]>
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 | ||
f549edcc GK |
24 | #include "config.h" |
25 | ||
0ad19a3f | 26 | #include <stdio.h> |
0ad19a3f | 27 | #include <string.h> |
28 | #include <stdlib.h> | |
29 | #include <dirent.h> | |
30 | #include <errno.h> | |
31 | #include <unistd.h> | |
32 | #include <signal.h> | |
b0a33c1e | 33 | #include <fcntl.h> |
c476bdce | 34 | #include <grp.h> |
37515ebd | 35 | #include <poll.h> |
0ad19a3f | 36 | #include <sys/param.h> |
37 | #include <sys/file.h> | |
f4d507d5 | 38 | #include <sys/mount.h> |
b4f8660e | 39 | #include <sys/stat.h> |
0ad19a3f | 40 | #include <sys/types.h> |
8173e600 | 41 | #include <sys/socket.h> |
0ad19a3f | 42 | #include <sys/prctl.h> |
ddceb1f9 | 43 | #include <sys/types.h> |
0ad19a3f | 44 | #include <sys/wait.h> |
b0a33c1e | 45 | #include <sys/un.h> |
8173e600 | 46 | #include <sys/syscall.h> |
ff218c25 | 47 | |
495d2046 SG |
48 | #if HAVE_SYS_CAPABILITY_H |
49 | #include <sys/capability.h> | |
50 | #endif | |
51 | ||
656994bb MH |
52 | #if !HAVE_DECL_PR_CAPBSET_DROP |
53 | #define PR_CAPBSET_DROP 24 | |
54 | #endif | |
55 | ||
63376d7d DL |
56 | #include "start.h" |
57 | #include "conf.h" | |
58 | #include "log.h" | |
563f2f2c | 59 | #include "cgroup.h" |
e2bcd7db | 60 | #include "error.h" |
b0a33c1e | 61 | #include "af_unix.h" |
62 | #include "mainloop.h" | |
63376d7d | 63 | #include "utils.h" |
565c2d76 | 64 | #include "lxcutmp.h" |
63376d7d | 65 | #include "monitor.h" |
96fa1ff0 | 66 | #include "commands.h" |
63376d7d | 67 | #include "console.h" |
3c22086f | 68 | #include "sync.h" |
f549edcc | 69 | #include "namespace.h" |
0d0527a9 | 70 | #include "lxcseccomp.h" |
8173e600 | 71 | #include "caps.h" |
76a26f55 | 72 | #include "bdev.h" |
fe4de9a6 | 73 | #include "lsm/lsm.h" |
f01f7975 | 74 | #include "lxclock.h" |
36eb9bde CLG |
75 | |
76 | lxc_log_define(lxc_start, lxc); | |
77 | ||
9f30a190 MM |
78 | const struct ns_info ns_info[LXC_NS_MAX] = { |
79 | [LXC_NS_MNT] = {"mnt", CLONE_NEWNS}, | |
80 | [LXC_NS_PID] = {"pid", CLONE_NEWPID}, | |
81 | [LXC_NS_UTS] = {"uts", CLONE_NEWUTS}, | |
82 | [LXC_NS_IPC] = {"ipc", CLONE_NEWIPC}, | |
83 | [LXC_NS_USER] = {"user", CLONE_NEWUSER}, | |
84 | [LXC_NS_NET] = {"net", CLONE_NEWNET} | |
85 | }; | |
86 | ||
f01f7975 | 87 | extern void mod_all_rdeps(struct lxc_container *c, bool inc); |
28272964 CB |
88 | static bool do_destroy_container(struct lxc_conf *conf); |
89 | static int lxc_rmdir_onedev_wrapper(void *data); | |
90 | static void lxc_destroy_container_on_signal(struct lxc_handler *handler, | |
91 | const char *name); | |
92 | ||
c8154066 SH |
93 | static void print_top_failing_dir(const char *path) |
94 | { | |
95 | size_t len = strlen(path); | |
96 | char *copy = alloca(len+1), *p, *e, saved; | |
97 | strcpy(copy, path); | |
98 | ||
99 | p = copy; | |
100 | e = copy + len; | |
101 | while (p < e) { | |
102 | while (p < e && *p == '/') p++; | |
103 | while (p < e && *p != '/') p++; | |
c8154066 SH |
104 | saved = *p; |
105 | *p = '\0'; | |
106 | if (access(copy, X_OK)) { | |
107 | SYSERROR("could not access %s. Please grant it 'x' " \ | |
108 | "access, or add an ACL for the container root.", | |
109 | copy); | |
110 | return; | |
111 | } | |
112 | *p = saved; | |
113 | } | |
114 | } | |
115 | ||
9f30a190 MM |
116 | static void close_ns(int ns_fd[LXC_NS_MAX]) { |
117 | int i; | |
118 | ||
9f30a190 MM |
119 | for (i = 0; i < LXC_NS_MAX; i++) { |
120 | if (ns_fd[i] > -1) { | |
121 | close(ns_fd[i]); | |
122 | ns_fd[i] = -1; | |
123 | } | |
124 | } | |
9f30a190 MM |
125 | } |
126 | ||
62d05d9b SH |
127 | /* |
128 | * preserve_ns: open /proc/@pid/ns/@ns for each namespace specified | |
129 | * in clone_flags. | |
130 | * Return true on success, false on failure. On failure, leave an error | |
131 | * message in *errmsg, which caller must free. | |
132 | */ | |
133 | static | |
134 | bool preserve_ns(int ns_fd[LXC_NS_MAX], int clone_flags, pid_t pid, char **errmsg) { | |
135 | int i, ret; | |
9f30a190 MM |
136 | char path[MAXPATHLEN]; |
137 | ||
9f30a190 MM |
138 | for (i = 0; i < LXC_NS_MAX; i++) |
139 | ns_fd[i] = -1; | |
140 | ||
035a38fc WB |
141 | snprintf(path, MAXPATHLEN, "/proc/%d/ns", pid); |
142 | if (access(path, X_OK)) { | |
62d05d9b SH |
143 | if (asprintf(errmsg, "Kernel does not support setns.") == -1) |
144 | *errmsg = NULL; | |
145 | return false; | |
cd43d2d1 SH |
146 | } |
147 | ||
9f30a190 MM |
148 | for (i = 0; i < LXC_NS_MAX; i++) { |
149 | if ((clone_flags & ns_info[i].clone_flag) == 0) | |
150 | continue; | |
035a38fc WB |
151 | snprintf(path, MAXPATHLEN, "/proc/%d/ns/%s", pid, |
152 | ns_info[i].proc_name); | |
9f30a190 | 153 | ns_fd[i] = open(path, O_RDONLY | O_CLOEXEC); |
9f30a190 MM |
154 | if (ns_fd[i] < 0) |
155 | goto error; | |
156 | } | |
157 | ||
62d05d9b | 158 | return true; |
9f30a190 MM |
159 | |
160 | error: | |
62d05d9b SH |
161 | if (errno == ENOENT) { |
162 | ret = asprintf(errmsg, "Kernel does not support setns for %s", | |
163 | ns_info[i].proc_name); | |
164 | } else { | |
165 | ret = asprintf(errmsg, "Failed to open %s: %s", | |
166 | path, strerror(errno)); | |
167 | } | |
168 | if (ret == -1) | |
169 | *errmsg = NULL; | |
9f30a190 | 170 | close_ns(ns_fd); |
62d05d9b | 171 | return false; |
9f30a190 MM |
172 | } |
173 | ||
174 | static int attach_ns(const int ns_fd[LXC_NS_MAX]) { | |
175 | int i; | |
176 | ||
177 | for (i = 0; i < LXC_NS_MAX; i++) { | |
178 | if (ns_fd[i] < 0) | |
179 | continue; | |
180 | ||
181 | if (setns(ns_fd[i], 0) != 0) | |
182 | goto error; | |
183 | } | |
184 | return 0; | |
185 | ||
186 | error: | |
187 | SYSERROR("failed to set namespace '%s'", ns_info[i].proc_name); | |
188 | return -1; | |
189 | } | |
190 | ||
80090207 CLG |
191 | static int match_fd(int fd) |
192 | { | |
193 | return (fd == 0 || fd == 1 || fd == 2); | |
194 | } | |
195 | ||
d2cf4c37 SH |
196 | /* |
197 | * Check for any fds we need to close | |
198 | * * if fd_to_ignore != -1, then if we find that fd open we will ignore it. | |
199 | * * By default we warn about open fds we find. | |
200 | * * If closeall is true, we will close open fds. | |
201 | * * If lxc-start was passed "-C", then conf->close_all_fds will be true, | |
202 | * in which case we also close all open fds. | |
203 | * * A daemonized container will always pass closeall=true. | |
204 | */ | |
205 | int lxc_check_inherited(struct lxc_conf *conf, bool closeall, int fd_to_ignore) | |
80090207 CLG |
206 | { |
207 | struct dirent dirent, *direntp; | |
208 | int fd, fddir; | |
209 | DIR *dir; | |
80090207 | 210 | |
d2cf4c37 SH |
211 | if (conf && conf->close_all_fds) |
212 | closeall = true; | |
213 | ||
