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_data sd_local;
 179         u32 ret = SECCOMP_RET_ALLOW;
 180         /* Make sure cross-thread synced filter points somewhere sane. */
 181         struct seccomp_filter *f =
 182                         lockless_dereference(current->seccomp.filter);
 183
 184         /* Ensure unexpected behavior doesn't result in failing open. */
 185         if (unlikely(WARN_ON(f == NULL)))
 186                 return SECCOMP_RET_KILL;
 187
 188         if (!sd) {
 189                 populate_seccomp_data(&sd_local);
 190                 sd = &sd_local;
 191         }
 192
 193         /*
 194          * All filters in the list are evaluated and the lowest BPF return
 195          * value always takes priority (ignoring the DATA).
 196          */
 197         for (; f; f = f->prev) {
 198                 u32 cur_ret = BPF_PROG_RUN(f->prog, (void *)sd);
 199
 200                 if ((cur_ret & SECCOMP_RET_ACTION) < (ret & SECCOMP_RET_ACTION))
 201                         ret = cur_ret;
 202         }
 203         return ret;
 204 }
 205 #endif /* CONFIG_SECCOMP_FILTER */
 206
 207 static inline bool seccomp_may_assign_mode(unsigned long seccomp_mode)
 208 {
 209         assert_spin_locked(&current->sighand->siglock);
 210
 211         if (current->seccomp.mode && current->seccomp.mode != seccomp_mode)
 212                 return false;
 213
 214         return true;
 215 }
 216
 217 static inline void seccomp_assign_mode(struct task_struct *task,
 218                                        unsigned long seccomp_mode)
 219 {
 220         assert_spin_locked(&task->sighand->siglock);
 221
 222         task->seccomp.mode = seccomp_mode;
 223         /*
 224          * Make sure TIF_SECCOMP cannot be set before the mode (and
 225          * filter) is set.
 226          */
 227         smp_mb__before_atomic();
 228         set_tsk_thread_flag(task, TIF_SECCOMP);
 229 }
 230
 231 #ifdef CONFIG_SECCOMP_FILTER
 232 /* Returns 1 if the parent is an ancestor of the child. */
 233 static int is_ancestor(struct seccomp_filter *parent,
 234                        struct seccomp_filter *child)
 235 {
 236         /* NULL is the root ancestor. */
 237         if (parent == NULL)
 238                 return 1;
 239         for (; child; child = child->prev)
 240                 if (child == parent)
 241                         return 1;
 242         return 0;
 243 }
 244
 245 /**
 246  * seccomp_can_sync_threads: checks if all threads can be synchronized
 247  *
 248  * Expects sighand and cred_guard_mutex locks to be held.
 249  *
 250  * Returns 0 on success, -ve on error, or the pid of a thread which was
 251  * either not in the correct seccomp mode or it did not have an ancestral
 252  * seccomp filter.
 253  */
 254 static inline pid_t seccomp_can_sync_threads(void)
 255 {
 256         struct task_struct *thread, *caller;
 257
 258         BUG_ON(!mutex_is_locked(&current->signal->cred_guard_mutex));
 259         assert_spin_locked(&current->sighand->siglock);
 260
 261         /* Validate all threads being eligible for synchronization. */
 262         caller = current;
 263         for_each_thread(caller, thread) {
 264                 pid_t failed;
 265
 266                 /* Skip current, since it is initiating the sync. */
 267                 if (thread == caller)
 268                         continue;
 269
 270                 if (thread->seccomp.mode == SECCOMP_MODE_DISABLED ||
 271                     (thread->seccomp.mode == SECCOMP_MODE_FILTER &&
 272                      is_ancestor(thread->seccomp.filter,
 273                                  caller->seccomp.filter)))
 274                         continue;
 275
 276                 /* Return the first thread that cannot be synchronized. */
 277                 failed = task_pid_vnr(thread);
 278                 /* If the pid cannot be resolved, then return -ESRCH */
 279                 if (unlikely(WARN_ON(failed == 0)))
 280                         failed = -ESRCH;
 281                 return failed;
 282         }
 283
 284         return 0;
 285 }
 286
 287 /**
 288  * seccomp_sync_threads: sets all threads to use current's filter
 289  *
 290  * Expects sighand and cred_guard_mutex locks to be held, and for
 291  * seccomp_can_sync_threads() to have returned success already
 292  * without dropping the locks.
