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/sched.h>
  20 #include <linux/seccomp.h>
  21 #include <linux/slab.h>
  22 #include <linux/syscalls.h>
  23
  24 #ifdef CONFIG_HAVE_ARCH_SECCOMP_FILTER
  25 #include <asm/syscall.h>
  26 #endif
  27
  28 #ifdef CONFIG_SECCOMP_FILTER
  29 #include <linux/filter.h>
  30 #include <linux/pid.h>
  31 #include <linux/ptrace.h>
  32 #include <linux/security.h>
  33 #include <linux/tracehook.h>
  34 #include <linux/uaccess.h>
  35
  36 /**
  37  * struct seccomp_filter - container for seccomp BPF programs
  38  *
  39  * @usage: reference count to manage the object lifetime.
  40  *         get/put helpers should be used when accessing an instance
  41  *         outside of a lifetime-guarded section.  In general, this
  42  *         is only needed for handling filters shared across tasks.
  43  * @prev: points to a previously installed, or inherited, filter
  44  * @len: the number of instructions in the program
  45  * @insnsi: the BPF program instructions to evaluate
  46  *
  47  * seccomp_filter objects are organized in a tree linked via the @prev
  48  * pointer.  For any task, it appears to be a singly-linked list starting
  49  * with current->seccomp.filter, the most recently attached or inherited filter.
  50  * However, multiple filters may share a @prev node, by way of fork(), which
  51  * results in a unidirectional tree existing in memory.  This is similar to
  52  * how namespaces work.
  53  *
  54  * seccomp_filter objects should never be modified after being attached
  55  * to a task_struct (other than @usage).
  56  */
  57 struct seccomp_filter {
  58         atomic_t usage;
  59         struct seccomp_filter *prev;
  60         struct bpf_prog *prog;
  61 };
  62
  63 /* Limit any path through the tree to 256KB worth of instructions. */
  64 #define MAX_INSNS_PER_PATH ((1 << 18) / sizeof(struct sock_filter))
  65
  66 /*
  67  * Endianness is explicitly ignored and left for BPF program authors to manage
  68  * as per the specific architecture.
  69  */
  70 static void populate_seccomp_data(struct seccomp_data *sd)
  71 {
  72         struct task_struct *task = current;
  73         struct pt_regs *regs = task_pt_regs(task);
  74         unsigned long args[6];
  75
  76         sd->nr = syscall_get_nr(task, regs);
  77         sd->arch = syscall_get_arch();
  78         syscall_get_arguments(task, regs, 0, 6, args);
  79         sd->args[0] = args[0];
  80         sd->args[1] = args[1];
  81         sd->args[2] = args[2];
  82         sd->args[3] = args[3];
  83         sd->args[4] = args[4];
  84         sd->args[5] = args[5];
  85         sd->instruction_pointer = KSTK_EIP(task);
  86 }
  87
  88 /**
  89  *      seccomp_check_filter - verify seccomp filter code
  90  *      @filter: filter to verify
  91  *      @flen: length of filter
  92  *
  93  * Takes a previously checked filter (by bpf_check_classic) and
  94  * redirects all filter code that loads struct sk_buff data
  95  * and related data through seccomp_bpf_load.  It also
  96  * enforces length and alignment checking of those loads.
  97  *
  98  * Returns 0 if the rule set is legal or -EINVAL if not.
