1 /* SPDX-License-Identifier: LGPL-2.1+ */
4 #include <linux/magic.h>
6 #include "alloc-util.h"
9 #include "format-util.h"
13 #include "mount-util.h"
14 #include "mountpoint-util.h"
15 #include "nspawn-mount.h"
16 #include "parse-util.h"
17 #include "path-util.h"
20 #include "sort-util.h"
21 #include "stat-util.h"
22 #include "string-util.h"
24 #include "tmpfile-util.h"
25 #include "user-util.h"
27 CustomMount
* custom_mount_add(CustomMount
**l
, size_t *n
, CustomMountType t
) {
33 assert(t
< _CUSTOM_MOUNT_TYPE_MAX
);
35 c
= reallocarray(*l
, *n
+ 1, sizeof(CustomMount
));
43 *ret
= (CustomMount
) { .type
= t
};
48 void custom_mount_free_all(CustomMount
*l
, size_t n
) {
51 for (i
= 0; i
< n
; i
++) {
52 CustomMount
*m
= l
+ i
;
59 (void) rm_rf(m
->work_dir
, REMOVE_ROOT
|REMOVE_PHYSICAL
);
63 if (m
->rm_rf_tmpdir
) {
64 (void) rm_rf(m
->rm_rf_tmpdir
, REMOVE_ROOT
|REMOVE_PHYSICAL
);
65 free(m
->rm_rf_tmpdir
);
69 free(m
->type_argument
);
75 static int custom_mount_compare(const CustomMount
*a
, const CustomMount
*b
) {
78 r
= path_compare(a
->destination
, b
->destination
);
82 return CMP(a
->type
, b
->type
);
85 static bool source_path_is_valid(const char *p
) {
91 return path_is_absolute(p
);
94 static char *resolve_source_path(const char *dest
, const char *source
) {
100 return path_join(dest
, source
+ 1);
102 return strdup(source
);
105 static int allocate_temporary_source(CustomMount
*m
) {
108 assert(!m
->rm_rf_tmpdir
);
110 m
->rm_rf_tmpdir
= strdup("/var/tmp/nspawn-temp-XXXXXX");
111 if (!m
->rm_rf_tmpdir
)
114 if (!mkdtemp(m
->rm_rf_tmpdir
)) {
115 m
->rm_rf_tmpdir
= mfree(m
->rm_rf_tmpdir
);
116 return log_error_errno(errno
, "Failed to acquire temporary directory: %m");
119 m
->source
= path_join(m
->rm_rf_tmpdir
, "src");
123 if (mkdir(m
->source
, 0755) < 0)
124 return log_error_errno(errno
, "Failed to create %s: %m", m
->source
);
129 int custom_mount_prepare_all(const char *dest
, CustomMount
*l
, size_t n
) {
133 /* Prepare all custom mounts. This will make source we know all temporary directories. This is called in the
134 * parent process, so that we know the temporary directories to remove on exit before we fork off the
139 /* Order the custom mounts, and make sure we have a working directory */
140 typesafe_qsort(l
, n
, custom_mount_compare
);
142 for (i
= 0; i
< n
; i
++) {
143 CustomMount
*m
= l
+ i
;
145 /* /proc we mount in the inner child, i.e. when we acquired CLONE_NEWPID. All other mounts we mount
146 * already in the outer child, so that the mounts are already established before CLONE_NEWPID and in
147 * particular CLONE_NEWUSER. This also means any custom mounts below /proc also need to be mounted in
148 * the inner child, not the outer one. Determine this here. */
149 m
->in_userns
= path_startswith(m
->destination
, "/proc");
151 if (m
->type
== CUSTOM_MOUNT_BIND
) {
155 s
= resolve_source_path(dest
, m
->source
);
159 free_and_replace(m
->source
, s
);
161 /* No source specified? In that case, use a throw-away temporary directory in /var/tmp */
163 r
= allocate_temporary_source(m
);
169 if (m
->type
== CUSTOM_MOUNT_OVERLAY
) {
172 STRV_FOREACH(j
, m
->lower
) {
175 s
= resolve_source_path(dest
, *j
);
179 free_and_replace(*j
, s
);
185 s
= resolve_source_path(dest
, m
->source
);
189 free_and_replace(m
->source
, s
);
191 r
= allocate_temporary_source(m
);
199 s
= resolve_source_path(dest
, m
->work_dir
);
203 free_and_replace(m
->work_dir
, s
);
205 r
= tempfn_random(m
->source
, NULL
, &m
->work_dir
);
207 return log_error_errno(r
, "Failed to acquire working directory: %m");
210 (void) mkdir_label(m
->work_dir
, 0700);
217 int bind_mount_parse(CustomMount
**l
, size_t *n
, const char *s
, bool read_only
) {
218 _cleanup_free_
char *source
= NULL
, *destination
= NULL
, *opts
= NULL
;
226 r
= extract_many_words(&p
, ":", EXTRACT_DONT_COALESCE_SEPARATORS
, &source
, &destination
, NULL
);
232 destination
= strdup(source
[0] == '+' ? source
+1 : source
);
236 if (r
== 2 && !isempty(p
)) {
243 source
= mfree(source
);
244 else if (!source_path_is_valid(source
))
247 if (!path_is_absolute(destination
))
250 m
= custom_mount_add(l
, n
, CUSTOM_MOUNT_BIND
);
