2 * lxc: linux Container library
4 * (C) Copyright IBM Corp. 2007, 2008
7 * Daniel Lezcano <daniel.lezcano at free.fr>
9 * This library is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public
11 * License as published by the Free Software Foundation; either
12 * version 2.1 of the License, or (at your option) any later version.
14 * This library is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with this library; if not, write to the Free Software
21 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
28 #include <linux/unistd.h>
34 #include <sys/mount.h>
35 #include <sys/param.h>
36 #include <sys/prctl.h>
37 #include <sys/socket.h>
38 #include <sys/syscall.h>
43 #include <lxc/lxccontainer.h>
56 #include "lxcseccomp.h"
59 #include "namespace.h"
63 #if HAVE_SYS_PERSONALITY_H
64 #include <sys/personality.h>
67 lxc_log_define(attach
, lxc
);
69 /* Define default options if no options are supplied by the user. */
70 static lxc_attach_options_t attach_static_default_options
= LXC_ATTACH_OPTIONS_DEFAULT
;
72 static struct lxc_proc_context_info
*lxc_proc_get_context_info(pid_t pid
)
77 char proc_fn
[LXC_PROC_STATUS_LEN
];
78 size_t line_bufsz
= 0;
80 struct lxc_proc_context_info
*info
= NULL
;
82 /* Read capabilities. */
83 ret
= snprintf(proc_fn
, LXC_PROC_STATUS_LEN
, "/proc/%d/status", pid
);
84 if (ret
< 0 || ret
>= LXC_PROC_STATUS_LEN
)
87 proc_file
= fopen(proc_fn
, "r");
89 SYSERROR("Could not open %s", proc_fn
);
93 info
= calloc(1, sizeof(*info
));
95 SYSERROR("Could not allocate memory");
102 while (getline(&line
, &line_bufsz
, proc_file
) != -1) {
103 ret
= sscanf(line
, "CapBnd: %llx", &info
->capability_mask
);
104 if (ret
!= EOF
&& ret
== 1) {
114 ERROR("Could not read capability bounding set from %s",
119 info
->lsm_label
= lsm_process_label_get(pid
);
120 info
->ns_inherited
= 0;
121 memset(info
->ns_fd
, -1, sizeof(int) * LXC_NS_MAX
);
130 static inline void lxc_proc_close_ns_fd(struct lxc_proc_context_info
*ctx
)
134 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
135 if (ctx
->ns_fd
[i
] < 0)
138 close(ctx
->ns_fd
[i
]);
139 ctx
->ns_fd
[i
] = -EBADF
;
143 static void lxc_proc_put_context_info(struct lxc_proc_context_info
*ctx
)
145 free(ctx
->lsm_label
);
146 ctx
->lsm_label
= NULL
;
148 if (ctx
->container
) {
149 lxc_container_put(ctx
->container
);
150 ctx
->container
= NULL
;
153 lxc_proc_close_ns_fd(ctx
);
158 * in_same_namespace - Check whether two processes are in the same namespace.
159 * @pid1 - PID of the first process.
160 * @pid2 - PID of the second process.
161 * @ns - Name of the namespace to check. Must correspond to one of the names
162 * for the namespaces as shown in /proc/<pid/ns/
164 * If the two processes are not in the same namespace returns an fd to the
165 * namespace of the second process identified by @pid2. If the two processes are
166 * in the same namespace returns -EINVAL, -1 if an error occurred.
168 static int in_same_namespace(pid_t pid1
, pid_t pid2
, const char *ns
)
170 int ns_fd1
= -1, ns_fd2
= -1, ret
= -1;
172 struct stat ns_st1
, ns_st2
;
174 ns_fd1
= lxc_preserve_ns(pid1
, ns
);
176 /* The kernel does not support this namespace. This is not an
185 ns_fd2
= lxc_preserve_ns(pid2
, ns
);
189 ret
= fstat(ns_fd1
, &ns_st1
);
193 ret
= fstat(ns_fd2
, &ns_st2
);
197 /* processes are in the same namespace */
198 if ((ns_st1
.st_dev
== ns_st2
.st_dev
) && (ns_st1
.st_ino
== ns_st2
.st_ino
)) {
203 /* processes are in different namespaces */
220 static int lxc_attach_to_ns(pid_t pid
, struct lxc_proc_context_info
*ctx
)
224 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
225 if (ctx
->ns_fd
[i
] < 0)
228 ret
= setns(ctx
->ns_fd
[i
], ns_info
[i
].clone_flag
);
230 SYSERROR("Failed to attach to %s namespace of %d",
231 ns_info
[i
].proc_name
, pid
);
235 DEBUG("Attached to %s namespace of %d", ns_info
[i
].proc_name
, pid
);
241 static int lxc_attach_remount_sys_proc(void)
245 ret
= unshare(CLONE_NEWNS
);
247 SYSERROR("Failed to unshare mount namespace");
251 if (detect_shared_rootfs()) {
252 if (mount(NULL
, "/", NULL
, MS_SLAVE
| MS_REC
, NULL
)) {
253 SYSERROR("Failed to make / rslave");
254 ERROR("Continuing...");
258 /* Assume /proc is always mounted, so remount it. */
259 ret
= umount2("/proc", MNT_DETACH
);
261 SYSERROR("Failed to unmount /proc");
265 ret
= mount("none", "/proc", "proc", 0, NULL
);
267 SYSERROR("Failed to remount /proc");
271 /* Try to umount /sys. If it's not a mount point, we'll get EINVAL, then
272 * we ignore it because it may not have been mounted in the first place.
