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
27 #define __STDC_FORMAT_MACROS /* Required for PRIu64 to work. */
41 #include <sys/mount.h>
42 #include <sys/param.h>
43 #include <sys/prctl.h>
45 #include <sys/types.h>
52 #include "namespace.h"
54 #include "raw_syscalls.h"
55 #include "syscall_wrappers.h"
59 #include "include/strlcpy.h"
63 #include "include/strlcat.h"
67 #define O_PATH 010000000
71 #define O_NOFOLLOW 00400000
74 lxc_log_define(utils
, lxc
);
77 * if path is btrfs, tries to remove it and any subvolumes beneath it
79 extern bool btrfs_try_remove_subvol(const char *path
);
81 static int _recursive_rmdir(const char *dirname
, dev_t pdev
,
82 const char *exclude
, int level
, bool onedev
)
84 struct dirent
*direntp
;
87 char pathname
[PATH_MAX
];
88 bool hadexclude
= false;
90 dir
= opendir(dirname
);
92 ERROR("failed to open %s", dirname
);
96 while ((direntp
= readdir(dir
))) {
100 if (!strcmp(direntp
->d_name
, ".") ||
101 !strcmp(direntp
->d_name
, ".."))
104 rc
= snprintf(pathname
, PATH_MAX
, "%s/%s", dirname
, direntp
->d_name
);
105 if (rc
< 0 || rc
>= PATH_MAX
) {
106 ERROR("pathname too long");
111 if (!level
&& exclude
&& !strcmp(direntp
->d_name
, exclude
)) {
112 ret
= rmdir(pathname
);
116 INFO("Not deleting snapshot %s", pathname
);
120 ret
= unlink(pathname
);
122 INFO("Failed to remove %s", pathname
);
125 SYSERROR("Failed to rmdir %s", pathname
);
133 ret
= lstat(pathname
, &mystat
);
135 ERROR("Failed to stat %s", pathname
);
140 if (onedev
&& mystat
.st_dev
!= pdev
) {
141 /* TODO should we be checking /proc/self/mountinfo for
142 * pathname and not doing this if found? */
143 if (btrfs_try_remove_subvol(pathname
))
144 INFO("Removed btrfs subvolume at %s\n", pathname
);
148 if (S_ISDIR(mystat
.st_mode
)) {
149 if (_recursive_rmdir(pathname
, pdev
, exclude
, level
+1, onedev
) < 0)
152 if (unlink(pathname
) < 0) {
153 SYSERROR("Failed to delete %s", pathname
);
159 if (rmdir(dirname
) < 0 && !btrfs_try_remove_subvol(dirname
) && !hadexclude
) {
160 ERROR("Failed to delete %s", dirname
);
166 ERROR("Failed to close directory %s", dirname
);
170 return failed
? -1 : 0;
173 /* In overlayfs, st_dev is unreliable. So on overlayfs we don't do the
176 static bool is_native_overlayfs(const char *path
)
178 if (has_fs_type(path
, OVERLAY_SUPER_MAGIC
) ||
179 has_fs_type(path
, OVERLAYFS_SUPER_MAGIC
))
185 /* returns 0 on success, -1 if there were any failures */
186 extern int lxc_rmdir_onedev(const char *path
, const char *exclude
)
191 if (is_native_overlayfs(path
))
194 if (lstat(path
, &mystat
) < 0) {
198 ERROR("Failed to stat %s", path
);
202 return _recursive_rmdir(path
, mystat
.st_dev
, exclude
, 0, onedev
);
205 /* borrowed from iproute2 */
206 extern int get_u16(unsigned short *val
, const char *arg
, int base
)
215 res
= strtoul(arg
, &ptr
, base
);
216 if (!ptr
|| ptr
== arg
|| *ptr
|| res
> 0xFFFF || errno
!= 0)
224 int mkdir_p(const char *dir
