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
30 #include <linux/unistd.h>
37 #include <sys/mount.h>
38 #include <sys/param.h>
39 #include <sys/prctl.h>
40 #include <sys/socket.h>
41 #include <sys/syscall.h>
46 #include <lxc/lxccontainer.h>
59 #include "lxcseccomp.h"
62 #include "namespace.h"
63 #include "raw_syscalls.h"
64 #include "syscall_wrappers.h"
68 #if HAVE_SYS_PERSONALITY_H
69 #include <sys/personality.h>
72 lxc_log_define(attach
, lxc
);
74 /* Define default options if no options are supplied by the user. */
75 static lxc_attach_options_t attach_static_default_options
= LXC_ATTACH_OPTIONS_DEFAULT
;
77 static struct lxc_proc_context_info
*lxc_proc_get_context_info(pid_t pid
)
82 char proc_fn
[LXC_PROC_STATUS_LEN
];
83 size_t line_bufsz
= 0;
85 struct lxc_proc_context_info
*info
= NULL
;
87 /* Read capabilities. */
88 ret
= snprintf(proc_fn
, LXC_PROC_STATUS_LEN
, "/proc/%d/status", pid
);
89 if (ret
< 0 || ret
>= LXC_PROC_STATUS_LEN
)
92 proc_file
= fopen(proc_fn
, "r");
94 SYSERROR("Could not open %s", proc_fn
);
98 info
= calloc(1, sizeof(*info
));
100 SYSERROR("Could not allocate memory");
107 while (getline(&line
, &line_bufsz
, proc_file
) != -1) {
108 ret
= sscanf(line
, "CapBnd: %llx", &info
->capability_mask
);
109 if (ret
!= EOF
&& ret
== 1) {
119 ERROR("Could not read capability bounding set from %s",
124 info
->lsm_label
= lsm_process_label_get(pid
);
125 info
->ns_inherited
= 0;
126 memset(info
->ns_fd
, -1, sizeof(int) * LXC_NS_MAX
);
135 static inline void lxc_proc_close_ns_fd(struct lxc_proc_context_info
*ctx
)
139 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
140 if (ctx
->ns_fd
[i
] < 0)
143 close(ctx
->ns_fd
[i
]);
144 ctx
->ns_fd
[i
] = -EBADF
;
148 static void lxc_proc_put_context_info(struct lxc_proc_context_info
*ctx
)
150 free(ctx
->lsm_label
);
151 ctx
->lsm_label
= NULL
;
153 if (ctx
->container
) {
154 lxc_container_put(ctx
->container
);
155 ctx
->container
= NULL
;
158 lxc_proc_close_ns_fd(ctx
);
163 * in_same_namespace - Check whether two processes are in the same namespace.
164 * @pid1 - PID of the first process.
165 * @pid2 - PID of the second process.
166 * @ns - Name of the namespace to check. Must correspond to one of the names
167 * for the namespaces as shown in /proc/<pid/ns/
169 * If the two processes are not in the same namespace returns an fd to the
170 * namespace of the second process identified by @pid2. If the two processes are
171 * in the same namespace returns -EINVAL, -1 if an error occurred.
173 static int in_same_namespace(pid_t pid1
, pid_t pid2
, const char *ns
)
175 int ns_fd1
= -1, ns_fd2
= -1, ret
= -1;
177 struct stat ns_st1
, ns_st2
;
179 ns_fd1
= lxc_preserve_ns(pid1
, ns
);
181 /* The kernel does not support this namespace. This is not an
190 ns_fd2
= lxc_preserve_ns(pid2
, ns
);
194 ret
= fstat(ns_fd1
, &ns_st1
);
198 ret
= fstat(ns_fd2
, &ns_st2
);
202 /* processes are in the same namespace */
203 if ((ns_st1
.st_dev
== ns_st2
.st_dev
) && (ns_st1
.st_ino
== ns_st2
.st_ino
)) {
208 /* processes are in different namespaces */
225 static int lxc_attach_to_ns(pid_t pid
, struct lxc_proc_context_info
*ctx
)
229 for (i
= 0; i
< LXC_NS_MAX
; i
++) {
230 if (ctx
->ns_fd
[i
] < 0)
233 ret
= setns(ctx
->ns_fd
[i
], ns_info
[i
].clone_flag
);
235 SYSERROR("Failed to attach to %s namespace of %d",
236 ns_info
[i
].proc_name
, pid
);
240 DEBUG("Attached to %s namespace of %d", ns_info
[i
].proc_name
, pid
);
246 static int lxc_attach_remount_sys_proc(void)
250 ret
= unshare(CLONE_NEWNS
);
252 SYSERROR("Failed to unshare mount namespace");
256 if (detect_shared_rootfs()) {
257 if (mount(NULL
, "/", NULL
, MS_SLAVE
| MS_REC
, NULL
)) {
258 SYSERROR("Failed to make / rslave");
259 ERROR("Continuing...");
263 /* Assume /proc is always mounted, so remount it. */
264 ret
= umount2("/proc", MNT_DETACH
);
266 SYSERROR("Failed to unmount /proc");
270 ret
= mount("none", "/proc", "proc", 0, NULL
);
272 SYSERROR("Failed to remount /proc");
276 /* Try to umount /sys. If it's not a mount point, we'll get EINVAL, then
277 * we ignore it because it may not have been mounted in the first place.