274 ret
= umount2("/sys", MNT_DETACH
);
275 if (ret
< 0 && errno
!= EINVAL
) {
276 SYSERROR("Failed to unmount /sys");
278 } else if (ret
== 0) {
280 ret
= mount("none", "/sys", "sysfs", 0, NULL
);
282 SYSERROR("Failed to remount /sys");
290 static int lxc_attach_drop_privs(struct lxc_proc_context_info
*ctx
)
294 last_cap
= lxc_caps_last_cap();
295 for (cap
= 0; cap
<= last_cap
; cap
++) {
296 if (ctx
->capability_mask
& (1LL << cap
))
299 if (prctl(PR_CAPBSET_DROP
, prctl_arg(cap
), prctl_arg(0),
300 prctl_arg(0), prctl_arg(0))) {
301 SYSERROR("Failed to drop capability %d", cap
);
305 TRACE("Dropped capability %d", cap
);
311 static int lxc_attach_set_environment(struct lxc_proc_context_info
*init_ctx
,
312 enum lxc_attach_env_policy_t policy
,
313 char **extra_env
, char **extra_keep
)
316 struct lxc_list
*iterator
;
318 if (policy
== LXC_ATTACH_CLEAR_ENV
) {
320 char **extra_keep_store
= NULL
;
325 for (count
= 0; extra_keep
[count
]; count
++)
328 extra_keep_store
= calloc(count
, sizeof(char *));
329 if (!extra_keep_store
)
332 for (i
= 0; i
< count
; i
++) {
333 char *v
= getenv(extra_keep
[i
]);
335 extra_keep_store
[i
] = strdup(v
);
336 if (!extra_keep_store
[i
]) {
338 free(extra_keep_store
[--i
]);
340 free(extra_keep_store
);
344 if (strcmp(extra_keep
[i
], "PATH") == 0)
351 if (extra_keep_store
) {
354 for (p
= extra_keep_store
; *p
; p
++)
357 free(extra_keep_store
);
360 ERROR("Failed to clear environment");
364 if (extra_keep_store
) {
367 for (i
= 0; extra_keep
[i
]; i
++) {
368 if (extra_keep_store
[i
]) {
369 ret
= setenv(extra_keep
[i
], extra_keep_store
[i
], 1);
371 SYSWARN("Failed to set environment variable");
374 free(extra_keep_store
[i
]);
377 free(extra_keep_store
);
380 /* Always set a default path; shells and execlp tend to be fine
381 * without it, but there is a disturbing number of C programs
382 * out there that just assume that getenv("PATH") is never NULL
383 * and then die a painful segfault death.
386 ret
= setenv("PATH", "/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin", 1);
388 SYSWARN("Failed to set environment variable");
392 ret
= putenv("container=lxc");
394 SYSWARN("Failed to set environment variable");
398 /* Set container environment variables.*/
399 if (init_ctx
&& init_ctx
->container
&& init_ctx
->container
->lxc_conf
) {
400 lxc_list_for_each(iterator
, &init_ctx
->container
->lxc_conf
->environment
) {
403 env_tmp
= strdup((char *)iterator
->elem
);
407 ret
= putenv(env_tmp
);
409 SYSERROR("Failed to set environment variable: %s", (char *)iterator
->elem
);
415 /* Set extra environment variables. */
417 for (; *extra_env
; extra_env
++) {
420 /* We just assume the user knows what they are doing, so
421 * we don't do any checks.
423 p
= strdup(*extra_env
);
429 SYSWARN("Failed to set environment variable");
436 static char *lxc_attach_getpwshell(uid_t uid
)
443 size_t line_bufsz
= 0;
444 char *line
= NULL
, *result
= NULL
;
446 /* We need to fork off a process that runs the getent program, and we
447 * need to capture its output, so we use a pipe for that purpose.
449 ret
= pipe2(pipes
, O_CLOEXEC
);
462 char *arguments
[] = {
471 /* We want to capture stdout. */
472 ret
= dup2(pipes
[1], STDOUT_FILENO
);
477 /* Get rid of stdin/stderr, so we try to associate it with
483 close(STDERR_FILENO
);
485 (void)dup3(fd
, STDIN_FILENO
, O_CLOEXEC
);
486 (void)dup3(fd
, STDOUT_FILENO
, O_CLOEXEC
);
490 /* Finish argument list. */
491 ret
= snprintf(uid_buf
, sizeof(uid_buf
), "%ld", (long)uid
);
492 if (ret
<= 0 || ret
>= sizeof(uid_buf
))
495 /* Try to run getent program. */
496 (void)execvp("getent", arguments
);
502 pipe_f
= fdopen(pipes
[0], "r");
503 while (getline(&line
, &line_bufsz
, pipe_f
) != -1) {
507 char *endptr
= NULL
, *saveptr
= NULL
;
509 /* If we already found something, just continue to read
510 * until the pipe doesn't deliver any more data, but
511 * don't modify the existing data structure.
