]>
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 | ||
20718e39 CB |
59 | /* |
60 | * The context used to attach to the container. | |
61 | * @attach_flags : the attach flags specified in lxc_attach_options_t | |
62 | * @init_pid : the PID of the container's init process | |
63 | * @dfd_init_pid : file descriptor to /proc/@init_pid | |
64 | * __Must be closed in attach_context_security_barrier()__! | |
65 | * @dfd_self_pid : file descriptor to /proc/self | |
66 | * __Must be closed in attach_context_security_barrier()__! | |
67 | * @setup_uid : if CLONE_NEWUSER is specified will contain the uid used | |
68 | * during attach setup. | |
69 | * @setup_gid : if CLONE_NEWUSER is specified will contain the gid used | |
70 | * during attach setup. | |
71 | * @target_uid : if CLONE_NEWUSER is specified the uid that the final program | |
72 | * will be run with. | |
73 | * @target_gid : if CLONE_NEWUSER is specified the gid that the final program | |
74 | * will be run with. | |
75 | * @lsm_label : LSM label to be used for the attaching process | |
76 | * @container : the container we're attaching o | |
77 | * @personality : the personality to use for the final program | |
78 | * @capability : the capability mask of the @init_pid | |
79 | * @ns_inherited : flags of namespaces that the final program will inherit from | |
80 | * @init_pid | |
81 | * @ns_fd : file descriptors to @init_pid's namespaces | |
82 | */ | |
ab919e5f | 83 | struct attach_context { |
afc691a0 | 84 | unsigned int attach_flags; |
500ed813 | 85 | int init_pid; |
25c659d5 CB |
86 | int dfd_init_pid; |
87 | int dfd_self_pid; | |
4475fabb CB |
88 | uid_t setup_uid; |
89 | gid_t setup_gid; | |
90 | uid_t target_uid; | |
91 | gid_t target_gid; | |
0e304baa CB |
92 | char *lsm_label; |
93 | struct lxc_container *container; | |
94 | signed long personality; | |
95 | unsigned long long capability_mask; | |
96 | int ns_inherited; | |
97 | int ns_fd[LXC_NS_MAX]; | |
98 | struct lsm_ops *lsm_ops; | |
99 | }; | |
100 | ||
d8764025 CB |
101 | static pid_t pidfd_get_pid(int pidfd) |
102 | { | |
103 | __do_free char *line = NULL; | |
104 | __do_fclose FILE *f = NULL; | |
105 | size_t len = 0; | |
106 | char path[STRLITERALLEN("/proc/self/fdinfo/") + | |
107 | INTTYPE_TO_STRLEN(int) + 1 ] = "/proc/self/fdinfo/"; | |
108 | int ret; | |
109 | ||
110 | if (pidfd < 0) | |
111 | return -EBADF; | |
112 | ||
113 | ret = snprintf(path + STRLITERALLEN("/proc/self/fdinfo/"), | |
114 | INTTYPE_TO_STRLEN(int), "%d", pidfd); | |
115 | if (ret < 0 || ret > (size_t)INTTYPE_TO_STRLEN(int)) | |
116 | return ret_errno(EIO); | |
117 | ||
118 | f = fopen_cloexec(path, "re"); | |
119 | if (!f) | |
120 | return -errno; | |
121 | ||
122 | while (getline(&line, &len, f) != -1) { | |
123 | const char *prefix = "Pid:\t"; | |
124 | const size_t prefix_len = STRLITERALLEN("Pid:\t"); | |
125 | int pid = -ESRCH; | |
126 | char *slider = line; | |
127 | ||
128 | if (strncmp(slider, prefix, prefix_len)) | |
129 | continue; | |
130 | ||
131 | slider += prefix_len; | |
132 | slider = lxc_trim_whitespace_in_place(slider); | |
133 | ||
134 | ret = lxc_safe_int(slider, &pid); | |
135 | if (ret) | |
136 | return -ret; | |
137 | ||
138 | return pid; | |
139 | } | |
140 | ||
141 | return ret_errno(ENOENT); | |
142 | } | |
143 | ||
6e36c297 CB |
144 | static inline bool sync_wake_pid(int fd, pid_t pid) |
145 | { | |
146 | return lxc_write_nointr(fd, &pid, sizeof(pid_t)) == sizeof(pid_t); | |
147 | } | |
148 | ||
149 | static inline bool sync_wait_pid(int fd, pid_t *pid) | |
150 | { | |
151 | return lxc_read_nointr(fd, pid, sizeof(pid_t)) == sizeof(pid_t); | |
152 | } | |
153 | ||
154 | static inline bool sync_wake_fd(int fd, int fd_send) | |
155 | { | |
156 | return lxc_abstract_unix_send_fds(fd, &fd_send, 1, NULL, 0) > 0; | |
157 | } | |
158 | ||
159 | static inline bool sync_wait_fd(int fd, int *fd_recv) | |
160 | { | |
161 | return lxc_abstract_unix_recv_fds(fd, fd_recv, 1, NULL, 0) > 0; | |
162 | } | |
163 | ||
afc691a0 CB |
164 | static bool attach_lsm(lxc_attach_options_t *options) |
165 | { | |
166 | return (options->namespaces & CLONE_NEWNS) && | |
167 | (options->attach_flags & (LXC_ATTACH_LSM | LXC_ATTACH_LSM_LABEL)); | |
168 | } | |
169 | ||
9745eb8a CB |
170 | static struct attach_context *alloc_attach_context(void) |
171 | { | |
581b849a CB |
172 | struct attach_context *ctx; |
173 | ||
174 | ctx = zalloc(sizeof(struct attach_context)); | |
175 | if (!ctx) | |
176 | return ret_set_errno(NULL, ENOMEM); | |
177 | ||
178 | ctx->dfd_self_pid = -EBADF; | |
179 | ctx->dfd_init_pid = -EBADF; | |
4475fabb CB |
180 | ctx->init_pid = -ESRCH; |
181 | ctx->setup_uid = LXC_INVALID_UID; | |
182 | ctx->setup_gid = LXC_INVALID_GID; | |
183 | ctx->target_uid = LXC_INVALID_UID; | |
184 | ctx->target_gid = LXC_INVALID_GID; | |
581b849a CB |
185 | |
186 | for (int i = 0; i < LXC_NS_MAX; i++) | |
187 | ctx->ns_fd[i] = -EBADF; | |
188 | ||
189 | return ctx; | |
9745eb8a CB |
190 | } |
191 | ||
ee142207 CB |
192 | static int get_personality(const char *name, const char *lxcpath, |
193 | signed long *personality) | |
d92c8e40 CB |
194 | { |
195 | __do_free char *p = NULL; | |
ee142207 | 196 | signed long per; |
d92c8e40 CB |
197 | |
198 | p = lxc_cmd_get_config_item(name, "lxc.arch", lxcpath); | |
ee142207 CB |
199 | if (!p) { |
200 | *personality = LXC_ARCH_UNCHANGED; | |
201 | return 0; | |
202 | } | |
d92c8e40 | 203 | |
ee142207 CB |
204 | per = lxc_config_parse_arch(p); |
205 | if (per == LXC_ARCH_UNCHANGED) | |
206 | return ret_errno(EINVAL); | |
207 | ||
208 | *personality = per; | |
209 | return 0; | |
d92c8e40 CB |
210 | } |
211 | ||
4475fabb CB |
212 | static int userns_setup_ids(struct attach_context *ctx, |
213 | lxc_attach_options_t *options, | |
214 | uid_t init_hostuid, gid_t init_hostgid) | |
215 | { | |
216 | __do_free char *line = NULL; | |
217 | __do_fclose FILE *f_gidmap = NULL, *f_uidmap = NULL; | |
218 | size_t len = 0; | |
219 | uid_t init_ns_uid = LXC_INVALID_UID; | |
220 | gid_t init_ns_gid = LXC_INVALID_GID; | |
221 | uid_t nsuid, hostuid, range_uid; | |
222 | gid_t nsgid, hostgid, range_gid; | |
223 | ||
224 | if (!(options->namespaces & CLONE_NEWUSER)) | |
225 | return 0; | |
226 | ||
227 | f_uidmap = fdopenat(ctx->dfd_init_pid, "uid_map", "re"); | |
228 | if (!f_uidmap) | |
229 | return log_error_errno(-errno, errno, "Failed to open uid_map"); | |
230 | ||
231 | while (getline(&line, &len, f_uidmap) != -1) { | |
232 | if (sscanf(line, "%u %u %u", &nsuid, &hostuid, &range_uid) != 3) | |
233 | continue; | |
234 | ||
235 | if (0 >= nsuid && 0 < nsuid + range_uid) { | |
236 | ctx->setup_uid = 0; | |
237 | TRACE("Container has mapping for uid 0"); | |
238 | break; | |
239 | } | |
240 | ||
241 | if (init_hostuid >= hostuid && init_hostuid < hostuid + range_uid) { | |
242 | init_ns_uid = (init_hostuid - hostuid) + nsuid; | |
243 | TRACE("Container runs with uid %d", init_ns_uid); | |
244 | } | |
245 | } | |
246 | ||
247 | f_gidmap = fdopenat(ctx->dfd_init_pid, "gid_map", "re"); | |
248 | if (!f_gidmap) | |
249 | return log_error_errno(-errno, errno, "Failed to open gid_map"); | |
250 | ||
251 | while (getline(&line, &len, f_gidmap) != -1) { | |
252 | if (sscanf(line, "%u %u %u", &nsgid, &hostgid, &range_gid) != 3) | |
253 | continue; | |
254 | ||
255 | if (0 >= nsgid && 0 < nsgid + range_gid) { | |
256 | ctx->setup_gid = 0; | |
257 | TRACE("Container has mapping for gid 0"); | |
258 | break; | |
259 | } | |
260 | ||
261 | if (init_hostgid >= hostgid && init_hostgid < hostgid + range_gid) { | |
262 | init_ns_gid = (init_hostgid - hostgid) + nsgid; | |
263 | TRACE("Container runs with gid %d", init_ns_gid); | |
264 | } | |
265 | } | |
266 | ||
267 | if (ctx->setup_uid == LXC_INVALID_UID) | |
268 | ctx->setup_uid = init_ns_uid; | |
269 | ||
270 | if (ctx->setup_gid == LXC_INVALID_UID) | |
271 | ctx->setup_gid = init_ns_gid; | |
272 | ||
273 | /* | |
274 | * TODO: we should also parse supplementary groups and use | |
275 | * setgroups() to set them. | |
276 | */ | |
277 | ||
278 | return 0; | |
279 | } | |
280 | ||
281 | static void userns_target_ids(struct attach_context *ctx, lxc_attach_options_t *options) | |
282 | { | |
283 | if (options->uid != LXC_INVALID_UID) | |
