]>
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 | |
f51c7eb4 CB |
119 | ret = strnprintf(path + STRLITERALLEN("fdinfo/"), INTTYPE_TO_STRLEN(int), "%d", pidfd); |
120 | if (ret < 0) | |
d8764025 CB |
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 | ||
6a6c7030 | 133 | if (!strnequal(slider, prefix, prefix_len)) |
d8764025 CB |
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 | 279 | |
4475fabb CB |
280 | return 0; |
281 | } | |
282 | ||
283 | static void userns_target_ids(struct attach_context *ctx, lxc_attach_options_t *options) | |
284 | { | |
285 | if (options->uid != LXC_INVALID_UID) | |
3ac4480a | 286 | ctx->target_ns_uid = options->uid; |
4475fabb | 287 | else if (options->namespaces & CLONE_NEWUSER) |
3ac4480a | 288 | ctx->target_ns_uid = ctx->setup_ns_uid; |
4475fabb | 289 | else |
3ac4480a | 290 | ctx->target_ns_uid = 0; |
4475fabb | 291 | |
3ac4480a | 292 | if (ctx->target_ns_uid == LXC_INVALID_UID) |
4475fabb CB |
293 | WARN("Invalid uid specified"); |
294 | ||
295 | if (options->gid != LXC_INVALID_GID) | |
3ac4480a | 296 | ctx->target_ns_gid = options->gid; |
4475fabb | 297 | else if (options->namespaces & CLONE_NEWUSER) |
3ac4480a | 298 | ctx->target_ns_gid = ctx->setup_ns_gid; |
4475fabb | 299 | else |
3ac4480a | 300 | ctx->target_ns_gid = 0; |
4475fabb | 301 | |
3ac4480a | 302 | if (ctx->target_ns_gid == LXC_INVALID_GID) |
4475fabb CB |
303 | WARN("Invalid gid specified"); |
304 | } | |
305 | ||
9680e7b0 CB |
306 | static int parse_init_status(struct attach_context *ctx, lxc_attach_options_t *options) |
307 | { | |
308 | __do_free char *line = NULL; | |
309 | __do_fclose FILE *f = NULL; | |
310 | size_t len = 0; | |
311 | bool caps_found = false; | |
4475fabb | 312 | int ret; |
9680e7b0 | 313 | |
72a19d2f | 314 | f = fdopen_at(ctx->dfd_init_pid, "status", "re", PROTECT_OPEN, PROTECT_LOOKUP_BENEATH); |
9680e7b0 | 315 | if (!f) |
4475fabb | 316 | return log_error_errno(-errno, errno, "Failed to open status file"); |
9680e7b0 CB |
317 | |
318 | while (getline(&line, &len, f) != -1) { | |
319 | signed long value = -1; | |
9680e7b0 | 320 | |
4475fabb CB |
321 | /* |
322 | * Format is: real, effective, saved set user, fs we only care | |
323 | * about real uid. | |
324 | */ | |
325 | ret = sscanf(line, "Uid: %ld", &value); | |
326 | if (ret != EOF && ret == 1) { | |
3ac4480a CB |
327 | ctx->target_host_uid = (uid_t)value; |
328 | TRACE("Container's init process runs with hostuid %d", ctx->target_host_uid); | |
4475fabb CB |
329 | goto next; |
330 | } | |
9680e7b0 | 331 | |
4475fabb CB |
332 | ret = sscanf(line, "Gid: %ld", &value); |
333 | if (ret != EOF && ret == 1) { | |
3ac4480a CB |
334 | ctx->target_host_gid = (gid_t)value; |
335 | TRACE("Container's init process runs with hostgid %d", ctx->target_host_gid); | |
4475fabb | 336 | goto next; |
9680e7b0 CB |
337 | } |
338 | ||
339 | ret = sscanf(line, "CapBnd: %llx", &ctx->capability_mask); | |
340 | if (ret != EOF && ret == 1) { | |
341 | caps_found = true; | |
342 | goto next; | |
343 | } | |
344 | ||
345 | next: | |
3ac4480a CB |
346 | if (ctx->target_host_uid != LXC_INVALID_UID && |
347 | ctx->target_host_gid != LXC_INVALID_GID && | |
4475fabb | 348 | caps_found) |
9680e7b0 CB |
349 | break; |
350 | ||
351 | } | |
352 | ||
3ac4480a | 353 | ret = userns_setup_ids(ctx, options); |
4475fabb CB |
354 | if (ret) |
355 | return log_error_errno(ret, errno, "Failed to get setup ids"); | |
356 | userns_target_ids(ctx, options); | |
357 | ||
9680e7b0 CB |
358 | return 0; |
359 | } | |
360 | ||
9b31ab58 CB |
361 | static bool pidfd_setns_supported(struct attach_context *ctx) |
362 | { | |
363 | int ret; | |
364 | ||
365 | /* | |
366 | * The ability to attach to time namespaces came after the introduction | |
367 | * of of using pidfds for attaching to namespaces. To avoid having to | |
368 | * special-case both CLONE_NEWUSER and CLONE_NEWTIME handling, let's | |
369 | * use CLONE_NEWTIME as gatekeeper. | |
370 | */ | |
371 | if (ctx->init_pidfd >= 0) | |
372 | ret = setns(ctx->init_pidfd, CLONE_NEWTIME); | |
373 | else | |
374 | ret = -EOPNOTSUPP; | |
375 | TRACE("Attaching to namespaces via pidfds %s", | |
376 | ret ? "unsupported" : "supported"); | |
377 | return ret == 0; | |
378 | } | |
379 | ||
500ed813 | 380 | static int get_attach_context(struct attach_context *ctx, |
afc691a0 CB |
381 | struct lxc_container *container, |
382 | lxc_attach_options_t *options) | |
e0732705 | 383 | { |
9680e7b0 | 384 | __do_free char *lsm_label = NULL; |
6f4f1937 | 385 | int ret; |
c538837d | 386 | char path[LXC_PROC_PID_LEN]; |
e0732705 | 387 | |
500ed813 | 388 | ctx->container = container; |
afc691a0 | 389 | ctx->attach_flags = options->attach_flags; |
500ed813 | 390 | |
6f0c2cea CB |
391 | ctx->dfd_self_pid = open_at(-EBADF, "/proc/self", |
392 | PROTECT_OPATH_FILE & ~O_NOFOLLOW, | |
393 | (PROTECT_LOOKUP_ABSOLUTE_WITH_SYMLINKS & ~RESOLVE_NO_XDEV), 0); | |
394 | if (ctx->dfd_self_pid < 0) | |
395 | return log_error_errno(-errno, errno, "Failed to open /proc/self"); | |
396 | ||
9b31ab58 CB |
397 | ctx->init_pidfd = lxc_cmd_get_init_pidfd(container->name, container->config_path); |
398 | if (ctx->init_pidfd >= 0) | |
399 | ctx->init_pid = pidfd_get_pid(ctx->dfd_self_pid, ctx->init_pidfd); | |
d8764025 CB |
400 | else |
401 | ctx->init_pid = lxc_cmd_get_init_pid(container->name, container->config_path); | |
402 | ||
500ed813 CB |
403 | if (ctx->init_pid < 0) |
404 | return log_error(-1, "Failed to get init pid"); | |
405 | ||
f51c7eb4 CB |
406 | ret = strnprintf(path, sizeof(path), "/proc/%d", ctx->init_pid); |
407 | if (ret < 0) | |
c538837d | 408 | return ret_errno(EIO); |
e0732705 | 409 | |
5129b2d3 CB |
410 | ctx->dfd_init_pid = open_at(-EBADF, path, |
411 | PROTECT_OPATH_DIRECTORY, | |
412 | (PROTECT_LOOKUP_ABSOLUTE & ~RESOLVE_NO_XDEV), 0); | |
9680e7b0 | 413 | if (ctx->dfd_init_pid < 0) |
4475fabb | 414 | return log_error_errno(-errno, errno, "Failed to open /proc/%d", ctx->init_pid); |
c538837d | 415 | |
9b31ab58 CB |
416 | if (ctx->init_pidfd >= 0) { |
417 | ret = lxc_raw_pidfd_send_signal(ctx->init_pidfd, 0, NULL, 0); | |
d8764025 CB |
418 | if (ret) |
419 | return log_error_errno(-errno, errno, "Container process exited or PID has been recycled"); | |
420 | else | |
421 | TRACE("Container process still running and PID was not recycled"); | |
9b31ab58 CB |
422 | |
423 | if (!pidfd_setns_supported(ctx)) { | |
424 | /* We can't risk leaking file descriptors during attach. */ | |
425 | if (close(ctx->init_pidfd)) | |
426 | return log_error_errno(-errno, errno, "Failed to close pidfd"); | |
427 | ||
428 | ctx->init_pidfd = -EBADF; | |
429 | TRACE("Attaching to namespaces via pidfds not supported"); | |
430 | } | |
d8764025 CB |
431 | } |
432 | ||
4475fabb CB |
433 | /* Determine which namespaces the container was created with. */ |
434 | if (options->namespaces == -1) { | |
435 | options->namespaces = lxc_cmd_get_clone_flags(container->name, container->config_path); | |
436 | if (options->namespaces == -1) | |
