]>
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. | |
3ac4480a CB |
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_ns_uid : if CLONE_NEWUSER is specified will contain the uid used | |
68 | * during attach setup. | |
69 | * @setup_ns_gid : if CLONE_NEWUSER is specified will contain the gid used | |
70 | * during attach setup. | |
71 | * @target_ns_uid : if CLONE_NEWUSER is specified the uid that the final | |
72 | * program will be run with. | |
73 | * @target_ns_gid : if CLONE_NEWUSER is specified the gid that the final | |
74 | * program will be run with. | |
75 | * @target_host_uid : if CLONE_NEWUSER is specified the uid that the final | |
76 | * program will be run with on the host. | |
77 | * @target_host_gid : if CLONE_NEWUSER is specified the gid that the final | |
78 | * program will be run with on the host. | |
79 | * @lsm_label : LSM label to be used for the attaching process | |
80 | * @container : the container we're attaching o | |
81 | * @personality : the personality to use for the final program | |
82 | * @capability : the capability mask of the @init_pid | |
83 | * @ns_inherited : flags of namespaces that the final program will inherit | |
84 | * from @init_pid | |
85 | * @ns_fd : file descriptors to @init_pid's namespaces | |
20718e39 | 86 | */ |
ab919e5f | 87 | struct attach_context { |
afc691a0 | 88 | unsigned int attach_flags; |
500ed813 | 89 | int init_pid; |
9b31ab58 | 90 | int init_pidfd; |
25c659d5 CB |
91 | int dfd_init_pid; |
92 | int dfd_self_pid; | |
3ac4480a CB |
93 | uid_t setup_ns_uid; |
94 | gid_t setup_ns_gid; | |
95 | uid_t target_ns_uid; | |
96 | gid_t target_ns_gid; | |
97 | uid_t target_host_uid; | |
98 | uid_t target_host_gid; | |
0e304baa CB |
99 | char *lsm_label; |
100 | struct lxc_container *container; | |
101 | signed long personality; | |
102 | unsigned long long capability_mask; | |
103 | int ns_inherited; | |
104 | int ns_fd[LXC_NS_MAX]; | |
105 | struct lsm_ops *lsm_ops; | |
106 | }; | |
107 | ||
6f0c2cea | 108 | static pid_t pidfd_get_pid(int dfd_init_pid, int pidfd) |
d8764025 CB |
109 | { |
110 | __do_free char *line = NULL; | |
111 | __do_fclose FILE *f = NULL; | |
112 | size_t len = 0; | |
6f0c2cea | 113 | char path[STRLITERALLEN("fdinfo/") + INTTYPE_TO_STRLEN(int) + 1 ] = "fdinfo/"; |
d8764025 CB |
114 | int ret; |
115 | ||
6f0c2cea CB |
116 | if (dfd_init_pid < 0 || pidfd < 0) |
117 | return ret_errno(EBADF); | |
d8764025 | 118 | |
6f0c2cea | 119 | ret = snprintf(path + STRLITERALLEN("fdinfo/"), INTTYPE_TO_STRLEN(int), "%d", pidfd); |
d8764025 CB |
120 | if (ret < 0 || ret > (size_t)INTTYPE_TO_STRLEN(int)) |
121 | return ret_errno(EIO); | |
122 | ||
6f0c2cea | 123 | f = fdopen_at(dfd_init_pid, path, "re", PROTECT_OPEN, PROTECT_LOOKUP_BENEATH); |
d8764025 CB |
124 | if (!f) |
125 | return -errno; | |
126 | ||
127 | while (getline(&line, &len, f) != -1) { | |
128 | const char *prefix = "Pid:\t"; | |
129 | const size_t prefix_len = STRLITERALLEN("Pid:\t"); | |
130 | int pid = -ESRCH; | |
131 | char *slider = line; | |
132 | ||
133 | if (strncmp(slider, prefix, prefix_len)) | |
134 | continue; | |
135 | ||
136 | slider += prefix_len; | |
137 | slider = lxc_trim_whitespace_in_place(slider); | |
138 | ||
139 | ret = lxc_safe_int(slider, &pid); | |
140 | if (ret) | |
141 | return -ret; | |
142 | ||
143 | return pid; | |
144 | } | |
145 | ||
146 | return ret_errno(ENOENT); | |
147 | } | |
148 | ||
6e36c297 CB |
149 | static inline bool sync_wake_pid(int fd, pid_t pid) |
150 | { | |
151 | return lxc_write_nointr(fd, &pid, sizeof(pid_t)) == sizeof(pid_t); | |
152 | } | |
153 | ||
154 | static inline bool sync_wait_pid(int fd, pid_t *pid) | |
155 | { | |
156 | return lxc_read_nointr(fd, pid, sizeof(pid_t)) == sizeof(pid_t); | |
157 | } | |
158 | ||
159 | static inline bool sync_wake_fd(int fd, int fd_send) | |
160 | { | |
161 | return lxc_abstract_unix_send_fds(fd, &fd_send, 1, NULL, 0) > 0; | |
162 | } | |
163 | ||
164 | static inline bool sync_wait_fd(int fd, int *fd_recv) | |
165 | { | |
166 | return lxc_abstract_unix_recv_fds(fd, fd_recv, 1, NULL, 0) > 0; | |
167 | } | |
168 | ||
afc691a0 CB |
169 | static bool attach_lsm(lxc_attach_options_t *options) |
170 | { | |
171 | return (options->namespaces & CLONE_NEWNS) && | |
172 | (options->attach_flags & (LXC_ATTACH_LSM | LXC_ATTACH_LSM_LABEL)); | |
173 | } | |
174 | ||
9745eb8a CB |
175 | static struct attach_context *alloc_attach_context(void) |
176 | { | |
581b849a CB |
177 | struct attach_context *ctx; |
178 | ||
179 | ctx = zalloc(sizeof(struct attach_context)); | |
180 | if (!ctx) | |
181 | return ret_set_errno(NULL, ENOMEM); | |
182 | ||
9b31ab58 CB |
183 | ctx->dfd_self_pid = -EBADF; |
184 | ctx->dfd_init_pid = -EBADF; | |
185 | ctx->init_pidfd = -EBADF; | |
186 | ctx->init_pid = -ESRCH; | |
187 | ctx->setup_ns_uid = LXC_INVALID_UID; | |
188 | ctx->setup_ns_gid = LXC_INVALID_GID; | |
189 | ctx->target_ns_uid = LXC_INVALID_UID; | |
190 | ctx->target_ns_gid = LXC_INVALID_GID; | |
191 | ctx->target_host_uid = LXC_INVALID_UID; | |
192 | ctx->target_host_gid = LXC_INVALID_GID; | |
581b849a CB |
193 | |
194 | for (int i = 0; i < LXC_NS_MAX; i++) | |
195 | ctx->ns_fd[i] = -EBADF; | |
196 | ||
197 | return ctx; | |
9745eb8a CB |
198 | } |
199 | ||
ee142207 CB |
200 | static int get_personality(const char *name, const char *lxcpath, |
201 | signed long *personality) | |
d92c8e40 CB |
202 | { |
203 | __do_free char *p = NULL; | |
ee142207 | 204 | signed long per; |
d92c8e40 CB |
205 | |
206 | p = lxc_cmd_get_config_item(name, "lxc.arch", lxcpath); | |
ee142207 CB |
207 | if (!p) { |
208 | *personality = LXC_ARCH_UNCHANGED; | |
209 | return 0; | |
210 | } | |
d92c8e40 | 211 | |
ee142207 CB |
212 | per = lxc_config_parse_arch(p); |
213 | if (per == LXC_ARCH_UNCHANGED) | |
214 | return ret_errno(EINVAL); | |
215 | ||
216 | *personality = per; | |
217 | return 0; | |
d92c8e40 CB |
218 | } |
219 | ||
4475fabb | 220 | static int userns_setup_ids(struct attach_context *ctx, |
3ac4480a | 221 | lxc_attach_options_t *options) |
4475fabb CB |
222 | { |
223 | __do_free char *line = NULL; | |
224 | __do_fclose FILE *f_gidmap = NULL, *f_uidmap = NULL; | |
225 | size_t len = 0; | |
226 | uid_t init_ns_uid = LXC_INVALID_UID; | |
227 | gid_t init_ns_gid = LXC_INVALID_GID; | |
228 | uid_t nsuid, hostuid, range_uid; | |
229 | gid_t nsgid, hostgid, range_gid; | |
230 | ||
231 | if (!(options->namespaces & CLONE_NEWUSER)) | |
232 | return 0; | |
233 | ||
72a19d2f | 234 | f_uidmap = fdopen_at(ctx->dfd_init_pid, "uid_map", "re", PROTECT_OPEN, PROTECT_LOOKUP_BENEATH); |
4475fabb CB |
235 | if (!f_uidmap) |
236 | return log_error_errno(-errno, errno, "Failed to open uid_map"); | |
237 | ||
238 | while (getline(&line, &len, f_uidmap) != -1) { | |
239 | if (sscanf(line, "%u %u %u", &nsuid, &hostuid, &range_uid) != 3) | |
240 | continue; | |
241 | ||
242 | if (0 >= nsuid && 0 < nsuid + range_uid) { | |
3ac4480a | 243 | ctx->setup_ns_uid = 0; |
4475fabb CB |
244 | TRACE("Container has mapping for uid 0"); |
245 | break; | |
246 | } | |
247 | ||
3ac4480a CB |
248 | if (ctx->target_host_uid >= hostuid && ctx->target_host_uid < hostuid + range_uid) { |
249 | init_ns_uid = (ctx->target_host_uid - hostuid) + nsuid; | |
4475fabb CB |
250 | TRACE("Container runs with uid %d", init_ns_uid); |
251 | } | |
252 | } | |
253 | ||
72a19d2f | 254 | f_gidmap = fdopen_at(ctx->dfd_init_pid, "gid_map", "re", PROTECT_OPEN, PROTECT_LOOKUP_BENEATH); |
4475fabb CB |
255 | if (!f_gidmap) |
256 | return log_error_errno(-errno, errno, "Failed to open gid_map"); | |
257 | ||
258 | while (getline(&line, &len, f_gidmap) != -1) { | |
259 | if (sscanf(line, "%u %u %u", &nsgid, &hostgid, &range_gid) != 3) | |
260 | continue; | |
261 | ||
262 | if (0 >= nsgid && 0 < nsgid + range_gid) { | |
3ac4480a | 263 | ctx->setup_ns_gid = 0; |
4475fabb CB |
264 | TRACE("Container has mapping for gid 0"); |
265 | break; | |
266 | } | |
267 | ||
3ac4480a CB |
268 | if (ctx->target_host_gid >= hostgid && ctx->target_host_gid < hostgid + range_gid) { |
269 | init_ns_gid = (ctx->target_host_gid - hostgid) + nsgid; | |
4475fabb CB |
270 | TRACE("Container runs with gid %d", init_ns_gid); |
271 | } | |
272 | } | |
273 | ||
3ac4480a CB |
274 | if (ctx->setup_ns_uid == LXC_INVALID_UID) |
275 | ctx->setup_ns_uid = init_ns_uid; | |
4475fabb | 276 | |
3ac4480a CB |
277 | if (ctx->setup_ns_gid == LXC_INVALID_UID) |
278 | ctx->setup_ns_gid = init_ns_gid; | |
4475fabb CB |
279 | |
280 | /* | |
281 | * TODO: we should also parse supplementary groups and use | |
282 | * setgroups() to set them. | |
283 | */ | |
284 | ||
285 | return 0; | |
286 | } | |
287 | ||
288 | static void userns_target_ids(struct attach_context *ctx, lxc_attach_options_t *options) | |
289 | { | |
290 | if (options->uid != LXC_INVALID_UID) | |
3ac4480a | 291 | ctx->target_ns_uid = options->uid; |
4475fabb | 292 | else if (options->namespaces & CLONE_NEWUSER) |