b119f362 | 214 | restart: |
80090207 CLG |
215 | dir = opendir("/proc/self/fd"); |
216 | if (!dir) { | |
217 | WARN("failed to open directory: %m"); | |
218 | return -1; | |
219 | } | |
220 | ||
221 | fddir = dirfd(dir); | |
222 | ||
223 | while (!readdir_r(dir, &dirent, &direntp)) { | |
80090207 CLG |
224 | if (!direntp) |
225 | break; | |
226 | ||
227 | if (!strcmp(direntp->d_name, ".")) | |
228 | continue; | |
229 | ||
230 | if (!strcmp(direntp->d_name, "..")) | |
231 | continue; | |
232 | ||
233 | fd = atoi(direntp->d_name); | |
234 | ||
f2faa8fa | 235 | if (fd == fddir || fd == lxc_log_fd || fd == fd_to_ignore) |
80090207 CLG |
236 | continue; |
237 | ||
858377e4 SH |
238 | if (current_config && fd == current_config->logfd) |
239 | continue; | |
240 | ||
80090207 CLG |
241 | if (match_fd(fd)) |
242 | continue; | |
80090207 | 243 | |
d2cf4c37 | 244 | if (closeall) { |
b119f362 SH |
245 | close(fd); |
246 | closedir(dir); | |
247 | INFO("closed inherited fd %d", fd); | |
248 | goto restart; | |
249 | } | |
92c7f629 | 250 | WARN("inherited fd %d", fd); |
80090207 CLG |
251 | } |
252 | ||
92c7f629 GK |
253 | closedir(dir); /* cannot fail */ |
254 | return 0; | |
80090207 CLG |
255 | } |
256 | ||
83ee7875 | 257 | static int setup_signal_fd(sigset_t *oldmask) |
b0a33c1e | 258 | { |
259 | sigset_t mask; | |
260 | int fd; | |
261 | ||
f3304a29 FW |
262 | /* Block everything except serious error signals */ |
263 | if (sigfillset(&mask) || | |
264 | sigdelset(&mask, SIGILL) || | |
265 | sigdelset(&mask, SIGSEGV) || | |
266 | sigdelset(&mask, SIGBUS) || | |
b5159817 | 267 | sigdelset(&mask, SIGWINCH) || |
f3304a29 FW |
268 | sigprocmask(SIG_BLOCK, &mask, oldmask)) { |
269 | SYSERROR("failed to set signal mask"); | |
b0a33c1e | 270 | return -1; |
271 | } | |
272 | ||
273 | fd = signalfd(-1, &mask, 0); | |
274 | if (fd < 0) { | |
36eb9bde | 275 | SYSERROR("failed to create the signal fd"); |
b0a33c1e | 276 | return -1; |
277 | } | |
278 | ||
279 | if (fcntl(fd, F_SETFD, FD_CLOEXEC)) { | |
36eb9bde | 280 | SYSERROR("failed to set sigfd to close-on-exec"); |
b0a33c1e | 281 | close(fd); |
282 | return -1; | |
283 | } | |
284 | ||
1ac470c0 DL |
285 | DEBUG("sigchild handler set"); |
286 | ||
b0a33c1e | 287 | return fd; |
288 | } | |
289 | ||
84c92abd | 290 | static int signal_handler(int fd, uint32_t events, void *data, |
b0a33c1e | 291 | struct lxc_epoll_descr *descr) |
292 | { | |
15cd25fd | 293 | struct signalfd_siginfo siginfo; |
80507ee8 | 294 | siginfo_t info; |
15cd25fd | 295 | int ret; |
82d89dce | 296 | pid_t *pid = data; |
80507ee8 | 297 | bool init_died = false; |
15cd25fd DL |
298 | |
299 | ret = read(fd, &siginfo, sizeof(siginfo)); | |
300 | if (ret < 0) { | |
f3304a29 | 301 | ERROR("failed to read signal info"); |
15cd25fd DL |
302 | return -1; |
303 | } | |
304 | ||
305 | if (ret != sizeof(siginfo)) { | |
306 | ERROR("unexpected siginfo size"); | |
307 | return -1; | |
308 | } | |
309 | ||
80507ee8 SH |
310 | // check whether init is running |
311 | info.si_pid = 0; | |
312 | ret = waitid(P_PID, *pid, &info, WEXITED | WNOWAIT | WNOHANG); | |
313 | if (ret == 0 && info.si_pid == *pid) { | |
314 | init_died = true; | |
315 | } | |
316 | ||
f3304a29 FW |
317 | if (siginfo.ssi_signo != SIGCHLD) { |
318 | kill(*pid, siginfo.ssi_signo); | |
319 | INFO("forwarded signal %d to pid %d", siginfo.ssi_signo, *pid); | |
80507ee8 | 320 | return init_died ? 1 : 0; |
f3304a29 FW |
321 | } |
322 | ||
15cd25fd DL |
323 | if (siginfo.ssi_code == CLD_STOPPED || |
324 | siginfo.ssi_code == CLD_CONTINUED) { | |
325 | INFO("container init process was stopped/continued"); | |
80507ee8 | 326 | return init_died ? 1 : 0; |
15cd25fd | 327 | } |
1ac470c0 | 328 | |
82d89dce DL |
329 | /* more robustness, protect ourself from a SIGCHLD sent |
330 | * by a process different from the container init | |
331 | */ | |
332 | if (siginfo.ssi_pid != *pid) { | |
333 | WARN("invalid pid for SIGCHLD"); | |
80507ee8 | 334 | return init_died ? 1 : 0; |
82d89dce DL |
335 | } |
336 | ||
15cd25fd | 337 | DEBUG("container init process exited"); |
b0a33c1e | 338 | return 1; |
339 | } | |
340 | ||
735f2c6e | 341 | int lxc_set_state(const char *name, struct lxc_handler *handler, lxc_state_t state) |
66aeffc7 DL |
342 | { |
343 | handler->state = state; | |
9123e471 | 344 | lxc_monitor_send_state(name, state, handler->lxcpath); |
66aeffc7 DL |
345 | return 0; |
346 | } | |
347 | ||
735f2c6e | 348 | int lxc_poll(const char *name, struct lxc_handler *handler) |
b0a33c1e | 349 | { |
ca5f7926 DL |
350 | int sigfd = handler->sigfd; |
351 | int pid = handler->pid; | |
b0a33c1e | 352 | struct lxc_epoll_descr descr; |
353 | ||
a9e61274 | 354 | if (lxc_mainloop_open(&descr)) { |
36eb9bde | 355 | ERROR("failed to create mainloop"); |
50c8bf05 | 356 | goto out_sigfd; |
b0a33c1e | 357 | } |
358 | ||
83ee7875 | 359 | if (lxc_mainloop_add_handler(&descr, sigfd, signal_handler, &pid)) { |
36eb9bde | 360 | ERROR("failed to add handler for the signal"); |
b0a33c1e | 361 | goto out_mainloop_open; |
362 | } | |
363 | ||
63376d7d DL |
364 | if (lxc_console_mainloop_add(&descr, handler)) { |
365 | ERROR("failed to add console handler to mainloop"); | |
366 | goto out_mainloop_open; | |
367 | } | |
368 | ||
ef6e34ee | 369 | if (lxc_cmd_mainloop_add(name, &descr, handler)) { |
563f2f2c | 370 | ERROR("failed to add command handler to mainloop"); |
96fa1ff0 | 371 | goto out_mainloop_open; |
563f2f2c DL |
372 | } |
373 | ||
828695d9 | 374 | if (handler->conf->need_utmp_watch) { |
495d2046 | 375 | #if HAVE_SYS_CAPABILITY_H |
828695d9 SH |
376 | if (lxc_utmp_mainloop_add(&descr, handler)) { |
377 | ERROR("failed to add utmp handler to mainloop"); | |
378 | goto out_mainloop_open; | |
379 | } | |
495d2046 | 380 | #else |
959aee9c | 381 | DEBUG("not starting utmp handler as cap_sys_boot cannot be dropped without capabilities support"); |
495d2046 | 382 | #endif |
563f2f2c | 383 | } |
b0a33c1e | 384 | |
e51d4895 | 385 | return lxc_mainloop(&descr, -1); |
b0a33c1e | 386 | |
387 | out_mainloop_open: | |
388 | lxc_mainloop_close(&descr); | |
b0a33c1e | 389 | out_sigfd: |
390 | close(sigfd); | |
c3e13372 | 391 | return -1; |
b0a33c1e | 392 | } |
393 | ||
13f5be62 | 394 | struct lxc_handler *lxc_init(const char *name, struct lxc_conf *conf, const char *lxcpath) |
59eb99ba | 395 | { |
b6b2b194 | 396 | int i; |
3a0f472d DL |
397 | struct lxc_handler *handler; |
398 | ||
399 | handler = malloc(sizeof(*handler)); | |
400 | if (!handler) | |
401 | return NULL; | |
59eb99ba DL |
402 | |
403 | memset(handler, 0, sizeof(*handler)); | |
404 | ||
e8bd4e43 | 405 | handler->ttysock[0] = handler->ttysock[1] = -1; |
