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/sched.h>
  21 #include <linux/sched/task_stack.h>
  22 #include <linux/seccomp.h>
  23 #include <linux/slab.h>
  24 #include <linux/syscalls.h>
  25
  26 #ifdef CONFIG_HAVE_ARCH_SECCOMP_FILTER
  27 #include <asm/syscall.h>
  28 #endif
  29
  30 #ifdef CONFIG_SECCOMP_FILTER
  31 #include <linux/filter.h>
  32 #include <linux/pid.h>
  33 #include <linux/ptrace.h>
  34 #include <linux/security.h>
  35 #include <linux/tracehook.h>
  36 #include <linux/uaccess.h>
  37
  38 /**
  39  * struct seccomp_filter - container for seccomp BPF programs
  40  *
  41  * @usage: reference count to manage the object lifetime.
  42  *         get/put helpers should be used when accessing an instance
  43  *         outside of a lifetime-guarded section.  In general, this
  44  *         is only needed for handling filters shared across tasks.
  45  * @prev: points to a previously installed, or inherited, filter
  46  * @prog: the BPF program to evaluate
  47  *
  48  * seccomp_filter objects are organized in a tree linked via the @prev
  49  * pointer.  For any task, it appears to be a singly-linked list starting
  50  * with current->seccomp.filter, the most recently attached or inherited filter.
  51  * However, multiple filters may share a @prev node, by way of fork(), which
  52  * results in a unidirectional tree existing in memory.  This is similar to
  53  * how namespaces work.
  54  *
  55  * seccomp_filter objects should never be modified after being attached
  56  * to a task_struct (other than @usage).
  57  */
  58 struct seccomp_filter {
  59         refcount_t usage;
  60         struct seccomp_filter *prev;
  61         struct bpf_prog *prog;
  62 };
  63
  64 /* Limit any path through the tree to 256KB worth of instructions. */
  65 #define MAX_INSNS_PER_PATH ((1 << 18) / sizeof(struct sock_filter))
  66
  67 /*
  68  * Endianness is explicitly ignored and left for BPF program authors to manage
  69  * as per the specific architecture.
  70  */
  71 static void populate_seccomp_data(struct seccomp_data *sd)
  72 {
  73         struct task_struct *task = current;
  74         struct pt_regs *regs = task_pt_regs(task);
  75         unsigned long args[6];
  76
  77         sd->nr = syscall_get_nr(task, regs);
  78         sd->arch = syscall_get_arch();
  79         syscall_get_arguments(task, regs, 0, 6, args);
  80         sd->args[0] = args[0];
  81         sd->args[1] = args[1];
  82         sd->args[2] = args[2];
  83         sd->args[3] = args[3];
  84         sd->args[4] = args[4];
  85         sd->args[5] = args[5];
  86         sd->instruction_pointer = KSTK_EIP(task);
  87 }
  88
  89 /**
  90  *      seccomp_check_filter - verify seccomp filter code
  91  *      @filter: filter to verify
  92  *      @flen: length of filter
  93  *
  94  * Takes a previously checked filter (by bpf_check_classic) and
  95  * redirects all filter code that loads struct sk_buff data
  96  * and related data through seccomp_bpf_load.  It also
  97  * enforces length and alignment checking of those loads.
  98  *
  99  * Returns 0 if the rule set is legal or -EINVAL if not.
