]>
Commit | Line | Data |
---|---|---|
cc73685d | 1 | /* SPDX-License-Identifier: LGPL-2.1+ */ |
e0732705 | 2 | |
d38dd64a CB |
3 | #ifndef _GNU_SOURCE |
4 | #define _GNU_SOURCE 1 | |
5 | #endif | |
e0732705 CS |
6 | #include <errno.h> |
7 | #include <fcntl.h> | |
c476bdce | 8 | #include <grp.h> |
604ca1c0 | 9 | #include <linux/unistd.h> |
6f4f1937 | 10 | #include <pwd.h> |
0bece477 | 11 | #include <pthread.h> |
6f4f1937 CB |
12 | #include <signal.h> |
13 | #include <stdio.h> | |
14 | #include <stdlib.h> | |
15 | #include <string.h> | |
6f4f1937 | 16 | #include <sys/mount.h> |
e0732705 CS |
17 | #include <sys/param.h> |
18 | #include <sys/prctl.h> | |
5ec27989 | 19 | #include <sys/socket.h> |
1ba0013f | 20 | #include <sys/syscall.h> |
905022f7 | 21 | #include <sys/wait.h> |
604ca1c0 CB |
22 | #include <termios.h> |
23 | #include <unistd.h> | |
6f4f1937 CB |
24 | |
25 | #include <lxc/lxccontainer.h> | |
e0732705 | 26 | |
81f466d0 | 27 | #include "af_unix.h" |
e0732705 CS |
28 | #include "attach.h" |
29 | #include "caps.h" | |
9c4693b8 | 30 | #include "cgroup.h" |
6f4f1937 | 31 | #include "commands.h" |
2c4ea790 | 32 | #include "conf.h" |
6f4f1937 | 33 | #include "config.h" |
9b8e3c96 | 34 | #include "confile.h" |
6f4f1937 CB |
35 | #include "log.h" |
36 | #include "lsm/lsm.h" | |
37 | #include "lxclock.h" | |
38 | #include "lxcseccomp.h" | |
604ca1c0 | 39 | #include "macro.h" |
ba2be1a8 | 40 | #include "mainloop.h" |
cd8f5663 | 41 | #include "memory_utils.h" |
657256e0 | 42 | #include "mount_utils.h" |
6f4f1937 | 43 | #include "namespace.h" |
f40988c7 | 44 | #include "process_utils.h" |
59524108 | 45 | #include "syscall_wrappers.h" |
0ed9b1bc | 46 | #include "terminal.h" |
6f4f1937 | 47 | #include "utils.h" |
9c4693b8 CS |
48 | |
49 | #if HAVE_SYS_PERSONALITY_H | |
50 | #include <sys/personality.h> | |
51 | #endif | |
e0732705 | 52 | |
ac2cecc4 | 53 | lxc_log_define(attach, lxc); |
e0732705 | 54 | |
ef05d368 CB |
55 | /* Define default options if no options are supplied by the user. */ |
56 | static lxc_attach_options_t attach_static_default_options = LXC_ATTACH_OPTIONS_DEFAULT; | |
57 | ||
ab919e5f | 58 | struct attach_context { |
500ed813 | 59 | int init_pid; |
0e304baa CB |
60 | char *lsm_label; |
61 | struct lxc_container *container; | |
62 | signed long personality; | |
63 | unsigned long long capability_mask; | |
64 | int ns_inherited; | |
65 | int ns_fd[LXC_NS_MAX]; | |
66 | struct lsm_ops *lsm_ops; | |
67 | }; | |
68 | ||
9745eb8a CB |
69 | static struct attach_context *alloc_attach_context(void) |
70 | { | |
71 | return zalloc(sizeof(struct attach_context)); | |
72 | } | |
73 | ||
ee142207 CB |
74 | static int get_personality(const char *name, const char *lxcpath, |
75 | signed long *personality) | |
d92c8e40 CB |
76 | { |
77 | __do_free char *p = NULL; | |
ee142207 | 78 | signed long per; |
d92c8e40 CB |
79 | |
80 | p = lxc_cmd_get_config_item(name, "lxc.arch", lxcpath); | |
ee142207 CB |
81 | if (!p) { |
82 | *personality = LXC_ARCH_UNCHANGED; | |
83 | return 0; | |
84 | } | |
d92c8e40 | 85 | |
ee142207 CB |
86 | per = lxc_config_parse_arch(p); |
87 | if (per == LXC_ARCH_UNCHANGED) | |
88 | return ret_errno(EINVAL); | |
89 | ||
90 | *personality = per; | |
91 | return 0; | |
d92c8e40 CB |
92 | } |
93 | ||
500ed813 CB |
94 | static int get_attach_context(struct attach_context *ctx, |
95 | struct lxc_container *container) | |
e0732705 | 96 | { |
cd8f5663 CB |
97 | __do_free char *line = NULL; |
98 | __do_fclose FILE *proc_file = NULL; | |
6f4f1937 CB |
99 | int ret; |
100 | bool found; | |
604ca1c0 | 101 | char proc_fn[LXC_PROC_STATUS_LEN]; |
e0732705 | 102 | size_t line_bufsz = 0; |
e0732705 | 103 | |
500ed813 CB |
104 | ctx->container = container; |
105 | ||
106 | ctx->init_pid = lxc_cmd_get_init_pid(container->name, container->config_path); | |
107 | if (ctx->init_pid < 0) | |
108 | return log_error(-1, "Failed to get init pid"); | |
109 | ||
8ce83369 | 110 | /* Read capabilities. */ |
500ed813 | 111 | ret = snprintf(proc_fn, LXC_PROC_STATUS_LEN, "/proc/%d/status", ctx->init_pid); |
604ca1c0 | 112 | if (ret < 0 || ret >= LXC_PROC_STATUS_LEN) |
9745eb8a | 113 | return -EIO; |
e0732705 | 114 | |
ffeeed8b CB |
115 | proc_file = fopen(proc_fn, "re"); |
116 | if (!proc_file) | |
9745eb8a | 117 | return log_error_errno(-errno, errno, "Failed to open %s", proc_fn); |
8ce83369 CB |
118 | |
119 | found = false; | |
ea918412 | 120 | |
e0732705 | 121 | while (getline(&line, &line_bufsz, proc_file) != -1) { |
9745eb8a | 122 | ret = sscanf(line, "CapBnd: %llx", &ctx->capability_mask); |
8ce83369 CB |
123 | if (ret != EOF && ret == 1) { |
124 | found = true; | |
e0732705 CS |
125 | break; |
126 | } | |
127 | } | |
128 | ||
ffeeed8b | 129 | if (!found) |
9745eb8a | 130 | return log_error_errno(-ENOENT, ENOENT, "Failed to read capability bounding set from %s", proc_fn); |
e0732705 | 131 | |
4eb19ac0 | 132 | ctx->lsm_ops = lsm_init_static(); |
d701d729 | 133 | |
500ed813 | 134 | ctx->lsm_label = ctx->lsm_ops->process_label_get(ctx->lsm_ops, ctx->init_pid); |
9745eb8a | 135 | ctx->ns_inherited = 0; |
b87ee312 | 136 | for (int i = 0; i < LXC_NS_MAX; i++) |
9745eb8a | 137 | ctx->ns_fd[i] = -EBADF; |
e0732705 | 138 | |
ee142207 CB |
139 | ret = get_personality(container->name, container->config_path, &ctx->personality); |
140 | if (ret) | |
141 | return log_error_errno(ret, errno, "Failed to get personality of the container"); | |
d92c8e40 | 142 | |
1874ef74 CB |
143 | if (!ctx->container->lxc_conf) { |
144 | ctx->container->lxc_conf = lxc_conf_init(); | |
145 | if (!ctx->container->lxc_conf) | |
146 | return log_error_errno(-ENOMEM, ENOMEM, "Failed to allocate new lxc config"); | |
147 | } | |
148 | ||
9745eb8a | 149 | return 0; |
e0732705 CS |
150 | } |
151 | ||
299d1198 CB |
152 | /** |
153 | * in_same_namespace - Check whether two processes are in the same namespace. | |
154 | * @pid1 - PID of the first process. | |
155 | * @pid2 - PID of the second process. | |
156 | * @ns - Name of the namespace to check. Must correspond to one of the names | |
157 | * for the namespaces as shown in /proc/<pid/ns/ | |
158 | * | |
159 | * If the two processes are not in the same namespace returns an fd to the | |
160 | * namespace of the second process identified by @pid2. If the two processes are | |
161 | * in the same namespace returns -EINVAL, -1 if an error occurred. | |
162 | */ | |
163 | static int in_same_namespace(pid_t pid1, pid_t pid2, const char *ns) | |
164 | { | |
644e7393 | 165 | __do_close int ns_fd1 = -EBADF, ns_fd2 = -EBADF; |
3cc629fe | 166 | int ret = -1; |
299d1198 CB |
167 | struct stat ns_st1, ns_st2; |
168 | ||
169 | ns_fd1 = lxc_preserve_ns(pid1, ns); | |
134284c3 CB |
170 | if (ns_fd1 < 0) { |
171 | /* The kernel does not support this namespace. This is not an | |
172 | * error. | |
173 | */ | |
174 | if (errno == ENOENT) | |
175 | return -EINVAL; | |
176 | ||
3cc629fe | 177 | return -1; |
134284c3 | 178 | } |
299d1198 CB |
179 | |
180 | ns_fd2 = lxc_preserve_ns(pid2, ns); | |
21d0acc2 | 181 | if (ns_fd2 < 0) |
3cc629fe | 182 | return -1; |
299d1198 CB |
183 | |
184 | ret = fstat(ns_fd1, &ns_st1); | |
21d0acc2 | 185 | if (ret < 0) |
3cc629fe | 186 | return -1; |
299d1198 CB |
187 | |
188 | ret = fstat(ns_fd2, &ns_st2); | |
21d0acc2 | 189 | if (ret < 0) |
3cc629fe | 190 | return -1; |
299d1198 CB |
191 | |
192 | /* processes are in the same namespace */ | |
3cc629fe CB |
193 | if ((ns_st1.st_dev == ns_st2.st_dev) && (ns_st1.st_ino == ns_st2.st_ino)) |
194 | return -EINVAL; | |