  99  */
 100 static int seccomp_check_filter(struct sock_filter *filter, unsigned int flen)
 101 {
 102         int pc;
 103         for (pc = 0; pc < flen; pc++) {
 104                 struct sock_filter *ftest = &filter[pc];
 105                 u16 code = ftest->code;
 106                 u32 k = ftest->k;
 107
 108                 switch (code) {
 109                 case BPF_LD | BPF_W | BPF_ABS:
 110                         ftest->code = BPF_LDX | BPF_W | BPF_ABS;
 111                         /* 32-bit aligned and not out of bounds. */
 112                         if (k >= sizeof(struct seccomp_data) || k & 3)
 113                                 return -EINVAL;
 114                         continue;
 115                 case BPF_LD | BPF_W | BPF_LEN:
 116                         ftest->code = BPF_LD | BPF_IMM;
 117                         ftest->k = sizeof(struct seccomp_data);
 118                         continue;
 119                 case BPF_LDX | BPF_W | BPF_LEN:
 120                         ftest->code = BPF_LDX | BPF_IMM;
 121                         ftest->k = sizeof(struct seccomp_data);
 122                         continue;
 123                 /* Explicitly include allowed calls. */
 124                 case BPF_RET | BPF_K:
 125                 case BPF_RET | BPF_A:
 126                 case BPF_ALU | BPF_ADD | BPF_K:
 127                 case BPF_ALU | BPF_ADD | BPF_X:
 128                 case BPF_ALU | BPF_SUB | BPF_K:
 129                 case BPF_ALU | BPF_SUB | BPF_X:
 130                 case BPF_ALU | BPF_MUL | BPF_K:
 131                 case BPF_ALU | BPF_MUL | BPF_X:
 132                 case BPF_ALU | BPF_DIV | BPF_K:
 133                 case BPF_ALU | BPF_DIV | BPF_X:
 134                 case BPF_ALU | BPF_AND | BPF_K:
 135                 case BPF_ALU | BPF_AND | BPF_X:
 136                 case BPF_ALU | BPF_OR | BPF_K:
 137                 case BPF_ALU | BPF_OR | BPF_X:
 138                 case BPF_ALU | BPF_XOR | BPF_K:
 139                 case BPF_ALU | BPF_XOR | BPF_X:
 140                 case BPF_ALU | BPF_LSH | BPF_K:
 141                 case BPF_ALU | BPF_LSH | BPF_X:
 142                 case BPF_ALU | BPF_RSH | BPF_K:
 143                 case BPF_ALU | BPF_RSH | BPF_X:
 144                 case BPF_ALU | BPF_NEG:
 145                 case BPF_LD | BPF_IMM:
 146                 case BPF_LDX | BPF_IMM:
 147                 case BPF_MISC | BPF_TAX:
 148                 case BPF_MISC | BPF_TXA:
 149                 case BPF_LD | BPF_MEM:
 150                 case BPF_LDX | BPF_MEM:
 151                 case BPF_ST:
 152                 case BPF_STX:
 153                 case BPF_JMP | BPF_JA:
 154                 case BPF_JMP | BPF_JEQ | BPF_K:
 155                 case BPF_JMP | BPF_JEQ | BPF_X:
 156                 case BPF_JMP | BPF_JGE | BPF_K:
 157                 case BPF_JMP | BPF_JGE | BPF_X:
 158                 case BPF_JMP | BPF_JGT | BPF_K:
 159                 case BPF_JMP | BPF_JGT | BPF_X:
 160                 case BPF_JMP | BPF_JSET | BPF_K:
 161                 case BPF_JMP | BPF_JSET | BPF_X:
 162                         continue;
 163                 default:
 164                         return -EINVAL;
 165                 }
 166         }
 167         return 0;
 168 }
 169
 170 /**
 171  * seccomp_run_filters - evaluates all seccomp filters against @syscall
 172  * @syscall: number of the current system call
 173  *
 174  * Returns valid seccomp BPF response codes.
 175  */
 176 static u32 seccomp_run_filters(struct seccomp_data *sd)
 177 {
 178         struct seccomp_filter *f = ACCESS_ONCE(current->seccomp.filter);
 179         struct seccomp_data sd_local;
 180         u32 ret = SECCOMP_RET_ALLOW;
 181
 182         /* Ensure unexpected behavior doesn't result in failing open. */
 183         if (unlikely(WARN_ON(f == NULL)))
 184                 return SECCOMP_RET_KILL;
 185
 186         /* Make sure cross-thread synced filter points somewhere sane. */
 187         smp_read_barrier_depends();
 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, (void *)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                  * Opt the other thread into seccomp if needed.
 322                  * As threads are considered to be trust-realm
 323                  * equivalent (see ptrace_may_access), it is safe to
 324                  * allow one thread to transition the other.