284 | ctx->target_uid = options->uid; | |
285 | else if (options->namespaces & CLONE_NEWUSER) | |
286 | ctx->target_uid = ctx->setup_uid; | |
287 | else | |
288 | ctx->target_uid = 0; | |
289 | ||
290 | if (ctx->target_uid == LXC_INVALID_UID) | |
291 | WARN("Invalid uid specified"); | |
292 | ||
293 | if (options->gid != LXC_INVALID_GID) | |
294 | ctx->target_gid = options->gid; | |
295 | else if (options->namespaces & CLONE_NEWUSER) | |
296 | ctx->target_gid = ctx->setup_gid; | |
297 | else | |
298 | ctx->target_uid = 0; | |
299 | ||
300 | if (ctx->target_gid == LXC_INVALID_GID) | |
301 | WARN("Invalid gid specified"); | |
302 | } | |
303 | ||
9680e7b0 CB |
304 | static int parse_init_status(struct attach_context *ctx, lxc_attach_options_t *options) |
305 | { | |
306 | __do_free char *line = NULL; | |
307 | __do_fclose FILE *f = NULL; | |
308 | size_t len = 0; | |
309 | bool caps_found = false; | |
4475fabb CB |
310 | int ret; |
311 | uid_t init_host_uid = LXC_INVALID_UID; | |
312 | gid_t init_host_gid = LXC_INVALID_GID; | |
313 | ||
9680e7b0 CB |
314 | |
315 | f = fdopenat(ctx->dfd_init_pid, "status", "re"); | |
316 | if (!f) | |
4475fabb | 317 | return log_error_errno(-errno, errno, "Failed to open status file"); |
9680e7b0 CB |
318 | |
319 | while (getline(&line, &len, f) != -1) { | |
320 | signed long value = -1; | |
9680e7b0 | 321 | |
4475fabb CB |
322 | /* |
323 | * Format is: real, effective, saved set user, fs we only care | |
324 | * about real uid. | |
325 | */ | |
326 | ret = sscanf(line, "Uid: %ld", &value); | |
327 | if (ret != EOF && ret == 1) { | |
328 | init_host_uid = (uid_t)value; | |
329 | TRACE("Container's init process runs with hostuid %d", init_host_uid); | |
330 | goto next; | |
331 | } | |
9680e7b0 | 332 | |
4475fabb CB |
333 | ret = sscanf(line, "Gid: %ld", &value); |
334 | if (ret != EOF && ret == 1) { | |
335 | init_host_gid = (gid_t)value; | |
336 | TRACE("Container's init process runs with hostgid %d", init_host_gid); | |
337 | goto next; | |
9680e7b0 CB |
338 | } |
339 | ||
340 | ret = sscanf(line, "CapBnd: %llx", &ctx->capability_mask); | |
341 | if (ret != EOF && ret == 1) { | |
342 | caps_found = true; | |
343 | goto next; | |
344 | } | |
345 | ||
346 | next: | |
4475fabb CB |
347 | if (init_host_uid != LXC_INVALID_UID && |
348 | init_host_gid != LXC_INVALID_GID && | |
349 | caps_found) | |
9680e7b0 CB |
350 | break; |
351 | ||
352 | } | |
353 | ||
4475fabb CB |
354 | ret = userns_setup_ids(ctx, options, init_host_uid, init_host_gid); |
355 | if (ret) | |
356 | return log_error_errno(ret, errno, "Failed to get setup ids"); | |
357 | userns_target_ids(ctx, options); | |
358 | ||
9680e7b0 CB |
359 | /* |
360 | * TODO: we should also parse supplementary groups and use | |
361 | * setgroups() to set them. | |
362 | */ | |
363 | ||
364 | return 0; | |
365 | } | |
366 | ||
500ed813 | 367 | static int get_attach_context(struct attach_context *ctx, |
afc691a0 CB |
368 | struct lxc_container *container, |
369 | lxc_attach_options_t *options) | |
e0732705 | 370 | { |
9680e7b0 CB |
371 | __do_close int init_pidfd = -EBADF; |
372 | __do_free char *lsm_label = NULL; | |
6f4f1937 | 373 | int ret; |
c538837d | 374 | char path[LXC_PROC_PID_LEN]; |
e0732705 | 375 | |
500ed813 | 376 | ctx->container = container; |
afc691a0 | 377 | ctx->attach_flags = options->attach_flags; |
500ed813 | 378 | |
d8764025 CB |
379 | init_pidfd = lxc_cmd_get_init_pidfd(container->name, container->config_path); |
380 | if (init_pidfd >= 0) | |
381 | ctx->init_pid = pidfd_get_pid(init_pidfd); | |
382 | else | |
383 | ctx->init_pid = lxc_cmd_get_init_pid(container->name, container->config_path); | |
384 | ||
500ed813 CB |
385 | if (ctx->init_pid < 0) |
386 | return log_error(-1, "Failed to get init pid"); | |
387 | ||
595798bb | 388 | ctx->dfd_self_pid = openat(-EBADF, "/proc/self", O_CLOEXEC | O_NOCTTY | O_PATH | O_DIRECTORY); |
9680e7b0 | 389 | if (ctx->dfd_self_pid < 0) |
4475fabb | 390 | return log_error_errno(-errno, errno, "Failed to open /proc/self"); |
25c659d5 | 391 | |
c538837d CB |
392 | ret = snprintf(path, sizeof(path), "/proc/%d", ctx->init_pid); |
393 | if (ret < 0 || ret >= sizeof(path)) | |
394 | return ret_errno(EIO); | |
e0732705 | 395 | |
9680e7b0 CB |
396 | ctx->dfd_init_pid = openat(-EBADF, path, O_CLOEXEC | O_NOCTTY | O_NOFOLLOW | O_PATH | O_DIRECTORY); |
397 | if (ctx->dfd_init_pid < 0) | |
4475fabb | 398 | return log_error_errno(-errno, errno, "Failed to open /proc/%d", ctx->init_pid); |
c538837d | 399 | |
d8764025 CB |
400 | if (init_pidfd >= 0) { |
401 | ret = lxc_raw_pidfd_send_signal(init_pidfd, 0, NULL, 0); | |
402 | if (ret) | |
403 | return log_error_errno(-errno, errno, "Container process exited or PID has been recycled"); | |
404 | else | |
405 | TRACE("Container process still running and PID was not recycled"); | |
406 | } | |
407 | ||
4475fabb CB |
408 | /* Determine which namespaces the container was created with. */ |
409 | if (options->namespaces == -1) { | |
410 | options->namespaces = lxc_cmd_get_clone_flags(container->name, container->config_path); | |
411 | if (options->namespaces == -1) | |
412 | return log_error_errno(-EINVAL, EINVAL, "Failed to automatically determine the namespaces which the container uses"); | |
413 | ||
414 | for (int i = 0; i < LXC_NS_MAX; i++) { | |
415 | if (ns_info[i].clone_flag & CLONE_NEWCGROUP) | |
416 | if (!(options->attach_flags & LXC_ATTACH_MOVE_TO_CGROUP) || | |
417 | !cgns_supported()) | |
418 | continue; | |
419 | ||
420 | if (ns_info[i].clone_flag & options->namespaces) | |
421 | continue; | |
422 | ||
423 | ctx->ns_inherited |= ns_info[i].clone_flag; | |
424 | } | |
425 | } | |
426 | ||
9680e7b0 CB |
427 | ret = parse_init_status(ctx, options); |
428 | if (ret) | |
4475fabb | 429 | return log_error_errno(-errno, errno, "Failed to open parse file"); |
e0732705 | 430 | |
4eb19ac0 | 431 | ctx->lsm_ops = lsm_init_static(); |
d701d729 | 432 | |
afc691a0 CB |
433 | if (attach_lsm(options)) { |
434 | if (ctx->attach_flags & LXC_ATTACH_LSM_LABEL) | |
435 | lsm_label = options->lsm_label; | |
436 | else | |
9680e7b0 | 437 | lsm_label = ctx->lsm_ops->process_label_get_at(ctx->lsm_ops, ctx->dfd_init_pid); |
afc691a0 CB |
438 | if (!lsm_label) |
439 | WARN("No security context received"); | |
440 | else | |
441 | INFO("Retrieved security context %s", lsm_label); | |
442 | } | |
9745eb8a | 443 | ctx->ns_inherited = 0; |
e0732705 | 444 | |
ee142207 CB |
445 | ret = get_personality(container->name, container->config_path, &ctx->personality); |
446 | if (ret) | |
447 | return log_error_errno(ret, errno, "Failed to get personality of the container"); | |
d92c8e40 | 448 | |
1874ef74 CB |
449 | if (!ctx->container->lxc_conf) { |
450 | ctx->container->lxc_conf = lxc_conf_init(); | |
451 | if (!ctx->container->lxc_conf) | |
452 | return log_error_errno(-ENOMEM, ENOMEM, "Failed to allocate new lxc config"); | |
453 | } | |
454 | ||
afc691a0 | 455 | ctx->lsm_label = move_ptr(lsm_label); |
9745eb8a | 456 | return 0; |
e0732705 CS |
457 | } |
458 | ||
25c659d5 | 459 | static int in_same_namespace(int ns_fd_pid1, int ns_fd_pid2, const char *ns_path) |
299d1198 | 460 | { |
644e7393 | 461 | __do_close int ns_fd1 = -EBADF, ns_fd2 = -EBADF; |
3cc629fe | 462 | int ret = -1; |
299d1198 CB |
463 | struct stat ns_st1, ns_st2; |
464 | ||
25c659d5 | 465 | ns_fd1 = openat(ns_fd_pid1, ns_path, O_CLOEXEC | O_NOCTTY | O_RDONLY); |
134284c3 | 466 | if (ns_fd1 < 0) { |
25c659d5 | 467 | /* The kernel does not support this namespace. This is not an error. */ |
134284c3 CB |
468 | if (errno == ENOENT) |
469 | return -EINVAL; | |
470 | ||
3cc629fe | 471 | return -1; |
134284c3 | 472 | } |
299d1198 | 473 | |
25c659d5 | 474 | ns_fd2 = openat(ns_fd_pid2, ns_path, O_CLOEXEC | O_NOCTTY | O_RDONLY); |
21d0acc2 | 475 | if (ns_fd2 < 0) |
3cc629fe | 476 | return -1; |
299d1198 CB |
477 | |
478 | ret = fstat(ns_fd1, &ns_st1); | |
21d0acc2 | 479 | if (ret < 0) |
3cc629fe | 480 | return -1; |
299d1198 CB |
481 | |
482 | ret = fstat(ns_fd2, &ns_st2); | |
21d0acc2 | 483 | if (ret < 0) |
3cc629fe | 484 | return -1; |
299d1198 CB |
485 | |
486 | /* processes are in the same namespace */ | |
25c659d5 CB |
487 | if ((ns_st1.st_dev == ns_st2.st_dev) && |
488 | (ns_st1.st_ino == ns_st2.st_ino)) | |
