int fdset_steal_first(FDSet *fds) {
void *p;
p = set_steal_first(MAKE_SET(fds));
if (!p)
return -ENOENT;
return PTR_TO_FD(p);
}
DnsTransaction* dns_transaction_free(DnsTransaction *t) {
DnsQuery *q;
DnsZoneItem *i;
if (!t)
return NULL;
sd_event_source_unref(t->timeout_event_source);
dns_packet_unref(t->sent);
dns_packet_unref(t->received);
dns_answer_unref(t->cached);
sd_event_source_unref(t->dns_udp_event_source);
safe_close(t->dns_udp_fd);
dns_server_unref(t->server);
dns_stream_free(t->stream);
if (t->scope) {
hashmap_remove(t->scope->transactions, t->key);
if (t->id != 0)
hashmap_remove(t->scope->manager->dns_transactions, UINT_TO_PTR(t->id));
}
dns_resource_key_unref(t->key);
while ((q = set_steal_first(t->queries)))
set_remove(q->transactions, t);
set_free(t->queries);
while ((i = set_steal_first(t->zone_items)))
i->probe_transaction = NULL;
set_free(t->zone_items);
free(t);
return NULL;
}
static void context_free(Context *context) {
sd_event_source *es;
Connection *c;
assert(context);
while ((es = set_steal_first(context->listen)))
sd_event_source_unref(es);
while ((c = set_first(context->connections)))
connection_free(c);
set_free(context->listen);
set_free(context->connections);
}
int automount_send_ready(Automount *a, int status) {
int ioctl_fd, r;
unsigned token;
assert(a);
assert(status <= 0);
if (set_isempty(a->tokens))
return 0;
ioctl_fd = open_ioctl_fd(UNIT(a)->manager->dev_autofs_fd, a->where, a->dev_id);
if (ioctl_fd < 0) {
r = ioctl_fd;
goto fail;
}
if (status)
log_debug_unit(UNIT(a)->id, "Sending failure: %s", strerror(-status));
else
log_debug_unit(UNIT(a)->id, "Sending success.");
r = 0;
/* Autofs thankfully does not hand out 0 as a token */
while ((token = PTR_TO_UINT(set_steal_first(a->tokens)))) {
int k;
/* Autofs fun fact II:
*
* if you pass a positive status code here, the kernel will
* freeze! Yay! */
k = autofs_send_ready(UNIT(a)->manager->dev_autofs_fd,
ioctl_fd,
token,
status);
if (k < 0)
r = k;
}
fail:
if (ioctl_fd >= 0)
close_nointr_nofail(ioctl_fd);
return r;
}
int introspect_write_child_nodes(struct introspect *i, Set *s, const char *prefix) {
char *node;
assert(i);
assert(prefix);
while ((node = set_steal_first(s))) {
const char *e;
e = object_path_startswith(node, prefix);
if (e && e[0])
fprintf(i->f, " <node name=\"%s\"/>\n", e);
free(node);
}
return 0;
}
void manager_etc_hosts_flush(Manager *m) {
EtcHostsItem *item;
EtcHostsItemByName *bn;
while ((item = set_steal_first(m->etc_hosts_by_address))) {
strv_free(item->names);
free(item);
}
while ((bn = hashmap_steal_first(m->etc_hosts_by_name))) {
free(bn->name);
free(bn->items);
free(bn);
}
m->etc_hosts_by_address = set_free(m->etc_hosts_by_address);
m->etc_hosts_by_name = hashmap_free(m->etc_hosts_by_name);
m->etc_hosts_mtime = USEC_INFINITY;
}
static int automount_send_ready(Automount *a, Set *tokens, int status) {
_cleanup_close_ int ioctl_fd = -1;
unsigned token;
int r;
assert(a);
assert(status <= 0);
if (set_isempty(tokens))
return 0;
ioctl_fd = open_ioctl_fd(UNIT(a)->manager->dev_autofs_fd, a->where, a->dev_id);
if (ioctl_fd < 0)
return ioctl_fd;
if (status != 0)
log_unit_debug_errno(UNIT(a), status, "Sending failure: %m");
else
log_unit_debug(UNIT(a), "Sending success.");
r = 0;
/* Autofs thankfully does not hand out 0 as a token */
while ((token = PTR_TO_UINT(set_steal_first(tokens)))) {
int k;
/* Autofs fun fact II:
*
* if you pass a positive status code here, the kernel will
* freeze! Yay! */
k = autofs_send_ready(UNIT(a)->manager->dev_autofs_fd,
ioctl_fd,
token,
status);
if (k < 0)
r = k;
}
return r;
}
void fdset_free(FDSet *s) {
void *p;
while ((p = set_steal_first(MAKE_SET(s)))) {
/* Valgrind's fd might have ended up in this set here,
* due to fdset_new_fill(). We'll ignore all failures
* here, so that the EBADFD that valgrind will return
* us on close() doesn't influence us */
/* When reloading duplicates of the private bus
* connection fds and suchlike are closed here, which
* has no effect at all, since they are only
* duplicates. So don't be surprised about these log