299d1198 CB |
195 | |
196 | /* processes are in different namespaces */ | |
3cc629fe | 197 | return move_fd(ns_fd2); |
299d1198 CB |
198 | } |
199 | ||
b7873c95 CB |
200 | static int get_attach_context_nsfds(struct attach_context *ctx, |
201 | lxc_attach_options_t *options) | |
202 | { | |
203 | ||
204 | pid_t pid_self = lxc_raw_getpid(); | |
205 | ||
206 | for (int i = 0; i < LXC_NS_MAX; i++) { | |
207 | int j; | |
208 | ||
209 | if (options->namespaces & ns_info[i].clone_flag) | |
210 | ctx->ns_fd[i] = lxc_preserve_ns(ctx->init_pid, ns_info[i].proc_name); | |
211 | else if (ctx->ns_inherited & ns_info[i].clone_flag) | |
212 | ctx->ns_fd[i] = in_same_namespace(pid_self, ctx->init_pid, ns_info[i].proc_name); | |
213 | else | |
214 | continue; | |
215 | ||
216 | if (ctx->ns_fd[i] >= 0) | |
217 | continue; | |
218 | ||
219 | if (ctx->ns_fd[i] == -EINVAL) { | |
220 | DEBUG("Inheriting %s namespace from %d", ns_info[i].proc_name, pid_self); | |
221 | ctx->ns_inherited &= ~ns_info[i].clone_flag; | |
222 | continue; | |
223 | } | |
224 | ||
225 | /* We failed to preserve the namespace. */ | |
226 | SYSERROR("Failed to attach to %s namespace of %d", ns_info[i].proc_name, pid_self); | |
227 | ||
228 | /* Close all already opened file descriptors before we return an | |
229 | * error, so we don't leak them. | |
230 | */ | |
231 | for (j = 0; j < i; j++) | |
232 | close_prot_errno_disarm(ctx->ns_fd[j]); | |
233 | ||
234 | return -1; | |
235 | } | |
236 | ||
237 | return 0; | |
238 | } | |
239 | ||
7e995801 | 240 | static inline void close_nsfds(struct attach_context *ctx) |
b7873c95 CB |
241 | { |
242 | for (int i = 0; i < LXC_NS_MAX; i++) | |
243 | close_prot_errno_disarm(ctx->ns_fd[i]); | |
244 | } | |
245 | ||
246 | static void put_attach_context(struct attach_context *ctx) | |
247 | { | |
248 | free_disarm(ctx->lsm_label); | |
249 | ||
250 | if (ctx->container) { | |
251 | lxc_container_put(ctx->container); | |
252 | ctx->container = NULL; | |
253 | } | |
254 | ||
7e995801 | 255 | close_nsfds(ctx); |
b7873c95 CB |
256 | free(ctx); |
257 | } | |
258 | ||
ab919e5f | 259 | static int lxc_attach_to_ns(pid_t pid, struct attach_context *ctx) |
99d50954 | 260 | { |
ffeeed8b CB |
261 | for (int i = 0; i < LXC_NS_MAX; i++) { |
262 | int ret; | |
99d50954 | 263 | |
877f3a04 | 264 | if (ctx->ns_fd[i] < 0) |
26818618 CB |
265 | continue; |
266 | ||
21d0acc2 | 267 | ret = setns(ctx->ns_fd[i], ns_info[i].clone_flag); |
ffeeed8b CB |
268 | if (ret < 0) |
269 | return log_error_errno(-1, | |
270 | errno, "Failed to attach to %s namespace of %d", | |
271 | ns_info[i].proc_name, pid); | |
99d50954 | 272 | |
299d1198 | 273 | DEBUG("Attached to %s namespace of %d", ns_info[i].proc_name, pid); |
99d50954 CS |
274 | } |
275 | ||
276 | return 0; | |
277 | } | |
278 | ||
e4103cf6 | 279 | int lxc_attach_remount_sys_proc(void) |
7a0b0b56 CS |
280 | { |
281 | int ret; | |
282 | ||
283 | ret = unshare(CLONE_NEWNS); | |
ffeeed8b CB |
284 | if (ret < 0) |
285 | return log_error_errno(-1, errno, "Failed to unshare mount namespace"); | |
7a0b0b56 | 286 | |
9e61fb1f CB |
287 | if (detect_shared_rootfs() && mount(NULL, "/", NULL, MS_SLAVE | MS_REC, NULL)) |
288 | SYSERROR("Failed to recursively turn root mount tree into dependent mount. Continuing..."); | |
2c6f3fc9 | 289 | |
8ce83369 | 290 | /* Assume /proc is always mounted, so remount it. */ |
7a0b0b56 | 291 | ret = umount2("/proc", MNT_DETACH); |
ffeeed8b CB |
292 | if (ret < 0) |
293 | return log_error_errno(-1, errno, "Failed to unmount /proc"); | |
7a0b0b56 | 294 | |
657256e0 | 295 | ret = mount_filesystem("proc", "/proc", 0); |
ffeeed8b CB |
296 | if (ret < 0) |
297 | return log_error_errno(-1, errno, "Failed to remount /proc"); | |
7a0b0b56 | 298 | |
ffeeed8b CB |
299 | /* |
300 | * Try to umount /sys. If it's not a mount point, we'll get EINVAL, then | |
8ce83369 | 301 | * we ignore it because it may not have been mounted in the first place. |
7a0b0b56 CS |
302 | */ |
303 | ret = umount2("/sys", MNT_DETACH); | |
ffeeed8b CB |
304 | if (ret < 0 && errno != EINVAL) |
305 | return log_error_errno(-1, errno, "Failed to unmount /sys"); | |
306 | ||
307 | /* Remount it. */ | |
657256e0 | 308 | if (ret == 0 && mount_filesystem("sysfs", "/sys", 0)) |
ffeeed8b | 309 | return log_error_errno(-1, errno, "Failed to remount /sys"); |
7a0b0b56 CS |
310 | |
311 | return 0; | |
312 | } | |
313 | ||
677e1d27 | 314 | static int drop_capabilities(struct attach_context *ctx) |
e0732705 | 315 | { |
ffeeed8b | 316 | int last_cap; |
e0732705 | 317 | |
6f4f1937 | 318 | last_cap = lxc_caps_last_cap(); |
ffeeed8b | 319 | for (int cap = 0; cap <= last_cap; cap++) { |
e0732705 CS |
320 | if (ctx->capability_mask & (1LL << cap)) |
321 | continue; | |
322 | ||
b81689a1 | 323 | if (prctl(PR_CAPBSET_DROP, prctl_arg(cap), prctl_arg(0), |
ffeeed8b CB |
324 | prctl_arg(0), prctl_arg(0))) |
325 | return log_error_errno(-1, errno, "Failed to drop capability %d", cap); | |
ea918412 | 326 | |
94ac256f | 327 | TRACE("Dropped capability %d", cap); |
e0732705 CS |
328 | } |
329 | ||
330 | return 0; | |
331 | } | |
905022f7 | 332 | |
ab919e5f | 333 | static int lxc_attach_set_environment(struct attach_context *ctx, |
7385273f | 334 | enum lxc_attach_env_policy_t policy, |
6f4f1937 | 335 | char **extra_env, char **extra_keep) |
b3a39ba6 | 336 | { |
3d55242a | 337 | int ret; |
7385273f | 338 | struct lxc_list *iterator; |
339 | ||
799f96fd | 340 | if (policy == LXC_ATTACH_CLEAR_ENV) { |
3d5e9f48 | 341 | int path_kept = 0; |
6f4f1937 | 342 | char **extra_keep_store = NULL; |
3d5e9f48 CS |
343 | |
344 | if (extra_keep) { | |
345 | size_t count, i; | |
346 | ||
3d55242a CB |
347 | for (count = 0; extra_keep[count]; count++) |
348 | ; | |
3d5e9f48 | 349 | |
89b7bfe3 | 350 | extra_keep_store = zalloc(count * sizeof(char *)); |
3d55242a | 351 | if (!extra_keep_store) |
3d5e9f48 | 352 | return -1; |
3d55242a | 353 | |
3d5e9f48 CS |
354 | for (i = 0; i < count; i++) { |
355 | char *v = getenv(extra_keep[i]); | |
356 | if (v) { | |
357 | extra_keep_store[i] = strdup(v); | |
358 | if (!extra_keep_store[i]) { | |
3d5e9f48 CS |
359 | while (i > 0) |
360 | free(extra_keep_store[--i]); | |
ea918412 | 361 | |
3d5e9f48 CS |
362 | free(extra_keep_store); |
363 | return -1; | |
364 | } | |
3d55242a | 365 | |
3d5e9f48 CS |
366 | if (strcmp(extra_keep[i], "PATH") == 0) |
367 | path_kept = 1; | |
368 | } | |
3d5e9f48 CS |
369 | } |
370 | } | |
371 | ||
799f96fd | 372 | if (clearenv()) { |
a9cab7e3 | 373 | if (extra_keep_store) { |
3d55242a CB |
374 | char **p; |
375 | ||
a9cab7e3 CS |
376 | for (p = extra_keep_store; *p; p++) |
377 | free(*p); | |
3d55242a | 378 | |
a9cab7e3 CS |
379 | free(extra_keep_store); |
380 | } | |
3d55242a | 381 | |
ffeeed8b | 382 | return log_error(-1, "Failed to clear environment"); |
3d5e9f48 CS |
383 | } |
384 | ||
385 | if (extra_keep_store) { | |
386 | size_t i; | |
6f4f1937 | 387 | |
3d5e9f48 | 388 | for (i = 0; extra_keep[i]; i++) { |
acd4922e | 389 | if (extra_keep_store[i]) { |
3d55242a CB |
390 | ret = setenv(extra_keep[i], extra_keep_store[i], 1); |
391 | if (ret < 0) | |
a24c5678 | 392 | SYSWARN("Failed to set environment variable"); |
acd4922e | 393 | } |
ea918412 | 394 | |
3d5e9f48 CS |
395 | free(extra_keep_store[i]); |
396 | } | |
ea918412 | 397 | |
3d5e9f48 CS |
398 | free(extra_keep_store); |
399 | } | |
400 | ||
8ce83369 CB |
401 | /* Always set a default path; shells and execlp tend to be fine |
402 | * without it, but there is a disturbing number of C programs | |