, mode_t mode
)
226 const char *tmp
= dir
;
227 const char *orig
= dir
;
232 dir
= tmp
+ strspn(tmp
, "/");
233 tmp
= dir
+ strcspn(dir
, "/");
236 makeme
= strndup(orig
, dir
- orig
);
240 ret
= mkdir(makeme
, mode
);
241 if (ret
< 0 && errno
!= EEXIST
) {
242 SYSERROR("Failed to create directory \"%s\"", makeme
);
248 } while (tmp
!= dir
);
259 if (stat(RUNTIME_PATH
, &sb
) < 0)
262 if (geteuid() == sb
.st_uid
|| getegid() == sb
.st_gid
) {
263 rundir
= strdup(RUNTIME_PATH
);
267 rundir
= getenv("XDG_RUNTIME_DIR");
269 rundir
= strdup(rundir
);
273 INFO("XDG_RUNTIME_DIR isn't set in the environment.");
274 homedir
= getenv("HOME");
276 ERROR("HOME isn't set in the environment.");
280 rundir
= malloc(sizeof(char) * (17 + strlen(homedir
)));
284 sprintf(rundir
, "%s/.cache/lxc/run/", homedir
);
289 int wait_for_pid(pid_t pid
)
294 ret
= waitpid(pid
, &status
, 0);
305 if (!WIFEXITED(status
) || WEXITSTATUS(status
) != 0)
311 int lxc_wait_for_pid_status(pid_t pid
)
316 ret
= waitpid(pid
, &status
, 0);
331 #include <gnutls/gnutls.h>
332 #include <gnutls/crypto.h>
334 __attribute__((constructor
))
335 static void gnutls_lxc_init(void)
337 gnutls_global_init();
340 int sha1sum_file(char *fnam
, unsigned char *digest
)
350 f
= fopen_cloexec(fnam
, "r");
352 SYSERROR("Error opening template");
356 if (fseek(f
, 0, SEEK_END
) < 0) {
357 SYSERROR("Error seeking to end of template");
362 if ((flen
= ftell(f
)) < 0) {
363 SYSERROR("Error telling size of template");
368 if (fseek(f
, 0, SEEK_SET
) < 0) {
369 SYSERROR("Error seeking to start of template");
374 if ((buf
= malloc(flen
+1)) == NULL
) {
375 SYSERROR("Out of memory");
380 if (fread(buf
, 1, flen
, f
) != flen
) {
381 SYSERROR("Failure reading template");
388 SYSERROR("Failre closing template");
394 ret
= gnutls_hash_fast(GNUTLS_DIG_SHA1
, buf
, flen
, (void *)digest
);
400 struct lxc_popen_FILE
*lxc_popen(const char *command
)
405 struct lxc_popen_FILE
*fp
= NULL
;
407 ret
= pipe2(pipe_fds
, O_CLOEXEC
);
420 /* duplicate stdout */
421 if (pipe_fds
[1] != STDOUT_FILENO
)
422 ret
= dup2(pipe_fds
[1], STDOUT_FILENO
);
424 ret
= fcntl(pipe_fds
[1], F_SETFD
, 0);
430 /* duplicate stderr */
431 if (pipe_fds
[1] != STDERR_FILENO
)
432 ret
= dup2(pipe_fds
[1], STDERR_FILENO
);
434 ret
= fcntl(pipe_fds
[1], F_SETFD
, 0);
439 /* unblock all signals */
440 ret
= sigfillset(&mask
);
444 ret
= pthread_sigmask(SIG_UNBLOCK
, &mask
, NULL
);
448 execl("/bin/sh", "sh", "-c", command
, (char *)NULL
);
455 fp
= malloc(sizeof(*fp
));
459 memset(fp
, 0, sizeof(*fp
));
461 fp
->child_pid
= child_pid
;
462 fp
->pipe
= pipe_fds
[0];
464 /* From now on, closing fp->f will also close fp->pipe. So only ever
465 * call fclose(fp->f).
467 fp
->f
= fdopen(pipe_fds
[0], "r");
474 /* We can only close pipe_fds[0] if fdopen() didn't succeed or wasn't
475 * called yet. Otherwise the fd belongs to the file opened by fdopen()
476 * since it isn't dup()ed.
