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                         kmod_loop_msg++;
 122                 }
 123                 atomic_dec(&kmod_concurrent);
 124                 return -ENOMEM;
 125         }
 126
 127         trace_module_request(module_name, wait, _RET_IP_);
 128
 129         ret = call_usermodehelper_fns(modprobe_path, argv, envp,
 130                         wait ? UMH_WAIT_PROC : UMH_WAIT_EXEC,
 131                         NULL, NULL, NULL);
 132
 133         atomic_dec(&kmod_concurrent);
 134         return ret;
 135 }
 136 EXPORT_SYMBOL(__request_module);
 137 #endif /* CONFIG_MODULES */
 138
 139 /*
 140  * This is the task which runs the usermode application
 141  */
 142 static int ____call_usermodehelper(void *data)
 143 {
 144         struct subprocess_info *sub_info = data;
 145         struct cred *new;
 146         int retval;
 147
 148         spin_lock_irq(&current->sighand->siglock);
 149         flush_signal_handlers(current, 1);
 150         spin_unlock_irq(&current->sighand->siglock);
 151
 152         /* We can run anywhere, unlike our parent keventd(). */
 153         set_cpus_allowed_ptr(current, cpu_all_mask);
 154
 155         /*
 156          * Our parent is keventd, which runs with elevated scheduling priority.
 157          * Avoid propagating that into the userspace child.
 158          */
 159         set_user_nice(current, 0);
 160
 161         retval = -ENOMEM;
 162         new = prepare_kernel_cred(current);
 163         if (!new)
 164                 goto fail;
 165
 166         spin_lock(&umh_sysctl_lock);
 167         new->cap_bset = cap_intersect(usermodehelper_bset, new->cap_bset);
 168         new->cap_inheritable = cap_intersect(usermodehelper_inheritable,
 169                                              new->cap_inheritable);
 170         spin_unlock(&umh_sysctl_lock);
 171
 172         if (sub_info->init) {
 173                 retval = sub_info->init(sub_info, new);
 174                 if (retval) {
 175                         abort_creds(new);
 176                         goto fail;
 177                 }
 178         }
 179
 180         commit_creds(new);
 181
 182         retval = kernel_execve(sub_info->path,
 183                                (const char *const *)sub_info->argv,
 184                                (const char *const *)sub_info->envp);
 185
 186         /* Exec failed? */
 187 fail:
 188         sub_info->retval = retval;
 189         do_exit(0);
 190 }
 191
 192 void call_usermodehelper_freeinfo(struct subprocess_info *info)
 193 {
 194         if (info->cleanup)
 195                 (*info->cleanup)(info);
 196         kfree(info);
 197 }
 198 EXPORT_SYMBOL(call_usermodehelper_freeinfo);
 199
 200 /* Keventd can't block, but this (a child) can. */
 201 static int wait_for_helper(void *data)
 202 {
 203         struct subprocess_info *sub_info = data;
 204         pid_t pid;
 205
 206         /* If SIGCLD is ignored sys_wait4 won't populate the status. */
 207         spin_lock_irq(&current->sighand->siglock);
 208         current->sighand->action[SIGCHLD-1].sa.sa_handler = SIG_DFL;
 209         spin_unlock_irq(&current->sighand->siglock);
 210
 211         pid = kernel_thread(____call_usermodehelper, sub_info, SIGCHLD);
 212         if (pid < 0) {
 213                 sub_info->retval = pid;
 214         } else {
 215                 int ret = -ECHILD;
 216                 /*
 217                  * Normally it is bogus to call wait4() from in-kernel because
 218                  * wait4() wants to write the exit code to a userspace address.
 219                  * But wait_for_helper() always runs as keventd, and put_user()
 220                  * to a kernel address works OK for kernel threads, due to their
 221                  * having an mm_segment_t which spans the entire address space.
 222                  *
 223                  * Thus the __user pointer cast is valid here.
 224                  */
 225                 sys_wait4(pid, (int __user *)&ret, 0, NULL);
 226
 227                 /*
 228                  * If ret is 0, either ____call_usermodehelper failed and the
 229                  * real error code is already in sub_info->retval or
 230                  * sub_info->retval is 0 anyway, so don't mess with it then.
 231                  */
 232                 if (ret)
 233                         sub_info->retval = ret;
 234         }
 235
 236         complete(sub_info->complete);
 237         return 0;
 238 }
 239
 240 /* This is run by khelper thread  */
 241 static void __call_usermodehelper(struct work_struct *work)
 242 {
 243         struct subprocess_info *sub_info =
 244                 container_of(work, struct subprocess_info, work);
 245         enum umh_wait wait = sub_info->wait;
 246         pid_t pid;
 247
 248         /* CLONE_VFORK: wait until the usermode helper has execve'd
 249          * successfully We need the data structures to stay around
 250          * until that is done.  */
 251         if (wait == UMH_WAIT_PROC)
 252                 pid = kernel_thread(wait_for_helper, sub_info,
 253                                     CLONE_FS | CLONE_FILES | SIGCHLD);
 254         else
 255                 pid = kernel_thread(____call_usermodehelper, sub_info,
 256                                     CLONE_VFORK | SIGCHLD);
 257
 258         switch (wait) {
 259         case UMH_NO_WAIT:
 260                 call_usermodehelper_freeinfo(sub_info);
 261                 break;
 262
 263         case UMH_WAIT_PROC:
 264                 if (pid > 0)
 265                         break;
 266                 /* FALLTHROUGH */
 267         case UMH_WAIT_EXEC:
 268                 if (pid < 0)
 269                         sub_info->retval = pid;
 270                 complete(sub_info->complete);
 271         }
 272 }
 273
 274 /*
 275  * If set, call_usermodehelper_exec() will exit immediately returning -EBUSY
 276  * (used for preventing user land processes from being created after the user
 277  * land has been frozen during a system-wide hibernation or suspend operation).
