#include "alloc-util.h"
#include "bus-util.h"
#include "cgroup-util.h"
+#include "conf-files.h"
+#include "cgroup-setup.h"
#include "dev-setup.h"
-#include "efivars.h"
+#include "dirent-util.h"
+#include "efi-loader.h"
+#include "fd-util.h"
#include "fileio.h"
#include "fs-util.h"
#include "label.h"
#include "log.h"
#include "macro.h"
-#include "missing.h"
#include "mkdir.h"
#include "mount-setup.h"
#include "mount-util.h"
+#include "mountpoint-util.h"
+#include "nulstr-util.h"
#include "path-util.h"
#include "set.h"
#include "smack-util.h"
#include "strv.h"
#include "user-util.h"
-#include "util.h"
#include "virt.h"
typedef enum MountMode {
#endif
static const MountPoint mount_table[] = {
- { "sysfs", "/sys", "sysfs", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
+ { "sysfs", "/sys", "sysfs", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
NULL, MNT_FATAL|MNT_IN_CONTAINER },
- { "proc", "/proc", "proc", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
+ { "proc", "/proc", "proc", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
NULL, MNT_FATAL|MNT_IN_CONTAINER },
- { "devtmpfs", "/dev", "devtmpfs", "mode=755", MS_NOSUID|MS_STRICTATIME,
+ { "devtmpfs", "/dev", "devtmpfs", "mode=755" TMPFS_LIMITS_DEV, MS_NOSUID|MS_NOEXEC|MS_STRICTATIME,
NULL, MNT_FATAL|MNT_IN_CONTAINER },
- { "securityfs", "/sys/kernel/security", "securityfs", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
+ { "securityfs", "/sys/kernel/security", "securityfs", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
NULL, MNT_NONE },
#if ENABLE_SMACK
- { "smackfs", "/sys/fs/smackfs", "smackfs", "smackfsdef=*", MS_NOSUID|MS_NOEXEC|MS_NODEV,
+ { "smackfs", "/sys/fs/smackfs", "smackfs", "smackfsdef=*", MS_NOSUID|MS_NOEXEC|MS_NODEV,
mac_smack_use, MNT_FATAL },
- { "tmpfs", "/dev/shm", "tmpfs", "mode=1777,smackfsroot=*", MS_NOSUID|MS_NODEV|MS_STRICTATIME,
+ { "tmpfs", "/dev/shm", "tmpfs", "mode=1777,smackfsroot=*" TMPFS_LIMITS_DEV_SHM, MS_NOSUID|MS_NODEV|MS_STRICTATIME,
mac_smack_use, MNT_FATAL },
#endif
- { "tmpfs", "/dev/shm", "tmpfs", "mode=1777", MS_NOSUID|MS_NODEV|MS_STRICTATIME,
+ { "tmpfs", "/dev/shm", "tmpfs", "mode=1777" TMPFS_LIMITS_DEV_SHM, MS_NOSUID|MS_NODEV|MS_STRICTATIME,
NULL, MNT_FATAL|MNT_IN_CONTAINER },
- { "devpts", "/dev/pts", "devpts", "mode=620,gid=" STRINGIFY(TTY_GID), MS_NOSUID|MS_NOEXEC,
+ { "devpts", "/dev/pts", "devpts", "mode=620,gid=" STRINGIFY(TTY_GID), MS_NOSUID|MS_NOEXEC,
NULL, MNT_IN_CONTAINER },
#if ENABLE_SMACK
- { "tmpfs", "/run", "tmpfs", "mode=755,smackfsroot=*", MS_NOSUID|MS_NODEV|MS_STRICTATIME,
+ { "tmpfs", "/run", "tmpfs", "mode=755,smackfsroot=*" TMPFS_LIMITS_RUN, MS_NOSUID|MS_NODEV|MS_STRICTATIME,
mac_smack_use, MNT_FATAL },
#endif
- { "tmpfs", "/run", "tmpfs", "mode=755", MS_NOSUID|MS_NODEV|MS_STRICTATIME,
+ { "tmpfs", "/run", "tmpfs", "mode=755" TMPFS_LIMITS_RUN, MS_NOSUID|MS_NODEV|MS_STRICTATIME,
NULL, MNT_FATAL|MNT_IN_CONTAINER },
- { "cgroup2", "/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 },
- { "cgroup2", "/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,
+ { "tmpfs", "/sys/fs/cgroup", "tmpfs", "mode=755" TMPFS_LIMITS_SYS_FS_CGROUP, MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME,