478 if (fp
&& !fp
->f
&& pipe_fds
[0] >= 0)
481 if (pipe_fds
[1] >= 0)
493 int lxc_pclose(struct lxc_popen_FILE
*fp
)
502 wait_pid
= waitpid(fp
->child_pid
, &wstatus
, 0);
503 } while (wait_pid
< 0 && errno
== EINTR
);
514 int randseed(bool srand_it
)
517 srand pre-seed function based on /dev/urandom
519 unsigned int seed
= time(NULL
) + getpid();
522 f
= fopen("/dev/urandom", "r");
524 int ret
= fread(&seed
, sizeof(seed
), 1, f
);
526 SYSDEBUG("unable to fread /dev/urandom, fallback to time+pid rand seed");
537 uid_t
get_ns_uid(uid_t orig
)
541 uid_t nsid
, hostid
, range
;
542 FILE *f
= fopen("/proc/self/uid_map", "r");
546 while (getline(&line
, &sz
, f
) != -1) {
547 if (sscanf(line
, "%u %u %u", &nsid
, &hostid
, &range
) != 3)
550 if (hostid
<= orig
&& hostid
+ range
> orig
) {
551 nsid
+= orig
- hostid
;
556 nsid
= LXC_INVALID_UID
;
564 gid_t
get_ns_gid(gid_t orig
)
568 gid_t nsid
, hostid
, range
;
569 FILE *f
= fopen("/proc/self/gid_map", "r");
573 while (getline(&line
, &sz
, f
) != -1) {
574 if (sscanf(line
, "%u %u %u", &nsid
, &hostid
, &range
) != 3)
577 if (hostid
<= orig
&& hostid
+ range
> orig
) {
578 nsid
+= orig
- hostid
;
583 nsid
= LXC_INVALID_GID
;
591 bool dir_exists(const char *path
)
596 ret
= stat(path
, &sb
);
598 /* Could be something other than eexist, just say "no". */
601 return S_ISDIR(sb
.st_mode
);
604 /* Note we don't use SHA-1 here as we don't want to depend on HAVE_GNUTLS.
605 * FNV has good anti collision properties and we're not worried
606 * about pre-image resistance or one-way-ness, we're just trying to make
607 * the name unique in the 108 bytes of space we have.
609 uint64_t fnv_64a_buf(void *buf
, size_t len
, uint64_t hval
)
613 for(bp
= buf
; bp
< (unsigned char *)buf
+ len
; bp
++)
615 /* xor the bottom with the current octet */
616 hval
^= (uint64_t)*bp
;
619 * multiply by the 64 bit FNV magic prime mod 2^64
621 hval
+= (hval
<< 1) + (hval
<< 4) + (hval
<< 5) +
622 (hval
<< 7) + (hval
<< 8) + (hval
<< 40);
628 bool is_shared_mountpoint(const char *path
)
630 char buf
[LXC_LINELEN
];
635 f
= fopen("/proc/self/mountinfo", "r");
639 while (fgets(buf
, LXC_LINELEN
, f
)) {
640 for (p
= buf
, i
= 0; p
&& i
< 4; i
++)
641 p
= strchr(p
+ 1, ' ');
645 p2
= strchr(p
+ 1, ' ');
650 if (strcmp(p
+ 1, path
) == 0) {
651 /* This is the path. Is it shared? */
652 p
= strchr(p2
+ 1, ' ');
653 if (p
&& strstr(p
, "shared:")) {
665 * Detect whether / is mounted MS_SHARED. The only way I know of to
666 * check that is through /proc/self/mountinfo.
667 * I'm only checking for /. If the container rootfs or mount location
668 * is MS_SHARED, but not '/', then you're out of luck - figuring that
669 * out would be too much work to be worth it.
671 int detect_shared_rootfs(void)
673 if (is_shared_mountpoint("/"))
678 bool switch_to_ns(pid_t pid
, const char *ns
)
681 char nspath
[PATH_MAX
];
683 /* Switch to new ns */
684 ret
= snprintf(nspath
, PATH_MAX
, "/proc/%d/ns/%s", pid
, ns
);
685 if (ret
< 0 || ret
>= PATH_MAX
)
688 fd
= open(nspath
, O_RDONLY
);
690 SYSERROR("Failed to open %s", nspath
);
696 SYSERROR("Failed to set process %d to %s of %d.", pid
, ns
, fd
);
706 * looking at fs/proc_namespace.c, it appears we can
707 * actually expect the rootfs entry to very specifically contain
708 * " - rootfs rootfs "
709 * IIUC, so long as we've chrooted so that rootfs is not our root,
710 * the rootfs entry should always be skipped in mountinfo contents.
712 bool detect_ramfs_rootfs(void)
720 f
= fopen("/proc/self/mountinfo", "r");
724 while (getline(&line
, &len
, f
) != -1) {
725 for (p
= line
, i
= 0; p
&& i
< 4; i
++)
726 p
= strchr(p
+ 1, ' ');
730 p2
= strchr(p
+ 1, ' ');
735 if (strcmp(p
+ 1, "/") == 0) {
736 /* This is '/'. Is it the ramfs? */
737 p
= strchr(p2
+ 1, '-');
738 if (p
&& strncmp(p
, "- rootfs rootfs ", 16) == 0) {
741 INFO("Rootfs is located on ramfs");
752 char *on_path(const char *cmd
, const char *rootfs
)
754 char *entry
= NULL
, *path
= NULL
;
755 char cmdpath
[PATH_MAX
];
758 path
= getenv("PATH");
766 lxc_iterate_parts (entry
, path
, ":") {
768 ret
= snprintf(cmdpath
, PATH_MAX
, "%s/%s/%s", rootfs
,
771 ret
= snprintf(cmdpath
, PATH_MAX
, "%s/%s", entry
, cmd
);
772 if (ret
< 0 || ret
>= PATH_MAX
)
775 if (access(cmdpath
, X_OK
) == 0) {
777 return strdup(cmdpath
);
785 bool cgns_supported(void)
787 return file_exists("/proc/self/ns/cgroup");
790 /* historically lxc-init has been under /usr/lib/lxc and under
791 * /usr/lib/$ARCH/lxc. It now lives as $prefix/sbin/init.lxc.