3ac4480a | 293 | ctx->target_ns_uid = ctx->setup_ns_uid; |
4475fabb | 294 | else |
3ac4480a | 295 | ctx->target_ns_uid = 0; |
4475fabb | 296 | |
3ac4480a | 297 | if (ctx->target_ns_uid == LXC_INVALID_UID) |
4475fabb CB |
298 | WARN("Invalid uid specified"); |
299 | ||
300 | if (options->gid != LXC_INVALID_GID) | |
3ac4480a | 301 | ctx->target_ns_gid = options->gid; |
4475fabb | 302 | else if (options->namespaces & CLONE_NEWUSER) |
3ac4480a | 303 | ctx->target_ns_gid = ctx->setup_ns_gid; |
4475fabb | 304 | else |
3ac4480a | 305 | ctx->target_ns_gid = 0; |
4475fabb | 306 | |
3ac4480a | 307 | if (ctx->target_ns_gid == LXC_INVALID_GID) |
4475fabb CB |
308 | WARN("Invalid gid specified"); |
309 | } | |
310 | ||
9680e7b0 CB |
311 | static int parse_init_status(struct attach_context *ctx, lxc_attach_options_t *options) |
312 | { | |
313 | __do_free char *line = NULL; | |
314 | __do_fclose FILE *f = NULL; | |
315 | size_t len = 0; | |
316 | bool caps_found = false; | |
4475fabb | 317 | int ret; |
9680e7b0 | 318 | |
72a19d2f | 319 | f = fdopen_at(ctx->dfd_init_pid, "status", "re", PROTECT_OPEN, PROTECT_LOOKUP_BENEATH); |
9680e7b0 | 320 | if (!f) |
4475fabb | 321 | return log_error_errno(-errno, errno, "Failed to open status file"); |
9680e7b0 CB |
322 | |
323 | while (getline(&line, &len, f) != -1) { | |
324 | signed long value = -1; | |
9680e7b0 | 325 | |
4475fabb CB |
326 | /* |
327 | * Format is: real, effective, saved set user, fs we only care | |
328 | * about real uid. | |
329 | */ | |
330 | ret = sscanf(line, "Uid: %ld", &value); | |
331 | if (ret != EOF && ret == 1) { | |
3ac4480a CB |
332 | ctx->target_host_uid = (uid_t)value; |
333 | TRACE("Container's init process runs with hostuid %d", ctx->target_host_uid); | |
4475fabb CB |
334 | goto next; |
335 | } | |
9680e7b0 | 336 | |
4475fabb CB |
337 | ret = sscanf(line, "Gid: %ld", &value); |
338 | if (ret != EOF && ret == 1) { | |
3ac4480a CB |
339 | ctx->target_host_gid = (gid_t)value; |
340 | TRACE("Container's init process runs with hostgid %d", ctx->target_host_gid); | |
4475fabb | 341 | goto next; |
9680e7b0 CB |
342 | } |
343 | ||
344 | ret = sscanf(line, "CapBnd: %llx", &ctx->capability_mask); | |
345 | if (ret != EOF && ret == 1) { | |
346 | caps_found = true; | |
347 | goto next; | |
348 | } | |
349 | ||
350 | next: | |
3ac4480a CB |
351 | if (ctx->target_host_uid != LXC_INVALID_UID && |
352 | ctx->target_host_gid != LXC_INVALID_GID && | |
4475fabb | 353 | caps_found) |
9680e7b0 CB |
354 | break; |
355 | ||
356 | } | |
357 | ||
3ac4480a | 358 | ret = userns_setup_ids(ctx, options); |
4475fabb CB |
359 | if (ret) |
360 | return log_error_errno(ret, errno, "Failed to get setup ids"); | |
361 | userns_target_ids(ctx, options); | |
362 | ||
9680e7b0 CB |
363 | /* |
364 | * TODO: we should also parse supplementary groups and use | |
365 | * setgroups() to set them. | |
366 | */ | |
367 | ||
368 | return 0; | |
369 | } | |
370 | ||
9b31ab58 CB |
371 | static bool pidfd_setns_supported(struct attach_context *ctx) |
372 | { | |
373 | int ret; | |
374 | ||
375 | /* | |
376 | * The ability to attach to time namespaces came after the introduction | |
377 | * of of using pidfds for attaching to namespaces. To avoid having to | |
378 | * special-case both CLONE_NEWUSER and CLONE_NEWTIME handling, let's | |
379 | * use CLONE_NEWTIME as gatekeeper. | |
380 | */ | |
381 | if (ctx->init_pidfd >= 0) | |
382 | ret = setns(ctx->init_pidfd, CLONE_NEWTIME); | |
383 | else | |
384 | ret = -EOPNOTSUPP; | |
385 | TRACE("Attaching to namespaces via pidfds %s", | |
386 | ret ? "unsupported" : "supported"); | |
387 | return ret == 0; | |
388 | } | |
389 | ||
500ed813 | 390 | static int get_attach_context(struct attach_context *ctx, |
afc691a0 CB |
391 | struct lxc_container *container, |
392 | lxc_attach_options_t *options) | |
e0732705 | 393 | { |
9680e7b0 | 394 | __do_free char *lsm_label = NULL; |
6f4f1937 | 395 | int ret; |
c538837d | 396 | char path[LXC_PROC_PID_LEN]; |
e0732705 | 397 | |
500ed813 | 398 | ctx->container = container; |
afc691a0 | 399 | ctx->attach_flags = options->attach_flags; |
500ed813 | 400 | |
6f0c2cea CB |
401 | ctx->dfd_self_pid = open_at(-EBADF, "/proc/self", |
402 | PROTECT_OPATH_FILE & ~O_NOFOLLOW, | |
403 | (PROTECT_LOOKUP_ABSOLUTE_WITH_SYMLINKS & ~RESOLVE_NO_XDEV), 0); | |
404 | if (ctx->dfd_self_pid < 0) | |
405 | return log_error_errno(-errno, errno, "Failed to open /proc/self"); | |
406 | ||
9b31ab58 CB |
407 | ctx->init_pidfd = lxc_cmd_get_init_pidfd(container->name, container->config_path); |
408 | if (ctx->init_pidfd >= 0) | |
409 | ctx->init_pid = pidfd_get_pid(ctx->dfd_self_pid, ctx->init_pidfd); | |
d8764025 CB |
410 | else |
411 | ctx->init_pid = lxc_cmd_get_init_pid(container->name, container->config_path); | |
412 | ||
500ed813 CB |
413 | if (ctx->init_pid < 0) |
414 | return log_error(-1, "Failed to get init pid"); | |
415 | ||
c538837d CB |
416 | ret = snprintf(path, sizeof(path), "/proc/%d", ctx->init_pid); |
417 | if (ret < 0 || ret >= sizeof(path)) | |
418 | return ret_errno(EIO); | |
e0732705 | 419 | |
5129b2d3 CB |
420 | ctx->dfd_init_pid = open_at(-EBADF, path, |
421 | PROTECT_OPATH_DIRECTORY, | |
422 | (PROTECT_LOOKUP_ABSOLUTE & ~RESOLVE_NO_XDEV), 0); | |
9680e7b0 | 423 | if (ctx->dfd_init_pid < 0) |
4475fabb | 424 | return log_error_errno(-errno, errno, "Failed to open /proc/%d", ctx->init_pid); |
c538837d | 425 | |
9b31ab58 CB |
426 | if (ctx->init_pidfd >= 0) { |
427 | ret = lxc_raw_pidfd_send_signal(ctx->init_pidfd, 0, NULL, 0); | |
d8764025 CB |
428 | if (ret) |
429 | return log_error_errno(-errno, errno, "Container process exited or PID has been recycled"); | |
430 | else | |
431 | TRACE("Container process still running and PID was not recycled"); | |
9b31ab58 CB |
432 | |
433 | if (!pidfd_setns_supported(ctx)) { | |
434 | /* We can't risk leaking file descriptors during attach. */ | |
435 | if (close(ctx->init_pidfd)) | |
436 | return log_error_errno(-errno, errno, "Failed to close pidfd"); | |
437 | ||
438 | ctx->init_pidfd = -EBADF; | |
439 | TRACE("Attaching to namespaces via pidfds not supported"); | |
440 | } | |
d8764025 CB |
441 | } |
442 | ||
4475fabb CB |
443 | /* Determine which namespaces the container was created with. */ |
444 | if (options->namespaces == -1) { | |
445 | options->namespaces = lxc_cmd_get_clone_flags(container->name, container->config_path); | |
446 | if (options->namespaces == -1) | |
447 | return log_error_errno(-EINVAL, EINVAL, "Failed to automatically determine the namespaces which the container uses"); | |
448 | ||
449 | for (int i = 0; i < LXC_NS_MAX; i++) { | |
450 | if (ns_info[i].clone_flag & CLONE_NEWCGROUP) | |
451 | if (!(options->attach_flags & LXC_ATTACH_MOVE_TO_CGROUP) || | |
452 | !cgns_supported()) | |
453 | continue; | |
454 | ||
455 | if (ns_info[i].clone_flag & options->namespaces) | |
456 | continue; | |
457 | ||
458 | ctx->ns_inherited |= ns_info[i].clone_flag; | |
459 | } | |
460 | } | |
461 | ||
9680e7b0 CB |
462 | ret = parse_init_status(ctx, options); |
463 | if (ret) | |
4475fabb | 464 | return log_error_errno(-errno, errno, "Failed to open parse file"); |
e0732705 | 465 | |
4eb19ac0 | 466 | ctx->lsm_ops = lsm_init_static(); |
d701d729 | 467 | |
afc691a0 CB |
468 | if (attach_lsm(options)) { |
469 | if (ctx->attach_flags & LXC_ATTACH_LSM_LABEL) | |
470 | lsm_label = options->lsm_label; | |
471 | else | |
9680e7b0 | 472 | lsm_label = ctx->lsm_ops->process_label_get_at(ctx->lsm_ops, ctx->dfd_init_pid); |
afc691a0 CB |
473 | if (!lsm_label) |
474 | WARN("No security context received"); | |
475 | else | |
476 | INFO("Retrieved security context %s", lsm_label); | |
477 | } | |
e0732705 | 478 | |
ee142207 CB |
479 | ret = get_personality(container->name, container->config_path, &ctx->personality); |
480 | if (ret) | |
481 | return log_error_errno(ret, errno, "Failed to get personality of the container"); | |
d92c8e40 | 482 | |
1874ef74 CB |
483 | if (!ctx->container->lxc_conf) { |
484 | ctx->container->lxc_conf = lxc_conf_init(); | |
485 | if (!ctx->container->lxc_conf) | |
486 | return log_error_errno(-ENOMEM, ENOMEM, "Failed to allocate new lxc config"); | |
487 | } | |
488 | ||
afc691a0 | 489 | ctx->lsm_label = move_ptr(lsm_label); |
9745eb8a | 490 | return 0; |
e0732705 CS |
491 | } |
492 | ||
9b31ab58 | 493 | static int same_nsfd(int dfd_pid1, int dfd_pid2, const char *ns_path) |
299d1198 | 494 | { |
9b31ab58 | 495 | int ret; |
299d1198 CB |
496 | struct stat ns_st1, ns_st2; |
497 | ||
9b31ab58 CB |
498 | ret = fstatat(dfd_pid1, ns_path, &ns_st1, 0); |
499 | if (ret) | |
3cc629fe | 500 | return -1; |
299d1198 | 501 | |
9b31ab58 CB |
502 | ret = fstatat(dfd_pid2, ns_path, &ns_st2, 0); |
503 | if (ret) | |
3cc629fe | 504 | return -1; |
299d1198 CB |
505 | |
506 | /* processes are in the same namespace */ | |
9b31ab58 CB |
507 | if ((ns_st1.st_dev == ns_st2.st_dev) && |
508 | (ns_st1.st_ino == ns_st2.st_ino)) | |
3cc629fe | 509 | return -EINVAL; |