279 ret
= umount2("/sys", MNT_DETACH
);
280 if (ret
< 0 && errno
!= EINVAL
) {
281 SYSERROR("Failed to unmount /sys");
283 } else if (ret
== 0) {
285 ret
= mount("none", "/sys", "sysfs", 0, NULL
);
287 SYSERROR("Failed to remount /sys");
295 static int lxc_attach_drop_privs(struct lxc_proc_context_info
*ctx
)
299 last_cap
= lxc_caps_last_cap();
300 for (cap
= 0; cap
<= last_cap
; cap
++) {
301 if (ctx
->capability_mask
& (1LL << cap
))
304 if (prctl(PR_CAPBSET_DROP
, prctl_arg(cap
), prctl_arg(0),
305 prctl_arg(0), prctl_arg(0))) {
306 SYSERROR("Failed to drop capability %d", cap
);
310 TRACE("Dropped capability %d", cap
);
316 static int lxc_attach_set_environment(struct lxc_proc_context_info
*init_ctx
,
317 enum lxc_attach_env_policy_t policy
,
318 char **extra_env
, char **extra_keep
)
321 struct lxc_list
*iterator
;
323 if (policy
== LXC_ATTACH_CLEAR_ENV
) {
325 char **extra_keep_store
= NULL
;
330 for (count
= 0; extra_keep
[count
]; count
++)
333 extra_keep_store
= calloc(count
, sizeof(char *));
334 if (!extra_keep_store
)
337 for (i
= 0; i
< count
; i
++) {
338 char *v
= getenv(extra_keep
[i
]);
340 extra_keep_store
[i
] = strdup(v
);
341 if (!extra_keep_store
[i
]) {
343 free(extra_keep_store
[--i
]);
345 free(extra_keep_store
);
349 if (strcmp(extra_keep
[i
], "PATH") == 0)
356 if (extra_keep_store
) {
359 for (p
= extra_keep_store
; *p
; p
++)
362 free(extra_keep_store
);
365 ERROR("Failed to clear environment");
369 if (extra_keep_store
) {
372 for (i
= 0; extra_keep
[i
]; i
++) {
373 if (extra_keep_store
[i
]) {
374 ret
= setenv(extra_keep
[i
], extra_keep_store
[i
], 1);
376 SYSWARN("Failed to set environment variable");
379 free(extra_keep_store
[i
]);
382 free(extra_keep_store
);
385 /* Always set a default path; shells and execlp tend to be fine
386 * without it, but there is a disturbing number of C programs
387 * out there that just assume that getenv("PATH") is never NULL
388 * and then die a painful segfault death.
391 ret
= setenv("PATH", "/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin", 1);
393 SYSWARN("Failed to set environment variable");
397 ret
= putenv("container=lxc");
399 SYSWARN("Failed to set environment variable");
403 /* Set container environment variables.*/
404 if (init_ctx
&& init_ctx
->container
&& init_ctx
->container
->lxc_conf
) {
405 lxc_list_for_each(iterator
, &init_ctx
->container
->lxc_conf
->environment
) {
408 env_tmp
= strdup((char *)iterator
->elem
);
412 ret
= putenv(env_tmp
);
414 SYSERROR("Failed to set environment variable: %s", (char *)iterator
->elem
);
420 /* Set extra environment variables. */
422 for (; *extra_env
; extra_env
++) {
425 /* We just assume the user knows what they are doing, so
426 * we don't do any checks.
