]>
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 | ||
656994bb MH |
53 | #if !HAVE_DECL_PR_CAPBSET_DROP |
54 | #define PR_CAPBSET_DROP 24 | |
55 | #endif | |
56 | ||
f3787121 CB |
57 | #include "af_unix.h" |
58 | #include "caps.h" | |
563f2f2c | 59 | #include "cgroup.h" |
f3787121 CB |
60 | #include "commands.h" |
61 | #include "conf.h" | |
62 | #include "console.h" | |
e2bcd7db | 63 | #include "error.h" |
f3787121 CB |
64 | #include "log.h" |
65 | #include "lxclock.h" | |
66 | #include "lxcseccomp.h" | |
565c2d76 | 67 | #include "lxcutmp.h" |
f3787121 | 68 | #include "mainloop.h" |
63376d7d | 69 | #include "monitor.h" |
f549edcc | 70 | #include "namespace.h" |
f3787121 CB |
71 | #include "start.h" |
72 | #include "sync.h" | |
73 | #include "utils.h" | |
4ec31c52 | 74 | #include "bdev/bdev.h" |
fe4de9a6 | 75 | #include "lsm/lsm.h" |
36eb9bde CLG |
76 | |
77 | lxc_log_define(lxc_start, lxc); | |
78 | ||
9f30a190 MM |
79 | const struct ns_info ns_info[LXC_NS_MAX] = { |
80 | [LXC_NS_MNT] = {"mnt", CLONE_NEWNS}, | |
81 | [LXC_NS_PID] = {"pid", CLONE_NEWPID}, | |
82 | [LXC_NS_UTS] = {"uts", CLONE_NEWUTS}, | |
83 | [LXC_NS_IPC] = {"ipc", CLONE_NEWIPC}, | |
84 | [LXC_NS_USER] = {"user", CLONE_NEWUSER}, | |
85 | [LXC_NS_NET] = {"net", CLONE_NEWNET} | |
86 | }; | |
87 | ||
f01f7975 | 88 | extern void mod_all_rdeps(struct lxc_container *c, bool inc); |
28272964 CB |
89 | static bool do_destroy_container(struct lxc_conf *conf); |
90 | static int lxc_rmdir_onedev_wrapper(void *data); | |
91 | static void lxc_destroy_container_on_signal(struct lxc_handler *handler, | |
92 | const char *name); | |
93 | ||
c8154066 SH |
94 | static void print_top_failing_dir(const char *path) |
95 | { | |
96 | size_t len = strlen(path); | |
f3787121 | 97 | char *copy = alloca(len + 1), *p, *e, saved; |
c8154066 SH |
98 | strcpy(copy, path); |
99 | ||
100 | p = copy; | |
101 | e = copy + len; | |
102 | while (p < e) { | |
f3787121 CB |
103 | while (p < e && *p == '/') |
104 | p++; | |
105 | while (p < e && *p != '/') | |
106 | p++; | |
c8154066 SH |
107 | saved = *p; |
108 | *p = '\0'; | |
109 | if (access(copy, X_OK)) { | |
110 | SYSERROR("could not access %s. Please grant it 'x' " \ | |
111 | "access, or add an ACL for the container root.", | |
112 | copy); | |
113 | return; | |
114 | } | |
115 | *p = saved; | |
116 | } | |
117 | } | |
118 | ||
9f30a190 MM |
119 | static void close_ns(int ns_fd[LXC_NS_MAX]) { |
120 | int i; | |
121 | ||
9f30a190 MM |
122 | for (i = 0; i < LXC_NS_MAX; i++) { |
123 | if (ns_fd[i] > -1) { | |
124 | close(ns_fd[i]); | |
125 | ns_fd[i] = -1; | |
126 | } | |
127 | } | |
9f30a190 MM |
128 | } |
129 | ||
62d05d9b SH |
130 | /* |
131 | * preserve_ns: open /proc/@pid/ns/@ns for each namespace specified | |
132 | * in clone_flags. | |
133 | * Return true on success, false on failure. On failure, leave an error | |
134 | * message in *errmsg, which caller must free. | |
135 | */ | |
f3787121 CB |
136 | static bool preserve_ns(int ns_fd[LXC_NS_MAX], int clone_flags, pid_t pid, |
137 | char **errmsg) | |
138 | { | |
62d05d9b | 139 | int i, ret; |
9f30a190 MM |
140 | char path[MAXPATHLEN]; |
141 | ||
9f30a190 MM |
142 | for (i = 0; i < LXC_NS_MAX; i++) |
143 | ns_fd[i] = -1; | |
144 | ||
035a38fc WB |
145 | snprintf(path, MAXPATHLEN, "/proc/%d/ns", pid); |
146 | if (access(path, X_OK)) { | |
62d05d9b SH |
147 | if (asprintf(errmsg, "Kernel does not support setns.") == -1) |
148 | *errmsg = NULL; | |
149 | return false; | |
cd43d2d1 SH |
150 | } |
151 | ||
9f30a190 MM |
152 | for (i = 0; i < LXC_NS_MAX; i++) { |
153 | if ((clone_flags & ns_info[i].clone_flag) == 0) | |
154 | continue; | |
035a38fc WB |
155 | snprintf(path, MAXPATHLEN, "/proc/%d/ns/%s", pid, |
156 | ns_info[i].proc_name); | |
9f30a190 | 157 | ns_fd[i] = open(path, O_RDONLY | O_CLOEXEC); |
9f30a190 MM |
158 | if (ns_fd[i] < 0) |
159 | goto error; | |
160 | } | |
161 | ||
62d05d9b | 162 | return true; |
9f30a190 MM |
163 | |
164 | error: | |
62d05d9b SH |
165 | if (errno == ENOENT) { |
166 | ret = asprintf(errmsg, "Kernel does not support setns for %s", | |
167 | ns_info[i].proc_name); | |
168 | } else { | |
169 | ret = asprintf(errmsg, "Failed to open %s: %s", | |
170 | path, strerror(errno)); | |
171 | } | |
172 | if (ret == -1) | |
173 | *errmsg = NULL; | |
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: | |
191 | SYSERROR("failed to set namespace '%s'", ns_info[i].proc_name); | |
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 | ||
d2cf4c37 SH |
200 | /* |
201 | * Check for any fds we need to close | |
202 | * * if fd_to_ignore != -1, then if we find that fd open we will ignore it. | |
203 | * * By default we warn about open fds we find. | |
204 | * * If closeall is true, we will close open fds. | |
205 | * * If lxc-start was passed "-C", then conf->close_all_fds will be true, | |
206 | * in which case we also close all open fds. | |
207 | * * A daemonized container will always pass closeall=true. | |
208 | */ | |
209 | int lxc_check_inherited(struct lxc_conf *conf, bool closeall, int fd_to_ignore) | |
80090207 CLG |
210 | { |
211 | struct dirent dirent, *direntp; | |
212 | int fd, fddir; | |
213 | DIR *dir; | |
80090207 | 214 | |
d2cf4c37 SH |
215 | if (conf && conf->close_all_fds) |
216 | closeall = true; | |
217 | ||
b119f362 | 218 | restart: |
80090207 CLG |
219 | dir = opendir("/proc/self/fd"); |
220 | if (!dir) { | |
221 | WARN("failed to open directory: %m"); | |
222 | return -1; | |
223 | } | |
224 | ||
225 | fddir = dirfd(dir); | |
226 | ||
227 | while (!readdir_r(dir, &dirent, &direntp)) { | |
80090207 CLG |
228 | if (!direntp) |
229 | break; | |
230 | ||
231 | if (!strcmp(direntp->d_name, ".")) | |
232 | continue; | |
233 | ||
234 | if (!strcmp(direntp->d_name, "..")) | |
235 | continue; | |
236 | ||
237 | fd = atoi(direntp->d_name); | |
238 | ||
f2faa8fa | 239 | if (fd == fddir || fd == lxc_log_fd || fd == fd_to_ignore) |
80090207 CLG |
240 | continue; |
241 | ||
858377e4 SH |
242 | if (current_config && fd == current_config->logfd) |
243 | continue; | |
244 | ||
80090207 CLG |
245 | if (match_fd(fd)) |
246 | continue; | |
80090207 | 247 | |
d2cf4c37 | 248 | if (closeall) { |
b119f362 SH |
249 | close(fd); |
250 | closedir(dir); | |
251 | INFO("closed inherited fd %d", fd); | |
252 | goto restart; | |
253 | } | |
92c7f629 | 254 | WARN("inherited fd %d", fd); |
80090207 CLG |
255 | } |
256 | ||
92c7f629 GK |
257 | closedir(dir); /* cannot fail */ |
258 | return 0; | |
80090207 CLG |
259 | } |
260 | ||
83ee7875 | 261 | static int setup_signal_fd(sigset_t *oldmask) |
b0a33c1e | 262 | { |