fae349da | 406 | handler->conf = conf; |
9123e471 | 407 | handler->lxcpath = lxcpath; |
5c068da9 | 408 | handler->pinfd = -1; |
fae349da | 409 | |
b6b2b194 WB |
410 | for (i = 0; i < LXC_NS_MAX; i++) |
411 | handler->nsfd[i] = -1; | |
412 | ||
fe4de9a6 DE |
413 | lsm_init(); |
414 | ||
3bdf52d7 DL |
415 | handler->name = strdup(name); |
416 | if (!handler->name) { | |
417 | ERROR("failed to allocate memory"); | |
418 | goto out_free; | |
419 | } | |
420 | ||
ef6e34ee | 421 | if (lxc_cmd_init(name, handler, lxcpath)) |
d2e30e99 DE |
422 | goto out_free_name; |
423 | ||
8f2c3a70 SH |
424 | if (lxc_read_seccomp_config(conf) != 0) { |
425 | ERROR("failed loading seccomp policy"); | |
d2e30e99 | 426 | goto out_close_maincmd_fd; |
8f2c3a70 SH |
427 | } |
428 | ||
051151de | 429 | /* Begin by setting the state to STARTING */ |
25c2aca5 | 430 | if (lxc_set_state(name, handler, STARTING)) { |
59eb99ba | 431 | ERROR("failed to set state '%s'", lxc_state2str(STARTING)); |
051151de | 432 | goto out_close_maincmd_fd; |
0ad19a3f | 433 | } |
434 | ||
f7bee6c6 | 435 | /* Start of environment variable setup for hooks */ |
ab7efcf5 | 436 | if (name && setenv("LXC_NAME", name, 1)) { |
f7bee6c6 MW |
437 | SYSERROR("failed to set environment variable for container name"); |
438 | } | |
ab7efcf5 | 439 | if (conf->rcfile && setenv("LXC_CONFIG_FILE", conf->rcfile, 1)) { |
f7bee6c6 MW |
440 | SYSERROR("failed to set environment variable for config path"); |
441 | } | |
ab7efcf5 | 442 | if (conf->rootfs.mount && setenv("LXC_ROOTFS_MOUNT", conf->rootfs.mount, 1)) { |
f7bee6c6 MW |
443 | SYSERROR("failed to set environment variable for rootfs mount"); |
444 | } | |
ab7efcf5 | 445 | if (conf->rootfs.path && setenv("LXC_ROOTFS_PATH", conf->rootfs.path, 1)) { |
f7bee6c6 MW |
446 | SYSERROR("failed to set environment variable for rootfs mount"); |
447 | } | |
448 | if (conf->console.path && setenv("LXC_CONSOLE", conf->console.path, 1)) { | |
449 | SYSERROR("failed to set environment variable for console path"); | |
450 | } | |
451 | if (conf->console.log_path && setenv("LXC_CONSOLE_LOGPATH", conf->console.log_path, 1)) { | |
452 | SYSERROR("failed to set environment variable for console log"); | |
453 | } | |
454 | /* End of environment variable setup for hooks */ | |
455 | ||
283678ed | 456 | if (run_lxc_hooks(name, "pre-start", conf, handler->lxcpath, NULL)) { |
773fb9ca SH |
457 | ERROR("failed to run pre-start hooks for container '%s'.", name); |
458 | goto out_aborting; | |
459 | } | |
26ddeedd | 460 | |
b0a33c1e | 461 | /* the signal fd has to be created before forking otherwise |
462 | * if the child process exits before we setup the signal fd, | |
463 | * the event will be lost and the command will be stuck */ | |
83ee7875 | 464 | handler->sigfd = setup_signal_fd(&handler->oldmask); |
59eb99ba | 465 | if (handler->sigfd < 0) { |
36eb9bde | 466 | ERROR("failed to set sigchild fd handler"); |
b5159817 DE |
467 | goto out_delete_tty; |
468 | } | |
469 | ||
470 | /* do this after setting up signals since it might unblock SIGWINCH */ | |
471 | if (lxc_console_create(conf)) { | |
472 | ERROR("failed to create console"); | |
473 | goto out_restore_sigmask; | |
b0a33c1e | 474 | } |
475 | ||
c4d10a05 SH |
476 | if (ttys_shift_ids(conf) < 0) { |
477 | ERROR("Failed to shift tty into container"); | |
478 | goto out_restore_sigmask; | |
479 | } | |
480 | ||
c3e13372 | 481 | INFO("'%s' is initialized", name); |
3a0f472d | 482 | return handler; |
59eb99ba | 483 | |
b5159817 DE |
484 | out_restore_sigmask: |
485 | sigprocmask(SIG_SETMASK, &handler->oldmask, NULL); | |
59eb99ba | 486 | out_delete_tty: |
fae349da | 487 | lxc_delete_tty(&conf->tty_info); |
59eb99ba | 488 | out_aborting: |
25c2aca5 | 489 | lxc_set_state(name, handler, ABORTING); |
d2e30e99 DE |
490 | out_close_maincmd_fd: |
491 | close(conf->maincmd_fd); | |
492 | conf->maincmd_fd = -1; | |
3bdf52d7 DL |
493 | out_free_name: |
494 | free(handler->name); | |
495 | handler->name = NULL; | |
3a0f472d DL |
496 | out_free: |
497 | free(handler); | |
c3e13372 | 498 | return NULL; |
59eb99ba DL |
499 | } |
500 | ||
3b72c4a0 | 501 | void lxc_fini(const char *name, struct lxc_handler *handler) |
59eb99ba | 502 | { |
b3286b62 WB |
503 | int i, rc; |
504 | pid_t self = getpid(); | |
505 | char *namespaces[LXC_NS_MAX+1]; | |
506 | size_t namespace_count = 0; | |
b6b2b194 | 507 | |
59eb99ba DL |
508 | /* The STOPPING state is there for future cleanup code |
509 | * which can take awhile | |
510 | */ | |
25c2aca5 | 511 | lxc_set_state(name, handler, STOPPING); |
b6b2b194 | 512 | |
b3286b62 WB |
513 | for (i = 0; i < LXC_NS_MAX; i++) { |
514 | if (handler->nsfd[i] != -1) { | |
515 | rc = asprintf(&namespaces[namespace_count], "%s:/proc/%d/fd/%d", | |
516 | ns_info[i].proc_name, self, handler->nsfd[i]); | |
517 | if (rc == -1) { | |
518 | SYSERROR("failed to allocate memory"); | |
519 | break; | |
520 | } | |
521 | ++namespace_count; | |
522 | } | |
523 | } | |
524 | namespaces[namespace_count] = NULL; | |
c154af98 SG |
525 | |
526 | if (handler->conf->reboot && setenv("LXC_TARGET", "reboot", 1)) { | |
527 | SYSERROR("failed to set environment variable for stop target"); | |
528 | } | |
529 | if (!handler->conf->reboot && setenv("LXC_TARGET", "stop", 1)) { | |
530 | SYSERROR("failed to set environment variable for stop target"); | |
531 | } | |
532 | ||
b3286b62 | 533 | if (run_lxc_hooks(name, "stop", handler->conf, handler->lxcpath, namespaces)) |
8438bfbd | 534 | ERROR("failed to run stop hooks for container '%s'.", name); |
c154af98 | 535 | |
b3286b62 WB |
536 | while (namespace_count--) |
537 | free(namespaces[namespace_count]); | |
b6b2b194 WB |
538 | for (i = 0; i < LXC_NS_MAX; i++) { |
539 | if (handler->nsfd[i] != -1) { | |
540 | close(handler->nsfd[i]); | |
541 | handler->nsfd[i] = -1; | |
542 | } | |
543 | } | |
25c2aca5 | 544 | lxc_set_state(name, handler, STOPPED); |
59eb99ba | 545 | |
283678ed | 546 | if (run_lxc_hooks(name, "post-stop", handler->conf, handler->lxcpath, NULL)) |
773fb9ca | 547 | ERROR("failed to run post-stop hooks for container '%s'.", name); |
26ddeedd | 548 | |
83ee7875 | 549 | /* reset mask set by setup_signal_fd */ |
8f64a3f6 MN |
550 | if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL)) |
551 | WARN("failed to restore sigprocmask"); | |
552 | ||
b5159817 | 553 | lxc_console_delete(&handler->conf->console); |
b2431939 | 554 | lxc_delete_tty(&handler->conf->tty_info); |
d2e30e99 DE |
555 | close(handler->conf->maincmd_fd); |
556 | handler->conf->maincmd_fd = -1; | |
3bdf52d7 | 557 | free(handler->name); |
e8bd4e43 SH |
558 | if (handler->ttysock[0] != -1) { |
559 | close(handler->ttysock[0]); | |