cg_is_legacy_wanted, MNT_FATAL|MNT_IN_CONTAINER },
- { "cgroup2", "/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 },
- { "cgroup2", "/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,
+ { "cgroup", "/sys/fs/cgroup/systemd", "cgroup", "none,name=systemd,xattr", MS_NOSUID|MS_NOEXEC|MS_NODEV,
cg_is_legacy_wanted, MNT_IN_CONTAINER },
- { "cgroup", "/sys/fs/cgroup/systemd", "cgroup", "none,name=systemd", MS_NOSUID|MS_NOEXEC|MS_NODEV,
+ { "cgroup", "/sys/fs/cgroup/systemd", "cgroup", "none,name=systemd", MS_NOSUID|MS_NOEXEC|MS_NODEV,
cg_is_legacy_wanted, MNT_FATAL|MNT_IN_CONTAINER },
- { "pstore", "/sys/fs/pstore", "pstore", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
+ { "pstore", "/sys/fs/pstore", "pstore", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
NULL, MNT_NONE },
#if ENABLE_EFI
- { "efivarfs", "/sys/firmware/efi/efivars", "efivarfs", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
+ { "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,
+ { "bpf", "/sys/fs/bpf", "bpf", "mode=700", MS_NOSUID|MS_NOEXEC|MS_NODEV,
NULL, MNT_NONE, },
};
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;
- bool has_argument = !!join_controllers;
int r;
if (!cg_is_legacy_wanted())
return 0;
/* Mount all available cgroup controllers that are built into the kernel. */
-
- if (!has_argument)
- /* The defaults:
- * mount "cpu" + "cpuacct" together, and "net_cls" + "net_prio".
- *
- * We'd like to add "cpuset" to the mix, but "cpuset" doesn't really
- * work for groups with no initialized attributes.
- */
- join_controllers = (char**[]) {
- STRV_MAKE("cpu", "cpuacct"),
- STRV_MAKE("net_cls", "net_prio"),
- NULL,
- };
-
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;
- for (k = join_controllers; *k; k++)
- if (strv_find(*k, controller))
- break;
-
- if (*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) {
- if (has_argument)
- 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
+ /* The simple case, where there's only one controller to mount together */
+ if (!options)
options = TAKE_PTR(controller);
- where = strappend("/sys/fs/cgroup/", options);
+ where = path_join("/sys/fs/cgroup", options);
if (!where)
return log_oom();
if (r < 0)
return r;
- if (r > 0 && *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. */
- (void) mount("tmpfs", "/sys/fs/cgroup", "tmpfs", MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME|MS_RDONLY, "mode=755");
+ /* 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" TMPFS_LIMITS_SYS_FS_CGROUP);
return 0;
}
return 0;
}
+
+static int relabel_extra(void) {
+ _cleanup_strv_free_ char **files = NULL;
+ char **file;
+ int r, c = 0;
+
+ /* Support for relabelling additional files or directories after loading the policy. For this, code in the
+ * initrd simply has to drop in *.relabel files into /run/systemd/relabel-extra.d/. We'll read all such files
+ * expecting one absolute path by line and will relabel each (and everyone below that in case the path refers
+ * to a directory). These drop-in files are supposed to be absolutely minimal, and do not understand comments
+ * and such. After the operation succeeded the files are removed, and the drop-in directory as well, if
+ * possible.