3cc629fe | 489 | return -EINVAL; |
299d1198 CB |
490 | |
491 | /* processes are in different namespaces */ | |
3cc629fe | 492 | return move_fd(ns_fd2); |
299d1198 CB |
493 | } |
494 | ||
b7873c95 CB |
495 | static int get_attach_context_nsfds(struct attach_context *ctx, |
496 | lxc_attach_options_t *options) | |
497 | { | |
b7873c95 CB |
498 | for (int i = 0; i < LXC_NS_MAX; i++) { |
499 | int j; | |
500 | ||
501 | if (options->namespaces & ns_info[i].clone_flag) | |
25c659d5 | 502 | ctx->ns_fd[i] = openat(ctx->dfd_init_pid, ns_info[i].proc_path, O_CLOEXEC | O_NOCTTY | O_RDONLY); |
b7873c95 | 503 | else if (ctx->ns_inherited & ns_info[i].clone_flag) |
25c659d5 | 504 | ctx->ns_fd[i] = in_same_namespace(ctx->dfd_self_pid, ctx->dfd_init_pid, ns_info[i].proc_path); |
b7873c95 CB |
505 | else |
506 | continue; | |
507 | ||
508 | if (ctx->ns_fd[i] >= 0) | |
509 | continue; | |
510 | ||
511 | if (ctx->ns_fd[i] == -EINVAL) { | |
25c659d5 | 512 | DEBUG("Inheriting %s namespace", ns_info[i].proc_name); |
b7873c95 CB |
513 | ctx->ns_inherited &= ~ns_info[i].clone_flag; |
514 | continue; | |
515 | } | |
516 | ||
517 | /* We failed to preserve the namespace. */ | |
25c659d5 | 518 | SYSERROR("Failed to preserve %s namespace of %d", ns_info[i].proc_name, ctx->init_pid); |
b7873c95 CB |
519 | |
520 | /* Close all already opened file descriptors before we return an | |
521 | * error, so we don't leak them. | |
522 | */ | |
523 | for (j = 0; j < i; j++) | |
524 | close_prot_errno_disarm(ctx->ns_fd[j]); | |
525 | ||
526 | return -1; | |
527 | } | |
528 | ||
529 | return 0; | |
530 | } | |
531 | ||
7e995801 | 532 | static inline void close_nsfds(struct attach_context *ctx) |
b7873c95 CB |
533 | { |
534 | for (int i = 0; i < LXC_NS_MAX; i++) | |
535 | close_prot_errno_disarm(ctx->ns_fd[i]); | |
536 | } | |
537 | ||
538 | static void put_attach_context(struct attach_context *ctx) | |
539 | { | |
c538837d | 540 | if (ctx) { |
afc691a0 CB |
541 | if (!(ctx->attach_flags & LXC_ATTACH_LSM_LABEL)) |
542 | free_disarm(ctx->lsm_label); | |
25c659d5 | 543 | close_prot_errno_disarm(ctx->dfd_init_pid); |
b7873c95 | 544 | |
c538837d CB |
545 | if (ctx->container) { |
546 | lxc_container_put(ctx->container); | |
547 | ctx->container = NULL; | |
548 | } | |
b7873c95 | 549 | |
c538837d CB |
550 | close_nsfds(ctx); |
551 | free(ctx); | |
552 | } | |
b7873c95 CB |
553 | } |
554 | ||
c538837d | 555 | static int attach_context_container(struct attach_context *ctx) |
99d50954 | 556 | { |
ffeeed8b CB |
557 | for (int i = 0; i < LXC_NS_MAX; i++) { |
558 | int ret; | |
99d50954 | 559 | |
877f3a04 | 560 | if (ctx->ns_fd[i] < 0) |
26818618 CB |
561 | continue; |
562 | ||
21d0acc2 | 563 | ret = setns(ctx->ns_fd[i], ns_info[i].clone_flag); |
ffeeed8b | 564 | if (ret < 0) |
39b3b69b CB |
565 | return log_error_errno(-1, errno, |
566 | "Failed to attach to %s namespace of %d", | |
567 | ns_info[i].proc_name, ctx->init_pid); | |
99d50954 | 568 | |
39b3b69b CB |
569 | DEBUG("Attached to %s namespace of %d", |
570 | ns_info[i].proc_name, ctx->init_pid); | |
99d50954 CS |
571 | } |
572 | ||
573 | return 0; | |
574 | } | |
575 | ||
c538837d CB |
576 | /* |
577 | * Place anything in here that needs to be get rid of before we move into the | |
578 | * container's context and fail hard if we can't. | |
579 | */ | |
580 | static bool attach_context_security_barrier(struct attach_context *ctx) | |
581 | { | |
582 | if (ctx) { | |
25c659d5 CB |
583 | if (close(ctx->dfd_self_pid)) |
584 | return false; | |
585 | ctx->dfd_self_pid = -EBADF; | |
586 | ||
587 | if (close(ctx->dfd_init_pid)) | |
c538837d | 588 | return false; |
25c659d5 | 589 | ctx->dfd_init_pid = -EBADF; |
c538837d CB |
590 | } |
591 | ||
592 | return true; | |
593 | } | |
594 | ||
e4103cf6 | 595 | int lxc_attach_remount_sys_proc(void) |
7a0b0b56 CS |
596 | { |
597 | int ret; | |
598 | ||
599 | ret = unshare(CLONE_NEWNS); | |
ffeeed8b CB |
600 | if (ret < 0) |
601 | return log_error_errno(-1, errno, "Failed to unshare mount namespace"); | |
7a0b0b56 | 602 | |
9e61fb1f CB |
603 | if (detect_shared_rootfs() && mount(NULL, "/", NULL, MS_SLAVE | MS_REC, NULL)) |
604 | SYSERROR("Failed to recursively turn root mount tree into dependent mount. Continuing..."); | |
2c6f3fc9 | 605 | |
8ce83369 | 606 | /* Assume /proc is always mounted, so remount it. */ |
7a0b0b56 | 607 | ret = umount2("/proc", MNT_DETACH); |
ffeeed8b CB |
608 | if (ret < 0) |
609 | return log_error_errno(-1, errno, "Failed to unmount /proc"); | |
7a0b0b56 | 610 | |
657256e0 | 611 | ret = mount_filesystem("proc", "/proc", 0); |
ffeeed8b CB |
612 | if (ret < 0) |
613 | return log_error_errno(-1, errno, "Failed to remount /proc"); | |
7a0b0b56 | 614 | |
ffeeed8b CB |
615 | /* |
616 | * Try to umount /sys. If it's not a mount point, we'll get EINVAL, then | |
8ce83369 | 617 | * we ignore it because it may not have been mounted in the first place. |
7a0b0b56 CS |
618 | */ |
619 | ret = umount2("/sys", MNT_DETACH); | |
ffeeed8b CB |
620 | if (ret < 0 && errno != EINVAL) |
621 | return log_error_errno(-1, errno, "Failed to unmount /sys"); | |
622 | ||
623 | /* Remount it. */ | |
657256e0 | 624 | if (ret == 0 && mount_filesystem("sysfs", "/sys", 0)) |
ffeeed8b | 625 | return log_error_errno(-1, errno, "Failed to remount /sys"); |
7a0b0b56 CS |
626 | |
627 | return 0; | |
628 | } | |
629 | ||
677e1d27 | 630 | static int drop_capabilities(struct attach_context *ctx) |
e0732705 | 631 | { |
ffeeed8b | 632 | int last_cap; |
e0732705 | 633 | |
6f4f1937 | 634 | last_cap = lxc_caps_last_cap(); |
ffeeed8b | 635 | for (int cap = 0; cap <= last_cap; cap++) { |
e0732705 CS |
636 | if (ctx->capability_mask & (1LL << cap)) |
637 | continue; | |
638 | ||
b81689a1 | 639 | if (prctl(PR_CAPBSET_DROP, prctl_arg(cap), prctl_arg(0), |
ffeeed8b CB |
640 | prctl_arg(0), prctl_arg(0))) |
641 | return log_error_errno(-1, errno, "Failed to drop capability %d", cap); | |
ea918412 | 642 | |
94ac256f | 643 | TRACE("Dropped capability %d", cap); |
e0732705 CS |
644 | } |
645 | ||
646 | return 0; | |
647 | } | |
905022f7 | 648 | |
ab919e5f | 649 | static int lxc_attach_set_environment(struct attach_context *ctx, |
7385273f | 650 | enum lxc_attach_env_policy_t policy, |
6f4f1937 | 651 | char **extra_env, char **extra_keep) |
b3a39ba6 | 652 | { |
3d55242a | 653 | int ret; |
7385273f | 654 | struct lxc_list *iterator; |
655 | ||
799f96fd | 656 | if (policy == LXC_ATTACH_CLEAR_ENV) { |
3d5e9f48 | 657 | int path_kept = 0; |
6f4f1937 | 658 | char **extra_keep_store = NULL; |
3d5e9f48 CS |
659 | |
660 | if (extra_keep) { | |
661 | size_t count, i; | |
662 | ||
3d55242a CB |
663 | for (count = 0; extra_keep[count]; count++) |
664 | ; | |
3d5e9f48 | 665 | |
89b7bfe3 | 666 | extra_keep_store = zalloc(count * sizeof(char *)); |
3d55242a | 667 | if (!extra_keep_store) |
3d5e9f48 | 668 | return -1; |
3d55242a | 669 | |
3d5e9f48 CS |
670 | for (i = 0; i < count; i++) { |
671 | char *v = getenv(extra_keep[i]); | |
672 | if (v) { | |
673 | extra_keep_store[i] = strdup(v); | |
674 | if (!extra_keep_store[i]) { | |
3d5e9f48 CS |
675 | while (i > 0) |
676 | free(extra_keep_store[--i]); | |
ea918412 | 677 | |
3d5e9f48 CS |
678 | free(extra_keep_store); |
679 | return -1; | |
680 | } | |
3d55242a | 681 | |
3d5e9f48 CS |
682 | if (strcmp(extra_keep[i], "PATH") == 0) |
683 | path_kept = 1; | |
684 | } | |
3d5e9f48 CS |
685 | } |
686 | } | |
687 | ||
799f96fd | 688 | if (clearenv()) { |
a9cab7e3 | 689 | if (extra_keep_store) { |
3d55242a CB |
690 | char **p; |
691 | ||
a9cab7e3 CS |
692 | for (p = extra_keep_store; *p; p++) |
693 | free(*p); | |
3d55242a | 694 | |
a9cab7e3 CS |
695 | free(extra_keep_store); |
696 | } | |
3d55242a | 697 | |
ffeeed8b | 698 | return log_error(-1, "Failed to clear environment"); |
3d5e9f48 CS |
699 | } |
700 | ||
701 | if (extra_keep_store) { | |
702 | size_t i; | |
6f4f1937 | 703 | |
3d5e9f48 | 704 | for (i = 0; extra_keep[i]; i++) { |
acd4922e | 705 | if (extra_keep_store[i]) { |
3d55242a CB |
706 | ret = setenv(extra_keep[i], extra_keep_store[i], 1); |
707 | if (ret < 0) | |
a24c5678 | 708 | SYSWARN("Failed to set environment variable"); |
acd4922e | 709 | } |
ea918412 | 710 | |
3d5e9f48 CS |
711 | free(extra_keep_store[i]); |
712 | } | |
ea918412 | 713 | |
3d5e9f48 CS |
