]>
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> |
27 | #undef _GNU_SOURCE | |
28 | #include <string.h> | |
29 | #include <stdlib.h> | |
30 | #include <dirent.h> | |
31 | #include <errno.h> | |
32 | #include <unistd.h> | |
33 | #include <signal.h> | |
b0a33c1e | 34 | #include <fcntl.h> |
35 | #include <termios.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> |
46 | #include <sys/poll.h> | |
8173e600 | 47 | #include <sys/syscall.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 | ||
63376d7d DL |
57 | #include "start.h" |
58 | #include "conf.h" | |
59 | #include "log.h" | |
563f2f2c | 60 | #include "cgroup.h" |
e2bcd7db | 61 | #include "error.h" |
b0a33c1e | 62 | #include "af_unix.h" |
63 | #include "mainloop.h" | |
63376d7d | 64 | #include "utils.h" |
565c2d76 | 65 | #include "lxcutmp.h" |
63376d7d | 66 | #include "monitor.h" |
96fa1ff0 | 67 | #include "commands.h" |
63376d7d | 68 | #include "console.h" |
3c22086f | 69 | #include "sync.h" |
f549edcc | 70 | #include "namespace.h" |
0d0527a9 | 71 | #include "lxcseccomp.h" |
8173e600 | 72 | #include "caps.h" |
025ed0f3 | 73 | #include "lxclock.h" |
fe4de9a6 | 74 | #include "lsm/lsm.h" |
36eb9bde CLG |
75 | |
76 | lxc_log_define(lxc_start, lxc); | |
77 | ||
80090207 CLG |
78 | static int match_fd(int fd) |
79 | { | |
80 | return (fd == 0 || fd == 1 || fd == 2); | |
81 | } | |
82 | ||
b119f362 | 83 | int lxc_check_inherited(struct lxc_conf *conf, int fd_to_ignore) |
80090207 CLG |
84 | { |
85 | struct dirent dirent, *direntp; | |
86 | int fd, fddir; | |
87 | DIR *dir; | |
80090207 | 88 | |
b119f362 | 89 | restart: |
025ed0f3 | 90 | process_lock(); |
80090207 | 91 | dir = opendir("/proc/self/fd"); |
025ed0f3 | 92 | process_unlock(); |
80090207 CLG |
93 | if (!dir) { |
94 | WARN("failed to open directory: %m"); | |
95 | return -1; | |
96 | } | |
97 | ||
98 | fddir = dirfd(dir); | |
99 | ||
100 | while (!readdir_r(dir, &dirent, &direntp)) { | |
80090207 CLG |
101 | if (!direntp) |
102 | break; | |
103 | ||
104 | if (!strcmp(direntp->d_name, ".")) | |
105 | continue; | |
106 | ||
107 | if (!strcmp(direntp->d_name, "..")) | |
108 | continue; | |
109 | ||
110 | fd = atoi(direntp->d_name); | |
111 | ||
f2faa8fa | 112 | if (fd == fddir || fd == lxc_log_fd || fd == fd_to_ignore) |
80090207 CLG |
113 | continue; |
114 | ||
115 | if (match_fd(fd)) | |
116 | continue; | |
80090207 | 117 | |
b119f362 | 118 | if (conf->close_all_fds) { |
025ed0f3 | 119 | process_lock(); |
b119f362 SH |
120 | close(fd); |
121 | closedir(dir); | |
025ed0f3 | 122 | process_unlock(); |
b119f362 SH |
123 | INFO("closed inherited fd %d", fd); |
124 | goto restart; | |
125 | } | |
92c7f629 | 126 | WARN("inherited fd %d", fd); |
80090207 CLG |
127 | } |
128 | ||
025ed0f3 | 129 | process_lock(); |
92c7f629 | 130 | closedir(dir); /* cannot fail */ |
025ed0f3 | 131 | process_unlock(); |
92c7f629 | 132 | return 0; |
80090207 CLG |
133 | } |
134 | ||
83ee7875 | 135 | static int setup_signal_fd(sigset_t *oldmask) |
b0a33c1e | 136 | { |
137 | sigset_t mask; | |
138 | int fd; | |
139 | ||
f3304a29 FW |
140 | /* Block everything except serious error signals */ |
141 | if (sigfillset(&mask) || | |
142 | sigdelset(&mask, SIGILL) || | |
143 | sigdelset(&mask, SIGSEGV) || | |
144 | sigdelset(&mask, SIGBUS) || | |
b5159817 | 145 | sigdelset(&mask, SIGWINCH) || |
f3304a29 FW |
146 | sigprocmask(SIG_BLOCK, &mask, oldmask)) { |
147 | SYSERROR("failed to set signal mask"); | |
b0a33c1e | 148 | return -1; |
149 | } | |
150 | ||
151 | fd = signalfd(-1, &mask, 0); | |
152 | if (fd < 0) { | |
36eb9bde | 153 | SYSERROR("failed to create the signal fd"); |
b0a33c1e | 154 | return -1; |
155 | } | |
156 | ||
157 | if (fcntl(fd, F_SETFD, FD_CLOEXEC)) { | |
36eb9bde | 158 | SYSERROR("failed to set sigfd to close-on-exec"); |
b0a33c1e | 159 | close(fd); |
160 | return -1; | |
161 | } | |
162 | ||
1ac470c0 DL |
163 | DEBUG("sigchild handler set"); |
164 | ||
b0a33c1e | 165 | return fd; |
166 | } | |
167 | ||
83ee7875 | 168 | static int signal_handler(int fd, void *data, |
b0a33c1e | 169 | struct lxc_epoll_descr *descr) |
170 | { | |
15cd25fd | 171 | struct signalfd_siginfo siginfo; |
80507ee8 | 172 | siginfo_t info; |
15cd25fd | 173 | int ret; |
82d89dce | 174 | pid_t *pid = data; |
80507ee8 | 175 | bool init_died = false; |
15cd25fd DL |
176 | |
177 | ret = read(fd, &siginfo, sizeof(siginfo)); | |
