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/atomic.h>
  17 #include <linux/audit.h>
  18 #include <linux/compat.h>
  19 #include <linux/nospec.h>
  20 #include <linux/prctl.h>
  21 #include <linux/sched.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  * @len: the number of instructions in the program
  47  * @insnsi: the BPF program instructions to evaluate
  48  *
  49  * seccomp_filter objects are organized in a tree linked via the @prev
  50  * pointer.  For any task, it appears to be a singly-linked list starting
  51  * with current->seccomp.filter, the most recently attached or inherited filter.
  52  * However, multiple filters may share a @prev node, by way of fork(), which
  53  * results in a unidirectional tree existing in memory.  This is similar to
  54  * how namespaces work.
  55  *
  56  * seccomp_filter objects should never be modified after being attached
  57  * to a task_struct (other than @usage).
  58  */
  59 struct seccomp_filter {
  60         atomic_t usage;
  61         struct seccomp_filter *prev;
  62         struct bpf_prog *prog;
  63 };
  64
  65 /* Limit any path through the tree to 256KB worth of instructions. */
  66 #define MAX_INSNS_PER_PATH ((1 << 18) / sizeof(struct sock_filter))
  67
  68 /*
  69  * Endianness is explicitly ignored and left for BPF program authors to manage
  70  * as per the specific architecture.
  71  */
  72 static void populate_seccomp_data(struct seccomp_data *sd)
  73 {
  74         struct task_struct *task = current;
  75         struct pt_regs *regs = task_pt_regs(task);
  76         unsigned long args[6];
  77
  78         sd->nr = syscall_get_nr(task, regs);
  79         sd->arch = syscall_get_arch();
  80         syscall_get_arguments(task, regs, 0, 6, args);
  81         sd->args[0] = args[0];
  82         sd->args[1] = args[1];
  83         sd->args[2] = args[2];
  84         sd->args[3] = args[3];
  85         sd->args[4] = args[4];
  86         sd->args[5] = args[5];
  87         sd->instruction_pointer = KSTK_EIP(task);
  88 }
  89
  90 /**
  91  *      seccomp_check_filter - verify seccomp filter code
  92  *      @filter: filter to verify
  93  *      @flen: length of filter
  94  *
  95  * Takes a previously checked filter (by bpf_check_classic) and
  96  * redirects all filter code that loads struct sk_buff data
  97  * and related data through seccomp_bpf_load.  It also
  98  * enforces length and alignment checking of those loads.
  99  *
 100  * Returns 0 if the rule set is legal or -EINVAL if not.
 101  */
 102 static int seccomp_check_filter(struct sock_filter *filter, unsigned int flen)
 103 {
 104         int pc;
 105         for (pc = 0; pc < flen; pc++) {
 106                 struct sock_filter *ftest = &filter[pc];
 107                 u16 code = ftest->code;
 108                 u32 k = ftest->k;
 109
 110                 switch (code) {
 111                 case BPF_LD | BPF_W | BPF_ABS:
 112                         ftest->code = BPF_LDX | BPF_W | BPF_ABS;
 113                         /* 32-bit aligned and not out of bounds. */
 114                         if (k >= sizeof(struct seccomp_data) || k & 3)
 115                                 return -EINVAL;
 116                         continue;
 117                 case BPF_LD | BPF_W | BPF_LEN:
 118                         ftest->code = BPF_LD | BPF_IMM;
 119                         ftest->k = sizeof(struct seccomp_data);
 120                         continue;
 121                 case BPF_LDX | BPF_W | BPF_LEN:
 122                         ftest->code = BPF_LDX | BPF_IMM;
 123                         ftest->k = sizeof(struct seccomp_data);
 124                         continue;
 125                 /* Explicitly include allowed calls. */
 126                 case BPF_RET | BPF_K:
 127                 case BPF_RET | BPF_A:
 128                 case BPF_ALU | BPF_ADD | BPF_K:
 129                 case BPF_ALU | BPF_ADD | BPF_X:
 130                 case BPF_ALU | BPF_SUB | BPF_K:
 131                 case BPF_ALU | BPF_SUB | BPF_X:
 132                 case BPF_ALU | BPF_MUL | BPF_K:
 133                 case BPF_ALU | BPF_MUL | BPF_X:
 134                 case BPF_ALU | BPF_DIV | BPF_K:
 135                 case BPF_ALU | BPF_DIV | BPF_X:
 136                 case BPF_ALU | BPF_AND | BPF_K:
 137                 case BPF_ALU | BPF_AND | BPF_X:
 138                 case BPF_ALU | BPF_OR | BPF_K:
 139                 case BPF_ALU | BPF_OR | BPF_X:
 140                 case BPF_ALU | BPF_XOR | BPF_K:
