SymbX/systemd
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Merge pull request #13246 from keszybz/add-SystemdOptions-efi-variable

Browse Source 
Add efi variable to augment /proc/cmdline
This commit is contained in:
Zbigniew Jędrzejewski-Szmek
2019-10-03 12:19:44 +02:00
committed by GitHub
parent 68e70ac2b2 4e5aa79185
commit 86e94d95d0
58 changed files with 1608 additions and 1380 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/basic/cgroup-util.c
+22 -873
View File
@@ -31,7 +31,6 @@
#include "mkdir.h"
#include "parse-util.h"
#include "path-util.h"
#include "proc-cmdline.h"
#include "process-util.h"
#include "set.h"
#include "special.h"
@@ -408,173 +407,6 @@ int cg_kill_recursive(
return ret;
}
int cg_migrate(
const char *cfrom,
const char *pfrom,
const char *cto,
const char *pto,
CGroupFlags flags) {
bool done = false;
_cleanup_set_free_ Set *s = NULL;
int r, ret = 0;
pid_t my_pid;
assert(cfrom);
assert(pfrom);
assert(cto);
assert(pto);
s = set_new(NULL);
if (!s)
return -ENOMEM;
my_pid = getpid_cached();
do {
_cleanup_fclose_ FILE *f = NULL;
pid_t pid = 0;
done = true;
r = cg_enumerate_processes(cfrom, pfrom, &f);
if (r < 0) {
if (ret >= 0 && r != -ENOENT)
return r;
return ret;
}
while ((r = cg_read_pid(f, &pid)) > 0) {
/* This might do weird stuff if we aren't a
* single-threaded program. However, we
* luckily know we are not */
if ((flags & CGROUP_IGNORE_SELF) && pid == my_pid)
continue;
if (set_get(s, PID_TO_PTR(pid)) == PID_TO_PTR(pid))
continue;
/* Ignore kernel threads. Since they can only
* exist in the root cgroup, we only check for
* them there. */
if (cfrom &&
empty_or_root(pfrom) &&
is_kernel_thread(pid) > 0)
continue;
r = cg_attach(cto, pto, pid);
if (r < 0) {
if (ret >= 0 && r != -ESRCH)
ret = r;
} else if (ret == 0)
ret = 1;
done = false;
r = set_put(s, PID_TO_PTR(pid));
if (r < 0) {
if (ret >= 0)
return r;
return ret;
}
}
if (r < 0) {
if (ret >= 0)
return r;
return ret;
}
} while (!done);
return ret;
}
int cg_migrate_recursive(
const char *cfrom,
const char *pfrom,
const char *cto,
const char *pto,
CGroupFlags flags) {
_cleanup_closedir_ DIR *d = NULL;
int r, ret = 0;
char *fn;
assert(cfrom);
assert(pfrom);
assert(cto);
assert(pto);
ret = cg_migrate(cfrom, pfrom, cto, pto, flags);
r = cg_enumerate_subgroups(cfrom, pfrom, &d);
if (r < 0) {
if (ret >= 0 && r != -ENOENT)
return r;
return ret;
}
while ((r = cg_read_subgroup(d, &fn)) > 0) {
_cleanup_free_ char *p = NULL;
p = path_join(empty_to_root(pfrom), fn);
free(fn);
if (!p)
return -ENOMEM;
r = cg_migrate_recursive(cfrom, p, cto, pto, flags);
if (r != 0 && ret >= 0)
ret = r;
}
if (r < 0 && ret >= 0)
ret = r;
if (flags & CGROUP_REMOVE) {
r = cg_rmdir(cfrom, pfrom);
if (r < 0 && ret >= 0 && !IN_SET(r, -ENOENT, -EBUSY))
return r;
}
return ret;
}
int cg_migrate_recursive_fallback(
const char *cfrom,
const char *pfrom,
const char *cto,
const char *pto,
CGroupFlags flags) {
int r;
assert(cfrom);
assert(pfrom);
assert(cto);
assert(pto);
r = cg_migrate_recursive(cfrom, pfrom, cto, pto, flags);
if (r < 0) {
char prefix[strlen(pto) + 1];
/* This didn't work? Then let's try all prefixes of the destination */
PATH_FOREACH_PREFIX(prefix, pto) {
int q;
q = cg_migrate_recursive(cfrom, pfrom, cto, prefix, flags);
if (q >= 0)
return q;
}
}
return r;
}
static const char *controller_to_dirname(const char *controller) {
const char *e;
@@ -740,253 +572,6 @@ int cg_get_path_and_check(const char *controller, const char *path, const char *
return cg_get_path(controller, path, suffix, fs);
}
static int trim_cb(const char *path, const struct stat *sb, int typeflag, struct FTW *ftwbuf) {
assert(path);
assert(sb);
assert(ftwbuf);
if (typeflag != FTW_DP)
return 0;
if (ftwbuf->level < 1)
return 0;
(void) rmdir(path);
return 0;
}
int cg_trim(const char *controller, const char *path, bool delete_root) {
_cleanup_free_ char *fs = NULL;
int r = 0, q;
assert(path);
r = cg_get_path(controller, path, NULL, &fs);
if (r < 0)
return r;
errno = 0;
if (nftw(fs, trim_cb, 64, FTW_DEPTH|FTW_MOUNT|FTW_PHYS) != 0) {
if (errno == ENOENT)
r = 0;
else
r = errno_or_else(EIO);
}
if (delete_root) {
if (rmdir(fs) < 0 && errno != ENOENT)
return -errno;
}
q = cg_hybrid_unified();
if (q < 0)
return q;
if (q > 0 && streq(controller, SYSTEMD_CGROUP_CONTROLLER)) {
q = cg_trim(SYSTEMD_CGROUP_CONTROLLER_LEGACY, path, delete_root);
if (q < 0)
log_warning_errno(q, "Failed to trim compat systemd cgroup %s: %m", path);
}
return r;
}
/* Create a cgroup in the hierarchy of controller.
