]>
Commit | Line | Data |
---|---|---|
1 | /* SPDX-License-Identifier: LGPL-2.1+ */ | |
2 | ||
3 | #ifndef _GNU_SOURCE | |
4 | #define _GNU_SOURCE 1 | |
5 | #endif | |
6 | #include <errno.h> | |
7 | #include <fcntl.h> | |
8 | #include <grp.h> | |
9 | #include <linux/unistd.h> | |
10 | #include <pwd.h> | |
11 | #include <pthread.h> | |
12 | #include <signal.h> | |
13 | #include <stdio.h> | |
14 | #include <stdlib.h> | |
15 | #include <string.h> | |
16 | #include <sys/mount.h> | |
17 | #include <sys/param.h> | |
18 | #include <sys/prctl.h> | |
19 | #include <sys/socket.h> | |
20 | #include <sys/syscall.h> | |
21 | #include <sys/wait.h> | |
22 | #include <termios.h> | |
23 | #include <unistd.h> | |
24 | ||
25 | #include <lxc/lxccontainer.h> | |
26 | ||
27 | #include "af_unix.h" | |
28 | #include "attach.h" | |
29 | #include "caps.h" | |
30 | #include "cgroup.h" | |
31 | #include "commands.h" | |
32 | #include "conf.h" | |
33 | #include "config.h" | |
34 | #include "confile.h" | |
35 | #include "log.h" | |
36 | #include "lsm/lsm.h" | |
37 | #include "lxclock.h" | |
38 | #include "lxcseccomp.h" | |
39 | #include "macro.h" | |
40 | #include "mainloop.h" | |
41 | #include "memory_utils.h" | |
42 | #include "mount_utils.h" | |
43 | #include "namespace.h" | |
44 | #include "process_utils.h" | |
45 | #include "sync.h" | |
46 | #include "syscall_wrappers.h" | |
47 | #include "terminal.h" | |
48 | #include "utils.h" | |
49 | ||
50 | #if HAVE_SYS_PERSONALITY_H | |
51 | #include <sys/personality.h> | |
52 | #endif | |
53 | ||
54 | lxc_log_define(attach, lxc); | |
55 | ||
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 | ||
59 | /* | |
60 | * The context used to attach to the container. | |
61 | * @attach_flags : the attach flags specified in lxc_attach_options_t | |
62 | * @init_pid : the PID of the container's init process | |
63 | * @dfd_init_pid : file descriptor to /proc/@init_pid | |
64 | * __Must be closed in attach_context_security_barrier()__! | |
65 | * @dfd_self_pid : file descriptor to /proc/self | |
66 | * __Must be closed in attach_context_security_barrier()__! | |
67 | * @setup_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 | |
86 | */ | |
87 | struct attach_context { | |
88 | unsigned int attach_flags; | |
89 | int init_pid; | |
90 | int init_pidfd; | |
91 | int dfd_init_pid; | |
92 | int dfd_self_pid; | |
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; | |
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 | ||
108 | static pid_t pidfd_get_pid(int dfd_init_pid, int pidfd) | |
109 | { | |
110 | __do_free char *line = NULL; | |
111 | __do_fclose FILE *f = NULL; | |
112 | size_t len = 0; | |
113 | char path[STRLITERALLEN("fdinfo/") + INTTYPE_TO_STRLEN(int) + 1 ] = "fdinfo/"; | |
114 | int ret; | |
115 | ||
116 | if (dfd_init_pid < 0 || pidfd < 0) | |
117 | return ret_errno(EBADF); | |
118 | ||
119 | ret = strnprintf(path + STRLITERALLEN("fdinfo/"), INTTYPE_TO_STRLEN(int), "%d", pidfd); | |
120 | if (ret < 0) | |
121 | return ret_errno(EIO); | |
122 | ||
123 | f = fdopen_at(dfd_init_pid, path, "re", PROTECT_OPEN, PROTECT_LOOKUP_BENEATH); | |
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 | ||
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 | ||
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 | ||
175 | static struct attach_context *alloc_attach_context(void) | |
176 | { | |
177 | struct attach_context *ctx; | |
178 | ||
179 | ctx = zalloc(sizeof(struct attach_context)); | |
180 | if (!ctx) | |
181 | return ret_set_errno(NULL, ENOMEM); | |
182 | ||
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; | |
193 | ||
194 | for (int i = 0; i < LXC_NS_MAX; i++) | |
195 | ctx->ns_fd[i] = -EBADF; | |
196 | ||
197 | return ctx; | |
198 | } | |
199 | ||
200 | static int get_personality(const char *name, const char *lxcpath, | |
201 | signed long *personality) | |
202 | { | |
203 | __do_free char *p = NULL; | |
204 | signed long per; | |
205 | ||
206 | p = lxc_cmd_get_config_item(name, "lxc.arch", lxcpath); | |
207 | if (!p) { | |
208 | *personality = LXC_ARCH_UNCHANGED; | |
209 | return 0; | |
210 | } | |
211 | ||
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; | |
218 | } | |
219 | ||
220 | static int userns_setup_ids(struct attach_context *ctx, | |
221 | lxc_attach_options_t *options) | |
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 | ||
234 | f_uidmap = fdopen_at(ctx->dfd_init_pid, "uid_map", "re", PROTECT_OPEN, PROTECT_LOOKUP_BENEATH); | |
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) { | |
243 | ctx->setup_ns_uid = 0; | |
244 | TRACE("Container has mapping for uid 0"); | |
245 | break; | |
246 | } | |
247 | ||
248 | if (ctx->target_host_uid >= hostuid && ctx->target_host_uid < hostuid + range_uid) { | |
249 | init_ns_uid = (ctx->target_host_uid - hostuid) + nsuid; | |
250 | TRACE("Container runs with uid %d", init_ns_uid); | |
251 | } | |
252 | } | |
253 | ||
254 | f_gidmap = fdopen_at(ctx->dfd_init_pid, "gid_map", "re", PROTECT_OPEN, PROTECT_LOOKUP_BENEATH); | |
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) { | |
263 | ctx->setup_ns_gid = 0; | |
264 | TRACE("Container has mapping for gid 0"); | |
265 | break; | |
266 | } | |
267 | ||
268 | if (ctx->target_host_gid >= hostgid && ctx->target_host_gid < hostgid + range_gid) { | |
269 | init_ns_gid = (ctx->target_host_gid - hostgid) + nsgid; | |
270 | TRACE("Container runs with gid %d", init_ns_gid); | |
271 | } | |
272 | } | |
273 | ||
274 | if (ctx->setup_ns_uid == LXC_INVALID_UID) | |