 278  */
 279 static int usermodehelper_disabled = 1;
 280
 281 /* Number of helpers running */
 282 static atomic_t running_helpers = ATOMIC_INIT(0);
 283
 284 /*
 285  * Wait queue head used by usermodehelper_pm_callback() to wait for all running
 286  * helpers to finish.
 287  */
 288 static DECLARE_WAIT_QUEUE_HEAD(running_helpers_waitq);
 289
 290 /*
 291  * Time to wait for running_helpers to become zero before the setting of
 292  * usermodehelper_disabled in usermodehelper_pm_callback() fails
 293  */
 294 #define RUNNING_HELPERS_TIMEOUT (5 * HZ)
 295
 296 /**
 297  * usermodehelper_disable - prevent new helpers from being started
 298  */
 299 int usermodehelper_disable(void)
 300 {
 301         long retval;
 302
 303         usermodehelper_disabled = 1;
 304         smp_mb();
 305         /*
 306          * From now on call_usermodehelper_exec() won't start any new
 307          * helpers, so it is sufficient if running_helpers turns out to
 308          * be zero at one point (it may be increased later, but that
 309          * doesn't matter).
 310          */
 311         retval = wait_event_timeout(running_helpers_waitq,
 312                                         atomic_read(&running_helpers) == 0,
 313                                         RUNNING_HELPERS_TIMEOUT);
 314         if (retval)
 315                 return 0;
 316
 317         usermodehelper_disabled = 0;
 318         return -EAGAIN;
 319 }
 320
 321 /**
 322  * usermodehelper_enable - allow new helpers to be started again
 323  */
 324 void usermodehelper_enable(void)
 325 {
 326         usermodehelper_disabled = 0;
 327 }
 328
 329 /**
 330  * usermodehelper_is_disabled - check if new helpers are allowed to be started
 331  */
 332 bool usermodehelper_is_disabled(void)
 333 {
 334         return usermodehelper_disabled;
 335 }
 336 EXPORT_SYMBOL_GPL(usermodehelper_is_disabled);
 337
 338 static void helper_lock(void)
 339 {
 340         atomic_inc(&running_helpers);
 341         smp_mb__after_atomic_inc();
 342 }
 343
 344 static void helper_unlock(void)
 345 {
 346         if (atomic_dec_and_test(&running_helpers))
 347                 wake_up(&running_helpers_waitq);
 348 }
 349
 350 /**
 351  * call_usermodehelper_setup - prepare to call a usermode helper
 352  * @path: path to usermode executable
 353  * @argv: arg vector for process
 354  * @envp: environment for process
 355  * @gfp_mask: gfp mask for memory allocation
 356  *
 357  * Returns either %NULL on allocation failure, or a subprocess_info
 358  * structure.  This should be passed to call_usermodehelper_exec to
 359  * exec the process and free the structure.
 360  */
 361 struct subprocess_info *call_usermodehelper_setup(char *path, char **argv,
 362                                                   char **envp, gfp_t gfp_mask)
 363 {
 364         struct subprocess_info *sub_info;
 365         sub_info = kzalloc(sizeof(struct subprocess_info), gfp_mask);
 366         if (!sub_info)
 367                 goto out;
 368
 369         INIT_WORK(&sub_info->work, __call_usermodehelper);
 370         sub_info->path = path;
 371         sub_info->argv = argv;
 372         sub_info->envp = envp;
 373   out:
 374         return sub_info;
 375 }
 376 EXPORT_SYMBOL(call_usermodehelper_setup);
 377
 378 /**
 379  * call_usermodehelper_setfns - set a cleanup/init function
 380  * @info: a subprocess_info returned by call_usermodehelper_setup
 381  * @cleanup: a cleanup function
 382  * @init: an init function
 383  * @data: arbitrary context sensitive data
 384  *
 385  * The init function is used to customize the helper process prior to
 386  * exec.  A non-zero return code causes the process to error out, exit,
 387  * and return the failure to the calling process
 388  *
 389  * The cleanup function is just before ethe subprocess_info is about to
 390  * be freed.  This can be used for freeing the argv and envp.  The
 391  * Function must be runnable in either a process context or the
 392  * context in which call_usermodehelper_exec is called.
