kernel/seccomp.c

   1 /*
   2  * linux/kernel/seccomp.c
   3  *
   4  * Copyright 2004-2005  Andrea Arcangeli <andrea@cpushare.com>
   5  *
   6  * Copyright (C) 2012 Google, Inc.
   7  * Will Drewry <wad@chromium.org>
   8  *
   9  * This defines a simple but solid secure-computing facility.
  10  *
  11  * Mode 1 uses a fixed list of allowed system calls.
  12  * Mode 2 allows user-defined system call filters in the form
  13  *        of Berkeley Packet Filters/Linux Socket Filters.
  14  */
  15
  16 #include <linux/refcount.h>
  17 #include <linux/audit.h>
  18 #include <linux/compat.h>
  19 #include <linux/coredump.h>
  20 #include <linux/kmemleak.h>
  21 #include <linux/sched.h>
  22 #include <linux/sched/task_stack.h>
  23 #include <linux/seccomp.h>
  24 #include <linux/slab.h>
  25 #include <linux/syscalls.h>
  26 #include <linux/sysctl.h>
  27
  28 #ifdef CONFIG_HAVE_ARCH_SECCOMP_FILTER
  29 #include <asm/syscall.h>
  30 #endif
  31
  32 #ifdef CONFIG_SECCOMP_FILTER
  33 #include <linux/filter.h>
  34 #include <linux/pid.h>
  35 #include <linux/ptrace.h>
  36 #include <linux/security.h>
  37 #include <linux/tracehook.h>
  38 #include <linux/uaccess.h>
  39
  40 /**
  41  * struct seccomp_filter - container for seccomp BPF programs
  42  *
  43  * @usage: reference count to manage the object lifetime.
  44  *         get/put helpers should be used when accessing an instance
  45  *         outside of a lifetime-guarded section.  In general, this
  46  *         is only needed for handling filters shared across tasks.
  47  * @log: true if all actions except for SECCOMP_RET_ALLOW should be logged
  48  * @prev: points to a previously installed, or inherited, filter
  49  * @prog: the BPF program to evaluate
  50  *
  51  * seccomp_filter objects are organized in a tree linked via the @prev
  52  * pointer.  For any task, it appears to be a singly-linked list starting
  53  * with current->seccomp.filter, the most recently attached or inherited filter.
  54  * However, multiple filters may share a @prev node, by way of fork(), which
  55  * results in a unidirectional tree existing in memory.  This is similar to
  56  * how namespaces work.
  57  *
  58  * seccomp_filter objects should never be modified after being attached
  59  * to a task_struct (other than @usage).
  60  */
  61 struct seccomp_filter {
  62         refcount_t usage;
  63         bool log;
  64         struct seccomp_filter *prev;
  65         struct bpf_prog *prog;
  66 };
  67
  68 /* Limit any path through the tree to 256KB worth of instructions. */
  69 #define MAX_INSNS_PER_PATH ((1 << 18) / sizeof(struct sock_filter))
  70
  71 /*
  72  * Endianness is explicitly ignored and left for BPF program authors to manage
  73  * as per the specific architecture.
  74  */
  75 static void populate_seccomp_data(struct seccomp_data *sd)
  76 {
  77         struct task_struct *task = current;
  78         struct pt_regs *regs = task_pt_regs(task);
  79         unsigned long args[6];
  80
  81         sd->nr = syscall_get_nr(task, regs);
  82         sd->arch = syscall_get_arch();
  83         syscall_get_arguments(task, regs, 0, 6, args);
  84         sd->args[0] = args[0];
  85         sd->args[1] = args[1];
  86         sd->args[2] = args[2];
  87         sd->args[3] = args[3];
  88         sd->args[4] = args[4];
  89         sd->args[5] = args[5];
  90         sd->instruction_pointer = KSTK_EIP(task);
  91 }
  92
  93 /**
  94  *      seccomp_check_filter - verify seccomp filter code
  95  *      @filter: filter to verify
  96  *      @flen: length of filter
  97  *
  98  * Takes a previously checked filter (by bpf_check_classic) and
  99  * redirects all filter code that loads struct sk_buff data
 100  * and related data through seccomp_bpf_load.  It also
 101  * enforces length and alignment checking of those loads.
 102  *
 103  * Returns 0 if the rule set is legal or -EINVAL if not.