519 /* Trim line on the right hand side. */
520 for (i
= strlen(line
); i
> 0 && (line
[i
- 1] == '\n' || line
[i
- 1] == '\r'); --i
)
523 /* Split into tokens: first: user name. */
524 token
= strtok_r(line
, ":", &saveptr
);
528 /* next: dummy password field */
529 token
= strtok_r(NULL
, ":", &saveptr
);
534 token
= strtok_r(NULL
, ":", &saveptr
);
535 value
= token
? strtol(token
, &endptr
, 10) : 0;
536 if (!token
|| !endptr
|| *endptr
|| value
== LONG_MIN
||
540 /* dummy sanity check: user id matches */
541 if ((uid_t
)value
!= uid
)
544 /* skip fields: gid, gecos, dir, go to next field 'shell' */
545 for (i
= 0; i
< 4; i
++) {
546 token
= strtok_r(NULL
, ":", &saveptr
);
555 result
= strdup(token
);
557 /* Sanity check that there are no fields after that. */
558 token
= strtok_r(NULL
, ":", &saveptr
);
568 ret
= wait_for_pid(pid
);
582 static void lxc_attach_get_init_uidgid(uid_t
*init_uid
, gid_t
*init_gid
)
585 char proc_fn
[LXC_PROC_STATUS_LEN
];
588 size_t line_bufsz
= 0;
590 uid_t uid
= (uid_t
)-1;
591 gid_t gid
= (gid_t
)-1;
593 ret
= snprintf(proc_fn
, LXC_PROC_STATUS_LEN
, "/proc/%d/status", 1);
594 if (ret
< 0 || ret
>= LXC_PROC_STATUS_LEN
)
597 proc_file
= fopen(proc_fn
, "r");
601 while (getline(&line
, &line_bufsz
, proc_file
) != -1) {
602 /* Format is: real, effective, saved set user, fs we only care
605 ret
= sscanf(line
, "Uid: %ld", &value
);
606 if (ret
!= EOF
&& ret
== 1) {
609 ret
= sscanf(line
, "Gid: %ld", &value
);
610 if (ret
!= EOF
&& ret
== 1)
614 if (uid
!= (uid_t
)-1 && gid
!= (gid_t
)-1)
621 /* Only override arguments if we found something. */
622 if (uid
!= (uid_t
)-1)
625 if (gid
!= (gid_t
)-1)
628 /* TODO: we should also parse supplementary groups and use
629 * setgroups() to set them.
633 static bool fetch_seccomp(struct lxc_container
*c
, lxc_attach_options_t
*options
)
639 if (!(options
->namespaces
& CLONE_NEWNS
) ||
640 !(options
->attach_flags
& LXC_ATTACH_LSM
)) {
641 free(c
->lxc_conf
->seccomp
);
642 c
->lxc_conf
->seccomp
= NULL
;
646 /* Remove current setting. */
647 if (!c
->set_config_item(c
, "lxc.seccomp.profile", "") &&
648 !c
->set_config_item(c
, "lxc.seccomp", ""))
651 /* Fetch the current profile path over the cmd interface. */
652 path
= c
->get_running_config_item(c
, "lxc.seccomp.profile");
654 INFO("Failed to retrieve lxc.seccomp.profile");
656 path
= c
->get_running_config_item(c
, "lxc.seccomp");
658 INFO("Failed to retrieve lxc.seccomp");
663 /* Copy the value into the new lxc_conf. */
664 bret
= c
->set_config_item(c
, "lxc.seccomp.profile", path
);
669 /* Attempt to parse the resulting config. */
670 ret
= lxc_read_seccomp_config(c
->lxc_conf
);
672 ERROR("Failed to retrieve seccomp policy");
676 INFO("Retrieved seccomp policy");
680 static bool no_new_privs(struct lxc_container
*c
, lxc_attach_options_t
*options
)
685 /* Remove current setting. */
686 if (!c
->set_config_item(c
, "lxc.no_new_privs", "")) {
687 INFO("Failed to unset lxc.no_new_privs");
691 /* Retrieve currently active setting. */
692 val
= c
->get_running_config_item(c
, "lxc.no_new_privs");
694 INFO("Failed to retrieve lxc.no_new_privs");
698 /* Set currently active setting. */
699 bret
= c
->set_config_item(c
, "lxc.no_new_privs", val
);
704 static signed long get_personality(const char *name
, const char *lxcpath
)
709 p
= lxc_cmd_get_config_item(name
, "lxc.arch", lxcpath
);
713 ret
= lxc_config_parse_arch(p
);
719 struct attach_clone_payload
{
721 int terminal_slave_fd
;
722 lxc_attach_options_t
*options
;
723 struct lxc_proc_context_info
*init_ctx
;
724 lxc_attach_exec_t exec_function
;
728 static void lxc_put_attach_clone_payload(struct attach_clone_payload
*p
)
730 if (p
->ipc_socket
>= 0) {
731 shutdown(p
->ipc_socket
, SHUT_RDWR
);
732 close(p
->ipc_socket
);
733 p
->ipc_socket
= -EBADF
;
736 if (p
->terminal_slave_fd
>= 0) {
737 close(p
->terminal_slave_fd
);
738 p
->terminal_slave_fd
= -EBADF
;
742 lxc_proc_put_context_info(p
->init_ctx
);
747 static int attach_child_main(struct attach_clone_payload
*payload
)
752 uid_t ns_root_uid
= 0;
753 gid_t ns_root_gid
= 0;
754 lxc_attach_options_t
* options
= payload
->options
;
755 struct lxc_proc_context_info
* init_ctx
