Elly Jones | cd7a904 | 2011-07-22 13:56:51 -0400 | [diff] [blame] | 1 | /* Copyright (c) 2011 The Chromium OS Authors. All rights reserved. |
| 2 | * Use of this source code is governed by a BSD-style license that can be |
Will Drewry | 32ac9f5 | 2011-08-18 21:36:27 -0500 | [diff] [blame] | 3 | * found in the LICENSE file. |
| 4 | */ |
Elly Jones | cd7a904 | 2011-07-22 13:56:51 -0400 | [diff] [blame] | 5 | |
| 6 | #define _BSD_SOURCE |
| 7 | #define _GNU_SOURCE |
Will Drewry | 32ac9f5 | 2011-08-18 21:36:27 -0500 | [diff] [blame] | 8 | #include <ctype.h> |
Elly Jones | cd7a904 | 2011-07-22 13:56:51 -0400 | [diff] [blame] | 9 | #include <errno.h> |
| 10 | #include <grp.h> |
| 11 | #include <inttypes.h> |
Will Drewry | fe4a372 | 2011-09-16 14:50:50 -0500 | [diff] [blame] | 12 | #include <limits.h> |
Elly Jones | cd7a904 | 2011-07-22 13:56:51 -0400 | [diff] [blame] | 13 | #include <linux/capability.h> |
| 14 | #include <linux/securebits.h> |
| 15 | #include <pwd.h> |
| 16 | #include <sched.h> |
| 17 | #include <signal.h> |
Will Drewry | 2f54b6a | 2011-09-16 13:45:31 -0500 | [diff] [blame] | 18 | #include <stdarg.h> |
Elly Jones | cd7a904 | 2011-07-22 13:56:51 -0400 | [diff] [blame] | 19 | #include <stdio.h> |
| 20 | #include <stdlib.h> |
| 21 | #include <string.h> |
| 22 | #include <syscall.h> |
| 23 | #include <sys/capability.h> |
| 24 | #include <sys/mount.h> |
Will Drewry | f89aef5 | 2011-09-16 16:48:57 -0500 | [diff] [blame] | 25 | #include <sys/param.h> |
Elly Jones | cd7a904 | 2011-07-22 13:56:51 -0400 | [diff] [blame] | 26 | #include <sys/prctl.h> |
| 27 | #include <sys/wait.h> |
| 28 | #include <syslog.h> |
| 29 | #include <unistd.h> |
| 30 | |
| 31 | #include "libminijail.h" |
Will Drewry | 32ac9f5 | 2011-08-18 21:36:27 -0500 | [diff] [blame] | 32 | #include "libsyscalls.h" |
Elly Jones | cd7a904 | 2011-07-22 13:56:51 -0400 | [diff] [blame] | 33 | #include "libminijail-private.h" |
| 34 | |
Will Drewry | 32ac9f5 | 2011-08-18 21:36:27 -0500 | [diff] [blame] | 35 | /* Until these are reliably available in linux/prctl.h */ |
| 36 | #ifndef PR_SET_SECCOMP_FILTER |
| 37 | # define PR_SECCOMP_FILTER_SYSCALL 0 |
| 38 | # define PR_SECCOMP_FILTER_EVENT 1 |
| 39 | # define PR_GET_SECCOMP_FILTER 35 |
| 40 | # define PR_SET_SECCOMP_FILTER 36 |
| 41 | # define PR_CLEAR_SECCOMP_FILTER 37 |
| 42 | #endif |
| 43 | |
Will Drewry | 32ac9f5 | 2011-08-18 21:36:27 -0500 | [diff] [blame] | 44 | #define die(_msg, ...) do { \ |
| 45 | syslog(LOG_ERR, "libminijail: " _msg, ## __VA_ARGS__); \ |
| 46 | abort(); \ |
| 47 | } while (0) |
Elly Jones | cd7a904 | 2011-07-22 13:56:51 -0400 | [diff] [blame] | 48 | |
Will Drewry | 32ac9f5 | 2011-08-18 21:36:27 -0500 | [diff] [blame] | 49 | #define pdie(_msg, ...) \ |
| 50 | die(_msg ": %s", ## __VA_ARGS__, strerror(errno)) |
| 51 | |
| 52 | #define warn(_msg, ...) \ |
| 53 | syslog(LOG_WARNING, "libminijail: " _msg, ## __VA_ARGS__) |
Elly Jones | cd7a904 | 2011-07-22 13:56:51 -0400 | [diff] [blame] | 54 | |
Will Drewry | f89aef5 | 2011-09-16 16:48:57 -0500 | [diff] [blame] | 55 | struct seccomp_filter { |
| 56 | int nr; |
| 57 | char *filter; |
| 58 | struct seccomp_filter *next, *prev; |
| 59 | }; |
| 60 | |
| 61 | struct minijail { |
| 62 | struct { |
| 63 | int uid : 1; |
| 64 | int gid : 1; |
| 65 | int caps : 1; |
| 66 | int vfs : 1; |
| 67 | int pids : 1; |
| 68 | int seccomp : 1; |
| 69 | int readonly : 1; |
| 70 | int usergroups : 1; |
| 71 | int ptrace : 1; |
| 72 | int seccomp_filter : 1; |
| 73 | } flags; |
| 74 | uid_t uid; |
| 75 | gid_t gid; |
| 76 | gid_t usergid; |
| 77 | char *user; |
| 78 | uint64_t caps; |
| 79 | pid_t initpid; |
| 80 | int filter_count; |
| 81 | struct seccomp_filter *filters; |
| 82 | }; |
| 83 | |
Elly Jones | cd7a904 | 2011-07-22 13:56:51 -0400 | [diff] [blame] | 84 | struct minijail *minijail_new(void) { |
| 85 | struct minijail *j = malloc(sizeof(*j)); |
| 86 | if (j) |
| 87 | memset(j, 0, sizeof(*j)); |
| 88 | return j; |
| 89 | } |
| 90 | |
| 91 | void minijail_change_uid(struct minijail *j, uid_t uid) { |
| 92 | if (uid == 0) |
| 93 | die("useless change to uid 0"); |