 104  */
 105 static int seccomp_check_filter(struct sock_filter *filter, unsigned int flen)
 106 {
 107         int pc;
 108         for (pc = 0; pc < flen; pc++) {
 109                 struct sock_filter *ftest = &filter[pc];
 110                 u16 code = ftest->code;
 111                 u32 k = ftest->k;
 112
 113                 switch (code) {
 114                 case BPF_LD | BPF_W | BPF_ABS:
 115                         ftest->code = BPF_LDX | BPF_W | BPF_ABS;
 116                         /* 32-bit aligned and not out of bounds. */
 117                         if (k >= sizeof(struct seccomp_data) || k & 3)
 118                                 return -EINVAL;
 119                         continue;
 120                 case BPF_LD | BPF_W | BPF_LEN:
 121                         ftest->code = BPF_LD | BPF_IMM;
 122                         ftest->k = sizeof(struct seccomp_data);
 123                         continue;
 124                 case BPF_LDX | BPF_W | BPF_LEN:
 125                         ftest->code = BPF_LDX | BPF_IMM;
 126                         ftest->k = sizeof(struct seccomp_data);
 127                         continue;
 128                 /* Explicitly include allowed calls. */
 129                 case BPF_RET | BPF_K:
 130                 case BPF_RET | BPF_A:
 131                 case BPF_ALU | BPF_ADD | BPF_K:
 132                 case BPF_ALU | BPF_ADD | BPF_X:
 133                 case BPF_ALU | BPF_SUB | BPF_K:
 134                 case BPF_ALU | BPF_SUB | BPF_X:
 135                 case BPF_ALU | BPF_MUL | BPF_K:
 136                 case BPF_ALU | BPF_MUL | BPF_X:
 137                 case BPF_ALU | BPF_DIV | BPF_K:
 138                 case BPF_ALU | BPF_DIV | BPF_X:
 139                 case BPF_ALU | BPF_AND | BPF_K:
 140                 case BPF_ALU | BPF_AND | BPF_X:
 141                 case BPF_ALU | BPF_OR | BPF_K:
 142                 case BPF_ALU | BPF_OR | BPF_X:
 143                 case BPF_ALU | BPF_XOR | BPF_K:
 144                 case BPF_ALU | BPF_XOR | BPF_X:
 145                 case BPF_ALU | BPF_LSH | BPF_K:
 146                 case BPF_ALU | BPF_LSH | BPF_X:
 147                 case BPF_ALU | BPF_RSH | BPF_K:
 148                 case BPF_ALU | BPF_RSH | BPF_X:
 149                 case BPF_ALU | BPF_NEG:
 150                 case BPF_LD | BPF_IMM:
 151                 case BPF_LDX | BPF_IMM:
 152                 case BPF_MISC | BPF_TAX:
 153                 case BPF_MISC | BPF_TXA:
 154                 case BPF_LD | BPF_MEM:
 155                 case BPF_LDX | BPF_MEM:
 156                 case BPF_ST:
 157                 case BPF_STX:
 158                 case BPF_JMP | BPF_JA:
 159                 case BPF_JMP | BPF_JEQ | BPF_K:
 160                 case BPF_JMP | BPF_JEQ | BPF_X:
 161                 case BPF_JMP | BPF_JGE | BPF_K:
 162                 case BPF_JMP | BPF_JGE | BPF_X:
 163                 case BPF_JMP | BPF_JGT | BPF_K:
 164                 case BPF_JMP | BPF_JGT | BPF_X:
 165                 case BPF_JMP | BPF_JSET | BPF_K:
 166                 case BPF_JMP | BPF_JSET | BPF_X:
 167                         continue;
 168                 default:
 169                         return -EINVAL;
 170                 }
 171         }
 172         return 0;
 173 }
 174
 175 /**
 176  * seccomp_run_filters - evaluates all seccomp filters against @sd
 177  * @sd: optional seccomp data to be passed to filters
 178  * @match: stores struct seccomp_filter that resulted in the return value,
 179  *         unless filter returned SECCOMP_RET_ALLOW, in which case it will
 180  *         be unchanged.
 181  *
 182  * Returns valid seccomp BPF response codes.
 183  */
 184 #define ACTION_ONLY(ret) ((s32)((ret) & (SECCOMP_RET_ACTION_FULL)))
 185 static u32 seccomp_run_filters(const struct seccomp_data *sd,
 186                                struct seccomp_filter **match)
 187 {
 188         struct seccomp_data sd_local;
 189         u32 ret = SECCOMP_RET_ALLOW;
 190         /* Make sure cross-thread synced filter points somewhere sane. */
 191         struct seccomp_filter *f =
 192                         lockless_dereference(current->seccomp.filter);
 193
 194         /* Ensure unexpected behavior doesn't result in failing open. */
 195         if (unlikely(WARN_ON(f == NULL)))
 196                 return SECCOMP_RET_KILL_PROCESS;
 197
 198         if (!sd) {
 199                 populate_seccomp_data(&sd_local);
 200                 sd = &sd_local;
 201         }
 202
 203         /*
 204          * All filters in the list are evaluated and the lowest BPF return
 205          * value always takes priority (ignoring the DATA).
 206          */
 207         for (; f; f = f->prev) {
 208                 u32 cur_ret = BPF_PROG_RUN(f->prog, sd);
 209
 210                 if (ACTION_ONLY(cur_ret) < ACTION_ONLY(ret)) {
 211                         ret = cur_ret;
 212                         *match = f;
 213                 }
 214         }
 215         return ret;
 216 }
 217 #endif /* CONFIG_SECCOMP_FILTER */
 218
 219 static inline bool seccomp_may_assign_mode(unsigned long seccomp_mode)
 220 {
 221         assert_spin_locked(&current->sighand->siglock);
 222
 223         if (current->seccomp.mode && current->seccomp.mode != seccomp_mode)
 224                 return false;
 225
 226         return true;
 227 }
 228
 229 static inline void seccomp_assign_mode(struct task_struct *task,
 230                                        unsigned long seccomp_mode)
 231 {
 232         assert_spin_locked(&task->sighand->siglock);
 233
 234         task->seccomp.mode = seccomp_mode;
 235         /*
 236          * Make sure TIF_SECCOMP cannot be set before the mode (and
 237          * filter) is set.
 238          */
 239         smp_mb__before_atomic();
 240         set_tsk_thread_flag(task, TIF_SECCOMP);
 241 }
 242
 243 #ifdef CONFIG_SECCOMP_FILTER
 244 /* Returns 1 if the parent is an ancestor of the child. */
 245 static int is_ancestor(struct seccomp_filter *parent,
 246                        struct seccomp_filter *child)
 247 {
 248         /* NULL is the root ancestor. */
 249         if (parent == NULL)
 250                 return 1;
 251         for (; child; child = child->prev)
 252                 if (child == parent)
 253                         return 1;
 254         return 0;
 255 }
 256
 257 /**
 258  * seccomp_can_sync_threads: checks if all threads can be synchronized
 259  *
 260  * Expects sighand and cred_guard_mutex locks to be held.
 261  *
 262  * Returns 0 on success, -ve on error, or the pid of a thread which was
 263  * either not in the correct seccomp mode or it did not have an ancestral
 264  * seccomp filter.