254 m
->source
= TAKE_PTR(source
);
255 m
->destination
= TAKE_PTR(destination
);
256 m
->read_only
= read_only
;
257 m
->options
= TAKE_PTR(opts
);
262 int tmpfs_mount_parse(CustomMount
**l
, size_t *n
, const char *s
) {
263 _cleanup_free_
char *path
= NULL
, *opts
= NULL
;
272 r
= extract_first_word(&p
, &path
, ":", EXTRACT_DONT_COALESCE_SEPARATORS
);
279 opts
= strdup("mode=0755");
285 if (!path_is_absolute(path
))
288 m
= custom_mount_add(l
, n
, CUSTOM_MOUNT_TMPFS
);
292 m
->destination
= TAKE_PTR(path
);
293 m
->options
= TAKE_PTR(opts
);
298 int overlay_mount_parse(CustomMount
**l
, size_t *n
, const char *s
, bool read_only
) {
299 _cleanup_free_
char *upper
= NULL
, *destination
= NULL
;
300 _cleanup_strv_free_
char **lower
= NULL
;
304 k
= strv_split_extract(&lower
, s
, ":", EXTRACT_DONT_COALESCE_SEPARATORS
);
308 return -EADDRNOTAVAIL
;
310 /* If two parameters are specified, the first one is the lower, the second one the upper directory. And
311 * we'll also define the destination mount point the same as the upper. */
313 if (!source_path_is_valid(lower
[0]) ||
314 !source_path_is_valid(lower
[1]))
317 upper
= TAKE_PTR(lower
[1]);
319 destination
= strdup(upper
[0] == '+' ? upper
+1 : upper
); /* take the destination without "+" prefix */
325 /* If more than two parameters are specified, the last one is the destination, the second to last one
326 * the "upper", and all before that the "lower" directories. */
328 destination
= lower
[k
- 1];
329 upper
= TAKE_PTR(lower
[k
- 2]);
331 STRV_FOREACH(i
, lower
)
332 if (!source_path_is_valid(*i
))
335 /* If the upper directory is unspecified, then let's create it automatically as a throw-away directory
338 upper
= mfree(upper
);
339 else if (!source_path_is_valid(upper
))
342 if (!path_is_absolute(destination
))
346 m
= custom_mount_add(l
, n
, CUSTOM_MOUNT_OVERLAY
);
350 m
->destination
= TAKE_PTR(destination
);
351 m
->source
= TAKE_PTR(upper
);
352 m
->lower
= TAKE_PTR(lower
);
353 m
->read_only
= read_only
;
358 int inaccessible_mount_parse(CustomMount
**l
, size_t *n
, const char *s
) {
359 _cleanup_free_
char *path
= NULL
;
366 if (!path_is_absolute(s
))
373 m
= custom_mount_add(l
, n
, CUSTOM_MOUNT_INACCESSIBLE
);
377 m
->destination
= TAKE_PTR(path
);
381 int tmpfs_patch_options(
384 const char *selinux_apifs_context
,
389 if (uid_shift
!= UID_INVALID
) {
390 if (asprintf(&buf
, "%s%suid=" UID_FMT
",gid=" UID_FMT
,
391 strempty(options
), options
? "," : "",
392 uid_shift
, uid_shift
) < 0)
399 if (selinux_apifs_context
) {
402 t
= strjoin(strempty(options
), options
? "," : "",
403 "context=\"", selinux_apifs_context
, "\"");
412 if (!buf
&& options
) {
413 buf
= strdup(options
);
422 int mount_sysfs(const char *dest
, MountSettingsMask mount_settings
) {
423 const char *full
, *top
, *x
;
425 unsigned long extra_flags
= 0;
427 top
= prefix_roota(dest
, "/sys");
428 r
= path_is_fs_type(top
, SYSFS_MAGIC
);
430 return log_error_errno(r
, "Failed to determine filesystem type of %s: %m", top
);
431 /* /sys might already be mounted as sysfs by the outer child in the
432 * !netns case. In this case, it's all good. Don't touch it because we
433 * don't have the right to do so, see https://github.com/systemd/systemd/issues/1555.
438 full
= prefix_roota(top
, "/full");
440 (void) mkdir(full
, 0755);
442 if (FLAGS_SET(mount_settings
, MOUNT_APPLY_APIVFS_RO
))
443 extra_flags
|= MS_RDONLY
;
445 r
= mount_verbose(LOG_ERR
, "sysfs", full
, "sysfs",
446 MS_NOSUID
|MS_NOEXEC
|MS_NODEV
|extra_flags
, NULL
);
450 FOREACH_STRING(x
, "block", "bus", "class", "dev", "devices", "kernel") {
451 _cleanup_free_
char *from
= NULL
, *to
= NULL
;
453 from
= path_join(full
, x
);
457 to
= path_join(top
, x
);
461 (void) mkdir(to
, 0755);
463 r
= mount_verbose(LOG_ERR
, from
, to
, NULL
, MS_BIND
, NULL
);
467 r
= mount_verbose(LOG_ERR
, NULL
, to
, NULL
,
468 MS_BIND
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
|MS_REMOUNT
|extra_flags
, NULL
);
473 r
= umount_verbose(full
);
478 return log_error_errno(errno
, "Failed to remove %s: %m", full
);
480 /* Create mountpoint for cgroups. Otherwise we are not allowed since we
481 * remount /sys read-only.
