/* SPDX-License-Identifier: LGPL-2.1+ */
-/***
- This file is part of systemd.
-
- Copyright 2010 Lennart Poettering
-
- systemd is free software; you can redistribute it and/or modify it
- under the terms of the GNU Lesser General Public License as published by
- the Free Software Foundation; either version 2.1 of the License, or
- (at your option) any later version.
-
- systemd is distributed in the hope that it will be useful, but
- WITHOUT ANY WARRANTY; without even the implied warranty of
- MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
- Lesser General Public License for more details.
-
- You should have received a copy of the GNU Lesser General Public License
- along with systemd; If not, see <http://www.gnu.org/licenses/>.
-***/
#include <errno.h>
#include <ftw.h>
#include <stdlib.h>
#include <sys/mount.h>
+#include <sys/statvfs.h>
#include <unistd.h>
#include "alloc-util.h"
#include "virt.h"
typedef enum MountMode {
- MNT_NONE = 0,
- MNT_FATAL = 1 << 0,
- MNT_IN_CONTAINER = 1 << 1,
- MNT_CHECK_WRITABLE = 1 << 2,
+ MNT_NONE = 0,
+ MNT_FATAL = 1 << 0,
+ MNT_IN_CONTAINER = 1 << 1,
+ MNT_CHECK_WRITABLE = 1 << 2,
} MountMode;
typedef struct MountPoint {
#endif
{ "tmpfs", "/run", "tmpfs", "mode=755", MS_NOSUID|MS_NODEV|MS_STRICTATIME,
NULL, MNT_FATAL|MNT_IN_CONTAINER },
- { "cgroup", "/sys/fs/cgroup", "cgroup2", "nsdelegate", MS_NOSUID|MS_NOEXEC|MS_NODEV,
+ { "cgroup2", "/sys/fs/cgroup", "cgroup2", "nsdelegate", MS_NOSUID|MS_NOEXEC|MS_NODEV,
cg_is_unified_wanted, MNT_IN_CONTAINER|MNT_CHECK_WRITABLE },
- { "cgroup", "/sys/fs/cgroup", "cgroup2", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
+ { "cgroup2", "/sys/fs/cgroup", "cgroup2", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
cg_is_unified_wanted, MNT_IN_CONTAINER|MNT_CHECK_WRITABLE },
{ "tmpfs", "/sys/fs/cgroup", "tmpfs", "mode=755", MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME,
cg_is_legacy_wanted, MNT_FATAL|MNT_IN_CONTAINER },
- { "cgroup", "/sys/fs/cgroup/unified", "cgroup2", "nsdelegate", MS_NOSUID|MS_NOEXEC|MS_NODEV,
+ { "cgroup2", "/sys/fs/cgroup/unified", "cgroup2", "nsdelegate", MS_NOSUID|MS_NOEXEC|MS_NODEV,
cg_is_hybrid_wanted, MNT_IN_CONTAINER|MNT_CHECK_WRITABLE },
- { "cgroup", "/sys/fs/cgroup/unified", "cgroup2", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
+ { "cgroup2", "/sys/fs/cgroup/unified", "cgroup2", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
cg_is_hybrid_wanted, MNT_IN_CONTAINER|MNT_CHECK_WRITABLE },
{ "cgroup", "/sys/fs/cgroup/systemd", "cgroup", "none,name=systemd,xattr", MS_NOSUID|MS_NOEXEC|MS_NODEV,
cg_is_legacy_wanted, MNT_IN_CONTAINER },
{ "efivarfs", "/sys/firmware/efi/efivars", "efivarfs", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
is_efi_boot, MNT_NONE },
#endif
+ { "bpf", "/sys/fs/bpf", "bpf", "mode=700", MS_NOSUID|MS_NOEXEC|MS_NODEV,
+ NULL, MNT_NONE, },
};
/* These are API file systems that might be mounted by other software,
/* Relabel first, just in case */
if (relabel)
- (void) label_fix(p->where, true, true);
+ (void) label_fix(p->where, LABEL_IGNORE_ENOENT|LABEL_IGNORE_EROFS);
r = path_is_mount_point(p->where, NULL, AT_SYMLINK_FOLLOW);
if (r < 0 && r != -ENOENT) {
/* Relabel again, since we now mounted something fresh here */
if (relabel)
- (void) label_fix(p->where, false, false);
+ (void) label_fix(p->where, 0);
if (p->mode & MNT_CHECK_WRITABLE) {
if (access(p->where, W_OK) < 0) {
return mount_points_setup(N_EARLY_MOUNT, false);
}
-int mount_cgroup_controllers(char ***join_controllers) {
+static const char *join_with(const char *controller) {
+
+ static const char* const pairs[] = {
+ "cpu", "cpuacct",
+ "net_cls", "net_prio",
+ NULL
+ };
+