263 | sigset_t mask; | |
264 | int fd; | |
265 | ||
f3304a29 FW |
266 | /* Block everything except serious error signals */ |
267 | if (sigfillset(&mask) || | |
268 | sigdelset(&mask, SIGILL) || | |
269 | sigdelset(&mask, SIGSEGV) || | |
270 | sigdelset(&mask, SIGBUS) || | |
b5159817 | 271 | sigdelset(&mask, SIGWINCH) || |
f3304a29 FW |
272 | sigprocmask(SIG_BLOCK, &mask, oldmask)) { |
273 | SYSERROR("failed to set signal mask"); | |
b0a33c1e | 274 | return -1; |
275 | } | |
276 | ||
277 | fd = signalfd(-1, &mask, 0); | |
278 | if (fd < 0) { | |
36eb9bde | 279 | SYSERROR("failed to create the signal fd"); |
b0a33c1e | 280 | return -1; |
281 | } | |
282 | ||
283 | if (fcntl(fd, F_SETFD, FD_CLOEXEC)) { | |
36eb9bde | 284 | SYSERROR("failed to set sigfd to close-on-exec"); |
b0a33c1e | 285 | close(fd); |
286 | return -1; | |
287 | } | |
288 | ||
1ac470c0 DL |
289 | DEBUG("sigchild handler set"); |
290 | ||
b0a33c1e | 291 | return fd; |
292 | } | |
293 | ||
84c92abd | 294 | static int signal_handler(int fd, uint32_t events, void *data, |
f3787121 | 295 | struct lxc_epoll_descr *descr) |
b0a33c1e | 296 | { |
15cd25fd | 297 | struct signalfd_siginfo siginfo; |
80507ee8 | 298 | siginfo_t info; |
15cd25fd | 299 | int ret; |
82d89dce | 300 | pid_t *pid = data; |
80507ee8 | 301 | bool init_died = false; |
15cd25fd DL |
302 | |
303 | ret = read(fd, &siginfo, sizeof(siginfo)); | |
304 | if (ret < 0) { | |
f3304a29 | 305 | ERROR("failed to read signal info"); |
15cd25fd DL |
306 | return -1; |
307 | } | |
308 | ||
309 | if (ret != sizeof(siginfo)) { | |
310 | ERROR("unexpected siginfo size"); | |
311 | return -1; | |
312 | } | |
313 | ||
80507ee8 SH |
314 | // check whether init is running |
315 | info.si_pid = 0; | |
316 | ret = waitid(P_PID, *pid, &info, WEXITED | WNOWAIT | WNOHANG); | |
317 | if (ret == 0 && info.si_pid == *pid) { | |
318 | init_died = true; | |
319 | } | |
320 | ||
f3304a29 FW |
321 | if (siginfo.ssi_signo != SIGCHLD) { |
322 | kill(*pid, siginfo.ssi_signo); | |
323 | INFO("forwarded signal %d to pid %d", siginfo.ssi_signo, *pid); | |
80507ee8 | 324 | return init_died ? 1 : 0; |
f3304a29 FW |
325 | } |
326 | ||
15cd25fd DL |
327 | if (siginfo.ssi_code == CLD_STOPPED || |
328 | siginfo.ssi_code == CLD_CONTINUED) { | |
329 | INFO("container init process was stopped/continued"); | |
80507ee8 | 330 | return init_died ? 1 : 0; |
15cd25fd | 331 | } |
1ac470c0 | 332 | |
82d89dce DL |
333 | /* more robustness, protect ourself from a SIGCHLD sent |
334 | * by a process different from the container init | |
335 | */ | |
336 | if (siginfo.ssi_pid != *pid) { | |
337 | WARN("invalid pid for SIGCHLD"); | |
80507ee8 | 338 | return init_died ? 1 : 0; |
82d89dce DL |
339 | } |
340 | ||
15cd25fd | 341 | DEBUG("container init process exited"); |
b0a33c1e | 342 | return 1; |
343 | } | |
344 | ||
735f2c6e | 345 | int lxc_set_state(const char *name, struct lxc_handler *handler, lxc_state_t state) |
66aeffc7 DL |
346 | { |
347 | handler->state = state; | |
9123e471 | 348 | lxc_monitor_send_state(name, state, handler->lxcpath); |
66aeffc7 DL |
349 | return 0; |
350 | } | |
351 | ||
735f2c6e | 352 | int lxc_poll(const char *name, struct lxc_handler *handler) |
b0a33c1e | 353 | { |
ca5f7926 DL |
354 | int sigfd = handler->sigfd; |
355 | int pid = handler->pid; | |
b0a33c1e | 356 | struct lxc_epoll_descr descr; |
357 | ||
a9e61274 | 358 | if (lxc_mainloop_open(&descr)) { |
36eb9bde | 359 | ERROR("failed to create mainloop"); |
50c8bf05 | 360 | goto out_sigfd; |
b0a33c1e | 361 | } |
362 | ||
83ee7875 | 363 | if (lxc_mainloop_add_handler(&descr, sigfd, signal_handler, &pid)) { |
36eb9bde | 364 | ERROR("failed to add handler for the signal"); |
b0a33c1e | 365 | goto out_mainloop_open; |
366 | } | |
367 | ||
63376d7d DL |
368 | if (lxc_console_mainloop_add(&descr, handler)) { |
369 | ERROR("failed to add console handler to mainloop"); | |
370 | goto out_mainloop_open; | |
371 | } | |
372 | ||
ef6e34ee | 373 | if (lxc_cmd_mainloop_add(name, &descr, handler)) { |
563f2f2c | 374 | ERROR("failed to add command handler to mainloop"); |
96fa1ff0 | 375 | goto out_mainloop_open; |
563f2f2c DL |
376 | } |
377 | ||
828695d9 | 378 | if (handler->conf->need_utmp_watch) { |
495d2046 | 379 | #if HAVE_SYS_CAPABILITY_H |
828695d9 SH |
380 | if (lxc_utmp_mainloop_add(&descr, handler)) { |
381 | ERROR("failed to add utmp handler to mainloop"); | |
382 | goto out_mainloop_open; | |
383 | } | |
495d2046 | 384 | #else |
959aee9c | 385 | DEBUG("not starting utmp handler as cap_sys_boot cannot be dropped without capabilities support"); |
495d2046 | 386 | #endif |
563f2f2c | 387 | } |
b0a33c1e | 388 | |
e51d4895 | 389 | return lxc_mainloop(&descr, -1); |
b0a33c1e | 390 | |
391 | out_mainloop_open: | |
392 | lxc_mainloop_close(&descr); | |
b0a33c1e | 393 | out_sigfd: |
394 | close(sigfd); | |
c3e13372 | 395 | return -1; |
b0a33c1e | 396 | } |
397 | ||
13f5be62 | 398 | struct lxc_handler *lxc_init(const char *name, struct lxc_conf *conf, const char *lxcpath) |
59eb99ba | 399 | { |
b6b2b194 | 400 | int i; |
3a0f472d DL |
401 | struct lxc_handler *handler; |
402 | ||
403 | handler = malloc(sizeof(*handler)); | |
404 | if (!handler) | |
405 | return NULL; | |
59eb99ba DL |
406 | |
407 | memset(handler, 0, sizeof(*handler)); | |
408 | ||
e8bd4e43 | 409 | handler->ttysock[0] = handler->ttysock[1] = -1; |
fae349da | 410 | handler->conf = conf; |
9123e471 | 411 | handler->lxcpath = lxcpath; |
5c068da9 | 412 | handler->pinfd = -1; |
fae349da | 413 | |
b6b2b194 WB |
414 | for (i = 0; i < LXC_NS_MAX; i++) |
415 | handler->nsfd[i] = -1; | |
416 | ||
fe4de9a6 DE |
417 | lsm_init(); |
418 | ||
3bdf52d7 DL |
419 | handler->name = strdup(name); |
420 | if (!handler->name) { | |
421 | ERROR("failed to allocate memory"); | |
422 | goto out_free; | |
423 | } | |
424 | ||
ef6e34ee | 425 | if (lxc_cmd_init(name, handler, lxcpath)) |
d2e30e99 DE |
426 | goto out_free_name; |
427 | ||
8f2c3a70 SH |
428 | if (lxc_read_seccomp_config(conf) != 0) { |
429 | ERROR("failed loading seccomp policy"); | |
d2e30e99 | 430 | goto out_close_maincmd_fd; |
8f2c3a70 SH |
431 | } |
432 | ||
051151de | 433 | /* Begin by setting the state to STARTING */ |
25c2aca5 | 434 | if (lxc_set_state(name, handler, STARTING)) { |
59eb99ba | 435 | ERROR("failed to set state '%s'", lxc_state2str(STARTING)); |
051151de | 436 | goto out_close_maincmd_fd; |
0ad19a3f | 437 | } |
438 | ||
f7bee6c6 | 439 | /* Start of environment variable setup for hooks */ |
ab7efcf5 | 440 | if (name && setenv("LXC_NAME", name, 1)) { |
f7bee6c6 MW |