 100  */
 101 static int seccomp_check_filter(struct sock_filter *filter, unsigned int flen)
 102 {
 103         int pc;
 104         for (pc = 0; pc < flen; pc++) {
 105                 struct sock_filter *ftest = &filter[pc];
 106                 u16 code = ftest->code;
 107                 u32 k = ftest->k;
 108
 109                 switch (code) {
 110                 case BPF_LD | BPF_W | BPF_ABS:
 111                         ftest->code = BPF_LDX | BPF_W | BPF_ABS;
 112                         /* 32-bit aligned and not out of bounds. */
 113                         if (k >= sizeof(struct seccomp_data) || k & 3)
 114                                 return -EINVAL;
 115                         continue;
 116                 case BPF_LD | BPF_W | BPF_LEN:
 117                         ftest->code = BPF_LD | BPF_IMM;
 118                         ftest->k = sizeof(struct seccomp_data);
 119                         continue;
 120                 case BPF_LDX | BPF_W | BPF_LEN:
 121                         ftest->code = BPF_LDX | BPF_IMM;
 122                         ftest->k = sizeof(struct seccomp_data);
 123                         continue;
 124                 /* Explicitly include allowed calls. */
 125                 case BPF_RET | BPF_K:
 126                 case BPF_RET | BPF_A:
 127                 case BPF_ALU | BPF_ADD | BPF_K:
 128                 case BPF_ALU | BPF_ADD | BPF_X:
 129                 case BPF_ALU | BPF_SUB | BPF_K:
 130                 case BPF_ALU | BPF_SUB | BPF_X:
 131                 case BPF_ALU | BPF_MUL | BPF_K:
 132                 case BPF_ALU | BPF_MUL | BPF_X:
 133                 case BPF_ALU | BPF_DIV | BPF_K:
 134                 case BPF_ALU | BPF_DIV | BPF_X:
 135                 case BPF_ALU | BPF_AND | BPF_K:
 136                 case BPF_ALU | BPF_AND | BPF_X:
 137                 case BPF_ALU | BPF_OR | BPF_K:
 138                 case BPF_ALU | BPF_OR | BPF_X:
 139                 case BPF_ALU | BPF_XOR | BPF_K:
 140                 case BPF_ALU | BPF_XOR | BPF_X:
 141                 case BPF_ALU | BPF_LSH | BPF_K:
 142                 case BPF_ALU | BPF_LSH | BPF_X:
 143                 case BPF_ALU | BPF_RSH | BPF_K:
 144                 case BPF_ALU | BPF_RSH | BPF_X:
 145                 case BPF_ALU | BPF_NEG:
 146                 case BPF_LD | BPF_IMM:
 147                 case BPF_LDX | BPF_IMM:
 148                 case BPF_MISC | BPF_TAX:
 149                 case BPF_MISC | BPF_TXA:
 150                 case BPF_LD | BPF_MEM:
 151                 case BPF_LDX | BPF_MEM:
 152                 case BPF_ST:
 153                 case BPF_STX:
 154                 case BPF_JMP | BPF_JA:
 155                 case BPF_JMP | BPF_JEQ | BPF_K:
 156                 case BPF_JMP | BPF_JEQ | BPF_X:
 157                 case BPF_JMP | BPF_JGE | BPF_K:
 158                 case BPF_JMP | BPF_JGE | BPF_X:
 159                 case BPF_JMP | BPF_JGT | BPF_K:
 160                 case BPF_JMP | BPF_JGT | BPF_X:
 161                 case BPF_JMP | BPF_JSET | BPF_K:
 162                 case BPF_JMP | BPF_JSET | BPF_X:
 163                         continue;
 164                 default:
 165                         return -EINVAL;
 166                 }
 167         }
 168         return 0;
 169 }
 170
 171 /**
 172  * seccomp_run_filters - evaluates all seccomp filters against @sd
 173  * @sd: optional seccomp data to be passed to filters
 174  *
 175  * Returns valid seccomp BPF response codes.
 176  */
 177 static u32 seccomp_run_filters(const struct seccomp_data *sd)
 178 {
 179         struct seccomp_data sd_local;
 180         u32 ret = SECCOMP_RET_ALLOW;
 181         /* Make sure cross-thread synced filter points somewhere sane. */
 182         struct seccomp_filter *f =
 183                         lockless_dereference(current->seccomp.filter);
 184
 185         /* Ensure unexpected behavior doesn't result in failing open. */
 186         if (unlikely(WARN_ON(f == NULL)))
 187                 return SECCOMP_RET_KILL;
 188
 189         if (!sd) {
 190                 populate_seccomp_data(&sd_local);
 191                 sd = &sd_local;
 192         }
 193
 194         /*
 195          * All filters in the list are evaluated and the lowest BPF return
 196          * value always takes priority (ignoring the DATA).
 197          */
 198         for (; f; f = f->prev) {
 199                 u32 cur_ret = BPF_PROG_RUN(f->prog, sd);
 200
 201                 if ((cur_ret & SECCOMP_RET_ACTION) < (ret & SECCOMP_RET_ACTION))
 202                         ret = cur_ret;
 203         }
 204         return ret;
 205 }
 206 #endif /* CONFIG_SECCOMP_FILTER */
 207
 208 static inline bool seccomp_may_assign_mode(unsigned long seccomp_mode)
 209 {
 210         assert_spin_locked(&current->sighand->siglock);
 211
 212         if (current->seccomp.mode && current->seccomp.mode != seccomp_mode)
 213                 return false;
 214
 215         return true;
 216 }
 217
 218 static inline void seccomp_assign_mode(struct task_struct *task,
 219                                        unsigned long seccomp_mode)
 220 {
 221         assert_spin_locked(&task->sighand->siglock);
 222
 223         task->seccomp.mode = seccomp_mode;
 224         /*
 225          * Make sure TIF_SECCOMP cannot be set before the mode (and
 226          * filter) is set.