403 | * out there that just assume that getenv("PATH") is never NULL | |
404 | * and then die a painful segfault death. | |
405 | */ | |
3d55242a CB |
406 | if (!path_kept) { |
407 | ret = setenv("PATH", "/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin", 1); | |
408 | if (ret < 0) | |
a24c5678 | 409 | SYSWARN("Failed to set environment variable"); |
3d55242a | 410 | } |
b3a39ba6 DW |
411 | } |
412 | ||
3d55242a | 413 | ret = putenv("container=lxc"); |
ffeeed8b | 414 | if (ret < 0) |
818a57fc | 415 | return log_warn(-1, "Failed to set environment variable"); |
b3a39ba6 | 416 | |
7385273f | 417 | /* Set container environment variables.*/ |
ab919e5f CB |
418 | if (ctx && ctx->container && ctx->container->lxc_conf) { |
419 | lxc_list_for_each(iterator, &ctx->container->lxc_conf->environment) { | |
3d55242a CB |
420 | char *env_tmp; |
421 | ||
422 | env_tmp = strdup((char *)iterator->elem); | |
423 | if (!env_tmp) | |
7385273f | 424 | return -1; |
7385273f | 425 | |
3d55242a | 426 | ret = putenv(env_tmp); |
ffeeed8b CB |
427 | if (ret < 0) |
428 | return log_error_errno(-1, errno, "Failed to set environment variable: %s", (char *)iterator->elem); | |
7385273f | 429 | } |
430 | } | |
431 | ||
8ce83369 | 432 | /* Set extra environment variables. */ |
3d5e9f48 CS |
433 | if (extra_env) { |
434 | for (; *extra_env; extra_env++) { | |
3d55242a | 435 | char *p; |
ea918412 | 436 | |
8ce83369 CB |
437 | /* We just assume the user knows what they are doing, so |
438 | * we don't do any checks. | |
439 | */ | |
3d55242a CB |
440 | p = strdup(*extra_env); |
441 | if (!p) | |
3d5e9f48 | 442 | return -1; |
3d55242a CB |
443 | |
444 | ret = putenv(p); | |
445 | if (ret < 0) | |
a24c5678 | 446 | SYSWARN("Failed to set environment variable"); |
3d5e9f48 CS |
447 | } |
448 | } | |
449 | ||
b3a39ba6 DW |
450 | return 0; |
451 | } | |
452 | ||
74a3920a | 453 | static char *lxc_attach_getpwshell(uid_t uid) |
905022f7 | 454 | { |
1b9c9f5b | 455 | __do_free char *line = NULL, *result = NULL; |
cd8f5663 | 456 | __do_fclose FILE *pipe_f = NULL; |
6f4f1937 | 457 | int fd, ret; |
905022f7 CS |
458 | pid_t pid; |
459 | int pipes[2]; | |
3fa23ac3 CB |
460 | bool found = false; |
461 | size_t line_bufsz = 0; | |
905022f7 | 462 | |
8ce83369 CB |
463 | /* We need to fork off a process that runs the getent program, and we |
464 | * need to capture its output, so we use a pipe for that purpose. | |
905022f7 | 465 | */ |
3fa23ac3 | 466 | ret = pipe2(pipes, O_CLOEXEC); |
905022f7 CS |
467 | if (ret < 0) |
468 | return NULL; | |
469 | ||
470 | pid = fork(); | |
471 | if (pid < 0) { | |
472 | close(pipes[0]); | |
473 | close(pipes[1]); | |
474 | return NULL; | |
475 | } | |
476 | ||
3fa23ac3 | 477 | if (!pid) { |
905022f7 CS |
478 | char uid_buf[32]; |
479 | char *arguments[] = { | |
480 | "getent", | |
481 | "passwd", | |
482 | uid_buf, | |
483 | NULL | |
484 | }; | |
485 | ||
486 | close(pipes[0]); | |
487 | ||
8ce83369 | 488 | /* We want to capture stdout. */ |
3fa23ac3 | 489 | ret = dup2(pipes[1], STDOUT_FILENO); |
905022f7 | 490 | close(pipes[1]); |
3fa23ac3 | 491 | if (ret < 0) |
ea918412 | 492 | _exit(EXIT_FAILURE); |
905022f7 | 493 | |
8ce83369 CB |
494 | /* Get rid of stdin/stderr, so we try to associate it with |
495 | * /dev/null. | |
905022f7 | 496 | */ |
3fa23ac3 | 497 | fd = open_devnull(); |
905022f7 | 498 | if (fd < 0) { |
3fa23ac3 CB |
499 | close(STDIN_FILENO); |
500 | close(STDERR_FILENO); | |
905022f7 | 501 | } else { |
3fa23ac3 | 502 | (void)dup3(fd, STDIN_FILENO, O_CLOEXEC); |
59f0e209 | 503 | (void)dup3(fd, STDERR_FILENO, O_CLOEXEC); |
905022f7 CS |
504 | close(fd); |
505 | } | |
506 | ||
8ce83369 | 507 | /* Finish argument list. */ |
3fa23ac3 CB |
508 | ret = snprintf(uid_buf, sizeof(uid_buf), "%ld", (long)uid); |
509 | if (ret <= 0 || ret >= sizeof(uid_buf)) | |
ea918412 | 510 | _exit(EXIT_FAILURE); |
905022f7 | 511 | |
8ce83369 | 512 | /* Try to run getent program. */ |
3fa23ac3 | 513 | (void)execvp("getent", arguments); |
ea918412 | 514 | _exit(EXIT_FAILURE); |
905022f7 | 515 | } |
3fa23ac3 CB |
516 | |
517 | close(pipes[1]); | |
518 | ||
4110345b | 519 | pipe_f = fdopen(pipes[0], "re"); |
cf4026f1 CB |
520 | if (!pipe_f) { |
521 | close(pipes[0]); | |
522 | goto reap_child; | |
523 | } | |
524 | /* Transfer ownership of pipes[0] to pipe_f. */ | |
525 | move_fd(pipes[0]); | |
526 | ||
3fa23ac3 CB |
527 | while (getline(&line, &line_bufsz, pipe_f) != -1) { |
528 | int i; | |
529 | long value; | |
530 | char *token; | |
531 | char *endptr = NULL, *saveptr = NULL; | |
532 | ||
533 | /* If we already found something, just continue to read | |
534 | * until the pipe doesn't deliver any more data, but | |
535 | * don't modify the existing data structure. | |
536 | */ | |
537 | if (found) | |
538 | continue; | |
539 | ||
18d4ffde | 540 | if (!line) |
541 | continue; | |
542 | ||
3fa23ac3 CB |
543 | /* Trim line on the right hand side. */ |
544 | for (i = strlen(line); i > 0 && (line[i - 1] == '\n' || line[i - 1] == '\r'); --i) | |
545 | line[i - 1] = '\0'; | |
546 | ||
547 | /* Split into tokens: first: user name. */ | |
548 | token = strtok_r(line, ":", &saveptr); | |
549 | if (!token) | |
550 | continue; | |
551 | ||
552 | /* next: dummy password field */ | |
553 | token = strtok_r(NULL, ":", &saveptr); | |
554 | if (!token) | |
555 | continue; | |
556 | ||
557 | /* next: user id */ | |
558 | token = strtok_r(NULL, ":", &saveptr); | |
559 | value = token ? strtol(token, &endptr, 10) : 0; | |
560 | if (!token || !endptr || *endptr || value == LONG_MIN || | |
ea918412 | 561 | value == LONG_MAX) |
3fa23ac3 CB |
562 | continue; |
563 | ||
564 | /* dummy sanity check: user id matches */ | |
565 | if ((uid_t)value != uid) | |
566 | continue; | |
567 | ||
568 | /* skip fields: gid, gecos, dir, go to next field 'shell' */ | |
569 | for (i = 0; i < 4; i++) { | |
570 | token = strtok_r(NULL, ":", &saveptr); | |
571 | if (!token) | |
572 | continue; | |
573 | } | |
ea918412 | 574 | |
3fa23ac3 CB |
575 | if (!token) |
576 | continue; | |
ea918412 | 577 | |
1b9c9f5b | 578 | free_disarm(result); |
3fa23ac3 CB |
579 | result = strdup(token); |
580 | ||
581 | /* Sanity check that there are no fields after that. */ | |
582 | token = strtok_r(NULL, ":", &saveptr); | |
583 | if (token) | |
584 | continue; | |
585 | ||
586 | found = true; | |
587 | } | |
ea918412 | 588 | |
cf4026f1 | 589 | reap_child: |
3fa23ac3 | 590 | ret = wait_for_pid(pid); |
1b9c9f5b | 591 | if (ret < 0) |
3fa23ac3 | 592 | return NULL; |
3fa23ac3 | 593 | |
1b9c9f5b | 594 | if (!found) |
3fa23ac3 | 595 | return NULL; |
3fa23ac3 | 596 | |
1b9c9f5b | 597 | return move_ptr(result); |
905022f7 | 598 | } |
cb3e61fa | 599 | |
6f4f1937 | 600 | static void lxc_attach_get_init_uidgid(uid_t *init_uid, gid_t *init_gid) |
cb3e61fa | 601 | { |
cd8f5663 CB |
602 | __do_free char *line = NULL; |
603 | __do_fclose FILE *proc_file = NULL; | |
604ca1c0 | 604 | char proc_fn[LXC_PROC_STATUS_LEN]; |
8ce83369 | 605 | int ret; |
cb3e61fa | 606 | size_t line_bufsz = 0; |
cb3e61fa | 607 | long value = -1; |
a5d657d3 CB |
608 | uid_t uid = LXC_INVALID_UID; |
609 | gid_t gid = LXC_INVALID_GID; | |
cb3e61fa | 610 | |
604ca1c0 CB |
611 | ret = snprintf(proc_fn, LXC_PROC_STATUS_LEN, "/proc/%d/status", 1); |
612 | if (ret < 0 || ret >= LXC_PROC_STATUS_LEN) | |
7fb45c93 | 613 | return; |
cb3e61fa | 614 | |
4110345b | 615 | proc_file = fopen(proc_fn, "re"); |
cb3e61fa CS |
616 | if (!proc_file) |
617 | return; | |
618 | ||
619 | while (getline(&line, &line_bufsz, proc_file) != -1) { | |