* messages. */
log_debug("Closing left-over fd %i", PTR_TO_FD(p));
close_nointr(PTR_TO_FD(p));
}
set_free(MAKE_SET(s));
}
int mount_cgroup_controllers(char ***join_controllers) {
int r;
FILE *f;
char buf[LINE_MAX];
Set *controllers;
/* Mount all available cgroup controllers that are built into the kernel. */
f = fopen("/proc/cgroups", "re");
if (!f) {
log_error("Failed to enumerate cgroup controllers: %m");
return 0;
}
controllers = set_new(string_hash_func, string_compare_func);
if (!controllers) {
r = log_oom();
goto finish;
}
/* Ignore the header line */
(void) fgets(buf, sizeof(buf), f);
for (;;) {
char *controller;
int enabled = 0;
if (fscanf(f, "%ms %*i %*i %i", &controller, &enabled) != 2) {
if (feof(f))
break;
log_error("Failed to parse /proc/cgroups.");
r = -EIO;
goto finish;
}
if (!enabled) {
free(controller);
continue;
}
r = set_put(controllers, controller);
if (r < 0) {
log_error("Failed to add controller to set.");
free(controller);
goto finish;
}
}
for (;;) {
MountPoint p;
char *controller, *where, *options;
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)) {
char *t;
t = set_remove(controllers, *i);
if (!t) {
free(*i);
continue;
}
free(t);
}
*(j++) = *i;
}
*j = NULL;
options = strv_join(*k, ",");
if (!options) {
free(controller);
r = log_oom();
goto finish;
}
} else {
options = controller;
controller = NULL;
}
where = strappend("/sys/fs/cgroup/", options);
if (!where) {
free(options);
r = log_oom();
goto finish;
}
zero(p);
p.what = "cgroup";
p.where = where;
p.type = "cgroup";
p.options = options;
p.flags = MS_NOSUID|MS_NOEXEC|MS_NODEV;
p.mode = MNT_IN_CONTAINER;
r = mount_one(&p, true);
free(controller);
free(where);
if (r < 0) {
free(options);
goto finish;
}
if (r > 0 && k && *k) {
char **i;
for (i = *k; *i; i++) {
char *t;
t = strappend("/sys/fs/cgroup/", *i);
if (!t) {
r = log_oom();
free(options);
goto finish;
}
r = symlink(options, t);
free(t);
if (r < 0 && errno != EEXIST) {
log_error("Failed to create symlink: %m");
r = -errno;
free(options);
goto finish;
}
}
}
free(options);
}
r = 0;
finish:
set_free_free(controllers);
fclose(f);
return r;
}
int mount_cgroup_controllers(char ***join_controllers) {
_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. */
controllers = set_new(&string_hash_ops);
if (!controllers)
return log_oom();
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;
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;
}
*j = NULL;
options = strv_join(*k, ",");
if (!options)
return log_oom();
} else {
options = controller;
controller = NULL;
}
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;
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 && errno != EEXIST)
return log_error_errno(errno, "Failed to create symlink %s: %m", t);
#ifdef SMACK_RUN_LABEL
r = mac_smack_copy(t, options);
if (r < 0 && r != -EOPNOTSUPP)
return log_error_errno(r, "Failed to copy smack label from %s to %s: %m", options, t);
#endif
}
}
}
/* 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 defined(HAVE_SELINUX) || defined(HAVE_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;
label_fix(fpath, false, false);
/* /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;
};
#endif
int mount_setup(bool loaded_policy) {
unsigned i;
int r = 0;
for (i = 0; i < ELEMENTSOF(mount_table); i ++) {
int j;
j = mount_one(mount_table + i, loaded_policy);
if (j != 0 && r >= 0)
r = j;
}
if (r < 0)
return r;
#if defined(HAVE_SELINUX) || defined(HAVE_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);
nftw("/dev", nftw_cb, 64, FTW_MOUNT|FTW_PHYS|FTW_ACTIONRETVAL);
nftw("/run", nftw_cb, 64, FTW_MOUNT|FTW_PHYS|FTW_ACTIONRETVAL);
after_relabel = now(CLOCK_MONOTONIC);
log_info("Relabelled /dev and /run 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. */
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. */
mkdir_label("/run/systemd", 0755);
mkdir_label("/run/systemd/system", 0755);
mkdir_label("/run/systemd/inaccessible", 0000);
return 0;
}
int mount_cgroup_controllers(char ***join_controllers) {
int r;
