]>
Commit | Line | Data |
---|---|---|
e0732705 CS |
1 | /* |
2 | * lxc: linux Container library | |
3 | * | |
4 | * (C) Copyright IBM Corp. 2007, 2008 | |
5 | * | |
6 | * Authors: | |
9afe19d6 | 7 | * Daniel Lezcano <daniel.lezcano at free.fr> |
e0732705 CS |
8 | * |
9 | * This library is free software; you can redistribute it and/or | |
10 | * modify it under the terms of the GNU Lesser General Public | |
11 | * License as published by the Free Software Foundation; either | |
12 | * version 2.1 of the License, or (at your option) any later version. | |
13 | * | |
14 | * This library is distributed in the hope that it will be useful, | |
15 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
16 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
17 | * Lesser General Public License for more details. | |
18 | * | |
19 | * You should have received a copy of the GNU Lesser General Public | |
20 | * License along with this library; if not, write to the Free Software | |
250b1eec | 21 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
e0732705 CS |
22 | */ |
23 | ||
24 | #define _GNU_SOURCE | |
25 | #include <unistd.h> | |
26 | #include <stdio.h> | |
27 | #include <string.h> | |
28 | #include <stdlib.h> | |
2eef2bda | 29 | #include <signal.h> |
e0732705 CS |
30 | #include <errno.h> |
31 | #include <fcntl.h> | |
c476bdce | 32 | #include <grp.h> |
e0732705 CS |
33 | #include <sys/param.h> |
34 | #include <sys/prctl.h> | |
7a0b0b56 | 35 | #include <sys/mount.h> |
5ec27989 | 36 | #include <sys/socket.h> |
1ba0013f | 37 | #include <sys/syscall.h> |
905022f7 | 38 | #include <sys/wait.h> |
910bb4fa | 39 | #include <linux/unistd.h> |
905022f7 | 40 | #include <pwd.h> |
e0732705 | 41 | |
955e2a02 | 42 | #ifndef HAVE_DECL_PR_CAPBSET_DROP |
e0732705 CS |
43 | #define PR_CAPBSET_DROP 24 |
44 | #endif | |
45 | ||
955e2a02 CB |
46 | #ifndef HAVE_DECL_PR_SET_NO_NEW_PRIVS |
47 | #define PR_SET_NO_NEW_PRIVS 38 | |
48 | #endif | |
49 | ||
50 | #ifndef HAVE_DECL_PR_GET_NO_NEW_PRIVS | |
51 | #define PR_GET_NO_NEW_PRIVS 39 | |
52 | #endif | |
53 | ||
e0732705 CS |
54 | #include "namespace.h" |
55 | #include "log.h" | |
81f466d0 | 56 | #include "af_unix.h" |
e0732705 CS |
57 | #include "attach.h" |
58 | #include "caps.h" | |
e0732705 | 59 | #include "config.h" |
6a44839f | 60 | #include "utils.h" |
9c4693b8 CS |
61 | #include "commands.h" |
62 | #include "cgroup.h" | |
025ed0f3 | 63 | #include "lxclock.h" |
2c4ea790 SH |
64 | #include "conf.h" |
65 | #include "lxcseccomp.h" | |
66 | #include <lxc/lxccontainer.h> | |
fe4de9a6 | 67 | #include "lsm/lsm.h" |
9b8e3c96 | 68 | #include "confile.h" |
9c4693b8 CS |
69 | |
70 | #if HAVE_SYS_PERSONALITY_H | |
71 | #include <sys/personality.h> | |
72 | #endif | |
e0732705 | 73 | |
a3da2f3b SG |
74 | #ifndef SOCK_CLOEXEC |
75 | # define SOCK_CLOEXEC 02000000 | |
76 | #endif | |
77 | ||
d6a3c917 SG |
78 | #ifndef MS_REC |
79 | #define MS_REC 16384 | |
80 | #endif | |
81 | ||
82 | #ifndef MS_SLAVE | |
83 | #define MS_SLAVE (1<<19) | |
84 | #endif | |
85 | ||
e0732705 CS |
86 | lxc_log_define(lxc_attach, lxc); |
87 | ||
82b1f317 CB |
88 | /* /proc/pid-to-str/current\0 = (5 + 21 + 7 + 1) */ |
89 | #define __LSMATTRLEN (5 + 21 + 7 + 1) | |
81f466d0 CB |
90 | static int lsm_openat(int procfd, pid_t pid, int on_exec) |
91 | { | |
92 | int ret = -1; | |
5c3fcae7 | 93 | int labelfd = -1; |
82b1f317 | 94 | const char *name; |
81f466d0 | 95 | char path[__LSMATTRLEN]; |
5c3fcae7 SG |
96 | |
97 | name = lsm_name(); | |
98 | ||
99 | if (strcmp(name, "nop") == 0) | |
81f466d0 | 100 | return 0; |
5c3fcae7 SG |
101 | |
102 | if (strcmp(name, "none") == 0) | |
81f466d0 | 103 | return 0; |
5c3fcae7 SG |
104 | |
105 | /* We don't support on-exec with AppArmor */ | |
106 | if (strcmp(name, "AppArmor") == 0) | |
107 | on_exec = 0; | |
108 | ||
82b1f317 | 109 | if (on_exec) |
81f466d0 | 110 | ret = snprintf(path, __LSMATTRLEN, "%d/attr/exec", pid); |
82b1f317 | 111 | else |
81f466d0 | 112 | ret = snprintf(path, __LSMATTRLEN, "%d/attr/current", pid); |
82b1f317 CB |
113 | if (ret < 0 || ret >= __LSMATTRLEN) |
114 | return -1; | |
5c3fcae7 | 115 | |
82b1f317 | 116 | labelfd = openat(procfd, path, O_RDWR); |
5c3fcae7 | 117 | if (labelfd < 0) { |
82b1f317 | 118 | SYSERROR("Unable to open file descriptor to set LSM label."); |
81f466d0 | 119 | return -1; |
5c3fcae7 SG |
120 | } |
121 | ||
81f466d0 CB |
122 | return labelfd; |
123 | } | |
124 | ||
125 | static int lsm_set_label_at(int lsm_labelfd, int on_exec, char *lsm_label) | |
126 | { | |
127 | int fret = -1; | |
128 | const char* name; | |
129 | char *command = NULL; | |
130 | ||
131 | name = lsm_name(); | |
132 | ||
133 | if (strcmp(name, "nop") == 0) | |
134 | return 0; | |
135 | ||
136 | if (strcmp(name, "none") == 0) | |
137 | return 0; | |
138 | ||
139 | /* We don't support on-exec with AppArmor */ | |
140 | if (strcmp(name, "AppArmor") == 0) | |
141 | on_exec = 0; | |
142 | ||
5c3fcae7 SG |
143 | if (strcmp(name, "AppArmor") == 0) { |
144 | int size; | |
145 | ||
146 | command = malloc(strlen(lsm_label) + strlen("changeprofile ") + 1); | |
147 | if (!command) { | |
148 | SYSERROR("Failed to write apparmor profile"); | |
5c3fcae7 SG |
149 | goto out; |
150 | } | |
151 | ||
152 | size = sprintf(command, "changeprofile %s", lsm_label); | |
153 | if (size < 0) { | |
154 | SYSERROR("Failed to write apparmor profile"); | |
5c3fcae7 SG |
155 | goto out; |
156 | } | |
157 | ||
81f466d0 CB |
158 | if (write(lsm_labelfd, command, size + 1) < 0) { |
159 | SYSERROR("Unable to set LSM label: %s.", command); | |
5c3fcae7 SG |
160 | goto out; |
161 | } | |
81f466d0 CB |
162 | INFO("Set LSM label to: %s.", command); |
163 | } else if (strcmp(name, "SELinux") == 0) { | |
164 | if (write(lsm_labelfd, lsm_label, strlen(lsm_label) + 1) < 0) { | |
5c3fcae7 | 165 | SYSERROR("Unable to set LSM label"); |
5c3fcae7 SG |
166 | goto out; |
167 | } | |
81f466d0 CB |
168 | INFO("Set LSM label to: %s.", lsm_label); |
169 | } else { | |
5c3fcae7 | 170 | ERROR("Unable to restore label for unknown LSM: %s", name); |
5c3fcae7 SG |
171 | goto out; |
172 | } | |
81f466d0 | 173 | fret = 0; |
5c3fcae7 SG |
174 | |
175 | out: | |
176 | free(command); | |
177 | ||
81f466d0 CB |
178 | if (lsm_labelfd != -1) |
179 | close(lsm_labelfd); | |