| 94 | j->uid = uid; |
| 95 | j->flags.uid = 1; |
| 96 | } |
| 97 | |
| 98 | void minijail_change_gid(struct minijail *j, gid_t gid) { |
| 99 | if (gid == 0) |
| 100 | die("useless change to gid 0"); |
| 101 | j->gid = gid; |
| 102 | j->flags.gid = 1; |
| 103 | } |
| 104 | |
| 105 | int minijail_change_user(struct minijail *j, const char *user) { |
Elly Jones | eb300c5 | 2011-09-22 14:35:43 -0400 | [diff] [blame] | 106 | char *buf = NULL; |
| 107 | struct passwd pw; |
| 108 | struct passwd *ppw = NULL; |
| 109 | ssize_t sz = sysconf(_SC_GETPW_R_SIZE_MAX); |
| 110 | if (sz == -1) |
| 111 | sz = 65536; /* your guess is as good as mine... */ |
| 112 | |
| 113 | /* sysconf(_SC_GETPW_R_SIZE_MAX), under glibc, is documented to return the |
| 114 | * maximum needed size of the buffer, so we don't have to search. */ |
| 115 | buf = malloc(sz); |
| 116 | if (!buf) |
| 117 | return -ENOMEM; |
| 118 | getpwnam_r(user, &pw, buf, sz, &ppw); |
| 119 | free(buf); |
| 120 | if (!ppw) |
Kees Cook | e805d37 | 2011-09-27 15:33:42 -0700 | [diff] [blame] | 121 | return -errno; |
Elly Jones | eb300c5 | 2011-09-22 14:35:43 -0400 | [diff] [blame] | 122 | minijail_change_uid(j, ppw->pw_uid); |
Will Drewry | 2ddaad0 | 2011-09-16 11:36:08 -0500 | [diff] [blame] | 123 | j->user = strdup(user); |
| 124 | if (!j->user) |
| 125 | return -ENOMEM; |
Elly Jones | eb300c5 | 2011-09-22 14:35:43 -0400 | [diff] [blame] | 126 | j->usergid = ppw->pw_gid; |
Elly Jones | cd7a904 | 2011-07-22 13:56:51 -0400 | [diff] [blame] | 127 | return 0; |
| 128 | } |
| 129 | |
| 130 | int minijail_change_group(struct minijail *j, const char *group) { |
Elly Jones | eb300c5 | 2011-09-22 14:35:43 -0400 | [diff] [blame] | 131 | char *buf = NULL; |
| 132 | struct group gr; |
| 133 | struct group *pgr = NULL; |
| 134 | ssize_t sz = sysconf(_SC_GETGR_R_SIZE_MAX); |
| 135 | if (sz == -1) |
| 136 | sz = 65536; /* and mine is as good as yours, really */ |
| 137 | |
| 138 | /* sysconf(_SC_GETGR_R_SIZE_MAX), under glibc, is documented to return the |
| 139 | * maximum needed size of the buffer, so we don't have to search. */ |
| 140 | buf = malloc(sz); |
| 141 | if (!buf) |
| 142 | return -ENOMEM; |
| 143 | getgrnam_r(group, &gr, buf, sz, &pgr); |
| 144 | free(buf); |
| 145 | if (!pgr) |
Kees Cook | e805d37 | 2011-09-27 15:33:42 -0700 | [diff] [blame] | 146 | return -errno; |
Elly Jones | eb300c5 | 2011-09-22 14:35:43 -0400 | [diff] [blame] | 147 | minijail_change_gid(j, pgr->gr_gid); |
Elly Jones | cd7a904 | 2011-07-22 13:56:51 -0400 | [diff] [blame] | 148 | return 0; |
| 149 | } |
| 150 | |
| 151 | void minijail_use_seccomp(struct minijail *j) { |
| 152 | j->flags.seccomp = 1; |
| 153 | } |
| 154 | |
Will Drewry | 32ac9f5 | 2011-08-18 21:36:27 -0500 | [diff] [blame] | 155 | void minijail_use_seccomp_filter(struct minijail *j) { |
| 156 | j->flags.seccomp_filter = 1; |
| 157 | } |
| 158 | |
Elly Jones | cd7a904 | 2011-07-22 13:56:51 -0400 | [diff] [blame] | 159 | void minijail_use_caps(struct minijail *j, uint64_t capmask) { |
| 160 | j->caps = capmask; |
| 161 | j->flags.caps = 1; |
| 162 | } |
| 163 | |
| 164 | void minijail_namespace_vfs(struct minijail *j) { |
| 165 | j->flags.vfs = 1; |
| 166 | } |
| 167 | |
| 168 | void minijail_namespace_pids(struct minijail *j) { |
| 169 | j->flags.pids = 1; |
| 170 | } |
| 171 | |
| 172 | void minijail_remount_readonly(struct minijail *j) { |
| 173 | j->flags.vfs = 1; |
| 174 | j->flags.readonly = 1; |
| 175 | } |
| 176 | |
| 177 | void minijail_inherit_usergroups(struct minijail *j) { |
| 178 | j->flags.usergroups = 1; |
| 179 | } |
| 180 | |
| 181 | void minijail_disable_ptrace(struct minijail *j) { |
| 182 | j->flags.ptrace = 1; |
| 183 | } |
| 184 | |
Will Drewry | 32ac9f5 | 2011-08-18 21:36:27 -0500 | [diff] [blame] | 185 | int minijail_add_seccomp_filter(struct minijail *j, int nr, |
| 186 | const char *filter) { |
| 187 | struct seccomp_filter *sf; |
| 188 | if (!filter || nr < 0) |
| 189 | return -EINVAL; |
| 190 | |
| 191 | sf = malloc(sizeof(*sf)); |
| 192 | if (!sf) |
| 193 | return -ENOMEM; |
| 194 | sf->nr = nr; |