= payload
->init_ctx
;
756 bool needs_lsm
= (options
->namespaces
& CLONE_NEWNS
) &&
757 (options
->attach_flags
& LXC_ATTACH_LSM
) &&
760 /* A description of the purpose of this functionality is provided in the
761 * lxc-attach(1) manual page. We have to remount here and not in the
762 * parent process, otherwise /proc may not properly reflect the new pid
765 if (!(options
->namespaces
& CLONE_NEWNS
) &&
766 (options
->attach_flags
& LXC_ATTACH_REMOUNT_PROC_SYS
)) {
767 ret
= lxc_attach_remount_sys_proc();
771 TRACE("Remounted \"/proc\" and \"/sys\"");
774 /* Now perform additional attachments. */
775 #if HAVE_SYS_PERSONALITY_H
776 if (options
->attach_flags
& LXC_ATTACH_SET_PERSONALITY
) {
777 long new_personality
;
779 if (options
->personality
< 0)
780 new_personality
= init_ctx
->personality
;
782 new_personality
= options
->personality
;
784 ret
= personality(new_personality
);
788 TRACE("Set new personality");
792 if (options
->attach_flags
& LXC_ATTACH_DROP_CAPABILITIES
) {
793 ret
= lxc_attach_drop_privs(init_ctx
);
797 TRACE("Dropped capabilities");
800 /* Always set the environment (specify (LXC_ATTACH_KEEP_ENV, NULL, NULL)
801 * if you want this to be a no-op).
803 ret
= lxc_attach_set_environment(init_ctx
,
805 options
->extra_env_vars
,
806 options
->extra_keep_env
);
810 TRACE("Set up environment");
812 /* This remark only affects fully unprivileged containers:
813 * Receive fd for LSM security module before we set{g,u}id(). The reason
814 * is that on set{g,u}id() the kernel will a) make us undumpable and b)
815 * we will change our effective uid. This means our effective uid will
816 * be different from the effective uid of the process that created us
817 * which means that this processs no longer has capabilities in our
818 * namespace including CAP_SYS_PTRACE. This means we will not be able to
819 * read and /proc/<pid> files for the process anymore when /proc is
820 * mounted with hidepid={1,2}. So let's get the lsm label fd before the
824 ret
= lxc_abstract_unix_recv_fds(payload
->ipc_socket
, &lsm_fd
, 1, NULL
, 0);
827 SYSERROR("Failed to receive lsm label fd");
832 TRACE("Received LSM label file descriptor %d from parent", lsm_fd
);
835 if (options
->stdin_fd
> 0 && isatty(options
->stdin_fd
)) {
836 ret
= lxc_make_controlling_terminal(options
->stdin_fd
);
841 if (options
->namespaces
& CLONE_NEWUSER
) {
842 /* Check whether nsuid 0 has a mapping. */
843 ns_root_uid
= get_ns_uid(0);
845 /* Check whether nsgid 0 has a mapping. */
846 ns_root_gid
= get_ns_gid(0);
848 /* If there's no mapping for nsuid 0 try to retrieve the nsuid
849 * init was started with.
851 if (ns_root_uid
== LXC_INVALID_UID
)
852 lxc_attach_get_init_uidgid(&ns_root_uid
, &ns_root_gid
);
854 if (ns_root_uid
== LXC_INVALID_UID
)
857 ret
= lxc_switch_uid_gid(ns_root_uid
, ns_root_gid
);
862 ret
= lxc_setgroups(0, NULL
);
863 if (ret
< 0 && errno
!= EPERM
)
867 if (options
->uid
!= LXC_INVALID_UID
)
868 new_uid
= options
->uid
;
870 new_uid
= ns_root_uid
;
872 if (options
->gid
!= LXC_INVALID_GID
)
873 new_gid
= options
->gid
;
875 new_gid
= ns_root_gid
;
877 if ((init_ctx
->container
&& init_ctx
->container
->lxc_conf
&&
878 init_ctx
->container
->lxc_conf
->no_new_privs
) ||
879 (options
->attach_flags
& LXC_ATTACH_NO_NEW_PRIVS
)) {
880 ret
= prctl(PR_SET_NO_NEW_PRIVS
, prctl_arg(1), prctl_arg(0),
881 prctl_arg(0), prctl_arg(0));
885 TRACE("Set PR_SET_NO_NEW_PRIVS");
891 /* Change into our new LSM profile. */
892 on_exec
= options
->attach_flags
& LXC_ATTACH_LSM_EXEC
? true : false;
894 ret
= lsm_process_label_set_at(lsm_fd
, init_ctx
->lsm_label
, on_exec
);
899 TRACE("Set %s LSM label to \"%s\"", lsm_name(), init_ctx
->lsm_label
);
902 if (init_ctx
->container
&& init_ctx
->container
->lxc_conf
&&
903 init_ctx
->container
->lxc_conf
->seccomp
) {
904 ret
= lxc_seccomp_load(init_ctx
->container
->lxc_conf
);
908 TRACE("Loaded seccomp profile");
911 shutdown(payload
->ipc_socket
, SHUT_RDWR
);
912 close(payload
->ipc_socket
);
913 payload
->ipc_socket
= -EBADF
;
914 lxc_proc_put_context_info(init_ctx
);
915 payload
->init_ctx
= NULL
;
917 /* The following is done after the communication socket is shut down.