5c3fcae7 | 180 | |
81f466d0 | 181 | return fret; |
5c3fcae7 SG |
182 | } |
183 | ||
74a3920a | 184 | static struct lxc_proc_context_info *lxc_proc_get_context_info(pid_t pid) |
e0732705 CS |
185 | { |
186 | struct lxc_proc_context_info *info = calloc(1, sizeof(*info)); | |
187 | FILE *proc_file; | |
188 | char proc_fn[MAXPATHLEN]; | |
460a1cf0 | 189 | char *line = NULL; |
e0732705 | 190 | size_t line_bufsz = 0; |
460a1cf0 | 191 | int ret, found; |
e0732705 CS |
192 | |
193 | if (!info) { | |
194 | SYSERROR("Could not allocate memory."); | |
195 | return NULL; | |
196 | } | |
197 | ||
198 | /* read capabilities */ | |
199 | snprintf(proc_fn, MAXPATHLEN, "/proc/%d/status", pid); | |
200 | ||
201 | proc_file = fopen(proc_fn, "r"); | |
202 | if (!proc_file) { | |
203 | SYSERROR("Could not open %s", proc_fn); | |
204 | goto out_error; | |
205 | } | |
206 | ||
207 | found = 0; | |
208 | while (getline(&line, &line_bufsz, proc_file) != -1) { | |
209 | ret = sscanf(line, "CapBnd: %llx", &info->capability_mask); | |
210 | if (ret != EOF && ret > 0) { | |
211 | found = 1; | |
212 | break; | |
213 | } | |
214 | } | |
215 | ||
f10fad2f | 216 | free(line); |
e0732705 CS |
217 | fclose(proc_file); |
218 | ||
219 | if (!found) { | |
220 | SYSERROR("Could not read capability bounding set from %s", proc_fn); | |
221 | errno = ENOENT; | |
222 | goto out_error; | |
223 | } | |
224 | ||
fe4de9a6 | 225 | info->lsm_label = lsm_process_label_get(pid); |
e0732705 | 226 | |
e0732705 CS |
227 | return info; |
228 | ||
229 | out_error: | |
460a1cf0 | 230 | free(info); |
e0732705 CS |
231 | return NULL; |
232 | } | |
233 | ||
fe4de9a6 DE |
234 | static void lxc_proc_put_context_info(struct lxc_proc_context_info *ctx) |
235 | { | |
f10fad2f | 236 | free(ctx->lsm_label); |
2c4ea790 SH |
237 | if (ctx->container) |
238 | lxc_container_put(ctx->container); | |
fe4de9a6 DE |
239 | free(ctx); |
240 | } | |
241 | ||
74a3920a | 242 | static int lxc_attach_to_ns(pid_t pid, int which) |
99d50954 | 243 | { |
26818618 | 244 | int fd[LXC_NS_MAX]; |
fc763ab7 CS |
245 | int i, j, saved_errno; |
246 | ||
99d50954 | 247 | |
a052913d | 248 | if (access("/proc/self/ns", X_OK)) { |
99d50954 CS |
249 | ERROR("Does this kernel version support 'attach' ?"); |
250 | return -1; | |
251 | } | |
252 | ||
26818618 | 253 | for (i = 0; i < LXC_NS_MAX; i++) { |
fc763ab7 CS |
254 | /* ignore if we are not supposed to attach to that |
255 | * namespace | |
256 | */ | |
26818618 | 257 | if (which != -1 && !(which & ns_info[i].clone_flag)) { |
fc763ab7 CS |
258 | fd[i] = -1; |
259 | continue; | |
260 | } | |
261 | ||
26818618 | 262 | fd[i] = lxc_preserve_ns(pid, ns_info[i].proc_name); |
99d50954 | 263 | if (fd[i] < 0) { |
fc763ab7 CS |
264 | saved_errno = errno; |
265 | ||
266 | /* close all already opened file descriptors before | |
267 | * we return an error, so we don't leak them | |
268 | */ | |
269 | for (j = 0; j < i; j++) | |
270 | close(fd[j]); | |
271 | ||
272 | errno = saved_errno; | |
26818618 | 273 | SYSERROR("failed to open namespace: '%s'.", ns_info[i].proc_name); |
99d50954 CS |
274 | return -1; |
275 | } | |
276 | } | |
277 | ||
26818618 CB |
278 | for (i = 0; i < LXC_NS_MAX; i++) { |
279 | if (fd[i] < 0) | |
280 | continue; | |
281 | ||
282 | if (setns(fd[i], 0) < 0) { | |
fc763ab7 CS |
283 | saved_errno = errno; |
284 | ||
26818618 | 285 | for (j = i; j < LXC_NS_MAX; j++) |
fc763ab7 CS |
286 | close(fd[j]); |
287 | ||
288 | errno = saved_errno; | |
26818618 | 289 | SYSERROR("Failed to attach to namespace \"%s\".", ns_info[i].proc_name); |
99d50954 CS |
290 | return -1; |
291 | } | |
292 | ||
26818618 CB |
293 | DEBUG("Attached to namespace \"%s\".", ns_info[i].proc_name); |
294 | ||
99d50954 CS |
295 | close(fd[i]); |
296 | } | |
297 | ||
298 | return 0; | |
299 | } | |
300 | ||
74a3920a | 301 | static int lxc_attach_remount_sys_proc(void) |
7a0b0b56 CS |
302 | { |
303 | int ret; | |
304 | ||
305 | ret = unshare(CLONE_NEWNS); | |
306 | if (ret < 0) { | |
307 | SYSERROR("failed to unshare mount namespace"); | |
308 | return -1; | |
309 | } | |
310 | ||
2c6f3fc9 SH |
311 | if (detect_shared_rootfs()) { |
312 | if (mount(NULL, "/", NULL, MS_SLAVE|MS_REC, NULL)) { | |
313 | SYSERROR("Failed to make / rslave"); | |
314 | ERROR("Continuing..."); | |
315 | } | |
316 | } | |
317 | ||
7a0b0b56 CS |
318 | /* assume /proc is always mounted, so remount it */ |
319 | ret = umount2("/proc", MNT_DETACH); | |
320 | if (ret < 0) { | |
321 | SYSERROR("failed to unmount /proc"); | |
322 | return -1; | |
323 | } | |
324 | ||
325 | ret = mount("none", "/proc", "proc", 0, NULL); | |
326 | if (ret < 0) { | |
327 | SYSERROR("failed to remount /proc"); | |
328 | return -1; | |
329 | } | |
330 | ||
331 | /* try to umount /sys - if it's not a mount point, | |
332 | * we'll get EINVAL, then we ignore it because it | |
333 | * may not have been mounted in the first place | |
334 | */ | |
335 | ret = umount2("/sys", MNT_DETACH); | |
336 | if (ret < 0 && errno != EINVAL) { | |
337 | SYSERROR("failed to unmount /sys"); | |
338 | return -1; | |
339 | } else if (ret == 0) { | |
340 | /* remount it */ | |
341 | ret = mount("none", "/sys", "sysfs", 0, NULL); | |
342 | if (ret < 0) { | |
343 | SYSERROR("failed to remount /sys"); | |
344 | return -1; | |
345 | } | |
346 | } | |
347 | ||
348 | return 0; | |
349 | } | |
350 | ||
74a3920a | 351 | static int lxc_attach_drop_privs(struct lxc_proc_context_info *ctx) |
e0732705 CS |
352 | { |
353 | int last_cap = lxc_caps_last_cap(); | |
354 | int cap; | |
355 | ||
356 | for (cap = 0; cap <= last_cap; cap++) { | |
357 | if (ctx->capability_mask & (1LL << cap)) | |
358 | continue; | |
359 | ||
360 | if (prctl(PR_CAPBSET_DROP, cap, 0, 0, 0)) { | |
361 | SYSERROR("failed to remove capability id %d", cap); | |
362 | return -1; | |
363 | } | |
364 | } | |
365 | ||
366 | return 0; | |
367 | } | |
905022f7 | 368 | |
74a3920a | 369 | static int lxc_attach_set_environment(enum lxc_attach_env_policy_t policy, char** extra_env, char** extra_keep) |
b3a39ba6 | 370 | { |
799f96fd | 371 | if (policy == LXC_ATTACH_CLEAR_ENV) { |
3d5e9f48 | 372 | char **extra_keep_store = NULL; |
3d5e9f48 CS |
373 | int path_kept = 0; |
374 | ||
375 | if (extra_keep) { | |
376 | size_t count, i; | |
