]>
Commit | Line | Data |
---|---|---|
0ad19a3f | 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> |
0ad19a3f | 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 |
0ad19a3f | 22 | */ |
23 | ||
f3787121 | 24 | #define _GNU_SOURCE |
f549edcc GK |
25 | #include "config.h" |
26 | ||
f3787121 | 27 | #include <alloca.h> |
0ad19a3f | 28 | #include <dirent.h> |
29 | #include <errno.h> | |
b0a33c1e | 30 | #include <fcntl.h> |
c476bdce | 31 | #include <grp.h> |
37515ebd | 32 | #include <poll.h> |
f3787121 CB |
33 | #include <signal.h> |
34 | #include <stdio.h> | |
35 | #include <stdlib.h> | |
36 | #include <string.h> | |
37 | #include <unistd.h> | |
0ad19a3f | 38 | #include <sys/file.h> |
f4d507d5 | 39 | #include <sys/mount.h> |
f3787121 | 40 | #include <sys/param.h> |
0ad19a3f | 41 | #include <sys/prctl.h> |
f3787121 CB |
42 | #include <sys/socket.h> |
43 | #include <sys/stat.h> | |
44 | #include <sys/syscall.h> | |
ddceb1f9 | 45 | #include <sys/types.h> |
b0a33c1e | 46 | #include <sys/un.h> |
f3787121 | 47 | #include <sys/wait.h> |
ff218c25 | 48 | |
495d2046 SG |
49 | #if HAVE_SYS_CAPABILITY_H |
50 | #include <sys/capability.h> | |
51 | #endif | |
52 | ||
955e2a02 | 53 | #ifndef HAVE_DECL_PR_CAPBSET_DROP |
656994bb MH |
54 | #define PR_CAPBSET_DROP 24 |
55 | #endif | |
56 | ||
955e2a02 CB |
57 | #ifndef HAVE_DECL_PR_SET_NO_NEW_PRIVS |
58 | #define PR_SET_NO_NEW_PRIVS 38 | |
59 | #endif | |
60 | ||
61 | #ifndef HAVE_DECL_PR_GET_NO_NEW_PRIVS | |
62 | #define PR_GET_NO_NEW_PRIVS 39 | |
63 | #endif | |
64 | ||
f3787121 | 65 | #include "af_unix.h" |
d8e48992 | 66 | #include "bdev.h" |
f3787121 | 67 | #include "caps.h" |
563f2f2c | 68 | #include "cgroup.h" |
f3787121 CB |
69 | #include "commands.h" |
70 | #include "conf.h" | |
71 | #include "console.h" | |
e2bcd7db | 72 | #include "error.h" |
f3787121 CB |
73 | #include "log.h" |
74 | #include "lxclock.h" | |
75 | #include "lxcseccomp.h" | |
565c2d76 | 76 | #include "lxcutmp.h" |
f3787121 | 77 | #include "mainloop.h" |
63376d7d | 78 | #include "monitor.h" |
f549edcc | 79 | #include "namespace.h" |
f3787121 CB |
80 | #include "start.h" |
81 | #include "sync.h" | |
82 | #include "utils.h" | |
fe4de9a6 | 83 | #include "lsm/lsm.h" |
36eb9bde CLG |
84 | |
85 | lxc_log_define(lxc_start, lxc); | |
86 | ||
9f30a190 MM |
87 | const struct ns_info ns_info[LXC_NS_MAX] = { |
88 | [LXC_NS_MNT] = {"mnt", CLONE_NEWNS}, | |
89 | [LXC_NS_PID] = {"pid", CLONE_NEWPID}, | |
90 | [LXC_NS_UTS] = {"uts", CLONE_NEWUTS}, | |
91 | [LXC_NS_IPC] = {"ipc", CLONE_NEWIPC}, | |
92 | [LXC_NS_USER] = {"user", CLONE_NEWUSER}, | |
ceecc92c CB |
93 | [LXC_NS_NET] = {"net", CLONE_NEWNET}, |
94 | [LXC_NS_CGROUP] = {"cgroup", CLONE_NEWCGROUP} | |
9f30a190 MM |
95 | }; |
96 | ||
f01f7975 | 97 | extern void mod_all_rdeps(struct lxc_container *c, bool inc); |
28272964 CB |
98 | static bool do_destroy_container(struct lxc_conf *conf); |
99 | static int lxc_rmdir_onedev_wrapper(void *data); | |
100 | static void lxc_destroy_container_on_signal(struct lxc_handler *handler, | |
101 | const char *name); | |
102 | ||
c8154066 SH |
103 | static void print_top_failing_dir(const char *path) |
104 | { | |
105 | size_t len = strlen(path); | |
f3787121 | 106 | char *copy = alloca(len + 1), *p, *e, saved; |
c8154066 SH |
107 | strcpy(copy, path); |
108 | ||
109 | p = copy; | |
110 | e = copy + len; | |
111 | while (p < e) { | |
f3787121 CB |
112 | while (p < e && *p == '/') |
113 | p++; | |
114 | while (p < e && *p != '/') | |
115 | p++; | |
c8154066 SH |
116 | saved = *p; |
117 | *p = '\0'; | |
118 | if (access(copy, X_OK)) { | |
408da065 CB |
119 | SYSERROR("Could not access %s. Please grant it x " |
120 | "access, or add an ACL for the container " | |
121 | "root.", copy); | |
c8154066 SH |
122 | return; |
123 | } | |
124 | *p = saved; | |
125 | } | |
126 | } | |
127 | ||
408da065 CB |
128 | static void close_ns(int ns_fd[LXC_NS_MAX]) |
129 | { | |
9f30a190 MM |
130 | int i; |
131 | ||
9f30a190 MM |
132 | for (i = 0; i < LXC_NS_MAX; i++) { |
133 | if (ns_fd[i] > -1) { | |
134 | close(ns_fd[i]); | |
135 | ns_fd[i] = -1; | |
136 | } | |
137 | } | |
9f30a190 MM |
138 | } |
139 | ||
4d8ac866 | 140 | /* preserve_ns: open /proc/@pid/ns/@ns for each namespace specified |
62d05d9b | 141 | * in clone_flags. |
4d8ac866 | 142 | * Return true on success, false on failure. |
62d05d9b | 143 | */ |
4d8ac866 | 144 | static bool preserve_ns(int ns_fd[LXC_NS_MAX], int clone_flags, pid_t pid) |
f3787121 | 145 | { |
62d05d9b | 146 | int i, ret; |
9f30a190 | 147 | |
9f30a190 MM |
148 | for (i = 0; i < LXC_NS_MAX; i++) |
149 | ns_fd[i] = -1; | |
150 | ||
4d8ac866 CB |
151 | ret = lxc_preserve_ns(pid, ""); |
152 | if (ret < 0) { | |
153 | SYSERROR("Kernel does not support attaching to namespaces."); | |
62d05d9b | 154 | return false; |
4d8ac866 CB |
155 | } else { |
156 | close(ret); | |
cd43d2d1 SH |
157 | } |
158 | ||
9f30a190 MM |
159 | for (i = 0; i < LXC_NS_MAX; i++) { |
160 | if ((clone_flags & ns_info[i].clone_flag) == 0) | |
161 | continue; | |
4d8ac866 | 162 | ns_fd[i] = lxc_preserve_ns(pid, ns_info[i].proc_name); |
9f30a190 MM |
163 | if (ns_fd[i] < 0) |
164 | goto error; | |
165 | } | |
166 | ||
62d05d9b | 167 | return true; |
9f30a190 MM |
168 | |
169 | error: | |
4d8ac866 CB |
170 | if (errno == ENOENT) |
171 | SYSERROR("Kernel does not support attaching to %s namespaces.", ns_info[i].proc_name); | |
172 | else | |
173 | SYSERROR("Failed to open file descriptor for %s namespace: %s.", ns_info[i].proc_name, strerror(errno)); | |
9f30a190 | 174 | close_ns(ns_fd); |
62d05d9b | 175 | return false; |
9f30a190 MM |
176 | } |
177 | ||
178 | static int attach_ns(const int ns_fd[LXC_NS_MAX]) { | |
179 | int i; | |
180 | ||
181 | for (i = 0; i < LXC_NS_MAX; i++) { | |
182 | if (ns_fd[i] < 0) | |
183 | continue; | |
184 | ||
185 | if (setns(ns_fd[i], 0) != 0) | |
186 | goto error; | |
187 | } | |
188 | return 0; | |
189 | ||
190 | error: | |
408da065 | 191 | SYSERROR("Failed to attach %s namespace.", ns_info[i].proc_name); |
9f30a190 MM |
192 | return -1; |
193 | } | |
194 | ||
80090207 CLG |
195 | static int match_fd(int fd) |
196 | { | |
197 | return (fd == 0 || fd == 1 || fd == 2); | |
198 | } | |
199 | ||
408da065 CB |
200 | /* Check for any fds we need to close. |
201 | * - If fd_to_ignore != -1, then if we find that fd open we will ignore it. | |
202 | * - By default we warn about open fds we find. | |
203 | * - If closeall is true, we will close open fds. | |
204 | * - If lxc-start was passed "-C", then conf->close_all_fds will be true, in | |
205 | * which case we also close all open fds. | |
206 | * - A daemonized container will always pass closeall=true. | |
d2cf4c37 SH |
207 | */ |
208 | int lxc_check_inherited(struct lxc_conf *conf, bool closeall, int fd_to_ignore) | |
80090207 | 209 | { |
74f96976 | 210 | struct dirent *direntp; |
80090207 CLG |
211 | int fd, fddir; |
212 | DIR *dir; | |
80090207 | 213 | |
d2cf4c37 SH |
214 | if (conf && conf->close_all_fds) |
215 | closeall = true; | |
216 | ||
b119f362 | 217 | restart: |
80090207 CLG |
218 | dir = opendir("/proc/self/fd"); |
219 | if (!dir) { | |
408da065 | 220 | WARN("Failed to open directory: %m."); |
80090207 CLG |
221 | return -1; |
222 | } | |
223 | ||
224 | fddir = dirfd(dir); | |
225 | ||
74f96976 | 226 | while ((direntp = readdir(dir))) { |
80090207 CLG |
227 | if (!direntp) |
228 | break; | |
229 | ||
230 | if (!strcmp(direntp->d_name, ".")) | |
231 | continue; | |
232 | ||
233 | if (!strcmp(direntp->d_name, "..")) | |
234 | continue; | |
235 | ||
236 | fd = atoi(direntp->d_name); | |
237 | ||