+ const char *const *x, *const *y;
+
+ assert(controller);
+
+ /* This will lookup which controller to mount another controller with. Input is a controller name, and output
+ * is the other controller name. The function works both ways: you can input one and get the other, and input
+ * the other to get the one. */
+
+ STRV_FOREACH_PAIR(x, y, pairs) {
+ if (streq(controller, *x))
+ return *y;
+ if (streq(controller, *y))
+ return *x;
+ }
+
+ return NULL;
+}
+
+static int symlink_controller(const char *target, const char *alias) {
+ const char *a;
+ int r;
+
+ assert(target);
+ assert(alias);
+
+ a = strjoina("/sys/fs/cgroup/", alias);
+
+ r = symlink_idempotent(target, a, false);
+ if (r < 0)
+ return log_error_errno(r, "Failed to create symlink %s: %m", a);
+
+#ifdef SMACK_RUN_LABEL
+ const char *p;
+
+ p = strjoina("/sys/fs/cgroup/", target);
+
+ r = mac_smack_copy(a, p);
+ if (r < 0 && r != -EOPNOTSUPP)
+ return log_error_errno(r, "Failed to copy smack label from %s to %s: %m", p, a);
+#endif
+
+ return 0;
+}
+
+int mount_cgroup_controllers(void) {
_cleanup_set_free_free_ Set *controllers = NULL;
int r;
return 0;
/* Mount all available cgroup controllers that are built into the kernel. */
-
r = cg_kernel_controllers(&controllers);
if (r < 0)
return log_error_errno(r, "Failed to enumerate cgroup controllers: %m");
for (;;) {
_cleanup_free_ char *options = NULL, *controller = NULL, *where = NULL;
+ const char *other_controller;
MountPoint p = {
.what = "cgroup",
.type = "cgroup",
.flags = MS_NOSUID|MS_NOEXEC|MS_NODEV,
.mode = MNT_IN_CONTAINER,
};
- char ***k = NULL;
controller = set_steal_first(controllers);
if (!controller)
break;
- if (join_controllers)
- for (k = join_controllers; *k; k++)
- if (strv_find(*k, controller))
- break;
-
- if (k && *k) {
- char **i, **j;
-
- for (i = *k, j = *k; *i; i++) {
-
- if (!streq(*i, controller)) {
- _cleanup_free_ char *t;
-
- t = set_remove(controllers, *i);
- if (!t) {
- free(*i);
- continue;
- }
- }
-
- *(j++) = *i;
+ /* Check if we shall mount this together with another controller */
+ other_controller = join_with(controller);
+ if (other_controller) {
+ _cleanup_free_ char *c = NULL;
+
+ /* Check if the other controller is actually available in the kernel too */
+ c = set_remove(controllers, other_controller);
+ if (c) {
+
+ /* Join the two controllers into one string, and maintain a stable ordering */
+ if (strcmp(controller, other_controller) < 0)
+ options = strjoin(controller, ",", other_controller);
+ else
+ options = strjoin(other_controller, ",", controller);
+ if (!options)
+ return log_oom();
}
-
- *j = NULL;
-
- options = strv_join(*k, ",");
- if (!options)
- return log_oom();
- } else {
- options = controller;
- controller = NULL;
}
+ /* The simple case, where there's only one controller to mount together */
+ if (!options)
+ options = TAKE_PTR(controller);
+
where = strappend("/sys/fs/cgroup/", options);
if (!where)
return log_oom();
if (r < 0)
return r;
- if (r > 0 && k && *k) {
- char **i;
-
- for (i = *k; *i; i++) {
- _cleanup_free_ char *t = NULL;
-
- t = strappend("/sys/fs/cgroup/", *i);
- if (!t)
- return log_oom();
-
- r = symlink(options, t);
- if (r >= 0) {
-#ifdef SMACK_RUN_LABEL
- _cleanup_free_ char *src;
- src = strappend("/sys/fs/cgroup/", options);
- if (!src)
- return log_oom();
- r = mac_smack_copy(t, src);
- if (r < 0 && r != -EOPNOTSUPP)
- return log_error_errno(r, "Failed to copy smack label from %s to %s: %m", src, t);
-#endif
- } else if (errno != EEXIST)
- return log_error_errno(errno, "Failed to create symlink %s: %m", t);
- }
- }
+ /* Create symlinks from the individual controller names, in case we have a joined mount */
+ if (controller)
