]>
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" |
a9f0cecf | 45 | #include "sync.h" |
59524108 | 46 | #include "syscall_wrappers.h" |
0ed9b1bc | 47 | #include "terminal.h" |
6f4f1937 | 48 | #include "utils.h" |
9c4693b8 CS |
49 | |
50 | #if HAVE_SYS_PERSONALITY_H | |
51 | #include <sys/personality.h> | |
52 | #endif | |
e0732705 | 53 | |
ac2cecc4 | 54 | lxc_log_define(attach, lxc); |
e0732705 | 55 | |
ef05d368 CB |
56 | /* Define default options if no options are supplied by the user. */ |
57 | static lxc_attach_options_t attach_static_default_options = LXC_ATTACH_OPTIONS_DEFAULT; | |
58 | ||
ab919e5f | 59 | struct attach_context { |
500ed813 | 60 | int init_pid; |
0e304baa CB |
61 | char *lsm_label; |
62 | struct lxc_container *container; | |
63 | signed long personality; | |
64 | unsigned long long capability_mask; | |
65 | int ns_inherited; | |
66 | int ns_fd[LXC_NS_MAX]; | |
67 | struct lsm_ops *lsm_ops; | |
68 | }; | |
69 | ||
6e36c297 CB |
70 | static inline bool sync_wake_pid(int fd, pid_t pid) |
71 | { | |
72 | return lxc_write_nointr(fd, &pid, sizeof(pid_t)) == sizeof(pid_t); | |
73 | } | |
74 | ||
75 | static inline bool sync_wait_pid(int fd, pid_t *pid) | |
76 | { | |
77 | return lxc_read_nointr(fd, pid, sizeof(pid_t)) == sizeof(pid_t); | |
78 | } | |
79 | ||
80 | static inline bool sync_wake_fd(int fd, int fd_send) | |
81 | { | |
82 | return lxc_abstract_unix_send_fds(fd, &fd_send, 1, NULL, 0) > 0; | |
83 | } | |
84 | ||
85 | static inline bool sync_wait_fd(int fd, int *fd_recv) | |
86 | { | |
87 | return lxc_abstract_unix_recv_fds(fd, fd_recv, 1, NULL, 0) > 0; | |
88 | } | |
89 | ||
9745eb8a CB |
90 | static struct attach_context *alloc_attach_context(void) |
91 | { | |
92 | return zalloc(sizeof(struct attach_context)); | |
93 | } | |
94 | ||
ee142207 CB |
95 | static int get_personality(const char *name, const char *lxcpath, |
96 | signed long *personality) | |
d92c8e40 CB |
97 | { |
98 | __do_free char *p = NULL; | |
ee142207 | 99 | signed long per; |
d92c8e40 CB |
100 | |
101 | p = lxc_cmd_get_config_item(name, "lxc.arch", lxcpath); | |
ee142207 CB |
102 | if (!p) { |
103 | *personality = LXC_ARCH_UNCHANGED; | |
104 | return 0; | |
105 | } | |
d92c8e40 | 106 | |
ee142207 CB |
107 | per = lxc_config_parse_arch(p); |
108 | if (per == LXC_ARCH_UNCHANGED) | |
109 | return ret_errno(EINVAL); | |
110 | ||
111 | *personality = per; | |
112 | return 0; | |
d92c8e40 CB |
113 | } |
114 | ||
500ed813 CB |
115 | static int get_attach_context(struct attach_context *ctx, |
116 | struct lxc_container *container) | |
e0732705 | 117 | { |
cd8f5663 CB |
118 | __do_free char *line = NULL; |
119 | __do_fclose FILE *proc_file = NULL; | |
6f4f1937 CB |
120 | int ret; |
121 | bool found; | |
604ca1c0 | 122 | char proc_fn[LXC_PROC_STATUS_LEN]; |
e0732705 | 123 | size_t line_bufsz = 0; |
e0732705 | 124 | |
500ed813 CB |
125 | ctx->container = container; |
126 | ||
127 | ctx->init_pid = lxc_cmd_get_init_pid(container->name, container->config_path); | |
128 | if (ctx->init_pid < 0) | |
129 | return log_error(-1, "Failed to get init pid"); | |
130 | ||
8ce83369 | 131 | /* Read capabilities. */ |
500ed813 | 132 | ret = snprintf(proc_fn, LXC_PROC_STATUS_LEN, "/proc/%d/status", ctx->init_pid); |
604ca1c0 | 133 | if (ret < 0 || ret >= LXC_PROC_STATUS_LEN) |
9745eb8a | 134 | return -EIO; |
e0732705 | 135 | |
ffeeed8b CB |
136 | proc_file = fopen(proc_fn, "re"); |
137 | if (!proc_file) | |
9745eb8a | 138 | return log_error_errno(-errno, errno, "Failed to open %s", proc_fn); |
8ce83369 CB |
139 | |
140 | found = false; | |
ea918412 | 141 | |
e0732705 | 142 | while (getline(&line, &line_bufsz, proc_file) != -1) { |
9745eb8a | 143 | ret = sscanf(line, "CapBnd: %llx", &ctx->capability_mask); |
8ce83369 CB |
144 | if (ret != EOF && ret == 1) { |
145 | found = true; | |
e0732705 CS |
146 | break; |
147 | } | |
148 | } | |
149 | ||
ffeeed8b | 150 | if (!found) |
9745eb8a | 151 | return log_error_errno(-ENOENT, ENOENT, "Failed to read capability bounding set from %s", proc_fn); |
e0732705 | 152 | |
4eb19ac0 | 153 | ctx->lsm_ops = lsm_init_static(); |
d701d729 | 154 | |
500ed813 | 155 | ctx->lsm_label = ctx->lsm_ops->process_label_get(ctx->lsm_ops, ctx->init_pid); |
9745eb8a | 156 | ctx->ns_inherited = 0; |
b87ee312 | 157 | for (int i = 0; i < LXC_NS_MAX; i++) |
9745eb8a | 158 | ctx->ns_fd[i] = -EBADF; |
e0732705 | 159 | |
ee142207 CB |
160 | ret = get_personality(container->name, container->config_path, &ctx->personality); |
161 | if (ret) | |
162 | return log_error_errno(ret, errno, "Failed to get personality of the container"); | |
d92c8e40 | 163 | |
1874ef74 CB |
164 | if (!ctx->container->lxc_conf) { |
165 | ctx->container->lxc_conf = lxc_conf_init(); | |
166 | if (!ctx->container->lxc_conf) | |
167 | return log_error_errno(-ENOMEM, ENOMEM, "Failed to allocate new lxc config"); | |
168 | } | |
169 | ||
9745eb8a | 170 | return 0; |
e0732705 CS |
171 | } |
172 | ||
299d1198 CB |
173 | /** |
174 | * in_same_namespace - Check whether two processes are in the same namespace. | |
175 | * @pid1 - PID of the first process. | |
176 | * @pid2 - PID of the second process. | |
177 | * @ns - Name of the namespace to check. Must correspond to one of the names | |
178 | * for the namespaces as shown in /proc/<pid/ns/ | |
179 | * | |
180 | * If the two processes are not in the same namespace returns an fd to the | |
181 | * namespace of the second process identified by @pid2. If the two processes are | |
182 | * in the same namespace returns -EINVAL, -1 if an error occurred. | |
183 | */ | |
184 | static int in_same_namespace(pid_t pid1, pid_t pid2, const char *ns) | |
185 | { | |
644e7393 | 186 | __do_close int ns_fd1 = -EBADF, ns_fd2 = -EBADF; |
3cc629fe | 187 | int ret = -1; |
299d1198 CB |
188 | struct stat ns_st1, ns_st2; |
189 | ||
190 | ns_fd1 = lxc_preserve_ns(pid1, ns); | |
134284c3 CB |
191 | if (ns_fd1 < 0) { |
192 | /* The kernel does not support this namespace. This is not an | |
193 | * error. | |
194 | */ | |
195 | if (errno == ENOENT) | |
196 | return -EINVAL; | |
197 | ||
3cc629fe | 198 | return -1; |
134284c3 | 199 | } |
299d1198 CB |
200 | |
201 | ns_fd2 = lxc_preserve_ns(pid2, ns); | |
21d0acc2 | 202 | if (ns_fd2 < 0) |
3cc629fe | 203 | return -1; |
299d1198 CB |
204 | |
205 | ret = fstat(ns_fd1, &ns_st1); | |
21d0acc2 | 206 | if (ret < 0) |
3cc629fe | 207 | return -1; |
299d1198 CB |
208 | |
209 | ret = fstat(ns_fd2, &ns_st2); | |
21d0acc2 | 210 | if (ret < 0) |
3cc629fe | 211 | return -1; |
299d1198 CB |
212 | |
213 | /* processes are in the same namespace */ | |
3cc629fe CB |
214 | if ((ns_st1.st_dev == ns_st2.st_dev) && (ns_st1.st_ino == ns_st2.st_ino)) |
215 | return -EINVAL; | |
299d1198 CB |
216 | |
217 | /* processes are in different namespaces */ | |
3cc629fe | 218 | return move_fd(ns_fd2); |
299d1198 CB |
219 | } |
220 | ||
b7873c95 CB |
221 | static int get_attach_context_nsfds(struct attach_context *ctx, |
222 | lxc_attach_options_t *options) | |
223 | { | |
224 | ||
225 | pid_t pid_self = lxc_raw_getpid(); | |
226 | ||
227 | for (int i = 0; i < LXC_NS_MAX; i++) { | |
228 | int j; | |
229 | ||
230 | if (options->namespaces & ns_info[i].clone_flag) | |
231 | ctx->ns_fd[i] = lxc_preserve_ns(ctx->init_pid, ns_info[i].proc_name); | |
232 | else if (ctx->ns_inherited & ns_info[i].clone_flag) | |