 141                 case BPF_ALU | BPF_XOR | BPF_X:
 142                 case BPF_ALU | BPF_LSH | BPF_K:
 143                 case BPF_ALU | BPF_LSH | BPF_X:
 144                 case BPF_ALU | BPF_RSH | BPF_K:
 145                 case BPF_ALU | BPF_RSH | BPF_X:
 146                 case BPF_ALU | BPF_NEG:
 147                 case BPF_LD | BPF_IMM:
 148                 case BPF_LDX | BPF_IMM:
 149                 case BPF_MISC | BPF_TAX:
 150                 case BPF_MISC | BPF_TXA:
 151                 case BPF_LD | BPF_MEM:
 152                 case BPF_LDX | BPF_MEM:
 153                 case BPF_ST:
 154                 case BPF_STX:
 155                 case BPF_JMP | BPF_JA:
 156                 case BPF_JMP | BPF_JEQ | BPF_K:
 157                 case BPF_JMP | BPF_JEQ | BPF_X:
 158                 case BPF_JMP | BPF_JGE | BPF_K:
 159                 case BPF_JMP | BPF_JGE | BPF_X:
 160                 case BPF_JMP | BPF_JGT | BPF_K:
 161                 case BPF_JMP | BPF_JGT | BPF_X:
 162                 case BPF_JMP | BPF_JSET | BPF_K:
 163                 case BPF_JMP | BPF_JSET | BPF_X:
 164                         continue;
 165                 default:
 166                         return -EINVAL;
 167                 }
 168         }
 169         return 0;
 170 }
 171
 172 /**
 173  * seccomp_run_filters - evaluates all seccomp filters against @syscall
 174  * @syscall: number of the current system call
 175  *
 176  * Returns valid seccomp BPF response codes.
 177  */
 178 static u32 seccomp_run_filters(struct seccomp_data *sd)
 179 {
 180         struct seccomp_data sd_local;
 181         u32 ret = SECCOMP_RET_ALLOW;
 182         /* Make sure cross-thread synced filter points somewhere sane. */
 183         struct seccomp_filter *f =
 184                         lockless_dereference(current->seccomp.filter);
 185
 186         /* Ensure unexpected behavior doesn't result in failing open. */
 187         if (unlikely(WARN_ON(f == NULL)))
 188                 return SECCOMP_RET_KILL;
 189
 190         if (!sd) {
 191                 populate_seccomp_data(&sd_local);
 192                 sd = &sd_local;
 193         }
 194
 195         /*
 196          * All filters in the list are evaluated and the lowest BPF return
 197          * value always takes priority (ignoring the DATA).
 198          */
 199         for (; f; f = f->prev) {
 200                 u32 cur_ret = BPF_PROG_RUN(f->prog, (void *)sd);
 201
 202                 if ((cur_ret & SECCOMP_RET_ACTION) < (ret & SECCOMP_RET_ACTION))
 203                         ret = cur_ret;
 204         }
 205         return ret;
 206 }
 207 #endif /* CONFIG_SECCOMP_FILTER */
 208
 209 static inline bool seccomp_may_assign_mode(unsigned long seccomp_mode)
 210 {
 211         assert_spin_locked(&current->sighand->siglock);
 212
 213         if (current->seccomp.mode && current->seccomp.mode != seccomp_mode)
 214                 return false;
 215
 216         return true;
 217 }
 218
 219 void __weak arch_seccomp_spec_mitigate(struct task_struct *task) { }
 220
 221 static inline void seccomp_assign_mode(struct task_struct *task,
 222                                        unsigned long seccomp_mode,
 223                                        unsigned long flags)
 224 {
 225         assert_spin_locked(&task->sighand->siglock);
 226
 227         task->seccomp.mode = seccomp_mode;
 228         /*
 229          * Make sure TIF_SECCOMP cannot be set before the mode (and
 230          * filter) is set.
 231          */
 232         smp_mb__before_atomic();
 233         /* Assume default seccomp processes want spec flaw mitigation. */
 234         if ((flags & SECCOMP_FILTER_FLAG_SPEC_ALLOW) == 0)
 235                 arch_seccomp_spec_mitigate(task);
 236         set_tsk_thread_flag(task, TIF_SECCOMP);
 237 }
 238
 239 #ifdef CONFIG_SECCOMP_FILTER
 240 /* Returns 1 if the parent is an ancestor of the child. */
 241 static int is_ancestor(struct seccomp_filter *parent,
 242                        struct seccomp_filter *child)
 243 {
 244         /* NULL is the root ancestor. */
 245         if (parent == NULL)
 246                 return 1;
 247         for (; child; child = child->prev)
 248                 if (child == parent)
 249                         return 1;
 250         return 0;
 251 }
 252
 253 /**
 254  * seccomp_can_sync_threads: checks if all threads can be synchronized
 255  *
 256  * Expects sighand and cred_guard_mutex locks to be held.