 293  *
 294  */
 295 static inline void seccomp_sync_threads(void)
 296 {
 297         struct task_struct *thread, *caller;
 298
 299         BUG_ON(!mutex_is_locked(&current->signal->cred_guard_mutex));
 300         assert_spin_locked(&current->sighand->siglock);
 301
 302         /* Synchronize all threads. */
 303         caller = current;
 304         for_each_thread(caller, thread) {
 305                 /* Skip current, since it needs no changes. */
 306                 if (thread == caller)
 307                         continue;
 308
 309                 /* Get a task reference for the new leaf node. */
 310                 get_seccomp_filter(caller);
 311                 /*
 312                  * Drop the task reference to the shared ancestor since
 313                  * current's path will hold a reference.  (This also
 314                  * allows a put before the assignment.)
 315                  */
 316                 put_seccomp_filter(thread);
 317                 smp_store_release(&thread->seccomp.filter,
 318                                   caller->seccomp.filter);
 319                 /*
 320                  * Opt the other thread into seccomp if needed.
 321                  * As threads are considered to be trust-realm
 322                  * equivalent (see ptrace_may_access), it is safe to
 323                  * allow one thread to transition the other.
 324                  */
 325                 if (thread->seccomp.mode == SECCOMP_MODE_DISABLED) {
 326                         /*
 327                          * Don't let an unprivileged task work around
 328                          * the no_new_privs restriction by creating
 329                          * a thread that sets it up, enters seccomp,
 330                          * then dies.
 331                          */
 332                         if (task_no_new_privs(caller))
 333                                 task_set_no_new_privs(thread);
 334
 335                         seccomp_assign_mode(thread, SECCOMP_MODE_FILTER);
 336                 }
 337         }
 338 }
 339
 340 /**
 341  * seccomp_prepare_filter: Prepares a seccomp filter for use.
 342  * @fprog: BPF program to install
 343  *
 344  * Returns filter on success or an ERR_PTR on failure.
 345  */
 346 static struct seccomp_filter *seccomp_prepare_filter(struct sock_fprog *fprog)
 347 {
 348         struct seccomp_filter *sfilter;
 349         int ret;
 350
 351         if (fprog->len == 0 || fprog->len > BPF_MAXINSNS)
 352                 return ERR_PTR(-EINVAL);
 353
 354         BUG_ON(INT_MAX / fprog->len < sizeof(struct sock_filter));
 355
 356         /*
 357          * Installing a seccomp filter requires that the task has
 358          * CAP_SYS_ADMIN in its namespace or be running with no_new_privs.
 359          * This avoids scenarios where unprivileged tasks can affect the
 360          * behavior of privileged children.
 361          */
 362         if (!task_no_new_privs(current) &&
 363             security_capable_noaudit(current_cred(), current_user_ns(),
 364                                      CAP_SYS_ADMIN) != 0)
 365                 return ERR_PTR(-EACCES);
 366
 367         /* Allocate a new seccomp_filter */
 368         sfilter = kzalloc(sizeof(*sfilter), GFP_KERNEL | __GFP_NOWARN);
 369         if (!sfilter)
 370                 return ERR_PTR(-ENOMEM);
 371
 372         ret = bpf_prog_create_from_user(&sfilter->prog, fprog,
 373                                         seccomp_check_filter);
 374         if (ret < 0) {
 375                 kfree(sfilter);
 376                 return ERR_PTR(ret);
 377         }
 378
 379         atomic_set(&sfilter->usage, 1);
 380
 381         return sfilter;
 382 }
 383
 384 /**
 385  * seccomp_prepare_user_filter - prepares a user-supplied sock_fprog
 386  * @user_filter: pointer to the user data containing a sock_fprog.
 387  *
 388  * Returns 0 on success and non-zero otherwise.