| 195 | sf->filter = strndup(filter, MINIJAIL_MAX_SECCOMP_FILTER_LINE); |
| 196 | if (!sf->filter) { |
| 197 | free(sf); |
| 198 | return -ENOMEM; |
| 199 | } |
| 200 | |
Will Drewry | f89aef5 | 2011-09-16 16:48:57 -0500 | [diff] [blame] | 201 | j->filter_count++; |
| 202 | |
Will Drewry | 32ac9f5 | 2011-08-18 21:36:27 -0500 | [diff] [blame] | 203 | if (!j->filters) { |
| 204 | j->filters = sf; |
| 205 | sf->next = sf; |
| 206 | sf->prev = sf; |
| 207 | return 0; |
| 208 | } |
| 209 | sf->next = j->filters; |
| 210 | sf->prev = j->filters->prev; |
| 211 | sf->prev->next = sf; |
| 212 | j->filters->prev = sf; |
| 213 | return 0; |
| 214 | } |
| 215 | |
| 216 | int minijail_lookup_syscall(const char *name) { |
| 217 | const struct syscall_entry *entry = syscall_table; |
| 218 | for (; entry->name && entry->nr >= 0; ++entry) |
| 219 | if (!strcmp(entry->name, name)) |
| 220 | return entry->nr; |
| 221 | return -1; |
| 222 | } |
| 223 | |
| 224 | static char *strip(char *s) { |
| 225 | char *end; |
| 226 | while (*s && isblank(*s)) |
| 227 | s++; |
| 228 | end = s + strlen(s) - 1; |
| 229 | while (*end && (isblank(*end) || *end == '\n')) |
| 230 | end--; |
| 231 | *(end+1) = '\0'; |
| 232 | return s; |
| 233 | } |
| 234 | |
| 235 | void minijail_parse_seccomp_filters(struct minijail *j, const char *path) { |
| 236 | FILE *file = fopen(path, "r"); |
| 237 | char line[MINIJAIL_MAX_SECCOMP_FILTER_LINE]; |
| 238 | int count = 1; |
| 239 | if (!file) |
| 240 | pdie("failed to open seccomp filters file"); |
| 241 | |
| 242 | /* Format is simple: |
| 243 | * syscall_name<COLON><FILTER STRING>[\n|EOF] |
| 244 | * #...comment... |
| 245 | * <empty line? |
| 246 | */ |
| 247 | while (fgets(line, sizeof(line), file)) { |
| 248 | char *filter = line; |
| 249 | char *name = strsep(&filter, ":"); |
| 250 | char *name_end = NULL; |
| 251 | int nr = -1; |
| 252 | |
| 253 | if (!name) |
| 254 | die("invalid filter on line %d", count); |
| 255 | |
| 256 | name = strip(name); |
| 257 | |
| 258 | if (!filter) { |
| 259 | if (strlen(name)) |
| 260 | die("invalid filter on line %d", count); |
| 261 | /* Allow empty lines */ |
| 262 | continue; |
| 263 | } |
| 264 | |
| 265 | /* Allow comment lines */ |
| 266 | if (*name == '#') |
| 267 | continue; |
| 268 | |
| 269 | filter = strip(filter); |
| 270 | |
| 271 | /* Take direct syscall numbers */ |
| 272 | nr = strtol(name, &name_end, 0); |
| 273 | /* Or fail-over to using names */ |
| 274 | if (*name_end != '\0') |
| 275 | nr = minijail_lookup_syscall(name); |
| 276 | if (nr < 0) |
| 277 | die("syscall '%s' unknown", name); |
| 278 | |
| 279 | if (minijail_add_seccomp_filter(j, nr, filter)) |
| 280 | pdie("failed to add filter for syscall '%s'", name); |
| 281 | } |
| 282 | fclose(file); |
| 283 | } |
| 284 | |
Will Drewry | f89aef5 | 2011-09-16 16:48:57 -0500 | [diff] [blame] | 285 | struct marshal_state { |
| 286 | size_t available; |
| 287 | size_t total; |
| 288 | char *buf; |
| 289 | }; |
| 290 | |
| 291 | static void marshal_state_init(struct marshal_state *state, |
| 292 | char *buf, |
| 293 | size_t available) { |
| 294 | state->available = available; |
| 295 | state->buf = buf; |
| 296 | state->total = 0; |
| 297 | } |
| 298 | |
| 299 | static void marshal_append(struct marshal_state *state, |
| 300 | char *src, |
| 301 | size_t length) { |
| 302 | size_t copy_len = MIN(state->available, length); |
| 303 | |
| 304 | /* Up to |available| will be written. */ |
| 305 | if (copy_len) { |
| 306 | memcpy(state->buf, src, copy_len); |
| 307 | state->buf += copy_len; |
| 308 | state->available -= copy_len; |
| 309 | } |
| 310 | /* |total| will contain the expected length. */ |
| 311 | state->total += length; |
| 312 | } |
| 313 | |
| 314 | static void minijail_marshal_helper(struct marshal_state *state, |
| 315 | const struct minijail *j) { |
| 316 | marshal_append(state, (char *) j, sizeof(*j)); |
Will Drewry | 2ddaad0 | 2011-09-16 11:36:08 -0500 | [diff] [blame] | 317 | if (j->user) |
Will Drewry | f89aef5 | 2011-09-16 16:48:57 -0500 | [diff] [blame] | 318 | marshal_append(state, j->user, strlen(j->user) + 1); |