 257  *
 258  * Returns 0 on success, -ve on error, or the pid of a thread which was
 259  * either not in the correct seccomp mode or it did not have an ancestral
 260  * seccomp filter.
 261  */
 262 static inline pid_t seccomp_can_sync_threads(void)
 263 {
 264         struct task_struct *thread, *caller;
 265
 266         BUG_ON(!mutex_is_locked(&current->signal->cred_guard_mutex));
 267         assert_spin_locked(&current->sighand->siglock);
 268
 269         /* Validate all threads being eligible for synchronization. */
 270         caller = current;
 271         for_each_thread(caller, thread) {
 272                 pid_t failed;
 273
 274                 /* Skip current, since it is initiating the sync. */
 275                 if (thread == caller)
 276                         continue;
 277
 278                 if (thread->seccomp.mode == SECCOMP_MODE_DISABLED ||
 279                     (thread->seccomp.mode == SECCOMP_MODE_FILTER &&
 280                      is_ancestor(thread->seccomp.filter,
 281                                  caller->seccomp.filter)))
 282                         continue;
 283
 284                 /* Return the first thread that cannot be synchronized. */
 285                 failed = task_pid_vnr(thread);
 286                 /* If the pid cannot be resolved, then return -ESRCH */
 287                 if (unlikely(WARN_ON(failed == 0)))
 288                         failed = -ESRCH;
 289                 return failed;
 290         }
 291
 292         return 0;
 293 }
 294
 295 /**
 296  * seccomp_sync_threads: sets all threads to use current's filter
 297  *
 298  * Expects sighand and cred_guard_mutex locks to be held, and for
 299  * seccomp_can_sync_threads() to have returned success already
 300  * without dropping the locks.
 301  *
 302  */
 303 static inline void seccomp_sync_threads(unsigned long flags)
 304 {
 305         struct task_struct *thread, *caller;
 306
 307         BUG_ON(!mutex_is_locked(&current->signal->cred_guard_mutex));
 308         assert_spin_locked(&current->sighand->siglock);
 309
 310         /* Synchronize all threads. */
 311         caller = current;
 312         for_each_thread(caller, thread) {
 313                 /* Skip current, since it needs no changes. */
 314                 if (thread == caller)
 315                         continue;
 316
 317                 /* Get a task reference for the new leaf node. */
 318                 get_seccomp_filter(caller);
 319                 /*
 320                  * Drop the task reference to the shared ancestor since
 321                  * current's path will hold a reference.  (This also
 322                  * allows a put before the assignment.)
 323                  */
 324                 put_seccomp_filter(thread);
 325                 smp_store_release(&thread->seccomp.filter,
 326                                   caller->seccomp.filter);
 327
 328                 /*
 329                  * Don't let an unprivileged task work around
 330                  * the no_new_privs restriction by creating
 331                  * a thread that sets it up, enters seccomp,
 332                  * then dies.
 333                  */
 334                 if (task_no_new_privs(caller))
 335                         task_set_no_new_privs(thread);
 336
 337                 /*
 338                  * Opt the other thread into seccomp if needed.
 339                  * As threads are considered to be trust-realm
 340                  * equivalent (see ptrace_may_access), it is safe to
 341                  * allow one thread to transition the other.
 342                  */
 343                 if (thread->seccomp.mode == SECCOMP_MODE_DISABLED)
 344                         seccomp_assign_mode(thread, SECCOMP_MODE_FILTER,
 345                                             flags);
 346         }
 347 }
 348
 349 /**
 350  * seccomp_prepare_filter: Prepares a seccomp filter for use.
 351  * @fprog: BPF program to install
 352  *
 353  * Returns filter on success or an ERR_PTR on failure.
 354  */
 355 static struct seccomp_filter *seccomp_prepare_filter(struct sock_fprog *fprog)
 356 {
 357         struct seccomp_filter *sfilter;
 358         int ret;
 359         const bool save_orig = config_enabled(CONFIG_CHECKPOINT_RESTORE);
 360
 361         if (fprog->len == 0 || fprog->len > BPF_MAXINSNS)
 362                 return ERR_PTR(-EINVAL);
 363
 364         BUG_ON(INT_MAX / fprog->len < sizeof(struct sock_filter));
 365
 366         /*
 367          * Installing a seccomp filter requires that the task has
 368          * CAP_SYS_ADMIN in its namespace or be running with no_new_privs.