* Returns 0 if the group already existed, 1 on success, negative otherwise.
*/
int cg_create(const char *controller, const char *path) {
_cleanup_free_ char *fs = NULL;
int r;
r = cg_get_path_and_check(controller, path, NULL, &fs);
if (r < 0)
return r;
r = mkdir_parents(fs, 0755);
if (r < 0)
return r;
r = mkdir_errno_wrapper(fs, 0755);
if (r == -EEXIST)
return 0;
if (r < 0)
return r;
r = cg_hybrid_unified();
if (r < 0)
return r;
if (r > 0 && streq(controller, SYSTEMD_CGROUP_CONTROLLER)) {
r = cg_create(SYSTEMD_CGROUP_CONTROLLER_LEGACY, path);
if (r < 0)
log_warning_errno(r, "Failed to create compat systemd cgroup %s: %m", path);
}
return 1;
}
int cg_create_and_attach(const char *controller, const char *path, pid_t pid) {
int r, q;
assert(pid >= 0);
r = cg_create(controller, path);
if (r < 0)
return r;
q = cg_attach(controller, path, pid);
if (q < 0)
return q;
/* This does not remove the cgroup on failure */
return r;
}
int cg_attach(const char *controller, const char *path, pid_t pid) {
_cleanup_free_ char *fs = NULL;
char c[DECIMAL_STR_MAX(pid_t) + 2];
int r;
assert(path);
assert(pid >= 0);
r = cg_get_path_and_check(controller, path, "cgroup.procs", &fs);
if (r < 0)
return r;
if (pid == 0)
pid = getpid_cached();
xsprintf(c, PID_FMT "\n", pid);
r = write_string_file(fs, c, WRITE_STRING_FILE_DISABLE_BUFFER);
if (r < 0)
return r;
r = cg_hybrid_unified();
if (r < 0)
return r;
if (r > 0 && streq(controller, SYSTEMD_CGROUP_CONTROLLER)) {
r = cg_attach(SYSTEMD_CGROUP_CONTROLLER_LEGACY, path, pid);
if (r < 0)
log_warning_errno(r, "Failed to attach "PID_FMT" to compat systemd cgroup %s: %m", pid, path);
}
return 0;
}
int cg_attach_fallback(const char *controller, const char *path, pid_t pid) {
int r;
assert(controller);
assert(path);
assert(pid >= 0);
r = cg_attach(controller, path, pid);
if (r < 0) {
char prefix[strlen(path) + 1];
/* This didn't work? Then let's try all prefixes of
* the destination */
PATH_FOREACH_PREFIX(prefix, path) {
int q;
q = cg_attach(controller, prefix, pid);
if (q >= 0)
return q;
}
}
return r;
}
int cg_set_access(
const char *controller,
const char *path,
uid_t uid,
gid_t gid) {
struct Attribute {
const char *name;
bool fatal;
};
/* cgroup v1, aka legacy/non-unified */
static const struct Attribute legacy_attributes[] = {
{ "cgroup.procs", true },
{ "tasks", false },
{ "cgroup.clone_children", false },
{},
};
/* cgroup v2, aka unified */
static const struct Attribute unified_attributes[] = {
{ "cgroup.procs", true },
{ "cgroup.subtree_control", true },
{ "cgroup.threads", false },
{},
};
static const struct Attribute* const attributes[] = {
[false] = legacy_attributes,
[true] = unified_attributes,
};
_cleanup_free_ char *fs = NULL;
const struct Attribute *i;
int r, unified;
assert(path);
if (uid == UID_INVALID && gid == GID_INVALID)
return 0;
unified = cg_unified_controller(controller);
if (unified < 0)
return unified;
/* Configure access to the cgroup itself */
r = cg_get_path(controller, path, NULL, &fs);
if (r < 0)
return r;
r = chmod_and_chown(fs, 0755, uid, gid);
if (r < 0)
return r;
/* Configure access to the cgroup's attributes */
for (i = attributes[unified]; i->name; i++) {
fs = mfree(fs);
r = cg_get_path(controller, path, i->name, &fs);
if (r < 0)
return r;
r = chmod_and_chown(fs, 0644, uid, gid);
if (r < 0) {
if (i->fatal)
return r;
log_debug_errno(r, "Failed to set access on cgroup %s, ignoring: %m", fs);
}
}
if (streq(controller, SYSTEMD_CGROUP_CONTROLLER)) {
r = cg_hybrid_unified();
if (r < 0)
return r;
if (r > 0) {
/* Always propagate access mode from unified to legacy controller */
r = cg_set_access(SYSTEMD_CGROUP_CONTROLLER_LEGACY, path, uid, gid);
if (r < 0)
log_debug_errno(r, "Failed to set access on compatibility systemd cgroup %s, ignoring: %m", path);
}
}
return 0;
}
int cg_set_xattr(const char *controller, const char *path, const char *name, const void *value, size_t size, int flags) {
_cleanup_free_ char *fs = NULL;
int r;
@@ -2141,194 +1726,6 @@ fail:
done:
memcpy(ret_values, v, sizeof(char*) * n);
return 0;
}
int cg_create_everywhere(CGroupMask supported, CGroupMask mask, const char *path) {
CGroupController c;
CGroupMask done;
bool created;
int r;
/* This one will create a cgroup in our private tree, but also
* duplicate it in the trees specified in mask, and remove it
* in all others.
