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
  22 /* #define SECCOMP_DEBUG 1 */
  23
  24 #ifdef CONFIG_SECCOMP_FILTER
  25 #include <asm/syscall.h>
  26 #include <linux/filter.h>
  27 #include <linux/ptrace.h>
  28 #include <linux/security.h>
  29 #include <linux/slab.h>
  30 #include <linux/tracehook.h>
  31 #include <linux/uaccess.h>
  32
  33 /**
  34  * struct seccomp_filter - container for seccomp BPF programs
  35  *
  36  * @usage: reference count to manage the object lifetime.
  37  *         get/put helpers should be used when accessing an instance
  38  *         outside of a lifetime-guarded section.  In general, this
  39  *         is only needed for handling filters shared across tasks.
  40  * @prev: points to a previously installed, or inherited, filter
  41  * @len: the number of instructions in the program
  42  * @insns: the BPF program instructions to evaluate
  43  *
  44  * seccomp_filter objects are organized in a tree linked via the @prev
  45  * pointer.  For any task, it appears to be a singly-linked list starting
  46  * with current->seccomp.filter, the most recently attached or inherited filter.
  47  * However, multiple filters may share a @prev node, by way of fork(), which
  48  * results in a unidirectional tree existing in memory.  This is similar to
  49  * how namespaces work.
  50  *
  51  * seccomp_filter objects should never be modified after being attached
  52  * to a task_struct (other than @usage).
  53  */
  54 struct seccomp_filter {
  55         atomic_t usage;
  56         struct seccomp_filter *prev;
  57         unsigned short len;  /* Instruction count */
  58         struct sock_filter insns[];
  59 };
  60
  61 /* Limit any path through the tree to 256KB worth of instructions. */
  62 #define MAX_INSNS_PER_PATH ((1 << 18) / sizeof(struct sock_filter))
  63
  64 /**
  65  * get_u32 - returns a u32 offset into data
  66  * @data: a unsigned 64 bit value
  67  * @index: 0 or 1 to return the first or second 32-bits
  68  *
  69  * This inline exists to hide the length of unsigned long.  If a 32-bit
  70  * unsigned long is passed in, it will be extended and the top 32-bits will be
  71  * 0. If it is a 64-bit unsigned long, then whatever data is resident will be
  72  * properly returned.
  73  *
  74  * Endianness is explicitly ignored and left for BPF program authors to manage
  75  * as per the specific architecture.
  76  */
  77 static inline u32 get_u32(u64 data, int index)
  78 {
  79         return ((u32 *)&data)[index];
  80 }
  81
  82 /* Helper for bpf_load below. */
  83 #define BPF_DATA(_name) offsetof(struct seccomp_data, _name)
  84 /**
  85  * bpf_load: checks and returns a pointer to the requested offset
  86  * @off: offset into struct seccomp_data to load from
  87  *
  88  * Returns the requested 32-bits of data.
  89  * seccomp_check_filter() should assure that @off is 32-bit aligned
  90  * and not out of bounds.  Failure to do so is a BUG.
  91  */
  92 u32 seccomp_bpf_load(int off)
  93 {
  94         struct pt_regs *regs = task_pt_regs(current);
  95         if (off == BPF_DATA(nr))
  96                 return syscall_get_nr(current, regs);
  97         if (off == BPF_DATA(arch))
  98                 return syscall_get_arch(current, regs);
  99         if (off >= BPF_DATA(args[0]) && off < BPF_DATA(args[6])) {
 100                 unsigned long value;
 101                 int arg = (off - BPF_DATA(args[0])) / sizeof(u64);
 102                 int index = !!(off % sizeof(u64));
 103                 syscall_get_arguments(current, regs, arg, 1, &value);
 104                 return get_u32(value, index);
 105         }
 106         if (off == BPF_DATA(instruction_pointer))
 107                 return get_u32(KSTK_EIP(current), 0);
 108         if (off == BPF_DATA(instruction_pointer) + sizeof(u32))
 109                 return get_u32(KSTK_EIP(current), 1);
 110         /* seccomp_check_filter should make this impossible. */
 111         BUG();
 112 }
 113
 114 /**
 115  *      seccomp_check_filter - verify seccomp filter code
 116  *      @filter: filter to verify
 117  *      @flen: length of filter
 118  *
 119  * Takes a previously checked filter (by sk_chk_filter) and
 120  * redirects all filter code that loads struct sk_buff data
 121  * and related data through seccomp_bpf_load.  It also
 122  * enforces length and alignment checking of those loads.