 369          * This avoids scenarios where unprivileged tasks can affect the
 370          * behavior of privileged children.
 371          */
 372         if (!task_no_new_privs(current) &&
 373             security_capable_noaudit(current_cred(), current_user_ns(),
 374                                      CAP_SYS_ADMIN) != 0)
 375                 return ERR_PTR(-EACCES);
 376
 377         /* Allocate a new seccomp_filter */
 378         sfilter = kzalloc(sizeof(*sfilter), GFP_KERNEL | __GFP_NOWARN);
 379         if (!sfilter)
 380                 return ERR_PTR(-ENOMEM);
 381
 382         ret = bpf_prog_create_from_user(&sfilter->prog, fprog,
 383                                         seccomp_check_filter, save_orig);
 384         if (ret < 0) {
 385                 kfree(sfilter);
 386                 return ERR_PTR(ret);
 387         }
 388
 389         atomic_set(&sfilter->usage, 1);
 390
 391         return sfilter;
 392 }
 393
 394 /**
 395  * seccomp_prepare_user_filter - prepares a user-supplied sock_fprog
 396  * @user_filter: pointer to the user data containing a sock_fprog.
 397  *
 398  * Returns 0 on success and non-zero otherwise.
 399  */
 400 static struct seccomp_filter *
 401 seccomp_prepare_user_filter(const char __user *user_filter)
 402 {
 403         struct sock_fprog fprog;
 404         struct seccomp_filter *filter = ERR_PTR(-EFAULT);
 405
 406 #ifdef CONFIG_COMPAT
 407         if (is_compat_task()) {
 408                 struct compat_sock_fprog fprog32;
 409                 if (copy_from_user(&fprog32, user_filter, sizeof(fprog32)))
 410                         goto out;
 411                 fprog.len = fprog32.len;
 412                 fprog.filter = compat_ptr(fprog32.filter);
 413         } else /* falls through to the if below. */
 414 #endif
 415         if (copy_from_user(&fprog, user_filter, sizeof(fprog)))
 416                 goto out;
 417         filter = seccomp_prepare_filter(&fprog);
 418 out:
 419         return filter;
 420 }
 421
 422 /**
 423  * seccomp_attach_filter: validate and attach filter
 424  * @flags:  flags to change filter behavior
 425  * @filter: seccomp filter to add to the current process
 426  *
 427  * Caller must be holding current->sighand->siglock lock.
 428  *
 429  * Returns 0 on success, -ve on error.
 430  */
 431 static long seccomp_attach_filter(unsigned int flags,
 432                                   struct seccomp_filter *filter)
 433 {
 434         unsigned long total_insns;
 435         struct seccomp_filter *walker;
 436
 437         assert_spin_locked(&current->sighand->siglock);
 438
 439         /* Validate resulting filter length. */
 440         total_insns = filter->prog->len;
 441         for (walker = current->seccomp.filter; walker; walker = walker->prev)
 442                 total_insns += walker->prog->len + 4;  /* 4 instr penalty */
 443         if (total_insns > MAX_INSNS_PER_PATH)
 444                 return -ENOMEM;
 445
 446         /* If thread sync has been requested, check that it is possible. */
 447         if (flags & SECCOMP_FILTER_FLAG_TSYNC) {
 448                 int ret;
 449
 450                 ret = seccomp_can_sync_threads();
 451                 if (ret)
 452                         return ret;
 453         }
 454
 455         /*
 456          * If there is an existing filter, make it the prev and don't drop its
 457          * task reference.
