]>
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> | |
29 | #include <errno.h> | |
30 | #include <fcntl.h> | |
c476bdce | 31 | #include <grp.h> |
e0732705 CS |
32 | #include <sys/param.h> |
33 | #include <sys/prctl.h> | |
7a0b0b56 | 34 | #include <sys/mount.h> |
5ec27989 | 35 | #include <sys/socket.h> |
1ba0013f | 36 | #include <sys/syscall.h> |
905022f7 | 37 | #include <sys/wait.h> |
910bb4fa | 38 | #include <linux/unistd.h> |
905022f7 | 39 | #include <pwd.h> |
e0732705 CS |
40 | |
41 | #if !HAVE_DECL_PR_CAPBSET_DROP | |
42 | #define PR_CAPBSET_DROP 24 | |
43 | #endif | |
44 | ||
45 | #include "namespace.h" | |
46 | #include "log.h" | |
47 | #include "attach.h" | |
48 | #include "caps.h" | |
e0732705 | 49 | #include "config.h" |
6a44839f | 50 | #include "utils.h" |
9c4693b8 CS |
51 | #include "commands.h" |
52 | #include "cgroup.h" | |
025ed0f3 | 53 | #include "lxclock.h" |
fe4de9a6 | 54 | #include "lsm/lsm.h" |
9c4693b8 CS |
55 | |
56 | #if HAVE_SYS_PERSONALITY_H | |
57 | #include <sys/personality.h> | |
58 | #endif | |
e0732705 | 59 | |
a3da2f3b SG |
60 | #ifndef SOCK_CLOEXEC |
61 | # define SOCK_CLOEXEC 02000000 | |
62 | #endif | |
63 | ||
e0732705 CS |
64 | lxc_log_define(lxc_attach, lxc); |
65 | ||
74a3920a | 66 | static struct lxc_proc_context_info *lxc_proc_get_context_info(pid_t pid) |
e0732705 CS |
67 | { |
68 | struct lxc_proc_context_info *info = calloc(1, sizeof(*info)); | |
69 | FILE *proc_file; | |
70 | char proc_fn[MAXPATHLEN]; | |
460a1cf0 | 71 | char *line = NULL; |
e0732705 | 72 | size_t line_bufsz = 0; |
460a1cf0 | 73 | int ret, found; |
e0732705 CS |
74 | |
75 | if (!info) { | |
76 | SYSERROR("Could not allocate memory."); | |
77 | return NULL; | |
78 | } | |
79 | ||
80 | /* read capabilities */ | |
81 | snprintf(proc_fn, MAXPATHLEN, "/proc/%d/status", pid); | |
82 | ||
83 | proc_file = fopen(proc_fn, "r"); | |
84 | if (!proc_file) { | |
85 | SYSERROR("Could not open %s", proc_fn); | |
86 | goto out_error; | |
87 | } | |
88 | ||
89 | found = 0; | |
90 | while (getline(&line, &line_bufsz, proc_file) != -1) { | |
91 | ret = sscanf(line, "CapBnd: %llx", &info->capability_mask); | |
92 | if (ret != EOF && ret > 0) { | |
93 | found = 1; | |
94 | break; | |
95 | } | |
96 | } | |
97 | ||
fa9ac567 SH |
98 | if (line) |
99 | free(line); | |
e0732705 CS |
100 | fclose(proc_file); |
101 | ||
102 | if (!found) { | |
103 | SYSERROR("Could not read capability bounding set from %s", proc_fn); | |
104 | errno = ENOENT; | |
105 | goto out_error; | |
106 | } | |
107 | ||
108 | /* read personality */ | |
109 | snprintf(proc_fn, MAXPATHLEN, "/proc/%d/personality", pid); | |
110 | ||
111 | proc_file = fopen(proc_fn, "r"); | |
112 | if (!proc_file) { | |
113 | SYSERROR("Could not open %s", proc_fn); | |
114 | goto out_error; | |
115 | } | |
116 | ||
117 | ret = fscanf(proc_file, "%lx", &info->personality); | |
118 | fclose(proc_file); | |
119 | ||
120 | if (ret == EOF || ret == 0) { | |
121 | SYSERROR("Could not read personality from %s", proc_fn); | |
122 | errno = ENOENT; | |
123 | goto out_error; | |
124 | } | |
fe4de9a6 | 125 | info->lsm_label = lsm_process_label_get(pid); |
e0732705 | 126 | |
e0732705 CS |
127 | return info; |
128 | ||
129 | out_error: | |
460a1cf0 | 130 | free(info); |
e0732705 CS |
131 | return NULL; |
132 | } | |
133 | ||
fe4de9a6 DE |
134 | static void lxc_proc_put_context_info(struct lxc_proc_context_info *ctx) |
135 | { | |
136 | if (ctx->lsm_label) | |
137 | free(ctx->lsm_label); | |
138 | free(ctx); | |
139 | } | |
140 | ||
74a3920a | 141 | static int lxc_attach_to_ns(pid_t pid, int which) |
99d50954 CS |
142 | { |
143 | char path[MAXPATHLEN]; | |
fc763ab7 CS |
144 | /* according to <http://article.gmane.org/gmane.linux.kernel.containers.lxc.devel/1429>, |
145 | * the file for user namepsaces in /proc/$pid/ns will be called | |
146 | * 'user' once the kernel supports it | |
147 | */ | |
f4364484 | 148 | static char *ns[] = { "user", "mnt", "pid", "uts", "ipc", "net" }; |
fc763ab7 | 149 | static int flags[] = { |
f4364484 SG |
150 | CLONE_NEWUSER, CLONE_NEWNS, CLONE_NEWPID, CLONE_NEWUTS, CLONE_NEWIPC, |
151 | CLONE_NEWNET | |
fc763ab7 CS |
152 | }; |
153 | static const int size = sizeof(ns) / sizeof(char *); | |
99d50954 | 154 | int fd[size]; |
fc763ab7 CS |
155 | int i, j, saved_errno; |
156 | ||
99d50954 CS |
157 | |
158 | snprintf(path, MAXPATHLEN, "/proc/%d/ns", pid); | |
159 | if (access(path, X_OK)) { | |
160 | ERROR("Does this kernel version support 'attach' ?"); | |
161 | return -1; | |
162 | } | |
163 | ||
164 | for (i = 0; i < size; i++) { | |