 389  */
 390 static struct seccomp_filter *
 391 seccomp_prepare_user_filter(const char __user *user_filter)
 392 {
 393         struct sock_fprog fprog;
 394         struct seccomp_filter *filter = ERR_PTR(-EFAULT);
 395
 396 #ifdef CONFIG_COMPAT
 397         if (is_compat_task()) {
 398                 struct compat_sock_fprog fprog32;
 399                 if (copy_from_user(&fprog32, user_filter, sizeof(fprog32)))
 400                         goto out;
 401                 fprog.len = fprog32.len;
 402                 fprog.filter = compat_ptr(fprog32.filter);
 403         } else /* falls through to the if below. */
 404 #endif
 405         if (copy_from_user(&fprog, user_filter, sizeof(fprog)))
 406                 goto out;
 407         filter = seccomp_prepare_filter(&fprog);
 408 out:
 409         return filter;
 410 }
 411
 412 /**
 413  * seccomp_attach_filter: validate and attach filter
 414  * @flags:  flags to change filter behavior
 415  * @filter: seccomp filter to add to the current process
 416  *
 417  * Caller must be holding current->sighand->siglock lock.
 418  *
 419  * Returns 0 on success, -ve on error.
 420  */
 421 static long seccomp_attach_filter(unsigned int flags,
 422                                   struct seccomp_filter *filter)
 423 {
 424         unsigned long total_insns;
 425         struct seccomp_filter *walker;
 426
 427         assert_spin_locked(&current->sighand->siglock);
 428
 429         /* Validate resulting filter length. */
 430         total_insns = filter->prog->len;
 431         for (walker = current->seccomp.filter; walker; walker = walker->prev)
 432                 total_insns += walker->prog->len + 4;  /* 4 instr penalty */
 433         if (total_insns > MAX_INSNS_PER_PATH)
 434                 return -ENOMEM;
 435
 436         /* If thread sync has been requested, check that it is possible. */
 437         if (flags & SECCOMP_FILTER_FLAG_TSYNC) {
 438                 int ret;
 439
 440                 ret = seccomp_can_sync_threads();
 441                 if (ret)
 442                         return ret;
 443         }
 444
 445         /*
 446          * If there is an existing filter, make it the prev and don't drop its
 447          * task reference.
 448          */
 449         filter->prev = current->seccomp.filter;
 450         current->seccomp.filter = filter;
 451
 452         /* Now that the new filter is in place, synchronize to all threads. */
 453         if (flags & SECCOMP_FILTER_FLAG_TSYNC)
 454                 seccomp_sync_threads();
 455
 456         return 0;
 457 }
 458
 459 /* get_seccomp_filter - increments the reference count of the filter on @tsk */
 460 void get_seccomp_filter(struct task_struct *tsk)
 461 {
 462         struct seccomp_filter *orig = tsk->seccomp.filter;
 463         if (!orig)
 464                 return;
 465         /* Reference count is bounded by the number of total processes. */
 466         atomic_inc(&orig->usage);
 467 }
 468
 469 static inline void seccomp_filter_free(struct seccomp_filter *filter)
 470 {
 471         if (filter) {
 472                 bpf_prog_free(filter->prog);
 473                 kfree(filter);
 474         }
 475 }
 476
 477 /* put_seccomp_filter - decrements the ref count of tsk->seccomp.filter */
 478 void put_seccomp_filter(struct task_struct *tsk)
 479 {
 480         struct seccomp_filter *orig = tsk->seccomp.filter;
 481         /* Clean up single-reference branches iteratively. */
 482         while (orig && atomic_dec_and_test(&orig->usage)) {
 483                 struct seccomp_filter *freeme = orig;
 484                 orig = orig->prev;
 485                 seccomp_filter_free(freeme);
 486         }
 487 }
 488
 489 /**
 490  * seccomp_send_sigsys - signals the task to allow in-process syscall emulation
 491  * @syscall: syscall number to send to userland
 492  * @reason: filter-supplied reason code to send to userland (via si_errno)
 493  *
 494  * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info.
 495  */
 496 static void seccomp_send_sigsys(int syscall, int reason)
 497 {
 498         struct siginfo info;
 499         memset(&info, 0, sizeof(info));
 500         info.si_signo = SIGSYS;
 501         info.si_code = SYS_SECCOMP;
 502         info.si_call_addr = (void __user *)KSTK_EIP(current);
 503         info.si_errno = reason;
 504         info.si_arch = syscall_get_arch();
 505         info.si_syscall = syscall;
 506         force_sig_info(SIGSYS, &info, current);
 507 }
 508 #endif  /* CONFIG_SECCOMP_FILTER */
 509
 510 /*
 511  * Secure computing mode 1 allows only read/write/exit/sigreturn.