 458          */
 459         filter->prev = current->seccomp.filter;
 460         current->seccomp.filter = filter;
 461
 462         /* Now that the new filter is in place, synchronize to all threads. */
 463         if (flags & SECCOMP_FILTER_FLAG_TSYNC)
 464                 seccomp_sync_threads(flags);
 465
 466         return 0;
 467 }
 468
 469 void __get_seccomp_filter(struct seccomp_filter *filter)
 470 {
 471         /* Reference count is bounded by the number of total processes. */
 472         atomic_inc(&filter->usage);
 473 }
 474
 475 /* get_seccomp_filter - increments the reference count of the filter on @tsk */
 476 void get_seccomp_filter(struct task_struct *tsk)
 477 {
 478         struct seccomp_filter *orig = tsk->seccomp.filter;
 479         if (!orig)
 480                 return;
 481         __get_seccomp_filter(orig);
 482 }
 483
 484 static inline void seccomp_filter_free(struct seccomp_filter *filter)
 485 {
 486         if (filter) {
 487                 bpf_prog_destroy(filter->prog);
 488                 kfree(filter);
 489         }
 490 }
 491
 492 static void __put_seccomp_filter(struct seccomp_filter *orig)
 493 {
 494         /* Clean up single-reference branches iteratively. */
 495         while (orig && atomic_dec_and_test(&orig->usage)) {
 496                 struct seccomp_filter *freeme = orig;
 497                 orig = orig->prev;
 498                 seccomp_filter_free(freeme);
 499         }
 500 }
 501
 502 /* put_seccomp_filter - decrements the ref count of tsk->seccomp.filter */
 503 void put_seccomp_filter(struct task_struct *tsk)
 504 {
 505         __put_seccomp_filter(tsk->seccomp.filter);
 506 }
 507
 508 /**
 509  * seccomp_send_sigsys - signals the task to allow in-process syscall emulation
 510  * @syscall: syscall number to send to userland
 511  * @reason: filter-supplied reason code to send to userland (via si_errno)
 512  *
 513  * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info.
 514  */
 515 static void seccomp_send_sigsys(int syscall, int reason)
 516 {
 517         struct siginfo info;
 518         memset(&info, 0, sizeof(info));
 519         info.si_signo = SIGSYS;
 520         info.si_code = SYS_SECCOMP;
 521         info.si_call_addr = (void __user *)KSTK_EIP(current);
 522         info.si_errno = reason;
 523         info.si_arch = syscall_get_arch();
 524         info.si_syscall = syscall;
 525         force_sig_info(SIGSYS, &info, current);
 526 }
 527 #endif  /* CONFIG_SECCOMP_FILTER */
 528
 529 /*
 530  * Secure computing mode 1 allows only read/write/exit/sigreturn.
 531  * To be fully secure this must be combined with rlimit
 532  * to limit the stack allocations too.
 533  */
 534 static int mode1_syscalls[] = {
 535         __NR_seccomp_read, __NR_seccomp_write, __NR_seccomp_exit, __NR_seccomp_sigreturn,
 536         0, /* null terminated */
 537 };
 538
 539 #ifdef CONFIG_COMPAT
 540 static int mode1_syscalls_32[] = {
 541         __NR_seccomp_read_32, __NR_seccomp_write_32, __NR_seccomp_exit_32, __NR_seccomp_sigreturn_32,
 542         0, /* null terminated */
 543 };
 544 #endif
 545
 546 static void __secure_computing_strict(int this_syscall)
 547 {
 548         int *syscall_whitelist = mode1_syscalls;
 549 #ifdef CONFIG_COMPAT
 550         if (is_compat_task())
 551                 syscall_whitelist = mode1_syscalls_32;
 552 #endif
 553         do {
 554                 if (*syscall_whitelist == this_syscall)
 555                         return;
 556         } while (*++syscall_whitelist);
 557
 558 #ifdef SECCOMP_DEBUG
 559         dump_stack();
 560 #endif
 561         audit_seccomp(this_syscall, SIGKILL, SECCOMP_RET_KILL);
 562         do_exit(SIGKILL);
 563 }
 564
 565 #ifndef CONFIG_HAVE_ARCH_SECCOMP_FILTER
 566 void secure_computing_strict(int this_syscall)
 567 {
 568         int mode = current->seccomp.mode;
 569
 570         if (config_enabled(CONFIG_CHECKPOINT_RESTORE) &&
 571             unlikely(current->ptrace & PT_SUSPEND_SECCOMP))
 572                 return;
 573
 574         if (mode == SECCOMP_MODE_DISABLED)
 575                 return;
 576         else if (mode == SECCOMP_MODE_STRICT)
 577                 __secure_computing_strict(this_syscall);
 578         else
 579                 BUG();
 580 }
 581 #else
 582 int __secure_computing(void)
 583 {
 584         u32 phase1_result = seccomp_phase1(NULL);
 585
 586         if (likely(phase1_result == SECCOMP_PHASE1_OK))
 587                 return 0;
 588         else if (likely(phase1_result == SECCOMP_PHASE1_SKIP))
 589                 return -1;
 590         else
 591                 return seccomp_phase2(phase1_result);
 592 }
 593
 594 #ifdef CONFIG_SECCOMP_FILTER
 595 static u32 __seccomp_phase1_filter(int this_syscall, struct seccomp_data *sd)
 596 {
 597         u32 filter_ret, action;
 598         int data;
 599
 600         /*
 601          * Make sure that any changes to mode from another thread have
 602          * been seen after TIF_SECCOMP was seen.