178 | if (ret < 0) { | |
f3304a29 | 179 | ERROR("failed to read signal info"); |
15cd25fd DL |
180 | return -1; |
181 | } | |
182 | ||
183 | if (ret != sizeof(siginfo)) { | |
184 | ERROR("unexpected siginfo size"); | |
185 | return -1; | |
186 | } | |
187 | ||
80507ee8 SH |
188 | // check whether init is running |
189 | info.si_pid = 0; | |
190 | ret = waitid(P_PID, *pid, &info, WEXITED | WNOWAIT | WNOHANG); | |
191 | if (ret == 0 && info.si_pid == *pid) { | |
192 | init_died = true; | |
193 | } | |
194 | ||
f3304a29 FW |
195 | if (siginfo.ssi_signo != SIGCHLD) { |
196 | kill(*pid, siginfo.ssi_signo); | |
197 | INFO("forwarded signal %d to pid %d", siginfo.ssi_signo, *pid); | |
80507ee8 | 198 | return init_died ? 1 : 0; |
f3304a29 FW |
199 | } |
200 | ||
15cd25fd DL |
201 | if (siginfo.ssi_code == CLD_STOPPED || |
202 | siginfo.ssi_code == CLD_CONTINUED) { | |
203 | INFO("container init process was stopped/continued"); | |
80507ee8 | 204 | return init_died ? 1 : 0; |
15cd25fd | 205 | } |
1ac470c0 | 206 | |
82d89dce DL |
207 | /* more robustness, protect ourself from a SIGCHLD sent |
208 | * by a process different from the container init | |
209 | */ | |
210 | if (siginfo.ssi_pid != *pid) { | |
211 | WARN("invalid pid for SIGCHLD"); | |
80507ee8 | 212 | return init_died ? 1 : 0; |
82d89dce DL |
213 | } |
214 | ||
15cd25fd | 215 | DEBUG("container init process exited"); |
b0a33c1e | 216 | return 1; |
217 | } | |
218 | ||
25c2aca5 | 219 | int lxc_set_state(const char *name, struct lxc_handler *handler, lxc_state_t state) |
66aeffc7 DL |
220 | { |
221 | handler->state = state; | |
9123e471 | 222 | lxc_monitor_send_state(name, state, handler->lxcpath); |
66aeffc7 DL |
223 | return 0; |
224 | } | |
225 | ||
1bc5cc8c | 226 | int lxc_poll(const char *name, struct lxc_handler *handler) |
b0a33c1e | 227 | { |
ca5f7926 DL |
228 | int sigfd = handler->sigfd; |
229 | int pid = handler->pid; | |
b0a33c1e | 230 | struct lxc_epoll_descr descr; |
231 | ||
a9e61274 | 232 | if (lxc_mainloop_open(&descr)) { |
36eb9bde | 233 | ERROR("failed to create mainloop"); |
50c8bf05 | 234 | goto out_sigfd; |
b0a33c1e | 235 | } |
236 | ||
83ee7875 | 237 | if (lxc_mainloop_add_handler(&descr, sigfd, signal_handler, &pid)) { |
36eb9bde | 238 | ERROR("failed to add handler for the signal"); |
b0a33c1e | 239 | goto out_mainloop_open; |
240 | } | |
241 | ||
63376d7d DL |
242 | if (lxc_console_mainloop_add(&descr, handler)) { |
243 | ERROR("failed to add console handler to mainloop"); | |
244 | goto out_mainloop_open; | |
245 | } | |
246 | ||
ef6e34ee | 247 | if (lxc_cmd_mainloop_add(name, &descr, handler)) { |
563f2f2c | 248 | ERROR("failed to add command handler to mainloop"); |
96fa1ff0 | 249 | goto out_mainloop_open; |
563f2f2c DL |
250 | } |
251 | ||
828695d9 | 252 | if (handler->conf->need_utmp_watch) { |
495d2046 | 253 | #if HAVE_SYS_CAPABILITY_H |
828695d9 SH |
254 | if (lxc_utmp_mainloop_add(&descr, handler)) { |
255 | ERROR("failed to add utmp handler to mainloop"); | |
256 | goto out_mainloop_open; | |
257 | } | |
495d2046 SG |
258 | #else |
259 | DEBUG("not starting utmp handler as cap_sys_boot cannot be dropped without capabilities support\n"); | |
260 | #endif | |
563f2f2c | 261 | } |
b0a33c1e | 262 | |
e51d4895 | 263 | return lxc_mainloop(&descr, -1); |
b0a33c1e | 264 | |
265 | out_mainloop_open: | |
266 | lxc_mainloop_close(&descr); | |
b0a33c1e | 267 | out_sigfd: |
025ed0f3 | 268 | process_lock(); |
b0a33c1e | 269 | close(sigfd); |
025ed0f3 | 270 | process_unlock(); |
c3e13372 | 271 | return -1; |
b0a33c1e | 272 | } |
273 | ||
13f5be62 | 274 | struct lxc_handler *lxc_init(const char *name, struct lxc_conf *conf, const char *lxcpath) |
59eb99ba | 275 | { |
3a0f472d DL |
276 | struct lxc_handler *handler; |
277 | ||
278 | handler = malloc(sizeof(*handler)); | |
279 | if (!handler) | |
280 | return NULL; | |
59eb99ba DL |
281 | |
282 | memset(handler, 0, sizeof(*handler)); | |
283 | ||
fae349da | 284 | handler->conf = conf; |
9123e471 | 285 | handler->lxcpath = lxcpath; |
5c068da9 | 286 | handler->pinfd = -1; |
fae349da | 287 | |
fe4de9a6 DE |
288 | lsm_init(); |
289 | ||
3bdf52d7 DL |
290 | handler->name = strdup(name); |
291 | if (!handler->name) { | |
292 | ERROR("failed to allocate memory"); | |
293 | goto out_free; | |
294 | } | |
295 | ||
ef6e34ee | 296 | if (lxc_cmd_init(name, handler, lxcpath)) |