441 | SYSERROR("failed to set environment variable for container name"); |
442 | } | |
ab7efcf5 | 443 | if (conf->rcfile && setenv("LXC_CONFIG_FILE", conf->rcfile, 1)) { |
f7bee6c6 MW |
444 | SYSERROR("failed to set environment variable for config path"); |
445 | } | |
ab7efcf5 | 446 | if (conf->rootfs.mount && setenv("LXC_ROOTFS_MOUNT", conf->rootfs.mount, 1)) { |
f7bee6c6 MW |
447 | SYSERROR("failed to set environment variable for rootfs mount"); |
448 | } | |
ab7efcf5 | 449 | if (conf->rootfs.path && setenv("LXC_ROOTFS_PATH", conf->rootfs.path, 1)) { |
f7bee6c6 MW |
450 | SYSERROR("failed to set environment variable for rootfs mount"); |
451 | } | |
452 | if (conf->console.path && setenv("LXC_CONSOLE", conf->console.path, 1)) { | |
453 | SYSERROR("failed to set environment variable for console path"); | |
454 | } | |
455 | if (conf->console.log_path && setenv("LXC_CONSOLE_LOGPATH", conf->console.log_path, 1)) { | |
456 | SYSERROR("failed to set environment variable for console log"); | |
457 | } | |
c4cafa08 SH |
458 | if (setenv("LXC_CGNS_AWARE", "1", 1)) { |
459 | SYSERROR("failed to set LXC_CGNS_AWARE environment variable"); | |
460 | } | |
f7bee6c6 MW |
461 | /* End of environment variable setup for hooks */ |
462 | ||
283678ed | 463 | if (run_lxc_hooks(name, "pre-start", conf, handler->lxcpath, NULL)) { |
773fb9ca SH |
464 | ERROR("failed to run pre-start hooks for container '%s'.", name); |
465 | goto out_aborting; | |
466 | } | |
26ddeedd | 467 | |
b0a33c1e | 468 | /* the signal fd has to be created before forking otherwise |
469 | * if the child process exits before we setup the signal fd, | |
470 | * the event will be lost and the command will be stuck */ | |
83ee7875 | 471 | handler->sigfd = setup_signal_fd(&handler->oldmask); |
59eb99ba | 472 | if (handler->sigfd < 0) { |
36eb9bde | 473 | ERROR("failed to set sigchild fd handler"); |
b5159817 DE |
474 | goto out_delete_tty; |
475 | } | |
476 | ||
477 | /* do this after setting up signals since it might unblock SIGWINCH */ | |
478 | if (lxc_console_create(conf)) { | |
479 | ERROR("failed to create console"); | |
480 | goto out_restore_sigmask; | |
b0a33c1e | 481 | } |
482 | ||
c4d10a05 SH |
483 | if (ttys_shift_ids(conf) < 0) { |
484 | ERROR("Failed to shift tty into container"); | |
485 | goto out_restore_sigmask; | |
486 | } | |
487 | ||
c3e13372 | 488 | INFO("'%s' is initialized", name); |
3a0f472d | 489 | return handler; |
59eb99ba | 490 | |
b5159817 DE |
491 | out_restore_sigmask: |
492 | sigprocmask(SIG_SETMASK, &handler->oldmask, NULL); | |
59eb99ba | 493 | out_delete_tty: |
fae349da | 494 | lxc_delete_tty(&conf->tty_info); |
59eb99ba | 495 | out_aborting: |
25c2aca5 | 496 | lxc_set_state(name, handler, ABORTING); |
d2e30e99 DE |
497 | out_close_maincmd_fd: |
498 | close(conf->maincmd_fd); | |
499 | conf->maincmd_fd = -1; | |
3bdf52d7 DL |
500 | out_free_name: |
501 | free(handler->name); | |
502 | handler->name = NULL; | |
3a0f472d DL |
503 | out_free: |
504 | free(handler); | |
c3e13372 | 505 | return NULL; |
59eb99ba DL |
506 | } |
507 | ||
3b72c4a0 | 508 | void lxc_fini(const char *name, struct lxc_handler *handler) |
59eb99ba | 509 | { |
b3286b62 WB |
510 | int i, rc; |
511 | pid_t self = getpid(); | |
512 | char *namespaces[LXC_NS_MAX+1]; | |
513 | size_t namespace_count = 0; | |
b6b2b194 | 514 | |
59eb99ba DL |
515 | /* The STOPPING state is there for future cleanup code |
516 | * which can take awhile | |
517 | */ | |
25c2aca5 | 518 | lxc_set_state(name, handler, STOPPING); |
b6b2b194 | 519 | |
b3286b62 WB |
520 | for (i = 0; i < LXC_NS_MAX; i++) { |
521 | if (handler->nsfd[i] != -1) { | |
522 | rc = asprintf(&namespaces[namespace_count], "%s:/proc/%d/fd/%d", | |
523 | ns_info[i].proc_name, self, handler->nsfd[i]); | |
524 | if (rc == -1) { | |
525 | SYSERROR("failed to allocate memory"); | |
526 | break; | |
527 | } | |
528 | ++namespace_count; | |
529 | } | |
530 | } | |
531 | namespaces[namespace_count] = NULL; | |
c154af98 SG |
532 | |
533 | if (handler->conf->reboot && setenv("LXC_TARGET", "reboot", 1)) { | |
534 | SYSERROR("failed to set environment variable for stop target"); | |
535 | } | |
536 | if (!handler->conf->reboot && setenv("LXC_TARGET", "stop", 1)) { | |
537 | SYSERROR("failed to set environment variable for stop target"); | |
538 | } | |
539 | ||
b3286b62 | 540 | if (run_lxc_hooks(name, "stop", handler->conf, handler->lxcpath, namespaces)) |
8438bfbd | 541 | ERROR("failed to run stop hooks for container '%s'.", name); |
c154af98 | 542 | |
b3286b62 WB |
543 | while (namespace_count--) |
544 | free(namespaces[namespace_count]); | |
b6b2b194 WB |
545 | for (i = 0; i < LXC_NS_MAX; i++) { |
546 | if (handler->nsfd[i] != -1) { | |
547 | close(handler->nsfd[i]); | |
548 | handler->nsfd[i] = -1; | |
549 | } | |
550 | } | |
25c2aca5 | 551 | lxc_set_state(name, handler, STOPPED); |
59eb99ba | 552 | |
f3787121 | 553 | if (run_lxc_hooks(name, "post-stop", handler->conf, handler->lxcpath, NULL)) { |
773fb9ca | 554 | ERROR("failed to run post-stop hooks for container '%s'.", name); |
f3787121 CB |
555 | if (handler->conf->reboot) { |
556 | WARN("Container will be stopped instead of rebooted."); | |
557 | handler->conf->reboot = 0; | |
558 | setenv("LXC_TARGET", "stop", 1); | |
559 | } | |
560 | } | |
26ddeedd | 561 | |
83ee7875 | 562 | /* reset mask set by setup_signal_fd */ |
8f64a3f6 MN |
563 | if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL)) |
564 | WARN("failed to restore sigprocmask"); | |
565 | ||
b5159817 | 566 | lxc_console_delete(&handler->conf->console); |
b2431939 | 567 | lxc_delete_tty(&handler->conf->tty_info); |
d2e30e99 DE |
568 | close(handler->conf->maincmd_fd); |
569 | handler->conf->maincmd_fd = -1; | |
3bdf52d7 | 570 | free(handler->name); |
e8bd4e43 SH |
571 | if (handler->ttysock[0] != -1) { |
572 | close(handler->ttysock[0]); | |
573 | close(handler->ttysock[1]); | |
574 | } | |
2c5f2ede CB |
575 | |
576 | if (handler->conf->ephemeral == 1 && handler->conf->reboot != 1) | |
28272964 | 577 | lxc_destroy_container_on_signal(handler, name); |
2c5f2ede | 578 | |
d4ef7c50 | 579 | cgroup_destroy(handler); |
b2431939 | 580 | free(handler); |
59eb99ba DL |
581 | } |
582 | ||
735f2c6e | 583 | void lxc_abort(const char *name, struct lxc_handler *handler) |
59eb99ba | 584 | { |
73e608b2 SH |
585 | int ret, status; |
586 | ||
25c2aca5 | 587 | lxc_set_state(name, handler, ABORTING); |
7d9fb3e9 DL |
588 | if (handler->pid > 0) |
589 | kill(handler->pid, SIGKILL); | |
73e608b2 | 590 | while ((ret = waitpid(-1, &status, 0)) > 0) ; |
59eb99ba DL |
591 | } |
592 | ||
828695d9 SH |
593 | #include <sys/reboot.h> |