483 x
= prefix_roota(top
, "/fs/cgroup");
484 (void) mkdir_p(x
, 0755);
486 return mount_verbose(LOG_ERR
, NULL
, top
, NULL
,
487 MS_BIND
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
|MS_REMOUNT
|extra_flags
, NULL
);
490 int mount_all(const char *dest
,
491 MountSettingsMask mount_settings
,
493 const char *selinux_apifs_context
) {
495 #define PROC_INACCESSIBLE_REG(path) \
496 { "/run/systemd/inaccessible/reg", (path), NULL, NULL, MS_BIND, \
497 MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO }, /* Bind mount first ... */ \
498 { NULL, (path), NULL, NULL, MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT, \
499 MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO } /* Then, make it r/o */
501 #define PROC_READ_ONLY(path) \
502 { (path), (path), NULL, NULL, MS_BIND, \
503 MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO }, /* Bind mount first ... */ \
504 { NULL, (path), NULL, NULL, MS_BIND|MS_RDONLY|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_REMOUNT, \
505 MOUNT_IN_USERNS|MOUNT_APPLY_APIVFS_RO } /* Then, make it r/o */
507 typedef struct MountPoint
{
513 MountSettingsMask mount_settings
;
516 static const MountPoint mount_table
[] = {
517 /* First we list inner child mounts (i.e. mounts applied *after* entering user namespacing) */
518 { "proc", "/proc", "proc", NULL
, MS_NOSUID
|MS_NOEXEC
|MS_NODEV
,
519 MOUNT_FATAL
|MOUNT_IN_USERNS
|MOUNT_MKDIR
},
521 { "/proc/sys", "/proc/sys", NULL
, NULL
, MS_BIND
,
522 MOUNT_FATAL
|MOUNT_IN_USERNS
|MOUNT_APPLY_APIVFS_RO
}, /* Bind mount first ... */
524 { "/proc/sys/net", "/proc/sys/net", NULL
, NULL
, MS_BIND
,
525 MOUNT_FATAL
|MOUNT_IN_USERNS
|MOUNT_APPLY_APIVFS_RO
|MOUNT_APPLY_APIVFS_NETNS
}, /* (except for this) */
527 { NULL
, "/proc/sys", NULL
, NULL
, MS_BIND
|MS_RDONLY
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
|MS_REMOUNT
,
528 MOUNT_FATAL
|MOUNT_IN_USERNS
|MOUNT_APPLY_APIVFS_RO
}, /* ... then, make it r/o */
530 /* Make these files inaccessible to container payloads: they potentially leak information about kernel
531 * internals or the host's execution environment to the container */
532 PROC_INACCESSIBLE_REG("/proc/kallsyms"),
533 PROC_INACCESSIBLE_REG("/proc/kcore"),
534 PROC_INACCESSIBLE_REG("/proc/keys"),
535 PROC_INACCESSIBLE_REG("/proc/sysrq-trigger"),
536 PROC_INACCESSIBLE_REG("/proc/timer_list"),
538 /* Make these directories read-only to container payloads: they show hardware information, and in some
539 * cases contain tunables the container really shouldn't have access to. */
540 PROC_READ_ONLY("/proc/acpi"),
541 PROC_READ_ONLY("/proc/apm"),
542 PROC_READ_ONLY("/proc/asound"),
543 PROC_READ_ONLY("/proc/bus"),
544 PROC_READ_ONLY("/proc/fs"),
545 PROC_READ_ONLY("/proc/irq"),
546 PROC_READ_ONLY("/proc/scsi"),
548 { "mqueue", "/dev/mqueue", "mqueue", NULL
, MS_NOSUID
|MS_NOEXEC
|MS_NODEV
,
549 MOUNT_IN_USERNS
|MOUNT_MKDIR
},
551 /* Then we list outer child mounts (i.e. mounts applied *before* entering user namespacing) */
552 { "tmpfs", "/tmp", "tmpfs", "mode=1777" TMPFS_LIMITS_TMP
, MS_NOSUID
|MS_NODEV
|MS_STRICTATIME
,
553 MOUNT_FATAL
|MOUNT_APPLY_TMPFS_TMP
|MOUNT_MKDIR
},
554 { "tmpfs", "/sys", "tmpfs", "mode=555" TMPFS_LIMITS_SYS
, MS_NOSUID
|MS_NOEXEC
|MS_NODEV
,
555 MOUNT_FATAL
|MOUNT_APPLY_APIVFS_NETNS
|MOUNT_MKDIR
},
556 { "sysfs", "/sys", "sysfs", NULL
, MS_RDONLY
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
,
557 MOUNT_FATAL
|MOUNT_APPLY_APIVFS_RO
|MOUNT_MKDIR
}, /* skipped if above was mounted */
558 { "sysfs", "/sys", "sysfs", NULL
, MS_NOSUID
|MS_NOEXEC
|MS_NODEV
,
559 MOUNT_FATAL
|MOUNT_MKDIR
}, /* skipped if above was mounted */
560 { "tmpfs", "/dev", "tmpfs", "mode=755" TMPFS_LIMITS_DEV
, MS_NOSUID
|MS_STRICTATIME
,
561 MOUNT_FATAL
|MOUNT_MKDIR
},
562 { "tmpfs", "/dev/shm", "tmpfs", "mode=1777" TMPFS_LIMITS_DEV_SHM
, MS_NOSUID
|MS_NODEV
|MS_STRICTATIME
,
563 MOUNT_FATAL
|MOUNT_MKDIR
},
564 { "tmpfs", "/run", "tmpfs", "mode=755" TMPFS_LIMITS_RUN
, MS_NOSUID
|MS_NODEV
|MS_STRICTATIME
,
565 MOUNT_FATAL
|MOUNT_MKDIR