fc763ab7 CS |
165 | /* ignore if we are not supposed to attach to that |
166 | * namespace | |
167 | */ | |
168 | if (which != -1 && !(which & flags[i])) { | |
169 | fd[i] = -1; | |
170 | continue; | |
171 | } | |
172 | ||
99d50954 | 173 | snprintf(path, MAXPATHLEN, "/proc/%d/ns/%s", pid, ns[i]); |
9c4693b8 | 174 | fd[i] = open(path, O_RDONLY | O_CLOEXEC); |
99d50954 | 175 | if (fd[i] < 0) { |
fc763ab7 CS |
176 | saved_errno = errno; |
177 | ||
178 | /* close all already opened file descriptors before | |
179 | * we return an error, so we don't leak them | |
180 | */ | |
181 | for (j = 0; j < i; j++) | |
182 | close(fd[j]); | |
183 | ||
184 | errno = saved_errno; | |
99d50954 CS |
185 | SYSERROR("failed to open '%s'", path); |
186 | return -1; | |
187 | } | |
188 | } | |
189 | ||
190 | for (i = 0; i < size; i++) { | |
fc763ab7 CS |
191 | if (fd[i] >= 0 && setns(fd[i], 0) != 0) { |
192 | saved_errno = errno; | |
193 | ||
194 | for (j = i; j < size; j++) | |
195 | close(fd[j]); | |
196 | ||
197 | errno = saved_errno; | |
99d50954 CS |
198 | SYSERROR("failed to set namespace '%s'", ns[i]); |
199 | return -1; | |
200 | } | |
201 | ||
202 | close(fd[i]); | |
203 | } | |
204 | ||
205 | return 0; | |
206 | } | |
207 | ||
74a3920a | 208 | static int lxc_attach_remount_sys_proc(void) |
7a0b0b56 CS |
209 | { |
210 | int ret; | |
211 | ||
212 | ret = unshare(CLONE_NEWNS); | |
213 | if (ret < 0) { | |
214 | SYSERROR("failed to unshare mount namespace"); | |
215 | return -1; | |
216 | } | |
217 | ||
218 | /* assume /proc is always mounted, so remount it */ | |
219 | ret = umount2("/proc", MNT_DETACH); | |
220 | if (ret < 0) { | |
221 | SYSERROR("failed to unmount /proc"); | |
222 | return -1; | |
223 | } | |
224 | ||
225 | ret = mount("none", "/proc", "proc", 0, NULL); | |
226 | if (ret < 0) { | |
227 | SYSERROR("failed to remount /proc"); | |
228 | return -1; | |
229 | } | |
230 | ||
231 | /* try to umount /sys - if it's not a mount point, | |
232 | * we'll get EINVAL, then we ignore it because it | |
233 | * may not have been mounted in the first place | |
234 | */ | |
235 | ret = umount2("/sys", MNT_DETACH); | |
236 | if (ret < 0 && errno != EINVAL) { | |
237 | SYSERROR("failed to unmount /sys"); | |
238 | return -1; | |
239 | } else if (ret == 0) { | |
240 | /* remount it */ | |
241 | ret = mount("none", "/sys", "sysfs", 0, NULL); | |
242 | if (ret < 0) { | |
243 | SYSERROR("failed to remount /sys"); | |
244 | return -1; | |
245 | } | |
246 | } | |
247 | ||
248 | return 0; | |
249 | } | |
250 | ||
74a3920a | 251 | static int lxc_attach_drop_privs(struct lxc_proc_context_info *ctx) |
e0732705 CS |
252 | { |
253 | int last_cap = lxc_caps_last_cap(); | |
254 | int cap; | |
255 | ||
256 | for (cap = 0; cap <= last_cap; cap++) { | |
257 | if (ctx->capability_mask & (1LL << cap)) | |
258 | continue; | |
259 | ||
260 | if (prctl(PR_CAPBSET_DROP, cap, 0, 0, 0)) { | |
261 | SYSERROR("failed to remove capability id %d", cap); | |
262 | return -1; | |
263 | } | |
264 | } | |
265 | ||
266 | return 0; | |
267 | } | |
905022f7 | 268 | |
74a3920a | 269 | static int lxc_attach_set_environment(enum lxc_attach_env_policy_t policy, char** extra_env, char** extra_keep) |
b3a39ba6 | 270 | { |
799f96fd | 271 | if (policy == LXC_ATTACH_CLEAR_ENV) { |
3d5e9f48 | 272 | char **extra_keep_store = NULL; |
3d5e9f48 CS |
273 | int path_kept = 0; |
274 | ||
275 | if (extra_keep) { | |
276 | size_t count, i; | |
277 | ||
278 | for (count = 0; extra_keep[count]; count++); | |
279 | ||
280 | extra_keep_store = calloc(count, sizeof(char *)); | |
281 | if (!extra_keep_store) { | |
282 | SYSERROR("failed to allocate memory for storing current " | |
283 | "environment variable values that will be kept"); | |
284 | return -1; | |
285 | } | |
286 | for (i = 0; i < count; i++) { | |
287 | char *v = getenv(extra_keep[i]); | |
288 | if (v) { | |
289 | extra_keep_store[i] = strdup(v); | |
290 | if (!extra_keep_store[i]) { | |
291 | SYSERROR("failed to allocate memory for storing current " | |
292 | "environment variable values that will be kept"); | |
293 | while (i > 0) | |
294 | free(extra_keep_store[--i]); | |
295 | free(extra_keep_store); | |
296 | return -1; | |
297 | } | |
298 | if (strcmp(extra_keep[i], "PATH") == 0) | |
299 | path_kept = 1; | |
300 | } | |
301 | /* calloc sets entire array to zero, so we don't | |
302 | * need an else */ | |
303 | } | |
304 | } | |
305 | ||
799f96fd | 306 | if (clearenv()) { |
a9cab7e3 | 307 | char **p; |
799f96fd | 308 | SYSERROR("failed to clear environment"); |
a9cab7e3 CS |
309 | if (extra_keep_store) { |