275 | ctx->setup_ns_uid = init_ns_uid; | |
276 | ||
277 | if (ctx->setup_ns_gid == LXC_INVALID_UID) | |
278 | ctx->setup_ns_gid = init_ns_gid; | |
279 | ||
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) | |
286 | ctx->target_ns_uid = options->uid; | |
287 | else if (options->namespaces & CLONE_NEWUSER) | |
288 | ctx->target_ns_uid = ctx->setup_ns_uid; | |
289 | else | |
290 | ctx->target_ns_uid = 0; | |
291 | ||
292 | if (ctx->target_ns_uid == LXC_INVALID_UID) | |
293 | WARN("Invalid uid specified"); | |
294 | ||
295 | if (options->gid != LXC_INVALID_GID) | |
296 | ctx->target_ns_gid = options->gid; | |
297 | else if (options->namespaces & CLONE_NEWUSER) | |
298 | ctx->target_ns_gid = ctx->setup_ns_gid; | |
299 | else | |
300 | ctx->target_ns_gid = 0; | |
301 | ||
302 | if (ctx->target_ns_gid == LXC_INVALID_GID) | |
303 | WARN("Invalid gid specified"); | |
304 | } | |
305 | ||
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; | |
312 | int ret; | |
313 | ||
314 | f = fdopen_at(ctx->dfd_init_pid, "status", "re", PROTECT_OPEN, PROTECT_LOOKUP_BENEATH); | |
315 | if (!f) | |
316 | return log_error_errno(-errno, errno, "Failed to open status file"); | |
317 | ||
318 | while (getline(&line, &len, f) != -1) { | |
319 | signed long value = -1; | |
320 | ||
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) { | |
327 | ctx->target_host_uid = (uid_t)value; | |
328 | TRACE("Container's init process runs with hostuid %d", ctx->target_host_uid); | |
329 | goto next; | |
330 | } | |
331 | ||
332 | ret = sscanf(line, "Gid: %ld", &value); | |
333 | if (ret != EOF && ret == 1) { | |
334 | ctx->target_host_gid = (gid_t)value; | |
335 | TRACE("Container's init process runs with hostgid %d", ctx->target_host_gid); | |
336 | goto next; | |
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: | |
346 | if (ctx->target_host_uid != LXC_INVALID_UID && | |
347 | ctx->target_host_gid != LXC_INVALID_GID && | |
348 | caps_found) | |
349 | break; | |
350 | ||
351 | } | |
352 | ||
353 | ret = userns_setup_ids(ctx, options); | |
354 | if (ret) | |
355 | return log_error_errno(ret, errno, "Failed to get setup ids"); | |
356 | userns_target_ids(ctx, options); | |
357 | ||
358 | return 0; | |
359 | } | |
360 | ||
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 | ||
380 | static int get_attach_context(struct attach_context *ctx, | |
381 | struct lxc_container *container, | |
382 | lxc_attach_options_t *options) | |
383 | { | |
384 | __do_free char *lsm_label = NULL; | |
385 | int ret; | |
386 | char path[LXC_PROC_PID_LEN]; | |
387 | ||
388 | ctx->container = container; | |
389 | ctx->attach_flags = options->attach_flags; | |
390 | ||
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 | ||
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); | |
400 | else | |
401 | ctx->init_pid = lxc_cmd_get_init_pid(container->name, container->config_path); | |
402 | ||
403 | if (ctx->init_pid < 0) | |
404 | return log_error(-1, "Failed to get init pid"); | |
405 | ||
406 | ret = strnprintf(path, sizeof(path), "/proc/%d", ctx->init_pid); | |
407 | if (ret < 0) | |
408 | return ret_errno(EIO); | |
409 | ||
410 | ctx->dfd_init_pid = open_at(-EBADF, path, | |
411 | PROTECT_OPATH_DIRECTORY, | |
412 | (PROTECT_LOOKUP_ABSOLUTE & ~RESOLVE_NO_XDEV), 0); | |
413 | if (ctx->dfd_init_pid < 0) | |
414 | return log_error_errno(-errno, errno, "Failed to open /proc/%d", ctx->init_pid); | |
415 | ||
416 | if (ctx->init_pidfd >= 0) { | |
417 | ret = lxc_raw_pidfd_send_signal(ctx->init_pidfd, 0, NULL, 0); | |
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"); | |
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 | } | |
431 | } | |
432 | ||
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 | ||
452 | ret = parse_init_status(ctx, options); | |
453 | if (ret) | |
454 | return log_error_errno(-errno, errno, "Failed to open parse file"); | |
455 | ||
456 | ctx->lsm_ops = lsm_init_static(); | |
457 | ||
458 | if (attach_lsm(options)) { | |
459 | if (ctx->attach_flags & LXC_ATTACH_LSM_LABEL) | |
460 | lsm_label = options->lsm_label; | |
461 | else | |
462 | lsm_label = ctx->lsm_ops->process_label_get_at(ctx->lsm_ops, ctx->dfd_init_pid); | |
463 | if (!lsm_label) | |
464 | WARN("No security context received"); | |
465 | else | |
466 | INFO("Retrieved security context %s", lsm_label); | |
467 | } | |
468 | ||
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"); | |
472 | ||
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 | ||
479 | ctx->lsm_label = move_ptr(lsm_label); | |
480 | return 0; | |
481 | } | |
482 | ||
483 | static int same_nsfd(int dfd_pid1, int dfd_pid2, const char *ns_path) | |
484 | { | |
485 | int ret; | |
486 | struct stat ns_st1, ns_st2; | |
487 | ||
488 | ret = fstatat(dfd_pid1, ns_path, &ns_st1, 0); | |
489 | if (ret) | |
490 | return -1; | |
491 | ||
492 | ret = fstatat(dfd_pid2, ns_path, &ns_st2, 0); | |
493 | if (ret) | |
494 | return -1; | |
495 | ||
496 | /* processes are in the same namespace */ | |
497 | if ((ns_st1.st_dev == ns_st2.st_dev) && | |
498 | (ns_st1.st_ino == ns_st2.st_ino)) | |
499 | return -EINVAL; | |
500 | ||
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 | ||
524 | /* processes are in different namespaces */ | |
525 | return move_fd(ns_fd2); | |
526 | } | |
527 | ||
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) | |
554 | { | |
555 | for (int i = 0; i < LXC_NS_MAX; i++) { | |
556 | int j; | |
557 | ||
558 | if (options->namespaces & ns_info[i].clone_flag) | |