233 | ctx->ns_fd[i] = in_same_namespace(pid_self, ctx->init_pid, ns_info[i].proc_name); | |
234 | else | |
235 | continue; | |
236 | ||
237 | if (ctx->ns_fd[i] >= 0) | |
238 | continue; | |
239 | ||
240 | if (ctx->ns_fd[i] == -EINVAL) { | |
241 | DEBUG("Inheriting %s namespace from %d", ns_info[i].proc_name, pid_self); | |
242 | ctx->ns_inherited &= ~ns_info[i].clone_flag; | |
243 | continue; | |
244 | } | |
245 | ||
246 | /* We failed to preserve the namespace. */ | |
247 | SYSERROR("Failed to attach to %s namespace of %d", ns_info[i].proc_name, pid_self); | |
248 | ||
249 | /* Close all already opened file descriptors before we return an | |
250 | * error, so we don't leak them. | |
251 | */ | |
252 | for (j = 0; j < i; j++) | |
253 | close_prot_errno_disarm(ctx->ns_fd[j]); | |
254 | ||
255 | return -1; | |
256 | } | |
257 | ||
258 | return 0; | |
259 | } | |
260 | ||
7e995801 | 261 | static inline void close_nsfds(struct attach_context *ctx) |
b7873c95 CB |
262 | { |
263 | for (int i = 0; i < LXC_NS_MAX; i++) | |
264 | close_prot_errno_disarm(ctx->ns_fd[i]); | |
265 | } | |
266 | ||
267 | static void put_attach_context(struct attach_context *ctx) | |
268 | { | |
269 | free_disarm(ctx->lsm_label); | |
270 | ||
271 | if (ctx->container) { | |
272 | lxc_container_put(ctx->container); | |
273 | ctx->container = NULL; | |
274 | } | |
275 | ||
7e995801 | 276 | close_nsfds(ctx); |
b7873c95 CB |
277 | free(ctx); |
278 | } | |
279 | ||
39b3b69b | 280 | static int attach_nsfds(struct attach_context *ctx) |
99d50954 | 281 | { |
ffeeed8b CB |
282 | for (int i = 0; i < LXC_NS_MAX; i++) { |
283 | int ret; | |
99d50954 | 284 | |
877f3a04 | 285 | if (ctx->ns_fd[i] < 0) |
26818618 CB |
286 | continue; |
287 | ||
21d0acc2 | 288 | ret = setns(ctx->ns_fd[i], ns_info[i].clone_flag); |
ffeeed8b | 289 | if (ret < 0) |
39b3b69b CB |
290 | return log_error_errno(-1, errno, |
291 | "Failed to attach to %s namespace of %d", | |
292 | ns_info[i].proc_name, ctx->init_pid); | |
99d50954 | 293 | |
39b3b69b CB |
294 | DEBUG("Attached to %s namespace of %d", |
295 | ns_info[i].proc_name, ctx->init_pid); | |
99d50954 CS |
296 | } |
297 | ||
298 | return 0; | |
299 | } | |
300 | ||
e4103cf6 | 301 | int lxc_attach_remount_sys_proc(void) |
7a0b0b56 CS |
302 | { |
303 | int ret; | |
304 | ||
305 | ret = unshare(CLONE_NEWNS); | |
ffeeed8b CB |
306 | if (ret < 0) |
307 | return log_error_errno(-1, errno, "Failed to unshare mount namespace"); | |
7a0b0b56 | 308 | |
9e61fb1f CB |
309 | if (detect_shared_rootfs() && mount(NULL, "/", NULL, MS_SLAVE | MS_REC, NULL)) |
310 | SYSERROR("Failed to recursively turn root mount tree into dependent mount. Continuing..."); | |
2c6f3fc9 | 311 | |
8ce83369 | 312 | /* Assume /proc is always mounted, so remount it. */ |
7a0b0b56 | 313 | ret = umount2("/proc", MNT_DETACH); |
ffeeed8b CB |
314 | if (ret < 0) |
315 | return log_error_errno(-1, errno, "Failed to unmount /proc"); | |
7a0b0b56 | 316 | |
657256e0 | 317 | ret = mount_filesystem("proc", "/proc", 0); |
ffeeed8b CB |
318 | if (ret < 0) |
319 | return log_error_errno(-1, errno, "Failed to remount /proc"); | |
7a0b0b56 | 320 | |
ffeeed8b CB |
321 | /* |
322 | * Try to umount /sys. If it's not a mount point, we'll get EINVAL, then | |
8ce83369 | 323 | * we ignore it because it may not have been mounted in the first place. |
7a0b0b56 CS |
324 | */ |
325 | ret = umount2("/sys", MNT_DETACH); | |
ffeeed8b CB |
326 | if (ret < 0 && errno != EINVAL) |
327 | return log_error_errno(-1, errno, "Failed to unmount /sys"); | |
328 | ||
329 | /* Remount it. */ | |
657256e0 | 330 | if (ret == 0 && mount_filesystem("sysfs", "/sys", 0)) |
ffeeed8b | 331 | return log_error_errno(-1, errno, "Failed to remount /sys"); |
7a0b0b56 CS |
332 | |
333 | return 0; | |
334 | } | |
335 | ||
677e1d27 | 336 | static int drop_capabilities(struct attach_context *ctx) |
e0732705 | 337 | { |
ffeeed8b | 338 | int last_cap; |
e0732705 | 339 | |
6f4f1937 | 340 | last_cap = lxc_caps_last_cap(); |
ffeeed8b | 341 | for (int cap = 0; cap <= last_cap; cap++) { |
e0732705 CS |
342 | if (ctx->capability_mask & (1LL << cap)) |
343 | continue; | |
344 | ||
b81689a1 | 345 | if (prctl(PR_CAPBSET_DROP, prctl_arg(cap), prctl_arg(0), |
ffeeed8b CB |
346 | prctl_arg(0), prctl_arg(0))) |
347 | return log_error_errno(-1, errno, "Failed to drop capability %d", cap); | |
ea918412 | 348 | |
94ac256f | 349 | TRACE("Dropped capability %d", cap); |
e0732705 CS |
350 | } |
351 | ||
352 | return 0; | |
353 | } | |
905022f7 | 354 | |
ab919e5f | 355 | static int lxc_attach_set_environment(struct attach_context *ctx, |
7385273f | 356 | enum lxc_attach_env_policy_t policy, |
6f4f1937 | 357 | char **extra_env, char **extra_keep) |
b3a39ba6 | 358 | { |
3d55242a | 359 | int ret; |
7385273f | 360 | struct lxc_list *iterator; |
361 | ||
799f96fd | 362 | if (policy == LXC_ATTACH_CLEAR_ENV) { |
3d5e9f48 | 363 | int path_kept = 0; |
6f4f1937 | 364 | char **extra_keep_store = NULL; |
3d5e9f48 CS |
365 | |
366 | if (extra_keep) { | |
367 | size_t count, i; | |
368 | ||
3d55242a CB |
369 | for (count = 0; extra_keep[count]; count++) |
370 | ; | |
3d5e9f48 | 371 | |
89b7bfe3 | 372 | extra_keep_store = zalloc(count * sizeof(char *)); |
3d55242a | 373 | if (!extra_keep_store) |
3d5e9f48 | 374 | return -1; |
3d55242a | 375 | |
3d5e9f48 CS |
376 | for (i = 0; i < count; i++) { |
377 | char *v = getenv(extra_keep[i]); | |
378 | if (v) { | |
379 | extra_keep_store[i] = strdup(v); | |
380 | if (!extra_keep_store[i]) { | |
3d5e9f48 CS |
381 | while (i > 0) |
382 | free(extra_keep_store[--i]); | |
ea918412 | 383 | |
3d5e9f48 CS |
384 | free(extra_keep_store); |
385 | return -1; | |
386 | } | |
3d55242a | 387 | |
3d5e9f48 CS |
388 | if (strcmp(extra_keep[i], "PATH") == 0) |
389 | path_kept = 1; | |
390 | } | |
3d5e9f48 CS |
391 | } |
392 | } | |
393 | ||
799f96fd | 394 | if (clearenv()) { |
a9cab7e3 | 395 | if (extra_keep_store) { |
3d55242a CB |
396 | char **p; |
397 | ||
a9cab7e3 CS |
398 | for (p = extra_keep_store; *p; p++) |
399 | free(*p); | |
3d55242a | 400 | |
a9cab7e3 CS |
401 | free(extra_keep_store); |
402 | } | |
3d55242a | 403 | |
ffeeed8b | 404 | return log_error(-1, "Failed to clear environment"); |
3d5e9f48 CS |
405 | } |
406 | ||
407 | if (extra_keep_store) { | |
408 | size_t i; | |
6f4f1937 | 409 | |
3d5e9f48 | 410 | for (i = 0; extra_keep[i]; i++) { |
acd4922e | 411 | if (extra_keep_store[i]) { |
3d55242a CB |
412 | ret = setenv(extra_keep[i], extra_keep_store[i], 1); |
413 | if (ret < 0) | |
a24c5678 | 414 | SYSWARN("Failed to set environment variable"); |
acd4922e | 415 | } |
ea918412 | 416 | |
3d5e9f48 CS |
417 | free(extra_keep_store[i]); |
418 | } | |
ea918412 | 419 | |
3d5e9f48 CS |
420 | free(extra_keep_store); |
421 | } | |
422 | ||
8ce83369 CB |
423 | /* Always set a default path; shells and execlp tend to be fine |
424 | * without it, but there is a disturbing number of C programs | |
425 | * out there that just assume that getenv("PATH") is never NULL | |
426 | * and then die a painful segfault death. | |
427 | */ | |
3d55242a CB |
428 | if (!path_kept) { |
429 | ret = setenv("PATH", "/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin", 1); | |
430 | if (ret < 0) | |
a24c5678 | 431 | SYSWARN("Failed to set environment variable"); |
3d55242a | 432 | } |
b3a39ba6 DW |
433 | } |
434 | ||