437 | return log_error_errno(-EINVAL, EINVAL, "Failed to automatically determine the namespaces which the container uses"); | |
438 | ||
439 | for (int i = 0; i < LXC_NS_MAX; i++) { | |
440 | if (ns_info[i].clone_flag & CLONE_NEWCGROUP) | |
441 | if (!(options->attach_flags & LXC_ATTACH_MOVE_TO_CGROUP) || | |
442 | !cgns_supported()) | |
443 | continue; | |
444 | ||
445 | if (ns_info[i].clone_flag & options->namespaces) | |
446 | continue; | |
447 | ||
448 | ctx->ns_inherited |= ns_info[i].clone_flag; | |
449 | } | |
450 | } | |
451 | ||
9680e7b0 CB |
452 | ret = parse_init_status(ctx, options); |
453 | if (ret) | |
4475fabb | 454 | return log_error_errno(-errno, errno, "Failed to open parse file"); |
e0732705 | 455 | |
4eb19ac0 | 456 | ctx->lsm_ops = lsm_init_static(); |
d701d729 | 457 | |
afc691a0 CB |
458 | if (attach_lsm(options)) { |
459 | if (ctx->attach_flags & LXC_ATTACH_LSM_LABEL) | |
460 | lsm_label = options->lsm_label; | |
461 | else | |
9680e7b0 | 462 | lsm_label = ctx->lsm_ops->process_label_get_at(ctx->lsm_ops, ctx->dfd_init_pid); |
afc691a0 CB |
463 | if (!lsm_label) |
464 | WARN("No security context received"); | |
465 | else | |
466 | INFO("Retrieved security context %s", lsm_label); | |
467 | } | |
e0732705 | 468 | |
ee142207 CB |
469 | ret = get_personality(container->name, container->config_path, &ctx->personality); |
470 | if (ret) | |
471 | return log_error_errno(ret, errno, "Failed to get personality of the container"); | |
d92c8e40 | 472 | |
1874ef74 CB |
473 | if (!ctx->container->lxc_conf) { |
474 | ctx->container->lxc_conf = lxc_conf_init(); | |
475 | if (!ctx->container->lxc_conf) | |
476 | return log_error_errno(-ENOMEM, ENOMEM, "Failed to allocate new lxc config"); | |
477 | } | |
478 | ||
afc691a0 | 479 | ctx->lsm_label = move_ptr(lsm_label); |
9745eb8a | 480 | return 0; |
e0732705 CS |
481 | } |
482 | ||
9b31ab58 | 483 | static int same_nsfd(int dfd_pid1, int dfd_pid2, const char *ns_path) |
299d1198 | 484 | { |
9b31ab58 | 485 | int ret; |
299d1198 CB |
486 | struct stat ns_st1, ns_st2; |
487 | ||
9b31ab58 CB |
488 | ret = fstatat(dfd_pid1, ns_path, &ns_st1, 0); |
489 | if (ret) | |
3cc629fe | 490 | return -1; |
299d1198 | 491 | |
9b31ab58 CB |
492 | ret = fstatat(dfd_pid2, ns_path, &ns_st2, 0); |
493 | if (ret) | |
3cc629fe | 494 | return -1; |
299d1198 CB |
495 | |
496 | /* processes are in the same namespace */ | |
9b31ab58 CB |
497 | if ((ns_st1.st_dev == ns_st2.st_dev) && |
498 | (ns_st1.st_ino == ns_st2.st_ino)) | |
3cc629fe | 499 | return -EINVAL; |
299d1198 | 500 | |
9b31ab58 CB |
501 | return 0; |
502 | } | |
503 | ||
504 | static int same_ns(int dfd_pid1, int dfd_pid2, const char *ns_path) | |
505 | { | |
506 | __do_close int ns_fd2 = -EBADF; | |
507 | int ret = -1; | |
508 | ||
509 | ns_fd2 = open_at(dfd_pid2, ns_path, PROTECT_OPEN_WITH_TRAILING_SYMLINKS, | |
510 | (PROTECT_LOOKUP_BENEATH_WITH_MAGICLINKS & | |
511 | ~(RESOLVE_NO_XDEV | RESOLVE_BENEATH)), 0); | |
512 | if (ns_fd2 < 0) { | |
513 | /* The kernel does not support this namespace. This is not an error. */ | |
514 | if (errno == ENOENT) | |
515 | return -EINVAL; | |
516 | return log_error_errno(-errno, errno, "Failed to open %d(%s)", | |
517 | dfd_pid2, ns_path); | |
518 | } | |
519 | ||
520 | ret = same_nsfd(dfd_pid1, dfd_pid2, ns_path); | |
521 | if (ret < 0) | |
522 | return ret; | |
523 | ||
299d1198 | 524 | /* processes are in different namespaces */ |
3cc629fe | 525 | return move_fd(ns_fd2); |
299d1198 CB |
526 | } |
527 | ||
9b31ab58 CB |
528 | static int __prepare_namespaces_pidfd(struct attach_context *ctx) |
529 | { | |
530 | for (int i = 0; i < LXC_NS_MAX; i++) { | |
531 | int ret; | |
532 | ||
533 | if (!(ctx->ns_inherited & ns_info[i].clone_flag)) | |
534 | continue; | |
535 | ||
536 | ret = same_nsfd(ctx->dfd_self_pid, | |
537 | ctx->dfd_init_pid, | |
538 | ns_info[i].proc_path); | |
539 | if (ret == -EINVAL) | |
540 | ctx->ns_inherited &= ~ns_info[i].clone_flag; | |
541 | else if (ret < 0) | |
542 | return log_error_errno(-1, errno, | |
543 | "Failed to determine whether %s namespace is shared", | |
544 | ns_info[i].proc_name); | |
545 | else | |
546 | TRACE("Shared %s namespace needs attach", ns_info[i].proc_name); | |
547 | } | |
548 | ||
549 | return 0; | |
550 | } | |
551 | ||
552 | static int __prepare_namespaces_nsfd(struct attach_context *ctx, | |
553 | lxc_attach_options_t *options) | |
b7873c95 | 554 | { |
b7873c95 CB |
555 | for (int i = 0; i < LXC_NS_MAX; i++) { |
556 | int j; | |
557 | ||
558 | if (options->namespaces & ns_info[i].clone_flag) | |
5129b2d3 CB |
559 | ctx->ns_fd[i] = open_at(ctx->dfd_init_pid, |
560 | ns_info[i].proc_path, | |
561 | PROTECT_OPEN_WITH_TRAILING_SYMLINKS, | |
9b31ab58 CB |
562 | (PROTECT_LOOKUP_BENEATH_WITH_MAGICLINKS & |
563 | ~(RESOLVE_NO_XDEV | RESOLVE_BENEATH)), | |
5129b2d3 | 564 | 0); |
b7873c95 | 565 | else if (ctx->ns_inherited & ns_info[i].clone_flag) |
5129b2d3 CB |
566 | ctx->ns_fd[i] = same_ns(ctx->dfd_self_pid, |
567 | ctx->dfd_init_pid, | |
568 | ns_info[i].proc_path); | |
b7873c95 CB |
569 | else |
570 | continue; | |
571 | ||
572 | if (ctx->ns_fd[i] >= 0) | |
573 | continue; | |
574 | ||
575 | if (ctx->ns_fd[i] == -EINVAL) { | |
b7873c95 CB |
576 | ctx->ns_inherited &= ~ns_info[i].clone_flag; |
577 | continue; | |
578 | } | |
579 | ||
580 | /* We failed to preserve the namespace. */ | |
9b31ab58 CB |
581 | SYSERROR("Failed to preserve %s namespace of %d", |
582 | ns_info[i].proc_name, ctx->init_pid); | |
b7873c95 CB |
583 | |
584 | /* Close all already opened file descriptors before we return an | |
585 | * error, so we don't leak them. | |
586 | */ | |
587 | for (j = 0; j < i; j++) | |
588 | close_prot_errno_disarm(ctx->ns_fd[j]); | |
589 | ||
590 | return -1; | |
591 | } | |
592 | ||
593 | return 0; | |
594 | } | |
595 | ||
9b31ab58 CB |
596 | static int prepare_namespaces(struct attach_context *ctx, |
597 | lxc_attach_options_t *options) | |
b7873c95 | 598 | { |
9b31ab58 CB |
599 | if (ctx->init_pidfd < 0) |
600 | return __prepare_namespaces_nsfd(ctx, options); | |
601 | ||
602 | return __prepare_namespaces_pidfd(ctx); | |
b7873c95 CB |
603 | } |
604 | ||
9b31ab58 | 605 | static inline void put_namespaces(struct attach_context *ctx) |
b7873c95 | 606 | { |
9b31ab58 CB |
607 | if (ctx->init_pidfd < 0) { |
608 | for (int i = 0; i < LXC_NS_MAX; i++) | |
609 | close_prot_errno_disarm(ctx->ns_fd[i]); | |
610 | } | |
611 | } | |
b7873c95 | 612 | |
9b31ab58 CB |
613 | static int __attach_namespaces_pidfd(struct attach_context *ctx, |
614 | lxc_attach_options_t *options) | |
615 | { | |
616 | unsigned int ns_flags = options->namespaces | ctx->ns_inherited; | |
617 | int ret; | |
b7873c95 | 618 | |
9b31ab58 CB |
619 | /* The common case is to attach to all namespaces. */ |
620 | ret = setns(ctx->init_pidfd, ns_flags); | |
621 | if (ret) | |
622 | return log_error_errno(-errno, errno, | |
623 | "Failed to attach to namespaces via pidfd"); | |
624 | ||