714 | free(extra_keep_store); |
715 | } | |
716 | ||
8ce83369 CB |
717 | /* Always set a default path; shells and execlp tend to be fine |
718 | * without it, but there is a disturbing number of C programs | |
719 | * out there that just assume that getenv("PATH") is never NULL | |
720 | * and then die a painful segfault death. | |
721 | */ | |
3d55242a CB |
722 | if (!path_kept) { |
723 | ret = setenv("PATH", "/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin", 1); | |
724 | if (ret < 0) | |
a24c5678 | 725 | SYSWARN("Failed to set environment variable"); |
3d55242a | 726 | } |
b3a39ba6 DW |
727 | } |
728 | ||
3d55242a | 729 | ret = putenv("container=lxc"); |
ffeeed8b | 730 | if (ret < 0) |
818a57fc | 731 | return log_warn(-1, "Failed to set environment variable"); |
b3a39ba6 | 732 | |
7385273f | 733 | /* Set container environment variables.*/ |
640952e5 | 734 | if (ctx->container->lxc_conf) { |
ab919e5f | 735 | lxc_list_for_each(iterator, &ctx->container->lxc_conf->environment) { |
3d55242a CB |
736 | char *env_tmp; |
737 | ||
738 | env_tmp = strdup((char *)iterator->elem); | |
739 | if (!env_tmp) | |
7385273f | 740 | return -1; |
7385273f | 741 | |
3d55242a | 742 | ret = putenv(env_tmp); |
ffeeed8b CB |
743 | if (ret < 0) |
744 | return log_error_errno(-1, errno, "Failed to set environment variable: %s", (char *)iterator->elem); | |
7385273f | 745 | } |
746 | } | |
747 | ||
8ce83369 | 748 | /* Set extra environment variables. */ |
3d5e9f48 CS |
749 | if (extra_env) { |
750 | for (; *extra_env; extra_env++) { | |
3d55242a | 751 | char *p; |
ea918412 | 752 | |
8ce83369 CB |
753 | /* We just assume the user knows what they are doing, so |
754 | * we don't do any checks. | |
755 | */ | |
3d55242a CB |
756 | p = strdup(*extra_env); |
757 | if (!p) | |
3d5e9f48 | 758 | return -1; |
3d55242a CB |
759 | |
760 | ret = putenv(p); | |
761 | if (ret < 0) | |
a24c5678 | 762 | SYSWARN("Failed to set environment variable"); |
3d5e9f48 CS |
763 | } |
764 | } | |
765 | ||
b3a39ba6 DW |
766 | return 0; |
767 | } | |
768 | ||
74a3920a | 769 | static char *lxc_attach_getpwshell(uid_t uid) |
905022f7 | 770 | { |
1b9c9f5b | 771 | __do_free char *line = NULL, *result = NULL; |
cd8f5663 | 772 | __do_fclose FILE *pipe_f = NULL; |
6f4f1937 | 773 | int fd, ret; |
905022f7 CS |
774 | pid_t pid; |
775 | int pipes[2]; | |
3fa23ac3 CB |
776 | bool found = false; |
777 | size_t line_bufsz = 0; | |
905022f7 | 778 | |
8ce83369 CB |
779 | /* We need to fork off a process that runs the getent program, and we |
780 | * need to capture its output, so we use a pipe for that purpose. | |
905022f7 | 781 | */ |
3fa23ac3 | 782 | ret = pipe2(pipes, O_CLOEXEC); |
905022f7 CS |
783 | if (ret < 0) |
784 | return NULL; | |
785 | ||
786 | pid = fork(); | |
787 | if (pid < 0) { | |
788 | close(pipes[0]); | |
789 | close(pipes[1]); | |
790 | return NULL; | |
791 | } | |
792 | ||
3fa23ac3 | 793 | if (!pid) { |
905022f7 CS |
794 | char uid_buf[32]; |
795 | char *arguments[] = { | |
796 | "getent", | |
797 | "passwd", | |
798 | uid_buf, | |
799 | NULL | |
800 | }; | |
801 | ||
802 | close(pipes[0]); | |
803 | ||
8ce83369 | 804 | /* We want to capture stdout. */ |
3fa23ac3 | 805 | ret = dup2(pipes[1], STDOUT_FILENO); |
905022f7 | 806 | close(pipes[1]); |
3fa23ac3 | 807 | if (ret < 0) |
ea918412 | 808 | _exit(EXIT_FAILURE); |
905022f7 | 809 | |
8ce83369 CB |
810 | /* Get rid of stdin/stderr, so we try to associate it with |
811 | * /dev/null. | |
905022f7 | 812 | */ |
3fa23ac3 | 813 | fd = open_devnull(); |
905022f7 | 814 | if (fd < 0) { |
3fa23ac3 CB |
815 | close(STDIN_FILENO); |
816 | close(STDERR_FILENO); | |
905022f7 | 817 | } else { |
3fa23ac3 | 818 | (void)dup3(fd, STDIN_FILENO, O_CLOEXEC); |
59f0e209 | 819 | (void)dup3(fd, STDERR_FILENO, O_CLOEXEC); |
905022f7 CS |
820 | close(fd); |
821 | } | |
822 | ||
8ce83369 | 823 | /* Finish argument list. */ |
3fa23ac3 CB |
824 | ret = snprintf(uid_buf, sizeof(uid_buf), "%ld", (long)uid); |
825 | if (ret <= 0 || ret >= sizeof(uid_buf)) | |
ea918412 | 826 | _exit(EXIT_FAILURE); |
905022f7 | 827 | |
8ce83369 | 828 | /* Try to run getent program. */ |
3fa23ac3 | 829 | (void)execvp("getent", arguments); |
ea918412 | 830 | _exit(EXIT_FAILURE); |
905022f7 | 831 | } |
3fa23ac3 CB |
832 | |
833 | close(pipes[1]); | |
834 | ||
4110345b | 835 | pipe_f = fdopen(pipes[0], "re"); |
cf4026f1 CB |
836 | if (!pipe_f) { |
837 | close(pipes[0]); | |
838 | goto reap_child; | |
839 | } | |
840 | /* Transfer ownership of pipes[0] to pipe_f. */ | |
841 | move_fd(pipes[0]); | |
842 | ||
3fa23ac3 CB |
843 | while (getline(&line, &line_bufsz, pipe_f) != -1) { |
844 | int i; | |
845 | long value; | |
846 | char *token; | |
847 | char *endptr = NULL, *saveptr = NULL; | |
848 | ||
849 | /* If we already found something, just continue to read | |
850 | * until the pipe doesn't deliver any more data, but | |
851 | * don't modify the existing data structure. | |
852 | */ | |
853 | if (found) | |
854 | continue; | |
855 | ||
18d4ffde | 856 | if (!line) |
857 | continue; | |
858 | ||
3fa23ac3 CB |
859 | /* Trim line on the right hand side. */ |
860 | for (i = strlen(line); i > 0 && (line[i - 1] == '\n' || line[i - 1] == '\r'); --i) | |
861 | line[i - 1] = '\0'; | |
862 | ||
863 | /* Split into tokens: first: user name. */ | |
864 | token = strtok_r(line, ":", &saveptr); | |
865 | if (!token) | |
866 | continue; | |
867 | ||
868 | /* next: dummy password field */ | |
869 | token = strtok_r(NULL, ":", &saveptr); | |
870 | if (!token) | |
871 | continue; | |
872 | ||
873 | /* next: user id */ | |
874 | token = strtok_r(NULL, ":", &saveptr); | |
875 | value = token ? strtol(token, &endptr, 10) : 0; | |
876 | if (!token || !endptr || *endptr || value == LONG_MIN || | |
ea918412 | 877 | value == LONG_MAX) |
3fa23ac3 CB |
878 | continue; |
879 | ||
880 | /* dummy sanity check: user id matches */ | |
881 | if ((uid_t)value != uid) | |
882 | continue; | |
883 | ||
884 | /* skip fields: gid, gecos, dir, go to next field 'shell' */ | |
885 | for (i = 0; i < 4; i++) { | |
886 | token = strtok_r(NULL, ":", &saveptr); | |
887 | if (!token) | |
888 | continue; | |
889 | } | |
ea918412 | 890 | |
3fa23ac3 CB |
891 | if (!token) |
892 | continue; | |
ea918412 | 893 | |
1b9c9f5b | 894 | free_disarm(result); |
3fa23ac3 CB |
895 | result = strdup(token); |
896 | ||
897 | /* Sanity check that there are no fields after that. */ | |
898 | token = strtok_r(NULL, ":", &saveptr); | |
899 | if (token) | |
900 | continue; | |
901 | ||
902 | found = true; | |
903 | } | |
ea918412 | 904 | |
cf4026f1 | 905 | reap_child: |
3fa23ac3 | 906 | ret = wait_for_pid(pid); |
1b9c9f5b | 907 | if (ret < 0) |
3fa23ac3 | 908 | return NULL; |
3fa23ac3 | 909 | |
1b9c9f5b | 910 | if (!found) |
3fa23ac3 | 911 | return NULL; |
3fa23ac3 | 912 | |
1b9c9f5b | 913 | return move_ptr(result); |
905022f7 | 914 | } |
cb3e61fa | 915 | |
d4db3d14 | 916 | static bool fetch_seccomp(struct lxc_container *c, lxc_attach_options_t *options) |
2c4ea790 | 917 | { |
cd8f5663 | 918 | __do_free char *path = NULL; |
d4db3d14 CB |
919 | int ret; |
920 | bool bret; | |
2eef2bda | 921 | |
afc691a0 | 922 | if (!attach_lsm(options)) { |
cca66e06 | 923 | free_disarm(c->lxc_conf->seccomp.seccomp); |
2c4ea790 | 924 | return true; |
bd4307f0 | 925 | } |
bd7b4e28 | 926 | |
afc691a0 | 927 | /* Remove current setting. */ |
d4db3d14 | 928 | if (!c->set_config_item(c, "lxc.seccomp.profile", "") && |
ea918412 | 929 | !c->set_config_item(c, "lxc.seccomp", "")) |
2c4ea790 | 930 | return false; |
bd7b4e28 | 931 | |
8ce83369 | 932 | /* Fetch the current profile path over the cmd interface. */ |
0b427da0 | 933 | path = c->get_running_config_item(c, "lxc.seccomp.profile"); |
bd7b4e28 | 934 | if (!path) { |
d4db3d14 | 935 | INFO("Failed to retrieve lxc.seccomp.profile"); |
ea918412 | 936 | |
0b427da0 | 937 | path = c->get_running_config_item(c, "lxc.seccomp"); |
cca66e06 CB |
938 | if (!path) |
939 | return log_info(true, "Failed to retrieve lxc.seccomp"); | |
bd7b4e28 SG |
940 | } |
941 | ||
8ce83369 | 942 | /* Copy the value into the new lxc_conf. */ |
d4db3d14 | 943 | bret = c->set_config_item(c, "lxc.seccomp.profile", path); |
d4db3d14 CB |
944 | if (!bret) |
945 | return false; | |
bd7b4e28 | 946 | |