 393  */
 394 void call_usermodehelper_setfns(struct subprocess_info *info,
 395                     int (*init)(struct subprocess_info *info, struct cred *new),
 396                     void (*cleanup)(struct subprocess_info *info),
 397                     void *data)
 398 {
 399         info->cleanup = cleanup;
 400         info->init = init;
 401         info->data = data;
 402 }
 403 EXPORT_SYMBOL(call_usermodehelper_setfns);
 404
 405 /**
 406  * call_usermodehelper_exec - start a usermode application
 407  * @sub_info: information about the subprocessa
 408  * @wait: wait for the application to finish and return status.
 409  *        when -1 don't wait at all, but you get no useful error back when
 410  *        the program couldn't be exec'ed. This makes it safe to call
 411  *        from interrupt context.
 412  *
 413  * Runs a user-space application.  The application is started
 414  * asynchronously if wait is not set, and runs as a child of keventd.
 415  * (ie. it runs with full root capabilities).
 416  */
 417 int call_usermodehelper_exec(struct subprocess_info *sub_info,
 418                              enum umh_wait wait)
 419 {
 420         DECLARE_COMPLETION_ONSTACK(done);
 421         int retval = 0;
 422
 423         helper_lock();
 424         if (sub_info->path[0] == '\0')
 425                 goto out;
 426
 427         if (!khelper_wq || usermodehelper_disabled) {
 428                 retval = -EBUSY;
 429                 goto out;
 430         }
 431
 432         sub_info->complete = &done;
 433         sub_info->wait = wait;
 434
 435         queue_work(khelper_wq, &sub_info->work);
 436         if (wait == UMH_NO_WAIT)        /* task has freed sub_info */
 437                 goto unlock;
 438         wait_for_completion(&done);
 439         retval = sub_info->retval;
 440
 441 out:
 442         call_usermodehelper_freeinfo(sub_info);
 443 unlock:
 444         helper_unlock();
 445         return retval;
 446 }
 447 EXPORT_SYMBOL(call_usermodehelper_exec);
 448
 449 static int proc_cap_handler(struct ctl_table *table, int write,
 450                          void __user *buffer, size_t *lenp, loff_t *ppos)
 451 {
 452         struct ctl_table t;
 453         unsigned long cap_array[_KERNEL_CAPABILITY_U32S];
 454         kernel_cap_t new_cap;
 455         int err, i;
 456
 457         if (write && (!capable(CAP_SETPCAP) ||
 458                       !capable(CAP_SYS_MODULE)))
 459                 return -EPERM;
 460
 461         /*
 462          * convert from the global kernel_cap_t to the ulong array to print to
 463          * userspace if this is a read.
 464          */
 465         spin_lock(&umh_sysctl_lock);
 466         for (i = 0; i < _KERNEL_CAPABILITY_U32S; i++)  {
 467                 if (table->data == CAP_BSET)
 468                         cap_array[i] = usermodehelper_bset.cap[i];
 469                 else if (table->data == CAP_PI)
 470                         cap_array[i] = usermodehelper_inheritable.cap[i];
 471                 else
 472                         BUG();
 473         }
 474         spin_unlock(&umh_sysctl_lock);
 475
 476         t = *table;
 477         t.data = &cap_array;
 478
 479         /*
 480          * actually read or write and array of ulongs from userspace.  Remember
 481          * these are least significant 32 bits first
 482          */
 483         err = proc_doulongvec_minmax(&t, write, buffer, lenp, ppos);
 484         if (err < 0)
 485                 return err;
 486
 487         /*
 488          * convert from the sysctl array of ulongs to the kernel_cap_t
 489          * internal representation
 490          */
 491         for (i = 0; i < _KERNEL_CAPABILITY_U32S; i++)
 492                 new_cap.cap[i] = cap_array[i];
 493
 494         /*
 495          * Drop everything not in the new_cap (but don't add things)
 496          */
 497         spin_lock(&umh_sysctl_lock);
 498         if (write) {
 499                 if (table->data == CAP_BSET)
 500                         usermodehelper_bset = cap_intersect(usermodehelper_bset, new_cap);
 501                 if (table->data == CAP_PI)
 502                         usermodehelper_inheritable = cap_intersect(usermodehelper_inheritable, new_cap);
 503         }
 504         spin_unlock(&umh_sysctl_lock);
 505
 506         return 0;
 507 }
 508
 509 struct ctl_table usermodehelper_table[] = {
 510         {
 511                 .procname       = "bset",
 512                 .data           = CAP_BSET,
 513                 .maxlen         = _KERNEL_CAPABILITY_U32S * sizeof(unsigned long),
 514                 .mode           = 0600,
 515                 .proc_handler   = proc_cap_handler,
 516         },
 517         {
 518                 .procname       = "inheritable",
 519                 .data           = CAP_PI,
 520                 .maxlen         = _KERNEL_CAPABILITY_U32S * sizeof(unsigned long),
 521                 .mode           = 0600,
 522                 .proc_handler   = proc_cap_handler,
 523         },
 524         { }
 525 };
 526
 527 void __init usermodehelper_init(void)
 528 {
 529         khelper_wq = create_singlethread_workqueue("khelper");
 530         BUG_ON(!khelper_wq);
 531 }