kernel/seccomp.c

   1 /*
   2  * linux/kernel/seccomp.c
   3  *
   4  * Copyright 2004-2005  Andrea Arcangeli <andrea@cpushare.com>
   5  *
   6  * Copyright (C) 2012 Google, Inc.
   7  * Will Drewry <wad@chromium.org>
   8  *
   9  * This defines a simple but solid secure-computing facility.
  10  *
  11  * Mode 1 uses a fixed list of allowed system calls.
  12  * Mode 2 allows user-defined system call filters in the form
  13  *        of Berkeley Packet Filters/Linux Socket Filters.
  14  */
  15
  16 #include <linux/atomic.h>
  17 #include <linux/audit.h>
  18 #include <linux/compat.h>
  19 #include <linux/nospec.h>
  20 #include <linux/prctl.h>
  21 #include <linux/sched.h>
  22 #include <linux/seccomp.h>
  23 #include <linux/syscalls.h>
  24
  25 /* #define SECCOMP_DEBUG 1 */
  26
  27 #ifdef CONFIG_SECCOMP_FILTER
  28 #include <asm/syscall.h>
  29 #include <linux/filter.h>
  30 #include <linux/ptrace.h>
  31 #include <linux/security.h>
  32 #include <linux/slab.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 sk_filter *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 sk_chk_filter) 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(int syscall)
 177 {
 178         struct seccomp_filter *f;
 179         struct seccomp_data sd;
 180         u32 ret = SECCOMP_RET_ALLOW;
 181
 182         /* Ensure unexpected behavior doesn't result in failing open. */
 183         if (WARN_ON(current->seccomp.filter == NULL))
 184                 return SECCOMP_RET_KILL;
 185
 186         populate_seccomp_data(&sd);
 187
 188         /*
 189          * All filters in the list are evaluated and the lowest BPF return
 190          * value always takes priority (ignoring the DATA).
 191          */
 192         for (f = current->seccomp.filter; f; f = f->prev) {
 193                 u32 cur_ret = SK_RUN_FILTER(f->prog, (void *)&sd);
 194
 195                 if ((cur_ret & SECCOMP_RET_ACTION) < (ret & SECCOMP_RET_ACTION))
 196                         ret = cur_ret;
 197         }
 198         return ret;
 199 }
 200 #endif /* CONFIG_SECCOMP_FILTER */
 201
 202 static inline bool seccomp_may_assign_mode(unsigned long seccomp_mode)
 203 {
 204         if (current->seccomp.mode && current->seccomp.mode != seccomp_mode)
 205                 return false;
 206
 207         return true;
 208 }
 209
 210 void __weak arch_seccomp_spec_mitigate(struct task_struct *task) { }
 211
 212 static inline void seccomp_assign_mode(unsigned long seccomp_mode,
 213                                        unsigned long flags)
 214 {
 215         current->seccomp.mode = seccomp_mode;
 216         /* Assume default seccomp processes want spec flaw mitigation. */
 217         if ((flags & SECCOMP_FILTER_FLAG_SPEC_ALLOW) == 0)
 218                 arch_seccomp_spec_mitigate(current);
 219         set_tsk_thread_flag(current, TIF_SECCOMP);
 220 }
 221
 222 #ifdef CONFIG_SECCOMP_FILTER
 223 /**
 224  * seccomp_attach_filter: Attaches a seccomp filter to current.
 225  * @fprog: BPF program to install
 226  *
 227  * Returns 0 on success or an errno on failure.
 228  */
 229 static long seccomp_attach_filter(struct sock_fprog *fprog)
 230 {
 231         struct seccomp_filter *filter;
 232         unsigned long fp_size = fprog->len * sizeof(struct sock_filter);
 233         unsigned long total_insns = fprog->len;
 234         struct sock_filter *fp;
 235         int new_len;
 236         long ret;
 237
 238         if (fprog->len == 0 || fprog->len > BPF_MAXINSNS)
 239                 return -EINVAL;
 240
 241         for (filter = current->seccomp.filter; filter; filter = filter->prev)
 242                 total_insns += filter->prog->len + 4;  /* include a 4 instr penalty */
 243         if (total_insns > MAX_INSNS_PER_PATH)
 244                 return -ENOMEM;
 245
 246         /*
 247          * Installing a seccomp filter requires that the task has
 248          * CAP_SYS_ADMIN in its namespace or be running with no_new_privs.
 249          * This avoids scenarios where unprivileged tasks can affect the
 250          * behavior of privileged children.