793 char *choose_init(const char *rootfs
)
796 const char *empty
= "",
798 int ret
, env_set
= 0;
800 if (!getenv("PATH")) {
801 if (setenv("PATH", "/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin", 0))
802 SYSERROR("Failed to setenv");
807 retv
= on_path("init.lxc", rootfs
);
810 if (unsetenv("PATH"))
811 SYSERROR("Failed to unsetenv");
817 retv
= malloc(PATH_MAX
);
826 ret
= snprintf(retv
, PATH_MAX
, "%s/%s/%s", tmp
, SBINDIR
, "/init.lxc");
827 if (ret
< 0 || ret
>= PATH_MAX
) {
828 ERROR("pathname too long");
832 if (access(retv
, X_OK
) == 0)
835 ret
= snprintf(retv
, PATH_MAX
, "%s/%s/%s", tmp
, LXCINITDIR
, "/lxc/lxc-init");
836 if (ret
< 0 || ret
>= PATH_MAX
) {
837 ERROR("pathname too long");
841 if (access(retv
, X_OK
) == 0)
844 ret
= snprintf(retv
, PATH_MAX
, "%s/usr/lib/lxc/lxc-init", tmp
);
845 if (ret
< 0 || ret
>= PATH_MAX
) {
846 ERROR("pathname too long");
850 if (access(retv
, X_OK
) == 0)
853 ret
= snprintf(retv
, PATH_MAX
, "%s/sbin/lxc-init", tmp
);
854 if (ret
< 0 || ret
>= PATH_MAX
) {
855 ERROR("pathname too long");
859 if (access(retv
, X_OK
) == 0)
863 * Last resort, look for the statically compiled init.lxc which we
864 * hopefully bind-mounted in.
865 * If we are called during container setup, and we get to this point,
866 * then the init.lxc.static from the host will need to be bind-mounted
867 * in. So we return NULL here to indicate that.
872 ret
= snprintf(retv
, PATH_MAX
, "/init.lxc.static");
873 if (ret
< 0 || ret
>= PATH_MAX
) {
874 WARN("Nonsense - name /lxc.init.static too long");
878 if (access(retv
, X_OK
) == 0)
887 * Given the '-t' template option to lxc-create, figure out what to
888 * do. If the template is a full executable path, use that. If it
889 * is something like 'sshd', then return $templatepath/lxc-sshd.
890 * On success return the template, on error return NULL.
892 char *get_template_path(const char *t
)
897 if (t
[0] == '/' && access(t
, X_OK
) == 0) {
902 len
= strlen(LXCTEMPLATEDIR
) + strlen(t
) + strlen("/lxc-") + 1;
908 ret
= snprintf(tpath
, len
, "%s/lxc-%s", LXCTEMPLATEDIR
, t
);
909 if (ret
< 0 || ret
>= len
) {
914 if (access(tpath
, X_OK
) < 0) {
915 SYSERROR("bad template: %s", t
);
924 * @path: a pathname where / replaced with '\0'.
925 * @offsetp: pointer to int showing which path segment was last seen.
926 * Updated on return to reflect the next segment.
927 * @fulllen: full original path length.
928 * Returns a pointer to the next path segment, or NULL if done.
930 static char *get_nextpath(char *path
, int *offsetp
, int fulllen
)
932 int offset
= *offsetp
;
934 if (offset
>= fulllen
)
937 while (offset
< fulllen
&& path
[offset
] != '\0')
940 while (offset
< fulllen
&& path
[offset
] == '\0')
944 return (offset
< fulllen
) ? &path
[offset
] : NULL
;
948 * Check that @subdir is a subdir of @dir. @len is the length of
949 * @dir (to avoid having to recalculate it).