 123  *
 124  * Returns 0 if the rule set is legal or -EINVAL if not.
 125  */
 126 static int seccomp_check_filter(struct sock_filter *filter, unsigned int flen)
 127 {
 128         int pc;
 129         for (pc = 0; pc < flen; pc++) {
 130                 struct sock_filter *ftest = &filter[pc];
 131                 u16 code = ftest->code;
 132                 u32 k = ftest->k;
 133
 134                 switch (code) {
 135                 case BPF_S_LD_W_ABS:
 136                         ftest->code = BPF_S_ANC_SECCOMP_LD_W;
 137                         /* 32-bit aligned and not out of bounds. */
 138                         if (k >= sizeof(struct seccomp_data) || k & 3)
 139                                 return -EINVAL;
 140                         continue;
 141                 case BPF_S_LD_W_LEN:
 142                         ftest->code = BPF_S_LD_IMM;
 143                         ftest->k = sizeof(struct seccomp_data);
 144                         continue;
 145                 case BPF_S_LDX_W_LEN:
 146                         ftest->code = BPF_S_LDX_IMM;
 147                         ftest->k = sizeof(struct seccomp_data);
 148                         continue;
 149                 /* Explicitly include allowed calls. */
 150                 case BPF_S_RET_K:
 151                 case BPF_S_RET_A:
 152                 case BPF_S_ALU_ADD_K:
 153                 case BPF_S_ALU_ADD_X:
 154                 case BPF_S_ALU_SUB_K:
 155                 case BPF_S_ALU_SUB_X:
 156                 case BPF_S_ALU_MUL_K:
 157                 case BPF_S_ALU_MUL_X:
 158                 case BPF_S_ALU_DIV_X:
 159                 case BPF_S_ALU_AND_K:
 160                 case BPF_S_ALU_AND_X:
 161                 case BPF_S_ALU_OR_K:
 162                 case BPF_S_ALU_OR_X:
 163                 case BPF_S_ALU_XOR_K:
 164                 case BPF_S_ALU_XOR_X:
 165                 case BPF_S_ALU_LSH_K:
 166                 case BPF_S_ALU_LSH_X:
 167                 case BPF_S_ALU_RSH_K:
 168                 case BPF_S_ALU_RSH_X:
 169                 case BPF_S_ALU_NEG:
 170                 case BPF_S_LD_IMM:
 171                 case BPF_S_LDX_IMM:
 172                 case BPF_S_MISC_TAX:
 173                 case BPF_S_MISC_TXA:
 174                 case BPF_S_ALU_DIV_K:
 175                 case BPF_S_LD_MEM:
 176                 case BPF_S_LDX_MEM:
 177                 case BPF_S_ST:
 178                 case BPF_S_STX:
 179                 case BPF_S_JMP_JA:
 180                 case BPF_S_JMP_JEQ_K:
 181                 case BPF_S_JMP_JEQ_X:
 182                 case BPF_S_JMP_JGE_K:
 183                 case BPF_S_JMP_JGE_X:
 184                 case BPF_S_JMP_JGT_K:
 185                 case BPF_S_JMP_JGT_X:
 186                 case BPF_S_JMP_JSET_K:
 187                 case BPF_S_JMP_JSET_X:
 188                         continue;
 189                 default:
 190                         return -EINVAL;
 191                 }
 192         }
 193         return 0;
 194 }
 195
 196 /**
 197  * seccomp_run_filters - evaluates all seccomp filters against @syscall
 198  * @syscall: number of the current system call
 199  *
 200  * Returns valid seccomp BPF response codes.