char buf[LINE_MAX];
_cleanup_set_free_free_ Set *controllers = NULL;
_cleanup_fclose_ FILE *f;
/* Mount all available cgroup controllers that are built into the kernel. */
f = fopen("/proc/cgroups", "re");
if (!f) {
log_error("Failed to enumerate cgroup controllers: %m");
return 0;
}
controllers = set_new(string_hash_func, string_compare_func);
if (!controllers)
return log_oom();
/* Ignore the header line */
(void) fgets(buf, sizeof(buf), f);
for (;;) {
char *controller;
int enabled = 0;
if (fscanf(f, "%ms %*i %*i %i", &controller, &enabled) != 2) {
if (feof(f))
break;
log_error("Failed to parse /proc/cgroups.");
return -EIO;
}
if (!enabled) {
free(controller);
continue;
}
r = set_consume(controllers, controller);
if (r < 0) {
log_error("Failed to add controller to set.");
return r;
}
}
for (;;) {
MountPoint p = {
.what = "cgroup",
.type = "cgroup",
.flags = MS_NOSUID|MS_NOEXEC|MS_NODEV,
.mode = MNT_IN_CONTAINER,
};
char ***k = NULL;
_cleanup_free_ char *options = NULL, *controller;
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)) {
char _cleanup_free_ *t;
t = set_remove(controllers, *i);
if (!t) {
free(*i);
continue;
}
}
*(j++) = *i;
}
*j = NULL;
options = strv_join(*k, ",");
if (!options)
return log_oom();
} else {
options = controller;
controller = NULL;
}
p.where = strappenda("/sys/fs/cgroup/", options);
p.options = options;
r = mount_one(&p, true);
if (r < 0)
return r;
if (r > 0 && k && *k) {
char **i;
for (i = *k; *i; i++) {
char *t = strappenda("/sys/fs/cgroup/", *i);
r = symlink(options, t);
if (r < 0 && errno != EEXIST) {
log_error("Failed to create symlink %s: %m", t);
return -errno;
}
}
}
}
return 0;
}
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;
label_fix(fpath, false, false);
/* /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;
};
int mount_setup(bool loaded_policy) {
int r;
unsigned i;
for (i = 0; i < ELEMENTSOF(mount_table); i ++) {
r = mount_one(mount_table + i, true);
if (r < 0)
return r;
}
/* 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);
nftw("/dev", nftw_cb, 64, FTW_MOUNT|FTW_PHYS|FTW_ACTIONRETVAL);
nftw("/run", nftw_cb, 64, FTW_MOUNT|FTW_PHYS|FTW_ACTIONRETVAL);
after_relabel = now(CLOCK_MONOTONIC);
log_info("Relabelled /dev and /run in %s.",
format_timespan(timespan, sizeof(timespan), after_relabel - before_relabel, 0));
}
/* Create a few default symlinks, which are normally created
* by udevd, but some scripts might need them before we start
* udevd. */
dev_setup(NULL);
/* 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. */
if (detect_container(NULL) <= 0)
if (mount(NULL, "/", NULL, MS_REC|MS_SHARED, NULL) < 0)
log_warning("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. */
mkdir_label("/run/systemd", 0755);
mkdir_label("/run/systemd/system", 0755);
mkdir_label("/run/systemd/inaccessible", 0000);
return 0;
}
/* Use this function only if you do not have direct access to /proc/self/mountinfo but the caller can open it
* for you. This is the case when /proc is masked or not mounted. Otherwise, use bind_remount_recursive. */
int bind_remount_recursive_with_mountinfo(
const char *prefix,
unsigned long new_flags,
unsigned long flags_mask,
char **blacklist,
FILE *proc_self_mountinfo) {
_cleanup_set_free_free_ Set *done = NULL;
_cleanup_free_ char *cleaned = NULL;
int r;
assert(proc_self_mountinfo);
/* Recursively remount a directory (and all its submounts) read-only or read-write. If the directory is already
* mounted, we reuse the mount and simply mark it MS_BIND|MS_RDONLY (or remove the MS_RDONLY for read-write
* operation). If it isn't we first make it one. Afterwards we apply MS_BIND|MS_RDONLY (or remove MS_RDONLY) to
* all submounts we can access, too. When mounts are stacked on the same mount point we only care for each
* individual "top-level" mount on each point, as we cannot influence/access the underlying mounts anyway. We
* do not have any effect on future submounts that might get propagated, they migt be writable. This includes
* future submounts that have been triggered via autofs.
