2017-11-18 17:09:20 +01:00
|
|
|
/* SPDX-License-Identifier: LGPL-2.1+ */
|
2010-03-31 16:29:55 +02:00
|
|
|
|
|
|
|
#include <errno.h>
|
2015-11-16 22:09:36 +01:00
|
|
|
#include <ftw.h>
|
2010-03-31 16:29:55 +02:00
|
|
|
#include <stdlib.h>
|
2015-11-16 22:09:36 +01:00
|
|
|
#include <sys/mount.h>
|
2018-03-28 13:36:33 +02:00
|
|
|
#include <sys/statvfs.h>
|
2010-10-27 05:47:48 +02:00
|
|
|
#include <unistd.h>
|
2010-03-31 16:29:55 +02:00
|
|
|
|
2015-10-27 03:01:06 +01:00
|
|
|
#include "alloc-util.h"
|
2015-07-05 12:24:29 +02:00
|
|
|
#include "bus-util.h"
|
2015-10-26 18:44:13 +01:00
|
|
|
#include "cgroup-util.h"
|
2019-08-06 20:22:23 +02:00
|
|
|
#include "conf-files.h"
|
2019-08-01 13:14:45 +02:00
|
|
|
#include "cgroup-setup.h"
|
2015-10-26 18:44:13 +01:00
|
|
|
#include "dev-setup.h"
|
2019-08-01 16:28:29 +02:00
|
|
|
#include "dirent-util.h"
|
|
|
|
#include "efi-loader.h"
|
2018-12-12 14:02:00 +01:00
|
|
|
#include "fd-util.h"
|
2017-11-22 17:20:35 +01:00
|
|
|
#include "fileio.h"
|
2016-07-06 09:48:58 +02:00
|
|
|
#include "fs-util.h"
|
2015-10-26 18:44:13 +01:00
|
|
|
#include "label.h"
|
2010-03-31 16:29:55 +02:00
|
|
|
#include "log.h"
|
2010-04-08 01:43:07 +02:00
|
|
|
#include "macro.h"
|
2012-04-10 21:54:31 +02:00
|
|
|
#include "mkdir.h"
|
2015-10-26 18:44:13 +01:00
|
|
|
#include "mount-setup.h"
|
2020-04-14 15:39:36 +02:00
|
|
|
#include "mount-util.h"
|
2018-11-29 10:24:39 +01:00
|
|
|
#include "mountpoint-util.h"
|
2019-03-14 13:14:33 +01:00
|
|
|
#include "nulstr-util.h"
|
2012-05-07 21:36:12 +02:00
|
|
|
#include "path-util.h"
|
2015-10-26 18:44:13 +01:00
|
|
|
#include "set.h"
|
2013-10-09 19:52:15 +02:00
|
|
|
#include "smack-util.h"
|
2015-10-26 18:44:13 +01:00
|
|
|
#include "strv.h"
|
2015-10-27 00:42:07 +01:00
|
|
|
#include "user-util.h"
|
2015-10-26 18:44:13 +01:00
|
|
|
#include "virt.h"
|
2011-01-04 01:58:38 +01:00
|
|
|
|
2012-11-04 17:03:48 +01:00
|
|
|
typedef enum MountMode {
|
2018-06-07 16:03:43 +02:00
|
|
|
MNT_NONE = 0,
|
|
|
|
MNT_FATAL = 1 << 0,
|
|
|
|
MNT_IN_CONTAINER = 1 << 1,
|
|
|
|
MNT_CHECK_WRITABLE = 1 << 2,
|
2012-11-04 17:03:48 +01:00
|
|
|
} MountMode;
|
|
|
|
|
2010-04-12 21:58:01 +02:00
|
|
|
typedef struct MountPoint {
|
|
|
|
const char *what;
|
|
|
|
const char *where;
|
|
|
|
const char *type;
|
|
|
|
const char *options;
|
|
|
|
unsigned long flags;
|
2012-11-04 17:03:48 +01:00
|
|
|
bool (*condition_fn)(void);
|
|
|
|
MountMode mode;
|
2010-04-12 21:58:01 +02:00
|
|
|
} MountPoint;
|
|
|
|
|
2011-07-29 01:48:18 +02:00
|
|
|
/* The first three entries we might need before SELinux is up. The
|
2012-03-15 19:06:10 +01:00
|
|
|
* fourth (securityfs) is needed by IMA to load a custom policy. The
|
2014-08-13 20:01:19 +02:00
|
|
|
* 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. */
|
2017-10-03 12:22:40 +02:00
|
|
|
#if ENABLE_SMACK
|
2013-03-07 20:06:58 +01:00
|
|
|
#define N_EARLY_MOUNT 5
|
2014-08-13 20:01:19 +02:00
|
|
|
#else
|
|
|
|
#define N_EARLY_MOUNT 4
|
|
|
|
#endif
|
2011-07-29 01:48:18 +02:00
|
|
|
|
2010-04-12 21:58:01 +02:00
|
|
|
static const MountPoint mount_table[] = {
|
2020-07-30 13:36:10 +02:00
|
|
|
{ "sysfs", "/sys", "sysfs", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
|
2014-11-26 15:43:20 +01:00
|
|
|
NULL, MNT_FATAL|MNT_IN_CONTAINER },
|
2020-07-30 13:36:10 +02:00
|
|
|
{ "proc", "/proc", "proc", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
|
2014-11-26 15:43:20 +01:00
|
|
|
NULL, MNT_FATAL|MNT_IN_CONTAINER },
|
2020-07-30 13:36:10 +02:00
|
|
|
