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/fs_struct.h>
  18 #include <linux/workqueue.h>
  19 #include <linux/security.h>
  20 #include <linux/mount.h>
  21 #include <linux/kernel.h>
  22 #include <linux/init.h>
  23 #include <linux/resource.h>
  24 #include <linux/notifier.h>
  25 #include <linux/suspend.h>
  26 #include <linux/rwsem.h>
  27 #include <linux/ptrace.h>
  28 #include <linux/async.h>
  29 #include <linux/uaccess.h>
  30 #include <linux/initrd.h>
  31 #include <linux/freezer.h>
  32
  33 #include <trace/events/module.h>
  34
  35 #define CAP_BSET        (void *)1
  36 #define CAP_PI          (void *)2
  37
  38 static kernel_cap_t usermodehelper_bset = CAP_FULL_SET;
  39 static kernel_cap_t usermodehelper_inheritable = CAP_FULL_SET;
  40 static DEFINE_SPINLOCK(umh_sysctl_lock);
  41 static DECLARE_RWSEM(umhelper_sem);
  42
  43 static void call_usermodehelper_freeinfo(struct subprocess_info *info)
  44 {
  45         if (info->cleanup)
  46                 (*info->cleanup)(info);
  47         kfree(info);
  48 }
  49
  50 static void umh_complete(struct subprocess_info *sub_info)
  51 {
  52         struct completion *comp = xchg(&sub_info->complete, NULL);
  53         /*
  54          * See call_usermodehelper_exec(). If xchg() returns NULL
  55          * we own sub_info, the UMH_KILLABLE caller has gone away
  56          * or the caller used UMH_NO_WAIT.
  57          */
  58         if (comp)
  59                 complete(comp);
  60         else
  61                 call_usermodehelper_freeinfo(sub_info);
  62 }
  63
  64 /*
  65  * This is the task which runs the usermode application
  66  */
  67 static int call_usermodehelper_exec_async(void *data)
  68 {
  69         struct subprocess_info *sub_info = data;
  70         struct cred *new;
  71         int retval;
  72
  73         spin_lock_irq(&current->sighand->siglock);
  74         flush_signal_handlers(current, 1);
  75         spin_unlock_irq(&current->sighand->siglock);
  76
  77         /*
  78          * Initial kernel threads share ther FS with init, in order to
  79          * get the init root directory. But we've now created a new
  80          * thread that is going to execve a user process and has its own
  81          * 'struct fs_struct'. Reset umask to the default.
  82          */
  83         current->fs->umask = 0022;
  84
  85         /*
  86          * Our parent (unbound workqueue) runs with elevated scheduling
  87          * priority. Avoid propagating that into the userspace child.
  88          */
  89         set_user_nice(current, 0);
  90
  91         retval = -ENOMEM;
  92         new = prepare_kernel_cred(current);
  93         if (!new)
  94                 goto out;
  95
  96         spin_lock(&umh_sysctl_lock);
  97         new->cap_bset = cap_intersect(usermodehelper_bset, new->cap_bset);
  98         new->cap_inheritable = cap_intersect(usermodehelper_inheritable,
  99                                              new->cap_inheritable);
 100         spin_unlock(&umh_sysctl_lock);
 101
 102         if (sub_info->init) {
 103                 retval = sub_info->init(sub_info, new);
 104                 if (retval) {
 105                         abort_creds(new);
 106                         goto out;
 107                 }
 108         }
 109
 110         commit_creds(new);
 111
 112         wait_for_initramfs();
 113         retval = kernel_execve(sub_info->path,
 114                                (const char *const *)sub_info->argv,
 115                                (const char *const *)sub_info->envp);
 116 out:
 117         sub_info->retval = retval;
 118         /*
 119          * call_usermodehelper_exec_sync() will call umh_complete
 120          * if UHM_WAIT_PROC.