299d1198 | 510 | |
9b31ab58 CB |
511 | return 0; |
512 | } | |
513 | ||
514 | static int same_ns(int dfd_pid1, int dfd_pid2, const char *ns_path) | |
515 | { | |
516 | __do_close int ns_fd2 = -EBADF; | |
517 | int ret = -1; | |
518 | ||
519 | ns_fd2 = open_at(dfd_pid2, ns_path, PROTECT_OPEN_WITH_TRAILING_SYMLINKS, | |
520 | (PROTECT_LOOKUP_BENEATH_WITH_MAGICLINKS & | |
521 | ~(RESOLVE_NO_XDEV | RESOLVE_BENEATH)), 0); | |
522 | if (ns_fd2 < 0) { | |
523 | /* The kernel does not support this namespace. This is not an error. */ | |
524 | if (errno == ENOENT) | |
525 | return -EINVAL; | |
526 | return log_error_errno(-errno, errno, "Failed to open %d(%s)", | |
527 | dfd_pid2, ns_path); | |
528 | } | |
529 | ||
530 | ret = same_nsfd(dfd_pid1, dfd_pid2, ns_path); | |
531 | if (ret < 0) | |
532 | return ret; | |
533 | ||
299d1198 | 534 | /* processes are in different namespaces */ |
3cc629fe | 535 | return move_fd(ns_fd2); |
299d1198 CB |
536 | } |
537 | ||
9b31ab58 CB |
538 | static int __prepare_namespaces_pidfd(struct attach_context *ctx) |
539 | { | |
540 | for (int i = 0; i < LXC_NS_MAX; i++) { | |
541 | int ret; | |
542 | ||
543 | if (!(ctx->ns_inherited & ns_info[i].clone_flag)) | |
544 | continue; | |
545 | ||
546 | ret = same_nsfd(ctx->dfd_self_pid, | |
547 | ctx->dfd_init_pid, | |
548 | ns_info[i].proc_path); | |
549 | if (ret == -EINVAL) | |
550 | ctx->ns_inherited &= ~ns_info[i].clone_flag; | |
551 | else if (ret < 0) | |
552 | return log_error_errno(-1, errno, | |
553 | "Failed to determine whether %s namespace is shared", | |
554 | ns_info[i].proc_name); | |
555 | else | |
556 | TRACE("Shared %s namespace needs attach", ns_info[i].proc_name); | |
557 | } | |
558 | ||
559 | return 0; | |
560 | } | |
561 | ||
562 | static int __prepare_namespaces_nsfd(struct attach_context *ctx, | |
563 | lxc_attach_options_t *options) | |
b7873c95 | 564 | { |
b7873c95 CB |
565 | for (int i = 0; i < LXC_NS_MAX; i++) { |
566 | int j; | |
567 | ||
568 | if (options->namespaces & ns_info[i].clone_flag) | |
5129b2d3 CB |
569 | ctx->ns_fd[i] = open_at(ctx->dfd_init_pid, |
570 | ns_info[i].proc_path, | |
571 | PROTECT_OPEN_WITH_TRAILING_SYMLINKS, | |
9b31ab58 CB |
572 | (PROTECT_LOOKUP_BENEATH_WITH_MAGICLINKS & |
573 | ~(RESOLVE_NO_XDEV | RESOLVE_BENEATH)), | |
5129b2d3 | 574 | 0); |
b7873c95 | 575 | else if (ctx->ns_inherited & ns_info[i].clone_flag) |
5129b2d3 CB |
576 | ctx->ns_fd[i] = same_ns(ctx->dfd_self_pid, |
577 | ctx->dfd_init_pid, | |
578 | ns_info[i].proc_path); | |
b7873c95 CB |
579 | else |
580 | continue; | |
581 | ||
582 | if (ctx->ns_fd[i] >= 0) | |
583 | continue; | |
584 | ||
585 | if (ctx->ns_fd[i] == -EINVAL) { | |
b7873c95 CB |
586 | ctx->ns_inherited &= ~ns_info[i].clone_flag; |
587 | continue; | |
588 | } | |
589 | ||
590 | /* We failed to preserve the namespace. */ | |
9b31ab58 CB |
591 | SYSERROR("Failed to preserve %s namespace of %d", |
592 | ns_info[i].proc_name, ctx->init_pid); | |
b7873c95 CB |
593 | |
594 | /* Close all already opened file descriptors before we return an | |
595 | * error, so we don't leak them. | |
596 | */ | |
597 | for (j = 0; j < i; j++) | |
598 | close_prot_errno_disarm(ctx->ns_fd[j]); | |
599 | ||
600 | return -1; | |
601 | } | |
602 | ||
603 | return 0; | |
604 | } | |
605 | ||
9b31ab58 CB |
606 | static int prepare_namespaces(struct attach_context *ctx, |
607 | lxc_attach_options_t *options) | |
b7873c95 | 608 | { |
9b31ab58 CB |
609 | if (ctx->init_pidfd < 0) |
610 | return __prepare_namespaces_nsfd(ctx, options); | |
611 | ||
612 | return __prepare_namespaces_pidfd(ctx); | |
b7873c95 CB |
613 | } |
614 | ||
9b31ab58 | 615 | static inline void put_namespaces(struct attach_context *ctx) |
b7873c95 | 616 | { |
9b31ab58 CB |
617 | if (ctx->init_pidfd < 0) { |
618 | for (int i = 0; i < LXC_NS_MAX; i++) | |
619 | close_prot_errno_disarm(ctx->ns_fd[i]); | |
620 | } | |
621 | } | |
b7873c95 | 622 | |
9b31ab58 CB |
623 | static int __attach_namespaces_pidfd(struct attach_context *ctx, |
624 | lxc_attach_options_t *options) | |
625 | { | |
626 | unsigned int ns_flags = options->namespaces | ctx->ns_inherited; | |
627 | int ret; | |
b7873c95 | 628 | |
9b31ab58 CB |
629 | /* The common case is to attach to all namespaces. */ |
630 | ret = setns(ctx->init_pidfd, ns_flags); | |
631 | if (ret) | |
632 | return log_error_errno(-errno, errno, | |
633 | "Failed to attach to namespaces via pidfd"); | |
634 | ||
635 | /* We can't risk leaking file descriptors into the container. */ | |
636 | if (close(ctx->init_pidfd)) | |
637 | return log_error_errno(-errno, errno, "Failed to close pidfd"); | |
638 | ctx->init_pidfd = -EBADF; | |
639 | ||
640 | return log_trace(0, "Attached to container namespaces via pidfd"); | |
b7873c95 CB |
641 | } |
642 | ||
9b31ab58 CB |
643 | static int __attach_namespaces_nsfd(struct attach_context *ctx, |
644 | lxc_attach_options_t *options) | |
99d50954 | 645 | { |
92466fe3 CB |
646 | int fret = 0; |
647 | ||
ffeeed8b CB |
648 | for (int i = 0; i < LXC_NS_MAX; i++) { |
649 | int ret; | |
99d50954 | 650 | |
877f3a04 | 651 | if (ctx->ns_fd[i] < 0) |
26818618 CB |
652 | continue; |
653 | ||
21d0acc2 | 654 | ret = setns(ctx->ns_fd[i], ns_info[i].clone_flag); |
92466fe3 | 655 | if (ret) |
9b31ab58 CB |
656 | return log_error_errno(-errno, errno, |
657 | "Failed to attach to %s namespace of %d", | |
658 | ns_info[i].proc_name, | |
659 | ctx->init_pid); | |
92466fe3 CB |
660 | |
661 | if (close(ctx->ns_fd[i])) { | |
662 | fret = -errno; | |
9b31ab58 CB |
663 | SYSERROR("Failed to close file descriptor for %s namespace", |
664 | ns_info[i].proc_name); | |
92466fe3 CB |
665 | } |
666 | ctx->ns_fd[i] = -EBADF; | |
99d50954 CS |
667 | } |
668 | ||
92466fe3 | 669 | return fret; |
99d50954 CS |
670 | } |
671 | ||
9b31ab58 CB |
672 | static int attach_namespaces(struct attach_context *ctx, |
673 | lxc_attach_options_t *options) | |
674 | { | |
675 | if (lxc_log_trace()) { | |
676 | for (int i = 0; i < LXC_NS_MAX; i++) { | |
677 | if (ns_info[i].clone_flag & options->namespaces) { | |
678 | TRACE("Attaching to %s namespace", ns_info[i].proc_name); | |
679 | continue; | |
680 | } | |
681 | if (ns_info[i].clone_flag & ctx->ns_inherited) { | |
682 | TRACE("Sharing %s namespace", ns_info[i].proc_name); | |
683 | continue; | |
684 | } | |
685 | TRACE("Inheriting %s namespace", ns_info[i].proc_name); | |
686 | } | |
687 | } | |
688 | ||
689 | if (ctx->init_pidfd < 0) | |
690 | return __attach_namespaces_nsfd(ctx, options); | |
691 | ||
692 | return __attach_namespaces_pidfd(ctx, options); | |
693 | } | |
694 | ||
695 | static void put_attach_context(struct attach_context *ctx) | |
696 | { | |
697 | if (ctx) { | |
698 | if (!(ctx->attach_flags & LXC_ATTACH_LSM_LABEL)) | |
699 | free_disarm(ctx->lsm_label); | |
700 | close_prot_errno_disarm(ctx->dfd_init_pid); | |
701 | ||
702 | if (ctx->container) { | |
703 | lxc_container_put(ctx->container); | |
704 | ctx->container = NULL; | |
705 | } | |
706 | ||
707 | put_namespaces(ctx); | |
708 | free(ctx); | |
709 | } | |
710 | } | |
711 | ||
c538837d CB |
712 | /* |
713 | * Place anything in here that needs to be get rid of before we move into the | |
714 | * container's context and fail hard if we can't. | |
715 | */ | |
716 | static bool attach_context_security_barrier(struct attach_context *ctx) | |
717 | { | |
718 | if (ctx) { | |
25c659d5 CB |
719 | if (close(ctx->dfd_self_pid)) |
720 | return false; | |
721 | ctx->dfd_self_pid = -EBADF; | |
722 | ||
723 | if (close(ctx->dfd_init_pid)) | |
c538837d | 724 | return false; |
25c659d5 | 725 | ctx->dfd_init_pid = -EBADF; |
c538837d CB |
726 | } |
727 | ||
728 | return true; | |
729 | } | |
730 | ||
e4103cf6 | 731 | int lxc_attach_remount_sys_proc(void) |
7a0b0b56 CS |
732 | { |
733 | int ret; | |
734 | ||
735 | ret = unshare(CLONE_NEWNS); | |
ffeeed8b CB |
736 | if (ret < 0) |
737 | return log_error_errno(-1, errno, "Failed to unshare mount namespace"); | |
7a0b0b56 | 738 | |
9e61fb1f CB |
739 | if (detect_shared_rootfs() && mount(NULL, "/", NULL, MS_SLAVE | MS_REC, NULL)) |
740 | SYSERROR("Failed to recursively turn root mount tree into dependent mount. Continuing..."); | |
2c6f3fc9 | 741 | |
8ce83369 | 742 | /* Assume /proc is always mounted, so remount it. */ |
7a0b0b56 | 743 | ret = umount2("/proc", MNT_DETACH); |
ffeeed8b CB |
744 | if (ret < 0) |
745 | return log_error_errno(-1, errno, "Failed to unmount /proc"); | |
7a0b0b56 | 746 | |
657256e0 | 747 | ret = mount_filesystem("proc", "/proc", 0); |
ffeeed8b CB |
748 | if (ret < 0) |
749 | return log_error_errno(-1, errno, "Failed to remount /proc"); | |
7a0b0b56 | 750 | |
ffeeed8b CB |
751 | /* |
752 | * Try to umount /sys. If it's not a mount point, we'll get EINVAL, then | |
8ce83369 | 753 | * we ignore it because it may not have been mounted in the first place. |
7a0b0b56 CS |
754 | */ |
755 | ret = umount2("/sys", MNT_DETACH); | |
ffeeed8b CB |
756 | if (ret < 0 && errno != EINVAL) |