{ "devtmpfs", "/dev", "devtmpfs", "mode=755" TMPFS_LIMITS_DEV, MS_NOSUID|MS_NOEXEC|MS_STRICTATIME,
|
2014-11-26 15:43:20 +01:00
|
|
|
NULL, MNT_FATAL|MNT_IN_CONTAINER },
|
2020-07-30 13:36:10 +02:00
|
|
|
{ "securityfs", "/sys/kernel/security", "securityfs", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
|
2014-11-26 15:43:20 +01:00
|
|
|
NULL, MNT_NONE },
|
2017-10-03 12:22:40 +02:00
|
|
|
#if ENABLE_SMACK
|
2020-07-30 13:36:10 +02:00
|
|
|
{ "smackfs", "/sys/fs/smackfs", "smackfs", "smackfsdef=*", MS_NOSUID|MS_NOEXEC|MS_NODEV,
|
2014-11-26 15:43:20 +01:00
|
|
|
mac_smack_use, MNT_FATAL },
|
2020-07-30 13:36:10 +02:00
|
|
|
{ "tmpfs", "/dev/shm", "tmpfs", "mode=1777,smackfsroot=*", MS_NOSUID|MS_NODEV|MS_STRICTATIME,
|
2014-11-26 15:43:20 +01:00
|
|
|
mac_smack_use, MNT_FATAL },
|
2013-09-26 23:41:09 +02:00
|
|
|
#endif
|
2020-07-30 13:36:10 +02:00
|
|
|
{ "tmpfs", "/dev/shm", "tmpfs", "mode=1777", MS_NOSUID|MS_NODEV|MS_STRICTATIME,
|
2014-11-26 15:43:20 +01:00
|
|
|
NULL, MNT_FATAL|MNT_IN_CONTAINER },
|
2020-07-30 13:36:10 +02:00
|
|
|
{ "devpts", "/dev/pts", "devpts", "mode=620,gid=" STRINGIFY(TTY_GID), MS_NOSUID|MS_NOEXEC,
|
2014-11-26 15:43:20 +01:00
|
|
|
NULL, MNT_IN_CONTAINER },
|
2017-10-03 12:22:40 +02:00
|
|
|
#if ENABLE_SMACK
|
2020-07-30 13:36:10 +02:00
|
|
|
{ "tmpfs", "/run", "tmpfs", "mode=755,smackfsroot=*" TMPFS_LIMITS_RUN, MS_NOSUID|MS_NODEV|MS_STRICTATIME,
|
2014-11-26 15:43:20 +01:00
|
|
|
mac_smack_use, MNT_FATAL },
|
2013-09-26 23:41:09 +02:00
|
|
|
#endif
|
2020-07-30 13:36:10 +02:00
|
|
|
{ "tmpfs", "/run", "tmpfs", "mode=755" TMPFS_LIMITS_RUN, MS_NOSUID|MS_NODEV|MS_STRICTATIME,
|
2014-11-26 15:43:20 +01:00
|
|
|
NULL, MNT_FATAL|MNT_IN_CONTAINER },
|
2020-07-23 12:56:32 +02:00
|
|
|
{ "cgroup2", "/sys/fs/cgroup", "cgroup2", "nsdelegate,memory_recursiveprot", MS_NOSUID|MS_NOEXEC|MS_NODEV,
|
|
|
|
cg_is_unified_wanted, MNT_IN_CONTAINER|MNT_CHECK_WRITABLE },
|
2020-07-30 13:36:10 +02:00
|
|
|
{ "cgroup2", "/sys/fs/cgroup", "cgroup2", "nsdelegate", MS_NOSUID|MS_NOEXEC|MS_NODEV,
|
2017-11-22 17:27:03 +01:00
|
|
|
cg_is_unified_wanted, MNT_IN_CONTAINER|MNT_CHECK_WRITABLE },
|
2020-07-30 13:36:10 +02:00
|
|
|
{ "cgroup2", "/sys/fs/cgroup", "cgroup2", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
|
2017-11-22 17:27:03 +01:00
|
|
|
cg_is_unified_wanted, MNT_IN_CONTAINER|MNT_CHECK_WRITABLE },
|
2020-07-30 13:36:10 +02:00
|
|
|
{ "tmpfs", "/sys/fs/cgroup", "tmpfs", "mode=755" TMPFS_LIMITS_SYS_FS_CGROUP, MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME,
|
core: unified cgroup hierarchy support
This patch set adds full support the new unified cgroup hierarchy logic
of modern kernels.
A new kernel command line option "systemd.unified_cgroup_hierarchy=1" is
added. If specified the unified hierarchy is mounted to /sys/fs/cgroup
instead of a tmpfs. No further hierarchies are mounted. The kernel
command line option defaults to off. We can turn it on by default as
soon as the kernel's APIs regarding this are stabilized (but even then
downstream distros might want to turn this off, as this will break any
tools that access cgroupfs directly).
It is possibly to choose for each boot individually whether the unified
or the legacy hierarchy is used. nspawn will by default provide the
legacy hierarchy to containers if the host is using it, and the unified
otherwise. However it is possible to run containers with the unified
hierarchy on a legacy host and vice versa, by setting the
$UNIFIED_CGROUP_HIERARCHY environment variable for nspawn to 1 or 0,
respectively.