 265  */
 266 static inline pid_t seccomp_can_sync_threads(void)
 267 {
 268         struct task_struct *thread, *caller;
 269
 270         BUG_ON(!mutex_is_locked(&current->signal->cred_guard_mutex));
 271         assert_spin_locked(&current->sighand->siglock);
 272
 273         /* Validate all threads being eligible for synchronization. */
 274         caller = current;
 275         for_each_thread(caller, thread) {
 276                 pid_t failed;
 277
 278                 /* Skip current, since it is initiating the sync. */
 279                 if (thread == caller)
 280                         continue;
 281
 282                 if (thread->seccomp.mode == SECCOMP_MODE_DISABLED ||
 283                     (thread->seccomp.mode == SECCOMP_MODE_FILTER &&
 284                      is_ancestor(thread->seccomp.filter,
 285                                  caller->seccomp.filter)))
 286                         continue;
 287
 288                 /* Return the first thread that cannot be synchronized. */
 289                 failed = task_pid_vnr(thread);
 290                 /* If the pid cannot be resolved, then return -ESRCH */
 291                 if (unlikely(WARN_ON(failed == 0)))
 292                         failed = -ESRCH;
 293                 return failed;
 294         }
 295
 296         return 0;
 297 }
 298
 299 /**
 300  * seccomp_sync_threads: sets all threads to use current's filter
 301  *
 302  * Expects sighand and cred_guard_mutex locks to be held, and for
 303  * seccomp_can_sync_threads() to have returned success already
 304  * without dropping the locks.
 305  *
 306  */
 307 static inline void seccomp_sync_threads(void)
 308 {
 309         struct task_struct *thread, *caller;
 310
 311         BUG_ON(!mutex_is_locked(&current->signal->cred_guard_mutex));
 312         assert_spin_locked(&current->sighand->siglock);
 313
 314         /* Synchronize all threads. */
 315         caller = current;
 316         for_each_thread(caller, thread) {
 317                 /* Skip current, since it needs no changes. */
 318                 if (thread == caller)
 319                         continue;
 320
 321                 /* Get a task reference for the new leaf node. */
 322                 get_seccomp_filter(caller);
 323                 /*
 324                  * Drop the task reference to the shared ancestor since
 325                  * current's path will hold a reference.  (This also
 326                  * allows a put before the assignment.)
 327                  */
 328                 put_seccomp_filter(thread);
 329                 smp_store_release(&thread->seccomp.filter,
 330                                   caller->seccomp.filter);
 331
 332                 /*
 333                  * Don't let an unprivileged task work around
 334                  * the no_new_privs restriction by creating
 335                  * a thread that sets it up, enters seccomp,
 336                  * then dies.
 337                  */
 338                 if (task_no_new_privs(caller))
 339                         task_set_no_new_privs(thread);
 340
 341                 /*
 342                  * Opt the other thread into seccomp if needed.
 343                  * As threads are considered to be trust-realm
 344                  * equivalent (see ptrace_may_access), it is safe to
 345                  * allow one thread to transition the other.
 346                  */
 347                 if (thread->seccomp.mode == SECCOMP_MODE_DISABLED)
 348                         seccomp_assign_mode(thread, SECCOMP_MODE_FILTER);
 349         }
 350 }
 351
 352 /**
 353  * seccomp_prepare_filter: Prepares a seccomp filter for use.
 354  * @fprog: BPF program to install
 355  *
 356  * Returns filter on success or an ERR_PTR on failure.
 357  */
 358 static struct seccomp_filter *seccomp_prepare_filter(struct sock_fprog *fprog)
 359 {
 360         struct seccomp_filter *sfilter;
 361         int ret;
 362         const bool save_orig = IS_ENABLED(CONFIG_CHECKPOINT_RESTORE);
 363
 364         if (fprog->len == 0 || fprog->len > BPF_MAXINSNS)
 365                 return ERR_PTR(-EINVAL);
 366
 367         BUG_ON(INT_MAX / fprog->len < sizeof(struct sock_filter));
 368
 369         /*
 370          * Installing a seccomp filter requires that the task has
 371          * CAP_SYS_ADMIN in its namespace or be running with no_new_privs.
 372          * This avoids scenarios where unprivileged tasks can affect the
 373          * behavior of privileged children.
 374          */
 375         if (!task_no_new_privs(current) &&
 376             security_capable_noaudit(current_cred(), current_user_ns(),
 377                                      CAP_SYS_ADMIN) != 0)
 378                 return ERR_PTR(-EACCES);
 379
 380         /* Allocate a new seccomp_filter */
 381         sfilter = kzalloc(sizeof(*sfilter), GFP_KERNEL | __GFP_NOWARN);
 382         if (!sfilter)
 383                 return ERR_PTR(-ENOMEM);
 384
 385         ret = bpf_prog_create_from_user(&sfilter->prog, fprog,
 386                                         seccomp_check_filter, save_orig);
 387         if (ret < 0) {
 388                 kfree(sfilter);
 389                 return ERR_PTR(ret);
 390         }
 391
 392         refcount_set(&sfilter->usage, 1);
 393
 394         return sfilter;
 395 }
 396
 397 /**
 398  * seccomp_prepare_user_filter - prepares a user-supplied sock_fprog
 399  * @user_filter: pointer to the user data containing a sock_fprog.
 400  *
 401  * Returns 0 on success and non-zero otherwise.
 402  */
 403 static struct seccomp_filter *
 404 seccomp_prepare_user_filter(const char __user *user_filter)
 405 {
 406         struct sock_fprog fprog;
 407         struct seccomp_filter *filter = ERR_PTR(-EFAULT);
 408
 409 #ifdef CONFIG_COMPAT
 410         if (in_compat_syscall()) {
 411                 struct compat_sock_fprog fprog32;
 412                 if (copy_from_user(&fprog32, user_filter, sizeof(fprog32)))
 413                         goto out;
 414                 fprog.len = fprog32.len;
 415                 fprog.filter = compat_ptr(fprog32.filter);
 416         } else /* falls through to the if below. */
 417 #endif
 418         if (copy_from_user(&fprog, user_filter, sizeof(fprog)))
 419                 goto out;
 420         filter = seccomp_prepare_filter(&fprog);
 421 out:
 422         return filter;
 423 }
 424
 425 /**
 426  * seccomp_attach_filter: validate and attach filter
 427  * @flags:  flags to change filter behavior
 428  * @filter: seccomp filter to add to the current process
 429  *
 430  * Caller must be holding current->sighand->siglock lock.