f2faa8fa | 238 | if (fd == fddir || fd == lxc_log_fd || fd == fd_to_ignore) |
80090207 CLG |
239 | continue; |
240 | ||
858377e4 SH |
241 | if (current_config && fd == current_config->logfd) |
242 | continue; | |
243 | ||
80090207 CLG |
244 | if (match_fd(fd)) |
245 | continue; | |
80090207 | 246 | |
d2cf4c37 | 247 | if (closeall) { |
b119f362 SH |
248 | close(fd); |
249 | closedir(dir); | |
408da065 | 250 | INFO("Closed inherited fd: %d.", fd); |
b119f362 SH |
251 | goto restart; |
252 | } | |
408da065 | 253 | WARN("Inherited fd: %d.", fd); |
80090207 CLG |
254 | } |
255 | ||
408da065 | 256 | /* Only enable syslog at this point to avoid the above logging function |
64c57ea1 BD |
257 | * to open a new fd and make the check_inherited function enter an |
258 | * infinite loop. | |
259 | */ | |
260 | lxc_log_enable_syslog(); | |
261 | ||
92c7f629 GK |
262 | closedir(dir); /* cannot fail */ |
263 | return 0; | |
80090207 CLG |
264 | } |
265 | ||
83ee7875 | 266 | static int setup_signal_fd(sigset_t *oldmask) |
b0a33c1e | 267 | { |
268 | sigset_t mask; | |
269 | int fd; | |
270 | ||
408da065 | 271 | /* Block everything except serious error signals. */ |
f3304a29 FW |
272 | if (sigfillset(&mask) || |
273 | sigdelset(&mask, SIGILL) || | |
274 | sigdelset(&mask, SIGSEGV) || | |
275 | sigdelset(&mask, SIGBUS) || | |
b5159817 | 276 | sigdelset(&mask, SIGWINCH) || |
f3304a29 | 277 | sigprocmask(SIG_BLOCK, &mask, oldmask)) { |
408da065 | 278 | SYSERROR("Failed to set signal mask."); |
b0a33c1e | 279 | return -1; |
280 | } | |
281 | ||
282 | fd = signalfd(-1, &mask, 0); | |
283 | if (fd < 0) { | |
408da065 | 284 | SYSERROR("Failed to create signal file descriptor."); |
b0a33c1e | 285 | return -1; |
286 | } | |
287 | ||
288 | if (fcntl(fd, F_SETFD, FD_CLOEXEC)) { | |
408da065 | 289 | SYSERROR("Failed to set FD_CLOEXEC on the signal file descriptor: %d.", fd); |
b0a33c1e | 290 | close(fd); |
291 | return -1; | |
292 | } | |
293 | ||
408da065 | 294 | DEBUG("Set SIGCHLD handler with file descriptor: %d.", fd); |
1ac470c0 | 295 | |
b0a33c1e | 296 | return fd; |
297 | } | |
298 | ||
84c92abd | 299 | static int signal_handler(int fd, uint32_t events, void *data, |
f3787121 | 300 | struct lxc_epoll_descr *descr) |
b0a33c1e | 301 | { |
15cd25fd | 302 | struct signalfd_siginfo siginfo; |
80507ee8 | 303 | siginfo_t info; |
15cd25fd | 304 | int ret; |
82d89dce | 305 | pid_t *pid = data; |
80507ee8 | 306 | bool init_died = false; |
15cd25fd DL |
307 | |
308 | ret = read(fd, &siginfo, sizeof(siginfo)); | |
309 | if (ret < 0) { | |
408da065 | 310 | ERROR("Failed to read signal info from signal file descriptor: %d.", fd); |
15cd25fd DL |
311 | return -1; |
312 | } | |
313 | ||
314 | if (ret != sizeof(siginfo)) { | |
408da065 | 315 | ERROR("Unexpected size for siginfo struct."); |
15cd25fd DL |
316 | return -1; |
317 | } | |
318 | ||
408da065 | 319 | /* Check whether init is running. */ |
80507ee8 SH |
320 | info.si_pid = 0; |
321 | ret = waitid(P_PID, *pid, &info, WEXITED | WNOWAIT | WNOHANG); | |
408da065 | 322 | if (ret == 0 && info.si_pid == *pid) |
80507ee8 | 323 | init_died = true; |
80507ee8 | 324 | |
f3304a29 FW |
325 | if (siginfo.ssi_signo != SIGCHLD) { |
326 | kill(*pid, siginfo.ssi_signo); | |
408da065 | 327 | INFO("Forwarded signal %d to pid %d.", siginfo.ssi_signo, *pid); |
80507ee8 | 328 | return init_died ? 1 : 0; |
f3304a29 FW |
329 | } |
330 | ||
408da065 CB |
331 | if (siginfo.ssi_code == CLD_STOPPED) { |
332 | INFO("Container init process was stopped."); | |
333 | return init_died ? 1 : 0; | |
334 | } else if (siginfo.ssi_code == CLD_CONTINUED) { | |
335 | INFO("Container init process was continued."); | |
80507ee8 | 336 | return init_died ? 1 : 0; |
15cd25fd | 337 | } |
1ac470c0 | 338 | |
408da065 CB |
339 | /* More robustness, protect ourself from a SIGCHLD sent |
340 | * by a process different from the container init. | |
82d89dce DL |
341 | */ |
342 | if (siginfo.ssi_pid != *pid) { | |
408da065 | 343 | WARN("Invalid pid for SIGCHLD. Received pid %d, expected pid %d.", siginfo.ssi_pid, *pid); |
80507ee8 | 344 | return init_died ? 1 : 0; |
82d89dce DL |
345 | } |
346 | ||
408da065 | 347 | DEBUG("Container init process %d exited.", *pid); |
b0a33c1e | 348 | return 1; |
349 | } | |
350 | ||
735f2c6e | 351 | int lxc_set_state(const char *name, struct lxc_handler *handler, lxc_state_t state) |
66aeffc7 DL |
352 | { |
353 | handler->state = state; | |
9123e471 | 354 | lxc_monitor_send_state(name, state, handler->lxcpath); |
66aeffc7 DL |
355 | return 0; |
356 | } | |
357 | ||
735f2c6e | 358 | int lxc_poll(const char *name, struct lxc_handler *handler) |
b0a33c1e | 359 | { |
ca5f7926 DL |
360 | int sigfd = handler->sigfd; |
361 | int pid = handler->pid; | |
b0a33c1e | 362 | struct lxc_epoll_descr descr; |
363 | ||
a9e61274 | 364 | if (lxc_mainloop_open(&descr)) { |
408da065 | 365 | ERROR("Failed to create LXC mainloop."); |
50c8bf05 | 366 | goto out_sigfd; |
b0a33c1e | 367 | } |
368 | ||
83ee7875 | 369 | if (lxc_mainloop_add_handler(&descr, sigfd, signal_handler, &pid)) { |
408da065 | 370 | ERROR("Failed to add signal handler with file descriptor %d to LXC mainloop.", sigfd); |
b0a33c1e | 371 | goto out_mainloop_open; |
372 | } | |
373 | ||
da41561c | 374 | if (lxc_console_mainloop_add(&descr, handler->conf)) { |
408da065 | 375 | ERROR("Failed to add console handler to LXC mainloop."); |
63376d7d DL |
376 | goto out_mainloop_open; |
377 | } | |
378 | ||
ef6e34ee | 379 | if (lxc_cmd_mainloop_add(name, &descr, handler)) { |
408da065 | 380 | ERROR("Failed to add command handler to LXC mainloop."); |
96fa1ff0 | 381 | goto out_mainloop_open; |
563f2f2c DL |
382 | } |
383 | ||
828695d9 | 384 | if (handler->conf->need_utmp_watch) { |
495d2046 | 385 | #if HAVE_SYS_CAPABILITY_H |
828695d9 | 386 | if (lxc_utmp_mainloop_add(&descr, handler)) { |
408da065 | 387 | ERROR("Failed to add utmp handler to LXC mainloop."); |
828695d9 SH |
388 | goto out_mainloop_open; |
389 | } | |
495d2046 | 390 | #else |
408da065 | 391 | DEBUG("Not starting utmp handler as CAP_SYS_BOOT cannot be dropped without capabilities support."); |
495d2046 | 392 | #endif |
563f2f2c | 393 | } |
b0a33c1e | 394 | |
e51d4895 | 395 | return lxc_mainloop(&descr, -1); |
b0a33c1e | 396 | |
397 | out_mainloop_open: | |
398 | lxc_mainloop_close(&descr); | |
408da065 | 399 | |
b0a33c1e | 400 | out_sigfd: |
401 | close(sigfd); | |
408da065 | 402 | |
c3e13372 | 403 | return -1; |
b0a33c1e | 404 | } |
405 | ||
13f5be62 | 406 | struct lxc_handler *lxc_init(const char *name, struct lxc_conf *conf, const char *lxcpath) |
59eb99ba | 407 | { |
b6b2b194 | 408 | int i; |
3a0f472d DL |
409 | struct lxc_handler *handler; |
410 | ||
411 | handler = malloc(sizeof(*handler)); | |
412 | if (!handler) | |
413 | return NULL; | |
59eb99ba DL |
414 | |
415 | memset(handler, 0, sizeof(*handler)); | |
416 | ||
e8bd4e43 | 417 | handler->ttysock[0] = handler->ttysock[1] = -1; |
fae349da | 418 | handler->conf = conf; |
9123e471 | 419 | handler->lxcpath = lxcpath; |
5c068da9 | 420 | handler->pinfd = -1; |
fae349da | 421 | |
b6b2b194 WB |
422 | for (i = 0; i < LXC_NS_MAX; i++) |
423 | handler->nsfd[i] = -1; | |
424 | ||
fe4de9a6 DE |
425 | lsm_init(); |
426 | ||
3bdf52d7 DL |
427 | handler->name = strdup(name); |
428 | if (!handler->name) { | |
408da065 | 429 | ERROR("Failed to allocate memory."); |
3bdf52d7 DL |
430 | goto out_free; |
431 | } | |
432 | ||
ef6e34ee | 433 | if (lxc_cmd_init(name, handler, lxcpath)) |
d2e30e99 DE |
434 | goto out_free_name; |
435 | ||
8f2c3a70 | 436 | if (lxc_read_seccomp_config(conf) != 0) { |
408da065 | 437 | ERROR("Failed loading seccomp policy."); |
