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         const bool save_orig = config_enabled(CONFIG_CHECKPOINT_RESTORE);
 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, save_orig);
 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_destroy(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 (config_enabled(CONFIG_CHECKPOINT_RESTORE) &&
 553             unlikely(current->ptrace & PT_SUSPEND_SECCOMP))
 554                 return;
 555
 556         if (mode == SECCOMP_MODE_DISABLED)
 557                 return;
 558         else if (mode == SECCOMP_MODE_STRICT)
 559                 __secure_computing_strict(this_syscall);
 560         else
 561                 BUG();
 562 }
 563 #else
 564 int __secure_computing(void)
 565 {
 566         u32 phase1_result = seccomp_phase1(NULL);
 567
 568         if (likely(phase1_result == SECCOMP_PHASE1_OK))
 569                 return 0;
 570         else if (likely(phase1_result == SECCOMP_PHASE1_SKIP))
 571                 return -1;
 572         else
 573                 return seccomp_phase2(phase1_result);
 574 }
 575
 576 #ifdef CONFIG_SECCOMP_FILTER
 577 static u32 __seccomp_phase1_filter(int this_syscall, struct seccomp_data *sd)
 578 {
 579         u32 filter_ret, action;
 580         int data;
 581
 582         /*
 583          * Make sure that any changes to mode from another thread have
 584          * been seen after TIF_SECCOMP was seen.
 585          */
 586         rmb();
 587
 588         filter_ret = seccomp_run_filters(sd);
 589         data = filter_ret & SECCOMP_RET_DATA;
 590         action = filter_ret & SECCOMP_RET_ACTION;
 591
 592         switch (action) {
 593         case SECCOMP_RET_ERRNO:
 594                 /* Set low-order bits as an errno, capped at MAX_ERRNO. */
 595                 if (data > MAX_ERRNO)
 596                         data = MAX_ERRNO;
 597                 syscall_set_return_value(current, task_pt_regs(current),
 598                                          -data, 0);
 599                 goto skip;
 600
 601         case SECCOMP_RET_TRAP:
 602                 /* Show the handler the original registers. */
 603                 syscall_rollback(current, task_pt_regs(current));
 604                 /* Let the filter pass back 16 bits of data. */
 605                 seccomp_send_sigsys(this_syscall, data);
 606                 goto skip;
 607
 608         case SECCOMP_RET_TRACE:
 609                 return filter_ret;  /* Save the rest for phase 2. */
 610
 611         case SECCOMP_RET_ALLOW:
 612                 return SECCOMP_PHASE1_OK;
 613
 614         case SECCOMP_RET_KILL:
 615         default:
 616                 audit_seccomp(this_syscall, SIGSYS, action);
 617                 do_exit(SIGSYS);
 618         }
 619
 620         unreachable();
 621
 622 skip:
 623         audit_seccomp(this_syscall, 0, action);
 624         return SECCOMP_PHASE1_SKIP;
 625 }
 626 #endif
 627
 628 /**
 629  * seccomp_phase1() - run fast path seccomp checks on the current syscall
 630  * @arg sd: The seccomp_data or NULL
 631  *
 632  * This only reads pt_regs via the syscall_xyz helpers.  The only change
 633  * it will make to pt_regs is via syscall_set_return_value, and it will
 634  * only do that if it returns SECCOMP_PHASE1_SKIP.
 635  *
 636  * If sd is provided, it will not read pt_regs at all.
 637  *
 638  * It may also call do_exit or force a signal; these actions must be
 639  * safe.
 640  *
 641  * If it returns SECCOMP_PHASE1_OK, the syscall passes checks and should
 642  * be processed normally.
 643  *
 644  * If it returns SECCOMP_PHASE1_SKIP, then the syscall should not be
 645  * invoked.  In this case, seccomp_phase1 will have set the return value
 646  * using syscall_set_return_value.
 647  *
 648  * If it returns anything else, then the return value should be passed
 649  * to seccomp_phase2 from a context in which ptrace hooks are safe.
 650  */
 651 u32 seccomp_phase1(struct seccomp_data *sd)
 652 {
 653         int mode = current->seccomp.mode;
 654         int this_syscall = sd ? sd->nr :
 655                 syscall_get_nr(current, task_pt_regs(current));
 656
 657         if (config_enabled(CONFIG_CHECKPOINT_RESTORE) &&
 658             unlikely(current->ptrace & PT_SUSPEND_SECCOMP))
 659                 return SECCOMP_PHASE1_OK;
 660
 661         switch (mode) {
 662         case SECCOMP_MODE_STRICT:
 663                 __secure_computing_strict(this_syscall);  /* may call do_exit */
 664                 return SECCOMP_PHASE1_OK;
 665 #ifdef CONFIG_SECCOMP_FILTER
 666         case SECCOMP_MODE_FILTER:
 667                 return __seccomp_phase1_filter(this_syscall, sd);
 668 #endif
 669         default:
 670                 BUG();
 671         }
 672 }
 673
 674 /**
 675  * seccomp_phase2() - finish slow path seccomp work for the current syscall
 676  * @phase1_result: The return value from seccomp_phase1()
 677  *
 678  * This must be called from a context in which ptrace hooks can be used.
