kernel/umh.c

   1 /*
   2  * umh - the kernel usermode helper
   3  */
   4 #include <linux/module.h>
   5 #include <linux/sched.h>
   6 #include <linux/sched/task.h>
   7 #include <linux/binfmts.h>
   8 #include <linux/syscalls.h>
   9 #include <linux/unistd.h>
  10 #include <linux/kmod.h>
  11 #include <linux/slab.h>
  12 #include <linux/completion.h>
  13 #include <linux/cred.h>
  14 #include <linux/file.h>
  15 #include <linux/fdtable.h>
  16 #include <linux/fs_struct.h>
  17 #include <linux/workqueue.h>
  18 #include <linux/security.h>
  19 #include <linux/mount.h>
  20 #include <linux/kernel.h>
  21 #include <linux/init.h>
  22 #include <linux/resource.h>
  23 #include <linux/notifier.h>
  24 #include <linux/suspend.h>
  25 #include <linux/rwsem.h>
  26 #include <linux/ptrace.h>
  27 #include <linux/async.h>
  28 #include <linux/uaccess.h>
  29
  30 #include <trace/events/module.h>
  31
  32 #define CAP_BSET        (void *)1
  33 #define CAP_PI          (void *)2
  34
  35 static kernel_cap_t usermodehelper_bset = CAP_FULL_SET;
  36 static kernel_cap_t usermodehelper_inheritable = CAP_FULL_SET;
  37 static DEFINE_SPINLOCK(umh_sysctl_lock);
  38 static DECLARE_RWSEM(umhelper_sem);
  39
  40 static void call_usermodehelper_freeinfo(struct subprocess_info *info)
  41 {
  42         if (info->cleanup)
  43                 (*info->cleanup)(info);
  44         kfree(info);
  45 }
  46
  47 static void umh_complete(struct subprocess_info *sub_info)
  48 {
  49         struct completion *comp = xchg(&sub_info->complete, NULL);
  50         /*
  51          * See call_usermodehelper_exec(). If xchg() returns NULL
  52          * we own sub_info, the UMH_KILLABLE caller has gone away
  53          * or the caller used UMH_NO_WAIT.
  54          */
  55         if (comp)
  56                 complete(comp);
  57         else
  58                 call_usermodehelper_freeinfo(sub_info);
  59 }
  60
  61 /*
  62  * This is the task which runs the usermode application
  63  */
  64 static int call_usermodehelper_exec_async(void *data)
  65 {
  66         struct subprocess_info *sub_info = data;
  67         struct cred *new;
  68         int retval;
  69
  70         spin_lock_irq(&current->sighand->siglock);
  71         flush_signal_handlers(current, 1);
  72         spin_unlock_irq(&current->sighand->siglock);
  73
  74         /*
  75          * Initial kernel threads share ther FS with init, in order to
  76          * get the init root directory. But we've now created a new
  77          * thread that is going to execve a user process and has its own
  78          * 'struct fs_struct'. Reset umask to the default.
  79          */
  80         current->fs->umask = 0022;
  81
  82         /*
  83          * Our parent (unbound workqueue) runs with elevated scheduling
  84          * priority. Avoid propagating that into the userspace child.
  85          */
  86         set_user_nice(current, 0);
  87
  88         retval = -ENOMEM;
  89         new = prepare_kernel_cred(current);
  90         if (!new)
  91                 goto out;
  92
  93         spin_lock(&umh_sysctl_lock);
  94         new->cap_bset = cap_intersect(usermodehelper_bset, new->cap_bset);
  95         new->cap_inheritable = cap_intersect(usermodehelper_inheritable,
  96                                              new->cap_inheritable);
  97         spin_unlock(&umh_sysctl_lock);
  98
  99         if (sub_info->init) {
 100                 retval = sub_info->init(sub_info, new);
 101                 if (retval) {
 102                         abort_creds(new);
 103                         goto out;
 104                 }
 105         }
 106
 107         commit_creds(new);
 108
 109         retval = do_execve(getname_kernel(sub_info->path),
 110                            (const char __user *const __user *)sub_info->argv,
 111                            (const char __user *const __user *)sub_info->envp);
 112 out:
 113         sub_info->retval = retval;
 114         /*
 115          * call_usermodehelper_exec_sync() will call umh_complete
 116          * if UHM_WAIT_PROC.