3d55242a | 435 | ret = putenv("container=lxc"); |
ffeeed8b | 436 | if (ret < 0) |
818a57fc | 437 | return log_warn(-1, "Failed to set environment variable"); |
b3a39ba6 | 438 | |
7385273f | 439 | /* Set container environment variables.*/ |
ab919e5f CB |
440 | if (ctx && ctx->container && ctx->container->lxc_conf) { |
441 | lxc_list_for_each(iterator, &ctx->container->lxc_conf->environment) { | |
3d55242a CB |
442 | char *env_tmp; |
443 | ||
444 | env_tmp = strdup((char *)iterator->elem); | |
445 | if (!env_tmp) | |
7385273f | 446 | return -1; |
7385273f | 447 | |
3d55242a | 448 | ret = putenv(env_tmp); |
ffeeed8b CB |
449 | if (ret < 0) |
450 | return log_error_errno(-1, errno, "Failed to set environment variable: %s", (char *)iterator->elem); | |
7385273f | 451 | } |
452 | } | |
453 | ||
8ce83369 | 454 | /* Set extra environment variables. */ |
3d5e9f48 CS |
455 | if (extra_env) { |
456 | for (; *extra_env; extra_env++) { | |
3d55242a | 457 | char *p; |
ea918412 | 458 | |
8ce83369 CB |
459 | /* We just assume the user knows what they are doing, so |
460 | * we don't do any checks. | |
461 | */ | |
3d55242a CB |
462 | p = strdup(*extra_env); |
463 | if (!p) | |
3d5e9f48 | 464 | return -1; |
3d55242a CB |
465 | |
466 | ret = putenv(p); | |
467 | if (ret < 0) | |
a24c5678 | 468 | SYSWARN("Failed to set environment variable"); |
3d5e9f48 CS |
469 | } |
470 | } | |
471 | ||
b3a39ba6 DW |
472 | return 0; |
473 | } | |
474 | ||
74a3920a | 475 | static char *lxc_attach_getpwshell(uid_t uid) |
905022f7 | 476 | { |
1b9c9f5b | 477 | __do_free char *line = NULL, *result = NULL; |
cd8f5663 | 478 | __do_fclose FILE *pipe_f = NULL; |
6f4f1937 | 479 | int fd, ret; |
905022f7 CS |
480 | pid_t pid; |
481 | int pipes[2]; | |
3fa23ac3 CB |
482 | bool found = false; |
483 | size_t line_bufsz = 0; | |
905022f7 | 484 | |
8ce83369 CB |
485 | /* We need to fork off a process that runs the getent program, and we |
486 | * need to capture its output, so we use a pipe for that purpose. | |
905022f7 | 487 | */ |
3fa23ac3 | 488 | ret = pipe2(pipes, O_CLOEXEC); |
905022f7 CS |
489 | if (ret < 0) |
490 | return NULL; | |
491 | ||
492 | pid = fork(); | |
493 | if (pid < 0) { | |
494 | close(pipes[0]); | |
495 | close(pipes[1]); | |
496 | return NULL; | |
497 | } | |
498 | ||
3fa23ac3 | 499 | if (!pid) { |
905022f7 CS |
500 | char uid_buf[32]; |
501 | char *arguments[] = { | |
502 | "getent", | |
503 | "passwd", | |
504 | uid_buf, | |
505 | NULL | |
506 | }; | |
507 | ||
508 | close(pipes[0]); | |
509 | ||
8ce83369 | 510 | /* We want to capture stdout. */ |
3fa23ac3 | 511 | ret = dup2(pipes[1], STDOUT_FILENO); |
905022f7 | 512 | close(pipes[1]); |
3fa23ac3 | 513 | if (ret < 0) |
ea918412 | 514 | _exit(EXIT_FAILURE); |
905022f7 | 515 | |
8ce83369 CB |
516 | /* Get rid of stdin/stderr, so we try to associate it with |
517 | * /dev/null. | |
905022f7 | 518 | */ |
3fa23ac3 | 519 | fd = open_devnull(); |
905022f7 | 520 | if (fd < 0) { |
3fa23ac3 CB |
521 | close(STDIN_FILENO); |
522 | close(STDERR_FILENO); | |
905022f7 | 523 | } else { |
3fa23ac3 | 524 | (void)dup3(fd, STDIN_FILENO, O_CLOEXEC); |
59f0e209 | 525 | (void)dup3(fd, STDERR_FILENO, O_CLOEXEC); |
905022f7 CS |
526 | close(fd); |
527 | } | |
528 | ||
8ce83369 | 529 | /* Finish argument list. */ |
3fa23ac3 CB |
530 | ret = snprintf(uid_buf, sizeof(uid_buf), "%ld", (long)uid); |
531 | if (ret <= 0 || ret >= sizeof(uid_buf)) | |
ea918412 | 532 | _exit(EXIT_FAILURE); |
905022f7 | 533 | |
8ce83369 | 534 | /* Try to run getent program. */ |
3fa23ac3 | 535 | (void)execvp("getent", arguments); |
ea918412 | 536 | _exit(EXIT_FAILURE); |
905022f7 | 537 | } |
3fa23ac3 CB |
538 | |
539 | close(pipes[1]); | |
540 | ||
4110345b | 541 | pipe_f = fdopen(pipes[0], "re"); |
cf4026f1 CB |
542 | if (!pipe_f) { |
543 | close(pipes[0]); | |
544 | goto reap_child; | |
545 | } | |
546 | /* Transfer ownership of pipes[0] to pipe_f. */ | |
547 | move_fd(pipes[0]); | |
548 | ||
3fa23ac3 CB |
549 | while (getline(&line, &line_bufsz, pipe_f) != -1) { |
550 | int i; | |
551 | long value; | |
552 | char *token; | |
553 | char *endptr = NULL, *saveptr = NULL; | |
554 | ||
555 | /* If we already found something, just continue to read | |
556 | * until the pipe doesn't deliver any more data, but | |
557 | * don't modify the existing data structure. | |
558 | */ | |
559 | if (found) | |
560 | continue; | |
561 | ||
18d4ffde | 562 | if (!line) |
563 | continue; | |
564 | ||
3fa23ac3 CB |
565 | /* Trim line on the right hand side. */ |
566 | for (i = strlen(line); i > 0 && (line[i - 1] == '\n' || line[i - 1] == '\r'); --i) | |
567 | line[i - 1] = '\0'; | |
568 | ||
569 | /* Split into tokens: first: user name. */ | |
570 | token = strtok_r(line, ":", &saveptr); | |
571 | if (!token) | |
572 | continue; | |
573 | ||
574 | /* next: dummy password field */ | |
575 | token = strtok_r(NULL, ":", &saveptr); | |
576 | if (!token) | |
577 | continue; | |
578 | ||
579 | /* next: user id */ | |
580 | token = strtok_r(NULL, ":", &saveptr); | |
581 | value = token ? strtol(token, &endptr, 10) : 0; | |
582 | if (!token || !endptr || *endptr || value == LONG_MIN || | |
ea918412 | 583 | value == LONG_MAX) |
3fa23ac3 CB |
584 | continue; |
585 | ||
586 | /* dummy sanity check: user id matches */ | |
587 | if ((uid_t)value != uid) | |
588 | continue; | |
589 | ||
590 | /* skip fields: gid, gecos, dir, go to next field 'shell' */ | |
591 | for (i = 0; i < 4; i++) { | |
592 | token = strtok_r(NULL, ":", &saveptr); | |
593 | if (!token) | |
594 | continue; | |
595 | } | |
ea918412 | 596 | |
3fa23ac3 CB |
597 | if (!token) |
598 | continue; | |
ea918412 | 599 | |
1b9c9f5b | 600 | free_disarm(result); |
3fa23ac3 CB |
601 | result = strdup(token); |
602 | ||
603 | /* Sanity check that there are no fields after that. */ | |
604 | token = strtok_r(NULL, ":", &saveptr); | |
605 | if (token) | |
606 | continue; | |
607 | ||
608 | found = true; | |
609 | } | |
ea918412 | 610 | |
cf4026f1 | 611 | reap_child: |
3fa23ac3 | 612 | ret = wait_for_pid(pid); |
1b9c9f5b | 613 | if (ret < 0) |
3fa23ac3 | 614 | return NULL; |
3fa23ac3 | 615 | |
1b9c9f5b | 616 | if (!found) |
3fa23ac3 | 617 | return NULL; |
3fa23ac3 | 618 | |
1b9c9f5b | 619 | return move_ptr(result); |
905022f7 | 620 | } |
cb3e61fa | 621 | |
6f4f1937 | 622 | static void lxc_attach_get_init_uidgid(uid_t *init_uid, gid_t *init_gid) |
cb3e61fa | 623 | { |
cd8f5663 CB |
624 | __do_free char *line = NULL; |
625 | __do_fclose FILE *proc_file = NULL; | |
604ca1c0 | 626 | char proc_fn[LXC_PROC_STATUS_LEN]; |
8ce83369 | 627 | int ret; |
cb3e61fa | 628 | size_t line_bufsz = 0; |
cb3e61fa | 629 | long value = -1; |
a5d657d3 CB |
630 | uid_t uid = LXC_INVALID_UID; |
631 | gid_t gid = LXC_INVALID_GID; | |
cb3e61fa | 632 | |
604ca1c0 CB |
633 | ret = snprintf(proc_fn, LXC_PROC_STATUS_LEN, "/proc/%d/status", 1); |
634 | if (ret < 0 || ret >= LXC_PROC_STATUS_LEN) | |
7fb45c93 | 635 | return; |
cb3e61fa | 636 | |
4110345b | 637 | proc_file = fopen(proc_fn, "re"); |
cb3e61fa CS |
638 | if (!proc_file) |
639 | return; | |
640 | ||
641 | while (getline(&line, &line_bufsz, proc_file) != -1) { | |
8ce83369 CB |
642 | /* Format is: real, effective, saved set user, fs we only care |
643 | * about real uid. | |
cb3e61fa CS |
644 | */ |
645 | ret = sscanf(line, "Uid: %ld", &value); | |
