kernel/umh.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * umh - the kernel usermode helper
   4  */
   5 #include <linux/module.h>
   6 #include <linux/sched.h>
   7 #include <linux/sched/task.h>
   8 #include <linux/binfmts.h>
   9 #include <linux/syscalls.h>
  10 #include <linux/unistd.h>
  11 #include <linux/kmod.h>
  12 #include <linux/slab.h>
  13 #include <linux/completion.h>
  14 #include <linux/cred.h>
  15 #include <linux/file.h>
  16 #include <linux/fdtable.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          * Our parent (unbound workqueue) runs with elevated scheduling
  76          * priority. Avoid propagating that into the userspace child.
  77          */
  78         set_user_nice(current, 0);
  79
  80         retval = -ENOMEM;
  81         new = prepare_kernel_cred(current);
  82         if (!new)
  83                 goto out;
  84
  85         spin_lock(&umh_sysctl_lock);
  86         new->cap_bset = cap_intersect(usermodehelper_bset, new->cap_bset);
  87         new->cap_inheritable = cap_intersect(usermodehelper_inheritable,
  88                                              new->cap_inheritable);
  89         spin_unlock(&umh_sysctl_lock);
  90
  91         if (sub_info->init) {
  92                 retval = sub_info->init(sub_info, new);
  93                 if (retval) {
  94                         abort_creds(new);
  95                         goto out;
  96                 }
  97         }
  98
  99         commit_creds(new);
 100
 101         retval = kernel_execve(sub_info->path,
 102                                (const char *const *)sub_info->argv,
 103                                (const char *const *)sub_info->envp);
 104 out:
 105         sub_info->retval = retval;
 106         /*
 107          * call_usermodehelper_exec_sync() will call umh_complete
 108          * if UHM_WAIT_PROC.
 109          */
 110         if (!(sub_info->wait & UMH_WAIT_PROC))
 111                 umh_complete(sub_info);
 112         if (!retval)
 113                 return 0;
 114         do_exit(0);
 115 }
 116
 117 /* Handles UMH_WAIT_PROC.  */
 118 static void call_usermodehelper_exec_sync(struct subprocess_info *sub_info)
 119 {
 120         pid_t pid;
 121
 122         /* If SIGCLD is ignored kernel_wait4 won't populate the status. */
 123         kernel_sigaction(SIGCHLD, SIG_DFL);
 124         pid = kernel_thread(call_usermodehelper_exec_async, sub_info, SIGCHLD);
 125         if (pid < 0) {
 126                 sub_info->retval = pid;
 127         } else {
 128                 int ret = -ECHILD;
 129                 /*
 130                  * Normally it is bogus to call wait4() from in-kernel because
 131                  * wait4() wants to write the exit code to a userspace address.
 132                  * But call_usermodehelper_exec_sync() always runs as kernel
 133                  * thread (workqueue) and put_user() to a kernel address works
 134                  * OK for kernel threads, due to their having an mm_segment_t
 135                  * which spans the entire address space.
 136                  *
 137                  * Thus the __user pointer cast is valid here.
 138                  */
 139                 kernel_wait4(pid, (int __user *)&ret, 0, NULL);
 140
 141                 /*
 142                  * If ret is 0, either call_usermodehelper_exec_async failed and
 143                  * the real error code is already in sub_info->retval or
 144                  * sub_info->retval is 0 anyway, so don't mess with it then.
 145                  */
 146                 if (ret)
 147                         sub_info->retval = ret;
 148         }
 149
 150         /* Restore default kernel sig handler */
 151         kernel_sigaction(SIGCHLD, SIG_IGN);
 152
 153         umh_complete(sub_info);
 154 }
 155
 156 /*
 157  * We need to create the usermodehelper kernel thread from a task that is affine
 158  * to an optimized set of CPUs (or nohz housekeeping ones) such that they
 159  * inherit a widest affinity irrespective of call_usermodehelper() callers with
 160  * possibly reduced affinity (eg: per-cpu workqueues). We don't want
 161  * usermodehelper targets to contend a busy CPU.