*
* Returns 0 if the group already existed in the systemd hierarchy,
* 1 on success, negative otherwise.
*/
/* First create the cgroup in our own hierarchy. */
r = cg_create(SYSTEMD_CGROUP_CONTROLLER, path);
if (r < 0)
return r;
created = r;
/* If we are in the unified hierarchy, we are done now */
r = cg_all_unified();
if (r < 0)
return r;
if (r > 0)
return created;
supported &= CGROUP_MASK_V1;
mask = CGROUP_MASK_EXTEND_JOINED(mask);
done = 0;
/* Otherwise, do the same in the other hierarchies */
for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c);
const char *n;
if (!FLAGS_SET(supported, bit))
continue;
if (FLAGS_SET(done, bit))
continue;
n = cgroup_controller_to_string(c);
if (FLAGS_SET(mask, bit))
(void) cg_create(n, path);
else
(void) cg_trim(n, path, true);
done |= CGROUP_MASK_EXTEND_JOINED(bit);
}
return created;
}
int cg_attach_everywhere(CGroupMask supported, const char *path, pid_t pid, cg_migrate_callback_t path_callback, void *userdata) {
CGroupController c;
CGroupMask done;
int r;
r = cg_attach(SYSTEMD_CGROUP_CONTROLLER, path, pid);
if (r < 0)
return r;
r = cg_all_unified();
if (r < 0)
return r;
if (r > 0)
return 0;
supported &= CGROUP_MASK_V1;
done = 0;
for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c);
const char *p = NULL;
if (!FLAGS_SET(supported, bit))
continue;
if (FLAGS_SET(done, bit))
continue;
if (path_callback)
p = path_callback(bit, userdata);
if (!p)
p = path;
(void) cg_attach_fallback(cgroup_controller_to_string(c), p, pid);
done |= CGROUP_MASK_EXTEND_JOINED(bit);
}
return 0;
}
int cg_attach_many_everywhere(CGroupMask supported, const char *path, Set* pids, cg_migrate_callback_t path_callback, void *userdata) {
Iterator i;
void *pidp;
int r = 0;
SET_FOREACH(pidp, pids, i) {
pid_t pid = PTR_TO_PID(pidp);
int q;
q = cg_attach_everywhere(supported, path, pid, path_callback, userdata);
if (q < 0 && r >= 0)
r = q;
}
return r;
}
int cg_migrate_everywhere(CGroupMask supported, const char *from, const char *to, cg_migrate_callback_t to_callback, void *userdata) {
CGroupController c;
CGroupMask done;
int r = 0, q;
if (!path_equal(from, to)) {
r = cg_migrate_recursive(SYSTEMD_CGROUP_CONTROLLER, from, SYSTEMD_CGROUP_CONTROLLER, to, CGROUP_REMOVE);
if (r < 0)
return r;
}
q = cg_all_unified();
if (q < 0)
return q;
if (q > 0)
return r;
supported &= CGROUP_MASK_V1;
done = 0;
for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c);
const char *p = NULL;
if (!FLAGS_SET(supported, bit))
continue;
if (FLAGS_SET(done, bit))
continue;
if (to_callback)
p = to_callback(bit, userdata);
if (!p)
p = to;
(void) cg_migrate_recursive_fallback(SYSTEMD_CGROUP_CONTROLLER, to, cgroup_controller_to_string(c), p, 0);
done |= CGROUP_MASK_EXTEND_JOINED(bit);
}
return r;
}
int cg_trim_everywhere(CGroupMask supported, const char *path, bool delete_root) {
CGroupController c;
CGroupMask done;
int r, q;
r = cg_trim(SYSTEMD_CGROUP_CONTROLLER, path, delete_root);
if (r < 0)
return r;
q = cg_all_unified();
if (q < 0)
return q;
if (q > 0)
return r;
supported &= CGROUP_MASK_V1;
done = 0;
for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c);
if (!FLAGS_SET(supported, bit))
continue;
if (FLAGS_SET(done, bit))
continue;
(void) cg_trim(cgroup_controller_to_string(c), path, delete_root);
done |= CGROUP_MASK_EXTEND_JOINED(bit);
}
return r;
}
int cg_mask_to_string(CGroupMask mask, char **ret) {
@@ -2523,20 +1920,20 @@ int cg_kernel_controllers(Set **ret) {
return 0;
}
static thread_local CGroupUnified unified_cache = CGROUP_UNIFIED_UNKNOWN;
/* The hybrid mode was initially implemented in v232 and simply mounted cgroup2 on /sys/fs/cgroup/systemd. This
* unfortunately broke other tools (such as docker) which expected the v1 "name=systemd" hierarchy on
* /sys/fs/cgroup/systemd. From v233 and on, the hybrid mode mountnbs v2 on /sys/fs/cgroup/unified and maintains
* "name=systemd" hierarchy on /sys/fs/cgroup/systemd for compatibility with other tools.