d2e30e99 DE |
297 | goto out_free_name; |
298 | ||
8f2c3a70 SH |
299 | if (lxc_read_seccomp_config(conf) != 0) { |
300 | ERROR("failed loading seccomp policy"); | |
d2e30e99 | 301 | goto out_close_maincmd_fd; |
8f2c3a70 SH |
302 | } |
303 | ||
051151de | 304 | /* Begin by setting the state to STARTING */ |
25c2aca5 | 305 | if (lxc_set_state(name, handler, STARTING)) { |
59eb99ba | 306 | ERROR("failed to set state '%s'", lxc_state2str(STARTING)); |
051151de | 307 | goto out_close_maincmd_fd; |
0ad19a3f | 308 | } |
309 | ||
f7bee6c6 MW |
310 | /* Start of environment variable setup for hooks */ |
311 | if (setenv("LXC_NAME", name, 1)) { | |
312 | SYSERROR("failed to set environment variable for container name"); | |
313 | } | |
314 | if (setenv("LXC_CONFIG_FILE", conf->rcfile, 1)) { | |
315 | SYSERROR("failed to set environment variable for config path"); | |
316 | } | |
317 | if (setenv("LXC_ROOTFS_MOUNT", conf->rootfs.mount, 1)) { | |
318 | SYSERROR("failed to set environment variable for rootfs mount"); | |
319 | } | |
320 | if (setenv("LXC_ROOTFS_PATH", conf->rootfs.path, 1)) { | |
321 | SYSERROR("failed to set environment variable for rootfs mount"); | |
322 | } | |
323 | if (conf->console.path && setenv("LXC_CONSOLE", conf->console.path, 1)) { | |
324 | SYSERROR("failed to set environment variable for console path"); | |
325 | } | |
326 | if (conf->console.log_path && setenv("LXC_CONSOLE_LOGPATH", conf->console.log_path, 1)) { | |
327 | SYSERROR("failed to set environment variable for console log"); | |
328 | } | |
329 | /* End of environment variable setup for hooks */ | |
330 | ||
283678ed | 331 | if (run_lxc_hooks(name, "pre-start", conf, handler->lxcpath, NULL)) { |
773fb9ca SH |
332 | ERROR("failed to run pre-start hooks for container '%s'.", name); |
333 | goto out_aborting; | |
334 | } | |
26ddeedd | 335 | |
fae349da | 336 | if (lxc_create_tty(name, conf)) { |
36eb9bde | 337 | ERROR("failed to create the ttys"); |
59eb99ba | 338 | goto out_aborting; |
b0a33c1e | 339 | } |
340 | ||
341 | /* the signal fd has to be created before forking otherwise | |
342 | * if the child process exits before we setup the signal fd, | |
343 | * the event will be lost and the command will be stuck */ | |
83ee7875 | 344 | handler->sigfd = setup_signal_fd(&handler->oldmask); |
59eb99ba | 345 | if (handler->sigfd < 0) { |
36eb9bde | 346 | ERROR("failed to set sigchild fd handler"); |
b5159817 DE |
347 | goto out_delete_tty; |
348 | } | |
349 | ||
350 | /* do this after setting up signals since it might unblock SIGWINCH */ | |
351 | if (lxc_console_create(conf)) { | |
352 | ERROR("failed to create console"); | |
353 | goto out_restore_sigmask; | |
b0a33c1e | 354 | } |
355 | ||
c3e13372 | 356 | INFO("'%s' is initialized", name); |
3a0f472d | 357 | return handler; |
59eb99ba | 358 | |
b5159817 DE |
359 | out_restore_sigmask: |
360 | sigprocmask(SIG_SETMASK, &handler->oldmask, NULL); | |
59eb99ba | 361 | out_delete_tty: |
fae349da | 362 | lxc_delete_tty(&conf->tty_info); |
59eb99ba | 363 | out_aborting: |
25c2aca5 | 364 | lxc_set_state(name, handler, ABORTING); |
d2e30e99 | 365 | out_close_maincmd_fd: |
025ed0f3 | 366 | process_lock(); |
d2e30e99 | 367 | close(conf->maincmd_fd); |
025ed0f3 | 368 | process_unlock(); |
d2e30e99 | 369 | conf->maincmd_fd = -1; |
3bdf52d7 DL |
370 | out_free_name: |
371 | free(handler->name); | |
372 | handler->name = NULL; | |
3a0f472d DL |
373 | out_free: |
374 | free(handler); | |
c3e13372 | 375 | return NULL; |
59eb99ba DL |
376 | } |
377 | ||
ae5c8b8e | 378 | static void lxc_fini(const char *name, struct lxc_handler *handler) |
59eb99ba DL |
379 | { |
380 | /* The STOPPING state is there for future cleanup code | |
381 | * which can take awhile | |
382 | */ | |
25c2aca5 MN |
383 | lxc_set_state(name, handler, STOPPING); |
384 | lxc_set_state(name, handler, STOPPED); | |
59eb99ba | 385 | |
283678ed | 386 | if (run_lxc_hooks(name, "post-stop", handler->conf, handler->lxcpath, NULL)) |
773fb9ca | 387 | ERROR("failed to run post-stop hooks for container '%s'.", name); |
26ddeedd | 388 | |
83ee7875 | 389 | /* reset mask set by setup_signal_fd */ |
8f64a3f6 MN |
390 | if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL)) |
391 | WARN("failed to restore sigprocmask"); | |
392 | ||
b5159817 | 393 | lxc_console_delete(&handler->conf->console); |