559 | ctx->ns_fd[i] = open_at(ctx->dfd_init_pid, | |
560 | ns_info[i].proc_path, | |
561 | PROTECT_OPEN_WITH_TRAILING_SYMLINKS, | |
562 | (PROTECT_LOOKUP_BENEATH_WITH_MAGICLINKS & | |
563 | ~(RESOLVE_NO_XDEV | RESOLVE_BENEATH)), | |
564 | 0); | |
565 | else if (ctx->ns_inherited & ns_info[i].clone_flag) | |
566 | ctx->ns_fd[i] = same_ns(ctx->dfd_self_pid, | |
567 | ctx->dfd_init_pid, | |
568 | ns_info[i].proc_path); | |
569 | else | |
570 | continue; | |
571 | ||
572 | if (ctx->ns_fd[i] >= 0) | |
573 | continue; | |
574 | ||
575 | if (ctx->ns_fd[i] == -EINVAL) { | |
576 | ctx->ns_inherited &= ~ns_info[i].clone_flag; | |
577 | continue; | |
578 | } | |
579 | ||
580 | /* We failed to preserve the namespace. */ | |
581 | SYSERROR("Failed to preserve %s namespace of %d", | |
582 | ns_info[i].proc_name, ctx->init_pid); | |
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 | ||
596 | static int prepare_namespaces(struct attach_context *ctx, | |
597 | lxc_attach_options_t *options) | |
598 | { | |
599 | if (ctx->init_pidfd < 0) | |
600 | return __prepare_namespaces_nsfd(ctx, options); | |
601 | ||
602 | return __prepare_namespaces_pidfd(ctx); | |
603 | } | |
604 | ||
605 | static inline void put_namespaces(struct attach_context *ctx) | |
606 | { | |
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 | } | |
612 | ||
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; | |
618 | ||
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"); | |
631 | } | |
632 | ||
633 | static int __attach_namespaces_nsfd(struct attach_context *ctx, | |
634 | lxc_attach_options_t *options) | |
635 | { | |
636 | int fret = 0; | |
637 | ||
638 | for (int i = 0; i < LXC_NS_MAX; i++) { | |
639 | int ret; | |
640 | ||
641 | if (ctx->ns_fd[i] < 0) | |
642 | continue; | |
643 | ||
644 | ret = setns(ctx->ns_fd[i], ns_info[i].clone_flag); | |
645 | if (ret) | |
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); | |
650 | ||
651 | if (close(ctx->ns_fd[i])) { | |
652 | fret = -errno; | |
653 | SYSERROR("Failed to close file descriptor for %s namespace", | |
654 | ns_info[i].proc_name); | |
655 | } | |
656 | ctx->ns_fd[i] = -EBADF; | |
657 | } | |
658 | ||
659 | return fret; | |
660 | } | |
661 | ||
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 | ||
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) { | |
709 | if (close(ctx->dfd_self_pid)) | |
710 | return false; | |
711 | ctx->dfd_self_pid = -EBADF; | |
712 | ||
713 | if (close(ctx->dfd_init_pid)) | |
714 | return false; | |
715 | ctx->dfd_init_pid = -EBADF; | |
716 | } | |
717 | ||
718 | return true; | |
719 | } | |
720 | ||
721 | int lxc_attach_remount_sys_proc(void) | |
722 | { | |
723 | int ret; | |
724 | ||
725 | ret = unshare(CLONE_NEWNS); | |
726 | if (ret < 0) | |
727 | return log_error_errno(-1, errno, "Failed to unshare mount namespace"); | |
728 | ||
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..."); | |
731 | ||
732 | /* Assume /proc is always mounted, so remount it. */ | |
733 | ret = umount2("/proc", MNT_DETACH); | |
734 | if (ret < 0) | |
735 | return log_error_errno(-1, errno, "Failed to unmount /proc"); | |
736 | ||
737 | ret = mount("none", "/proc", "proc", 0, NULL); | |
738 | if (ret < 0) | |
739 | return log_error_errno(-1, errno, "Failed to remount /proc"); | |
740 | ||
741 | /* | |
742 | * Try to umount /sys. If it's not a mount point, we'll get EINVAL, then | |
743 | * we ignore it because it may not have been mounted in the first place. | |
744 | */ | |
745 | ret = umount2("/sys", MNT_DETACH); | |
746 | if (ret < 0 && errno != EINVAL) | |
747 | return log_error_errno(-1, errno, "Failed to unmount /sys"); | |
748 | ||
749 | /* Remount it. */ | |
750 | if (ret == 0 && mount("none", "/sys", "sysfs", 0, NULL)) | |
751 | return log_error_errno(-1, errno, "Failed to remount /sys"); | |
752 | ||
753 | return 0; | |
754 | } | |
755 | ||
756 | static int drop_capabilities(struct attach_context *ctx) | |
757 | { | |
758 | int last_cap; | |
759 | ||
760 | last_cap = lxc_caps_last_cap(); | |
761 | for (int cap = 0; cap <= last_cap; cap++) { | |
762 | if (ctx->capability_mask & (1LL << cap)) | |
763 | continue; | |
764 | ||
765 | if (prctl(PR_CAPBSET_DROP, prctl_arg(cap), prctl_arg(0), | |
766 | prctl_arg(0), prctl_arg(0))) | |
767 | return log_error_errno(-1, errno, "Failed to drop capability %d", cap); | |
768 | ||
769 | TRACE("Dropped capability %d", cap); | |
770 | } | |
771 | ||
772 | return 0; | |
773 | } | |
774 | ||
775 | static int lxc_attach_set_environment(struct attach_context *ctx, | |
776 | enum lxc_attach_env_policy_t policy, | |
777 | char **extra_env, char **extra_keep) | |
778 | { | |
779 | int ret; | |
780 | struct lxc_list *iterator; | |
781 | ||
782 | if (policy == LXC_ATTACH_CLEAR_ENV) { | |
783 | int path_kept = 0; | |
784 | char **extra_keep_store = NULL; | |
785 | ||
786 | if (extra_keep) { | |
787 | size_t count, i; | |
788 | ||
789 | for (count = 0; extra_keep[count]; count++) | |
790 | ; | |
791 | ||
792 | extra_keep_store = zalloc(count * sizeof(char *)); | |
793 | if (!extra_keep_store) | |
794 | return -1; | |
795 | ||
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]) { | |
801 | while (i > 0) | |
802 | free(extra_keep_store[--i]); | |
803 | ||
804 | free(extra_keep_store); | |
805 | return -1; | |
806 | } | |
807 | ||
808 | if (strcmp(extra_keep[i], "PATH") == 0) | |
809 | path_kept = 1; | |
810 | } | |
811 | } | |
812 | } | |
813 | ||
814 | if (clearenv()) { | |