 227          */
 228         smp_mb__before_atomic();
 229         set_tsk_thread_flag(task, TIF_SECCOMP);
 230 }
 231
 232 #ifdef CONFIG_SECCOMP_FILTER
 233 /* Returns 1 if the parent is an ancestor of the child. */
 234 static int is_ancestor(struct seccomp_filter *parent,
 235                        struct seccomp_filter *child)
 236 {
 237         /* NULL is the root ancestor. */
 238         if (parent == NULL)
 239                 return 1;
 240         for (; child; child = child->prev)
 241                 if (child == parent)
 242                         return 1;
 243         return 0;
 244 }
 245
 246 /**
 247  * seccomp_can_sync_threads: checks if all threads can be synchronized
 248  *
 249  * Expects sighand and cred_guard_mutex locks to be held.
 250  *
 251  * Returns 0 on success, -ve on error, or the pid of a thread which was
 252  * either not in the correct seccomp mode or it did not have an ancestral
 253  * seccomp filter.
 254  */
 255 static inline pid_t seccomp_can_sync_threads(void)
 256 {
 257         struct task_struct *thread, *caller;
 258
 259         BUG_ON(!mutex_is_locked(&current->signal->cred_guard_mutex));
 260         assert_spin_locked(&current->sighand->siglock);
 261
 262         /* Validate all threads being eligible for synchronization. */
 263         caller = current;
 264         for_each_thread(caller, thread) {
 265                 pid_t failed;
 266
 267                 /* Skip current, since it is initiating the sync. */
 268                 if (thread == caller)
 269                         continue;
 270
 271                 if (thread->seccomp.mode == SECCOMP_MODE_DISABLED ||
 272                     (thread->seccomp.mode == SECCOMP_MODE_FILTER &&
 273                      is_ancestor(thread->seccomp.filter,
 274                                  caller->seccomp.filter)))
 275                         continue;
 276
 277                 /* Return the first thread that cannot be synchronized. */
 278                 failed = task_pid_vnr(thread);
 279                 /* If the pid cannot be resolved, then return -ESRCH */
 280                 if (unlikely(WARN_ON(failed == 0)))
 281                         failed = -ESRCH;
 282                 return failed;
 283         }
 284
 285         return 0;
 286 }
 287
 288 /**
 289  * seccomp_sync_threads: sets all threads to use current's filter
 290  *
 291  * Expects sighand and cred_guard_mutex locks to be held, and for
 292  * seccomp_can_sync_threads() to have returned success already
 293  * without dropping the locks.
 294  *
 295  */
 296 static inline void seccomp_sync_threads(void)
 297 {
 298         struct task_struct *thread, *caller;
 299
 300         BUG_ON(!mutex_is_locked(&current->signal->cred_guard_mutex));
 301         assert_spin_locked(&current->sighand->siglock);
 302
 303         /* Synchronize all threads. */
 304         caller = current;
 305         for_each_thread(caller, thread) {
 306                 /* Skip current, since it needs no changes. */
 307                 if (thread == caller)
 308                         continue;
 309
 310                 /* Get a task reference for the new leaf node. */
 311                 get_seccomp_filter(caller);
 312                 /*
 313                  * Drop the task reference to the shared ancestor since
 314                  * current's path will hold a reference.  (This also
 315                  * allows a put before the assignment.)
 316                  */
 317                 put_seccomp_filter(thread);
 318                 smp_store_release(&thread->seccomp.filter,
 319                                   caller->seccomp.filter);
 320
 321                 /*
 322                  * Don't let an unprivileged task work around
 323                  * the no_new_privs restriction by creating
 324                  * a thread that sets it up, enters seccomp,
 325                  * then dies.
 326                  */
 327                 if (task_no_new_privs(caller))
 328                         task_set_no_new_privs(thread);
 329
 330                 /*
 331                  * Opt the other thread into seccomp if needed.
 332                  * As threads are considered to be trust-realm
 333                  * equivalent (see ptrace_may_access), it is safe to
 334                  * allow one thread to transition the other.