 512  * To be fully secure this must be combined with rlimit
 513  * to limit the stack allocations too.
 514  */
 515 static int mode1_syscalls[] = {
 516         __NR_seccomp_read, __NR_seccomp_write, __NR_seccomp_exit, __NR_seccomp_sigreturn,
 517         0, /* null terminated */
 518 };
 519
 520 #ifdef CONFIG_COMPAT
 521 static int mode1_syscalls_32[] = {
 522         __NR_seccomp_read_32, __NR_seccomp_write_32, __NR_seccomp_exit_32, __NR_seccomp_sigreturn_32,
 523         0, /* null terminated */
 524 };
 525 #endif
 526
 527 static void __secure_computing_strict(int this_syscall)
 528 {
 529         int *syscall_whitelist = mode1_syscalls;
 530 #ifdef CONFIG_COMPAT
 531         if (is_compat_task())
 532                 syscall_whitelist = mode1_syscalls_32;
 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 (config_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 int __secure_computing(void)
 564 {
 565         u32 phase1_result = seccomp_phase1(NULL);
 566
 567         if (likely(phase1_result == SECCOMP_PHASE1_OK))
 568                 return 0;
 569         else if (likely(phase1_result == SECCOMP_PHASE1_SKIP))
 570                 return -1;
 571         else
 572                 return seccomp_phase2(phase1_result);
 573 }
 574
 575 #ifdef CONFIG_SECCOMP_FILTER
 576 static u32 __seccomp_phase1_filter(int this_syscall, struct seccomp_data *sd)
 577 {
 578         u32 filter_ret, action;
 579         int data;
 580
 581         /*
 582          * Make sure that any changes to mode from another thread have
 583          * been seen after TIF_SECCOMP was seen.
 584          */
 585         rmb();
 586
 587         filter_ret = seccomp_run_filters(sd);
 588         data = filter_ret & SECCOMP_RET_DATA;
 589         action = filter_ret & SECCOMP_RET_ACTION;
 590
 591         switch (action) {
 592         case SECCOMP_RET_ERRNO:
 593                 /* Set low-order bits as an errno, capped at MAX_ERRNO. */
 594                 if (data > MAX_ERRNO)
 595                         data = MAX_ERRNO;
 596                 syscall_set_return_value(current, task_pt_regs(current),
 597                                          -data, 0);
 598                 goto skip;
 599
 600         case SECCOMP_RET_TRAP:
 601                 /* Show the handler the original registers. */
 602                 syscall_rollback(current, task_pt_regs(current));
 603                 /* Let the filter pass back 16 bits of data. */
 604                 seccomp_send_sigsys(this_syscall, data);
 605                 goto skip;
 606
 607         case SECCOMP_RET_TRACE:
 608                 return filter_ret;  /* Save the rest for phase 2. */
 609
 610         case SECCOMP_RET_ALLOW:
 611                 return SECCOMP_PHASE1_OK;
 612
 613         case SECCOMP_RET_KILL:
 614         default:
 615                 audit_seccomp(this_syscall, SIGSYS, action);
 616                 do_exit(SIGSYS);
 617         }
 618
 619         unreachable();
 620
 621 skip:
 622         audit_seccomp(this_syscall, 0, action);
 623         return SECCOMP_PHASE1_SKIP;
 624 }
 625 #endif
 626
 627 /**
 628  * seccomp_phase1() - run fast path seccomp checks on the current syscall
 629  * @arg sd: The seccomp_data or NULL
 630  *
 631  * This only reads pt_regs via the syscall_xyz helpers.  The only change
 632  * it will make to pt_regs is via syscall_set_return_value, and it will
 633  * only do that if it returns SECCOMP_PHASE1_SKIP.
 634  *
 635  * If sd is provided, it will not read pt_regs at all.
 636  *
 637  * It may also call do_exit or force a signal; these actions must be
 638  * safe.
 639  *
 640  * If it returns SECCOMP_PHASE1_OK, the syscall passes checks and should
 641  * be processed normally.
 642  *
 643  * If it returns SECCOMP_PHASE1_SKIP, then the syscall should not be
 644  * invoked.  In this case, seccomp_phase1 will have set the return value
 645  * using syscall_set_return_value.
