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
  73         sd->nr = syscall_get_nr(task, regs);
  74         sd->arch = syscall_get_arch();
  75
  76         /* Unroll syscall_get_args to help gcc on arm. */
  77         syscall_get_arguments(task, regs, 0, 1, (unsigned long *) &sd->args[0]);
  78         syscall_get_arguments(task, regs, 1, 1, (unsigned long *) &sd->args[1]);
  79         syscall_get_arguments(task, regs, 2, 1, (unsigned long *) &sd->args[2]);
  80         syscall_get_arguments(task, regs, 3, 1, (unsigned long *) &sd->args[3]);
  81         syscall_get_arguments(task, regs, 4, 1, (unsigned long *) &sd->args[4]);
  82         syscall_get_arguments(task, regs, 5, 1, (unsigned long *) &sd->args[5]);
  83
  84         sd->instruction_pointer = KSTK_EIP(task);
  85 }
  86
  87 /**
  88  *      seccomp_check_filter - verify seccomp filter code
  89  *      @filter: filter to verify
  90  *      @flen: length of filter
  91  *
  92  * Takes a previously checked filter (by sk_chk_filter) and
  93  * redirects all filter code that loads struct sk_buff data
  94  * and related data through seccomp_bpf_load.  It also
  95  * enforces length and alignment checking of those loads.
  96  *
  97  * Returns 0 if the rule set is legal or -EINVAL if not.
  98  */
  99 static int seccomp_check_filter(struct sock_filter *filter, unsigned int flen)
 100 {
 101         int pc;
 102         for (pc = 0; pc < flen; pc++) {
 103                 struct sock_filter *ftest = &filter[pc];
 104                 u16 code = ftest->code;
 105                 u32 k = ftest->k;
 106
 107                 switch (code) {
 108                 case BPF_S_LD_W_ABS:
 109                         ftest->code = BPF_LDX | BPF_W | BPF_ABS;
 110                         /* 32-bit aligned and not out of bounds. */
 111                         if (k >= sizeof(struct seccomp_data) || k & 3)
 112                                 return -EINVAL;
 113                         continue;
 114                 case BPF_S_LD_W_LEN:
 115                         ftest->code = BPF_LD | BPF_IMM;
 116                         ftest->k = sizeof(struct seccomp_data);
 117                         continue;
 118                 case BPF_S_LDX_W_LEN:
 119                         ftest->code = BPF_LDX | BPF_IMM;
 120                         ftest->k = sizeof(struct seccomp_data);
 121                         continue;
 122                 /* Explicitly include allowed calls. */
 123                 case BPF_S_RET_K:
 124                 case BPF_S_RET_A:
 125                 case BPF_S_ALU_ADD_K:
 126                 case BPF_S_ALU_ADD_X:
 127                 case BPF_S_ALU_SUB_K:
 128                 case BPF_S_ALU_SUB_X:
 129                 case BPF_S_ALU_MUL_K:
 130                 case BPF_S_ALU_MUL_X:
 131                 case BPF_S_ALU_DIV_X:
 132                 case BPF_S_ALU_AND_K:
 133                 case BPF_S_ALU_AND_X:
 134                 case BPF_S_ALU_OR_K:
 135                 case BPF_S_ALU_OR_X:
 136                 case BPF_S_ALU_XOR_K:
 137                 case BPF_S_ALU_XOR_X:
 138                 case BPF_S_ALU_LSH_K:
 139                 case BPF_S_ALU_LSH_X:
 140                 case BPF_S_ALU_RSH_K:
 141                 case BPF_S_ALU_RSH_X:
 142                 case BPF_S_ALU_NEG:
 143                 case BPF_S_LD_IMM:
 144                 case BPF_S_LDX_IMM:
 145                 case BPF_S_MISC_TAX:
 146                 case BPF_S_MISC_TXA:
 147                 case BPF_S_ALU_DIV_K:
 148                 case BPF_S_LD_MEM:
 149                 case BPF_S_LDX_MEM:
 150                 case BPF_S_ST:
 151                 case BPF_S_STX:
 152                 case BPF_S_JMP_JA:
 153                 case BPF_S_JMP_JEQ_K:
 154                 case BPF_S_JMP_JEQ_X:
 155                 case BPF_S_JMP_JGE_K:
 156                 case BPF_S_JMP_JGE_X:
 157                 case BPF_S_JMP_JGT_K:
 158                 case BPF_S_JMP_JGT_X:
 159                 case BPF_S_JMP_JSET_K:
 160                 case BPF_S_JMP_JSET_X:
 161                         sk_decode_filter(ftest, ftest);
 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_int_seccomp(&sd, f->insnsi);
 194                 if ((cur_ret & SECCOMP_RET_ACTION) < (ret & SECCOMP_RET_ACTION))
 195                         ret = cur_ret;
 196         }
 197         return ret;
 198 }
 199
 200 /**
 201  * seccomp_attach_filter: Attaches a seccomp filter to current.