625 | /* We can't risk leaking file descriptors into the container. */ | |
626 | if (close(ctx->init_pidfd)) | |
627 | return log_error_errno(-errno, errno, "Failed to close pidfd"); | |
628 | ctx->init_pidfd = -EBADF; | |
629 | ||
630 | return log_trace(0, "Attached to container namespaces via pidfd"); | |
b7873c95 CB |
631 | } |
632 | ||
9b31ab58 CB |
633 | static int __attach_namespaces_nsfd(struct attach_context *ctx, |
634 | lxc_attach_options_t *options) | |
99d50954 | 635 | { |
92466fe3 CB |
636 | int fret = 0; |
637 | ||
ffeeed8b CB |
638 | for (int i = 0; i < LXC_NS_MAX; i++) { |
639 | int ret; | |
99d50954 | 640 | |
877f3a04 | 641 | if (ctx->ns_fd[i] < 0) |
26818618 CB |
642 | continue; |
643 | ||
21d0acc2 | 644 | ret = setns(ctx->ns_fd[i], ns_info[i].clone_flag); |
92466fe3 | 645 | if (ret) |
9b31ab58 CB |
646 | return log_error_errno(-errno, errno, |
647 | "Failed to attach to %s namespace of %d", | |
648 | ns_info[i].proc_name, | |
649 | ctx->init_pid); | |
92466fe3 CB |
650 | |
651 | if (close(ctx->ns_fd[i])) { | |
652 | fret = -errno; | |
9b31ab58 CB |
653 | SYSERROR("Failed to close file descriptor for %s namespace", |
654 | ns_info[i].proc_name); | |
92466fe3 CB |
655 | } |
656 | ctx->ns_fd[i] = -EBADF; | |
99d50954 CS |
657 | } |
658 | ||
92466fe3 | 659 | return fret; |
99d50954 CS |
660 | } |
661 | ||
9b31ab58 CB |
662 | static int attach_namespaces(struct attach_context *ctx, |
663 | lxc_attach_options_t *options) | |
664 | { | |
665 | if (lxc_log_trace()) { | |
666 | for (int i = 0; i < LXC_NS_MAX; i++) { | |
667 | if (ns_info[i].clone_flag & options->namespaces) { | |
668 | TRACE("Attaching to %s namespace", ns_info[i].proc_name); | |
669 | continue; | |
670 | } | |
671 | if (ns_info[i].clone_flag & ctx->ns_inherited) { | |
672 | TRACE("Sharing %s namespace", ns_info[i].proc_name); | |
673 | continue; | |
674 | } | |
675 | TRACE("Inheriting %s namespace", ns_info[i].proc_name); | |
676 | } | |
677 | } | |
678 | ||
679 | if (ctx->init_pidfd < 0) | |
680 | return __attach_namespaces_nsfd(ctx, options); | |
681 | ||
682 | return __attach_namespaces_pidfd(ctx, options); | |
683 | } | |
684 | ||
685 | static void put_attach_context(struct attach_context *ctx) | |
686 | { | |
687 | if (ctx) { | |
688 | if (!(ctx->attach_flags & LXC_ATTACH_LSM_LABEL)) | |
689 | free_disarm(ctx->lsm_label); | |
690 | close_prot_errno_disarm(ctx->dfd_init_pid); | |
691 | ||
692 | if (ctx->container) { | |
693 | lxc_container_put(ctx->container); | |
694 | ctx->container = NULL; | |
695 | } | |
696 | ||
697 | put_namespaces(ctx); | |
698 | free(ctx); | |
699 | } | |
700 | } | |
701 | ||
c538837d CB |
702 | /* |
703 | * Place anything in here that needs to be get rid of before we move into the | |
704 | * container's context and fail hard if we can't. | |
705 | */ | |
706 | static bool attach_context_security_barrier(struct attach_context *ctx) | |
707 | { | |
708 | if (ctx) { | |
25c659d5 CB |
709 | if (close(ctx->dfd_self_pid)) |
710 | return false; | |
711 | ctx->dfd_self_pid = -EBADF; | |
712 | ||
713 | if (close(ctx->dfd_init_pid)) | |
c538837d | 714 | return false; |
25c659d5 | 715 | ctx->dfd_init_pid = -EBADF; |
c538837d CB |
716 | } |
717 | ||
718 | return true; | |
719 | } | |
720 | ||
e4103cf6 | 721 | int lxc_attach_remount_sys_proc(void) |
7a0b0b56 CS |
722 | { |
723 | int ret; | |
724 | ||
725 | ret = unshare(CLONE_NEWNS); | |
ffeeed8b CB |
726 | if (ret < 0) |
727 | return log_error_errno(-1, errno, "Failed to unshare mount namespace"); | |
7a0b0b56 | 728 | |
9e61fb1f CB |
729 | if (detect_shared_rootfs() && mount(NULL, "/", NULL, MS_SLAVE | MS_REC, NULL)) |
730 | SYSERROR("Failed to recursively turn root mount tree into dependent mount. Continuing..."); | |
2c6f3fc9 | 731 | |
8ce83369 | 732 | /* Assume /proc is always mounted, so remount it. */ |
7a0b0b56 | 733 | ret = umount2("/proc", MNT_DETACH); |
ffeeed8b CB |
734 | if (ret < 0) |
735 | return log_error_errno(-1, errno, "Failed to unmount /proc"); | |
7a0b0b56 | 736 | |
0d50f288 | 737 | ret = mount("none", "/proc", "proc", 0, NULL); |
ffeeed8b CB |
738 | if (ret < 0) |
739 | return log_error_errno(-1, errno, "Failed to remount /proc"); | |
7a0b0b56 | 740 | |
ffeeed8b CB |
741 | /* |
742 | * Try to umount /sys. If it's not a mount point, we'll get EINVAL, then | |
8ce83369 | 743 | * we ignore it because it may not have been mounted in the first place. |
7a0b0b56 CS |
744 | */ |
745 | ret = umount2("/sys", MNT_DETACH); | |
ffeeed8b CB |
746 | if (ret < 0 && errno != EINVAL) |
747 | return log_error_errno(-1, errno, "Failed to unmount /sys"); | |
748 | ||
749 | /* Remount it. */ | |
0d50f288 | 750 | if (ret == 0 && mount("none", "/sys", "sysfs", 0, NULL)) |
ffeeed8b | 751 | return log_error_errno(-1, errno, "Failed to remount /sys"); |
7a0b0b56 CS |
752 | |
753 | return 0; | |
754 | } | |
755 | ||
677e1d27 | 756 | static int drop_capabilities(struct attach_context *ctx) |
e0732705 | 757 | { |
ffeeed8b | 758 | int last_cap; |
e0732705 | 759 | |
6f4f1937 | 760 | last_cap = lxc_caps_last_cap(); |
ffeeed8b | 761 | for (int cap = 0; cap <= last_cap; cap++) { |
e0732705 CS |
762 | if (ctx->capability_mask & (1LL << cap)) |
763 | continue; | |
764 | ||
b81689a1 | 765 | if (prctl(PR_CAPBSET_DROP, prctl_arg(cap), prctl_arg(0), |
ffeeed8b CB |
766 | prctl_arg(0), prctl_arg(0))) |
767 | return log_error_errno(-1, errno, "Failed to drop capability %d", cap); | |
ea918412 | 768 | |
94ac256f | 769 | TRACE("Dropped capability %d", cap); |
e0732705 CS |
770 | } |
771 | ||
772 | return 0; | |
773 | } | |
905022f7 | 774 | |
ab919e5f | 775 | static int lxc_attach_set_environment(struct attach_context *ctx, |
7385273f | 776 | enum lxc_attach_env_policy_t policy, |
6f4f1937 | 777 | char **extra_env, char **extra_keep) |
b3a39ba6 | 778 | { |
3d55242a | 779 | int ret; |
7385273f | 780 | struct lxc_list *iterator; |
781 | ||
799f96fd | 782 | if (policy == LXC_ATTACH_CLEAR_ENV) { |
3d5e9f48 | 783 | int path_kept = 0; |
6f4f1937 | 784 | char **extra_keep_store = NULL; |
3d5e9f48 CS |
785 | |
786 | if (extra_keep) { | |
787 | size_t count, i; | |
788 | ||
3d55242a CB |
789 | for (count = 0; extra_keep[count]; count++) |
790 | ; | |
3d5e9f48 | 791 | |
89b7bfe3 | 792 | extra_keep_store = zalloc(count * sizeof(char *)); |
3d55242a | 793 | if (!extra_keep_store) |
3d5e9f48 | 794 | return -1; |
3d55242a | 795 | |
3d5e9f48 CS |
796 | for (i = 0; i < count; i++) { |
797 | char *v = getenv(extra_keep[i]); | |
798 | if (v) { | |
799 | extra_keep_store[i] = strdup(v); | |
800 | if (!extra_keep_store[i]) { | |
3d5e9f48 CS |
801 | while (i > 0) |
802 | free(extra_keep_store[--i]); | |
ea918412 | 803 | |
3d5e9f48 CS |
804 | free(extra_keep_store); |
805 | return -1; | |
806 | } | |
3d55242a | 807 | |
e8c43357 | 808 | if (strequal(extra_keep[i], "PATH")) |
3d5e9f48 CS |
809 | path_kept = 1; |
810 | } | |
3d5e9f48 CS |
811 | } |
812 | } | |
813 | ||
799f96fd | 814 | if (clearenv()) { |
a9cab7e3 | 815 | if (extra_keep_store) { |
3d55242a CB |
816 | char **p; |
817 | ||