560 | close(handler->ttysock[1]); | |
561 | } | |
776d170c | 562 | if (handler->conf->ephemeral == 1 && handler->conf->reboot != 1) { |
28272964 CB |
563 | lxc_destroy_container_on_signal(handler, name); |
564 | } | |
d4ef7c50 | 565 | cgroup_destroy(handler); |
b2431939 | 566 | free(handler); |
59eb99ba DL |
567 | } |
568 | ||
735f2c6e | 569 | void lxc_abort(const char *name, struct lxc_handler *handler) |
59eb99ba | 570 | { |
73e608b2 SH |
571 | int ret, status; |
572 | ||
25c2aca5 | 573 | lxc_set_state(name, handler, ABORTING); |
7d9fb3e9 DL |
574 | if (handler->pid > 0) |
575 | kill(handler->pid, SIGKILL); | |
73e608b2 | 576 | while ((ret = waitpid(-1, &status, 0)) > 0) ; |
59eb99ba DL |
577 | } |
578 | ||
828695d9 SH |
579 | #include <sys/reboot.h> |
580 | #include <linux/reboot.h> | |
581 | ||
e2fa1520 SH |
582 | /* |
583 | * reboot(LINUX_REBOOT_CMD_CAD_ON) will return -EINVAL | |
584 | * in a child pid namespace if container reboot support exists. | |
585 | * Otherwise, it will either succeed or return -EPERM. | |
586 | */ | |
587 | static int container_reboot_supported(void *arg) | |
828695d9 | 588 | { |
d028235d | 589 | int *cmd = arg; |
828695d9 | 590 | int ret; |
828695d9 | 591 | |
d028235d | 592 | ret = reboot(*cmd); |
e2fa1520 SH |
593 | if (ret == -1 && errno == EINVAL) |
594 | return 1; | |
595 | return 0; | |
596 | } | |
597 | ||
b60ed720 | 598 | static int must_drop_cap_sys_boot(struct lxc_conf *conf) |
e2fa1520 | 599 | { |
025ed0f3 | 600 | FILE *f; |
b60ed720 | 601 | int ret, cmd, v, flags; |
d028235d SG |
602 | long stack_size = 4096; |
603 | void *stack = alloca(stack_size); | |
604 | int status; | |
605 | pid_t pid; | |
e2fa1520 | 606 | |
025ed0f3 | 607 | f = fopen("/proc/sys/kernel/ctrl-alt-del", "r"); |
e2fa1520 SH |
608 | if (!f) { |
609 | DEBUG("failed to open /proc/sys/kernel/ctrl-alt-del"); | |
828695d9 | 610 | return 1; |
e2fa1520 | 611 | } |
828695d9 SH |
612 | |
613 | ret = fscanf(f, "%d", &v); | |
614 | fclose(f); | |
e2fa1520 SH |
615 | if (ret != 1) { |
616 | DEBUG("Failed to read /proc/sys/kernel/ctrl-alt-del"); | |
828695d9 | 617 | return 1; |
e2fa1520 SH |
618 | } |
619 | cmd = v ? LINUX_REBOOT_CMD_CAD_ON : LINUX_REBOOT_CMD_CAD_OFF; | |
620 | ||
b60ed720 SH |
621 | flags = CLONE_NEWPID | SIGCHLD; |
622 | if (!lxc_list_empty(&conf->id_map)) | |
623 | flags |= CLONE_NEWUSER; | |
624 | ||
7f145a6d | 625 | #ifdef __ia64__ |
959aee9c | 626 | pid = __clone2(container_reboot_supported, stack, stack_size, flags, &cmd); |
7f145a6d | 627 | #else |
959aee9c SG |
628 | stack += stack_size; |
629 | pid = clone(container_reboot_supported, stack, flags, &cmd); | |
7f145a6d | 630 | #endif |
959aee9c | 631 | if (pid < 0) { |
c08220e9 | 632 | if (flags & CLONE_NEWUSER) |
4a7e5f4f | 633 | ERROR("failed to clone (%#x): %s (includes CLONE_NEWUSER)", flags, strerror(errno)); |
c08220e9 | 634 | else |
4a7e5f4f | 635 | ERROR("failed to clone (%#x): %s", flags, strerror(errno)); |
959aee9c SG |
636 | return -1; |
637 | } | |
638 | if (wait(&status) < 0) { | |
639 | SYSERROR("unexpected wait error: %m"); | |
640 | return -1; | |
641 | } | |
e2fa1520 SH |
642 | |
643 | if (WEXITSTATUS(status) != 1) | |
828695d9 | 644 | return 1; |
e2fa1520 | 645 | |
828695d9 SH |
646 | return 0; |
647 | } | |
648 | ||
658979c5 SH |
649 | /* |
650 | * netpipe is used in the unprivileged case to transfer the ifindexes | |
651 | * from parent to child | |
652 | */ | |
653 | static int netpipe = -1; | |
654 | ||
655 | static inline int count_veths(struct lxc_list *network) | |
656 | { | |
657 | struct lxc_list *iterator; | |
658 | struct lxc_netdev *netdev; | |
659 | int count = 0; | |
660 | ||
661 | lxc_list_for_each(iterator, network) { | |
662 | netdev = iterator->elem; | |
663 | if (netdev->type != LXC_NET_VETH) | |
664 | continue; | |
665 | count++; | |
666 | } | |
667 | return count; | |
668 | } | |
669 | ||
670 | static int read_unpriv_netifindex(struct lxc_list *network) | |
671 | { | |
672 | struct lxc_list *iterator; | |
673 | struct lxc_netdev *netdev; | |
674 | ||
675 | if (netpipe == -1) | |
676 | return 0; | |
677 | lxc_list_for_each(iterator, network) { | |
678 | netdev = iterator->elem; | |
679 | if (netdev->type != LXC_NET_VETH) | |
680 | continue; | |
681 | if (!(netdev->name = malloc(IFNAMSIZ))) { | |
682 | ERROR("Out of memory"); | |
683 | close(netpipe); | |
684 | return -1; | |
685 | } | |
686 | if (read(netpipe, netdev->name, IFNAMSIZ) != IFNAMSIZ) { | |
687 | close(netpipe); | |
688 | return -1; | |
689 | } | |
690 | } | |
691 | close(netpipe); | |
692 | return 0; | |
693 | } | |
694 | ||
ffe1e01a | 695 | static int do_start(void *data) |
50e98013 | 696 | { |
7c661726 | 697 | struct lxc_list *iterator; |
23c53af9 | 698 | struct lxc_handler *handler = data; |
50e98013 DL |
699 | |
700 | if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL)) { | |
701 | SYSERROR("failed to set sigprocmask"); | |
9d7f9e52 | 702 | return -1; |
50e98013 DL |
703 | } |
704 | ||
d028235d | 705 | /* This prctl must be before the synchro, so if the parent |
743ecd2e DL |
706 | * dies before we set the parent death signal, we will detect |
707 | * its death with the synchro right after, otherwise we have | |
708 | * a window where the parent can exit before we set the pdeath | |
709 | * signal leading to a unsupervized container. | |
710 | */ | |
711 | if (prctl(PR_SET_PDEATHSIG, SIGKILL, 0, 0, 0)) { | |
712 | SYSERROR("failed to set pdeath signal"); | |
713 | return -1; | |
714 | } | |
715 | ||
3c22086f | 716 | lxc_sync_fini_parent(handler); |
50e98013 | 717 | |
2b0e17e4 | 718 | /* don't leak the pinfd to the container */ |
025ed0f3 | 719 | if (handler->pinfd >= 0) { |
0d03360a | 720 | close(handler->pinfd); |
025ed0f3 | 721 | } |
2b0e17e4 | 722 | |
3c22086f CLG |
723 | /* Tell the parent task it can begin to configure the |
724 | * container and wait for it to finish | |
725 | */ | |
726 | if (lxc_sync_barrier_parent(handler, LXC_SYNC_CONFIGURE)) | |
9d7f9e52 | 727 | return -1; |
50e98013 | 728 | |
658979c5 SH |
729 | if (read_unpriv_netifindex(&handler->conf->network) < 0) |
730 | goto out_warn_father; | |
731 | ||
f6d3e3e4 SH |
732 | /* |
733 | * if we are in a new user namespace, become root there to have | |
62d00539 PT |
734 | * privilege over our namespace. When using lxc-execute we default to root, |
735 | * but this can be overriden using the lxc.init_uid and lxc.init_gid | |
736 | * configuration options. | |
f6d3e3e4 SH |
737 | */ |
738 | if (!lxc_list_empty(&handler->conf->id_map)) { | |
c5cd20ce PT |
739 | gid_t new_gid = 0; |
740 | if (handler->conf->is_execute && handler->conf->init_gid) | |
741 | new_gid = handler->conf->init_gid; | |