b2431939 | 394 | lxc_delete_tty(&handler->conf->tty_info); |
025ed0f3 | 395 | process_lock(); |
d2e30e99 | 396 | close(handler->conf->maincmd_fd); |
025ed0f3 | 397 | process_unlock(); |
d2e30e99 | 398 | handler->conf->maincmd_fd = -1; |
3bdf52d7 | 399 | free(handler->name); |
ae5c8b8e | 400 | if (handler->cgroup) { |
33ad9f1a | 401 | lxc_cgroup_process_info_free_and_remove(handler->cgroup); |
ae5c8b8e SH |
402 | handler->cgroup = NULL; |
403 | } | |
b2431939 | 404 | free(handler); |
59eb99ba DL |
405 | } |
406 | ||
1bc5cc8c | 407 | void lxc_abort(const char *name, struct lxc_handler *handler) |
59eb99ba | 408 | { |
73e608b2 SH |
409 | int ret, status; |
410 | ||
25c2aca5 | 411 | lxc_set_state(name, handler, ABORTING); |
7d9fb3e9 DL |
412 | if (handler->pid > 0) |
413 | kill(handler->pid, SIGKILL); | |
73e608b2 | 414 | while ((ret = waitpid(-1, &status, 0)) > 0) ; |
59eb99ba DL |
415 | } |
416 | ||
828695d9 SH |
417 | #include <sys/reboot.h> |
418 | #include <linux/reboot.h> | |
419 | ||
e2fa1520 SH |
420 | /* |
421 | * reboot(LINUX_REBOOT_CMD_CAD_ON) will return -EINVAL | |
422 | * in a child pid namespace if container reboot support exists. | |
423 | * Otherwise, it will either succeed or return -EPERM. | |
424 | */ | |
425 | static int container_reboot_supported(void *arg) | |
828695d9 | 426 | { |
e2fa1520 | 427 | int *cmd = arg; |
828695d9 | 428 | int ret; |
828695d9 | 429 | |
e2fa1520 SH |
430 | ret = reboot(*cmd); |
431 | if (ret == -1 && errno == EINVAL) | |
432 | return 1; | |
433 | return 0; | |
434 | } | |
435 | ||
b60ed720 | 436 | static int must_drop_cap_sys_boot(struct lxc_conf *conf) |
e2fa1520 | 437 | { |
025ed0f3 | 438 | FILE *f; |
b60ed720 | 439 | int ret, cmd, v, flags; |
e2fa1520 | 440 | long stack_size = 4096; |
7f145a6d | 441 | void *stack = alloca(stack_size); |
e2fa1520 SH |
442 | int status; |
443 | pid_t pid; | |
444 | ||
025ed0f3 SH |
445 | process_lock(); |
446 | f = fopen("/proc/sys/kernel/ctrl-alt-del", "r"); | |
447 | process_unlock(); | |
e2fa1520 SH |
448 | if (!f) { |
449 | DEBUG("failed to open /proc/sys/kernel/ctrl-alt-del"); | |
828695d9 | 450 | return 1; |
e2fa1520 | 451 | } |
828695d9 SH |
452 | |
453 | ret = fscanf(f, "%d", &v); | |
025ed0f3 | 454 | process_lock(); |
828695d9 | 455 | fclose(f); |
025ed0f3 | 456 | process_unlock(); |
e2fa1520 SH |
457 | if (ret != 1) { |
458 | DEBUG("Failed to read /proc/sys/kernel/ctrl-alt-del"); | |
828695d9 | 459 | return 1; |
e2fa1520 SH |
460 | } |
461 | cmd = v ? LINUX_REBOOT_CMD_CAD_ON : LINUX_REBOOT_CMD_CAD_OFF; | |
462 | ||
b60ed720 SH |
463 | flags = CLONE_NEWPID | SIGCHLD; |
464 | if (!lxc_list_empty(&conf->id_map)) | |
465 | flags |= CLONE_NEWUSER; | |
466 | ||
7f145a6d | 467 | #ifdef __ia64__ |
b60ed720 | 468 | pid = __clone2(container_reboot_supported, stack, stack_size, flags, &cmd); |
7f145a6d DS |
469 | #else |
470 | stack += stack_size; | |
b60ed720 | 471 | pid = clone(container_reboot_supported, stack, flags, &cmd); |
7f145a6d | 472 | #endif |
e2fa1520 SH |
473 | if (pid < 0) { |
474 | SYSERROR("failed to clone\n"); | |
475 | return -1; | |
476 | } | |
477 | if (wait(&status) < 0) { | |
478 | SYSERROR("unexpected wait error: %m\n"); | |
479 | return -1; | |
480 | } | |
481 | ||
482 | if (WEXITSTATUS(status) != 1) | |
828695d9 | 483 | return 1; |
e2fa1520 | 484 | |
828695d9 SH |
485 | return 0; |
486 | } | |
487 | ||
ffe1e01a | 488 | static int do_start(void *data) |
50e98013 | 489 | { |
23c53af9 | 490 | struct lxc_handler *handler = data; |
50e98013 DL |
491 | |
492 | if (sigprocmask(SIG_SETMASK, &handler->oldmask, NULL)) { | |
493 | SYSERROR("failed to set sigprocmask"); | |
9d7f9e52 | 494 | return -1; |
50e98013 DL |
495 | } |
496 | ||
743ecd2e DL |
497 | /* This prctl must be before the synchro, so if the parent |
498 | * dies before we set the parent death signal, we will detect | |
499 | * its death with the synchro right after, otherwise we have | |
500 | * a window where the parent can exit before we set the pdeath | |
501 | * signal leading to a unsupervized container. | |
502 | */ | |
503 | if (prctl(PR_SET_PDEATHSIG, SIGKILL, 0, 0, 0)) { | |
504 | SYSERROR("failed to set pdeath signal"); | |
505 | return -1; | |
506 | } | |
507 | ||
3c22086f | 508 | lxc_sync_fini_parent(handler); |
50e98013 | 509 | |
2b0e17e4 | 510 | /* don't leak the pinfd to the container */ |