a9cab7e3 CS |
818 | for (p = extra_keep_store; *p; p++) |
819 | free(*p); | |
3d55242a | 820 | |
a9cab7e3 CS |
821 | free(extra_keep_store); |
822 | } | |
3d55242a | 823 | |
ffeeed8b | 824 | return log_error(-1, "Failed to clear environment"); |
3d5e9f48 CS |
825 | } |
826 | ||
827 | if (extra_keep_store) { | |
828 | size_t i; | |
6f4f1937 | 829 | |
3d5e9f48 | 830 | for (i = 0; extra_keep[i]; i++) { |
acd4922e | 831 | if (extra_keep_store[i]) { |
3d55242a CB |
832 | ret = setenv(extra_keep[i], extra_keep_store[i], 1); |
833 | if (ret < 0) | |
a24c5678 | 834 | SYSWARN("Failed to set environment variable"); |
acd4922e | 835 | } |
ea918412 | 836 | |
3d5e9f48 CS |
837 | free(extra_keep_store[i]); |
838 | } | |
ea918412 | 839 | |
3d5e9f48 CS |
840 | free(extra_keep_store); |
841 | } | |
842 | ||
8ce83369 CB |
843 | /* Always set a default path; shells and execlp tend to be fine |
844 | * without it, but there is a disturbing number of C programs | |
845 | * out there that just assume that getenv("PATH") is never NULL | |
846 | * and then die a painful segfault death. | |
847 | */ | |
3d55242a CB |
848 | if (!path_kept) { |
849 | ret = setenv("PATH", "/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin", 1); | |
850 | if (ret < 0) | |
a24c5678 | 851 | SYSWARN("Failed to set environment variable"); |
3d55242a | 852 | } |
b3a39ba6 DW |
853 | } |
854 | ||
3d55242a | 855 | ret = putenv("container=lxc"); |
ffeeed8b | 856 | if (ret < 0) |
818a57fc | 857 | return log_warn(-1, "Failed to set environment variable"); |
b3a39ba6 | 858 | |
7385273f | 859 | /* Set container environment variables.*/ |
640952e5 | 860 | if (ctx->container->lxc_conf) { |
ab919e5f | 861 | lxc_list_for_each(iterator, &ctx->container->lxc_conf->environment) { |
3d55242a CB |
862 | char *env_tmp; |
863 | ||
864 | env_tmp = strdup((char *)iterator->elem); | |
865 | if (!env_tmp) | |
7385273f | 866 | return -1; |
7385273f | 867 | |
3d55242a | 868 | ret = putenv(env_tmp); |
ffeeed8b CB |
869 | if (ret < 0) |
870 | return log_error_errno(-1, errno, "Failed to set environment variable: %s", (char *)iterator->elem); | |
7385273f | 871 | } |
872 | } | |
873 | ||
8ce83369 | 874 | /* Set extra environment variables. */ |
3d5e9f48 CS |
875 | if (extra_env) { |
876 | for (; *extra_env; extra_env++) { | |
3d55242a | 877 | char *p; |
ea918412 | 878 | |
8ce83369 CB |
879 | /* We just assume the user knows what they are doing, so |
880 | * we don't do any checks. | |
881 | */ | |
3d55242a CB |
882 | p = strdup(*extra_env); |
883 | if (!p) | |
3d5e9f48 | 884 | return -1; |
3d55242a CB |
885 | |
886 | ret = putenv(p); | |
887 | if (ret < 0) | |
a24c5678 | 888 | SYSWARN("Failed to set environment variable"); |
3d5e9f48 CS |
889 | } |
890 | } | |
891 | ||
b3a39ba6 DW |
892 | return 0; |
893 | } | |
894 | ||
74a3920a | 895 | static char *lxc_attach_getpwshell(uid_t uid) |
905022f7 | 896 | { |
1b9c9f5b | 897 | __do_free char *line = NULL, *result = NULL; |
cd8f5663 | 898 | __do_fclose FILE *pipe_f = NULL; |
6f4f1937 | 899 | int fd, ret; |
905022f7 CS |
900 | pid_t pid; |
901 | int pipes[2]; | |
3fa23ac3 CB |
902 | bool found = false; |
903 | size_t line_bufsz = 0; | |
905022f7 | 904 | |
8ce83369 CB |
905 | /* We need to fork off a process that runs the getent program, and we |
906 | * need to capture its output, so we use a pipe for that purpose. | |
905022f7 | 907 | */ |
3fa23ac3 | 908 | ret = pipe2(pipes, O_CLOEXEC); |
905022f7 CS |
909 | if (ret < 0) |
910 | return NULL; | |
911 | ||
912 | pid = fork(); | |
913 | if (pid < 0) { | |
914 | close(pipes[0]); | |
915 | close(pipes[1]); | |
916 | return NULL; | |
917 | } | |
918 | ||
3fa23ac3 | 919 | if (!pid) { |
905022f7 CS |
920 | char uid_buf[32]; |
921 | char *arguments[] = { | |
922 | "getent", | |
923 | "passwd", | |
924 | uid_buf, | |
925 | NULL | |
926 | }; | |
927 | ||
928 | close(pipes[0]); | |
929 | ||
8ce83369 | 930 | /* We want to capture stdout. */ |
3fa23ac3 | 931 | ret = dup2(pipes[1], STDOUT_FILENO); |
905022f7 | 932 | close(pipes[1]); |
3fa23ac3 | 933 | if (ret < 0) |
ea918412 | 934 | _exit(EXIT_FAILURE); |
905022f7 | 935 | |
8ce83369 CB |
936 | /* Get rid of stdin/stderr, so we try to associate it with |
937 | * /dev/null. | |
905022f7 | 938 | */ |
3fa23ac3 | 939 | fd = open_devnull(); |
905022f7 | 940 | if (fd < 0) { |
3fa23ac3 CB |
941 | close(STDIN_FILENO); |
942 | close(STDERR_FILENO); | |
905022f7 | 943 | } else { |
3fa23ac3 | 944 | (void)dup3(fd, STDIN_FILENO, O_CLOEXEC); |
59f0e209 | 945 | (void)dup3(fd, STDERR_FILENO, O_CLOEXEC); |
905022f7 CS |
946 | close(fd); |
947 | } | |
948 | ||
8ce83369 | 949 | /* Finish argument list. */ |
f51c7eb4 CB |
950 | ret = strnprintf(uid_buf, sizeof(uid_buf), "%ld", (long)uid); |
951 | if (ret <= 0) | |
ea918412 | 952 | _exit(EXIT_FAILURE); |
905022f7 | 953 | |
8ce83369 | 954 | /* Try to run getent program. */ |
3fa23ac3 | 955 | (void)execvp("getent", arguments); |
ea918412 | 956 | _exit(EXIT_FAILURE); |
905022f7 | 957 | } |
3fa23ac3 CB |
958 | |
959 | close(pipes[1]); | |
960 | ||
4110345b | 961 | pipe_f = fdopen(pipes[0], "re"); |
cf4026f1 CB |
962 | if (!pipe_f) { |
963 | close(pipes[0]); | |
964 | goto reap_child; | |
965 | } | |
966 | /* Transfer ownership of pipes[0] to pipe_f. */ | |
967 | move_fd(pipes[0]); | |
968 | ||
3fa23ac3 CB |
969 | while (getline(&line, &line_bufsz, pipe_f) != -1) { |
970 | int i; | |
971 | long value; | |
972 | char *token; | |
973 | char *endptr = NULL, *saveptr = NULL; | |
974 | ||
975 | /* If we already found something, just continue to read | |
976 | * until the pipe doesn't deliver any more data, but | |
977 | * don't modify the existing data structure. | |
978 | */ | |
979 | if (found) | |
980 | continue; | |
981 | ||
18d4ffde | 982 | if (!line) |
983 | continue; | |
984 | ||
3fa23ac3 CB |
985 | /* Trim line on the right hand side. */ |
986 | for (i = strlen(line); i > 0 && (line[i - 1] == '\n' || line[i - 1] == '\r'); --i) | |
987 | line[i - 1] = '\0'; | |
988 | ||
989 | /* Split into tokens: first: user name. */ | |
990 | token = strtok_r(line, ":", &saveptr); | |
991 | if (!token) | |
992 | continue; | |
993 | ||
994 | /* next: dummy password field */ | |
995 | token = strtok_r(NULL, ":", &saveptr); | |
996 | if (!token) | |
997 | continue; | |
998 | ||
999 | /* next: user id */ | |
1000 | token = strtok_r(NULL, ":", &saveptr); | |
1001 | value = token ? strtol(token, &endptr, 10) : 0; | |
1002 | if (!token || !endptr || *endptr || value == LONG_MIN || | |
ea918412 | 1003 | value == LONG_MAX) |
3fa23ac3 CB |
1004 | continue; |
1005 | ||
1006 | /* dummy sanity check: user id matches */ | |
1007 | if ((uid_t)value != uid) | |
1008 | continue; | |
1009 | ||
1010 | /* skip fields: gid, gecos, dir, go to next field 'shell' */ | |
1011 | for (i = 0; i < 4; i++) { | |
1012 | token = strtok_r(NULL, ":", &saveptr); | |
1013 | if (!token) | |
1014 | continue; | |
1015 | } | |
ea918412 | 1016 | |
3fa23ac3 CB |
1017 | if (!token) |
1018 | continue; | |
ea918412 | 1019 | |