},
566 { "/run/host", "/run/host", NULL
, NULL
, MS_BIND
,
567 MOUNT_FATAL
|MOUNT_MKDIR
|MOUNT_PREFIX_ROOT
}, /* Prepare this so that we can make it read-only when we are done */
568 { "/etc/os-release", "/run/host/os-release", NULL
, NULL
, MS_BIND
,
569 MOUNT_TOUCH
}, /* As per kernel interface requirements, bind mount first (creating mount points) and make read-only later */
570 { "/usr/lib/os-release", "/run/host/os-release", NULL
, NULL
, MS_BIND
,
571 MOUNT_FATAL
}, /* If /etc/os-release doesn't exist use the version in /usr/lib as fallback */
572 { NULL
, "/run/host/os-release", NULL
, NULL
, MS_BIND
|MS_RDONLY
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
|MS_REMOUNT
,
574 { NULL
, "/run/host", NULL
, NULL
, MS_BIND
|MS_RDONLY
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
|MS_REMOUNT
,
575 MOUNT_FATAL
|MOUNT_IN_USERNS
},
577 { "/sys/fs/selinux", "/sys/fs/selinux", NULL
, NULL
, MS_BIND
,
578 MOUNT_MKDIR
}, /* Bind mount first (mkdir/chown the mount point in case /sys/ is mounted as minimal skeleton tmpfs) */
579 { NULL
, "/sys/fs/selinux", NULL
, NULL
, MS_BIND
|MS_RDONLY
|MS_NOSUID
|MS_NOEXEC
|MS_NODEV
|MS_REMOUNT
,
580 0 }, /* Then, make it r/o (don't mkdir/chown the mount point here, the previous entry already did that) */
584 bool use_userns
= FLAGS_SET(mount_settings
, MOUNT_USE_USERNS
);
585 bool netns
= FLAGS_SET(mount_settings
, MOUNT_APPLY_APIVFS_NETNS
);
586 bool ro
= FLAGS_SET(mount_settings
, MOUNT_APPLY_APIVFS_RO
);
587 bool in_userns
= FLAGS_SET(mount_settings
, MOUNT_IN_USERNS
);
588 bool tmpfs_tmp
= FLAGS_SET(mount_settings
, MOUNT_APPLY_TMPFS_TMP
);
592 for (k
= 0; k
< ELEMENTSOF(mount_table
); k
++) {
593 _cleanup_free_
char *where
= NULL
, *options
= NULL
, *prefixed
= NULL
;
594 bool fatal
= FLAGS_SET(mount_table
[k
].mount_settings
, MOUNT_FATAL
);
597 if (in_userns
!= FLAGS_SET(mount_table
[k
].mount_settings
, MOUNT_IN_USERNS
))
600 if (!netns
&& FLAGS_SET(mount_table
[k
].mount_settings
, MOUNT_APPLY_APIVFS_NETNS
))
603 if (!ro
&& FLAGS_SET(mount_table
[k
].mount_settings
, MOUNT_APPLY_APIVFS_RO
))
606 if (!tmpfs_tmp
&& FLAGS_SET(mount_table
[k
].mount_settings
, MOUNT_APPLY_TMPFS_TMP
))
609 r
= chase_symlinks(mount_table
[k
].where
, dest
, CHASE_NONEXISTENT
|CHASE_PREFIX_ROOT
, &where
, NULL
);
611 return log_error_errno(r
, "Failed to resolve %s/%s: %m", dest
, mount_table
[k
].where
);
613 /* Skip this entry if it is not a remount. */
614 if (mount_table
[k
].what
) {
615 r
= path_is_mount_point(where
, NULL
, 0);
616 if (r
< 0 && r
!= -ENOENT
)
617 return log_error_errno(r
, "Failed to detect whether %s is a mount point: %m", where
);
622 if ((mount_table
[k
].mount_settings
& (MOUNT_MKDIR
|MOUNT_TOUCH
)) != 0) {
623 uid_t u
= (use_userns
&& !in_userns
) ? uid_shift
: UID_INVALID
;
625 if (FLAGS_SET(mount_table
[k
].mount_settings
, MOUNT_TOUCH
))
626 r
= mkdir_parents_safe(dest
, where
, 0755, u
, u
, 0);
628 r
= mkdir_p_safe(dest
, where
, 0755, u
, u
, 0);
629 if (r
< 0 && r
!= -EEXIST
) {
630 if (fatal
&& r
!= -EROFS
)
631 return log_error_errno(r
, "Failed to create directory %s: %m", where
);
633 log_debug_errno(r
, "Failed to create directory %s: %m", where
);
635 /* If we failed mkdir() or chown() due to the root directory being read only,
636 * attempt to mount this fs anyway and let mount_verbose log any errors */
642 if (FLAGS_SET(mount_table
[k
].mount_settings
, MOUNT_TOUCH
)) {
644 if (r
< 0 && r
!= -EEXIST
) {
645 if (fatal
&& r
!= -EROFS
)
646 return log_error_errno(r
, "Failed to create file %s: %m", where
);
648 log_debug_errno(r
, "Failed to create file %s: %m", where
);
654 o
= mount_table
[k
].options
;
655 if (streq_ptr(mount_table
[k
].type
, "tmpfs")) {
656 r
= tmpfs_patch_options(o
, in_userns
? 0 : uid_shift
, selinux_apifs_context
, &options
);
663 if (FLAGS_SET(mount_table
[k
].mount_settings
, MOUNT_PREFIX_ROOT
)) {
664 /* Optionally prefix the mount source with the root dir. This is useful in bind
665 * mounts to be created within the container image before we transition into it. Note