 646  *
 647  * If it returns anything else, then the return value should be passed
 648  * to seccomp_phase2 from a context in which ptrace hooks are safe.
 649  */
 650 u32 seccomp_phase1(struct seccomp_data *sd)
 651 {
 652         int mode = current->seccomp.mode;
 653         int this_syscall = sd ? sd->nr :
 654                 syscall_get_nr(current, task_pt_regs(current));
 655
 656         if (config_enabled(CONFIG_CHECKPOINT_RESTORE) &&
 657             unlikely(current->ptrace & PT_SUSPEND_SECCOMP))
 658                 return SECCOMP_PHASE1_OK;
 659
 660         switch (mode) {
 661         case SECCOMP_MODE_STRICT:
 662                 __secure_computing_strict(this_syscall);  /* may call do_exit */
 663                 return SECCOMP_PHASE1_OK;
 664 #ifdef CONFIG_SECCOMP_FILTER
 665         case SECCOMP_MODE_FILTER:
 666                 return __seccomp_phase1_filter(this_syscall, sd);
 667 #endif
 668         default:
 669                 BUG();
 670         }
 671 }
 672
 673 /**
 674  * seccomp_phase2() - finish slow path seccomp work for the current syscall
 675  * @phase1_result: The return value from seccomp_phase1()
 676  *
 677  * This must be called from a context in which ptrace hooks can be used.
 678  *
 679  * Returns 0 if the syscall should be processed or -1 to skip the syscall.
 680  */
 681 int seccomp_phase2(u32 phase1_result)
 682 {
 683         struct pt_regs *regs = task_pt_regs(current);
 684         u32 action = phase1_result & SECCOMP_RET_ACTION;
 685         int data = phase1_result & SECCOMP_RET_DATA;
 686
 687         BUG_ON(action != SECCOMP_RET_TRACE);
 688
 689         audit_seccomp(syscall_get_nr(current, regs), 0, action);
 690
 691         /* Skip these calls if there is no tracer. */
 692         if (!ptrace_event_enabled(current, PTRACE_EVENT_SECCOMP)) {
 693                 syscall_set_return_value(current, regs,
 694                                          -ENOSYS, 0);
 695                 return -1;
 696         }
 697
 698         /* Allow the BPF to provide the event message */
 699         ptrace_event(PTRACE_EVENT_SECCOMP, data);
 700         /*
 701          * The delivery of a fatal signal during event
 702          * notification may silently skip tracer notification.
 703          * Terminating the task now avoids executing a system
 704          * call that may not be intended.
 705          */
 706         if (fatal_signal_pending(current))
 707                 do_exit(SIGSYS);
 708         if (syscall_get_nr(current, regs) < 0)
 709                 return -1;  /* Explicit request to skip. */
 710
 711         return 0;
 712 }
 713 #endif /* CONFIG_HAVE_ARCH_SECCOMP_FILTER */
 714
 715 long prctl_get_seccomp(void)
 716 {
 717         return current->seccomp.mode;
 718 }
 719
 720 /**
 721  * seccomp_set_mode_strict: internal function for setting strict seccomp
 722  *
 723  * Once current->seccomp.mode is non-zero, it may not be changed.
 724  *
 725  * Returns 0 on success or -EINVAL on failure.
 726  */
 727 static long seccomp_set_mode_strict(void)
 728 {
 729         const unsigned long seccomp_mode = SECCOMP_MODE_STRICT;
 730         long ret = -EINVAL;
 731
 732         spin_lock_irq(&current->sighand->siglock);
 733
 734         if (!seccomp_may_assign_mode(seccomp_mode))
 735                 goto out;
 736
 737 #ifdef TIF_NOTSC
 738         disable_TSC();
 739 #endif
 740         seccomp_assign_mode(current, seccomp_mode);
 741         ret = 0;
 742
 743 out:
 744         spin_unlock_irq(&current->sighand->siglock);
 745
 746         return ret;
 747 }
 748
 749 #ifdef CONFIG_SECCOMP_FILTER
 750 /**
 751  * seccomp_set_mode_filter: internal function for setting seccomp filter
 752  * @flags:  flags to change filter behavior
 753  * @filter: struct sock_fprog containing filter
 754  *
 755  * This function may be called repeatedly to install additional filters.