377 | ||
378 | for (count = 0; extra_keep[count]; count++); | |
379 | ||
380 | extra_keep_store = calloc(count, sizeof(char *)); | |
381 | if (!extra_keep_store) { | |
382 | SYSERROR("failed to allocate memory for storing current " | |
383 | "environment variable values that will be kept"); | |
384 | return -1; | |
385 | } | |
386 | for (i = 0; i < count; i++) { | |
387 | char *v = getenv(extra_keep[i]); | |
388 | if (v) { | |
389 | extra_keep_store[i] = strdup(v); | |
390 | if (!extra_keep_store[i]) { | |
391 | SYSERROR("failed to allocate memory for storing current " | |
392 | "environment variable values that will be kept"); | |
393 | while (i > 0) | |
394 | free(extra_keep_store[--i]); | |
395 | free(extra_keep_store); | |
396 | return -1; | |
397 | } | |
398 | if (strcmp(extra_keep[i], "PATH") == 0) | |
399 | path_kept = 1; | |
400 | } | |
401 | /* calloc sets entire array to zero, so we don't | |
402 | * need an else */ | |
403 | } | |
404 | } | |
405 | ||
799f96fd | 406 | if (clearenv()) { |
a9cab7e3 | 407 | char **p; |
799f96fd | 408 | SYSERROR("failed to clear environment"); |
a9cab7e3 CS |
409 | if (extra_keep_store) { |
410 | for (p = extra_keep_store; *p; p++) | |
411 | free(*p); | |
412 | free(extra_keep_store); | |
413 | } | |
3d5e9f48 CS |
414 | return -1; |
415 | } | |
416 | ||
417 | if (extra_keep_store) { | |
418 | size_t i; | |
419 | for (i = 0; extra_keep[i]; i++) { | |
acd4922e SG |
420 | if (extra_keep_store[i]) { |
421 | if (setenv(extra_keep[i], extra_keep_store[i], 1) < 0) | |
422 | SYSERROR("Unable to set environment variable"); | |
423 | } | |
3d5e9f48 CS |
424 | free(extra_keep_store[i]); |
425 | } | |
426 | free(extra_keep_store); | |
427 | } | |
428 | ||
429 | /* always set a default path; shells and execlp tend | |
430 | * to be fine without it, but there is a disturbing | |
431 | * number of C programs out there that just assume | |
432 | * that getenv("PATH") is never NULL and then die a | |
433 | * painful segfault death. */ | |
cfa70b88 | 434 | if (!path_kept) |
511a6936 | 435 | setenv("PATH", "/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin", 1); |
b3a39ba6 DW |
436 | } |
437 | ||
438 | if (putenv("container=lxc")) { | |
439 | SYSERROR("failed to set environment variable"); | |
440 | return -1; | |
441 | } | |
442 | ||
3d5e9f48 CS |
443 | /* set extra environment variables */ |
444 | if (extra_env) { | |
445 | for (; *extra_env; extra_env++) { | |
446 | /* duplicate the string, just to be on | |
447 | * the safe side, because putenv does not | |
448 | * do it for us */ | |
449 | char *p = strdup(*extra_env); | |
450 | /* we just assume the user knows what they | |
451 | * are doing, so we don't do any checks */ | |
452 | if (!p) { | |
453 | SYSERROR("failed to allocate memory for additional environment " | |
454 | "variables"); | |
455 | return -1; | |
456 | } | |
457 | putenv(p); | |
458 | } | |
459 | } | |
460 | ||
b3a39ba6 DW |
461 | return 0; |
462 | } | |
463 | ||
74a3920a | 464 | static char *lxc_attach_getpwshell(uid_t uid) |
905022f7 CS |
465 | { |
466 | /* local variables */ | |
467 | pid_t pid; | |
468 | int pipes[2]; | |
469 | int ret; | |
470 | int fd; | |
471 | char *result = NULL; | |
472 | ||
473 | /* we need to fork off a process that runs the | |
474 | * getent program, and we need to capture its | |
475 | * output, so we use a pipe for that purpose | |
476 | */ | |
477 | ret = pipe(pipes); | |
478 | if (ret < 0) | |
479 | return NULL; | |
480 | ||
481 | pid = fork(); | |
482 | if (pid < 0) { | |
483 | close(pipes[0]); | |
484 | close(pipes[1]); | |
485 | return NULL; | |
486 | } | |
487 | ||
488 | if (pid) { | |
489 | /* parent process */ | |
490 | FILE *pipe_f; | |
491 | char *line = NULL; | |
492 | size_t line_bufsz = 0; | |
493 | int found = 0; | |
494 | int status; | |
495 | ||
496 | close(pipes[1]); | |
497 | ||
498 | pipe_f = fdopen(pipes[0], "r"); | |
499 | while (getline(&line, &line_bufsz, pipe_f) != -1) { | |
500 | char *token; | |
501 | char *saveptr = NULL; | |
502 | long value; | |
503 | char *endptr = NULL; | |
504 | int i; | |
505 | ||
506 | /* if we already found something, just continue | |
507 | * to read until the pipe doesn't deliver any more | |
508 | * data, but don't modify the existing data | |
509 | * structure | |
510 | */ | |
511 | if (found) | |
512 | continue; | |
513 | ||
514 | /* trim line on the right hand side */ | |
bbb8a488 | 515 | for (i = strlen(line); i > 0 && (line[i - 1] == '\n' || line[i - 1] == '\r'); --i) |
905022f7 CS |
516 | line[i - 1] = '\0'; |
517 | ||
518 | /* split into tokens: first user name */ | |
519 | token = strtok_r(line, ":", &saveptr); | |
520 | if (!token) | |
521 | continue; | |
522 | /* next: dummy password field */ | |
523 | token = strtok_r(NULL, ":", &saveptr); | |
524 | if (!token) | |
525 | continue; | |
526 | /* next: user id */ | |
527 | token = strtok_r(NULL, ":", &saveptr); | |
528 | value = token ? strtol(token, &endptr, 10) : 0; | |
529 | if (!token || !endptr || *endptr || value == LONG_MIN || value == LONG_MAX) | |
530 | continue; | |
531 | /* dummy sanity check: user id matches */ | |
532 | if ((uid_t) value != uid) | |
533 | continue; | |
534 | /* skip fields: gid, gecos, dir, go to next field 'shell' */ | |
535 | for (i = 0; i < 4; i++) { | |
536 | token = strtok_r(NULL, ":", &saveptr); | |
537 | if (!token) | |
538 | break; | |
539 | } | |
540 | if (!token) | |
541 | continue; | |
f10fad2f | 542 | free(result); |
905022f7 CS |
543 | result = strdup(token); |
544 | ||
545 | /* sanity check that there are no fields after that */ | |
546 | token = strtok_r(NULL, ":", &saveptr); | |
547 | if (token) | |
548 | continue; | |
549 | ||
550 | found = 1; | |
551 | } | |
552 | ||
553 | free(line); | |
554 | fclose(pipe_f); | |
555 | again: | |
556 | if (waitpid(pid, &status, 0) < 0) { | |
557 | if (errno == EINTR) | |
558 | goto again; | |
559 | return NULL; | |
560 | } | |
561 | ||
562 | /* some sanity checks: if anything even hinted at going | |
563 | * wrong: we can't be sure we have a valid result, so | |
564 | * we assume we don't | |
565 | */ | |
566 | ||
567 | if (!WIFEXITED(status)) | |
568 | return NULL; | |
569 | ||
570 | if (WEXITSTATUS(status) != 0) | |
571 | return NULL; | |
572 | ||
573 | if (!found) | |