594 | #include <linux/reboot.h> | |
595 | ||
e2fa1520 SH |
596 | /* |
597 | * reboot(LINUX_REBOOT_CMD_CAD_ON) will return -EINVAL | |
598 | * in a child pid namespace if container reboot support exists. | |
599 | * Otherwise, it will either succeed or return -EPERM. | |
600 | */ | |
601 | static int container_reboot_supported(void *arg) | |
828695d9 | 602 | { |
d028235d | 603 | int *cmd = arg; |
828695d9 | 604 | int ret; |
828695d9 | 605 | |
d028235d | 606 | ret = reboot(*cmd); |
e2fa1520 SH |
607 | if (ret == -1 && errno == EINVAL) |
608 | return 1; | |
609 | return 0; | |
610 | } | |
611 | ||
b60ed720 | 612 | static int must_drop_cap_sys_boot(struct lxc_conf *conf) |
e2fa1520 | 613 | { |
025ed0f3 | 614 | FILE *f; |
b60ed720 | 615 | int ret, cmd, v, flags; |
d028235d SG |
616 | long stack_size = 4096; |
617 | void *stack = alloca(stack_size); | |
618 | int status; | |
619 | pid_t pid; | |
e2fa1520 | 620 | |
025ed0f3 | 621 | f = fopen("/proc/sys/kernel/ctrl-alt-del", "r"); |
e2fa1520 SH |
622 | if (!f) { |
623 | DEBUG("failed to open /proc/sys/kernel/ctrl-alt-del"); | |
828695d9 | 624 | return 1; |
e2fa1520 | 625 | } |
828695d9 SH |
626 | |
627 | ret = fscanf(f, "%d", &v); | |
628 | fclose(f); | |
e2fa1520 SH |
629 | if (ret != 1) { |
630 | DEBUG("Failed to read /proc/sys/kernel/ctrl-alt-del"); | |
828695d9 | 631 | return 1; |
e2fa1520 SH |
632 | } |
633 | cmd = v ? LINUX_REBOOT_CMD_CAD_ON : LINUX_REBOOT_CMD_CAD_OFF; | |
634 | ||
b60ed720 SH |
635 | flags = CLONE_NEWPID | SIGCHLD; |
636 | if (!lxc_list_empty(&conf->id_map)) | |
637 | flags |= CLONE_NEWUSER; | |
638 | ||
7f145a6d | 639 | #ifdef __ia64__ |
959aee9c | 640 | pid = __clone2(container_reboot_supported, stack, stack_size, flags, &cmd); |
7f145a6d | 641 | #else |
959aee9c SG |
642 | stack += stack_size; |
643 | pid = clone(container_reboot_supported, stack, flags, &cmd); | |
7f145a6d | 644 | #endif |
959aee9c | 645 | if (pid < 0) { |
c08220e9 | 646 | if (flags & CLONE_NEWUSER) |
4a7e5f4f | 647 | ERROR("failed to clone (%#x): %s (includes CLONE_NEWUSER)", flags, strerror(errno)); |
c08220e9 | 648 | else |
4a7e5f4f | 649 | ERROR("failed to clone (%#x): %s", flags, strerror(errno)); |
959aee9c SG |
650 | return -1; |
651 | } | |
652 | if (wait(&status) < 0) { | |
653 | SYSERROR("unexpected wait error: %m"); | |
654 | return -1; | |
655 | } | |
e2fa1520 SH |
656 | |
657 | if (WEXITSTATUS(status) != 1) | |
828695d9 | 658 | return 1; |
e2fa1520 | 659 | |
828695d9 SH |
660 | return 0; |
661 | } | |
662 | ||
658979c5 SH |
663 | /* |
664 | * netpipe is used in the unprivileged case to transfer the ifindexes | |
665 | * from parent to child | |
666 | */ | |
667 | static int netpipe = -1; | |
668 | ||
669 | static inline int count_veths(struct lxc_list *network) | |
670 | { | |
671 | struct lxc_list *iterator; | |
672 | struct lxc_netdev *netdev; | |
673 | int count = 0; | |
674 | ||
675 | lxc_list_for_each(iterator, network) { | |
676 | netdev = iterator->elem; | |
677 | if (netdev->type != LXC_NET_VETH) | |
678 | continue; | |
679 | count++; | |
680 | } | |
681 | return count; | |
682 | } | |
683 | ||
684 | static int read_unpriv_netifindex(struct lxc_list *network) | |
685 | { | |
686 | struct lxc_list *iterator; | |
687 | struct lxc_netdev *netdev; | |
688 | ||
689 | if (netpipe == -1) | |
690 | return 0; | |
691 | lxc_list_for_each(iterator, network) { | |
692 | netdev = iterator->elem; | |
693 | if (netdev->type != LXC_NET_VETH) | |
694 | continue; | |
695 | if (!(netdev->name = malloc(IFNAMSIZ))) { | |
696 | ERROR("Out of memory"); | |
697 | close(netpipe); | |
698 | return -1; | |
699 | } | |
700 | if (read(netpipe, netdev->name, IFNAMSIZ) != IFNAMSIZ) { | |
701 | close(netpipe); | |
702 | return -1; | |
703 | } | |
704 | } | |
705 | close(netpipe); | |
706 | return 0; | |
707 | } | |
708 | ||
ffe1e01a | 709 | static int do_start(void *data) |
50e98013 | 710 | { |
7c661726 | 711 | struct lxc_list *iterator; |
23c53af9 | 712 | struct lxc_handler *handler = data; |
50e98013 DL |
713 | |
714 | if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL)) { | |
715 | SYSERROR("failed to set sigprocmask"); | |
9d7f9e52 | 716 | return -1; |
50e98013 DL |
717 | } |
718 | ||
d028235d | 719 | /* This prctl must be before the synchro, so if the parent |
743ecd2e DL |
720 | * dies before we set the parent death signal, we will detect |
721 | * its death with the synchro right after, otherwise we have | |
722 | * a window where the parent can exit before we set the pdeath | |
723 | * signal leading to a unsupervized container. | |
724 | */ | |
725 | if (prctl(PR_SET_PDEATHSIG, SIGKILL, 0, 0, 0)) { | |
726 | SYSERROR("failed to set pdeath signal"); | |
727 | return -1; | |
728 | } | |
729 | ||
3c22086f | 730 | lxc_sync_fini_parent(handler); |
50e98013 | 731 | |
2b0e17e4 | 732 | /* don't leak the pinfd to the container */ |
025ed0f3 | 733 | if (handler->pinfd >= 0) { |
0d03360a | 734 | close(handler->pinfd); |
025ed0f3 | 735 | } |
2b0e17e4 | 736 | |
3c22086f CLG |
737 | /* Tell the parent task it can begin to configure the |
738 | * container and wait for it to finish | |
739 | */ | |
740 | if (lxc_sync_barrier_parent(handler, LXC_SYNC_CONFIGURE)) | |
9d7f9e52 | 741 | return -1; |
50e98013 | 742 | |
658979c5 SH |
743 | if (read_unpriv_netifindex(&handler->conf->network) < 0) |
744 | goto out_warn_father; | |
745 | ||
f6d3e3e4 SH |
746 | /* |
747 | * if we are in a new user namespace, become root there to have | |
62d00539 PT |
748 | * privilege over our namespace. When using lxc-execute we default to root, |
749 | * but this can be overriden using the lxc.init_uid and lxc.init_gid | |
750 | * configuration options. | |
f6d3e3e4 SH |
751 | */ |
752 | if (!lxc_list_empty(&handler->conf->id_map)) { | |
c5cd20ce PT |
753 | gid_t new_gid = 0; |
754 | if (handler->conf->is_execute && handler->conf->init_gid) | |
755 | new_gid = handler->conf->init_gid; | |
756 | ||
757 | uid_t new_uid = 0; | |
758 | if (handler->conf->is_execute && handler->conf->init_uid) | |
759 | new_uid = handler->conf->init_uid; | |
760 | ||
56f8ff00 PT |
761 | NOTICE("switching to gid/uid %d/%d in new user namespace", new_gid, new_uid); |
762 | if (setgid(new_gid)) { | |
f6d3e3e4 SH |
763 | SYSERROR("setgid"); |
764 | goto out_warn_father; | |
765 | } | |
56f8ff00 | 766 | if (setuid(new_uid)) { |
f6d3e3e4 SH |
767 | SYSERROR("setuid"); |
768 | goto out_warn_father; | |
769 | } | |
c476bdce SH |
770 | if (setgroups(0, NULL)) { |
771 | SYSERROR("setgroups"); | |
772 | goto out_warn_father; | |
773 | } | |
f6d3e3e4 SH |
774 | } |
775 | ||