8ce83369 | 646 | if (ret != EOF && ret == 1) { |
6f4f1937 | 647 | uid = (uid_t)value; |
cb3e61fa CS |
648 | } else { |
649 | ret = sscanf(line, "Gid: %ld", &value); | |
8ce83369 | 650 | if (ret != EOF && ret == 1) |
6f4f1937 | 651 | gid = (gid_t)value; |
cb3e61fa | 652 | } |
ea918412 | 653 | |
a5d657d3 | 654 | if (uid != LXC_INVALID_UID && gid != LXC_INVALID_GID) |
cb3e61fa CS |
655 | break; |
656 | } | |
657 | ||
8ce83369 | 658 | /* Only override arguments if we found something. */ |
a5d657d3 | 659 | if (uid != LXC_INVALID_UID) |
cb3e61fa | 660 | *init_uid = uid; |
ea918412 | 661 | |
a5d657d3 | 662 | if (gid != LXC_INVALID_GID) |
cb3e61fa CS |
663 | *init_gid = gid; |
664 | ||
665 | /* TODO: we should also parse supplementary groups and use | |
8ce83369 CB |
666 | * setgroups() to set them. |
667 | */ | |
cb3e61fa | 668 | } |
9c4693b8 | 669 | |
d4db3d14 | 670 | static bool fetch_seccomp(struct lxc_container *c, lxc_attach_options_t *options) |
2c4ea790 | 671 | { |
cd8f5663 | 672 | __do_free char *path = NULL; |
d4db3d14 CB |
673 | int ret; |
674 | bool bret; | |
2eef2bda | 675 | |
6f4f1937 CB |
676 | if (!(options->namespaces & CLONE_NEWNS) || |
677 | !(options->attach_flags & LXC_ATTACH_LSM)) { | |
cca66e06 | 678 | free_disarm(c->lxc_conf->seccomp.seccomp); |
2c4ea790 | 679 | return true; |
bd4307f0 | 680 | } |
bd7b4e28 | 681 | |
2e812c16 | 682 | /* Remove current setting. */ |
d4db3d14 | 683 | if (!c->set_config_item(c, "lxc.seccomp.profile", "") && |
ea918412 | 684 | !c->set_config_item(c, "lxc.seccomp", "")) |
2c4ea790 | 685 | return false; |
bd7b4e28 | 686 | |
8ce83369 | 687 | /* Fetch the current profile path over the cmd interface. */ |
0b427da0 | 688 | path = c->get_running_config_item(c, "lxc.seccomp.profile"); |
bd7b4e28 | 689 | if (!path) { |
d4db3d14 | 690 | INFO("Failed to retrieve lxc.seccomp.profile"); |
ea918412 | 691 | |
0b427da0 | 692 | path = c->get_running_config_item(c, "lxc.seccomp"); |
cca66e06 CB |
693 | if (!path) |
694 | return log_info(true, "Failed to retrieve lxc.seccomp"); | |
bd7b4e28 SG |
695 | } |
696 | ||
8ce83369 | 697 | /* Copy the value into the new lxc_conf. */ |
d4db3d14 | 698 | bret = c->set_config_item(c, "lxc.seccomp.profile", path); |
d4db3d14 CB |
699 | if (!bret) |
700 | return false; | |
bd7b4e28 | 701 | |
8ce83369 | 702 | /* Attempt to parse the resulting config. */ |
d4db3d14 | 703 | ret = lxc_read_seccomp_config(c->lxc_conf); |
cca66e06 CB |
704 | if (ret < 0) |
705 | return log_error(false, "Failed to retrieve seccomp policy"); | |
2c4ea790 | 706 | |
cca66e06 | 707 | return log_info(true, "Retrieved seccomp policy"); |
2e812c16 CB |
708 | } |
709 | ||
6f4f1937 | 710 | static bool no_new_privs(struct lxc_container *c, lxc_attach_options_t *options) |
2e812c16 | 711 | { |
cd8f5663 | 712 | __do_free char *val = NULL; |
2e812c16 | 713 | |
2e812c16 | 714 | /* Remove current setting. */ |
02d3b72b CB |
715 | if (!c->set_config_item(c, "lxc.no_new_privs", "")) |
716 | return log_info(false, "Failed to unset lxc.no_new_privs"); | |
2e812c16 CB |
717 | |
718 | /* Retrieve currently active setting. */ | |
719 | val = c->get_running_config_item(c, "lxc.no_new_privs"); | |
02d3b72b CB |
720 | if (!val) |
721 | return log_info(false, "Failed to retrieve lxc.no_new_privs"); | |
2e812c16 CB |
722 | |
723 | /* Set currently active setting. */ | |
cd8f5663 | 724 | return c->set_config_item(c, "lxc.no_new_privs", val); |
2c4ea790 SH |
725 | } |
726 | ||
338b230f | 727 | struct attach_payload { |
a998454a | 728 | int ipc_socket; |
cecf3e83 | 729 | int terminal_pts_fd; |
a998454a | 730 | lxc_attach_options_t *options; |
ab919e5f | 731 | struct attach_context *ctx; |
a998454a CB |
732 | lxc_attach_exec_t exec_function; |
733 | void *exec_payload; | |
734 | }; | |
735 | ||
338b230f | 736 | static void put_attach_payload(struct attach_payload *p) |
ba2be1a8 | 737 | { |
81102768 | 738 | close_prot_errno_disarm(p->ipc_socket); |
cecf3e83 | 739 | close_prot_errno_disarm(p->terminal_pts_fd); |
ab919e5f | 740 | if (p->ctx) { |
dd53c8af | 741 | put_attach_context(p->ctx); |
ab919e5f | 742 | p->ctx = NULL; |
b21da190 | 743 | } |
ba2be1a8 CB |
744 | } |
745 | ||
338b230f | 746 | __noreturn static void do_attach(struct attach_payload *ap) |
a998454a | 747 | { |
427a8067 | 748 | int lsm_fd, ret; |
a998454a CB |
749 | uid_t new_uid; |
750 | gid_t new_gid; | |
936efc72 CB |
751 | uid_t ns_root_uid = 0; |
752 | gid_t ns_root_gid = 0; | |
338b230f CB |
753 | lxc_attach_options_t* options = ap->options; |
754 | struct attach_context *ctx = ap->ctx; | |
ab919e5f | 755 | struct lxc_conf *conf = ctx->container->lxc_conf; |
57de839f CB |
756 | bool needs_lsm = (options->namespaces & CLONE_NEWNS) && |
757 | (options->attach_flags & LXC_ATTACH_LSM) && | |
ab919e5f | 758 | ctx->lsm_label; |
8455e39e | 759 | char *lsm_label = NULL; |
a998454a CB |
760 | |
761 | /* A description of the purpose of this functionality is provided in the | |
762 | * lxc-attach(1) manual page. We have to remount here and not in the | |
763 | * parent process, otherwise /proc may not properly reflect the new pid | |
764 | * namespace. | |
765 | */ | |
766 | if (!(options->namespaces & CLONE_NEWNS) && | |
767 | (options->attach_flags & LXC_ATTACH_REMOUNT_PROC_SYS)) { | |
768 | ret = lxc_attach_remount_sys_proc(); | |
b75c344c CB |
769 | if (ret < 0) |
770 | goto on_error; | |
ea918412 | 771 | |
b75c344c | 772 | TRACE("Remounted \"/proc\" and \"/sys\""); |
a998454a CB |
773 | } |
774 | ||
5b514ce3 | 775 | /* Now perform additional attachments. */ |
a998454a | 776 | #if HAVE_SYS_PERSONALITY_H |
a998454a | 777 | if (options->attach_flags & LXC_ATTACH_SET_PERSONALITY) { |
b75c344c CB |
778 | long new_personality; |
779 | ||
780 | if (options->personality < 0) | |
ab919e5f | 781 | new_personality = ctx->personality; |
b75c344c CB |
782 | else |
783 | new_personality = options->personality; | |
ea918412 | 784 | |
ee142207 CB |
785 | if (new_personality != LXC_ARCH_UNCHANGED) { |
786 | ret = personality(new_personality); | |
787 | if (ret < 0) | |
788 | goto on_error; | |
ea918412 | 789 | |
ee142207 CB |
790 | TRACE("Set new personality"); |
791 | } | |
a998454a CB |
792 | } |
793 | #endif | |
794 | ||
795 | if (options->attach_flags & LXC_ATTACH_DROP_CAPABILITIES) { | |
677e1d27 | 796 | ret = drop_capabilities(ctx); |
b75c344c CB |
797 | if (ret < 0) |
798 | goto on_error; | |
ea918412 | 799 | |
b75c344c | 800 | TRACE("Dropped capabilities"); |
a998454a CB |
801 | } |
802 | ||
803 | /* Always set the environment (specify (LXC_ATTACH_KEEP_ENV, NULL, NULL) | |
804 | * if you want this to be a no-op). | |
805 | */ | |
ab919e5f | 806 | ret = lxc_attach_set_environment(ctx, |
7385273f | 807 | options->env_policy, |
a998454a CB |
808 | options->extra_env_vars, |
809 | options->extra_keep_env); | |
b75c344c CB |
810 | if (ret < 0) |
811 | goto on_error; | |
ea918412 | 812 | |
b75c344c | 813 | TRACE("Set up environment"); |
a998454a | 814 | |
57de839f CB |
815 | /* This remark only affects fully unprivileged containers: |
816 | * Receive fd for LSM security module before we set{g,u}id(). The reason | |
817 | * is that on set{g,u}id() the kernel will a) make us undumpable and b) | |
818 | * we will change our effective uid. This means our effective uid will | |
819 | * be different from the effective uid of the process that created us | |