 431  *
 432  * Returns 0 on success, -ve on error.
 433  */
 434 static long seccomp_attach_filter(unsigned int flags,
 435                                   struct seccomp_filter *filter)
 436 {
 437         unsigned long total_insns;
 438         struct seccomp_filter *walker;
 439
 440         assert_spin_locked(&current->sighand->siglock);
 441
 442         /* Validate resulting filter length. */
 443         total_insns = filter->prog->len;
 444         for (walker = current->seccomp.filter; walker; walker = walker->prev)
 445                 total_insns += walker->prog->len + 4;  /* 4 instr penalty */
 446         if (total_insns > MAX_INSNS_PER_PATH)
 447                 return -ENOMEM;
 448
 449         /* If thread sync has been requested, check that it is possible. */
 450         if (flags & SECCOMP_FILTER_FLAG_TSYNC) {
 451                 int ret;
 452
 453                 ret = seccomp_can_sync_threads();
 454                 if (ret)
 455                         return ret;
 456         }
 457
 458         /* Set log flag, if present. */
 459         if (flags & SECCOMP_FILTER_FLAG_LOG)
 460                 filter->log = true;
 461
 462         /*
 463          * If there is an existing filter, make it the prev and don't drop its
 464          * task reference.
 465          */
 466         filter->prev = current->seccomp.filter;
 467         current->seccomp.filter = filter;
 468
 469         /* Now that the new filter is in place, synchronize to all threads. */
 470         if (flags & SECCOMP_FILTER_FLAG_TSYNC)
 471                 seccomp_sync_threads();
 472
 473         return 0;
 474 }
 475
 476 /* get_seccomp_filter - increments the reference count of the filter on @tsk */
 477 void get_seccomp_filter(struct task_struct *tsk)
 478 {
 479         struct seccomp_filter *orig = tsk->seccomp.filter;
 480         if (!orig)
 481                 return;
 482         /* Reference count is bounded by the number of total processes. */
 483         refcount_inc(&orig->usage);
 484 }
 485
 486 static inline void seccomp_filter_free(struct seccomp_filter *filter)
 487 {
 488         if (filter) {
 489                 bpf_prog_destroy(filter->prog);
 490                 kfree(filter);
 491         }
 492 }
 493
 494 /* put_seccomp_filter - decrements the ref count of tsk->seccomp.filter */
 495 void put_seccomp_filter(struct task_struct *tsk)
 496 {
 497         struct seccomp_filter *orig = tsk->seccomp.filter;
 498         /* Clean up single-reference branches iteratively. */
 499         while (orig && refcount_dec_and_test(&orig->usage)) {
 500                 struct seccomp_filter *freeme = orig;
 501                 orig = orig->prev;
 502                 seccomp_filter_free(freeme);
 503         }
 504 }
 505
 506 static void seccomp_init_siginfo(siginfo_t *info, int syscall, int reason)
 507 {
 508         memset(info, 0, sizeof(*info));
 509         info->si_signo = SIGSYS;
 510         info->si_code = SYS_SECCOMP;
 511         info->si_call_addr = (void __user *)KSTK_EIP(current);
 512         info->si_errno = reason;
 513         info->si_arch = syscall_get_arch();
 514         info->si_syscall = syscall;
 515 }
 516
 517 /**
 518  * seccomp_send_sigsys - signals the task to allow in-process syscall emulation
 519  * @syscall: syscall number to send to userland
 520  * @reason: filter-supplied reason code to send to userland (via si_errno)
 521  *
 522  * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info.
 523  */
 524 static void seccomp_send_sigsys(int syscall, int reason)
 525 {
 526         struct siginfo info;
 527         seccomp_init_siginfo(&info, syscall, reason);
 528         force_sig_info(SIGSYS, &info, current);
 529 }
 530 #endif  /* CONFIG_SECCOMP_FILTER */
 531
 532 /* For use with seccomp_actions_logged */
 533 #define SECCOMP_LOG_KILL_PROCESS        (1 << 0)
 534 #define SECCOMP_LOG_KILL_THREAD         (1 << 1)
 535 #define SECCOMP_LOG_TRAP                (1 << 2)
 536 #define SECCOMP_LOG_ERRNO               (1 << 3)
 537 #define SECCOMP_LOG_TRACE               (1 << 4)
 538 #define SECCOMP_LOG_LOG                 (1 << 5)
 539 #define SECCOMP_LOG_ALLOW               (1 << 6)
 540
 541 static u32 seccomp_actions_logged = SECCOMP_LOG_KILL_PROCESS |
 542                                     SECCOMP_LOG_KILL_THREAD  |
 543                                     SECCOMP_LOG_TRAP  |
 544                                     SECCOMP_LOG_ERRNO |
 545                                     SECCOMP_LOG_TRACE |
 546                                     SECCOMP_LOG_LOG;
 547
 548 static inline void seccomp_log(unsigned long syscall, long signr, u32 action,
 549                                bool requested)
 550 {
 551         bool log = false;
 552
 553         switch (action) {
 554         case SECCOMP_RET_ALLOW:
 555                 break;
 556         case SECCOMP_RET_TRAP:
 557                 log = requested && seccomp_actions_logged & SECCOMP_LOG_TRAP;
 558                 break;
 559         case SECCOMP_RET_ERRNO:
 560                 log = requested && seccomp_actions_logged & SECCOMP_LOG_ERRNO;
 561                 break;
 562         case SECCOMP_RET_TRACE:
 563                 log = requested && seccomp_actions_logged & SECCOMP_LOG_TRACE;
 564                 break;
 565         case SECCOMP_RET_LOG:
 566                 log = seccomp_actions_logged & SECCOMP_LOG_LOG;
 567                 break;
 568         case SECCOMP_RET_KILL_THREAD:
 569                 log = seccomp_actions_logged & SECCOMP_LOG_KILL_THREAD;
 570                 break;
 571         case SECCOMP_RET_KILL_PROCESS:
 572         default:
 573                 log = seccomp_actions_logged & SECCOMP_LOG_KILL_PROCESS;
 574         }
 575
 576         /*
 577          * Force an audit message to be emitted when the action is RET_KILL_*,
 578          * RET_LOG, or the FILTER_FLAG_LOG bit was set and the action is
 579          * allowed to be logged by the admin.