 117          */
 118         if (!(sub_info->wait & UMH_WAIT_PROC))
 119                 umh_complete(sub_info);
 120         if (!retval)
 121                 return 0;
 122         do_exit(0);
 123 }
 124
 125 /* Handles UMH_WAIT_PROC.  */
 126 static void call_usermodehelper_exec_sync(struct subprocess_info *sub_info)
 127 {
 128         pid_t pid;
 129
 130         /* If SIGCLD is ignored sys_wait4 won't populate the status. */
 131         kernel_sigaction(SIGCHLD, SIG_DFL);
 132         pid = kernel_thread(call_usermodehelper_exec_async, sub_info, SIGCHLD);
 133         if (pid < 0) {
 134                 sub_info->retval = pid;
 135         } else {
 136                 int ret = -ECHILD;
 137                 /*
 138                  * Normally it is bogus to call wait4() from in-kernel because
 139                  * wait4() wants to write the exit code to a userspace address.
 140                  * But call_usermodehelper_exec_sync() always runs as kernel
 141                  * thread (workqueue) and put_user() to a kernel address works
 142                  * OK for kernel threads, due to their having an mm_segment_t
 143                  * which spans the entire address space.
 144                  *
 145                  * Thus the __user pointer cast is valid here.
 146                  */
 147                 sys_wait4(pid, (int __user *)&ret, 0, NULL);
 148
 149                 /*
 150                  * If ret is 0, either call_usermodehelper_exec_async failed and
 151                  * the real error code is already in sub_info->retval or
 152                  * sub_info->retval is 0 anyway, so don't mess with it then.
 153                  */
 154                 if (ret)
 155                         sub_info->retval = ret;
 156         }
 157
 158         /* Restore default kernel sig handler */
 159         kernel_sigaction(SIGCHLD, SIG_IGN);
 160
 161         umh_complete(sub_info);
 162 }
 163
 164 /*
 165  * We need to create the usermodehelper kernel thread from a task that is affine
 166  * to an optimized set of CPUs (or nohz housekeeping ones) such that they
 167  * inherit a widest affinity irrespective of call_usermodehelper() callers with
 168  * possibly reduced affinity (eg: per-cpu workqueues). We don't want
 169  * usermodehelper targets to contend a busy CPU.
 170  *
 171  * Unbound workqueues provide such wide affinity and allow to block on
 172  * UMH_WAIT_PROC requests without blocking pending request (up to some limit).
 173  *
 174  * Besides, workqueues provide the privilege level that caller might not have
 175  * to perform the usermodehelper request.
 176  *
 177  */
 178 static void call_usermodehelper_exec_work(struct work_struct *work)
 179 {
 180         struct subprocess_info *sub_info =
 181                 container_of(work, struct subprocess_info, work);
 182
 183         if (sub_info->wait & UMH_WAIT_PROC) {
 184                 call_usermodehelper_exec_sync(sub_info);
 185         } else {
 186                 pid_t pid;
 187                 /*
 188                  * Use CLONE_PARENT to reparent it to kthreadd; we do not
 189                  * want to pollute current->children, and we need a parent
 190                  * that always ignores SIGCHLD to ensure auto-reaping.
 191                  */
 192                 pid = kernel_thread(call_usermodehelper_exec_async, sub_info,
 193                                     CLONE_PARENT | SIGCHLD);
 194                 if (pid < 0) {
 195                         sub_info->retval = pid;
 196                         umh_complete(sub_info);
 197                 }
 198         }
 199 }
 200
 201 /*
 202  * If set, call_usermodehelper_exec() will exit immediately returning -EBUSY
 203  * (used for preventing user land processes from being created after the user
 204  * land has been frozen during a system-wide hibernation or suspend operation).
 205  * Should always be manipulated under umhelper_sem acquired for write.
 206  */
 207 static enum umh_disable_depth usermodehelper_disabled = UMH_DISABLED;
 208
 209 /* Number of helpers running */
 210 static atomic_t running_helpers = ATOMIC_INIT(0);
 211
 212 /*
 213  * Wait queue head used by usermodehelper_disable() to wait for all running
 214  * helpers to finish.
 215  */
 216 static DECLARE_WAIT_QUEUE_HEAD(running_helpers_waitq);
 217
 218 /*
 219  * Used by usermodehelper_read_lock_wait() to wait for usermodehelper_disabled
 220  * to become 'false'.