757 | return log_error_errno(-1, errno, "Failed to unmount /sys"); | |
758 | ||
759 | /* Remount it. */ | |
657256e0 | 760 | if (ret == 0 && mount_filesystem("sysfs", "/sys", 0)) |
ffeeed8b | 761 | return log_error_errno(-1, errno, "Failed to remount /sys"); |
7a0b0b56 CS |
762 | |
763 | return 0; | |
764 | } | |
765 | ||
677e1d27 | 766 | static int drop_capabilities(struct attach_context *ctx) |
e0732705 | 767 | { |
ffeeed8b | 768 | int last_cap; |
e0732705 | 769 | |
6f4f1937 | 770 | last_cap = lxc_caps_last_cap(); |
ffeeed8b | 771 | for (int cap = 0; cap <= last_cap; cap++) { |
e0732705 CS |
772 | if (ctx->capability_mask & (1LL << cap)) |
773 | continue; | |
774 | ||
b81689a1 | 775 | if (prctl(PR_CAPBSET_DROP, prctl_arg(cap), prctl_arg(0), |
ffeeed8b CB |
776 | prctl_arg(0), prctl_arg(0))) |
777 | return log_error_errno(-1, errno, "Failed to drop capability %d", cap); | |
ea918412 | 778 | |
94ac256f | 779 | TRACE("Dropped capability %d", cap); |
e0732705 CS |
780 | } |
781 | ||
782 | return 0; | |
783 | } | |
905022f7 | 784 | |
ab919e5f | 785 | static int lxc_attach_set_environment(struct attach_context *ctx, |
7385273f | 786 | enum lxc_attach_env_policy_t policy, |
6f4f1937 | 787 | char **extra_env, char **extra_keep) |
b3a39ba6 | 788 | { |
3d55242a | 789 | int ret; |
7385273f | 790 | struct lxc_list *iterator; |
791 | ||
799f96fd | 792 | if (policy == LXC_ATTACH_CLEAR_ENV) { |
3d5e9f48 | 793 | int path_kept = 0; |
6f4f1937 | 794 | char **extra_keep_store = NULL; |
3d5e9f48 CS |
795 | |
796 | if (extra_keep) { | |
797 | size_t count, i; | |
798 | ||
3d55242a CB |
799 | for (count = 0; extra_keep[count]; count++) |
800 | ; | |
3d5e9f48 | 801 | |
89b7bfe3 | 802 | extra_keep_store = zalloc(count * sizeof(char *)); |
3d55242a | 803 | if (!extra_keep_store) |
3d5e9f48 | 804 | return -1; |
3d55242a | 805 | |
3d5e9f48 CS |
806 | for (i = 0; i < count; i++) { |
807 | char *v = getenv(extra_keep[i]); | |
808 | if (v) { | |
809 | extra_keep_store[i] = strdup(v); | |
810 | if (!extra_keep_store[i]) { | |
3d5e9f48 CS |
811 | while (i > 0) |
812 | free(extra_keep_store[--i]); | |
ea918412 | 813 | |
3d5e9f48 CS |
814 | free(extra_keep_store); |
815 | return -1; | |
816 | } | |
3d55242a | 817 | |
3d5e9f48 CS |
818 | if (strcmp(extra_keep[i], "PATH") == 0) |
819 | path_kept = 1; | |
820 | } | |
3d5e9f48 CS |
821 | } |
822 | } | |
823 | ||
799f96fd | 824 | if (clearenv()) { |
a9cab7e3 | 825 | if (extra_keep_store) { |
3d55242a CB |
826 | char **p; |
827 | ||
a9cab7e3 CS |
828 | for (p = extra_keep_store; *p; p++) |
829 | free(*p); | |
3d55242a | 830 | |
a9cab7e3 CS |
831 | free(extra_keep_store); |
832 | } | |
3d55242a | 833 | |
ffeeed8b | 834 | return log_error(-1, "Failed to clear environment"); |
3d5e9f48 CS |
835 | } |
836 | ||
837 | if (extra_keep_store) { | |
838 | size_t i; | |
6f4f1937 | 839 | |
3d5e9f48 | 840 | for (i = 0; extra_keep[i]; i++) { |
acd4922e | 841 | if (extra_keep_store[i]) { |
3d55242a CB |
842 | ret = setenv(extra_keep[i], extra_keep_store[i], 1); |
843 | if (ret < 0) | |
a24c5678 | 844 | SYSWARN("Failed to set environment variable"); |
acd4922e | 845 | } |
ea918412 | 846 | |
3d5e9f48 CS |
847 | free(extra_keep_store[i]); |
848 | } | |
ea918412 | 849 | |
3d5e9f48 CS |
850 | free(extra_keep_store); |
851 | } | |
852 | ||
8ce83369 CB |
853 | /* Always set a default path; shells and execlp tend to be fine |
854 | * without it, but there is a disturbing number of C programs | |
855 | * out there that just assume that getenv("PATH") is never NULL | |
856 | * and then die a painful segfault death. | |
857 | */ | |
3d55242a CB |
858 | if (!path_kept) { |
859 | ret = setenv("PATH", "/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin", 1); | |
860 | if (ret < 0) | |
a24c5678 | 861 | SYSWARN("Failed to set environment variable"); |
3d55242a | 862 | } |
b3a39ba6 DW |
863 | } |
864 | ||
3d55242a | 865 | ret = putenv("container=lxc"); |
ffeeed8b | 866 | if (ret < 0) |
818a57fc | 867 | return log_warn(-1, "Failed to set environment variable"); |
b3a39ba6 | 868 | |
7385273f | 869 | /* Set container environment variables.*/ |
640952e5 | 870 | if (ctx->container->lxc_conf) { |
ab919e5f | 871 | lxc_list_for_each(iterator, &ctx->container->lxc_conf->environment) { |
3d55242a CB |
872 | char *env_tmp; |
873 | ||
874 | env_tmp = strdup((char *)iterator->elem); | |
875 | if (!env_tmp) | |
7385273f | 876 | return -1; |
7385273f | 877 | |
3d55242a | 878 | ret = putenv(env_tmp); |
ffeeed8b CB |
879 | if (ret < 0) |
880 | return log_error_errno(-1, errno, "Failed to set environment variable: %s", (char *)iterator->elem); | |
7385273f | 881 | } |
882 | } | |
883 | ||
8ce83369 | 884 | /* Set extra environment variables. */ |
3d5e9f48 CS |
885 | if (extra_env) { |
886 | for (; *extra_env; extra_env++) { | |
3d55242a | 887 | char *p; |
ea918412 | 888 | |
8ce83369 CB |
889 | /* We just assume the user knows what they are doing, so |
890 | * we don't do any checks. | |
891 | */ | |
3d55242a CB |
892 | p = strdup(*extra_env); |
893 | if (!p) | |
3d5e9f48 | 894 | return -1; |
3d55242a CB |
895 | |
896 | ret = putenv(p); | |
897 | if (ret < 0) | |
a24c5678 | 898 | SYSWARN("Failed to set environment variable"); |
3d5e9f48 CS |
899 | } |
900 | } | |
901 | ||
b3a39ba6 DW |
902 | return 0; |
903 | } | |
904 | ||
74a3920a | 905 | static char *lxc_attach_getpwshell(uid_t uid) |
905022f7 | 906 | { |
1b9c9f5b | 907 | __do_free char *line = NULL, *result = NULL; |
cd8f5663 | 908 | __do_fclose FILE *pipe_f = NULL; |
6f4f1937 | 909 | int fd, ret; |
905022f7 CS |
910 | pid_t pid; |
911 | int pipes[2]; | |
3fa23ac3 CB |
912 | bool found = false; |
913 | size_t line_bufsz = 0; | |
905022f7 | 914 | |
8ce83369 CB |
915 | /* We need to fork off a process that runs the getent program, and we |
916 | * need to capture its output, so we use a pipe for that purpose. | |
905022f7 | 917 | */ |
3fa23ac3 | 918 | ret = pipe2(pipes, O_CLOEXEC); |
905022f7 CS |
919 | if (ret < 0) |
920 | return NULL; | |
921 | ||
922 | pid = fork(); | |
923 | if (pid < 0) { | |
924 | close(pipes[0]); | |
925 | close(pipes[1]); | |
926 | return NULL; | |
927 | } | |
928 | ||
3fa23ac3 | 929 | if (!pid) { |
905022f7 CS |
930 | char uid_buf[32]; |
931 | char *arguments[] = { | |
932 | "getent", | |
933 | "passwd", | |
934 | uid_buf, | |
935 | NULL | |
936 | }; | |
937 | ||
938 | close(pipes[0]); | |
939 | ||
8ce83369 | 940 | /* We want to capture stdout. */ |
3fa23ac3 | 941 | ret = dup2(pipes[1], STDOUT_FILENO); |
905022f7 | 942 | close(pipes[1]); |
3fa23ac3 | 943 | if (ret < 0) |
ea918412 | 944 | _exit(EXIT_FAILURE); |
905022f7 | 945 | |
8ce83369 CB |
946 | /* Get rid of stdin/stderr, so we try to associate it with |
947 | * /dev/null. | |
905022f7 | 948 | */ |
3fa23ac3 | 949 | fd = open_devnull(); |
905022f7 | 950 | if (fd < 0) { |
3fa23ac3 CB |
951 | close(STDIN_FILENO); |
952 | close(STDERR_FILENO); | |
905022f7 | 953 | } else { |
3fa23ac3 | 954 | (void)dup3(fd, STDIN_FILENO, O_CLOEXEC); |
59f0e209 | 955 | (void)dup3(fd, STDERR_FILENO, O_CLOEXEC); |
905022f7 CS |
956 | close(fd); |
957 | } | |
958 | ||
8ce83369 | 959 | /* Finish argument list. */ |
3fa23ac3 CB |
960 | ret = snprintf(uid_buf, sizeof(uid_buf), "%ld", (long)uid); |
961 | if (ret <= 0 || ret >= sizeof(uid_buf)) | |
ea918412 | 962 | _exit(EXIT_FAILURE); |
905022f7 | 963 | |
8ce83369 | 964 | /* Try to run getent program. */ |
3fa23ac3 | 965 | (void)execvp("getent", arguments); |
ea918412 | 966 | _exit(EXIT_FAILURE); |
905022f7 | 967 | } |
3fa23ac3 CB |
968 | |
969 | close(pipes[1]); | |
970 | ||
4110345b | 971 | pipe_f = fdopen(pipes[0], "re"); |
cf4026f1 CB |
972 | if (!pipe_f) { |
973 | close(pipes[0]); | |
974 | goto reap_child; | |
975 | } | |
976 | /* Transfer ownership of pipes[0] to pipe_f. */ | |
977 | move_fd(pipes[0]); | |
978 | ||
3fa23ac3 CB |
979 | while (getline(&line, &line_bufsz, pipe_f) != -1) { |
980 | int i; | |
981 | long value; | |
982 | char *token; | |
983 | char *endptr = NULL, *saveptr = NULL; | |
984 | ||
985 | /* If we already found something, just continue to read | |
986 | * until the pipe doesn't deliver any more data, but | |
987 | * don't modify the existing data structure. | |
988 | */ | |
989 | if (found) | |
990 | continue; | |
991 | ||
18d4ffde | 992 | if (!line) |
993 | continue; | |
994 | ||
3fa23ac3 CB |
995 | /* Trim line on the right hand side. */ |
996 | for (i = strlen(line); i > 0 && (line[i - 1] == '\n' || line[i - 1] == '\r'); --i) | |
997 | line[i - 1] = '\0'; | |
998 | ||
999 | /* Split into tokens: first: user name. */ | |
1000 | token = strtok_r(line, ":", &saveptr); | |
1001 | if (!token) | |
1002 | continue; | |
1003 | ||
1004 | /* next: dummy password field */ | |
1005 | token = strtok_r(NULL, ":", &saveptr); | |
1006 | if (!token) | |
1007 | continue; | |
1008 | ||
1009 | /* next: user id */ | |
1010 | token = strtok_r(NULL, ":", &saveptr); | |
1011 | value = token ? strtol(token, &endptr, 10) : 0; | |