574 | return NULL; | |
575 | ||
576 | return result; | |
577 | } else { | |
578 | /* child process */ | |
579 | char uid_buf[32]; | |
580 | char *arguments[] = { | |
581 | "getent", | |
582 | "passwd", | |
583 | uid_buf, | |
584 | NULL | |
585 | }; | |
586 | ||
587 | close(pipes[0]); | |
588 | ||
589 | /* we want to capture stdout */ | |
590 | dup2(pipes[1], 1); | |
591 | close(pipes[1]); | |
592 | ||
593 | /* get rid of stdin/stderr, so we try to associate it | |
594 | * with /dev/null | |
595 | */ | |
596 | fd = open("/dev/null", O_RDWR); | |
597 | if (fd < 0) { | |
598 | close(0); | |
599 | close(2); | |
600 | } else { | |
601 | dup2(fd, 0); | |
602 | dup2(fd, 2); | |
603 | close(fd); | |
604 | } | |
605 | ||
606 | /* finish argument list */ | |
607 | ret = snprintf(uid_buf, sizeof(uid_buf), "%ld", (long) uid); | |
608 | if (ret <= 0) | |
609 | exit(-1); | |
610 | ||
611 | /* try to run getent program */ | |
612 | (void) execvp("getent", arguments); | |
613 | exit(-1); | |
614 | } | |
615 | } | |
cb3e61fa | 616 | |
74a3920a | 617 | static void lxc_attach_get_init_uidgid(uid_t* init_uid, gid_t* init_gid) |
cb3e61fa CS |
618 | { |
619 | FILE *proc_file; | |
620 | char proc_fn[MAXPATHLEN]; | |
621 | char *line = NULL; | |
622 | size_t line_bufsz = 0; | |
623 | int ret; | |
624 | long value = -1; | |
625 | uid_t uid = (uid_t)-1; | |
626 | gid_t gid = (gid_t)-1; | |
627 | ||
628 | /* read capabilities */ | |
629 | snprintf(proc_fn, MAXPATHLEN, "/proc/%d/status", 1); | |
630 | ||
631 | proc_file = fopen(proc_fn, "r"); | |
632 | if (!proc_file) | |
633 | return; | |
634 | ||
635 | while (getline(&line, &line_bufsz, proc_file) != -1) { | |
636 | /* format is: real, effective, saved set user, fs | |
637 | * we only care about real uid | |
638 | */ | |
639 | ret = sscanf(line, "Uid: %ld", &value); | |
640 | if (ret != EOF && ret > 0) { | |
641 | uid = (uid_t) value; | |
642 | } else { | |
643 | ret = sscanf(line, "Gid: %ld", &value); | |
644 | if (ret != EOF && ret > 0) | |
645 | gid = (gid_t) value; | |
646 | } | |
647 | if (uid != (uid_t)-1 && gid != (gid_t)-1) | |
648 | break; | |
649 | } | |
650 | ||
651 | fclose(proc_file); | |
652 | free(line); | |
653 | ||
654 | /* only override arguments if we found something */ | |
655 | if (uid != (uid_t)-1) | |
656 | *init_uid = uid; | |
657 | if (gid != (gid_t)-1) | |
658 | *init_gid = gid; | |
659 | ||
660 | /* TODO: we should also parse supplementary groups and use | |
661 | * setgroups() to set them */ | |
662 | } | |
9c4693b8 CS |
663 | |
664 | struct attach_clone_payload { | |
665 | int ipc_socket; | |
666 | lxc_attach_options_t* options; | |
667 | struct lxc_proc_context_info* init_ctx; | |
668 | lxc_attach_exec_t exec_function; | |
669 | void* exec_payload; | |
670 | }; | |
671 | ||
672 | static int attach_child_main(void* data); | |
673 | ||
674 | /* help the optimizer along if it doesn't know that exit always exits */ | |
5dcc1ca6 | 675 | #define rexit(c) do { int __c = (c); _exit(__c); return __c; } while(0) |
9c4693b8 CS |
676 | |
677 | /* define default options if no options are supplied by the user */ | |
678 | static lxc_attach_options_t attach_static_default_options = LXC_ATTACH_OPTIONS_DEFAULT; | |
679 | ||
bd4307f0 | 680 | static bool fetch_seccomp(struct lxc_container *c, |
ff07d7bb | 681 | lxc_attach_options_t *options) |
2c4ea790 | 682 | { |
bd7b4e28 | 683 | char *path; |
2eef2bda | 684 | |
bd4307f0 CB |
685 | if (!(options->namespaces & CLONE_NEWNS) || !(options->attach_flags & LXC_ATTACH_LSM)) { |
686 | free(c->lxc_conf->seccomp); | |
687 | c->lxc_conf->seccomp = NULL; | |
2c4ea790 | 688 | return true; |
bd4307f0 | 689 | } |
bd7b4e28 | 690 | |
2e812c16 | 691 | /* Remove current setting. */ |
bd7b4e28 | 692 | if (!c->set_config_item(c, "lxc.seccomp", "")) { |
2c4ea790 | 693 | return false; |
bd7b4e28 SG |
694 | } |
695 | ||
696 | /* Fetch the current profile path over the cmd interface */ | |
697 | path = c->get_running_config_item(c, "lxc.seccomp"); | |
698 | if (!path) { | |
bd4307f0 | 699 | INFO("Failed to get running config item for lxc.seccomp."); |
bd7b4e28 SG |
700 | return true; |
701 | } | |
702 | ||
703 | /* Copy the value into the new lxc_conf */ | |
704 | if (!c->set_config_item(c, "lxc.seccomp", path)) { | |
705 | free(path); | |
706 | return false; | |
707 | } | |
708 | free(path); | |
709 | ||
710 | /* Attempt to parse the resulting config */ | |
2c4ea790 | 711 | if (lxc_read_seccomp_config(c->lxc_conf) < 0) { |
442f5c0f | 712 | ERROR("Error reading seccomp policy"); |
2c4ea790 SH |
713 | return false; |
714 | } | |
715 | ||
2e812c16 CB |
716 | INFO("Retrieved seccomp policy."); |
717 | return true; | |
718 | } | |
719 | ||
bd4307f0 | 720 | static bool no_new_privs(struct lxc_container *c, |
2e812c16 CB |
721 | lxc_attach_options_t *options) |
722 | { | |
2e812c16 CB |
723 | char *val; |
724 | ||
2e812c16 CB |
725 | /* Remove current setting. */ |
726 | if (!c->set_config_item(c, "lxc.no_new_privs", "")) { | |
727 | return false; | |
728 | } | |
729 | ||
730 | /* Retrieve currently active setting. */ | |
731 | val = c->get_running_config_item(c, "lxc.no_new_privs"); | |
732 | if (!val) { | |
733 | INFO("Failed to get running config item for lxc.no_new_privs."); | |
734 | return false; | |
735 | } | |
736 | ||
737 | /* Set currently active setting. */ | |
738 | if (!c->set_config_item(c, "lxc.no_new_privs", val)) { | |
739 | free(val); | |
740 | return false; | |
741 | } | |
742 | free(val); | |
743 | ||
2c4ea790 SH |
744 | return true; |
745 | } | |
746 | ||
9b8e3c96 SH |
747 | static signed long get_personality(const char *name, const char *lxcpath) |
748 | { | |
0d7cf7e9 | 749 | char *p = lxc_cmd_get_config_item(name, "lxc.arch", lxcpath); |
9b8e3c96 SH |
750 | signed long ret; |
751 | ||
752 | if (!p) | |
753 | return -1; | |
754 | ret = lxc_config_parse_arch(p); | |
755 | free(p); | |
756 | return ret; | |
757 | } | |
758 | ||
9c4693b8 CS |
759 | int lxc_attach(const char* name, const char* lxcpath, lxc_attach_exec_t exec_function, void* exec_payload, lxc_attach_options_t* options, pid_t* attached_process) |
760 | { | |
761 | int ret, status; | |
f4364484 | 762 | pid_t init_pid, pid, attached_pid, expected; |
9c4693b8 CS |
763 | struct lxc_proc_context_info *init_ctx; |