/* The hybrid mode was initially implemented in v232 and simply mounted cgroup2 on
* /sys/fs/cgroup/systemd. This unfortunately broke other tools (such as docker) which expected the v1
* "name=systemd" hierarchy on /sys/fs/cgroup/systemd. From v233 and on, the hybrid mode mounts v2 on
* /sys/fs/cgroup/unified and maintains "name=systemd" hierarchy on /sys/fs/cgroup/systemd for compatibility
* with other tools.
*
* To keep live upgrade working, we detect and support v232 layout. When v232 layout is detected, to keep cgroup v2
* process management but disable the compat dual layout, we return %true on
* cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) and %false on cg_hybrid_unified().
* To keep live upgrade working, we detect and support v232 layout. When v232 layout is detected, to keep
* cgroup v2 process management but disable the compat dual layout, we return true on
* cg_unified_controller(SYSTEMD_CGROUP_CONTROLLER) and false on cg_hybrid_unified().
*/
static thread_local bool unified_systemd_v232;
static int cg_unified_update(void) {
int cg_unified_cached(bool flush) {
static thread_local CGroupUnified unified_cache = CGROUP_UNIFIED_UNKNOWN;
struct statfs fs;
@@ -2545,8 +1942,10 @@ static int cg_unified_update(void) {
* have any other trouble determining if the unified hierarchy
* is supported. */
if (unified_cache >= CGROUP_UNIFIED_NONE)
return 0;
if (flush)
unified_cache = CGROUP_UNIFIED_UNKNOWN;
else if (unified_cache >= CGROUP_UNIFIED_NONE)
return unified_cache;
if (statfs("/sys/fs/cgroup/", &fs) < 0)
return log_debug_errno(errno, "statfs(\"/sys/fs/cgroup/\") failed: %m");
@@ -2582,20 +1981,20 @@ static int cg_unified_update(void) {
"Unknown filesystem type %llx mounted on /sys/fs/cgroup.",
(unsigned long long)fs.f_type);
return 0;
return unified_cache;
}
int cg_unified_controller(const char *controller) {
int r;
r = cg_unified_update();
r = cg_unified_cached(false);
if (r < 0)
return r;
if (unified_cache == CGROUP_UNIFIED_NONE)
if (r == CGROUP_UNIFIED_NONE)
return false;
if (unified_cache >= CGROUP_UNIFIED_ALL)
if (r >= CGROUP_UNIFIED_ALL)
return true;
return streq_ptr(controller, SYSTEMD_CGROUP_CONTROLLER);
@@ -2604,231 +2003,21 @@ int cg_unified_controller(const char *controller) {
int cg_all_unified(void) {
int r;
r = cg_unified_update();
r = cg_unified_cached(false);
if (r < 0)
return r;
return unified_cache >= CGROUP_UNIFIED_ALL;
return r >= CGROUP_UNIFIED_ALL;
}
int cg_hybrid_unified(void) {
int r;
r = cg_unified_update();
r = cg_unified_cached(false);
if (r < 0)
return r;
return unified_cache == CGROUP_UNIFIED_SYSTEMD && !unified_systemd_v232;
}
int cg_unified_flush(void) {
unified_cache = CGROUP_UNIFIED_UNKNOWN;
return cg_unified_update();
}
int cg_enable_everywhere(
CGroupMask supported,
CGroupMask mask,
const char *p,
CGroupMask *ret_result_mask) {
_cleanup_fclose_ FILE *f = NULL;
_cleanup_free_ char *fs = NULL;
CGroupController c;
CGroupMask ret = 0;
int r;
assert(p);
if (supported == 0) {
if (ret_result_mask)
*ret_result_mask = 0;
return 0;
}
r = cg_all_unified();
if (r < 0)
return r;
if (r == 0) {
/* On the legacy hierarchy there's no concept of "enabling" controllers in cgroups defined. Let's claim
* complete success right away. (If you wonder why we return the full mask here, rather than zero: the
* caller tends to use the returned mask later on to compare if all controllers where properly joined,
* and if not requeues realization. This use is the primary purpose of the return value, hence let's
* minimize surprises here and reduce triggers for re-realization by always saying we fully
* succeeded.) */
if (ret_result_mask)
*ret_result_mask = mask & supported & CGROUP_MASK_V2; /* If you wonder why we mask this with
* CGROUP_MASK_V2: The 'supported' mask
* might contain pure-V1 or BPF
* controllers, and we never want to
* claim that we could enable those with
* cgroup.subtree_control */
return 0;
}