The unified hierarchy provides reliable cgroup empty notifications for
the first time, via inotify. To make use of this we maintain one
manager-wide inotify fd, and each cgroup to it.
This patch also removes cg_delete() which is unused now.
On kernel 4.2 only the "memory" controller is compatible with the
unified hierarchy, hence that's the only controller systemd exposes when
booted in unified heirarchy mode.
This introduces a new enum for enumerating supported controllers, plus a
related enum for the mask bits mapping to it. The core is changed to
make use of this everywhere.
This moves PID 1 into a new "init.scope" implicit scope unit in the root
slice. This is necessary since on the unified hierarchy cgroups may
either contain subgroups or processes but not both. PID 1 hence has to
move out of the root cgroup (strictly speaking the root cgroup is the
only one where processes and subgroups are still allowed, but in order
to support containers nicey, we move PID 1 into the new scope in all
cases.) This new unit is also used on legacy hierarchy setups. It's
actually pretty useful on all systems, as it can then be used to filter
journal messages coming from PID 1, and so on.
The root slice ("-.slice") is now implicitly created and started (and
does not require a unit file on disk anymore), since
that's where "init.scope" is located and the slice needs to be started
before the scope can.
To check whether we are in unified or legacy hierarchy mode we use
statfs() on /sys/fs/cgroup. If the .f_type field reports tmpfs we are in
legacy mode, if it reports cgroupfs we are in unified mode.
This patch set carefuly makes sure that cgls and cgtop continue to work
as desired.
When invoking nspawn as a service it will implicitly create two
subcgroups in the cgroup it is using, one to move the nspawn process
into, the other to move the actual container processes into. This is
done because of the requirement that cgroups may either contain
processes or other subgroups.
2015-09-01 19:22:36 +02:00
|
|
|
cg_is_legacy_wanted, MNT_FATAL|MNT_IN_CONTAINER },
|
2020-07-30 13:36:10 +02:00
|
|
|
{ "cgroup2", "/sys/fs/cgroup/unified", "cgroup2", "nsdelegate", MS_NOSUID|MS_NOEXEC|MS_NODEV,
|
2017-11-22 17:20:35 +01:00
|
|
|
cg_is_hybrid_wanted, MNT_IN_CONTAINER|MNT_CHECK_WRITABLE },
|
2020-07-30 13:36:10 +02:00
|
|
|
{ "cgroup2", "/sys/fs/cgroup/unified", "cgroup2", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
|
2017-11-22 17:20:35 +01:00
|
|
|
cg_is_hybrid_wanted, MNT_IN_CONTAINER|MNT_CHECK_WRITABLE },
|
2020-07-30 13:36:10 +02:00
|
|
|
{ "cgroup", "/sys/fs/cgroup/systemd", "cgroup", "none,name=systemd,xattr", MS_NOSUID|MS_NOEXEC|MS_NODEV,
|
2016-11-21 20:45:53 +01:00
|
|
|
cg_is_legacy_wanted, MNT_IN_CONTAINER },
|
2020-07-30 13:36:10 +02:00
|
|
|
{ "cgroup", "/sys/fs/cgroup/systemd", "cgroup", "none,name=systemd", MS_NOSUID|MS_NOEXEC|MS_NODEV,
|
2016-11-21 20:45:53 +01:00
|
|
|
cg_is_legacy_wanted, MNT_FATAL|MNT_IN_CONTAINER },
|
2020-07-30 13:36:10 +02:00
|
|
|
{ "pstore", "/sys/fs/pstore", "pstore", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
|
2014-11-26 15:43:20 +01:00
|
|
|
NULL, MNT_NONE },
|
2017-10-03 10:41:51 +02:00
|
|
|
#if ENABLE_EFI
|
2020-07-30 13:36:10 +02:00
|
|
|
{ "efivarfs", "/sys/firmware/efi/efivars", "efivarfs", NULL, MS_NOSUID|MS_NOEXEC|MS_NODEV,
|
2014-11-26 15:43:20 +01:00
|
|
|
is_efi_boot, MNT_NONE },
|
2013-02-27 13:07:09 +01:00
|
|
|
#endif
|
2020-07-30 13:36:10 +02:00
|
|
|
{ "bpf", "/sys/fs/bpf", "bpf", "mode=700", MS_NOSUID|MS_NOEXEC|MS_NODEV,
|
2018-02-16 14:59:26 +01:00
|
|
|
NULL, MNT_NONE, },
|
sd-bus: sync with kdbus upstream (ABI break)
kdbus has seen a larger update than expected lately, most notably with
kdbusfs, a file system to expose the kdbus control files:
* Each time a file system of this type is mounted, a new kdbus
domain is created.
* The layout inside each mount point is the same as before, except
that domains are not hierarchically nested anymore.
* Domains are therefore also unnamed now.
* Unmounting a kdbusfs will automatically also detroy the
associated domain.
* Hence, the action of creating a kdbus domain is now as
privileged as mounting a filesystem.
* This way, we can get around creating dev nodes for everything,
which is last but not least something that is not limited by
20-bit minor numbers.
The kdbus specific bits in nspawn have all been dropped now, as nspawn
can rely on the container OS to set up its own kdbus domain, simply by
mounting a new instance.