99b71824 | 776 | if (access(handler->lxcpath, X_OK)) { |
c8154066 SH |
777 | print_top_failing_dir(handler->lxcpath); |
778 | goto out_warn_father; | |
779 | } | |
780 | ||
495d2046 | 781 | #if HAVE_SYS_CAPABILITY_H |
69182a31 | 782 | if (handler->conf->need_utmp_watch) { |
828695d9 SH |
783 | if (prctl(PR_CAPBSET_DROP, CAP_SYS_BOOT, 0, 0, 0)) { |
784 | SYSERROR("failed to remove CAP_SYS_BOOT capability"); | |
c4ea60df | 785 | goto out_warn_father; |
828695d9 | 786 | } |
959aee9c | 787 | DEBUG("Dropped cap_sys_boot"); |
e2fa1520 | 788 | } |
495d2046 | 789 | #endif |
e2fa1520 SH |
790 | |
791 | /* Setup the container, ip, names, utsname, ... */ | |
d4ef7c50 | 792 | if (lxc_setup(handler)) { |
e2fa1520 SH |
793 | ERROR("failed to setup the container"); |
794 | goto out_warn_father; | |
795 | } | |
50e98013 | 796 | |
544a48a0 SH |
797 | /* ask father to setup cgroups and wait for him to finish */ |
798 | if (lxc_sync_barrier_parent(handler, LXC_SYNC_CGROUP)) | |
799 | return -1; | |
800 | ||
72863294 | 801 | /* Set the label to change to when we exec(2) the container's init */ |
7aff4f43 | 802 | if (lsm_process_label_set(NULL, handler->conf, 1, 1) < 0) |
e075f5d9 | 803 | goto out_warn_father; |
5112cd70 | 804 | |
0d9acb99 DE |
805 | /* Some init's such as busybox will set sane tty settings on stdin, |
806 | * stdout, stderr which it thinks is the console. We already set them | |
807 | * the way we wanted on the real terminal, and we want init to do its | |
808 | * setup on its console ie. the pty allocated in lxc_console_create() | |
809 | * so make sure that that pty is stdin,stdout,stderr. | |
810 | */ | |
39a78bbe | 811 | if (lxc_console_set_stdfds(handler->conf->console.slave) < 0) |
0d9acb99 DE |
812 | goto out_warn_father; |
813 | ||
5112cd70 SH |
814 | /* If we mounted a temporary proc, then unmount it now */ |
815 | tmp_proc_unmount(handler->conf); | |
e075f5d9 | 816 | |
8f2c3a70 SH |
817 | if (lxc_seccomp_load(handler->conf) != 0) |
818 | goto out_warn_father; | |
819 | ||
283678ed | 820 | if (run_lxc_hooks(handler->name, "start", handler->conf, handler->lxcpath, NULL)) { |
773fb9ca SH |
821 | ERROR("failed to run start hooks for container '%s'.", handler->name); |
822 | goto out_warn_father; | |
823 | } | |
fc25b815 | 824 | |
f7bee6c6 | 825 | /* The clearenv() and putenv() calls have been moved here |
ec64264d | 826 | * to allow us to use environment variables passed to the various |
f7bee6c6 MW |
827 | * hooks, such as the start hook above. Not all of the |
828 | * variables like CONFIG_PATH or ROOTFS are valid in this | |
829 | * context but others are. */ | |
830 | if (clearenv()) { | |
831 | SYSERROR("failed to clear environment"); | |
832 | /* don't error out though */ | |
833 | } | |
834 | ||
7c661726 MP |
835 | lxc_list_for_each(iterator, &handler->conf->environment) { |
836 | if (putenv((char *)iterator->elem)) { | |
837 | SYSERROR("failed to set environment variable '%s'", (char *)iterator->elem); | |
838 | goto out_warn_father; | |
839 | } | |
840 | } | |
841 | ||
f7bee6c6 | 842 | if (putenv("container=lxc")) { |
7c661726 | 843 | SYSERROR("failed to set environment variable 'container=lxc'"); |
c4ea60df | 844 | goto out_warn_father; |
f7bee6c6 MW |
845 | } |
846 | ||
393903d1 SH |
847 | if (handler->conf->pty_names) { |
848 | if (putenv(handler->conf->pty_names)) { | |
849 | SYSERROR("failed to set environment variable for container ptys"); | |
850 | goto out_warn_father; | |
851 | } | |
852 | } | |
853 | ||
773fb9ca | 854 | close(handler->sigfd); |
26ddeedd | 855 | |
69aeabac TA |
856 | if (handler->backgrounded && null_stdfds() < 0) |
857 | goto out_warn_father; | |
507cee36 | 858 | |
12983ba4 SH |
859 | if (cgns_supported() && unshare(CLONE_NEWCGROUP) != 0) { |
860 | SYSERROR("Failed to unshare cgroup namespace"); | |
861 | goto out_warn_father; | |
862 | } | |
863 | ||
8c9a7665 TA |
864 | setsid(); |
865 | ||
e6126dbe MN |
866 | /* after this call, we are in error because this |
867 | * ops should not return as it execs */ | |
c4ea60df | 868 | handler->ops->start(handler, handler->data); |
50e98013 DL |
869 | |
870 | out_warn_father: | |
d1ccb562 AM |
871 | /* we want the parent to know something went wrong, so we return a special |
872 | * error code. */ | |
873 | lxc_sync_wake_parent(handler, LXC_SYNC_ERROR); | |
50e98013 DL |
874 | return -1; |
875 | } | |
876 | ||
74a3920a | 877 | static int save_phys_nics(struct lxc_conf *conf) |
7b35f3d6 SH |
878 | { |
879 | struct lxc_list *iterator; | |
40f2f8a2 | 880 | int am_root = (getuid() == 0); |
7b35f3d6 | 881 | |
40f2f8a2 LQ |
882 | if (!am_root) |
883 | return 0; | |
2c5f2ede | 884 | |
7b35f3d6 SH |
885 | lxc_list_for_each(iterator, &conf->network) { |
886 | struct lxc_netdev *netdev = iterator->elem; | |
887 | ||
888 | if (netdev->type != LXC_NET_PHYS) | |
889 | continue; | |
890 | conf->saved_nics = realloc(conf->saved_nics, | |
891 | (conf->num_savednics+1)*sizeof(struct saved_nic)); | |
892 | if (!conf->saved_nics) { | |
893 | SYSERROR("failed to allocate memory"); | |
894 | return -1; | |
895 | } | |
896 | conf->saved_nics[conf->num_savednics].ifindex = netdev->ifindex; | |
897 | conf->saved_nics[conf->num_savednics].orig_name = strdup(netdev->link); | |
898 | if (!conf->saved_nics[conf->num_savednics].orig_name) { | |
899 | SYSERROR("failed to allocate memory"); | |
900 | return -1; | |
901 | } | |
959aee9c | 902 | INFO("stored saved_nic #%d idx %d name %s", conf->num_savednics, |
7b35f3d6 SH |
903 | conf->saved_nics[conf->num_savednics].ifindex, |
904 | conf->saved_nics[conf->num_savednics].orig_name); | |
905 | conf->num_savednics++; | |
906 | } | |
907 | ||
908 | return 0; | |
909 | } | |
910 | ||
e8bd4e43 SH |
911 | static int recv_fd(int sock, int *fd) |
912 | { | |
913 | if (lxc_abstract_unix_recv_fd(sock, fd, NULL, 0) < 0) { | |
914 | SYSERROR("Error receiving tty fd from child"); | |
915 | return -1; | |
916 | } | |
917 | if (*fd == -1) | |
918 | return -1; | |
919 | return 0; | |
920 | } | |
921 | ||
922 | static int recv_ttys_from_child(struct lxc_handler *handler) | |
923 | { | |
924 | struct lxc_conf *conf = handler->conf; | |
925 | int i, sock = handler->ttysock[1]; | |
926 | struct lxc_tty_info *tty_info = &conf->tty_info; | |
927 | ||
928 | if (!conf->tty) | |
929 | return 0; | |
930 | ||
931 | tty_info->pty_info = malloc(sizeof(*tty_info->pty_info)*conf->tty); | |
932 | if (!tty_info->pty_info) { | |
933 | SYSERROR("failed to allocate pty_info"); | |
934 | return -1; | |
935 | } | |
936 | ||
937 | for (i = 0; i < conf->tty; i++) { | |
938 | struct lxc_pty_info *pty_info = &tty_info->pty_info[i]; | |
