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_int insnsi[];
  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  * Endianness is explicitly ignored and left for BPF program authors to manage
  66  * as per the specific architecture.
  67  */
  68 static void populate_seccomp_data(struct seccomp_data *sd)
  69 {
  70         struct task_struct *task = current;
  71         struct pt_regs *regs = task_pt_regs(task);
  72         unsigned long args[6];
  73
  74         sd->nr = syscall_get_nr(task, regs);
  75         sd->arch = syscall_get_arch();
  76         syscall_get_arguments(task, regs, 0, 6, args);
  77         sd->args[0] = args[0];
  78         sd->args[1] = args[1];
  79         sd->args[2] = args[2];
  80         sd->args[3] = args[3];
  81         sd->args[4] = args[4];
  82         sd->args[5] = args[5];
  83         sd->instruction_pointer = KSTK_EIP(task);
  84 }
  85
  86 /**
  87  *      seccomp_check_filter - verify seccomp filter code
  88  *      @filter: filter to verify
  89  *      @flen: length of filter
  90  *
  91  * Takes a previously checked filter (by sk_chk_filter) and
  92  * redirects all filter code that loads struct sk_buff data
  93  * and related data through seccomp_bpf_load.  It also
  94  * enforces length and alignment checking of those loads.
  95  *
  96  * Returns 0 if the rule set is legal or -EINVAL if not.
  97  */
  98 static int seccomp_check_filter(struct sock_filter *filter, unsigned int flen)
  99 {
 100         int pc;
 101         for (pc = 0; pc < flen; pc++) {
 102                 struct sock_filter *ftest = &filter[pc];
 103                 u16 code = ftest->code;
 104                 u32 k = ftest->k;
 105
 106                 switch (code) {
 107                 case BPF_S_LD_W_ABS:
 108                         ftest->code = BPF_LDX | BPF_W | BPF_ABS;
 109                         /* 32-bit aligned and not out of bounds. */
 110                         if (k >= sizeof(struct seccomp_data) || k & 3)
 111                                 return -EINVAL;
 112                         continue;
 113                 case BPF_S_LD_W_LEN:
 114                         ftest->code = BPF_LD | BPF_IMM;
 115                         ftest->k = sizeof(struct seccomp_data);
 116                         continue;
 117                 case BPF_S_LDX_W_LEN:
 118                         ftest->code = BPF_LDX | BPF_IMM;
 119                         ftest->k = sizeof(struct seccomp_data);
 120                         continue;
 121                 /* Explicitly include allowed calls. */
 122                 case BPF_S_RET_K:
 123                 case BPF_S_RET_A:
 124                 case BPF_S_ALU_ADD_K:
 125                 case BPF_S_ALU_ADD_X:
 126                 case BPF_S_ALU_SUB_K:
 127                 case BPF_S_ALU_SUB_X:
 128                 case BPF_S_ALU_MUL_K:
 129                 case BPF_S_ALU_MUL_X:
 130                 case BPF_S_ALU_DIV_X:
 131                 case BPF_S_ALU_AND_K:
 132                 case BPF_S_ALU_AND_X:
 133                 case BPF_S_ALU_OR_K:
 134                 case BPF_S_ALU_OR_X:
 135                 case BPF_S_ALU_XOR_K:
 136                 case BPF_S_ALU_XOR_X:
 137                 case BPF_S_ALU_LSH_K:
 138                 case BPF_S_ALU_LSH_X:
 139                 case BPF_S_ALU_RSH_K:
 140                 case BPF_S_ALU_RSH_X:
 141                 case BPF_S_ALU_NEG:
 142                 case BPF_S_LD_IMM:
 143                 case BPF_S_LDX_IMM:
 144                 case BPF_S_MISC_TAX:
 145                 case BPF_S_MISC_TXA:
 146                 case BPF_S_ALU_DIV_K:
 147                 case BPF_S_LD_MEM:
 148                 case BPF_S_LDX_MEM:
 149                 case BPF_S_ST:
 150                 case BPF_S_STX:
 151                 case BPF_S_JMP_JA:
 152                 case BPF_S_JMP_JEQ_K:
 153                 case BPF_S_JMP_JEQ_X:
 154                 case BPF_S_JMP_JGE_K:
 155                 case BPF_S_JMP_JGE_X:
 156                 case BPF_S_JMP_JGT_K:
 157                 case BPF_S_JMP_JGT_X:
 158                 case BPF_S_JMP_JSET_K:
 159                 case BPF_S_JMP_JSET_X:
 160                         sk_decode_filter(ftest, ftest);
 161                         continue;
 162                 default:
 163                         return -EINVAL;
 164                 }
 165         }
 166         return 0;
 167 }
 168
 169 /**
 170  * seccomp_run_filters - evaluates all seccomp filters against @syscall
 171  * @syscall: number of the current system call
 172  *
 173  * Returns valid seccomp BPF response codes.
