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 do_wait 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                 kernel_wait(pid, &sub_info->retval);
 129
 130         /* Restore default kernel sig handler */
 131         kernel_sigaction(SIGCHLD, SIG_IGN);
 132         umh_complete(sub_info);
 133 }
 134
 135 /*
 136  * We need to create the usermodehelper kernel thread from a task that is affine
 137  * to an optimized set of CPUs (or nohz housekeeping ones) such that they
 138  * inherit a widest affinity irrespective of call_usermodehelper() callers with
 139  * possibly reduced affinity (eg: per-cpu workqueues). We don't want
 140  * usermodehelper targets to contend a busy CPU.
 141  *
 142  * Unbound workqueues provide such wide affinity and allow to block on
 143  * UMH_WAIT_PROC requests without blocking pending request (up to some limit).
 144  *
 145  * Besides, workqueues provide the privilege level that caller might not have
 146  * to perform the usermodehelper request.
 147  *
 148  */
 149 static void call_usermodehelper_exec_work(struct work_struct *work)
 150 {
 151         struct subprocess_info *sub_info =
 152                 container_of(work, struct subprocess_info, work);
 153
 154         if (sub_info->wait & UMH_WAIT_PROC) {
 155                 call_usermodehelper_exec_sync(sub_info);
 156         } else {
 157                 pid_t pid;
 158                 /*
 159                  * Use CLONE_PARENT to reparent it to kthreadd; we do not
 160                  * want to pollute current->children, and we need a parent
 161                  * that always ignores SIGCHLD to ensure auto-reaping.
 162                  */
 163                 pid = kernel_thread(call_usermodehelper_exec_async, sub_info,
 164                                     CLONE_PARENT | SIGCHLD);
 165                 if (pid < 0) {
 166                         sub_info->retval = pid;
 167                         umh_complete(sub_info);
 168                 }
 169         }
 170 }
 171
 172 /*
 173  * If set, call_usermodehelper_exec() will exit immediately returning -EBUSY
 174  * (used for preventing user land processes from being created after the user
 175  * land has been frozen during a system-wide hibernation or suspend operation).
 176  * Should always be manipulated under umhelper_sem acquired for write.
 177  */
 178 static enum umh_disable_depth usermodehelper_disabled = UMH_DISABLED;
 179
 180 /* Number of helpers running */
 181 static atomic_t running_helpers = ATOMIC_INIT(0);
 182
 183 /*
 184  * Wait queue head used by usermodehelper_disable() to wait for all running
 185  * helpers to finish.
 186  */
 187 static DECLARE_WAIT_QUEUE_HEAD(running_helpers_waitq);
 188
 189 /*
 190  * Used by usermodehelper_read_lock_wait() to wait for usermodehelper_disabled
 191  * to become 'false'.
 192  */
 193 static DECLARE_WAIT_QUEUE_HEAD(usermodehelper_disabled_waitq);
 194
 195 /*
 196  * Time to wait for running_helpers to become zero before the setting of
 197  * usermodehelper_disabled in usermodehelper_disable() fails
 198  */
 199 #define RUNNING_HELPERS_TIMEOUT (5 * HZ)
 200
 201 int usermodehelper_read_trylock(void)
 202 {
 203         DEFINE_WAIT(wait);
 204         int ret = 0;
 205
 206         down_read(&umhelper_sem);
 207         for (;;) {
 208                 prepare_to_wait(&usermodehelper_disabled_waitq, &wait,
 209                                 TASK_INTERRUPTIBLE);
 210                 if (!usermodehelper_disabled)
 211                         break;
 212
 213                 if (usermodehelper_disabled == UMH_DISABLED)
 214                         ret = -EAGAIN;
 215
 216                 up_read(&umhelper_sem);
 217
 218                 if (ret)
 219                         break;
 220
 221                 schedule();
 222                 try_to_freeze();
 223
 224                 down_read(&umhelper_sem);
 225         }
 226         finish_wait(&usermodehelper_disabled_waitq, &wait);
 227         return ret;
 228 }
 229 EXPORT_SYMBOL_GPL(usermodehelper_read_trylock);
 230
 231 long usermodehelper_read_lock_wait(long timeout)
 232 {
 233         DEFINE_WAIT(wait);
 234
 235         if (timeout < 0)
 236                 return -EINVAL;
 237
 238         down_read(&umhelper_sem);
 239         for (;;) {
 240                 prepare_to_wait(&usermodehelper_disabled_waitq, &wait,
 241                                 TASK_UNINTERRUPTIBLE);
 242                 if (!usermodehelper_disabled)
 243                         break;
 244
 245                 up_read(&umhelper_sem);
 246
 247                 timeout = schedule_timeout(timeout);
 248                 if (!timeout)
 249                         break;
 250
 251                 down_read(&umhelper_sem);
 252         }
 253         finish_wait(&usermodehelper_disabled_waitq, &wait);
 254         return timeout;
 255 }
 256 EXPORT_SYMBOL_GPL(usermodehelper_read_lock_wait);
 257
 258 void usermodehelper_read_unlock(void)
 259 {
 260         up_read(&umhelper_sem);
 261 }
 262 EXPORT_SYMBOL_GPL(usermodehelper_read_unlock);
 263
 264 /**
 265  * __usermodehelper_set_disable_depth - Modify usermodehelper_disabled.