 121          */
 122         if (!(sub_info->wait & UMH_WAIT_PROC))
 123                 umh_complete(sub_info);
 124         if (!retval)
 125                 return 0;
 126         do_exit(0);
 127 }
 128
 129 /* Handles UMH_WAIT_PROC.  */
 130 static void call_usermodehelper_exec_sync(struct subprocess_info *sub_info)
 131 {
 132         pid_t pid;
 133
 134         /* If SIGCLD is ignored do_wait won't populate the status. */
 135         kernel_sigaction(SIGCHLD, SIG_DFL);
 136         pid = user_mode_thread(call_usermodehelper_exec_async, sub_info, SIGCHLD);
 137         if (pid < 0)
 138                 sub_info->retval = pid;
 139         else
 140                 kernel_wait(pid, &sub_info->retval);
 141
 142         /* Restore default kernel sig handler */
 143         kernel_sigaction(SIGCHLD, SIG_IGN);
 144         umh_complete(sub_info);
 145 }
 146
 147 /*
 148  * We need to create the usermodehelper kernel thread from a task that is affine
 149  * to an optimized set of CPUs (or nohz housekeeping ones) such that they
 150  * inherit a widest affinity irrespective of call_usermodehelper() callers with
 151  * possibly reduced affinity (eg: per-cpu workqueues). We don't want
 152  * usermodehelper targets to contend a busy CPU.
 153  *
 154  * Unbound workqueues provide such wide affinity and allow to block on
 155  * UMH_WAIT_PROC requests without blocking pending request (up to some limit).
 156  *
 157  * Besides, workqueues provide the privilege level that caller might not have
 158  * to perform the usermodehelper request.
 159  *
 160  */
 161 static void call_usermodehelper_exec_work(struct work_struct *work)
 162 {
 163         struct subprocess_info *sub_info =
 164                 container_of(work, struct subprocess_info, work);
 165
 166         if (sub_info->wait & UMH_WAIT_PROC) {
 167                 call_usermodehelper_exec_sync(sub_info);
 168         } else {
 169                 pid_t pid;
 170                 /*
 171                  * Use CLONE_PARENT to reparent it to kthreadd; we do not
 172                  * want to pollute current->children, and we need a parent
 173                  * that always ignores SIGCHLD to ensure auto-reaping.
 174                  */
 175                 pid = user_mode_thread(call_usermodehelper_exec_async, sub_info,
 176                                        CLONE_PARENT | SIGCHLD);
 177                 if (pid < 0) {
 178                         sub_info->retval = pid;
 179                         umh_complete(sub_info);
 180                 }
 181         }
 182 }
 183
 184 /*
 185  * If set, call_usermodehelper_exec() will exit immediately returning -EBUSY
 186  * (used for preventing user land processes from being created after the user
 187  * land has been frozen during a system-wide hibernation or suspend operation).
 188  * Should always be manipulated under umhelper_sem acquired for write.
 189  */
 190 static enum umh_disable_depth usermodehelper_disabled = UMH_DISABLED;
 191
 192 /* Number of helpers running */
 193 static atomic_t running_helpers = ATOMIC_INIT(0);
 194
 195 /*
 196  * Wait queue head used by usermodehelper_disable() to wait for all running
 197  * helpers to finish.
 198  */
 199 static DECLARE_WAIT_QUEUE_HEAD(running_helpers_waitq);
 200
 201 /*
 202  * Used by usermodehelper_read_lock_wait() to wait for usermodehelper_disabled
 203  * to become 'false'.
 204  */
 205 static DECLARE_WAIT_QUEUE_HEAD(usermodehelper_disabled_waitq);
 206
 207 /*
 208  * Time to wait for running_helpers to become zero before the setting of
 209  * usermodehelper_disabled in usermodehelper_disable() fails
 210  */
 211 #define RUNNING_HELPERS_TIMEOUT (5 * HZ)
 212
 213 int usermodehelper_read_trylock(void)
 214 {
 215         DEFINE_WAIT(wait);
 216         int ret = 0;
 217
 218         down_read(&umhelper_sem);
 219         for (;;) {
 220                 prepare_to_wait(&usermodehelper_disabled_waitq, &wait,
 221                                 TASK_INTERRUPTIBLE);
 222                 if (!usermodehelper_disabled)
 223                         break;
 224
 225                 if (usermodehelper_disabled == UMH_DISABLED)
 226                         ret = -EAGAIN;
 227
 228                 up_read(&umhelper_sem);
 229
 230                 if (ret)
 231                         break;
 232
 233                 schedule();
 234                 try_to_freeze();
 235
 236                 down_read(&umhelper_sem);
 237         }
 238         finish_wait(&usermodehelper_disabled_waitq, &wait);
 239         return ret;
 240 }
 241 EXPORT_SYMBOL_GPL(usermodehelper_read_trylock);
 242
 243 long usermodehelper_read_lock_wait(long timeout)
 244 {
 245         DEFINE_WAIT(wait);
 246
 247         if (timeout < 0)
 248                 return -EINVAL;
 249
 250         down_read(&umhelper_sem);
 251         for (;;) {
 252                 prepare_to_wait(&usermodehelper_disabled_waitq, &wait,
 253                                 TASK_UNINTERRUPTIBLE);
 254                 if (!usermodehelper_disabled)
 255                         break;
 256
 257                 up_read(&umhelper_sem);
 258
 259                 timeout = schedule_timeout(timeout);
 260                 if (!timeout)
 261                         break;
 262
 263                 down_read(&umhelper_sem);
 264         }
 265         finish_wait(&usermodehelper_disabled_waitq, &wait);
 266         return timeout;
 267 }
 268 EXPORT_SYMBOL_GPL(usermodehelper_read_lock_wait);
 269
 270 void usermodehelper_read_unlock(void)
 271 {
 272         up_read(&umhelper_sem);
 273 }
 274 EXPORT_SYMBOL_GPL(usermodehelper_read_unlock);
 275
 276 /**
 277  * __usermodehelper_set_disable_depth - Modify usermodehelper_disabled.