8ce83369 | 947 | /* Attempt to parse the resulting config. */ |
d4db3d14 | 948 | ret = lxc_read_seccomp_config(c->lxc_conf); |
cca66e06 CB |
949 | if (ret < 0) |
950 | return log_error(false, "Failed to retrieve seccomp policy"); | |
2c4ea790 | 951 | |
cca66e06 | 952 | return log_info(true, "Retrieved seccomp policy"); |
2e812c16 CB |
953 | } |
954 | ||
6f4f1937 | 955 | static bool no_new_privs(struct lxc_container *c, lxc_attach_options_t *options) |
2e812c16 | 956 | { |
cd8f5663 | 957 | __do_free char *val = NULL; |
2e812c16 | 958 | |
2e812c16 | 959 | /* Remove current setting. */ |
02d3b72b CB |
960 | if (!c->set_config_item(c, "lxc.no_new_privs", "")) |
961 | return log_info(false, "Failed to unset lxc.no_new_privs"); | |
2e812c16 CB |
962 | |
963 | /* Retrieve currently active setting. */ | |
964 | val = c->get_running_config_item(c, "lxc.no_new_privs"); | |
02d3b72b CB |
965 | if (!val) |
966 | return log_info(false, "Failed to retrieve lxc.no_new_privs"); | |
2e812c16 CB |
967 | |
968 | /* Set currently active setting. */ | |
cd8f5663 | 969 | return c->set_config_item(c, "lxc.no_new_privs", val); |
2c4ea790 SH |
970 | } |
971 | ||
338b230f | 972 | struct attach_payload { |
a998454a | 973 | int ipc_socket; |
cecf3e83 | 974 | int terminal_pts_fd; |
a998454a | 975 | lxc_attach_options_t *options; |
ab919e5f | 976 | struct attach_context *ctx; |
a998454a CB |
977 | lxc_attach_exec_t exec_function; |
978 | void *exec_payload; | |
979 | }; | |
980 | ||
338b230f | 981 | static void put_attach_payload(struct attach_payload *p) |
ba2be1a8 | 982 | { |
afc691a0 CB |
983 | if (p) { |
984 | close_prot_errno_disarm(p->ipc_socket); | |
985 | close_prot_errno_disarm(p->terminal_pts_fd); | |
dd53c8af | 986 | put_attach_context(p->ctx); |
ab919e5f | 987 | p->ctx = NULL; |
b21da190 | 988 | } |
ba2be1a8 CB |
989 | } |
990 | ||
338b230f | 991 | __noreturn static void do_attach(struct attach_payload *ap) |
a998454a | 992 | { |
afc691a0 CB |
993 | lxc_attach_exec_t attach_function = move_ptr(ap->exec_function); |
994 | void *attach_function_args = move_ptr(ap->exec_payload); | |
427a8067 | 995 | int lsm_fd, ret; |
338b230f CB |
996 | lxc_attach_options_t* options = ap->options; |
997 | struct attach_context *ctx = ap->ctx; | |
ab919e5f | 998 | struct lxc_conf *conf = ctx->container->lxc_conf; |
a998454a CB |
999 | |
1000 | /* A description of the purpose of this functionality is provided in the | |
1001 | * lxc-attach(1) manual page. We have to remount here and not in the | |
1002 | * parent process, otherwise /proc may not properly reflect the new pid | |
1003 | * namespace. | |
1004 | */ | |
1005 | if (!(options->namespaces & CLONE_NEWNS) && | |
1006 | (options->attach_flags & LXC_ATTACH_REMOUNT_PROC_SYS)) { | |
1007 | ret = lxc_attach_remount_sys_proc(); | |
b75c344c CB |
1008 | if (ret < 0) |
1009 | goto on_error; | |
ea918412 | 1010 | |
b75c344c | 1011 | TRACE("Remounted \"/proc\" and \"/sys\""); |
a998454a CB |
1012 | } |
1013 | ||
5b514ce3 | 1014 | /* Now perform additional attachments. */ |
a998454a | 1015 | #if HAVE_SYS_PERSONALITY_H |
a998454a | 1016 | if (options->attach_flags & LXC_ATTACH_SET_PERSONALITY) { |
b75c344c CB |
1017 | long new_personality; |
1018 | ||
1019 | if (options->personality < 0) | |
ab919e5f | 1020 | new_personality = ctx->personality; |
b75c344c CB |
1021 | else |
1022 | new_personality = options->personality; | |
ea918412 | 1023 | |
ee142207 CB |
1024 | if (new_personality != LXC_ARCH_UNCHANGED) { |
1025 | ret = personality(new_personality); | |
1026 | if (ret < 0) | |
1027 | goto on_error; | |
ea918412 | 1028 | |
ee142207 CB |
1029 | TRACE("Set new personality"); |
1030 | } | |
a998454a CB |
1031 | } |
1032 | #endif | |
1033 | ||
1034 | if (options->attach_flags & LXC_ATTACH_DROP_CAPABILITIES) { | |
677e1d27 | 1035 | ret = drop_capabilities(ctx); |
b75c344c CB |
1036 | if (ret < 0) |
1037 | goto on_error; | |
ea918412 | 1038 | |
b75c344c | 1039 | TRACE("Dropped capabilities"); |
a998454a CB |
1040 | } |
1041 | ||
1042 | /* Always set the environment (specify (LXC_ATTACH_KEEP_ENV, NULL, NULL) | |
1043 | * if you want this to be a no-op). | |
1044 | */ | |
ab919e5f | 1045 | ret = lxc_attach_set_environment(ctx, |
7385273f | 1046 | options->env_policy, |
a998454a CB |
1047 | options->extra_env_vars, |
1048 | options->extra_keep_env); | |
b75c344c CB |
1049 | if (ret < 0) |
1050 | goto on_error; | |
ea918412 | 1051 | |
b75c344c | 1052 | TRACE("Set up environment"); |
a998454a | 1053 | |
afc691a0 CB |
1054 | /* |
1055 | * This remark only affects fully unprivileged containers: | |
57de839f CB |
1056 | * Receive fd for LSM security module before we set{g,u}id(). The reason |
1057 | * is that on set{g,u}id() the kernel will a) make us undumpable and b) | |
1058 | * we will change our effective uid. This means our effective uid will | |
1059 | * be different from the effective uid of the process that created us | |
1060 | * which means that this processs no longer has capabilities in our | |
1061 | * namespace including CAP_SYS_PTRACE. This means we will not be able to | |
1062 | * read and /proc/<pid> files for the process anymore when /proc is | |
1063 | * mounted with hidepid={1,2}. So let's get the lsm label fd before the | |
1064 | * set{g,u}id(). | |
1065 | */ | |
afc691a0 | 1066 | if (attach_lsm(options) && ctx->lsm_label) { |
f8e88e94 | 1067 | if (!sync_wait_fd(ap->ipc_socket, ATTACH_SYNC_LSM(&lsm_fd))) { |
6e36c297 | 1068 | SYSERROR("Failed to receive lsm label fd"); |
b75c344c | 1069 | goto on_error; |
9044b79e | 1070 | } |
1071 | ||
57de839f CB |
1072 | TRACE("Received LSM label file descriptor %d from parent", lsm_fd); |
1073 | } | |
1074 | ||
08ea9270 | 1075 | if (options->stdin_fd > 0 && isatty(options->stdin_fd)) { |
cd0a2b2f | 1076 | ret = lxc_make_controlling_terminal(options->stdin_fd); |
08ea9270 CB |
1077 | if (ret < 0) |
1078 | goto on_error; | |
1079 | } | |
1080 | ||
b58214ac CB |
1081 | if (!lxc_setgroups(0, NULL) && errno != EPERM) |
1082 | goto on_error; | |
1083 | ||
4475fabb CB |
1084 | if (options->namespaces & CLONE_NEWUSER) |
1085 | if (!lxc_switch_uid_gid(ctx->setup_uid, ctx->setup_gid)) | |
b75c344c | 1086 | goto on_error; |
936efc72 | 1087 | |
afc691a0 | 1088 | if (attach_lsm(options) && ctx->lsm_label) { |
d3ba7c98 | 1089 | bool on_exec; |
a998454a CB |
1090 | |
1091 | /* Change into our new LSM profile. */ | |
d3ba7c98 | 1092 | on_exec = options->attach_flags & LXC_ATTACH_LSM_EXEC ? true : false; |
afc691a0 | 1093 | ret = ctx->lsm_ops->process_label_set_at(ctx->lsm_ops, lsm_fd, ctx->lsm_label, on_exec); |
cb2420df | 1094 | close_prot_errno_disarm(lsm_fd); |
b75c344c CB |
1095 | if (ret < 0) |
1096 | goto on_error; | |
ea918412 | 1097 | |
ab919e5f | 1098 | TRACE("Set %s LSM label to \"%s\"", ctx->lsm_ops->name, ctx->lsm_label); |
a998454a CB |
1099 | } |
1100 | ||
640952e5 | 1101 | if (conf->no_new_privs || (options->attach_flags & LXC_ATTACH_NO_NEW_PRIVS)) { |
6ce8e678 AL |
1102 | ret = prctl(PR_SET_NO_NEW_PRIVS, prctl_arg(1), prctl_arg(0), |
1103 | prctl_arg(0), prctl_arg(0)); | |
1104 | if (ret < 0) | |
1105 | goto on_error; | |
1106 | ||
1107 | TRACE("Set PR_SET_NO_NEW_PRIVS"); | |
1108 | } | |
1109 | ||
640952e5 | 1110 | if (conf->seccomp.seccomp) { |
cdb2a47f | 1111 | ret = lxc_seccomp_load(conf); |
b75c344c CB |
1112 | if (ret < 0) |
1113 | goto on_error; | |
ea918412 | 1114 | |
b75c344c | 1115 | TRACE("Loaded seccomp profile"); |
cdb2a47f | 1116 | |
338b230f | 1117 | ret = lxc_seccomp_send_notifier_fd(&conf->seccomp, ap->ipc_socket); |
c3e3c21a CB |
1118 | if (ret < 0) |
1119 | goto on_error; | |
a998454a | 1120 | } |
ea918412 | 1121 | |
a998454a CB |
1122 | /* The following is done after the communication socket is shut down. |
1123 | * That way, all errors that might (though unlikely) occur up until this | |
1124 | * point will have their messages printed to the original stderr (if | |
1125 | * logging is so configured) and not the fd the user supplied, if any. | |
1126 | */ | |
1127 | ||
1128 | /* Fd handling for stdin, stdout and stderr; ignore errors here, user | |
1129 | * may want to make sure the fds are closed, for example. | |
1130 | */ | |
08ea9270 | 1131 | if (options->stdin_fd >= 0 && options->stdin_fd != STDIN_FILENO) |
40301d48 | 1132 | if (dup2(options->stdin_fd, STDIN_FILENO) < 0) |