 266  * @depth: New value to assign to usermodehelper_disabled.
 267  *
 268  * Change the value of usermodehelper_disabled (under umhelper_sem locked for
 269  * writing) and wakeup tasks waiting for it to change.
 270  */
 271 void __usermodehelper_set_disable_depth(enum umh_disable_depth depth)
 272 {
 273         down_write(&umhelper_sem);
 274         usermodehelper_disabled = depth;
 275         wake_up(&usermodehelper_disabled_waitq);
 276         up_write(&umhelper_sem);
 277 }
 278
 279 /**
 280  * __usermodehelper_disable - Prevent new helpers from being started.
 281  * @depth: New value to assign to usermodehelper_disabled.
 282  *
 283  * Set usermodehelper_disabled to @depth and wait for running helpers to exit.
 284  */
 285 int __usermodehelper_disable(enum umh_disable_depth depth)
 286 {
 287         long retval;
 288
 289         if (!depth)
 290                 return -EINVAL;
 291
 292         down_write(&umhelper_sem);
 293         usermodehelper_disabled = depth;
 294         up_write(&umhelper_sem);
 295
 296         /*
 297          * From now on call_usermodehelper_exec() won't start any new
 298          * helpers, so it is sufficient if running_helpers turns out to
 299          * be zero at one point (it may be increased later, but that
 300          * doesn't matter).
 301          */
 302         retval = wait_event_timeout(running_helpers_waitq,
 303                                         atomic_read(&running_helpers) == 0,
 304                                         RUNNING_HELPERS_TIMEOUT);
 305         if (retval)
 306                 return 0;
 307
 308         __usermodehelper_set_disable_depth(UMH_ENABLED);
 309         return -EAGAIN;
 310 }
 311
 312 static void helper_lock(void)
 313 {
 314         atomic_inc(&running_helpers);
 315         smp_mb__after_atomic();
 316 }
 317
 318 static void helper_unlock(void)
 319 {
 320         if (atomic_dec_and_test(&running_helpers))
 321                 wake_up(&running_helpers_waitq);
 322 }
 323
 324 /**
 325  * call_usermodehelper_setup - prepare to call a usermode helper
 326  * @path: path to usermode executable
 327  * @argv: arg vector for process
 328  * @envp: environment for process
 329  * @gfp_mask: gfp mask for memory allocation
 330  * @cleanup: a cleanup function
 331  * @init: an init function
 332  * @data: arbitrary context sensitive data
 333  *
 334  * Returns either %NULL on allocation failure, or a subprocess_info
 335  * structure.  This should be passed to call_usermodehelper_exec to
 336  * exec the process and free the structure.
 337  *
 338  * The init function is used to customize the helper process prior to
 339  * exec.  A non-zero return code causes the process to error out, exit,
 340  * and return the failure to the calling process
 341  *
 342  * The cleanup function is just before ethe subprocess_info is about to
 343  * be freed.  This can be used for freeing the argv and envp.  The
 344  * Function must be runnable in either a process context or the
 345  * context in which call_usermodehelper_exec is called.