8ce83369 CB |
620 | /* Format is: real, effective, saved set user, fs we only care |
621 | * about real uid. | |
cb3e61fa CS |
622 | */ |
623 | ret = sscanf(line, "Uid: %ld", &value); | |
8ce83369 | 624 | if (ret != EOF && ret == 1) { |
6f4f1937 | 625 | uid = (uid_t)value; |
cb3e61fa CS |
626 | } else { |
627 | ret = sscanf(line, "Gid: %ld", &value); | |
8ce83369 | 628 | if (ret != EOF && ret == 1) |
6f4f1937 | 629 | gid = (gid_t)value; |
cb3e61fa | 630 | } |
ea918412 | 631 | |
a5d657d3 | 632 | if (uid != LXC_INVALID_UID && gid != LXC_INVALID_GID) |
cb3e61fa CS |
633 | break; |
634 | } | |
635 | ||
8ce83369 | 636 | /* Only override arguments if we found something. */ |
a5d657d3 | 637 | if (uid != LXC_INVALID_UID) |
cb3e61fa | 638 | *init_uid = uid; |
ea918412 | 639 | |
a5d657d3 | 640 | if (gid != LXC_INVALID_GID) |
cb3e61fa CS |
641 | *init_gid = gid; |
642 | ||
643 | /* TODO: we should also parse supplementary groups and use | |
8ce83369 CB |
644 | * setgroups() to set them. |
645 | */ | |
cb3e61fa | 646 | } |
9c4693b8 | 647 | |
d4db3d14 | 648 | static bool fetch_seccomp(struct lxc_container *c, lxc_attach_options_t *options) |
2c4ea790 | 649 | { |
cd8f5663 | 650 | __do_free char *path = NULL; |
d4db3d14 CB |
651 | int ret; |
652 | bool bret; | |
2eef2bda | 653 | |
6f4f1937 CB |
654 | if (!(options->namespaces & CLONE_NEWNS) || |
655 | !(options->attach_flags & LXC_ATTACH_LSM)) { | |
cca66e06 | 656 | free_disarm(c->lxc_conf->seccomp.seccomp); |
2c4ea790 | 657 | return true; |
bd4307f0 | 658 | } |
bd7b4e28 | 659 | |
2e812c16 | 660 | /* Remove current setting. */ |
d4db3d14 | 661 | if (!c->set_config_item(c, "lxc.seccomp.profile", "") && |
ea918412 | 662 | !c->set_config_item(c, "lxc.seccomp", "")) |
2c4ea790 | 663 | return false; |
bd7b4e28 | 664 | |
8ce83369 | 665 | /* Fetch the current profile path over the cmd interface. */ |
0b427da0 | 666 | path = c->get_running_config_item(c, "lxc.seccomp.profile"); |
bd7b4e28 | 667 | if (!path) { |
d4db3d14 | 668 | INFO("Failed to retrieve lxc.seccomp.profile"); |
ea918412 | 669 | |
0b427da0 | 670 | path = c->get_running_config_item(c, "lxc.seccomp"); |
cca66e06 CB |
671 | if (!path) |
672 | return log_info(true, "Failed to retrieve lxc.seccomp"); | |
bd7b4e28 SG |
673 | } |
674 | ||
8ce83369 | 675 | /* Copy the value into the new lxc_conf. */ |
d4db3d14 | 676 | bret = c->set_config_item(c, "lxc.seccomp.profile", path); |
d4db3d14 CB |
677 | if (!bret) |
678 | return false; | |
bd7b4e28 | 679 | |
8ce83369 | 680 | /* Attempt to parse the resulting config. */ |
d4db3d14 | 681 | ret = lxc_read_seccomp_config(c->lxc_conf); |
cca66e06 CB |
682 | if (ret < 0) |
683 | return log_error(false, "Failed to retrieve seccomp policy"); | |
2c4ea790 | 684 | |
cca66e06 | 685 | return log_info(true, "Retrieved seccomp policy"); |
2e812c16 CB |
686 | } |
687 | ||
6f4f1937 | 688 | static bool no_new_privs(struct lxc_container *c, lxc_attach_options_t *options) |
2e812c16 | 689 | { |
cd8f5663 | 690 | __do_free char *val = NULL; |
2e812c16 | 691 | |
2e812c16 | 692 | /* Remove current setting. */ |
02d3b72b CB |
693 | if (!c->set_config_item(c, "lxc.no_new_privs", "")) |
694 | return log_info(false, "Failed to unset lxc.no_new_privs"); | |
2e812c16 CB |
695 | |
696 | /* Retrieve currently active setting. */ | |
697 | val = c->get_running_config_item(c, "lxc.no_new_privs"); | |
02d3b72b CB |
698 | if (!val) |
699 | return log_info(false, "Failed to retrieve lxc.no_new_privs"); | |
2e812c16 CB |
700 | |
701 | /* Set currently active setting. */ | |
cd8f5663 | 702 | return c->set_config_item(c, "lxc.no_new_privs", val); |
2c4ea790 SH |
703 | } |
704 | ||
a998454a CB |
705 | struct attach_clone_payload { |
706 | int ipc_socket; | |
cecf3e83 | 707 | int terminal_pts_fd; |
a998454a | 708 | lxc_attach_options_t *options; |
ab919e5f | 709 | struct attach_context *ctx; |
a998454a CB |
710 | lxc_attach_exec_t exec_function; |
711 | void *exec_payload; | |
712 | }; | |
713 | ||
ba2be1a8 CB |
714 | static void lxc_put_attach_clone_payload(struct attach_clone_payload *p) |
715 | { | |
81102768 | 716 | close_prot_errno_disarm(p->ipc_socket); |
cecf3e83 | 717 | close_prot_errno_disarm(p->terminal_pts_fd); |
ab919e5f | 718 | if (p->ctx) { |
dd53c8af | 719 | put_attach_context(p->ctx); |
ab919e5f | 720 | p->ctx = NULL; |
b21da190 | 721 | } |
ba2be1a8 CB |
722 | } |
723 | ||
dab02267 | 724 | __noreturn static void do_attach(struct attach_clone_payload *payload) |
a998454a | 725 | { |
427a8067 | 726 | int lsm_fd, ret; |
a998454a CB |
727 | uid_t new_uid; |
728 | gid_t new_gid; | |
936efc72 CB |
729 | uid_t ns_root_uid = 0; |
730 | gid_t ns_root_gid = 0; | |
a998454a | 731 | lxc_attach_options_t* options = payload->options; |
ab919e5f CB |
732 | struct attach_context *ctx = payload->ctx; |
733 | struct lxc_conf *conf = ctx->container->lxc_conf; | |
57de839f CB |
734 | bool needs_lsm = (options->namespaces & CLONE_NEWNS) && |
735 | (options->attach_flags & LXC_ATTACH_LSM) && | |
ab919e5f | 736 | ctx->lsm_label; |
8455e39e | 737 | char *lsm_label = NULL; |
a998454a CB |
738 | |
739 | /* A description of the purpose of this functionality is provided in the | |
740 | * lxc-attach(1) manual page. We have to remount here and not in the | |
741 | * parent process, otherwise /proc may not properly reflect the new pid | |
742 | * namespace. | |
743 | */ | |
744 | if (!(options->namespaces & CLONE_NEWNS) && | |
745 | (options->attach_flags & LXC_ATTACH_REMOUNT_PROC_SYS)) { | |
746 | ret = lxc_attach_remount_sys_proc(); | |
b75c344c CB |
747 | if (ret < 0) |
748 | goto on_error; | |
ea918412 | 749 | |
b75c344c | 750 | TRACE("Remounted \"/proc\" and \"/sys\""); |
a998454a CB |
751 | } |
752 | ||
5b514ce3 | 753 | /* Now perform additional attachments. */ |
a998454a | 754 | #if HAVE_SYS_PERSONALITY_H |
a998454a | 755 | if (options->attach_flags & LXC_ATTACH_SET_PERSONALITY) { |
b75c344c CB |
756 | long new_personality; |
757 | ||
758 | if (options->personality < 0) | |
ab919e5f | 759 | new_personality = ctx->personality; |
b75c344c CB |
760 | else |
761 | new_personality = options->personality; | |
ea918412 | 762 | |
ee142207 CB |
763 | if (new_personality != LXC_ARCH_UNCHANGED) { |
764 | ret = personality(new_personality); | |
765 | if (ret < 0) | |
766 | goto on_error; | |
ea918412 | 767 | |
ee142207 CB |
768 | TRACE("Set new personality"); |
769 | } | |
a998454a CB |
770 | } |
771 | #endif | |
772 | ||
773 | if (options->attach_flags & LXC_ATTACH_DROP_CAPABILITIES) { | |
677e1d27 | 774 | ret = drop_capabilities(ctx); |
b75c344c CB |
775 | if (ret < 0) |
776 | goto on_error; | |
ea918412 | 777 | |
b75c344c | 778 | TRACE("Dropped capabilities"); |
a998454a CB |
779 | } |
780 | ||
781 | /* Always set the environment (specify (LXC_ATTACH_KEEP_ENV, NULL, NULL) | |
782 | * if you want this to be a no-op). | |
783 | */ | |
ab919e5f | 784 | ret = lxc_attach_set_environment(ctx, |
7385273f | 785 | options->env_policy, |
a998454a CB |
786 | options->extra_env_vars, |
787 | options->extra_keep_env); | |
b75c344c CB |
788 | if (ret < 0) |
789 | goto on_error; | |
ea918412 | 790 | |
b75c344c | 791 | TRACE("Set up environment"); |
a998454a | 792 | |
57de839f CB |
793 | /* This remark only affects fully unprivileged containers: |
794 | * Receive fd for LSM security module before we set{g,u}id(). The reason | |
795 | * is that on set{g,u}id() the kernel will a) make us undumpable and b) | |
796 | * we will change our effective uid. This means our effective uid will | |