 679  *
 680  * Returns 0 if the syscall should be processed or -1 to skip the syscall.
 681  */
 682 int seccomp_phase2(u32 phase1_result)
 683 {
 684         struct pt_regs *regs = task_pt_regs(current);
 685         u32 action = phase1_result & SECCOMP_RET_ACTION;
 686         int data = phase1_result & SECCOMP_RET_DATA;
 687
 688         BUG_ON(action != SECCOMP_RET_TRACE);
 689
 690         audit_seccomp(syscall_get_nr(current, regs), 0, action);
 691
 692         /* Skip these calls if there is no tracer. */
 693         if (!ptrace_event_enabled(current, PTRACE_EVENT_SECCOMP)) {
 694                 syscall_set_return_value(current, regs,
 695                                          -ENOSYS, 0);
 696                 return -1;
 697         }
 698
 699         /* Allow the BPF to provide the event message */
 700         ptrace_event(PTRACE_EVENT_SECCOMP, data);
 701         /*
 702          * The delivery of a fatal signal during event
 703          * notification may silently skip tracer notification.
 704          * Terminating the task now avoids executing a system
 705          * call that may not be intended.
 706          */
 707         if (fatal_signal_pending(current))
 708                 do_exit(SIGSYS);
 709         if (syscall_get_nr(current, regs) < 0)
 710                 return -1;  /* Explicit request to skip. */
 711
 712         return 0;
 713 }
 714 #endif /* CONFIG_HAVE_ARCH_SECCOMP_FILTER */
 715
 716 long prctl_get_seccomp(void)
 717 {
 718         return current->seccomp.mode;
 719 }
 720
 721 /**
 722  * seccomp_set_mode_strict: internal function for setting strict seccomp
 723  *
 724  * Once current->seccomp.mode is non-zero, it may not be changed.
 725  *
 726  * Returns 0 on success or -EINVAL on failure.
 727  */
 728 static long seccomp_set_mode_strict(void)
 729 {
 730         const unsigned long seccomp_mode = SECCOMP_MODE_STRICT;
 731         long ret = -EINVAL;
 732
 733         spin_lock_irq(&current->sighand->siglock);
 734
 735         if (!seccomp_may_assign_mode(seccomp_mode))
 736                 goto out;
 737
 738 #ifdef TIF_NOTSC
 739         disable_TSC();
 740 #endif
 741         seccomp_assign_mode(current, seccomp_mode);
 742         ret = 0;
 743
 744 out:
 745         spin_unlock_irq(&current->sighand->siglock);
 746
 747         return ret;
 748 }
 749
 750 #ifdef CONFIG_SECCOMP_FILTER
 751 /**
 752  * seccomp_set_mode_filter: internal function for setting seccomp filter
 753  * @flags:  flags to change filter behavior
 754  * @filter: struct sock_fprog containing filter
 755  *
 756  * This function may be called repeatedly to install additional filters.
 757  * Every filter successfully installed will be evaluated (in reverse order)
 758  * for each system call the task makes.
 759  *
 760  * Once current->seccomp.mode is non-zero, it may not be changed.
 761  *
 762  * Returns 0 on success or -EINVAL on failure.
 763  */
 764 static long seccomp_set_mode_filter(unsigned int flags,
 765                                     const char __user *filter)
 766 {
 767         const unsigned long seccomp_mode = SECCOMP_MODE_FILTER;
 768         struct seccomp_filter *prepared = NULL;
 769         long ret = -EINVAL;
 770
 771         /* Validate flags. */
 772         if (flags & ~SECCOMP_FILTER_FLAG_MASK)
 773                 return -EINVAL;
 774
 775         /* Prepare the new filter before holding any locks. */
 776         prepared = seccomp_prepare_user_filter(filter);
 777         if (IS_ERR(prepared))
 778                 return PTR_ERR(prepared);
 779
 780         /*
 781          * Make sure we cannot change seccomp or nnp state via TSYNC
 782          * while another thread is in the middle of calling exec.