 162  *
 163  * Unbound workqueues provide such wide affinity and allow to block on
 164  * UMH_WAIT_PROC requests without blocking pending request (up to some limit).
 165  *
 166  * Besides, workqueues provide the privilege level that caller might not have
 167  * to perform the usermodehelper request.
 168  *
 169  */
 170 static void call_usermodehelper_exec_work(struct work_struct *work)
 171 {
 172         struct subprocess_info *sub_info =
 173                 container_of(work, struct subprocess_info, work);
 174
 175         if (sub_info->wait & UMH_WAIT_PROC) {
 176                 call_usermodehelper_exec_sync(sub_info);
 177         } else {
 178                 pid_t pid;
 179                 /*
 180                  * Use CLONE_PARENT to reparent it to kthreadd; we do not
 181                  * want to pollute current->children, and we need a parent
 182                  * that always ignores SIGCHLD to ensure auto-reaping.
 183                  */
 184                 pid = kernel_thread(call_usermodehelper_exec_async, sub_info,
 185                                     CLONE_PARENT | SIGCHLD);
 186                 if (pid < 0) {
 187                         sub_info->retval = pid;
 188                         umh_complete(sub_info);
 189                 }
 190         }
 191 }
 192
 193 /*
 194  * If set, call_usermodehelper_exec() will exit immediately returning -EBUSY
 195  * (used for preventing user land processes from being created after the user
 196  * land has been frozen during a system-wide hibernation or suspend operation).
 197  * Should always be manipulated under umhelper_sem acquired for write.
 198  */
 199 static enum umh_disable_depth usermodehelper_disabled = UMH_DISABLED;
 200
 201 /* Number of helpers running */
 202 static atomic_t running_helpers = ATOMIC_INIT(0);
 203
 204 /*
 205  * Wait queue head used by usermodehelper_disable() to wait for all running
 206  * helpers to finish.
 207  */
 208 static DECLARE_WAIT_QUEUE_HEAD(running_helpers_waitq);
 209
 210 /*
 211  * Used by usermodehelper_read_lock_wait() to wait for usermodehelper_disabled
 212  * to become 'false'.
 213  */
 214 static DECLARE_WAIT_QUEUE_HEAD(usermodehelper_disabled_waitq);
 215
 216 /*
 217  * Time to wait for running_helpers to become zero before the setting of
 218  * usermodehelper_disabled in usermodehelper_disable() fails
 219  */
 220 #define RUNNING_HELPERS_TIMEOUT (5 * HZ)
 221
 222 int usermodehelper_read_trylock(void)
 223 {
 224         DEFINE_WAIT(wait);
 225         int ret = 0;
 226
 227         down_read(&umhelper_sem);
 228         for (;;) {
 229                 prepare_to_wait(&usermodehelper_disabled_waitq, &wait,
 230                                 TASK_INTERRUPTIBLE);
 231                 if (!usermodehelper_disabled)
 232                         break;
 233
 234                 if (usermodehelper_disabled == UMH_DISABLED)
 235                         ret = -EAGAIN;
 236
 237                 up_read(&umhelper_sem);
 238
 239                 if (ret)
 240                         break;
 241
 242                 schedule();
 243                 try_to_freeze();
 244
 245                 down_read(&umhelper_sem);
 246         }
 247         finish_wait(&usermodehelper_disabled_waitq, &wait);
 248         return ret;
 249 }
 250 EXPORT_SYMBOL_GPL(usermodehelper_read_trylock);
 251
 252 long usermodehelper_read_lock_wait(long timeout)
 253 {
 254         DEFINE_WAIT(wait);
 255
 256         if (timeout < 0)
 257                 return -EINVAL;
 258
 259         down_read(&umhelper_sem);
 260         for (;;) {
 261                 prepare_to_wait(&usermodehelper_disabled_waitq, &wait,
 262                                 TASK_UNINTERRUPTIBLE);
 263                 if (!usermodehelper_disabled)
 264                         break;
 265
 266                 up_read(&umhelper_sem);
 267
 268                 timeout = schedule_timeout(timeout);
 269                 if (!timeout)
 270                         break;
 271
 272                 down_read(&umhelper_sem);
 273         }
 274         finish_wait(&usermodehelper_disabled_waitq, &wait);
 275         return timeout;
 276 }
 277 EXPORT_SYMBOL_GPL(usermodehelper_read_lock_wait);
 278
 279 void usermodehelper_read_unlock(void)
 280 {
 281         up_read(&umhelper_sem);
 282 }
 283 EXPORT_SYMBOL_GPL(usermodehelper_read_unlock);
 284
 285 /**
 286  * __usermodehelper_set_disable_depth - Modify usermodehelper_disabled.