797 | * be different from the effective uid of the process that created us | |
798 | * which means that this processs no longer has capabilities in our | |
799 | * namespace including CAP_SYS_PTRACE. This means we will not be able to | |
800 | * read and /proc/<pid> files for the process anymore when /proc is | |
801 | * mounted with hidepid={1,2}. So let's get the lsm label fd before the | |
802 | * set{g,u}id(). | |
803 | */ | |
804 | if (needs_lsm) { | |
b75c344c | 805 | ret = lxc_abstract_unix_recv_fds(payload->ipc_socket, &lsm_fd, 1, NULL, 0); |
9044b79e | 806 | if (ret <= 0) { |
807 | if (ret < 0) | |
808 | SYSERROR("Failed to receive lsm label fd"); | |
809 | ||
b75c344c | 810 | goto on_error; |
9044b79e | 811 | } |
812 | ||
57de839f CB |
813 | TRACE("Received LSM label file descriptor %d from parent", lsm_fd); |
814 | } | |
815 | ||
08ea9270 | 816 | if (options->stdin_fd > 0 && isatty(options->stdin_fd)) { |
cd0a2b2f | 817 | ret = lxc_make_controlling_terminal(options->stdin_fd); |
08ea9270 CB |
818 | if (ret < 0) |
819 | goto on_error; | |
820 | } | |
821 | ||
b58214ac CB |
822 | if (!lxc_setgroups(0, NULL) && errno != EPERM) |
823 | goto on_error; | |
824 | ||
936efc72 CB |
825 | if (options->namespaces & CLONE_NEWUSER) { |
826 | /* Check whether nsuid 0 has a mapping. */ | |
827 | ns_root_uid = get_ns_uid(0); | |
ea918412 | 828 | |
936efc72 CB |
829 | /* Check whether nsgid 0 has a mapping. */ |
830 | ns_root_gid = get_ns_gid(0); | |
a998454a | 831 | |
936efc72 CB |
832 | /* If there's no mapping for nsuid 0 try to retrieve the nsuid |
833 | * init was started with. | |
834 | */ | |
835 | if (ns_root_uid == LXC_INVALID_UID) | |
836 | lxc_attach_get_init_uidgid(&ns_root_uid, &ns_root_gid); | |
ea918412 | 837 | |
936efc72 CB |
838 | if (ns_root_uid == LXC_INVALID_UID) |
839 | goto on_error; | |
a998454a | 840 | |
464c4611 | 841 | if (!lxc_switch_uid_gid(ns_root_uid, ns_root_gid)) |
b75c344c | 842 | goto on_error; |
a998454a CB |
843 | } |
844 | ||
936efc72 CB |
845 | /* Set {u,g}id. */ |
846 | if (options->uid != LXC_INVALID_UID) | |
847 | new_uid = options->uid; | |
848 | else | |
849 | new_uid = ns_root_uid; | |
850 | ||
851 | if (options->gid != LXC_INVALID_GID) | |
852 | new_gid = options->gid; | |
853 | else | |
854 | new_gid = ns_root_gid; | |
855 | ||
57de839f | 856 | if (needs_lsm) { |
d3ba7c98 | 857 | bool on_exec; |
a998454a CB |
858 | |
859 | /* Change into our new LSM profile. */ | |
d3ba7c98 | 860 | on_exec = options->attach_flags & LXC_ATTACH_LSM_EXEC ? true : false; |
65129087 CB |
861 | if (options->attach_flags & LXC_ATTACH_LSM_LABEL) |
862 | lsm_label = options->lsm_label; | |
863 | if (!lsm_label) | |
ab919e5f CB |
864 | lsm_label = ctx->lsm_label; |
865 | ret = ctx->lsm_ops->process_label_set_at(ctx->lsm_ops, lsm_fd, | |
866 | lsm_label, on_exec); | |
cb2420df | 867 | close_prot_errno_disarm(lsm_fd); |
b75c344c CB |
868 | if (ret < 0) |
869 | goto on_error; | |
ea918412 | 870 | |
ab919e5f | 871 | TRACE("Set %s LSM label to \"%s\"", ctx->lsm_ops->name, ctx->lsm_label); |
a998454a CB |
872 | } |
873 | ||
ab919e5f | 874 | if ((ctx->container && conf && conf->no_new_privs) || |
6ce8e678 AL |
875 | (options->attach_flags & LXC_ATTACH_NO_NEW_PRIVS)) { |
876 | ret = prctl(PR_SET_NO_NEW_PRIVS, prctl_arg(1), prctl_arg(0), | |
877 | prctl_arg(0), prctl_arg(0)); | |
878 | if (ret < 0) | |
879 | goto on_error; | |
880 | ||
881 | TRACE("Set PR_SET_NO_NEW_PRIVS"); | |
882 | } | |
883 | ||
ab919e5f | 884 | if (ctx->container && conf && conf->seccomp.seccomp) { |
cdb2a47f | 885 | ret = lxc_seccomp_load(conf); |
b75c344c CB |
886 | if (ret < 0) |
887 | goto on_error; | |
ea918412 | 888 | |
b75c344c | 889 | TRACE("Loaded seccomp profile"); |
cdb2a47f | 890 | |
c3e3c21a CB |
891 | ret = lxc_seccomp_send_notifier_fd(&conf->seccomp, payload->ipc_socket); |
892 | if (ret < 0) | |
893 | goto on_error; | |
a998454a | 894 | } |
ea918412 | 895 | |
578d4b62 | 896 | close_prot_errno_disarm(payload->ipc_socket); |
dd53c8af | 897 | put_attach_context(ctx); |
ab919e5f | 898 | payload->ctx = NULL; |
a998454a CB |
899 | |
900 | /* The following is done after the communication socket is shut down. | |
901 | * That way, all errors that might (though unlikely) occur up until this | |
902 | * point will have their messages printed to the original stderr (if | |
903 | * logging is so configured) and not the fd the user supplied, if any. | |
904 | */ | |
905 | ||
906 | /* Fd handling for stdin, stdout and stderr; ignore errors here, user | |
907 | * may want to make sure the fds are closed, for example. | |
908 | */ | |
08ea9270 | 909 | if (options->stdin_fd >= 0 && options->stdin_fd != STDIN_FILENO) |
b4959848 CB |
910 | if (dup2(options->stdin_fd, STDIN_FILENO)) |
911 | DEBUG("Failed to replace stdin with %d", options->stdin_fd); | |
08ea9270 CB |
912 | |
913 | if (options->stdout_fd >= 0 && options->stdout_fd != STDOUT_FILENO) | |
b4959848 CB |
914 | if (dup2(options->stdout_fd, STDOUT_FILENO)) |
915 | DEBUG("Failed to replace stdout with %d", options->stdin_fd); | |
08ea9270 CB |
916 | |
917 | if (options->stderr_fd >= 0 && options->stderr_fd != STDERR_FILENO) | |
b4959848 CB |
918 | if (dup2(options->stderr_fd, STDERR_FILENO)) |
919 | DEBUG("Failed to replace stderr with %d", options->stdin_fd); | |
a998454a CB |
920 | |
921 | /* close the old fds */ | |
08ea9270 | 922 | if (options->stdin_fd > STDERR_FILENO) |
a998454a | 923 | close(options->stdin_fd); |
08ea9270 CB |
924 | |
925 | if (options->stdout_fd > STDERR_FILENO) | |
a998454a | 926 | close(options->stdout_fd); |
08ea9270 CB |
927 | |
928 | if (options->stderr_fd > STDERR_FILENO) | |
a998454a CB |
929 | close(options->stderr_fd); |
930 | ||
427a8067 CB |
931 | /* |
932 | * Try to remove FD_CLOEXEC flag from stdin/stdout/stderr, but also | |
a998454a CB |
933 | * here, ignore errors. |
934 | */ | |
427a8067 | 935 | for (int fd = STDIN_FILENO; fd <= STDERR_FILENO; fd++) { |
3f62938a | 936 | ret = fd_cloexec(fd, false); |
b75c344c CB |
937 | if (ret < 0) { |
938 | SYSERROR("Failed to clear FD_CLOEXEC from file descriptor %d", fd); | |
939 | goto on_error; | |
940 | } | |
a998454a CB |
941 | } |
942 | ||
9e84479f | 943 | if (options->attach_flags & LXC_ATTACH_TERMINAL) { |
cecf3e83 | 944 | ret = lxc_terminal_prepare_login(payload->terminal_pts_fd); |
ba2be1a8 | 945 | if (ret < 0) { |
cecf3e83 | 946 | SYSERROR("Failed to prepare terminal file descriptor %d", payload->terminal_pts_fd); |
ba2be1a8 CB |
947 | goto on_error; |
948 | } | |
ea918412 | 949 | |
cecf3e83 | 950 | TRACE("Prepared terminal file descriptor %d", payload->terminal_pts_fd); |
ba2be1a8 CB |
951 | } |
952 | ||
936efc72 CB |
953 | /* Avoid unnecessary syscalls. */ |
954 | if (new_uid == ns_root_uid) | |
955 | new_uid = LXC_INVALID_UID; | |
956 | ||
957 | if (new_gid == ns_root_gid) | |
958 | new_gid = LXC_INVALID_GID; | |
c353b0b9 | 959 | |
6aff5157 | 960 | /* Make sure that the processes STDIO is correctly owned by the user that we are switching to */ |
c353b0b9 CB |
961 | ret = fix_stdio_permissions(new_uid); |
962 | if (ret) | |
a2c26bef | 963 | INFO("Failed to adjust stdio permissions"); |
936efc72 | 964 | |
464c4611 | 965 | if (!lxc_switch_uid_gid(new_uid, new_gid)) |
936efc72 CB |
966 | goto on_error; |
967 | ||
a998454a | 968 | /* We're done, so we can now do whatever the user intended us to do. */ |
c7ac2e1c | 969 | _exit(payload->exec_function(payload->exec_payload)); |
b75c344c CB |
970 | |
971 | on_error: | |
ba2be1a8 | 972 | lxc_put_attach_clone_payload(payload); |