 251          */
 252         if (!task_no_new_privs(current) &&
 253             security_capable_noaudit(current_cred(), current_user_ns(),
 254                                      CAP_SYS_ADMIN) != 0)
 255                 return -EACCES;
 256
 257         fp = kzalloc(fp_size, GFP_KERNEL|__GFP_NOWARN);
 258         if (!fp)
 259                 return -ENOMEM;
 260
 261         /* Copy the instructions from fprog. */
 262         ret = -EFAULT;
 263         if (copy_from_user(fp, fprog->filter, fp_size))
 264                 goto free_prog;
 265
 266         /* Check and rewrite the fprog via the skb checker */
 267         ret = sk_chk_filter(fp, fprog->len);
 268         if (ret)
 269                 goto free_prog;
 270
 271         /* Check and rewrite the fprog for seccomp use */
 272         ret = seccomp_check_filter(fp, fprog->len);
 273         if (ret)
 274                 goto free_prog;
 275
 276         /* Convert 'sock_filter' insns to 'sock_filter_int' insns */
 277         ret = sk_convert_filter(fp, fprog->len, NULL, &new_len);
 278         if (ret)
 279                 goto free_prog;
 280
 281         /* Allocate a new seccomp_filter */
 282         ret = -ENOMEM;
 283         filter = kzalloc(sizeof(struct seccomp_filter),
 284                          GFP_KERNEL|__GFP_NOWARN);
 285         if (!filter)
 286                 goto free_prog;
 287
 288         filter->prog = kzalloc(sk_filter_size(new_len),
 289                                GFP_KERNEL|__GFP_NOWARN);
 290         if (!filter->prog)
 291                 goto free_filter;
 292
 293         ret = sk_convert_filter(fp, fprog->len, filter->prog->insnsi, &new_len);
 294         if (ret)
 295                 goto free_filter_prog;
 296         kfree(fp);
 297
 298         atomic_set(&filter->usage, 1);
 299         filter->prog->len = new_len;
 300
 301         sk_filter_select_runtime(filter->prog);
 302
 303         /*
 304          * If there is an existing filter, make it the prev and don't drop its
 305          * task reference.
 306          */
 307         filter->prev = current->seccomp.filter;
 308         current->seccomp.filter = filter;
 309         return 0;
 310
 311 free_filter_prog:
 312         kfree(filter->prog);
 313 free_filter:
 314         kfree(filter);
 315 free_prog:
 316         kfree(fp);
 317         return ret;
 318 }
 319
 320 /**
 321  * seccomp_attach_user_filter - attaches a user-supplied sock_fprog
 322  * @user_filter: pointer to the user data containing a sock_fprog.
 323  *
 324  * Returns 0 on success and non-zero otherwise.
 325  */
 326 static long seccomp_attach_user_filter(const char __user *user_filter)
 327 {
 328         struct sock_fprog fprog;
 329         long ret = -EFAULT;
 330
 331 #ifdef CONFIG_COMPAT
 332         if (is_compat_task()) {
 333                 struct compat_sock_fprog fprog32;
 334                 if (copy_from_user(&fprog32, user_filter, sizeof(fprog32)))
 335                         goto out;
 336                 fprog.len = fprog32.len;
 337                 fprog.filter = compat_ptr(fprog32.filter);
 338         } else /* falls through to the if below. */
 339 #endif
 340         if (copy_from_user(&fprog, user_filter, sizeof(fprog)))
 341                 goto out;
 342         ret = seccomp_attach_filter(&fprog);
 343 out:
 344         return ret;
 345 }
 346
 347 /* get_seccomp_filter - increments the reference count of the filter on @tsk */
 348 void get_seccomp_filter(struct task_struct *tsk)
 349 {
 350         struct seccomp_filter *orig = tsk->seccomp.filter;
 351         if (!orig)
 352                 return;
 353         /* Reference count is bounded by the number of total processes. */
 354         atomic_inc(&orig->usage);
 355 }
 356
 357 /* put_seccomp_filter - decrements the ref count of tsk->seccomp.filter */
 358 void put_seccomp_filter(struct task_struct *tsk)
 359 {
 360         struct seccomp_filter *orig = tsk->seccomp.filter;
 361         /* Clean up single-reference branches iteratively. */
 362         while (orig && atomic_dec_and_test(&orig->usage)) {
 363                 struct seccomp_filter *freeme = orig;
 364                 orig = orig->prev;
 365                 sk_filter_free(freeme->prog);
 366                 kfree(freeme);
 367         }
 368 }
 369
 370 /**
 371  * seccomp_send_sigsys - signals the task to allow in-process syscall emulation
 372  * @syscall: syscall number to send to userland
 373  * @reason: filter-supplied reason code to send to userland (via si_errno)
 374  *
 375  * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info.