1b9c9f5b | 1020 | free_disarm(result); |
3fa23ac3 CB |
1021 | result = strdup(token); |
1022 | ||
1023 | /* Sanity check that there are no fields after that. */ | |
1024 | token = strtok_r(NULL, ":", &saveptr); | |
1025 | if (token) | |
1026 | continue; | |
1027 | ||
1028 | found = true; | |
1029 | } | |
ea918412 | 1030 | |
cf4026f1 | 1031 | reap_child: |
3fa23ac3 | 1032 | ret = wait_for_pid(pid); |
1b9c9f5b | 1033 | if (ret < 0) |
3fa23ac3 | 1034 | return NULL; |
3fa23ac3 | 1035 | |
1b9c9f5b | 1036 | if (!found) |
3fa23ac3 | 1037 | return NULL; |
3fa23ac3 | 1038 | |
1b9c9f5b | 1039 | return move_ptr(result); |
905022f7 | 1040 | } |
cb3e61fa | 1041 | |
d4db3d14 | 1042 | static bool fetch_seccomp(struct lxc_container *c, lxc_attach_options_t *options) |
2c4ea790 | 1043 | { |
cd8f5663 | 1044 | __do_free char *path = NULL; |
d4db3d14 CB |
1045 | int ret; |
1046 | bool bret; | |
2eef2bda | 1047 | |
afc691a0 | 1048 | if (!attach_lsm(options)) { |
cca66e06 | 1049 | free_disarm(c->lxc_conf->seccomp.seccomp); |
2c4ea790 | 1050 | return true; |
bd4307f0 | 1051 | } |
bd7b4e28 | 1052 | |
afc691a0 | 1053 | /* Remove current setting. */ |
d4db3d14 | 1054 | if (!c->set_config_item(c, "lxc.seccomp.profile", "") && |
ea918412 | 1055 | !c->set_config_item(c, "lxc.seccomp", "")) |
2c4ea790 | 1056 | return false; |
bd7b4e28 | 1057 | |
8ce83369 | 1058 | /* Fetch the current profile path over the cmd interface. */ |
0b427da0 | 1059 | path = c->get_running_config_item(c, "lxc.seccomp.profile"); |
bd7b4e28 | 1060 | if (!path) { |
d4db3d14 | 1061 | INFO("Failed to retrieve lxc.seccomp.profile"); |
ea918412 | 1062 | |
0b427da0 | 1063 | path = c->get_running_config_item(c, "lxc.seccomp"); |
cca66e06 CB |
1064 | if (!path) |
1065 | return log_info(true, "Failed to retrieve lxc.seccomp"); | |
bd7b4e28 SG |
1066 | } |
1067 | ||
8ce83369 | 1068 | /* Copy the value into the new lxc_conf. */ |
d4db3d14 | 1069 | bret = c->set_config_item(c, "lxc.seccomp.profile", path); |
d4db3d14 CB |
1070 | if (!bret) |
1071 | return false; | |
bd7b4e28 | 1072 | |
8ce83369 | 1073 | /* Attempt to parse the resulting config. */ |
d4db3d14 | 1074 | ret = lxc_read_seccomp_config(c->lxc_conf); |
cca66e06 CB |
1075 | if (ret < 0) |
1076 | return log_error(false, "Failed to retrieve seccomp policy"); | |
2c4ea790 | 1077 | |
cca66e06 | 1078 | return log_info(true, "Retrieved seccomp policy"); |
2e812c16 CB |
1079 | } |
1080 | ||
6f4f1937 | 1081 | static bool no_new_privs(struct lxc_container *c, lxc_attach_options_t *options) |
2e812c16 | 1082 | { |
cd8f5663 | 1083 | __do_free char *val = NULL; |
2e812c16 | 1084 | |
2e812c16 | 1085 | /* Remove current setting. */ |
02d3b72b CB |
1086 | if (!c->set_config_item(c, "lxc.no_new_privs", "")) |
1087 | return log_info(false, "Failed to unset lxc.no_new_privs"); | |
2e812c16 CB |
1088 | |
1089 | /* Retrieve currently active setting. */ | |
1090 | val = c->get_running_config_item(c, "lxc.no_new_privs"); | |
02d3b72b CB |
1091 | if (!val) |
1092 | return log_info(false, "Failed to retrieve lxc.no_new_privs"); | |
2e812c16 CB |
1093 | |
1094 | /* Set currently active setting. */ | |
cd8f5663 | 1095 | return c->set_config_item(c, "lxc.no_new_privs", val); |
2c4ea790 SH |
1096 | } |
1097 | ||
338b230f | 1098 | struct attach_payload { |
a998454a | 1099 | int ipc_socket; |
cecf3e83 | 1100 | int terminal_pts_fd; |
a998454a | 1101 | lxc_attach_options_t *options; |
ab919e5f | 1102 | struct attach_context *ctx; |
a998454a CB |
1103 | lxc_attach_exec_t exec_function; |
1104 | void *exec_payload; | |
1105 | }; | |
1106 | ||
338b230f | 1107 | static void put_attach_payload(struct attach_payload *p) |
ba2be1a8 | 1108 | { |
afc691a0 CB |
1109 | if (p) { |
1110 | close_prot_errno_disarm(p->ipc_socket); | |
1111 | close_prot_errno_disarm(p->terminal_pts_fd); | |
dd53c8af | 1112 | put_attach_context(p->ctx); |
ab919e5f | 1113 | p->ctx = NULL; |
b21da190 | 1114 | } |
ba2be1a8 CB |
1115 | } |
1116 | ||
338b230f | 1117 | __noreturn static void do_attach(struct attach_payload *ap) |
a998454a | 1118 | { |
afc691a0 CB |
1119 | lxc_attach_exec_t attach_function = move_ptr(ap->exec_function); |
1120 | void *attach_function_args = move_ptr(ap->exec_payload); | |
427a8067 | 1121 | int lsm_fd, ret; |
338b230f CB |
1122 | lxc_attach_options_t* options = ap->options; |
1123 | struct attach_context *ctx = ap->ctx; | |
ab919e5f | 1124 | struct lxc_conf *conf = ctx->container->lxc_conf; |
a998454a CB |
1125 | |
1126 | /* A description of the purpose of this functionality is provided in the | |
1127 | * lxc-attach(1) manual page. We have to remount here and not in the | |
1128 | * parent process, otherwise /proc may not properly reflect the new pid | |
1129 | * namespace. | |
1130 | */ | |
1131 | if (!(options->namespaces & CLONE_NEWNS) && | |
1132 | (options->attach_flags & LXC_ATTACH_REMOUNT_PROC_SYS)) { | |
1133 | ret = lxc_attach_remount_sys_proc(); | |
b75c344c CB |
1134 | if (ret < 0) |
1135 | goto on_error; | |
ea918412 | 1136 | |
b75c344c | 1137 | TRACE("Remounted \"/proc\" and \"/sys\""); |
a998454a CB |
1138 | } |
1139 | ||
5b514ce3 | 1140 | /* Now perform additional attachments. */ |
a998454a | 1141 | #if HAVE_SYS_PERSONALITY_H |
a998454a | 1142 | if (options->attach_flags & LXC_ATTACH_SET_PERSONALITY) { |
b75c344c CB |
1143 | long new_personality; |
1144 | ||
1145 | if (options->personality < 0) | |
ab919e5f | 1146 | new_personality = ctx->personality; |
b75c344c CB |
1147 | else |
1148 | new_personality = options->personality; | |
ea918412 | 1149 | |
ee142207 CB |
1150 | if (new_personality != LXC_ARCH_UNCHANGED) { |
1151 | ret = personality(new_personality); | |
1152 | if (ret < 0) | |
1153 | goto on_error; | |
ea918412 | 1154 | |
ee142207 CB |
1155 | TRACE("Set new personality"); |
1156 | } | |
a998454a CB |
1157 | } |
1158 | #endif | |
1159 | ||
1160 | if (options->attach_flags & LXC_ATTACH_DROP_CAPABILITIES) { | |
677e1d27 | 1161 | ret = drop_capabilities(ctx); |
b75c344c CB |
1162 | if (ret < 0) |
1163 | goto on_error; | |
ea918412 | 1164 | |
b75c344c | 1165 | TRACE("Dropped capabilities"); |
a998454a CB |
1166 | } |
1167 | ||
1168 | /* Always set the environment (specify (LXC_ATTACH_KEEP_ENV, NULL, NULL) | |
1169 | * if you want this to be a no-op). | |
1170 | */ | |
ab919e5f | 1171 | ret = lxc_attach_set_environment(ctx, |
7385273f | 1172 | options->env_policy, |
a998454a CB |
1173 | options->extra_env_vars, |
1174 | options->extra_keep_env); | |
b75c344c CB |
1175 | if (ret < 0) |
1176 | goto on_error; | |
ea918412 | 1177 | |
b75c344c | 1178 | TRACE("Set up environment"); |
a998454a | 1179 | |
afc691a0 CB |
1180 | /* |
1181 | * This remark only affects fully unprivileged containers: | |
57de839f CB |
1182 | * Receive fd for LSM security module before we set{g,u}id(). The reason |
1183 | * is that on set{g,u}id() the kernel will a) make us undumpable and b) | |
1184 | * we will change our effective uid. This means our effective uid will | |
1185 | * be different from the effective uid of the process that created us | |