dab02267 | 973 | ERROR("Failed to attach to container"); |
c7ac2e1c | 974 | _exit(EXIT_FAILURE); |
a998454a CB |
975 | } |
976 | ||
f797f05e | 977 | static int lxc_attach_terminal(const char *name, const char *lxcpath, struct lxc_conf *conf, |
9e84479f | 978 | struct lxc_terminal *terminal) |
ba2be1a8 CB |
979 | { |
980 | int ret; | |
981 | ||
9e84479f | 982 | lxc_terminal_init(terminal); |
ba2be1a8 | 983 | |
8ea93a0f | 984 | ret = lxc_terminal_create(name, lxcpath, conf, terminal); |
c2af3a15 CB |
985 | if (ret < 0) |
986 | return log_error(-1, "Failed to create terminal"); | |
ba2be1a8 | 987 | |
ba2be1a8 | 988 | return 0; |
ba2be1a8 CB |
989 | } |
990 | ||
9e84479f CB |
991 | static int lxc_attach_terminal_mainloop_init(struct lxc_terminal *terminal, |
992 | struct lxc_epoll_descr *descr) | |
ba2be1a8 CB |
993 | { |
994 | int ret; | |
995 | ||
996 | ret = lxc_mainloop_open(descr); | |
c2af3a15 CB |
997 | if (ret < 0) |
998 | return log_error(-1, "Failed to create mainloop"); | |
ba2be1a8 | 999 | |
9e84479f | 1000 | ret = lxc_terminal_mainloop_add(descr, terminal); |
ba2be1a8 | 1001 | if (ret < 0) { |
ba2be1a8 | 1002 | lxc_mainloop_close(descr); |
c2af3a15 | 1003 | return log_error(-1, "Failed to add handlers to mainloop"); |
ba2be1a8 CB |
1004 | } |
1005 | ||
1006 | return 0; | |
1007 | } | |
1008 | ||
36a94ce8 | 1009 | static inline void lxc_attach_terminal_close_ptx(struct lxc_terminal *terminal) |
ba2be1a8 | 1010 | { |
36a94ce8 | 1011 | close_prot_errno_disarm(terminal->ptx); |
ba2be1a8 CB |
1012 | } |
1013 | ||
cecf3e83 | 1014 | static inline void lxc_attach_terminal_close_pts(struct lxc_terminal *terminal) |
ba2be1a8 | 1015 | { |
41808e20 | 1016 | close_prot_errno_disarm(terminal->pty); |
ba2be1a8 CB |
1017 | } |
1018 | ||
9e84479f | 1019 | static inline void lxc_attach_terminal_close_peer(struct lxc_terminal *terminal) |
ba2be1a8 | 1020 | { |
19a3e906 | 1021 | close_prot_errno_disarm(terminal->peer); |
ba2be1a8 CB |
1022 | } |
1023 | ||
9e84479f | 1024 | static inline void lxc_attach_terminal_close_log(struct lxc_terminal *terminal) |
ba2be1a8 | 1025 | { |
19a3e906 | 1026 | close_prot_errno_disarm(terminal->log_fd); |
ba2be1a8 CB |
1027 | } |
1028 | ||
908fbc1a CB |
1029 | int lxc_attach(struct lxc_container *container, lxc_attach_exec_t exec_function, |
1030 | void *exec_payload, lxc_attach_options_t *options, | |
1031 | pid_t *attached_process) | |
9c4693b8 | 1032 | { |
6f9fe5d0 CB |
1033 | int ret_parent = -1; |
1034 | struct attach_clone_payload payload = {}; | |
1035 | struct lxc_epoll_descr descr = {}; | |
877f3a04 | 1036 | int i, ret, status; |
6f9fe5d0 | 1037 | char *name, *lxcpath, *new_cwd; |
9c4693b8 | 1038 | int ipc_sockets[2]; |
500ed813 | 1039 | pid_t attached_pid, pid, to_cleanup_pid; |
ab919e5f | 1040 | struct attach_context *ctx; |
9e84479f | 1041 | struct lxc_terminal terminal; |
1cce35e6 | 1042 | struct lxc_conf *conf; |
9c4693b8 | 1043 | |
908fbc1a | 1044 | if (!container) |
540a2f70 | 1045 | return ret_set_errno(-1, EINVAL); |
908fbc1a CB |
1046 | |
1047 | if (!lxc_container_get(container)) | |
540a2f70 | 1048 | return ret_set_errno(-1, EINVAL); |
908fbc1a CB |
1049 | |
1050 | name = container->name; | |
1051 | lxcpath = container->config_path; | |
1052 | ||
9c4693b8 CS |
1053 | if (!options) |
1054 | options = &attach_static_default_options; | |
1055 | ||
9745eb8a | 1056 | ctx = alloc_attach_context(); |
ab919e5f | 1057 | if (!ctx) { |
9745eb8a CB |
1058 | lxc_container_put(container); |
1059 | return log_error_errno(-ENOMEM, ENOMEM, "Failed to allocate attach context"); | |
1060 | } | |
1061 | ||
500ed813 | 1062 | ret = get_attach_context(ctx, container); |
9745eb8a | 1063 | if (ret) { |
500ed813 | 1064 | ERROR("Failed to get attach context"); |
7e995801 | 1065 | put_attach_context(ctx); |
9c4693b8 CS |
1066 | return -1; |
1067 | } | |
1068 | ||
ab919e5f | 1069 | conf = ctx->container->lxc_conf; |
ba773996 | 1070 | |
ab919e5f | 1071 | if (!fetch_seccomp(ctx->container, options)) |
ae026f55 | 1072 | WARN("Failed to get seccomp policy"); |
2c4ea790 | 1073 | |
ab919e5f | 1074 | if (!no_new_privs(ctx->container, options)) |
ae026f55 | 1075 | WARN("Could not determine whether PR_SET_NO_NEW_PRIVS is set"); |
2e812c16 | 1076 | |
8ce83369 CB |
1077 | /* Determine which namespaces the container was created with |
1078 | * by asking lxc-start, if necessary. | |
9c4693b8 CS |
1079 | */ |
1080 | if (options->namespaces == -1) { | |
1081 | options->namespaces = lxc_cmd_get_clone_flags(name, lxcpath); | |
1082 | /* call failed */ | |
1083 | if (options->namespaces == -1) { | |
8ce83369 | 1084 | ERROR("Failed to automatically determine the " |
877f3a04 | 1085 | "namespaces which the container uses"); |
dd53c8af | 1086 | put_attach_context(ctx); |
9c4693b8 CS |
1087 | return -1; |
1088 | } | |
877f3a04 CB |
1089 | |
1090 | for (i = 0; i < LXC_NS_MAX; i++) { | |
1091 | if (ns_info[i].clone_flag & CLONE_NEWCGROUP) | |
1092 | if (!(options->attach_flags & LXC_ATTACH_MOVE_TO_CGROUP) || | |
1093 | !cgns_supported()) | |
1094 | continue; | |
1095 | ||
1096 | if (ns_info[i].clone_flag & options->namespaces) | |
1097 | continue; | |
1098 | ||
ab919e5f | 1099 | ctx->ns_inherited |= ns_info[i].clone_flag; |
877f3a04 CB |
1100 | } |
1101 | } | |
1102 | ||
b7873c95 CB |
1103 | ret = get_attach_context_nsfds(ctx, options); |
1104 | if (ret) { | |
1105 | ERROR("Failed to get namespace file descriptors"); | |
1106 | lxc_container_put(container); | |
877f3a04 | 1107 | return -1; |
9c4693b8 CS |
1108 | } |
1109 | ||
9e84479f | 1110 | if (options->attach_flags & LXC_ATTACH_TERMINAL) { |
f797f05e | 1111 | ret = lxc_attach_terminal(name, lxcpath, conf, &terminal); |
ba2be1a8 | 1112 | if (ret < 0) { |
9e84479f | 1113 | ERROR("Failed to setup new terminal"); |
dd53c8af | 1114 | put_attach_context(ctx); |
ba2be1a8 CB |
1115 | return -1; |
1116 | } | |
1117 | ||
9e84479f | 1118 | terminal.log_fd = options->log_fd; |
c948657b | 1119 | } else { |
9e84479f | 1120 | lxc_terminal_init(&terminal); |
ba2be1a8 CB |
1121 | } |
1122 | ||
8ce83369 CB |
1123 | /* Create a socket pair for IPC communication; set SOCK_CLOEXEC in order |
1124 | * to make sure we don't irritate other threads that want to fork+exec | |
1125 | * away | |
9c4693b8 CS |
1126 | * |
1127 | * IMPORTANT: if the initial process is multithreaded and another call | |
1128 | * just fork()s away without exec'ing directly after, the socket fd will | |
1129 | * exist in the forked process from the other thread and any close() in | |
8ce83369 | 1130 | * our own child process will not really cause the socket to close |
1d801260 | 1131 | * properly, potentially causing the parent to hang. |
9c4693b8 CS |
1132 | * |
1133 | * For this reason, while IPC is still active, we have to use shutdown() | |
8ce83369 CB |
1134 | * if the child exits prematurely in order to signal that the socket is |
1135 | * closed and cannot assume that the child exiting will automatically do | |
1136 | * that. | |
9c4693b8 CS |
1137 | * |
1138 | * IPC mechanism: (X is receiver) | |
1139 | * initial process intermediate attached | |
1140 | * X <--- send pid of | |
1141 | * attached proc, | |
1142 | * then exit | |
1143 | * send 0 ------------------------------------> X | |
1144 | * [do initialization] | |
1145 | * X <------------------------------------ send 1 | |
1146 | * [add to cgroup, ...] | |
1147 | * send 2 ------------------------------------> X | |
81f466d0 CB |
1148 | * [set LXC_ATTACH_NO_NEW_PRIVS] |
1149 | * X <------------------------------------ send 3 | |