d2e30e99 | 438 | goto out_close_maincmd_fd; |
8f2c3a70 SH |
439 | } |
440 | ||
408da065 | 441 | /* Begin by setting the state to STARTING. */ |
25c2aca5 | 442 | if (lxc_set_state(name, handler, STARTING)) { |
408da065 | 443 | ERROR("Failed to set state for container \"%s\" to \"%s\".", name, lxc_state2str(STARTING)); |
051151de | 444 | goto out_close_maincmd_fd; |
0ad19a3f | 445 | } |
446 | ||
408da065 CB |
447 | /* Start of environment variable setup for hooks. */ |
448 | if (name && setenv("LXC_NAME", name, 1)) | |
449 | SYSERROR("Failed to set environment variable: LXC_NAME=%s.", name); | |
450 | ||
451 | if (conf->rcfile && setenv("LXC_CONFIG_FILE", conf->rcfile, 1)) | |
452 | SYSERROR("Failed to set environment variable: LXC_CONFIG_FILE=%s.", conf->rcfile); | |
453 | ||
454 | if (conf->rootfs.mount && setenv("LXC_ROOTFS_MOUNT", conf->rootfs.mount, 1)) | |
455 | SYSERROR("Failed to set environment variable: LXC_ROOTFS_MOUNT=%s.", conf->rootfs.mount); | |
456 | ||
457 | if (conf->rootfs.path && setenv("LXC_ROOTFS_PATH", conf->rootfs.path, 1)) | |
458 | SYSERROR("Failed to set environment variable: LXC_ROOTFS_PATH=%s.", conf->rootfs.path); | |
459 | ||
460 | if (conf->console.path && setenv("LXC_CONSOLE", conf->console.path, 1)) | |
461 | SYSERROR("Failed to set environment variable: LXC_CONSOLE=%s.", conf->console.path); | |
462 | ||
463 | if (conf->console.log_path && setenv("LXC_CONSOLE_LOGPATH", conf->console.log_path, 1)) | |
464 | SYSERROR("Failed to set environment variable: LXC_CONSOLE_LOGPATH=%s.", conf->console.log_path); | |
465 | ||
466 | if (setenv("LXC_CGNS_AWARE", "1", 1)) | |
467 | SYSERROR("Failed to set environment variable LXC_CGNS_AWARE=1."); | |
468 | /* End of environment variable setup for hooks. */ | |
f7bee6c6 | 469 | |
283678ed | 470 | if (run_lxc_hooks(name, "pre-start", conf, handler->lxcpath, NULL)) { |
408da065 | 471 | ERROR("Failed to run lxc.hook.pre-start for container \"%s\".", name); |
773fb9ca SH |
472 | goto out_aborting; |
473 | } | |
26ddeedd | 474 | |
408da065 CB |
475 | /* The signal fd has to be created before forking otherwise if the child |
476 | * process exits before we setup the signal fd, the event will be lost | |
477 | * and the command will be stuck. | |
478 | */ | |
83ee7875 | 479 | handler->sigfd = setup_signal_fd(&handler->oldmask); |
59eb99ba | 480 | if (handler->sigfd < 0) { |
408da065 | 481 | ERROR("Failed to setup SIGCHLD fd handler."); |
b5159817 DE |
482 | goto out_delete_tty; |
483 | } | |
484 | ||
408da065 | 485 | /* Do this after setting up signals since it might unblock SIGWINCH. */ |
b5159817 | 486 | if (lxc_console_create(conf)) { |
408da065 | 487 | ERROR("Failed to create console for container \"%s\".", name); |
b5159817 | 488 | goto out_restore_sigmask; |
b0a33c1e | 489 | } |
490 | ||
c4d10a05 | 491 | if (ttys_shift_ids(conf) < 0) { |
408da065 | 492 | ERROR("Failed to shift tty into container."); |
c4d10a05 SH |
493 | goto out_restore_sigmask; |
494 | } | |
495 | ||
408da065 | 496 | INFO("Container \"%s\" is initialized.", name); |
3a0f472d | 497 | return handler; |
59eb99ba | 498 | |
b5159817 DE |
499 | out_restore_sigmask: |
500 | sigprocmask(SIG_SETMASK, &handler->oldmask, NULL); | |
59eb99ba | 501 | out_delete_tty: |
fae349da | 502 | lxc_delete_tty(&conf->tty_info); |
59eb99ba | 503 | out_aborting: |
25c2aca5 | 504 | lxc_set_state(name, handler, ABORTING); |
d2e30e99 DE |
505 | out_close_maincmd_fd: |
506 | close(conf->maincmd_fd); | |
507 | conf->maincmd_fd = -1; | |
3bdf52d7 DL |
508 | out_free_name: |
509 | free(handler->name); | |
510 | handler->name = NULL; | |
3a0f472d DL |
511 | out_free: |
512 | free(handler); | |
c3e13372 | 513 | return NULL; |
59eb99ba DL |
514 | } |
515 | ||
3b72c4a0 | 516 | void lxc_fini(const char *name, struct lxc_handler *handler) |
59eb99ba | 517 | { |
b3286b62 WB |
518 | int i, rc; |
519 | pid_t self = getpid(); | |
520 | char *namespaces[LXC_NS_MAX+1]; | |
521 | size_t namespace_count = 0; | |
b6b2b194 | 522 | |
408da065 CB |
523 | /* The STOPPING state is there for future cleanup code which can take |
524 | * awhile. | |
59eb99ba | 525 | */ |
25c2aca5 | 526 | lxc_set_state(name, handler, STOPPING); |
b6b2b194 | 527 | |
b3286b62 WB |
528 | for (i = 0; i < LXC_NS_MAX; i++) { |
529 | if (handler->nsfd[i] != -1) { | |
530 | rc = asprintf(&namespaces[namespace_count], "%s:/proc/%d/fd/%d", | |
531 | ns_info[i].proc_name, self, handler->nsfd[i]); | |
532 | if (rc == -1) { | |
408da065 | 533 | SYSERROR("Failed to allocate memory."); |
b3286b62 WB |
534 | break; |
535 | } | |
536 | ++namespace_count; | |
537 | } | |
538 | } | |
539 | namespaces[namespace_count] = NULL; | |
c154af98 | 540 | |
408da065 CB |
541 | if (handler->conf->reboot && setenv("LXC_TARGET", "reboot", 1)) |
542 | SYSERROR("Failed to set environment variable: LXC_TARGET=reboot."); | |
543 | ||
544 | if (!handler->conf->reboot && setenv("LXC_TARGET", "stop", 1)) | |
545 | SYSERROR("Failed to set environment variable: LXC_TARGET=stop."); | |
c154af98 | 546 | |
b3286b62 | 547 | if (run_lxc_hooks(name, "stop", handler->conf, handler->lxcpath, namespaces)) |
408da065 | 548 | ERROR("Failed to run lxc.hook.stop for container \"%s\".", name); |
c154af98 | 549 | |
b3286b62 WB |
550 | while (namespace_count--) |
551 | free(namespaces[namespace_count]); | |
b6b2b194 WB |
552 | for (i = 0; i < LXC_NS_MAX; i++) { |
553 | if (handler->nsfd[i] != -1) { | |
554 | close(handler->nsfd[i]); | |
555 | handler->nsfd[i] = -1; | |
556 | } | |
557 | } | |
738d0deb CB |
558 | |
559 | if (handler->netnsfd >= 0) { | |
560 | close(handler->netnsfd); | |
561 | handler->netnsfd = -1; | |
562 | } | |
563 | ||
25c2aca5 | 564 | lxc_set_state(name, handler, STOPPED); |
59eb99ba | 565 | |
f3787121 | 566 | if (run_lxc_hooks(name, "post-stop", handler->conf, handler->lxcpath, NULL)) { |
408da065 | 567 | ERROR("Failed to run lxc.hook.post-stop for container \"%s\".", name); |
f3787121 CB |
568 | if (handler->conf->reboot) { |
569 | WARN("Container will be stopped instead of rebooted."); | |
570 | handler->conf->reboot = 0; | |
408da065 CB |
571 | if (setenv("LXC_TARGET", "stop", 1)) |
572 | WARN("Failed to set environment variable: LXC_TARGET=stop."); | |
f3787121 CB |
573 | } |
574 | } | |
26ddeedd | 575 | |
408da065 | 576 | /* Reset mask set by setup_signal_fd. */ |
8f64a3f6 | 577 | if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL)) |
408da065 | 578 | WARN("Failed to restore signal mask."); |
8f64a3f6 | 579 | |
b5159817 | 580 | lxc_console_delete(&handler->conf->console); |
b2431939 | 581 | lxc_delete_tty(&handler->conf->tty_info); |
d2e30e99 DE |
582 | close(handler->conf->maincmd_fd); |
583 | handler->conf->maincmd_fd = -1; | |
3bdf52d7 | 584 | free(handler->name); |
e8bd4e43 SH |
585 | if (handler->ttysock[0] != -1) { |
586 | close(handler->ttysock[0]); | |
587 | close(handler->ttysock[1]); | |
588 | } | |
2c5f2ede CB |
589 | |
590 | if (handler->conf->ephemeral == 1 && handler->conf->reboot != 1) | |
28272964 | 591 | lxc_destroy_container_on_signal(handler, name); |
2c5f2ede | 592 | |
d4ef7c50 | 593 | cgroup_destroy(handler); |
b2431939 | 594 | free(handler); |
59eb99ba DL |
595 | } |
596 | ||
735f2c6e | 597 | void lxc_abort(const char *name, struct lxc_handler *handler) |
59eb99ba | 598 | { |
73e608b2 SH |
599 | int ret, status; |
600 | ||
25c2aca5 | 601 | lxc_set_state(name, handler, ABORTING); |
7d9fb3e9 DL |
602 | if (handler->pid > 0) |
603 | kill(handler->pid, SIGKILL); | |
408da065 CB |
604 | while ((ret = waitpid(-1, &status, 0)) > 0) { |
605 | ; | |
606 | } | |
59eb99ba DL |
607 | } |
608 | ||
828695d9 SH |
609 | #include <sys/reboot.h> |