 202  * @fprog: BPF program to install
 203  *
 204  * Returns 0 on success or an errno on failure.
 205  */
 206 static long seccomp_attach_filter(struct sock_fprog *fprog)
 207 {
 208         struct seccomp_filter *filter;
 209         unsigned long fp_size = fprog->len * sizeof(struct sock_filter);
 210         unsigned long total_insns = fprog->len;
 211         struct sock_filter *fp;
 212         int new_len;
 213         long ret;
 214
 215         if (fprog->len == 0 || fprog->len > BPF_MAXINSNS)
 216                 return -EINVAL;
 217
 218         for (filter = current->seccomp.filter; filter; filter = filter->prev)
 219                 total_insns += filter->len + 4;  /* include a 4 instr penalty */
 220         if (total_insns > MAX_INSNS_PER_PATH)
 221                 return -ENOMEM;
 222
 223         /*
 224          * Installing a seccomp filter requires that the task have
 225          * CAP_SYS_ADMIN in its namespace or be running with no_new_privs.
 226          * This avoids scenarios where unprivileged tasks can affect the
 227          * behavior of privileged children.
 228          */
 229         if (!current->no_new_privs &&
 230             security_capable_noaudit(current_cred(), current_user_ns(),
 231                                      CAP_SYS_ADMIN) != 0)
 232                 return -EACCES;
 233
 234         fp = kzalloc(fp_size, GFP_KERNEL|__GFP_NOWARN);
 235         if (!fp)
 236                 return -ENOMEM;
 237
 238         /* Copy the instructions from fprog. */
 239         ret = -EFAULT;
 240         if (copy_from_user(fp, fprog->filter, fp_size))
 241                 goto free_prog;
 242
 243         /* Check and rewrite the fprog via the skb checker */
 244         ret = sk_chk_filter(fp, fprog->len);
 245         if (ret)
 246                 goto free_prog;
 247
 248         /* Check and rewrite the fprog for seccomp use */
 249         ret = seccomp_check_filter(fp, fprog->len);
 250         if (ret)
 251                 goto free_prog;
 252
 253         /* Convert 'sock_filter' insns to 'sock_filter_int' insns */
 254         ret = sk_convert_filter(fp, fprog->len, NULL, &new_len);
 255         if (ret)
 256                 goto free_prog;
 257
 258         /* Allocate a new seccomp_filter */
 259         filter = kzalloc(sizeof(struct seccomp_filter) +
 260                          sizeof(struct sock_filter_int) * new_len,
 261                          GFP_KERNEL|__GFP_NOWARN);
 262         if (!filter)
 263                 goto free_prog;
 264
 265         ret = sk_convert_filter(fp, fprog->len, filter->insnsi, &new_len);
 266         if (ret)
 267                 goto free_filter;
 268
 269         atomic_set(&filter->usage, 1);
 270         filter->len = new_len;
 271
 272         /*
 273          * If there is an existing filter, make it the prev and don't drop its
 274          * task reference.
 275          */
 276         filter->prev = current->seccomp.filter;
 277         current->seccomp.filter = filter;
 278         return 0;
 279
 280 free_filter:
 281         kfree(filter);
 282 free_prog:
 283         kfree(fp);
 284         return ret;
 285 }
 286
 287 /**
 288  * seccomp_attach_user_filter - attaches a user-supplied sock_fprog
 289  * @user_filter: pointer to the user data containing a sock_fprog.
 290  *
 291  * Returns 0 on success and non-zero otherwise.