 783          */
 784         if (flags & SECCOMP_FILTER_FLAG_TSYNC &&
 785             mutex_lock_killable(&current->signal->cred_guard_mutex))
 786                 goto out_free;
 787
 788         spin_lock_irq(&current->sighand->siglock);
 789
 790         if (!seccomp_may_assign_mode(seccomp_mode))
 791                 goto out;
 792
 793         ret = seccomp_attach_filter(flags, prepared);
 794         if (ret)
 795                 goto out;
 796         /* Do not free the successfully attached filter. */
 797         prepared = NULL;
 798
 799         seccomp_assign_mode(current, seccomp_mode);
 800 out:
 801         spin_unlock_irq(&current->sighand->siglock);
 802         if (flags & SECCOMP_FILTER_FLAG_TSYNC)
 803                 mutex_unlock(&current->signal->cred_guard_mutex);
 804 out_free:
 805         seccomp_filter_free(prepared);
 806         return ret;
 807 }
 808 #else
 809 static inline long seccomp_set_mode_filter(unsigned int flags,
 810                                            const char __user *filter)
 811 {
 812         return -EINVAL;
 813 }
 814 #endif
 815
 816 /* Common entry point for both prctl and syscall. */
 817 static long do_seccomp(unsigned int op, unsigned int flags,
 818                        const char __user *uargs)
 819 {
 820         switch (op) {
 821         case SECCOMP_SET_MODE_STRICT:
 822                 if (flags != 0 || uargs != NULL)
 823                         return -EINVAL;
 824                 return seccomp_set_mode_strict();
 825         case SECCOMP_SET_MODE_FILTER:
 826                 return seccomp_set_mode_filter(flags, uargs);
 827         default:
 828                 return -EINVAL;
 829         }
 830 }
 831
 832 SYSCALL_DEFINE3(seccomp, unsigned int, op, unsigned int, flags,
 833                          const char __user *, uargs)
 834 {
 835         return do_seccomp(op, flags, uargs);
 836 }
 837
 838 /**
 839  * prctl_set_seccomp: configures current->seccomp.mode
 840  * @seccomp_mode: requested mode to use
 841  * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER
 842  *
 843  * Returns 0 on success or -EINVAL on failure.
 844  */
 845 long prctl_set_seccomp(unsigned long seccomp_mode, char __user *filter)
 846 {
 847         unsigned int op;
 848         char __user *uargs;
 849
 850         switch (seccomp_mode) {
 851         case SECCOMP_MODE_STRICT:
 852                 op = SECCOMP_SET_MODE_STRICT;
 853                 /*
 854                  * Setting strict mode through prctl always ignored filter,
 855                  * so make sure it is always NULL here to pass the internal
 856                  * check in do_seccomp().
 857                  */
 858                 uargs = NULL;
 859                 break;
 860         case SECCOMP_MODE_FILTER:
 861                 op = SECCOMP_SET_MODE_FILTER;
 862                 uargs = filter;
 863                 break;
 864         default:
 865                 return -EINVAL;
 866         }
 867
 868         /* prctl interface doesn't have flags, so they are always zero. */
 869         return do_seccomp(op, 0, uargs);
 870 }
 871
 872 #if defined(CONFIG_SECCOMP_FILTER) && defined(CONFIG_CHECKPOINT_RESTORE)
 873 long seccomp_get_filter(struct task_struct *task, unsigned long filter_off,
 874                         void __user *data)
 875 {
 876         struct seccomp_filter *filter;
 877         struct sock_fprog_kern *fprog;
 878         long ret;
 879         unsigned long count = 0;
 880
 881         if (!capable(CAP_SYS_ADMIN) ||
 882             current->seccomp.mode != SECCOMP_MODE_DISABLED) {
 883                 return -EACCES;
 884         }
 885
 886         spin_lock_irq(&task->sighand->siglock);
 887         if (task->seccomp.mode != SECCOMP_MODE_FILTER) {
 888                 ret = -EINVAL;
 889                 goto out;
 890         }
 891
 892         filter = task->seccomp.filter;
 893         while (filter) {
 894                 filter = filter->prev;
 895                 count++;
 896         }
 897
 898         if (filter_off >= count) {
 899                 ret = -ENOENT;
 900                 goto out;
 901         }
 902         count -= filter_off;
 903
 904         filter = task->seccomp.filter;
 905         while (filter && count > 1) {
 906                 filter = filter->prev;
 907                 count--;
 908         }
 909
 910         if (WARN_ON(count != 1 || !filter)) {
 911                 /* The filter tree shouldn't shrink while we're using it. */
 912                 ret = -ENOENT;
 913                 goto out;
 914         }
 915
 916         fprog = filter->prog->orig_prog;
 917         if (!fprog) {
 918                 /* This must be a new non-cBPF filter, since we save every
 919                  * every cBPF filter's orig_prog above when
 920                  * CONFIG_CHECKPOINT_RESTORE is enabled.
 921                  */
 922                 ret = -EMEDIUMTYPE;
 923                 goto out;
 924         }
 925
 926         ret = fprog->len;
 927         if (!data)
 928                 goto out;
 929
 930         get_seccomp_filter(task);
 931         spin_unlock_irq(&task->sighand->siglock);
 932
 933         if (copy_to_user(data, fprog->filter, bpf_classic_proglen(fprog)))
 934                 ret = -EFAULT;
 935
 936         put_seccomp_filter(task);
 937         return ret;
 938
 939 out:
 940         spin_unlock_irq(&task->sighand->siglock);
 941         return ret;
 942 }
 943 #endif