610 | #include <linux/reboot.h> | |
611 | ||
408da065 CB |
612 | /* reboot(LINUX_REBOOT_CMD_CAD_ON) will return -EINVAL in a child pid namespace |
613 | * if container reboot support exists. Otherwise, it will either succeed or | |
614 | * return -EPERM. | |
e2fa1520 SH |
615 | */ |
616 | static int container_reboot_supported(void *arg) | |
828695d9 | 617 | { |
d028235d | 618 | int *cmd = arg; |
828695d9 | 619 | int ret; |
828695d9 | 620 | |
d028235d | 621 | ret = reboot(*cmd); |
e2fa1520 SH |
622 | if (ret == -1 && errno == EINVAL) |
623 | return 1; | |
624 | return 0; | |
625 | } | |
626 | ||
b60ed720 | 627 | static int must_drop_cap_sys_boot(struct lxc_conf *conf) |
e2fa1520 | 628 | { |
025ed0f3 | 629 | FILE *f; |
b60ed720 | 630 | int ret, cmd, v, flags; |
d028235d SG |
631 | long stack_size = 4096; |
632 | void *stack = alloca(stack_size); | |
633 | int status; | |
634 | pid_t pid; | |
e2fa1520 | 635 | |
025ed0f3 | 636 | f = fopen("/proc/sys/kernel/ctrl-alt-del", "r"); |
e2fa1520 SH |
637 | if (!f) { |
638 | DEBUG("failed to open /proc/sys/kernel/ctrl-alt-del"); | |
828695d9 | 639 | return 1; |
e2fa1520 | 640 | } |
828695d9 SH |
641 | |
642 | ret = fscanf(f, "%d", &v); | |
643 | fclose(f); | |
e2fa1520 | 644 | if (ret != 1) { |
408da065 | 645 | DEBUG("Failed to read /proc/sys/kernel/ctrl-alt-del."); |
828695d9 | 646 | return 1; |
e2fa1520 SH |
647 | } |
648 | cmd = v ? LINUX_REBOOT_CMD_CAD_ON : LINUX_REBOOT_CMD_CAD_OFF; | |
649 | ||
b60ed720 SH |
650 | flags = CLONE_NEWPID | SIGCHLD; |
651 | if (!lxc_list_empty(&conf->id_map)) | |
652 | flags |= CLONE_NEWUSER; | |
653 | ||
7f145a6d | 654 | #ifdef __ia64__ |
959aee9c | 655 | pid = __clone2(container_reboot_supported, stack, stack_size, flags, &cmd); |
7f145a6d | 656 | #else |
959aee9c SG |
657 | stack += stack_size; |
658 | pid = clone(container_reboot_supported, stack, flags, &cmd); | |
7f145a6d | 659 | #endif |
959aee9c | 660 | if (pid < 0) { |
c08220e9 | 661 | if (flags & CLONE_NEWUSER) |
408da065 | 662 | ERROR("Failed to clone (%#x): %s (includes CLONE_NEWUSER).", flags, strerror(errno)); |
c08220e9 | 663 | else |
408da065 | 664 | ERROR("Failed to clone (%#x): %s.", flags, strerror(errno)); |
959aee9c SG |
665 | return -1; |
666 | } | |
667 | if (wait(&status) < 0) { | |
408da065 | 668 | SYSERROR("Unexpected wait error: %m."); |
959aee9c SG |
669 | return -1; |
670 | } | |
e2fa1520 SH |
671 | |
672 | if (WEXITSTATUS(status) != 1) | |
828695d9 | 673 | return 1; |
e2fa1520 | 674 | |
828695d9 SH |
675 | return 0; |
676 | } | |
677 | ||
408da065 CB |
678 | /* netpipe is used in the unprivileged case to transfer the ifindexes from |
679 | * parent to child | |
658979c5 SH |
680 | */ |
681 | static int netpipe = -1; | |
682 | ||
683 | static inline int count_veths(struct lxc_list *network) | |
684 | { | |
685 | struct lxc_list *iterator; | |
686 | struct lxc_netdev *netdev; | |
687 | int count = 0; | |
688 | ||
689 | lxc_list_for_each(iterator, network) { | |
690 | netdev = iterator->elem; | |
691 | if (netdev->type != LXC_NET_VETH) | |
692 | continue; | |
693 | count++; | |
694 | } | |
695 | return count; | |
696 | } | |
697 | ||
698 | static int read_unpriv_netifindex(struct lxc_list *network) | |
699 | { | |
700 | struct lxc_list *iterator; | |
701 | struct lxc_netdev *netdev; | |
702 | ||
703 | if (netpipe == -1) | |
704 | return 0; | |
705 | lxc_list_for_each(iterator, network) { | |
706 | netdev = iterator->elem; | |
707 | if (netdev->type != LXC_NET_VETH) | |
708 | continue; | |
709 | if (!(netdev->name = malloc(IFNAMSIZ))) { | |
408da065 | 710 | ERROR("Out of memory."); |
658979c5 SH |
711 | close(netpipe); |
712 | return -1; | |
713 | } | |
714 | if (read(netpipe, netdev->name, IFNAMSIZ) != IFNAMSIZ) { | |
715 | close(netpipe); | |
716 | return -1; | |
717 | } | |
718 | } | |
719 | close(netpipe); | |
720 | return 0; | |
721 | } | |
722 | ||
ffe1e01a | 723 | static int do_start(void *data) |
50e98013 | 724 | { |
7c661726 | 725 | struct lxc_list *iterator; |
23c53af9 | 726 | struct lxc_handler *handler = data; |
7a55c157 TA |
727 | int devnull_fd = -1, ret; |
728 | char path[PATH_MAX]; | |
50e98013 DL |
729 | |
730 | if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL)) { | |
408da065 | 731 | SYSERROR("Failed to set signal mask."); |
9d7f9e52 | 732 | return -1; |
50e98013 DL |
733 | } |
734 | ||
408da065 CB |
735 | /* This prctl must be before the synchro, so if the parent dies before |
736 | * we set the parent death signal, we will detect its death with the | |
737 | * synchro right after, otherwise we have a window where the parent can | |
738 | * exit before we set the pdeath signal leading to a unsupervized | |
739 | * container. | |
743ecd2e DL |
740 | */ |
741 | if (prctl(PR_SET_PDEATHSIG, SIGKILL, 0, 0, 0)) { | |
408da065 | 742 | SYSERROR("Failed to set PR_SET_PDEATHSIG to SIGKILL."); |
743ecd2e DL |
743 | return -1; |
744 | } | |
745 | ||
3c22086f | 746 | lxc_sync_fini_parent(handler); |
50e98013 | 747 | |
408da065 | 748 | /* Don't leak the pinfd to the container. */ |
025ed0f3 | 749 | if (handler->pinfd >= 0) { |
0d03360a | 750 | close(handler->pinfd); |
025ed0f3 | 751 | } |
2b0e17e4 | 752 | |
5b1e83cb SH |
753 | if (lxc_sync_wait_parent(handler, LXC_SYNC_STARTUP)) |
754 | return -1; | |
755 | ||
408da065 CB |
756 | /* Unshare CLONE_NEWNET after CLONE_NEWUSER. See |
757 | * https://github.com/lxc/lxd/issues/1978. | |
758 | */ | |
5b1e83cb | 759 | if ((handler->clone_flags & (CLONE_NEWNET | CLONE_NEWUSER)) == |
408da065 | 760 | (CLONE_NEWNET | CLONE_NEWUSER)) { |
5b1e83cb SH |
761 | ret = unshare(CLONE_NEWNET); |
762 | if (ret < 0) { | |
408da065 | 763 | SYSERROR("Failed to unshare CLONE_NEWNET."); |
5b1e83cb SH |
764 | goto out_warn_father; |
765 | } | |
408da065 | 766 | INFO("Unshared CLONE_NEWNET."); |
5b1e83cb SH |
767 | } |
768 | ||
408da065 CB |
769 | /* Tell the parent task it can begin to configure the container and wait |
770 | * for it to finish. | |
3c22086f CLG |
771 | */ |
772 | if (lxc_sync_barrier_parent(handler, LXC_SYNC_CONFIGURE)) | |
9d7f9e52 | 773 | return -1; |
50e98013 | 774 | |
658979c5 SH |
775 | if (read_unpriv_netifindex(&handler->conf->network) < 0) |
776 | goto out_warn_father; | |
777 | ||
408da065 CB |
778 | /* If we are in a new user namespace, become root there to have |
779 | * privilege over our namespace. When using lxc-execute we default to | |
780 | * root, but this can be overriden using the lxc.init_uid and | |
781 | * lxc.init_gid configuration options. | |
f6d3e3e4 SH |
782 | */ |
783 | if (!lxc_list_empty(&handler->conf->id_map)) { | |
c5cd20ce PT |
784 | gid_t new_gid = 0; |
785 | if (handler->conf->is_execute && handler->conf->init_gid) | |
786 | new_gid = handler->conf->init_gid; | |
787 | ||
788 | uid_t new_uid = 0; | |
789 | if (handler->conf->is_execute && handler->conf->init_uid) | |
790 | new_uid = handler->conf->init_uid; | |
791 | ||
408da065 | 792 | NOTICE("Switching to uid=%d and gid=%d in new user namespace.", new_uid, new_gid); |
56f8ff00 | 793 | if (setgid(new_gid)) { |
408da065 | 794 | SYSERROR("Failed to setgid()."); |
f6d3e3e4 SH |
795 | goto out_warn_father; |
796 | } | |
56f8ff00 | 797 | if (setuid(new_uid)) { |
408da065 | 798 | SYSERROR("Failed to setuid()."); |
f6d3e3e4 SH |
799 | goto out_warn_father; |
800 | } | |
c476bdce | 801 | if (setgroups(0, NULL)) { |
408da065 | 802 | SYSERROR("Failed to setgroups()."); |
c476bdce SH |
803 | goto out_warn_father; |
804 | } | |
f6d3e3e4 SH |
805 | } |
806 | ||
99b71824 | 807 | if (access(handler->lxcpath, X_OK)) { |
c8154066 SH |
808 | print_top_failing_dir(handler->lxcpath); |
809 | goto out_warn_father; | |