 346  */
 347 struct subprocess_info *call_usermodehelper_setup(const char *path, char **argv,
 348                 char **envp, gfp_t gfp_mask,
 349                 int (*init)(struct subprocess_info *info, struct cred *new),
 350                 void (*cleanup)(struct subprocess_info *info),
 351                 void *data)
 352 {
 353         struct subprocess_info *sub_info;
 354         sub_info = kzalloc(sizeof(struct subprocess_info), gfp_mask);
 355         if (!sub_info)
 356                 goto out;
 357
 358         INIT_WORK(&sub_info->work, call_usermodehelper_exec_work);
 359
 360 #ifdef CONFIG_STATIC_USERMODEHELPER
 361         sub_info->path = CONFIG_STATIC_USERMODEHELPER_PATH;
 362 #else
 363         sub_info->path = path;
 364 #endif
 365         sub_info->argv = argv;
 366         sub_info->envp = envp;
 367
 368         sub_info->cleanup = cleanup;
 369         sub_info->init = init;
 370         sub_info->data = data;
 371   out:
 372         return sub_info;
 373 }
 374 EXPORT_SYMBOL(call_usermodehelper_setup);
 375
 376 /**
 377  * call_usermodehelper_exec - start a usermode application
 378  * @sub_info: information about the subprocessa
 379  * @wait: wait for the application to finish and return status.
 380  *        when UMH_NO_WAIT don't wait at all, but you get no useful error back
 381  *        when the program couldn't be exec'ed. This makes it safe to call
 382  *        from interrupt context.
 383  *
 384  * Runs a user-space application.  The application is started
 385  * asynchronously if wait is not set, and runs as a child of system workqueues.
 386  * (ie. it runs with full root capabilities and optimized affinity).
 387  *
 388  * Note: successful return value does not guarantee the helper was called at
 389  * all. You can't rely on sub_info->{init,cleanup} being called even for
 390  * UMH_WAIT_* wait modes as STATIC_USERMODEHELPER_PATH="" turns all helpers
 391  * into a successful no-op.
 392  */
 393 int call_usermodehelper_exec(struct subprocess_info *sub_info, int wait)
 394 {
 395         DECLARE_COMPLETION_ONSTACK(done);
 396         int retval = 0;
 397
 398         if (!sub_info->path) {
 399                 call_usermodehelper_freeinfo(sub_info);
 400                 return -EINVAL;
 401         }
 402         helper_lock();
 403         if (usermodehelper_disabled) {
 404                 retval = -EBUSY;
 405                 goto out;
 406         }
 407
 408         /*
 409          * If there is no binary for us to call, then just return and get out of
 410          * here.  This allows us to set STATIC_USERMODEHELPER_PATH to "" and
 411          * disable all call_usermodehelper() calls.
 412          */
 413         if (strlen(sub_info->path) == 0)
 414                 goto out;
 415
 416         /*
 417          * Set the completion pointer only if there is a waiter.
 418          * This makes it possible to use umh_complete to free
 419          * the data structure in case of UMH_NO_WAIT.
 420          */
 421         sub_info->complete = (wait == UMH_NO_WAIT) ? NULL : &done;
 422         sub_info->wait = wait;
 423
 424         queue_work(system_unbound_wq, &sub_info->work);
 425         if (wait == UMH_NO_WAIT)        /* task has freed sub_info */
 426                 goto unlock;
 427
 428         if (wait & UMH_KILLABLE) {
 429                 retval = wait_for_completion_killable(&done);
 430                 if (!retval)
 431                         goto wait_done;
 432
 433                 /* umh_complete() will see NULL and free sub_info */
 434                 if (xchg(&sub_info->complete, NULL))
 435                         goto unlock;
 436                 /* fallthrough, umh_complete() was already called */
 437         }
 438
 439         wait_for_completion(&done);
 440 wait_done:
 441         retval = sub_info->retval;
 442 out:
 443         call_usermodehelper_freeinfo(sub_info);
 444 unlock:
 445         helper_unlock();
 446         return retval;
 447 }
 448 EXPORT_SYMBOL(call_usermodehelper_exec);
 449
 450 /**
 451  * call_usermodehelper() - prepare and start a usermode application
 452  * @path: path to usermode executable
 453  * @argv: arg vector for process
 454  * @envp: environment for process
 455  * @wait: wait for the application to finish and return status.