764 | char* cwd; | |
765 | char* new_cwd; | |
766 | int ipc_sockets[2]; | |
9b8e3c96 | 767 | signed long personality; |
9c4693b8 CS |
768 | |
769 | if (!options) | |
770 | options = &attach_static_default_options; | |
771 | ||
772 | init_pid = lxc_cmd_get_init_pid(name, lxcpath); | |
773 | if (init_pid < 0) { | |
774 | ERROR("failed to get the init pid"); | |
775 | return -1; | |
776 | } | |
777 | ||
778 | init_ctx = lxc_proc_get_context_info(init_pid); | |
779 | if (!init_ctx) { | |
780 | ERROR("failed to get context of the init process, pid = %ld", (long)init_pid); | |
781 | return -1; | |
782 | } | |
783 | ||
9b8e3c96 SH |
784 | personality = get_personality(name, lxcpath); |
785 | if (init_ctx->personality < 0) { | |
786 | ERROR("Failed to get personality of the container"); | |
787 | lxc_proc_put_context_info(init_ctx); | |
788 | return -1; | |
789 | } | |
790 | init_ctx->personality = personality; | |
791 | ||
ff07d7bb CB |
792 | init_ctx->container = lxc_container_new(name, lxcpath); |
793 | if (!init_ctx->container) | |
794 | return -1; | |
795 | ||
bd4307f0 | 796 | if (!fetch_seccomp(init_ctx->container, options)) |
2c4ea790 SH |
797 | WARN("Failed to get seccomp policy"); |
798 | ||
bd4307f0 | 799 | if (!no_new_privs(init_ctx->container, options)) |
2e812c16 CB |
800 | WARN("Could not determine whether PR_SET_NO_NEW_PRIVS is set."); |
801 | ||
9c4693b8 CS |
802 | cwd = getcwd(NULL, 0); |
803 | ||
804 | /* determine which namespaces the container was created with | |
805 | * by asking lxc-start, if necessary | |
806 | */ | |
807 | if (options->namespaces == -1) { | |
808 | options->namespaces = lxc_cmd_get_clone_flags(name, lxcpath); | |
809 | /* call failed */ | |
810 | if (options->namespaces == -1) { | |
811 | ERROR("failed to automatically determine the " | |
812 | "namespaces which the container unshared"); | |
813 | free(cwd); | |
fe4de9a6 | 814 | lxc_proc_put_context_info(init_ctx); |
9c4693b8 CS |
815 | return -1; |
816 | } | |
817 | } | |
818 | ||
819 | /* create a socket pair for IPC communication; set SOCK_CLOEXEC in order | |
820 | * to make sure we don't irritate other threads that want to fork+exec away | |
821 | * | |
822 | * IMPORTANT: if the initial process is multithreaded and another call | |
823 | * just fork()s away without exec'ing directly after, the socket fd will | |
824 | * exist in the forked process from the other thread and any close() in | |
825 | * our own child process will not really cause the socket to close properly, | |
826 | * potentiall causing the parent to hang. | |
827 | * | |
828 | * For this reason, while IPC is still active, we have to use shutdown() | |
829 | * if the child exits prematurely in order to signal that the socket | |
830 | * is closed and cannot assume that the child exiting will automatically | |
831 | * do that. | |
832 | * | |
833 | * IPC mechanism: (X is receiver) | |
834 | * initial process intermediate attached | |
835 | * X <--- send pid of | |
836 | * attached proc, | |
837 | * then exit | |
838 | * send 0 ------------------------------------> X | |
839 | * [do initialization] | |
840 | * X <------------------------------------ send 1 | |
841 | * [add to cgroup, ...] | |
842 | * send 2 ------------------------------------> X | |
81f466d0 CB |
843 | * [set LXC_ATTACH_NO_NEW_PRIVS] |
844 | * X <------------------------------------ send 3 | |
845 | * [open LSM label fd] | |
846 | * send 4 ------------------------------------> X | |
847 | * [set LSM label] | |
9c4693b8 CS |
848 | * close socket close socket |
849 | * run program | |
850 | */ | |
851 | ret = socketpair(PF_LOCAL, SOCK_STREAM | SOCK_CLOEXEC, 0, ipc_sockets); | |
852 | if (ret < 0) { | |
853 | SYSERROR("could not set up required IPC mechanism for attaching"); | |
854 | free(cwd); | |
fe4de9a6 | 855 | lxc_proc_put_context_info(init_ctx); |
9c4693b8 CS |
856 | return -1; |
857 | } | |
858 | ||
859 | /* create intermediate subprocess, three reasons: | |
860 | * 1. runs all pthread_atfork handlers and the | |
861 | * child will no longer be threaded | |
862 | * (we can't properly setns() in a threaded process) | |
863 | * 2. we can't setns() in the child itself, since | |
864 | * we want to make sure we are properly attached to | |
865 | * the pidns | |
866 | * 3. also, the initial thread has to put the attached | |
867 | * process into the cgroup, which we can only do if | |
868 | * we didn't already setns() (otherwise, user | |
869 | * namespaces will hate us) | |
870 | */ | |
871 | pid = fork(); | |
872 | ||
873 | if (pid < 0) { | |
874 | SYSERROR("failed to create first subprocess"); | |
875 | free(cwd); | |
fe4de9a6 | 876 | lxc_proc_put_context_info(init_ctx); |
9c4693b8 CS |
877 | return -1; |
878 | } | |
879 | ||
880 | if (pid) { | |
81f466d0 | 881 | int procfd = -1; |
9c4693b8 | 882 | pid_t to_cleanup_pid = pid; |
9c4693b8 | 883 | |
ec64264d | 884 | /* initial thread, we close the socket that is for the |
9c4693b8 CS |
885 | * subprocesses |
886 | */ | |
887 | close(ipc_sockets[1]); | |
888 | free(cwd); | |
889 | ||
f4364484 SG |
890 | /* attach to cgroup, if requested */ |
891 | if (options->attach_flags & LXC_ATTACH_MOVE_TO_CGROUP) { | |
4fb3cba5 | 892 | if (!cgroup_attach(name, lxcpath, pid)) |
f4364484 | 893 | goto cleanup_error; |
f4364484 SG |
894 | } |
895 | ||
81f466d0 CB |
896 | /* Open /proc before setns() to the containers namespace so we |
897 | * don't rely on any information from inside the container. | |
898 | */ | |
899 | procfd = open("/proc", O_DIRECTORY | O_RDONLY | O_CLOEXEC); | |
900 | if (procfd < 0) { | |
901 | SYSERROR("Unable to open /proc."); | |
902 | goto cleanup_error; | |
903 | } | |
904 | ||
f4364484 SG |
905 | /* Let the child process know to go ahead */ |
906 | status = 0; | |
907 | ret = lxc_write_nointr(ipc_sockets[0], &status, sizeof(status)); | |
908 | if (ret <= 0) { | |
909 | ERROR("error using IPC to notify attached process for initialization (0)"); | |
910 | goto cleanup_error; | |
911 | } | |
912 | ||
9c4693b8 CS |
913 | /* get pid from intermediate process */ |
914 | ret = lxc_read_nointr_expect(ipc_sockets[0], &attached_pid, sizeof(attached_pid), NULL); | |
915 | if (ret <= 0) { | |
916 | if (ret != 0) | |
917 | ERROR("error using IPC to receive pid of attached process"); | |