951 static bool is_subdir(const char *subdir
, const char *dir
, size_t len
)
953 size_t subdirlen
= strlen(subdir
);
958 if (strncmp(subdir
, dir
, len
) != 0)
961 if (dir
[len
-1] == '/')
964 if (subdir
[len
] == '/' || subdirlen
== len
)
971 * Check if the open fd is a symlink. Return -ELOOP if it is. Return
972 * -ENOENT if we couldn't fstat. Return 0 if the fd is ok.
974 static int check_symlink(int fd
)
979 ret
= fstat(fd
, &sb
);
983 if (S_ISLNK(sb
.st_mode
))
990 * Open a file or directory, provided that it contains no symlinks.
992 * CAVEAT: This function must not be used for other purposes than container
993 * setup before executing the container's init
995 static int open_if_safe(int dirfd
, const char *nextpath
)
997 int newfd
= openat(dirfd
, nextpath
, O_RDONLY
| O_NOFOLLOW
);
998 if (newfd
>= 0) /* Was not a symlink, all good. */
1004 if (errno
== EPERM
|| errno
== EACCES
) {
1005 /* We're not root (cause we got EPERM) so try opening with
1008 newfd
= openat(dirfd
, nextpath
, O_PATH
| O_NOFOLLOW
);
1010 /* O_PATH will return an fd for symlinks. We know
1011 * nextpath wasn't a symlink at last openat, so if fd is
1012 * now a link, then something * fishy is going on.
1014 int ret
= check_symlink(newfd
);
1026 * Open a path intending for mounting, ensuring that the final path
1027 * is inside the container's rootfs.
1029 * CAVEAT: This function must not be used for other purposes than container
1030 * setup before executing the container's init
1032 * @target: path to be opened
1033 * @prefix_skip: a part of @target in which to ignore symbolic links. This
1034 * would be the container's rootfs.
1036 * Return an open fd for the path, or <0 on error.
1038 static int open_without_symlink(const char *target
, const char *prefix_skip
)
1040 int curlen
= 0, dirfd
, fulllen
, i
;
1043 fulllen
= strlen(target
);
1045 /* make sure prefix-skip makes sense */
1046 if (prefix_skip
&& strlen(prefix_skip
) > 0) {
1047 curlen
= strlen(prefix_skip
);
1048 if (!is_subdir(target
, prefix_skip
, curlen
)) {
1049 ERROR("WHOA there - target '%s' didn't start with prefix '%s'",
1050 target
, prefix_skip
);
1055 * get_nextpath() expects the curlen argument to be
1056 * on a (turned into \0) / or before it, so decrement
1057 * curlen to make sure that happens
1066 /* Make a copy of target which we can hack up, and tokenize it */
1067 if ((dup
= strdup(target
)) == NULL
) {
1068 SYSERROR("Out of memory checking for symbolic link");
1072 for (i
= 0; i
< fulllen
; i
++) {
1077 dirfd
= open(prefix_skip
, O_RDONLY
);
1082 int newfd
, saved_errno
;
1085 if ((nextpath
= get_nextpath(dup
, &curlen
, fulllen
)) == NULL
)
1088 newfd
= open_if_safe(dirfd
, nextpath
);
1089 saved_errno
= errno
;
1094 errno
= saved_errno
;
1096 SYSERROR("%s in %s was a symbolic link!", nextpath
, target
);
1108 * Safely mount a path into a container, ensuring that the mount target
1109 * is under the container's @rootfs. (If @rootfs is NULL, then the container
1110 * uses the host's /)
1112 * CAVEAT: This function must not be used for other purposes than container
1113 * setup before executing the container's init
1115 int safe_mount(const char *src
, const char *dest
, const char *fstype
,
1116 unsigned long flags
, const void *data
, const char *rootfs
)
1118 int destfd
, ret
, saved_errno
;
1119 /* Only needs enough for /proc/self/fd/<fd>. */
1120 char srcbuf
[50], destbuf
[50];
1122 const char *mntsrc
= src
;
1127 /* todo - allow symlinks for relative paths if 'allowsymlinks' option is passed */
1128 if (flags
& MS_BIND
&& src
&& src
[0] != '/') {
1129 INFO("this is a relative bind mount");
1131 srcfd
= open_without_symlink(src
, NULL
);
1135 ret
= snprintf(srcbuf
, 50, "/proc/self/fd/%d", srcfd
);
1136 if (ret
< 0 || ret
> 50) {
1138 ERROR("Out of memory");
1144 destfd
= open_without_symlink(dest
, rootfs
);
1147 saved_errno
= errno
;
1149 errno
= saved_errno
;
1155 ret
= snprintf(destbuf
, 50, "/proc/self/fd/%d", destfd
);
1156 if (ret
< 0 || ret
> 50) {
1161 ERROR("Out of memory");
1165 ret
= mount(mntsrc
, destbuf
, fstype
, flags
, data
);
1166 saved_errno
= errno
;
1172 errno
= saved_errno
;
1173 SYSERROR("Failed to mount %s onto %s", src
? src
: "(null)", dest
);
1181 * Mount a proc under @rootfs if proc self points to a pid other than
1182 * my own. This is needed to have a known-good proc mount for setting
1183 * up LSMs both at container startup and attach.
