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 <asm/uaccess.h>
  41
  42 #include <trace/events/module.h>
  43
  44 extern int max_threads;
  45
  46 static struct workqueue_struct *khelper_wq;
  47
  48 #define CAP_BSET        (void *)1
  49 #define CAP_PI          (void *)2
  50
  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);
  55
  56 #ifdef CONFIG_MODULES
  57
  58 /*
  59         modprobe_path is set via /proc/sys.
  60 */
  61 char modprobe_path[KMOD_PATH_LEN] = "/sbin/modprobe";
  62
  63 /**
  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
  68  *
  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.
  74  *
  75  * If module auto-loading support is disabled then this function
  76  * becomes a no-operation.
  77  */
  78 int __request_module(bool wait, const char *fmt, ...)
  79 {
  80         va_list args;
  81         char module_name[MODULE_NAME_LEN];
  82         unsigned int max_modprobes;
  83         int ret;
  84         char *argv[] = { modprobe_path, "-q", "--", module_name, NULL };
  85         static char *envp[] = { "HOME=/",
  86                                 "TERM=linux",
  87                                 "PATH=/sbin:/usr/sbin:/bin:/usr/bin",
  88                                 NULL };
  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;
  92
  93         va_start(args, fmt);
  94         ret = vsnprintf(module_name, MODULE_NAME_LEN, fmt, args);
  95         va_end(args);
  96         if (ret >= MODULE_NAME_LEN)
  97                 return -ENAMETOOLONG;
  98
  99         ret = security_kernel_module_request(module_name);
 100         if (ret)
 101                 return ret;
 102
 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.
 110          * KAO.
 111          *
 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
 114          */
 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) {
 120                         printk(KERN_ERR
 121                                "request_module: runaway loop modprobe %s\n",
 122                                module_name);
 123                         kmod_loop_msg++;
 124                 }
 125                 atomic_dec(&kmod_concurrent);
 126                 return -ENOMEM;
 127         }
 128
 129         trace_module_request(module_name, wait, _RET_IP_);
 130
 131         ret = call_usermodehelper_fns(modprobe_path, argv, envp,
 132                         wait ? UMH_WAIT_PROC : UMH_WAIT_EXEC,
 133                         NULL, NULL, NULL);
 134
 135         atomic_dec(&kmod_concurrent);
 136         return ret;
 137 }
 138 EXPORT_SYMBOL(__request_module);
 139 #endif /* CONFIG_MODULES */
 140
 141 /*
 142  * This is the task which runs the usermode application
 143  */
 144 static int ____call_usermodehelper(void *data)
 145 {
 146         struct subprocess_info *sub_info = data;
 147         struct cred *new;
 148         int retval;
 149
 150         spin_lock_irq(&current->sighand->siglock);
 151         flush_signal_handlers(current, 1);
 152         spin_unlock_irq(&current->sighand->siglock);
 153
 154         /* We can run anywhere, unlike our parent keventd(). */
 155         set_cpus_allowed_ptr(current, cpu_all_mask);
 156
 157         /*
 158          * Our parent is keventd, which runs with elevated scheduling priority.
 159          * Avoid propagating that into the userspace child.
 160          */
 161         set_user_nice(current, 0);
 162
 163         retval = -ENOMEM;
 164         new = prepare_kernel_cred(current);
 165         if (!new)
 166                 goto fail;
 167
 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);
 173
 174         if (sub_info->init) {
 175                 retval = sub_info->init(sub_info, new);
 176                 if (retval) {
 177                         abort_creds(new);
 178                         goto fail;
 179                 }
 180         }
 181
 182         commit_creds(new);
 183
 184         retval = kernel_execve(sub_info->path,
 185                                (const char *const *)sub_info->argv,
 186                                (const char *const *)sub_info->envp);
 187
 188         /* Exec failed? */
 189 fail:
 190         sub_info->retval = retval;
 191         do_exit(0);
 192 }
 193
 194 void call_usermodehelper_freeinfo(struct subprocess_info *info)
 195 {
 196         if (info->cleanup)
 197                 (*info->cleanup)(info);
 198         kfree(info);
 199 }
 200 EXPORT_SYMBOL(call_usermodehelper_freeinfo);
 201
 202 /* Keventd can't block, but this (a child) can. */
 203 static int wait_for_helper(void *data)
 204 {
 205         struct subprocess_info *sub_info = data;
 206         pid_t pid;
 207
 208         /* If SIGCLD is ignored sys_wait4 won't populate the status. */
 209         spin_lock_irq(&current->sighand->siglock);
 210         current->sighand->action[SIGCHLD-1].sa.sa_handler = SIG_DFL;
 211         spin_unlock_irq(&current->sighand->siglock);
 212
 213         pid = kernel_thread(____call_usermodehelper, sub_info, SIGCHLD);
 214         if (pid < 0) {
 215                 sub_info->retval = pid;
 216         } else {
 217                 int ret = -ECHILD;
 218                 /*
 219                  * Normally it is bogus to call wait4() from in-kernel because
 220                  * wait4() wants to write the exit code to a userspace address.