 325                  */
 326                 if (thread->seccomp.mode == SECCOMP_MODE_DISABLED) {
 327                         /*
 328                          * Don't let an unprivileged task work around
 329                          * the no_new_privs restriction by creating
 330                          * a thread that sets it up, enters seccomp,
 331                          * then dies.
 332                          */
 333                         if (task_no_new_privs(caller))
 334                                 task_set_no_new_privs(thread);
 335
 336                         seccomp_assign_mode(thread, SECCOMP_MODE_FILTER);
 337                 }
 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
 352         if (fprog->len == 0 || fprog->len > BPF_MAXINSNS)
 353                 return ERR_PTR(-EINVAL);
 354
 355         BUG_ON(INT_MAX / fprog->len < sizeof(struct sock_filter));
 356
 357         /*
 358          * Installing a seccomp filter requires that the task has
 359          * CAP_SYS_ADMIN in its namespace or be running with no_new_privs.
 360          * This avoids scenarios where unprivileged tasks can affect the
 361          * behavior of privileged children.
 362          */
 363         if (!task_no_new_privs(current) &&
 364             security_capable_noaudit(current_cred(), current_user_ns(),
 365                                      CAP_SYS_ADMIN) != 0)
 366                 return ERR_PTR(-EACCES);
 367
 368         /* Allocate a new seccomp_filter */
 369         sfilter = kzalloc(sizeof(*sfilter), GFP_KERNEL | __GFP_NOWARN);
 370         if (!sfilter)
 371                 return ERR_PTR(-ENOMEM);
 372
 373         ret = bpf_prog_create_from_user(&sfilter->prog, fprog,
 374                                         seccomp_check_filter);
 375         if (ret < 0) {
 376                 kfree(sfilter);
 377                 return ERR_PTR(ret);
 378         }
 379
 380         atomic_set(&sfilter->usage, 1);
 381
 382         return sfilter;
 383 }
 384
 385 /**
 386  * seccomp_prepare_user_filter - prepares a user-supplied sock_fprog
 387  * @user_filter: pointer to the user data containing a sock_fprog.
 388  *
 389  * Returns 0 on success and non-zero otherwise.
 390  */
 391 static struct seccomp_filter *
 392 seccomp_prepare_user_filter(const char __user *user_filter)
 393 {
 394         struct sock_fprog fprog;
 395         struct seccomp_filter *filter = ERR_PTR(-EFAULT);
 396
 397 #ifdef CONFIG_COMPAT
 398         if (is_compat_task()) {
 399                 struct compat_sock_fprog fprog32;
 400                 if (copy_from_user(&fprog32, user_filter, sizeof(fprog32)))
 401                         goto out;
 402                 fprog.len = fprog32.len;
 403                 fprog.filter = compat_ptr(fprog32.filter);
 404         } else /* falls through to the if below. */
 405 #endif
 406         if (copy_from_user(&fprog, user_filter, sizeof(fprog)))
 407                 goto out;
 408         filter = seccomp_prepare_filter(&fprog);
 409 out:
 410         return filter;
 411 }
 412
 413 /**
 414  * seccomp_attach_filter: validate and attach filter
 415  * @flags:  flags to change filter behavior
 416  * @filter: seccomp filter to add to the current process
 417  *
 418  * Caller must be holding current->sighand->siglock lock.
 419  *
 420  * Returns 0 on success, -ve on error.
 421  */
 422 static long seccomp_attach_filter(unsigned int flags,
 423                                   struct seccomp_filter *filter)
 424 {
 425         unsigned long total_insns;
 426         struct seccomp_filter *walker;
 427
 428         assert_spin_locked(&current->sighand->siglock);
 429
 430         /* Validate resulting filter length. */
 431         total_insns = filter->prog->len;
 432         for (walker = current->seccomp.filter; walker; walker = walker->prev)
 433                 total_insns += walker->prog->len + 4;  /* 4 instr penalty */
 434         if (total_insns > MAX_INSNS_PER_PATH)
 435                 return -ENOMEM;
 436
 437         /* If thread sync has been requested, check that it is possible. */
 438         if (flags & SECCOMP_FILTER_FLAG_TSYNC) {
 439                 int ret;
 440
 441                 ret = seccomp_can_sync_threads();
 442                 if (ret)
 443                         return ret;
 444         }
 445
 446         /*
 447          * If there is an existing filter, make it the prev and don't drop its
 448          * task reference.