815 | if (extra_keep_store) { | |
816 | char **p; | |
817 | ||
818 | for (p = extra_keep_store; *p; p++) | |
819 | free(*p); | |
820 | ||
821 | free(extra_keep_store); | |
822 | } | |
823 | ||
824 | return log_error(-1, "Failed to clear environment"); | |
825 | } | |
826 | ||
827 | if (extra_keep_store) { | |
828 | size_t i; | |
829 | ||
830 | for (i = 0; extra_keep[i]; i++) { | |
831 | if (extra_keep_store[i]) { | |
832 | ret = setenv(extra_keep[i], extra_keep_store[i], 1); | |
833 | if (ret < 0) | |
834 | SYSWARN("Failed to set environment variable"); | |
835 | } | |
836 | ||
837 | free(extra_keep_store[i]); | |
838 | } | |
839 | ||
840 | free(extra_keep_store); | |
841 | } | |
842 | ||
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 | */ | |
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) | |
851 | SYSWARN("Failed to set environment variable"); | |
852 | } | |
853 | } | |
854 | ||
855 | ret = putenv("container=lxc"); | |
856 | if (ret < 0) | |
857 | return log_warn(-1, "Failed to set environment variable"); | |
858 | ||
859 | /* Set container environment variables.*/ | |
860 | if (ctx->container->lxc_conf) { | |
861 | lxc_list_for_each(iterator, &ctx->container->lxc_conf->environment) { | |
862 | char *env_tmp; | |
863 | ||
864 | env_tmp = strdup((char *)iterator->elem); | |
865 | if (!env_tmp) | |
866 | return -1; | |
867 | ||
868 | ret = putenv(env_tmp); | |
869 | if (ret < 0) | |
870 | return log_error_errno(-1, errno, "Failed to set environment variable: %s", (char *)iterator->elem); | |
871 | } | |
872 | } | |
873 | ||
874 | /* Set extra environment variables. */ | |
875 | if (extra_env) { | |
876 | for (; *extra_env; extra_env++) { | |
877 | char *p; | |
878 | ||
879 | /* We just assume the user knows what they are doing, so | |
880 | * we don't do any checks. | |
881 | */ | |
882 | p = strdup(*extra_env); | |
883 | if (!p) | |
884 | return -1; | |
885 | ||
886 | ret = putenv(p); | |
887 | if (ret < 0) | |
888 | SYSWARN("Failed to set environment variable"); | |
889 | } | |
890 | } | |
891 | ||
892 | return 0; | |
893 | } | |
894 | ||
895 | static char *lxc_attach_getpwshell(uid_t uid) | |
896 | { | |
897 | __do_free char *line = NULL, *result = NULL; | |
898 | __do_fclose FILE *pipe_f = NULL; | |
899 | int fd, ret; | |
900 | pid_t pid; | |
901 | int pipes[2]; | |
902 | bool found = false; | |
903 | size_t line_bufsz = 0; | |
904 | ||
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. | |
907 | */ | |
908 | ret = pipe2(pipes, O_CLOEXEC); | |
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 | ||
919 | if (!pid) { | |
920 | char uid_buf[32]; | |
921 | char *arguments[] = { | |
922 | "getent", | |
923 | "passwd", | |
924 | uid_buf, | |
925 | NULL | |
926 | }; | |
927 | ||
928 | close(pipes[0]); | |
929 | ||
930 | /* We want to capture stdout. */ | |
931 | ret = dup2(pipes[1], STDOUT_FILENO); | |
932 | close(pipes[1]); | |
933 | if (ret < 0) | |
934 | _exit(EXIT_FAILURE); | |
935 | ||
936 | /* Get rid of stdin/stderr, so we try to associate it with | |
937 | * /dev/null. | |
938 | */ | |
939 | fd = open_devnull(); | |
940 | if (fd < 0) { | |
941 | close(STDIN_FILENO); | |
942 | close(STDERR_FILENO); | |
943 | } else { | |
944 | (void)dup3(fd, STDIN_FILENO, O_CLOEXEC); | |
945 | (void)dup3(fd, STDERR_FILENO, O_CLOEXEC); | |
946 | close(fd); | |
947 | } | |
948 | ||
949 | /* Finish argument list. */ | |
950 | ret = strnprintf(uid_buf, sizeof(uid_buf), "%ld", (long)uid); | |
951 | if (ret <= 0) | |
952 | _exit(EXIT_FAILURE); | |
953 | ||
954 | /* Try to run getent program. */ | |
955 | (void)execvp("getent", arguments); | |
956 | _exit(EXIT_FAILURE); | |
957 | } | |
958 | ||
959 | close(pipes[1]); | |
960 | ||
961 | pipe_f = fdopen(pipes[0], "re"); | |
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 | ||
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 | ||
982 | if (!line) | |
983 | continue; | |
984 | ||
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 || | |
1003 | value == LONG_MAX) | |
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 | } | |
1016 | ||
1017 | if (!token) | |
1018 | continue; | |
1019 | ||
1020 | free_disarm(result); | |
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 | } | |
1030 | ||
1031 | reap_child: | |
1032 | ret = wait_for_pid(pid); | |
1033 | if (ret < 0) | |
1034 | return NULL; | |
1035 | ||
1036 | if (!found) | |
1037 | return NULL; | |
1038 | ||
1039 | return move_ptr(result); | |
1040 | } | |
1041 | ||
1042 | static bool fetch_seccomp(struct lxc_container *c, lxc_attach_options_t *options) | |
1043 | { | |
1044 | __do_free char *path = NULL; | |
1045 | int ret; | |
1046 | bool bret; | |
1047 | ||
1048 | if (!attach_lsm(options)) { | |
1049 | free_disarm(c->lxc_conf->seccomp.seccomp); | |
1050 | return true; | |
1051 | } | |
1052 | ||
1053 | /* Remove current setting. */ | |
1054 | if (!c->set_config_item(c, "lxc.seccomp.profile", "") && | |
1055 | !c->set_config_item(c, "lxc.seccomp", "")) | |
1056 | return false; | |
1057 | ||
1058 | /* Fetch the current profile path over the cmd interface. */ | |
1059 | path = c->get_running_config_item(c, "lxc.seccomp.profile"); | |
1060 | if (!path) { | |
1061 | INFO("Failed to retrieve lxc.seccomp.profile"); | |
1062 | ||
1063 | path = c->get_running_config_item(c, "lxc.seccomp"); | |
1064 | if (!path) | |
1065 | return log_info(true, "Failed to retrieve lxc.seccomp"); | |
1066 | } | |
1067 | ||
1068 | /* Copy the value into the new lxc_conf. */ | |