 580          */
 581         if (log)
 582                 return __audit_seccomp(syscall, signr, action);
 583
 584         /*
 585          * Let the audit subsystem decide if the action should be audited based
 586          * on whether the current task itself is being audited.
 587          */
 588         return audit_seccomp(syscall, signr, action);
 589 }
 590
 591 /*
 592  * Secure computing mode 1 allows only read/write/exit/sigreturn.
 593  * To be fully secure this must be combined with rlimit
 594  * to limit the stack allocations too.
 595  */
 596 static const int mode1_syscalls[] = {
 597         __NR_seccomp_read, __NR_seccomp_write, __NR_seccomp_exit, __NR_seccomp_sigreturn,
 598         0, /* null terminated */
 599 };
 600
 601 static void __secure_computing_strict(int this_syscall)
 602 {
 603         const int *syscall_whitelist = mode1_syscalls;
 604 #ifdef CONFIG_COMPAT
 605         if (in_compat_syscall())
 606                 syscall_whitelist = get_compat_mode1_syscalls();
 607 #endif
 608         do {
 609                 if (*syscall_whitelist == this_syscall)
 610                         return;
 611         } while (*++syscall_whitelist);
 612
 613 #ifdef SECCOMP_DEBUG
 614         dump_stack();
 615 #endif
 616         seccomp_log(this_syscall, SIGKILL, SECCOMP_RET_KILL_THREAD, true);
 617         do_exit(SIGKILL);
 618 }
 619
 620 #ifndef CONFIG_HAVE_ARCH_SECCOMP_FILTER
 621 void secure_computing_strict(int this_syscall)
 622 {
 623         int mode = current->seccomp.mode;
 624
 625         if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) &&
 626             unlikely(current->ptrace & PT_SUSPEND_SECCOMP))
 627                 return;
 628
 629         if (mode == SECCOMP_MODE_DISABLED)
 630                 return;
 631         else if (mode == SECCOMP_MODE_STRICT)
 632                 __secure_computing_strict(this_syscall);
 633         else
 634                 BUG();
 635 }
 636 #else
 637
 638 #ifdef CONFIG_SECCOMP_FILTER
 639 static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd,
 640                             const bool recheck_after_trace)
 641 {
 642         u32 filter_ret, action;
 643         struct seccomp_filter *match = NULL;
 644         int data;
 645
 646         /*
 647          * Make sure that any changes to mode from another thread have
 648          * been seen after TIF_SECCOMP was seen.
 649          */
 650         rmb();
 651
 652         filter_ret = seccomp_run_filters(sd, &match);
 653         data = filter_ret & SECCOMP_RET_DATA;
 654         action = filter_ret & SECCOMP_RET_ACTION_FULL;
 655
 656         switch (action) {
 657         case SECCOMP_RET_ERRNO:
 658                 /* Set low-order bits as an errno, capped at MAX_ERRNO. */
 659                 if (data > MAX_ERRNO)
 660                         data = MAX_ERRNO;
 661                 syscall_set_return_value(current, task_pt_regs(current),
 662                                          -data, 0);
 663                 goto skip;
 664
 665         case SECCOMP_RET_TRAP:
 666                 /* Show the handler the original registers. */
 667                 syscall_rollback(current, task_pt_regs(current));
 668                 /* Let the filter pass back 16 bits of data. */
 669                 seccomp_send_sigsys(this_syscall, data);
 670                 goto skip;
 671
 672         case SECCOMP_RET_TRACE:
 673                 /* We've been put in this state by the ptracer already. */
 674                 if (recheck_after_trace)
 675                         return 0;
 676
 677                 /* ENOSYS these calls if there is no tracer attached. */
 678                 if (!ptrace_event_enabled(current, PTRACE_EVENT_SECCOMP)) {
 679                         syscall_set_return_value(current,
 680                                                  task_pt_regs(current),
 681                                                  -ENOSYS, 0);
 682                         goto skip;
 683                 }
 684
 685                 /* Allow the BPF to provide the event message */
 686                 ptrace_event(PTRACE_EVENT_SECCOMP, data);
 687                 /*
 688                  * The delivery of a fatal signal during event
 689                  * notification may silently skip tracer notification,
 690                  * which could leave us with a potentially unmodified
 691                  * syscall that the tracer would have liked to have
 692                  * changed. Since the process is about to die, we just
 693                  * force the syscall to be skipped and let the signal
 694                  * kill the process and correctly handle any tracer exit
 695                  * notifications.
 696                  */
 697                 if (fatal_signal_pending(current))
 698                         goto skip;
 699                 /* Check if the tracer forced the syscall to be skipped. */
 700                 this_syscall = syscall_get_nr(current, task_pt_regs(current));
 701                 if (this_syscall < 0)
 702                         goto skip;
 703
 704                 /*
 705                  * Recheck the syscall, since it may have changed. This
 706                  * intentionally uses a NULL struct seccomp_data to force
 707                  * a reload of all registers. This does not goto skip since
 708                  * a skip would have already been reported.
 709                  */
 710                 if (__seccomp_filter(this_syscall, NULL, true))
 711                         return -1;
 712
 713                 return 0;
 714
 715         case SECCOMP_RET_LOG:
 716                 seccomp_log(this_syscall, 0, action, true);
 717                 return 0;
 718
 719         case SECCOMP_RET_ALLOW:
 720                 /*
 721                  * Note that the "match" filter will always be NULL for
 722                  * this action since SECCOMP_RET_ALLOW is the starting
 723                  * state in seccomp_run_filters().