+ */
+
+ r = conf_files_list(&files, ".relabel", NULL,
+ CONF_FILES_FILTER_MASKED | CONF_FILES_REGULAR,
+ "/run/systemd/relabel-extra.d/");
+ if (r < 0)
+ return log_error_errno(r, "Failed to enumerate /run/systemd/relabel-extra.d/, ignoring: %m");
+
+ STRV_FOREACH(file, files) {
+ _cleanup_fclose_ FILE *f = NULL;
+
+ f = fopen(*file, "re");
+ if (!f) {
+ log_warning_errno(errno, "Failed to open %s, ignoring: %m", *file);
+ continue;
+ }
+
+ for (;;) {
+ _cleanup_free_ char *line = NULL;
+
+ r = read_line(f, LONG_LINE_MAX, &line);
+ if (r < 0) {
+ log_warning_errno(r, "Failed to read %s, ignoring: %m", *file);
+ break;
+ }
+ if (r == 0) /* EOF */
+ break;
+
+ path_simplify(line, true);
+
+ if (!path_is_normalized(line)) {
+ log_warning("Path to relabel is not normalized, ignoring: %s", line);
+ continue;
+ }
+
+ if (!path_is_absolute(line)) {
+ log_warning("Path to relabel is not absolute, ignoring: %s", line);
+ continue;
+ }
+
+ log_debug("Relabelling additional file/directory '%s'.", line);
+ (void) label_fix(line, 0);
+ (void) nftw(line, nftw_cb, 64, FTW_MOUNT|FTW_PHYS|FTW_ACTIONRETVAL);
+ c++;
+ }
+
+ if (unlink(*file) < 0)
+ log_warning_errno(errno, "Failed to remove %s, ignoring: %m", *file);
+ }
+
+ /* Remove when we complete things. */
+ if (rmdir("/run/systemd/relabel-extra.d") < 0 &&
+ errno != ENOENT)
+ log_warning_errno(errno, "Failed to remove /run/systemd/relabel-extra.d/ directory: %m");
+
+ return c;
+}
#endif
-int mount_setup(bool loaded_policy) {
+int mount_setup(bool loaded_policy, bool leave_propagation) {
int r = 0;
r = mount_points_setup(ELEMENTSOF(mount_table), loaded_policy);
if (loaded_policy) {
usec_t before_relabel, after_relabel;
char timespan[FORMAT_TIMESPAN_MAX];
+ const char *i;
+ int n_extra;
before_relabel = now(CLOCK_MONOTONIC);
- (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);
+ FOREACH_STRING(i, "/dev", "/dev/shm", "/run")
+ (void) nftw(i, nftw_cb, 64, FTW_MOUNT|FTW_PHYS|FTW_ACTIONRETVAL);
- r = relabel_cgroup_filesystems();
- if (r < 0)
- return r;
+ (void) relabel_cgroup_filesystems();
+
+ n_extra = relabel_extra();
after_relabel = now(CLOCK_MONOTONIC);
- log_info("Relabelled /dev, /run and /sys/fs/cgroup in %s.",
+ log_info("Relabelled /dev, /dev/shm, /run, /sys/fs/cgroup%s in %s.",
+ n_extra > 0 ? ", additional files" : "",
format_timespan(timespan, sizeof(timespan), after_relabel - before_relabel, 0));
}
#endif
* needed. Note that we set this only when we are invoked directly by the kernel. If we are invoked by a
* container manager we assume the container manager knows what it is doing (for example, because it set up
* some directories with different propagation modes). */
- if (detect_container() <= 0)
+ if (detect_container() <= 0 && !leave_propagation)
if (mount(NULL, "/", NULL, MS_REC|MS_SHARED, NULL) < 0)
log_warning_errno(errno, "Failed to set up the root directory for shared mount propagation: %m");
(void) mkdir_label("/run/systemd", 0755);
(void) mkdir_label("/run/systemd/system", 0755);
- /* Set up inaccessible (and empty) file nodes of all types */
- (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) mkfifo("/run/systemd/inaccessible/fifo", 0000);
- (void) mknod("/run/systemd/inaccessible/sock", S_IFSOCK | 0000, 0);
-
- /* The following two are likely to fail if we lack the privs for it (for example in an userns environment, if
- * CAP_SYS_MKNOD is missing, or if a device node policy prohibit major/minor of 0 device nodes to be
- * created). But that's entirely fine. Consumers of these files should carry fallback to use a different node
- * then, for example /run/systemd/inaccessible/sock, which is close enough in behaviour and semantics for most
- * uses. */
- (void) mknod("/run/systemd/inaccessible/chr", S_IFCHR | 0000, makedev(0, 0));
- (void) mknod("/run/systemd/inaccessible/blk", S_IFBLK | 0000, makedev(0, 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;
}