 449          */
 450         filter->prev = current->seccomp.filter;
 451         current->seccomp.filter = filter;
 452
 453         /* Now that the new filter is in place, synchronize to all threads. */
 454         if (flags & SECCOMP_FILTER_FLAG_TSYNC)
 455                 seccomp_sync_threads();
 456
 457         return 0;
 458 }
 459
 460 /* get_seccomp_filter - increments the reference count of the filter on @tsk */
 461 void get_seccomp_filter(struct task_struct *tsk)
 462 {
 463         struct seccomp_filter *orig = tsk->seccomp.filter;
 464         if (!orig)
 465                 return;
 466         /* Reference count is bounded by the number of total processes. */
 467         atomic_inc(&orig->usage);
 468 }
 469
 470 static inline void seccomp_filter_free(struct seccomp_filter *filter)
 471 {
 472         if (filter) {
 473                 bpf_prog_free(filter->prog);
 474                 kfree(filter);
 475         }
 476 }
 477
 478 /* put_seccomp_filter - decrements the ref count of tsk->seccomp.filter */
 479 void put_seccomp_filter(struct task_struct *tsk)
 480 {
 481         struct seccomp_filter *orig = tsk->seccomp.filter;
 482         /* Clean up single-reference branches iteratively. */
 483         while (orig && atomic_dec_and_test(&orig->usage)) {
 484                 struct seccomp_filter *freeme = orig;
 485                 orig = orig->prev;
 486                 seccomp_filter_free(freeme);
 487         }
 488 }
 489
 490 /**
 491  * seccomp_send_sigsys - signals the task to allow in-process syscall emulation
 492  * @syscall: syscall number to send to userland
 493  * @reason: filter-supplied reason code to send to userland (via si_errno)
 494  *
 495  * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info.
 496  */
 497 static void seccomp_send_sigsys(int syscall, int reason)
 498 {
 499         struct siginfo info;
 500         memset(&info, 0, sizeof(info));
 501         info.si_signo = SIGSYS;
 502         info.si_code = SYS_SECCOMP;
 503         info.si_call_addr = (void __user *)KSTK_EIP(current);
 504         info.si_errno = reason;
 505         info.si_arch = syscall_get_arch();
 506         info.si_syscall = syscall;
 507         force_sig_info(SIGSYS, &info, current);
 508 }
 509 #endif  /* CONFIG_SECCOMP_FILTER */
 510
 511 /*
 512  * Secure computing mode 1 allows only read/write/exit/sigreturn.
 513  * To be fully secure this must be combined with rlimit
 514  * to limit the stack allocations too.