428 p
= strdup(*extra_env
);
434 SYSWARN("Failed to set environment variable");
441 static char *lxc_attach_getpwshell(uid_t uid
)
448 size_t line_bufsz
= 0;
449 char *line
= NULL
, *result
= NULL
;
451 /* We need to fork off a process that runs the getent program, and we
452 * need to capture its output, so we use a pipe for that purpose.
454 ret
= pipe2(pipes
, O_CLOEXEC
);
467 char *arguments
[] = {
476 /* We want to capture stdout. */
477 ret
= dup2(pipes
[1], STDOUT_FILENO
);
482 /* Get rid of stdin/stderr, so we try to associate it with
488 close(STDERR_FILENO
);
490 (void)dup3(fd
, STDIN_FILENO
, O_CLOEXEC
);
491 (void)dup3(fd
, STDOUT_FILENO
, O_CLOEXEC
);
495 /* Finish argument list. */
496 ret
= snprintf(uid_buf
, sizeof(uid_buf
), "%ld", (long)uid
);
497 if (ret
<= 0 || ret
>= sizeof(uid_buf
))
500 /* Try to run getent program. */
501 (void)execvp("getent", arguments
);
507 pipe_f
= fdopen(pipes
[0], "r");
508 while (getline(&line
, &line_bufsz
, pipe_f
) != -1) {
512 char *endptr
= NULL
, *saveptr
= NULL
;
514 /* If we already found something, just continue to read
515 * until the pipe doesn't deliver any more data, but
516 * don't modify the existing data structure.
524 /* Trim line on the right hand side. */
525 for (i
= strlen(line
); i
> 0 && (line
[i
- 1] == '\n' || line
[i
- 1] == '\r'); --i
)
528 /* Split into tokens: first: user name. */
529 token
= strtok_r(line
, ":", &saveptr
);
533 /* next: dummy password field */
534 token
= strtok_r(NULL
, ":", &saveptr
);
539 token
= strtok_r(NULL
, ":", &saveptr
);
540 value
= token
? strtol(token
, &endptr
, 10) : 0;
541 if (!token
|| !endptr
|| *endptr
|| value
== LONG_MIN
||
545 /* dummy sanity check: user id matches */
546 if ((uid_t
)value
!= uid
)
549 /* skip fields: gid, gecos, dir, go to next field 'shell' */
550 for (i
= 0; i
< 4; i
++) {
551 token
= strtok_r(NULL
, ":", &saveptr
);
560 result
= strdup(token
);
562 /* Sanity check that there are no fields after that. */
563 token
= strtok_r(NULL
, ":", &saveptr
);
573 ret
= wait_for_pid(pid
);
587 static void lxc_attach_get_init_uidgid(uid_t
*init_uid
, gid_t
*init_gid
)
590 char proc_fn
[LXC_PROC_STATUS_LEN
];
593 size_t line_bufsz
= 0;
595 uid_t uid
= (uid_t
)-1;
596 gid_t gid
= (gid_t
)-1;
598 ret
= snprintf(proc_fn
, LXC_PROC_STATUS_LEN
, "/proc/%d/status", 1);
599 if (ret
< 0 || ret
>= LXC_PROC_STATUS_LEN
)
602 proc_file
= fopen(proc_fn
, "r");
606 while (getline(&line
, &line_bufsz
, proc_file
) != -1) {
607 /* Format is: real, effective, saved set user, fs we only care
610 ret
= sscanf(line
, "Uid: %ld", &value
);
611 if (ret
!= EOF
&& ret
== 1) {
614 ret
= sscanf(line
, "Gid: %ld", &value
);
615 if (ret
!= EOF
&& ret
== 1)
619 if (uid
!= (uid_t
)-1 && gid
!= (gid_t
)-1)
626 /* Only override arguments if we found something. */
627 if (uid
!= (uid_t
)-1)
630 if (gid
!= (gid_t
)-1)
633 /* TODO: we should also parse supplementary groups and use
634 * setgroups() to set them.