| 319 | if (j->flags.seccomp_filter && j->filters) { |
| 320 | struct seccomp_filter *f = j->filters; |
| 321 | do { |
| 322 | marshal_append(state, (char *) &f->nr, sizeof(f->nr)); |
| 323 | marshal_append(state, f->filter, strlen(f->filter) + 1); |
| 324 | f = f->next; |
| 325 | } while (f != j->filters); |
| 326 | } |
| 327 | } |
| 328 | |
| 329 | size_t minijail_size(const struct minijail *j) { |
| 330 | struct marshal_state state; |
| 331 | marshal_state_init(&state, NULL, 0); |
| 332 | minijail_marshal_helper(&state, j); |
| 333 | return state.total; |
Will Drewry | 2ddaad0 | 2011-09-16 11:36:08 -0500 | [diff] [blame] | 334 | } |
| 335 | |
Will Drewry | 2ddaad0 | 2011-09-16 11:36:08 -0500 | [diff] [blame] | 336 | int minijail_marshal(const struct minijail *j, char *buf, size_t available) { |
Will Drewry | f89aef5 | 2011-09-16 16:48:57 -0500 | [diff] [blame] | 337 | struct marshal_state state; |
| 338 | marshal_state_init(&state, buf, available); |
| 339 | minijail_marshal_helper(&state, j); |
| 340 | return (state.total > available); |
Will Drewry | 2ddaad0 | 2011-09-16 11:36:08 -0500 | [diff] [blame] | 341 | } |
| 342 | |
| 343 | int minijail_unmarshal(struct minijail *j, char *serialized, size_t length) { |
| 344 | if (length < sizeof(*j)) |
| 345 | return -EINVAL; |
| 346 | memcpy((void *) j, serialized, sizeof(*j)); |
| 347 | serialized += sizeof(*j); |
| 348 | length -= sizeof(*j); |
Will Drewry | f89aef5 | 2011-09-16 16:48:57 -0500 | [diff] [blame] | 349 | |
Will Drewry | 2ddaad0 | 2011-09-16 11:36:08 -0500 | [diff] [blame] | 350 | if (j->user) { /* stale pointer */ |
| 351 | if (!length) |
| 352 | return -EINVAL; |
| 353 | j->user = strndup(serialized, length); |
| 354 | length -= strlen(j->user) + 1; |
Will Drewry | f89aef5 | 2011-09-16 16:48:57 -0500 | [diff] [blame] | 355 | serialized += strlen(j->user) + 1; |
| 356 | } |
| 357 | |
| 358 | if (j->flags.seccomp_filter && j->filter_count) { |
| 359 | int count = j->filter_count; |
| 360 | /* Let add_seccomp_filter recompute the value. */ |
| 361 | j->filter_count = 0; |
| 362 | j->filters = NULL; /* Don't follow the stale pointer. */ |
| 363 | for ( ; count > 0; --count) { |
| 364 | int *nr = (int *) serialized; |
| 365 | char *filter; |
| 366 | if (length < sizeof(*nr)) |
| 367 | return -EINVAL; |
| 368 | length -= sizeof(*nr); |
| 369 | serialized += sizeof(*nr); |
| 370 | if (!length) |
| 371 | return -EINVAL; |
| 372 | filter = serialized; |
| 373 | if (minijail_add_seccomp_filter(j, *nr, filter)) |
| 374 | return -EINVAL; |
| 375 | length -= strlen(filter) + 1; |
| 376 | serialized += strlen(filter) + 1; |
| 377 | } |
Will Drewry | 2ddaad0 | 2011-09-16 11:36:08 -0500 | [diff] [blame] | 378 | } |
| 379 | return 0; |
| 380 | } |
| 381 | |
Will Drewry | fe4a372 | 2011-09-16 14:50:50 -0500 | [diff] [blame] | 382 | void minijail_preenter(struct minijail *j) { |
| 383 | /* Strip out options which are minijail_run() only. */ |
| 384 | j->flags.vfs = 0; |
| 385 | j->flags.readonly = 0; |
| 386 | j->flags.pids = 0; |
| 387 | } |
| 388 | |
| 389 | void minijail_preexec(struct minijail *j) { |
| 390 | int vfs = j->flags.vfs; |
| 391 | int readonly = j->flags.readonly; |
Will Drewry | 2ddaad0 | 2011-09-16 11:36:08 -0500 | [diff] [blame] | 392 | if (j->user) |
| 393 | free(j->user); |
| 394 | j->user = NULL; |
Will Drewry | fe4a372 | 2011-09-16 14:50:50 -0500 | [diff] [blame] | 395 | memset(&j->flags, 0, sizeof(j->flags)); |
| 396 | /* Now restore anything we meant to keep. */ |
| 397 | j->flags.vfs = vfs; |
| 398 | j->flags.readonly = readonly; |
| 399 | /* Note, pidns will already have been used before this call. */ |
Will Drewry | 2ddaad0 | 2011-09-16 11:36:08 -0500 | [diff] [blame] | 400 | } |
| 401 | |
Elly Jones | cd7a904 | 2011-07-22 13:56:51 -0400 | [diff] [blame] | 402 | static int remount_readonly(void) { |
| 403 | const char *kProcPath = "/proc"; |
| 404 | const unsigned int kSafeFlags = MS_NODEV | MS_NOEXEC | MS_NOSUID; |
| 405 | /* Right now, we're holding a reference to our parent's old mount of /proc in |