 287  * @depth: New value to assign to usermodehelper_disabled.
 288  *
 289  * Change the value of usermodehelper_disabled (under umhelper_sem locked for
 290  * writing) and wakeup tasks waiting for it to change.
 291  */
 292 void __usermodehelper_set_disable_depth(enum umh_disable_depth depth)
 293 {
 294         down_write(&umhelper_sem);
 295         usermodehelper_disabled = depth;
 296         wake_up(&usermodehelper_disabled_waitq);
 297         up_write(&umhelper_sem);
 298 }
 299
 300 /**
 301  * __usermodehelper_disable - Prevent new helpers from being started.
 302  * @depth: New value to assign to usermodehelper_disabled.
 303  *
 304  * Set usermodehelper_disabled to @depth and wait for running helpers to exit.
 305  */
 306 int __usermodehelper_disable(enum umh_disable_depth depth)
 307 {
 308         long retval;
 309
 310         if (!depth)
 311                 return -EINVAL;
 312
 313         down_write(&umhelper_sem);
 314         usermodehelper_disabled = depth;
 315         up_write(&umhelper_sem);
 316
 317         /*
 318          * From now on call_usermodehelper_exec() won't start any new
 319          * helpers, so it is sufficient if running_helpers turns out to
 320          * be zero at one point (it may be increased later, but that
 321          * doesn't matter).
 322          */
 323         retval = wait_event_timeout(running_helpers_waitq,
 324                                         atomic_read(&running_helpers) == 0,
 325                                         RUNNING_HELPERS_TIMEOUT);
 326         if (retval)
 327                 return 0;
 328
 329         __usermodehelper_set_disable_depth(UMH_ENABLED);
 330         return -EAGAIN;
 331 }
 332
 333 static void helper_lock(void)
 334 {
 335         atomic_inc(&running_helpers);
 336         smp_mb__after_atomic();
 337 }
 338
 339 static void helper_unlock(void)
 340 {
 341         if (atomic_dec_and_test(&running_helpers))
 342                 wake_up(&running_helpers_waitq);
 343 }
 344
 345 /**
 346  * call_usermodehelper_setup - prepare to call a usermode helper
 347  * @path: path to usermode executable
 348  * @argv: arg vector for process
 349  * @envp: environment for process
 350  * @gfp_mask: gfp mask for memory allocation
 351  * @cleanup: a cleanup function
 352  * @init: an init function
 353  * @data: arbitrary context sensitive data
 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  * The init function is used to customize the helper process prior to
 360  * exec.  A non-zero return code causes the process to error out, exit,
 361  * and return the failure to the calling process
 362  *
 363  * The cleanup function is just before ethe subprocess_info is about to
 364  * be freed.  This can be used for freeing the argv and envp.  The
 365  * Function must be runnable in either a process context or the
 366  * context in which call_usermodehelper_exec is called.
