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_LSH_K:
 164                 case BPF_S_ALU_LSH_X:
 165                 case BPF_S_ALU_RSH_K:
 166                 case BPF_S_ALU_RSH_X:
 167                 case BPF_S_ALU_NEG:
 168                 case BPF_S_LD_IMM:
 169                 case BPF_S_LDX_IMM:
 170                 case BPF_S_MISC_TAX:
 171                 case BPF_S_MISC_TXA:
 172                 case BPF_S_ALU_DIV_K:
 173                 case BPF_S_LD_MEM:
 174                 case BPF_S_LDX_MEM:
 175                 case BPF_S_ST:
 176                 case BPF_S_STX:
 177                 case BPF_S_JMP_JA:
 178                 case BPF_S_JMP_JEQ_K:
 179                 case BPF_S_JMP_JEQ_X:
 180                 case BPF_S_JMP_JGE_K:
 181                 case BPF_S_JMP_JGE_X:
 182                 case BPF_S_JMP_JGT_K:
 183                 case BPF_S_JMP_JGT_X:
 184                 case BPF_S_JMP_JSET_K:
 185                 case BPF_S_JMP_JSET_X:
 186                         continue;
 187                 default:
 188                         return -EINVAL;
 189                 }
 190         }
 191         return 0;
 192 }
 193
 194 /**
 195  * seccomp_run_filters - evaluates all seccomp filters against @syscall
 196  * @syscall: number of the current system call
 197  *
 198  * Returns valid seccomp BPF response codes.
 199  */
 200 static u32 seccomp_run_filters(int syscall)
 201 {
 202         struct seccomp_filter *f;
 203         u32 ret = SECCOMP_RET_ALLOW;
 204
 205         /* Ensure unexpected behavior doesn't result in failing open. */
 206         if (WARN_ON(current->seccomp.filter == NULL))
 207                 return SECCOMP_RET_KILL;
 208
 209         /*
 210          * All filters in the list are evaluated and the lowest BPF return
 211          * value always takes priority (ignoring the DATA).
 212          */
 213         for (f = current->seccomp.filter; f; f = f->prev) {
 214                 u32 cur_ret = sk_run_filter(NULL, f->insns);
 215                 if ((cur_ret & SECCOMP_RET_ACTION) < (ret & SECCOMP_RET_ACTION))
 216                         ret = cur_ret;
 217         }
 218         return ret;
 219 }
 220
 221 /**
 222  * seccomp_attach_filter: Attaches a seccomp filter to current.
 223  * @fprog: BPF program to install
 224  *
 225  * Returns 0 on success or an errno on failure.
 226  */
 227 static long seccomp_attach_filter(struct sock_fprog *fprog)
 228 {
 229         struct seccomp_filter *filter;
 230         unsigned long fp_size = fprog->len * sizeof(struct sock_filter);
 231         unsigned long total_insns = fprog->len;
 232         long ret;
 233
 234         if (fprog->len == 0 || fprog->len > BPF_MAXINSNS)
 235                 return -EINVAL;
 236
 237         for (filter = current->seccomp.filter; filter; filter = filter->prev)
 238                 total_insns += filter->len + 4;  /* include a 4 instr penalty */
 239         if (total_insns > MAX_INSNS_PER_PATH)
 240                 return -ENOMEM;
 241
 242         /*
 243          * Installing a seccomp filter requires that the task have
 244          * CAP_SYS_ADMIN in its namespace or be running with no_new_privs.
 245          * This avoids scenarios where unprivileged tasks can affect the
 246          * behavior of privileged children.
 247          */
 248         if (!current->no_new_privs &&
 249             security_capable_noaudit(current_cred(), current_user_ns(),
 250                                      CAP_SYS_ADMIN) != 0)
 251                 return -EACCES;
 252
 253         /* Allocate a new seccomp_filter */
 254         filter = kzalloc(sizeof(struct seccomp_filter) + fp_size,
 255                          GFP_KERNEL|__GFP_NOWARN);
 256         if (!filter)
 257                 return -ENOMEM;
 258         atomic_set(&filter->usage, 1);
 259         filter->len = fprog->len;
 260
 261         /* Copy the instructions from fprog. */
 262         ret = -EFAULT;
 263         if (copy_from_user(filter->insns, fprog->filter, fp_size))
 264                 goto fail;
 265
 266         /* Check and rewrite the fprog via the skb checker */
 267         ret = sk_chk_filter(filter->insns, filter->len);
 268         if (ret)
 269                 goto fail;
 270
 271         /* Check and rewrite the fprog for seccomp use */
 272         ret = seccomp_check_filter(filter->insns, filter->len);
 273         if (ret)
 274                 goto fail;
 275
 276         /*
 277          * If there is an existing filter, make it the prev and don't drop its
 278          * task reference.