 174  */
 175 static u32 seccomp_run_filters(int syscall)
 176 {
 177         struct seccomp_filter *f;
 178         struct seccomp_data sd;
 179         u32 ret = SECCOMP_RET_ALLOW;
 180
 181         /* Ensure unexpected behavior doesn't result in failing open. */
 182         if (WARN_ON(current->seccomp.filter == NULL))
 183                 return SECCOMP_RET_KILL;
 184
 185         populate_seccomp_data(&sd);
 186
 187         /*
 188          * All filters in the list are evaluated and the lowest BPF return
 189          * value always takes priority (ignoring the DATA).
 190          */
 191         for (f = current->seccomp.filter; f; f = f->prev) {
 192                 u32 cur_ret = sk_run_filter_int_seccomp(&sd, f->insnsi);
 193                 if ((cur_ret & SECCOMP_RET_ACTION) < (ret & SECCOMP_RET_ACTION))
 194                         ret = cur_ret;
 195         }
 196         return ret;
 197 }
 198
 199 /**
 200  * seccomp_attach_filter: Attaches a seccomp filter to current.
 201  * @fprog: BPF program to install
 202  *
 203  * Returns 0 on success or an errno on failure.
 204  */
 205 static long seccomp_attach_filter(struct sock_fprog *fprog)
 206 {
 207         struct seccomp_filter *filter;
 208         unsigned long fp_size = fprog->len * sizeof(struct sock_filter);
 209         unsigned long total_insns = fprog->len;
 210         struct sock_filter *fp;
 211         int new_len;
 212         long ret;
 213
 214         if (fprog->len == 0 || fprog->len > BPF_MAXINSNS)
 215                 return -EINVAL;
 216
 217         for (filter = current->seccomp.filter; filter; filter = filter->prev)
 218                 total_insns += filter->len + 4;  /* include a 4 instr penalty */
 219         if (total_insns > MAX_INSNS_PER_PATH)
 220                 return -ENOMEM;
 221
 222         /*
 223          * Installing a seccomp filter requires that the task have
 224          * CAP_SYS_ADMIN in its namespace or be running with no_new_privs.
 225          * This avoids scenarios where unprivileged tasks can affect the
 226          * behavior of privileged children.
 227          */
 228         if (!current->no_new_privs &&
 229             security_capable_noaudit(current_cred(), current_user_ns(),
 230                                      CAP_SYS_ADMIN) != 0)
 231                 return -EACCES;
 232
 233         fp = kzalloc(fp_size, GFP_KERNEL|__GFP_NOWARN);
 234         if (!fp)
 235                 return -ENOMEM;
 236
 237         /* Copy the instructions from fprog. */
 238         ret = -EFAULT;
 239         if (copy_from_user(fp, fprog->filter, fp_size))
 240                 goto free_prog;
 241
 242         /* Check and rewrite the fprog via the skb checker */
 243         ret = sk_chk_filter(fp, fprog->len);
 244         if (ret)
 245                 goto free_prog;
 246
 247         /* Check and rewrite the fprog for seccomp use */
 248         ret = seccomp_check_filter(fp, fprog->len);
 249         if (ret)
 250                 goto free_prog;
 251
 252         /* Convert 'sock_filter' insns to 'sock_filter_int' insns */
 253         ret = sk_convert_filter(fp, fprog->len, NULL, &new_len);
 254         if (ret)
 255                 goto free_prog;
 256
 257         /* Allocate a new seccomp_filter */
 258         filter = kzalloc(sizeof(struct seccomp_filter) +
 259                          sizeof(struct sock_filter_int) * new_len,
 260                          GFP_KERNEL|__GFP_NOWARN);
 261         if (!filter)
 262                 goto free_prog;
 263
 264         ret = sk_convert_filter(fp, fprog->len, filter->insnsi, &new_len);
 265         if (ret)
 266                 goto free_filter;
 267
 268         atomic_set(&filter->usage, 1);
 269         filter->len = new_len;
 270
 271         /*
 272          * If there is an existing filter, make it the prev and don't drop its
 273          * task reference.