 221                  * But wait_for_helper() always runs as keventd, and put_user()
 222                  * to a kernel address works OK for kernel threads, due to their
 223                  * having an mm_segment_t which spans the entire address space.
 224                  *
 225                  * Thus the __user pointer cast is valid here.
 226                  */
 227                 sys_wait4(pid, (int __user *)&ret, 0, NULL);
 228
 229                 /*
 230                  * If ret is 0, either ____call_usermodehelper failed and the
 231                  * real error code is already in sub_info->retval or
 232                  * sub_info->retval is 0 anyway, so don't mess with it then.
 233                  */
 234                 if (ret)
 235                         sub_info->retval = ret;
 236         }
 237
 238         complete(sub_info->complete);
 239         return 0;
 240 }
 241
 242 /* This is run by khelper thread  */
 243 static void __call_usermodehelper(struct work_struct *work)
 244 {
 245         struct subprocess_info *sub_info =
 246                 container_of(work, struct subprocess_info, work);
 247         enum umh_wait wait = sub_info->wait;
 248         pid_t pid;
 249
 250         /* CLONE_VFORK: wait until the usermode helper has execve'd
 251          * successfully We need the data structures to stay around
 252          * until that is done.  */
 253         if (wait == UMH_WAIT_PROC)
 254                 pid = kernel_thread(wait_for_helper, sub_info,
 255                                     CLONE_FS | CLONE_FILES | SIGCHLD);
 256         else
 257                 pid = kernel_thread(____call_usermodehelper, sub_info,
 258                                     CLONE_VFORK | SIGCHLD);
 259
 260         switch (wait) {
 261         case UMH_NO_WAIT:
 262                 call_usermodehelper_freeinfo(sub_info);
 263                 break;
 264
 265         case UMH_WAIT_PROC:
 266                 if (pid > 0)
 267                         break;
 268                 /* FALLTHROUGH */
 269         case UMH_WAIT_EXEC:
 270                 if (pid < 0)
 271                         sub_info->retval = pid;
 272                 complete(sub_info->complete);
 273         }
 274 }
 275
 276 /*
 277  * If set, call_usermodehelper_exec() will exit immediately returning -EBUSY
 278  * (used for preventing user land processes from being created after the user
 279  * land has been frozen during a system-wide hibernation or suspend operation).
 280  * Should always be manipulated under umhelper_sem acquired for write.
 281  */
 282 static int usermodehelper_disabled = 1;
 283
 284 /* Number of helpers running */
 285 static atomic_t running_helpers = ATOMIC_INIT(0);
 286
 287 /*
 288  * Wait queue head used by usermodehelper_disable() to wait for all running
 289  * helpers to finish.
 290  */
 291 static DECLARE_WAIT_QUEUE_HEAD(running_helpers_waitq);
 292
 293 /*
 294  * Time to wait for running_helpers to become zero before the setting of
 295  * usermodehelper_disabled in usermodehelper_disable() fails
 296  */
 297 #define RUNNING_HELPERS_TIMEOUT (5 * HZ)
 298
 299 void read_lock_usermodehelper(void)
 300 {
 301         down_read(&umhelper_sem);
 302 }
 303 EXPORT_SYMBOL_GPL(read_lock_usermodehelper);
 304
 305 void read_unlock_usermodehelper(void)
 306 {
 307         up_read(&umhelper_sem);
 308 }
 309 EXPORT_SYMBOL_GPL(read_unlock_usermodehelper);
 310
 311 /**
 312  * usermodehelper_disable - prevent new helpers from being started
 313  */
 314 int usermodehelper_disable(void)
 315 {
 316         long retval;
 317
 318         down_write(&umhelper_sem);
 319         usermodehelper_disabled = 1;
 320         up_write(&umhelper_sem);
 321
 322         /*
 323          * From now on call_usermodehelper_exec() won't start any new
 324          * helpers, so it is sufficient if running_helpers turns out to
 325          * be zero at one point (it may be increased later, but that
 326          * doesn't matter).