 515  */
 516 static int mode1_syscalls[] = {
 517         __NR_seccomp_read, __NR_seccomp_write, __NR_seccomp_exit, __NR_seccomp_sigreturn,
 518         0, /* null terminated */
 519 };
 520
 521 #ifdef CONFIG_COMPAT
 522 static int mode1_syscalls_32[] = {
 523         __NR_seccomp_read_32, __NR_seccomp_write_32, __NR_seccomp_exit_32, __NR_seccomp_sigreturn_32,
 524         0, /* null terminated */
 525 };
 526 #endif
 527
 528 static void __secure_computing_strict(int this_syscall)
 529 {
 530         int *syscall_whitelist = mode1_syscalls;
 531 #ifdef CONFIG_COMPAT
 532         if (is_compat_task())
 533                 syscall_whitelist = mode1_syscalls_32;
 534 #endif
 535         do {
 536                 if (*syscall_whitelist == this_syscall)
 537                         return;
 538         } while (*++syscall_whitelist);
 539
 540 #ifdef SECCOMP_DEBUG
 541         dump_stack();
 542 #endif
 543         audit_seccomp(this_syscall, SIGKILL, SECCOMP_RET_KILL);
 544         do_exit(SIGKILL);
 545 }
 546
 547 #ifndef CONFIG_HAVE_ARCH_SECCOMP_FILTER
 548 void secure_computing_strict(int this_syscall)
 549 {
 550         int mode = current->seccomp.mode;
 551
 552         if (mode == 0)
 553                 return;
 554         else if (mode == SECCOMP_MODE_STRICT)
 555                 __secure_computing_strict(this_syscall);
 556         else
 557                 BUG();
 558 }
 559 #else
 560 int __secure_computing(void)
 561 {
 562         u32 phase1_result = seccomp_phase1(NULL);
 563
 564         if (likely(phase1_result == SECCOMP_PHASE1_OK))
 565                 return 0;
 566         else if (likely(phase1_result == SECCOMP_PHASE1_SKIP))
 567                 return -1;
 568         else
 569                 return seccomp_phase2(phase1_result);
 570 }
 571
 572 #ifdef CONFIG_SECCOMP_FILTER
 573 static u32 __seccomp_phase1_filter(int this_syscall, struct seccomp_data *sd)
 574 {
 575         u32 filter_ret, action;
 576         int data;
 577
 578         /*
 579          * Make sure that any changes to mode from another thread have
 580          * been seen after TIF_SECCOMP was seen.
 581          */
 582         rmb();
 583
 584         filter_ret = seccomp_run_filters(sd);
 585         data = filter_ret & SECCOMP_RET_DATA;
 586         action = filter_ret & SECCOMP_RET_ACTION;
 587
 588         switch (action) {
 589         case SECCOMP_RET_ERRNO:
 590                 /* Set low-order bits as an errno, capped at MAX_ERRNO. */
 591                 if (data > MAX_ERRNO)
 592                         data = MAX_ERRNO;
 593                 syscall_set_return_value(current, task_pt_regs(current),
 594                                          -data, 0);
 595                 goto skip;
 596
 597         case SECCOMP_RET_TRAP:
 598                 /* Show the handler the original registers. */
 599                 syscall_rollback(current, task_pt_regs(current));
 600                 /* Let the filter pass back 16 bits of data. */
 601                 seccomp_send_sigsys(this_syscall, data);
 602                 goto skip;
 603
 604         case SECCOMP_RET_TRACE:
 605                 return filter_ret;  /* Save the rest for phase 2. */
 606
 607         case SECCOMP_RET_ALLOW:
 608                 return SECCOMP_PHASE1_OK;
 609
 610         case SECCOMP_RET_KILL:
 611         default:
 612                 audit_seccomp(this_syscall, SIGSYS, action);
 613                 do_exit(SIGSYS);
 614         }
 615
 616         unreachable();
 617
 618 skip:
 619         audit_seccomp(this_syscall, 0, action);
 620         return SECCOMP_PHASE1_SKIP;
 621 }
 622 #endif
 623
 624 /**
 625  * seccomp_phase1() - run fast path seccomp checks on the current syscall
 626  * @arg sd: The seccomp_data or NULL
 627  *
 628  * This only reads pt_regs via the syscall_xyz helpers.  The only change
 629  * it will make to pt_regs is via syscall_set_return_value, and it will
 630  * only do that if it returns SECCOMP_PHASE1_SKIP.
 631  *
 632  * If sd is provided, it will not read pt_regs at all.
 633  *
 634  * It may also call do_exit or force a signal; these actions must be
 635  * safe.
 636  *
 637  * If it returns SECCOMP_PHASE1_OK, the syscall passes checks and should
 638  * be processed normally.
 639  *
 640  * If it returns SECCOMP_PHASE1_SKIP, then the syscall should not be
 641  * invoked.  In this case, seccomp_phase1 will have set the return value
 642  * using syscall_set_return_value.
 643  *
 644  * If it returns anything else, then the return value should be passed
 645  * to seccomp_phase2 from a context in which ptrace hooks are safe.