1186 | * which means that this processs no longer has capabilities in our | |
1187 | * namespace including CAP_SYS_PTRACE. This means we will not be able to | |
1188 | * read and /proc/<pid> files for the process anymore when /proc is | |
1189 | * mounted with hidepid={1,2}. So let's get the lsm label fd before the | |
1190 | * set{g,u}id(). | |
1191 | */ | |
afc691a0 | 1192 | if (attach_lsm(options) && ctx->lsm_label) { |
f8e88e94 | 1193 | if (!sync_wait_fd(ap->ipc_socket, ATTACH_SYNC_LSM(&lsm_fd))) { |
6e36c297 | 1194 | SYSERROR("Failed to receive lsm label fd"); |
b75c344c | 1195 | goto on_error; |
9044b79e | 1196 | } |
1197 | ||
57de839f CB |
1198 | TRACE("Received LSM label file descriptor %d from parent", lsm_fd); |
1199 | } | |
1200 | ||
08ea9270 | 1201 | if (options->stdin_fd > 0 && isatty(options->stdin_fd)) { |
cd0a2b2f | 1202 | ret = lxc_make_controlling_terminal(options->stdin_fd); |
08ea9270 CB |
1203 | if (ret < 0) |
1204 | goto on_error; | |
1205 | } | |
1206 | ||
9475d2b9 CB |
1207 | if ((options->attach_flags & LXC_ATTACH_SETGROUPS) && |
1208 | options->groups.size > 0) { | |
8caac583 RJ |
1209 | if (!lxc_setgroups(options->groups.list, options->groups.size)) |
1210 | goto on_error; | |
1211 | } else { | |
1212 | if (!lxc_drop_groups() && errno != EPERM) | |
1213 | goto on_error; | |
1214 | } | |
b58214ac | 1215 | |
4475fabb | 1216 | if (options->namespaces & CLONE_NEWUSER) |
3ac4480a | 1217 | if (!lxc_switch_uid_gid(ctx->setup_ns_uid, ctx->setup_ns_gid)) |
b75c344c | 1218 | goto on_error; |
936efc72 | 1219 | |
afc691a0 | 1220 | if (attach_lsm(options) && ctx->lsm_label) { |
d3ba7c98 | 1221 | bool on_exec; |
a998454a CB |
1222 | |
1223 | /* Change into our new LSM profile. */ | |
d3ba7c98 | 1224 | on_exec = options->attach_flags & LXC_ATTACH_LSM_EXEC ? true : false; |
afc691a0 | 1225 | ret = ctx->lsm_ops->process_label_set_at(ctx->lsm_ops, lsm_fd, ctx->lsm_label, on_exec); |
cb2420df | 1226 | close_prot_errno_disarm(lsm_fd); |
b75c344c CB |
1227 | if (ret < 0) |
1228 | goto on_error; | |
ea918412 | 1229 | |
ab919e5f | 1230 | TRACE("Set %s LSM label to \"%s\"", ctx->lsm_ops->name, ctx->lsm_label); |
a998454a CB |
1231 | } |
1232 | ||
640952e5 | 1233 | if (conf->no_new_privs || (options->attach_flags & LXC_ATTACH_NO_NEW_PRIVS)) { |
6ce8e678 AL |
1234 | ret = prctl(PR_SET_NO_NEW_PRIVS, prctl_arg(1), prctl_arg(0), |
1235 | prctl_arg(0), prctl_arg(0)); | |
1236 | if (ret < 0) | |
1237 | goto on_error; | |
1238 | ||
1239 | TRACE("Set PR_SET_NO_NEW_PRIVS"); | |
1240 | } | |
1241 | ||
a998454a CB |
1242 | /* The following is done after the communication socket is shut down. |
1243 | * That way, all errors that might (though unlikely) occur up until this | |
1244 | * point will have their messages printed to the original stderr (if | |
1245 | * logging is so configured) and not the fd the user supplied, if any. | |
1246 | */ | |
1247 | ||
1248 | /* Fd handling for stdin, stdout and stderr; ignore errors here, user | |
1249 | * may want to make sure the fds are closed, for example. | |
1250 | */ | |
08ea9270 | 1251 | if (options->stdin_fd >= 0 && options->stdin_fd != STDIN_FILENO) |
40301d48 | 1252 | if (dup2(options->stdin_fd, STDIN_FILENO) < 0) |
a7563434 | 1253 | SYSDEBUG("Failed to replace stdin with %d", options->stdin_fd); |
08ea9270 CB |
1254 | |
1255 | if (options->stdout_fd >= 0 && options->stdout_fd != STDOUT_FILENO) | |
40301d48 | 1256 | if (dup2(options->stdout_fd, STDOUT_FILENO) < 0) |
93b9960a | 1257 | SYSDEBUG("Failed to replace stdout with %d", options->stdout_fd); |
08ea9270 CB |
1258 | |
1259 | if (options->stderr_fd >= 0 && options->stderr_fd != STDERR_FILENO) | |
40301d48 | 1260 | if (dup2(options->stderr_fd, STDERR_FILENO) < 0) |
93b9960a | 1261 | SYSDEBUG("Failed to replace stderr with %d", options->stderr_fd); |
a998454a CB |
1262 | |
1263 | /* close the old fds */ | |
08ea9270 | 1264 | if (options->stdin_fd > STDERR_FILENO) |
a998454a | 1265 | close(options->stdin_fd); |
08ea9270 CB |
1266 | |
1267 | if (options->stdout_fd > STDERR_FILENO) | |
a998454a | 1268 | close(options->stdout_fd); |
08ea9270 CB |
1269 | |
1270 | if (options->stderr_fd > STDERR_FILENO) | |
a998454a CB |
1271 | close(options->stderr_fd); |
1272 | ||
427a8067 CB |
1273 | /* |
1274 | * Try to remove FD_CLOEXEC flag from stdin/stdout/stderr, but also | |
a998454a CB |
1275 | * here, ignore errors. |
1276 | */ | |
427a8067 | 1277 | for (int fd = STDIN_FILENO; fd <= STDERR_FILENO; fd++) { |
3f62938a | 1278 | ret = fd_cloexec(fd, false); |
b75c344c CB |
1279 | if (ret < 0) { |
1280 | SYSERROR("Failed to clear FD_CLOEXEC from file descriptor %d", fd); | |
1281 | goto on_error; | |
1282 | } | |
a998454a CB |
1283 | } |
1284 | ||
9e84479f | 1285 | if (options->attach_flags & LXC_ATTACH_TERMINAL) { |
338b230f | 1286 | ret = lxc_terminal_prepare_login(ap->terminal_pts_fd); |
ba2be1a8 | 1287 | if (ret < 0) { |
338b230f | 1288 | SYSERROR("Failed to prepare terminal file descriptor %d", ap->terminal_pts_fd); |
ba2be1a8 CB |
1289 | goto on_error; |
1290 | } | |
ea918412 | 1291 | |
338b230f | 1292 | TRACE("Prepared terminal file descriptor %d", ap->terminal_pts_fd); |
ba2be1a8 CB |
1293 | } |
1294 | ||
4475fabb | 1295 | /* Avoid unnecessary syscalls. */ |
3ac4480a CB |
1296 | if (ctx->setup_ns_uid == ctx->target_ns_uid) |
1297 | ctx->target_ns_uid = LXC_INVALID_UID; | |
1298 | ||
1299 | if (ctx->setup_ns_gid == ctx->target_ns_gid) | |
1300 | ctx->target_ns_gid = LXC_INVALID_GID; | |
4475fabb | 1301 | |
3ac4480a CB |
1302 | /* |
1303 | * Make sure that the processes STDIO is correctly owned by the user | |
1304 | * that we are switching to. | |
1305 | */ | |
1306 | ret = fix_stdio_permissions(ctx->target_ns_uid); | |
1307 | if (ret) | |
1308 | INFO("Failed to adjust stdio permissions"); | |
4475fabb | 1309 | |
e18aba7d CB |
1310 | if (conf->seccomp.seccomp) { |
1311 | ret = lxc_seccomp_load(conf); | |
1312 | if (ret < 0) | |
1313 | goto on_error; | |
1314 | ||
1315 | TRACE("Loaded seccomp profile"); | |
1316 | ||
1317 | ret = lxc_seccomp_send_notifier_fd(&conf->seccomp, ap->ipc_socket); | |
1318 | if (ret < 0) | |
1319 | goto on_error; | |
c5bac506 | 1320 | lxc_seccomp_close_notifier_fd(&conf->seccomp); |
e18aba7d CB |
1321 | } |
1322 | ||
3ac4480a | 1323 | if (!lxc_switch_uid_gid(ctx->target_ns_uid, ctx->target_ns_gid)) |
936efc72 CB |
1324 | goto on_error; |
1325 | ||
cd5f35ec CB |
1326 | put_attach_payload(ap); |
1327 | ||
a998454a | 1328 | /* We're done, so we can now do whatever the user intended us to do. */ |
afc691a0 | 1329 | _exit(attach_function(attach_function_args)); |
b75c344c CB |
1330 | |
1331 | on_error: | |
dab02267 | 1332 | ERROR("Failed to attach to container"); |
cd5f35ec | 1333 | put_attach_payload(ap); |
c7ac2e1c | 1334 | _exit(EXIT_FAILURE); |
a998454a CB |
1335 | } |
1336 | ||
f797f05e | 1337 | static int lxc_attach_terminal(const char *name, const char *lxcpath, struct lxc_conf *conf, |
9e84479f | 1338 | struct lxc_terminal *terminal) |