939 | pty_info->busy = 0; | |
940 | if (recv_fd(sock, &pty_info->slave) < 0 || | |
941 | recv_fd(sock, &pty_info->master) < 0) { | |
942 | ERROR("Error receiving tty info from child"); | |
943 | return -1; | |
944 | } | |
945 | } | |
946 | tty_info->nbtty = conf->tty; | |
947 | ||
948 | return 0; | |
949 | } | |
950 | ||
f813849c TA |
951 | void resolve_clone_flags(struct lxc_handler *handler) |
952 | { | |
953 | handler->clone_flags = CLONE_NEWPID | CLONE_NEWNS; | |
954 | ||
955 | if (!lxc_list_empty(&handler->conf->id_map)) { | |
956 | INFO("Cloning a new user namespace"); | |
957 | handler->clone_flags |= CLONE_NEWUSER; | |
958 | } | |
959 | ||
960 | if (handler->conf->inherit_ns_fd[LXC_NS_NET] == -1) { | |
961 | if (!lxc_requests_empty_network(handler)) | |
962 | handler->clone_flags |= CLONE_NEWNET; | |
963 | } else { | |
964 | INFO("Inheriting a net namespace"); | |
965 | } | |
966 | ||
967 | if (handler->conf->inherit_ns_fd[LXC_NS_IPC] == -1) { | |
968 | handler->clone_flags |= CLONE_NEWIPC; | |
969 | } else { | |
970 | INFO("Inheriting an IPC namespace"); | |
971 | } | |
972 | ||
973 | if (handler->conf->inherit_ns_fd[LXC_NS_UTS] == -1) { | |
974 | handler->clone_flags |= CLONE_NEWUTS; | |
975 | } else { | |
976 | INFO("Inheriting a UTS namespace"); | |
977 | } | |
978 | } | |
979 | ||
74a3920a | 980 | static int lxc_spawn(struct lxc_handler *handler) |
59eb99ba | 981 | { |
b98f7d6e | 982 | int failed_before_rename = 0; |
ffe1e01a | 983 | const char *name = handler->name; |
62d05d9b | 984 | char *errmsg = NULL; |
7e4dfe0b | 985 | bool cgroups_connected = false; |
9f30a190 MM |
986 | int saved_ns_fd[LXC_NS_MAX]; |
987 | int preserve_mask = 0, i; | |
658979c5 | 988 | int netpipepair[2], nveths; |
9f30a190 | 989 | |
2f2623ec SH |
990 | netpipe = -1; |
991 | ||
9f30a190 | 992 | for (i = 0; i < LXC_NS_MAX; i++) |
6c544cb3 | 993 | if (handler->conf->inherit_ns_fd[i] != -1) |
9f30a190 | 994 | preserve_mask |= ns_info[i].clone_flag; |
50e98013 | 995 | |
3c22086f | 996 | if (lxc_sync_init(handler)) |
9d7f9e52 | 997 | return -1; |
0ad19a3f | 998 | |
e8bd4e43 SH |
999 | if (socketpair(AF_UNIX, SOCK_DGRAM, 0, handler->ttysock) < 0) { |
1000 | lxc_sync_fini(handler); | |
1001 | return -1; | |
1002 | } | |
1003 | ||
f813849c | 1004 | resolve_clone_flags(handler); |
9f30a190 | 1005 | |
f813849c | 1006 | if (handler->clone_flags & CLONE_NEWNET) { |
26b797f3 SH |
1007 | if (!lxc_list_empty(&handler->conf->network)) { |
1008 | ||
9f30a190 MM |
1009 | /* Find gateway addresses from the link device, which is |
1010 | * no longer accessible inside the container. Do this | |
1011 | * before creating network interfaces, since goto | |
1012 | * out_delete_net does not work before lxc_clone. */ | |
1013 | if (lxc_find_gateway_addresses(handler)) { | |
1014 | ERROR("failed to find gateway addresses"); | |
1015 | lxc_sync_fini(handler); | |
1016 | return -1; | |
1017 | } | |
1018 | ||
1019 | /* that should be done before the clone because we will | |
1020 | * fill the netdev index and use them in the child | |
1021 | */ | |
1022 | if (lxc_create_network(handler)) { | |
38521a3b | 1023 | ERROR("failed to create the network"); |
9f30a190 MM |
1024 | lxc_sync_fini(handler); |
1025 | return -1; | |
1026 | } | |
19a26f82 MK |
1027 | } |
1028 | ||
9f30a190 MM |
1029 | if (save_phys_nics(handler->conf)) { |
1030 | ERROR("failed to save physical nic info"); | |
1031 | goto out_abort; | |
82d5ae15 DL |
1032 | } |
1033 | } | |
1034 | ||
d4ef7c50 SH |
1035 | if (!cgroup_init(handler)) { |
1036 | ERROR("failed initializing cgroup support"); | |
33ad9f1a CS |
1037 | goto out_delete_net; |
1038 | } | |
1039 | ||
7e4dfe0b SH |
1040 | cgroups_connected = true; |
1041 | ||
d4ef7c50 SH |
1042 | if (!cgroup_create(handler)) { |
1043 | ERROR("failed creating cgroups"); | |
47d8fb3b CS |
1044 | goto out_delete_net; |
1045 | } | |
1046 | ||
0c547523 SH |
1047 | /* |
1048 | * if the rootfs is not a blockdev, prevent the container from | |
1049 | * marking it readonly. | |
5e32a990 ÇO |
1050 | * |
1051 | * if the container is unprivileged then skip rootfs pinning | |
0c547523 | 1052 | */ |
5e32a990 ÇO |
1053 | if (lxc_list_empty(&handler->conf->id_map)) { |
1054 | handler->pinfd = pin_rootfs(handler->conf->rootfs.path); | |
1055 | if (handler->pinfd == -1) | |
1056 | INFO("failed to pin the container's rootfs"); | |
1057 | } | |
0c547523 | 1058 | |
62d05d9b SH |
1059 | if (!preserve_ns(saved_ns_fd, preserve_mask, getpid(), &errmsg)) { |
1060 | SYSERROR("Failed to preserve requested namespaces: %s", | |
1061 | errmsg ? errmsg : "(Out of memory)"); | |
1062 | free(errmsg); | |
cd43d2d1 | 1063 | goto out_delete_net; |
62d05d9b | 1064 | } |
cd43d2d1 SH |
1065 | if (attach_ns(handler->conf->inherit_ns_fd) < 0) |
1066 | goto out_delete_net; | |
9f30a190 | 1067 | |
658979c5 SH |
1068 | if (am_unpriv() && (nveths = count_veths(&handler->conf->network))) { |
1069 | if (pipe(netpipepair) < 0) { | |
1070 | SYSERROR("Error creating pipe"); | |
1071 | goto out_delete_net; | |
1072 | } | |
1073 | /* store netpipe in the global var for do_start's use */ | |
1074 | netpipe = netpipepair[0]; | |
1075 | } | |
1076 | ||
0ad19a3f | 1077 | /* Create a process in a new set of namespaces */ |
d5088cf2 | 1078 | handler->pid = lxc_clone(do_start, handler, handler->clone_flags); |
59eb99ba | 1079 | if (handler->pid < 0) { |
36eb9bde | 1080 | SYSERROR("failed to fork into a new namespace"); |
7fef7a06 | 1081 | goto out_delete_net; |
0ad19a3f | 1082 | } |
1083 | ||
330cbe36 | 1084 | if (!preserve_ns(handler->nsfd, handler->clone_flags | preserve_mask, handler->pid, &errmsg)) { |
62d05d9b SH |
1085 | INFO("Failed to store namespace references for stop hook: %s", |
1086 | errmsg ? errmsg : "(Out of memory)"); | |
1087 | free(errmsg); | |
b6b2b194 WB |
1088 | } |
1089 | ||
cd43d2d1 SH |
1090 | if (attach_ns(saved_ns_fd)) |
1091 | WARN("failed to restore saved namespaces"); | |
9f30a190 | 1092 | |
3c22086f CLG |
1093 | lxc_sync_fini_child(handler); |
1094 | ||
1095 | if (lxc_sync_wait_child(handler, LXC_SYNC_CONFIGURE)) | |
99a6af52 | 1096 | failed_before_rename = 1; |
0ad19a3f | 1097 | |
d4ef7c50 SH |
1098 | if (!cgroup_create_legacy(handler)) { |
1099 | ERROR("failed to setup the legacy cgroups for %s", name); | |
ae5c8b8e | 1100 | goto out_delete_net; |
33ad9f1a | 1101 | } |
9daf6f5d SH |
1102 | if (!cgroup_setup_limits(handler, false)) { |
1103 | ERROR("failed to setup the cgroup limits for '%s'", name); | |
6031a6e5 DE |
1104 | goto out_delete_net; |
1105 | } | |
1106 | ||
d4ef7c50 | 1107 | if (!cgroup_enter(handler)) |