 603          */
 604         rmb();
 605
 606         filter_ret = seccomp_run_filters(sd);
 607         data = filter_ret & SECCOMP_RET_DATA;
 608         action = filter_ret & SECCOMP_RET_ACTION;
 609
 610         switch (action) {
 611         case SECCOMP_RET_ERRNO:
 612                 /* Set low-order bits as an errno, capped at MAX_ERRNO. */
 613                 if (data > MAX_ERRNO)
 614                         data = MAX_ERRNO;
 615                 syscall_set_return_value(current, task_pt_regs(current),
 616                                          -data, 0);
 617                 goto skip;
 618
 619         case SECCOMP_RET_TRAP:
 620                 /* Show the handler the original registers. */
 621                 syscall_rollback(current, task_pt_regs(current));
 622                 /* Let the filter pass back 16 bits of data. */
 623                 seccomp_send_sigsys(this_syscall, data);
 624                 goto skip;
 625
 626         case SECCOMP_RET_TRACE:
 627                 return filter_ret;  /* Save the rest for phase 2. */
 628
 629         case SECCOMP_RET_ALLOW:
 630                 return SECCOMP_PHASE1_OK;
 631
 632         case SECCOMP_RET_KILL:
 633         default:
 634                 audit_seccomp(this_syscall, SIGSYS, action);
 635                 do_exit(SIGSYS);
 636         }
 637
 638         unreachable();
 639
 640 skip:
 641         audit_seccomp(this_syscall, 0, action);
 642         return SECCOMP_PHASE1_SKIP;
 643 }
 644 #endif
 645
 646 /**
 647  * seccomp_phase1() - run fast path seccomp checks on the current syscall
 648  * @arg sd: The seccomp_data or NULL
 649  *
 650  * This only reads pt_regs via the syscall_xyz helpers.  The only change
 651  * it will make to pt_regs is via syscall_set_return_value, and it will
 652  * only do that if it returns SECCOMP_PHASE1_SKIP.
 653  *
 654  * If sd is provided, it will not read pt_regs at all.
 655  *
 656  * It may also call do_exit or force a signal; these actions must be
 657  * safe.
 658  *
 659  * If it returns SECCOMP_PHASE1_OK, the syscall passes checks and should
 660  * be processed normally.
 661  *
 662  * If it returns SECCOMP_PHASE1_SKIP, then the syscall should not be
 663  * invoked.  In this case, seccomp_phase1 will have set the return value
 664  * using syscall_set_return_value.
 665  *
 666  * If it returns anything else, then the return value should be passed
 667  * to seccomp_phase2 from a context in which ptrace hooks are safe.
 668  */
 669 u32 seccomp_phase1(struct seccomp_data *sd)
 670 {
 671         int mode = current->seccomp.mode;
 672         int this_syscall = sd ? sd->nr :
 673                 syscall_get_nr(current, task_pt_regs(current));
 674
 675         if (config_enabled(CONFIG_CHECKPOINT_RESTORE) &&
 676             unlikely(current->ptrace & PT_SUSPEND_SECCOMP))
 677                 return SECCOMP_PHASE1_OK;
 678
 679         switch (mode) {
 680         case SECCOMP_MODE_STRICT:
 681                 __secure_computing_strict(this_syscall);  /* may call do_exit */
 682                 return SECCOMP_PHASE1_OK;
 683 #ifdef CONFIG_SECCOMP_FILTER
 684         case SECCOMP_MODE_FILTER:
 685                 return __seccomp_phase1_filter(this_syscall, sd);
 686 #endif
 687         default:
 688                 BUG();
 689         }
 690 }
 691
 692 /**
 693  * seccomp_phase2() - finish slow path seccomp work for the current syscall
 694  * @phase1_result: The return value from seccomp_phase1()
 695  *
 696  * This must be called from a context in which ptrace hooks can be used.
 697  *
 698  * Returns 0 if the syscall should be processed or -1 to skip the syscall.