 724                  */
 725                 return 0;
 726
 727         case SECCOMP_RET_KILL_THREAD:
 728         case SECCOMP_RET_KILL_PROCESS:
 729         default:
 730                 seccomp_log(this_syscall, SIGSYS, action, true);
 731                 /* Dump core only if this is the last remaining thread. */
 732                 if (action == SECCOMP_RET_KILL_PROCESS ||
 733                     get_nr_threads(current) == 1) {
 734                         siginfo_t info;
 735
 736                         /* Show the original registers in the dump. */
 737                         syscall_rollback(current, task_pt_regs(current));
 738                         /* Trigger a manual coredump since do_exit skips it. */
 739                         seccomp_init_siginfo(&info, this_syscall, data);
 740                         do_coredump(&info);
 741                 }
 742                 if (action == SECCOMP_RET_KILL_PROCESS)
 743                         do_group_exit(SIGSYS);
 744                 else
 745                         do_exit(SIGSYS);
 746         }
 747
 748         unreachable();
 749
 750 skip:
 751         seccomp_log(this_syscall, 0, action, match ? match->log : false);
 752         return -1;
 753 }
 754 #else
 755 static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd,
 756                             const bool recheck_after_trace)
 757 {
 758         BUG();
 759 }
 760 #endif
 761
 762 int __secure_computing(const struct seccomp_data *sd)
 763 {
 764         int mode = current->seccomp.mode;
 765         int this_syscall;
 766
 767         if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) &&
 768             unlikely(current->ptrace & PT_SUSPEND_SECCOMP))
 769                 return 0;
 770
 771         this_syscall = sd ? sd->nr :
 772                 syscall_get_nr(current, task_pt_regs(current));
 773
 774         switch (mode) {
 775         case SECCOMP_MODE_STRICT:
 776                 __secure_computing_strict(this_syscall);  /* may call do_exit */
 777                 return 0;
 778         case SECCOMP_MODE_FILTER:
 779                 return __seccomp_filter(this_syscall, sd, false);
 780         default:
 781                 BUG();
 782         }
 783 }
 784 #endif /* CONFIG_HAVE_ARCH_SECCOMP_FILTER */
 785
 786 long prctl_get_seccomp(void)
 787 {
 788         return current->seccomp.mode;
 789 }
 790
 791 /**
 792  * seccomp_set_mode_strict: internal function for setting strict seccomp
 793  *
 794  * Once current->seccomp.mode is non-zero, it may not be changed.
 795  *
 796  * Returns 0 on success or -EINVAL on failure.
 797  */
 798 static long seccomp_set_mode_strict(void)
 799 {
 800         const unsigned long seccomp_mode = SECCOMP_MODE_STRICT;
 801         long ret = -EINVAL;
 802
 803         spin_lock_irq(&current->sighand->siglock);
 804
 805         if (!seccomp_may_assign_mode(seccomp_mode))
 806                 goto out;
 807
 808 #ifdef TIF_NOTSC
 809         disable_TSC();
 810 #endif
 811         seccomp_assign_mode(current, seccomp_mode);
 812         ret = 0;
 813
 814 out:
 815         spin_unlock_irq(&current->sighand->siglock);
 816
 817         return ret;
 818 }
 819
 820 #ifdef CONFIG_SECCOMP_FILTER
 821 /**
 822  * seccomp_set_mode_filter: internal function for setting seccomp filter
 823  * @flags:  flags to change filter behavior
 824  * @filter: struct sock_fprog containing filter
 825  *
 826  * This function may be called repeatedly to install additional filters.
 827  * Every filter successfully installed will be evaluated (in reverse order)
 828  * for each system call the task makes.
 829  *
 830  * Once current->seccomp.mode is non-zero, it may not be changed.
 831  *
 832  * Returns 0 on success or -EINVAL on failure.
 833  */
 834 static long seccomp_set_mode_filter(unsigned int flags,
 835                                     const char __user *filter)
 836 {
 837         const unsigned long seccomp_mode = SECCOMP_MODE_FILTER;
 838         struct seccomp_filter *prepared = NULL;
 839         long ret = -EINVAL;
 840
 841         /* Validate flags. */
 842         if (flags & ~SECCOMP_FILTER_FLAG_MASK)
 843                 return -EINVAL;
 844
 845         /* Prepare the new filter before holding any locks. */
 846         prepared = seccomp_prepare_user_filter(filter);
 847         if (IS_ERR(prepared))
 848                 return PTR_ERR(prepared);
 849
 850         /*
 851          * Make sure we cannot change seccomp or nnp state via TSYNC
 852          * while another thread is in the middle of calling exec.