1069 | bret = c->set_config_item(c, "lxc.seccomp.profile", path); | |
1070 | if (!bret) | |
1071 | return false; | |
1072 | ||
1073 | /* Attempt to parse the resulting config. */ | |
1074 | ret = lxc_read_seccomp_config(c->lxc_conf); | |
1075 | if (ret < 0) | |
1076 | return log_error(false, "Failed to retrieve seccomp policy"); | |
1077 | ||
1078 | return log_info(true, "Retrieved seccomp policy"); | |
1079 | } | |
1080 | ||
1081 | static bool no_new_privs(struct lxc_container *c, lxc_attach_options_t *options) | |
1082 | { | |
1083 | __do_free char *val = NULL; | |
1084 | ||
1085 | /* Remove current setting. */ | |
1086 | if (!c->set_config_item(c, "lxc.no_new_privs", "")) | |
1087 | return log_info(false, "Failed to unset lxc.no_new_privs"); | |
1088 | ||
1089 | /* Retrieve currently active setting. */ | |
1090 | val = c->get_running_config_item(c, "lxc.no_new_privs"); | |
1091 | if (!val) | |
1092 | return log_info(false, "Failed to retrieve lxc.no_new_privs"); | |
1093 | ||
1094 | /* Set currently active setting. */ | |
1095 | return c->set_config_item(c, "lxc.no_new_privs", val); | |
1096 | } | |
1097 | ||
1098 | struct attach_payload { | |
1099 | int ipc_socket; | |
1100 | int terminal_pts_fd; | |
1101 | lxc_attach_options_t *options; | |
1102 | struct attach_context *ctx; | |
1103 | lxc_attach_exec_t exec_function; | |
1104 | void *exec_payload; | |
1105 | }; | |
1106 | ||
1107 | static void put_attach_payload(struct attach_payload *p) | |
1108 | { | |
1109 | if (p) { | |
1110 | close_prot_errno_disarm(p->ipc_socket); | |
1111 | close_prot_errno_disarm(p->terminal_pts_fd); | |
1112 | put_attach_context(p->ctx); | |
1113 | p->ctx = NULL; | |
1114 | } | |
1115 | } | |
1116 | ||
1117 | __noreturn static void do_attach(struct attach_payload *ap) | |
1118 | { | |
1119 | lxc_attach_exec_t attach_function = move_ptr(ap->exec_function); | |
1120 | void *attach_function_args = move_ptr(ap->exec_payload); | |
1121 | int lsm_fd, ret; | |
1122 | lxc_attach_options_t* options = ap->options; | |
1123 | struct attach_context *ctx = ap->ctx; | |
1124 | struct lxc_conf *conf = ctx->container->lxc_conf; | |
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(); | |
1134 | if (ret < 0) | |
1135 | goto on_error; | |
1136 | ||
1137 | TRACE("Remounted \"/proc\" and \"/sys\""); | |
1138 | } | |
1139 | ||
1140 | /* Now perform additional attachments. */ | |
1141 | #if HAVE_SYS_PERSONALITY_H | |
1142 | if (options->attach_flags & LXC_ATTACH_SET_PERSONALITY) { | |
1143 | long new_personality; | |
1144 | ||
1145 | if (options->personality < 0) | |
1146 | new_personality = ctx->personality; | |
1147 | else | |
1148 | new_personality = options->personality; | |
1149 | ||
1150 | if (new_personality != LXC_ARCH_UNCHANGED) { | |
1151 | ret = personality(new_personality); | |
1152 | if (ret < 0) | |
1153 | goto on_error; | |
1154 | ||
1155 | TRACE("Set new personality"); | |
1156 | } | |
1157 | } | |
1158 | #endif | |
1159 | ||
1160 | if (options->attach_flags & LXC_ATTACH_DROP_CAPABILITIES) { | |
1161 | ret = drop_capabilities(ctx); | |
1162 | if (ret < 0) | |
1163 | goto on_error; | |
1164 | ||
1165 | TRACE("Dropped capabilities"); | |
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 | */ | |
1171 | ret = lxc_attach_set_environment(ctx, | |
1172 | options->env_policy, | |
1173 | options->extra_env_vars, | |
1174 | options->extra_keep_env); | |
1175 | if (ret < 0) | |
1176 | goto on_error; | |
1177 | ||
1178 | TRACE("Set up environment"); | |
1179 | ||
1180 | /* | |
1181 | * This remark only affects fully unprivileged containers: | |
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 | */ | |
1192 | if (attach_lsm(options) && ctx->lsm_label) { | |
1193 | if (!sync_wait_fd(ap->ipc_socket, ATTACH_SYNC_LSM(&lsm_fd))) { | |
1194 | SYSERROR("Failed to receive lsm label fd"); | |
1195 | goto on_error; | |
1196 | } | |
1197 | ||
1198 | TRACE("Received LSM label file descriptor %d from parent", lsm_fd); | |
1199 | } | |
1200 | ||
1201 | if (options->stdin_fd > 0 && isatty(options->stdin_fd)) { | |
1202 | ret = lxc_make_controlling_terminal(options->stdin_fd); | |
1203 | if (ret < 0) | |
1204 | goto on_error; | |
1205 | } | |
1206 | ||
1207 | if ((options->attach_flags & LXC_ATTACH_SETGROUPS) && | |
1208 | options->groups.size > 0) { | |
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 | } | |
1215 | ||
1216 | if (options->namespaces & CLONE_NEWUSER) | |
1217 | if (!lxc_switch_uid_gid(ctx->setup_ns_uid, ctx->setup_ns_gid)) | |
1218 | goto on_error; | |
1219 | ||
1220 | if (attach_lsm(options) && ctx->lsm_label) { | |
1221 | bool on_exec; | |
1222 | ||
1223 | /* Change into our new LSM profile. */ | |
1224 | on_exec = options->attach_flags & LXC_ATTACH_LSM_EXEC ? true : false; | |
1225 | ret = ctx->lsm_ops->process_label_set_at(ctx->lsm_ops, lsm_fd, ctx->lsm_label, on_exec); | |
1226 | close_prot_errno_disarm(lsm_fd); | |
1227 | if (ret < 0) | |
1228 | goto on_error; | |
1229 | ||
1230 | TRACE("Set %s LSM label to \"%s\"", ctx->lsm_ops->name, ctx->lsm_label); | |
1231 | } | |
1232 | ||
1233 | if (conf->no_new_privs || (options->attach_flags & LXC_ATTACH_NO_NEW_PRIVS)) { | |
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 | ||
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 | */ | |
1251 | if (options->stdin_fd >= 0 && options->stdin_fd != STDIN_FILENO) | |
1252 | if (dup2(options->stdin_fd, STDIN_FILENO) < 0) | |
1253 | SYSDEBUG("Failed to replace stdin with %d", options->stdin_fd); | |
1254 | ||