918 | goto cleanup_error; | |
919 | } | |
920 | ||
2eef2bda | 921 | /* ignore SIGKILL (CTRL-C) and SIGQUIT (CTRL-\) - issue #313 */ |
62183f1a SH |
922 | if (options->stdin_fd == 0) { |
923 | signal(SIGINT, SIG_IGN); | |
924 | signal(SIGQUIT, SIG_IGN); | |
925 | } | |
2eef2bda | 926 | |
9c4693b8 CS |
927 | /* reap intermediate process */ |
928 | ret = wait_for_pid(pid); | |
929 | if (ret < 0) | |
930 | goto cleanup_error; | |
931 | ||
932 | /* we will always have to reap the grandchild now */ | |
933 | to_cleanup_pid = attached_pid; | |
934 | ||
935 | /* tell attached process it may start initializing */ | |
936 | status = 0; | |
937 | ret = lxc_write_nointr(ipc_sockets[0], &status, sizeof(status)); | |
938 | if (ret <= 0) { | |
939 | ERROR("error using IPC to notify attached process for initialization (0)"); | |
940 | goto cleanup_error; | |
941 | } | |
942 | ||
943 | /* wait for the attached process to finish initializing */ | |
944 | expected = 1; | |
945 | ret = lxc_read_nointr_expect(ipc_sockets[0], &status, sizeof(status), &expected); | |
946 | if (ret <= 0) { | |
947 | if (ret != 0) | |
81f466d0 CB |
948 | ERROR("error using IPC to receive notification " |
949 | "from attached process (1)"); | |
9c4693b8 CS |
950 | goto cleanup_error; |
951 | } | |
952 | ||
9c4693b8 CS |
953 | /* tell attached process we're done */ |
954 | status = 2; | |
955 | ret = lxc_write_nointr(ipc_sockets[0], &status, sizeof(status)); | |
956 | if (ret <= 0) { | |
81f466d0 CB |
957 | ERROR("Error using IPC to notify attached process for " |
958 | "initialization (2): %s.", strerror(errno)); | |
9c4693b8 CS |
959 | goto cleanup_error; |
960 | } | |
961 | ||
81f466d0 CB |
962 | /* Wait for the (grand)child to tell us that it's ready to set |
963 | * up its LSM labels. | |
964 | */ | |
965 | expected = 3; | |
966 | ret = lxc_read_nointr_expect(ipc_sockets[0], &status, sizeof(status), &expected); | |
967 | if (ret <= 0) { | |
968 | ERROR("Error using IPC for the child to tell us to open LSM fd (3): %s.", | |
969 | strerror(errno)); | |
970 | goto cleanup_error; | |
971 | } | |
972 | ||
973 | /* Open LSM fd and send it to child. */ | |
974 | if ((options->namespaces & CLONE_NEWNS) && (options->attach_flags & LXC_ATTACH_LSM) && init_ctx->lsm_label) { | |
975 | int on_exec, labelfd; | |
976 | on_exec = options->attach_flags & LXC_ATTACH_LSM_EXEC ? 1 : 0; | |
977 | /* Open fd for the LSM security module. */ | |
978 | labelfd = lsm_openat(procfd, attached_pid, on_exec); | |
979 | if (labelfd < 0) | |
980 | goto cleanup_error; | |
981 | ||
982 | /* Send child fd of the LSM security module to write to. */ | |
983 | ret = lxc_abstract_unix_send_fd(ipc_sockets[0], labelfd, NULL, 0); | |
984 | if (ret <= 0) { | |
985 | ERROR("Error using IPC to send child LSM fd (4): %s.", | |
986 | strerror(errno)); | |
987 | goto cleanup_error; | |
988 | } | |
989 | } | |
990 | ||
9c4693b8 CS |
991 | /* now shut down communication with child, we're done */ |
992 | shutdown(ipc_sockets[0], SHUT_RDWR); | |
993 | close(ipc_sockets[0]); | |
fe4de9a6 | 994 | lxc_proc_put_context_info(init_ctx); |
9c4693b8 CS |
995 | |
996 | /* we're done, the child process should now execute whatever | |
997 | * it is that the user requested. The parent can now track it | |
998 | * with waitpid() or similar. | |
999 | */ | |
1000 | ||
1001 | *attached_process = attached_pid; | |
1002 | return 0; | |
1003 | ||
1004 | cleanup_error: | |
1005 | /* first shut down the socket, then wait for the pid, | |
1006 | * otherwise the pid we're waiting for may never exit | |
1007 | */ | |
81f466d0 CB |
1008 | if (procfd >= 0) |
1009 | close(procfd); | |
9c4693b8 CS |
1010 | shutdown(ipc_sockets[0], SHUT_RDWR); |
1011 | close(ipc_sockets[0]); | |
1012 | if (to_cleanup_pid) | |
1013 | (void) wait_for_pid(to_cleanup_pid); | |
fe4de9a6 | 1014 | lxc_proc_put_context_info(init_ctx); |
9c4693b8 CS |
1015 | return -1; |
1016 | } | |
1017 | ||
1018 | /* first subprocess begins here, we close the socket that is for the | |
1019 | * initial thread | |
1020 | */ | |
1021 | close(ipc_sockets[0]); | |
1022 | ||
f4364484 SG |
1023 | /* Wait for the parent to have setup cgroups */ |
1024 | expected = 0; | |
1025 | status = -1; | |
1026 | ret = lxc_read_nointr_expect(ipc_sockets[1], &status, sizeof(status), &expected); | |
1027 | if (ret <= 0) { | |
1028 | ERROR("error communicating with child process"); | |
1029 | shutdown(ipc_sockets[1], SHUT_RDWR); | |
1030 | rexit(-1); | |
1031 | } | |
1032 | ||
dac862c0 | 1033 | if ((options->attach_flags & LXC_ATTACH_MOVE_TO_CGROUP) && cgns_supported()) |
1034 | options->namespaces |= CLONE_NEWCGROUP; | |
fe3c80af | 1035 | |
9c4693b8 CS |
1036 | /* attach now, create another subprocess later, since pid namespaces |
1037 | * only really affect the children of the current process | |
1038 | */ | |
1039 | ret = lxc_attach_to_ns(init_pid, options->namespaces); | |
1040 | if (ret < 0) { | |
1041 | ERROR("failed to enter the namespace"); | |
1042 | shutdown(ipc_sockets[1], SHUT_RDWR); | |
1043 | rexit(-1); | |
1044 | } | |
1045 | ||
1046 | /* attach succeeded, try to cwd */ | |
1047 | if (options->initial_cwd) | |
1048 | new_cwd = options->initial_cwd; | |
1049 | else | |
1050 | new_cwd = cwd; | |
1051 | ret = chdir(new_cwd); | |
1052 | if (ret < 0) | |
1053 | WARN("could not change directory to '%s'", new_cwd); | |
1054 | free(cwd); | |
1055 | ||
1056 | /* now create the real child process */ | |
1057 | { | |
1058 | struct attach_clone_payload payload = { | |
1059 | .ipc_socket = ipc_sockets[1], | |
1060 | .options = options, | |
1061 | .init_ctx = init_ctx, | |
1062 | .exec_function = exec_function, | |
5c3fcae7 | 1063 | .exec_payload = exec_payload, |
9c4693b8 CS |
1064 | }; |
1065 | /* We use clone_parent here to make this subprocess a direct child of | |
1066 | * the initial process. Then this intermediate process can exit and | |
1067 | * the parent can directly track the attached process. | |
1068 | */ | |
1069 | pid = lxc_clone(attach_child_main, &payload, CLONE_PARENT); | |
1070 | } | |
1071 | ||
1072 | /* shouldn't happen, clone() should always return positive pid */ | |
1073 | if (pid <= 0) { | |
1074 | SYSERROR("failed to create subprocess"); | |