310 | for (p = extra_keep_store; *p; p++) | |
311 | free(*p); | |
312 | free(extra_keep_store); | |
313 | } | |
3d5e9f48 CS |
314 | return -1; |
315 | } | |
316 | ||
317 | if (extra_keep_store) { | |
318 | size_t i; | |
319 | for (i = 0; extra_keep[i]; i++) { | |
320 | if (extra_keep_store[i]) | |
321 | setenv(extra_keep[i], extra_keep_store[i], 1); | |
322 | free(extra_keep_store[i]); | |
323 | } | |
324 | free(extra_keep_store); | |
325 | } | |
326 | ||
327 | /* always set a default path; shells and execlp tend | |
328 | * to be fine without it, but there is a disturbing | |
329 | * number of C programs out there that just assume | |
330 | * that getenv("PATH") is never NULL and then die a | |
331 | * painful segfault death. */ | |
332 | if (!path_kept) { | |
511a6936 SG |
333 | #ifdef HAVE_CONFSTR |
334 | size_t n; | |
335 | char *path_env; | |
336 | ||
3d5e9f48 CS |
337 | n = confstr(_CS_PATH, NULL, 0); |
338 | path_env = malloc(n); | |
339 | if (path_env) { | |
340 | confstr(_CS_PATH, path_env, n); | |
341 | setenv("PATH", path_env, 1); | |
342 | free(path_env); | |
343 | } | |
344 | /* don't error out, this is just an extra service */ | |
511a6936 SG |
345 | #else |
346 | setenv("PATH", "/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin", 1); | |
347 | #endif | |
799f96fd | 348 | } |
b3a39ba6 DW |
349 | } |
350 | ||
351 | if (putenv("container=lxc")) { | |
352 | SYSERROR("failed to set environment variable"); | |
353 | return -1; | |
354 | } | |
355 | ||
3d5e9f48 CS |
356 | /* set extra environment variables */ |
357 | if (extra_env) { | |
358 | for (; *extra_env; extra_env++) { | |
359 | /* duplicate the string, just to be on | |
360 | * the safe side, because putenv does not | |
361 | * do it for us */ | |
362 | char *p = strdup(*extra_env); | |
363 | /* we just assume the user knows what they | |
364 | * are doing, so we don't do any checks */ | |
365 | if (!p) { | |
366 | SYSERROR("failed to allocate memory for additional environment " | |
367 | "variables"); | |
368 | return -1; | |
369 | } | |
370 | putenv(p); | |
371 | } | |
372 | } | |
373 | ||
b3a39ba6 DW |
374 | return 0; |
375 | } | |
376 | ||
74a3920a | 377 | static char *lxc_attach_getpwshell(uid_t uid) |
905022f7 CS |
378 | { |
379 | /* local variables */ | |
380 | pid_t pid; | |
381 | int pipes[2]; | |
382 | int ret; | |
383 | int fd; | |
384 | char *result = NULL; | |
385 | ||
386 | /* we need to fork off a process that runs the | |
387 | * getent program, and we need to capture its | |
388 | * output, so we use a pipe for that purpose | |
389 | */ | |
390 | ret = pipe(pipes); | |
391 | if (ret < 0) | |
392 | return NULL; | |
393 | ||
394 | pid = fork(); | |
395 | if (pid < 0) { | |
396 | close(pipes[0]); | |
397 | close(pipes[1]); | |
398 | return NULL; | |
399 | } | |
400 | ||
401 | if (pid) { | |
402 | /* parent process */ | |
403 | FILE *pipe_f; | |
404 | char *line = NULL; | |
405 | size_t line_bufsz = 0; | |
406 | int found = 0; | |
407 | int status; | |
408 | ||
409 | close(pipes[1]); | |
410 | ||
411 | pipe_f = fdopen(pipes[0], "r"); | |
412 | while (getline(&line, &line_bufsz, pipe_f) != -1) { | |
413 | char *token; | |
414 | char *saveptr = NULL; | |
415 | long value; | |
416 | char *endptr = NULL; | |
417 | int i; | |
418 | ||
419 | /* if we already found something, just continue | |
420 | * to read until the pipe doesn't deliver any more | |
421 | * data, but don't modify the existing data | |
422 | * structure | |
423 | */ | |
424 | if (found) | |
425 | continue; | |
426 | ||
427 | /* trim line on the right hand side */ | |
bbb8a488 | 428 | for (i = strlen(line); i > 0 && (line[i - 1] == '\n' || line[i - 1] == '\r'); --i) |
905022f7 CS |
429 | line[i - 1] = '\0'; |
430 | ||
431 | /* split into tokens: first user name */ | |
432 | token = strtok_r(line, ":", &saveptr); | |
433 | if (!token) | |
434 | continue; | |
435 | /* next: dummy password field */ | |
436 | token = strtok_r(NULL, ":", &saveptr); | |
437 | if (!token) | |
438 | continue; | |
439 | /* next: user id */ | |
440 | token = strtok_r(NULL, ":", &saveptr); | |
441 | value = token ? strtol(token, &endptr, 10) : 0; | |
442 | if (!token || !endptr || *endptr || value == LONG_MIN || value == LONG_MAX) | |
443 | continue; | |
444 | /* dummy sanity check: user id matches */ | |
445 | if ((uid_t) value != uid) | |
446 | continue; | |
447 | /* skip fields: gid, gecos, dir, go to next field 'shell' */ | |
448 | for (i = 0; i < 4; i++) { | |
449 | token = strtok_r(NULL, ":", &saveptr); | |
450 | if (!token) | |
451 | break; | |