A new set of mounts has been added to mount things *after* the kernel
modules have been loaded. For now, only kdbus is in this set, which is
invoked with mount_setup_late().
2014-11-13 20:33:03 +01:00
|
|
|
};
|
|
|
|
|
2010-07-16 02:56:57 +02:00
|
|
|
/* These are API file systems that might be mounted by other software,
|
2010-11-24 22:27:45 +01:00
|
|
|
* we just list them here so that we know that we should ignore them */
|
2010-07-16 02:56:57 +02:00
|
|
|
|
2012-04-22 01:00:57 +02:00
|
|
|
static const char ignore_paths[] =
|
|
|
|
/* SELinux file systems */
|
|
|
|
"/sys/fs/selinux\0"
|
|
|
|
/* Container bind mounts */
|
|
|
|
"/proc/sys\0"
|
|
|
|
"/dev/console\0"
|
2012-09-18 12:05:47 +02:00
|
|
|
"/proc/kmsg\0";
|
2010-07-16 02:56:57 +02:00
|
|
|
|
2010-04-10 17:42:00 +02:00
|
|
|
bool mount_point_is_api(const char *path) {
|
|
|
|
unsigned i;
|
|
|
|
|
|
|
|
/* Checks if this mount point is considered "API", and hence
|
|
|
|
* should be ignored */
|
|
|
|
|
2010-04-12 21:58:01 +02:00
|
|
|
for (i = 0; i < ELEMENTSOF(mount_table); i ++)
|
2010-08-20 03:26:15 +02:00
|
|
|
if (path_equal(path, mount_table[i].where))
|
2010-04-10 17:42:00 +02:00
|
|
|
return true;
|
|
|
|
|
2010-11-11 11:15:16 +01:00
|
|
|
return path_startswith(path, "/sys/fs/cgroup/");
|
|
|
|
}
|
|
|
|
|
|
|
|
bool mount_point_ignore(const char *path) {
|
2012-04-22 01:00:57 +02:00
|
|
|
const char *i;
|
2010-11-11 11:15:16 +01:00
|
|
|
|
2012-04-22 01:00:57 +02:00
|
|
|
NULSTR_FOREACH(i, ignore_paths)
|
|
|
|
if (path_equal(path, i))
|
2010-07-16 02:56:57 +02:00
|
|
|
return true;
|
|
|
|
|
2010-11-11 11:15:16 +01:00
|
|
|
return false;
|
2010-04-10 17:42:00 +02:00
|
|
|
}
|
|
|
|
|
2011-07-29 01:48:18 +02:00
|
|
|
static int mount_one(const MountPoint *p, bool relabel) {
|
2017-12-15 17:37:16 +01:00
|
|
|
int r, priority;
|
2010-03-31 16:29:55 +02:00
|
|
|
|
2010-04-12 21:58:01 +02:00
|
|
|
assert(p);
|
2010-03-31 16:29:55 +02:00
|
|
|
|
2017-12-15 17:37:16 +01:00
|
|
|
priority = (p->mode & MNT_FATAL) ? LOG_ERR : LOG_DEBUG;
|
|
|
|
|
2012-11-04 17:03:48 +01:00
|
|
|
if (p->condition_fn && !p->condition_fn())
|
|
|
|
return 0;
|
|
|
|
|
2011-04-06 23:38:01 +02:00
|
|
|
/* Relabel first, just in case */
|
2011-07-29 01:48:18 +02:00
|
|
|
if (relabel)
|
2018-03-27 07:38:26 +02:00
|
|
|
(void) label_fix(p->where, LABEL_IGNORE_ENOENT|LABEL_IGNORE_EROFS);
|
2011-04-06 23:38:01 +02:00
|
|
|
|
2016-11-18 21:35:21 +01:00
|
|
|
r = path_is_mount_point(p->where, NULL, AT_SYMLINK_FOLLOW);
|
2016-02-03 23:53:08 +01:00
|
|
|
if (r < 0 && r != -ENOENT) {
|
2017-12-15 17:37:16 +01:00
|
|
|
log_full_errno(priority, r, "Failed to determine whether %s is a mount point: %m", p->where);
|
2016-02-03 23:53:08 +01:00
|
|
|
return (p->mode & MNT_FATAL) ? r : 0;
|
|
|
|
}
|
2010-03-31 16:29:55 +02:00
|
|
|
if (r > 0)
|
2011-04-06 23:38:01 +02:00
|
|
|
return 0;
|
2010-03-31 16:29:55 +02:00
|
|
|
|
2012-09-18 12:05:47 +02:00
|
|
|
/* Skip securityfs in a container */
|
2015-09-07 13:42:47 +02:00
|
|
|
if (!(p->mode & MNT_IN_CONTAINER) && detect_container() > 0)
|
2012-09-18 12:05:47 +02:00
|
|
|
return 0;
|
|
|
|
|
2010-04-06 21:55:58 +02:00
|
|
|
/* The access mode here doesn't really matter too much, since
|
|
|
|
* the mounted file system will take precedence anyway. */
|
2014-03-24 21:04:02 +01:00
|
|
|
if (relabel)
|
2016-02-03 23:53:08 +01:00
|
|
|
(void) mkdir_p_label(p->where, 0755);
|
2014-03-24 21:04:02 +01:00
|
|
|
else
|
2016-02-03 23:53:08 +01:00
|
|
|
(void) mkdir_p(p->where, 0755);
|
2010-04-06 21:55:58 +02:00
|
|
|
|
2010-03-31 16:29:55 +02:00
|
|
|
log_debug("Mounting %s to %s of type %s with options %s.",
|
2010-04-12 21:58:01 +02:00
|
|
|
p->what,
|
|
|
|
p->where,
|
|
|
|
p->type,
|
|
|
|
strna(p->options));
|
|
|
|
|
|
|
|
if (mount(p->what,
|
|
|
|
p->where,
|
|
|
|
p->type,
|
|
|
|
p->flags,
|
|
|
|
p->options) < 0) {
|
2017-12-15 17:37:16 +01:00
|
|
|
log_full_errno(priority, errno, "Failed to mount %s at %s: %m", p->type, p->where);
|
2012-11-04 17:03:48 +01:00
|
|
|
