143fadf369
This removes the ability to configure which cgroup controllers to mount together. Instead, we'll now hardcode that "cpu" and "cpuacct" are mounted together as well as "net_cls" and "net_prio". The concept of mounting controllers together has no future as it does not exist to cgroupsv2. Moreover, the current logic is systematically broken, as revealed by the discussions in #10507. Also, we surveyed Red Hat customers and couldn't find a single user of the concept (which isn't particularly surprising, as it is broken...) This reduced the (already way too complex) cgroup handling for us, since we now know whenever we make a change to a cgroup for one controller to which other controllers it applies.
480 lines
19 KiB
C
480 lines
19 KiB
C
/* SPDX-License-Identifier: LGPL-2.1+ */
|
|
|
|
#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 "bus-util.h"
|
|
#include "cgroup-util.h"
|
|
#include "dev-setup.h"
|
|
#include "efivars.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 "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 {
|
|
MNT_NONE = 0,
|
|
MNT_FATAL = 1 << 0,
|
|
MNT_IN_CONTAINER = 1 << 1,
|
|
MNT_CHECK_WRITABLE = 1 << 2,
|
|
} MountMode;
|
|
|
|
typedef struct MountPoint {
|
|
const char *what;
|
|
const char *where;
|
|
const char *type;
|
|
const char *options;
|
|
unsigned long flags;
|
|
bool (*condition_fn)(void);
|
|
MountMode mode;
|
|
} MountPoint;
|
|
|
|
/* The first three entries we might need before SELinux is up. The
|
|
* fourth (securityfs) is needed by IMA to load a custom policy. The
|
|
* other ones we can delay until SELinux and IMA are loaded. When
|
|
* SMACK is enabled we need smackfs, too, so it's a fifth one. */
|
|
#if ENABLE_SMACK
|
|
#define N_EARLY_MOUNT 5
|
|
#else
|
|
#define N_EARLY_MOUNT 4
|
|
#endif
|
|
|
|
static const MountPoint mount_table[] = {
|
|
{ "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,
|
|
NULL, MNT_FATAL|MNT_IN_CONTAINER },
|
|
{ "devtmpfs", "/dev", "devtmpfs", "mode=755", MS_NOSUID|MS_STRICTATIME,
|
|
NULL, MNT_FATAL|MNT_IN_CONTAINER },
|
|
{ "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,
|
|
mac_smack_use, MNT_FATAL },
|
|
{ "tmpfs", "/dev/shm", "tmpfs", "mode=1777,smackfsroot=*", MS_NOSUID|MS_NODEV|MS_STRICTATIME,
|
|
mac_smack_use, MNT_FATAL },
|
|
#endif
|
|
{ "tmpfs", "/dev/shm", "tmpfs", "mode=1777", 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,
|
|
NULL, MNT_IN_CONTAINER },
|
|
#if ENABLE_SMACK
|
|
{ "tmpfs", "/run", "tmpfs", "mode=755,smackfsroot=*", MS_NOSUID|MS_NODEV|MS_STRICTATIME,
|
|
mac_smack_use, MNT_FATAL },
|
|
#endif
|
|
{ "tmpfs", "/run", "tmpfs", "mode=755", MS_NOSUID|MS_NODEV|MS_STRICTATIME,
|
|
NULL, MNT_FATAL|MNT_IN_CONTAINER },
|
|
{ "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,
|
|
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 },
|
|
{ "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,
|
|
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 },
|
|
{ "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,
|
|
NULL, MNT_NONE },
|
|
#if ENABLE_EFI
|
|
{ "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,
|
|
* we just list them here so that we know that we should ignore them */
|
|
|
|
static const char ignore_paths[] =
|
|
/* SELinux file systems */
|
|
"/sys/fs/selinux\0"
|
|
/* Container bind mounts */
|
|
"/proc/sys\0"
|
|
"/dev/console\0"
|
|
"/proc/kmsg\0";
|
|
|
|
bool mount_point_is_api(const char *path) {
|
|
unsigned i;
|
|
|
|
/* Checks if this mount point is considered "API", and hence
|
|
* should be ignored */
|
|
|
|
for (i = 0; i < ELEMENTSOF(mount_table); i ++)
|
|
if (path_equal(path, mount_table[i].where))
|
|
return true;
|
|
|
|
return path_startswith(path, "/sys/fs/cgroup/");
|
|
}
|
|
|
|
bool mount_point_ignore(const char *path) {
|
|
const char *i;
|
|
|
|
NULSTR_FOREACH(i, ignore_paths)
|
|
if (path_equal(path, i))
|
|
return true;
|
|
|
|
return false;
|
|
}
|
|
|
|
static int mount_one(const MountPoint *p, bool relabel) {
|
|
int r, priority;
|
|
|
|
assert(p);
|
|
|
|
priority = (p->mode & MNT_FATAL) ? LOG_ERR : LOG_DEBUG;
|
|
|
|
if (p->condition_fn && !p->condition_fn())
|
|
return 0;
|
|
|
|
/* Relabel first, just in case */
|
|
if (relabel)
|
|
(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) {
|
|
log_full_errno(priority, r, "Failed to determine whether %s is a mount point: %m", p->where);
|
|
return (p->mode & MNT_FATAL) ? r : 0;
|
|
}
|
|
if (r > 0)
|
|
return 0;
|
|
|
|
/* Skip securityfs in a container */
|
|
if (!(p->mode & MNT_IN_CONTAINER) && detect_container() > 0)
|
|
return 0;
|
|
|
|
/* The access mode here doesn't really matter too much, since