a7563434 | 1133 | SYSDEBUG("Failed to replace stdin with %d", options->stdin_fd); |
08ea9270 CB |
1134 | |
1135 | if (options->stdout_fd >= 0 && options->stdout_fd != STDOUT_FILENO) | |
40301d48 | 1136 | if (dup2(options->stdout_fd, STDOUT_FILENO) < 0) |
93b9960a | 1137 | SYSDEBUG("Failed to replace stdout with %d", options->stdout_fd); |
08ea9270 CB |
1138 | |
1139 | if (options->stderr_fd >= 0 && options->stderr_fd != STDERR_FILENO) | |
40301d48 | 1140 | if (dup2(options->stderr_fd, STDERR_FILENO) < 0) |
93b9960a | 1141 | SYSDEBUG("Failed to replace stderr with %d", options->stderr_fd); |
a998454a CB |
1142 | |
1143 | /* close the old fds */ | |
08ea9270 | 1144 | if (options->stdin_fd > STDERR_FILENO) |
a998454a | 1145 | close(options->stdin_fd); |
08ea9270 CB |
1146 | |
1147 | if (options->stdout_fd > STDERR_FILENO) | |
a998454a | 1148 | close(options->stdout_fd); |
08ea9270 CB |
1149 | |
1150 | if (options->stderr_fd > STDERR_FILENO) | |
a998454a CB |
1151 | close(options->stderr_fd); |
1152 | ||
427a8067 CB |
1153 | /* |
1154 | * Try to remove FD_CLOEXEC flag from stdin/stdout/stderr, but also | |
a998454a CB |
1155 | * here, ignore errors. |
1156 | */ | |
427a8067 | 1157 | for (int fd = STDIN_FILENO; fd <= STDERR_FILENO; fd++) { |
3f62938a | 1158 | ret = fd_cloexec(fd, false); |
b75c344c CB |
1159 | if (ret < 0) { |
1160 | SYSERROR("Failed to clear FD_CLOEXEC from file descriptor %d", fd); | |
1161 | goto on_error; | |
1162 | } | |
a998454a CB |
1163 | } |
1164 | ||
9e84479f | 1165 | if (options->attach_flags & LXC_ATTACH_TERMINAL) { |
338b230f | 1166 | ret = lxc_terminal_prepare_login(ap->terminal_pts_fd); |
ba2be1a8 | 1167 | if (ret < 0) { |
338b230f | 1168 | SYSERROR("Failed to prepare terminal file descriptor %d", ap->terminal_pts_fd); |
ba2be1a8 CB |
1169 | goto on_error; |
1170 | } | |
ea918412 | 1171 | |
338b230f | 1172 | TRACE("Prepared terminal file descriptor %d", ap->terminal_pts_fd); |
ba2be1a8 CB |
1173 | } |
1174 | ||
afc691a0 CB |
1175 | put_attach_payload(ap); |
1176 | ||
6aff5157 | 1177 | /* Make sure that the processes STDIO is correctly owned by the user that we are switching to */ |
4475fabb | 1178 | ret = fix_stdio_permissions(ctx->target_uid); |
c353b0b9 | 1179 | if (ret) |
a2c26bef | 1180 | INFO("Failed to adjust stdio permissions"); |
936efc72 | 1181 | |
4475fabb CB |
1182 | /* Avoid unnecessary syscalls. */ |
1183 | if (ctx->setup_uid == ctx->target_uid) | |
1184 | ctx->target_uid = LXC_INVALID_UID; | |
1185 | ||
1186 | if (ctx->setup_gid == ctx->target_gid) | |
1187 | ctx->target_gid = LXC_INVALID_GID; | |
1188 | ||
1189 | if (!lxc_switch_uid_gid(ctx->target_uid, ctx->target_gid)) | |
936efc72 CB |
1190 | goto on_error; |
1191 | ||
a998454a | 1192 | /* We're done, so we can now do whatever the user intended us to do. */ |
afc691a0 | 1193 | _exit(attach_function(attach_function_args)); |
b75c344c CB |
1194 | |
1195 | on_error: | |
dab02267 | 1196 | ERROR("Failed to attach to container"); |
c7ac2e1c | 1197 | _exit(EXIT_FAILURE); |
a998454a CB |
1198 | } |
1199 | ||
f797f05e | 1200 | static int lxc_attach_terminal(const char *name, const char *lxcpath, struct lxc_conf *conf, |
9e84479f | 1201 | struct lxc_terminal *terminal) |
ba2be1a8 CB |
1202 | { |
1203 | int ret; | |
1204 | ||
9e84479f | 1205 | lxc_terminal_init(terminal); |
ba2be1a8 | 1206 | |
8ea93a0f | 1207 | ret = lxc_terminal_create(name, lxcpath, conf, terminal); |
c2af3a15 CB |
1208 | if (ret < 0) |
1209 | return log_error(-1, "Failed to create terminal"); | |
ba2be1a8 | 1210 | |
ba2be1a8 | 1211 | return 0; |
ba2be1a8 CB |
1212 | } |
1213 | ||
9e84479f CB |
1214 | static int lxc_attach_terminal_mainloop_init(struct lxc_terminal *terminal, |
1215 | struct lxc_epoll_descr *descr) | |
ba2be1a8 CB |
1216 | { |
1217 | int ret; | |
1218 | ||
1219 | ret = lxc_mainloop_open(descr); | |
c2af3a15 CB |
1220 | if (ret < 0) |
1221 | return log_error(-1, "Failed to create mainloop"); | |
ba2be1a8 | 1222 | |
9e84479f | 1223 | ret = lxc_terminal_mainloop_add(descr, terminal); |
ba2be1a8 | 1224 | if (ret < 0) { |
ba2be1a8 | 1225 | lxc_mainloop_close(descr); |
c2af3a15 | 1226 | return log_error(-1, "Failed to add handlers to mainloop"); |
ba2be1a8 CB |
1227 | } |
1228 | ||
1229 | return 0; | |
1230 | } | |
1231 | ||
36a94ce8 | 1232 | static inline void lxc_attach_terminal_close_ptx(struct lxc_terminal *terminal) |
ba2be1a8 | 1233 | { |
36a94ce8 | 1234 | close_prot_errno_disarm(terminal->ptx); |
ba2be1a8 CB |
1235 | } |
1236 | ||
cecf3e83 | 1237 | static inline void lxc_attach_terminal_close_pts(struct lxc_terminal *terminal) |
ba2be1a8 | 1238 | { |
41808e20 | 1239 | close_prot_errno_disarm(terminal->pty); |
ba2be1a8 CB |
1240 | } |
1241 | ||
9e84479f | 1242 | static inline void lxc_attach_terminal_close_peer(struct lxc_terminal *terminal) |
ba2be1a8 | 1243 | { |
19a3e906 | 1244 | close_prot_errno_disarm(terminal->peer); |
ba2be1a8 CB |
1245 | } |
1246 | ||
9e84479f | 1247 | static inline void lxc_attach_terminal_close_log(struct lxc_terminal *terminal) |
ba2be1a8 | 1248 | { |
19a3e906 | 1249 | close_prot_errno_disarm(terminal->log_fd); |
ba2be1a8 CB |
1250 | } |
1251 | ||
908fbc1a CB |
1252 | int lxc_attach(struct lxc_container *container, lxc_attach_exec_t exec_function, |
1253 | void *exec_payload, lxc_attach_options_t *options, | |
1254 | pid_t *attached_process) | |
9c4693b8 | 1255 | { |
6f9fe5d0 | 1256 | int ret_parent = -1; |
6f9fe5d0 | 1257 | struct lxc_epoll_descr descr = {}; |
a9f0cecf | 1258 | int ret; |
26abd7ea | 1259 | char *name, *lxcpath; |
9c4693b8 | 1260 | int ipc_sockets[2]; |
500ed813 | 1261 | pid_t attached_pid, pid, to_cleanup_pid; |
ab919e5f | 1262 | struct attach_context *ctx; |
9e84479f | 1263 | struct lxc_terminal terminal; |
1cce35e6 | 1264 | struct lxc_conf *conf; |
9c4693b8 | 1265 | |
908fbc1a | 1266 | if (!container) |
540a2f70 | 1267 | return ret_set_errno(-1, EINVAL); |
908fbc1a CB |
1268 | |
1269 | if (!lxc_container_get(container)) | |
540a2f70 | 1270 | return ret_set_errno(-1, EINVAL); |
908fbc1a CB |
1271 | |
1272 | name = container->name; | |
1273 | lxcpath = container->config_path; | |
1274 | ||
afc691a0 | 1275 | if (!options) { |
9c4693b8 | 1276 | options = &attach_static_default_options; |
afc691a0 CB |
1277 | options->lsm_label = NULL; |
1278 | } | |
9c4693b8 | 1279 | |
9745eb8a | 1280 | ctx = alloc_attach_context(); |
ab919e5f | 1281 | if (!ctx) { |
9745eb8a CB |
1282 | lxc_container_put(container); |
1283 | return log_error_errno(-ENOMEM, ENOMEM, "Failed to allocate attach context"); | |
1284 | } | |
1285 | ||
afc691a0 | 1286 | ret = get_attach_context(ctx, container, options); |
9745eb8a | 1287 | if (ret) { |
7e995801 | 1288 | put_attach_context(ctx); |
74ce42b5 | 1289 | return log_error(-1, "Failed to get attach context"); |
9c4693b8 CS |
1290 | } |
1291 | ||
ab919e5f | 1292 | conf = ctx->container->lxc_conf; |
ba773996 | 1293 | |
ab919e5f | 1294 | if (!fetch_seccomp(ctx->container, options)) |
ae026f55 | 1295 | WARN("Failed to get seccomp policy"); |
2c4ea790 | 1296 | |
ab919e5f | 1297 | if (!no_new_privs(ctx->container, options)) |
ae026f55 | 1298 | WARN("Could not determine whether PR_SET_NO_NEW_PRIVS is set"); |
2e812c16 | 1299 | |
b7873c95 CB |
1300 | ret = get_attach_context_nsfds(ctx, options); |
1301 | if (ret) { | |
b7873c95 | 1302 | lxc_container_put(container); |
74ce42b5 | 1303 | return log_error(-1, "Failed to get namespace file descriptors"); |
9c4693b8 CS |
1304 | } |
1305 | ||
9e84479f | 1306 | if (options->attach_flags & LXC_ATTACH_TERMINAL) { |
f797f05e | 1307 | ret = lxc_attach_terminal(name, lxcpath, conf, &terminal); |
ba2be1a8 | 1308 | if (ret < 0) { |
dd53c8af | 1309 | put_attach_context(ctx); |
74ce42b5 | 1310 | return log_error(-1, "Failed to setup new terminal"); |
ba2be1a8 CB |
1311 | } |
1312 | ||
9e84479f | 1313 | terminal.log_fd = options->log_fd; |
c948657b | 1314 | } else { |
9e84479f | 1315 | lxc_terminal_init(&terminal); |
ba2be1a8 CB |
1316 | } |
1317 | ||
8ce83369 CB |
1318 | /* Create a socket pair for IPC communication; set SOCK_CLOEXEC in order |
1319 | * to make sure we don't irritate other threads that want to fork+exec | |
1320 | * away | |
9c4693b8 CS |
1321 | * |
1322 | * IMPORTANT: if the initial process is multithreaded and another call | |