 367  */
 368 struct subprocess_info *call_usermodehelper_setup(const char *path, char **argv,
 369                 char **envp, gfp_t gfp_mask,
 370                 int (*init)(struct subprocess_info *info, struct cred *new),
 371                 void (*cleanup)(struct subprocess_info *info),
 372                 void *data)
 373 {
 374         struct subprocess_info *sub_info;
 375         sub_info = kzalloc(sizeof(struct subprocess_info), gfp_mask);
 376         if (!sub_info)
 377                 goto out;
 378
 379         INIT_WORK(&sub_info->work, call_usermodehelper_exec_work);
 380
 381 #ifdef CONFIG_STATIC_USERMODEHELPER
 382         sub_info->path = CONFIG_STATIC_USERMODEHELPER_PATH;
 383 #else
 384         sub_info->path = path;
 385 #endif
 386         sub_info->argv = argv;
 387         sub_info->envp = envp;
 388
 389         sub_info->cleanup = cleanup;
 390         sub_info->init = init;
 391         sub_info->data = data;
 392   out:
 393         return sub_info;
 394 }
 395 EXPORT_SYMBOL(call_usermodehelper_setup);
 396
 397 /**
 398  * call_usermodehelper_exec - start a usermode application
 399  * @sub_info: information about the subprocessa
 400  * @wait: wait for the application to finish and return status.
 401  *        when UMH_NO_WAIT don't wait at all, but you get no useful error back
 402  *        when the program couldn't be exec'ed. This makes it safe to call
 403  *        from interrupt context.
 404  *
 405  * Runs a user-space application.  The application is started
 406  * asynchronously if wait is not set, and runs as a child of system workqueues.
 407  * (ie. it runs with full root capabilities and optimized affinity).
 408  *
 409  * Note: successful return value does not guarantee the helper was called at
 410  * all. You can't rely on sub_info->{init,cleanup} being called even for
 411  * UMH_WAIT_* wait modes as STATIC_USERMODEHELPER_PATH="" turns all helpers
 412  * into a successful no-op.
 413  */
 414 int call_usermodehelper_exec(struct subprocess_info *sub_info, int wait)
 415 {
 416         DECLARE_COMPLETION_ONSTACK(done);
 417         int retval = 0;
 418
 419         if (!sub_info->path) {
 420                 call_usermodehelper_freeinfo(sub_info);
 421                 return -EINVAL;
 422         }
 423         helper_lock();
 424         if (usermodehelper_disabled) {
 425                 retval = -EBUSY;
 426                 goto out;
 427         }
 428
 429         /*
 430          * If there is no binary for us to call, then just return and get out of
 431          * here.  This allows us to set STATIC_USERMODEHELPER_PATH to "" and
 432          * disable all call_usermodehelper() calls.
 433          */
 434         if (strlen(sub_info->path) == 0)
 435                 goto out;
 436
 437         /*
 438          * Set the completion pointer only if there is a waiter.
 439          * This makes it possible to use umh_complete to free
 440          * the data structure in case of UMH_NO_WAIT.
 441          */
 442         sub_info->complete = (wait == UMH_NO_WAIT) ? NULL : &done;
 443         sub_info->wait = wait;
 444
 445         queue_work(system_unbound_wq, &sub_info->work);
 446         if (wait == UMH_NO_WAIT)        /* task has freed sub_info */
 447                 goto unlock;
 448
 449         if (wait & UMH_KILLABLE) {
 450                 retval = wait_for_completion_killable(&done);
 451                 if (!retval)
 452                         goto wait_done;
 453
 454                 /* umh_complete() will see NULL and free sub_info */
 455                 if (xchg(&sub_info->complete, NULL))
 456                         goto unlock;
 457                 /* fallthrough, umh_complete() was already called */
 458         }
 459
 460         wait_for_completion(&done);
 461 wait_done:
 462         retval = sub_info->retval;
 463 out:
 464         call_usermodehelper_freeinfo(sub_info);
 465 unlock:
 466         helper_unlock();
 467         return retval;
 468 }
 469 EXPORT_SYMBOL(call_usermodehelper_exec);
 470
 471 /**
 472  * call_usermodehelper() - prepare and start a usermode application
 473  * @path: path to usermode executable
 474  * @argv: arg vector for process
 475  * @envp: environment for process
 476  * @wait: wait for the application to finish and return status.