| 406 | * our namespace, which means using MS_REMOUNT here would mutate our parent's |
| 407 | * mount as well, even though we're in a VFS namespace (!). Instead, remove |
| 408 | * their mount from our namespace and make our own. */ |
| 409 | if (umount(kProcPath)) |
Kees Cook | e805d37 | 2011-09-27 15:33:42 -0700 | [diff] [blame] | 410 | return -errno; |
Elly Jones | cd7a904 | 2011-07-22 13:56:51 -0400 | [diff] [blame] | 411 | if (mount("", kProcPath, "proc", kSafeFlags | MS_RDONLY, "")) |
Kees Cook | e805d37 | 2011-09-27 15:33:42 -0700 | [diff] [blame] | 412 | return -errno; |
Elly Jones | cd7a904 | 2011-07-22 13:56:51 -0400 | [diff] [blame] | 413 | return 0; |
| 414 | } |
| 415 | |
| 416 | static void drop_caps(const struct minijail *j) { |
| 417 | cap_t caps = cap_get_proc(); |
| 418 | cap_value_t raise_flag[1]; |
| 419 | unsigned int i; |
| 420 | if (!caps) |
| 421 | die("can't get process caps"); |
| 422 | if (cap_clear_flag(caps, CAP_INHERITABLE)) |
| 423 | die("can't clear inheritable caps"); |
| 424 | if (cap_clear_flag(caps, CAP_EFFECTIVE)) |
| 425 | die("can't clear effective caps"); |
| 426 | if (cap_clear_flag(caps, CAP_PERMITTED)) |
| 427 | die("can't clear permitted caps"); |
| 428 | for (i = 0; i < sizeof(j->caps) * 8 && cap_valid((int)i); ++i) { |
| 429 | if (i != CAP_SETPCAP && !(j->caps & (1 << i))) |
| 430 | continue; |
| 431 | raise_flag[0] = i; |
| 432 | if (cap_set_flag(caps, CAP_EFFECTIVE, 1, raise_flag, CAP_SET)) |
| 433 | die("can't add effective cap"); |
| 434 | if (cap_set_flag(caps, CAP_PERMITTED, 1, raise_flag, CAP_SET)) |
| 435 | die("can't add permitted cap"); |
| 436 | if (cap_set_flag(caps, CAP_INHERITABLE, 1, raise_flag, CAP_SET)) |
| 437 | die("can't add inheritable cap"); |
| 438 | } |
| 439 | if (cap_set_proc(caps)) |
| 440 | die("can't apply cleaned capset"); |
| 441 | cap_free(caps); |
| 442 | for (i = 0; i < sizeof(j->caps) * 8 && cap_valid((int)i); ++i) { |
| 443 | if (j->caps & (1 << i)) |
| 444 | continue; |
| 445 | if (prctl(PR_CAPBSET_DROP, i)) |
| 446 | pdie("prctl(PR_CAPBSET_DROP)"); |
| 447 | } |
| 448 | } |
| 449 | |
Will Drewry | 32ac9f5 | 2011-08-18 21:36:27 -0500 | [diff] [blame] | 450 | static int setup_seccomp_filters(const struct minijail *j) { |
| 451 | const struct seccomp_filter *sf = j->filters; |
| 452 | int ret = 0; |
| 453 | int broaden = 0; |
| 454 | |
| 455 | /* No filters installed isn't necessarily an error. */ |
| 456 | if (!sf) |
| 457 | return ret; |
| 458 | |
| 459 | do { |
| 460 | errno = 0; |
| 461 | ret = prctl(PR_SET_SECCOMP_FILTER, PR_SECCOMP_FILTER_SYSCALL, |
| 462 | sf->nr, broaden ? "1" : sf->filter); |
| 463 | if (ret) { |
| 464 | switch (errno) { |
| 465 | case ENOSYS: |
| 466 | /* TODO(wad) make this a config option */ |
| 467 | if (broaden) |
| 468 | die("CONFIG_SECCOMP_FILTER is not supported by your kernel"); |
| 469 | warn("missing CONFIG_FTRACE_SYSCALLS; relaxing the filter for %d", |
| 470 | sf->nr); |
| 471 | broaden = 1; |
| 472 | continue; |
| 473 | case E2BIG: |
| 474 | warn("seccomp filter too long: %d", sf->nr); |
| 475 | pdie("filter too long"); |
| 476 | case ENOSPC: |
| 477 | pdie("too many seccomp filters"); |
| 478 | case EPERM: |
| 479 | warn("syscall filter disallowed for %d", sf->nr); |
| 480 | pdie("failed to install seccomp filter"); |
| 481 | case EINVAL: |
| 482 | warn("seccomp filter or call method is invalid. %d:'%s'", |
| 483 | sf->nr, sf->filter); |
| 484 | default: |
| 485 | pdie("failed to install seccomp filter"); |
| 486 | } |
| 487 | } |
| 488 | sf = sf->next; |
| 489 | broaden = 0; |
| 490 | } while (sf != j->filters); |
| 491 | return ret; |
| 492 | } |
| 493 | |
Elly Jones | cd7a904 | 2011-07-22 13:56:51 -0400 | [diff] [blame] | 494 | void minijail_enter(const struct minijail *j) { |
| 495 | if (j->flags.pids) |
| 496 | die("tried to enter a pid-namespaced jail; try minijail_run()?"); |
| 497 | |
Will Drewry | f89aef5 | 2011-09-16 16:48:57 -0500 | [diff] [blame] | 498 | if (j->flags.seccomp_filter && setup_seccomp_filters(j)) |
| 499 | pdie("failed to configure seccomp filters"); |
Will Drewry | 32ac9f5 | 2011-08-18 21:36:27 -0500 | [diff] [blame] | 500 | |