918 * That way, all errors that might (though unlikely) occur up until this
919 * point will have their messages printed to the original stderr (if
920 * logging is so configured) and not the fd the user supplied, if any.
923 /* Fd handling for stdin, stdout and stderr; ignore errors here, user
924 * may want to make sure the fds are closed, for example.
926 if (options
->stdin_fd
>= 0 && options
->stdin_fd
!= STDIN_FILENO
)
927 (void)dup2(options
->stdin_fd
, STDIN_FILENO
);
929 if (options
->stdout_fd
>= 0 && options
->stdout_fd
!= STDOUT_FILENO
)
930 (void)dup2(options
->stdout_fd
, STDOUT_FILENO
);
932 if (options
->stderr_fd
>= 0 && options
->stderr_fd
!= STDERR_FILENO
)
933 (void)dup2(options
->stderr_fd
, STDERR_FILENO
);
935 /* close the old fds */
936 if (options
->stdin_fd
> STDERR_FILENO
)
937 close(options
->stdin_fd
);
939 if (options
->stdout_fd
> STDERR_FILENO
)
940 close(options
->stdout_fd
);
942 if (options
->stderr_fd
> STDERR_FILENO
)
943 close(options
->stderr_fd
);
945 /* Try to remove FD_CLOEXEC flag from stdin/stdout/stderr, but also
946 * here, ignore errors.
948 for (fd
= STDIN_FILENO
; fd
<= STDERR_FILENO
; fd
++) {
949 ret
= fd_cloexec(fd
, false);
951 SYSERROR("Failed to clear FD_CLOEXEC from file descriptor %d", fd
);
956 if (options
->attach_flags
& LXC_ATTACH_TERMINAL
) {
957 ret
= lxc_terminal_prepare_login(payload
->terminal_slave_fd
);
959 SYSERROR("Failed to prepare terminal file descriptor %d", payload
->terminal_slave_fd
);
963 TRACE("Prepared terminal file descriptor %d", payload
->terminal_slave_fd
);
966 /* Avoid unnecessary syscalls. */
967 if (new_uid
== ns_root_uid
)
968 new_uid
= LXC_INVALID_UID
;
970 if (new_gid
== ns_root_gid
)
971 new_gid
= LXC_INVALID_GID
;
973 ret
= lxc_switch_uid_gid(new_uid
, new_gid
);
977 /* We're done, so we can now do whatever the user intended us to do. */
978 _exit(payload
->exec_function(payload
->exec_payload
));
981 lxc_put_attach_clone_payload(payload
);
985 static int lxc_attach_terminal(struct lxc_conf
*conf
,
986 struct lxc_terminal
*terminal
)
990 lxc_terminal_init(terminal
);
992 ret
= lxc_terminal_create(terminal
);
994 ERROR("Failed to create terminal");
998 /* Shift ttys to container. */
999 ret
= lxc_terminal_map_ids(conf
, terminal
);
1001 ERROR("Failed to chown terminal");
1008 lxc_terminal_delete(terminal
);
1009 lxc_terminal_conf_free(terminal
);
1013 static int lxc_attach_terminal_mainloop_init(struct lxc_terminal
*terminal
,
1014 struct lxc_epoll_descr
*descr
)
1018 ret
= lxc_mainloop_open(descr
);
1020 ERROR("Failed to create mainloop");
1024 ret
= lxc_terminal_mainloop_add(descr
, terminal
);
1026 ERROR("Failed to add handlers to mainloop");
1027 lxc_mainloop_close(descr
);
1034 static inline void lxc_attach_terminal_close_master(struct lxc_terminal
*terminal
)
1036 if (terminal
->master
< 0)
1039 close(terminal
->master
);
1040 terminal
->master
= -EBADF
;
1043 static inline void lxc_attach_terminal_close_slave(struct lxc_terminal
*terminal
)
1045 if (terminal
->slave
< 0)
1048 close(terminal
->slave
);
1049 terminal
->slave
= -EBADF
;
1052 static inline void lxc_attach_terminal_close_peer(struct lxc_terminal
*terminal
)
1054 if (terminal
->peer
< 0)
1057 close(terminal
->peer
);
1058 terminal
->peer
= -EBADF
;
1061 static inline void lxc_attach_terminal_close_log(struct lxc_terminal
*terminal
)
1063 if (terminal
->log_fd
< 0)
1066 close(terminal
->log_fd
);
1067 terminal
->log_fd
= -EBADF
;
1070 int lxc_attach(const char *name
, const char *lxcpath
,
1071 lxc_attach_exec_t exec_function
, void *exec_payload
,
1072 lxc_attach_options_t
*options
, pid_t
*attached_process
)
1076 char *cwd
, *new_cwd
;
1077 signed long personality
;
1078 pid_t attached_pid
, init_pid
, pid
;
1079 struct lxc_proc_context_info
*init_ctx
;
1080 struct lxc_terminal terminal
;
1081 struct lxc_conf
*conf
;
1082 struct attach_clone_payload payload
= {0};
1084 ret
= access("/proc/self/ns", X_OK
);
1086 SYSERROR("Does this kernel version support namespaces?");
1091 options
= &attach_static_default_options
;
1093 init_pid
= lxc_cmd_get_init_pid(name
, lxcpath
);
1095 ERROR("Failed to get init pid");
1099 init_ctx
= lxc_proc_get_context_info(init_pid
);
1101 ERROR("Failed to get context of init process: %ld", (long)init_pid
);
1105 personality
= get_personality(name
, lxcpath
);
1106 if (init_ctx
->personality
< 0) {
1107 ERROR("Failed to get personality of the container");
1108 lxc_proc_put_context_info(init_ctx
);
1111 init_ctx
->personality
= personality
;
1113 init_ctx
->container
= lxc_container_new(name
, lxcpath
);
1114 if (!init_ctx
->container
) {
1115 lxc_proc_put_context_info(init_ctx
);
1119 if (!init_ctx
->container
->lxc_conf
) {
1120 init_ctx
->container
->lxc_conf
= lxc_conf_init();
1121 if (!init_ctx
->container
->lxc_conf
) {
1122 lxc_proc_put_context_info(init_ctx
);
1126 conf
= init_ctx
->container
->lxc_conf
;
1128 if (!fetch_seccomp(init_ctx
->container
, options
))
1129 WARN("Failed to get seccomp policy");
1131 if (!no_new_privs(init_ctx
->container
, options
))
1132 WARN("Could not determine whether PR_SET_NO_NEW_PRIVS is set");
1134 cwd
= getcwd(NULL
, 0);
1136 /* Determine which namespaces the container was created with
1137 * by asking lxc-start, if necessary.