 335                  */
 336                 if (thread->seccomp.mode == SECCOMP_MODE_DISABLED)
 337                         seccomp_assign_mode(thread, SECCOMP_MODE_FILTER);
 338         }
 339 }
 340
 341 /**
 342  * seccomp_prepare_filter: Prepares a seccomp filter for use.
 343  * @fprog: BPF program to install
 344  *
 345  * Returns filter on success or an ERR_PTR on failure.
 346  */
 347 static struct seccomp_filter *seccomp_prepare_filter(struct sock_fprog *fprog)
 348 {
 349         struct seccomp_filter *sfilter;
 350         int ret;
 351         const bool save_orig = IS_ENABLED(CONFIG_CHECKPOINT_RESTORE);
 352
 353         if (fprog->len == 0 || fprog->len > BPF_MAXINSNS)
 354                 return ERR_PTR(-EINVAL);
 355
 356         BUG_ON(INT_MAX / fprog->len < sizeof(struct sock_filter));
 357
 358         /*
 359          * Installing a seccomp filter requires that the task has
 360          * CAP_SYS_ADMIN in its namespace or be running with no_new_privs.
 361          * This avoids scenarios where unprivileged tasks can affect the
 362          * behavior of privileged children.
 363          */
 364         if (!task_no_new_privs(current) &&
 365             security_capable_noaudit(current_cred(), current_user_ns(),
 366                                      CAP_SYS_ADMIN) != 0)
 367                 return ERR_PTR(-EACCES);
 368
 369         /* Allocate a new seccomp_filter */
 370         sfilter = kzalloc(sizeof(*sfilter), GFP_KERNEL | __GFP_NOWARN);
 371         if (!sfilter)
 372                 return ERR_PTR(-ENOMEM);
 373
 374         ret = bpf_prog_create_from_user(&sfilter->prog, fprog,
 375                                         seccomp_check_filter, save_orig);
 376         if (ret < 0) {
 377                 kfree(sfilter);
 378                 return ERR_PTR(ret);
 379         }
 380
 381         refcount_set(&sfilter->usage, 1);
 382
 383         return sfilter;
 384 }
 385
 386 /**
 387  * seccomp_prepare_user_filter - prepares a user-supplied sock_fprog
 388  * @user_filter: pointer to the user data containing a sock_fprog.
 389  *
 390  * Returns 0 on success and non-zero otherwise.
 391  */
 392 static struct seccomp_filter *
 393 seccomp_prepare_user_filter(const char __user *user_filter)
 394 {
 395         struct sock_fprog fprog;
 396         struct seccomp_filter *filter = ERR_PTR(-EFAULT);
 397
 398 #ifdef CONFIG_COMPAT
 399         if (in_compat_syscall()) {
 400                 struct compat_sock_fprog fprog32;
 401                 if (copy_from_user(&fprog32, user_filter, sizeof(fprog32)))
 402                         goto out;
 403                 fprog.len = fprog32.len;
 404                 fprog.filter = compat_ptr(fprog32.filter);
 405         } else /* falls through to the if below. */
 406 #endif
 407         if (copy_from_user(&fprog, user_filter, sizeof(fprog)))
 408                 goto out;
 409         filter = seccomp_prepare_filter(&fprog);
 410 out:
 411         return filter;
 412 }
 413
 414 /**
 415  * seccomp_attach_filter: validate and attach filter
 416  * @flags:  flags to change filter behavior
 417  * @filter: seccomp filter to add to the current process
 418  *
 419  * Caller must be holding current->sighand->siglock lock.
 420  *
 421  * Returns 0 on success, -ve on error.
 422  */
 423 static long seccomp_attach_filter(unsigned int flags,
 424                                   struct seccomp_filter *filter)
 425 {
 426         unsigned long total_insns;
 427         struct seccomp_filter *walker;
 428
 429         assert_spin_locked(&current->sighand->siglock);
 430
 431         /* Validate resulting filter length. */
 432         total_insns = filter->prog->len;
 433         for (walker = current->seccomp.filter; walker; walker = walker->prev)
 434                 total_insns += walker->prog->len + 4;  /* 4 instr penalty */
 435         if (total_insns > MAX_INSNS_PER_PATH)
 436                 return -ENOMEM;
 437
 438         /* If thread sync has been requested, check that it is possible. */
 439         if (flags & SECCOMP_FILTER_FLAG_TSYNC) {
 440                 int ret;
 441
 442                 ret = seccomp_can_sync_threads();
 443                 if (ret)
 444                         return ret;
 445         }
 446
 447         /*
 448          * If there is an existing filter, make it the prev and don't drop its
 449          * task reference.