*
* If the "blacklist" parameter is specified it may contain a list of subtrees to exclude from the
* remount operation. Note that we'll ignore the blacklist for the top-level path. */
cleaned = strdup(prefix);
if (!cleaned)
return -ENOMEM;
path_simplify(cleaned, false);
done = set_new(&path_hash_ops);
if (!done)
return -ENOMEM;
for (;;) {
_cleanup_set_free_free_ Set *todo = NULL;
bool top_autofs = false;
char *x;
unsigned long orig_flags;
todo = set_new(&path_hash_ops);
if (!todo)
return -ENOMEM;
rewind(proc_self_mountinfo);
for (;;) {
_cleanup_free_ char *path = NULL, *p = NULL, *type = NULL;
int k;
k = fscanf(proc_self_mountinfo,
"%*s " /* (1) mount id */
"%*s " /* (2) parent id */
"%*s " /* (3) major:minor */
"%*s " /* (4) root */
"%ms " /* (5) mount point */
"%*s" /* (6) mount options (superblock) */
"%*[^-]" /* (7) optional fields */
"- " /* (8) separator */
"%ms " /* (9) file system type */
"%*s" /* (10) mount source */
"%*s" /* (11) mount options (bind mount) */
"%*[^\n]", /* some rubbish at the end */
&path,
&type);
if (k != 2) {
if (k == EOF)
break;
continue;
}
r = cunescape(path, UNESCAPE_RELAX, &p);
if (r < 0)
return r;
if (!path_startswith(p, cleaned))
continue;
/* Ignore this mount if it is blacklisted, but only if it isn't the top-level mount we shall
* operate on. */
if (!path_equal(cleaned, p)) {
bool blacklisted = false;
char **i;
STRV_FOREACH(i, blacklist) {
if (path_equal(*i, cleaned))
continue;
if (!path_startswith(*i, cleaned))
continue;
if (path_startswith(p, *i)) {
blacklisted = true;
log_debug("Not remounting %s blacklisted by %s, called for %s", p, *i, cleaned);
break;
}
}
if (blacklisted)
continue;
}
/* Let's ignore autofs mounts. If they aren't
* triggered yet, we want to avoid triggering
* them, as we don't make any guarantees for
* future submounts anyway. If they are
* already triggered, then we will find
* another entry for this. */
if (streq(type, "autofs")) {
top_autofs = top_autofs || path_equal(cleaned, p);
continue;
}
if (!set_contains(done, p)) {
r = set_consume(todo, p);
p = NULL;
if (r == -EEXIST)
continue;
if (r < 0)
return r;
}
}
/* If we have no submounts to process anymore and if
* the root is either already done, or an autofs, we
* are done */
if (set_isempty(todo) &&
(top_autofs || set_contains(done, cleaned)))
return 0;
if (!set_contains(done, cleaned) &&
!set_contains(todo, cleaned)) {
/* The prefix directory itself is not yet a mount, make it one. */
if (mount(cleaned, cleaned, NULL, MS_BIND|MS_REC, NULL) < 0)
return -errno;
orig_flags = 0;
(void) get_mount_flags(cleaned, &orig_flags);
orig_flags &= ~MS_RDONLY;
if (mount(NULL, cleaned, NULL, (orig_flags & ~flags_mask)|MS_BIND|MS_REMOUNT|new_flags, NULL) < 0)
return -errno;
log_debug("Made top-level directory %s a mount point.", prefix);
r = set_put_strdup(done, cleaned);
if (r < 0)
return r;
}
while ((x = set_steal_first(todo))) {
r = set_consume(done, x);
if (IN_SET(r, 0, -EEXIST))
continue;
if (r < 0)
return r;
/* Deal with mount points that are obstructed by a later mount */
r = path_is_mount_point(x, NULL, 0);
if (IN_SET(r, 0, -ENOENT))
continue;
if (IN_SET(r, -EACCES, -EPERM)) {
/* Even if root user invoke this, submounts under private FUSE or NFS mount points
* may not be acceessed. E.g.,
*
* $ bindfs --no-allow-other ~/mnt/mnt ~/mnt/mnt
* $ bindfs --no-allow-other ~/mnt ~/mnt
*
* Then, root user cannot access the mount point ~/mnt/mnt.