820 | * which means that this processs no longer has capabilities in our | |
821 | * namespace including CAP_SYS_PTRACE. This means we will not be able to | |
822 | * read and /proc/<pid> files for the process anymore when /proc is | |
823 | * mounted with hidepid={1,2}. So let's get the lsm label fd before the | |
824 | * set{g,u}id(). | |
825 | */ | |
826 | if (needs_lsm) { | |
f8e88e94 | 827 | if (!sync_wait_fd(ap->ipc_socket, ATTACH_SYNC_LSM(&lsm_fd))) { |
6e36c297 | 828 | SYSERROR("Failed to receive lsm label fd"); |
b75c344c | 829 | goto on_error; |
9044b79e | 830 | } |
831 | ||
57de839f CB |
832 | TRACE("Received LSM label file descriptor %d from parent", lsm_fd); |
833 | } | |
834 | ||
08ea9270 | 835 | if (options->stdin_fd > 0 && isatty(options->stdin_fd)) { |
cd0a2b2f | 836 | ret = lxc_make_controlling_terminal(options->stdin_fd); |
08ea9270 CB |
837 | if (ret < 0) |
838 | goto on_error; | |
839 | } | |
840 | ||
b58214ac CB |
841 | if (!lxc_setgroups(0, NULL) && errno != EPERM) |
842 | goto on_error; | |
843 | ||
936efc72 CB |
844 | if (options->namespaces & CLONE_NEWUSER) { |
845 | /* Check whether nsuid 0 has a mapping. */ | |
846 | ns_root_uid = get_ns_uid(0); | |
ea918412 | 847 | |
936efc72 CB |
848 | /* Check whether nsgid 0 has a mapping. */ |
849 | ns_root_gid = get_ns_gid(0); | |
a998454a | 850 | |
936efc72 CB |
851 | /* If there's no mapping for nsuid 0 try to retrieve the nsuid |
852 | * init was started with. | |
853 | */ | |
854 | if (ns_root_uid == LXC_INVALID_UID) | |
855 | lxc_attach_get_init_uidgid(&ns_root_uid, &ns_root_gid); | |
ea918412 | 856 | |
936efc72 CB |
857 | if (ns_root_uid == LXC_INVALID_UID) |
858 | goto on_error; | |
a998454a | 859 | |
464c4611 | 860 | if (!lxc_switch_uid_gid(ns_root_uid, ns_root_gid)) |
b75c344c | 861 | goto on_error; |
a998454a CB |
862 | } |
863 | ||
936efc72 CB |
864 | /* Set {u,g}id. */ |
865 | if (options->uid != LXC_INVALID_UID) | |
866 | new_uid = options->uid; | |
867 | else | |
868 | new_uid = ns_root_uid; | |
869 | ||
870 | if (options->gid != LXC_INVALID_GID) | |
871 | new_gid = options->gid; | |
872 | else | |
873 | new_gid = ns_root_gid; | |
874 | ||
57de839f | 875 | if (needs_lsm) { |
d3ba7c98 | 876 | bool on_exec; |
a998454a CB |
877 | |
878 | /* Change into our new LSM profile. */ | |
d3ba7c98 | 879 | on_exec = options->attach_flags & LXC_ATTACH_LSM_EXEC ? true : false; |
65129087 CB |
880 | if (options->attach_flags & LXC_ATTACH_LSM_LABEL) |
881 | lsm_label = options->lsm_label; | |
882 | if (!lsm_label) | |
ab919e5f CB |
883 | lsm_label = ctx->lsm_label; |
884 | ret = ctx->lsm_ops->process_label_set_at(ctx->lsm_ops, lsm_fd, | |
885 | lsm_label, on_exec); | |
cb2420df | 886 | close_prot_errno_disarm(lsm_fd); |
b75c344c CB |
887 | if (ret < 0) |
888 | goto on_error; | |
ea918412 | 889 | |
ab919e5f | 890 | TRACE("Set %s LSM label to \"%s\"", ctx->lsm_ops->name, ctx->lsm_label); |
a998454a CB |
891 | } |
892 | ||
ab919e5f | 893 | if ((ctx->container && conf && conf->no_new_privs) || |
6ce8e678 AL |
894 | (options->attach_flags & LXC_ATTACH_NO_NEW_PRIVS)) { |
895 | ret = prctl(PR_SET_NO_NEW_PRIVS, prctl_arg(1), prctl_arg(0), | |
896 | prctl_arg(0), prctl_arg(0)); | |
897 | if (ret < 0) | |
898 | goto on_error; | |
899 | ||
900 | TRACE("Set PR_SET_NO_NEW_PRIVS"); | |
901 | } | |
902 | ||
ab919e5f | 903 | if (ctx->container && conf && conf->seccomp.seccomp) { |
cdb2a47f | 904 | ret = lxc_seccomp_load(conf); |
b75c344c CB |
905 | if (ret < 0) |
906 | goto on_error; | |
ea918412 | 907 | |
b75c344c | 908 | TRACE("Loaded seccomp profile"); |
cdb2a47f | 909 | |
338b230f | 910 | ret = lxc_seccomp_send_notifier_fd(&conf->seccomp, ap->ipc_socket); |
c3e3c21a CB |
911 | if (ret < 0) |
912 | goto on_error; | |
a998454a | 913 | } |
ea918412 | 914 | |
338b230f | 915 | close_prot_errno_disarm(ap->ipc_socket); |
dd53c8af | 916 | put_attach_context(ctx); |
338b230f | 917 | ap->ctx = NULL; |
a998454a CB |
918 | |
919 | /* The following is done after the communication socket is shut down. | |
920 | * That way, all errors that might (though unlikely) occur up until this | |
921 | * point will have their messages printed to the original stderr (if | |
922 | * logging is so configured) and not the fd the user supplied, if any. | |
923 | */ | |
924 | ||
925 | /* Fd handling for stdin, stdout and stderr; ignore errors here, user | |
926 | * may want to make sure the fds are closed, for example. | |
927 | */ | |
08ea9270 | 928 | if (options->stdin_fd >= 0 && options->stdin_fd != STDIN_FILENO) |
b4959848 CB |
929 | if (dup2(options->stdin_fd, STDIN_FILENO)) |
930 | DEBUG("Failed to replace stdin with %d", options->stdin_fd); | |
08ea9270 CB |
931 | |
932 | if (options->stdout_fd >= 0 && options->stdout_fd != STDOUT_FILENO) | |
b4959848 CB |
933 | if (dup2(options->stdout_fd, STDOUT_FILENO)) |
934 | DEBUG("Failed to replace stdout with %d", options->stdin_fd); | |
08ea9270 CB |
935 | |
936 | if (options->stderr_fd >= 0 && options->stderr_fd != STDERR_FILENO) | |
b4959848 CB |
937 | if (dup2(options->stderr_fd, STDERR_FILENO)) |
938 | DEBUG("Failed to replace stderr with %d", options->stdin_fd); | |
a998454a CB |
939 | |
940 | /* close the old fds */ | |
08ea9270 | 941 | if (options->stdin_fd > STDERR_FILENO) |
a998454a | 942 | close(options->stdin_fd); |
08ea9270 CB |
943 | |
944 | if (options->stdout_fd > STDERR_FILENO) | |
a998454a | 945 | close(options->stdout_fd); |
08ea9270 CB |
946 | |
947 | if (options->stderr_fd > STDERR_FILENO) | |
a998454a CB |
948 | close(options->stderr_fd); |
949 | ||
427a8067 CB |
950 | /* |
951 | * Try to remove FD_CLOEXEC flag from stdin/stdout/stderr, but also | |
a998454a CB |
952 | * here, ignore errors. |
953 | */ | |
427a8067 | 954 | for (int fd = STDIN_FILENO; fd <= STDERR_FILENO; fd++) { |
3f62938a | 955 | ret = fd_cloexec(fd, false); |
b75c344c CB |
956 | if (ret < 0) { |
957 | SYSERROR("Failed to clear FD_CLOEXEC from file descriptor %d", fd); | |
958 | goto on_error; | |
959 | } | |
a998454a CB |
960 | } |
961 | ||
9e84479f | 962 | if (options->attach_flags & LXC_ATTACH_TERMINAL) { |
338b230f | 963 | ret = lxc_terminal_prepare_login(ap->terminal_pts_fd); |
ba2be1a8 | 964 | if (ret < 0) { |
338b230f | 965 | SYSERROR("Failed to prepare terminal file descriptor %d", ap->terminal_pts_fd); |
ba2be1a8 CB |
966 | goto on_error; |
967 | } | |
ea918412 | 968 | |
338b230f | 969 | TRACE("Prepared terminal file descriptor %d", ap->terminal_pts_fd); |
ba2be1a8 CB |
970 | } |
971 | ||
936efc72 CB |
972 | /* Avoid unnecessary syscalls. */ |
973 | if (new_uid == ns_root_uid) | |
974 | new_uid = LXC_INVALID_UID; | |
975 | ||
976 | if (new_gid == ns_root_gid) | |
977 | new_gid = LXC_INVALID_GID; | |
c353b0b9 | 978 | |
6aff5157 | 979 | /* Make sure that the processes STDIO is correctly owned by the user that we are switching to */ |
c353b0b9 CB |
980 | ret = fix_stdio_permissions(new_uid); |
981 | if (ret) | |
a2c26bef | 982 | INFO("Failed to adjust stdio permissions"); |
936efc72 | 983 | |
464c4611 | 984 | if (!lxc_switch_uid_gid(new_uid, new_gid)) |
936efc72 CB |
985 | goto on_error; |
986 | ||
a998454a | 987 | /* We're done, so we can now do whatever the user intended us to do. */ |
338b230f | 988 | _exit(ap->exec_function(ap->exec_payload)); |
b75c344c CB |
989 | |
990 | on_error: | |
338b230f | 991 | put_attach_payload(ap); |