ba2be1a8 CB |
1339 | { |
1340 | int ret; | |
1341 | ||
9e84479f | 1342 | lxc_terminal_init(terminal); |
ba2be1a8 | 1343 | |
8ea93a0f | 1344 | ret = lxc_terminal_create(name, lxcpath, conf, terminal); |
c2af3a15 CB |
1345 | if (ret < 0) |
1346 | return log_error(-1, "Failed to create terminal"); | |
ba2be1a8 | 1347 | |
ba2be1a8 | 1348 | return 0; |
ba2be1a8 CB |
1349 | } |
1350 | ||
9e84479f CB |
1351 | static int lxc_attach_terminal_mainloop_init(struct lxc_terminal *terminal, |
1352 | struct lxc_epoll_descr *descr) | |
ba2be1a8 CB |
1353 | { |
1354 | int ret; | |
1355 | ||
1356 | ret = lxc_mainloop_open(descr); | |
c2af3a15 CB |
1357 | if (ret < 0) |
1358 | return log_error(-1, "Failed to create mainloop"); | |
ba2be1a8 | 1359 | |
9e84479f | 1360 | ret = lxc_terminal_mainloop_add(descr, terminal); |
ba2be1a8 | 1361 | if (ret < 0) { |
ba2be1a8 | 1362 | lxc_mainloop_close(descr); |
c2af3a15 | 1363 | return log_error(-1, "Failed to add handlers to mainloop"); |
ba2be1a8 CB |
1364 | } |
1365 | ||
1366 | return 0; | |
1367 | } | |
1368 | ||
36a94ce8 | 1369 | static inline void lxc_attach_terminal_close_ptx(struct lxc_terminal *terminal) |
ba2be1a8 | 1370 | { |
36a94ce8 | 1371 | close_prot_errno_disarm(terminal->ptx); |
ba2be1a8 CB |
1372 | } |
1373 | ||
cecf3e83 | 1374 | static inline void lxc_attach_terminal_close_pts(struct lxc_terminal *terminal) |
ba2be1a8 | 1375 | { |
41808e20 | 1376 | close_prot_errno_disarm(terminal->pty); |
ba2be1a8 CB |
1377 | } |
1378 | ||
9e84479f | 1379 | static inline void lxc_attach_terminal_close_peer(struct lxc_terminal *terminal) |
ba2be1a8 | 1380 | { |
19a3e906 | 1381 | close_prot_errno_disarm(terminal->peer); |
ba2be1a8 CB |
1382 | } |
1383 | ||
9e84479f | 1384 | static inline void lxc_attach_terminal_close_log(struct lxc_terminal *terminal) |
ba2be1a8 | 1385 | { |
19a3e906 | 1386 | close_prot_errno_disarm(terminal->log_fd); |
ba2be1a8 CB |
1387 | } |
1388 | ||
908fbc1a CB |
1389 | int lxc_attach(struct lxc_container *container, lxc_attach_exec_t exec_function, |
1390 | void *exec_payload, lxc_attach_options_t *options, | |
1391 | pid_t *attached_process) | |
9c4693b8 | 1392 | { |
6f9fe5d0 | 1393 | int ret_parent = -1; |
6f9fe5d0 | 1394 | struct lxc_epoll_descr descr = {}; |
a9f0cecf | 1395 | int ret; |
26abd7ea | 1396 | char *name, *lxcpath; |
9c4693b8 | 1397 | int ipc_sockets[2]; |
500ed813 | 1398 | pid_t attached_pid, pid, to_cleanup_pid; |
ab919e5f | 1399 | struct attach_context *ctx; |
9e84479f | 1400 | struct lxc_terminal terminal; |
1cce35e6 | 1401 | struct lxc_conf *conf; |
9c4693b8 | 1402 | |
908fbc1a | 1403 | if (!container) |
540a2f70 | 1404 | return ret_set_errno(-1, EINVAL); |
908fbc1a CB |
1405 | |
1406 | if (!lxc_container_get(container)) | |
540a2f70 | 1407 | return ret_set_errno(-1, EINVAL); |
908fbc1a CB |
1408 | |
1409 | name = container->name; | |
1410 | lxcpath = container->config_path; | |
1411 | ||
afc691a0 | 1412 | if (!options) { |
9c4693b8 | 1413 | options = &attach_static_default_options; |
afc691a0 CB |
1414 | options->lsm_label = NULL; |
1415 | } | |
9c4693b8 | 1416 | |
9745eb8a | 1417 | ctx = alloc_attach_context(); |
ab919e5f | 1418 | if (!ctx) { |
9745eb8a CB |
1419 | lxc_container_put(container); |
1420 | return log_error_errno(-ENOMEM, ENOMEM, "Failed to allocate attach context"); | |
1421 | } | |
1422 | ||
afc691a0 | 1423 | ret = get_attach_context(ctx, container, options); |
9745eb8a | 1424 | if (ret) { |
7e995801 | 1425 | put_attach_context(ctx); |
74ce42b5 | 1426 | return log_error(-1, "Failed to get attach context"); |
9c4693b8 CS |
1427 | } |
1428 | ||
ab919e5f | 1429 | conf = ctx->container->lxc_conf; |
ba773996 | 1430 | |
ab919e5f | 1431 | if (!fetch_seccomp(ctx->container, options)) |
ae026f55 | 1432 | WARN("Failed to get seccomp policy"); |
2c4ea790 | 1433 | |
ab919e5f | 1434 | if (!no_new_privs(ctx->container, options)) |
ae026f55 | 1435 | WARN("Could not determine whether PR_SET_NO_NEW_PRIVS is set"); |
2e812c16 | 1436 | |
9b31ab58 | 1437 | ret = prepare_namespaces(ctx, options); |
b7873c95 | 1438 | if (ret) { |
52ed870e | 1439 | put_attach_context(ctx); |
74ce42b5 | 1440 | return log_error(-1, "Failed to get namespace file descriptors"); |
9c4693b8 CS |
1441 | } |
1442 | ||
9e84479f | 1443 | if (options->attach_flags & LXC_ATTACH_TERMINAL) { |
f797f05e | 1444 | ret = lxc_attach_terminal(name, lxcpath, conf, &terminal); |
ba2be1a8 | 1445 | if (ret < 0) { |
dd53c8af | 1446 | put_attach_context(ctx); |
74ce42b5 | 1447 | return log_error(-1, "Failed to setup new terminal"); |
ba2be1a8 CB |
1448 | } |
1449 | ||
9e84479f | 1450 | terminal.log_fd = options->log_fd; |
c948657b | 1451 | } else { |
9e84479f | 1452 | lxc_terminal_init(&terminal); |
ba2be1a8 CB |
1453 | } |
1454 | ||
8ce83369 CB |
1455 | /* Create a socket pair for IPC communication; set SOCK_CLOEXEC in order |
1456 | * to make sure we don't irritate other threads that want to fork+exec | |
1457 | * away | |
9c4693b8 CS |
1458 | * |
1459 | * IMPORTANT: if the initial process is multithreaded and another call | |
1460 | * just fork()s away without exec'ing directly after, the socket fd will | |
1461 | * exist in the forked process from the other thread and any close() in | |
8ce83369 | 1462 | * our own child process will not really cause the socket to close |
1d801260 | 1463 | * properly, potentially causing the parent to hang. |
9c4693b8 CS |
1464 | * |
1465 | * For this reason, while IPC is still active, we have to use shutdown() | |
8ce83369 CB |
1466 | * if the child exits prematurely in order to signal that the socket is |
1467 | * closed and cannot assume that the child exiting will automatically do | |
1468 | * that. | |
9c4693b8 CS |
1469 | * |
1470 | * IPC mechanism: (X is receiver) | |
bd6a2355 | 1471 | * initial process transient process attached process |
9c4693b8 CS |
1472 | * X <--- send pid of |
1473 | * attached proc, | |
1474 | * then exit | |
1475 | * send 0 ------------------------------------> X | |
1476 | * [do initialization] | |
1477 | * X <------------------------------------ send 1 | |
1478 | * [add to cgroup, ...] | |
1479 | * send 2 ------------------------------------> X | |
81f466d0 CB |
1480 | * [set LXC_ATTACH_NO_NEW_PRIVS] |
1481 | * X <------------------------------------ send 3 | |
1482 | * [open LSM label fd] | |
1483 | * send 4 ------------------------------------> X | |
1484 | * [set LSM label] | |
9c4693b8 CS |
1485 | * close socket close socket |
1486 | * run program | |
1487 | */ | |
1488 | ret = socketpair(PF_LOCAL, SOCK_STREAM | SOCK_CLOEXEC, 0, ipc_sockets); | |
1489 | if (ret < 0) { | |
dd53c8af | 1490 | put_attach_context(ctx); |
74ce42b5 | 1491 | return log_error_errno(-1, errno, "Could not set up required IPC mechanism for attaching"); |
9c4693b8 CS |
1492 | } |
1493 | ||
bd6a2355 | 1494 | /* Create transient process, two reasons: |
e3f0e436 | 1495 | * 1. We can't setns() in the child itself, since we want to make |
8ce83369 | 1496 | * sure we are properly attached to the pidns. |