r = cg_get_path(SYSTEMD_CGROUP_CONTROLLER, p, "cgroup.subtree_control", &fs);
if (r < 0)
return r;
for (c = 0; c < _CGROUP_CONTROLLER_MAX; c++) {
CGroupMask bit = CGROUP_CONTROLLER_TO_MASK(c);
const char *n;
if (!FLAGS_SET(CGROUP_MASK_V2, bit))
continue;
if (!FLAGS_SET(supported, bit))
continue;
n = cgroup_controller_to_string(c);
{
char s[1 + strlen(n) + 1];
s[0] = FLAGS_SET(mask, bit) ? '+' : '-';
strcpy(s + 1, n);
if (!f) {
f = fopen(fs, "we");
if (!f)
return log_debug_errno(errno, "Failed to open cgroup.subtree_control file of %s: %m", p);
}
r = write_string_stream(f, s, WRITE_STRING_FILE_DISABLE_BUFFER);
if (r < 0) {
log_debug_errno(r, "Failed to %s controller %s for %s (%s): %m",
FLAGS_SET(mask, bit) ? "enable" : "disable", n, p, fs);
clearerr(f);
/* If we can't turn off a controller, leave it on in the reported resulting mask. This
* happens for example when we attempt to turn off a controller up in the tree that is
* used down in the tree. */
if (!FLAGS_SET(mask, bit) && r == -EBUSY) /* You might wonder why we check for EBUSY
* only here, and not follow the same logic
* for other errors such as EINVAL or
* EOPNOTSUPP or anything else. That's
* because EBUSY indicates that the
* controllers is currently enabled and
* cannot be disabled because something down
* the hierarchy is still using it. Any other
* error most likely means something like "I
* never heard of this controller" or
* similar. In the former case it's hence
* safe to assume the controller is still on
* after the failed operation, while in the
* latter case it's safer to assume the
* controller is unknown and hence certainly
* not enabled. */
ret |= bit;
} else {
/* Otherwise, if we managed to turn on a controller, set the bit reflecting that. */
if (FLAGS_SET(mask, bit))
ret |= bit;
}
}
}
/* Let's return the precise set of controllers now enabled for the cgroup. */
if (ret_result_mask)
*ret_result_mask = ret;
return 0;
}
bool cg_is_unified_wanted(void) {
static thread_local int wanted = -1;
int r;
bool b;
const bool is_default = DEFAULT_HIERARCHY == CGROUP_UNIFIED_ALL;
_cleanup_free_ char *c = NULL;
/* If we have a cached value, return that. */
if (wanted >= 0)
return wanted;
/* If the hierarchy is already mounted, then follow whatever
* was chosen for it. */
if (cg_unified_flush() >= 0)
return (wanted = unified_cache >= CGROUP_UNIFIED_ALL);
/* If we were explicitly passed systemd.unified_cgroup_hierarchy,
* respect that. */
r = proc_cmdline_get_bool("systemd.unified_cgroup_hierarchy", &b);
if (r > 0)
return (wanted = b);
/* If we passed cgroup_no_v1=all with no other instructions, it seems
* highly unlikely that we want to use hybrid or legacy hierarchy. */
r = proc_cmdline_get_key("cgroup_no_v1", 0, &c);
if (r > 0 && streq_ptr(c, "all"))
return (wanted = true);
return (wanted = is_default);
}
bool cg_is_legacy_wanted(void) {
static thread_local int wanted = -1;
/* If we have a cached value, return that. */
if (wanted >= 0)
return wanted;
/* Check if we have cgroup v2 already mounted. */
if (cg_unified_flush() >= 0 &&
unified_cache == CGROUP_UNIFIED_ALL)
return (wanted = false);
/* Otherwise, assume that at least partial legacy is wanted,
* since cgroup v2 should already be mounted at this point. */
return (wanted = true);
}
bool cg_is_hybrid_wanted(void) {
static thread_local int wanted = -1;
int r;
bool b;
const bool is_default = DEFAULT_HIERARCHY >= CGROUP_UNIFIED_SYSTEMD;
/* We default to true if the default is "hybrid", obviously,
* but also when the default is "unified", because if we get
* called, it means that unified hierarchy was not mounted. */
/* If we have a cached value, return that. */
if (wanted >= 0)
return wanted;
/* If the hierarchy is already mounted, then follow whatever
* was chosen for it. */
if (cg_unified_flush() >= 0 &&
unified_cache == CGROUP_UNIFIED_ALL)
return (wanted = false);
/* Otherwise, let's see what the kernel command line has to say.