1323 | * just fork()s away without exec'ing directly after, the socket fd will | |
1324 | * exist in the forked process from the other thread and any close() in | |
8ce83369 | 1325 | * our own child process will not really cause the socket to close |
1d801260 | 1326 | * properly, potentially causing the parent to hang. |
9c4693b8 CS |
1327 | * |
1328 | * For this reason, while IPC is still active, we have to use shutdown() | |
8ce83369 CB |
1329 | * if the child exits prematurely in order to signal that the socket is |
1330 | * closed and cannot assume that the child exiting will automatically do | |
1331 | * that. | |
9c4693b8 CS |
1332 | * |
1333 | * IPC mechanism: (X is receiver) | |
1334 | * initial process intermediate attached | |
1335 | * X <--- send pid of | |
1336 | * attached proc, | |
1337 | * then exit | |
1338 | * send 0 ------------------------------------> X | |
1339 | * [do initialization] | |
1340 | * X <------------------------------------ send 1 | |
1341 | * [add to cgroup, ...] | |
1342 | * send 2 ------------------------------------> X | |
81f466d0 CB |
1343 | * [set LXC_ATTACH_NO_NEW_PRIVS] |
1344 | * X <------------------------------------ send 3 | |
1345 | * [open LSM label fd] | |
1346 | * send 4 ------------------------------------> X | |
1347 | * [set LSM label] | |
9c4693b8 CS |
1348 | * close socket close socket |
1349 | * run program | |
1350 | */ | |
1351 | ret = socketpair(PF_LOCAL, SOCK_STREAM | SOCK_CLOEXEC, 0, ipc_sockets); | |
1352 | if (ret < 0) { | |
dd53c8af | 1353 | put_attach_context(ctx); |
74ce42b5 | 1354 | return log_error_errno(-1, errno, "Could not set up required IPC mechanism for attaching"); |
9c4693b8 CS |
1355 | } |
1356 | ||
e3f0e436 CB |
1357 | /* Create intermediate subprocess, two reasons: |
1358 | * 1. We can't setns() in the child itself, since we want to make | |
8ce83369 | 1359 | * sure we are properly attached to the pidns. |
e3f0e436 | 1360 | * 2. Also, the initial thread has to put the attached process |
8ce83369 CB |
1361 | * into the cgroup, which we can only do if we didn't already |
1362 | * setns() (otherwise, user namespaces will hate us). | |
9c4693b8 CS |
1363 | */ |
1364 | pid = fork(); | |
9c4693b8 | 1365 | if (pid < 0) { |
dd53c8af | 1366 | put_attach_context(ctx); |
74ce42b5 | 1367 | return log_error_errno(-1, errno, "Failed to create first subprocess"); |
9c4693b8 CS |
1368 | } |
1369 | ||
4f25e72f | 1370 | if (pid == 0) { |
26abd7ea | 1371 | char *cwd, *new_cwd; |
a588a482 | 1372 | |
ba2be1a8 | 1373 | /* close unneeded file descriptors */ |
4f25e72f | 1374 | close_prot_errno_disarm(ipc_sockets[0]); |
2202afc9 | 1375 | |
4f25e72f CB |
1376 | if (options->attach_flags & LXC_ATTACH_TERMINAL) { |
1377 | lxc_attach_terminal_close_ptx(&terminal); | |
1378 | lxc_attach_terminal_close_peer(&terminal); | |
1379 | lxc_attach_terminal_close_log(&terminal); | |
f4364484 SG |
1380 | } |
1381 | ||
4f25e72f | 1382 | /* Wait for the parent to have setup cgroups. */ |
6e48e7c5 | 1383 | if (!sync_wait(ipc_sockets[1], ATTACH_SYNC_CGROUP)) { |
4f25e72f | 1384 | shutdown(ipc_sockets[1], SHUT_RDWR); |
dd53c8af | 1385 | put_attach_context(ctx); |
4f25e72f | 1386 | _exit(EXIT_FAILURE); |
bb2ada6f CB |
1387 | } |
1388 | ||
c538837d CB |
1389 | if (!attach_context_security_barrier(ctx)) { |
1390 | shutdown(ipc_sockets[1], SHUT_RDWR); | |
1391 | put_attach_context(ctx); | |
1392 | _exit(EXIT_FAILURE); | |
1393 | } | |
1394 | ||
4f25e72f CB |
1395 | TRACE("Intermediate process starting to initialize"); |
1396 | ||
a588a482 CB |
1397 | cwd = getcwd(NULL, 0); |
1398 | ||
c538837d CB |
1399 | /* |
1400 | * Attach now, create another subprocess later, since pid | |
1401 | * namespaces only really affect the children of the current | |
1402 | * process. | |
1403 | * | |
1404 | * Note that this is a crucial barrier. We're no moving into | |
1405 | * the container's context so we need to make sure to not leak | |
1406 | * anything sensitive. That especially means things such as | |
1407 | * open file descriptors! | |
4f25e72f | 1408 | */ |
c538837d | 1409 | ret = attach_context_container(ctx); |
4f25e72f CB |
1410 | if (ret < 0) { |
1411 | ERROR("Failed to enter namespaces"); | |
1412 | shutdown(ipc_sockets[1], SHUT_RDWR); | |
dd53c8af | 1413 | put_attach_context(ctx); |
4f25e72f | 1414 | _exit(EXIT_FAILURE); |
ba2be1a8 CB |
1415 | } |
1416 | ||
4f25e72f | 1417 | /* close namespace file descriptors */ |
7e995801 | 1418 | close_nsfds(ctx); |
ea918412 | 1419 | |
4f25e72f CB |
1420 | /* Attach succeeded, try to cwd. */ |
1421 | if (options->initial_cwd) | |
1422 | new_cwd = options->initial_cwd; | |
1423 | else | |
1424 | new_cwd = cwd; | |
1425 | if (new_cwd) { | |
1426 | ret = chdir(new_cwd); | |
1427 | if (ret < 0) | |
1428 | WARN("Could not change directory to \"%s\"", new_cwd); | |
ba2be1a8 | 1429 | } |
a588a482 | 1430 | free_disarm(cwd); |
c6d09e15 | 1431 | |
4f25e72f | 1432 | /* Create attached process. */ |
4f25e72f CB |
1433 | pid = lxc_raw_clone(CLONE_PARENT, NULL); |
1434 | if (pid < 0) { | |
1435 | SYSERROR("Failed to clone attached process"); | |
1436 | shutdown(ipc_sockets[1], SHUT_RDWR); | |
dd53c8af | 1437 | put_attach_context(ctx); |
4f25e72f CB |
1438 | _exit(EXIT_FAILURE); |
1439 | } | |
f4364484 | 1440 | |
4f25e72f | 1441 | if (pid == 0) { |
338b230f | 1442 | struct attach_payload ap = { |
a64902ab CB |
1443 | .ipc_socket = ipc_sockets[1], |
1444 | .options = options, | |
1445 | .ctx = ctx, | |
1446 | .terminal_pts_fd = terminal.pty, | |
1447 | .exec_function = exec_function, | |
1448 | .exec_payload = exec_payload, | |
1449 | }; | |
1450 | ||
4f25e72f CB |
1451 | if (options->attach_flags & LXC_ATTACH_TERMINAL) { |
1452 | ret = lxc_terminal_signal_sigmask_safe_blocked(&terminal); | |
1453 | if (ret < 0) { | |
1454 | SYSERROR("Failed to reset signal mask"); | |
1455 | _exit(EXIT_FAILURE); | |
1456 | } | |
1457 | } | |
ea918412 | 1458 | |
a64902ab | 1459 | /* Does not return. */ |
338b230f | 1460 | do_attach(&ap); |
62183f1a | 1461 | } |
2eef2bda | 1462 | |
4f25e72f CB |
1463 | if (options->attach_flags & LXC_ATTACH_TERMINAL) |
1464 | lxc_attach_terminal_close_pts(&terminal); | |
ea918412 | 1465 | |
4f25e72f | 1466 | /* Tell grandparent the pid of the pid of the newly created child. */ |
f8e88e94 | 1467 | if (!sync_wake_pid(ipc_sockets[1], ATTACH_SYNC_PID(pid))) { |
4f25e72f CB |
1468 | /* If this really happens here, this is very unfortunate, since |
1469 | * the parent will not know the pid of the attached process and | |
1470 | * will not be able to wait for it (and we won't either due to | |
1471 | * CLONE_PARENT) so the parent won't be able to reap it and the | |
1472 | * attached process will remain a zombie. | |
1473 | */ | |
1474 | shutdown(ipc_sockets[1], SHUT_RDWR); | |
dd53c8af | 1475 | put_attach_context(ctx); |
4f25e72f CB |
1476 | _exit(EXIT_FAILURE); |
1477 | } | |
9c4693b8 | 1478 | |
4f25e72f | 1479 | TRACE("Sending pid %d of attached process", pid); |
9c4693b8 | 1480 | |
4f25e72f | 1481 | /* The rest is in the hands of the initial and the attached process. */ |
dd53c8af | 1482 | put_attach_context(ctx); |
4f25e72f CB |
1483 | _exit(EXIT_SUCCESS); |
1484 | } | |
6f4f1937 | 1485 | |
4f25e72f | 1486 | to_cleanup_pid = pid; |
ea918412 | 1487 | |
4f25e72f | 1488 | /* close unneeded file descriptors */ |
cb2420df | 1489 | close_prot_errno_disarm(ipc_sockets[1]); |
7e995801 | 1490 | close_nsfds(ctx); |
4f25e72f CB |
1491 | if (options->attach_flags & LXC_ATTACH_TERMINAL) |
1492 | lxc_attach_terminal_close_pts(&terminal); | |
81f466d0 | 1493 | |
4f25e72f CB |
1494 | /* Attach to cgroup, if requested. */ |
1495 | if (options->attach_flags & LXC_ATTACH_MOVE_TO_CGROUP) { | |
1496 | /* | |
1497 | * If this is the unified hierarchy cgroup_attach() is | |
1498 | * enough. | |
1499 | */ | |
1500 | ret = cgroup_attach(conf, name, lxcpath, pid); | |
1501 | if (ret) { | |
1502 | call_cleaner(cgroup_exit) struct cgroup_ops *cgroup_ops = NULL; | |
9044b79e | 1503 | |
4f25e72f CB |
1504 | cgroup_ops = cgroup_init(conf); |
1505 | if (!cgroup_ops) | |
1506 | goto on_error; | |
9044b79e | 1507 | |
4f25e72f CB |
1508 | if (!cgroup_ops->attach(cgroup_ops, conf, name, lxcpath, pid)) |
1509 | goto on_error; | |
81f466d0 | 1510 | } |