1185 * @rootfs : the rootfs where proc should be mounted
1187 * Returns < 0 on failure, 0 if the correct proc was already mounted
1188 * and 1 if a new proc was mounted.
1190 * NOTE: not to be called from inside the container namespace!
1192 int lxc_mount_proc_if_needed(const char *rootfs
)
1194 char path
[PATH_MAX
];
1195 int link_to_pid
, linklen
, mypid
, ret
;
1196 char link
[INTTYPE_TO_STRLEN(pid_t
)] = {0};
1198 ret
= snprintf(path
, PATH_MAX
, "%s/proc/self", rootfs
);
1199 if (ret
< 0 || ret
>= PATH_MAX
) {
1200 SYSERROR("proc path name too long");
1204 linklen
= readlink(path
, link
, sizeof(link
));
1206 ret
= snprintf(path
, PATH_MAX
, "%s/proc", rootfs
);
1207 if (ret
< 0 || ret
>= PATH_MAX
) {
1208 SYSERROR("proc path name too long");
1212 /* /proc not mounted */
1214 if (mkdir(path
, 0755) && errno
!= EEXIST
)
1218 } else if (linklen
>= sizeof(link
)) {
1219 link
[linklen
- 1] = '\0';
1220 ERROR("readlink returned truncated content: \"%s\"", link
);
1224 mypid
= lxc_raw_getpid();
1225 INFO("I am %d, /proc/self points to \"%s\"", mypid
, link
);
1227 if (lxc_safe_int(link
, &link_to_pid
) < 0)
1230 /* correct procfs is already mounted */
1231 if (link_to_pid
== mypid
)
1234 ret
= umount2(path
, MNT_DETACH
);
1236 WARN("failed to umount \"%s\" with MNT_DETACH", path
);
1239 /* rootfs is NULL */
1240 if (!strcmp(rootfs
, ""))
1241 ret
= mount("proc", path
, "proc", 0, NULL
);
1243 ret
= safe_mount("proc", path
, "proc", 0, NULL
, rootfs
);
1247 INFO("mounted /proc in container for security transition");
1251 int open_devnull(void)
1253 int fd
= open("/dev/null", O_RDWR
);
1256 SYSERROR("Can't open /dev/null");
1261 int set_stdfds(int fd
)
1268 ret
= dup2(fd
, STDIN_FILENO
);
1272 ret
= dup2(fd
, STDOUT_FILENO
);
1276 ret
= dup2(fd
, STDERR_FILENO
);
1283 int null_stdfds(void)
1288 fd
= open_devnull();
1290 ret
= set_stdfds(fd
);
1297 /* Check whether a signal is blocked by a process. */
1298 /* /proc/pid-to-str/status\0 = (5 + 21 + 7 + 1) */
1299 #define __PROC_STATUS_LEN (6 + INTTYPE_TO_STRLEN(pid_t) + 7 + 1)
1300 bool task_blocks_signal(pid_t pid
, int signal
)
1303 char status
[__PROC_STATUS_LEN
];
1305 uint64_t sigblk
= 0, one
= 1;
1310 ret
= snprintf(status
, __PROC_STATUS_LEN
, "/proc/%d/status", pid
);
1311 if (ret
< 0 || ret
>= __PROC_STATUS_LEN
)
1314 f
= fopen(status
, "r");
1318 while (getline(&line
, &n
, f
) != -1) {
1321 if (strncmp(line
, "SigBlk:", 7))
1324 numstr
= lxc_trim_whitespace_in_place(line
+ 7);
1325 ret
= lxc_safe_uint64(numstr
, &sigblk
, 16);
1332 if (sigblk
& (one
<< (signal
- 1)))
1341 int lxc_preserve_ns(const int pid
, const char *ns
)
1344 /* 5 /proc + 21 /int_as_str + 3 /ns + 20 /NS_NAME + 1 \0 */
1345 #define __NS_PATH_LEN 50
1346 char path
[__NS_PATH_LEN
];
1348 /* This way we can use this function to also check whether namespaces
1349 * are supported by the kernel by passing in the NULL or the empty