1255 | if (options->stdout_fd >= 0 && options->stdout_fd != STDOUT_FILENO) | |
1256 | if (dup2(options->stdout_fd, STDOUT_FILENO) < 0) | |
1257 | SYSDEBUG("Failed to replace stdout with %d", options->stdout_fd); | |
1258 | ||
1259 | if (options->stderr_fd >= 0 && options->stderr_fd != STDERR_FILENO) | |
1260 | if (dup2(options->stderr_fd, STDERR_FILENO) < 0) | |
1261 | SYSDEBUG("Failed to replace stderr with %d", options->stderr_fd); | |
1262 | ||
1263 | /* close the old fds */ | |
1264 | if (options->stdin_fd > STDERR_FILENO) | |
1265 | close(options->stdin_fd); | |
1266 | ||
1267 | if (options->stdout_fd > STDERR_FILENO) | |
1268 | close(options->stdout_fd); | |
1269 | ||
1270 | if (options->stderr_fd > STDERR_FILENO) | |
1271 | close(options->stderr_fd); | |
1272 | ||
1273 | /* | |
1274 | * Try to remove FD_CLOEXEC flag from stdin/stdout/stderr, but also | |
1275 | * here, ignore errors. | |
1276 | */ | |
1277 | for (int fd = STDIN_FILENO; fd <= STDERR_FILENO; fd++) { | |
1278 | ret = fd_cloexec(fd, false); | |
1279 | if (ret < 0) { | |
1280 | SYSERROR("Failed to clear FD_CLOEXEC from file descriptor %d", fd); | |
1281 | goto on_error; | |
1282 | } | |
1283 | } | |
1284 | ||
1285 | if (options->attach_flags & LXC_ATTACH_TERMINAL) { | |
1286 | ret = lxc_terminal_prepare_login(ap->terminal_pts_fd); | |
1287 | if (ret < 0) { | |
1288 | SYSERROR("Failed to prepare terminal file descriptor %d", ap->terminal_pts_fd); | |
1289 | goto on_error; | |
1290 | } | |
1291 | ||
1292 | TRACE("Prepared terminal file descriptor %d", ap->terminal_pts_fd); | |
1293 | } | |
1294 | ||
1295 | /* Avoid unnecessary syscalls. */ | |
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; | |
1301 | ||
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"); | |
1309 | ||
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; | |
1320 | lxc_seccomp_close_notifier_fd(&conf->seccomp); | |
1321 | } | |
1322 | ||
1323 | if (!lxc_switch_uid_gid(ctx->target_ns_uid, ctx->target_ns_gid)) | |
1324 | goto on_error; | |
1325 | ||
1326 | put_attach_payload(ap); | |
1327 | ||
1328 | /* We're done, so we can now do whatever the user intended us to do. */ | |
1329 | _exit(attach_function(attach_function_args)); | |
1330 | ||
1331 | on_error: | |
1332 | ERROR("Failed to attach to container"); | |
1333 | put_attach_payload(ap); | |
1334 | _exit(EXIT_FAILURE); | |
1335 | } | |
1336 | ||
1337 | static int lxc_attach_terminal(const char *name, const char *lxcpath, struct lxc_conf *conf, | |
1338 | struct lxc_terminal *terminal) | |
1339 | { | |
1340 | int ret; | |
1341 | ||
1342 | lxc_terminal_init(terminal); | |
1343 | ||
1344 | ret = lxc_terminal_create(name, lxcpath, conf, terminal); | |
1345 | if (ret < 0) | |
1346 | return log_error(-1, "Failed to create terminal"); | |
1347 | ||
1348 | return 0; | |
1349 | } | |
1350 | ||
1351 | static int lxc_attach_terminal_mainloop_init(struct lxc_terminal *terminal, | |
1352 | struct lxc_epoll_descr *descr) | |
1353 | { | |
1354 | int ret; | |
1355 | ||
1356 | ret = lxc_mainloop_open(descr); | |
1357 | if (ret < 0) | |
1358 | return log_error(-1, "Failed to create mainloop"); | |
1359 | ||
1360 | ret = lxc_terminal_mainloop_add(descr, terminal); | |
1361 | if (ret < 0) { | |
1362 | lxc_mainloop_close(descr); | |
1363 | return log_error(-1, "Failed to add handlers to mainloop"); | |
1364 | } | |
1365 | ||
1366 | return 0; | |
1367 | } | |
1368 | ||
1369 | static inline void lxc_attach_terminal_close_ptx(struct lxc_terminal *terminal) | |
1370 | { | |
1371 | close_prot_errno_disarm(terminal->ptx); | |
1372 | } | |
1373 | ||
1374 | static inline void lxc_attach_terminal_close_pts(struct lxc_terminal *terminal) | |
1375 | { | |
1376 | close_prot_errno_disarm(terminal->pty); | |
1377 | } | |
1378 | ||
1379 | static inline void lxc_attach_terminal_close_peer(struct lxc_terminal *terminal) | |
1380 | { | |
1381 | close_prot_errno_disarm(terminal->peer); | |
1382 | } | |
1383 | ||
1384 | static inline void lxc_attach_terminal_close_log(struct lxc_terminal *terminal) | |
1385 | { | |
1386 | close_prot_errno_disarm(terminal->log_fd); | |
1387 | } | |
1388 | ||
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) | |
1392 | { | |
1393 | int ret_parent = -1; | |
1394 | struct lxc_epoll_descr descr = {}; | |
1395 | int ret; | |
1396 | char *name, *lxcpath; | |
1397 | int ipc_sockets[2]; | |
1398 | pid_t attached_pid, pid, to_cleanup_pid; | |
1399 | struct attach_context *ctx; | |
1400 | struct lxc_terminal terminal; | |
1401 | struct lxc_conf *conf; | |
1402 | ||
1403 | if (!container) | |
1404 | return ret_set_errno(-1, EINVAL); | |
1405 | ||
1406 | if (!lxc_container_get(container)) | |
1407 | return ret_set_errno(-1, EINVAL); | |
1408 | ||
1409 | name = container->name; | |
1410 | lxcpath = container->config_path; | |
1411 | ||
1412 | if (!options) { | |
1413 | options = &attach_static_default_options; | |
1414 | options->lsm_label = NULL; | |
1415 | } | |
1416 | ||
1417 | ctx = alloc_attach_context(); | |
1418 | if (!ctx) { | |
1419 | lxc_container_put(container); | |
1420 | return log_error_errno(-ENOMEM, ENOMEM, "Failed to allocate attach context"); | |
1421 | } | |
1422 | ||
1423 | ret = get_attach_context(ctx, container, options); | |
1424 | if (ret) { | |
1425 | put_attach_context(ctx); | |
1426 | return log_error(-1, "Failed to get attach context"); | |
1427 | } | |
1428 | ||
1429 | conf = ctx->container->lxc_conf; | |
1430 | ||
1431 | if (!fetch_seccomp(ctx->container, options)) | |