 376  */
 377 static void seccomp_send_sigsys(int syscall, int reason)
 378 {
 379         struct siginfo info;
 380         memset(&info, 0, sizeof(info));
 381         info.si_signo = SIGSYS;
 382         info.si_code = SYS_SECCOMP;
 383         info.si_call_addr = (void __user *)KSTK_EIP(current);
 384         info.si_errno = reason;
 385         info.si_arch = syscall_get_arch();
 386         info.si_syscall = syscall;
 387         force_sig_info(SIGSYS, &info, current);
 388 }
 389 #endif  /* CONFIG_SECCOMP_FILTER */
 390
 391 /*
 392  * Secure computing mode 1 allows only read/write/exit/sigreturn.
 393  * To be fully secure this must be combined with rlimit
 394  * to limit the stack allocations too.
 395  */
 396 static int mode1_syscalls[] = {
 397         __NR_seccomp_read, __NR_seccomp_write, __NR_seccomp_exit, __NR_seccomp_sigreturn,
 398         0, /* null terminated */
 399 };
 400
 401 #ifdef CONFIG_COMPAT
 402 static int mode1_syscalls_32[] = {
 403         __NR_seccomp_read_32, __NR_seccomp_write_32, __NR_seccomp_exit_32, __NR_seccomp_sigreturn_32,
 404         0, /* null terminated */
 405 };
 406 #endif
 407
 408 int __secure_computing(int this_syscall)
 409 {
 410         int mode = current->seccomp.mode;
 411         int exit_sig = 0;
 412         int *syscall;
 413         u32 ret;
 414
 415         switch (mode) {
 416         case SECCOMP_MODE_STRICT:
 417                 syscall = mode1_syscalls;
 418 #ifdef CONFIG_COMPAT
 419                 if (is_compat_task())
 420                         syscall = mode1_syscalls_32;
 421 #endif
 422                 do {
 423                         if (*syscall == this_syscall)
 424                                 return 0;
 425                 } while (*++syscall);
 426                 exit_sig = SIGKILL;
 427                 ret = SECCOMP_RET_KILL;
 428                 break;
 429 #ifdef CONFIG_SECCOMP_FILTER
 430         case SECCOMP_MODE_FILTER: {
 431                 int data;
 432                 struct pt_regs *regs = task_pt_regs(current);
 433                 ret = seccomp_run_filters(this_syscall);
 434                 data = ret & SECCOMP_RET_DATA;
 435                 ret &= SECCOMP_RET_ACTION;
 436                 switch (ret) {
 437                 case SECCOMP_RET_ERRNO:
 438                         /* Set the low-order 16-bits as a errno. */
 439                         syscall_set_return_value(current, regs,
 440                                                  -data, 0);
 441                         goto skip;
 442                 case SECCOMP_RET_TRAP:
 443                         /* Show the handler the original registers. */
 444                         syscall_rollback(current, regs);
 445                         /* Let the filter pass back 16 bits of data. */
 446                         seccomp_send_sigsys(this_syscall, data);
 447                         goto skip;
 448                 case SECCOMP_RET_TRACE:
 449                         /* Skip these calls if there is no tracer. */
 450                         if (!ptrace_event_enabled(current, PTRACE_EVENT_SECCOMP)) {
 451                                 syscall_set_return_value(current, regs,
 452                                                          -ENOSYS, 0);
 453                                 goto skip;
 454                         }
 455                         /* Allow the BPF to provide the event message */
 456                         ptrace_event(PTRACE_EVENT_SECCOMP, data);
 457                         /*
 458                          * The delivery of a fatal signal during event
 459                          * notification may silently skip tracer notification.
 460                          * Terminating the task now avoids executing a system
 461                          * call that may not be intended.
 462                          */
 463                         if (fatal_signal_pending(current))
 464                                 break;
 465                         if (syscall_get_nr(current, regs) < 0)
 466                                 goto skip;  /* Explicit request to skip. */
 467
 468                         return 0;
 469                 case SECCOMP_RET_ALLOW:
 470                         return 0;
 471                 case SECCOMP_RET_KILL:
 472                 default:
 473                         break;
 474                 }
 475                 exit_sig = SIGSYS;
 476                 break;
 477         }
 478 #endif
 479         default:
 480                 BUG();
 481         }
 482
 483 #ifdef SECCOMP_DEBUG
 484         dump_stack();
 485 #endif
 486         audit_seccomp(this_syscall, exit_sig, ret);
 487         do_exit(exit_sig);
 488 #ifdef CONFIG_SECCOMP_FILTER
 489 skip:
 490         audit_seccomp(this_syscall, exit_sig, ret);
 491 #endif
 492         return -1;
 493 }
 494
 495 long prctl_get_seccomp(void)
 496 {
 497         return current->seccomp.mode;
 498 }
 499
 500 /**
 501  * seccomp_set_mode_strict: internal function for setting strict seccomp
 502  *
 503  * Once current->seccomp.mode is non-zero, it may not be changed.