Elly Jones | cd7a904 | 2011-07-22 13:56:51 -0400 | [diff] [blame] | 501 | if (j->flags.usergroups && !j->user) |
| 502 | die("usergroup inheritance without username"); |
| 503 | |
| 504 | /* We can't recover from failures if we've dropped privileges partially, |
| 505 | * so we don't even try. If any of our operations fail, we abort() the |
| 506 | * entire process. */ |
| 507 | if (j->flags.vfs && unshare(CLONE_NEWNS)) |
| 508 | pdie("unshare"); |
| 509 | |
| 510 | if (j->flags.readonly && remount_readonly()) |
| 511 | pdie("remount"); |
| 512 | |
| 513 | if (j->flags.caps) { |
| 514 | /* POSIX capabilities are a bit tricky. If we drop our capability to change |
| 515 | * uids, our attempt to use setuid() below will fail. Hang on to root caps |
| 516 | * across setuid(), then lock securebits. */ |
| 517 | if (prctl(PR_SET_KEEPCAPS, 1)) |
| 518 | pdie("prctl(PR_SET_KEEPCAPS)"); |
| 519 | if (prctl(PR_SET_SECUREBITS, SECURE_ALL_BITS | SECURE_ALL_LOCKS)) |
| 520 | pdie("prctl(PR_SET_SECUREBITS)"); |
| 521 | } |
| 522 | |
Will Drewry | c6c8643 | 2011-09-18 14:37:22 -0500 | [diff] [blame] | 523 | if (j->flags.usergroups) { |
| 524 | if (initgroups(j->user, j->usergid)) |
| 525 | pdie("initgroups"); |
| 526 | } else { |
| 527 | /* Only attempt to clear supplemental groups if we are changing users. */ |
| 528 | if ((j->uid || j->gid) && setgroups(0, NULL)) |
| 529 | pdie("setgroups"); |
Will Drewry | 32ac9f5 | 2011-08-18 21:36:27 -0500 | [diff] [blame] | 530 | } |
Elly Jones | cd7a904 | 2011-07-22 13:56:51 -0400 | [diff] [blame] | 531 | |
| 532 | if (j->flags.gid && setresgid(j->gid, j->gid, j->gid)) |
| 533 | pdie("setresgid"); |
| 534 | |
| 535 | if (j->flags.uid && setresuid(j->uid, j->uid, j->uid)) |
| 536 | pdie("setresuid"); |
| 537 | |
| 538 | if (j->flags.caps) |
| 539 | drop_caps(j); |
| 540 | |
| 541 | /* seccomp has to come last since it cuts off all the other |
| 542 | * privilege-dropping syscalls :) */ |
Will Drewry | 32ac9f5 | 2011-08-18 21:36:27 -0500 | [diff] [blame] | 543 | if (j->flags.seccomp_filter && prctl(PR_SET_SECCOMP, 13)) |
| 544 | pdie("prctl(PR_SET_SECCOMP, 13)"); |
| 545 | |
Elly Jones | cd7a904 | 2011-07-22 13:56:51 -0400 | [diff] [blame] | 546 | if (j->flags.seccomp && prctl(PR_SET_SECCOMP, 1)) |
| 547 | pdie("prctl(PR_SET_SECCOMP)"); |
| 548 | } |
| 549 | |
| 550 | static int init_exitstatus = 0; |
| 551 | |
| 552 | static void init_term(int __attribute__((unused)) sig) { |
| 553 | _exit(init_exitstatus); |
| 554 | } |
| 555 | |
| 556 | static int init(pid_t rootpid) { |
| 557 | pid_t pid; |
| 558 | int status; |
| 559 | signal(SIGTERM, init_term); /* so that we exit with the right status */ |
Will Drewry | f89aef5 | 2011-09-16 16:48:57 -0500 | [diff] [blame] | 560 | /* TODO(wad) self jail with seccomp_filters here. */ |
Elly Jones | cd7a904 | 2011-07-22 13:56:51 -0400 | [diff] [blame] | 561 | while ((pid = wait(&status)) > 0) { |
| 562 | /* This loop will only end when either there are no processes left inside |
| 563 | * our pid namespace or we get a signal. */ |
| 564 | if (pid == rootpid) |
| 565 | init_exitstatus = status; |
| 566 | } |
| 567 | if (!WIFEXITED(init_exitstatus)) |
| 568 | _exit(MINIJAIL_ERR_INIT); |
| 569 | _exit(WEXITSTATUS(init_exitstatus)); |
| 570 | } |
| 571 | |
Will Drewry | fe4a372 | 2011-09-16 14:50:50 -0500 | [diff] [blame] | 572 | int minijail_from_fd(int fd, struct minijail *j) { |
| 573 | size_t sz = 0; |
| 574 | size_t bytes = read(fd, &sz, sizeof(sz)); |
| 575 | char *buf; |
Will Drewry | 2f54b6a | 2011-09-16 13:45:31 -0500 | [diff] [blame] | 576 | int r; |
Will Drewry | fe4a372 | 2011-09-16 14:50:50 -0500 | [diff] [blame] | 577 | if (sizeof(sz) != bytes) |
| 578 | return -EINVAL; |
| 579 | if (sz > USHRT_MAX) /* Arbitrary sanity check */ |
Will Drewry | 2f54b6a | 2011-09-16 13:45:31 -0500 | [diff] [blame] | 580 | return -E2BIG; |
Will Drewry | fe4a372 | 2011-09-16 14:50:50 -0500 | [diff] [blame] | 581 | buf = malloc(sz); |
| 582 | if (!buf) |
| 583 | return -ENOMEM; |
| 584 | bytes = read(fd, buf, sz); |
| 585 | if (bytes != sz) { |