 646  */
 647 u32 seccomp_phase1(struct seccomp_data *sd)
 648 {
 649         int mode = current->seccomp.mode;
 650         int this_syscall = sd ? sd->nr :
 651                 syscall_get_nr(current, task_pt_regs(current));
 652
 653         switch (mode) {
 654         case SECCOMP_MODE_STRICT:
 655                 __secure_computing_strict(this_syscall);  /* may call do_exit */
 656                 return SECCOMP_PHASE1_OK;
 657 #ifdef CONFIG_SECCOMP_FILTER
 658         case SECCOMP_MODE_FILTER:
 659                 return __seccomp_phase1_filter(this_syscall, sd);
 660 #endif
 661         default:
 662                 BUG();
 663         }
 664 }
 665
 666 /**
 667  * seccomp_phase2() - finish slow path seccomp work for the current syscall
 668  * @phase1_result: The return value from seccomp_phase1()
 669  *
 670  * This must be called from a context in which ptrace hooks can be used.
 671  *
 672  * Returns 0 if the syscall should be processed or -1 to skip the syscall.
 673  */
 674 int seccomp_phase2(u32 phase1_result)
 675 {
 676         struct pt_regs *regs = task_pt_regs(current);
 677         u32 action = phase1_result & SECCOMP_RET_ACTION;
 678         int data = phase1_result & SECCOMP_RET_DATA;
 679
 680         BUG_ON(action != SECCOMP_RET_TRACE);
 681
 682         audit_seccomp(syscall_get_nr(current, regs), 0, action);
 683
 684         /* Skip these calls if there is no tracer. */
 685         if (!ptrace_event_enabled(current, PTRACE_EVENT_SECCOMP)) {
 686                 syscall_set_return_value(current, regs,
 687                                          -ENOSYS, 0);
 688                 return -1;
 689         }
 690
 691         /* Allow the BPF to provide the event message */
 692         ptrace_event(PTRACE_EVENT_SECCOMP, data);
 693         /*
 694          * The delivery of a fatal signal during event
 695          * notification may silently skip tracer notification.
 696          * Terminating the task now avoids executing a system
 697          * call that may not be intended.
 698          */
 699         if (fatal_signal_pending(current))
 700                 do_exit(SIGSYS);
 701         if (syscall_get_nr(current, regs) < 0)
 702                 return -1;  /* Explicit request to skip. */
 703
 704         return 0;
 705 }
 706 #endif /* CONFIG_HAVE_ARCH_SECCOMP_FILTER */
 707
 708 long prctl_get_seccomp(void)
 709 {
 710         return current->seccomp.mode;
 711 }
 712
 713 /**
 714  * seccomp_set_mode_strict: internal function for setting strict seccomp
 715  *
 716  * Once current->seccomp.mode is non-zero, it may not be changed.
 717  *
 718  * Returns 0 on success or -EINVAL on failure.
 719  */
 720 static long seccomp_set_mode_strict(void)
 721 {
 722         const unsigned long seccomp_mode = SECCOMP_MODE_STRICT;
 723         long ret = -EINVAL;
 724
 725         spin_lock_irq(&current->sighand->siglock);
 726
 727         if (!seccomp_may_assign_mode(seccomp_mode))
 728                 goto out;
 729
 730 #ifdef TIF_NOTSC
 731         disable_TSC();
 732 #endif
 733         seccomp_assign_mode(current, seccomp_mode);
 734         ret = 0;
 735
 736 out:
 737         spin_unlock_irq(&current->sighand->siglock);
 738
 739         return ret;
 740 }
 741
 742 #ifdef CONFIG_SECCOMP_FILTER
 743 /**
 744  * seccomp_set_mode_filter: internal function for setting seccomp filter
 745  * @flags:  flags to change filter behavior
 746  * @filter: struct sock_fprog containing filter
 747  *
 748  * This function may be called repeatedly to install additional filters.
 749  * Every filter successfully installed will be evaluated (in reverse order)
 750  * for each system call the task makes.
 751  *
 752  * Once current->seccomp.mode is non-zero, it may not be changed.