1012 | if (!token || !endptr || *endptr || value == LONG_MIN || | |
ea918412 | 1013 | value == LONG_MAX) |
3fa23ac3 CB |
1014 | continue; |
1015 | ||
1016 | /* dummy sanity check: user id matches */ | |
1017 | if ((uid_t)value != uid) | |
1018 | continue; | |
1019 | ||
1020 | /* skip fields: gid, gecos, dir, go to next field 'shell' */ | |
1021 | for (i = 0; i < 4; i++) { | |
1022 | token = strtok_r(NULL, ":", &saveptr); | |
1023 | if (!token) | |
1024 | continue; | |
1025 | } | |
ea918412 | 1026 | |
3fa23ac3 CB |
1027 | if (!token) |
1028 | continue; | |
ea918412 | 1029 | |
1b9c9f5b | 1030 | free_disarm(result); |
3fa23ac3 CB |
1031 | result = strdup(token); |
1032 | ||
1033 | /* Sanity check that there are no fields after that. */ | |
1034 | token = strtok_r(NULL, ":", &saveptr); | |
1035 | if (token) | |
1036 | continue; | |
1037 | ||
1038 | found = true; | |
1039 | } | |
ea918412 | 1040 | |
cf4026f1 | 1041 | reap_child: |
3fa23ac3 | 1042 | ret = wait_for_pid(pid); |
1b9c9f5b | 1043 | if (ret < 0) |
3fa23ac3 | 1044 | return NULL; |
3fa23ac3 | 1045 | |
1b9c9f5b | 1046 | if (!found) |
3fa23ac3 | 1047 | return NULL; |
3fa23ac3 | 1048 | |
1b9c9f5b | 1049 | return move_ptr(result); |
905022f7 | 1050 | } |
cb3e61fa | 1051 | |
d4db3d14 | 1052 | static bool fetch_seccomp(struct lxc_container *c, lxc_attach_options_t *options) |
2c4ea790 | 1053 | { |
cd8f5663 | 1054 | __do_free char *path = NULL; |
d4db3d14 CB |
1055 | int ret; |
1056 | bool bret; | |
2eef2bda | 1057 | |
afc691a0 | 1058 | if (!attach_lsm(options)) { |
cca66e06 | 1059 | free_disarm(c->lxc_conf->seccomp.seccomp); |
2c4ea790 | 1060 | return true; |
bd4307f0 | 1061 | } |
bd7b4e28 | 1062 | |
afc691a0 | 1063 | /* Remove current setting. */ |
d4db3d14 | 1064 | if (!c->set_config_item(c, "lxc.seccomp.profile", "") && |
ea918412 | 1065 | !c->set_config_item(c, "lxc.seccomp", "")) |
2c4ea790 | 1066 | return false; |
bd7b4e28 | 1067 | |
8ce83369 | 1068 | /* Fetch the current profile path over the cmd interface. */ |
0b427da0 | 1069 | path = c->get_running_config_item(c, "lxc.seccomp.profile"); |
bd7b4e28 | 1070 | if (!path) { |
d4db3d14 | 1071 | INFO("Failed to retrieve lxc.seccomp.profile"); |
ea918412 | 1072 | |
0b427da0 | 1073 | path = c->get_running_config_item(c, "lxc.seccomp"); |
cca66e06 CB |
1074 | if (!path) |
1075 | return log_info(true, "Failed to retrieve lxc.seccomp"); | |
bd7b4e28 SG |
1076 | } |
1077 | ||
8ce83369 | 1078 | /* Copy the value into the new lxc_conf. */ |
d4db3d14 | 1079 | bret = c->set_config_item(c, "lxc.seccomp.profile", path); |
d4db3d14 CB |
1080 | if (!bret) |
1081 | return false; | |
bd7b4e28 | 1082 | |
8ce83369 | 1083 | /* Attempt to parse the resulting config. */ |
d4db3d14 | 1084 | ret = lxc_read_seccomp_config(c->lxc_conf); |
cca66e06 CB |
1085 | if (ret < 0) |
1086 | return log_error(false, "Failed to retrieve seccomp policy"); | |
2c4ea790 | 1087 | |
cca66e06 | 1088 | return log_info(true, "Retrieved seccomp policy"); |
2e812c16 CB |
1089 | } |
1090 | ||
6f4f1937 | 1091 | static bool no_new_privs(struct lxc_container *c, lxc_attach_options_t *options) |
2e812c16 | 1092 | { |
cd8f5663 | 1093 | __do_free char *val = NULL; |
2e812c16 | 1094 | |
2e812c16 | 1095 | /* Remove current setting. */ |
02d3b72b CB |
1096 | if (!c->set_config_item(c, "lxc.no_new_privs", "")) |
1097 | return log_info(false, "Failed to unset lxc.no_new_privs"); | |
2e812c16 CB |
1098 | |
1099 | /* Retrieve currently active setting. */ | |
1100 | val = c->get_running_config_item(c, "lxc.no_new_privs"); | |
02d3b72b CB |
1101 | if (!val) |
1102 | return log_info(false, "Failed to retrieve lxc.no_new_privs"); | |
2e812c16 CB |
1103 | |
1104 | /* Set currently active setting. */ | |
cd8f5663 | 1105 | return c->set_config_item(c, "lxc.no_new_privs", val); |
2c4ea790 SH |
1106 | } |
1107 | ||
338b230f | 1108 | struct attach_payload { |
a998454a | 1109 | int ipc_socket; |
cecf3e83 | 1110 | int terminal_pts_fd; |
a998454a | 1111 | lxc_attach_options_t *options; |
ab919e5f | 1112 | struct attach_context *ctx; |
a998454a CB |
1113 | lxc_attach_exec_t exec_function; |
1114 | void *exec_payload; | |
1115 | }; | |
1116 | ||
338b230f | 1117 | static void put_attach_payload(struct attach_payload *p) |
ba2be1a8 | 1118 | { |
afc691a0 CB |
1119 | if (p) { |
1120 | close_prot_errno_disarm(p->ipc_socket); | |
1121 | close_prot_errno_disarm(p->terminal_pts_fd); | |
dd53c8af | 1122 | put_attach_context(p->ctx); |
ab919e5f | 1123 | p->ctx = NULL; |
b21da190 | 1124 | } |
ba2be1a8 CB |
1125 | } |
1126 | ||
338b230f | 1127 | __noreturn static void do_attach(struct attach_payload *ap) |
a998454a | 1128 | { |
afc691a0 CB |
1129 | lxc_attach_exec_t attach_function = move_ptr(ap->exec_function); |
1130 | void *attach_function_args = move_ptr(ap->exec_payload); | |
427a8067 | 1131 | int lsm_fd, ret; |
338b230f CB |
1132 | lxc_attach_options_t* options = ap->options; |
1133 | struct attach_context *ctx = ap->ctx; | |
ab919e5f | 1134 | struct lxc_conf *conf = ctx->container->lxc_conf; |
a998454a CB |
1135 | |
1136 | /* A description of the purpose of this functionality is provided in the | |
1137 | * lxc-attach(1) manual page. We have to remount here and not in the | |
1138 | * parent process, otherwise /proc may not properly reflect the new pid | |
1139 | * namespace. | |
1140 | */ | |
1141 | if (!(options->namespaces & CLONE_NEWNS) && | |
1142 | (options->attach_flags & LXC_ATTACH_REMOUNT_PROC_SYS)) { | |
1143 | ret = lxc_attach_remount_sys_proc(); | |
b75c344c CB |
1144 | if (ret < 0) |
1145 | goto on_error; | |
ea918412 | 1146 | |
b75c344c | 1147 | TRACE("Remounted \"/proc\" and \"/sys\""); |
a998454a CB |
1148 | } |
1149 | ||
5b514ce3 | 1150 | /* Now perform additional attachments. */ |
a998454a | 1151 | #if HAVE_SYS_PERSONALITY_H |
a998454a | 1152 | if (options->attach_flags & LXC_ATTACH_SET_PERSONALITY) { |
b75c344c CB |
1153 | long new_personality; |
1154 | ||
1155 | if (options->personality < 0) | |
ab919e5f | 1156 | new_personality = ctx->personality; |
b75c344c CB |
1157 | else |
1158 | new_personality = options->personality; | |
ea918412 | 1159 | |
ee142207 CB |
1160 | if (new_personality != LXC_ARCH_UNCHANGED) { |
1161 | ret = personality(new_personality); | |
1162 | if (ret < 0) | |
1163 | goto on_error; | |
ea918412 | 1164 | |
ee142207 CB |
1165 | TRACE("Set new personality"); |
1166 | } | |
a998454a CB |
1167 | } |
1168 | #endif | |
1169 | ||
1170 | if (options->attach_flags & LXC_ATTACH_DROP_CAPABILITIES) { | |
677e1d27 | 1171 | ret = drop_capabilities(ctx); |
b75c344c CB |
1172 | if (ret < 0) |
1173 | goto on_error; | |
ea918412 | 1174 | |
b75c344c | 1175 | TRACE("Dropped capabilities"); |
a998454a CB |
1176 | } |
1177 | ||
1178 | /* Always set the environment (specify (LXC_ATTACH_KEEP_ENV, NULL, NULL) | |
1179 | * if you want this to be a no-op). | |
1180 | */ | |
ab919e5f | 1181 | ret = lxc_attach_set_environment(ctx, |
7385273f | 1182 | options->env_policy, |
a998454a CB |
1183 | options->extra_env_vars, |
1184 | options->extra_keep_env); | |
b75c344c CB |
1185 | if (ret < 0) |
1186 | goto on_error; | |
ea918412 | 1187 | |
b75c344c | 1188 | TRACE("Set up environment"); |
a998454a | 1189 | |
afc691a0 CB |
1190 | /* |
1191 | * This remark only affects fully unprivileged containers: | |
57de839f CB |
1192 | * Receive fd for LSM security module before we set{g,u}id(). The reason |
1193 | * is that on set{g,u}id() the kernel will a) make us undumpable and b) | |
1194 | * we will change our effective uid. This means our effective uid will | |
1195 | * be different from the effective uid of the process that created us | |
1196 | * which means that this processs no longer has capabilities in our | |
1197 | * namespace including CAP_SYS_PTRACE. This means we will not be able to | |
1198 | * read and /proc/<pid> files for the process anymore when /proc is | |
1199 | * mounted with hidepid={1,2}. So let's get the lsm label fd before the | |
1200 | * set{g,u}id(). | |
1201 | */ | |
afc691a0 | 1202 | if (attach_lsm(options) && ctx->lsm_label) { |
f8e88e94 | 1203 | if (!sync_wait_fd(ap->ipc_socket, ATTACH_SYNC_LSM(&lsm_fd))) { |
6e36c297 | 1204 | SYSERROR("Failed to receive lsm label fd"); |
b75c344c | 1205 | goto on_error; |
9044b79e | 1206 | } |
1207 | ||
57de839f CB |
1208 | TRACE("Received LSM label file descriptor %d from parent", lsm_fd); |
1209 | } | |
1210 | ||
08ea9270 | 1211 | if (options->stdin_fd > 0 && isatty(options->stdin_fd)) { |
cd0a2b2f | 1212 | ret = lxc_make_controlling_terminal(options->stdin_fd); |
08ea9270 CB |
1213 | if (ret < 0) |
1214 | goto on_error; | |
1215 | } | |
1216 | ||
b58214ac CB |
1217 | if (!lxc_setgroups(0, NULL) && errno != EPERM) |
1218 | goto on_error; | |
1219 | ||
4475fabb | 1220 | if (options->namespaces & CLONE_NEWUSER) |
3ac4480a | 1221 | if (!lxc_switch_uid_gid(ctx->setup_ns_uid, ctx->setup_ns_gid)) |
b75c344c | 1222 | goto on_error; |
936efc72 | 1223 | |
afc691a0 | 1224 | if (attach_lsm(options) && ctx->lsm_label) { |