| 586 | free(buf); |
| 587 | return -EINVAL; |
| 588 | } |
| 589 | r = minijail_unmarshal(j, buf, sz); |
| 590 | free(buf); |
| 591 | return r; |
Will Drewry | 2f54b6a | 2011-09-16 13:45:31 -0500 | [diff] [blame] | 592 | } |
| 593 | |
Will Drewry | fe4a372 | 2011-09-16 14:50:50 -0500 | [diff] [blame] | 594 | int minijail_to_fd(struct minijail *j, int fd) { |
| 595 | char *buf; |
| 596 | size_t sz = minijail_size(j); |
| 597 | ssize_t written; |
| 598 | int r; |
Elly Jones | cd7a904 | 2011-07-22 13:56:51 -0400 | [diff] [blame] | 599 | |
Will Drewry | fe4a372 | 2011-09-16 14:50:50 -0500 | [diff] [blame] | 600 | if (!sz) |
| 601 | return -EINVAL; |
| 602 | buf = malloc(sz); |
| 603 | if ((r = minijail_marshal(j, buf, sz))) { |
| 604 | free(buf); |
| 605 | return r; |
| 606 | } |
| 607 | /* Sends [size][minijail]. */ |
| 608 | written = write(fd, &sz, sizeof(sz)); |
| 609 | if (written != sizeof(sz)) { |
| 610 | free(buf); |
| 611 | return -EFAULT; |
| 612 | } |
| 613 | written = write(fd, buf, sz); |
| 614 | if (written < 0 || (size_t) written != sz) { |
| 615 | free(buf); |
| 616 | return -EFAULT; |
| 617 | } |
| 618 | free(buf); |
| 619 | return 0; |
Will Drewry | 2f54b6a | 2011-09-16 13:45:31 -0500 | [diff] [blame] | 620 | } |
Elly Jones | cd7a904 | 2011-07-22 13:56:51 -0400 | [diff] [blame] | 621 | |
Will Drewry | 2f54b6a | 2011-09-16 13:45:31 -0500 | [diff] [blame] | 622 | static int setup_preload(void) { |
| 623 | char *oldenv = getenv(kLdPreloadEnvVar) ? : ""; |
| 624 | char *newenv = malloc(strlen(oldenv) + 2 + strlen(PRELOADPATH)); |
| 625 | if (!newenv) |
Elly Jones | cd7a904 | 2011-07-22 13:56:51 -0400 | [diff] [blame] | 626 | return -ENOMEM; |
Elly Jones | cd7a904 | 2011-07-22 13:56:51 -0400 | [diff] [blame] | 627 | |
| 628 | /* Only insert a separating space if we have something to separate... */ |
| 629 | sprintf(newenv, "%s%s%s", oldenv, strlen(oldenv) ? " " : "", PRELOADPATH); |
| 630 | |
| 631 | /* setenv() makes a copy of the string we give it */ |
Ben Chan | 541c7e5 | 2011-08-26 14:55:53 -0700 | [diff] [blame] | 632 | setenv(kLdPreloadEnvVar, newenv, 1); |
Elly Jones | cd7a904 | 2011-07-22 13:56:51 -0400 | [diff] [blame] | 633 | free(newenv); |
Elly Jones | cd7a904 | 2011-07-22 13:56:51 -0400 | [diff] [blame] | 634 | return 0; |
| 635 | } |
| 636 | |
Will Drewry | f89aef5 | 2011-09-16 16:48:57 -0500 | [diff] [blame] | 637 | static int setup_pipe(int fds[2]) { |
| 638 | int r = pipe(fds); |
| 639 | char fd_buf[11]; |
| 640 | if (r) |
| 641 | return r; |
| 642 | r = snprintf(fd_buf, sizeof(fd_buf), "%d", fds[0]); |
| 643 | if (r <= 0) |
| 644 | return -EINVAL; |
| 645 | setenv(kFdEnvVar, fd_buf, 1); |
| 646 | return 0; |
| 647 | } |
| 648 | |
Elly Jones | cd7a904 | 2011-07-22 13:56:51 -0400 | [diff] [blame] | 649 | int minijail_run(struct minijail *j, const char *filename, char *const argv[]) { |
| 650 | unsigned int pidns = j->flags.pids ? CLONE_NEWPID : 0; |
Ben Chan | 541c7e5 | 2011-08-26 14:55:53 -0700 | [diff] [blame] | 651 | char *oldenv, *oldenv_copy = NULL; |
Will Drewry | f89aef5 | 2011-09-16 16:48:57 -0500 | [diff] [blame] | 652 | pid_t child_pid; |
Will Drewry | 2f54b6a | 2011-09-16 13:45:31 -0500 | [diff] [blame] | 653 | int pipe_fds[2]; |
Will Drewry | f89aef5 | 2011-09-16 16:48:57 -0500 | [diff] [blame] | 654 | int ret; |
Ben Chan | 541c7e5 | 2011-08-26 14:55:53 -0700 | [diff] [blame] | 655 | |
| 656 | oldenv = getenv(kLdPreloadEnvVar); |
| 657 | if (oldenv) { |
| 658 | oldenv_copy = strdup(oldenv); |
| 659 | if (!oldenv_copy) |
| 660 | return -ENOMEM; |
| 661 | } |
Will Drewry | f89aef5 | 2011-09-16 16:48:57 -0500 | [diff] [blame] | 662 | |
| 663 | if (setup_preload()) |
| 664 | return -EFAULT; |
Will Drewry | 2f54b6a | 2011-09-16 13:45:31 -0500 | [diff] [blame] | 665 | |
| 666 | /* Before we fork(2) and execve(2) the child process, we need to open |
| 667 | * a pipe(2) to send the minijail configuration over. |
| 668 | */ |
Will Drewry | f89aef5 | 2011-09-16 16:48:57 -0500 | [diff] [blame] | 669 | if (setup_pipe(pipe_fds)) |
| 670 | return -EFAULT; |