 853          */
 854         if (flags & SECCOMP_FILTER_FLAG_TSYNC &&
 855             mutex_lock_killable(&current->signal->cred_guard_mutex))
 856                 goto out_free;
 857
 858         spin_lock_irq(&current->sighand->siglock);
 859
 860         if (!seccomp_may_assign_mode(seccomp_mode))
 861                 goto out;
 862
 863         ret = seccomp_attach_filter(flags, prepared);
 864         if (ret)
 865                 goto out;
 866         /* Do not free the successfully attached filter. */
 867         prepared = NULL;
 868
 869         seccomp_assign_mode(current, seccomp_mode);
 870 out:
 871         spin_unlock_irq(&current->sighand->siglock);
 872         if (flags & SECCOMP_FILTER_FLAG_TSYNC)
 873                 mutex_unlock(&current->signal->cred_guard_mutex);
 874 out_free:
 875         seccomp_filter_free(prepared);
 876         return ret;
 877 }
 878 #else
 879 static inline long seccomp_set_mode_filter(unsigned int flags,
 880                                            const char __user *filter)
 881 {
 882         return -EINVAL;
 883 }
 884 #endif
 885
 886 static long seccomp_get_action_avail(const char __user *uaction)
 887 {
 888         u32 action;
 889
 890         if (copy_from_user(&action, uaction, sizeof(action)))
 891                 return -EFAULT;
 892
 893         switch (action) {
 894         case SECCOMP_RET_KILL_PROCESS:
 895         case SECCOMP_RET_KILL_THREAD:
 896         case SECCOMP_RET_TRAP:
 897         case SECCOMP_RET_ERRNO:
 898         case SECCOMP_RET_TRACE:
 899         case SECCOMP_RET_LOG:
 900         case SECCOMP_RET_ALLOW:
 901                 break;
 902         default:
 903                 return -EOPNOTSUPP;
 904         }
 905
 906         return 0;
 907 }
 908
 909 /* Common entry point for both prctl and syscall. */
 910 static long do_seccomp(unsigned int op, unsigned int flags,
 911                        const char __user *uargs)
 912 {
 913         switch (op) {
 914         case SECCOMP_SET_MODE_STRICT:
 915                 if (flags != 0 || uargs != NULL)
 916                         return -EINVAL;
 917                 return seccomp_set_mode_strict();
 918         case SECCOMP_SET_MODE_FILTER:
 919                 return seccomp_set_mode_filter(flags, uargs);
 920         case SECCOMP_GET_ACTION_AVAIL:
 921                 if (flags != 0)
 922                         return -EINVAL;
 923
 924                 return seccomp_get_action_avail(uargs);
 925         default:
 926                 return -EINVAL;
 927         }
 928 }
 929
 930 SYSCALL_DEFINE3(seccomp, unsigned int, op, unsigned int, flags,
 931                          const char __user *, uargs)
 932 {
 933         return do_seccomp(op, flags, uargs);
 934 }
 935
 936 /**
 937  * prctl_set_seccomp: configures current->seccomp.mode
 938  * @seccomp_mode: requested mode to use
 939  * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER
 940  *
 941  * Returns 0 on success or -EINVAL on failure.
 942  */
 943 long prctl_set_seccomp(unsigned long seccomp_mode, char __user *filter)
 944 {
 945         unsigned int op;
 946         char __user *uargs;
 947
 948         switch (seccomp_mode) {
 949         case SECCOMP_MODE_STRICT:
 950                 op = SECCOMP_SET_MODE_STRICT;
 951                 /*
 952                  * Setting strict mode through prctl always ignored filter,
 953                  * so make sure it is always NULL here to pass the internal
 954                  * check in do_seccomp().
 955                  */
 956                 uargs = NULL;
 957                 break;
 958         case SECCOMP_MODE_FILTER:
 959                 op = SECCOMP_SET_MODE_FILTER;
 960                 uargs = filter;
 961                 break;
 962         default:
 963                 return -EINVAL;
 964         }
 965
 966         /* prctl interface doesn't have flags, so they are always zero. */
 967         return do_seccomp(op, 0, uargs);
 968 }
 969
 970 #if defined(CONFIG_SECCOMP_FILTER) && defined(CONFIG_CHECKPOINT_RESTORE)
 971 long seccomp_get_filter(struct task_struct *task, unsigned long filter_off,
 972                         void __user *data)
 973 {
 974         struct seccomp_filter *filter;
 975         struct sock_fprog_kern *fprog;
 976         long ret;
 977         unsigned long count = 0;
 978
 979         if (!capable(CAP_SYS_ADMIN) ||
 980             current->seccomp.mode != SECCOMP_MODE_DISABLED) {
 981                 return -EACCES;
 982         }
 983
 984         spin_lock_irq(&task->sighand->siglock);
 985         if (task->seccomp.mode != SECCOMP_MODE_FILTER) {
 986                 ret = -EINVAL;
 987                 goto out;
 988         }
 989
 990         filter = task->seccomp.filter;
 991         while (filter) {
 992                 filter = filter->prev;
 993                 count++;
 994         }
 995
 996         if (filter_off >= count) {
 997                 ret = -ENOENT;
 998                 goto out;
 999         }
1000         count -= filter_off;
1001
1002         filter = task->seccomp.filter;
1003         while (filter && count > 1) {
1004                 filter = filter->prev;
1005                 count--;
1006         }
1007
1008         if (WARN_ON(count != 1 || !filter)) {
1009                 /* The filter tree shouldn't shrink while we're using it. */
1010                 ret = -ENOENT;
1011                 goto out;
1012         }
1013
1014         fprog = filter->prog->orig_prog;
1015         if (!fprog) {
1016                 /* This must be a new non-cBPF filter, since we save
1017                  * every cBPF filter's orig_prog above when
1018                  * CONFIG_CHECKPOINT_RESTORE is enabled.