 753  *
 754  * Returns 0 on success or -EINVAL on failure.
 755  */
 756 static long seccomp_set_mode_filter(unsigned int flags,
 757                                     const char __user *filter)
 758 {
 759         const unsigned long seccomp_mode = SECCOMP_MODE_FILTER;
 760         struct seccomp_filter *prepared = NULL;
 761         long ret = -EINVAL;
 762
 763         /* Validate flags. */
 764         if (flags & ~SECCOMP_FILTER_FLAG_MASK)
 765                 return -EINVAL;
 766
 767         /* Prepare the new filter before holding any locks. */
 768         prepared = seccomp_prepare_user_filter(filter);
 769         if (IS_ERR(prepared))
 770                 return PTR_ERR(prepared);
 771
 772         /*
 773          * Make sure we cannot change seccomp or nnp state via TSYNC
 774          * while another thread is in the middle of calling exec.
 775          */
 776         if (flags & SECCOMP_FILTER_FLAG_TSYNC &&
 777             mutex_lock_killable(&current->signal->cred_guard_mutex))
 778                 goto out_free;
 779
 780         spin_lock_irq(&current->sighand->siglock);
 781
 782         if (!seccomp_may_assign_mode(seccomp_mode))
 783                 goto out;
 784
 785         ret = seccomp_attach_filter(flags, prepared);
 786         if (ret)
 787                 goto out;
 788         /* Do not free the successfully attached filter. */
 789         prepared = NULL;
 790
 791         seccomp_assign_mode(current, seccomp_mode);
 792 out:
 793         spin_unlock_irq(&current->sighand->siglock);
 794         if (flags & SECCOMP_FILTER_FLAG_TSYNC)
 795                 mutex_unlock(&current->signal->cred_guard_mutex);
 796 out_free:
 797         seccomp_filter_free(prepared);
 798         return ret;
 799 }
 800 #else
 801 static inline long seccomp_set_mode_filter(unsigned int flags,
 802                                            const char __user *filter)
 803 {
 804         return -EINVAL;
 805 }
 806 #endif
 807
 808 /* Common entry point for both prctl and syscall. */
 809 static long do_seccomp(unsigned int op, unsigned int flags,
 810                        const char __user *uargs)
 811 {
 812         switch (op) {
 813         case SECCOMP_SET_MODE_STRICT:
 814                 if (flags != 0 || uargs != NULL)
 815                         return -EINVAL;
 816                 return seccomp_set_mode_strict();
 817         case SECCOMP_SET_MODE_FILTER:
 818                 return seccomp_set_mode_filter(flags, uargs);
 819         default:
 820                 return -EINVAL;
 821         }
 822 }
 823
 824 SYSCALL_DEFINE3(seccomp, unsigned int, op, unsigned int, flags,
 825                          const char __user *, uargs)
 826 {
 827         return do_seccomp(op, flags, uargs);
 828 }
 829
 830 /**
 831  * prctl_set_seccomp: configures current->seccomp.mode
 832  * @seccomp_mode: requested mode to use
 833  * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER
 834  *
 835  * Returns 0 on success or -EINVAL on failure.
 836  */
 837 long prctl_set_seccomp(unsigned long seccomp_mode, char __user *filter)
 838 {
 839         unsigned int op;
 840         char __user *uargs;
 841
 842         switch (seccomp_mode) {
 843         case SECCOMP_MODE_STRICT:
 844                 op = SECCOMP_SET_MODE_STRICT;
 845                 /*
 846                  * Setting strict mode through prctl always ignored filter,
 847                  * so make sure it is always NULL here to pass the internal
 848                  * check in do_seccomp().
 849                  */
 850                 uargs = NULL;
 851                 break;
 852         case SECCOMP_MODE_FILTER:
 853                 op = SECCOMP_SET_MODE_FILTER;
 854                 uargs = filter;
 855                 break;
 856         default:
 857                 return -EINVAL;
 858         }
 859
 860         /* prctl interface doesn't have flags, so they are always zero. */
 861         return do_seccomp(op, 0, uargs);
 862 }