 201  */
 202 static u32 seccomp_run_filters(int syscall)
 203 {
 204         struct seccomp_filter *f;
 205         u32 ret = SECCOMP_RET_ALLOW;
 206
 207         /* Ensure unexpected behavior doesn't result in failing open. */
 208         if (WARN_ON(current->seccomp.filter == NULL))
 209                 return SECCOMP_RET_KILL;
 210
 211         /*
 212          * All filters in the list are evaluated and the lowest BPF return
 213          * value always takes priority (ignoring the DATA).
 214          */
 215         for (f = current->seccomp.filter; f; f = f->prev) {
 216                 u32 cur_ret = sk_run_filter(NULL, f->insns);
 217                 if ((cur_ret & SECCOMP_RET_ACTION) < (ret & SECCOMP_RET_ACTION))
 218                         ret = cur_ret;
 219         }
 220         return ret;
 221 }
 222
 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         long ret;
 235
 236         if (fprog->len == 0 || fprog->len > BPF_MAXINSNS)
 237                 return -EINVAL;
 238
 239         for (filter = current->seccomp.filter; filter; filter = filter->prev)
 240                 total_insns += filter->len + 4;  /* include a 4 instr penalty */
 241         if (total_insns > MAX_INSNS_PER_PATH)
 242                 return -ENOMEM;
 243
 244         /*
 245          * Installing a seccomp filter requires that the task have
 246          * CAP_SYS_ADMIN in its namespace or be running with no_new_privs.
 247          * This avoids scenarios where unprivileged tasks can affect the
 248          * behavior of privileged children.
 249          */
 250         if (!current->no_new_privs &&
 251             security_capable_noaudit(current_cred(), current_user_ns(),
 252                                      CAP_SYS_ADMIN) != 0)
 253                 return -EACCES;
 254
 255         /* Allocate a new seccomp_filter */
 256         filter = kzalloc(sizeof(struct seccomp_filter) + fp_size,
 257                          GFP_KERNEL|__GFP_NOWARN);
 258         if (!filter)
 259                 return -ENOMEM;
 260         atomic_set(&filter->usage, 1);
 261         filter->len = fprog->len;
 262
 263         /* Copy the instructions from fprog. */
 264         ret = -EFAULT;
 265         if (copy_from_user(filter->insns, fprog->filter, fp_size))
 266                 goto fail;
 267
 268         /* Check and rewrite the fprog via the skb checker */
 269         ret = sk_chk_filter(filter->insns, filter->len);
 270         if (ret)
 271                 goto fail;
 272
 273         /* Check and rewrite the fprog for seccomp use */
 274         ret = seccomp_check_filter(filter->insns, filter->len);
 275         if (ret)
 276                 goto fail;
 277
 278         /*
 279          * If there is an existing filter, make it the prev and don't drop its
 280          * task reference.
 281          */
 282         filter->prev = current->seccomp.filter;
 283         current->seccomp.filter = filter;
 284         return 0;
 285 fail:
 286         kfree(filter);
 287         return ret;
 288 }
 289
 290 /**
 291  * seccomp_attach_user_filter - attaches a user-supplied sock_fprog
 292  * @user_filter: pointer to the user data containing a sock_fprog.
 293  *
 294  * Returns 0 on success and non-zero otherwise.