1019                  */
1020                 ret = -EMEDIUMTYPE;
1021                 goto out;
1022         }
1023
1024         ret = fprog->len;
1025         if (!data)
1026                 goto out;
1027
1028         get_seccomp_filter(task);
1029         spin_unlock_irq(&task->sighand->siglock);
1030
1031         if (copy_to_user(data, fprog->filter, bpf_classic_proglen(fprog)))
1032                 ret = -EFAULT;
1033
1034         put_seccomp_filter(task);
1035         return ret;
1036
1037 out:
1038         spin_unlock_irq(&task->sighand->siglock);
1039         return ret;
1040 }
1041 #endif
1042
1043 #ifdef CONFIG_SYSCTL
1044
1045 /* Human readable action names for friendly sysctl interaction */
1046 #define SECCOMP_RET_KILL_PROCESS_NAME   "kill_process"
1047 #define SECCOMP_RET_KILL_THREAD_NAME    "kill_thread"
1048 #define SECCOMP_RET_TRAP_NAME           "trap"
1049 #define SECCOMP_RET_ERRNO_NAME          "errno"
1050 #define SECCOMP_RET_TRACE_NAME          "trace"
1051 #define SECCOMP_RET_LOG_NAME            "log"
1052 #define SECCOMP_RET_ALLOW_NAME          "allow"
1053
1054 static const char seccomp_actions_avail[] =
1055                                 SECCOMP_RET_KILL_PROCESS_NAME   " "
1056                                 SECCOMP_RET_KILL_THREAD_NAME    " "
1057                                 SECCOMP_RET_TRAP_NAME           " "
1058                                 SECCOMP_RET_ERRNO_NAME          " "
1059                                 SECCOMP_RET_TRACE_NAME          " "
1060                                 SECCOMP_RET_LOG_NAME            " "
1061                                 SECCOMP_RET_ALLOW_NAME;
1062
1063 struct seccomp_log_name {
1064         u32             log;
1065         const char      *name;
1066 };
1067
1068 static const struct seccomp_log_name seccomp_log_names[] = {
1069         { SECCOMP_LOG_KILL_PROCESS, SECCOMP_RET_KILL_PROCESS_NAME },
1070         { SECCOMP_LOG_KILL_THREAD, SECCOMP_RET_KILL_THREAD_NAME },
1071         { SECCOMP_LOG_TRAP, SECCOMP_RET_TRAP_NAME },
1072         { SECCOMP_LOG_ERRNO, SECCOMP_RET_ERRNO_NAME },
1073         { SECCOMP_LOG_TRACE, SECCOMP_RET_TRACE_NAME },
1074         { SECCOMP_LOG_LOG, SECCOMP_RET_LOG_NAME },
1075         { SECCOMP_LOG_ALLOW, SECCOMP_RET_ALLOW_NAME },
1076         { }
1077 };
1078
1079 static bool seccomp_names_from_actions_logged(char *names, size_t size,
1080                                               u32 actions_logged)
1081 {
1082         const struct seccomp_log_name *cur;
1083         bool append_space = false;
1084
1085         for (cur = seccomp_log_names; cur->name && size; cur++) {
1086                 ssize_t ret;
1087
1088                 if (!(actions_logged & cur->log))
1089                         continue;
1090
1091                 if (append_space) {
1092                         ret = strscpy(names, " ", size);
1093                         if (ret < 0)
1094                                 return false;
1095
1096                         names += ret;
1097                         size -= ret;
1098                 } else
1099                         append_space = true;
1100
1101                 ret = strscpy(names, cur->name, size);
1102                 if (ret < 0)
1103                         return false;
1104
1105                 names += ret;
1106                 size -= ret;
1107         }
1108
1109         return true;
1110 }
1111
1112 static bool seccomp_action_logged_from_name(u32 *action_logged,
1113                                             const char *name)
1114 {
1115         const struct seccomp_log_name *cur;
1116
1117         for (cur = seccomp_log_names; cur->name; cur++) {
1118                 if (!strcmp(cur->name, name)) {
1119                         *action_logged = cur->log;
1120                         return true;
1121                 }
1122         }
1123
1124         return false;
1125 }
1126
1127 static bool seccomp_actions_logged_from_names(u32 *actions_logged, char *names)
1128 {
1129         char *name;
1130
1131         *actions_logged = 0;
1132         while ((name = strsep(&names, " ")) && *name) {
1133                 u32 action_logged = 0;
1134
1135                 if (!seccomp_action_logged_from_name(&action_logged, name))
1136                         return false;
1137
1138                 *actions_logged |= action_logged;
1139         }
1140
1141         return true;
1142 }
1143
1144 static int seccomp_actions_logged_handler(struct ctl_table *ro_table, int write,
1145                                           void __user *buffer, size_t *lenp,
1146                                           loff_t *ppos)
1147 {
1148         char names[sizeof(seccomp_actions_avail)];
1149         struct ctl_table table;
1150         int ret;
1151
1152         if (write && !capable(CAP_SYS_ADMIN))
1153                 return -EPERM;
1154
1155         memset(names, 0, sizeof(names));
1156
1157         if (!write) {
1158                 if (!seccomp_names_from_actions_logged(names, sizeof(names),
1159                                                        seccomp_actions_logged))
1160                         return -EINVAL;
1161         }
1162
1163         table = *ro_table;
1164         table.data = names;
1165         table.maxlen = sizeof(names);
1166         ret = proc_dostring(&table, write, buffer, lenp, ppos);
1167         if (ret)
1168                 return ret;
1169
1170         if (write) {
1171                 u32 actions_logged;
1172
1173                 if (!seccomp_actions_logged_from_names(&actions_logged,
1174                                                        table.data))
1175                         return -EINVAL;
1176
1177                 if (actions_logged & SECCOMP_LOG_ALLOW)
1178                         return -EINVAL;
1179
1180                 seccomp_actions_logged = actions_logged;
1181         }
1182
1183         return 0;
1184 }
1185
1186 static struct ctl_path seccomp_sysctl_path[] = {
1187         { .procname = "kernel", },
1188         { .procname = "seccomp", },
1189         { }
1190 };
1191
1192 static struct ctl_table seccomp_sysctl_table[] = {
1193         {
1194                 .procname       = "actions_avail",
1195                 .data           = (void *) &seccomp_actions_avail,
1196                 .maxlen         = sizeof(seccomp_actions_avail),
1197                 .mode           = 0444,
1198                 .proc_handler   = proc_dostring,
1199         },
1200         {
1201                 .procname       = "actions_logged",
1202                 .mode           = 0644,
1203                 .proc_handler   = seccomp_actions_logged_handler,
1204         },
1205         { }
1206 };
1207
1208 static int __init seccomp_sysctl_init(void)
1209 {
1210         struct ctl_table_header *hdr;
1211
1212         hdr = register_sysctl_paths(seccomp_sysctl_path, seccomp_sysctl_table);
1213         if (!hdr)
1214                 pr_warn("seccomp: sysctl registration failed\n");
1215         else
1216                 kmemleak_not_leak(hdr);
1217
1218         return 0;
1219 }
1220
1221 device_initcall(seccomp_sysctl_init)
1222
1223 #endif /* CONFIG_SYSCTL */