025ed0f3 SH |
511 | if (handler->pinfd >= 0) { |
512 | process_lock(); | |
0d03360a | 513 | close(handler->pinfd); |
025ed0f3 SH |
514 | process_unlock(); |
515 | } | |
2b0e17e4 | 516 | |
3c22086f CLG |
517 | /* Tell the parent task it can begin to configure the |
518 | * container and wait for it to finish | |
519 | */ | |
520 | if (lxc_sync_barrier_parent(handler, LXC_SYNC_CONFIGURE)) | |
9d7f9e52 | 521 | return -1; |
50e98013 | 522 | |
f6d3e3e4 SH |
523 | /* |
524 | * if we are in a new user namespace, become root there to have | |
525 | * privilege over our namespace | |
526 | */ | |
527 | if (!lxc_list_empty(&handler->conf->id_map)) { | |
528 | NOTICE("switching to gid/uid 0 in new user namespace"); | |
529 | if (setgid(0)) { | |
530 | SYSERROR("setgid"); | |
531 | goto out_warn_father; | |
532 | } | |
533 | if (setuid(0)) { | |
534 | SYSERROR("setuid"); | |
535 | goto out_warn_father; | |
536 | } | |
537 | } | |
538 | ||
495d2046 | 539 | #if HAVE_SYS_CAPABILITY_H |
69182a31 | 540 | if (handler->conf->need_utmp_watch) { |
828695d9 SH |
541 | if (prctl(PR_CAPBSET_DROP, CAP_SYS_BOOT, 0, 0, 0)) { |
542 | SYSERROR("failed to remove CAP_SYS_BOOT capability"); | |
c4ea60df | 543 | goto out_warn_father; |
828695d9 | 544 | } |
e2fa1520 | 545 | DEBUG("Dropped cap_sys_boot\n"); |
e2fa1520 | 546 | } |
495d2046 | 547 | #endif |
e2fa1520 SH |
548 | |
549 | /* Setup the container, ip, names, utsname, ... */ | |
368bbc02 | 550 | if (lxc_setup(handler->name, handler->conf, handler->lxcpath, handler->cgroup)) { |
e2fa1520 SH |
551 | ERROR("failed to setup the container"); |
552 | goto out_warn_father; | |
553 | } | |
50e98013 | 554 | |
544a48a0 SH |
555 | /* ask father to setup cgroups and wait for him to finish */ |
556 | if (lxc_sync_barrier_parent(handler, LXC_SYNC_CGROUP)) | |
557 | return -1; | |
558 | ||
fe4de9a6 DE |
559 | /* XXX: hmm apparmor switches right away since it uses |
560 | * aa_change_profile() and not aa_change_onexec(). SELinux on the other | |
561 | * hand is going to transition on exec(). Is it bad to run the stuff | |
562 | * between here and exec() in the more privileged context? | |
563 | */ | |
564 | if (lsm_process_label_set(handler->conf->lsm_aa_profile ? | |
565 | handler->conf->lsm_aa_profile : | |
566 | handler->conf->lsm_se_context, 1) < 0) | |
e075f5d9 | 567 | goto out_warn_father; |
fe4de9a6 | 568 | lsm_proc_unmount(handler->conf); |
e075f5d9 | 569 | |
8f2c3a70 SH |
570 | if (lxc_seccomp_load(handler->conf) != 0) |
571 | goto out_warn_father; | |
572 | ||
283678ed | 573 | if (run_lxc_hooks(handler->name, "start", handler->conf, handler->lxcpath, NULL)) { |
773fb9ca SH |
574 | ERROR("failed to run start hooks for container '%s'.", handler->name); |
575 | goto out_warn_father; | |
576 | } | |
fc25b815 | 577 | |
f7bee6c6 MW |
578 | /* The clearenv() and putenv() calls have been moved here |
579 | * to allow us to use enviroment variables passed to the various | |
580 | * hooks, such as the start hook above. Not all of the | |
581 | * variables like CONFIG_PATH or ROOTFS are valid in this | |
582 | * context but others are. */ | |
583 | if (clearenv()) { | |
584 | SYSERROR("failed to clear environment"); | |
585 | /* don't error out though */ | |
586 | } | |
587 | ||
588 | if (putenv("container=lxc")) { | |
589 | SYSERROR("failed to set environment variable"); | |
c4ea60df | 590 | goto out_warn_father; |
f7bee6c6 MW |
591 | } |
592 | ||
025ed0f3 | 593 | process_lock(); |
773fb9ca | 594 | close(handler->sigfd); |
025ed0f3 | 595 | process_unlock(); |
26ddeedd | 596 | |
e6126dbe MN |
597 | /* after this call, we are in error because this |
598 | * ops should not return as it execs */ | |
c4ea60df | 599 | handler->ops->start(handler, handler->data); |
50e98013 DL |
600 | |
601 | out_warn_father: | |
544a48a0 SH |
602 | /* we want the parent to know something went wrong, so any |
603 | * value other than what it expects is ok. */ | |
3c22086f | 604 | lxc_sync_wake_parent(handler, LXC_SYNC_POST_CONFIGURE); |
50e98013 DL |
605 | return -1; |
606 | } | |
607 | ||
7b35f3d6 SH |
608 | int save_phys_nics(struct lxc_conf *conf) |
609 | { | |
610 | struct lxc_list *iterator; | |
611 | ||
612 | lxc_list_for_each(iterator, &conf->network) { | |
613 | struct lxc_netdev *netdev = iterator->elem; | |
614 | ||
615 | if (netdev->type != LXC_NET_PHYS) | |
616 | continue; | |