+ (void) symlink_controller(options, controller);
+ if (other_controller)
+ (void) symlink_controller(options, other_controller);
}
- /* Now that we mounted everything, let's make the tmpfs the
- * cgroup file systems are mounted into read-only. */
+ /* Now that we mounted everything, let's make the tmpfs the cgroup file systems are mounted into read-only. */
(void) mount("tmpfs", "/sys/fs/cgroup", "tmpfs", MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME|MS_RDONLY, "mode=755");
return 0;
if (_unlikely_(ftwbuf->level == 0))
return FTW_CONTINUE;
- label_fix(fpath, false, false);
+ (void) label_fix(fpath, 0);
/* /run/initramfs is static data and big, no need to
* dynamically relabel its contents at boot... */
return FTW_CONTINUE;
};
+
+static int relabel_cgroup_filesystems(void) {
+ int r;
+ struct statfs st;
+
+ r = cg_all_unified();
+ if (r == 0) {
+ /* Temporarily remount the root cgroup filesystem to give it a proper label. Do this
+ only when the filesystem has been already populated by a previous instance of systemd
+ running from initrd. Otherwise don't remount anything and leave the filesystem read-write
+ for the cgroup filesystems to be mounted inside. */
+ if (statfs("/sys/fs/cgroup", &st) < 0)
+ return log_error_errno(errno, "Failed to determine mount flags for /sys/fs/cgroup: %m");
+
+ if (st.f_flags & ST_RDONLY)
+ (void) mount(NULL, "/sys/fs/cgroup", NULL, MS_REMOUNT, NULL);
+
+ (void) label_fix("/sys/fs/cgroup", 0);
+ (void) nftw("/sys/fs/cgroup", nftw_cb, 64, FTW_MOUNT|FTW_PHYS|FTW_ACTIONRETVAL);
+
+ if (st.f_flags & ST_RDONLY)
+ (void) mount(NULL, "/sys/fs/cgroup", NULL, MS_REMOUNT|MS_RDONLY, NULL);
+
+ } else if (r < 0)
+ return log_error_errno(r, "Failed to determine whether we are in all unified mode: %m");
+
+ return 0;
+}
#endif
int mount_setup(bool loaded_policy) {
before_relabel = now(CLOCK_MONOTONIC);
- nftw("/dev", nftw_cb, 64, FTW_MOUNT|FTW_PHYS|FTW_ACTIONRETVAL);
- nftw("/dev/shm", nftw_cb, 64, FTW_MOUNT|FTW_PHYS|FTW_ACTIONRETVAL);
- nftw("/run", nftw_cb, 64, FTW_MOUNT|FTW_PHYS|FTW_ACTIONRETVAL);
+ (void) nftw("/dev", nftw_cb, 64, FTW_MOUNT|FTW_PHYS|FTW_ACTIONRETVAL);
+ (void) nftw("/dev/shm", nftw_cb, 64, FTW_MOUNT|FTW_PHYS|FTW_ACTIONRETVAL);
+ (void) nftw("/run", nftw_cb, 64, FTW_MOUNT|FTW_PHYS|FTW_ACTIONRETVAL);
- /* Temporarily remount the root cgroup filesystem to give it a proper label. */
- r = cg_all_unified();
- if (r == 0) {
- (void) mount(NULL, "/sys/fs/cgroup", NULL, MS_REMOUNT, NULL);
- label_fix("/sys/fs/cgroup", false, false);
- nftw("/sys/fs/cgroup", nftw_cb, 64, FTW_MOUNT|FTW_PHYS|FTW_ACTIONRETVAL);
- (void) mount(NULL, "/sys/fs/cgroup", NULL, MS_REMOUNT|MS_RDONLY, NULL);
- } else if (r < 0)
- return log_error_errno(r, "Failed to determine whether we are in all unified mode: %m");
+ r = relabel_cgroup_filesystems();
+ if (r < 0)
+ return r;
after_relabel = now(CLOCK_MONOTONIC);
(void) mkdir_label("/run/systemd", 0755);
(void) mkdir_label("/run/systemd/system", 0755);
- /* Set up inaccessible items */
- (void) mkdir_label("/run/systemd/inaccessible", 0000);
- (void) mknod("/run/systemd/inaccessible/reg", S_IFREG | 0000, 0);
- (void) mkdir_label("/run/systemd/inaccessible/dir", 0000);
- (void) mknod("/run/systemd/inaccessible/chr", S_IFCHR | 0000, makedev(0, 0));
- (void) mknod("/run/systemd/inaccessible/blk", S_IFBLK | 0000, makedev(0, 0));
- (void) mkfifo("/run/systemd/inaccessible/fifo", 0000);
- (void) mknod("/run/systemd/inaccessible/sock", S_IFSOCK | 0000, 0);
+ /* Also create /run/systemd/inaccessible nodes, so that we always have something to mount inaccessible nodes
+ * from. */
+ (void) make_inaccessible_nodes(NULL, UID_INVALID, GID_INVALID);
return 0;
}