1139 if (options
->namespaces
== -1) {
1140 options
->namespaces
= lxc_cmd_get_clone_flags(name
, lxcpath
);
1142 if (options
->namespaces
== -1) {
1143 ERROR("Failed to automatically determine the "
1144 "namespaces which the container uses");
1146 lxc_proc_put_context_info(init_ctx
);
1150 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
1151 if (ns_info
[i
].clone_flag
& CLONE_NEWCGROUP
)
1152 if (!(options
->attach_flags
& LXC_ATTACH_MOVE_TO_CGROUP
) ||
1156 if (ns_info
[i
].clone_flag
& options
->namespaces
)
1159 init_ctx
->ns_inherited
|= ns_info
[i
].clone_flag
;
1163 pid
= lxc_raw_getpid();
1165 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
1168 if (options
->namespaces
& ns_info
[i
].clone_flag
)
1169 init_ctx
->ns_fd
[i
] = lxc_preserve_ns(init_pid
, ns_info
[i
].proc_name
);
1170 else if (init_ctx
->ns_inherited
& ns_info
[i
].clone_flag
)
1171 init_ctx
->ns_fd
[i
] = in_same_namespace(pid
, init_pid
, ns_info
[i
].proc_name
);
1175 if (init_ctx
->ns_fd
[i
] >= 0)
1178 if (init_ctx
->ns_fd
[i
] == -EINVAL
) {
1179 DEBUG("Inheriting %s namespace from %d",
1180 ns_info
[i
].proc_name
, pid
);
1181 init_ctx
->ns_inherited
&= ~ns_info
[i
].clone_flag
;
1185 /* We failed to preserve the namespace. */
1186 SYSERROR("Failed to attach to %s namespace of %d",
1187 ns_info
[i
].proc_name
, pid
);
1189 /* Close all already opened file descriptors before we return an
1190 * error, so we don't leak them.
1192 for (j
= 0; j
< i
; j
++)
1193 close(init_ctx
->ns_fd
[j
]);
1196 lxc_proc_put_context_info(init_ctx
);
1200 if (options
->attach_flags
& LXC_ATTACH_TERMINAL
) {
1201 ret
= lxc_attach_terminal(conf
, &terminal
);
1203 ERROR("Failed to setup new terminal");
1205 lxc_proc_put_context_info(init_ctx
);
1209 terminal
.log_fd
= options
->log_fd
;
1211 lxc_terminal_init(&terminal
);
1214 /* Create a socket pair for IPC communication; set SOCK_CLOEXEC in order
1215 * to make sure we don't irritate other threads that want to fork+exec
1218 * IMPORTANT: if the initial process is multithreaded and another call
1219 * just fork()s away without exec'ing directly after, the socket fd will
1220 * exist in the forked process from the other thread and any close() in
1221 * our own child process will not really cause the socket to close
1222 * properly, potentiall causing the parent to hang.
1224 * For this reason, while IPC is still active, we have to use shutdown()
1225 * if the child exits prematurely in order to signal that the socket is
1226 * closed and cannot assume that the child exiting will automatically do
1229 * IPC mechanism: (X is receiver)
1230 * initial process intermediate attached
1231 * X <--- send pid of
1234 * send 0 ------------------------------------> X
1235 * [do initialization]
1236 * X <------------------------------------ send 1
1237 * [add to cgroup, ...]
1238 * send 2 ------------------------------------> X
1239 * [set LXC_ATTACH_NO_NEW_PRIVS]
1240 * X <------------------------------------ send 3
1241 * [open LSM label fd]
1242 * send 4 ------------------------------------> X
1244 * close socket close socket
1247 ret
= socketpair(PF_LOCAL
, SOCK_STREAM
| SOCK_CLOEXEC
, 0, ipc_sockets
);
1249 SYSERROR("Could not set up required IPC mechanism for attaching");
1251 lxc_proc_put_context_info(init_ctx
);
1255 /* Create intermediate subprocess, two reasons:
1256 * 1. We can't setns() in the child itself, since we want to make
1257 * sure we are properly attached to the pidns.