742 | ||
743 | uid_t new_uid = 0; | |
744 | if (handler->conf->is_execute && handler->conf->init_uid) | |
745 | new_uid = handler->conf->init_uid; | |
746 | ||
56f8ff00 PT |
747 | NOTICE("switching to gid/uid %d/%d in new user namespace", new_gid, new_uid); |
748 | if (setgid(new_gid)) { | |
f6d3e3e4 SH |
749 | SYSERROR("setgid"); |
750 | goto out_warn_father; | |
751 | } | |
56f8ff00 | 752 | if (setuid(new_uid)) { |
f6d3e3e4 SH |
753 | SYSERROR("setuid"); |
754 | goto out_warn_father; | |
755 | } | |
c476bdce SH |
756 | if (setgroups(0, NULL)) { |
757 | SYSERROR("setgroups"); | |
758 | goto out_warn_father; | |
759 | } | |
f6d3e3e4 SH |
760 | } |
761 | ||
99b71824 | 762 | if (access(handler->lxcpath, X_OK)) { |
c8154066 SH |
763 | print_top_failing_dir(handler->lxcpath); |
764 | goto out_warn_father; | |
765 | } | |
766 | ||
495d2046 | 767 | #if HAVE_SYS_CAPABILITY_H |
69182a31 | 768 | if (handler->conf->need_utmp_watch) { |
828695d9 SH |
769 | if (prctl(PR_CAPBSET_DROP, CAP_SYS_BOOT, 0, 0, 0)) { |
770 | SYSERROR("failed to remove CAP_SYS_BOOT capability"); | |
c4ea60df | 771 | goto out_warn_father; |
828695d9 | 772 | } |
959aee9c | 773 | DEBUG("Dropped cap_sys_boot"); |
e2fa1520 | 774 | } |
495d2046 | 775 | #endif |
e2fa1520 SH |
776 | |
777 | /* Setup the container, ip, names, utsname, ... */ | |
d4ef7c50 | 778 | if (lxc_setup(handler)) { |
e2fa1520 SH |
779 | ERROR("failed to setup the container"); |
780 | goto out_warn_father; | |
781 | } | |
50e98013 | 782 | |
544a48a0 SH |
783 | /* ask father to setup cgroups and wait for him to finish */ |
784 | if (lxc_sync_barrier_parent(handler, LXC_SYNC_CGROUP)) | |
785 | return -1; | |
786 | ||
72863294 | 787 | /* Set the label to change to when we exec(2) the container's init */ |
7aff4f43 | 788 | if (lsm_process_label_set(NULL, handler->conf, 1, 1) < 0) |
e075f5d9 | 789 | goto out_warn_father; |
5112cd70 | 790 | |
0d9acb99 DE |
791 | /* Some init's such as busybox will set sane tty settings on stdin, |
792 | * stdout, stderr which it thinks is the console. We already set them | |
793 | * the way we wanted on the real terminal, and we want init to do its | |
794 | * setup on its console ie. the pty allocated in lxc_console_create() | |
795 | * so make sure that that pty is stdin,stdout,stderr. | |
796 | */ | |
797 | if (lxc_console_set_stdfds(handler) < 0) | |
798 | goto out_warn_father; | |
799 | ||
5112cd70 SH |
800 | /* If we mounted a temporary proc, then unmount it now */ |
801 | tmp_proc_unmount(handler->conf); | |
e075f5d9 | 802 | |
8f2c3a70 SH |
803 | if (lxc_seccomp_load(handler->conf) != 0) |
804 | goto out_warn_father; | |
805 | ||
283678ed | 806 | if (run_lxc_hooks(handler->name, "start", handler->conf, handler->lxcpath, NULL)) { |
773fb9ca SH |
807 | ERROR("failed to run start hooks for container '%s'.", handler->name); |
808 | goto out_warn_father; | |
809 | } | |
fc25b815 | 810 | |
f7bee6c6 | 811 | /* The clearenv() and putenv() calls have been moved here |
ec64264d | 812 | * to allow us to use environment variables passed to the various |
f7bee6c6 MW |
813 | * hooks, such as the start hook above. Not all of the |
814 | * variables like CONFIG_PATH or ROOTFS are valid in this | |
815 | * context but others are. */ | |
816 | if (clearenv()) { | |
817 | SYSERROR("failed to clear environment"); | |
818 | /* don't error out though */ | |
819 | } | |
820 | ||
7c661726 MP |
821 | lxc_list_for_each(iterator, &handler->conf->environment) { |
822 | if (putenv((char *)iterator->elem)) { | |
823 | SYSERROR("failed to set environment variable '%s'", (char *)iterator->elem); | |
824 | goto out_warn_father; | |
825 | } | |
826 | } | |
827 | ||
f7bee6c6 | 828 | if (putenv("container=lxc")) { |
7c661726 | 829 | SYSERROR("failed to set environment variable 'container=lxc'"); |
c4ea60df | 830 | goto out_warn_father; |
f7bee6c6 MW |
831 | } |
832 | ||
393903d1 SH |
833 | if (handler->conf->pty_names) { |
834 | if (putenv(handler->conf->pty_names)) { | |
835 | SYSERROR("failed to set environment variable for container ptys"); | |
836 | goto out_warn_father; | |
837 | } | |
838 | } | |
839 | ||
773fb9ca | 840 | close(handler->sigfd); |
26ddeedd | 841 | |
69aeabac TA |
842 | if (handler->backgrounded && null_stdfds() < 0) |
843 | goto out_warn_father; | |
507cee36 | 844 | |
e6126dbe MN |
845 | /* after this call, we are in error because this |
846 | * ops should not return as it execs */ | |
c4ea60df | 847 | handler->ops->start(handler, handler->data); |
50e98013 DL |
848 | |
849 | out_warn_father: | |
544a48a0 SH |
850 | /* we want the parent to know something went wrong, so any |
851 | * value other than what it expects is ok. */ | |
3c22086f | 852 | lxc_sync_wake_parent(handler, LXC_SYNC_POST_CONFIGURE); |
50e98013 DL |
853 | return -1; |
854 | } | |
855 | ||
74a3920a | 856 | static int save_phys_nics(struct lxc_conf *conf) |
7b35f3d6 SH |
857 | { |
858 | struct lxc_list *iterator; | |
859 | ||
860 | lxc_list_for_each(iterator, &conf->network) { | |
861 | struct lxc_netdev *netdev = iterator->elem; | |
862 | ||
863 | if (netdev->type != LXC_NET_PHYS) | |
864 | continue; | |
865 | conf->saved_nics = realloc(conf->saved_nics, | |
866 | (conf->num_savednics+1)*sizeof(struct saved_nic)); | |
867 | if (!conf->saved_nics) { | |
868 | SYSERROR("failed to allocate memory"); | |
869 | return -1; | |
870 | } | |
871 | conf->saved_nics[conf->num_savednics].ifindex = netdev->ifindex; | |
872 | conf->saved_nics[conf->num_savednics].orig_name = strdup(netdev->link); | |
873 | if (!conf->saved_nics[conf->num_savednics].orig_name) { | |
874 | SYSERROR("failed to allocate memory"); | |
875 | return -1; | |
876 | } | |
959aee9c | 877 | INFO("stored saved_nic #%d idx %d name %s", conf->num_savednics, |
7b35f3d6 SH |
878 | conf->saved_nics[conf->num_savednics].ifindex, |
879 | conf->saved_nics[conf->num_savednics].orig_name); | |
880 | conf->num_savednics++; | |
881 | } | |
882 | ||
883 | return 0; | |
884 | } | |
885 | ||
e8bd4e43 SH |
886 | static int recv_fd(int sock, int *fd) |
887 | { | |
888 | if (lxc_abstract_unix_recv_fd(sock, fd, NULL, 0) < 0) { | |
889 | SYSERROR("Error receiving tty fd from child"); | |
890 | return -1; | |
891 | } | |
892 | if (*fd == -1) | |
893 | return -1; | |
894 | return 0; | |
895 | } | |
896 | ||
897 | static int recv_ttys_from_child(struct lxc_handler *handler) | |
898 | { | |
899 | struct lxc_conf *conf = handler->conf; | |
900 | int i, sock = handler->ttysock[1]; | |
901 | struct lxc_tty_info *tty_info = &conf->tty_info; | |
902 | ||
903 | if (!conf->tty) | |
904 | return 0; | |
905 | ||
906 | tty_info->pty_info = malloc(sizeof(*tty_info->pty_info)*conf->tty); | |