1432 | WARN("Failed to get seccomp policy"); | |
1433 | ||
1434 | if (!no_new_privs(ctx->container, options)) | |
1435 | WARN("Could not determine whether PR_SET_NO_NEW_PRIVS is set"); | |
1436 | ||
1437 | ret = prepare_namespaces(ctx, options); | |
1438 | if (ret) { | |
1439 | put_attach_context(ctx); | |
1440 | return log_error(-1, "Failed to get namespace file descriptors"); | |
1441 | } | |
1442 | ||
1443 | if (options->attach_flags & LXC_ATTACH_TERMINAL) { | |
1444 | ret = lxc_attach_terminal(name, lxcpath, conf, &terminal); | |
1445 | if (ret < 0) { | |
1446 | put_attach_context(ctx); | |
1447 | return log_error(-1, "Failed to setup new terminal"); | |
1448 | } | |
1449 | ||
1450 | terminal.log_fd = options->log_fd; | |
1451 | } else { | |
1452 | lxc_terminal_init(&terminal); | |
1453 | } | |
1454 | ||
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 | |
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 | |
1462 | * our own child process will not really cause the socket to close | |
1463 | * properly, potentially causing the parent to hang. | |
1464 | * | |
1465 | * For this reason, while IPC is still active, we have to use shutdown() | |
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. | |
1469 | * | |
1470 | * IPC mechanism: (X is receiver) | |
1471 | * initial process transient process attached process | |
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 | |
1480 | * [set LXC_ATTACH_NO_NEW_PRIVS] | |
1481 | * X <------------------------------------ send 3 | |
1482 | * [open LSM label fd] | |
1483 | * send 4 ------------------------------------> X | |
1484 | * [set LSM label] | |
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) { | |
1490 | put_attach_context(ctx); | |
1491 | return log_error_errno(-1, errno, "Could not set up required IPC mechanism for attaching"); | |
1492 | } | |
1493 | ||
1494 | /* Create transient process, two reasons: | |
1495 | * 1. We can't setns() in the child itself, since we want to make | |
1496 | * sure we are properly attached to the pidns. | |
1497 | * 2. Also, the initial thread has to put the attached process | |
1498 | * into the cgroup, which we can only do if we didn't already | |
1499 | * setns() (otherwise, user namespaces will hate us). | |
1500 | */ | |
1501 | pid = fork(); | |
1502 | if (pid < 0) { | |
1503 | put_attach_context(ctx); | |
1504 | return log_error_errno(-1, errno, "Failed to create first subprocess"); | |
1505 | } | |
1506 | ||
1507 | if (pid == 0) { | |
1508 | char *cwd, *new_cwd; | |
1509 | ||
1510 | /* close unneeded file descriptors */ | |
1511 | close_prot_errno_disarm(ipc_sockets[0]); | |
1512 | ||
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); | |
1517 | } | |
1518 | ||
1519 | /* Wait for the parent to have setup cgroups. */ | |
1520 | if (!sync_wait(ipc_sockets[1], ATTACH_SYNC_CGROUP)) { | |
1521 | shutdown(ipc_sockets[1], SHUT_RDWR); | |
1522 | put_attach_context(ctx); | |
1523 | _exit(EXIT_FAILURE); | |
1524 | } | |
1525 | ||
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 | ||
1532 | cwd = getcwd(NULL, 0); | |
1533 | ||
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! | |
1543 | */ | |
1544 | ret = attach_namespaces(ctx, options); | |
1545 | if (ret < 0) { | |
1546 | ERROR("Failed to enter namespaces"); | |
1547 | shutdown(ipc_sockets[1], SHUT_RDWR); | |
1548 | put_attach_context(ctx); | |
1549 | _exit(EXIT_FAILURE); | |
1550 | } | |
1551 | ||
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); | |
1561 | } | |
1562 | free_disarm(cwd); | |
1563 | ||
1564 | /* Create attached process. */ | |
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); | |
1569 | put_attach_context(ctx); | |
1570 | _exit(EXIT_FAILURE); | |
1571 | } | |
1572 | ||
1573 | if (pid == 0) { | |
1574 | struct attach_payload ap = { | |
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 | ||
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 | } | |
1590 | ||
1591 | /* Does not return. */ | |
1592 | do_attach(&ap); | |
1593 | } | |
1594 | TRACE("Attached process %d started initializing", pid); | |
1595 | ||
1596 | if (options->attach_flags & LXC_ATTACH_TERMINAL) | |
1597 | lxc_attach_terminal_close_pts(&terminal); | |
1598 | ||
1599 | /* Tell grandparent the pid of the pid of the newly created child. */ | |
1600 | if (!sync_wake_pid(ipc_sockets[1], ATTACH_SYNC_PID(pid))) { | |
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); | |
1608 | put_attach_context(ctx); | |
1609 | _exit(EXIT_FAILURE); | |
1610 | } | |
1611 | ||
1612 | /* The rest is in the hands of the initial and the attached process. */ | |
1613 | put_attach_context(ctx); | |
1614 | _exit(EXIT_SUCCESS); | |
1615 | } | |
1616 | TRACE("Transient process %d started initializing", pid); | |
1617 | ||
1618 | to_cleanup_pid = pid; | |
1619 | ||
1620 | /* close unneeded file descriptors */ | |
1621 | close_prot_errno_disarm(ipc_sockets[1]); | |
1622 | put_namespaces(ctx); | |
1623 | if (options->attach_flags & LXC_ATTACH_TERMINAL) | |
1624 | lxc_attach_terminal_close_pts(&terminal); | |
1625 | ||
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); | |
1633 | if (ret) { | |
1634 | call_cleaner(cgroup_exit) struct cgroup_ops *cgroup_ops = NULL; | |
1635 | ||
1636 | if (ret != -ENOCGROUP2) { | |
1637 | SYSERROR("Failed to attach cgroup"); | |
1638 | goto on_error; | |
1639 | } | |
1640 | ||
1641 | cgroup_ops = cgroup_init(conf); | |
1642 | if (!cgroup_ops) | |
1643 | goto on_error; | |
1644 | ||
1645 | if (!cgroup_ops->attach(cgroup_ops, conf, name, lxcpath, pid)) | |