 279          */
 280         filter->prev = current->seccomp.filter;
 281         current->seccomp.filter = filter;
 282         return 0;
 283 fail:
 284         kfree(filter);
 285         return ret;
 286 }
 287
 288 /**
 289  * seccomp_attach_user_filter - attaches a user-supplied sock_fprog
 290  * @user_filter: pointer to the user data containing a sock_fprog.
 291  *
 292  * Returns 0 on success and non-zero otherwise.
 293  */
 294 long seccomp_attach_user_filter(char __user *user_filter)
 295 {
 296         struct sock_fprog fprog;
 297         long ret = -EFAULT;
 298
 299 #ifdef CONFIG_COMPAT
 300         if (is_compat_task()) {
 301                 struct compat_sock_fprog fprog32;
 302                 if (copy_from_user(&fprog32, user_filter, sizeof(fprog32)))
 303                         goto out;
 304                 fprog.len = fprog32.len;
 305                 fprog.filter = compat_ptr(fprog32.filter);
 306         } else /* falls through to the if below. */
 307 #endif
 308         if (copy_from_user(&fprog, user_filter, sizeof(fprog)))
 309                 goto out;
 310         ret = seccomp_attach_filter(&fprog);
 311 out:
 312         return ret;
 313 }
 314
 315 /* get_seccomp_filter - increments the reference count of the filter on @tsk */
 316 void get_seccomp_filter(struct task_struct *tsk)
 317 {
 318         struct seccomp_filter *orig = tsk->seccomp.filter;
 319         if (!orig)
 320                 return;
 321         /* Reference count is bounded by the number of total processes. */
 322         atomic_inc(&orig->usage);
 323 }
 324
 325 /* put_seccomp_filter - decrements the ref count of tsk->seccomp.filter */
 326 void put_seccomp_filter(struct task_struct *tsk)
 327 {
 328         struct seccomp_filter *orig = tsk->seccomp.filter;
 329         /* Clean up single-reference branches iteratively. */
 330         while (orig && atomic_dec_and_test(&orig->usage)) {
 331                 struct seccomp_filter *freeme = orig;
 332                 orig = orig->prev;
 333                 kfree(freeme);
 334         }
 335 }
 336
 337 /**
 338  * seccomp_send_sigsys - signals the task to allow in-process syscall emulation
 339  * @syscall: syscall number to send to userland
 340  * @reason: filter-supplied reason code to send to userland (via si_errno)
 341  *
 342  * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info.
 343  */
 344 static void seccomp_send_sigsys(int syscall, int reason)
 345 {
 346         struct siginfo info;
 347         memset(&info, 0, sizeof(info));
 348         info.si_signo = SIGSYS;
 349         info.si_code = SYS_SECCOMP;
 350         info.si_call_addr = (void __user *)KSTK_EIP(current);
 351         info.si_errno = reason;
 352         info.si_arch = syscall_get_arch(current, task_pt_regs(current));
 353         info.si_syscall = syscall;
 354         force_sig_info(SIGSYS, &info, current);
 355 }
 356 #endif  /* CONFIG_SECCOMP_FILTER */
 357
 358 /*
 359  * Secure computing mode 1 allows only read/write/exit/sigreturn.
 360  * To be fully secure this must be combined with rlimit
 361  * to limit the stack allocations too.