 477  *        when UMH_NO_WAIT don't wait at all, but you get no useful error back
 478  *        when the program couldn't be exec'ed. This makes it safe to call
 479  *        from interrupt context.
 480  *
 481  * This function is the equivalent to use call_usermodehelper_setup() and
 482  * call_usermodehelper_exec().
 483  */
 484 int call_usermodehelper(const char *path, char **argv, char **envp, int wait)
 485 {
 486         struct subprocess_info *info;
 487         gfp_t gfp_mask = (wait == UMH_NO_WAIT) ? GFP_ATOMIC : GFP_KERNEL;
 488
 489         info = call_usermodehelper_setup(path, argv, envp, gfp_mask,
 490                                          NULL, NULL, NULL);
 491         if (info == NULL)
 492                 return -ENOMEM;
 493
 494         return call_usermodehelper_exec(info, wait);
 495 }
 496 EXPORT_SYMBOL(call_usermodehelper);
 497
 498 static int proc_cap_handler(struct ctl_table *table, int write,
 499                          void *buffer, size_t *lenp, loff_t *ppos)
 500 {
 501         struct ctl_table t;
 502         unsigned long cap_array[_KERNEL_CAPABILITY_U32S];
 503         kernel_cap_t new_cap;
 504         int err, i;
 505
 506         if (write && (!capable(CAP_SETPCAP) ||
 507                       !capable(CAP_SYS_MODULE)))
 508                 return -EPERM;
 509
 510         /*
 511          * convert from the global kernel_cap_t to the ulong array to print to
 512          * userspace if this is a read.
 513          */
 514         spin_lock(&umh_sysctl_lock);
 515         for (i = 0; i < _KERNEL_CAPABILITY_U32S; i++)  {
 516                 if (table->data == CAP_BSET)
 517                         cap_array[i] = usermodehelper_bset.cap[i];
 518                 else if (table->data == CAP_PI)
 519                         cap_array[i] = usermodehelper_inheritable.cap[i];
 520                 else
 521                         BUG();
 522         }
 523         spin_unlock(&umh_sysctl_lock);
 524
 525         t = *table;
 526         t.data = &cap_array;
 527
 528         /*
 529          * actually read or write and array of ulongs from userspace.  Remember
 530          * these are least significant 32 bits first
 531          */
 532         err = proc_doulongvec_minmax(&t, write, buffer, lenp, ppos);
 533         if (err < 0)
 534                 return err;
 535
 536         /*
 537          * convert from the sysctl array of ulongs to the kernel_cap_t
 538          * internal representation
 539          */
 540         for (i = 0; i < _KERNEL_CAPABILITY_U32S; i++)
 541                 new_cap.cap[i] = cap_array[i];
 542
 543         /*
 544          * Drop everything not in the new_cap (but don't add things)
 545          */
 546         if (write) {
 547                 spin_lock(&umh_sysctl_lock);
 548                 if (table->data == CAP_BSET)
 549                         usermodehelper_bset = cap_intersect(usermodehelper_bset, new_cap);
 550                 if (table->data == CAP_PI)
 551                         usermodehelper_inheritable = cap_intersect(usermodehelper_inheritable, new_cap);
 552                 spin_unlock(&umh_sysctl_lock);
 553         }
 554
 555         return 0;
 556 }
 557
 558 struct ctl_table usermodehelper_table[] = {
 559         {
 560                 .procname       = "bset",
 561                 .data           = CAP_BSET,
 562                 .maxlen         = _KERNEL_CAPABILITY_U32S * sizeof(unsigned long),
 563                 .mode           = 0600,
 564                 .proc_handler   = proc_cap_handler,
 565         },
 566         {
 567                 .procname       = "inheritable",
 568                 .data           = CAP_PI,
 569                 .maxlen         = _KERNEL_CAPABILITY_U32S * sizeof(unsigned long),
 570                 .mode           = 0600,
 571                 .proc_handler   = proc_cap_handler,
 572         },
 573         { }
 574 };