907 | if (!tty_info->pty_info) { | |
908 | SYSERROR("failed to allocate pty_info"); | |
909 | return -1; | |
910 | } | |
911 | ||
912 | for (i = 0; i < conf->tty; i++) { | |
913 | struct lxc_pty_info *pty_info = &tty_info->pty_info[i]; | |
914 | pty_info->busy = 0; | |
915 | if (recv_fd(sock, &pty_info->slave) < 0 || | |
916 | recv_fd(sock, &pty_info->master) < 0) { | |
917 | ERROR("Error receiving tty info from child"); | |
918 | return -1; | |
919 | } | |
920 | } | |
921 | tty_info->nbtty = conf->tty; | |
922 | ||
923 | return 0; | |
924 | } | |
925 | ||
f813849c TA |
926 | void resolve_clone_flags(struct lxc_handler *handler) |
927 | { | |
928 | handler->clone_flags = CLONE_NEWPID | CLONE_NEWNS; | |
929 | ||
930 | if (!lxc_list_empty(&handler->conf->id_map)) { | |
931 | INFO("Cloning a new user namespace"); | |
932 | handler->clone_flags |= CLONE_NEWUSER; | |
933 | } | |
934 | ||
935 | if (handler->conf->inherit_ns_fd[LXC_NS_NET] == -1) { | |
936 | if (!lxc_requests_empty_network(handler)) | |
937 | handler->clone_flags |= CLONE_NEWNET; | |
938 | } else { | |
939 | INFO("Inheriting a net namespace"); | |
940 | } | |
941 | ||
942 | if (handler->conf->inherit_ns_fd[LXC_NS_IPC] == -1) { | |
943 | handler->clone_flags |= CLONE_NEWIPC; | |
944 | } else { | |
945 | INFO("Inheriting an IPC namespace"); | |
946 | } | |
947 | ||
948 | if (handler->conf->inherit_ns_fd[LXC_NS_UTS] == -1) { | |
949 | handler->clone_flags |= CLONE_NEWUTS; | |
950 | } else { | |
951 | INFO("Inheriting a UTS namespace"); | |
952 | } | |
953 | } | |
954 | ||
74a3920a | 955 | static int lxc_spawn(struct lxc_handler *handler) |
59eb99ba | 956 | { |
b98f7d6e | 957 | int failed_before_rename = 0; |
ffe1e01a | 958 | const char *name = handler->name; |
62d05d9b | 959 | char *errmsg = NULL; |
7e4dfe0b | 960 | bool cgroups_connected = false; |
9f30a190 MM |
961 | int saved_ns_fd[LXC_NS_MAX]; |
962 | int preserve_mask = 0, i; | |
658979c5 | 963 | int netpipepair[2], nveths; |
9f30a190 | 964 | |
2f2623ec SH |
965 | netpipe = -1; |
966 | ||
9f30a190 | 967 | for (i = 0; i < LXC_NS_MAX; i++) |
6c544cb3 | 968 | if (handler->conf->inherit_ns_fd[i] != -1) |
9f30a190 | 969 | preserve_mask |= ns_info[i].clone_flag; |
50e98013 | 970 | |
3c22086f | 971 | if (lxc_sync_init(handler)) |
9d7f9e52 | 972 | return -1; |
0ad19a3f | 973 | |
e8bd4e43 SH |
974 | if (socketpair(AF_UNIX, SOCK_DGRAM, 0, handler->ttysock) < 0) { |
975 | lxc_sync_fini(handler); | |
976 | return -1; | |
977 | } | |
978 | ||
f813849c | 979 | resolve_clone_flags(handler); |
9f30a190 | 980 | |
f813849c | 981 | if (handler->clone_flags & CLONE_NEWNET) { |
26b797f3 SH |
982 | if (!lxc_list_empty(&handler->conf->network)) { |
983 | ||
9f30a190 MM |
984 | /* Find gateway addresses from the link device, which is |
985 | * no longer accessible inside the container. Do this | |
986 | * before creating network interfaces, since goto | |
987 | * out_delete_net does not work before lxc_clone. */ | |
988 | if (lxc_find_gateway_addresses(handler)) { | |
989 | ERROR("failed to find gateway addresses"); | |
990 | lxc_sync_fini(handler); | |
991 | return -1; | |
992 | } | |
993 | ||
994 | /* that should be done before the clone because we will | |
995 | * fill the netdev index and use them in the child | |
996 | */ | |
997 | if (lxc_create_network(handler)) { | |
38521a3b | 998 | ERROR("failed to create the network"); |
9f30a190 MM |
999 | lxc_sync_fini(handler); |
1000 | return -1; | |
1001 | } | |
19a26f82 MK |
1002 | } |
1003 | ||
9f30a190 MM |
1004 | if (save_phys_nics(handler->conf)) { |
1005 | ERROR("failed to save physical nic info"); | |
1006 | goto out_abort; | |
82d5ae15 DL |
1007 | } |
1008 | } | |
1009 | ||
d4ef7c50 SH |
1010 | if (!cgroup_init(handler)) { |
1011 | ERROR("failed initializing cgroup support"); | |
33ad9f1a CS |
1012 | goto out_delete_net; |
1013 | } | |
1014 | ||
7e4dfe0b SH |
1015 | cgroups_connected = true; |
1016 | ||
d4ef7c50 SH |
1017 | if (!cgroup_create(handler)) { |
1018 | ERROR("failed creating cgroups"); | |
47d8fb3b CS |
1019 | goto out_delete_net; |
1020 | } | |
1021 | ||
0c547523 SH |
1022 | /* |
1023 | * if the rootfs is not a blockdev, prevent the container from | |
1024 | * marking it readonly. | |
5e32a990 ÇO |
1025 | * |
1026 | * if the container is unprivileged then skip rootfs pinning | |
0c547523 | 1027 | */ |
5e32a990 ÇO |
1028 | if (lxc_list_empty(&handler->conf->id_map)) { |
1029 | handler->pinfd = pin_rootfs(handler->conf->rootfs.path); | |
1030 | if (handler->pinfd == -1) | |
1031 | INFO("failed to pin the container's rootfs"); | |
1032 | } | |
0c547523 | 1033 | |
62d05d9b SH |
1034 | if (!preserve_ns(saved_ns_fd, preserve_mask, getpid(), &errmsg)) { |
1035 | SYSERROR("Failed to preserve requested namespaces: %s", | |
1036 | errmsg ? errmsg : "(Out of memory)"); | |
1037 | free(errmsg); | |
cd43d2d1 | 1038 | goto out_delete_net; |
62d05d9b | 1039 | } |
cd43d2d1 SH |
1040 | if (attach_ns(handler->conf->inherit_ns_fd) < 0) |
1041 | goto out_delete_net; | |
9f30a190 | 1042 | |
658979c5 SH |
1043 | if (am_unpriv() && (nveths = count_veths(&handler->conf->network))) { |
1044 | if (pipe(netpipepair) < 0) { | |
1045 | SYSERROR("Error creating pipe"); | |
1046 | goto out_delete_net; | |
1047 | } | |
1048 | /* store netpipe in the global var for do_start's use */ | |
1049 | netpipe = netpipepair[0]; | |
1050 | } | |
1051 | ||
0ad19a3f | 1052 | /* Create a process in a new set of namespaces */ |
d5088cf2 | 1053 | handler->pid = lxc_clone(do_start, handler, handler->clone_flags); |
59eb99ba | 1054 | if (handler->pid < 0) { |
36eb9bde | 1055 | SYSERROR("failed to fork into a new namespace"); |
7fef7a06 | 1056 | goto out_delete_net; |
0ad19a3f | 1057 | } |
1058 | ||
62d05d9b SH |
1059 | if (preserve_ns(handler->nsfd, handler->clone_flags, handler->pid, |
1060 | &errmsg) < 0) { | |
1061 | INFO("Failed to store namespace references for stop hook: %s", | |
1062 | errmsg ? errmsg : "(Out of memory)"); | |
1063 | free(errmsg); | |
b6b2b194 WB |
1064 | } |
1065 | ||
cd43d2d1 SH |
1066 | if (attach_ns(saved_ns_fd)) |
1067 | WARN("failed to restore saved namespaces"); | |
9f30a190 | 1068 | |
3c22086f CLG |
1069 | lxc_sync_fini_child(handler); |
1070 | ||
1071 | if (lxc_sync_wait_child(handler, LXC_SYNC_CONFIGURE)) | |
99a6af52 | 1072 | failed_before_rename = 1; |
0ad19a3f | 1073 | |
d4ef7c50 SH |
1074 | if (!cgroup_create_legacy(handler)) { |
1075 | ERROR("failed to setup the legacy cgroups for %s", name); | |
ae5c8b8e | 1076 | goto out_delete_net; |
33ad9f1a | 1077 | } |
9daf6f5d SH |
1078 | if (!cgroup_setup_limits(handler, false)) { |