d3ba7c98 | 1225 | bool on_exec; |
a998454a CB |
1226 | |
1227 | /* Change into our new LSM profile. */ | |
d3ba7c98 | 1228 | on_exec = options->attach_flags & LXC_ATTACH_LSM_EXEC ? true : false; |
afc691a0 | 1229 | ret = ctx->lsm_ops->process_label_set_at(ctx->lsm_ops, lsm_fd, ctx->lsm_label, on_exec); |
cb2420df | 1230 | close_prot_errno_disarm(lsm_fd); |
b75c344c CB |
1231 | if (ret < 0) |
1232 | goto on_error; | |
ea918412 | 1233 | |
ab919e5f | 1234 | TRACE("Set %s LSM label to \"%s\"", ctx->lsm_ops->name, ctx->lsm_label); |
a998454a CB |
1235 | } |
1236 | ||
640952e5 | 1237 | if (conf->no_new_privs || (options->attach_flags & LXC_ATTACH_NO_NEW_PRIVS)) { |
6ce8e678 AL |
1238 | ret = prctl(PR_SET_NO_NEW_PRIVS, prctl_arg(1), prctl_arg(0), |
1239 | prctl_arg(0), prctl_arg(0)); | |
1240 | if (ret < 0) | |
1241 | goto on_error; | |
1242 | ||
1243 | TRACE("Set PR_SET_NO_NEW_PRIVS"); | |
1244 | } | |
1245 | ||
a998454a CB |
1246 | /* The following is done after the communication socket is shut down. |
1247 | * That way, all errors that might (though unlikely) occur up until this | |
1248 | * point will have their messages printed to the original stderr (if | |
1249 | * logging is so configured) and not the fd the user supplied, if any. | |
1250 | */ | |
1251 | ||
1252 | /* Fd handling for stdin, stdout and stderr; ignore errors here, user | |
1253 | * may want to make sure the fds are closed, for example. | |
1254 | */ | |
08ea9270 | 1255 | if (options->stdin_fd >= 0 && options->stdin_fd != STDIN_FILENO) |
40301d48 | 1256 | if (dup2(options->stdin_fd, STDIN_FILENO) < 0) |
a7563434 | 1257 | SYSDEBUG("Failed to replace stdin with %d", options->stdin_fd); |
08ea9270 CB |
1258 | |
1259 | if (options->stdout_fd >= 0 && options->stdout_fd != STDOUT_FILENO) | |
40301d48 | 1260 | if (dup2(options->stdout_fd, STDOUT_FILENO) < 0) |
93b9960a | 1261 | SYSDEBUG("Failed to replace stdout with %d", options->stdout_fd); |
08ea9270 CB |
1262 | |
1263 | if (options->stderr_fd >= 0 && options->stderr_fd != STDERR_FILENO) | |
40301d48 | 1264 | if (dup2(options->stderr_fd, STDERR_FILENO) < 0) |
93b9960a | 1265 | SYSDEBUG("Failed to replace stderr with %d", options->stderr_fd); |
a998454a CB |
1266 | |
1267 | /* close the old fds */ | |
08ea9270 | 1268 | if (options->stdin_fd > STDERR_FILENO) |
a998454a | 1269 | close(options->stdin_fd); |
08ea9270 CB |
1270 | |
1271 | if (options->stdout_fd > STDERR_FILENO) | |
a998454a | 1272 | close(options->stdout_fd); |
08ea9270 CB |
1273 | |
1274 | if (options->stderr_fd > STDERR_FILENO) | |
a998454a CB |
1275 | close(options->stderr_fd); |
1276 | ||
427a8067 CB |
1277 | /* |
1278 | * Try to remove FD_CLOEXEC flag from stdin/stdout/stderr, but also | |
a998454a CB |
1279 | * here, ignore errors. |
1280 | */ | |
427a8067 | 1281 | for (int fd = STDIN_FILENO; fd <= STDERR_FILENO; fd++) { |
3f62938a | 1282 | ret = fd_cloexec(fd, false); |
b75c344c CB |
1283 | if (ret < 0) { |
1284 | SYSERROR("Failed to clear FD_CLOEXEC from file descriptor %d", fd); | |
1285 | goto on_error; | |
1286 | } | |
a998454a CB |
1287 | } |
1288 | ||
9e84479f | 1289 | if (options->attach_flags & LXC_ATTACH_TERMINAL) { |
338b230f | 1290 | ret = lxc_terminal_prepare_login(ap->terminal_pts_fd); |
ba2be1a8 | 1291 | if (ret < 0) { |
338b230f | 1292 | SYSERROR("Failed to prepare terminal file descriptor %d", ap->terminal_pts_fd); |
ba2be1a8 CB |
1293 | goto on_error; |
1294 | } | |
ea918412 | 1295 | |
338b230f | 1296 | TRACE("Prepared terminal file descriptor %d", ap->terminal_pts_fd); |
ba2be1a8 CB |
1297 | } |
1298 | ||
4475fabb | 1299 | /* Avoid unnecessary syscalls. */ |
3ac4480a CB |
1300 | if (ctx->setup_ns_uid == ctx->target_ns_uid) |
1301 | ctx->target_ns_uid = LXC_INVALID_UID; | |
1302 | ||
1303 | if (ctx->setup_ns_gid == ctx->target_ns_gid) | |
1304 | ctx->target_ns_gid = LXC_INVALID_GID; | |
4475fabb | 1305 | |
3ac4480a CB |
1306 | /* |
1307 | * Make sure that the processes STDIO is correctly owned by the user | |
1308 | * that we are switching to. | |
1309 | */ | |
1310 | ret = fix_stdio_permissions(ctx->target_ns_uid); | |
1311 | if (ret) | |
1312 | INFO("Failed to adjust stdio permissions"); | |
4475fabb | 1313 | |
e18aba7d CB |
1314 | if (conf->seccomp.seccomp) { |
1315 | ret = lxc_seccomp_load(conf); | |
1316 | if (ret < 0) | |
1317 | goto on_error; | |
1318 | ||
1319 | TRACE("Loaded seccomp profile"); | |
1320 | ||
1321 | ret = lxc_seccomp_send_notifier_fd(&conf->seccomp, ap->ipc_socket); | |
1322 | if (ret < 0) | |
1323 | goto on_error; | |
c5bac506 | 1324 | lxc_seccomp_close_notifier_fd(&conf->seccomp); |
e18aba7d CB |
1325 | } |
1326 | ||
3ac4480a | 1327 | if (!lxc_switch_uid_gid(ctx->target_ns_uid, ctx->target_ns_gid)) |
936efc72 CB |
1328 | goto on_error; |
1329 | ||
cd5f35ec CB |
1330 | put_attach_payload(ap); |
1331 | ||
a998454a | 1332 | /* We're done, so we can now do whatever the user intended us to do. */ |
afc691a0 | 1333 | _exit(attach_function(attach_function_args)); |
b75c344c CB |
1334 | |
1335 | on_error: | |
dab02267 | 1336 | ERROR("Failed to attach to container"); |
cd5f35ec | 1337 | put_attach_payload(ap); |
c7ac2e1c | 1338 | _exit(EXIT_FAILURE); |
a998454a CB |
1339 | } |
1340 | ||
f797f05e | 1341 | static int lxc_attach_terminal(const char *name, const char *lxcpath, struct lxc_conf *conf, |
9e84479f | 1342 | struct lxc_terminal *terminal) |
ba2be1a8 CB |
1343 | { |
1344 | int ret; | |
1345 | ||
9e84479f | 1346 | lxc_terminal_init(terminal); |
ba2be1a8 | 1347 | |
8ea93a0f | 1348 | ret = lxc_terminal_create(name, lxcpath, conf, terminal); |
c2af3a15 CB |
1349 | if (ret < 0) |
1350 | return log_error(-1, "Failed to create terminal"); | |
ba2be1a8 | 1351 | |
ba2be1a8 | 1352 | return 0; |
ba2be1a8 CB |
1353 | } |
1354 | ||
9e84479f CB |
1355 | static int lxc_attach_terminal_mainloop_init(struct lxc_terminal *terminal, |
1356 | struct lxc_epoll_descr *descr) | |
ba2be1a8 CB |
1357 | { |
1358 | int ret; | |
1359 | ||
1360 | ret = lxc_mainloop_open(descr); | |
c2af3a15 CB |
1361 | if (ret < 0) |
1362 | return log_error(-1, "Failed to create mainloop"); | |
ba2be1a8 | 1363 | |
9e84479f | 1364 | ret = lxc_terminal_mainloop_add(descr, terminal); |
ba2be1a8 | 1365 | if (ret < 0) { |
ba2be1a8 | 1366 | lxc_mainloop_close(descr); |
c2af3a15 | 1367 | return log_error(-1, "Failed to add handlers to mainloop"); |
ba2be1a8 CB |
1368 | } |
1369 | ||
1370 | return 0; | |
1371 | } | |
1372 | ||
36a94ce8 | 1373 | static inline void lxc_attach_terminal_close_ptx(struct lxc_terminal *terminal) |
ba2be1a8 | 1374 | { |
36a94ce8 | 1375 | close_prot_errno_disarm(terminal->ptx); |
ba2be1a8 CB |
1376 | } |
1377 | ||
cecf3e83 | 1378 | static inline void lxc_attach_terminal_close_pts(struct lxc_terminal *terminal) |
ba2be1a8 | 1379 | { |
41808e20 | 1380 | close_prot_errno_disarm(terminal->pty); |
ba2be1a8 CB |
1381 | } |
1382 | ||
9e84479f | 1383 | static inline void lxc_attach_terminal_close_peer(struct lxc_terminal *terminal) |
ba2be1a8 | 1384 | { |
19a3e906 | 1385 | close_prot_errno_disarm(terminal->peer); |
ba2be1a8 CB |
1386 | } |
1387 | ||
9e84479f | 1388 | static inline void lxc_attach_terminal_close_log(struct lxc_terminal *terminal) |
ba2be1a8 | 1389 | { |
19a3e906 | 1390 | close_prot_errno_disarm(terminal->log_fd); |
ba2be1a8 CB |
1391 | } |
1392 | ||
908fbc1a CB |
1393 | int lxc_attach(struct lxc_container *container, lxc_attach_exec_t exec_function, |
1394 | void *exec_payload, lxc_attach_options_t *options, | |
1395 | pid_t *attached_process) | |
9c4693b8 | 1396 | { |
6f9fe5d0 | 1397 | int ret_parent = -1; |
6f9fe5d0 | 1398 | struct lxc_epoll_descr descr = {}; |
a9f0cecf | 1399 | int ret; |
26abd7ea | 1400 | char *name, *lxcpath; |
9c4693b8 | 1401 | int ipc_sockets[2]; |
500ed813 | 1402 | pid_t attached_pid, pid, to_cleanup_pid; |
ab919e5f | 1403 | struct attach_context *ctx; |
9e84479f | 1404 | struct lxc_terminal terminal; |
1cce35e6 | 1405 | struct lxc_conf *conf; |
9c4693b8 | 1406 | |
908fbc1a | 1407 | if (!container) |
540a2f70 | 1408 | return ret_set_errno(-1, EINVAL); |
908fbc1a CB |
1409 | |
1410 | if (!lxc_container_get(container)) | |
540a2f70 | 1411 | return ret_set_errno(-1, EINVAL); |
908fbc1a CB |
1412 | |
1413 | name = container->name; | |
1414 | lxcpath = container->config_path; | |
1415 | ||
afc691a0 | 1416 | if (!options) { |
9c4693b8 | 1417 | options = &attach_static_default_options; |
afc691a0 CB |
1418 | options->lsm_label = NULL; |
1419 | } | |
9c4693b8 | 1420 | |
9745eb8a | 1421 | ctx = alloc_attach_context(); |
ab919e5f | 1422 | if (!ctx) { |
9745eb8a CB |
1423 | lxc_container_put(container); |
1424 | return log_error_errno(-ENOMEM, ENOMEM, "Failed to allocate attach context"); | |
1425 | } | |
1426 | ||
afc691a0 | 1427 | ret = get_attach_context(ctx, container, options); |
9745eb8a | 1428 | if (ret) { |
7e995801 | 1429 | put_attach_context(ctx); |
74ce42b5 | 1430 | return log_error(-1, "Failed to get attach context"); |
9c4693b8 CS |
1431 | } |