 450          */
 451         filter->prev = current->seccomp.filter;
 452         current->seccomp.filter = filter;
 453
 454         /* Now that the new filter is in place, synchronize to all threads. */
 455         if (flags & SECCOMP_FILTER_FLAG_TSYNC)
 456                 seccomp_sync_threads();
 457
 458         return 0;
 459 }
 460
 461 /* get_seccomp_filter - increments the reference count of the filter on @tsk */
 462 void get_seccomp_filter(struct task_struct *tsk)
 463 {
 464         struct seccomp_filter *orig = tsk->seccomp.filter;
 465         if (!orig)
 466                 return;
 467         /* Reference count is bounded by the number of total processes. */
 468         refcount_inc(&orig->usage);
 469 }
 470
 471 static inline void seccomp_filter_free(struct seccomp_filter *filter)
 472 {
 473         if (filter) {
 474                 bpf_prog_destroy(filter->prog);
 475                 kfree(filter);
 476         }
 477 }
 478
 479 /* put_seccomp_filter - decrements the ref count of tsk->seccomp.filter */
 480 void put_seccomp_filter(struct task_struct *tsk)
 481 {
 482         struct seccomp_filter *orig = tsk->seccomp.filter;
 483         /* Clean up single-reference branches iteratively. */
 484         while (orig && refcount_dec_and_test(&orig->usage)) {
 485                 struct seccomp_filter *freeme = orig;
 486                 orig = orig->prev;
 487                 seccomp_filter_free(freeme);
 488         }
 489 }
 490
 491 static void seccomp_init_siginfo(siginfo_t *info, int syscall, int reason)
 492 {
 493         memset(info, 0, sizeof(*info));
 494         info->si_signo = SIGSYS;
 495         info->si_code = SYS_SECCOMP;
 496         info->si_call_addr = (void __user *)KSTK_EIP(current);
 497         info->si_errno = reason;
 498         info->si_arch = syscall_get_arch();
 499         info->si_syscall = syscall;
 500 }
 501
 502 /**
 503  * seccomp_send_sigsys - signals the task to allow in-process syscall emulation
 504  * @syscall: syscall number to send to userland
 505  * @reason: filter-supplied reason code to send to userland (via si_errno)
 506  *
 507  * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info.
 508  */
 509 static void seccomp_send_sigsys(int syscall, int reason)
 510 {
 511         struct siginfo info;
 512         seccomp_init_siginfo(&info, syscall, reason);
 513         force_sig_info(SIGSYS, &info, current);
 514 }
 515 #endif  /* CONFIG_SECCOMP_FILTER */
 516
 517 /*
 518  * Secure computing mode 1 allows only read/write/exit/sigreturn.
 519  * To be fully secure this must be combined with rlimit
 520  * to limit the stack allocations too.
 521  */
 522 static const int mode1_syscalls[] = {
 523         __NR_seccomp_read, __NR_seccomp_write, __NR_seccomp_exit, __NR_seccomp_sigreturn,
 524         0, /* null terminated */
 525 };
 526
 527 static void __secure_computing_strict(int this_syscall)
 528 {
 529         const int *syscall_whitelist = mode1_syscalls;
 530 #ifdef CONFIG_COMPAT
 531         if (in_compat_syscall())
 532                 syscall_whitelist = get_compat_mode1_syscalls();
 533 #endif
 534         do {
 535                 if (*syscall_whitelist == this_syscall)
 536                         return;
 537         } while (*++syscall_whitelist);
 538
 539 #ifdef SECCOMP_DEBUG
 540         dump_stack();
 541 #endif
 542         audit_seccomp(this_syscall, SIGKILL, SECCOMP_RET_KILL);
 543         do_exit(SIGKILL);
 544 }
 545
 546 #ifndef CONFIG_HAVE_ARCH_SECCOMP_FILTER
 547 void secure_computing_strict(int this_syscall)
 548 {
 549         int mode = current->seccomp.mode;
 550
 551         if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) &&
 552             unlikely(current->ptrace & PT_SUSPEND_SECCOMP))
 553                 return;
 554
 555         if (mode == SECCOMP_MODE_DISABLED)
 556                 return;
 557         else if (mode == SECCOMP_MODE_STRICT)
 558                 __secure_computing_strict(this_syscall);
 559         else
 560                 BUG();
 561 }
 562 #else
 563
 564 #ifdef CONFIG_SECCOMP_FILTER
 565 static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd,
 566                             const bool recheck_after_trace)
 567 {
 568         u32 filter_ret, action;
 569         int data;
 570
 571         /*
 572          * Make sure that any changes to mode from another thread have
 573          * been seen after TIF_SECCOMP was seen.