 456  *        when UMH_NO_WAIT don't wait at all, but you get no useful error back
 457  *        when the program couldn't be exec'ed. This makes it safe to call
 458  *        from interrupt context.
 459  *
 460  * This function is the equivalent to use call_usermodehelper_setup() and
 461  * call_usermodehelper_exec().
 462  */
 463 int call_usermodehelper(const char *path, char **argv, char **envp, int wait)
 464 {
 465         struct subprocess_info *info;
 466         gfp_t gfp_mask = (wait == UMH_NO_WAIT) ? GFP_ATOMIC : GFP_KERNEL;
 467
 468         info = call_usermodehelper_setup(path, argv, envp, gfp_mask,
 469                                          NULL, NULL, NULL);
 470         if (info == NULL)
 471                 return -ENOMEM;
 472
 473         return call_usermodehelper_exec(info, wait);
 474 }
 475 EXPORT_SYMBOL(call_usermodehelper);
 476
 477 static int proc_cap_handler(struct ctl_table *table, int write,
 478                          void *buffer, size_t *lenp, loff_t *ppos)
 479 {
 480         struct ctl_table t;
 481         unsigned long cap_array[_KERNEL_CAPABILITY_U32S];
 482         kernel_cap_t new_cap;
 483         int err, i;
 484
 485         if (write && (!capable(CAP_SETPCAP) ||
 486                       !capable(CAP_SYS_MODULE)))
 487                 return -EPERM;
 488
 489         /*
 490          * convert from the global kernel_cap_t to the ulong array to print to
 491          * userspace if this is a read.
 492          */
 493         spin_lock(&umh_sysctl_lock);
 494         for (i = 0; i < _KERNEL_CAPABILITY_U32S; i++)  {
 495                 if (table->data == CAP_BSET)
 496                         cap_array[i] = usermodehelper_bset.cap[i];
 497                 else if (table->data == CAP_PI)
 498                         cap_array[i] = usermodehelper_inheritable.cap[i];
 499                 else
 500                         BUG();
 501         }
 502         spin_unlock(&umh_sysctl_lock);
 503
 504         t = *table;
 505         t.data = &cap_array;
 506
 507         /*
 508          * actually read or write and array of ulongs from userspace.  Remember
 509          * these are least significant 32 bits first
 510          */
 511         err = proc_doulongvec_minmax(&t, write, buffer, lenp, ppos);
 512         if (err < 0)
 513                 return err;
 514
 515         /*
 516          * convert from the sysctl array of ulongs to the kernel_cap_t
 517          * internal representation
 518          */
 519         for (i = 0; i < _KERNEL_CAPABILITY_U32S; i++)
 520                 new_cap.cap[i] = cap_array[i];
 521
 522         /*
 523          * Drop everything not in the new_cap (but don't add things)
 524          */
 525         if (write) {
 526                 spin_lock(&umh_sysctl_lock);
 527                 if (table->data == CAP_BSET)
 528                         usermodehelper_bset = cap_intersect(usermodehelper_bset, new_cap);
 529                 if (table->data == CAP_PI)
 530                         usermodehelper_inheritable = cap_intersect(usermodehelper_inheritable, new_cap);
 531                 spin_unlock(&umh_sysctl_lock);
 532         }
 533
 534         return 0;
 535 }
 536
 537 struct ctl_table usermodehelper_table[] = {
 538         {
 539                 .procname       = "bset",
 540                 .data           = CAP_BSET,
 541                 .maxlen         = _KERNEL_CAPABILITY_U32S * sizeof(unsigned long),
 542                 .mode           = 0600,
 543                 .proc_handler   = proc_cap_handler,
 544         },
 545         {
 546                 .procname       = "inheritable",
 547                 .data           = CAP_PI,
 548                 .maxlen         = _KERNEL_CAPABILITY_U32S * sizeof(unsigned long),
 549                 .mode           = 0600,
 550                 .proc_handler   = proc_cap_handler,
 551         },
 552         { }
 553 };