* In such cases, the submounts are ignored, as we have no way to manage them. */
log_debug_errno(r, "Failed to determine '%s' is mount point or not, ignoring: %m", x);
continue;
}
if (r < 0)
return r;
/* Try to reuse the original flag set */
orig_flags = 0;
(void) get_mount_flags(x, &orig_flags);
orig_flags &= ~MS_RDONLY;
if (mount(NULL, x, NULL, (orig_flags & ~flags_mask)|MS_BIND|MS_REMOUNT|new_flags, NULL) < 0)
return -errno;
log_debug("Remounted %s read-only.", x);
}
}
/* Mount legacy cgroup hierarchy when cgroup namespaces are unsupported. */
static int mount_legacy_cgns_unsupported(
const char *dest,
CGroupUnified unified_requested,
bool userns,
uid_t uid_shift,
uid_t uid_range,
const char *selinux_apifs_context) {
_cleanup_set_free_free_ Set *controllers = NULL;
const char *cgroup_root;
int r;
cgroup_root = prefix_roota(dest, "/sys/fs/cgroup");
(void) mkdir_p(cgroup_root, 0755);
/* Mount a tmpfs to /sys/fs/cgroup if it's not mounted there yet. */
r = path_is_mount_point(cgroup_root, dest, AT_SYMLINK_FOLLOW);
if (r < 0)
return log_error_errno(r, "Failed to determine if /sys/fs/cgroup is already mounted: %m");
if (r == 0) {
_cleanup_free_ char *options = NULL;
r = tmpfs_patch_options("mode=755", uid_shift == 0 ? UID_INVALID : uid_shift, selinux_apifs_context, &options);
if (r < 0)
return log_oom();
r = mount_verbose(LOG_ERR, "tmpfs", cgroup_root, "tmpfs",
MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME, options);
if (r < 0)
return r;
}
r = cg_all_unified();
if (r < 0)
return r;
if (r > 0)
goto skip_controllers;
r = cg_kernel_controllers(&controllers);
if (r < 0)
return log_error_errno(r, "Failed to determine cgroup controllers: %m");
for (;;) {
_cleanup_free_ char *controller = NULL, *origin = NULL, *combined = NULL;
controller = set_steal_first(controllers);
if (!controller)
break;
origin = prefix_root("/sys/fs/cgroup/", controller);
if (!origin)
return log_oom();
r = readlink_malloc(origin, &combined);
if (r == -EINVAL) {
/* Not a symbolic link, but directly a single cgroup hierarchy */
r = mount_legacy_cgroup_hierarchy(dest, controller, controller, true);
if (r < 0)
return r;
} else if (r < 0)
return log_error_errno(r, "Failed to read link %s: %m", origin);
else {
_cleanup_free_ char *target = NULL;
target = prefix_root(dest, origin);
if (!target)
return log_oom();
/* A symbolic link, a combination of controllers in one hierarchy */
if (!filename_is_valid(combined)) {
log_warning("Ignoring invalid combined hierarchy %s.", combined);
continue;
}
r = mount_legacy_cgroup_hierarchy(dest, combined, combined, true);
if (r < 0)
return r;
r = symlink_idempotent(combined, target, false);
if (r == -EINVAL)
return log_error_errno(r, "Invalid existing symlink for combined hierarchy: %m");
if (r < 0)
return log_error_errno(r, "Failed to create symlink for combined hierarchy: %m");
}
}
skip_controllers:
if (unified_requested >= CGROUP_UNIFIED_SYSTEMD) {
r = mount_legacy_cgroup_hierarchy(dest, SYSTEMD_CGROUP_CONTROLLER_HYBRID, "unified", false);
if (r < 0)
return r;
}
r = mount_legacy_cgroup_hierarchy(dest, SYSTEMD_CGROUP_CONTROLLER_LEGACY, "systemd", false);
if (r < 0)
return r;
return mount_verbose(LOG_ERR, NULL, cgroup_root, NULL,
MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME|MS_RDONLY, "mode=755");
}
/* Mount a legacy cgroup hierarchy when cgroup namespaces are supported. */
static int mount_legacy_cgns_supported(
const char *dest,
CGroupUnified unified_requested,
bool userns,
uid_t uid_shift,
uid_t uid_range,
const char *selinux_apifs_context) {
_cleanup_set_free_free_ Set *controllers = NULL;
const char *cgroup_root = "/sys/fs/cgroup", *c;
int r;
(void) mkdir_p(cgroup_root, 0755);
/* Mount a tmpfs to /sys/fs/cgroup if it's not mounted there yet. */
r = path_is_mount_point(cgroup_root, dest, AT_SYMLINK_FOLLOW);
if (r < 0)
return log_error_errno(r, "Failed to determine if /sys/fs/cgroup is already mounted: %m");
if (r == 0) {
_cleanup_free_ char *options = NULL;
/* When cgroup namespaces are enabled and user namespaces are
* used then the mount of the cgroupfs is done *inside* the new
* user namespace. We're root in the new user namespace and the
* kernel will happily translate our uid/gid to the correct
* uid/gid as seen from e.g. /proc/1/mountinfo. So we simply
* pass uid 0 and not uid_shift to tmpfs_patch_options().