 295  */
 296 long seccomp_attach_user_filter(char __user *user_filter)
 297 {
 298         struct sock_fprog fprog;
 299         long ret = -EFAULT;
 300
 301 #ifdef CONFIG_COMPAT
 302         if (is_compat_task()) {
 303                 struct compat_sock_fprog fprog32;
 304                 if (copy_from_user(&fprog32, user_filter, sizeof(fprog32)))
 305                         goto out;
 306                 fprog.len = fprog32.len;
 307                 fprog.filter = compat_ptr(fprog32.filter);
 308         } else /* falls through to the if below. */
 309 #endif
 310         if (copy_from_user(&fprog, user_filter, sizeof(fprog)))
 311                 goto out;
 312         ret = seccomp_attach_filter(&fprog);
 313 out:
 314         return ret;
 315 }
 316
 317 /* get_seccomp_filter - increments the reference count of the filter on @tsk */
 318 void get_seccomp_filter(struct task_struct *tsk)
 319 {
 320         struct seccomp_filter *orig = tsk->seccomp.filter;
 321         if (!orig)
 322                 return;
 323         /* Reference count is bounded by the number of total processes. */
 324         atomic_inc(&orig->usage);
 325 }
 326
 327 /* put_seccomp_filter - decrements the ref count of tsk->seccomp.filter */
 328 void put_seccomp_filter(struct task_struct *tsk)
 329 {
 330         struct seccomp_filter *orig = tsk->seccomp.filter;
 331         /* Clean up single-reference branches iteratively. */
 332         while (orig && atomic_dec_and_test(&orig->usage)) {
 333                 struct seccomp_filter *freeme = orig;
 334                 orig = orig->prev;
 335                 kfree(freeme);
 336         }
 337 }
 338
 339 /**
 340  * seccomp_send_sigsys - signals the task to allow in-process syscall emulation
 341  * @syscall: syscall number to send to userland
 342  * @reason: filter-supplied reason code to send to userland (via si_errno)
 343  *
 344  * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info.
 345  */
 346 static void seccomp_send_sigsys(int syscall, int reason)
 347 {
 348         struct siginfo info;
 349         memset(&info, 0, sizeof(info));
 350         info.si_signo = SIGSYS;
 351         info.si_code = SYS_SECCOMP;
 352         info.si_call_addr = (void __user *)KSTK_EIP(current);
 353         info.si_errno = reason;
 354         info.si_arch = syscall_get_arch(current, task_pt_regs(current));
 355         info.si_syscall = syscall;
 356         force_sig_info(SIGSYS, &info, current);
 357 }
 358 #endif  /* CONFIG_SECCOMP_FILTER */
 359
 360 /*
 361  * Secure computing mode 1 allows only read/write/exit/sigreturn.
 362  * To be fully secure this must be combined with rlimit
 363  * to limit the stack allocations too.
 364  */
 365 static int mode1_syscalls[] = {
 366         __NR_seccomp_read, __NR_seccomp_write, __NR_seccomp_exit, __NR_seccomp_sigreturn,
 367         0, /* null terminated */
 368 };
 369
 370 #ifdef CONFIG_COMPAT
 371 static int mode1_syscalls_32[] = {
 372         __NR_seccomp_read_32, __NR_seccomp_write_32, __NR_seccomp_exit_32, __NR_seccomp_sigreturn_32,
 373         0, /* null terminated */
 374 };
 375 #endif
 376
 377 int __secure_computing(int this_syscall)
 378 {
 379         int mode = current->seccomp.mode;
 380         int exit_sig = 0;
 381         int *syscall;
 382         u32 ret;
 383
 384         switch (mode) {
 385         case SECCOMP_MODE_STRICT:
 386                 syscall = mode1_syscalls;
 387 #ifdef CONFIG_COMPAT
 388                 if (is_compat_task())
 389                         syscall = mode1_syscalls_32;
 390 #endif
 391                 do {
 392                         if (*syscall == this_syscall)
 393                                 return 0;
 394                 } while (*++syscall);
 395                 exit_sig = SIGKILL;
 396                 ret = SECCOMP_RET_KILL;
 397                 break;
 398 #ifdef CONFIG_SECCOMP_FILTER
 399         case SECCOMP_MODE_FILTER: {
 400                 int data;
 401                 struct pt_regs *regs = task_pt_regs(current);
 402                 ret = seccomp_run_filters(this_syscall);
 403                 data = ret & SECCOMP_RET_DATA;
 404                 ret &= SECCOMP_RET_ACTION;
 405                 switch (ret) {