452 | } | |
453 | if (!token) | |
454 | continue; | |
53a54099 SH |
455 | if (result) |
456 | free(result); | |
905022f7 CS |
457 | result = strdup(token); |
458 | ||
459 | /* sanity check that there are no fields after that */ | |
460 | token = strtok_r(NULL, ":", &saveptr); | |
461 | if (token) | |
462 | continue; | |
463 | ||
464 | found = 1; | |
465 | } | |
466 | ||
467 | free(line); | |
468 | fclose(pipe_f); | |
469 | again: | |
470 | if (waitpid(pid, &status, 0) < 0) { | |
471 | if (errno == EINTR) | |
472 | goto again; | |
473 | return NULL; | |
474 | } | |
475 | ||
476 | /* some sanity checks: if anything even hinted at going | |
477 | * wrong: we can't be sure we have a valid result, so | |
478 | * we assume we don't | |
479 | */ | |
480 | ||
481 | if (!WIFEXITED(status)) | |
482 | return NULL; | |
483 | ||
484 | if (WEXITSTATUS(status) != 0) | |
485 | return NULL; | |
486 | ||
487 | if (!found) | |
488 | return NULL; | |
489 | ||
490 | return result; | |
491 | } else { | |
492 | /* child process */ | |
493 | char uid_buf[32]; | |
494 | char *arguments[] = { | |
495 | "getent", | |
496 | "passwd", | |
497 | uid_buf, | |
498 | NULL | |
499 | }; | |
500 | ||
501 | close(pipes[0]); | |
502 | ||
503 | /* we want to capture stdout */ | |
504 | dup2(pipes[1], 1); | |
505 | close(pipes[1]); | |
506 | ||
507 | /* get rid of stdin/stderr, so we try to associate it | |
508 | * with /dev/null | |
509 | */ | |
510 | fd = open("/dev/null", O_RDWR); | |
511 | if (fd < 0) { | |
512 | close(0); | |
513 | close(2); | |
514 | } else { | |
515 | dup2(fd, 0); | |
516 | dup2(fd, 2); | |
517 | close(fd); | |
518 | } | |
519 | ||
520 | /* finish argument list */ | |
521 | ret = snprintf(uid_buf, sizeof(uid_buf), "%ld", (long) uid); | |
522 | if (ret <= 0) | |
523 | exit(-1); | |
524 | ||
525 | /* try to run getent program */ | |
526 | (void) execvp("getent", arguments); | |
527 | exit(-1); | |
528 | } | |
529 | } | |
cb3e61fa | 530 | |
74a3920a | 531 | static void lxc_attach_get_init_uidgid(uid_t* init_uid, gid_t* init_gid) |
cb3e61fa CS |
532 | { |
533 | FILE *proc_file; | |
534 | char proc_fn[MAXPATHLEN]; | |
535 | char *line = NULL; | |
536 | size_t line_bufsz = 0; | |
537 | int ret; | |
538 | long value = -1; | |
539 | uid_t uid = (uid_t)-1; | |
540 | gid_t gid = (gid_t)-1; | |
541 | ||
542 | /* read capabilities */ | |
543 | snprintf(proc_fn, MAXPATHLEN, "/proc/%d/status", 1); | |
544 | ||
545 | proc_file = fopen(proc_fn, "r"); | |
546 | if (!proc_file) | |
547 | return; | |
548 | ||
549 | while (getline(&line, &line_bufsz, proc_file) != -1) { | |
550 | /* format is: real, effective, saved set user, fs | |
551 | * we only care about real uid | |
552 | */ | |
553 | ret = sscanf(line, "Uid: %ld", &value); | |
554 | if (ret != EOF && ret > 0) { | |
555 | uid = (uid_t) value; | |
556 | } else { | |
557 | ret = sscanf(line, "Gid: %ld", &value); | |
558 | if (ret != EOF && ret > 0) | |
559 | gid = (gid_t) value; | |
560 | } | |
561 | if (uid != (uid_t)-1 && gid != (gid_t)-1) | |
562 | break; | |
563 | } | |
564 | ||
565 | fclose(proc_file); | |
566 | free(line); | |
567 | ||
568 | /* only override arguments if we found something */ | |
569 | if (uid != (uid_t)-1) | |
570 | *init_uid = uid; | |
571 | if (gid != (gid_t)-1) | |
572 | *init_gid = gid; | |
573 | ||
574 | /* TODO: we should also parse supplementary groups and use | |
575 | * setgroups() to set them */ | |
576 | } | |
9c4693b8 CS |
577 | |
578 | struct attach_clone_payload { | |
579 | int ipc_socket; | |
580 | lxc_attach_options_t* options; | |
581 | struct lxc_proc_context_info* init_ctx; | |
582 | lxc_attach_exec_t exec_function; | |
583 | void* exec_payload; | |
584 | }; | |
585 | ||
586 | static int attach_child_main(void* data); | |
587 | ||
588 | /* help the optimizer along if it doesn't know that exit always exits */ | |
589 | #define rexit(c) do { int __c = (c); exit(__c); return __c; } while(0) | |
590 | ||
591 | /* define default options if no options are supplied by the user */ | |
592 | static lxc_attach_options_t attach_static_default_options = LXC_ATTACH_OPTIONS_DEFAULT; | |
593 | ||
594 | 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) | |
595 | { | |
596 | int ret, status; | |
f4364484 | 597 | pid_t init_pid, pid, attached_pid, expected; |
9c4693b8 CS |
598 | struct lxc_proc_context_info *init_ctx; |
599 | char* cwd; | |
600 | char* new_cwd; | |
601 | int ipc_sockets[2]; | |
602 | ||
603 | if (!options) | |
604 | options = &attach_static_default_options; | |
605 | ||
606 | init_pid = lxc_cmd_get_init_pid(name, lxcpath); | |