638 static bool fetch_seccomp(struct lxc_container
*c
, lxc_attach_options_t
*options
)
644 if (!(options
->namespaces
& CLONE_NEWNS
) ||
645 !(options
->attach_flags
& LXC_ATTACH_LSM
)) {
646 free(c
->lxc_conf
->seccomp
);
647 c
->lxc_conf
->seccomp
= NULL
;
651 /* Remove current setting. */
652 if (!c
->set_config_item(c
, "lxc.seccomp.profile", "") &&
653 !c
->set_config_item(c
, "lxc.seccomp", ""))
656 /* Fetch the current profile path over the cmd interface. */
657 path
= c
->get_running_config_item(c
, "lxc.seccomp.profile");
659 INFO("Failed to retrieve lxc.seccomp.profile");
661 path
= c
->get_running_config_item(c
, "lxc.seccomp");
663 INFO("Failed to retrieve lxc.seccomp");
668 /* Copy the value into the new lxc_conf. */
669 bret
= c
->set_config_item(c
, "lxc.seccomp.profile", path
);
674 /* Attempt to parse the resulting config. */
675 ret
= lxc_read_seccomp_config(c
->lxc_conf
);
677 ERROR("Failed to retrieve seccomp policy");
681 INFO("Retrieved seccomp policy");
685 static bool no_new_privs(struct lxc_container
*c
, lxc_attach_options_t
*options
)
690 /* Remove current setting. */
691 if (!c
->set_config_item(c
, "lxc.no_new_privs", "")) {
692 INFO("Failed to unset lxc.no_new_privs");
696 /* Retrieve currently active setting. */
697 val
= c
->get_running_config_item(c
, "lxc.no_new_privs");
699 INFO("Failed to retrieve lxc.no_new_privs");
703 /* Set currently active setting. */
704 bret
= c
->set_config_item(c
, "lxc.no_new_privs", val
);
709 static signed long get_personality(const char *name
, const char *lxcpath
)
714 p
= lxc_cmd_get_config_item(name
, "lxc.arch", lxcpath
);
718 ret
= lxc_config_parse_arch(p
);
724 struct attach_clone_payload
{
726 int terminal_slave_fd
;
727 lxc_attach_options_t
*options
;
728 struct lxc_proc_context_info
*init_ctx
;
729 lxc_attach_exec_t exec_function
;
733 static void lxc_put_attach_clone_payload(struct attach_clone_payload
*p
)
735 if (p
->ipc_socket
>= 0) {
736 close(p
->ipc_socket
);
737 p
->ipc_socket
= -EBADF
;
740 if (p
->terminal_slave_fd
>= 0) {
741 close(p
->terminal_slave_fd
);
742 p
->terminal_slave_fd
= -EBADF
;
746 lxc_proc_put_context_info(p
->init_ctx
);
751 static int attach_child_main(struct attach_clone_payload
*payload
)
756 uid_t ns_root_uid
= 0;
757 gid_t ns_root_gid
= 0;
758 lxc_attach_options_t
* options
= payload
->options
;
759 struct lxc_proc_context_info
* init_ctx
= payload
->init_ctx
;
760 bool needs_lsm
= (options
->namespaces
& CLONE_NEWNS
) &&
761 (options
->attach_flags
& LXC_ATTACH_LSM
) &&
764 /* A description of the purpose of this functionality is provided in the
765 * lxc-attach(1) manual page. We have to remount here and not in the
766 * parent process, otherwise /proc may not properly reflect the new pid
769 if (!(options
->namespaces
& CLONE_NEWNS
) &&
770 (options
->attach_flags
& LXC_ATTACH_REMOUNT_PROC_SYS
)) {
771 ret
= lxc_attach_remount_sys_proc();
775 TRACE("Remounted \"/proc\" and \"/sys\"");
778 /* Now perform additional attachments. */
779 #if HAVE_SYS_PERSONALITY_H
780 if (options
->attach_flags
& LXC_ATTACH_SET_PERSONALITY
) {
781 long new_personality
;
783 if (options
->personality
< 0)
784 new_personality
= init_ctx
->personality
;
786 new_personality
= options
->personality
;
788 ret
= personality(new_personality
);
792 TRACE("Set new personality");
796 if (options
->attach_flags
& LXC_ATTACH_DROP_CAPABILITIES
) {
797 ret
= lxc_attach_drop_privs(init_ctx
);
801 TRACE("Dropped capabilities");
804 /* Always set the environment (specify (LXC_ATTACH_KEEP_ENV, NULL, NULL)
805 * if you want this to be a no-op).