 221  */
 222 static DECLARE_WAIT_QUEUE_HEAD(usermodehelper_disabled_waitq);
 223
 224 /*
 225  * Time to wait for running_helpers to become zero before the setting of
 226  * usermodehelper_disabled in usermodehelper_disable() fails
 227  */
 228 #define RUNNING_HELPERS_TIMEOUT (5 * HZ)
 229
 230 int usermodehelper_read_trylock(void)
 231 {
 232         DEFINE_WAIT(wait);
 233         int ret = 0;
 234
 235         down_read(&umhelper_sem);
 236         for (;;) {
 237                 prepare_to_wait(&usermodehelper_disabled_waitq, &wait,
 238                                 TASK_INTERRUPTIBLE);
 239                 if (!usermodehelper_disabled)
 240                         break;
 241
 242                 if (usermodehelper_disabled == UMH_DISABLED)
 243                         ret = -EAGAIN;
 244
 245                 up_read(&umhelper_sem);
 246
 247                 if (ret)
 248                         break;
 249
 250                 schedule();
 251                 try_to_freeze();
 252
 253                 down_read(&umhelper_sem);
 254         }
 255         finish_wait(&usermodehelper_disabled_waitq, &wait);
 256         return ret;
 257 }
 258 EXPORT_SYMBOL_GPL(usermodehelper_read_trylock);
 259
 260 long usermodehelper_read_lock_wait(long timeout)
 261 {
 262         DEFINE_WAIT(wait);
 263
 264         if (timeout < 0)
 265                 return -EINVAL;
 266
 267         down_read(&umhelper_sem);
 268         for (;;) {
 269                 prepare_to_wait(&usermodehelper_disabled_waitq, &wait,
 270                                 TASK_UNINTERRUPTIBLE);
 271                 if (!usermodehelper_disabled)
 272                         break;
 273
 274                 up_read(&umhelper_sem);
 275
 276                 timeout = schedule_timeout(timeout);
 277                 if (!timeout)
 278                         break;
 279
 280                 down_read(&umhelper_sem);
 281         }
 282         finish_wait(&usermodehelper_disabled_waitq, &wait);
 283         return timeout;
 284 }
 285 EXPORT_SYMBOL_GPL(usermodehelper_read_lock_wait);
 286
 287 void usermodehelper_read_unlock(void)
 288 {
 289         up_read(&umhelper_sem);
 290 }
 291 EXPORT_SYMBOL_GPL(usermodehelper_read_unlock);
 292
 293 /**
 294  * __usermodehelper_set_disable_depth - Modify usermodehelper_disabled.
 295  * @depth: New value to assign to usermodehelper_disabled.
 296  *
 297  * Change the value of usermodehelper_disabled (under umhelper_sem locked for
 298  * writing) and wakeup tasks waiting for it to change.
 299  */
 300 void __usermodehelper_set_disable_depth(enum umh_disable_depth depth)
 301 {
 302         down_write(&umhelper_sem);
 303         usermodehelper_disabled = depth;
 304         wake_up(&usermodehelper_disabled_waitq);
 305         up_write(&umhelper_sem);
 306 }
 307
 308 /**
 309  * __usermodehelper_disable - Prevent new helpers from being started.
 310  * @depth: New value to assign to usermodehelper_disabled.
 311  *
 312  * Set usermodehelper_disabled to @depth and wait for running helpers to exit.
 313  */
 314 int __usermodehelper_disable(enum umh_disable_depth depth)
 315 {
 316         long retval;
 317
 318         if (!depth)
 319                 return -EINVAL;
 320
 321         down_write(&umhelper_sem);
 322         usermodehelper_disabled = depth;
 323         up_write(&umhelper_sem);
 324
 325         /*
 326          * From now on call_usermodehelper_exec() won't start any new
 327          * helpers, so it is sufficient if running_helpers turns out to
 328          * be zero at one point (it may be increased later, but that
 329          * doesn't matter).
 330          */
 331         retval = wait_event_timeout(running_helpers_waitq,
 332                                         atomic_read(&running_helpers) == 0,
 333                                         RUNNING_HELPERS_TIMEOUT);
 334         if (retval)
 335                 return 0;
 336
 337         __usermodehelper_set_disable_depth(UMH_ENABLED);
 338         return -EAGAIN;
 339 }
 340
 341 static void helper_lock(void)
 342 {
 343         atomic_inc(&running_helpers);
 344         smp_mb__after_atomic();
 345 }
 346
 347 static void helper_unlock(void)
 348 {
 349         if (atomic_dec_and_test(&running_helpers))
 350                 wake_up(&running_helpers_waitq);
 351 }
 352
 353 /**
 354  * call_usermodehelper_setup - prepare to call a usermode helper
 355  * @path: path to usermode executable
 356  * @argv: arg vector for process
 357  * @envp: environment for process
 358  * @gfp_mask: gfp mask for memory allocation
 359  * @cleanup: a cleanup function
 360  * @init: an init function
 361  * @data: arbitrary context sensitive data
 362  *
 363  * Returns either %NULL on allocation failure, or a subprocess_info
 364  * structure.  This should be passed to call_usermodehelper_exec to
 365  * exec the process and free the structure.