dab02267 | 992 | ERROR("Failed to attach to container"); |
c7ac2e1c | 993 | _exit(EXIT_FAILURE); |
a998454a CB |
994 | } |
995 | ||
f797f05e | 996 | static int lxc_attach_terminal(const char *name, const char *lxcpath, struct lxc_conf *conf, |
9e84479f | 997 | struct lxc_terminal *terminal) |
ba2be1a8 CB |
998 | { |
999 | int ret; | |
1000 | ||
9e84479f | 1001 | lxc_terminal_init(terminal); |
ba2be1a8 | 1002 | |
8ea93a0f | 1003 | ret = lxc_terminal_create(name, lxcpath, conf, terminal); |
c2af3a15 CB |
1004 | if (ret < 0) |
1005 | return log_error(-1, "Failed to create terminal"); | |
ba2be1a8 | 1006 | |
ba2be1a8 | 1007 | return 0; |
ba2be1a8 CB |
1008 | } |
1009 | ||
9e84479f CB |
1010 | static int lxc_attach_terminal_mainloop_init(struct lxc_terminal *terminal, |
1011 | struct lxc_epoll_descr *descr) | |
ba2be1a8 CB |
1012 | { |
1013 | int ret; | |
1014 | ||
1015 | ret = lxc_mainloop_open(descr); | |
c2af3a15 CB |
1016 | if (ret < 0) |
1017 | return log_error(-1, "Failed to create mainloop"); | |
ba2be1a8 | 1018 | |
9e84479f | 1019 | ret = lxc_terminal_mainloop_add(descr, terminal); |
ba2be1a8 | 1020 | if (ret < 0) { |
ba2be1a8 | 1021 | lxc_mainloop_close(descr); |
c2af3a15 | 1022 | return log_error(-1, "Failed to add handlers to mainloop"); |
ba2be1a8 CB |
1023 | } |
1024 | ||
1025 | return 0; | |
1026 | } | |
1027 | ||
36a94ce8 | 1028 | static inline void lxc_attach_terminal_close_ptx(struct lxc_terminal *terminal) |
ba2be1a8 | 1029 | { |
36a94ce8 | 1030 | close_prot_errno_disarm(terminal->ptx); |
ba2be1a8 CB |
1031 | } |
1032 | ||
cecf3e83 | 1033 | static inline void lxc_attach_terminal_close_pts(struct lxc_terminal *terminal) |
ba2be1a8 | 1034 | { |
41808e20 | 1035 | close_prot_errno_disarm(terminal->pty); |
ba2be1a8 CB |
1036 | } |
1037 | ||
9e84479f | 1038 | static inline void lxc_attach_terminal_close_peer(struct lxc_terminal *terminal) |
ba2be1a8 | 1039 | { |
19a3e906 | 1040 | close_prot_errno_disarm(terminal->peer); |
ba2be1a8 CB |
1041 | } |
1042 | ||
9e84479f | 1043 | static inline void lxc_attach_terminal_close_log(struct lxc_terminal *terminal) |
ba2be1a8 | 1044 | { |
19a3e906 | 1045 | close_prot_errno_disarm(terminal->log_fd); |
ba2be1a8 CB |
1046 | } |
1047 | ||
908fbc1a CB |
1048 | int lxc_attach(struct lxc_container *container, lxc_attach_exec_t exec_function, |
1049 | void *exec_payload, lxc_attach_options_t *options, | |
1050 | pid_t *attached_process) | |
9c4693b8 | 1051 | { |
6f9fe5d0 | 1052 | int ret_parent = -1; |
6f9fe5d0 | 1053 | struct lxc_epoll_descr descr = {}; |
a9f0cecf | 1054 | int ret; |
26abd7ea | 1055 | char *name, *lxcpath; |
9c4693b8 | 1056 | int ipc_sockets[2]; |
500ed813 | 1057 | pid_t attached_pid, pid, to_cleanup_pid; |
ab919e5f | 1058 | struct attach_context *ctx; |
9e84479f | 1059 | struct lxc_terminal terminal; |
1cce35e6 | 1060 | struct lxc_conf *conf; |
9c4693b8 | 1061 | |
908fbc1a | 1062 | if (!container) |
540a2f70 | 1063 | return ret_set_errno(-1, EINVAL); |
908fbc1a CB |
1064 | |
1065 | if (!lxc_container_get(container)) | |
540a2f70 | 1066 | return ret_set_errno(-1, EINVAL); |
908fbc1a CB |
1067 | |
1068 | name = container->name; | |
1069 | lxcpath = container->config_path; | |
1070 | ||
9c4693b8 CS |
1071 | if (!options) |
1072 | options = &attach_static_default_options; | |
1073 | ||
9745eb8a | 1074 | ctx = alloc_attach_context(); |
ab919e5f | 1075 | if (!ctx) { |
9745eb8a CB |
1076 | lxc_container_put(container); |
1077 | return log_error_errno(-ENOMEM, ENOMEM, "Failed to allocate attach context"); | |
1078 | } | |
1079 | ||
500ed813 | 1080 | ret = get_attach_context(ctx, container); |
9745eb8a | 1081 | if (ret) { |
7e995801 | 1082 | put_attach_context(ctx); |
74ce42b5 | 1083 | return log_error(-1, "Failed to get attach context"); |
9c4693b8 CS |
1084 | } |
1085 | ||
ab919e5f | 1086 | conf = ctx->container->lxc_conf; |
ba773996 | 1087 | |
ab919e5f | 1088 | if (!fetch_seccomp(ctx->container, options)) |
ae026f55 | 1089 | WARN("Failed to get seccomp policy"); |
2c4ea790 | 1090 | |
ab919e5f | 1091 | if (!no_new_privs(ctx->container, options)) |
ae026f55 | 1092 | WARN("Could not determine whether PR_SET_NO_NEW_PRIVS is set"); |
2e812c16 | 1093 | |
74ce42b5 | 1094 | /* Determine which namespaces the container was created with. */ |
9c4693b8 CS |
1095 | if (options->namespaces == -1) { |
1096 | options->namespaces = lxc_cmd_get_clone_flags(name, lxcpath); | |
9c4693b8 | 1097 | if (options->namespaces == -1) { |
dd53c8af | 1098 | put_attach_context(ctx); |
74ce42b5 | 1099 | return log_error(-1, "Failed to automatically determine the namespaces which the container uses"); |
9c4693b8 | 1100 | } |
877f3a04 | 1101 | |
74ce42b5 | 1102 | for (int i = 0; i < LXC_NS_MAX; i++) { |
877f3a04 CB |
1103 | if (ns_info[i].clone_flag & CLONE_NEWCGROUP) |
1104 | if (!(options->attach_flags & LXC_ATTACH_MOVE_TO_CGROUP) || | |
1105 | !cgns_supported()) | |
1106 | continue; | |
1107 | ||
1108 | if (ns_info[i].clone_flag & options->namespaces) | |
1109 | continue; | |
1110 | ||
ab919e5f | 1111 | ctx->ns_inherited |= ns_info[i].clone_flag; |
877f3a04 CB |
1112 | } |
1113 | } | |
1114 | ||
b7873c95 CB |
1115 | ret = get_attach_context_nsfds(ctx, options); |
1116 | if (ret) { | |
b7873c95 | 1117 | lxc_container_put(container); |
74ce42b5 | 1118 | return log_error(-1, "Failed to get namespace file descriptors"); |
9c4693b8 CS |
1119 | } |
1120 | ||
9e84479f | 1121 | if (options->attach_flags & LXC_ATTACH_TERMINAL) { |
f797f05e | 1122 | ret = lxc_attach_terminal(name, lxcpath, conf, &terminal); |
ba2be1a8 | 1123 | if (ret < 0) { |
dd53c8af | 1124 | put_attach_context(ctx); |
74ce42b5 | 1125 | return log_error(-1, "Failed to setup new terminal"); |
ba2be1a8 CB |
1126 | } |
1127 | ||
9e84479f | 1128 | terminal.log_fd = options->log_fd; |
c948657b | 1129 | } else { |
9e84479f | 1130 | lxc_terminal_init(&terminal); |
ba2be1a8 CB |
1131 | } |
1132 | ||
8ce83369 CB |
1133 | /* Create a socket pair for IPC communication; set SOCK_CLOEXEC in order |
1134 | * to make sure we don't irritate other threads that want to fork+exec | |
1135 | * away | |
9c4693b8 CS |
1136 | * |
1137 | * IMPORTANT: if the initial process is multithreaded and another call | |
1138 | * just fork()s away without exec'ing directly after, the socket fd will | |
1139 | * exist in the forked process from the other thread and any close() in | |
8ce83369 | 1140 | * our own child process will not really cause the socket to close |
1d801260 | 1141 | * properly, potentially causing the parent to hang. |
9c4693b8 CS |
1142 | * |
1143 | * For this reason, while IPC is still active, we have to use shutdown() | |
8ce83369 CB |
1144 | * if the child exits prematurely in order to signal that the socket is |
1145 | * closed and cannot assume that the child exiting will automatically do | |
1146 | * that. | |
9c4693b8 CS |
1147 | * |
1148 | * IPC mechanism: (X is receiver) | |
1149 | * initial process intermediate attached | |
1150 | * X <--- send pid of | |
1151 | * attached proc, | |
1152 | * then exit | |
1153 | * send 0 ------------------------------------> X | |
1154 | * [do initialization] | |
1155 | * X <------------------------------------ send 1 | |
1156 | * [add to cgroup, ...] | |
1157 | * send 2 ------------------------------------> X | |
81f466d0 CB |
1158 | * [set LXC_ATTACH_NO_NEW_PRIVS] |