 504  *
 505  * Returns 0 on success or -EINVAL on failure.
 506  */
 507 static long seccomp_set_mode_strict(void)
 508 {
 509         const unsigned long seccomp_mode = SECCOMP_MODE_STRICT;
 510         long ret = -EINVAL;
 511
 512         if (!seccomp_may_assign_mode(seccomp_mode))
 513                 goto out;
 514
 515 #ifdef TIF_NOTSC
 516         disable_TSC();
 517 #endif
 518         seccomp_assign_mode(seccomp_mode, 0);
 519         ret = 0;
 520
 521 out:
 522
 523         return ret;
 524 }
 525
 526 #ifdef CONFIG_SECCOMP_FILTER
 527 /**
 528  * seccomp_set_mode_filter: internal function for setting seccomp filter
 529  * @flags:  flags to change filter behavior
 530  * @filter: struct sock_fprog containing filter
 531  *
 532  * This function may be called repeatedly to install additional filters.
 533  * Every filter successfully installed will be evaluated (in reverse order)
 534  * for each system call the task makes.
 535  *
 536  * Once current->seccomp.mode is non-zero, it may not be changed.
 537  *
 538  * Returns 0 on success or -EINVAL on failure.
 539  */
 540 static long seccomp_set_mode_filter(unsigned int flags,
 541                                     const char __user *filter)
 542 {
 543         const unsigned long seccomp_mode = SECCOMP_MODE_FILTER;
 544         long ret = -EINVAL;
 545
 546         /* Validate flags. */
 547         if (flags & ~SECCOMP_FILTER_FLAG_MASK)
 548                 goto out;
 549
 550         if (!seccomp_may_assign_mode(seccomp_mode))
 551                 goto out;
 552
 553         ret = seccomp_attach_user_filter(filter);
 554         if (ret)
 555                 goto out;
 556
 557         seccomp_assign_mode(seccomp_mode, flags);
 558 out:
 559         return ret;
 560 }
 561 #else
 562 static inline long seccomp_set_mode_filter(unsigned int flags,
 563                                            const char __user *filter)
 564 {
 565         return -EINVAL;
 566 }
 567 #endif
 568
 569 /* Common entry point for both prctl and syscall. */
 570 static long do_seccomp(unsigned int op, unsigned int flags,
 571                        const char __user *uargs)
 572 {
 573         switch (op) {
 574         case SECCOMP_SET_MODE_STRICT:
 575                 if (flags != 0 || uargs != NULL)
 576                         return -EINVAL;
 577                 return seccomp_set_mode_strict();
 578         case SECCOMP_SET_MODE_FILTER:
 579                 return seccomp_set_mode_filter(flags, uargs);
 580         default:
 581                 return -EINVAL;
 582         }
 583 }
 584
 585 SYSCALL_DEFINE3(seccomp, unsigned int, op, unsigned int, flags,
 586                          const char __user *, uargs)
 587 {
 588         return do_seccomp(op, flags, uargs);
 589 }
 590
 591 /**
 592  * prctl_set_seccomp: configures current->seccomp.mode
 593  * @seccomp_mode: requested mode to use
 594  * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER
 595  *
 596  * Returns 0 on success or -EINVAL on failure.
 597  */
 598 long prctl_set_seccomp(unsigned long seccomp_mode, char __user *filter)
 599 {
 600         unsigned int op;
 601         char __user *uargs;
 602
 603         switch (seccomp_mode) {
 604         case SECCOMP_MODE_STRICT:
 605                 op = SECCOMP_SET_MODE_STRICT;
 606                 /*
 607                  * Setting strict mode through prctl always ignored filter,
 608                  * so make sure it is always NULL here to pass the internal
 609                  * check in do_seccomp().
 610                  */
 611                 uargs = NULL;
 612                 break;
 613         case SECCOMP_MODE_FILTER:
 614                 op = SECCOMP_SET_MODE_FILTER;
 615                 uargs = filter;
 616                 break;
 617         default:
 618                 return -EINVAL;
 619         }
 620
 621         /* prctl interface doesn't have flags, so they are always zero. */
 622         return do_seccomp(op, 0, uargs);
 623 }