617 | conf->saved_nics = realloc(conf->saved_nics, | |
618 | (conf->num_savednics+1)*sizeof(struct saved_nic)); | |
619 | if (!conf->saved_nics) { | |
620 | SYSERROR("failed to allocate memory"); | |
621 | return -1; | |
622 | } | |
623 | conf->saved_nics[conf->num_savednics].ifindex = netdev->ifindex; | |
624 | conf->saved_nics[conf->num_savednics].orig_name = strdup(netdev->link); | |
625 | if (!conf->saved_nics[conf->num_savednics].orig_name) { | |
626 | SYSERROR("failed to allocate memory"); | |
627 | return -1; | |
628 | } | |
629 | INFO("stored saved_nic #%d idx %d name %s\n", conf->num_savednics, | |
630 | conf->saved_nics[conf->num_savednics].ifindex, | |
631 | conf->saved_nics[conf->num_savednics].orig_name); | |
632 | conf->num_savednics++; | |
633 | } | |
634 | ||
635 | return 0; | |
636 | } | |
637 | ||
23c53af9 | 638 | int lxc_spawn(struct lxc_handler *handler) |
59eb99ba | 639 | { |
b98f7d6e | 640 | int failed_before_rename = 0; |
ffe1e01a | 641 | const char *name = handler->name; |
33ad9f1a CS |
642 | struct cgroup_meta_data *cgroup_meta = NULL; |
643 | const char *cgroup_pattern = NULL; | |
50e98013 | 644 | |
3c22086f | 645 | if (lxc_sync_init(handler)) |
9d7f9e52 | 646 | return -1; |
0ad19a3f | 647 | |
d5088cf2 | 648 | handler->clone_flags = CLONE_NEWUTS|CLONE_NEWPID|CLONE_NEWIPC|CLONE_NEWNS; |
f6d3e3e4 SH |
649 | if (!lxc_list_empty(&handler->conf->id_map)) { |
650 | INFO("Cloning a new user namespace"); | |
651 | handler->clone_flags |= CLONE_NEWUSER; | |
652 | } | |
fae349da | 653 | if (!lxc_list_empty(&handler->conf->network)) { |
82d5ae15 | 654 | |
d5088cf2 | 655 | handler->clone_flags |= CLONE_NEWNET; |
0ad19a3f | 656 | |
19a26f82 MK |
657 | /* Find gateway addresses from the link device, which is |
658 | * no longer accessible inside the container. Do this | |
659 | * before creating network interfaces, since goto | |
660 | * out_delete_net does not work before lxc_clone. */ | |
661 | if (lxc_find_gateway_addresses(handler)) { | |
662 | ERROR("failed to find gateway addresses"); | |
663 | lxc_sync_fini(handler); | |
664 | return -1; | |
665 | } | |
666 | ||
82d5ae15 DL |
667 | /* that should be done before the clone because we will |
668 | * fill the netdev index and use them in the child | |
669 | */ | |
e3b4c4c4 | 670 | if (lxc_create_network(handler)) { |
82d5ae15 | 671 | ERROR("failed to create the network"); |
3c22086f | 672 | lxc_sync_fini(handler); |
32e1c760 | 673 | return -1; |
82d5ae15 DL |
674 | } |
675 | } | |
676 | ||
7b35f3d6 SH |
677 | if (save_phys_nics(handler->conf)) { |
678 | ERROR("failed to save physical nic info"); | |
679 | goto out_abort; | |
680 | } | |
681 | ||
33ad9f1a CS |
682 | cgroup_meta = lxc_cgroup_load_meta(); |
683 | if (!cgroup_meta) { | |
684 | ERROR("failed to detect cgroup metadata"); | |
685 | goto out_delete_net; | |
686 | } | |
687 | ||
688 | /* if we are running as root, use system cgroup pattern, otherwise | |
689 | * just create a cgroup under the current one. But also fall back to | |
690 | * that if for some reason reading the configuration fails and no | |
691 | * default value is available | |
692 | */ | |
693 | if (getuid() == 0) | |
694 | cgroup_pattern = lxc_global_config_value("cgroup.pattern"); | |
695 | if (!cgroup_pattern) | |
696 | cgroup_pattern = "%n"; | |
697 | ||
47d8fb3b CS |
698 | /* Create cgroup before doing clone(), so the child will know from |
699 | * handler which cgroup it is going to be put in later. | |
700 | */ | |
701 | if ((handler->cgroup = lxc_cgroup_create(name, cgroup_pattern, cgroup_meta, NULL)) == NULL) { | |
702 | ERROR("failed to create cgroups for '%s'", name); | |
703 | goto out_delete_net; | |
704 | } | |
705 | ||
0c547523 SH |
706 | /* |
707 | * if the rootfs is not a blockdev, prevent the container from | |
708 | * marking it readonly. | |
709 | */ | |
710 | ||
2b0e17e4 | 711 | handler->pinfd = pin_rootfs(handler->conf->rootfs.path); |
00ec333b SH |
712 | if (handler->pinfd == -1) |
713 | INFO("failed to pin the container's rootfs"); | |
0c547523 | 714 | |
0ad19a3f | 715 | /* Create a process in a new set of namespaces */ |
d5088cf2 | 716 | handler->pid = lxc_clone(do_start, handler, handler->clone_flags); |
59eb99ba | 717 | if (handler->pid < 0) { |
36eb9bde | 718 | SYSERROR("failed to fork into a new namespace"); |
7fef7a06 | 719 | goto out_delete_net; |
0ad19a3f | 720 | } |