 274          */
 275         filter->prev = current->seccomp.filter;
 276         current->seccomp.filter = filter;
 277         return 0;
 278
 279 free_filter:
 280         kfree(filter);
 281 free_prog:
 282         kfree(fp);
 283         return ret;
 284 }
 285
 286 /**
 287  * seccomp_attach_user_filter - attaches a user-supplied sock_fprog
 288  * @user_filter: pointer to the user data containing a sock_fprog.
 289  *
 290  * Returns 0 on success and non-zero otherwise.
 291  */
 292 static long seccomp_attach_user_filter(char __user *user_filter)
 293 {
 294         struct sock_fprog fprog;
 295         long ret = -EFAULT;
 296
 297 #ifdef CONFIG_COMPAT
 298         if (is_compat_task()) {
 299                 struct compat_sock_fprog fprog32;
 300                 if (copy_from_user(&fprog32, user_filter, sizeof(fprog32)))
 301                         goto out;
 302                 fprog.len = fprog32.len;
 303                 fprog.filter = compat_ptr(fprog32.filter);
 304         } else /* falls through to the if below. */
 305 #endif
 306         if (copy_from_user(&fprog, user_filter, sizeof(fprog)))
 307                 goto out;
 308         ret = seccomp_attach_filter(&fprog);
 309 out:
 310         return ret;
 311 }
 312
 313 /* get_seccomp_filter - increments the reference count of the filter on @tsk */
 314 void get_seccomp_filter(struct task_struct *tsk)
 315 {
 316         struct seccomp_filter *orig = tsk->seccomp.filter;
 317         if (!orig)
 318                 return;
 319         /* Reference count is bounded by the number of total processes. */
 320         atomic_inc(&orig->usage);
 321 }
 322
 323 /* put_seccomp_filter - decrements the ref count of tsk->seccomp.filter */
 324 void put_seccomp_filter(struct task_struct *tsk)
 325 {
 326         struct seccomp_filter *orig = tsk->seccomp.filter;
 327         /* Clean up single-reference branches iteratively. */
 328         while (orig && atomic_dec_and_test(&orig->usage)) {
 329                 struct seccomp_filter *freeme = orig;
 330                 orig = orig->prev;
 331                 kfree(freeme);
 332         }
 333 }
 334
 335 /**
 336  * seccomp_send_sigsys - signals the task to allow in-process syscall emulation
 337  * @syscall: syscall number to send to userland
 338  * @reason: filter-supplied reason code to send to userland (via si_errno)
 339  *
 340  * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info.
 341  */
 342 static void seccomp_send_sigsys(int syscall, int reason)
 343 {
 344         struct siginfo info;
 345         memset(&info, 0, sizeof(info));
 346         info.si_signo = SIGSYS;
 347         info.si_code = SYS_SECCOMP;
 348         info.si_call_addr = (void __user *)KSTK_EIP(current);
 349         info.si_errno = reason;
 350         info.si_arch = syscall_get_arch();
 351         info.si_syscall = syscall;
 352         force_sig_info(SIGSYS, &info, current);
 353 }
 354 #endif  /* CONFIG_SECCOMP_FILTER */
 355
 356 /*
 357  * Secure computing mode 1 allows only read/write/exit/sigreturn.
 358  * To be fully secure this must be combined with rlimit
 359  * to limit the stack allocations too.