666 * that MOUNT_IN_USERNS is run after we transitioned hence prefixing is not ncessary
668 r
= chase_symlinks(mount_table
[k
].what
, dest
, CHASE_PREFIX_ROOT
, &prefixed
, NULL
);
670 return log_error_errno(r
, "Failed to resolve %s/%s: %m", dest
, mount_table
[k
].what
);
673 r
= mount_verbose(fatal
? LOG_ERR
: LOG_DEBUG
,
674 prefixed
?: mount_table
[k
].what
,
677 mount_table
[k
].flags
,
686 static int parse_mount_bind_options(const char *options
, unsigned long *mount_flags
, char **mount_opts
) {
687 const char *p
= options
;
688 unsigned long flags
= *mount_flags
;
695 _cleanup_free_
char *word
= NULL
;
697 r
= extract_first_word(&p
, &word
, ",", 0);
699 return log_error_errno(r
, "Failed to extract mount option: %m");
703 if (streq(word
, "rbind"))
705 else if (streq(word
, "norbind"))
708 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
709 "Invalid bind mount option: %s",
714 *mount_flags
= flags
;
715 /* in the future mount_opts will hold string options for mount(2) */
721 static int mount_bind(const char *dest
, CustomMount
*m
) {
722 _cleanup_free_
char *mount_opts
= NULL
, *where
= NULL
;
723 unsigned long mount_flags
= MS_BIND
| MS_REC
;
724 struct stat source_st
, dest_st
;
731 r
= parse_mount_bind_options(m
->options
, &mount_flags
, &mount_opts
);
736 if (stat(m
->source
, &source_st
) < 0)
737 return log_error_errno(errno
, "Failed to stat %s: %m", m
->source
);
739 r
= chase_symlinks(m
->destination
, dest
, CHASE_PREFIX_ROOT
|CHASE_NONEXISTENT
, &where
, NULL
);
741 return log_error_errno(r
, "Failed to resolve %s/%s: %m", dest
, m
->destination
);
742 if (r
> 0) { /* Path exists already? */
744 if (stat(where
, &dest_st
) < 0)
745 return log_error_errno(errno
, "Failed to stat %s: %m", where
);
747 if (S_ISDIR(source_st
.st_mode
) && !S_ISDIR(dest_st
.st_mode
))
748 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
749 "Cannot bind mount directory %s on file %s.",
752 if (!S_ISDIR(source_st
.st_mode
) && S_ISDIR(dest_st
.st_mode
))
753 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
754 "Cannot bind mount file %s on directory %s.",
757 } else { /* Path doesn't exist yet? */
758 r
= mkdir_parents_label(where
, 0755);
760 return log_error_errno(r
, "Failed to make parents of %s: %m", where
);
762 /* Create the mount point. Any non-directory file can be
763 * mounted on any non-directory file (regular, fifo, socket,
766 if (S_ISDIR(source_st
.st_mode
))
767 r
= mkdir_label(where
, 0755);
771 return log_error_errno(r
, "Failed to create mount point %s: %m", where
);
774 r
= mount_verbose(LOG_ERR
, m
->source
, where
, NULL
, mount_flags
, mount_opts
);
779 r
= bind_remount_recursive(where
, MS_RDONLY
, MS_RDONLY
, NULL
);
781 return log_error_errno(r
, "Read-only bind mount failed: %m");
787 static int mount_tmpfs(const char *dest
, CustomMount
*m
, uid_t uid_shift
, const char *selinux_apifs_context
) {
790 _cleanup_free_
char *buf
= NULL
, *where
= NULL
;
796 r
= chase_symlinks(m
->destination
, dest
, CHASE_PREFIX_ROOT
|CHASE_NONEXISTENT
, &where
, NULL
);
798 return log_error_errno(r
, "Failed to resolve %s/%s: %m", dest
, m
->destination
);
799 if (r
== 0) { /* Doesn't exist yet? */
800 r
= mkdir_p_label(where
, 0755);
802 return log_error_errno(r
, "Creating mount point for tmpfs %s failed: %m", where
);
805 r
= tmpfs_patch_options(m
->options
, uid_shift
== 0 ? UID_INVALID
: uid_shift
, selinux_apifs_context
, &buf
);
808 options
= r
> 0 ? buf
: m
->options
;
810 return mount_verbose(LOG_ERR
, "tmpfs", where
, "tmpfs", MS_NODEV
|MS_STRICTATIME
, options
);
813 static char *joined_and_escaped_lower_dirs(char **lower
) {
814 _cleanup_strv_free_
char **sv
= NULL
;
816 sv
= strv_copy(lower
);
822 if (!strv_shell_escape(sv
, ",:"))
825 return strv_join(sv
, ":");
828 static int mount_overlay(const char *dest
, CustomMount
*m
) {
829 _cleanup_free_
char *lower
= NULL
, *where
= NULL
, *escaped_source
= NULL
;
836 r
= chase_symlinks(m
->destination
, dest
, CHASE_PREFIX_ROOT
|CHASE_NONEXISTENT