607 | if (init_pid < 0) { | |
608 | ERROR("failed to get the init pid"); | |
609 | return -1; | |
610 | } | |
611 | ||
612 | init_ctx = lxc_proc_get_context_info(init_pid); | |
613 | if (!init_ctx) { | |
614 | ERROR("failed to get context of the init process, pid = %ld", (long)init_pid); | |
615 | return -1; | |
616 | } | |
617 | ||
618 | cwd = getcwd(NULL, 0); | |
619 | ||
620 | /* determine which namespaces the container was created with | |
621 | * by asking lxc-start, if necessary | |
622 | */ | |
623 | if (options->namespaces == -1) { | |
624 | options->namespaces = lxc_cmd_get_clone_flags(name, lxcpath); | |
625 | /* call failed */ | |
626 | if (options->namespaces == -1) { | |
627 | ERROR("failed to automatically determine the " | |
628 | "namespaces which the container unshared"); | |
629 | free(cwd); | |
fe4de9a6 | 630 | lxc_proc_put_context_info(init_ctx); |
9c4693b8 CS |
631 | return -1; |
632 | } | |
633 | } | |
634 | ||
635 | /* create a socket pair for IPC communication; set SOCK_CLOEXEC in order | |
636 | * to make sure we don't irritate other threads that want to fork+exec away | |
637 | * | |
638 | * IMPORTANT: if the initial process is multithreaded and another call | |
639 | * just fork()s away without exec'ing directly after, the socket fd will | |
640 | * exist in the forked process from the other thread and any close() in | |
641 | * our own child process will not really cause the socket to close properly, | |
642 | * potentiall causing the parent to hang. | |
643 | * | |
644 | * For this reason, while IPC is still active, we have to use shutdown() | |
645 | * if the child exits prematurely in order to signal that the socket | |
646 | * is closed and cannot assume that the child exiting will automatically | |
647 | * do that. | |
648 | * | |
649 | * IPC mechanism: (X is receiver) | |
650 | * initial process intermediate attached | |
651 | * X <--- send pid of | |
652 | * attached proc, | |
653 | * then exit | |
654 | * send 0 ------------------------------------> X | |
655 | * [do initialization] | |
656 | * X <------------------------------------ send 1 | |
657 | * [add to cgroup, ...] | |
658 | * send 2 ------------------------------------> X | |
659 | * close socket close socket | |
660 | * run program | |
661 | */ | |
662 | ret = socketpair(PF_LOCAL, SOCK_STREAM | SOCK_CLOEXEC, 0, ipc_sockets); | |
663 | if (ret < 0) { | |
664 | SYSERROR("could not set up required IPC mechanism for attaching"); | |
665 | free(cwd); | |
fe4de9a6 | 666 | lxc_proc_put_context_info(init_ctx); |
9c4693b8 CS |
667 | return -1; |
668 | } | |
669 | ||
670 | /* create intermediate subprocess, three reasons: | |
671 | * 1. runs all pthread_atfork handlers and the | |
672 | * child will no longer be threaded | |
673 | * (we can't properly setns() in a threaded process) | |
674 | * 2. we can't setns() in the child itself, since | |
675 | * we want to make sure we are properly attached to | |
676 | * the pidns | |
677 | * 3. also, the initial thread has to put the attached | |
678 | * process into the cgroup, which we can only do if | |
679 | * we didn't already setns() (otherwise, user | |
680 | * namespaces will hate us) | |
681 | */ | |
682 | pid = fork(); | |
683 | ||
684 | if (pid < 0) { | |
685 | SYSERROR("failed to create first subprocess"); | |
686 | free(cwd); | |
fe4de9a6 | 687 | lxc_proc_put_context_info(init_ctx); |
9c4693b8 CS |
688 | return -1; |
689 | } | |
690 | ||
691 | if (pid) { | |
692 | pid_t to_cleanup_pid = pid; | |
9c4693b8 CS |
693 | |
694 | /* inital thread, we close the socket that is for the | |
695 | * subprocesses | |
696 | */ | |
697 | close(ipc_sockets[1]); | |
698 | free(cwd); | |
699 | ||
f4364484 SG |
700 | /* attach to cgroup, if requested */ |
701 | if (options->attach_flags & LXC_ATTACH_MOVE_TO_CGROUP) { | |
4fb3cba5 | 702 | if (!cgroup_attach(name, lxcpath, pid)) |
f4364484 | 703 | goto cleanup_error; |
f4364484 SG |
704 | } |
705 | ||
706 | /* Let the child process know to go ahead */ | |
707 | status = 0; | |
708 | ret = lxc_write_nointr(ipc_sockets[0], &status, sizeof(status)); | |
709 | if (ret <= 0) { | |
710 | ERROR("error using IPC to notify attached process for initialization (0)"); | |
711 | goto cleanup_error; | |
712 | } | |
713 | ||
9c4693b8 CS |
714 | /* get pid from intermediate process */ |
715 | ret = lxc_read_nointr_expect(ipc_sockets[0], &attached_pid, sizeof(attached_pid), NULL); | |
716 | if (ret <= 0) { | |
717 | if (ret != 0) | |
718 | ERROR("error using IPC to receive pid of attached process"); | |
719 | goto cleanup_error; | |