* Since checking is expensive, cache a non-error result. */
r = proc_cmdline_get_bool("systemd.legacy_systemd_cgroup_controller", &b);
/* The meaning of the kernel option is reversed wrt. to the return value
* of this function, hence the negation. */
return (wanted = r > 0 ? !b : is_default);
}
int cg_weight_parse(const char *s, uint64_t *ret) {
uint64_t u;
int r;
if (isempty(s)) {
*ret = CGROUP_WEIGHT_INVALID;
return 0;
}
r = safe_atou64(s, &u);
if (r < 0)
return r;
if (u < CGROUP_WEIGHT_MIN || u > CGROUP_WEIGHT_MAX)
return -ERANGE;
*ret = u;
return 0;
return r == CGROUP_UNIFIED_SYSTEMD && !unified_systemd_v232;
}
const uint64_t cgroup_io_limit_defaults[_CGROUP_IO_LIMIT_TYPE_MAX] = {
@@ -2847,46 +2036,6 @@ static const char* const cgroup_io_limit_type_table[_CGROUP_IO_LIMIT_TYPE_MAX] =
DEFINE_STRING_TABLE_LOOKUP(cgroup_io_limit_type, CGroupIOLimitType);
int cg_cpu_shares_parse(const char *s, uint64_t *ret) {
uint64_t u;
int r;
if (isempty(s)) {
*ret = CGROUP_CPU_SHARES_INVALID;
return 0;
}
r = safe_atou64(s, &u);
if (r < 0)
return r;
if (u < CGROUP_CPU_SHARES_MIN || u > CGROUP_CPU_SHARES_MAX)
return -ERANGE;
*ret = u;
return 0;
}
int cg_blkio_weight_parse(const char *s, uint64_t *ret) {
uint64_t u;
int r;
if (isempty(s)) {
*ret = CGROUP_BLKIO_WEIGHT_INVALID;
return 0;
}
r = safe_atou64(s, &u);
if (r < 0)
return r;
if (u < CGROUP_BLKIO_WEIGHT_MIN || u > CGROUP_BLKIO_WEIGHT_MAX)
return -ERANGE;
*ret = u;
return 0;
}
bool is_cgroup_fs(const struct statfs *s) {
return is_fs_type(s, CGROUP_SUPER_MAGIC) ||
is_fs_type(s, CGROUP2_SUPER_MAGIC);
src/basic/cgroup-util.h
+4 -27
View File
@@ -174,10 +174,6 @@ typedef int (*cg_kill_log_func_t)(pid_t pid, int sig, void *userdata);
int cg_kill(const char *controller, const char *path, int sig, CGroupFlags flags, Set *s, cg_kill_log_func_t kill_log, void *userdata);
int cg_kill_recursive(const char *controller, const char *path, int sig, CGroupFlags flags, Set *s, cg_kill_log_func_t kill_log, void *userdata);
int cg_migrate(const char *cfrom, const char *pfrom, const char *cto, const char *pto, CGroupFlags flags);
int cg_migrate_recursive(const char *cfrom, const char *pfrom, const char *cto, const char *pto, CGroupFlags flags);
int cg_migrate_recursive_fallback(const char *cfrom, const char *pfrom, const char *cto, const char *pto, CGroupFlags flags);
int cg_split_spec(const char *spec, char **controller, char **path);
int cg_mangle_path(const char *path, char **result);
@@ -186,15 +182,8 @@ int cg_get_path_and_check(const char *controller, const char *path, const char *
int cg_pid_get_path(const char *controller, pid_t pid, char **path);
int cg_trim(const char *controller, const char *path, bool delete_root);
int cg_rmdir(const char *controller, const char *path);
int cg_create(const char *controller, const char *path);
int cg_attach(const char *controller, const char *path, pid_t pid);
int cg_attach_fallback(const char *controller, const char *path, pid_t pid);
int cg_create_and_attach(const char *controller, const char *path, pid_t pid);
int cg_set_attribute(const char *controller, const char *path, const char *attribute, const char *value);
int cg_get_attribute(const char *controller, const char *path, const char *attribute, char **ret);
int cg_get_keyed_attribute(const char *controller, const char *path, const char *attribute, char **keys, char **values);
@@ -242,13 +231,6 @@ int cg_slice_to_path(const char *unit, char **ret);
typedef const char* (*cg_migrate_callback_t)(CGroupMask mask, void *userdata);
int cg_create_everywhere(CGroupMask supported, CGroupMask mask, const char *path);
int cg_attach_everywhere(CGroupMask supported, const char *path, pid_t pid, cg_migrate_callback_t callback, void *userdata);
int cg_attach_many_everywhere(CGroupMask supported, const char *path, Set* pids, cg_migrate_callback_t callback, void *userdata);
int cg_migrate_everywhere(CGroupMask supported, const char *from, const char *to, cg_migrate_callback_t callback, void *userdata);
int cg_trim_everywhere(CGroupMask supported, const char *path, bool delete_root);
int cg_enable_everywhere(CGroupMask supported, CGroupMask mask, const char *p, CGroupMask *ret_result_mask);
int cg_mask_supported(CGroupMask *ret);
int cg_mask_from_string(const char *s, CGroupMask *ret);
int cg_mask_to_string(CGroupMask mask, char **ret);
@@ -260,18 +242,13 @@ bool cg_ns_supported(void);
int cg_all_unified(void);
int cg_hybrid_unified(void);
int cg_unified_controller(const char *controller);
int cg_unified_flush(void);
bool cg_is_unified_wanted(void);
bool cg_is_legacy_wanted(void);
bool cg_is_hybrid_wanted(void);