, &where
, NULL
);
838 return log_error_errno(r
, "Failed to resolve %s/%s: %m", dest
, m
->destination
);
839 if (r
== 0) { /* Doesn't exist yet? */
840 r
= mkdir_label(where
, 0755);
842 return log_error_errno(r
, "Creating mount point for overlay %s failed: %m", where
);
845 (void) mkdir_p_label(m
->source
, 0755);
847 lower
= joined_and_escaped_lower_dirs(m
->lower
);
851 escaped_source
= shell_escape(m
->source
, ",:");
856 options
= strjoina("lowerdir=", escaped_source
, ":", lower
);
858 _cleanup_free_
char *escaped_work_dir
= NULL
;
860 escaped_work_dir
= shell_escape(m
->work_dir
, ",:");
861 if (!escaped_work_dir
)
864 options
= strjoina("lowerdir=", lower
, ",upperdir=", escaped_source
, ",workdir=", escaped_work_dir
);
867 return mount_verbose(LOG_ERR
, "overlay", where
, "overlay", m
->read_only
? MS_RDONLY
: 0, options
);
870 static int mount_inaccessible(const char *dest
, CustomMount
*m
) {
871 _cleanup_free_
char *where
= NULL
, *source
= NULL
;
878 r
= chase_symlinks_and_stat(m
->destination
, dest
, CHASE_PREFIX_ROOT
, &where
, &st
, NULL
);
880 log_full_errno(m
->graceful
? LOG_DEBUG
: LOG_ERR
, r
, "Failed to resolve %s/%s: %m", dest
, m
->destination
);
881 return m
->graceful
? 0 : r
;
884 r
= mode_to_inaccessible_node(NULL
, st
.st_mode
, &source
);
886 return m
->graceful
? 0 : r
;
888 r
= mount_verbose(m
->graceful
? LOG_DEBUG
: LOG_ERR
, source
, where
, NULL
, MS_BIND
, NULL
);
890 return m
->graceful
? 0 : r
;
892 r
= mount_verbose(m
->graceful
? LOG_DEBUG
: LOG_ERR
, NULL
, where
, NULL
, MS_BIND
|MS_RDONLY
|MS_REMOUNT
, NULL
);
894 (void) umount_verbose(where
);
895 return m
->graceful
? 0 : r
;
901 static int mount_arbitrary(const char *dest
, CustomMount
*m
) {
902 _cleanup_free_
char *where
= NULL
;
908 r
= chase_symlinks(m
->destination
, dest
, CHASE_PREFIX_ROOT
|CHASE_NONEXISTENT
, &where
, NULL
);
910 return log_error_errno(r
, "Failed to resolve %s/%s: %m", dest
, m
->destination
);
911 if (r
== 0) { /* Doesn't exist yet? */
912 r
= mkdir_p_label(where
, 0755);
914 return log_error_errno(r
, "Creating mount point for mount %s failed: %m", where
);
917 return mount_verbose(LOG_ERR
, m
->source
, where
, m
->type_argument
, 0, m
->options
);
922 CustomMount
*mounts
, size_t n
,
924 const char *selinux_apifs_context
,
925 MountSettingsMask mount_settings
) {
932 for (i
= 0; i
< n
; i
++) {
933 CustomMount
*m
= mounts
+ i
;
935 if (FLAGS_SET(mount_settings
, MOUNT_IN_USERNS
) != m
->in_userns
)
938 if (FLAGS_SET(mount_settings
, MOUNT_ROOT_ONLY
) && !path_equal(m
->destination
, "/"))
941 if (FLAGS_SET(mount_settings
, MOUNT_NON_ROOT_ONLY
) && path_equal(m
->destination
, "/"))
946 case CUSTOM_MOUNT_BIND
:
947 r
= mount_bind(dest
, m
);
950 case CUSTOM_MOUNT_TMPFS
:
951 r
= mount_tmpfs(dest
, m
, uid_shift
, selinux_apifs_context
);
954 case CUSTOM_MOUNT_OVERLAY
:
955 r
= mount_overlay(dest
, m
);
958 case CUSTOM_MOUNT_INACCESSIBLE
:
959 r
= mount_inaccessible(dest
, m
);
962 case CUSTOM_MOUNT_ARBITRARY
:
963 r
= mount_arbitrary(dest
, m
);
967 assert_not_reached("Unknown custom mount type");
977 bool has_custom_root_mount(const CustomMount
*mounts
, size_t n
) {
980 for (i
= 0; i
< n
; i
++) {
981 const CustomMount
*m
= mounts
+ i
;
983 if (path_equal(m
->destination
, "/"))
990 static int setup_volatile_state(const char *directory
, uid_t uid_shift
, const char *selinux_apifs_context
) {
992 _cleanup_free_
char *buf
= NULL
;
993 const char *p
, *options
;
998 /* --volatile=state means we simply overmount /var with a tmpfs, and the rest read-only. */
1000 r
= bind_remount_recursive(directory
, MS_RDONLY
, MS_RDONLY
, NULL
);
1002 return log_error_errno(r
, "Failed to remount %s read-only: %m", directory
);
1004 p
= prefix_roota(directory
, "/var");
1006 if (r
< 0 && errno
!= EEXIST
)
1007 return log_error_errno(errno
, "Failed to create %s: %m", directory
);
1009 options
= "mode=755" TMPFS_LIMITS_VOLATILE_STATE
;
1010 r
= tmpfs_patch_options(options
, uid_shift
== 0 ? UID_INVALID
: uid_shift
, selinux_apifs_context