 278  * @depth: New value to assign to usermodehelper_disabled.
 279  *
 280  * Change the value of usermodehelper_disabled (under umhelper_sem locked for
 281  * writing) and wakeup tasks waiting for it to change.
 282  */
 283 void __usermodehelper_set_disable_depth(enum umh_disable_depth depth)
 284 {
 285         down_write(&umhelper_sem);
 286         usermodehelper_disabled = depth;
 287         wake_up(&usermodehelper_disabled_waitq);
 288         up_write(&umhelper_sem);
 289 }
 290
 291 /**
 292  * __usermodehelper_disable - Prevent new helpers from being started.
 293  * @depth: New value to assign to usermodehelper_disabled.
 294  *
 295  * Set usermodehelper_disabled to @depth and wait for running helpers to exit.
 296  */
 297 int __usermodehelper_disable(enum umh_disable_depth depth)
 298 {
 299         long retval;
 300
 301         if (!depth)
 302                 return -EINVAL;
 303
 304         down_write(&umhelper_sem);
 305         usermodehelper_disabled = depth;
 306         up_write(&umhelper_sem);
 307
 308         /*
 309          * From now on call_usermodehelper_exec() won't start any new
 310          * helpers, so it is sufficient if running_helpers turns out to
 311          * be zero at one point (it may be increased later, but that
 312          * doesn't matter).
 313          */
 314         retval = wait_event_timeout(running_helpers_waitq,
 315                                         atomic_read(&running_helpers) == 0,
 316                                         RUNNING_HELPERS_TIMEOUT);
 317         if (retval)
 318                 return 0;
 319
 320         __usermodehelper_set_disable_depth(UMH_ENABLED);
 321         return -EAGAIN;
 322 }
 323
 324 static void helper_lock(void)
 325 {
 326         atomic_inc(&running_helpers);
 327         smp_mb__after_atomic();
 328 }
 329
 330 static void helper_unlock(void)
 331 {
 332         if (atomic_dec_and_test(&running_helpers))
 333                 wake_up(&running_helpers_waitq);
 334 }
 335
 336 /**
 337  * call_usermodehelper_setup - prepare to call a usermode helper
 338  * @path: path to usermode executable
 339  * @argv: arg vector for process
 340  * @envp: environment for process
 341  * @gfp_mask: gfp mask for memory allocation
 342  * @init: an init function
 343  * @cleanup: a cleanup function
 344  * @data: arbitrary context sensitive data
 345  *
 346  * Returns either %NULL on allocation failure, or a subprocess_info
 347  * structure.  This should be passed to call_usermodehelper_exec to
 348  * exec the process and free the structure.
 349  *
 350  * The init function is used to customize the helper process prior to
 351  * exec.  A non-zero return code causes the process to error out, exit,
 352  * and return the failure to the calling process
 353  *
 354  * The cleanup function is just before the subprocess_info is about to
 355  * be freed.  This can be used for freeing the argv and envp.  The
 356  * Function must be runnable in either a process context or the
 357  * context in which call_usermodehelper_exec is called.