 292  */
 293 static long seccomp_attach_user_filter(char __user *user_filter)
 294 {
 295         struct sock_fprog fprog;
 296         long ret = -EFAULT;
 297
 298 #ifdef CONFIG_COMPAT
 299         if (is_compat_task()) {
 300                 struct compat_sock_fprog fprog32;
 301                 if (copy_from_user(&fprog32, user_filter, sizeof(fprog32)))
 302                         goto out;
 303                 fprog.len = fprog32.len;
 304                 fprog.filter = compat_ptr(fprog32.filter);
 305         } else /* falls through to the if below. */
 306 #endif
 307         if (copy_from_user(&fprog, user_filter, sizeof(fprog)))
 308                 goto out;
 309         ret = seccomp_attach_filter(&fprog);
 310 out:
 311         return ret;
 312 }
 313
 314 /* get_seccomp_filter - increments the reference count of the filter on @tsk */
 315 void get_seccomp_filter(struct task_struct *tsk)
 316 {
 317         struct seccomp_filter *orig = tsk->seccomp.filter;
 318         if (!orig)
 319                 return;
 320         /* Reference count is bounded by the number of total processes. */
 321         atomic_inc(&orig->usage);
 322 }
 323
 324 /* put_seccomp_filter - decrements the ref count of tsk->seccomp.filter */
 325 void put_seccomp_filter(struct task_struct *tsk)
 326 {
 327         struct seccomp_filter *orig = tsk->seccomp.filter;
 328         /* Clean up single-reference branches iteratively. */
 329         while (orig && atomic_dec_and_test(&orig->usage)) {
 330                 struct seccomp_filter *freeme = orig;
 331                 orig = orig->prev;
 332                 kfree(freeme);
 333         }
 334 }
 335
 336 /**
 337  * seccomp_send_sigsys - signals the task to allow in-process syscall emulation
 338  * @syscall: syscall number to send to userland
 339  * @reason: filter-supplied reason code to send to userland (via si_errno)
 340  *
 341  * Forces a SIGSYS with a code of SYS_SECCOMP and related sigsys info.
 342  */
 343 static void seccomp_send_sigsys(int syscall, int reason)
 344 {
 345         struct siginfo info;
 346         memset(&info, 0, sizeof(info));
 347         info.si_signo = SIGSYS;
 348         info.si_code = SYS_SECCOMP;
 349         info.si_call_addr = (void __user *)KSTK_EIP(current);
 350         info.si_errno = reason;
 351         info.si_arch = syscall_get_arch();
 352         info.si_syscall = syscall;
 353         force_sig_info(SIGSYS, &info, current);
 354 }
 355 #endif  /* CONFIG_SECCOMP_FILTER */
 356
 357 /*
 358  * Secure computing mode 1 allows only read/write/exit/sigreturn.
 359  * To be fully secure this must be combined with rlimit
 360  * to limit the stack allocations too.
 361  */
 362 static int mode1_syscalls[] = {
 363         __NR_seccomp_read, __NR_seccomp_write, __NR_seccomp_exit, __NR_seccomp_sigreturn,
 364         0, /* null terminated */
 365 };
 366
 367 #ifdef CONFIG_COMPAT
 368 static int mode1_syscalls_32[] = {
 369         __NR_seccomp_read_32, __NR_seccomp_write_32, __NR_seccomp_exit_32, __NR_seccomp_sigreturn_32,
 370         0, /* null terminated */
 371 };
 372 #endif
 373
 374 int __secure_computing(int this_syscall)
 375 {
 376         int mode = current->seccomp.mode;
 377         int exit_sig = 0;
 378         int *syscall;
 379         u32 ret;
 380
 381         switch (mode) {
 382         case SECCOMP_MODE_STRICT:
 383                 syscall = mode1_syscalls;
 384 #ifdef CONFIG_COMPAT
 385                 if (is_compat_task())
 386                         syscall = mode1_syscalls_32;
 387 #endif
 388                 do {
 389                         if (*syscall == this_syscall)
 390                                 return 0;
 391                 } while (*++syscall);
 392                 exit_sig = SIGKILL;
 393                 ret = SECCOMP_RET_KILL;
 394                 break;
 395 #ifdef CONFIG_SECCOMP_FILTER
 396         case SECCOMP_MODE_FILTER: {
 397                 int data;
 398                 struct pt_regs *regs = task_pt_regs(current);
 399                 ret = seccomp_run_filters(this_syscall);
 400                 data = ret & SECCOMP_RET_DATA;
 401                 ret &= SECCOMP_RET_ACTION;
 402                 switch (ret) {