1352 ret
= snprintf(path
, __NS_PATH_LEN
, "/proc/%d/ns%s%s", pid
,
1353 !ns
|| strcmp(ns
, "") == 0 ? "" : "/",
1354 !ns
|| strcmp(ns
, "") == 0 ? "" : ns
);
1355 if (ret
< 0 || (size_t)ret
>= __NS_PATH_LEN
) {
1360 return open(path
, O_RDONLY
| O_CLOEXEC
);
1363 bool lxc_switch_uid_gid(uid_t uid
, gid_t gid
)
1367 if (gid
!= LXC_INVALID_GID
) {
1370 SYSERROR("Failed to switch to gid %d", gid
);
1373 NOTICE("Switched to gid %d", gid
);
1376 if (uid
!= LXC_INVALID_UID
) {
1379 SYSERROR("Failed to switch to uid %d", uid
);
1382 NOTICE("Switched to uid %d", uid
);
1388 /* Simple covenience function which enables uniform logging. */
1389 bool lxc_setgroups(int size
, gid_t list
[])
1391 if (setgroups(size
, list
) < 0) {
1392 SYSERROR("Failed to setgroups()");
1395 NOTICE("Dropped additional groups");
1400 static int lxc_get_unused_loop_dev_legacy(char *loop_name
)
1403 struct loop_info64 lo64
;
1405 int dfd
= -1, fd
= -1, ret
= -1;
1407 dir
= opendir("/dev");
1411 while ((dp
= readdir(dir
))) {
1412 if (strncmp(dp
->d_name
, "loop", 4) != 0)
1419 fd
= openat(dfd
, dp
->d_name
, O_RDWR
);
1423 ret
= ioctl(fd
, LOOP_GET_STATUS64
, &lo64
);
1425 if (ioctl(fd
, LOOP_GET_STATUS64
, &lo64
) == 0 ||
1433 ret
= snprintf(loop_name
, LO_NAME_SIZE
, "/dev/%s", dp
->d_name
);
1434 if (ret
< 0 || ret
>= LO_NAME_SIZE
) {
1451 static int lxc_get_unused_loop_dev(char *name_loop
)
1454 int fd_ctl
= -1, fd_tmp
= -1;
1456 fd_ctl
= open("/dev/loop-control", O_RDWR
| O_CLOEXEC
);
1460 loop_nr
= ioctl(fd_ctl
, LOOP_CTL_GET_FREE
);
1464 ret
= snprintf(name_loop
, LO_NAME_SIZE
, "/dev/loop%d", loop_nr
);
1465 if (ret
< 0 || ret
>= LO_NAME_SIZE
)
1468 fd_tmp
= open(name_loop
, O_RDWR
| O_CLOEXEC
);
1477 int lxc_prepare_loop_dev(const char *source
, char *loop_dev
, int flags
)
1480 struct loop_info64 lo64
;
1481 int fd_img
= -1, fret
= -1, fd_loop
= -1;
1483 fd_loop
= lxc_get_unused_loop_dev(loop_dev
);
1485 if (fd_loop
== -ENODEV
)
1486 fd_loop
= lxc_get_unused_loop_dev_legacy(loop_dev
);
1491 fd_img
= open(source
, O_RDWR
| O_CLOEXEC
);
1495 ret
= ioctl(fd_loop
, LOOP_SET_FD
, fd_img
);
1499 memset(&lo64
, 0, sizeof(lo64
));
1500 lo64
.lo_flags
= flags
;
1502 ret
= ioctl(fd_loop
, LOOP_SET_STATUS64
, &lo64
);
1512 if (fret
< 0 && fd_loop
>= 0) {
1520 int lxc_unstack_mountpoint(const char *path
, bool lazy
)
1526 ret
= umount2(path
, lazy
? MNT_DETACH
: 0);
1528 /* We consider anything else than EINVAL deadly to prevent going
1529 * into an infinite loop. (The other alternative is constantly
1530 * parsing /proc/self/mountinfo which is yucky and probably
1533 if (errno
!= EINVAL
)
1536 /* Just stop counting when this happens. That'd just be so
1537 * stupid that we won't even bother trying to report back the
1538 * correct value anymore.
1540 if (umounts
!= INT_MAX
)
1543 /* We succeeded in umounting. Make sure that there's no other
1544 * mountpoint stacked underneath.