int cg_unified_cached(bool flush);
static inline int cg_unified(void) {
return cg_unified_cached(true);
}
const char* cgroup_controller_to_string(CGroupController c) _const_;
CGroupController cgroup_controller_from_string(const char *s) _pure_;
int cg_weight_parse(const char *s, uint64_t *ret);
int cg_cpu_shares_parse(const char *s, uint64_t *ret);
int cg_blkio_weight_parse(const char *s, uint64_t *ret);
bool is_cgroup_fs(const struct statfs *s);
bool fd_is_cgroup_fs(int fd);
src/basic/efivars.c
+250
View File
@@ -0,0 +1,250 @@
/* SPDX-License-Identifier: LGPL-2.1+ */
#include <errno.h>
#include <fcntl.h>
#include <limits.h>
#include <linux/fs.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sys/stat.h>
#include <unistd.h>
#include "sd-id128.h"
#include "alloc-util.h"
#include "chattr-util.h"
#include "efivars.h"
#include "fd-util.h"
#include "io-util.h"
#include "macro.h"
#include "stdio-util.h"
#include "strv.h"
#include "time-util.h"
#include "utf8.h"
#if ENABLE_EFI
char* efi_variable_path(sd_id128_t vendor, const char *name) {
char *p;
if (asprintf(&p,
"/sys/firmware/efi/efivars/%s-" SD_ID128_UUID_FORMAT_STR,
name, SD_ID128_FORMAT_VAL(vendor)) < 0)
return NULL;
return p;
}
int efi_get_variable(
sd_id128_t vendor,
const char *name,
uint32_t *ret_attribute,
void **ret_value,
size_t *ret_size) {
_cleanup_close_ int fd = -1;
_cleanup_free_ char *p = NULL;
_cleanup_free_ void *buf = NULL;
struct stat st;
uint32_t a;
ssize_t n;
assert(name);
p = efi_variable_path(vendor, name);
if (!p)
return -ENOMEM;
if (!ret_value && !ret_size && !ret_attribute) {
/* If caller is not interested in anything, just check if the variable exists and is readable
* to us. */
if (access(p, R_OK) < 0)
return -errno;
return 0;
}
fd = open(p, O_RDONLY|O_NOCTTY|O_CLOEXEC);
if (fd < 0)
return -errno;
if (fstat(fd, &st) < 0)
return -errno;
if (st.st_size < 4)
return -ENODATA;
if (st.st_size > 4*1024*1024 + 4)
return -E2BIG;
if (ret_value || ret_attribute) {
n = read(fd, &a, sizeof(a));
if (n < 0)
return -errno;
if (n != sizeof(a))
return -EIO;
}
if (ret_value) {
buf = malloc(st.st_size - 4 + 2);
if (!buf)
return -ENOMEM;
n = read(fd, buf, (size_t) st.st_size - 4);
if (n < 0)
return -errno;
if (n != st.st_size - 4)
return -EIO;
/* Always NUL terminate (2 bytes, to protect UTF-16) */
((char*) buf)[st.st_size - 4] = 0;
((char*) buf)[st.st_size - 4 + 1] = 0;
}
/* Note that efivarfs interestingly doesn't require ftruncate() to update an existing EFI variable
* with a smaller value. */
if (ret_attribute)
*ret_attribute = a;
if (ret_value)
*ret_value = TAKE_PTR(buf);
if (ret_size)
*ret_size = (size_t) st.st_size - 4;
return 0;
}
int efi_get_variable_string(sd_id128_t vendor, const char *name, char **p) {
_cleanup_free_ void *s = NULL;
size_t ss = 0;
int r;
char *x;
r = efi_get_variable(vendor, name, NULL, &s, &ss);
if (r < 0)
return r;
x = utf16_to_utf8(s, ss);
if (!x)
return -ENOMEM;
*p = x;
return 0;
}
int efi_set_variable(
sd_id128_t vendor,
const char *name,
const void *value,
size_t size) {
struct var {
uint32_t attr;
char buf[];
} _packed_ * _cleanup_free_ buf = NULL;
_cleanup_free_ char *p = NULL;
_cleanup_close_ int fd = -1;
bool saved_flags_valid = false;
unsigned saved_flags;
int r;
assert(name);
assert(value || size == 0);
p = efi_variable_path(vendor, name);
if (!p)
return -ENOMEM;
/* Newer efivarfs protects variables that are not in a whitelist with FS_IMMUTABLE_FL by default, to protect
* them for accidental removal and modification. We are not changing these variables accidentally however,
* hence let's unset the bit first. */
r = chattr_path(p, 0, FS_IMMUTABLE_FL, &saved_flags);
if (r < 0 && r != -ENOENT)
log_debug_errno(r, "Failed to drop FS_IMMUTABLE_FL flag from '%s', ignoring: %m", p);
saved_flags_valid = r >= 0;
if (size == 0) {
if (unlink(p) < 0) {
r = -errno;
goto finish;
}
return 0;
}
fd = open(p, O_WRONLY|O_CREAT|O_NOCTTY|O_CLOEXEC, 0644);
if (fd < 0) {
r = -errno;
goto finish;
}
buf = malloc(sizeof(uint32_t) + size);
if (!buf) {
r = -ENOMEM;
goto finish;
}
buf->attr = EFI_VARIABLE_NON_VOLATILE|EFI_VARIABLE_BOOTSERVICE_ACCESS|EFI_VARIABLE_RUNTIME_ACCESS;
memcpy(buf->buf, value, size);
r = loop_write(fd, buf, sizeof(uint32_t) + size, false);
if (r < 0)
goto finish;
r = 0;
finish:
if (saved_flags_valid) {
int q;
/* Restore the original flags field, just in case */
if (fd < 0)
q = chattr_path(p, saved_flags, FS_IMMUTABLE_FL, NULL);
else
q = chattr_fd(fd, saved_flags, FS_IMMUTABLE_FL, NULL);
if (q < 0)
log_debug_errno(q, "Failed to restore FS_IMMUTABLE_FL on '%s', ignoring: %m", p);
}
return r;
}