*/
r = tmpfs_patch_options("mode=755", 0, selinux_apifs_context, &options);
if (r < 0)
return log_oom();
r = mount_verbose(LOG_ERR, "tmpfs", cgroup_root, "tmpfs",
MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME, options);
if (r < 0)
return r;
}
r = cg_all_unified();
if (r < 0)
return r;
if (r > 0)
goto skip_controllers;
r = get_process_controllers(&controllers);
if (r < 0)
return log_error_errno(r, "Failed to determine cgroup controllers: %m");
for (;;) {
_cleanup_free_ const char *controller = NULL;
controller = set_steal_first(controllers);
if (!controller)
break;
r = mount_legacy_cgroup_hierarchy("", controller, controller, !userns);
if (r < 0)
return r;
/* When multiple hierarchies are co-mounted, make their
* constituting individual hierarchies a symlink to the
* co-mount.
*/
c = controller;
for (;;) {
_cleanup_free_ char *target = NULL, *tok = NULL;
r = extract_first_word(&c, &tok, ",", 0);
if (r < 0)
return log_error_errno(r, "Failed to extract co-mounted cgroup controller: %m");
if (r == 0)
break;
if (streq(controller, tok))
break;
target = prefix_root("/sys/fs/cgroup/", tok);
if (!target)
return log_oom();
r = symlink_idempotent(controller, target, false);
if (r == -EINVAL)
return log_error_errno(r, "Invalid existing symlink for combined hierarchy: %m");
if (r < 0)
return log_error_errno(r, "Failed to create symlink for combined hierarchy: %m");
}
}
skip_controllers:
if (unified_requested >= CGROUP_UNIFIED_SYSTEMD) {
r = mount_legacy_cgroup_hierarchy("", SYSTEMD_CGROUP_CONTROLLER_HYBRID, "unified", false);
if (r < 0)
return r;
}
r = mount_legacy_cgroup_hierarchy("", SYSTEMD_CGROUP_CONTROLLER_LEGACY, "systemd", false);
if (r < 0)
return r;
if (!userns)
return mount_verbose(LOG_ERR, NULL, cgroup_root, NULL,
MS_REMOUNT|MS_NOSUID|MS_NOEXEC|MS_NODEV|MS_STRICTATIME|MS_RDONLY, "mode=755");
return 0;
}
static void test_struct(void) {
Prioq *q;
Set *s;
unsigned previous = 0, i;
int r;
srand(0);
q = prioq_new(test_compare);
assert_se(q);
s = set_new(test_hash, test_compare);
assert_se(s);
for (i = 0; i < SET_SIZE; i++) {
struct test *t;
t = new0(struct test, 1);
assert_se(t);
t->value = (unsigned) rand();
r = prioq_put(q, t, &t->idx);
assert_se(r >= 0);
if (i % 4 == 0) {
r = set_consume(s, t);
assert_se(r >= 0);
}
}
for (;;) {
struct test *t;
t = set_steal_first(s);
if (!t)
break;
r = prioq_remove(q, t, &t->idx);
assert_se(r > 0);
free(t);
}
for (i = 0; i < SET_SIZE * 3 / 4; i++) {
struct test *t;
assert_se(prioq_size(q) == (SET_SIZE * 3 / 4) - i);
t = prioq_pop(q);
assert_se(t);
assert_se(previous <= t->value);
previous = t->value;
free(t);
}
assert_se(prioq_isempty(q));
prioq_free(q);
assert_se(set_isempty(s));
set_free(s);
}