1258 * 2. Also, the initial thread has to put the attached process
1259 * into the cgroup, which we can only do if we didn't already
1260 * setns() (otherwise, user namespaces will hate us).
1264 SYSERROR("Failed to create first subprocess");
1266 lxc_proc_put_context_info(init_ctx
);
1271 int ret_parent
= -1;
1272 pid_t to_cleanup_pid
= pid
;
1273 struct lxc_epoll_descr descr
= {0};
1275 /* close unneeded file descriptors */
1276 close(ipc_sockets
[1]);
1278 lxc_proc_close_ns_fd(init_ctx
);
1279 if (options
->attach_flags
& LXC_ATTACH_TERMINAL
)
1280 lxc_attach_terminal_close_slave(&terminal
);
1282 /* Attach to cgroup, if requested. */
1283 if (options
->attach_flags
& LXC_ATTACH_MOVE_TO_CGROUP
) {
1284 struct cgroup_ops
*cgroup_ops
;
1286 cgroup_ops
= cgroup_init(conf
);
1290 if (!cgroup_ops
->attach(cgroup_ops
, name
, lxcpath
, pid
))
1293 cgroup_exit(cgroup_ops
);
1294 TRACE("Moved intermediate process %d into container's cgroups", pid
);
1297 /* Setup /proc limits */
1298 if (!lxc_list_empty(&conf
->procs
)) {
1299 ret
= setup_proc_filesystem(&conf
->procs
, pid
);
1304 /* Setup resource limits */
1305 if (!lxc_list_empty(&conf
->limits
)) {
1306 ret
= setup_resource_limits(&conf
->limits
, pid
);
1311 if (options
->attach_flags
& LXC_ATTACH_TERMINAL
) {
1312 ret
= lxc_attach_terminal_mainloop_init(&terminal
, &descr
);
1316 TRACE("Initialized terminal mainloop");
1319 /* Let the child process know to go ahead. */
1321 ret
= lxc_write_nointr(ipc_sockets
[0], &status
, sizeof(status
));
1322 if (ret
!= sizeof(status
))
1323 goto close_mainloop
;
1325 TRACE("Told intermediate process to start initializing");
1327 /* Get pid of attached process from intermediate process. */
1328 ret
= lxc_read_nointr(ipc_sockets
[0], &attached_pid
, sizeof(attached_pid
));
1329 if (ret
!= sizeof(attached_pid
))
1330 goto close_mainloop
;
1332 TRACE("Received pid %d of attached process in parent pid namespace", attached_pid
);
1334 /* Ignore SIGKILL (CTRL-C) and SIGQUIT (CTRL-\) - issue #313. */
1335 if (options
->stdin_fd
== 0) {
1336 signal(SIGINT
, SIG_IGN
);
1337 signal(SIGQUIT
, SIG_IGN
);
1340 /* Reap intermediate process. */
1341 ret
= wait_for_pid(pid
);
1343 goto close_mainloop
;
1345 TRACE("Intermediate process %d exited", pid
);
1347 /* We will always have to reap the attached process now. */
1348 to_cleanup_pid
= attached_pid
;
1350 /* Open LSM fd and send it to child. */
1351 if ((options
->namespaces
& CLONE_NEWNS
) &&
1352 (options
->attach_flags
& LXC_ATTACH_LSM
) &&
1353 init_ctx
->lsm_label
) {
1358 on_exec
= options
->attach_flags
& LXC_ATTACH_LSM_EXEC
? true : false;
1359 labelfd
= lsm_process_label_fd_get(attached_pid
, on_exec
);
1361 goto close_mainloop
;
1363 TRACE("Opened LSM label file descriptor %d", labelfd
);
1365 /* Send child fd of the LSM security module to write to. */
1366 ret
= lxc_abstract_unix_send_fds(ipc_sockets
[0], &labelfd
, 1, NULL
, 0);
1369 SYSERROR("Failed to send lsm label fd");
1372 goto close_mainloop
;
1376 TRACE("Sent LSM label file descriptor %d to child", labelfd
);
1379 /* We're done, the child process should now execute whatever it
1380 * is that the user requested. The parent can now track it with
1381 * waitpid() or similar.