 358  */
 359 struct subprocess_info *call_usermodehelper_setup(const char *path, char **argv,
 360                 char **envp, gfp_t gfp_mask,
 361                 int (*init)(struct subprocess_info *info, struct cred *new),
 362                 void (*cleanup)(struct subprocess_info *info),
 363                 void *data)
 364 {
 365         struct subprocess_info *sub_info;
 366         sub_info = kzalloc(sizeof(struct subprocess_info), gfp_mask);
 367         if (!sub_info)
 368                 goto out;
 369
 370         INIT_WORK(&sub_info->work, call_usermodehelper_exec_work);
 371
 372 #ifdef CONFIG_STATIC_USERMODEHELPER
 373         sub_info->path = CONFIG_STATIC_USERMODEHELPER_PATH;
 374 #else
 375         sub_info->path = path;
 376 #endif
 377         sub_info->argv = argv;
 378         sub_info->envp = envp;
 379
 380         sub_info->cleanup = cleanup;
 381         sub_info->init = init;
 382         sub_info->data = data;
 383   out:
 384         return sub_info;
 385 }
 386 EXPORT_SYMBOL(call_usermodehelper_setup);
 387
 388 /**
 389  * call_usermodehelper_exec - start a usermode application
 390  * @sub_info: information about the subprocess
 391  * @wait: wait for the application to finish and return status.
 392  *        when UMH_NO_WAIT don't wait at all, but you get no useful error back
 393  *        when the program couldn't be exec'ed. This makes it safe to call
 394  *        from interrupt context.
 395  *
 396  * Runs a user-space application.  The application is started
 397  * asynchronously if wait is not set, and runs as a child of system workqueues.
 398  * (ie. it runs with full root capabilities and optimized affinity).
 399  *
 400  * Note: successful return value does not guarantee the helper was called at
 401  * all. You can't rely on sub_info->{init,cleanup} being called even for
 402  * UMH_WAIT_* wait modes as STATIC_USERMODEHELPER_PATH="" turns all helpers
 403  * into a successful no-op.
 404  */
 405 int call_usermodehelper_exec(struct subprocess_info *sub_info, int wait)
 406 {
 407         unsigned int state = TASK_UNINTERRUPTIBLE;
 408         DECLARE_COMPLETION_ONSTACK(done);
 409         int retval = 0;
 410
 411         if (!sub_info->path) {
 412                 call_usermodehelper_freeinfo(sub_info);
 413                 return -EINVAL;
 414         }
 415         helper_lock();
 416         if (usermodehelper_disabled) {
 417                 retval = -EBUSY;
 418                 goto out;
 419         }
 420
 421         /*
 422          * If there is no binary for us to call, then just return and get out of
 423          * here.  This allows us to set STATIC_USERMODEHELPER_PATH to "" and
 424          * disable all call_usermodehelper() calls.
 425          */
 426         if (strlen(sub_info->path) == 0)
 427                 goto out;
 428
 429         /*
 430          * Set the completion pointer only if there is a waiter.
 431          * This makes it possible to use umh_complete to free
 432          * the data structure in case of UMH_NO_WAIT.
 433          */
 434         sub_info->complete = (wait == UMH_NO_WAIT) ? NULL : &done;
 435         sub_info->wait = wait;
 436
 437         queue_work(system_unbound_wq, &sub_info->work);
 438         if (wait == UMH_NO_WAIT)        /* task has freed sub_info */
 439                 goto unlock;
 440
 441         if (wait & UMH_FREEZABLE)
 442                 state |= TASK_FREEZABLE;
 443
 444         if (wait & UMH_KILLABLE) {
 445                 retval = wait_for_completion_state(&done, state | TASK_KILLABLE);
 446                 if (!retval)
 447                         goto wait_done;
 448
 449                 /* umh_complete() will see NULL and free sub_info */
 450                 if (xchg(&sub_info->complete, NULL))
 451                         goto unlock;
 452
 453                 /*
 454                  * fallthrough; in case of -ERESTARTSYS now do uninterruptible
 455                  * wait_for_completion_state(). Since umh_complete() shall call
 456                  * complete() in a moment if xchg() above returned NULL, this
 457                  * uninterruptible wait_for_completion_state() will not block
 458                  * SIGKILL'ed processes for long.