720 | } | |
721 | ||
722 | /* reap intermediate process */ | |
723 | ret = wait_for_pid(pid); | |
724 | if (ret < 0) | |
725 | goto cleanup_error; | |
726 | ||
727 | /* we will always have to reap the grandchild now */ | |
728 | to_cleanup_pid = attached_pid; | |
729 | ||
730 | /* tell attached process it may start initializing */ | |
731 | status = 0; | |
732 | ret = lxc_write_nointr(ipc_sockets[0], &status, sizeof(status)); | |
733 | if (ret <= 0) { | |
734 | ERROR("error using IPC to notify attached process for initialization (0)"); | |
735 | goto cleanup_error; | |
736 | } | |
737 | ||
738 | /* wait for the attached process to finish initializing */ | |
739 | expected = 1; | |
740 | ret = lxc_read_nointr_expect(ipc_sockets[0], &status, sizeof(status), &expected); | |
741 | if (ret <= 0) { | |
742 | if (ret != 0) | |
743 | ERROR("error using IPC to receive notification from attached process (1)"); | |
744 | goto cleanup_error; | |
745 | } | |
746 | ||
9c4693b8 CS |
747 | /* tell attached process we're done */ |
748 | status = 2; | |
749 | ret = lxc_write_nointr(ipc_sockets[0], &status, sizeof(status)); | |
750 | if (ret <= 0) { | |
751 | ERROR("error using IPC to notify attached process for initialization (2)"); | |
752 | goto cleanup_error; | |
753 | } | |
754 | ||
755 | /* now shut down communication with child, we're done */ | |
756 | shutdown(ipc_sockets[0], SHUT_RDWR); | |
757 | close(ipc_sockets[0]); | |
fe4de9a6 | 758 | lxc_proc_put_context_info(init_ctx); |
9c4693b8 CS |
759 | |
760 | /* we're done, the child process should now execute whatever | |
761 | * it is that the user requested. The parent can now track it | |
762 | * with waitpid() or similar. | |
763 | */ | |
764 | ||
765 | *attached_process = attached_pid; | |
766 | return 0; | |
767 | ||
768 | cleanup_error: | |
769 | /* first shut down the socket, then wait for the pid, | |
770 | * otherwise the pid we're waiting for may never exit | |
771 | */ | |
772 | shutdown(ipc_sockets[0], SHUT_RDWR); | |
773 | close(ipc_sockets[0]); | |
774 | if (to_cleanup_pid) | |
775 | (void) wait_for_pid(to_cleanup_pid); | |
fe4de9a6 | 776 | lxc_proc_put_context_info(init_ctx); |
9c4693b8 CS |
777 | return -1; |
778 | } | |
779 | ||
780 | /* first subprocess begins here, we close the socket that is for the | |
781 | * initial thread | |
782 | */ | |
783 | close(ipc_sockets[0]); | |
784 | ||
f4364484 SG |
785 | /* Wait for the parent to have setup cgroups */ |
786 | expected = 0; | |
787 | status = -1; | |
788 | ret = lxc_read_nointr_expect(ipc_sockets[1], &status, sizeof(status), &expected); | |
789 | if (ret <= 0) { | |
790 | ERROR("error communicating with child process"); | |
791 | shutdown(ipc_sockets[1], SHUT_RDWR); | |
792 | rexit(-1); | |
793 | } | |
794 | ||
9c4693b8 CS |
795 | /* attach now, create another subprocess later, since pid namespaces |
796 | * only really affect the children of the current process | |
797 | */ | |
798 | ret = lxc_attach_to_ns(init_pid, options->namespaces); | |
799 | if (ret < 0) { | |
800 | ERROR("failed to enter the namespace"); | |
801 | shutdown(ipc_sockets[1], SHUT_RDWR); | |
802 | rexit(-1); | |
803 | } | |
804 | ||
805 | /* attach succeeded, try to cwd */ | |
806 | if (options->initial_cwd) | |
807 | new_cwd = options->initial_cwd; | |
808 | else | |
809 | new_cwd = cwd; | |
810 | ret = chdir(new_cwd); | |
811 | if (ret < 0) | |
812 | WARN("could not change directory to '%s'", new_cwd); | |
813 | free(cwd); | |
814 | ||
815 | /* now create the real child process */ | |
816 | { | |
817 | struct attach_clone_payload payload = { | |
818 | .ipc_socket = ipc_sockets[1], | |
819 | .options = options, | |
820 | .init_ctx = init_ctx, | |
821 | .exec_function = exec_function, | |
822 | .exec_payload = exec_payload | |
823 | }; | |
824 | /* We use clone_parent here to make this subprocess a direct child of | |
825 | * the initial process. Then this intermediate process can exit and | |
826 | * the parent can directly track the attached process. | |
827 | */ | |
828 | pid = lxc_clone(attach_child_main, &payload, CLONE_PARENT); | |
829 | } | |
830 | ||
831 | /* shouldn't happen, clone() should always return positive pid */ | |
832 | if (pid <= 0) { | |
833 | SYSERROR("failed to create subprocess"); | |
834 | shutdown(ipc_sockets[1], SHUT_RDWR); | |
835 | rexit(-1); | |
836 | } | |
837 | ||
838 | /* tell grandparent the pid of the pid of the newly created child */ | |
839 | ret = lxc_write_nointr(ipc_sockets[1], &pid, sizeof(pid)); | |
840 | if (ret != sizeof(pid)) { | |