 699  */
 700 int seccomp_phase2(u32 phase1_result)
 701 {
 702         struct pt_regs *regs = task_pt_regs(current);
 703         u32 action = phase1_result & SECCOMP_RET_ACTION;
 704         int data = phase1_result & SECCOMP_RET_DATA;
 705
 706         BUG_ON(action != SECCOMP_RET_TRACE);
 707
 708         audit_seccomp(syscall_get_nr(current, regs), 0, action);
 709
 710         /* Skip these calls if there is no tracer. */
 711         if (!ptrace_event_enabled(current, PTRACE_EVENT_SECCOMP)) {
 712                 syscall_set_return_value(current, regs,
 713                                          -ENOSYS, 0);
 714                 return -1;
 715         }
 716
 717         /* Allow the BPF to provide the event message */
 718         ptrace_event(PTRACE_EVENT_SECCOMP, data);
 719         /*
 720          * The delivery of a fatal signal during event
 721          * notification may silently skip tracer notification.
 722          * Terminating the task now avoids executing a system
 723          * call that may not be intended.
 724          */
 725         if (fatal_signal_pending(current))
 726                 do_exit(SIGSYS);
 727         if (syscall_get_nr(current, regs) < 0)
 728                 return -1;  /* Explicit request to skip. */
 729
 730         return 0;
 731 }
 732 #endif /* CONFIG_HAVE_ARCH_SECCOMP_FILTER */
 733
 734 long prctl_get_seccomp(void)
 735 {
 736         return current->seccomp.mode;
 737 }
 738
 739 /**
 740  * seccomp_set_mode_strict: internal function for setting strict seccomp
 741  *
 742  * Once current->seccomp.mode is non-zero, it may not be changed.
 743  *
 744  * Returns 0 on success or -EINVAL on failure.
 745  */
 746 static long seccomp_set_mode_strict(void)
 747 {
 748         const unsigned long seccomp_mode = SECCOMP_MODE_STRICT;
 749         long ret = -EINVAL;
 750
 751         spin_lock_irq(&current->sighand->siglock);
 752
 753         if (!seccomp_may_assign_mode(seccomp_mode))
 754                 goto out;
 755
 756 #ifdef TIF_NOTSC
 757         disable_TSC();
 758 #endif
 759         seccomp_assign_mode(current, seccomp_mode, 0);
 760         ret = 0;
 761
 762 out:
 763         spin_unlock_irq(&current->sighand->siglock);
 764
 765         return ret;
 766 }
 767
 768 #ifdef CONFIG_SECCOMP_FILTER
 769 /**
 770  * seccomp_set_mode_filter: internal function for setting seccomp filter
 771  * @flags:  flags to change filter behavior
 772  * @filter: struct sock_fprog containing filter
 773  *
 774  * This function may be called repeatedly to install additional filters.
 775  * Every filter successfully installed will be evaluated (in reverse order)
 776  * for each system call the task makes.
 777  *
 778  * Once current->seccomp.mode is non-zero, it may not be changed.
 779  *
 780  * Returns 0 on success or -EINVAL on failure.
 781  */
 782 static long seccomp_set_mode_filter(unsigned int flags,
 783                                     const char __user *filter)
 784 {
 785         const unsigned long seccomp_mode = SECCOMP_MODE_FILTER;
 786         struct seccomp_filter *prepared = NULL;
 787         long ret = -EINVAL;
 788
 789         /* Validate flags. */
 790         if (flags & ~SECCOMP_FILTER_FLAG_MASK)
 791                 return -EINVAL;
 792
 793         /* Prepare the new filter before holding any locks. */
 794         prepared = seccomp_prepare_user_filter(filter);
 795         if (IS_ERR(prepared))
 796                 return PTR_ERR(prepared);
 797
 798         /*
 799          * Make sure we cannot change seccomp or nnp state via TSYNC
 800          * while another thread is in the middle of calling exec.