, &buf
);
1016 return mount_verbose(LOG_ERR
, "tmpfs", p
, "tmpfs", MS_STRICTATIME
, options
);
1019 static int setup_volatile_yes(const char *directory
, uid_t uid_shift
, const char *selinux_apifs_context
) {
1021 bool tmpfs_mounted
= false, bind_mounted
= false;
1022 char template[] = "/tmp/nspawn-volatile-XXXXXX";
1023 _cleanup_free_
char *buf
= NULL
, *bindir
= NULL
;
1024 const char *f
, *t
, *options
;
1030 /* --volatile=yes means we mount a tmpfs to the root dir, and the original /usr to use inside it, and
1031 * that read-only. Before we start setting this up let's validate if the image has the /usr merge
1032 * implemented, and let's output a friendly log message if it hasn't. */
1034 bindir
= path_join(directory
, "/bin");
1037 if (lstat(bindir
, &st
) < 0) {
1038 if (errno
!= ENOENT
)
1039 return log_error_errno(errno
, "Failed to stat /bin directory below image: %m");
1041 /* ENOENT is fine, just means the image is probably just a naked /usr and we can create the
1043 } else if (S_ISDIR(st
.st_mode
))
1044 return log_error_errno(SYNTHETIC_ERRNO(EISDIR
),
1045 "Sorry, --volatile=yes mode is not supported with OS images that have not merged /bin/, /sbin/, /lib/, /lib64/ into /usr/. "
1046 "Please work with your distribution and help them adopt the merged /usr scheme.");
1047 else if (!S_ISLNK(st
.st_mode
))
1048 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
1049 "Error starting image: if --volatile=yes is used /bin must be a symlink (for merged /usr support) or non-existent (in which case a symlink is created automatically).");
1051 if (!mkdtemp(template))
1052 return log_error_errno(errno
, "Failed to create temporary directory: %m");
1054 options
= "mode=755" TMPFS_LIMITS_ROOTFS
;
1055 r
= tmpfs_patch_options(options
, uid_shift
== 0 ? UID_INVALID
: uid_shift
, selinux_apifs_context
, &buf
);
1061 r
= mount_verbose(LOG_ERR
, "tmpfs", template, "tmpfs", MS_STRICTATIME
, options
);
1065 tmpfs_mounted
= true;
1067 f
= prefix_roota(directory
, "/usr");
1068 t
= prefix_roota(template, "/usr");
1071 if (r
< 0 && errno
!= EEXIST
) {
1072 r
= log_error_errno(errno
, "Failed to create %s: %m", t
);
1076 r
= mount_verbose(LOG_ERR
, f
, t
, NULL
, MS_BIND
|MS_REC
, NULL
);
1080 bind_mounted
= true;
1082 r
= bind_remount_recursive(t
, MS_RDONLY
, MS_RDONLY
, NULL
);
1084 log_error_errno(r
, "Failed to remount %s read-only: %m", t
);
1088 r
= mount_verbose(LOG_ERR
, template, directory
, NULL
, MS_MOVE
, NULL
);
1092 (void) rmdir(template);
1098 (void) umount_verbose(t
);
1101 (void) umount_verbose(template);
1102 (void) rmdir(template);
1106 static int setup_volatile_overlay(const char *directory
, uid_t uid_shift
, const char *selinux_apifs_context
) {
1108 _cleanup_free_
char *buf
= NULL
, *escaped_directory
= NULL
, *escaped_upper
= NULL
, *escaped_work
= NULL
;
1109 char template[] = "/tmp/nspawn-volatile-XXXXXX";
1110 const char *upper
, *work
, *options
;
1111 bool tmpfs_mounted
= false;
1116 /* --volatile=overlay means we mount an overlayfs to the root dir. */
1118 if (!mkdtemp(template))
1119 return log_error_errno(errno
, "Failed to create temporary directory: %m");
1121 options
= "mode=755" TMPFS_LIMITS_ROOTFS
;
1122 r
= tmpfs_patch_options(options
, uid_shift
== 0 ? UID_INVALID
: uid_shift
, selinux_apifs_context
, &buf
);
1128 r
= mount_verbose(LOG_ERR
, "tmpfs", template, "tmpfs", MS_STRICTATIME
, options
);
1132 tmpfs_mounted
= true;
1134 upper
= strjoina(template, "/upper");
1135 work
= strjoina(template, "/work");
1137 if (mkdir(upper
, 0755) < 0) {
1138 r
= log_error_errno(errno
, "Failed to create %s: %m", upper
);
1141 if (mkdir(work
, 0755) < 0) {
1142 r
= log_error_errno(errno
, "Failed to create %s: %m", work
);
1146 /* And now, let's overmount the root dir with an overlayfs that uses the root dir as lower dir. It's kinda nice
1147 * that the kernel allows us to do that without going through some mount point rearrangements. */
1149 escaped_directory
= shell_escape(directory
, ",:");
1150 escaped_upper