807 ret
= lxc_attach_set_environment(init_ctx
,
809 options
->extra_env_vars
,
810 options
->extra_keep_env
);
814 TRACE("Set up environment");
816 /* This remark only affects fully unprivileged containers:
817 * Receive fd for LSM security module before we set{g,u}id(). The reason
818 * is that on set{g,u}id() the kernel will a) make us undumpable and b)
819 * we will change our effective uid. This means our effective uid will
820 * be different from the effective uid of the process that created us
821 * which means that this processs no longer has capabilities in our
822 * namespace including CAP_SYS_PTRACE. This means we will not be able to
823 * read and /proc/<pid> files for the process anymore when /proc is
824 * mounted with hidepid={1,2}. So let's get the lsm label fd before the
828 ret
= lxc_abstract_unix_recv_fds(payload
->ipc_socket
, &lsm_fd
, 1, NULL
, 0);
831 SYSERROR("Failed to receive lsm label fd");
836 TRACE("Received LSM label file descriptor %d from parent", lsm_fd
);
839 if (options
->stdin_fd
> 0 && isatty(options
->stdin_fd
)) {
840 ret
= lxc_make_controlling_terminal(options
->stdin_fd
);
845 if (options
->namespaces
& CLONE_NEWUSER
) {
846 /* Check whether nsuid 0 has a mapping. */
847 ns_root_uid
= get_ns_uid(0);
849 /* Check whether nsgid 0 has a mapping. */
850 ns_root_gid
= get_ns_gid(0);
852 /* If there's no mapping for nsuid 0 try to retrieve the nsuid
853 * init was started with.
855 if (ns_root_uid
== LXC_INVALID_UID
)
856 lxc_attach_get_init_uidgid(&ns_root_uid
, &ns_root_gid
);
858 if (ns_root_uid
== LXC_INVALID_UID
)
861 if (!lxc_switch_uid_gid(ns_root_uid
, ns_root_gid
))
865 if (!lxc_setgroups(0, NULL
) && errno
!= EPERM
)
869 if (options
->uid
!= LXC_INVALID_UID
)
870 new_uid
= options
->uid
;
872 new_uid
= ns_root_uid
;
874 if (options
->gid
!= LXC_INVALID_GID
)
875 new_gid
= options
->gid
;
877 new_gid
= ns_root_gid
;
879 if ((init_ctx
->container
&& init_ctx
->container
->lxc_conf
&&
880 init_ctx
->container
->lxc_conf
->no_new_privs
) ||
881 (options
->attach_flags
& LXC_ATTACH_NO_NEW_PRIVS
)) {
882 ret
= prctl(PR_SET_NO_NEW_PRIVS
, prctl_arg(1), prctl_arg(0),
883 prctl_arg(0), prctl_arg(0));
887 TRACE("Set PR_SET_NO_NEW_PRIVS");
893 /* Change into our new LSM profile. */
894 on_exec
= options
->attach_flags
& LXC_ATTACH_LSM_EXEC
? true : false;
896 ret
= lsm_process_label_set_at(lsm_fd
, init_ctx
->lsm_label
, on_exec
);
901 TRACE("Set %s LSM label to \"%s\"", lsm_name(), init_ctx
->lsm_label
);
904 if (init_ctx
->container
&& init_ctx
->container
->lxc_conf
&&
905 init_ctx
->container
->lxc_conf
->seccomp
) {
906 ret
= lxc_seccomp_load(init_ctx
->container
->lxc_conf
);
910 TRACE("Loaded seccomp profile");
913 close(payload
->ipc_socket
);
914 payload
->ipc_socket
= -EBADF
;
915 lxc_proc_put_context_info(init_ctx
);
916 payload
->init_ctx
= NULL
;
918 /* The following is done after the communication socket is shut down.
919 * That way, all errors that might (though unlikely) occur up until this
920 * point will have their messages printed to the original stderr (if
921 * logging is so configured) and not the fd the user supplied, if any.
924 /* Fd handling for stdin, stdout and stderr; ignore errors here, user
925 * may want to make sure the fds are closed, for example.
927 if (options
->stdin_fd
>= 0 && options
->stdin_fd
!= STDIN_FILENO
)
928 (void)dup2(options
->stdin_fd
, STDIN_FILENO
);
930 if (options
->stdout_fd
>= 0 && options
->stdout_fd
!= STDOUT_FILENO
)
931 (void)dup2(options
->stdout_fd
, STDOUT_FILENO
);
933 if (options
->stderr_fd
>= 0 && options
->stderr_fd
!= STDERR_FILENO
)
934 (void)dup2(options
->stderr_fd
, STDERR_FILENO
);
936 /* close the old fds */
937 if (options
->stdin_fd
> STDERR_FILENO
)
938 close(options
->stdin_fd
);
940 if (options
->stdout_fd
> STDERR_FILENO
)
941 close(options
->stdout_fd
);
943 if (options
->stderr_fd
> STDERR_FILENO
)
944 close(options
->stderr_fd
);
946 /* Try to remove FD_CLOEXEC flag from stdin/stdout/stderr, but also
947 * here, ignore errors.