810 | } | |
811 | ||
495d2046 | 812 | #if HAVE_SYS_CAPABILITY_H |
69182a31 | 813 | if (handler->conf->need_utmp_watch) { |
828695d9 | 814 | if (prctl(PR_CAPBSET_DROP, CAP_SYS_BOOT, 0, 0, 0)) { |
408da065 | 815 | SYSERROR("Failed to remove the CAP_SYS_BOOT capability."); |
c4ea60df | 816 | goto out_warn_father; |
828695d9 | 817 | } |
408da065 | 818 | DEBUG("Dropped the CAP_SYS_BOOT capability."); |
e2fa1520 | 819 | } |
495d2046 | 820 | #endif |
e2fa1520 | 821 | |
a91cde21 | 822 | ret = snprintf(path, sizeof(path), "%s/dev/null", handler->conf->rootfs.mount); |
408da065 | 823 | if (ret < 0 || ret >= sizeof(path)) |
7a55c157 | 824 | goto out_warn_father; |
7a55c157 TA |
825 | |
826 | /* In order to checkpoint restore, we need to have everything in the | |
827 | * same mount namespace. However, some containers may not have a | |
828 | * reasonable /dev (in particular, they may not have /dev/null), so we | |
829 | * can't set init's std fds to /dev/null by opening it from inside the | |
830 | * container. | |
831 | * | |
832 | * If that's the case, fall back to using the host's /dev/null. This | |
833 | * means that migration won't work, but at least we won't spew output | |
834 | * where it isn't wanted. | |
835 | */ | |
836 | if (handler->backgrounded && !handler->conf->autodev && access(path, F_OK) < 0) { | |
c44de748 AM |
837 | devnull_fd = open_devnull(); |
838 | ||
839 | if (devnull_fd < 0) | |
840 | goto out_warn_father; | |
408da065 CB |
841 | WARN("Using /dev/null from the host for container init's " |
842 | "standard file descriptors. Migration will not work."); | |
c44de748 AM |
843 | } |
844 | ||
e2fa1520 | 845 | /* Setup the container, ip, names, utsname, ... */ |
d4ef7c50 | 846 | if (lxc_setup(handler)) { |
408da065 | 847 | ERROR("Failed to setup container \"%s\".", handler->name); |
e2fa1520 SH |
848 | goto out_warn_father; |
849 | } | |
50e98013 | 850 | |
408da065 | 851 | /* Ask father to setup cgroups and wait for him to finish. */ |
544a48a0 | 852 | if (lxc_sync_barrier_parent(handler, LXC_SYNC_CGROUP)) |
c44de748 | 853 | goto out_error; |
544a48a0 | 854 | |
deefdf8a CB |
855 | /* Unshare cgroup namespace after we have setup our cgroups. If we do it |
856 | * earlier we end up with a wrong view of /proc/self/cgroup. For | |
857 | * example, assume we unshare(CLONE_NEWCGROUP) first, and then create | |
858 | * the cgroup for the container, say /sys/fs/cgroup/cpuset/lxc/c, then | |
859 | * /proc/self/cgroup would show us: | |
860 | * | |
861 | * 8:cpuset:/lxc/c | |
862 | * | |
863 | * whereas it should actually show | |
864 | * | |
865 | * 8:cpuset:/ | |
866 | */ | |
867 | if (cgns_supported()) { | |
868 | if (unshare(CLONE_NEWCGROUP) < 0) { | |
869 | INFO("Failed to unshare CLONE_NEWCGROUP."); | |
870 | goto out_warn_father; | |
871 | } | |
872 | INFO("Unshared CLONE_NEWCGROUP."); | |
873 | } | |
874 | ||
408da065 | 875 | /* Set the label to change to when we exec(2) the container's init. */ |
7aff4f43 | 876 | if (lsm_process_label_set(NULL, handler->conf, 1, 1) < 0) |
e075f5d9 | 877 | goto out_warn_father; |
5112cd70 | 878 | |
029cdff5 | 879 | /* Set PR_SET_NO_NEW_PRIVS after we changed the lsm label. If we do it |
408da065 CB |
880 | * before we aren't allowed anymore. |
881 | */ | |
029cdff5 CB |
882 | if (handler->conf->no_new_privs) { |
883 | if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) < 0) { | |
884 | SYSERROR("Could not set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges."); | |
885 | goto out_warn_father; | |
886 | } | |
887 | DEBUG("Set PR_SET_NO_NEW_PRIVS to block execve() gainable privileges."); | |
888 | } | |
889 | ||
0d9acb99 DE |
890 | /* Some init's such as busybox will set sane tty settings on stdin, |
891 | * stdout, stderr which it thinks is the console. We already set them | |
892 | * the way we wanted on the real terminal, and we want init to do its | |
408da065 CB |
893 | * setup on its console ie. the pty allocated in lxc_console_create() so |
894 | * make sure that that pty is stdin,stdout,stderr. | |
0d9acb99 | 895 | */ |
39a78bbe | 896 | if (lxc_console_set_stdfds(handler->conf->console.slave) < 0) |
0d9acb99 DE |
897 | goto out_warn_father; |
898 | ||
408da065 | 899 | /* If we mounted a temporary proc, then unmount it now. */ |
5112cd70 | 900 | tmp_proc_unmount(handler->conf); |
e075f5d9 | 901 | |
8f2c3a70 SH |
902 | if (lxc_seccomp_load(handler->conf) != 0) |
903 | goto out_warn_father; | |
904 | ||
283678ed | 905 | if (run_lxc_hooks(handler->name, "start", handler->conf, handler->lxcpath, NULL)) { |
408da065 | 906 | ERROR("Failed to run lxc.hook.start for container \"%s\".", handler->name); |
773fb9ca SH |
907 | goto out_warn_father; |
908 | } | |
fc25b815 | 909 | |
408da065 CB |
910 | /* The clearenv() and putenv() calls have been moved here to allow us to |
911 | * use environment variables passed to the various hooks, such as the | |
912 | * start hook above. Not all of the variables like CONFIG_PATH or ROOTFS | |
913 | * are valid in this context but others are. | |
914 | */ | |
f7bee6c6 | 915 | if (clearenv()) { |
408da065 CB |
916 | SYSERROR("Failed to clear environment."); |
917 | /* Don't error out though. */ | |
f7bee6c6 MW |
918 | } |
919 | ||
7c661726 MP |
920 | lxc_list_for_each(iterator, &handler->conf->environment) { |
921 | if (putenv((char *)iterator->elem)) { | |
408da065 | 922 | SYSERROR("Failed to set environment variable: %s.", (char *)iterator->elem); |
7c661726 MP |
923 | goto out_warn_father; |
924 | } | |
925 | } | |
926 | ||
f7bee6c6 | 927 | if (putenv("container=lxc")) { |
408da065 | 928 | SYSERROR("Failed to set environment variable: container=lxc."); |
c4ea60df | 929 | goto out_warn_father; |
f7bee6c6 MW |
930 | } |
931 | ||
393903d1 SH |
932 | if (handler->conf->pty_names) { |
933 | if (putenv(handler->conf->pty_names)) { | |
408da065 | 934 | SYSERROR("Failed to set environment variable for container ptys."); |
393903d1 SH |
935 | goto out_warn_father; |
936 | } | |
937 | } | |
938 | ||
773fb9ca | 939 | close(handler->sigfd); |
26ddeedd | 940 | |
7a55c157 TA |
941 | if (devnull_fd < 0) { |
942 | devnull_fd = open_devnull(); | |
943 | ||
944 | if (devnull_fd < 0) | |
945 | goto out_warn_father; | |
946 | } | |
947 | ||
c44de748 | 948 | if (handler->backgrounded && set_stdfds(devnull_fd)) |
69aeabac | 949 | goto out_warn_father; |
507cee36 | 950 | |
c44de748 AM |
951 | if (devnull_fd >= 0) { |
952 | close(devnull_fd); | |
953 | devnull_fd = -1; | |
954 | } | |
955 | ||
8c9a7665 TA |
956 | setsid(); |
957 | ||
408da065 CB |
958 | /* After this call, we are in error because this ops should not return |
959 | * as it execs. | |
960 | */ | |
c4ea60df | 961 | handler->ops->start(handler, handler->data); |
50e98013 DL |
962 | |
963 | out_warn_father: | |
408da065 CB |
964 | /* We want the parent to know something went wrong, so we return a |
965 | * special error code. | |
966 | */ | |
d1ccb562 | 967 | lxc_sync_wake_parent(handler, LXC_SYNC_ERROR); |
c44de748 AM |
968 | |
969 | out_error: | |
970 | if (devnull_fd >= 0) | |
971 | close(devnull_fd); | |
972 | ||
50e98013 DL |
973 | return -1; |
974 | } | |
975 | ||
74a3920a | 976 | static int save_phys_nics(struct lxc_conf *conf) |
7b35f3d6 SH |
977 | { |
978 | struct lxc_list *iterator; | |
40f2f8a2 | 979 | int am_root = (getuid() == 0); |
7b35f3d6 | 980 | |
40f2f8a2 LQ |
981 | if (!am_root) |
982 | return 0; | |
2c5f2ede | 983 | |
7b35f3d6 SH |
984 | lxc_list_for_each(iterator, &conf->network) { |
985 | struct lxc_netdev *netdev = iterator->elem; | |
986 | ||
987 | if (netdev->type != LXC_NET_PHYS) | |
988 | continue; | |