 574          */
 575         rmb();
 576
 577         filter_ret = seccomp_run_filters(sd);
 578         data = filter_ret & SECCOMP_RET_DATA;
 579         action = filter_ret & SECCOMP_RET_ACTION;
 580
 581         switch (action) {
 582         case SECCOMP_RET_ERRNO:
 583                 /* Set low-order bits as an errno, capped at MAX_ERRNO. */
 584                 if (data > MAX_ERRNO)
 585                         data = MAX_ERRNO;
 586                 syscall_set_return_value(current, task_pt_regs(current),
 587                                          -data, 0);
 588                 goto skip;
 589
 590         case SECCOMP_RET_TRAP:
 591                 /* Show the handler the original registers. */
 592                 syscall_rollback(current, task_pt_regs(current));
 593                 /* Let the filter pass back 16 bits of data. */
 594                 seccomp_send_sigsys(this_syscall, data);
 595                 goto skip;
 596
 597         case SECCOMP_RET_TRACE:
 598                 /* We've been put in this state by the ptracer already. */
 599                 if (recheck_after_trace)
 600                         return 0;
 601
 602                 /* ENOSYS these calls if there is no tracer attached. */
 603                 if (!ptrace_event_enabled(current, PTRACE_EVENT_SECCOMP)) {
 604                         syscall_set_return_value(current,
 605                                                  task_pt_regs(current),
 606                                                  -ENOSYS, 0);
 607                         goto skip;
 608                 }
 609
 610                 /* Allow the BPF to provide the event message */
 611                 ptrace_event(PTRACE_EVENT_SECCOMP, data);
 612                 /*
 613                  * The delivery of a fatal signal during event
 614                  * notification may silently skip tracer notification,
 615                  * which could leave us with a potentially unmodified
 616                  * syscall that the tracer would have liked to have
 617                  * changed. Since the process is about to die, we just
 618                  * force the syscall to be skipped and let the signal
 619                  * kill the process and correctly handle any tracer exit
 620                  * notifications.
 621                  */
 622                 if (fatal_signal_pending(current))
 623                         goto skip;
 624                 /* Check if the tracer forced the syscall to be skipped. */
 625                 this_syscall = syscall_get_nr(current, task_pt_regs(current));
 626                 if (this_syscall < 0)
 627                         goto skip;
 628
 629                 /*
 630                  * Recheck the syscall, since it may have changed. This
 631                  * intentionally uses a NULL struct seccomp_data to force
 632                  * a reload of all registers. This does not goto skip since
 633                  * a skip would have already been reported.
 634                  */
 635                 if (__seccomp_filter(this_syscall, NULL, true))
 636                         return -1;
 637
 638                 return 0;
 639
 640         case SECCOMP_RET_ALLOW:
 641                 return 0;
 642
 643         case SECCOMP_RET_KILL:
 644         default:
 645                 audit_seccomp(this_syscall, SIGSYS, action);
 646                 /* Dump core only if this is the last remaining thread. */
 647                 if (get_nr_threads(current) == 1) {
 648                         siginfo_t info;
 649
 650                         /* Show the original registers in the dump. */
 651                         syscall_rollback(current, task_pt_regs(current));
 652                         /* Trigger a manual coredump since do_exit skips it. */
 653                         seccomp_init_siginfo(&info, this_syscall, data);
 654                         do_coredump(&info);
 655                 }
 656                 do_exit(SIGSYS);
 657         }
 658
 659         unreachable();
 660
 661 skip:
 662         audit_seccomp(this_syscall, 0, action);
 663         return -1;
 664 }
 665 #else
 666 static int __seccomp_filter(int this_syscall, const struct seccomp_data *sd,
 667                             const bool recheck_after_trace)
 668 {
 669         BUG();
 670 }
 671 #endif
 672
 673 int __secure_computing(const struct seccomp_data *sd)
 674 {
 675         int mode = current->seccomp.mode;
 676         int this_syscall;
 677
 678         if (IS_ENABLED(CONFIG_CHECKPOINT_RESTORE) &&
 679             unlikely(current->ptrace & PT_SUSPEND_SECCOMP))
 680                 return 0;
 681
 682         this_syscall = sd ? sd->nr :
 683                 syscall_get_nr(current, task_pt_regs(current));
 684
 685         switch (mode) {
 686         case SECCOMP_MODE_STRICT:
 687                 __secure_computing_strict(this_syscall);  /* may call do_exit */
 688                 return 0;
 689         case SECCOMP_MODE_FILTER:
 690                 return __seccomp_filter(this_syscall, sd, false);
 691         default:
 692                 BUG();
 693         }
 694 }
 695 #endif /* CONFIG_HAVE_ARCH_SECCOMP_FILTER */
 696
 697 long prctl_get_seccomp(void)
 698 {
 699         return current->seccomp.mode;
 700 }
 701
 702 /**
 703  * seccomp_set_mode_strict: internal function for setting strict seccomp
 704  *
 705  * Once current->seccomp.mode is non-zero, it may not be changed.