return (p->mode & MNT_FATAL) ? -errno : 0;
|
2010-03-31 16:29:55 +02:00
|
|
|
}
|
|
|
|
|
2011-04-06 23:38:01 +02:00
|
|
|
/* Relabel again, since we now mounted something fresh here */
|
2011-07-29 01:48:18 +02:00
|
|
|
if (relabel)
|
2018-03-27 07:38:26 +02:00
|
|
|
(void) label_fix(p->where, 0);
|
2010-09-15 01:38:07 +02:00
|
|
|
|
2017-11-22 17:20:35 +01:00
|
|
|
if (p->mode & MNT_CHECK_WRITABLE) {
|
2017-12-15 17:37:16 +01:00
|
|
|
if (access(p->where, W_OK) < 0) {
|
|
|
|
r = -errno;
|
|
|
|
|
2017-11-22 17:20:35 +01:00
|
|
|
(void) umount(p->where);
|
2017-11-24 21:05:16 +01:00
|
|
|
(void) rmdir(p->where);
|
2017-12-15 17:37:16 +01:00
|
|
|
|
|
|
|
log_full_errno(priority, r, "Mount point %s not writable after mounting: %m", p->where);
|
2017-11-22 17:20:35 +01:00
|
|
|
return (p->mode & MNT_FATAL) ? r : 0;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2011-08-23 00:37:35 +02:00
|
|
|
return 1;
|
2010-03-31 16:29:55 +02:00
|
|
|
}
|
|
|
|
|
2016-02-02 20:03:12 +01:00
|
|
|
static int mount_points_setup(unsigned n, bool loaded_policy) {
|
2011-07-29 01:48:18 +02:00
|
|
|
unsigned i;
|
|
|
|
int r = 0;
|
|
|
|
|
2016-02-02 20:03:12 +01:00
|
|
|
for (i = 0; i < n; i ++) {
|
2011-07-29 01:48:18 +02:00
|
|
|
int j;
|
|
|
|
|
2016-02-02 20:03:12 +01:00
|
|
|
j = mount_one(mount_table + i, loaded_policy);
|
2015-09-21 19:59:07 +02:00
|
|
|
if (j != 0 && r >= 0)
|
2011-07-29 01:48:18 +02:00
|
|
|
r = j;
|
|
|
|
}
|
|
|
|
|
|
|
|
return r;
|
|
|
|
}
|
|
|
|
|
2016-02-02 20:03:12 +01:00
|
|
|
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);
|
|
|
|
}
|
|
|
|
|
2018-11-15 21:07:43 +01:00
|
|
|
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) {
|
2013-04-23 14:28:10 +02:00
|
|
|
_cleanup_set_free_free_ Set *controllers = NULL;
|
2014-03-18 04:06:36 +01:00
|
|
|
int r;
|
2010-04-16 23:22:32 +02:00
|
|
|
|
core: unified cgroup hierarchy support
This patch set adds full support the new unified cgroup hierarchy logic
of modern kernels.
A new kernel command line option "systemd.unified_cgroup_hierarchy=1" is
added. If specified the unified hierarchy is mounted to /sys/fs/cgroup
instead of a tmpfs. No further hierarchies are mounted. The kernel
command line option defaults to off. We can turn it on by default as
soon as the kernel's APIs regarding this are stabilized (but even then
downstream distros might want to turn this off, as this will break any
tools that access cgroupfs directly).
It is possibly to choose for each boot individually whether the unified
or the legacy hierarchy is used. nspawn will by default provide the
legacy hierarchy to containers if the host is using it, and the unified
otherwise. However it is possible to run containers with the unified
hierarchy on a legacy host and vice versa, by setting the
$UNIFIED_CGROUP_HIERARCHY environment variable for nspawn to 1 or 0,
respectively.
The unified hierarchy provides reliable cgroup empty notifications for
the first time, via inotify. To make use of this we maintain one
manager-wide inotify fd, and each cgroup to it.
This patch also removes cg_delete() which is unused now.
On kernel 4.2 only the "memory" controller is compatible with the
unified hierarchy, hence that's the only controller systemd exposes when
booted in unified heirarchy mode.
This introduces a new enum for enumerating supported controllers, plus a
related enum for the mask bits mapping to it. The core is changed to
make use of this everywhere.
This moves PID 1 into a new "init.scope" implicit scope unit in the root
slice. This is necessary since on the unified hierarchy cgroups may
either contain subgroups or processes but not both. PID 1 hence has to
move out of the root cgroup (strictly speaking the root cgroup is the
only one where processes and subgroups are still allowed, but in order
to support containers nicey, we move PID 1 into the new scope in all
cases.) This new unit is also used on legacy hierarchy setups. It's
actually pretty useful on all systems, as it can then be used to filter
journal messages coming from PID 1, and so on.