989 | conf->saved_nics = realloc(conf->saved_nics, | |
990 | (conf->num_savednics+1)*sizeof(struct saved_nic)); | |
408da065 | 991 | if (!conf->saved_nics) |
7b35f3d6 | 992 | return -1; |
7b35f3d6 SH |
993 | conf->saved_nics[conf->num_savednics].ifindex = netdev->ifindex; |
994 | conf->saved_nics[conf->num_savednics].orig_name = strdup(netdev->link); | |
408da065 | 995 | if (!conf->saved_nics[conf->num_savednics].orig_name) |
7b35f3d6 | 996 | return -1; |
408da065 | 997 | INFO("Stored saved_nic #%d idx %d name %s.", conf->num_savednics, |
7b35f3d6 SH |
998 | conf->saved_nics[conf->num_savednics].ifindex, |
999 | conf->saved_nics[conf->num_savednics].orig_name); | |
1000 | conf->num_savednics++; | |
1001 | } | |
1002 | ||
1003 | return 0; | |
1004 | } | |
1005 | ||
e8bd4e43 SH |
1006 | static int recv_fd(int sock, int *fd) |
1007 | { | |
1008 | if (lxc_abstract_unix_recv_fd(sock, fd, NULL, 0) < 0) { | |
408da065 | 1009 | SYSERROR("Error receiving tty file descriptor from child process."); |
e8bd4e43 SH |
1010 | return -1; |
1011 | } | |
1012 | if (*fd == -1) | |
1013 | return -1; | |
1014 | return 0; | |
1015 | } | |
1016 | ||
1017 | static int recv_ttys_from_child(struct lxc_handler *handler) | |
1018 | { | |
1019 | struct lxc_conf *conf = handler->conf; | |
1020 | int i, sock = handler->ttysock[1]; | |
1021 | struct lxc_tty_info *tty_info = &conf->tty_info; | |
1022 | ||
1023 | if (!conf->tty) | |
1024 | return 0; | |
1025 | ||
408da065 CB |
1026 | tty_info->pty_info = malloc(sizeof(*tty_info->pty_info) * conf->tty); |
1027 | if (!tty_info->pty_info) | |
e8bd4e43 | 1028 | return -1; |
e8bd4e43 SH |
1029 | |
1030 | for (i = 0; i < conf->tty; i++) { | |
1031 | struct lxc_pty_info *pty_info = &tty_info->pty_info[i]; | |
1032 | pty_info->busy = 0; | |
1033 | if (recv_fd(sock, &pty_info->slave) < 0 || | |
408da065 CB |
1034 | recv_fd(sock, &pty_info->master) < 0) { |
1035 | ERROR("Error receiving tty info from child process."); | |
e8bd4e43 SH |
1036 | return -1; |
1037 | } | |
1038 | } | |
1039 | tty_info->nbtty = conf->tty; | |
1040 | ||
1041 | return 0; | |
1042 | } | |
1043 | ||
f813849c TA |
1044 | void resolve_clone_flags(struct lxc_handler *handler) |
1045 | { | |
1046 | handler->clone_flags = CLONE_NEWPID | CLONE_NEWNS; | |
1047 | ||
408da065 | 1048 | if (!lxc_list_empty(&handler->conf->id_map)) |
f813849c | 1049 | handler->clone_flags |= CLONE_NEWUSER; |
f813849c TA |
1050 | |
1051 | if (handler->conf->inherit_ns_fd[LXC_NS_NET] == -1) { | |
408da065 | 1052 | if (!lxc_requests_empty_network(handler)) |
f813849c TA |
1053 | handler->clone_flags |= CLONE_NEWNET; |
1054 | } else { | |
198cbbaa | 1055 | INFO("Inheriting a NET namespace."); |
f813849c TA |
1056 | } |
1057 | ||
408da065 | 1058 | if (handler->conf->inherit_ns_fd[LXC_NS_IPC] == -1) |
f813849c | 1059 | handler->clone_flags |= CLONE_NEWIPC; |
408da065 | 1060 | else |
198cbbaa | 1061 | INFO("Inheriting an IPC namespace."); |
f813849c | 1062 | |
408da065 | 1063 | if (handler->conf->inherit_ns_fd[LXC_NS_UTS] == -1) |
f813849c | 1064 | handler->clone_flags |= CLONE_NEWUTS; |
408da065 | 1065 | else |
198cbbaa | 1066 | INFO("Inheriting a UTS namespace."); |
f813849c TA |
1067 | } |
1068 | ||
74a3920a | 1069 | static int lxc_spawn(struct lxc_handler *handler) |
59eb99ba | 1070 | { |
b98f7d6e | 1071 | int failed_before_rename = 0; |
ffe1e01a | 1072 | const char *name = handler->name; |
7e4dfe0b | 1073 | bool cgroups_connected = false; |
9f30a190 | 1074 | int saved_ns_fd[LXC_NS_MAX]; |
5b1e83cb | 1075 | int preserve_mask = 0, i, flags; |
658979c5 | 1076 | int netpipepair[2], nveths; |
9f30a190 | 1077 | |
2f2623ec SH |
1078 | netpipe = -1; |
1079 | ||
9f30a190 | 1080 | for (i = 0; i < LXC_NS_MAX; i++) |
6c544cb3 | 1081 | if (handler->conf->inherit_ns_fd[i] != -1) |
9f30a190 | 1082 | preserve_mask |= ns_info[i].clone_flag; |
50e98013 | 1083 | |
3c22086f | 1084 | if (lxc_sync_init(handler)) |
9d7f9e52 | 1085 | return -1; |
0ad19a3f | 1086 | |
e8bd4e43 SH |
1087 | if (socketpair(AF_UNIX, SOCK_DGRAM, 0, handler->ttysock) < 0) { |
1088 | lxc_sync_fini(handler); | |
1089 | return -1; | |
1090 | } | |
1091 | ||
f813849c | 1092 | resolve_clone_flags(handler); |
9f30a190 | 1093 | |
f813849c | 1094 | if (handler->clone_flags & CLONE_NEWNET) { |
26b797f3 SH |
1095 | if (!lxc_list_empty(&handler->conf->network)) { |
1096 | ||
9f30a190 MM |
1097 | /* Find gateway addresses from the link device, which is |
1098 | * no longer accessible inside the container. Do this | |
1099 | * before creating network interfaces, since goto | |
408da065 CB |
1100 | * out_delete_net does not work before lxc_clone. |
1101 | */ | |
9f30a190 | 1102 | if (lxc_find_gateway_addresses(handler)) { |
408da065 | 1103 | ERROR("Failed to find gateway addresses."); |
9f30a190 MM |
1104 | lxc_sync_fini(handler); |
1105 | return -1; | |
1106 | } | |
1107 | ||
408da065 CB |
1108 | /* That should be done before the clone because we will |
1109 | * fill the netdev index and use them in the child. | |
9f30a190 MM |
1110 | */ |
1111 | if (lxc_create_network(handler)) { | |
408da065 | 1112 | ERROR("Failed to create the network."); |
9f30a190 MM |
1113 | lxc_sync_fini(handler); |
1114 | return -1; | |
1115 | } | |
19a26f82 MK |
1116 | } |
1117 | ||
9f30a190 | 1118 | if (save_phys_nics(handler->conf)) { |
408da065 | 1119 | ERROR("Failed to save physical nic info."); |
9f30a190 | 1120 | goto out_abort; |
82d5ae15 DL |
1121 | } |
1122 | } | |
1123 | ||
d4ef7c50 | 1124 | if (!cgroup_init(handler)) { |
408da065 | 1125 | ERROR("Failed initializing cgroup support."); |
33ad9f1a CS |
1126 | goto out_delete_net; |
1127 | } | |
1128 | ||
7e4dfe0b SH |
1129 | cgroups_connected = true; |
1130 | ||
d4ef7c50 | 1131 | if (!cgroup_create(handler)) { |
408da065 | 1132 | ERROR("Failed creating cgroups."); |
47d8fb3b CS |
1133 | goto out_delete_net; |
1134 | } | |
1135 | ||
408da065 CB |
1136 | /* If the rootfs is not a blockdev, prevent the container from marking |
1137 | * it readonly. | |
1138 | * If the container is unprivileged then skip rootfs pinning. | |
0c547523 | 1139 | */ |
5e32a990 ÇO |
1140 | if (lxc_list_empty(&handler->conf->id_map)) { |
1141 | handler->pinfd = pin_rootfs(handler->conf->rootfs.path); | |
1142 | if (handler->pinfd == -1) | |
408da065 | 1143 | INFO("Failed to pin the rootfs for container \"%s\".", handler->name); |
5e32a990 | 1144 | } |
0c547523 | 1145 | |
4d8ac866 | 1146 | if (!preserve_ns(saved_ns_fd, preserve_mask, getpid())) |
cd43d2d1 | 1147 | goto out_delete_net; |
4d8ac866 | 1148 | |
cd43d2d1 SH |
1149 | if (attach_ns(handler->conf->inherit_ns_fd) < 0) |
1150 | goto out_delete_net; | |
9f30a190 | 1151 | |
658979c5 SH |
1152 | if (am_unpriv() && (nveths = count_veths(&handler->conf->network))) { |
1153 | if (pipe(netpipepair) < 0) { | |
408da065 | 1154 | SYSERROR("Failed to create pipe."); |
658979c5 SH |
1155 | goto out_delete_net; |
1156 | } | |
408da065 | 1157 | /* Store netpipe in the global var for do_start's use. */ |
658979c5 SH |
1158 | netpipe = netpipepair[0]; |
1159 | } | |
1160 | ||
408da065 | 1161 | /* Create a process in a new set of namespaces. */ |
5b1e83cb | 1162 | flags = handler->clone_flags; |
408da065 CB |
1163 | if (handler->clone_flags & CLONE_NEWUSER) { |
1164 | /* If CLONE_NEWUSER and CLONE_NEWNET was requested, we need to | |
1165 | * clone a new user namespace first and only later unshare our | |
1166 | * network namespace to ensure that network devices ownership is | |
1167 | * set up correctly. | |
1168 | */ | |
5b1e83cb | 1169 | flags &= ~CLONE_NEWNET; |
408da065 | 1170 | } |
9fac8fbb | 1171 | handler->pid = lxc_clone(do_start, handler, flags); |