 360  */
 361 static int mode1_syscalls[] = {
 362         __NR_seccomp_read, __NR_seccomp_write, __NR_seccomp_exit, __NR_seccomp_sigreturn,
 363         0, /* null terminated */
 364 };
 365
 366 #ifdef CONFIG_COMPAT
 367 static int mode1_syscalls_32[] = {
 368         __NR_seccomp_read_32, __NR_seccomp_write_32, __NR_seccomp_exit_32, __NR_seccomp_sigreturn_32,
 369         0, /* null terminated */
 370 };
 371 #endif
 372
 373 int __secure_computing(int this_syscall)
 374 {
 375         int mode = current->seccomp.mode;
 376         int exit_sig = 0;
 377         int *syscall;
 378         u32 ret;
 379
 380         switch (mode) {
 381         case SECCOMP_MODE_STRICT:
 382                 syscall = mode1_syscalls;
 383 #ifdef CONFIG_COMPAT
 384                 if (is_compat_task())
 385                         syscall = mode1_syscalls_32;
 386 #endif
 387                 do {
 388                         if (*syscall == this_syscall)
 389                                 return 0;
 390                 } while (*++syscall);
 391                 exit_sig = SIGKILL;
 392                 ret = SECCOMP_RET_KILL;
 393                 break;
 394 #ifdef CONFIG_SECCOMP_FILTER
 395         case SECCOMP_MODE_FILTER: {
 396                 int data;
 397                 struct pt_regs *regs = task_pt_regs(current);
 398                 ret = seccomp_run_filters(this_syscall);
 399                 data = ret & SECCOMP_RET_DATA;
 400                 ret &= SECCOMP_RET_ACTION;
 401                 switch (ret) {
 402                 case SECCOMP_RET_ERRNO:
 403                         /* Set the low-order 16-bits as a errno. */
 404                         syscall_set_return_value(current, regs,
 405                                                  -data, 0);
 406                         goto skip;
 407                 case SECCOMP_RET_TRAP:
 408                         /* Show the handler the original registers. */
 409                         syscall_rollback(current, regs);
 410                         /* Let the filter pass back 16 bits of data. */
 411                         seccomp_send_sigsys(this_syscall, data);
 412                         goto skip;
 413                 case SECCOMP_RET_TRACE:
 414                         /* Skip these calls if there is no tracer. */
 415                         if (!ptrace_event_enabled(current, PTRACE_EVENT_SECCOMP)) {
 416                                 syscall_set_return_value(current, regs,
 417                                                          -ENOSYS, 0);
 418                                 goto skip;
 419                         }
 420                         /* Allow the BPF to provide the event message */
 421                         ptrace_event(PTRACE_EVENT_SECCOMP, data);
 422                         /*
 423                          * The delivery of a fatal signal during event
 424                          * notification may silently skip tracer notification.
 425                          * Terminating the task now avoids executing a system
 426                          * call that may not be intended.
 427                          */
 428                         if (fatal_signal_pending(current))
 429                                 break;
 430                         if (syscall_get_nr(current, regs) < 0)
 431                                 goto skip;  /* Explicit request to skip. */
 432
 433                         return 0;
 434                 case SECCOMP_RET_ALLOW:
 435                         return 0;
 436                 case SECCOMP_RET_KILL:
 437                 default:
 438                         break;
 439                 }
 440                 exit_sig = SIGSYS;
 441                 break;
 442         }
 443 #endif
 444         default:
 445                 BUG();
 446         }
 447
 448 #ifdef SECCOMP_DEBUG
 449         dump_stack();
 450 #endif
 451         audit_seccomp(this_syscall, exit_sig, ret);
 452         do_exit(exit_sig);
 453 #ifdef CONFIG_SECCOMP_FILTER
 454 skip:
 455         audit_seccomp(this_syscall, exit_sig, ret);
 456 #endif
 457         return -1;
 458 }
 459
 460 long prctl_get_seccomp(void)
 461 {
 462         return current->seccomp.mode;
 463 }
 464
 465 /**
 466  * prctl_set_seccomp: configures current->seccomp.mode
 467  * @seccomp_mode: requested mode to use
 468  * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER
 469  *
 470  * This function may be called repeatedly with a @seccomp_mode of
 471  * SECCOMP_MODE_FILTER to install additional filters.  Every filter
 472  * successfully installed will be evaluated (in reverse order) for each system
 473  * call the task makes.
 474  *
 475  * Once current->seccomp.mode is non-zero, it may not be changed.
 476  *
 477  * Returns 0 on success or -EINVAL on failure.
 478  */
 479 long prctl_set_seccomp(unsigned long seccomp_mode, char __user *filter)
 480 {
 481         long ret = -EINVAL;
 482
 483         if (current->seccomp.mode &&
 484             current->seccomp.mode != seccomp_mode)
 485                 goto out;
 486
 487         switch (seccomp_mode) {
 488         case SECCOMP_MODE_STRICT:
 489                 ret = 0;
 490 #ifdef TIF_NOTSC
 491                 disable_TSC();
 492 #endif
 493                 break;
 494 #ifdef CONFIG_SECCOMP_FILTER
 495         case SECCOMP_MODE_FILTER:
 496                 ret = seccomp_attach_user_filter(filter);
 497                 if (ret)
 498                         goto out;
 499                 break;
 500 #endif
 501         default:
 502                 goto out;
 503         }
 504
 505         current->seccomp.mode = seccomp_mode;
 506         set_thread_flag(TIF_SECCOMP);
 507 out:
 508         return ret;
 509 }