841 | /* if this really happens here, this is very unfortunate, since the | |
842 | * parent will not know the pid of the attached process and will | |
843 | * not be able to wait for it (and we won't either due to CLONE_PARENT) | |
844 | * so the parent won't be able to reap it and the attached process | |
845 | * will remain a zombie | |
846 | */ | |
847 | ERROR("error using IPC to notify main process of pid of the attached process"); | |
848 | shutdown(ipc_sockets[1], SHUT_RDWR); | |
849 | rexit(-1); | |
850 | } | |
851 | ||
852 | /* the rest is in the hands of the initial and the attached process */ | |
853 | rexit(0); | |
854 | } | |
855 | ||
74a3920a | 856 | static int attach_child_main(void* data) |
9c4693b8 CS |
857 | { |
858 | struct attach_clone_payload* payload = (struct attach_clone_payload*)data; | |
859 | int ipc_socket = payload->ipc_socket; | |
860 | lxc_attach_options_t* options = payload->options; | |
861 | struct lxc_proc_context_info* init_ctx = payload->init_ctx; | |
1a2e58cf | 862 | #if HAVE_SYS_PERSONALITY_H |
9c4693b8 | 863 | long new_personality; |
1a2e58cf | 864 | #endif |
9c4693b8 CS |
865 | int ret; |
866 | int status; | |
867 | int expected; | |
868 | long flags; | |
869 | int fd; | |
870 | uid_t new_uid; | |
871 | gid_t new_gid; | |
872 | ||
873 | /* wait for the initial thread to signal us that it's ready | |
874 | * for us to start initializing | |
875 | */ | |
876 | expected = 0; | |
877 | status = -1; | |
878 | ret = lxc_read_nointr_expect(ipc_socket, &status, sizeof(status), &expected); | |
879 | if (ret <= 0) { | |
880 | ERROR("error using IPC to receive notification from initial process (0)"); | |
881 | shutdown(ipc_socket, SHUT_RDWR); | |
882 | rexit(-1); | |
883 | } | |
884 | ||
9c4693b8 CS |
885 | /* A description of the purpose of this functionality is |
886 | * provided in the lxc-attach(1) manual page. We have to | |
887 | * remount here and not in the parent process, otherwise | |
888 | * /proc may not properly reflect the new pid namespace. | |
889 | */ | |
890 | if (!(options->namespaces & CLONE_NEWNS) && (options->attach_flags & LXC_ATTACH_REMOUNT_PROC_SYS)) { | |
891 | ret = lxc_attach_remount_sys_proc(); | |
892 | if (ret < 0) { | |
893 | shutdown(ipc_socket, SHUT_RDWR); | |
894 | rexit(-1); | |
895 | } | |
896 | } | |
897 | ||
898 | /* now perform additional attachments*/ | |
899 | #if HAVE_SYS_PERSONALITY_H | |
900 | if (options->personality < 0) | |
901 | new_personality = init_ctx->personality; | |
902 | else | |
903 | new_personality = options->personality; | |
904 | ||
905 | if (options->attach_flags & LXC_ATTACH_SET_PERSONALITY) { | |
906 | ret = personality(new_personality); | |
907 | if (ret < 0) { | |
908 | SYSERROR("could not ensure correct architecture"); | |
909 | shutdown(ipc_socket, SHUT_RDWR); | |
910 | rexit(-1); | |
911 | } | |
912 | } | |
913 | #endif | |
914 | ||
915 | if (options->attach_flags & LXC_ATTACH_DROP_CAPABILITIES) { | |
916 | ret = lxc_attach_drop_privs(init_ctx); | |
917 | if (ret < 0) { | |
918 | ERROR("could not drop privileges"); | |
919 | shutdown(ipc_socket, SHUT_RDWR); | |
920 | rexit(-1); | |
921 | } | |
922 | } | |
923 | ||
924 | /* always set the environment (specify (LXC_ATTACH_KEEP_ENV, NULL, NULL) if you want this to be a no-op) */ | |
925 | ret = lxc_attach_set_environment(options->env_policy, options->extra_env_vars, options->extra_keep_env); | |
926 | if (ret < 0) { | |
927 | ERROR("could not set initial environment for attached process"); | |
928 | shutdown(ipc_socket, SHUT_RDWR); | |
929 | rexit(-1); | |
930 | } | |
931 | ||
932 | /* set user / group id */ | |
933 | new_uid = 0; | |
934 | new_gid = 0; | |
935 | /* ignore errors, we will fall back to root in that case | |
936 | * (/proc was not mounted etc.) | |
937 | */ | |
938 | if (options->namespaces & CLONE_NEWUSER) | |
939 | lxc_attach_get_init_uidgid(&new_uid, &new_gid); | |
940 | ||
941 | if (options->uid != (uid_t)-1) | |
942 | new_uid = options->uid; | |
943 | if (options->gid != (gid_t)-1) | |
944 | new_gid = options->gid; | |
945 | ||
946 | /* try to set the uid/gid combination */ | |
c476bdce SH |
947 | if ((new_gid != 0 || options->namespaces & CLONE_NEWUSER)) { |
948 | if (setgid(new_gid) || setgroups(0, NULL)) { | |
949 | SYSERROR("switching to container gid"); | |
950 | shutdown(ipc_socket, SHUT_RDWR); | |
951 | rexit(-1); | |
952 | } | |
9c4693b8 CS |
953 | } |
954 | if ((new_uid != 0 || options->namespaces & CLONE_NEWUSER) && setuid(new_uid)) { | |
955 | SYSERROR("switching to container uid"); | |
956 | shutdown(ipc_socket, SHUT_RDWR); | |
957 | rexit(-1); | |