7fef7a06 | 1108 | goto out_delete_net; |
218d4250 | 1109 | |
0996e18a SH |
1110 | if (!cgroup_chown(handler)) |
1111 | goto out_delete_net; | |
1112 | ||
99a6af52 MN |
1113 | if (failed_before_rename) |
1114 | goto out_delete_net; | |
1115 | ||
0ad19a3f | 1116 | /* Create the network configuration */ |
d5088cf2 | 1117 | if (handler->clone_flags & CLONE_NEWNET) { |
c43cbc04 SH |
1118 | if (lxc_assign_network(handler->lxcpath, handler->name, |
1119 | &handler->conf->network, handler->pid)) { | |
82d5ae15 | 1120 | ERROR("failed to create the configured network"); |
7fef7a06 | 1121 | goto out_delete_net; |
82d5ae15 | 1122 | } |
0ad19a3f | 1123 | } |
1124 | ||
658979c5 SH |
1125 | if (netpipe != -1) { |
1126 | struct lxc_list *iterator; | |
1127 | struct lxc_netdev *netdev; | |
1128 | ||
1129 | close(netpipe); | |
1130 | lxc_list_for_each(iterator, &handler->conf->network) { | |
1131 | netdev = iterator->elem; | |
1132 | if (netdev->type != LXC_NET_VETH) | |
1133 | continue; | |
1134 | if (write(netpipepair[1], netdev->name, IFNAMSIZ) != IFNAMSIZ) { | |
1135 | ERROR("Error writing veth name to container"); | |
1136 | goto out_delete_net; | |
1137 | } | |
1138 | } | |
1139 | close(netpipepair[1]); | |
1140 | } | |
1141 | ||
f6d3e3e4 SH |
1142 | /* map the container uids - the container became an invalid |
1143 | * userid the moment it was cloned with CLONE_NEWUSER - this | |
1144 | * call doesn't change anything immediately, but allows the | |
1145 | * container to setuid(0) (0 being mapped to something else on | |
1146 | * the host) later to become a valid uid again */ | |
1147 | if (lxc_map_ids(&handler->conf->id_map, handler->pid)) { | |
1148 | ERROR("failed to set up id mapping"); | |
1149 | goto out_delete_net; | |
1150 | } | |
1151 | ||
544a48a0 SH |
1152 | /* Tell the child to continue its initialization. we'll get |
1153 | * LXC_SYNC_CGROUP when it is ready for us to setup cgroups | |
3c22086f CLG |
1154 | */ |
1155 | if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CONFIGURE)) | |
544a48a0 SH |
1156 | goto out_delete_net; |
1157 | ||
9daf6f5d | 1158 | if (!cgroup_setup_limits(handler, true)) { |
b98f7d6e SH |
1159 | ERROR("failed to setup the devices cgroup for '%s'", name); |
1160 | goto out_delete_net; | |
544a48a0 SH |
1161 | } |
1162 | ||
73d28d42 | 1163 | cgroup_disconnect(); |
7e4dfe0b | 1164 | cgroups_connected = false; |
73d28d42 | 1165 | |
e8bd4e43 SH |
1166 | /* read tty fds allocated by child */ |
1167 | if (recv_ttys_from_child(handler) < 0) { | |
1168 | ERROR("failed to receive tty info from child"); | |
1169 | goto out_delete_net; | |
1170 | } | |
1171 | ||
544a48a0 SH |
1172 | /* Tell the child to complete its initialization and wait for |
1173 | * it to exec or return an error. (the child will never | |
1174 | * return LXC_SYNC_POST_CGROUP+1. It will either close the | |
1175 | * sync pipe, causing lxc_sync_barrier_child to return | |
1176 | * success, or return a different value, causing us to error | |
1177 | * out). | |
1178 | */ | |
1179 | if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CGROUP)) | |
3c22086f | 1180 | return -1; |
0ad19a3f | 1181 | |
cc28d0b0 SH |
1182 | if (detect_shared_rootfs()) |
1183 | umount2(handler->conf->rootfs.mount, MNT_DETACH); | |
1184 | ||
23c53af9 | 1185 | if (handler->ops->post_start(handler, handler->data)) |
e6126dbe MN |
1186 | goto out_abort; |
1187 | ||
25c2aca5 | 1188 | if (lxc_set_state(name, handler, RUNNING)) { |
59eb99ba DL |
1189 | ERROR("failed to set state to %s", |
1190 | lxc_state2str(RUNNING)); | |
1191 | goto out_abort; | |
3f21c114 | 1192 | } |
22ebac19 | 1193 | |
3c22086f | 1194 | lxc_sync_fini(handler); |
0c547523 | 1195 | |
e6126dbe | 1196 | return 0; |
1ac470c0 | 1197 | |
7fef7a06 | 1198 | out_delete_net: |
7e4dfe0b SH |
1199 | if (cgroups_connected) |
1200 | cgroup_disconnect(); | |
d5088cf2 | 1201 | if (handler->clone_flags & CLONE_NEWNET) |
74a2b586 | 1202 | lxc_delete_network(handler); |
59eb99ba DL |
1203 | out_abort: |
1204 | lxc_abort(name, handler); | |
3c22086f | 1205 | lxc_sync_fini(handler); |
5c068da9 SH |
1206 | if (handler->pinfd >= 0) { |
1207 | close(handler->pinfd); | |
1208 | handler->pinfd = -1; | |
1209 | } | |
1210 | ||
b79fcd86 | 1211 | return -1; |
59eb99ba | 1212 | } |
0ad19a3f | 1213 | |
2af6bd1b SH |
1214 | int get_netns_fd(int pid) |
1215 | { | |
1216 | char path[MAXPATHLEN]; | |
1217 | int ret, fd; | |
1218 | ||
1219 | ret = snprintf(path, MAXPATHLEN, "/proc/%d/ns/net", pid); | |
1220 | if (ret < 0 || ret >= MAXPATHLEN) { | |
1221 | WARN("Failed to pin netns file for pid %d", pid); | |
1222 | return -1; | |
1223 | } | |
1224 | ||
1225 | fd = open(path, O_RDONLY); | |
1226 | if (fd < 0) { | |
1227 | WARN("Failed to pin netns file %s for pid %d: %s", | |
1228 | path, pid, strerror(errno)); | |
1229 | return -1; | |
1230 | } | |
1231 | return fd; | |
1232 | } | |
1233 | ||
ee70bf78 | 1234 | int __lxc_start(const char *name, struct lxc_conf *conf, |
507cee36 TA |
1235 | struct lxc_operations* ops, void *data, const char *lxcpath, |
1236 | bool backgrounded) | |
59eb99ba | 1237 | { |
3a0f472d | 1238 | struct lxc_handler *handler; |
e043236e | 1239 | int err = -1; |
59eb99ba | 1240 | int status; |
2af6bd1b | 1241 | int netnsfd = -1; |
80090207 | 1242 | |
13f5be62 | 1243 | handler = lxc_init(name, conf, lxcpath); |
3a0f472d | 1244 | if (!handler) { |
59eb99ba | 1245 | ERROR("failed to initialize the container"); |
66aeffc7 | 1246 | return -1; |
0ad19a3f | 1247 | } |
ee70bf78 CLG |
1248 | handler->ops = ops; |
1249 | handler->data = data; | |
507cee36 | 1250 | handler->backgrounded = backgrounded; |
e6126dbe | 1251 | |
b60ed720 | 1252 | if (must_drop_cap_sys_boot(handler->conf)) { |
495d2046 | 1253 | #if HAVE_SYS_CAPABILITY_H |
959aee9c | 1254 | DEBUG("Dropping cap_sys_boot"); |
495d2046 | 1255 | #else |
959aee9c | 1256 | DEBUG("Can't drop cap_sys_boot as capabilities aren't supported"); |
495d2046 | 1257 | #endif |
69182a31 | 1258 | } else { |
959aee9c | 1259 | DEBUG("Not dropping cap_sys_boot or watching utmp"); |
69182a31 SH |
1260 | handler->conf->need_utmp_watch = 0; |
1261 | } | |
1262 | ||
76a26f55 SH |
1263 | if (!attach_block_device(handler->conf)) { |
1264 | ERROR("Failure attaching block device"); | |
1265 | goto out_fini_nonet; | |
1266 | } | |
1267 | ||
35120d9c SH |
1268 | if (geteuid() == 0 && !lxc_list_empty(&conf->id_map)) { |
1269 | /* if the backing store is a device, mount it here and now */ | |
1270 | if (rootfs_is_blockdev(conf)) { | |
1271 | if (unshare(CLONE_NEWNS) < 0) { | |
1272 | ERROR("Error unsharing mounts"); | |
1273 | goto out_fini_nonet; | |