1646 | goto on_error; | |
1647 | } | |
1648 | ||
1649 | TRACE("Moved transient process %d into container cgroup", pid); | |
1650 | } | |
1651 | ||
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; | |
1657 | ||
1658 | TRACE("Setup /proc/%d settings", pid); | |
1659 | } | |
1660 | ||
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; | |
1666 | ||
1667 | TRACE("Setup resource limits"); | |
1668 | } | |
1669 | ||
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; | |
1674 | ||
1675 | TRACE("Initialized terminal mainloop"); | |
1676 | } | |
1677 | ||
1678 | /* Let the child process know to go ahead. */ | |
1679 | if (!sync_wake(ipc_sockets[0], ATTACH_SYNC_CGROUP)) | |
1680 | goto close_mainloop; | |
1681 | ||
1682 | TRACE("Told transient process to start initializing"); | |
1683 | ||
1684 | /* Get pid of attached process from transient process. */ | |
1685 | if (!sync_wait_pid(ipc_sockets[0], ATTACH_SYNC_PID(&attached_pid))) | |
1686 | goto close_mainloop; | |
1687 | ||
1688 | TRACE("Received pid %d of attached process in parent pid namespace", attached_pid); | |
1689 | ||
1690 | /* Ignore SIGKILL (CTRL-C) and SIGQUIT (CTRL-\) - issue #313. */ | |
1691 | if (options->stdin_fd == STDIN_FILENO) { | |
1692 | signal(SIGINT, SIG_IGN); | |
1693 | signal(SIGQUIT, SIG_IGN); | |
1694 | } | |
1695 | ||
1696 | /* Reap transient process. */ | |
1697 | ret = wait_for_pid(pid); | |
1698 | if (ret < 0) | |
1699 | goto close_mainloop; | |
1700 | ||
1701 | TRACE("Transient process %d exited", pid); | |
1702 | ||
1703 | /* We will always have to reap the attached process now. */ | |
1704 | to_cleanup_pid = attached_pid; | |
1705 | ||
1706 | /* Open LSM fd and send it to child. */ | |
1707 | if (attach_lsm(options) && ctx->lsm_label) { | |
1708 | __do_close int labelfd = -EBADF; | |
1709 | bool on_exec; | |
1710 | ||
1711 | on_exec = options->attach_flags & LXC_ATTACH_LSM_EXEC ? true : false; | |
1712 | labelfd = ctx->lsm_ops->process_label_fd_get(ctx->lsm_ops, attached_pid, on_exec); | |
1713 | if (labelfd < 0) | |
1714 | goto close_mainloop; | |
1715 | ||
1716 | TRACE("Opened LSM label file descriptor %d", labelfd); | |
1717 | ||
1718 | /* Send child fd of the LSM security module to write to. */ | |
1719 | if (!sync_wake_fd(ipc_sockets[0], ATTACH_SYNC_LSM(labelfd))) { | |
1720 | SYSERROR("Failed to send lsm label fd"); | |
1721 | goto close_mainloop; | |
1722 | } | |
1723 | ||
1724 | TRACE("Sent LSM label file descriptor %d to child", labelfd); | |
1725 | } | |
1726 | ||
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; | |
1731 | ||
1732 | ret = lxc_seccomp_add_notifier(name, lxcpath, &conf->seccomp); | |
1733 | if (ret < 0) | |
1734 | goto close_mainloop; | |
1735 | } | |
1736 | ||
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 | */ | |
1741 | ||
1742 | *attached_process = attached_pid; | |
1743 | ||
1744 | /* Now shut down communication with child, we're done. */ | |
1745 | shutdown(ipc_sockets[0], SHUT_RDWR); | |
1746 | close_prot_errno_disarm(ipc_sockets[0]); | |
1747 | ||
1748 | ret_parent = 0; | |
1749 | to_cleanup_pid = -1; | |
1750 | ||
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 | } | |
1757 | } | |
1758 | ||
1759 | close_mainloop: | |
1760 | if (options->attach_flags & LXC_ATTACH_TERMINAL) | |
1761 | lxc_mainloop_close(&descr); | |
1762 | ||
1763 | on_error: | |
1764 | if (ipc_sockets[0] >= 0) { | |
1765 | shutdown(ipc_sockets[0], SHUT_RDWR); | |
1766 | close_prot_errno_disarm(ipc_sockets[0]); | |
1767 | } | |
1768 | ||
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 | } | |
1776 | ||
1777 | put_attach_context(ctx); | |
1778 | return ret_parent; | |
1779 | } | |
1780 | ||
1781 | int lxc_attach_run_command(void *payload) | |
1782 | { | |
1783 | int ret = -1; | |
1784 | lxc_attach_command_t *cmd = payload; | |
1785 | ||
1786 | ret = execvp(cmd->program, cmd->argv); | |
1787 | if (ret < 0) { | |
1788 | switch (errno) { | |
1789 | case ENOEXEC: | |
1790 | ret = 126; | |
1791 | break; | |
1792 | case ENOENT: | |
1793 | ret = 127; | |
1794 | break; | |
1795 | } | |
1796 | } | |
1797 | ||
1798 | return log_error_errno(ret, errno, "Failed to exec \"%s\"", cmd->program); | |
1799 | } | |
1800 | ||
1801 | int lxc_attach_run_shell(void* payload) | |
1802 | { | |
1803 | __do_free char *buf = NULL; | |
1804 | uid_t uid; | |
1805 | struct passwd pwent; | |
1806 | struct passwd *pwentp = NULL; | |
1807 | char *user_shell; | |
1808 | size_t bufsize; | |
1809 | int ret; | |
1810 | ||
1811 | /* Ignore payload parameter. */ | |
1812 | (void)payload; | |
1813 | ||
1814 | uid = getuid(); | |
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) | |
1825 | WARN("Could not find matched password record"); | |
1826 | ||
1827 | WARN("Failed to get password record - %u", uid); | |
1828 | } | |
1829 | } | |
1830 | ||
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. | |
1836 | */ | |
1837 | if (!pwentp) | |
1838 | user_shell = lxc_attach_getpwshell(uid); | |
1839 | else | |
1840 | user_shell = pwent.pw_shell; | |
1841 | ||
1842 | if (user_shell) | |
1843 | execlp(user_shell, user_shell, (char *)NULL); | |
1844 | ||
1845 | /* Executed if either no passwd entry or execvp fails, we will fall back | |
1846 | * on /bin/sh as a default shell. | |
1847 | */ | |
1848 | execlp("/bin/sh", "/bin/sh", (char *)NULL); | |
1849 | ||
1850 | SYSERROR("Failed to execute shell"); | |
1851 | if (!pwentp) | |
1852 | free(user_shell); | |
1853 | ||
1854 | return -1; | |
1855 | } |