 327          */
 328         retval = wait_event_timeout(running_helpers_waitq,
 329                                         atomic_read(&running_helpers) == 0,
 330                                         RUNNING_HELPERS_TIMEOUT);
 331         if (retval)
 332                 return 0;
 333
 334         down_write(&umhelper_sem);
 335         usermodehelper_disabled = 0;
 336         up_write(&umhelper_sem);
 337         return -EAGAIN;
 338 }
 339
 340 /**
 341  * usermodehelper_enable - allow new helpers to be started again
 342  */
 343 void usermodehelper_enable(void)
 344 {
 345         down_write(&umhelper_sem);
 346         usermodehelper_disabled = 0;
 347         up_write(&umhelper_sem);
 348 }
 349
 350 /**
 351  * usermodehelper_is_disabled - check if new helpers are allowed to be started
 352  */
 353 bool usermodehelper_is_disabled(void)
 354 {
 355         return usermodehelper_disabled;
 356 }
 357 EXPORT_SYMBOL_GPL(usermodehelper_is_disabled);
 358
 359 static void helper_lock(void)
 360 {
 361         atomic_inc(&running_helpers);
 362         smp_mb__after_atomic_inc();
 363 }
 364
 365 static void helper_unlock(void)
 366 {
 367         if (atomic_dec_and_test(&running_helpers))
 368                 wake_up(&running_helpers_waitq);
 369 }
 370
 371 /**
 372  * call_usermodehelper_setup - prepare to call a usermode helper
 373  * @path: path to usermode executable
 374  * @argv: arg vector for process
 375  * @envp: environment for process
 376  * @gfp_mask: gfp mask for memory allocation
 377  *
 378  * Returns either %NULL on allocation failure, or a subprocess_info
 379  * structure.  This should be passed to call_usermodehelper_exec to
 380  * exec the process and free the structure.
 381  */
 382 struct subprocess_info *call_usermodehelper_setup(char *path, char **argv,
 383                                                   char **envp, gfp_t gfp_mask)
 384 {
 385         struct subprocess_info *sub_info;
 386         sub_info = kzalloc(sizeof(struct subprocess_info), gfp_mask);
 387         if (!sub_info)
 388                 goto out;
 389
 390         INIT_WORK(&sub_info->work, __call_usermodehelper);
 391         sub_info->path = path;
 392         sub_info->argv = argv;
 393         sub_info->envp = envp;
 394   out:
 395         return sub_info;
 396 }
 397 EXPORT_SYMBOL(call_usermodehelper_setup);
 398
 399 /**
 400  * call_usermodehelper_setfns - set a cleanup/init function
 401  * @info: a subprocess_info returned by call_usermodehelper_setup
 402  * @cleanup: a cleanup function
 403  * @init: an init function
 404  * @data: arbitrary context sensitive data
 405  *
 406  * The init function is used to customize the helper process prior to
 407  * exec.  A non-zero return code causes the process to error out, exit,
 408  * and return the failure to the calling process
 409  *
 410  * The cleanup function is just before ethe subprocess_info is about to
 411  * be freed.  This can be used for freeing the argv and envp.  The
 412  * Function must be runnable in either a process context or the
 413  * context in which call_usermodehelper_exec is called.
 414  */
 415 void call_usermodehelper_setfns(struct subprocess_info *info,
 416                     int (*init)(struct subprocess_info *info, struct cred *new),
 417                     void (*cleanup)(struct subprocess_info *info),
 418                     void *data)
 419 {
 420         info->cleanup = cleanup;
 421         info->init = init;
 422         info->data = data;
 423 }
 424 EXPORT_SYMBOL(call_usermodehelper_setfns);
 425
 426 /**
 427  * call_usermodehelper_exec - start a usermode application
 428  * @sub_info: information about the subprocessa
 429  * @wait: wait for the application to finish and return status.
 430  *        when -1 don't wait at all, but you get no useful error back when
 431  *        the program couldn't be exec'ed. This makes it safe to call
 432  *        from interrupt context.
 433  *
 434  * Runs a user-space application.  The application is started
 435  * asynchronously if wait is not set, and runs as a child of keventd.