1552 int run_command(char *buf
, size_t buf_size
, int (*child_fn
)(void *), void *args
)
1555 int ret
, fret
, pipefd
[2];
1558 /* Make sure our callers do not receive uninitialized memory. */
1559 if (buf_size
> 0 && buf
)
1562 if (pipe(pipefd
) < 0) {
1563 SYSERROR("failed to create pipe");
1567 child
= lxc_raw_clone(0);
1571 SYSERROR("failed to create new process");
1576 /* Close the read-end of the pipe. */
1579 /* Redirect std{err,out} to write-end of the
1582 ret
= dup2(pipefd
[1], STDOUT_FILENO
);
1584 ret
= dup2(pipefd
[1], STDERR_FILENO
);
1586 /* Close the write-end of the pipe. */
1590 SYSERROR("failed to duplicate std{err,out} file descriptor");
1591 _exit(EXIT_FAILURE
);
1594 /* Does not return. */
1596 ERROR("failed to exec command");
1597 _exit(EXIT_FAILURE
);
1600 /* close the write-end of the pipe */
1603 if (buf
&& buf_size
> 0) {
1604 bytes
= lxc_read_nointr(pipefd
[0], buf
, buf_size
- 1);
1606 buf
[bytes
- 1] = '\0';
1609 fret
= wait_for_pid(child
);
1610 /* close the read-end of the pipe */
1616 bool lxc_nic_exists(char *nic
)
1618 #define __LXC_SYS_CLASS_NET_LEN 15 + IFNAMSIZ + 1
1619 char path
[__LXC_SYS_CLASS_NET_LEN
];
1623 if (!strcmp(nic
, "none"))
1626 ret
= snprintf(path
, __LXC_SYS_CLASS_NET_LEN
, "/sys/class/net/%s", nic
);
1627 if (ret
< 0 || (size_t)ret
>= __LXC_SYS_CLASS_NET_LEN
)
1630 ret
= stat(path
, &sb
);
1637 uint64_t lxc_find_next_power2(uint64_t n
)
1639 /* 0 is not valid input. We return 0 to the caller since 0 is not a
1640 * valid power of two.
1655 int lxc_set_death_signal(int signal
, pid_t parent
)
1660 ret
= prctl(PR_SET_PDEATHSIG
, prctl_arg(signal
), prctl_arg(0),
1661 prctl_arg(0), prctl_arg(0));
1663 /* Check whether we have been orphaned. */
1664 ppid
= (pid_t
)syscall(SYS_getppid
);
1665 if (ppid
!= parent
) {
1666 ret
= raise(SIGKILL
);
1672 SYSERROR("Failed to set PR_SET_PDEATHSIG to %d", signal
);
1679 int fd_cloexec(int fd
, bool cloexec
)
1683 oflags
= fcntl(fd
, F_GETFD
, 0);
1688 nflags
= oflags
| FD_CLOEXEC
;
1690 nflags
= oflags
& ~FD_CLOEXEC
;
1692 if (nflags
== oflags
)
1695 if (fcntl(fd
, F_SETFD
, nflags
) < 0)
1701 int recursive_destroy(char *dirname
)
1704 struct dirent
*direntp
;
1708 dir
= opendir(dirname
);
1712 while ((direntp
= readdir(dir
))) {
1716 if (!strcmp(direntp
->d_name
, ".") ||
1717 !strcmp(direntp
->d_name
, ".."))
1720 pathname
= must_make_path(dirname
, direntp
->d_name
, NULL
);
1722 ret
= lstat(pathname
, &mystat
);
1725 WARN("Failed to stat \"%s\"", pathname
);
1731 if (!S_ISDIR(mystat
.st_mode
))
1734 ret
= recursive_destroy(pathname
);
1742 ret
= rmdir(dirname
);
1745 SYSWARN("Failed to delete \"%s\"", dirname
);
1750 ret
= closedir(dir
);
1753 SYSWARN("Failed to delete \"%s\"", dirname
);
1761 int lxc_setup_keyring(void)
1763 key_serial_t keyring
;
1766 /* Try to allocate a new session keyring for the container to prevent
1767 * information leaks.
1769 keyring
= keyctl(KEYCTL_JOIN_SESSION_KEYRING
, prctl_arg(0),
1770 prctl_arg(0), prctl_arg(0), prctl_arg(0));
1774 DEBUG("The keyctl() syscall is not supported or blocked");
1779 DEBUG("Failed to access kernel keyring. Continuing...");
1782 SYSERROR("Failed to create kernel keyring");