1075 | shutdown(ipc_sockets[1], SHUT_RDWR); | |
1076 | rexit(-1); | |
1077 | } | |
1078 | ||
1079 | /* tell grandparent the pid of the pid of the newly created child */ | |
1080 | ret = lxc_write_nointr(ipc_sockets[1], &pid, sizeof(pid)); | |
1081 | if (ret != sizeof(pid)) { | |
1082 | /* if this really happens here, this is very unfortunate, since the | |
1083 | * parent will not know the pid of the attached process and will | |
1084 | * not be able to wait for it (and we won't either due to CLONE_PARENT) | |
1085 | * so the parent won't be able to reap it and the attached process | |
1086 | * will remain a zombie | |
1087 | */ | |
1088 | ERROR("error using IPC to notify main process of pid of the attached process"); | |
1089 | shutdown(ipc_sockets[1], SHUT_RDWR); | |
1090 | rexit(-1); | |
1091 | } | |
1092 | ||
1093 | /* the rest is in the hands of the initial and the attached process */ | |
1094 | rexit(0); | |
1095 | } | |
1096 | ||
74a3920a | 1097 | static int attach_child_main(void* data) |
9c4693b8 CS |
1098 | { |
1099 | struct attach_clone_payload* payload = (struct attach_clone_payload*)data; | |
1100 | int ipc_socket = payload->ipc_socket; | |
1101 | lxc_attach_options_t* options = payload->options; | |
1102 | struct lxc_proc_context_info* init_ctx = payload->init_ctx; | |
1a2e58cf | 1103 | #if HAVE_SYS_PERSONALITY_H |
9c4693b8 | 1104 | long new_personality; |
1a2e58cf | 1105 | #endif |
9c4693b8 CS |
1106 | int ret; |
1107 | int status; | |
1108 | int expected; | |
1109 | long flags; | |
1110 | int fd; | |
81f466d0 | 1111 | int lsm_labelfd; |
9c4693b8 CS |
1112 | uid_t new_uid; |
1113 | gid_t new_gid; | |
1114 | ||
1115 | /* wait for the initial thread to signal us that it's ready | |
1116 | * for us to start initializing | |
1117 | */ | |
1118 | expected = 0; | |
1119 | status = -1; | |
1120 | ret = lxc_read_nointr_expect(ipc_socket, &status, sizeof(status), &expected); | |
1121 | if (ret <= 0) { | |
81f466d0 | 1122 | ERROR("Error using IPC to receive notification from initial process (0): %s.", strerror(errno)); |
9c4693b8 CS |
1123 | shutdown(ipc_socket, SHUT_RDWR); |
1124 | rexit(-1); | |
1125 | } | |
1126 | ||
9c4693b8 CS |
1127 | /* A description of the purpose of this functionality is |
1128 | * provided in the lxc-attach(1) manual page. We have to | |
1129 | * remount here and not in the parent process, otherwise | |
1130 | * /proc may not properly reflect the new pid namespace. | |
1131 | */ | |
1132 | if (!(options->namespaces & CLONE_NEWNS) && (options->attach_flags & LXC_ATTACH_REMOUNT_PROC_SYS)) { | |
1133 | ret = lxc_attach_remount_sys_proc(); | |
1134 | if (ret < 0) { | |
1135 | shutdown(ipc_socket, SHUT_RDWR); | |
1136 | rexit(-1); | |
1137 | } | |
1138 | } | |
1139 | ||
1140 | /* now perform additional attachments*/ | |
1141 | #if HAVE_SYS_PERSONALITY_H | |
1142 | if (options->personality < 0) | |
1143 | new_personality = init_ctx->personality; | |
1144 | else | |
1145 | new_personality = options->personality; | |
1146 | ||
1147 | if (options->attach_flags & LXC_ATTACH_SET_PERSONALITY) { | |
1148 | ret = personality(new_personality); | |
1149 | if (ret < 0) { | |
1150 | SYSERROR("could not ensure correct architecture"); | |
1151 | shutdown(ipc_socket, SHUT_RDWR); | |
1152 | rexit(-1); | |
1153 | } | |
1154 | } | |
1155 | #endif | |
1156 | ||
1157 | if (options->attach_flags & LXC_ATTACH_DROP_CAPABILITIES) { | |
1158 | ret = lxc_attach_drop_privs(init_ctx); | |
1159 | if (ret < 0) { | |
1160 | ERROR("could not drop privileges"); | |
1161 | shutdown(ipc_socket, SHUT_RDWR); | |
1162 | rexit(-1); | |
1163 | } | |
1164 | } | |
1165 | ||
1166 | /* always set the environment (specify (LXC_ATTACH_KEEP_ENV, NULL, NULL) if you want this to be a no-op) */ | |
1167 | ret = lxc_attach_set_environment(options->env_policy, options->extra_env_vars, options->extra_keep_env); | |
1168 | if (ret < 0) { | |
1169 | ERROR("could not set initial environment for attached process"); | |
1170 | shutdown(ipc_socket, SHUT_RDWR); | |
1171 | rexit(-1); | |
1172 | } | |
1173 | ||
1174 | /* set user / group id */ | |
1175 | new_uid = 0; | |
1176 | new_gid = 0; | |
1177 | /* ignore errors, we will fall back to root in that case | |
1178 | * (/proc was not mounted etc.) | |
1179 | */ | |
1180 | if (options->namespaces & CLONE_NEWUSER) | |
1181 | lxc_attach_get_init_uidgid(&new_uid, &new_gid); | |
1182 | ||
1183 | if (options->uid != (uid_t)-1) | |
1184 | new_uid = options->uid; | |
1185 | if (options->gid != (gid_t)-1) | |
1186 | new_gid = options->gid; | |
1187 | ||
82e28fe0 | 1188 | /* setup the control tty */ |
d3b63011 | 1189 | if (options->stdin_fd && isatty(options->stdin_fd)) { |
82e28fe0 SG |
1190 | if (setsid() < 0) { |
1191 | SYSERROR("unable to setsid"); | |
1192 | shutdown(ipc_socket, SHUT_RDWR); | |
1193 | rexit(-1); | |
1194 | } | |
1195 | ||
1196 | if (ioctl(options->stdin_fd, TIOCSCTTY, (char *)NULL) < 0) { | |
1197 | SYSERROR("unable to TIOCSTTY"); | |
1198 | shutdown(ipc_socket, SHUT_RDWR); | |
1199 | rexit(-1); | |
1200 | } | |
1201 | } | |
1202 | ||
9c4693b8 | 1203 | /* try to set the uid/gid combination */ |
c476bdce SH |
1204 | if ((new_gid != 0 || options->namespaces & CLONE_NEWUSER)) { |
1205 | if (setgid(new_gid) || setgroups(0, NULL)) { | |
1206 | SYSERROR("switching to container gid"); | |
1207 | shutdown(ipc_socket, SHUT_RDWR); | |
1208 | rexit(-1); | |
1209 | } | |
9c4693b8 CS |
1210 | } |
1211 | if ((new_uid != 0 || options->namespaces & CLONE_NEWUSER) && setuid(new_uid)) { | |
1212 | SYSERROR("switching to container uid"); | |
1213 | shutdown(ipc_socket, SHUT_RDWR); | |
1214 | rexit(-1); | |
1215 | } | |
1216 | ||
1217 | /* tell initial process it may now put us into the cgroups */ | |
1218 | status = 1; | |
1219 | ret = lxc_write_nointr(ipc_socket, &status, sizeof(status)); | |
1220 | if (ret != sizeof(status)) { | |
81f466d0 | 1221 | ERROR("Error using IPC to notify initial process for initialization (1): %s.", strerror(errno)); |
9c4693b8 CS |
1222 | shutdown(ipc_socket, SHUT_RDWR); |
1223 | rexit(-1); | |
1224 | } | |
1225 | ||
1226 | /* wait for the initial thread to signal us that it has done | |
1227 | * everything for us when it comes to cgroups etc. | |
1228 | */ | |
1229 | expected = 2; | |
1230 | status = -1; | |