1150 | * [open LSM label fd] | |
1151 | * send 4 ------------------------------------> X | |
1152 | * [set LSM label] | |
9c4693b8 CS |
1153 | * close socket close socket |
1154 | * run program | |
1155 | */ | |
1156 | ret = socketpair(PF_LOCAL, SOCK_STREAM | SOCK_CLOEXEC, 0, ipc_sockets); | |
1157 | if (ret < 0) { | |
ae026f55 | 1158 | SYSERROR("Could not set up required IPC mechanism for attaching"); |
dd53c8af | 1159 | put_attach_context(ctx); |
9c4693b8 CS |
1160 | return -1; |
1161 | } | |
1162 | ||
e3f0e436 CB |
1163 | /* Create intermediate subprocess, two reasons: |
1164 | * 1. We can't setns() in the child itself, since we want to make | |
8ce83369 | 1165 | * sure we are properly attached to the pidns. |
e3f0e436 | 1166 | * 2. Also, the initial thread has to put the attached process |
8ce83369 CB |
1167 | * into the cgroup, which we can only do if we didn't already |
1168 | * setns() (otherwise, user namespaces will hate us). | |
9c4693b8 CS |
1169 | */ |
1170 | pid = fork(); | |
9c4693b8 | 1171 | if (pid < 0) { |
ae026f55 | 1172 | SYSERROR("Failed to create first subprocess"); |
dd53c8af | 1173 | put_attach_context(ctx); |
9c4693b8 CS |
1174 | return -1; |
1175 | } | |
1176 | ||
4f25e72f | 1177 | if (pid == 0) { |
a588a482 CB |
1178 | char *cwd; |
1179 | ||
ba2be1a8 | 1180 | /* close unneeded file descriptors */ |
4f25e72f | 1181 | close_prot_errno_disarm(ipc_sockets[0]); |
2202afc9 | 1182 | |
4f25e72f CB |
1183 | if (options->attach_flags & LXC_ATTACH_TERMINAL) { |
1184 | lxc_attach_terminal_close_ptx(&terminal); | |
1185 | lxc_attach_terminal_close_peer(&terminal); | |
1186 | lxc_attach_terminal_close_log(&terminal); | |
f4364484 SG |
1187 | } |
1188 | ||
4f25e72f CB |
1189 | /* Wait for the parent to have setup cgroups. */ |
1190 | ret = lxc_read_nointr(ipc_sockets[1], &status, sizeof(status)); | |
1191 | if (ret != sizeof(status)) { | |
1192 | shutdown(ipc_sockets[1], SHUT_RDWR); | |
dd53c8af | 1193 | put_attach_context(ctx); |
4f25e72f | 1194 | _exit(EXIT_FAILURE); |
bb2ada6f CB |
1195 | } |
1196 | ||
4f25e72f CB |
1197 | TRACE("Intermediate process starting to initialize"); |
1198 | ||
a588a482 CB |
1199 | cwd = getcwd(NULL, 0); |
1200 | ||
4f25e72f CB |
1201 | /* Attach now, create another subprocess later, since pid namespaces |
1202 | * only really affect the children of the current process. | |
1203 | */ | |
500ed813 | 1204 | ret = lxc_attach_to_ns(ctx->init_pid, ctx); |
4f25e72f CB |
1205 | if (ret < 0) { |
1206 | ERROR("Failed to enter namespaces"); | |
1207 | shutdown(ipc_sockets[1], SHUT_RDWR); | |
dd53c8af | 1208 | put_attach_context(ctx); |
4f25e72f | 1209 | _exit(EXIT_FAILURE); |
ba2be1a8 CB |
1210 | } |
1211 | ||
4f25e72f | 1212 | /* close namespace file descriptors */ |
7e995801 | 1213 | close_nsfds(ctx); |
ea918412 | 1214 | |
4f25e72f CB |
1215 | /* Attach succeeded, try to cwd. */ |
1216 | if (options->initial_cwd) | |
1217 | new_cwd = options->initial_cwd; | |
1218 | else | |
1219 | new_cwd = cwd; | |
1220 | if (new_cwd) { | |
1221 | ret = chdir(new_cwd); | |
1222 | if (ret < 0) | |
1223 | WARN("Could not change directory to \"%s\"", new_cwd); | |
ba2be1a8 | 1224 | } |
a588a482 | 1225 | free_disarm(cwd); |
c6d09e15 | 1226 | |
4f25e72f | 1227 | /* Create attached process. */ |
76783714 CB |
1228 | payload.ipc_socket = ipc_sockets[1]; |
1229 | payload.options = options; | |
ab919e5f | 1230 | payload.ctx = ctx; |
4f25e72f | 1231 | payload.terminal_pts_fd = terminal.pty; |
76783714 CB |
1232 | payload.exec_function = exec_function; |
1233 | payload.exec_payload = exec_payload; | |
4f25e72f CB |
1234 | |
1235 | pid = lxc_raw_clone(CLONE_PARENT, NULL); | |
1236 | if (pid < 0) { | |
1237 | SYSERROR("Failed to clone attached process"); | |
1238 | shutdown(ipc_sockets[1], SHUT_RDWR); | |
dd53c8af | 1239 | put_attach_context(ctx); |
4f25e72f CB |
1240 | _exit(EXIT_FAILURE); |
1241 | } | |
f4364484 | 1242 | |
4f25e72f CB |
1243 | if (pid == 0) { |
1244 | if (options->attach_flags & LXC_ATTACH_TERMINAL) { | |
1245 | ret = lxc_terminal_signal_sigmask_safe_blocked(&terminal); | |
1246 | if (ret < 0) { | |
1247 | SYSERROR("Failed to reset signal mask"); | |
1248 | _exit(EXIT_FAILURE); | |
1249 | } | |
1250 | } | |
ea918412 | 1251 | |
dab02267 | 1252 | do_attach(&payload); |
62183f1a | 1253 | } |
2eef2bda | 1254 | |
4f25e72f CB |
1255 | if (options->attach_flags & LXC_ATTACH_TERMINAL) |
1256 | lxc_attach_terminal_close_pts(&terminal); | |
ea918412 | 1257 | |
4f25e72f CB |
1258 | /* Tell grandparent the pid of the pid of the newly created child. */ |
1259 | ret = lxc_write_nointr(ipc_sockets[1], &pid, sizeof(pid)); | |
1260 | if (ret != sizeof(pid)) { | |
1261 | /* If this really happens here, this is very unfortunate, since | |
1262 | * the parent will not know the pid of the attached process and | |
1263 | * will not be able to wait for it (and we won't either due to | |
1264 | * CLONE_PARENT) so the parent won't be able to reap it and the | |
1265 | * attached process will remain a zombie. | |
1266 | */ | |
1267 | shutdown(ipc_sockets[1], SHUT_RDWR); | |
dd53c8af | 1268 | put_attach_context(ctx); |
4f25e72f CB |
1269 | _exit(EXIT_FAILURE); |
1270 | } | |
9c4693b8 | 1271 | |
4f25e72f | 1272 | TRACE("Sending pid %d of attached process", pid); |
9c4693b8 | 1273 | |
4f25e72f | 1274 | /* The rest is in the hands of the initial and the attached process. */ |
dd53c8af | 1275 | put_attach_context(ctx); |
4f25e72f CB |
1276 | _exit(EXIT_SUCCESS); |
1277 | } | |
6f4f1937 | 1278 | |
4f25e72f | 1279 | to_cleanup_pid = pid; |
ea918412 | 1280 | |
4f25e72f | 1281 | /* close unneeded file descriptors */ |
cb2420df | 1282 | close_prot_errno_disarm(ipc_sockets[1]); |
7e995801 | 1283 | close_nsfds(ctx); |
4f25e72f CB |
1284 | if (options->attach_flags & LXC_ATTACH_TERMINAL) |
1285 | lxc_attach_terminal_close_pts(&terminal); | |
81f466d0 | 1286 | |
4f25e72f CB |
1287 | /* Attach to cgroup, if requested. */ |
1288 | if (options->attach_flags & LXC_ATTACH_MOVE_TO_CGROUP) { | |
1289 | /* | |
1290 | * If this is the unified hierarchy cgroup_attach() is | |
1291 | * enough. | |
1292 | */ | |
1293 | ret = cgroup_attach(conf, name, lxcpath, pid); | |
1294 | if (ret) { | |
1295 | call_cleaner(cgroup_exit) struct cgroup_ops *cgroup_ops = NULL; | |
9044b79e | 1296 | |
4f25e72f CB |
1297 | cgroup_ops = cgroup_init(conf); |
1298 | if (!cgroup_ops) | |
1299 | goto on_error; | |
9044b79e | 1300 | |
4f25e72f CB |
1301 | if (!cgroup_ops->attach(cgroup_ops, conf, name, lxcpath, pid)) |
1302 | goto on_error; | |
81f466d0 | 1303 | } |
4f25e72f CB |
1304 | TRACE("Moved intermediate process %d into container's cgroups", pid); |
1305 | } | |
81f466d0 | 1306 | |
4f25e72f CB |
1307 | /* Setup /proc limits */ |
1308 | if (!lxc_list_empty(&conf->procs)) { | |
1309 | ret = setup_proc_filesystem(&conf->procs, pid); | |
1310 | if (ret < 0) | |
1311 | goto on_error; | |
4f3b6a85 CB |
1312 | |
1313 | TRACE("Setup /proc/%d settings", pid); | |
4f25e72f | 1314 | } |
cdb2a47f | 1315 | |
4f25e72f CB |
1316 | /* Setup resource limits */ |
1317 | if (!lxc_list_empty(&conf->limits)) { | |
1318 | ret = setup_resource_limits(&conf->limits, pid); | |
1319 | if (ret < 0) | |
1320 | goto on_error; | |
4f3b6a85 CB |
1321 | |
1322 | TRACE("Setup resource limits"); | |
4f25e72f | 1323 | } |
cdb2a47f | 1324 | |
4f25e72f CB |
1325 | if (options->attach_flags & LXC_ATTACH_TERMINAL) { |
1326 | ret = lxc_attach_terminal_mainloop_init(&terminal, &descr); | |
1327 | if (ret < 0) | |
1328 | goto on_error; | |