1159 | * X <------------------------------------ send 3 | |
1160 | * [open LSM label fd] | |
1161 | * send 4 ------------------------------------> X | |
1162 | * [set LSM label] | |
9c4693b8 CS |
1163 | * close socket close socket |
1164 | * run program | |
1165 | */ | |
1166 | ret = socketpair(PF_LOCAL, SOCK_STREAM | SOCK_CLOEXEC, 0, ipc_sockets); | |
1167 | if (ret < 0) { | |
dd53c8af | 1168 | put_attach_context(ctx); |
74ce42b5 | 1169 | return log_error_errno(-1, errno, "Could not set up required IPC mechanism for attaching"); |
9c4693b8 CS |
1170 | } |
1171 | ||
e3f0e436 CB |
1172 | /* Create intermediate subprocess, two reasons: |
1173 | * 1. We can't setns() in the child itself, since we want to make | |
8ce83369 | 1174 | * sure we are properly attached to the pidns. |
e3f0e436 | 1175 | * 2. Also, the initial thread has to put the attached process |
8ce83369 CB |
1176 | * into the cgroup, which we can only do if we didn't already |
1177 | * setns() (otherwise, user namespaces will hate us). | |
9c4693b8 CS |
1178 | */ |
1179 | pid = fork(); | |
9c4693b8 | 1180 | if (pid < 0) { |
dd53c8af | 1181 | put_attach_context(ctx); |
74ce42b5 | 1182 | return log_error_errno(-1, errno, "Failed to create first subprocess"); |
9c4693b8 CS |
1183 | } |
1184 | ||
4f25e72f | 1185 | if (pid == 0) { |
26abd7ea | 1186 | char *cwd, *new_cwd; |
a588a482 | 1187 | |
ba2be1a8 | 1188 | /* close unneeded file descriptors */ |
4f25e72f | 1189 | close_prot_errno_disarm(ipc_sockets[0]); |
2202afc9 | 1190 | |
4f25e72f CB |
1191 | if (options->attach_flags & LXC_ATTACH_TERMINAL) { |
1192 | lxc_attach_terminal_close_ptx(&terminal); | |
1193 | lxc_attach_terminal_close_peer(&terminal); | |
1194 | lxc_attach_terminal_close_log(&terminal); | |
f4364484 SG |
1195 | } |
1196 | ||
4f25e72f | 1197 | /* Wait for the parent to have setup cgroups. */ |
6e48e7c5 | 1198 | if (!sync_wait(ipc_sockets[1], ATTACH_SYNC_CGROUP)) { |
4f25e72f | 1199 | shutdown(ipc_sockets[1], SHUT_RDWR); |
dd53c8af | 1200 | put_attach_context(ctx); |
4f25e72f | 1201 | _exit(EXIT_FAILURE); |
bb2ada6f CB |
1202 | } |
1203 | ||
4f25e72f CB |
1204 | TRACE("Intermediate process starting to initialize"); |
1205 | ||
a588a482 CB |
1206 | cwd = getcwd(NULL, 0); |
1207 | ||
4f25e72f CB |
1208 | /* Attach now, create another subprocess later, since pid namespaces |
1209 | * only really affect the children of the current process. | |
1210 | */ | |
39b3b69b | 1211 | ret = attach_nsfds(ctx); |
4f25e72f CB |
1212 | if (ret < 0) { |
1213 | ERROR("Failed to enter namespaces"); | |
1214 | shutdown(ipc_sockets[1], SHUT_RDWR); | |
dd53c8af | 1215 | put_attach_context(ctx); |
4f25e72f | 1216 | _exit(EXIT_FAILURE); |
ba2be1a8 CB |
1217 | } |
1218 | ||
4f25e72f | 1219 | /* close namespace file descriptors */ |
7e995801 | 1220 | close_nsfds(ctx); |
ea918412 | 1221 | |
4f25e72f CB |
1222 | /* Attach succeeded, try to cwd. */ |
1223 | if (options->initial_cwd) | |
1224 | new_cwd = options->initial_cwd; | |
1225 | else | |
1226 | new_cwd = cwd; | |
1227 | if (new_cwd) { | |
1228 | ret = chdir(new_cwd); | |
1229 | if (ret < 0) | |
1230 | WARN("Could not change directory to \"%s\"", new_cwd); | |
ba2be1a8 | 1231 | } |
a588a482 | 1232 | free_disarm(cwd); |
c6d09e15 | 1233 | |
4f25e72f | 1234 | /* Create attached process. */ |
4f25e72f CB |
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 | 1243 | if (pid == 0) { |
338b230f | 1244 | struct attach_payload ap = { |
a64902ab CB |
1245 | .ipc_socket = ipc_sockets[1], |
1246 | .options = options, | |
1247 | .ctx = ctx, | |
1248 | .terminal_pts_fd = terminal.pty, | |
1249 | .exec_function = exec_function, | |
1250 | .exec_payload = exec_payload, | |
1251 | }; | |
1252 | ||
4f25e72f CB |
1253 | if (options->attach_flags & LXC_ATTACH_TERMINAL) { |
1254 | ret = lxc_terminal_signal_sigmask_safe_blocked(&terminal); | |
1255 | if (ret < 0) { | |
1256 | SYSERROR("Failed to reset signal mask"); | |
1257 | _exit(EXIT_FAILURE); | |
1258 | } | |
1259 | } | |
ea918412 | 1260 | |
a64902ab | 1261 | /* Does not return. */ |
338b230f | 1262 | do_attach(&ap); |
62183f1a | 1263 | } |
2eef2bda | 1264 | |
4f25e72f CB |
1265 | if (options->attach_flags & LXC_ATTACH_TERMINAL) |
1266 | lxc_attach_terminal_close_pts(&terminal); | |
ea918412 | 1267 | |
4f25e72f | 1268 | /* Tell grandparent the pid of the pid of the newly created child. */ |
f8e88e94 | 1269 | if (!sync_wake_pid(ipc_sockets[1], ATTACH_SYNC_PID(pid))) { |
4f25e72f CB |
1270 | /* If this really happens here, this is very unfortunate, since |
1271 | * the parent will not know the pid of the attached process and | |
1272 | * will not be able to wait for it (and we won't either due to | |
1273 | * CLONE_PARENT) so the parent won't be able to reap it and the | |
1274 | * attached process will remain a zombie. | |
1275 | */ | |
1276 | shutdown(ipc_sockets[1], SHUT_RDWR); | |
dd53c8af | 1277 | put_attach_context(ctx); |
4f25e72f CB |
1278 | _exit(EXIT_FAILURE); |
1279 | } | |
9c4693b8 | 1280 | |
4f25e72f | 1281 | TRACE("Sending pid %d of attached process", pid); |
9c4693b8 | 1282 | |
4f25e72f | 1283 | /* The rest is in the hands of the initial and the attached process. */ |
dd53c8af | 1284 | put_attach_context(ctx); |
4f25e72f CB |
1285 | _exit(EXIT_SUCCESS); |
1286 | } | |
6f4f1937 | 1287 | |
4f25e72f | 1288 | to_cleanup_pid = pid; |
ea918412 | 1289 | |
4f25e72f | 1290 | /* close unneeded file descriptors */ |
cb2420df | 1291 | close_prot_errno_disarm(ipc_sockets[1]); |
7e995801 | 1292 | close_nsfds(ctx); |
4f25e72f CB |
1293 | if (options->attach_flags & LXC_ATTACH_TERMINAL) |
1294 | lxc_attach_terminal_close_pts(&terminal); | |
81f466d0 | 1295 | |
4f25e72f CB |
1296 | /* Attach to cgroup, if requested. */ |
1297 | if (options->attach_flags & LXC_ATTACH_MOVE_TO_CGROUP) { | |
1298 | /* | |
1299 | * If this is the unified hierarchy cgroup_attach() is | |
1300 | * enough. | |
1301 | */ | |
1302 | ret = cgroup_attach(conf, name, lxcpath, pid); | |
1303 | if (ret) { | |
1304 | call_cleaner(cgroup_exit) struct cgroup_ops *cgroup_ops = NULL; | |
9044b79e | 1305 | |
4f25e72f CB |
1306 | cgroup_ops = cgroup_init(conf); |
1307 | if (!cgroup_ops) | |
1308 | goto on_error; | |
9044b79e | 1309 | |
4f25e72f CB |
1310 | if (!cgroup_ops->attach(cgroup_ops, conf, name, lxcpath, pid)) |
1311 | goto on_error; | |
81f466d0 | 1312 | } |
4f25e72f CB |
1313 | TRACE("Moved intermediate process %d into container's cgroups", pid); |
1314 | } | |
81f466d0 | 1315 | |
4f25e72f CB |
1316 | /* Setup /proc limits */ |
1317 | if (!lxc_list_empty(&conf->procs)) { | |
1318 | ret = setup_proc_filesystem(&conf->procs, pid); | |
1319 | if (ret < 0) | |
1320 | goto on_error; | |
4f3b6a85 CB |
1321 | |
1322 | TRACE("Setup /proc/%d settings", pid); | |
4f25e72f | 1323 | } |
cdb2a47f | 1324 | |
4f25e72f CB |
1325 | /* Setup resource limits */ |
1326 | if (!lxc_list_empty(&conf->limits)) { | |
1327 | ret = setup_resource_limits(&conf->limits, pid); | |
1328 | if (ret < 0) | |
1329 | goto on_error; | |
4f3b6a85 CB |
1330 | |
1331 | TRACE("Setup resource limits"); | |
4f25e72f | 1332 | } |
cdb2a47f | 1333 | |
4f25e72f CB |
1334 | if (options->attach_flags & LXC_ATTACH_TERMINAL) { |
1335 | ret = lxc_attach_terminal_mainloop_init(&terminal, &descr); | |
1336 | if (ret < 0) | |
1337 | goto on_error; | |
9c4693b8 | 1338 | |
4f25e72f CB |
1339 | TRACE("Initialized terminal mainloop"); |