1231 | ret = lxc_read_nointr_expect(ipc_socket, &status, sizeof(status), &expected); | |
1232 | if (ret <= 0) { | |
81f466d0 CB |
1233 | ERROR("Error using IPC to receive message from initial process " |
1234 | "that it is done pre-initializing (2): %s", | |
1235 | strerror(errno)); | |
9c4693b8 CS |
1236 | shutdown(ipc_socket, SHUT_RDWR); |
1237 | rexit(-1); | |
1238 | } | |
1239 | ||
2e812c16 CB |
1240 | if ((init_ctx->container && init_ctx->container->lxc_conf && |
1241 | init_ctx->container->lxc_conf->no_new_privs) || | |
1242 | (options->attach_flags & LXC_ATTACH_NO_NEW_PRIVS)) { | |
1243 | if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) < 0) { | |
1244 | SYSERROR("PR_SET_NO_NEW_PRIVS could not be set. " | |
1245 | "Process can use execve() gainable " | |
1246 | "privileges."); | |
81f466d0 | 1247 | shutdown(ipc_socket, SHUT_RDWR); |
2e812c16 CB |
1248 | rexit(-1); |
1249 | } | |
1250 | INFO("PR_SET_NO_NEW_PRIVS is set. Process cannot use execve() " | |
1251 | "gainable privileges."); | |
1252 | } | |
1253 | ||
81f466d0 CB |
1254 | /* Tell the (grand)parent to send us LSM label fd. */ |
1255 | status = 3; | |
1256 | ret = lxc_write_nointr(ipc_socket, &status, sizeof(status)); | |
1257 | if (ret <= 0) { | |
1258 | ERROR("Error using IPC to tell parent to set up LSM labels (3): %s.", strerror(errno)); | |
1259 | shutdown(ipc_socket, SHUT_RDWR); | |
1260 | rexit(-1); | |
1261 | } | |
1262 | ||
5c3fcae7 | 1263 | if ((options->namespaces & CLONE_NEWNS) && (options->attach_flags & LXC_ATTACH_LSM) && init_ctx->lsm_label) { |
72863294 | 1264 | int on_exec; |
81f466d0 CB |
1265 | /* Receive fd for LSM security module. */ |
1266 | ret = lxc_abstract_unix_recv_fd(ipc_socket, &lsm_labelfd, NULL, 0); | |
1267 | if (ret <= 0) { | |
1268 | ERROR("Error using IPC for parent to tell us LSM label fd (4): %s.", strerror(errno)); | |
1269 | shutdown(ipc_socket, SHUT_RDWR); | |
1270 | rexit(-1); | |
1271 | } | |
72863294 | 1272 | |
81f466d0 | 1273 | /* Change into our new LSM profile. */ |
72863294 | 1274 | on_exec = options->attach_flags & LXC_ATTACH_LSM_EXEC ? 1 : 0; |
81f466d0 CB |
1275 | if (lsm_set_label_at(lsm_labelfd, on_exec, init_ctx->lsm_label) < 0) { |
1276 | SYSERROR("Failed to set LSM label."); | |
1277 | shutdown(ipc_socket, SHUT_RDWR); | |
1278 | close(lsm_labelfd); | |
72863294 DE |
1279 | rexit(-1); |
1280 | } | |
81f466d0 | 1281 | close(lsm_labelfd); |
72863294 | 1282 | } |
81f466d0 | 1283 | |
2c4ea790 | 1284 | if (init_ctx->container && init_ctx->container->lxc_conf && |
bd4307f0 CB |
1285 | init_ctx->container->lxc_conf->seccomp && |
1286 | (lxc_seccomp_load(init_ctx->container->lxc_conf) != 0)) { | |
2c4ea790 | 1287 | ERROR("Loading seccomp policy"); |
81f466d0 | 1288 | shutdown(ipc_socket, SHUT_RDWR); |
2c4ea790 SH |
1289 | rexit(-1); |
1290 | } | |
81f466d0 CB |
1291 | |
1292 | shutdown(ipc_socket, SHUT_RDWR); | |
1293 | close(ipc_socket); | |
fe4de9a6 | 1294 | lxc_proc_put_context_info(init_ctx); |
9c4693b8 CS |
1295 | |
1296 | /* The following is done after the communication socket is | |
1297 | * shut down. That way, all errors that might (though | |
1298 | * unlikely) occur up until this point will have their messages | |
1299 | * printed to the original stderr (if logging is so configured) | |
1300 | * and not the fd the user supplied, if any. | |
1301 | */ | |
1302 | ||
1303 | /* fd handling for stdin, stdout and stderr; | |
1304 | * ignore errors here, user may want to make sure | |
1305 | * the fds are closed, for example */ | |
1306 | if (options->stdin_fd >= 0 && options->stdin_fd != 0) | |
1307 | dup2(options->stdin_fd, 0); | |
1308 | if (options->stdout_fd >= 0 && options->stdout_fd != 1) | |
1309 | dup2(options->stdout_fd, 1); | |
1310 | if (options->stderr_fd >= 0 && options->stderr_fd != 2) | |
1311 | dup2(options->stderr_fd, 2); | |
1312 | ||
1313 | /* close the old fds */ | |
1314 | if (options->stdin_fd > 2) | |
1315 | close(options->stdin_fd); | |
1316 | if (options->stdout_fd > 2) | |
1317 | close(options->stdout_fd); | |
1318 | if (options->stderr_fd > 2) | |
1319 | close(options->stderr_fd); | |
1320 | ||
1321 | /* try to remove CLOEXEC flag from stdin/stdout/stderr, | |
1322 | * but also here, ignore errors */ | |
1323 | for (fd = 0; fd <= 2; fd++) { | |
1324 | flags = fcntl(fd, F_GETFL); | |
1325 | if (flags < 0) | |
1326 | continue; | |
26818618 CB |
1327 | if (flags & FD_CLOEXEC) |
1328 | if (fcntl(fd, F_SETFL, flags & ~FD_CLOEXEC) < 0) | |
71b2940d | 1329 | SYSERROR("Unable to clear CLOEXEC from fd"); |
9c4693b8 CS |
1330 | } |
1331 | ||
1332 | /* we're done, so we can now do whatever the user intended us to do */ | |
1333 | rexit(payload->exec_function(payload->exec_payload)); | |
1334 | } | |
1335 | ||
1336 | int lxc_attach_run_command(void* payload) | |
1337 | { | |
1338 | lxc_attach_command_t* cmd = (lxc_attach_command_t*)payload; | |
1339 | ||
1340 | execvp(cmd->program, cmd->argv); | |
1341 | SYSERROR("failed to exec '%s'", cmd->program); | |
1342 | return -1; | |
1343 | } | |
1344 | ||
1345 | int lxc_attach_run_shell(void* payload) | |
1346 | { | |
1347 | uid_t uid; | |
1348 | struct passwd *passwd; | |
1349 | char *user_shell; | |
1350 | ||
1351 | /* ignore payload parameter */ | |
1352 | (void)payload; | |
1353 | ||
1354 | uid = getuid(); | |
1355 | passwd = getpwuid(uid); | |
1356 | ||
1357 | /* this probably happens because of incompatible nss | |
1358 | * implementations in host and container (remember, this | |
1359 | * code is still using the host's glibc but our mount | |
1360 | * namespace is in the container) | |
1361 | * we may try to get the information by spawning a | |
1362 | * [getent passwd uid] process and parsing the result | |
1363 | */ | |
1364 | if (!passwd) | |
1365 | user_shell = lxc_attach_getpwshell(uid); | |
1366 | else | |
1367 | user_shell = passwd->pw_shell; | |
1368 | ||
1369 | if (user_shell) | |
acf47e1b | 1370 | execlp(user_shell, user_shell, (char *)NULL); |
9c4693b8 CS |
1371 | |
1372 | /* executed if either no passwd entry or execvp fails, | |
1373 | * we will fall back on /bin/sh as a default shell | |
1374 | */ | |
acf47e1b | 1375 | execlp("/bin/sh", "/bin/sh", (char *)NULL); |
9c4693b8 CS |
1376 | SYSERROR("failed to exec shell"); |
1377 | return -1; | |
1378 | } |