The root slice ("-.slice") is now implicitly created and started (and
does not require a unit file on disk anymore), since
that's where "init.scope" is located and the slice needs to be started
before the scope can.
To check whether we are in unified or legacy hierarchy mode we use
statfs() on /sys/fs/cgroup. If the .f_type field reports tmpfs we are in
legacy mode, if it reports cgroupfs we are in unified mode.
This patch set carefuly makes sure that cgls and cgtop continue to work
as desired.
When invoking nspawn as a service it will implicitly create two
subcgroups in the cgroup it is using, one to move the nspawn process
into, the other to move the actual container processes into. This is
done because of the requirement that cgroups may either contain
processes or other subgroups.
2015-09-01 19:22:36 +02:00
|
|
|
if (!cg_is_legacy_wanted())
|
|
|
|
return 0;
|
|
|
|
|
2010-05-18 03:10:17 +02:00
|
|
|
/* Mount all available cgroup controllers that are built into the kernel. */
|
2017-11-17 16:27:13 +01:00
|
|
|
r = cg_kernel_controllers(&controllers);
|
2014-12-30 01:57:23 +01:00
|
|
|
if (r < 0)
|
|
|
|
return log_error_errno(r, "Failed to enumerate cgroup controllers: %m");
|
2011-08-23 00:37:35 +02:00
|
|
|
|
|
|
|
for (;;) {
|
2014-03-18 04:06:36 +01:00
|
|
|
_cleanup_free_ char *options = NULL, *controller = NULL, *where = NULL;
|
2018-11-15 21:07:43 +01:00
|
|
|
const char *other_controller;
|
2013-04-23 14:28:10 +02:00
|
|
|
MountPoint p = {
|
|
|
|
.what = "cgroup",
|
|
|
|
.type = "cgroup",
|
|
|
|
.flags = MS_NOSUID|MS_NOEXEC|MS_NODEV,
|
|
|
|
.mode = MNT_IN_CONTAINER,
|
|
|
|
};
|
2011-08-23 00:37:35 +02:00
|
|
|
|
|
|
|
controller = set_steal_first(controllers);
|
|
|
|
if (!controller)
|
|
|
|
break;
|
|
|
|
|
2018-11-15 21:07:43 +01:00
|
|
|
/* 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();
|
2011-08-23 00:37:35 +02:00
|
|
|
}
|
2018-11-15 21:07:43 +01:00
|
|
|
}
|
2011-08-23 00:37:35 +02:00
|
|
|
|
2018-11-15 21:07:43 +01:00
|
|
|
/* The simple case, where there's only one controller to mount together */
|
|
|
|
if (!options)
|
2018-03-22 16:53:26 +01:00
|
|
|
options = TAKE_PTR(controller);
|
2011-08-23 00:37:35 +02:00
|
|
|
|
2019-07-11 19:14:16 +02:00
|
|
|
where = path_join("/sys/fs/cgroup", options);
|
2014-03-18 04:06:36 +01:00
|
|
|
if (!where)
|
|
|
|
return log_oom();
|
|
|
|
|
|
|
|
p.where = where;
|
2011-08-23 00:37:35 +02:00
|
|
|
p.options = options;
|
2010-04-16 23:22:32 +02:00
|
|
|
|
2011-07-29 01:48:18 +02:00
|
|
|
r = mount_one(&p, true);
|
2013-04-23 14:28:10 +02:00
|
|
|
if (r < 0)
|
|
|
|
return r;
|
2011-08-23 00:37:35 +02:00
|
|
|
|
2018-11-15 21:07:43 +01:00
|
|
|
/* 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);
|
2010-04-16 23:22:32 +02:00
|
|
|
}
|
|
|
|
|
2018-11-15 21:07:43 +01:00
|
|
|
/* Now that we mounted everything, let's make the tmpfs the cgroup file systems are mounted into read-only. */
|
2020-04-14 15:39:36 +02:00
|
|
|
(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);
|
2014-03-18 04:06:54 +01:00
|
|
|
|
2013-04-23 14:28:10 +02:00
|
|
|
return 0;
|
2010-04-16 23:22:32 +02:00
|
|
|
}
|
|
|
|
|
2017-10-03 12:22:40 +02:00
|
|
|
#if HAVE_SELINUX || ENABLE_SMACK
|
2010-11-08 04:59:39 +01:00
|
|
|
static int nftw_cb(
|
|
|
|
const char *fpath,
|
|
|
|
const struct stat *sb,
|
|
|
|
int tflag,
|
|
|
|
struct FTW *ftwbuf) {
|
|
|
|
|
2010-11-11 01:22:42 +01:00
|
|
|
/* No need to label /dev twice in a row... */
|
2011-08-30 00:16:00 +02:00
|
|
|
if (_unlikely_(ftwbuf->level == 0))
|
|
|
|
return FTW_CONTINUE;
|
|
|
|
|
2018-03-27 07:38:26 +02:00
|
|
|
(void) label_fix(fpath, 0);
|
2011-08-30 18:49:17 +02:00
|
|
|
|
2011-08-30 00:16:00 +02:00
|
|
|