949 for (fd
= STDIN_FILENO
; fd
<= STDERR_FILENO
; fd
++) {
950 ret
= fd_cloexec(fd
, false);
952 SYSERROR("Failed to clear FD_CLOEXEC from file descriptor %d", fd
);
957 if (options
->attach_flags
& LXC_ATTACH_TERMINAL
) {
958 ret
= lxc_terminal_prepare_login(payload
->terminal_slave_fd
);
960 SYSERROR("Failed to prepare terminal file descriptor %d", payload
->terminal_slave_fd
);
964 TRACE("Prepared terminal file descriptor %d", payload
->terminal_slave_fd
);
967 /* Avoid unnecessary syscalls. */
968 if (new_uid
== ns_root_uid
)
969 new_uid
= LXC_INVALID_UID
;
971 if (new_gid
== ns_root_gid
)
972 new_gid
= LXC_INVALID_GID
;
974 if (!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, potentially 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 if (options
->attach_flags
& LXC_ATTACH_TERMINAL
) {
1488 ret
= pthread_sigmask(SIG_SETMASK
,
1489 &terminal
.tty_state
->oldmask
, NULL
);
1491 SYSERROR("Failed to reset signal mask");
1492 _exit(EXIT_FAILURE
);
1496 ret
= attach_child_main(&payload
);
1498 ERROR("Failed to exec");
1500 _exit(EXIT_FAILURE
);
1503 if (options
->attach_flags
& LXC_ATTACH_TERMINAL
)
1504 lxc_attach_terminal_close_slave(&terminal
);
1506 /* Tell grandparent the pid of the pid of the newly created child. */
1507 ret
= lxc_write_nointr(ipc_sockets
[1], &pid
, sizeof(pid
));
1508 if (ret
!= sizeof(pid
)) {
1509 /* If this really happens here, this is very unfortunate, since
1510 * the parent will not know the pid of the attached process and
1511 * will not be able to wait for it (and we won't either due to
1512 * CLONE_PARENT) so the parent won't be able to reap it and the
1513 * attached process will remain a zombie.
1515 shutdown(ipc_sockets
[1], SHUT_RDWR
);
1516 lxc_proc_put_context_info(init_ctx
);
1517 _exit(EXIT_FAILURE
);
1520 TRACE("Sending pid %d of attached process", pid
);
1522 /* The rest is in the hands of the initial and the attached process. */
1523 lxc_proc_put_context_info(init_ctx
);
1527 int lxc_attach_run_command(void *payload
)
1530 lxc_attach_command_t
*cmd
= payload
;
1532 ret
= execvp(cmd
->program
, cmd
->argv
);
1544 SYSERROR("Failed to exec \"%s\"", cmd
->program
);
1548 int lxc_attach_run_shell(void* payload
)
1551 struct passwd pwent
;
1552 struct passwd
*pwentp
= NULL
;
1558 /* Ignore payload parameter. */
1563 bufsize
= sysconf(_SC_GETPW_R_SIZE_MAX
);
1567 buf
= malloc(bufsize
);
1569 ret
= getpwuid_r(uid
, &pwent
, buf
, bufsize
, &pwentp
);
1572 WARN("Could not find matched password record");
1574 WARN("Failed to get password record - %u", uid
);
1578 /* This probably happens because of incompatible nss implementations in
1579 * host and container (remember, this code is still using the host's
1580 * glibc but our mount namespace is in the container) we may try to get
1581 * the information by spawning a [getent passwd uid] process and parsing
1585 user_shell
= lxc_attach_getpwshell(uid
);
1587 user_shell
= pwent
.pw_shell
;
1590 execlp(user_shell
, user_shell
, (char *)NULL
);
1592 /* Executed if either no passwd entry or execvp fails, we will fall back
1593 * on /bin/sh as a default shell.
1595 execlp("/bin/sh", "/bin/sh", (char *)NULL
);
1597 SYSERROR("Failed to execute shell");