1079 | ERROR("failed to setup the cgroup limits for '%s'", name); | |
6031a6e5 DE |
1080 | goto out_delete_net; |
1081 | } | |
1082 | ||
d4ef7c50 | 1083 | if (!cgroup_enter(handler)) |
7fef7a06 | 1084 | goto out_delete_net; |
218d4250 | 1085 | |
0996e18a SH |
1086 | if (!cgroup_chown(handler)) |
1087 | goto out_delete_net; | |
1088 | ||
99a6af52 MN |
1089 | if (failed_before_rename) |
1090 | goto out_delete_net; | |
1091 | ||
0ad19a3f | 1092 | /* Create the network configuration */ |
d5088cf2 | 1093 | if (handler->clone_flags & CLONE_NEWNET) { |
fae349da | 1094 | if (lxc_assign_network(&handler->conf->network, handler->pid)) { |
82d5ae15 | 1095 | ERROR("failed to create the configured network"); |
7fef7a06 | 1096 | goto out_delete_net; |
82d5ae15 | 1097 | } |
0ad19a3f | 1098 | } |
1099 | ||
658979c5 SH |
1100 | if (netpipe != -1) { |
1101 | struct lxc_list *iterator; | |
1102 | struct lxc_netdev *netdev; | |
1103 | ||
1104 | close(netpipe); | |
1105 | lxc_list_for_each(iterator, &handler->conf->network) { | |
1106 | netdev = iterator->elem; | |
1107 | if (netdev->type != LXC_NET_VETH) | |
1108 | continue; | |
1109 | if (write(netpipepair[1], netdev->name, IFNAMSIZ) != IFNAMSIZ) { | |
1110 | ERROR("Error writing veth name to container"); | |
1111 | goto out_delete_net; | |
1112 | } | |
1113 | } | |
1114 | close(netpipepair[1]); | |
1115 | } | |
1116 | ||
f6d3e3e4 SH |
1117 | /* map the container uids - the container became an invalid |
1118 | * userid the moment it was cloned with CLONE_NEWUSER - this | |
1119 | * call doesn't change anything immediately, but allows the | |
1120 | * container to setuid(0) (0 being mapped to something else on | |
1121 | * the host) later to become a valid uid again */ | |
1122 | if (lxc_map_ids(&handler->conf->id_map, handler->pid)) { | |
1123 | ERROR("failed to set up id mapping"); | |
1124 | goto out_delete_net; | |
1125 | } | |
1126 | ||
544a48a0 SH |
1127 | /* Tell the child to continue its initialization. we'll get |
1128 | * LXC_SYNC_CGROUP when it is ready for us to setup cgroups | |
3c22086f CLG |
1129 | */ |
1130 | if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CONFIGURE)) | |
544a48a0 SH |
1131 | goto out_delete_net; |
1132 | ||
9daf6f5d | 1133 | if (!cgroup_setup_limits(handler, true)) { |
b98f7d6e SH |
1134 | ERROR("failed to setup the devices cgroup for '%s'", name); |
1135 | goto out_delete_net; | |
544a48a0 SH |
1136 | } |
1137 | ||
73d28d42 | 1138 | cgroup_disconnect(); |
7e4dfe0b | 1139 | cgroups_connected = false; |
73d28d42 | 1140 | |
e8bd4e43 SH |
1141 | /* read tty fds allocated by child */ |
1142 | if (recv_ttys_from_child(handler) < 0) { | |
1143 | ERROR("failed to receive tty info from child"); | |
1144 | goto out_delete_net; | |
1145 | } | |
1146 | ||
544a48a0 SH |
1147 | /* Tell the child to complete its initialization and wait for |
1148 | * it to exec or return an error. (the child will never | |
1149 | * return LXC_SYNC_POST_CGROUP+1. It will either close the | |
1150 | * sync pipe, causing lxc_sync_barrier_child to return | |
1151 | * success, or return a different value, causing us to error | |
1152 | * out). | |
1153 | */ | |
1154 | if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CGROUP)) | |
3c22086f | 1155 | return -1; |
0ad19a3f | 1156 | |
cc28d0b0 SH |
1157 | if (detect_shared_rootfs()) |
1158 | umount2(handler->conf->rootfs.mount, MNT_DETACH); | |
1159 | ||
23c53af9 | 1160 | if (handler->ops->post_start(handler, handler->data)) |
e6126dbe MN |
1161 | goto out_abort; |
1162 | ||
25c2aca5 | 1163 | if (lxc_set_state(name, handler, RUNNING)) { |
59eb99ba DL |
1164 | ERROR("failed to set state to %s", |
1165 | lxc_state2str(RUNNING)); | |
1166 | goto out_abort; | |
3f21c114 | 1167 | } |
22ebac19 | 1168 | |
3c22086f | 1169 | lxc_sync_fini(handler); |
0c547523 | 1170 | |
e6126dbe | 1171 | return 0; |
1ac470c0 | 1172 | |
7fef7a06 | 1173 | out_delete_net: |
7e4dfe0b SH |
1174 | if (cgroups_connected) |
1175 | cgroup_disconnect(); | |
d5088cf2 | 1176 | if (handler->clone_flags & CLONE_NEWNET) |
74a2b586 | 1177 | lxc_delete_network(handler); |
59eb99ba DL |
1178 | out_abort: |
1179 | lxc_abort(name, handler); | |
3c22086f | 1180 | lxc_sync_fini(handler); |
5c068da9 SH |
1181 | if (handler->pinfd >= 0) { |
1182 | close(handler->pinfd); | |
1183 | handler->pinfd = -1; | |
1184 | } | |
1185 | ||
b79fcd86 | 1186 | return -1; |
59eb99ba | 1187 | } |
0ad19a3f | 1188 | |
2af6bd1b SH |
1189 | int get_netns_fd(int pid) |
1190 | { | |
1191 | char path[MAXPATHLEN]; | |
1192 | int ret, fd; | |
1193 | ||
1194 | ret = snprintf(path, MAXPATHLEN, "/proc/%d/ns/net", pid); | |
1195 | if (ret < 0 || ret >= MAXPATHLEN) { | |
1196 | WARN("Failed to pin netns file for pid %d", pid); | |
1197 | return -1; | |
1198 | } | |
1199 | ||
1200 | fd = open(path, O_RDONLY); | |
1201 | if (fd < 0) { | |
1202 | WARN("Failed to pin netns file %s for pid %d: %s", | |
1203 | path, pid, strerror(errno)); | |
1204 | return -1; | |
1205 | } | |
1206 | return fd; | |
1207 | } | |
1208 | ||
ee70bf78 | 1209 | int __lxc_start(const char *name, struct lxc_conf *conf, |
507cee36 TA |
1210 | struct lxc_operations* ops, void *data, const char *lxcpath, |
1211 | bool backgrounded) | |
59eb99ba | 1212 | { |
3a0f472d | 1213 | struct lxc_handler *handler; |
e043236e | 1214 | int err = -1; |
59eb99ba | 1215 | int status; |
2af6bd1b | 1216 | int netnsfd = -1; |
80090207 | 1217 | |
13f5be62 | 1218 | handler = lxc_init(name, conf, lxcpath); |
3a0f472d | 1219 | if (!handler) { |
59eb99ba | 1220 | ERROR("failed to initialize the container"); |
66aeffc7 | 1221 | return -1; |
0ad19a3f | 1222 | } |
ee70bf78 CLG |
1223 | handler->ops = ops; |
1224 | handler->data = data; | |
507cee36 | 1225 | handler->backgrounded = backgrounded; |
e6126dbe | 1226 | |
b60ed720 | 1227 | if (must_drop_cap_sys_boot(handler->conf)) { |
495d2046 | 1228 | #if HAVE_SYS_CAPABILITY_H |
959aee9c | 1229 | DEBUG("Dropping cap_sys_boot"); |
495d2046 | 1230 | #else |
959aee9c | 1231 | DEBUG("Can't drop cap_sys_boot as capabilities aren't supported"); |
495d2046 | 1232 | #endif |
69182a31 | 1233 | } else { |
959aee9c | 1234 | DEBUG("Not dropping cap_sys_boot or watching utmp"); |
69182a31 SH |
1235 | handler->conf->need_utmp_watch = 0; |
1236 | } | |
1237 | ||
76a26f55 SH |
1238 | if (!attach_block_device(handler->conf)) { |
1239 | ERROR("Failure attaching block device"); | |
1240 | goto out_fini_nonet; | |
1241 | } | |
1242 | ||
35120d9c SH |
1243 | if (geteuid() == 0 && !lxc_list_empty(&conf->id_map)) { |
1244 | /* if the backing store is a device, mount it here and now */ | |
1245 | if (rootfs_is_blockdev(conf)) { | |
1246 | if (unshare(CLONE_NEWNS) < 0) { | |