 366  *
 367  * The init function is used to customize the helper process prior to
 368  * exec.  A non-zero return code causes the process to error out, exit,
 369  * and return the failure to the calling process
 370  *
 371  * The cleanup function is just before ethe subprocess_info is about to
 372  * be freed.  This can be used for freeing the argv and envp.  The
 373  * Function must be runnable in either a process context or the
 374  * context in which call_usermodehelper_exec is called.
 375  */
 376 struct subprocess_info *call_usermodehelper_setup(const char *path, char **argv,
 377                 char **envp, gfp_t gfp_mask,
 378                 int (*init)(struct subprocess_info *info, struct cred *new),
 379                 void (*cleanup)(struct subprocess_info *info),
 380                 void *data)
 381 {
 382         struct subprocess_info *sub_info;
 383         sub_info = kzalloc(sizeof(struct subprocess_info), gfp_mask);
 384         if (!sub_info)
 385                 goto out;
 386
 387         INIT_WORK(&sub_info->work, call_usermodehelper_exec_work);
 388
 389 #ifdef CONFIG_STATIC_USERMODEHELPER
 390         sub_info->path = CONFIG_STATIC_USERMODEHELPER_PATH;
 391 #else
 392         sub_info->path = path;
 393 #endif
 394         sub_info->argv = argv;
 395         sub_info->envp = envp;
 396
 397         sub_info->cleanup = cleanup;
 398         sub_info->init = init;
 399         sub_info->data = data;
 400   out:
 401         return sub_info;
 402 }
 403 EXPORT_SYMBOL(call_usermodehelper_setup);
 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 UMH_NO_WAIT don't wait at all, but you get no useful error back
 410  *        when 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 system workqueues.
 415  * (ie. it runs with full root capabilities and optimized affinity).
 416  *
 417  * Note: successful return value does not guarantee the helper was called at
 418  * all. You can't rely on sub_info->{init,cleanup} being called even for
 419  * UMH_WAIT_* wait modes as STATIC_USERMODEHELPER_PATH="" turns all helpers
 420  * into a successful no-op.
 421  */
 422 int call_usermodehelper_exec(struct subprocess_info *sub_info, int wait)
 423 {
 424         DECLARE_COMPLETION_ONSTACK(done);
 425         int retval = 0;
 426
 427         if (!sub_info->path) {
 428                 call_usermodehelper_freeinfo(sub_info);
 429                 return -EINVAL;
 430         }
 431         helper_lock();
 432         if (usermodehelper_disabled) {
 433                 retval = -EBUSY;
 434                 goto out;
 435         }
 436
 437         /*
 438          * If there is no binary for us to call, then just return and get out of
 439          * here.  This allows us to set STATIC_USERMODEHELPER_PATH to "" and
 440          * disable all call_usermodehelper() calls.
 441          */
 442         if (strlen(sub_info->path) == 0)
 443                 goto out;
 444
 445         /*
 446          * Set the completion pointer only if there is a waiter.
 447          * This makes it possible to use umh_complete to free
 448          * the data structure in case of UMH_NO_WAIT.
 449          */
 450         sub_info->complete = (wait == UMH_NO_WAIT) ? NULL : &done;
 451         sub_info->wait = wait;
 452
 453         queue_work(system_unbound_wq, &sub_info->work);
 454         if (wait == UMH_NO_WAIT)        /* task has freed sub_info */
 455                 goto unlock;
 456
 457         if (wait & UMH_KILLABLE) {
 458                 retval = wait_for_completion_killable(&done);
 459                 if (!retval)
 460                         goto wait_done;
 461
 462                 /* umh_complete() will see NULL and free sub_info */
 463                 if (xchg(&sub_info->complete, NULL))
 464                         goto unlock;
 465                 /* fallthrough, umh_complete() was already called */
 466         }
 467
 468         wait_for_completion(&done);
 469 wait_done:
 470         retval = sub_info->retval;
 471 out:
 472         call_usermodehelper_freeinfo(sub_info);
 473 unlock:
 474         helper_unlock();
 475         return retval;
 476 }
 477 EXPORT_SYMBOL(call_usermodehelper_exec);
 478
 479 /**
 480  * call_usermodehelper() - prepare and start a usermode application
 481  * @path: path to usermode executable
 482  * @argv: arg vector for process
 483  * @envp: environment for process
 484  * @wait: wait for the application to finish and return status.