 706  *
 707  * Returns 0 on success or -EINVAL on failure.
 708  */
 709 static long seccomp_set_mode_strict(void)
 710 {
 711         const unsigned long seccomp_mode = SECCOMP_MODE_STRICT;
 712         long ret = -EINVAL;
 713
 714         spin_lock_irq(&current->sighand->siglock);
 715
 716         if (!seccomp_may_assign_mode(seccomp_mode))
 717                 goto out;
 718
 719 #ifdef TIF_NOTSC
 720         disable_TSC();
 721 #endif
 722         seccomp_assign_mode(current, seccomp_mode);
 723         ret = 0;
 724
 725 out:
 726         spin_unlock_irq(&current->sighand->siglock);
 727
 728         return ret;
 729 }
 730
 731 #ifdef CONFIG_SECCOMP_FILTER
 732 /**
 733  * seccomp_set_mode_filter: internal function for setting seccomp filter
 734  * @flags:  flags to change filter behavior
 735  * @filter: struct sock_fprog containing filter
 736  *
 737  * This function may be called repeatedly to install additional filters.
 738  * Every filter successfully installed will be evaluated (in reverse order)
 739  * for each system call the task makes.
 740  *
 741  * Once current->seccomp.mode is non-zero, it may not be changed.
 742  *
 743  * Returns 0 on success or -EINVAL on failure.
 744  */
 745 static long seccomp_set_mode_filter(unsigned int flags,
 746                                     const char __user *filter)
 747 {
 748         const unsigned long seccomp_mode = SECCOMP_MODE_FILTER;
 749         struct seccomp_filter *prepared = NULL;
 750         long ret = -EINVAL;
 751
 752         /* Validate flags. */
 753         if (flags & ~SECCOMP_FILTER_FLAG_MASK)
 754                 return -EINVAL;
 755
 756         /* Prepare the new filter before holding any locks. */
 757         prepared = seccomp_prepare_user_filter(filter);
 758         if (IS_ERR(prepared))
 759                 return PTR_ERR(prepared);
 760
 761         /*
 762          * Make sure we cannot change seccomp or nnp state via TSYNC
 763          * while another thread is in the middle of calling exec.
 764          */
 765         if (flags & SECCOMP_FILTER_FLAG_TSYNC &&
 766             mutex_lock_killable(&current->signal->cred_guard_mutex))
 767                 goto out_free;
 768
 769         spin_lock_irq(&current->sighand->siglock);
 770
 771         if (!seccomp_may_assign_mode(seccomp_mode))
 772                 goto out;
 773
 774         ret = seccomp_attach_filter(flags, prepared);
 775         if (ret)
 776                 goto out;
 777         /* Do not free the successfully attached filter. */
 778         prepared = NULL;
 779
 780         seccomp_assign_mode(current, seccomp_mode);
 781 out:
 782         spin_unlock_irq(&current->sighand->siglock);
 783         if (flags & SECCOMP_FILTER_FLAG_TSYNC)
 784                 mutex_unlock(&current->signal->cred_guard_mutex);
 785 out_free:
 786         seccomp_filter_free(prepared);
 787         return ret;
 788 }
 789 #else
 790 static inline long seccomp_set_mode_filter(unsigned int flags,
 791                                            const char __user *filter)
 792 {
 793         return -EINVAL;
 794 }
 795 #endif
 796
 797 /* Common entry point for both prctl and syscall. */
 798 static long do_seccomp(unsigned int op, unsigned int flags,
 799                        const char __user *uargs)
 800 {
 801         switch (op) {
 802         case SECCOMP_SET_MODE_STRICT:
 803                 if (flags != 0 || uargs != NULL)
 804                         return -EINVAL;
 805                 return seccomp_set_mode_strict();
 806         case SECCOMP_SET_MODE_FILTER:
 807                 return seccomp_set_mode_filter(flags, uargs);
 808         default:
 809                 return -EINVAL;
 810         }
 811 }
 812
 813 SYSCALL_DEFINE3(seccomp, unsigned int, op, unsigned int, flags,
 814                          const char __user *, uargs)
 815 {
 816         return do_seccomp(op, flags, uargs);
 817 }
 818
 819 /**
 820  * prctl_set_seccomp: configures current->seccomp.mode
 821  * @seccomp_mode: requested mode to use
 822  * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER
 823  *
 824  * Returns 0 on success or -EINVAL on failure.