721 | ||
3c22086f CLG |
722 | lxc_sync_fini_child(handler); |
723 | ||
724 | if (lxc_sync_wait_child(handler, LXC_SYNC_CONFIGURE)) | |
99a6af52 | 725 | failed_before_rename = 1; |
0ad19a3f | 726 | |
47d8fb3b CS |
727 | /* In case there is still legacy ns cgroup support in the kernel. |
728 | * Should be removed at some later point in time. | |
729 | */ | |
730 | if (lxc_cgroup_create_legacy(handler->cgroup, name, handler->pid) < 0) { | |
731 | ERROR("failed to create legacy ns cgroups for '%s'", name); | |
ae5c8b8e | 732 | goto out_delete_net; |
33ad9f1a | 733 | } |
e51d4895 | 734 | |
33ad9f1a | 735 | if (lxc_setup_cgroup_without_devices(handler, &handler->conf->cgroup)) { |
6031a6e5 DE |
736 | ERROR("failed to setup the cgroups for '%s'", name); |
737 | goto out_delete_net; | |
738 | } | |
739 | ||
33ad9f1a | 740 | if (lxc_cgroup_enter(handler->cgroup, handler->pid, false) < 0) |
7fef7a06 | 741 | goto out_delete_net; |
218d4250 | 742 | |
99a6af52 MN |
743 | if (failed_before_rename) |
744 | goto out_delete_net; | |
745 | ||
0ad19a3f | 746 | /* Create the network configuration */ |
d5088cf2 | 747 | if (handler->clone_flags & CLONE_NEWNET) { |
fae349da | 748 | if (lxc_assign_network(&handler->conf->network, handler->pid)) { |
82d5ae15 | 749 | ERROR("failed to create the configured network"); |
7fef7a06 | 750 | goto out_delete_net; |
82d5ae15 | 751 | } |
0ad19a3f | 752 | } |
753 | ||
f6d3e3e4 SH |
754 | /* map the container uids - the container became an invalid |
755 | * userid the moment it was cloned with CLONE_NEWUSER - this | |
756 | * call doesn't change anything immediately, but allows the | |
757 | * container to setuid(0) (0 being mapped to something else on | |
758 | * the host) later to become a valid uid again */ | |
759 | if (lxc_map_ids(&handler->conf->id_map, handler->pid)) { | |
760 | ERROR("failed to set up id mapping"); | |
761 | goto out_delete_net; | |
762 | } | |
763 | ||
544a48a0 SH |
764 | /* Tell the child to continue its initialization. we'll get |
765 | * LXC_SYNC_CGROUP when it is ready for us to setup cgroups | |
3c22086f CLG |
766 | */ |
767 | if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CONFIGURE)) | |
544a48a0 SH |
768 | goto out_delete_net; |
769 | ||
33ad9f1a | 770 | if (lxc_setup_cgroup_devices(handler, &handler->conf->cgroup)) { |
b98f7d6e SH |
771 | ERROR("failed to setup the devices cgroup for '%s'", name); |
772 | goto out_delete_net; | |
544a48a0 SH |
773 | } |
774 | ||
544a48a0 SH |
775 | /* Tell the child to complete its initialization and wait for |
776 | * it to exec or return an error. (the child will never | |
777 | * return LXC_SYNC_POST_CGROUP+1. It will either close the | |
778 | * sync pipe, causing lxc_sync_barrier_child to return | |
779 | * success, or return a different value, causing us to error | |
780 | * out). | |
781 | */ | |
782 | if (lxc_sync_barrier_child(handler, LXC_SYNC_POST_CGROUP)) | |
3c22086f | 783 | return -1; |
0ad19a3f | 784 | |
cc28d0b0 SH |
785 | if (detect_shared_rootfs()) |
786 | umount2(handler->conf->rootfs.mount, MNT_DETACH); | |
787 | ||
f6d3e3e4 SH |
788 | /* If child is in a fresh user namespace, chown his ptys for |
789 | * it */ | |
790 | if (uid_shift_ttys(handler->pid, handler->conf)) | |
791 | DEBUG("Failed to chown ptys.\n"); | |
792 | ||
23c53af9 | 793 | if (handler->ops->post_start(handler, handler->data)) |
e6126dbe MN |
794 | goto out_abort; |
795 | ||
25c2aca5 | 796 | if (lxc_set_state(name, handler, RUNNING)) { |
59eb99ba DL |
797 | ERROR("failed to set state to %s", |
798 | lxc_state2str(RUNNING)); | |
799 | goto out_abort; | |
3f21c114 | 800 | } |
22ebac19 | 801 | |
33ad9f1a | 802 | lxc_cgroup_put_meta(cgroup_meta); |
3c22086f | 803 | lxc_sync_fini(handler); |
0c547523 | 804 | |
e6126dbe | 805 | return 0; |
1ac470c0 | 806 | |
7fef7a06 | 807 | out_delete_net: |
d5088cf2 | 808 | if (handler->clone_flags & CLONE_NEWNET) |
74a2b586 | 809 | lxc_delete_network(handler); |
59eb99ba | 810 | out_abort: |
33ad9f1a | 811 | lxc_cgroup_put_meta(cgroup_meta); |
59eb99ba | 812 | lxc_abort(name, handler); |
3c22086f | 813 | lxc_sync_fini(handler); |
5c068da9 | 814 | if (handler->pinfd >= 0) { |
025ed0f3 | 815 | process_lock(); |
5c068da9 | 816 | close(handler->pinfd); |
025ed0f3 | 817 | process_unlock(); |
5c068da9 SH |
818 | handler->pinfd = -1; |
819 | } | |
820 | ||
b79fcd86 | 821 | return -1; |