4f25e72f CB |
1511 | TRACE("Moved intermediate process %d into container's cgroups", pid); |
1512 | } | |
81f466d0 | 1513 | |
4f25e72f CB |
1514 | /* Setup /proc limits */ |
1515 | if (!lxc_list_empty(&conf->procs)) { | |
1516 | ret = setup_proc_filesystem(&conf->procs, pid); | |
1517 | if (ret < 0) | |
1518 | goto on_error; | |
4f3b6a85 CB |
1519 | |
1520 | TRACE("Setup /proc/%d settings", pid); | |
4f25e72f | 1521 | } |
cdb2a47f | 1522 | |
4f25e72f CB |
1523 | /* Setup resource limits */ |
1524 | if (!lxc_list_empty(&conf->limits)) { | |
1525 | ret = setup_resource_limits(&conf->limits, pid); | |
1526 | if (ret < 0) | |
1527 | goto on_error; | |
4f3b6a85 CB |
1528 | |
1529 | TRACE("Setup resource limits"); | |
4f25e72f | 1530 | } |
cdb2a47f | 1531 | |
4f25e72f CB |
1532 | if (options->attach_flags & LXC_ATTACH_TERMINAL) { |
1533 | ret = lxc_attach_terminal_mainloop_init(&terminal, &descr); | |
1534 | if (ret < 0) | |
1535 | goto on_error; | |
9c4693b8 | 1536 | |
4f25e72f CB |
1537 | TRACE("Initialized terminal mainloop"); |
1538 | } | |
9c4693b8 | 1539 | |
4f25e72f | 1540 | /* Let the child process know to go ahead. */ |
6e48e7c5 | 1541 | if (!sync_wake(ipc_sockets[0], ATTACH_SYNC_CGROUP)) |
4f25e72f | 1542 | goto close_mainloop; |
ba2be1a8 | 1543 | |
4f25e72f | 1544 | TRACE("Told intermediate process to start initializing"); |
ea918412 | 1545 | |
4f25e72f | 1546 | /* Get pid of attached process from intermediate process. */ |
f8e88e94 | 1547 | if (!sync_wait_pid(ipc_sockets[0], ATTACH_SYNC_PID(&attached_pid))) |
4f25e72f | 1548 | goto close_mainloop; |
ba2be1a8 | 1549 | |
4f25e72f | 1550 | TRACE("Received pid %d of attached process in parent pid namespace", attached_pid); |
ba2be1a8 | 1551 | |
4f25e72f | 1552 | /* Ignore SIGKILL (CTRL-C) and SIGQUIT (CTRL-\) - issue #313. */ |
5d2b46fb | 1553 | if (options->stdin_fd == STDIN_FILENO) { |
4f25e72f CB |
1554 | signal(SIGINT, SIG_IGN); |
1555 | signal(SIGQUIT, SIG_IGN); | |
1556 | } | |
ba2be1a8 | 1557 | |
4f25e72f CB |
1558 | /* Reap intermediate process. */ |
1559 | ret = wait_for_pid(pid); | |
1560 | if (ret < 0) | |
1561 | goto close_mainloop; | |
ba2be1a8 | 1562 | |
4f25e72f | 1563 | TRACE("Intermediate process %d exited", pid); |
ea918412 | 1564 | |
4f25e72f CB |
1565 | /* We will always have to reap the attached process now. */ |
1566 | to_cleanup_pid = attached_pid; | |
9c4693b8 | 1567 | |
4f25e72f | 1568 | /* Open LSM fd and send it to child. */ |
afc691a0 | 1569 | if (attach_lsm(options) && ctx->lsm_label) { |
ad001fb6 | 1570 | __do_close int labelfd = -EBADF; |
4f25e72f | 1571 | bool on_exec; |
ea918412 | 1572 | |
4f25e72f | 1573 | on_exec = options->attach_flags & LXC_ATTACH_LSM_EXEC ? true : false; |
afc691a0 | 1574 | labelfd = ctx->lsm_ops->process_label_fd_get(ctx->lsm_ops, attached_pid, on_exec); |
4f25e72f CB |
1575 | if (labelfd < 0) |
1576 | goto close_mainloop; | |
9c4693b8 | 1577 | |
4f25e72f | 1578 | TRACE("Opened LSM label file descriptor %d", labelfd); |
ea918412 | 1579 | |
4f25e72f | 1580 | /* Send child fd of the LSM security module to write to. */ |
f8e88e94 | 1581 | if (!sync_wake_fd(ipc_sockets[0], ATTACH_SYNC_LSM(labelfd))) { |
6e36c297 | 1582 | SYSERROR("Failed to send lsm label fd"); |
4f25e72f CB |
1583 | goto close_mainloop; |
1584 | } | |
1585 | ||
4f25e72f | 1586 | TRACE("Sent LSM label file descriptor %d to child", labelfd); |
9c4693b8 | 1587 | } |
ea918412 | 1588 | |
4f25e72f CB |
1589 | if (conf->seccomp.seccomp) { |
1590 | ret = lxc_seccomp_recv_notifier_fd(&conf->seccomp, ipc_sockets[0]); | |
1591 | if (ret < 0) | |
1592 | goto close_mainloop; | |
9c4693b8 | 1593 | |
4f25e72f | 1594 | ret = lxc_seccomp_add_notifier(name, lxcpath, &conf->seccomp); |
d6d979bc | 1595 | if (ret < 0) |
4f25e72f | 1596 | goto close_mainloop; |
d6d979bc | 1597 | } |
9c4693b8 | 1598 | |
4f25e72f CB |
1599 | /* We're done, the child process should now execute whatever it |
1600 | * is that the user requested. The parent can now track it with | |
1601 | * waitpid() or similar. | |
1602 | */ | |
9c4693b8 | 1603 | |
4f25e72f | 1604 | *attached_process = attached_pid; |
a998454a | 1605 | |
4f25e72f CB |
1606 | /* Now shut down communication with child, we're done. */ |
1607 | shutdown(ipc_sockets[0], SHUT_RDWR); | |
cb2420df | 1608 | close_prot_errno_disarm(ipc_sockets[0]); |
f157b056 | 1609 | |
4f25e72f CB |
1610 | ret_parent = 0; |
1611 | to_cleanup_pid = -1; | |
ea918412 | 1612 | |
4f25e72f CB |
1613 | if (options->attach_flags & LXC_ATTACH_TERMINAL) { |
1614 | ret = lxc_mainloop(&descr, -1); | |
1615 | if (ret < 0) { | |
1616 | ret_parent = -1; | |
1617 | to_cleanup_pid = attached_pid; | |
1618 | } | |
a998454a | 1619 | } |
ea918412 | 1620 | |
4f25e72f | 1621 | close_mainloop: |
9e84479f | 1622 | if (options->attach_flags & LXC_ATTACH_TERMINAL) |
4f25e72f | 1623 | lxc_mainloop_close(&descr); |
9c4693b8 | 1624 | |
4f25e72f CB |
1625 | on_error: |
1626 | if (ipc_sockets[0] >= 0) { | |
1627 | shutdown(ipc_sockets[0], SHUT_RDWR); | |
cb2420df | 1628 | close_prot_errno_disarm(ipc_sockets[0]); |
9c4693b8 | 1629 | } |
ea918412 | 1630 | |
4f25e72f CB |
1631 | if (to_cleanup_pid > 0) |
1632 | (void)wait_for_pid(to_cleanup_pid); | |
1633 | ||
1634 | if (options->attach_flags & LXC_ATTACH_TERMINAL) { | |
1635 | lxc_terminal_delete(&terminal); | |
1636 | lxc_terminal_conf_free(&terminal); | |
1637 | } | |
9c4693b8 | 1638 | |
dd53c8af | 1639 | put_attach_context(ctx); |
4f25e72f | 1640 | return ret_parent; |
9c4693b8 CS |
1641 | } |
1642 | ||
06346bb0 | 1643 | int lxc_attach_run_command(void *payload) |
9c4693b8 | 1644 | { |
06346bb0 CB |
1645 | int ret = -1; |
1646 | lxc_attach_command_t *cmd = payload; | |
9c4693b8 | 1647 | |
06346bb0 CB |
1648 | ret = execvp(cmd->program, cmd->argv); |
1649 | if (ret < 0) { | |
1650 | switch (errno) { | |
1651 | case ENOEXEC: | |
1652 | ret = 126; | |
cf0fd972 | 1653 | break; |
06346bb0 CB |
1654 | case ENOENT: |
1655 | ret = 127; | |
cf0fd972 | 1656 | break; |
06346bb0 CB |
1657 | } |
1658 | } | |
ea918412 | 1659 | |
c2af3a15 | 1660 | return log_error_errno(ret, errno, "Failed to exec \"%s\"", cmd->program); |
9c4693b8 CS |
1661 | } |
1662 | ||
1663 | int lxc_attach_run_shell(void* payload) | |
1664 | { | |
cd8f5663 | 1665 | __do_free char *buf = NULL; |
9c4693b8 | 1666 | uid_t uid; |
cb7aa5e8 DJ |
1667 | struct passwd pwent; |
1668 | struct passwd *pwentp = NULL; | |
9c4693b8 | 1669 | char *user_shell; |
cb7aa5e8 DJ |
1670 | size_t bufsize; |
1671 | int ret; | |
9c4693b8 | 1672 | |
8ce83369 | 1673 | /* Ignore payload parameter. */ |
9c4693b8 CS |
1674 | (void)payload; |
1675 | ||
1676 | uid = getuid(); | |
cb7aa5e8 DJ |
1677 | |
1678 | bufsize = sysconf(_SC_GETPW_R_SIZE_MAX); | |
1679 | if (bufsize == -1) | |
1680 | bufsize = 1024; | |
1681 | ||
1682 | buf = malloc(bufsize); | |
1683 | if (buf) { | |
1684 | ret = getpwuid_r(uid, &pwent, buf, bufsize, &pwentp); | |
1685 | if (!pwentp) { | |
1686 | if (ret == 0) | |
ea918412 | 1687 | WARN("Could not find matched password record"); |
cb7aa5e8 DJ |
1688 | |
1689 | WARN("Failed to get password record - %u", uid); | |
1690 | } | |
1691 | } | |
9c4693b8 | 1692 | |
8ce83369 CB |
1693 | /* This probably happens because of incompatible nss implementations in |
1694 | * host and container (remember, this code is still using the host's | |
1695 | * glibc but our mount namespace is in the container) we may try to get | |
1696 | * the information by spawning a [getent passwd uid] process and parsing | |
1697 | * the result. | |
9c4693b8 | 1698 | */ |
cb7aa5e8 | 1699 | if (!pwentp) |
9c4693b8 CS |
1700 | user_shell = lxc_attach_getpwshell(uid); |
1701 | else | |
cb7aa5e8 | 1702 | user_shell = pwent.pw_shell; |
ea918412 | 1703 | |
9c4693b8 | 1704 | if (user_shell) |
acf47e1b | 1705 | execlp(user_shell, user_shell, (char *)NULL); |
9c4693b8 | 1706 | |
8ce83369 CB |
1707 | /* Executed if either no passwd entry or execvp fails, we will fall back |
1708 | * on /bin/sh as a default shell. | |
9c4693b8 | 1709 | */ |
acf47e1b | 1710 | execlp("/bin/sh", "/bin/sh", (char *)NULL); |
ea918412 | 1711 | |
edeb1836 | 1712 | SYSERROR("Failed to execute shell"); |
cb7aa5e8 | 1713 | if (!pwentp) |
edeb1836 | 1714 | free(user_shell); |
ea918412 | 1715 | |
9c4693b8 CS |
1716 | return -1; |
1717 | } |