958 | } | |
959 | ||
960 | /* tell initial process it may now put us into the cgroups */ | |
961 | status = 1; | |
962 | ret = lxc_write_nointr(ipc_socket, &status, sizeof(status)); | |
963 | if (ret != sizeof(status)) { | |
964 | ERROR("error using IPC to notify initial process for initialization (1)"); | |
965 | shutdown(ipc_socket, SHUT_RDWR); | |
966 | rexit(-1); | |
967 | } | |
968 | ||
969 | /* wait for the initial thread to signal us that it has done | |
970 | * everything for us when it comes to cgroups etc. | |
971 | */ | |
972 | expected = 2; | |
973 | status = -1; | |
974 | ret = lxc_read_nointr_expect(ipc_socket, &status, sizeof(status), &expected); | |
975 | if (ret <= 0) { | |
976 | ERROR("error using IPC to receive final notification from initial process (2)"); | |
977 | shutdown(ipc_socket, SHUT_RDWR); | |
978 | rexit(-1); | |
979 | } | |
980 | ||
981 | shutdown(ipc_socket, SHUT_RDWR); | |
982 | close(ipc_socket); | |
72863294 DE |
983 | |
984 | /* set new apparmor profile/selinux context */ | |
985 | if ((options->namespaces & CLONE_NEWNS) && (options->attach_flags & LXC_ATTACH_LSM)) { | |
986 | int on_exec; | |
987 | ||
988 | on_exec = options->attach_flags & LXC_ATTACH_LSM_EXEC ? 1 : 0; | |
989 | ret = lsm_process_label_set(init_ctx->lsm_label, 0, on_exec); | |
990 | if (ret < 0) { | |
991 | rexit(-1); | |
992 | } | |
993 | } | |
fe4de9a6 | 994 | lxc_proc_put_context_info(init_ctx); |
9c4693b8 CS |
995 | |
996 | /* The following is done after the communication socket is | |
997 | * shut down. That way, all errors that might (though | |
998 | * unlikely) occur up until this point will have their messages | |
999 | * printed to the original stderr (if logging is so configured) | |
1000 | * and not the fd the user supplied, if any. | |
1001 | */ | |
1002 | ||
1003 | /* fd handling for stdin, stdout and stderr; | |
1004 | * ignore errors here, user may want to make sure | |
1005 | * the fds are closed, for example */ | |
1006 | if (options->stdin_fd >= 0 && options->stdin_fd != 0) | |
1007 | dup2(options->stdin_fd, 0); | |
1008 | if (options->stdout_fd >= 0 && options->stdout_fd != 1) | |
1009 | dup2(options->stdout_fd, 1); | |
1010 | if (options->stderr_fd >= 0 && options->stderr_fd != 2) | |
1011 | dup2(options->stderr_fd, 2); | |
1012 | ||
1013 | /* close the old fds */ | |
1014 | if (options->stdin_fd > 2) | |
1015 | close(options->stdin_fd); | |
1016 | if (options->stdout_fd > 2) | |
1017 | close(options->stdout_fd); | |
1018 | if (options->stderr_fd > 2) | |
1019 | close(options->stderr_fd); | |
1020 | ||
1021 | /* try to remove CLOEXEC flag from stdin/stdout/stderr, | |
1022 | * but also here, ignore errors */ | |
1023 | for (fd = 0; fd <= 2; fd++) { | |
1024 | flags = fcntl(fd, F_GETFL); | |
1025 | if (flags < 0) | |
1026 | continue; | |
1027 | if (flags & FD_CLOEXEC) | |
1028 | fcntl(fd, F_SETFL, flags & ~FD_CLOEXEC); | |
1029 | } | |
1030 | ||
1031 | /* we're done, so we can now do whatever the user intended us to do */ | |
1032 | rexit(payload->exec_function(payload->exec_payload)); | |
1033 | } | |
1034 | ||
1035 | int lxc_attach_run_command(void* payload) | |
1036 | { | |
1037 | lxc_attach_command_t* cmd = (lxc_attach_command_t*)payload; | |
1038 | ||
1039 | execvp(cmd->program, cmd->argv); | |
1040 | SYSERROR("failed to exec '%s'", cmd->program); | |
1041 | return -1; | |
1042 | } | |
1043 | ||
1044 | int lxc_attach_run_shell(void* payload) | |
1045 | { | |
1046 | uid_t uid; | |
1047 | struct passwd *passwd; | |
1048 | char *user_shell; | |
1049 | ||
1050 | /* ignore payload parameter */ | |
1051 | (void)payload; | |
1052 | ||
1053 | uid = getuid(); | |
1054 | passwd = getpwuid(uid); | |
1055 | ||
1056 | /* this probably happens because of incompatible nss | |
1057 | * implementations in host and container (remember, this | |
1058 | * code is still using the host's glibc but our mount | |
1059 | * namespace is in the container) | |
1060 | * we may try to get the information by spawning a | |
1061 | * [getent passwd uid] process and parsing the result | |
1062 | */ | |
1063 | if (!passwd) | |
1064 | user_shell = lxc_attach_getpwshell(uid); | |
1065 | else | |
1066 | user_shell = passwd->pw_shell; | |
1067 | ||
1068 | if (user_shell) | |
1069 | execlp(user_shell, user_shell, NULL); | |
1070 | ||
1071 | /* executed if either no passwd entry or execvp fails, | |
1072 | * we will fall back on /bin/sh as a default shell | |
1073 | */ | |
1074 | execlp("/bin/sh", "/bin/sh", NULL); | |
1075 | SYSERROR("failed to exec shell"); | |
1076 | return -1; | |
1077 | } |