Elly Jones | cd7a904 | 2011-07-22 13:56:51 -0400 | [diff] [blame] | 671 | |
Will Drewry | f89aef5 | 2011-09-16 16:48:57 -0500 | [diff] [blame] | 672 | child_pid = syscall(SYS_clone, pidns | SIGCHLD, NULL); |
| 673 | if (child_pid < 0) { |
| 674 | free(oldenv_copy); |
| 675 | return child_pid; |
| 676 | } |
| 677 | |
| 678 | if (child_pid) { |
| 679 | /* Restore parent's LD_PRELOAD. */ |
Ben Chan | 541c7e5 | 2011-08-26 14:55:53 -0700 | [diff] [blame] | 680 | if (oldenv_copy) { |
| 681 | setenv(kLdPreloadEnvVar, oldenv_copy, 1); |
| 682 | free(oldenv_copy); |
| 683 | } else { |
| 684 | unsetenv(kLdPreloadEnvVar); |
| 685 | } |
Will Drewry | 2f54b6a | 2011-09-16 13:45:31 -0500 | [diff] [blame] | 686 | unsetenv(kFdEnvVar); |
Will Drewry | f89aef5 | 2011-09-16 16:48:57 -0500 | [diff] [blame] | 687 | j->initpid = child_pid; |
| 688 | close(pipe_fds[0]); /* read endpoint */ |
| 689 | ret = minijail_to_fd(j, pipe_fds[1]); |
| 690 | close(pipe_fds[1]); /* write endpoint */ |
| 691 | if (ret) { |
Will Drewry | 2f54b6a | 2011-09-16 13:45:31 -0500 | [diff] [blame] | 692 | kill(j->initpid, SIGKILL); |
| 693 | die("failed to send marshalled minijail"); |
| 694 | } |
Elly Jones | cd7a904 | 2011-07-22 13:56:51 -0400 | [diff] [blame] | 695 | return 0; |
| 696 | } |
Ben Chan | 541c7e5 | 2011-08-26 14:55:53 -0700 | [diff] [blame] | 697 | free(oldenv_copy); |
| 698 | |
Will Drewry | fe4a372 | 2011-09-16 14:50:50 -0500 | [diff] [blame] | 699 | /* Drop everything that cannot be inherited across execve. */ |
| 700 | minijail_preexec(j); |
Elly Jones | cd7a904 | 2011-07-22 13:56:51 -0400 | [diff] [blame] | 701 | /* Jail this process and its descendants... */ |
| 702 | minijail_enter(j); |
| 703 | |
| 704 | if (pidns) { |
| 705 | /* pid namespace: this process will become init inside the new namespace, so |
| 706 | * fork off a child to actually run the program (we don't want all programs |
| 707 | * we might exec to have to know how to be init). */ |
Will Drewry | f89aef5 | 2011-09-16 16:48:57 -0500 | [diff] [blame] | 708 | child_pid = fork(); |
| 709 | if (child_pid < 0) |
| 710 | _exit(child_pid); |
| 711 | else if (child_pid > 0) |
| 712 | init(child_pid); /* never returns */ |
Elly Jones | cd7a904 | 2011-07-22 13:56:51 -0400 | [diff] [blame] | 713 | } |
| 714 | |
| 715 | /* If we aren't pid-namespaced: |
| 716 | * calling process |
| 717 | * -> execve()-ing process |
| 718 | * If we are: |
| 719 | * calling process |
| 720 | * -> init()-ing process |
| 721 | * -> execve()-ing process |
| 722 | */ |
| 723 | _exit(execve(filename, argv, environ)); |
| 724 | } |
| 725 | |
| 726 | int minijail_kill(struct minijail *j) { |
| 727 | int st; |
| 728 | if (kill(j->initpid, SIGTERM)) |
Kees Cook | e805d37 | 2011-09-27 15:33:42 -0700 | [diff] [blame] | 729 | return -errno; |
Elly Jones | cd7a904 | 2011-07-22 13:56:51 -0400 | [diff] [blame] | 730 | if (waitpid(j->initpid, &st, 0) < 0) |
Kees Cook | e805d37 | 2011-09-27 15:33:42 -0700 | [diff] [blame] | 731 | return -errno; |
Elly Jones | cd7a904 | 2011-07-22 13:56:51 -0400 | [diff] [blame] | 732 | return st; |
| 733 | } |
| 734 | |
| 735 | int minijail_wait(struct minijail *j) { |
| 736 | int st; |
| 737 | if (waitpid(j->initpid, &st, 0) < 0) |
Kees Cook | e805d37 | 2011-09-27 15:33:42 -0700 | [diff] [blame] | 738 | return -errno; |
Elly Jones | cd7a904 | 2011-07-22 13:56:51 -0400 | [diff] [blame] | 739 | if (!WIFEXITED(st)) |
| 740 | return MINIJAIL_ERR_JAIL; |
| 741 | return WEXITSTATUS(st); |
| 742 | } |
| 743 | |
| 744 | void minijail_destroy(struct minijail *j) { |
Will Drewry | 32ac9f5 | 2011-08-18 21:36:27 -0500 | [diff] [blame] | 745 | struct seccomp_filter *f = j->filters; |
| 746 | /* Unlink the tail and head */ |
| 747 | if (f) |
| 748 | f->prev->next = NULL; |
| 749 | while (f) { |
| 750 | struct seccomp_filter *next = f->next; |
| 751 | free(f->filter); |
| 752 | free(f); |
| 753 | f = next; |
| 754 | } |
Will Drewry | 2ddaad0 | 2011-09-16 11:36:08 -0500 | [diff] [blame] | 755 | if (j->user) |
| 756 | free(j->user); |
Elly Jones | cd7a904 | 2011-07-22 13:56:51 -0400 | [diff] [blame] | 757 | free(j); |
| 758 | } |