 756  * Every filter successfully installed will be evaluated (in reverse order)
 757  * for each system call the task makes.
 758  *
 759  * Once current->seccomp.mode is non-zero, it may not be changed.
 760  *
 761  * Returns 0 on success or -EINVAL on failure.
 762  */
 763 static long seccomp_set_mode_filter(unsigned int flags,
 764                                     const char __user *filter)
 765 {
 766         const unsigned long seccomp_mode = SECCOMP_MODE_FILTER;
 767         struct seccomp_filter *prepared = NULL;
 768         long ret = -EINVAL;
 769
 770         /* Validate flags. */
 771         if (flags & ~SECCOMP_FILTER_FLAG_MASK)
 772                 return -EINVAL;
 773
 774         /* Prepare the new filter before holding any locks. */
 775         prepared = seccomp_prepare_user_filter(filter);
 776         if (IS_ERR(prepared))
 777                 return PTR_ERR(prepared);
 778
 779         /*
 780          * Make sure we cannot change seccomp or nnp state via TSYNC
 781          * while another thread is in the middle of calling exec.
 782          */
 783         if (flags & SECCOMP_FILTER_FLAG_TSYNC &&
 784             mutex_lock_killable(&current->signal->cred_guard_mutex))
 785                 goto out_free;
 786
 787         spin_lock_irq(&current->sighand->siglock);
 788
 789         if (!seccomp_may_assign_mode(seccomp_mode))
 790                 goto out;
 791
 792         ret = seccomp_attach_filter(flags, prepared);
 793         if (ret)
 794                 goto out;
 795         /* Do not free the successfully attached filter. */
 796         prepared = NULL;
 797
 798         seccomp_assign_mode(current, seccomp_mode);
 799 out:
 800         spin_unlock_irq(&current->sighand->siglock);
 801         if (flags & SECCOMP_FILTER_FLAG_TSYNC)
 802                 mutex_unlock(&current->signal->cred_guard_mutex);
 803 out_free:
 804         seccomp_filter_free(prepared);
 805         return ret;
 806 }
 807 #else
 808 static inline long seccomp_set_mode_filter(unsigned int flags,
 809                                            const char __user *filter)
 810 {
 811         return -EINVAL;
 812 }
 813 #endif
 814
 815 /* Common entry point for both prctl and syscall. */
 816 static long do_seccomp(unsigned int op, unsigned int flags,
 817                        const char __user *uargs)
 818 {
 819         switch (op) {
 820         case SECCOMP_SET_MODE_STRICT:
 821                 if (flags != 0 || uargs != NULL)
 822                         return -EINVAL;
 823                 return seccomp_set_mode_strict();
 824         case SECCOMP_SET_MODE_FILTER:
 825                 return seccomp_set_mode_filter(flags, uargs);
 826         default:
 827                 return -EINVAL;
 828         }
 829 }
 830
 831 SYSCALL_DEFINE3(seccomp, unsigned int, op, unsigned int, flags,
 832                          const char __user *, uargs)
 833 {
 834         return do_seccomp(op, flags, uargs);
 835 }
 836
 837 /**
 838  * prctl_set_seccomp: configures current->seccomp.mode
 839  * @seccomp_mode: requested mode to use
 840  * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER
 841  *
 842  * Returns 0 on success or -EINVAL on failure.
 843  */
 844 long prctl_set_seccomp(unsigned long seccomp_mode, char __user *filter)
 845 {
 846         unsigned int op;
 847         char __user *uargs;
 848
 849         switch (seccomp_mode) {
 850         case SECCOMP_MODE_STRICT:
 851                 op = SECCOMP_SET_MODE_STRICT;
 852                 /*
 853                  * Setting strict mode through prctl always ignored filter,
 854                  * so make sure it is always NULL here to pass the internal
 855                  * check in do_seccomp().
 856                  */
 857                 uargs = NULL;
 858                 break;
 859         case SECCOMP_MODE_FILTER:
 860                 op = SECCOMP_SET_MODE_FILTER;
 861                 uargs = filter;
 862                 break;
 863         default:
 864                 return -EINVAL;
 865         }
 866
 867         /* prctl interface doesn't have flags, so they are always zero. */
 868         return do_seccomp(op, 0, uargs);
 869 }