1247 | ERROR("Error unsharing mounts"); | |
1248 | goto out_fini_nonet; | |
1249 | } | |
6a0c909a | 1250 | remount_all_slave(); |
35120d9c SH |
1251 | if (do_rootfs_setup(conf, name, lxcpath) < 0) { |
1252 | ERROR("Error setting up rootfs mount as root before spawn"); | |
1253 | goto out_fini_nonet; | |
1254 | } | |
1255 | INFO("Set up container rootfs as host root"); | |
1256 | } | |
1257 | } | |
1258 | ||
23c53af9 | 1259 | err = lxc_spawn(handler); |
59eb99ba | 1260 | if (err) { |
ee70bf78 | 1261 | ERROR("failed to spawn '%s'", name); |
76a26f55 | 1262 | goto out_detach_blockdev; |
0ad19a3f | 1263 | } |
1264 | ||
8bee8851 WB |
1265 | handler->conf->reboot = 0; |
1266 | ||
2af6bd1b SH |
1267 | netnsfd = get_netns_fd(handler->pid); |
1268 | ||
3a0f472d | 1269 | err = lxc_poll(name, handler); |
e043236e | 1270 | if (err) { |
59eb99ba | 1271 | ERROR("mainloop exited with an error"); |
4b2f98f5 SH |
1272 | if (netnsfd >= 0) |
1273 | close(netnsfd); | |
59eb99ba DL |
1274 | goto out_abort; |
1275 | } | |
0ad19a3f | 1276 | |
3a0f472d | 1277 | while (waitpid(handler->pid, &status, 0) < 0 && errno == EINTR) |
1bc5cc8c | 1278 | continue; |
e043236e | 1279 | |
8b004f07 SH |
1280 | /* |
1281 | * If the child process exited but was not signaled, | |
1282 | * it didn't call reboot. This should mean it was an | |
1283 | * lxc-execute which simply exited. In any case, treat | |
1284 | * it as a 'halt' | |
1285 | */ | |
d028235d | 1286 | if (WIFSIGNALED(status)) { |
8b004f07 SH |
1287 | switch(WTERMSIG(status)) { |
1288 | case SIGINT: /* halt */ | |
1289 | DEBUG("Container halting"); | |
1290 | break; | |
1291 | case SIGHUP: /* reboot */ | |
1292 | DEBUG("Container rebooting"); | |
1293 | handler->conf->reboot = 1; | |
1294 | break; | |
c2b9bd9e SH |
1295 | case SIGSYS: /* seccomp */ |
1296 | DEBUG("Container violated its seccomp policy"); | |
1297 | break; | |
8b004f07 | 1298 | default: |
959aee9c | 1299 | DEBUG("unknown exit status for init: %d", WTERMSIG(status)); |
8b004f07 SH |
1300 | break; |
1301 | } | |
d028235d | 1302 | } |
828695d9 | 1303 | |
ce5782df | 1304 | DEBUG("Pushing physical nics back to host namespace"); |
2af6bd1b | 1305 | lxc_rename_phys_nics_on_shutdown(netnsfd, handler->conf); |
ce5782df MH |
1306 | |
1307 | DEBUG("Tearing down virtual network devices used by container"); | |
1308 | lxc_delete_network(handler); | |
1309 | ||
4b2f98f5 SH |
1310 | if (netnsfd >= 0) |
1311 | close(netnsfd); | |
7b35f3d6 | 1312 | |
5c068da9 SH |
1313 | if (handler->pinfd >= 0) { |
1314 | close(handler->pinfd); | |
1315 | handler->pinfd = -1; | |
1316 | } | |
1317 | ||
1787abca | 1318 | lxc_monitor_send_exit_code(name, status, handler->lxcpath); |
3a0f472d | 1319 | err = lxc_error_set_and_log(handler->pid, status); |
9d7f9e52 | 1320 | out_fini: |
74a2b586 JK |
1321 | lxc_delete_network(handler); |
1322 | ||
76a26f55 SH |
1323 | out_detach_blockdev: |
1324 | detach_block_device(handler->conf); | |
1325 | ||
74a2b586 | 1326 | out_fini_nonet: |
3a0f472d | 1327 | lxc_fini(name, handler); |
0ad19a3f | 1328 | return err; |
1329 | ||
59eb99ba | 1330 | out_abort: |
3a0f472d | 1331 | lxc_abort(name, handler); |
9d7f9e52 | 1332 | goto out_fini; |
0ad19a3f | 1333 | } |
ee70bf78 CLG |
1334 | |
1335 | struct start_args { | |
1336 | char *const *argv; | |
1337 | }; | |
1338 | ||
1339 | static int start(struct lxc_handler *handler, void* data) | |
1340 | { | |
1341 | struct start_args *arg = data; | |
1342 | ||
1343 | NOTICE("exec'ing '%s'", arg->argv[0]); | |
1344 | ||
1345 | execvp(arg->argv[0], arg->argv); | |
1346 | SYSERROR("failed to exec %s", arg->argv[0]); | |
1347 | return 0; | |
1348 | } | |
1349 | ||
1350 | static int post_start(struct lxc_handler *handler, void* data) | |
1351 | { | |
1352 | struct start_args *arg = data; | |
1353 | ||
1354 | NOTICE("'%s' started with pid '%d'", arg->argv[0], handler->pid); | |
1355 | return 0; | |
1356 | } | |
1357 | ||
1358 | static struct lxc_operations start_ops = { | |
1359 | .start = start, | |
1360 | .post_start = post_start | |
1361 | }; | |
1362 | ||
13f5be62 | 1363 | int lxc_start(const char *name, char *const argv[], struct lxc_conf *conf, |
507cee36 | 1364 | const char *lxcpath, bool backgrounded) |
ee70bf78 CLG |
1365 | { |
1366 | struct start_args start_arg = { | |
1367 | .argv = argv, | |
1368 | }; | |
1369 | ||
828695d9 | 1370 | conf->need_utmp_watch = 1; |
507cee36 | 1371 | return __lxc_start(name, conf, &start_ops, &start_arg, lxcpath, backgrounded); |
ee70bf78 | 1372 | } |
28272964 CB |
1373 | |
1374 | static void lxc_destroy_container_on_signal(struct lxc_handler *handler, | |
1375 | const char *name) | |
1376 | { | |
1377 | char destroy[MAXPATHLEN]; | |
1378 | bool bret = true; | |
1379 | int ret = 0; | |
f01f7975 | 1380 | struct lxc_container *c; |
df31363a | 1381 | if (handler->conf->rootfs.path && handler->conf->rootfs.mount) { |
28272964 CB |
1382 | bret = do_destroy_container(handler->conf); |
1383 | if (!bret) { | |
1384 | ERROR("Error destroying rootfs for %s", name); | |
1385 | return; | |
1386 | } | |
1387 | } | |
1388 | INFO("Destroyed rootfs for %s", name); | |
1389 | ||
1390 | ret = snprintf(destroy, MAXPATHLEN, "%s/%s", handler->lxcpath, name); | |
1391 | if (ret < 0 || ret >= MAXPATHLEN) { | |
1392 | ERROR("Error printing path for %s", name); | |
1393 | ERROR("Error destroying directory for %s", name); | |
1394 | return; | |
1395 | } | |
1396 | ||
f01f7975 CB |
1397 | c = lxc_container_new(name, handler->lxcpath); |
1398 | if (c) { | |
1399 | if (container_disk_lock(c)) { | |
1400 | INFO("Could not update lxc_snapshots file"); | |
1401 | lxc_container_put(c); | |
1402 | } else { | |
1403 | mod_all_rdeps(c, false); | |
1404 | container_disk_unlock(c); | |
1405 | lxc_container_put(c); | |
1406 | } | |
1407 | } | |
1408 | ||
28272964 CB |
1409 | if (am_unpriv()) |
1410 | ret = userns_exec_1(handler->conf, lxc_rmdir_onedev_wrapper, destroy); | |
1411 | else | |
1412 | ret = lxc_rmdir_onedev(destroy, NULL); | |
1413 | ||
1414 | if (ret < 0) { | |
1415 | ERROR("Error destroying directory for %s", name); | |
1416 | return; | |
1417 | } | |
1418 | INFO("Destroyed directory for %s", name); | |
1419 | } | |
1420 | ||
1421 | static int lxc_rmdir_onedev_wrapper(void *data) | |
1422 | { | |
1423 | char *arg = (char *) data; | |
1424 | return lxc_rmdir_onedev(arg, NULL); | |
1425 | } | |
1426 | ||
1427 | static bool do_destroy_container(struct lxc_conf *conf) { | |
d028235d SG |
1428 | if (am_unpriv()) { |
1429 | if (userns_exec_1(conf, bdev_destroy_wrapper, conf) < 0) | |
1430 | return false; | |
1431 | return true; | |
1432 | } | |
1433 | return bdev_destroy(conf); | |
28272964 CB |
1434 | } |
1435 |