 436  * (ie. it runs with full root capabilities).
 437  */
 438 int call_usermodehelper_exec(struct subprocess_info *sub_info,
 439                              enum umh_wait wait)
 440 {
 441         DECLARE_COMPLETION_ONSTACK(done);
 442         int retval = 0;
 443
 444         helper_lock();
 445         if (sub_info->path[0] == '\0')
 446                 goto out;
 447
 448         if (!khelper_wq || usermodehelper_disabled) {
 449                 retval = -EBUSY;
 450                 goto out;
 451         }
 452
 453         sub_info->complete = &done;
 454         sub_info->wait = wait;
 455
 456         queue_work(khelper_wq, &sub_info->work);
 457         if (wait == UMH_NO_WAIT)        /* task has freed sub_info */
 458                 goto unlock;
 459         wait_for_completion(&done);
 460         retval = sub_info->retval;
 461
 462 out:
 463         call_usermodehelper_freeinfo(sub_info);
 464 unlock:
 465         helper_unlock();
 466         return retval;
 467 }
 468 EXPORT_SYMBOL(call_usermodehelper_exec);
 469
 470 static int proc_cap_handler(struct ctl_table *table, int write,
 471                          void __user *buffer, size_t *lenp, loff_t *ppos)
 472 {
 473         struct ctl_table t;
 474         unsigned long cap_array[_KERNEL_CAPABILITY_U32S];
 475         kernel_cap_t new_cap;
 476         int err, i;
 477
 478         if (write && (!capable(CAP_SETPCAP) ||
 479                       !capable(CAP_SYS_MODULE)))
 480                 return -EPERM;
 481
 482         /*
 483          * convert from the global kernel_cap_t to the ulong array to print to
 484          * userspace if this is a read.
 485          */
 486         spin_lock(&umh_sysctl_lock);
 487         for (i = 0; i < _KERNEL_CAPABILITY_U32S; i++)  {
 488                 if (table->data == CAP_BSET)
 489                         cap_array[i] = usermodehelper_bset.cap[i];
 490                 else if (table->data == CAP_PI)
 491                         cap_array[i] = usermodehelper_inheritable.cap[i];
 492                 else
 493                         BUG();
 494         }
 495         spin_unlock(&umh_sysctl_lock);
 496
 497         t = *table;
 498         t.data = &cap_array;
 499
 500         /*
 501          * actually read or write and array of ulongs from userspace.  Remember
 502          * these are least significant 32 bits first
 503          */
 504         err = proc_doulongvec_minmax(&t, write, buffer, lenp, ppos);
 505         if (err < 0)
 506                 return err;
 507
 508         /*
 509          * convert from the sysctl array of ulongs to the kernel_cap_t
 510          * internal representation
 511          */
 512         for (i = 0; i < _KERNEL_CAPABILITY_U32S; i++)
 513                 new_cap.cap[i] = cap_array[i];
 514
 515         /*
 516          * Drop everything not in the new_cap (but don't add things)
 517          */
 518         spin_lock(&umh_sysctl_lock);
 519         if (write) {
 520                 if (table->data == CAP_BSET)
 521                         usermodehelper_bset = cap_intersect(usermodehelper_bset, new_cap);
 522                 if (table->data == CAP_PI)
 523                         usermodehelper_inheritable = cap_intersect(usermodehelper_inheritable, new_cap);
 524         }
 525         spin_unlock(&umh_sysctl_lock);
 526
 527         return 0;
 528 }
 529
 530 struct ctl_table usermodehelper_table[] = {
 531         {
 532                 .procname       = "bset",
 533                 .data           = CAP_BSET,
 534                 .maxlen         = _KERNEL_CAPABILITY_U32S * sizeof(unsigned long),
 535                 .mode           = 0600,
 536                 .proc_handler   = proc_cap_handler,
 537         },
 538         {
 539                 .procname       = "inheritable",
 540                 .data           = CAP_PI,
 541                 .maxlen         = _KERNEL_CAPABILITY_U32S * sizeof(unsigned long),
 542                 .mode           = 0600,
 543                 .proc_handler   = proc_cap_handler,
 544         },
 545         { }
 546 };
 547
 548 void __init usermodehelper_init(void)
 549 {
 550         khelper_wq = create_singlethread_workqueue("khelper");
 551         BUG_ON(!khelper_wq);
 552 }