 362  */
 363 static int mode1_syscalls[] = {
 364         __NR_seccomp_read, __NR_seccomp_write, __NR_seccomp_exit, __NR_seccomp_sigreturn,
 365         0, /* null terminated */
 366 };
 367
 368 #ifdef CONFIG_COMPAT
 369 static int mode1_syscalls_32[] = {
 370         __NR_seccomp_read_32, __NR_seccomp_write_32, __NR_seccomp_exit_32, __NR_seccomp_sigreturn_32,
 371         0, /* null terminated */
 372 };
 373 #endif
 374
 375 int __secure_computing(int this_syscall)
 376 {
 377         int mode = current->seccomp.mode;
 378         int exit_sig = 0;
 379         int *syscall;
 380         u32 ret;
 381
 382         switch (mode) {
 383         case SECCOMP_MODE_STRICT:
 384                 syscall = mode1_syscalls;
 385 #ifdef CONFIG_COMPAT
 386                 if (is_compat_task())
 387                         syscall = mode1_syscalls_32;
 388 #endif
 389                 do {
 390                         if (*syscall == this_syscall)
 391                                 return 0;
 392                 } while (*++syscall);
 393                 exit_sig = SIGKILL;
 394                 ret = SECCOMP_RET_KILL;
 395                 break;
 396 #ifdef CONFIG_SECCOMP_FILTER
 397         case SECCOMP_MODE_FILTER: {
 398                 int data;
 399                 struct pt_regs *regs = task_pt_regs(current);
 400                 ret = seccomp_run_filters(this_syscall);
 401                 data = ret & SECCOMP_RET_DATA;
 402                 ret &= SECCOMP_RET_ACTION;
 403                 switch (ret) {
 404                 case SECCOMP_RET_ERRNO:
 405                         /* Set the low-order 16-bits as a errno. */
 406                         syscall_set_return_value(current, regs,
 407                                                  -data, 0);
 408                         goto skip;
 409                 case SECCOMP_RET_TRAP:
 410                         /* Show the handler the original registers. */
 411                         syscall_rollback(current, regs);
 412                         /* Let the filter pass back 16 bits of data. */
 413                         seccomp_send_sigsys(this_syscall, data);
 414                         goto skip;
 415                 case SECCOMP_RET_TRACE:
 416                         /* Skip these calls if there is no tracer. */
 417                         if (!ptrace_event_enabled(current, PTRACE_EVENT_SECCOMP)) {
 418                                 syscall_set_return_value(current, regs,
 419                                                          -ENOSYS, 0);
 420                                 goto skip;
 421                         }
 422                         /* Allow the BPF to provide the event message */
 423                         ptrace_event(PTRACE_EVENT_SECCOMP, data);
 424                         /*
 425                          * The delivery of a fatal signal during event
 426                          * notification may silently skip tracer notification.
 427                          * Terminating the task now avoids executing a system
 428                          * call that may not be intended.
 429                          */
 430                         if (fatal_signal_pending(current))
 431                                 break;
 432                         if (syscall_get_nr(current, regs) < 0)
 433                                 goto skip;  /* Explicit request to skip. */
 434
 435                         return 0;
 436                 case SECCOMP_RET_ALLOW:
 437                         return 0;
 438                 case SECCOMP_RET_KILL:
 439                 default:
 440                         break;
 441                 }
 442                 exit_sig = SIGSYS;
 443                 break;
 444         }
 445 #endif
 446         default:
 447                 BUG();
 448         }
 449
 450 #ifdef SECCOMP_DEBUG
 451         dump_stack();
 452 #endif
 453         audit_seccomp(this_syscall, exit_sig, ret);
 454         do_exit(exit_sig);
 455 #ifdef CONFIG_SECCOMP_FILTER
 456 skip:
 457         audit_seccomp(this_syscall, exit_sig, ret);
 458 #endif
 459         return -1;
 460 }
 461
 462 long prctl_get_seccomp(void)
 463 {
 464         return current->seccomp.mode;
 465 }
 466
 467 /**
 468  * prctl_set_seccomp: configures current->seccomp.mode
 469  * @seccomp_mode: requested mode to use
 470  * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER
 471  *
 472  * This function may be called repeatedly with a @seccomp_mode of
 473  * SECCOMP_MODE_FILTER to install additional filters.  Every filter
 474  * successfully installed will be evaluated (in reverse order) for each system
 475  * call the task makes.
 476  *
 477  * Once current->seccomp.mode is non-zero, it may not be changed.
 478  *
 479  * Returns 0 on success or -EINVAL on failure.
 480  */
 481 long prctl_set_seccomp(unsigned long seccomp_mode, char __user *filter)
 482 {
 483         long ret = -EINVAL;
 484
 485         if (current->seccomp.mode &&
 486             current->seccomp.mode != seccomp_mode)
 487                 goto out;
 488
 489         switch (seccomp_mode) {
 490         case SECCOMP_MODE_STRICT:
 491                 ret = 0;
 492 #ifdef TIF_NOTSC
 493                 disable_TSC();
 494 #endif
 495                 break;
 496 #ifdef CONFIG_SECCOMP_FILTER
 497         case SECCOMP_MODE_FILTER:
 498                 ret = seccomp_attach_user_filter(filter);
 499                 if (ret)
 500                         goto out;
 501                 break;
 502 #endif
 503         default:
 504                 goto out;
 505         }
 506
 507         current->seccomp.mode = seccomp_mode;
 508         set_thread_flag(TIF_SECCOMP);
 509 out:
 510         return ret;
 511 }