e3f0e436 | 1497 | * 2. Also, the initial thread has to put the attached process |
8ce83369 CB |
1498 | * into the cgroup, which we can only do if we didn't already |
1499 | * setns() (otherwise, user namespaces will hate us). | |
9c4693b8 CS |
1500 | */ |
1501 | pid = fork(); | |
9c4693b8 | 1502 | if (pid < 0) { |
dd53c8af | 1503 | put_attach_context(ctx); |
74ce42b5 | 1504 | return log_error_errno(-1, errno, "Failed to create first subprocess"); |
9c4693b8 CS |
1505 | } |
1506 | ||
4f25e72f | 1507 | if (pid == 0) { |
26abd7ea | 1508 | char *cwd, *new_cwd; |
a588a482 | 1509 | |
ba2be1a8 | 1510 | /* close unneeded file descriptors */ |
4f25e72f | 1511 | close_prot_errno_disarm(ipc_sockets[0]); |
2202afc9 | 1512 | |
4f25e72f CB |
1513 | if (options->attach_flags & LXC_ATTACH_TERMINAL) { |
1514 | lxc_attach_terminal_close_ptx(&terminal); | |
1515 | lxc_attach_terminal_close_peer(&terminal); | |
1516 | lxc_attach_terminal_close_log(&terminal); | |
f4364484 SG |
1517 | } |
1518 | ||
4f25e72f | 1519 | /* Wait for the parent to have setup cgroups. */ |
6e48e7c5 | 1520 | if (!sync_wait(ipc_sockets[1], ATTACH_SYNC_CGROUP)) { |
4f25e72f | 1521 | shutdown(ipc_sockets[1], SHUT_RDWR); |
dd53c8af | 1522 | put_attach_context(ctx); |
4f25e72f | 1523 | _exit(EXIT_FAILURE); |
bb2ada6f CB |
1524 | } |
1525 | ||
c538837d CB |
1526 | if (!attach_context_security_barrier(ctx)) { |
1527 | shutdown(ipc_sockets[1], SHUT_RDWR); | |
1528 | put_attach_context(ctx); | |
1529 | _exit(EXIT_FAILURE); | |
1530 | } | |
1531 | ||
a588a482 CB |
1532 | cwd = getcwd(NULL, 0); |
1533 | ||
c538837d CB |
1534 | /* |
1535 | * Attach now, create another subprocess later, since pid | |
1536 | * namespaces only really affect the children of the current | |
1537 | * process. | |
1538 | * | |
1539 | * Note that this is a crucial barrier. We're no moving into | |
1540 | * the container's context so we need to make sure to not leak | |
1541 | * anything sensitive. That especially means things such as | |
1542 | * open file descriptors! | |
4f25e72f | 1543 | */ |
9b31ab58 | 1544 | ret = attach_namespaces(ctx, options); |
4f25e72f CB |
1545 | if (ret < 0) { |
1546 | ERROR("Failed to enter namespaces"); | |
1547 | shutdown(ipc_sockets[1], SHUT_RDWR); | |
dd53c8af | 1548 | put_attach_context(ctx); |
4f25e72f | 1549 | _exit(EXIT_FAILURE); |
ba2be1a8 CB |
1550 | } |
1551 | ||
4f25e72f CB |
1552 | /* Attach succeeded, try to cwd. */ |
1553 | if (options->initial_cwd) | |
1554 | new_cwd = options->initial_cwd; | |
1555 | else | |
1556 | new_cwd = cwd; | |
1557 | if (new_cwd) { | |
1558 | ret = chdir(new_cwd); | |
1559 | if (ret < 0) | |
1560 | WARN("Could not change directory to \"%s\"", new_cwd); | |
ba2be1a8 | 1561 | } |
a588a482 | 1562 | free_disarm(cwd); |
c6d09e15 | 1563 | |
4f25e72f | 1564 | /* Create attached process. */ |
4f25e72f CB |
1565 | pid = lxc_raw_clone(CLONE_PARENT, NULL); |
1566 | if (pid < 0) { | |
1567 | SYSERROR("Failed to clone attached process"); | |
1568 | shutdown(ipc_sockets[1], SHUT_RDWR); | |
dd53c8af | 1569 | put_attach_context(ctx); |
4f25e72f CB |
1570 | _exit(EXIT_FAILURE); |
1571 | } | |
f4364484 | 1572 | |
4f25e72f | 1573 | if (pid == 0) { |
338b230f | 1574 | struct attach_payload ap = { |
a64902ab CB |
1575 | .ipc_socket = ipc_sockets[1], |
1576 | .options = options, | |
1577 | .ctx = ctx, | |
1578 | .terminal_pts_fd = terminal.pty, | |
1579 | .exec_function = exec_function, | |
1580 | .exec_payload = exec_payload, | |
1581 | }; | |
1582 | ||
4f25e72f CB |
1583 | if (options->attach_flags & LXC_ATTACH_TERMINAL) { |
1584 | ret = lxc_terminal_signal_sigmask_safe_blocked(&terminal); | |
1585 | if (ret < 0) { | |
1586 | SYSERROR("Failed to reset signal mask"); | |
1587 | _exit(EXIT_FAILURE); | |
1588 | } | |
1589 | } | |
ea918412 | 1590 | |
a64902ab | 1591 | /* Does not return. */ |
338b230f | 1592 | do_attach(&ap); |
62183f1a | 1593 | } |
bd6a2355 | 1594 | TRACE("Attached process %d started initializing", pid); |
2eef2bda | 1595 | |
4f25e72f CB |
1596 | if (options->attach_flags & LXC_ATTACH_TERMINAL) |
1597 | lxc_attach_terminal_close_pts(&terminal); | |
ea918412 | 1598 | |
4f25e72f | 1599 | /* Tell grandparent the pid of the pid of the newly created child. */ |
f8e88e94 | 1600 | if (!sync_wake_pid(ipc_sockets[1], ATTACH_SYNC_PID(pid))) { |
4f25e72f CB |
1601 | /* If this really happens here, this is very unfortunate, since |
1602 | * the parent will not know the pid of the attached process and | |
1603 | * will not be able to wait for it (and we won't either due to | |
1604 | * CLONE_PARENT) so the parent won't be able to reap it and the | |
1605 | * attached process will remain a zombie. | |
1606 | */ | |
1607 | shutdown(ipc_sockets[1], SHUT_RDWR); | |
dd53c8af | 1608 | put_attach_context(ctx); |
4f25e72f CB |
1609 | _exit(EXIT_FAILURE); |
1610 | } | |
9c4693b8 | 1611 | |
4f25e72f | 1612 | /* The rest is in the hands of the initial and the attached process. */ |
dd53c8af | 1613 | put_attach_context(ctx); |
4f25e72f CB |
1614 | _exit(EXIT_SUCCESS); |
1615 | } | |
bd6a2355 | 1616 | TRACE("Transient process %d started initializing", pid); |
6f4f1937 | 1617 | |
4f25e72f | 1618 | to_cleanup_pid = pid; |
ea918412 | 1619 | |
4f25e72f | 1620 | /* close unneeded file descriptors */ |
cb2420df | 1621 | close_prot_errno_disarm(ipc_sockets[1]); |
9b31ab58 | 1622 | put_namespaces(ctx); |
4f25e72f CB |
1623 | if (options->attach_flags & LXC_ATTACH_TERMINAL) |
1624 | lxc_attach_terminal_close_pts(&terminal); | |
81f466d0 | 1625 | |
4f25e72f CB |
1626 | /* Attach to cgroup, if requested. */ |
1627 | if (options->attach_flags & LXC_ATTACH_MOVE_TO_CGROUP) { | |
1628 | /* | |
1629 | * If this is the unified hierarchy cgroup_attach() is | |
1630 | * enough. | |
1631 | */ | |
1632 | ret = cgroup_attach(conf, name, lxcpath, pid); | |
9a57778b | 1633 | if (ret) { |
4f25e72f | 1634 | call_cleaner(cgroup_exit) struct cgroup_ops *cgroup_ops = NULL; |
9044b79e | 1635 | |
9a57778b CB |
1636 | if (ret != -ENOCGROUP2) { |
1637 | SYSERROR("Failed to attach cgroup"); | |
1638 | goto on_error; | |
1639 | } | |
1640 | ||
4f25e72f CB |
1641 | cgroup_ops = cgroup_init(conf); |
1642 | if (!cgroup_ops) | |
1643 | goto on_error; | |
9044b79e | 1644 | |
4f25e72f CB |
1645 | if (!cgroup_ops->attach(cgroup_ops, conf, name, lxcpath, pid)) |
1646 | goto on_error; | |
81f466d0 | 1647 | } |
9a57778b | 1648 | |
bd6a2355 | 1649 | TRACE("Moved transient process %d into container cgroup", pid); |
4f25e72f | 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 | |
bd6a2355 | 1682 | TRACE("Told transient process to start initializing"); |
ea918412 | 1683 | |
bd6a2355 | 1684 | /* Get pid of attached process from transient 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 | |
bd6a2355 | 1696 | /* Reap transient process. */ |
4f25e72f CB |
1697 | ret = wait_for_pid(pid); |
1698 | if (ret < 0) | |
1699 | goto close_mainloop; | |
ba2be1a8 | 1700 | |
bd6a2355 | 1701 | TRACE("Transient 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 | } |