/* /run/initramfs is static data and big, no need to
|
2011-08-30 18:49:17 +02:00
|
|
|
* dynamically relabel its contents at boot... */
|
2011-08-30 00:16:00 +02:00
|
|
|
if (_unlikely_(ftwbuf->level == 1 &&
|
|
|
|
tflag == FTW_D &&
|
|
|
|
streq(fpath, "/run/initramfs")))
|
|
|
|
return FTW_SKIP_SUBTREE;
|
2010-11-11 01:22:42 +01:00
|
|
|
|
2011-08-30 00:16:00 +02:00
|
|
|
return FTW_CONTINUE;
|
2010-11-08 04:59:39 +01:00
|
|
|
};
|
2018-03-28 13:36:33 +02:00
|
|
|
|
|
|
|
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. */
|
2018-03-28 17:07:42 +02:00
|
|
|
if (statfs("/sys/fs/cgroup", &st) < 0)
|
2018-03-28 13:36:33 +02:00
|
|
|
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);
|
2018-03-28 17:07:42 +02:00
|
|
|
(void) nftw("/sys/fs/cgroup", nftw_cb, 64, FTW_MOUNT|FTW_PHYS|FTW_ACTIONRETVAL);
|
2018-03-28 13:36:33 +02:00
|
|
|
|
|
|
|
if (st.f_flags & ST_RDONLY)
|
|
|
|
(void) mount(NULL, "/sys/fs/cgroup", NULL, MS_REMOUNT|MS_RDONLY, NULL);
|
2018-03-28 17:07:42 +02:00
|
|
|
|
2018-03-28 13:36:33 +02:00
|
|
|
} else if (r < 0)
|
|
|
|
return log_error_errno(r, "Failed to determine whether we are in all unified mode: %m");
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
2018-12-12 14:02:00 +01:00
|
|
|
|
|
|
|
static int relabel_extra(void) {
|
2019-08-06 20:22:23 +02:00
|
|
|
_cleanup_strv_free_ char **files = NULL;
|
|
|
|
char **file;
|
2018-12-12 14:02:00 +01:00
|
|
|
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.
|
|
|
|
*/
|
|
|
|
|
2019-08-06 20:22:23 +02:00
|
|
|
r = conf_files_list(&files, ".relabel", NULL,
|
|
|
|
CONF_FILES_FILTER_MASKED | CONF_FILES_REGULAR,
|
2019-08-19 16:31:27 +02:00
|
|
|
"/run/systemd/relabel-extra.d/");
|
2019-08-06 20:22:23 +02:00
|
|
|
if (r < 0)
|
|
|
|
return log_error_errno(r, "Failed to enumerate /run/systemd/relabel-extra.d/, ignoring: %m");
|
2018-12-12 14:02:00 +01:00
|
|
|
|
2019-08-06 20:22:23 +02:00
|
|
|
STRV_FOREACH(file, files) {
|
2018-12-12 14:02:00 +01:00
|
|
|
_cleanup_fclose_ FILE *f = NULL;
|
|
|
|
|
2019-08-06 20:22:23 +02:00
|
|
|
f = fopen(*file, "re");
|
2018-12-12 14:02:00 +01:00
|
|
|
if (!f) {
|
2019-08-06 20:22:23 +02:00
|
|
|
log_warning_errno(errno, "Failed to open %s, ignoring: %m", *file);
|
2018-12-12 14:02:00 +01:00
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
|
|
|
for (;;) {
|
|
|
|
_cleanup_free_ char *line = NULL;
|
|
|
|
|
|
|
|
r = read_line(f, LONG_LINE_MAX, &line);
|
|
|
|
if (r < 0) {
|
2019-08-06 20:22:23 +02:00
|
|
|
log_warning_errno(r, "Failed to read %s, ignoring: %m", *file);
|
2018-12-12 14:02:00 +01:00
|
|
|
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);
|
2019-09-11 05:23:59 +02:00
|
|
|
(void) label_fix(line, 0);
|
2018-12-12 14:02:00 +01:00
|
|
|
(void) nftw(line, nftw_cb, 64, FTW_MOUNT|FTW_PHYS|FTW_ACTIONRETVAL);
|
|
|
|
c++;
|
|
|
|
}
|
|
|
|
|
2019-08-06 20:22:23 +02:00
|
|
|
if (unlink(*file) < 0)
|
|
|
|
log_warning_errno(errno, "Failed to remove %s, ignoring: %m", *file);
|
2018-12-12 14:02:00 +01:00
|
|
|
}
|
|
|
|
|
2019-08-06 20:22:23 +02:00
|
|
|
/* Remove when we complete things. */
|
2019-09-18 17:52:30 +02:00
|
|
|
if (rmdir("/run/systemd/relabel-extra.d") < 0 &&
|
|
|
|
errno != ENOENT)
|
2018-12-12 14:02:00 +01:00
|
|
|
log_warning_errno(errno, "Failed to remove /run/systemd/relabel-extra.d/ directory: %m");
|
|
|
|
|
|
|
|
return c;
|
|
|
|
}
|
2014-09-18 15:24:55 +02:00
|
|
|
#endif
|
2010-11-08 04:59:39 +01:00
|
|
|
|
2020-03-23 03:42:46 +01:00
|
|
|
int mount_setup(bool loaded_policy, bool leave_propagation) {
|
2020-08-19 17:47:32 +02:00
|
|
|
int r;
|
2010-03-31 16:29:55 +02:00
|
|
|
|
2016-02-02 20:03:12 +01:00
|
|
|
r = mount_points_setup(ELEMENTSOF(mount_table), loaded_policy);
|
2014-11-26 15:43:20 +01:00