1384 *attached_process
= attached_pid
;
1386 /* Now shut down communication with child, we're done. */
1387 shutdown(ipc_sockets
[0], SHUT_RDWR
);
1388 close(ipc_sockets
[0]);
1389 ipc_sockets
[0] = -1;
1392 to_cleanup_pid
= -1;
1394 if (options
->attach_flags
& LXC_ATTACH_TERMINAL
) {
1395 ret
= lxc_mainloop(&descr
, -1);
1398 to_cleanup_pid
= attached_pid
;
1403 if (options
->attach_flags
& LXC_ATTACH_TERMINAL
)
1404 lxc_mainloop_close(&descr
);
1407 if (ipc_sockets
[0] >= 0) {
1408 shutdown(ipc_sockets
[0], SHUT_RDWR
);
1409 close(ipc_sockets
[0]);
1412 if (to_cleanup_pid
> 0)
1413 (void)wait_for_pid(to_cleanup_pid
);
1415 if (options
->attach_flags
& LXC_ATTACH_TERMINAL
) {
1416 lxc_terminal_delete(&terminal
);
1417 lxc_terminal_conf_free(&terminal
);
1420 lxc_proc_put_context_info(init_ctx
);
1424 /* close unneeded file descriptors */
1425 close(ipc_sockets
[0]);
1426 ipc_sockets
[0] = -EBADF
;
1428 if (options
->attach_flags
& LXC_ATTACH_TERMINAL
) {
1429 lxc_attach_terminal_close_master(&terminal
);
1430 lxc_attach_terminal_close_peer(&terminal
);
1431 lxc_attach_terminal_close_log(&terminal
);
1434 /* Wait for the parent to have setup cgroups. */
1435 ret
= lxc_read_nointr(ipc_sockets
[1], &status
, sizeof(status
));
1436 if (ret
!= sizeof(status
)) {
1437 shutdown(ipc_sockets
[1], SHUT_RDWR
);
1438 lxc_proc_put_context_info(init_ctx
);
1439 _exit(EXIT_FAILURE
);
1442 TRACE("Intermediate process starting to initialize");
1444 /* Attach now, create another subprocess later, since pid namespaces
1445 * only really affect the children of the current process.
1447 ret
= lxc_attach_to_ns(init_pid
, init_ctx
);
1449 ERROR("Failed to enter namespaces");
1450 shutdown(ipc_sockets
[1], SHUT_RDWR
);
1451 lxc_proc_put_context_info(init_ctx
);
1452 _exit(EXIT_FAILURE
);
1455 /* close namespace file descriptors */
1456 lxc_proc_close_ns_fd(init_ctx
);
1458 /* Attach succeeded, try to cwd. */
1459 if (options
->initial_cwd
)
1460 new_cwd
= options
->initial_cwd
;
1464 ret
= chdir(new_cwd
);
1466 WARN("Could not change directory to \"%s\"", new_cwd
);
1470 /* Create attached process. */
1471 payload
.ipc_socket
= ipc_sockets
[1];
1472 payload
.options
= options
;
1473 payload
.init_ctx
= init_ctx
;
1474 payload
.terminal_slave_fd
= terminal
.slave
;
1475 payload
.exec_function
= exec_function
;
1476 payload
.exec_payload
= exec_payload
;
1478 pid
= lxc_raw_clone(CLONE_PARENT
);
1480 SYSERROR("Failed to clone attached process");
1481 shutdown(ipc_sockets
[1], SHUT_RDWR
);
1482 lxc_proc_put_context_info(init_ctx
);
1483 _exit(EXIT_FAILURE
);
1487 ret
= attach_child_main(&payload
);
1489 ERROR("Failed to exec");
1491 _exit(EXIT_FAILURE
);
1494 if (options
->attach_flags
& LXC_ATTACH_TERMINAL
)
1495 lxc_attach_terminal_close_slave(&terminal
);
1497 /* Tell grandparent the pid of the pid of the newly created child. */
1498 ret
= lxc_write_nointr(ipc_sockets
[1], &pid
, sizeof(pid
));
1499 if (ret
!= sizeof(pid
)) {
1500 /* If this really happens here, this is very unfortunate, since
1501 * the parent will not know the pid of the attached process and
1502 * will not be able to wait for it (and we won't either due to
1503 * CLONE_PARENT) so the parent won't be able to reap it and the
1504 * attached process will remain a zombie.
1506 shutdown(ipc_sockets
[1], SHUT_RDWR
);
1507 lxc_proc_put_context_info(init_ctx
);
1508 _exit(EXIT_FAILURE
);
1511 TRACE("Sending pid %d of attached process", pid
);
1513 /* The rest is in the hands of the initial and the attached process. */
1514 lxc_proc_put_context_info(init_ctx
);
1518 int lxc_attach_run_command(void* payload
)
1520 lxc_attach_command_t
* cmd
= (lxc_attach_command_t
*)payload
;
1522 execvp(cmd
->program
, cmd
->argv
);
1524 SYSERROR("Failed to exec \"%s\"", cmd
->program
);
1528 int lxc_attach_run_shell(void* payload
)
1531 struct passwd pwent
;
1532 struct passwd
*pwentp
= NULL
;
1538 /* Ignore payload parameter. */
1543 bufsize
= sysconf(_SC_GETPW_R_SIZE_MAX
);
1547 buf
= malloc(bufsize
);
1549 ret
= getpwuid_r(uid
, &pwent
, buf
, bufsize
, &pwentp
);
1552 WARN("Could not find matched password record");
1554 WARN("Failed to get password record - %u", uid
);
1558 /* This probably happens because of incompatible nss implementations in
1559 * host and container (remember, this code is still using the host's
1560 * glibc but our mount namespace is in the container) we may try to get
1561 * the information by spawning a [getent passwd uid] process and parsing
1565 user_shell
= lxc_attach_getpwshell(uid
);
1567 user_shell
= pwent
.pw_shell
;
1570 execlp(user_shell
, user_shell
, (char *)NULL
);
1572 /* Executed if either no passwd entry or execvp fails, we will fall back
1573 * on /bin/sh as a default shell.
1575 execlp("/bin/sh", "/bin/sh", (char *)NULL
);
1577 SYSERROR("Failed to execute shell");