1432 | ||
ab919e5f | 1433 | conf = ctx->container->lxc_conf; |
ba773996 | 1434 | |
ab919e5f | 1435 | if (!fetch_seccomp(ctx->container, options)) |
ae026f55 | 1436 | WARN("Failed to get seccomp policy"); |
2c4ea790 | 1437 | |
ab919e5f | 1438 | if (!no_new_privs(ctx->container, options)) |
ae026f55 | 1439 | WARN("Could not determine whether PR_SET_NO_NEW_PRIVS is set"); |
2e812c16 | 1440 | |
9b31ab58 | 1441 | ret = prepare_namespaces(ctx, options); |
b7873c95 | 1442 | if (ret) { |
52ed870e | 1443 | put_attach_context(ctx); |
74ce42b5 | 1444 | return log_error(-1, "Failed to get namespace file descriptors"); |
9c4693b8 CS |
1445 | } |
1446 | ||
9e84479f | 1447 | if (options->attach_flags & LXC_ATTACH_TERMINAL) { |
f797f05e | 1448 | ret = lxc_attach_terminal(name, lxcpath, conf, &terminal); |
ba2be1a8 | 1449 | if (ret < 0) { |
dd53c8af | 1450 | put_attach_context(ctx); |
74ce42b5 | 1451 | return log_error(-1, "Failed to setup new terminal"); |
ba2be1a8 CB |
1452 | } |
1453 | ||
9e84479f | 1454 | terminal.log_fd = options->log_fd; |
c948657b | 1455 | } else { |
9e84479f | 1456 | lxc_terminal_init(&terminal); |
ba2be1a8 CB |
1457 | } |
1458 | ||
8ce83369 CB |
1459 | /* Create a socket pair for IPC communication; set SOCK_CLOEXEC in order |
1460 | * to make sure we don't irritate other threads that want to fork+exec | |
1461 | * away | |
9c4693b8 CS |
1462 | * |
1463 | * IMPORTANT: if the initial process is multithreaded and another call | |
1464 | * just fork()s away without exec'ing directly after, the socket fd will | |
1465 | * exist in the forked process from the other thread and any close() in | |
8ce83369 | 1466 | * our own child process will not really cause the socket to close |
1d801260 | 1467 | * properly, potentially causing the parent to hang. |
9c4693b8 CS |
1468 | * |
1469 | * For this reason, while IPC is still active, we have to use shutdown() | |
8ce83369 CB |
1470 | * if the child exits prematurely in order to signal that the socket is |
1471 | * closed and cannot assume that the child exiting will automatically do | |
1472 | * that. | |
9c4693b8 CS |
1473 | * |
1474 | * IPC mechanism: (X is receiver) | |
1475 | * initial process intermediate attached | |
1476 | * X <--- send pid of | |
1477 | * attached proc, | |
1478 | * then exit | |
1479 | * send 0 ------------------------------------> X | |
1480 | * [do initialization] | |
1481 | * X <------------------------------------ send 1 | |
1482 | * [add to cgroup, ...] | |
1483 | * send 2 ------------------------------------> X | |
81f466d0 CB |
1484 | * [set LXC_ATTACH_NO_NEW_PRIVS] |
1485 | * X <------------------------------------ send 3 | |
1486 | * [open LSM label fd] | |
1487 | * send 4 ------------------------------------> X | |
1488 | * [set LSM label] | |
9c4693b8 CS |
1489 | * close socket close socket |
1490 | * run program | |
1491 | */ | |
1492 | ret = socketpair(PF_LOCAL, SOCK_STREAM | SOCK_CLOEXEC, 0, ipc_sockets); | |
1493 | if (ret < 0) { | |
dd53c8af | 1494 | put_attach_context(ctx); |
74ce42b5 | 1495 | return log_error_errno(-1, errno, "Could not set up required IPC mechanism for attaching"); |
9c4693b8 CS |
1496 | } |
1497 | ||
e3f0e436 CB |
1498 | /* Create intermediate subprocess, two reasons: |
1499 | * 1. We can't setns() in the child itself, since we want to make | |
8ce83369 | 1500 | * sure we are properly attached to the pidns. |
e3f0e436 | 1501 | * 2. Also, the initial thread has to put the attached process |
8ce83369 CB |
1502 | * into the cgroup, which we can only do if we didn't already |
1503 | * setns() (otherwise, user namespaces will hate us). | |
9c4693b8 CS |
1504 | */ |
1505 | pid = fork(); | |
9c4693b8 | 1506 | if (pid < 0) { |
dd53c8af | 1507 | put_attach_context(ctx); |
74ce42b5 | 1508 | return log_error_errno(-1, errno, "Failed to create first subprocess"); |
9c4693b8 CS |
1509 | } |
1510 | ||
4f25e72f | 1511 | if (pid == 0) { |
26abd7ea | 1512 | char *cwd, *new_cwd; |
a588a482 | 1513 | |
ba2be1a8 | 1514 | /* close unneeded file descriptors */ |
4f25e72f | 1515 | close_prot_errno_disarm(ipc_sockets[0]); |
2202afc9 | 1516 | |
4f25e72f CB |
1517 | if (options->attach_flags & LXC_ATTACH_TERMINAL) { |
1518 | lxc_attach_terminal_close_ptx(&terminal); | |
1519 | lxc_attach_terminal_close_peer(&terminal); | |
1520 | lxc_attach_terminal_close_log(&terminal); | |
f4364484 SG |
1521 | } |
1522 | ||
4f25e72f | 1523 | /* Wait for the parent to have setup cgroups. */ |
6e48e7c5 | 1524 | if (!sync_wait(ipc_sockets[1], ATTACH_SYNC_CGROUP)) { |
4f25e72f | 1525 | shutdown(ipc_sockets[1], SHUT_RDWR); |
dd53c8af | 1526 | put_attach_context(ctx); |
4f25e72f | 1527 | _exit(EXIT_FAILURE); |
bb2ada6f CB |
1528 | } |
1529 | ||
c538837d CB |
1530 | if (!attach_context_security_barrier(ctx)) { |
1531 | shutdown(ipc_sockets[1], SHUT_RDWR); | |
1532 | put_attach_context(ctx); | |
1533 | _exit(EXIT_FAILURE); | |
1534 | } | |
1535 | ||
4f25e72f CB |
1536 | TRACE("Intermediate process starting to initialize"); |
1537 | ||
a588a482 CB |
1538 | cwd = getcwd(NULL, 0); |
1539 | ||
c538837d CB |
1540 | /* |
1541 | * Attach now, create another subprocess later, since pid | |
1542 | * namespaces only really affect the children of the current | |
1543 | * process. | |
1544 | * | |
1545 | * Note that this is a crucial barrier. We're no moving into | |
1546 | * the container's context so we need to make sure to not leak | |
1547 | * anything sensitive. That especially means things such as | |
1548 | * open file descriptors! | |
4f25e72f | 1549 | */ |
9b31ab58 | 1550 | ret = attach_namespaces(ctx, options); |
4f25e72f CB |
1551 | if (ret < 0) { |
1552 | ERROR("Failed to enter namespaces"); | |
1553 | shutdown(ipc_sockets[1], SHUT_RDWR); | |
dd53c8af | 1554 | put_attach_context(ctx); |
4f25e72f | 1555 | _exit(EXIT_FAILURE); |
ba2be1a8 CB |
1556 | } |
1557 | ||
4f25e72f CB |
1558 | /* Attach succeeded, try to cwd. */ |
1559 | if (options->initial_cwd) | |
1560 | new_cwd = options->initial_cwd; | |
1561 | else | |
1562 | new_cwd = cwd; | |
1563 | if (new_cwd) { | |
1564 | ret = chdir(new_cwd); | |
1565 | if (ret < 0) | |
1566 | WARN("Could not change directory to \"%s\"", new_cwd); | |
ba2be1a8 | 1567 | } |
a588a482 | 1568 | free_disarm(cwd); |
c6d09e15 | 1569 | |
4f25e72f | 1570 | /* Create attached process. */ |
4f25e72f CB |
1571 | pid = lxc_raw_clone(CLONE_PARENT, NULL); |
1572 | if (pid < 0) { | |
1573 | SYSERROR("Failed to clone attached process"); | |
1574 | shutdown(ipc_sockets[1], SHUT_RDWR); | |
dd53c8af | 1575 | put_attach_context(ctx); |
4f25e72f CB |
1576 | _exit(EXIT_FAILURE); |
1577 | } | |
f4364484 | 1578 | |
4f25e72f | 1579 | if (pid == 0) { |
338b230f | 1580 | struct attach_payload ap = { |
a64902ab CB |
1581 | .ipc_socket = ipc_sockets[1], |
1582 | .options = options, | |
1583 | .ctx = ctx, | |
1584 | .terminal_pts_fd = terminal.pty, | |
1585 | .exec_function = exec_function, | |
1586 | .exec_payload = exec_payload, | |
1587 | }; | |
1588 | ||
4f25e72f CB |
1589 | if (options->attach_flags & LXC_ATTACH_TERMINAL) { |
1590 | ret = lxc_terminal_signal_sigmask_safe_blocked(&terminal); | |
1591 | if (ret < 0) { | |
1592 | SYSERROR("Failed to reset signal mask"); | |
1593 | _exit(EXIT_FAILURE); | |
1594 | } | |
1595 | } | |
ea918412 | 1596 | |
a64902ab | 1597 | /* Does not return. */ |
338b230f | 1598 | do_attach(&ap); |
62183f1a | 1599 | } |
2eef2bda | 1600 | |
4f25e72f CB |
1601 | if (options->attach_flags & LXC_ATTACH_TERMINAL) |
1602 | lxc_attach_terminal_close_pts(&terminal); | |
ea918412 | 1603 | |
4f25e72f | 1604 | /* Tell grandparent the pid of the pid of the newly created child. */ |
f8e88e94 | 1605 | if (!sync_wake_pid(ipc_sockets[1], ATTACH_SYNC_PID(pid))) { |
4f25e72f CB |
1606 | /* If this really happens here, this is very unfortunate, since |
1607 | * the parent will not know the pid of the attached process and | |
1608 | * will not be able to wait for it (and we won't either due to | |
1609 | * CLONE_PARENT) so the parent won't be able to reap it and the | |
1610 | * attached process will remain a zombie. | |
1611 | */ | |
1612 | shutdown(ipc_sockets[1], SHUT_RDWR); | |
dd53c8af | 1613 | put_attach_context(ctx); |
4f25e72f CB |
1614 | _exit(EXIT_FAILURE); |
1615 | } | |
9c4693b8 | 1616 | |
4f25e72f | 1617 | TRACE("Sending pid %d of attached process", pid); |
9c4693b8 | 1618 | |
4f25e72f | 1619 | /* The rest is in the hands of the initial and the attached process. */ |
dd53c8af | 1620 | put_attach_context(ctx); |
4f25e72f CB |
1621 | _exit(EXIT_SUCCESS); |
1622 | } | |
6f4f1937 | 1623 | |
4f25e72f | 1624 | to_cleanup_pid = pid; |
ea918412 | 1625 | |
4f25e72f | 1626 | /* close unneeded file descriptors */ |
cb2420df | 1627 | close_prot_errno_disarm(ipc_sockets[1]); |
9b31ab58 | 1628 | put_namespaces(ctx); |
4f25e72f CB |
1629 | if (options->attach_flags & LXC_ATTACH_TERMINAL) |
1630 | lxc_attach_terminal_close_pts(&terminal); | |
81f466d0 | 1631 | |
4f25e72f CB |