 459                  */
 460         }
 461         wait_for_completion_state(&done, state);
 462
 463 wait_done:
 464         retval = sub_info->retval;
 465 out:
 466         call_usermodehelper_freeinfo(sub_info);
 467 unlock:
 468         helper_unlock();
 469         return retval;
 470 }
 471 EXPORT_SYMBOL(call_usermodehelper_exec);
 472
 473 /**
 474  * call_usermodehelper() - prepare and start a usermode application
 475  * @path: path to usermode executable
 476  * @argv: arg vector for process
 477  * @envp: environment for process
 478  * @wait: wait for the application to finish and return status.
 479  *        when UMH_NO_WAIT don't wait at all, but you get no useful error back
 480  *        when the program couldn't be exec'ed. This makes it safe to call
 481  *        from interrupt context.
 482  *
 483  * This function is the equivalent to use call_usermodehelper_setup() and
 484  * call_usermodehelper_exec().
 485  */
 486 int call_usermodehelper(const char *path, char **argv, char **envp, int wait)
 487 {
 488         struct subprocess_info *info;
 489         gfp_t gfp_mask = (wait == UMH_NO_WAIT) ? GFP_ATOMIC : GFP_KERNEL;
 490
 491         info = call_usermodehelper_setup(path, argv, envp, gfp_mask,
 492                                          NULL, NULL, NULL);
 493         if (info == NULL)
 494                 return -ENOMEM;
 495
 496         return call_usermodehelper_exec(info, wait);
 497 }
 498 EXPORT_SYMBOL(call_usermodehelper);
 499
 500 static int proc_cap_handler(struct ctl_table *table, int write,
 501                          void *buffer, size_t *lenp, loff_t *ppos)
 502 {
 503         struct ctl_table t;
 504         unsigned long cap_array[_KERNEL_CAPABILITY_U32S];
 505         kernel_cap_t new_cap;
 506         int err, i;
 507
 508         if (write && (!capable(CAP_SETPCAP) ||
 509                       !capable(CAP_SYS_MODULE)))
 510                 return -EPERM;
 511
 512         /*
 513          * convert from the global kernel_cap_t to the ulong array to print to
 514          * userspace if this is a read.
 515          */
 516         spin_lock(&umh_sysctl_lock);
 517         for (i = 0; i < _KERNEL_CAPABILITY_U32S; i++)  {
 518                 if (table->data == CAP_BSET)
 519                         cap_array[i] = usermodehelper_bset.cap[i];
 520                 else if (table->data == CAP_PI)
 521                         cap_array[i] = usermodehelper_inheritable.cap[i];
 522                 else
 523                         BUG();
 524         }
 525         spin_unlock(&umh_sysctl_lock);
 526
 527         t = *table;
 528         t.data = &cap_array;
 529
 530         /*
 531          * actually read or write and array of ulongs from userspace.  Remember
 532          * these are least significant 32 bits first
 533          */
 534         err = proc_doulongvec_minmax(&t, write, buffer, lenp, ppos);
 535         if (err < 0)
 536                 return err;
 537
 538         /*
 539          * convert from the sysctl array of ulongs to the kernel_cap_t
 540          * internal representation
 541          */
 542         for (i = 0; i < _KERNEL_CAPABILITY_U32S; i++)
 543                 new_cap.cap[i] = cap_array[i];
 544
 545         /*
 546          * Drop everything not in the new_cap (but don't add things)
 547          */
 548         if (write) {
 549                 spin_lock(&umh_sysctl_lock);
 550                 if (table->data == CAP_BSET)
 551                         usermodehelper_bset = cap_intersect(usermodehelper_bset, new_cap);
 552                 if (table->data == CAP_PI)
 553                         usermodehelper_inheritable = cap_intersect(usermodehelper_inheritable, new_cap);
 554                 spin_unlock(&umh_sysctl_lock);
 555         }
 556
 557         return 0;
 558 }
 559
 560 struct ctl_table usermodehelper_table[] = {
 561         {
 562                 .procname       = "bset",
 563                 .data           = CAP_BSET,
 564                 .maxlen         = _KERNEL_CAPABILITY_U32S * sizeof(unsigned long),
 565                 .mode           = 0600,
 566                 .proc_handler   = proc_cap_handler,
 567         },
 568         {
 569                 .procname       = "inheritable",
 570                 .data           = CAP_PI,
 571                 .maxlen         = _KERNEL_CAPABILITY_U32S * sizeof(unsigned long),
 572                 .mode           = 0600,
 573                 .proc_handler   = proc_cap_handler,
 574         },
 575         { }
 576 };