59eb99ba | 1172 | if (handler->pid < 0) { |
408da065 | 1173 | SYSERROR("Failed to clone a new set of namespaces."); |
7fef7a06 | 1174 | goto out_delete_net; |
0ad19a3f | 1175 | } |
408da065 | 1176 | INFO("Cloned a set of new namespaces."); |
0ad19a3f | 1177 | |
4d8ac866 CB |
1178 | if (!preserve_ns(handler->nsfd, handler->clone_flags | preserve_mask, handler->pid)) |
1179 | INFO("Failed to preserve namespace for lxc.hook.stop."); | |
b6b2b194 | 1180 | |
cd43d2d1 | 1181 | if (attach_ns(saved_ns_fd)) |
408da065 | 1182 | WARN("Failed to restore saved namespaces."); |
9f30a190 | 1183 | |
3c22086f CLG |
1184 | lxc_sync_fini_child(handler); |
1185 | ||
408da065 CB |
1186 | /* Map the container uids. The container became an invalid userid the |
1187 | * moment it was cloned with CLONE_NEWUSER. This call doesn't change | |
1188 | * anything immediately, but allows the container to setuid(0) (0 being | |
1189 | * mapped to something else on the host.) later to become a valid uid | |
1190 | * again. | |
1191 | */ | |
5b1e83cb | 1192 | if (lxc_map_ids(&handler->conf->id_map, handler->pid)) { |
408da065 | 1193 | ERROR("Failed to set up id mapping."); |
5b1e83cb SH |
1194 | goto out_delete_net; |
1195 | } | |
1196 | ||
1197 | if (lxc_sync_wake_child(handler, LXC_SYNC_STARTUP)) { | |
99a6af52 | 1198 | failed_before_rename = 1; |
5b1e83cb SH |
1199 | goto out_delete_net; |
1200 | } | |
1201 | ||
1202 | if (lxc_sync_wait_child(handler, LXC_SYNC_CONFIGURE)) { | |
1203 | failed_before_rename = 1; | |
1204 | goto out_delete_net; | |
1205 | } | |
0ad19a3f | 1206 | |
d4ef7c50 | 1207 | if (!cgroup_create_legacy(handler)) { |
408da065 | 1208 | ERROR("Failed to setup legacy cgroups for container \"%s\".", name); |
ae5c8b8e | 1209 | goto out_delete_net; |
33ad9f1a | 1210 | } |
9daf6f5d | 1211 | if (!cgroup_setup_limits(handler, false)) { |
408da065 | 1212 | ERROR("Failed to setup cgroup limits for container \"%s\".", name); |
6031a6e5 DE |
1213 | goto out_delete_net; |
1214 | } | |
1215 | ||
d4ef7c50 | 1216 | if (!cgroup_enter(handler)) |
7fef7a06 | 1217 | goto out_delete_net; |
218d4250 | 1218 | |
0996e18a SH |
1219 | if (!cgroup_chown(handler)) |
1220 | goto out_delete_net; | |
1221 | ||
99a6af52 MN |
1222 | if (failed_before_rename) |
1223 | goto out_delete_net; | |
1224 | ||
408da065 | 1225 | /* Create the network configuration. */ |
d5088cf2 | 1226 | if (handler->clone_flags & CLONE_NEWNET) { |
c43cbc04 | 1227 | if (lxc_assign_network(handler->lxcpath, handler->name, |
408da065 CB |
1228 | &handler->conf->network, handler->pid)) { |
1229 | ERROR("Failed to create the configured network."); | |
7fef7a06 | 1230 | goto out_delete_net; |
82d5ae15 | 1231 | } |
0ad19a3f | 1232 | } |
1233 | ||
658979c5 SH |
1234 | if (netpipe != -1) { |
1235 | struct lxc_list *iterator; | |
1236 | struct lxc_netdev *netdev; | |
1237 | ||
1238 | close(netpipe); | |
1239 | lxc_list_for_each(iterator, &handler->conf->network) { | |
1240 | netdev = iterator->elem; | |
1241 | if (netdev->type != LXC_NET_VETH) | |
1242 | continue; | |
1243 | if (write(netpipepair[1], netdev->name, IFNAMSIZ) != IFNAMSIZ) { | |
408da065 | 1244 | ERROR("Error writing veth name to container."); |
658979c5 SH |
1245 | goto out_delete_net; |
1246 | } | |
1247 | } | |
1248 | close(netpipepair[1]); | |
1249 | } | |
1250 | ||
408da065 CB |
1251 | /* Tell the child to continue its initialization. We'll get |
1252 | * LXC_SYNC_CGROUP when it is ready for us to setup cgroups. | |
3c22086f CLG |
1253 | */ |
1254 | if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CONFIGURE)) | |
544a48a0 SH |
1255 | goto out_delete_net; |
1256 | ||
9daf6f5d | 1257 | if (!cgroup_setup_limits(handler, true)) { |
408da065 | 1258 | ERROR("Failed to setup the devices cgroup for container \"%s\".", name); |
b98f7d6e | 1259 | goto out_delete_net; |
544a48a0 SH |
1260 | } |
1261 | ||
73d28d42 | 1262 | cgroup_disconnect(); |
7e4dfe0b | 1263 | cgroups_connected = false; |
73d28d42 | 1264 | |
408da065 | 1265 | /* Read tty fds allocated by child. */ |
e8bd4e43 | 1266 | if (recv_ttys_from_child(handler) < 0) { |
408da065 | 1267 | ERROR("Failed to receive tty info from child process."); |
e8bd4e43 SH |
1268 | goto out_delete_net; |
1269 | } | |
1270 | ||
408da065 CB |
1271 | /* Tell the child to complete its initialization and wait for it to exec |
1272 | * or return an error. (The child will never return | |
1273 | * LXC_SYNC_POST_CGROUP+1. It will either close the sync pipe, causing | |
1274 | * lxc_sync_barrier_child to return success, or return a different | |
1275 | * value, causing us to error out). | |
544a48a0 SH |
1276 | */ |
1277 | if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CGROUP)) | |
3c22086f | 1278 | return -1; |
0ad19a3f | 1279 | |
cc28d0b0 SH |
1280 | if (detect_shared_rootfs()) |
1281 | umount2(handler->conf->rootfs.mount, MNT_DETACH); | |
1282 | ||
23c53af9 | 1283 | if (handler->ops->post_start(handler, handler->data)) |
e6126dbe MN |
1284 | goto out_abort; |
1285 | ||
25c2aca5 | 1286 | if (lxc_set_state(name, handler, RUNNING)) { |
408da065 CB |
1287 | ERROR("Failed to set state for container \"%s\" to \"%s\".", name, |
1288 | lxc_state2str(RUNNING)); | |
59eb99ba | 1289 | goto out_abort; |
3f21c114 | 1290 | } |
22ebac19 | 1291 | |
3c22086f | 1292 | lxc_sync_fini(handler); |
738d0deb | 1293 | handler->netnsfd = lxc_preserve_ns(handler->pid, "net"); |
0c547523 | 1294 | |
e6126dbe | 1295 | return 0; |
1ac470c0 | 1296 | |
7fef7a06 | 1297 | out_delete_net: |
7e4dfe0b SH |
1298 | if (cgroups_connected) |
1299 | cgroup_disconnect(); | |
d5088cf2 | 1300 | if (handler->clone_flags & CLONE_NEWNET) |
74a2b586 | 1301 | lxc_delete_network(handler); |
59eb99ba DL |
1302 | out_abort: |
1303 | lxc_abort(name, handler); | |
3c22086f | 1304 | lxc_sync_fini(handler); |
5c068da9 SH |
1305 | if (handler->pinfd >= 0) { |
1306 | close(handler->pinfd); | |
1307 | handler->pinfd = -1; | |
1308 | } | |
1309 | ||
b79fcd86 | 1310 | return -1; |
59eb99ba | 1311 | } |
0ad19a3f | 1312 | |
ee70bf78 | 1313 | int __lxc_start(const char *name, struct lxc_conf *conf, |
507cee36 TA |
1314 | struct lxc_operations* ops, void *data, const char *lxcpath, |
1315 | bool backgrounded) | |
59eb99ba | 1316 | { |
3a0f472d | 1317 | struct lxc_handler *handler; |
e043236e | 1318 | int err = -1; |
59eb99ba | 1319 | int status; |
358daf49 | 1320 | bool removed_all_netdevs = true; |
80090207 | 1321 | |
13f5be62 | 1322 | handler = lxc_init(name, conf, lxcpath); |
3a0f472d | 1323 | if (!handler) { |
408da065 | 1324 | ERROR("Failed to initialize container \"%s\".", name); |
66aeffc7 | 1325 | return -1; |
0ad19a3f | 1326 | } |
ee70bf78 CLG |
1327 | handler->ops = ops; |
1328 | handler->data = data; | |
507cee36 | 1329 | handler->backgrounded = backgrounded; |
738d0deb | 1330 | handler->netnsfd = -1; |
e6126dbe | 1331 | |
b60ed720 | 1332 | if (must_drop_cap_sys_boot(handler->conf)) { |
495d2046 | 1333 | #if HAVE_SYS_CAPABILITY_H |
408da065 | 1334 | DEBUG("Dropping CAP_SYS_BOOT capability."); |
495d2046 | 1335 | #else |
408da065 | 1336 | DEBUG("Not dropping CAP_SYS_BOOT capability as capabilities aren't supported."); |
495d2046 | 1337 | #endif |
69182a31 | 1338 | } else { |
408da065 | 1339 | DEBUG("Not dropping CAP_SYS_BOOT or watching utmp."); |
69182a31 SH |
1340 | handler->conf->need_utmp_watch = 0; |
1341 | } | |
1342 | ||
76a26f55 | 1343 | if (!attach_block_device(handler->conf)) { |
408da065 | 1344 | ERROR("Failed to attach block device."); |
76a26f55 SH |
1345 | goto out_fini_nonet; |
1346 | } | |
1347 | ||
35120d9c | 1348 | if (geteuid() == 0 && !lxc_list_empty(&conf->id_map)) { |