1340 | } | |
9c4693b8 | 1341 | |
4f25e72f | 1342 | /* Let the child process know to go ahead. */ |
6e48e7c5 | 1343 | if (!sync_wake(ipc_sockets[0], ATTACH_SYNC_CGROUP)) |
4f25e72f | 1344 | goto close_mainloop; |
ba2be1a8 | 1345 | |
4f25e72f | 1346 | TRACE("Told intermediate process to start initializing"); |
ea918412 | 1347 | |
4f25e72f | 1348 | /* Get pid of attached process from intermediate process. */ |
f8e88e94 | 1349 | if (!sync_wait_pid(ipc_sockets[0], ATTACH_SYNC_PID(&attached_pid))) |
4f25e72f | 1350 | goto close_mainloop; |
ba2be1a8 | 1351 | |
4f25e72f | 1352 | TRACE("Received pid %d of attached process in parent pid namespace", attached_pid); |
ba2be1a8 | 1353 | |
4f25e72f | 1354 | /* Ignore SIGKILL (CTRL-C) and SIGQUIT (CTRL-\) - issue #313. */ |
5d2b46fb | 1355 | if (options->stdin_fd == STDIN_FILENO) { |
4f25e72f CB |
1356 | signal(SIGINT, SIG_IGN); |
1357 | signal(SIGQUIT, SIG_IGN); | |
1358 | } | |
ba2be1a8 | 1359 | |
4f25e72f CB |
1360 | /* Reap intermediate process. */ |
1361 | ret = wait_for_pid(pid); | |
1362 | if (ret < 0) | |
1363 | goto close_mainloop; | |
ba2be1a8 | 1364 | |
4f25e72f | 1365 | TRACE("Intermediate process %d exited", pid); |
ea918412 | 1366 | |
4f25e72f CB |
1367 | /* We will always have to reap the attached process now. */ |
1368 | to_cleanup_pid = attached_pid; | |
9c4693b8 | 1369 | |
4f25e72f CB |
1370 | /* Open LSM fd and send it to child. */ |
1371 | if ((options->namespaces & CLONE_NEWNS) && | |
ab919e5f | 1372 | (options->attach_flags & LXC_ATTACH_LSM) && ctx->lsm_label) { |
ad001fb6 | 1373 | __do_close int labelfd = -EBADF; |
4f25e72f | 1374 | bool on_exec; |
ea918412 | 1375 | |
4f25e72f | 1376 | on_exec = options->attach_flags & LXC_ATTACH_LSM_EXEC ? true : false; |
ab919e5f CB |
1377 | labelfd = ctx->lsm_ops->process_label_fd_get(ctx->lsm_ops, |
1378 | attached_pid, on_exec); | |
4f25e72f CB |
1379 | if (labelfd < 0) |
1380 | goto close_mainloop; | |
9c4693b8 | 1381 | |
4f25e72f | 1382 | TRACE("Opened LSM label file descriptor %d", labelfd); |
ea918412 | 1383 | |
4f25e72f | 1384 | /* Send child fd of the LSM security module to write to. */ |
f8e88e94 | 1385 | if (!sync_wake_fd(ipc_sockets[0], ATTACH_SYNC_LSM(labelfd))) { |
6e36c297 | 1386 | SYSERROR("Failed to send lsm label fd"); |
4f25e72f CB |
1387 | goto close_mainloop; |
1388 | } | |
1389 | ||
4f25e72f | 1390 | TRACE("Sent LSM label file descriptor %d to child", labelfd); |
9c4693b8 | 1391 | } |
ea918412 | 1392 | |
4f25e72f CB |
1393 | if (conf->seccomp.seccomp) { |
1394 | ret = lxc_seccomp_recv_notifier_fd(&conf->seccomp, ipc_sockets[0]); | |
1395 | if (ret < 0) | |
1396 | goto close_mainloop; | |
9c4693b8 | 1397 | |
4f25e72f | 1398 | ret = lxc_seccomp_add_notifier(name, lxcpath, &conf->seccomp); |
d6d979bc | 1399 | if (ret < 0) |
4f25e72f | 1400 | goto close_mainloop; |
d6d979bc | 1401 | } |
9c4693b8 | 1402 | |
4f25e72f CB |
1403 | /* We're done, the child process should now execute whatever it |
1404 | * is that the user requested. The parent can now track it with | |
1405 | * waitpid() or similar. | |
1406 | */ | |
9c4693b8 | 1407 | |
4f25e72f | 1408 | *attached_process = attached_pid; |
a998454a | 1409 | |
4f25e72f CB |
1410 | /* Now shut down communication with child, we're done. */ |
1411 | shutdown(ipc_sockets[0], SHUT_RDWR); | |
cb2420df | 1412 | close_prot_errno_disarm(ipc_sockets[0]); |
f157b056 | 1413 | |
4f25e72f CB |
1414 | ret_parent = 0; |
1415 | to_cleanup_pid = -1; | |
ea918412 | 1416 | |
4f25e72f CB |
1417 | if (options->attach_flags & LXC_ATTACH_TERMINAL) { |
1418 | ret = lxc_mainloop(&descr, -1); | |
1419 | if (ret < 0) { | |
1420 | ret_parent = -1; | |
1421 | to_cleanup_pid = attached_pid; | |
1422 | } | |
a998454a | 1423 | } |
ea918412 | 1424 | |
4f25e72f | 1425 | close_mainloop: |
9e84479f | 1426 | if (options->attach_flags & LXC_ATTACH_TERMINAL) |
4f25e72f | 1427 | lxc_mainloop_close(&descr); |
9c4693b8 | 1428 | |
4f25e72f CB |
1429 | on_error: |
1430 | if (ipc_sockets[0] >= 0) { | |
1431 | shutdown(ipc_sockets[0], SHUT_RDWR); | |
cb2420df | 1432 | close_prot_errno_disarm(ipc_sockets[0]); |
9c4693b8 | 1433 | } |
ea918412 | 1434 | |
4f25e72f CB |
1435 | if (to_cleanup_pid > 0) |
1436 | (void)wait_for_pid(to_cleanup_pid); | |
1437 | ||
1438 | if (options->attach_flags & LXC_ATTACH_TERMINAL) { | |
1439 | lxc_terminal_delete(&terminal); | |
1440 | lxc_terminal_conf_free(&terminal); | |
1441 | } | |
9c4693b8 | 1442 | |
dd53c8af | 1443 | put_attach_context(ctx); |
4f25e72f | 1444 | return ret_parent; |
9c4693b8 CS |
1445 | } |
1446 | ||
06346bb0 | 1447 | int lxc_attach_run_command(void *payload) |
9c4693b8 | 1448 | { |
06346bb0 CB |
1449 | int ret = -1; |
1450 | lxc_attach_command_t *cmd = payload; | |
9c4693b8 | 1451 | |
06346bb0 CB |
1452 | ret = execvp(cmd->program, cmd->argv); |
1453 | if (ret < 0) { | |
1454 | switch (errno) { | |
1455 | case ENOEXEC: | |
1456 | ret = 126; | |
cf0fd972 | 1457 | break; |
06346bb0 CB |
1458 | case ENOENT: |
1459 | ret = 127; | |
cf0fd972 | 1460 | break; |
06346bb0 CB |
1461 | } |
1462 | } | |
ea918412 | 1463 | |
c2af3a15 | 1464 | return log_error_errno(ret, errno, "Failed to exec \"%s\"", cmd->program); |
9c4693b8 CS |
1465 | } |
1466 | ||
1467 | int lxc_attach_run_shell(void* payload) | |
1468 | { | |
cd8f5663 | 1469 | __do_free char *buf = NULL; |
9c4693b8 | 1470 | uid_t uid; |
cb7aa5e8 DJ |
1471 | struct passwd pwent; |
1472 | struct passwd *pwentp = NULL; | |
9c4693b8 | 1473 | char *user_shell; |
cb7aa5e8 DJ |
1474 | size_t bufsize; |
1475 | int ret; | |
9c4693b8 | 1476 | |
8ce83369 | 1477 | /* Ignore payload parameter. */ |
9c4693b8 CS |
1478 | (void)payload; |
1479 | ||
1480 | uid = getuid(); | |
cb7aa5e8 DJ |
1481 | |
1482 | bufsize = sysconf(_SC_GETPW_R_SIZE_MAX); | |
1483 | if (bufsize == -1) | |
1484 | bufsize = 1024; | |
1485 | ||
1486 | buf = malloc(bufsize); | |
1487 | if (buf) { | |
1488 | ret = getpwuid_r(uid, &pwent, buf, bufsize, &pwentp); | |
1489 | if (!pwentp) { | |
1490 | if (ret == 0) | |
ea918412 | 1491 | WARN("Could not find matched password record"); |
cb7aa5e8 DJ |
1492 | |
1493 | WARN("Failed to get password record - %u", uid); | |
1494 | } | |
1495 | } | |
9c4693b8 | 1496 | |
8ce83369 CB |
1497 | /* This probably happens because of incompatible nss implementations in |
1498 | * host and container (remember, this code is still using the host's | |
1499 | * glibc but our mount namespace is in the container) we may try to get | |
1500 | * the information by spawning a [getent passwd uid] process and parsing | |
1501 | * the result. | |
9c4693b8 | 1502 | */ |
cb7aa5e8 | 1503 | if (!pwentp) |
9c4693b8 CS |
1504 | user_shell = lxc_attach_getpwshell(uid); |
1505 | else | |
cb7aa5e8 | 1506 | user_shell = pwent.pw_shell; |
ea918412 | 1507 | |
9c4693b8 | 1508 | if (user_shell) |
acf47e1b | 1509 | execlp(user_shell, user_shell, (char *)NULL); |
9c4693b8 | 1510 | |
8ce83369 CB |
1511 | /* Executed if either no passwd entry or execvp fails, we will fall back |
1512 | * on /bin/sh as a default shell. | |
9c4693b8 | 1513 | */ |
acf47e1b | 1514 | execlp("/bin/sh", "/bin/sh", (char *)NULL); |
ea918412 | 1515 | |
edeb1836 | 1516 | SYSERROR("Failed to execute shell"); |
cb7aa5e8 | 1517 | if (!pwentp) |
edeb1836 | 1518 | free(user_shell); |
ea918412 | 1519 | |
9c4693b8 CS |
1520 | return -1; |
1521 | } |