59eb99ba | 822 | } |
0ad19a3f | 823 | |
ee70bf78 | 824 | int __lxc_start(const char *name, struct lxc_conf *conf, |
13f5be62 | 825 | struct lxc_operations* ops, void *data, const char *lxcpath) |
59eb99ba | 826 | { |
3a0f472d | 827 | struct lxc_handler *handler; |
e043236e | 828 | int err = -1; |
59eb99ba | 829 | int status; |
80090207 | 830 | |
13f5be62 | 831 | handler = lxc_init(name, conf, lxcpath); |
3a0f472d | 832 | if (!handler) { |
59eb99ba | 833 | ERROR("failed to initialize the container"); |
66aeffc7 | 834 | return -1; |
0ad19a3f | 835 | } |
ee70bf78 CLG |
836 | handler->ops = ops; |
837 | handler->data = data; | |
e6126dbe | 838 | |
b60ed720 | 839 | if (must_drop_cap_sys_boot(handler->conf)) { |
495d2046 | 840 | #if HAVE_SYS_CAPABILITY_H |
f51db2b3 | 841 | DEBUG("Dropping cap_sys_boot\n"); |
495d2046 SG |
842 | #else |
843 | DEBUG("Can't drop cap_sys_boot as capabilities aren't supported\n"); | |
844 | #endif | |
69182a31 SH |
845 | } else { |
846 | DEBUG("Not dropping cap_sys_boot or watching utmp\n"); | |
847 | handler->conf->need_utmp_watch = 0; | |
848 | } | |
849 | ||
23c53af9 | 850 | err = lxc_spawn(handler); |
59eb99ba | 851 | if (err) { |
ee70bf78 | 852 | ERROR("failed to spawn '%s'", name); |
74a2b586 | 853 | goto out_fini_nonet; |
0ad19a3f | 854 | } |
855 | ||
3a0f472d | 856 | err = lxc_poll(name, handler); |
e043236e | 857 | if (err) { |
59eb99ba DL |
858 | ERROR("mainloop exited with an error"); |
859 | goto out_abort; | |
860 | } | |
0ad19a3f | 861 | |
3a0f472d | 862 | while (waitpid(handler->pid, &status, 0) < 0 && errno == EINTR) |
1bc5cc8c | 863 | continue; |
e043236e | 864 | |
8b004f07 SH |
865 | /* |
866 | * If the child process exited but was not signaled, | |
867 | * it didn't call reboot. This should mean it was an | |
868 | * lxc-execute which simply exited. In any case, treat | |
869 | * it as a 'halt' | |
870 | */ | |
871 | if (WIFSIGNALED(status)) { | |
872 | switch(WTERMSIG(status)) { | |
873 | case SIGINT: /* halt */ | |
874 | DEBUG("Container halting"); | |
875 | break; | |
876 | case SIGHUP: /* reboot */ | |
877 | DEBUG("Container rebooting"); | |
878 | handler->conf->reboot = 1; | |
879 | break; | |
880 | default: | |
881 | DEBUG("unknown exit status for init: %d\n", WTERMSIG(status)); | |
882 | break; | |
883 | } | |
828695d9 SH |
884 | } |
885 | ||
7b35f3d6 SH |
886 | lxc_rename_phys_nics_on_shutdown(handler->conf); |
887 | ||
5c068da9 | 888 | if (handler->pinfd >= 0) { |
025ed0f3 | 889 | process_lock(); |
5c068da9 | 890 | close(handler->pinfd); |
025ed0f3 | 891 | process_unlock(); |
5c068da9 SH |
892 | handler->pinfd = -1; |
893 | } | |
894 | ||
3a0f472d | 895 | err = lxc_error_set_and_log(handler->pid, status); |
9d7f9e52 | 896 | out_fini: |
74a2b586 JK |
897 | lxc_delete_network(handler); |
898 | ||
899 | out_fini_nonet: | |
3a0f472d | 900 | lxc_fini(name, handler); |
0ad19a3f | 901 | return err; |
902 | ||
59eb99ba | 903 | out_abort: |
3a0f472d | 904 | lxc_abort(name, handler); |
9d7f9e52 | 905 | goto out_fini; |
0ad19a3f | 906 | } |
ee70bf78 CLG |
907 | |
908 | struct start_args { | |
909 | char *const *argv; | |
910 | }; | |
911 | ||
912 | static int start(struct lxc_handler *handler, void* data) | |
913 | { | |
914 | struct start_args *arg = data; | |
915 | ||
916 | NOTICE("exec'ing '%s'", arg->argv[0]); | |
917 | ||
918 | execvp(arg->argv[0], arg->argv); | |
919 | SYSERROR("failed to exec %s", arg->argv[0]); | |
920 | return 0; | |
921 | } | |
922 | ||
923 | static int post_start(struct lxc_handler *handler, void* data) | |
924 | { | |
925 | struct start_args *arg = data; | |
926 | ||
927 | NOTICE("'%s' started with pid '%d'", arg->argv[0], handler->pid); | |
928 | return 0; | |
929 | } | |
930 | ||
931 | static struct lxc_operations start_ops = { | |
932 | .start = start, | |
933 | .post_start = post_start | |
934 | }; | |
935 | ||
13f5be62 SH |
936 | int lxc_start(const char *name, char *const argv[], struct lxc_conf *conf, |
937 | const char *lxcpath) | |
ee70bf78 CLG |
938 | { |
939 | struct start_args start_arg = { | |
940 | .argv = argv, | |
941 | }; | |
942 | ||
b119f362 | 943 | if (lxc_check_inherited(conf, -1)) |
ee70bf78 CLG |
944 | return -1; |
945 | ||
828695d9 | 946 | conf->need_utmp_watch = 1; |
13f5be62 | 947 | return __lxc_start(name, conf, &start_ops, &start_arg, lxcpath); |
ee70bf78 | 948 | } |