= shell_escape(upper
, ",:");
1151 escaped_work
= shell_escape(work
, ",:");
1152 if (!escaped_directory
|| !escaped_upper
|| !escaped_work
) {
1157 options
= strjoina("lowerdir=", escaped_directory
, ",upperdir=", escaped_upper
, ",workdir=", escaped_work
);
1158 r
= mount_verbose(LOG_ERR
, "overlay", directory
, "overlay", 0, options
);
1162 (void) umount_verbose(template);
1164 (void) rmdir(template);
1168 int setup_volatile_mode(
1169 const char *directory
,
1172 const char *selinux_apifs_context
) {
1177 return setup_volatile_yes(directory
, uid_shift
, selinux_apifs_context
);
1179 case VOLATILE_STATE
:
1180 return setup_volatile_state(directory
, uid_shift
, selinux_apifs_context
);
1182 case VOLATILE_OVERLAY
:
1183 return setup_volatile_overlay(directory
, uid_shift
, selinux_apifs_context
);
1190 /* Expects *pivot_root_new and *pivot_root_old to be initialised to allocated memory or NULL. */
1191 int pivot_root_parse(char **pivot_root_new
, char **pivot_root_old
, const char *s
) {
1192 _cleanup_free_
char *root_new
= NULL
, *root_old
= NULL
;
1196 assert(pivot_root_new
);
1197 assert(pivot_root_old
);
1199 r
= extract_first_word(&p
, &root_new
, ":", EXTRACT_DONT_COALESCE_SEPARATORS
);
1208 root_old
= strdup(p
);
1213 if (!path_is_absolute(root_new
))
1215 if (root_old
&& !path_is_absolute(root_old
))
1218 free_and_replace(*pivot_root_new
, root_new
);
1219 free_and_replace(*pivot_root_old
, root_old
);
1224 int setup_pivot_root(const char *directory
, const char *pivot_root_new
, const char *pivot_root_old
) {
1225 _cleanup_free_
char *directory_pivot_root_new
= NULL
;
1226 _cleanup_free_
char *pivot_tmp_pivot_root_old
= NULL
;
1227 char pivot_tmp
[] = "/tmp/nspawn-pivot-XXXXXX";
1228 bool remove_pivot_tmp
= false;
1233 if (!pivot_root_new
)
1236 /* Pivot pivot_root_new to / and the existing / to pivot_root_old.
1237 * If pivot_root_old is NULL, the existing / disappears.
1238 * This requires a temporary directory, pivot_tmp, which is
1239 * not a child of either.
1241 * This is typically used for OSTree-style containers, where
1242 * the root partition contains several sysroots which could be
1243 * run. Normally, one would be chosen by the bootloader and
1244 * pivoted to / by initramfs.
1246 * For example, for an OSTree deployment, pivot_root_new
1247 * would be: /ostree/deploy/$os/deploy/$checksum. Note that this
1248 * code doesn’t do the /var mount which OSTree expects: use
1249 * --bind +/sysroot/ostree/deploy/$os/var:/var for that.
1251 * So in the OSTree case, we’ll end up with something like:
1252 * - directory = /tmp/nspawn-root-123456
1253 * - pivot_root_new = /ostree/deploy/os/deploy/123abc
1254 * - pivot_root_old = /sysroot
1255 * - directory_pivot_root_new =
1256 * /tmp/nspawn-root-123456/ostree/deploy/os/deploy/123abc
1257 * - pivot_tmp = /tmp/nspawn-pivot-123456
1258 * - pivot_tmp_pivot_root_old = /tmp/nspawn-pivot-123456/sysroot
1260 * Requires all file systems at directory and below to be mounted
1261 * MS_PRIVATE or MS_SLAVE so they can be moved.
1263 directory_pivot_root_new
= path_join(directory
, pivot_root_new
);
1264 if (!directory_pivot_root_new
)
1267 /* Remount directory_pivot_root_new to make it movable. */
1268 r
= mount_verbose(LOG_ERR
, directory_pivot_root_new
, directory_pivot_root_new
, NULL
, MS_BIND
, NULL
);
1272 if (pivot_root_old
) {
1273 if (!mkdtemp(pivot_tmp
)) {
1274 r
= log_error_errno(errno
, "Failed to create temporary directory: %m");
1278 remove_pivot_tmp
= true;
1279 pivot_tmp_pivot_root_old
= path_join(pivot_tmp
, pivot_root_old
);
1280 if (!pivot_tmp_pivot_root_old
) {
1285 r
= mount_verbose(LOG_ERR
, directory_pivot_root_new
, pivot_tmp
, NULL
, MS_MOVE
, NULL
);
1289 r
= mount_verbose(LOG_ERR
, directory
, pivot_tmp_pivot_root_old
, NULL
, MS_MOVE
, NULL
);
1293 r
= mount_verbose(LOG_ERR
, pivot_tmp
, directory
, NULL
, MS_MOVE
, NULL
);
1297 r
= mount_verbose(LOG_ERR
, directory_pivot_root_new
, directory
, NULL
, MS_MOVE
, NULL
);
1303 if (remove_pivot_tmp
)
1304 (void) rmdir(pivot_tmp
);