 801          */
 802         if (flags & SECCOMP_FILTER_FLAG_TSYNC &&
 803             mutex_lock_killable(&current->signal->cred_guard_mutex))
 804                 goto out_free;
 805
 806         spin_lock_irq(&current->sighand->siglock);
 807
 808         if (!seccomp_may_assign_mode(seccomp_mode))
 809                 goto out;
 810
 811         ret = seccomp_attach_filter(flags, prepared);
 812         if (ret)
 813                 goto out;
 814         /* Do not free the successfully attached filter. */
 815         prepared = NULL;
 816
 817         seccomp_assign_mode(current, seccomp_mode, flags);
 818 out:
 819         spin_unlock_irq(&current->sighand->siglock);
 820         if (flags & SECCOMP_FILTER_FLAG_TSYNC)
 821                 mutex_unlock(&current->signal->cred_guard_mutex);
 822 out_free:
 823         seccomp_filter_free(prepared);
 824         return ret;
 825 }
 826 #else
 827 static inline long seccomp_set_mode_filter(unsigned int flags,
 828                                            const char __user *filter)
 829 {
 830         return -EINVAL;
 831 }
 832 #endif
 833
 834 /* Common entry point for both prctl and syscall. */
 835 static long do_seccomp(unsigned int op, unsigned int flags,
 836                        const char __user *uargs)
 837 {
 838         switch (op) {
 839         case SECCOMP_SET_MODE_STRICT:
 840                 if (flags != 0 || uargs != NULL)
 841                         return -EINVAL;
 842                 return seccomp_set_mode_strict();
 843         case SECCOMP_SET_MODE_FILTER:
 844                 return seccomp_set_mode_filter(flags, uargs);
 845         default:
 846                 return -EINVAL;
 847         }
 848 }
 849
 850 SYSCALL_DEFINE3(seccomp, unsigned int, op, unsigned int, flags,
 851                          const char __user *, uargs)
 852 {
 853         return do_seccomp(op, flags, uargs);
 854 }
 855
 856 /**
 857  * prctl_set_seccomp: configures current->seccomp.mode
 858  * @seccomp_mode: requested mode to use
 859  * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER
 860  *
 861  * Returns 0 on success or -EINVAL on failure.
 862  */
 863 long prctl_set_seccomp(unsigned long seccomp_mode, char __user *filter)
 864 {
 865         unsigned int op;
 866         char __user *uargs;
 867
 868         switch (seccomp_mode) {
 869         case SECCOMP_MODE_STRICT:
 870                 op = SECCOMP_SET_MODE_STRICT;
 871                 /*
 872                  * Setting strict mode through prctl always ignored filter,
 873                  * so make sure it is always NULL here to pass the internal
 874                  * check in do_seccomp().
 875                  */
 876                 uargs = NULL;
 877                 break;
 878         case SECCOMP_MODE_FILTER:
 879                 op = SECCOMP_SET_MODE_FILTER;
 880                 uargs = filter;
 881                 break;
 882         default:
 883                 return -EINVAL;
 884         }
 885
 886         /* prctl interface doesn't have flags, so they are always zero. */
 887         return do_seccomp(op, 0, uargs);
 888 }
 889
 890 #if defined(CONFIG_SECCOMP_FILTER) && defined(CONFIG_CHECKPOINT_RESTORE)
 891 long seccomp_get_filter(struct task_struct *task, unsigned long filter_off,
 892                         void __user *data)
 893 {
 894         struct seccomp_filter *filter;
 895         struct sock_fprog_kern *fprog;
 896         long ret;
 897         unsigned long count = 0;
 898
 899         if (!capable(CAP_SYS_ADMIN) ||
 900             current->seccomp.mode != SECCOMP_MODE_DISABLED) {
 901                 return -EACCES;
 902         }
 903
 904         spin_lock_irq(&task->sighand->siglock);
 905         if (task->seccomp.mode != SECCOMP_MODE_FILTER) {
 906                 ret = -EINVAL;
 907                 goto out;
 908         }
 909
 910         filter = task->seccomp.filter;
 911         while (filter) {
 912                 filter = filter->prev;
 913                 count++;
 914         }
 915
 916         if (filter_off >= count) {
 917                 ret = -ENOENT;
 918                 goto out;
 919         }
 920         count -= filter_off;
 921
 922         filter = task->seccomp.filter;
 923         while (filter && count > 1) {
 924                 filter = filter->prev;
 925                 count--;
 926         }
 927
 928         if (WARN_ON(count != 1 || !filter)) {
 929                 /* The filter tree shouldn't shrink while we're using it. */
 930                 ret = -ENOENT;
 931                 goto out;
 932         }
 933
 934         fprog = filter->prog->orig_prog;
 935         if (!fprog) {
 936                 /* This must be a new non-cBPF filter, since we save every
 937                  * every cBPF filter's orig_prog above when
 938                  * CONFIG_CHECKPOINT_RESTORE is enabled.
 939                  */
 940                 ret = -EMEDIUMTYPE;
 941                 goto out;
 942         }
 943
 944         ret = fprog->len;
 945         if (!data)
 946                 goto out;
 947
 948         __get_seccomp_filter(filter);
 949         spin_unlock_irq(&task->sighand->siglock);
 950
 951         if (copy_to_user(data, fprog->filter, bpf_classic_proglen(fprog)))
 952                 ret = -EFAULT;
 953
 954         __put_seccomp_filter(filter);
 955         return ret;
 956
 957 out:
 958         spin_unlock_irq(&task->sighand->siglock);
 959         return ret;
 960 }
 961 #endif