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
2017-10-03 12:22:40 +02:00
|
|
|
#if HAVE_SELINUX || ENABLE_SMACK
|
2011-04-07 21:22:41 +02:00
|
|
|
/* 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. */
|
2011-07-28 23:52:23 +02:00
|
|
|
if (loaded_policy) {
|
|
|
|
usec_t before_relabel, after_relabel;
|
|
|
|
char timespan[FORMAT_TIMESPAN_MAX];
|
2018-12-12 14:41:56 +01:00
|
|
|
const char *i;
|
2018-12-12 14:02:00 +01:00
|
|
|
int n_extra;
|
2011-07-28 23:52:23 +02:00
|
|
|
|
|
|
|
before_relabel = now(CLOCK_MONOTONIC);
|
|
|
|
|
2018-12-12 14:41:56 +01:00
|
|
|
FOREACH_STRING(i, "/dev", "/dev/shm", "/run")
|
|
|
|
(void) nftw(i, nftw_cb, 64, FTW_MOUNT|FTW_PHYS|FTW_ACTIONRETVAL);
|
2011-07-28 23:52:23 +02:00
|
|
|
|
2018-12-12 20:45:24 +01:00
|
|
|
(void) relabel_cgroup_filesystems();
|
2017-11-30 11:59:29 +01:00
|
|
|
|
2018-12-12 14:02:00 +01:00
|
|
|
n_extra = relabel_extra();
|
|
|
|
|
2011-07-28 23:52:23 +02:00
|
|
|
after_relabel = now(CLOCK_MONOTONIC);
|
|
|
|
|
2018-12-12 14:02:00 +01:00
|
|
|
log_info("Relabelled /dev, /dev/shm, /run, /sys/fs/cgroup%s in %s.",
|
|
|
|
n_extra > 0 ? ", additional files" : "",
|
2013-04-04 02:56:56 +02:00
|
|
|
format_timespan(timespan, sizeof(timespan), after_relabel - before_relabel, 0));
|
2011-04-04 16:56:51 +02:00
|
|
|
}
|
2014-09-18 15:24:55 +02:00
|
|
|
#endif
|
2010-11-08 04:59:39 +01:00
|
|
|
|
2010-10-27 05:47:48 +02:00
|
|
|
/* Create a few default symlinks, which are normally created
|
2011-04-07 21:22:41 +02:00
|
|
|
* by udevd, but some scripts might need them before we start
|
2010-10-27 05:47:48 +02:00
|
|
|
* udevd. */
|
2015-05-21 16:30:58 +02:00
|
|
|
dev_setup(NULL, UID_INVALID, GID_INVALID);
|
2010-10-27 05:47:48 +02:00
|
|
|
|
2016-12-08 10:51:32 +01:00
|
|
|
/* 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). */
|
2020-03-23 03:42:46 +01:00
|
|
|
if (detect_container() <= 0 && !leave_propagation)
|
2012-09-18 12:05:47 +02:00
|
|
|
if (mount(NULL, "/", NULL, MS_REC|MS_SHARED, NULL) < 0)
|
2014-11-28 19:29:59 +01:00
|
|
|
log_warning_errno(errno, "Failed to set up the root directory for shared mount propagation: %m");
|
2012-08-06 18:28:42 +02:00
|
|
|
|
2016-12-08 10:51:32 +01:00
|
|
|
/* 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. */
|
2016-07-06 09:48:58 +02:00
|
|
|
(void) mkdir_label("/run/systemd", 0755);
|
|
|
|
(void) mkdir_label("/run/systemd/system", 0755);
|
2016-12-08 10:51:32 +01:00
|
|
|
|
2020-08-14 15:54:48 +02:00
|
|
|
/* Make sure we have a mount point to hide in sandboxes */
|
|
|
|
(void) mkdir_label("/run/credentials", 0755);
|
|
|
|
|
2020-06-09 16:22:24 +02:00
|
|
|
/* Also create /run/systemd/inaccessible nodes, so that we always have something to mount
|
2020-08-14 18:56:54 +02:00
|
|
|
* inaccessible nodes from. If we run in a container the host might have created these for us already
|
2020-09-14 15:32:37 +02:00
|
|
|
* in /run/host/inaccessible/. Use those if we can, since that way we likely get access to block/char
|
2020-08-14 18:56:54 +02:00
|
|
|
* device nodes that are inaccessible, and if userns is used to nodes that are on mounts owned by a
|
|
|
|
* userns outside the container and thus nicely read-only and not remountable. */
|
|
|
|
if (access("/run/host/inaccessible/", F_OK) < 0) {
|
|
|
|
if (errno != ENOENT)
|
|
|
|
log_debug_errno(errno, "Failed to check if /run/host/inaccessible exists, ignoring: %m");
|
|
|
|
|
|
|
|
(void) make_inaccessible_nodes("/run/systemd", UID_INVALID, GID_INVALID);
|
|
|
|
} else
|
|
|
|
(void) symlink("../host/inaccessible", "/run/systemd/inaccessible");
|
2018-05-03 16:10:35 +02:00
|
|
|
|
2011-08-23 00:37:35 +02:00
|
|
|
return 0;
|
2010-03-31 16:29:55 +02:00
|
|
|
}
|