 403                 case SECCOMP_RET_ERRNO:
 404                         /* Set the low-order 16-bits as a errno. */
 405                         syscall_set_return_value(current, regs,
 406                                                  -data, 0);
 407                         goto skip;
 408                 case SECCOMP_RET_TRAP:
 409                         /* Show the handler the original registers. */
 410                         syscall_rollback(current, regs);
 411                         /* Let the filter pass back 16 bits of data. */
 412                         seccomp_send_sigsys(this_syscall, data);
 413                         goto skip;
 414                 case SECCOMP_RET_TRACE:
 415                         /* Skip these calls if there is no tracer. */
 416                         if (!ptrace_event_enabled(current, PTRACE_EVENT_SECCOMP)) {
 417                                 syscall_set_return_value(current, regs,
 418                                                          -ENOSYS, 0);
 419                                 goto skip;
 420                         }
 421                         /* Allow the BPF to provide the event message */
 422                         ptrace_event(PTRACE_EVENT_SECCOMP, data);
 423                         /*
 424                          * The delivery of a fatal signal during event
 425                          * notification may silently skip tracer notification.
 426                          * Terminating the task now avoids executing a system
 427                          * call that may not be intended.
 428                          */
 429                         if (fatal_signal_pending(current))
 430                                 break;
 431                         if (syscall_get_nr(current, regs) < 0)
 432                                 goto skip;  /* Explicit request to skip. */
 433
 434                         return 0;
 435                 case SECCOMP_RET_ALLOW:
 436                         return 0;
 437                 case SECCOMP_RET_KILL:
 438                 default:
 439                         break;
 440                 }
 441                 exit_sig = SIGSYS;
 442                 break;
 443         }
 444 #endif
 445         default:
 446                 BUG();
 447         }
 448
 449 #ifdef SECCOMP_DEBUG
 450         dump_stack();
 451 #endif
 452         audit_seccomp(this_syscall, exit_sig, ret);
 453         do_exit(exit_sig);
 454 #ifdef CONFIG_SECCOMP_FILTER
 455 skip:
 456         audit_seccomp(this_syscall, exit_sig, ret);
 457 #endif
 458         return -1;
 459 }
 460
 461 long prctl_get_seccomp(void)
 462 {
 463         return current->seccomp.mode;
 464 }
 465
 466 /**
 467  * prctl_set_seccomp: configures current->seccomp.mode
 468  * @seccomp_mode: requested mode to use
 469  * @filter: optional struct sock_fprog for use with SECCOMP_MODE_FILTER
 470  *
 471  * This function may be called repeatedly with a @seccomp_mode of
 472  * SECCOMP_MODE_FILTER to install additional filters.  Every filter
 473  * successfully installed will be evaluated (in reverse order) for each system
 474  * call the task makes.
 475  *
 476  * Once current->seccomp.mode is non-zero, it may not be changed.
 477  *
 478  * Returns 0 on success or -EINVAL on failure.
 479  */
 480 long prctl_set_seccomp(unsigned long seccomp_mode, char __user *filter)
 481 {
 482         long ret = -EINVAL;
 483
 484         if (current->seccomp.mode &&
 485             current->seccomp.mode != seccomp_mode)
 486                 goto out;
 487
 488         switch (seccomp_mode) {
 489         case SECCOMP_MODE_STRICT:
 490                 ret = 0;
 491 #ifdef TIF_NOTSC
 492                 disable_TSC();
 493 #endif
 494                 break;
 495 #ifdef CONFIG_SECCOMP_FILTER
 496         case SECCOMP_MODE_FILTER:
 497                 ret = seccomp_attach_user_filter(filter);
 498                 if (ret)
 499                         goto out;
 500                 break;
 501 #endif
 502         default:
 503                 goto out;
 504         }
 505
 506         current->seccomp.mode = seccomp_mode;
 507         set_thread_flag(TIF_SECCOMP);
 508 out:
 509         return ret;
 510 }