 406                 case SECCOMP_RET_ERRNO:
 407                         /* Set the low-order 16-bits as a errno. */
 408                         syscall_set_return_value(current, regs,
 409                                                  -data, 0);
 410                         goto skip;
 411                 case SECCOMP_RET_TRAP:
 412                         /* Show the handler the original registers. */
 413                         syscall_rollback(current, regs);
 414                         /* Let the filter pass back 16 bits of data. */
 415                         seccomp_send_sigsys(this_syscall, data);
 416                         goto skip;
 417                 case SECCOMP_RET_TRACE:
 418                         /* Skip these calls if there is no tracer. */
 419                         if (!ptrace_event_enabled(current, PTRACE_EVENT_SECCOMP)) {
 420                                 syscall_set_return_value(current, regs,
 421                                                          -ENOSYS, 0);
 422                                 goto skip;
 423                         }
 424                         /* Allow the BPF to provide the event message */
 425                         ptrace_event(PTRACE_EVENT_SECCOMP, data);
 426                         /*
 427                          * The delivery of a fatal signal during event
 428                          * notification may silently skip tracer notification.
 429                          * Terminating the task now avoids executing a system
 430                          * call that may not be intended.
 431                          */
 432                         if (fatal_signal_pending(current))
 433                                 break;
 434                         if (syscall_get_nr(current, regs) < 0)
 435                                 goto skip;  /* Explicit request to skip. */
 436
 437                         return 0;
 438                 case SECCOMP_RET_ALLOW:
 439                         return 0;
 440                 case SECCOMP_RET_KILL:
 441                 default:
 442                         break;
 443                 }
 444                 exit_sig = SIGSYS;
 445                 break;
 446         }
 447 #endif
 448         default:
 449                 BUG();
 450         }
 451
 452 #ifdef SECCOMP_DEBUG
 453         dump_stack();
 454 #endif
 455         audit_seccomp(this_syscall, exit_sig, ret);
 456         do_exit(exit_sig);
 457 #ifdef CONFIG_SECCOMP_FILTER
 458 skip:
 459         audit_seccomp(this_syscall, exit_sig, ret);
 460 #endif
 461         return -1;
 462 }
 463
 464 long prctl_get_seccomp(void)
 465 {
 466         return current->seccomp.mode;
 467 }
 468
 469 /**
 470  * prctl_set_seccomp: configures current->seccomp.mode
 471  * @seccomp_mode: requested mode to use
 472  * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER
 473  *
 474  * This function may be called repeatedly with a @seccomp_mode of
 475  * SECCOMP_MODE_FILTER to install additional filters.  Every filter
 476  * successfully installed will be evaluated (in reverse order) for each system
 477  * call the task makes.
 478  *
 479  * Once current->seccomp.mode is non-zero, it may not be changed.
 480  *
 481  * Returns 0 on success or -EINVAL on failure.
 482  */
 483 long prctl_set_seccomp(unsigned long seccomp_mode, char __user *filter)
 484 {
 485         long ret = -EINVAL;
 486
 487         if (current->seccomp.mode &&
 488             current->seccomp.mode != seccomp_mode)
 489                 goto out;
 490
 491         switch (seccomp_mode) {
 492         case SECCOMP_MODE_STRICT:
 493                 ret = 0;
 494 #ifdef TIF_NOTSC
 495                 disable_TSC();
 496 #endif
 497                 break;
 498 #ifdef CONFIG_SECCOMP_FILTER
 499         case SECCOMP_MODE_FILTER:
 500                 ret = seccomp_attach_user_filter(filter);
 501                 if (ret)
 502                         goto out;
 503                 break;
 504 #endif
 505         default:
 506                 goto out;
 507         }
 508
 509         current->seccomp.mode = seccomp_mode;
 510         set_thread_flag(TIF_SECCOMP);
 511 out:
 512         return ret;
 513 }