 825  */
 826 long prctl_set_seccomp(unsigned long seccomp_mode, char __user *filter)
 827 {
 828         unsigned int op;
 829         char __user *uargs;
 830
 831         switch (seccomp_mode) {
 832         case SECCOMP_MODE_STRICT:
 833                 op = SECCOMP_SET_MODE_STRICT;
 834                 /*
 835                  * Setting strict mode through prctl always ignored filter,
 836                  * so make sure it is always NULL here to pass the internal
 837                  * check in do_seccomp().
 838                  */
 839                 uargs = NULL;
 840                 break;
 841         case SECCOMP_MODE_FILTER:
 842                 op = SECCOMP_SET_MODE_FILTER;
 843                 uargs = filter;
 844                 break;
 845         default:
 846                 return -EINVAL;
 847         }
 848
 849         /* prctl interface doesn't have flags, so they are always zero. */
 850         return do_seccomp(op, 0, uargs);
 851 }
 852
 853 #if defined(CONFIG_SECCOMP_FILTER) && defined(CONFIG_CHECKPOINT_RESTORE)
 854 long seccomp_get_filter(struct task_struct *task, unsigned long filter_off,
 855                         void __user *data)
 856 {
 857         struct seccomp_filter *filter;
 858         struct sock_fprog_kern *fprog;
 859         long ret;
 860         unsigned long count = 0;
 861
 862         if (!capable(CAP_SYS_ADMIN) ||
 863             current->seccomp.mode != SECCOMP_MODE_DISABLED) {
 864                 return -EACCES;
 865         }
 866
 867         spin_lock_irq(&task->sighand->siglock);
 868         if (task->seccomp.mode != SECCOMP_MODE_FILTER) {
 869                 ret = -EINVAL;
 870                 goto out;
 871         }
 872
 873         filter = task->seccomp.filter;
 874         while (filter) {
 875                 filter = filter->prev;
 876                 count++;
 877         }
 878
 879         if (filter_off >= count) {
 880                 ret = -ENOENT;
 881                 goto out;
 882         }
 883         count -= filter_off;
 884
 885         filter = task->seccomp.filter;
 886         while (filter && count > 1) {
 887                 filter = filter->prev;
 888                 count--;
 889         }
 890
 891         if (WARN_ON(count != 1 || !filter)) {
 892                 /* The filter tree shouldn't shrink while we're using it. */
 893                 ret = -ENOENT;
 894                 goto out;
 895         }
 896
 897         fprog = filter->prog->orig_prog;
 898         if (!fprog) {
 899                 /* This must be a new non-cBPF filter, since we save
 900                  * every cBPF filter's orig_prog above when
 901                  * CONFIG_CHECKPOINT_RESTORE is enabled.
 902                  */
 903                 ret = -EMEDIUMTYPE;
 904                 goto out;
 905         }
 906
 907         ret = fprog->len;
 908         if (!data)
 909                 goto out;
 910
 911         get_seccomp_filter(task);
 912         spin_unlock_irq(&task->sighand->siglock);
 913
 914         if (copy_to_user(data, fprog->filter, bpf_classic_proglen(fprog)))
 915                 ret = -EFAULT;
 916
 917         put_seccomp_filter(task);
 918         return ret;
 919
 920 out:
 921         spin_unlock_irq(&task->sighand->siglock);
 922         return ret;
 923 }
 924 #endif