1632 | /* Attach to cgroup, if requested. */ |
1633 | if (options->attach_flags & LXC_ATTACH_MOVE_TO_CGROUP) { | |
1634 | /* | |
1635 | * If this is the unified hierarchy cgroup_attach() is | |
1636 | * enough. | |
1637 | */ | |
1638 | ret = cgroup_attach(conf, name, lxcpath, pid); | |
29619d41 | 1639 | if (ret == -ENOCGROUP2) { |
4f25e72f | 1640 | call_cleaner(cgroup_exit) struct cgroup_ops *cgroup_ops = NULL; |
9044b79e | 1641 | |
4f25e72f CB |
1642 | cgroup_ops = cgroup_init(conf); |
1643 | if (!cgroup_ops) | |
1644 | goto on_error; | |
9044b79e | 1645 | |
4f25e72f CB |
1646 | if (!cgroup_ops->attach(cgroup_ops, conf, name, lxcpath, pid)) |
1647 | goto on_error; | |
81f466d0 | 1648 | } |
4f25e72f CB |
1649 | TRACE("Moved intermediate process %d into container's cgroups", pid); |
1650 | } | |
81f466d0 | 1651 | |
4f25e72f CB |
1652 | /* Setup /proc limits */ |
1653 | if (!lxc_list_empty(&conf->procs)) { | |
1654 | ret = setup_proc_filesystem(&conf->procs, pid); | |
1655 | if (ret < 0) | |
1656 | goto on_error; | |
4f3b6a85 CB |
1657 | |
1658 | TRACE("Setup /proc/%d settings", pid); | |
4f25e72f | 1659 | } |
cdb2a47f | 1660 | |
4f25e72f CB |
1661 | /* Setup resource limits */ |
1662 | if (!lxc_list_empty(&conf->limits)) { | |
1663 | ret = setup_resource_limits(&conf->limits, pid); | |
1664 | if (ret < 0) | |
1665 | goto on_error; | |
4f3b6a85 CB |
1666 | |
1667 | TRACE("Setup resource limits"); | |
4f25e72f | 1668 | } |
cdb2a47f | 1669 | |
4f25e72f CB |
1670 | if (options->attach_flags & LXC_ATTACH_TERMINAL) { |
1671 | ret = lxc_attach_terminal_mainloop_init(&terminal, &descr); | |
1672 | if (ret < 0) | |
1673 | goto on_error; | |
9c4693b8 | 1674 | |
4f25e72f CB |
1675 | TRACE("Initialized terminal mainloop"); |
1676 | } | |
9c4693b8 | 1677 | |
4f25e72f | 1678 | /* Let the child process know to go ahead. */ |
6e48e7c5 | 1679 | if (!sync_wake(ipc_sockets[0], ATTACH_SYNC_CGROUP)) |
4f25e72f | 1680 | goto close_mainloop; |
ba2be1a8 | 1681 | |
4f25e72f | 1682 | TRACE("Told intermediate process to start initializing"); |
ea918412 | 1683 | |
4f25e72f | 1684 | /* Get pid of attached process from intermediate process. */ |
f8e88e94 | 1685 | if (!sync_wait_pid(ipc_sockets[0], ATTACH_SYNC_PID(&attached_pid))) |
4f25e72f | 1686 | goto close_mainloop; |
ba2be1a8 | 1687 | |
4f25e72f | 1688 | TRACE("Received pid %d of attached process in parent pid namespace", attached_pid); |
ba2be1a8 | 1689 | |
4f25e72f | 1690 | /* Ignore SIGKILL (CTRL-C) and SIGQUIT (CTRL-\) - issue #313. */ |
5d2b46fb | 1691 | if (options->stdin_fd == STDIN_FILENO) { |
4f25e72f CB |
1692 | signal(SIGINT, SIG_IGN); |
1693 | signal(SIGQUIT, SIG_IGN); | |
1694 | } | |
ba2be1a8 | 1695 | |
4f25e72f CB |
1696 | /* Reap intermediate process. */ |
1697 | ret = wait_for_pid(pid); | |
1698 | if (ret < 0) | |
1699 | goto close_mainloop; | |
ba2be1a8 | 1700 | |
4f25e72f | 1701 | TRACE("Intermediate process %d exited", pid); |
ea918412 | 1702 | |
4f25e72f CB |
1703 | /* We will always have to reap the attached process now. */ |
1704 | to_cleanup_pid = attached_pid; | |
9c4693b8 | 1705 | |
4f25e72f | 1706 | /* Open LSM fd and send it to child. */ |
afc691a0 | 1707 | if (attach_lsm(options) && ctx->lsm_label) { |
ad001fb6 | 1708 | __do_close int labelfd = -EBADF; |
4f25e72f | 1709 | bool on_exec; |
ea918412 | 1710 | |
4f25e72f | 1711 | on_exec = options->attach_flags & LXC_ATTACH_LSM_EXEC ? true : false; |
afc691a0 | 1712 | labelfd = ctx->lsm_ops->process_label_fd_get(ctx->lsm_ops, attached_pid, on_exec); |
4f25e72f CB |
1713 | if (labelfd < 0) |
1714 | goto close_mainloop; | |
9c4693b8 | 1715 | |
4f25e72f | 1716 | TRACE("Opened LSM label file descriptor %d", labelfd); |
ea918412 | 1717 | |
4f25e72f | 1718 | /* Send child fd of the LSM security module to write to. */ |
f8e88e94 | 1719 | if (!sync_wake_fd(ipc_sockets[0], ATTACH_SYNC_LSM(labelfd))) { |
6e36c297 | 1720 | SYSERROR("Failed to send lsm label fd"); |
4f25e72f CB |
1721 | goto close_mainloop; |
1722 | } | |
1723 | ||
4f25e72f | 1724 | TRACE("Sent LSM label file descriptor %d to child", labelfd); |
9c4693b8 | 1725 | } |
ea918412 | 1726 | |
4f25e72f CB |
1727 | if (conf->seccomp.seccomp) { |
1728 | ret = lxc_seccomp_recv_notifier_fd(&conf->seccomp, ipc_sockets[0]); | |
1729 | if (ret < 0) | |
1730 | goto close_mainloop; | |
9c4693b8 | 1731 | |
4f25e72f | 1732 | ret = lxc_seccomp_add_notifier(name, lxcpath, &conf->seccomp); |
d6d979bc | 1733 | if (ret < 0) |
4f25e72f | 1734 | goto close_mainloop; |
d6d979bc | 1735 | } |
9c4693b8 | 1736 | |
4f25e72f CB |
1737 | /* We're done, the child process should now execute whatever it |
1738 | * is that the user requested. The parent can now track it with | |
1739 | * waitpid() or similar. | |
1740 | */ | |
9c4693b8 | 1741 | |
4f25e72f | 1742 | *attached_process = attached_pid; |
a998454a | 1743 | |
4f25e72f CB |
1744 | /* Now shut down communication with child, we're done. */ |
1745 | shutdown(ipc_sockets[0], SHUT_RDWR); | |
cb2420df | 1746 | close_prot_errno_disarm(ipc_sockets[0]); |
f157b056 | 1747 | |
4f25e72f CB |
1748 | ret_parent = 0; |
1749 | to_cleanup_pid = -1; | |
ea918412 | 1750 | |
4f25e72f CB |
1751 | if (options->attach_flags & LXC_ATTACH_TERMINAL) { |
1752 | ret = lxc_mainloop(&descr, -1); | |
1753 | if (ret < 0) { | |
1754 | ret_parent = -1; | |
1755 | to_cleanup_pid = attached_pid; | |
1756 | } | |
a998454a | 1757 | } |
ea918412 | 1758 | |
4f25e72f | 1759 | close_mainloop: |
9e84479f | 1760 | if (options->attach_flags & LXC_ATTACH_TERMINAL) |
4f25e72f | 1761 | lxc_mainloop_close(&descr); |
9c4693b8 | 1762 | |
4f25e72f CB |
1763 | on_error: |
1764 | if (ipc_sockets[0] >= 0) { | |
1765 | shutdown(ipc_sockets[0], SHUT_RDWR); | |
cb2420df | 1766 | close_prot_errno_disarm(ipc_sockets[0]); |
9c4693b8 | 1767 | } |
ea918412 | 1768 | |
4f25e72f CB |
1769 | if (to_cleanup_pid > 0) |
1770 | (void)wait_for_pid(to_cleanup_pid); | |
1771 | ||
1772 | if (options->attach_flags & LXC_ATTACH_TERMINAL) { | |
1773 | lxc_terminal_delete(&terminal); | |
1774 | lxc_terminal_conf_free(&terminal); | |
1775 | } | |
9c4693b8 | 1776 | |
dd53c8af | 1777 | put_attach_context(ctx); |
4f25e72f | 1778 | return ret_parent; |
9c4693b8 CS |
1779 | } |
1780 | ||
06346bb0 | 1781 | int lxc_attach_run_command(void *payload) |
9c4693b8 | 1782 | { |
06346bb0 CB |
1783 | int ret = -1; |
1784 | lxc_attach_command_t *cmd = payload; | |
9c4693b8 | 1785 | |
06346bb0 CB |
1786 | ret = execvp(cmd->program, cmd->argv); |
1787 | if (ret < 0) { | |
1788 | switch (errno) { | |
1789 | case ENOEXEC: | |
1790 | ret = 126; | |
cf0fd972 | 1791 | break; |
06346bb0 CB |
1792 | case ENOENT: |
1793 | ret = 127; | |
cf0fd972 | 1794 | break; |
06346bb0 CB |
1795 | } |
1796 | } | |
ea918412 | 1797 | |
c2af3a15 | 1798 | return log_error_errno(ret, errno, "Failed to exec \"%s\"", cmd->program); |
9c4693b8 CS |
1799 | } |
1800 | ||
1801 | int lxc_attach_run_shell(void* payload) | |
1802 | { | |
cd8f5663 | 1803 | __do_free char *buf = NULL; |
9c4693b8 | 1804 | uid_t uid; |
cb7aa5e8 DJ |
1805 | struct passwd pwent; |
1806 | struct passwd *pwentp = NULL; | |
9c4693b8 | 1807 | char *user_shell; |
cb7aa5e8 DJ |
1808 | size_t bufsize; |
1809 | int ret; | |
9c4693b8 | 1810 | |
8ce83369 | 1811 | /* Ignore payload parameter. */ |
9c4693b8 CS |
1812 | (void)payload; |
1813 | ||
1814 | uid = getuid(); | |
cb7aa5e8 DJ |
1815 | |
1816 | bufsize = sysconf(_SC_GETPW_R_SIZE_MAX); | |
1817 | if (bufsize == -1) | |
1818 | bufsize = 1024; | |
1819 | ||
1820 | buf = malloc(bufsize); | |
1821 | if (buf) { | |
1822 | ret = getpwuid_r(uid, &pwent, buf, bufsize, &pwentp); | |
1823 | if (!pwentp) { | |
1824 | if (ret == 0) | |
ea918412 | 1825 | WARN("Could not find matched password record"); |
cb7aa5e8 DJ |
1826 | |
1827 | WARN("Failed to get password record - %u", uid); | |
1828 | } | |
1829 | } | |
9c4693b8 | 1830 | |
8ce83369 CB |
1831 | /* This probably happens because of incompatible nss implementations in |
1832 | * host and container (remember, this code is still using the host's | |
1833 | * glibc but our mount namespace is in the container) we may try to get | |
1834 | * the information by spawning a [getent passwd uid] process and parsing | |
1835 | * the result. | |
9c4693b8 | 1836 | */ |
cb7aa5e8 | 1837 | if (!pwentp) |
9c4693b8 CS |
1838 | user_shell = lxc_attach_getpwshell(uid); |
1839 | else | |
cb7aa5e8 | 1840 | user_shell = pwent.pw_shell; |
ea918412 | 1841 | |
9c4693b8 | 1842 | if (user_shell) |
acf47e1b | 1843 | execlp(user_shell, user_shell, (char *)NULL); |
9c4693b8 | 1844 | |
8ce83369 CB |
1845 | /* Executed if either no passwd entry or execvp fails, we will fall back |
1846 | * on /bin/sh as a default shell. | |
9c4693b8 | 1847 | */ |
acf47e1b | 1848 | execlp("/bin/sh", "/bin/sh", (char *)NULL); |
ea918412 | 1849 | |
edeb1836 | 1850 | SYSERROR("Failed to execute shell"); |
cb7aa5e8 | 1851 | if (!pwentp) |
edeb1836 | 1852 | free(user_shell); |
ea918412 | 1853 | |
9c4693b8 CS |
1854 | return -1; |
1855 | } |