408da065 | 1349 | /* If the backing store is a device, mount it here and now. */ |
35120d9c SH |
1350 | if (rootfs_is_blockdev(conf)) { |
1351 | if (unshare(CLONE_NEWNS) < 0) { | |
408da065 | 1352 | ERROR("Failed to unshare CLONE_NEWNS."); |
35120d9c SH |
1353 | goto out_fini_nonet; |
1354 | } | |
408da065 CB |
1355 | INFO("Unshared CLONE_NEWNS."); |
1356 | ||
6a0c909a | 1357 | remount_all_slave(); |
35120d9c | 1358 | if (do_rootfs_setup(conf, name, lxcpath) < 0) { |
408da065 | 1359 | ERROR("Error setting up rootfs mount as root before spawn."); |
35120d9c SH |
1360 | goto out_fini_nonet; |
1361 | } | |
408da065 | 1362 | INFO("Set up container rootfs as host root."); |
35120d9c SH |
1363 | } |
1364 | } | |
1365 | ||
23c53af9 | 1366 | err = lxc_spawn(handler); |
59eb99ba | 1367 | if (err) { |
408da065 | 1368 | ERROR("Failed to spawn container \"%s\".", name); |
76a26f55 | 1369 | goto out_detach_blockdev; |
0ad19a3f | 1370 | } |
1371 | ||
8bee8851 WB |
1372 | handler->conf->reboot = 0; |
1373 | ||
3a0f472d | 1374 | err = lxc_poll(name, handler); |
e043236e | 1375 | if (err) { |
408da065 | 1376 | ERROR("LXC mainloop exited with error: %d.", err); |
738d0deb CB |
1377 | if (handler->netnsfd >= 0) { |
1378 | close(handler->netnsfd); | |
1379 | handler->netnsfd = -1; | |
1380 | } | |
59eb99ba DL |
1381 | goto out_abort; |
1382 | } | |
0ad19a3f | 1383 | |
3a0f472d | 1384 | while (waitpid(handler->pid, &status, 0) < 0 && errno == EINTR) |
1bc5cc8c | 1385 | continue; |
e043236e | 1386 | |
408da065 CB |
1387 | /* If the child process exited but was not signaled, it didn't call |
1388 | * reboot. This should mean it was an lxc-execute which simply exited. | |
1389 | * In any case, treat it as a 'halt'. | |
8b004f07 | 1390 | */ |
d028235d | 1391 | if (WIFSIGNALED(status)) { |
8b004f07 SH |
1392 | switch(WTERMSIG(status)) { |
1393 | case SIGINT: /* halt */ | |
408da065 | 1394 | DEBUG("Container \"%s\" is halting.", name); |
8b004f07 SH |
1395 | break; |
1396 | case SIGHUP: /* reboot */ | |
408da065 | 1397 | DEBUG("Container \"%s\" is rebooting.", name); |
8b004f07 SH |
1398 | handler->conf->reboot = 1; |
1399 | break; | |
c2b9bd9e | 1400 | case SIGSYS: /* seccomp */ |
408da065 | 1401 | DEBUG("Container \"%s\" violated its seccomp policy.", name); |
c2b9bd9e | 1402 | break; |
8b004f07 | 1403 | default: |
408da065 | 1404 | DEBUG("Unknown exit status for container \"%s\" init %d.", name, WTERMSIG(status)); |
8b004f07 SH |
1405 | break; |
1406 | } | |
d028235d | 1407 | } |
828695d9 | 1408 | |
ce5782df | 1409 | DEBUG("Pushing physical nics back to host namespace"); |
738d0deb | 1410 | lxc_restore_phys_nics_to_netns(handler->netnsfd, handler->conf); |
ce5782df | 1411 | |
408da065 | 1412 | DEBUG("Tearing down virtual network devices used by container \"%s\".", name); |
358daf49 | 1413 | removed_all_netdevs = lxc_delete_network(handler); |
ce5782df | 1414 | |
5c068da9 SH |
1415 | if (handler->pinfd >= 0) { |
1416 | close(handler->pinfd); | |
1417 | handler->pinfd = -1; | |
1418 | } | |
1419 | ||
1787abca | 1420 | lxc_monitor_send_exit_code(name, status, handler->lxcpath); |
3a0f472d | 1421 | err = lxc_error_set_and_log(handler->pid, status); |
9d7f9e52 | 1422 | out_fini: |
358daf49 | 1423 | if (!removed_all_netdevs) { |
408da065 | 1424 | DEBUG("Failed tearing down network devices used by container. Trying again!"); |
358daf49 CB |
1425 | removed_all_netdevs = lxc_delete_network(handler); |
1426 | if (!removed_all_netdevs) | |
1427 | DEBUG("Failed tearing down network devices used by container. Not trying again!"); | |
1428 | } | |
74a2b586 | 1429 | |
76a26f55 SH |
1430 | out_detach_blockdev: |
1431 | detach_block_device(handler->conf); | |
1432 | ||
74a2b586 | 1433 | out_fini_nonet: |
3a0f472d | 1434 | lxc_fini(name, handler); |
0ad19a3f | 1435 | return err; |
1436 | ||
59eb99ba | 1437 | out_abort: |
3a0f472d | 1438 | lxc_abort(name, handler); |
9d7f9e52 | 1439 | goto out_fini; |
0ad19a3f | 1440 | } |
ee70bf78 CLG |
1441 | |
1442 | struct start_args { | |
1443 | char *const *argv; | |
1444 | }; | |
1445 | ||
1446 | static int start(struct lxc_handler *handler, void* data) | |
1447 | { | |
1448 | struct start_args *arg = data; | |
1449 | ||
408da065 | 1450 | NOTICE("Exec'ing \"%s\".", arg->argv[0]); |
ee70bf78 CLG |
1451 | |
1452 | execvp(arg->argv[0], arg->argv); | |
408da065 | 1453 | SYSERROR("Failed to exec \"%s\".", arg->argv[0]); |
ee70bf78 CLG |
1454 | return 0; |
1455 | } | |
1456 | ||
1457 | static int post_start(struct lxc_handler *handler, void* data) | |
1458 | { | |
1459 | struct start_args *arg = data; | |
1460 | ||
408da065 | 1461 | NOTICE("Started \"%s\" with pid \"%d\".", arg->argv[0], handler->pid); |
ee70bf78 CLG |
1462 | return 0; |
1463 | } | |
1464 | ||
1465 | static struct lxc_operations start_ops = { | |
1466 | .start = start, | |
1467 | .post_start = post_start | |
1468 | }; | |
1469 | ||
13f5be62 | 1470 | int lxc_start(const char *name, char *const argv[], struct lxc_conf *conf, |
507cee36 | 1471 | const char *lxcpath, bool backgrounded) |
ee70bf78 CLG |
1472 | { |
1473 | struct start_args start_arg = { | |
1474 | .argv = argv, | |
1475 | }; | |
1476 | ||
828695d9 | 1477 | conf->need_utmp_watch = 1; |
507cee36 | 1478 | return __lxc_start(name, conf, &start_ops, &start_arg, lxcpath, backgrounded); |
ee70bf78 | 1479 | } |
28272964 CB |
1480 | |
1481 | static void lxc_destroy_container_on_signal(struct lxc_handler *handler, | |
1482 | const char *name) | |
1483 | { | |
1484 | char destroy[MAXPATHLEN]; | |
1485 | bool bret = true; | |
1486 | int ret = 0; | |
f01f7975 | 1487 | struct lxc_container *c; |
df31363a | 1488 | if (handler->conf->rootfs.path && handler->conf->rootfs.mount) { |
28272964 CB |
1489 | bret = do_destroy_container(handler->conf); |
1490 | if (!bret) { | |
408da065 | 1491 | ERROR("Error destroying rootfs for container \"%s\".", name); |
28272964 CB |
1492 | return; |
1493 | } | |
1494 | } | |
408da065 | 1495 | INFO("Destroyed rootfs for container \"%s\".", name); |
28272964 CB |
1496 | |
1497 | ret = snprintf(destroy, MAXPATHLEN, "%s/%s", handler->lxcpath, name); | |
1498 | if (ret < 0 || ret >= MAXPATHLEN) { | |
408da065 | 1499 | ERROR("Error destroying directory for container \"%s\".", name); |
28272964 CB |
1500 | return; |
1501 | } | |
1502 | ||
f01f7975 CB |
1503 | c = lxc_container_new(name, handler->lxcpath); |
1504 | if (c) { | |
1505 | if (container_disk_lock(c)) { | |
408da065 | 1506 | INFO("Could not update lxc_snapshots file."); |
f01f7975 CB |
1507 | lxc_container_put(c); |
1508 | } else { | |
1509 | mod_all_rdeps(c, false); | |
1510 | container_disk_unlock(c); | |
1511 | lxc_container_put(c); | |
1512 | } | |
1513 | } | |
1514 | ||
28272964 CB |
1515 | if (am_unpriv()) |
1516 | ret = userns_exec_1(handler->conf, lxc_rmdir_onedev_wrapper, destroy); | |
1517 | else | |
1518 | ret = lxc_rmdir_onedev(destroy, NULL); | |
1519 | ||
1520 | if (ret < 0) { | |
408da065 | 1521 | ERROR("Error destroying directory for container \"%s\".", name); |
28272964 CB |
1522 | return; |
1523 | } | |
408da065 | 1524 | INFO("Destroyed directory for container \"%s\".", name); |
28272964 CB |
1525 | } |
1526 | ||
1527 | static int lxc_rmdir_onedev_wrapper(void *data) | |
1528 | { | |
1529 | char *arg = (char *) data; | |
1530 | return lxc_rmdir_onedev(arg, NULL); | |
1531 | } | |
1532 | ||
1533 | static bool do_destroy_container(struct lxc_conf *conf) { | |
d028235d SG |
1534 | if (am_unpriv()) { |
1535 | if (userns_exec_1(conf, bdev_destroy_wrapper, conf) < 0) | |
1536 | return false; | |
1537 | return true; | |
1538 | } | |
1539 | return bdev_destroy(conf); | |
28272964 | 1540 | } |