = CAP_PI,
 518                 .maxlen         = _KERNEL_CAPABILITY_U32S * sizeof(unsigned long),
 519                 .mode           = 0600,
 520                 .proc_handler   = proc_cap_handler,
 521         },
 522         { }
 523 };
 524
 525 void __init usermodehelper_init(void)
 526 {
 527         khelper_wq = create_singlethread_workqueue("khelper");
 528         BUG_ON(!khelper_wq);
 529 }