 485  *        when UMH_NO_WAIT don't wait at all, but you get no useful error back
 486  *        when the program couldn't be exec'ed. This makes it safe to call
 487  *        from interrupt context.
 488  *
 489  * This function is the equivalent to use call_usermodehelper_setup() and
 490  * call_usermodehelper_exec().
 491  */
 492 int call_usermodehelper(const char *path, char **argv, char **envp, int wait)
 493 {
 494         struct subprocess_info *info;
 495         gfp_t gfp_mask = (wait == UMH_NO_WAIT) ? GFP_ATOMIC : GFP_KERNEL;
 496
 497         info = call_usermodehelper_setup(path, argv, envp, gfp_mask,
 498                                          NULL, NULL, NULL);
 499         if (info == NULL)
 500                 return -ENOMEM;
 501
 502         return call_usermodehelper_exec(info, wait);
 503 }
 504 EXPORT_SYMBOL(call_usermodehelper);
 505
 506 static int proc_cap_handler(struct ctl_table *table, int write,
 507                          void __user *buffer, size_t *lenp, loff_t *ppos)
 508 {
 509         struct ctl_table t;
 510         unsigned long cap_array[_KERNEL_CAPABILITY_U32S];
 511         kernel_cap_t new_cap;
 512         int err, i;
 513
 514         if (write && (!capable(CAP_SETPCAP) ||
 515                       !capable(CAP_SYS_MODULE)))
 516                 return -EPERM;
 517
 518         /*
 519          * convert from the global kernel_cap_t to the ulong array to print to
 520          * userspace if this is a read.
 521          */
 522         spin_lock(&umh_sysctl_lock);
 523         for (i = 0; i < _KERNEL_CAPABILITY_U32S; i++)  {
 524                 if (table->data == CAP_BSET)
 525                         cap_array[i] = usermodehelper_bset.cap[i];
 526                 else if (table->data == CAP_PI)
 527                         cap_array[i] = usermodehelper_inheritable.cap[i];
 528                 else
 529                         BUG();
 530         }
 531         spin_unlock(&umh_sysctl_lock);
 532
 533         t = *table;
 534         t.data = &cap_array;
 535
 536         /*
 537          * actually read or write and array of ulongs from userspace.  Remember
 538          * these are least significant 32 bits first
 539          */
 540         err = proc_doulongvec_minmax(&t, write, buffer, lenp, ppos);
 541         if (err < 0)
 542                 return err;
 543
 544         /*
 545          * convert from the sysctl array of ulongs to the kernel_cap_t
 546          * internal representation
 547          */
 548         for (i = 0; i < _KERNEL_CAPABILITY_U32S; i++)
 549                 new_cap.cap[i] = cap_array[i];
 550
 551         /*
 552          * Drop everything not in the new_cap (but don't add things)
 553          */
 554         spin_lock(&umh_sysctl_lock);
 555         if (write) {
 556                 if (table->data == CAP_BSET)
 557                         usermodehelper_bset = cap_intersect(usermodehelper_bset, new_cap);
 558                 if (table->data == CAP_PI)
 559                         usermodehelper_inheritable = cap_intersect(usermodehelper_inheritable, new_cap);
 560         }
 561         spin_unlock(&umh_sysctl_lock);
 562
 563         return 0;
 564 }
 565
 566 struct ctl_table usermodehelper_table[] = {
 567         {
 568                 .procname       = "bset",
 569                 .data           = CAP_BSET,
 570                 .maxlen         = _KERNEL_CAPABILITY_U32S * sizeof(unsigned long),
 571                 .mode           = 0600,
 572                 .proc_handler   = proc_cap_handler,
 573         },
 574         {
 575                 .procname       = "inheritable",
 576                 .data           = CAP_PI,
 577                 .maxlen         = _KERNEL_CAPABILITY_U32S * sizeof(unsigned long),
 578                 .mode           = 0600,
 579                 .proc_handler   = proc_cap_handler,
 580         },
 581         { }
 582 };