9c4693b8 | 1329 | |
4f25e72f CB |
1330 | TRACE("Initialized terminal mainloop"); |
1331 | } | |
9c4693b8 | 1332 | |
4f25e72f CB |
1333 | /* Let the child process know to go ahead. */ |
1334 | status = 0; | |
1335 | ret = lxc_write_nointr(ipc_sockets[0], &status, sizeof(status)); | |
1336 | if (ret != sizeof(status)) | |
1337 | goto close_mainloop; | |
ba2be1a8 | 1338 | |
4f25e72f | 1339 | TRACE("Told intermediate process to start initializing"); |
ea918412 | 1340 | |
4f25e72f CB |
1341 | /* Get pid of attached process from intermediate process. */ |
1342 | ret = lxc_read_nointr(ipc_sockets[0], &attached_pid, sizeof(attached_pid)); | |
1343 | if (ret != sizeof(attached_pid)) | |
1344 | goto close_mainloop; | |
ba2be1a8 | 1345 | |
4f25e72f | 1346 | TRACE("Received pid %d of attached process in parent pid namespace", attached_pid); |
ba2be1a8 | 1347 | |
4f25e72f CB |
1348 | /* Ignore SIGKILL (CTRL-C) and SIGQUIT (CTRL-\) - issue #313. */ |
1349 | if (options->stdin_fd == 0) { | |
1350 | signal(SIGINT, SIG_IGN); | |
1351 | signal(SIGQUIT, SIG_IGN); | |
1352 | } | |
ba2be1a8 | 1353 | |
4f25e72f CB |
1354 | /* Reap intermediate process. */ |
1355 | ret = wait_for_pid(pid); | |
1356 | if (ret < 0) | |
1357 | goto close_mainloop; | |
ba2be1a8 | 1358 | |
4f25e72f | 1359 | TRACE("Intermediate process %d exited", pid); |
ea918412 | 1360 | |
4f25e72f CB |
1361 | /* We will always have to reap the attached process now. */ |
1362 | to_cleanup_pid = attached_pid; | |
9c4693b8 | 1363 | |
4f25e72f CB |
1364 | /* Open LSM fd and send it to child. */ |
1365 | if ((options->namespaces & CLONE_NEWNS) && | |
ab919e5f | 1366 | (options->attach_flags & LXC_ATTACH_LSM) && ctx->lsm_label) { |
ad001fb6 | 1367 | __do_close int labelfd = -EBADF; |
4f25e72f | 1368 | bool on_exec; |
ea918412 | 1369 | |
4f25e72f CB |
1370 | ret = -1; |
1371 | on_exec = options->attach_flags & LXC_ATTACH_LSM_EXEC ? true : false; | |
ab919e5f CB |
1372 | labelfd = ctx->lsm_ops->process_label_fd_get(ctx->lsm_ops, |
1373 | attached_pid, on_exec); | |
4f25e72f CB |
1374 | if (labelfd < 0) |
1375 | goto close_mainloop; | |
9c4693b8 | 1376 | |
4f25e72f | 1377 | TRACE("Opened LSM label file descriptor %d", labelfd); |
ea918412 | 1378 | |
4f25e72f CB |
1379 | /* Send child fd of the LSM security module to write to. */ |
1380 | ret = lxc_abstract_unix_send_fds(ipc_sockets[0], &labelfd, 1, NULL, 0); | |
1381 | if (ret <= 0) { | |
1382 | if (ret < 0) | |
1383 | SYSERROR("Failed to send lsm label fd"); | |
4f25e72f CB |
1384 | goto close_mainloop; |
1385 | } | |
1386 | ||
4f25e72f | 1387 | TRACE("Sent LSM label file descriptor %d to child", labelfd); |
9c4693b8 | 1388 | } |
ea918412 | 1389 | |
4f25e72f CB |
1390 | if (conf->seccomp.seccomp) { |
1391 | ret = lxc_seccomp_recv_notifier_fd(&conf->seccomp, ipc_sockets[0]); | |
1392 | if (ret < 0) | |
1393 | goto close_mainloop; | |
9c4693b8 | 1394 | |
4f25e72f | 1395 | ret = lxc_seccomp_add_notifier(name, lxcpath, &conf->seccomp); |
d6d979bc | 1396 | if (ret < 0) |
4f25e72f | 1397 | goto close_mainloop; |
d6d979bc | 1398 | } |
9c4693b8 | 1399 | |
4f25e72f CB |
1400 | /* We're done, the child process should now execute whatever it |
1401 | * is that the user requested. The parent can now track it with | |
1402 | * waitpid() or similar. | |
1403 | */ | |
9c4693b8 | 1404 | |
4f25e72f | 1405 | *attached_process = attached_pid; |
a998454a | 1406 | |
4f25e72f CB |
1407 | /* Now shut down communication with child, we're done. */ |
1408 | shutdown(ipc_sockets[0], SHUT_RDWR); | |
cb2420df | 1409 | close_prot_errno_disarm(ipc_sockets[0]); |
f157b056 | 1410 | |
4f25e72f CB |
1411 | ret_parent = 0; |
1412 | to_cleanup_pid = -1; | |
ea918412 | 1413 | |
4f25e72f CB |
1414 | if (options->attach_flags & LXC_ATTACH_TERMINAL) { |
1415 | ret = lxc_mainloop(&descr, -1); | |
1416 | if (ret < 0) { | |
1417 | ret_parent = -1; | |
1418 | to_cleanup_pid = attached_pid; | |
1419 | } | |
a998454a | 1420 | } |
ea918412 | 1421 | |
4f25e72f | 1422 | close_mainloop: |
9e84479f | 1423 | if (options->attach_flags & LXC_ATTACH_TERMINAL) |
4f25e72f | 1424 | lxc_mainloop_close(&descr); |
9c4693b8 | 1425 | |
4f25e72f CB |
1426 | on_error: |
1427 | if (ipc_sockets[0] >= 0) { | |
1428 | shutdown(ipc_sockets[0], SHUT_RDWR); | |
cb2420df | 1429 | close_prot_errno_disarm(ipc_sockets[0]); |
9c4693b8 | 1430 | } |
ea918412 | 1431 | |
4f25e72f CB |
1432 | if (to_cleanup_pid > 0) |
1433 | (void)wait_for_pid(to_cleanup_pid); | |
1434 | ||
1435 | if (options->attach_flags & LXC_ATTACH_TERMINAL) { | |
1436 | lxc_terminal_delete(&terminal); | |
1437 | lxc_terminal_conf_free(&terminal); | |
1438 | } | |
9c4693b8 | 1439 | |
dd53c8af | 1440 | put_attach_context(ctx); |
4f25e72f | 1441 | return ret_parent; |
9c4693b8 CS |
1442 | } |
1443 | ||
06346bb0 | 1444 | int lxc_attach_run_command(void *payload) |
9c4693b8 | 1445 | { |
06346bb0 CB |
1446 | int ret = -1; |
1447 | lxc_attach_command_t *cmd = payload; | |
9c4693b8 | 1448 | |
06346bb0 CB |
1449 | ret = execvp(cmd->program, cmd->argv); |
1450 | if (ret < 0) { | |
1451 | switch (errno) { | |
1452 | case ENOEXEC: | |
1453 | ret = 126; | |
cf0fd972 | 1454 | break; |
06346bb0 CB |
1455 | case ENOENT: |
1456 | ret = 127; | |
cf0fd972 | 1457 | break; |
06346bb0 CB |
1458 | } |
1459 | } | |
ea918412 | 1460 | |
c2af3a15 | 1461 | return log_error_errno(ret, errno, "Failed to exec \"%s\"", cmd->program); |
9c4693b8 CS |
1462 | } |
1463 | ||
1464 | int lxc_attach_run_shell(void* payload) | |
1465 | { | |
cd8f5663 | 1466 | __do_free char *buf = NULL; |
9c4693b8 | 1467 | uid_t uid; |
cb7aa5e8 DJ |
1468 | struct passwd pwent; |
1469 | struct passwd *pwentp = NULL; | |
9c4693b8 | 1470 | char *user_shell; |
cb7aa5e8 DJ |
1471 | size_t bufsize; |
1472 | int ret; | |
9c4693b8 | 1473 | |
8ce83369 | 1474 | /* Ignore payload parameter. */ |
9c4693b8 CS |
1475 | (void)payload; |
1476 | ||
1477 | uid = getuid(); | |
cb7aa5e8 DJ |
1478 | |
1479 | bufsize = sysconf(_SC_GETPW_R_SIZE_MAX); | |
1480 | if (bufsize == -1) | |
1481 | bufsize = 1024; | |
1482 | ||
1483 | buf = malloc(bufsize); | |
1484 | if (buf) { | |
1485 | ret = getpwuid_r(uid, &pwent, buf, bufsize, &pwentp); | |
1486 | if (!pwentp) { | |
1487 | if (ret == 0) | |
ea918412 | 1488 | WARN("Could not find matched password record"); |
cb7aa5e8 DJ |
1489 | |
1490 | WARN("Failed to get password record - %u", uid); | |
1491 | } | |
1492 | } | |
9c4693b8 | 1493 | |
8ce83369 CB |
1494 | /* This probably happens because of incompatible nss implementations in |
1495 | * host and container (remember, this code is still using the host's | |
1496 | * glibc but our mount namespace is in the container) we may try to get | |
1497 | * the information by spawning a [getent passwd uid] process and parsing | |
1498 | * the result. | |
9c4693b8 | 1499 | */ |
cb7aa5e8 | 1500 | if (!pwentp) |
9c4693b8 CS |
1501 | user_shell = lxc_attach_getpwshell(uid); |
1502 | else | |
cb7aa5e8 | 1503 | user_shell = pwent.pw_shell; |
ea918412 | 1504 | |
9c4693b8 | 1505 | if (user_shell) |
acf47e1b | 1506 | execlp(user_shell, user_shell, (char *)NULL); |
9c4693b8 | 1507 | |
8ce83369 CB |
1508 | /* Executed if either no passwd entry or execvp fails, we will fall back |
1509 | * on /bin/sh as a default shell. | |
9c4693b8 | 1510 | */ |
acf47e1b | 1511 | execlp("/bin/sh", "/bin/sh", (char *)NULL); |
ea918412 | 1512 | |
edeb1836 | 1513 | SYSERROR("Failed to execute shell"); |
cb7aa5e8 | 1514 | if (!pwentp) |
edeb1836 | 1515 | free(user_shell); |
ea918412 | 1516 | |
9c4693b8 CS |
1517 | return -1; |
1518 | } |