|
|
* the mounted file system will take precedence anyway. */
|
|
if (relabel)
|
|
(void) mkdir_p_label(p->where, 0755);
|
|
else
|
|
(void) mkdir_p(p->where, 0755);
|
|
|
|
log_debug("Mounting %s to %s of type %s with options %s.",
|
|
p->what,
|
|
p->where,
|
|
p->type,
|
|
strna(p->options));
|
|
|
|
if (mount(p->what,
|
|
p->where,
|
|
p->type,
|
|
p->flags,
|
|
p->options) < 0) {
|
|
log_full_errno(priority, errno, "Failed to mount %s at %s: %m", p->type, p->where);
|
|
return (p->mode & MNT_FATAL) ? -errno : 0;
|
|
}
|
|
|
|
/* Relabel again, since we now mounted something fresh here */
|
|
if (relabel)
|
|
(void) label_fix(p->where, 0);
|
|
|
|
if (p->mode & MNT_CHECK_WRITABLE) {
|
|
if (access(p->where, W_OK) < 0) {
|
|
r = -errno;
|
|
|
|
(void) umount(p->where);
|
|
(void) rmdir(p->where);
|
|
|
|
log_full_errno(priority, r, "Mount point %s not writable after mounting: %m", p->where);
|
|
return (p->mode & MNT_FATAL) ? r : 0;
|
|
}
|
|
}
|
|
|
|
return 1;
|
|
}
|
|
|
|
static int mount_points_setup(unsigned n, bool loaded_policy) {
|
|
unsigned i;
|
|
int r = 0;
|
|
|
|
for (i = 0; i < n; i ++) {
|
|
int j;
|
|
|
|
j = mount_one(mount_table + i, loaded_policy);
|
|
if (j != 0 && r >= 0)
|
|
r = j;
|
|
}
|
|
|
|
return r;
|
|
}
|
|
|
|
int mount_setup_early(void) {
|
|
assert_cc(N_EARLY_MOUNT <= ELEMENTSOF(mount_table));
|
|
|
|
/* Do a minimal mount of /proc and friends to enable the most
|
|
* basic stuff, such as SELinux */
|
|
return mount_points_setup(N_EARLY_MOUNT, false);
|
|
}
|
|
|
|
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;
|
|
|
|
if (!cg_is_legacy_wanted())
|
|
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,
|
|
};
|
|
|
|
controller = set_steal_first(controllers);
|
|
if (!controller)
|
|
break;
|
|
|
|
/* 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();
|
|
}
|
|
}
|
|
|
|
/* 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();
|
|
|
|
p.where = where;
|
|
p.options = options;
|
|
|
|
r = mount_one(&p, true);
|
|
if (r < 0)
|
|
return r;
|
|
|
|
/* 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");
|
|
|
|
return 0;
|
|
}
|
|
|
|
#if HAVE_SELINUX || ENABLE_SMACK
|
|
static int nftw_cb(
|
|
const char *fpath,
|
|
const struct stat *sb,
|
|
int tflag,
|
|
struct FTW *ftwbuf) {
|
|
|
|
/* No need to label /dev twice in a row... */
|
|
if (_unlikely_(ftwbuf->level == 0))
|
|
return FTW_CONTINUE;
|
|
|
|
(void) label_fix(fpath, 0);
|
|
|
|
/* /run/initramfs is static data and big, no need to
|
|
* dynamically relabel its contents at boot... */
|
|
if (_unlikely_(ftwbuf->level == 1 &&
|
|
tflag == FTW_D &&
|
|
streq(fpath, "/run/initramfs")))
|
|
return FTW_SKIP_SUBTREE;
|
|
|
|
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) {
|
|
int r = 0;
|
|
|
|
r = mount_points_setup(ELEMENTSOF(mount_table), loaded_policy);
|
|
if (r < 0)
|
|
return r;
|
|
|
|
#if HAVE_SELINUX || ENABLE_SMACK
|
|
/* Nodes in devtmpfs and /run need to be manually updated for
|
|
* the appropriate labels, after mounting. The other virtual
|
|
* API file systems like /sys and /proc do not need that, they
|
|
* use the same label for all their files. */
|
|
if (loaded_policy) {
|
|
usec_t before_relabel, after_relabel;
|
|
char timespan[FORMAT_TIMESPAN_MAX];
|
|
|
|
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);
|
|
|
|
r = relabel_cgroup_filesystems();
|
|
if (r < 0)
|
|
return r;
|
|
|
|
after_relabel = now(CLOCK_MONOTONIC);
|
|
|
|
log_info("Relabelled /dev, /run and /sys/fs/cgroup in %s.",
|
|
format_timespan(timespan, sizeof(timespan), after_relabel - before_relabel, 0));
|
|
}
|
|
#endif
|
|
|
|
/* Create a few default symlinks, which are normally created
|
|
* by udevd, but some scripts might need them before we start
|
|
* udevd. */
|
|
dev_setup(NULL, UID_INVALID, GID_INVALID);
|
|
|
|
/* Mark the root directory as shared in regards to mount propagation. The kernel defaults to "private", but we
|
|
* think it makes more sense to have a default of "shared" so that nspawn and the container tools work out of
|
|
* the box. If specific setups need other settings they can reset the propagation mode to private if
|
|
* 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 (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");
|
|
|
|
/* Create a few directories we always want around, Note that sd_booted() checks for /run/systemd/system, so
|
|
* this mkdir really needs to stay for good, otherwise software that copied sd-daemon.c into their sources will
|
|
* misdetect systemd. */
|
|
(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));
|
|
|
|
return 0;
|
|
}
|