1274 | } | |
6a0c909a | 1275 | remount_all_slave(); |
35120d9c SH |
1276 | if (do_rootfs_setup(conf, name, lxcpath) < 0) { |
1277 | ERROR("Error setting up rootfs mount as root before spawn"); | |
1278 | goto out_fini_nonet; | |
1279 | } | |
1280 | INFO("Set up container rootfs as host root"); | |
1281 | } | |
1282 | } | |
1283 | ||
23c53af9 | 1284 | err = lxc_spawn(handler); |
59eb99ba | 1285 | if (err) { |
ee70bf78 | 1286 | ERROR("failed to spawn '%s'", name); |
76a26f55 | 1287 | goto out_detach_blockdev; |
0ad19a3f | 1288 | } |
1289 | ||
8bee8851 WB |
1290 | handler->conf->reboot = 0; |
1291 | ||
2af6bd1b SH |
1292 | netnsfd = get_netns_fd(handler->pid); |
1293 | ||
3a0f472d | 1294 | err = lxc_poll(name, handler); |
e043236e | 1295 | if (err) { |
59eb99ba | 1296 | ERROR("mainloop exited with an error"); |
4b2f98f5 SH |
1297 | if (netnsfd >= 0) |
1298 | close(netnsfd); | |
59eb99ba DL |
1299 | goto out_abort; |
1300 | } | |
0ad19a3f | 1301 | |
3a0f472d | 1302 | while (waitpid(handler->pid, &status, 0) < 0 && errno == EINTR) |
1bc5cc8c | 1303 | continue; |
e043236e | 1304 | |
8b004f07 SH |
1305 | /* |
1306 | * If the child process exited but was not signaled, | |
1307 | * it didn't call reboot. This should mean it was an | |
1308 | * lxc-execute which simply exited. In any case, treat | |
1309 | * it as a 'halt' | |
1310 | */ | |
d028235d | 1311 | if (WIFSIGNALED(status)) { |
8b004f07 SH |
1312 | switch(WTERMSIG(status)) { |
1313 | case SIGINT: /* halt */ | |
1314 | DEBUG("Container halting"); | |
1315 | break; | |
1316 | case SIGHUP: /* reboot */ | |
1317 | DEBUG("Container rebooting"); | |
1318 | handler->conf->reboot = 1; | |
1319 | break; | |
c2b9bd9e SH |
1320 | case SIGSYS: /* seccomp */ |
1321 | DEBUG("Container violated its seccomp policy"); | |
1322 | break; | |
8b004f07 | 1323 | default: |
959aee9c | 1324 | DEBUG("unknown exit status for init: %d", WTERMSIG(status)); |
8b004f07 SH |
1325 | break; |
1326 | } | |
d028235d | 1327 | } |
828695d9 | 1328 | |
ce5782df | 1329 | DEBUG("Pushing physical nics back to host namespace"); |
2af6bd1b | 1330 | lxc_rename_phys_nics_on_shutdown(netnsfd, handler->conf); |
ce5782df MH |
1331 | |
1332 | DEBUG("Tearing down virtual network devices used by container"); | |
1333 | lxc_delete_network(handler); | |
1334 | ||
4b2f98f5 SH |
1335 | if (netnsfd >= 0) |
1336 | close(netnsfd); | |
7b35f3d6 | 1337 | |
5c068da9 SH |
1338 | if (handler->pinfd >= 0) { |
1339 | close(handler->pinfd); | |
1340 | handler->pinfd = -1; | |
1341 | } | |
1342 | ||
1787abca | 1343 | lxc_monitor_send_exit_code(name, status, handler->lxcpath); |
3a0f472d | 1344 | err = lxc_error_set_and_log(handler->pid, status); |
9d7f9e52 | 1345 | out_fini: |
74a2b586 JK |
1346 | lxc_delete_network(handler); |
1347 | ||
76a26f55 SH |
1348 | out_detach_blockdev: |
1349 | detach_block_device(handler->conf); | |
1350 | ||
74a2b586 | 1351 | out_fini_nonet: |
3a0f472d | 1352 | lxc_fini(name, handler); |
0ad19a3f | 1353 | return err; |
1354 | ||
59eb99ba | 1355 | out_abort: |
3a0f472d | 1356 | lxc_abort(name, handler); |
9d7f9e52 | 1357 | goto out_fini; |
0ad19a3f | 1358 | } |
ee70bf78 CLG |
1359 | |
1360 | struct start_args { | |
1361 | char *const *argv; | |
1362 | }; | |
1363 | ||
1364 | static int start(struct lxc_handler *handler, void* data) | |
1365 | { | |
1366 | struct start_args *arg = data; | |
1367 | ||
1368 | NOTICE("exec'ing '%s'", arg->argv[0]); | |
1369 | ||
1370 | execvp(arg->argv[0], arg->argv); | |
1371 | SYSERROR("failed to exec %s", arg->argv[0]); | |
1372 | return 0; | |
1373 | } | |
1374 | ||
1375 | static int post_start(struct lxc_handler *handler, void* data) | |
1376 | { | |
1377 | struct start_args *arg = data; | |
1378 | ||
1379 | NOTICE("'%s' started with pid '%d'", arg->argv[0], handler->pid); | |
1380 | return 0; | |
1381 | } | |
1382 | ||
1383 | static struct lxc_operations start_ops = { | |
1384 | .start = start, | |
1385 | .post_start = post_start | |
1386 | }; | |
1387 | ||
13f5be62 | 1388 | int lxc_start(const char *name, char *const argv[], struct lxc_conf *conf, |
507cee36 | 1389 | const char *lxcpath, bool backgrounded) |
ee70bf78 CLG |
1390 | { |
1391 | struct start_args start_arg = { | |
1392 | .argv = argv, | |
1393 | }; | |
1394 | ||
828695d9 | 1395 | conf->need_utmp_watch = 1; |
507cee36 | 1396 | return __lxc_start(name, conf, &start_ops, &start_arg, lxcpath, backgrounded); |
ee70bf78 | 1397 | } |
28272964 CB |
1398 | |
1399 | static void lxc_destroy_container_on_signal(struct lxc_handler *handler, | |
1400 | const char *name) | |
1401 | { | |
1402 | char destroy[MAXPATHLEN]; | |
1403 | bool bret = true; | |
1404 | int ret = 0; | |
f01f7975 | 1405 | struct lxc_container *c; |
df31363a | 1406 | if (handler->conf->rootfs.path && handler->conf->rootfs.mount) { |
28272964 CB |
1407 | bret = do_destroy_container(handler->conf); |
1408 | if (!bret) { | |
1409 | ERROR("Error destroying rootfs for %s", name); | |
1410 | return; | |
1411 | } | |
1412 | } | |
1413 | INFO("Destroyed rootfs for %s", name); | |
1414 | ||
1415 | ret = snprintf(destroy, MAXPATHLEN, "%s/%s", handler->lxcpath, name); | |
1416 | if (ret < 0 || ret >= MAXPATHLEN) { | |
1417 | ERROR("Error printing path for %s", name); | |
1418 | ERROR("Error destroying directory for %s", name); | |
1419 | return; | |
1420 | } | |
1421 | ||
f01f7975 CB |
1422 | c = lxc_container_new(name, handler->lxcpath); |
1423 | if (c) { | |
1424 | if (container_disk_lock(c)) { | |
1425 | INFO("Could not update lxc_snapshots file"); | |
1426 | lxc_container_put(c); | |
1427 | } else { | |
1428 | mod_all_rdeps(c, false); | |
1429 | container_disk_unlock(c); | |
1430 | lxc_container_put(c); | |
1431 | } | |
1432 | } | |
1433 | ||
28272964 CB |
1434 | if (am_unpriv()) |
1435 | ret = userns_exec_1(handler->conf, lxc_rmdir_onedev_wrapper, destroy); | |
1436 | else | |
1437 | ret = lxc_rmdir_onedev(destroy, NULL); | |
1438 | ||
1439 | if (ret < 0) { | |
1440 | ERROR("Error destroying directory for %s", name); | |
1441 | return; | |
1442 | } | |
1443 | INFO("Destroyed directory for %s", name); | |
1444 | } | |
1445 | ||
1446 | static int lxc_rmdir_onedev_wrapper(void *data) | |
1447 | { | |
1448 | char *arg = (char *) data; | |
1449 | return lxc_rmdir_onedev(arg, NULL); | |
1450 | } | |
1451 | ||
1452 | static bool do_destroy_container(struct lxc_conf *conf) { | |
d028235d SG |
1453 | if (am_unpriv()) { |
1454 | if (userns_exec_1(conf, bdev_destroy_wrapper, conf) < 0) | |
1455 | return false; | |
1456 | return true; | |
1457 | } | |
1458 | return bdev_destroy(conf); | |
28272964 | 1459 | } |