int efi_set_variable_string(sd_id128_t vendor, const char *name, const char *v) {
_cleanup_free_ char16_t *u16 = NULL;
u16 = utf8_to_utf16(v, strlen(v));
if (!u16)
return -ENOMEM;
return efi_set_variable(vendor, name, u16, (char16_strlen(u16) + 1) * sizeof(char16_t));
}
int efi_systemd_options_variable(char **line) {
const char *e;
int r;
assert(line);
/* For testing purposes it is sometimes useful to be able to override this */
e = secure_getenv("SYSTEMD_EFI_OPTIONS");
if (e) {
char *m;
m = strdup(e);
if (!m)
return -ENOMEM;
*line = m;
return 0;
}
r = efi_get_variable_string(EFI_VENDOR_SYSTEMD, "SystemdOptions", line);
if (r == -ENOENT)
return -ENODATA;
return r;
}
#endif
src/basic/efivars.h
+59
View File
@@ -0,0 +1,59 @@
/* SPDX-License-Identifier: LGPL-2.1+ */
#pragma once
#if !ENABLE_EFI
# include <errno.h>
#endif
#include <stdbool.h>
#include <stddef.h>
#include <stdint.h>
#include "sd-id128.h"
#include "efi/loader-features.h"
#include "time-util.h"
#define EFI_VENDOR_LOADER SD_ID128_MAKE(4a,67,b0,82,0a,4c,41,cf,b6,c7,44,0b,29,bb,8c,4f)
#define EFI_VENDOR_GLOBAL SD_ID128_MAKE(8b,e4,df,61,93,ca,11,d2,aa,0d,00,e0,98,03,2b,8c)
#define EFI_VENDOR_SYSTEMD SD_ID128_MAKE(8c,f2,64,4b,4b,0b,42,8f,93,87,6d,87,60,50,dc,67)
#define EFI_VARIABLE_NON_VOLATILE 0x0000000000000001
#define EFI_VARIABLE_BOOTSERVICE_ACCESS 0x0000000000000002
#define EFI_VARIABLE_RUNTIME_ACCESS 0x0000000000000004
#if ENABLE_EFI
char* efi_variable_path(sd_id128_t vendor, const char *name);
int efi_get_variable(sd_id128_t vendor, const char *name, uint32_t *attribute, void **value, size_t *size);
int efi_get_variable_string(sd_id128_t vendor, const char *name, char **p);
int efi_set_variable(sd_id128_t vendor, const char *name, const void *value, size_t size);
int efi_set_variable_string(sd_id128_t vendor, const char *name, const char *p);
int efi_systemd_options_variable(char **line);
#else
static inline char* efi_variable_path(sd_id128_t vendor, const char *name) {
return NULL;
}
static inline int efi_get_variable(sd_id128_t vendor, const char *name, uint32_t *attribute, void **value, size_t *size) {
return -EOPNOTSUPP;
}
static inline int efi_get_variable_string(sd_id128_t vendor, const char *name, char **p) {
return -EOPNOTSUPP;
}
static inline int efi_set_variable(sd_id128_t vendor, const char *name, const void *value, size_t size) {
return -EOPNOTSUPP;
}
static inline int efi_set_variable_string(sd_id128_t vendor, const char *name, const char *p) {
return -EOPNOTSUPP;
}
static inline int efi_systemd_options_variable(char **line) {
return -ENODATA;
}
#endif
src/basic/meson.build
+2
View File
@@ -39,6 +39,8 @@ basic_sources = files('''
device-nodes.h
dirent-util.c
dirent-util.h
efivars.c
efivars.h
env-file.c
env-file.h
env-util.c
src/basic/proc-cmdline.c
+58 -79
View File
@@ -5,6 +5,7 @@
#include <string.h>
#include "alloc-util.h"
#include "efivars.h"
#include "extract-word.h"
#include "fileio.h"
#include "macro.h"
@@ -117,6 +118,17 @@ int proc_cmdline_parse(proc_cmdline_parse_t parse_item, void *data, ProcCmdlineF
assert(parse_item);
/* We parse the EFI variable first, because later settings have higher priority. */
r = efi_systemd_options_variable(&line);
if (r < 0 && r != -ENODATA)
log_debug_errno(r, "Failed to get SystemdOptions EFI variable, ignoring: %m");
r = proc_cmdline_parse_given(line, parse_item, data, flags);
if (r < 0)
return r;
line = mfree(line);
r = proc_cmdline(&line);
if (r < 0)
return r;
@@ -156,34 +168,14 @@ bool proc_cmdline_key_streq(const char *x, const char *y) {
return true;
}
int proc_cmdline_get_key(const char *key, ProcCmdlineFlags flags, char **ret_value) {
_cleanup_free_ char *line = NULL, *ret = NULL;
static int cmdline_get_key(const char *line, const char *key, ProcCmdlineFlags flags, char **ret_value) {
_cleanup_free_ char *ret = NULL;
bool found = false;
const char *p;
int r;
/* Looks for a specific key on the kernel command line. Supports three modes:
*
* a) The "ret_value" parameter is used. In this case a parameter beginning with the "key" string followed by
* "=" is searched for, and the value following it is returned in "ret_value".
*
* b) as above, but the PROC_CMDLINE_VALUE_OPTIONAL flag is set. In this case if the key is found as a separate
* word (i.e. not followed by "=" but instead by whitespace or the end of the command line), then this is
* also accepted, and "value" is returned as NULL.
*
* c) The "ret_value" parameter is NULL. In this case a search for the exact "key" parameter is performed.
*
* In all three cases, > 0 is returned if the key is found, 0 if not. */
if (isempty(key))
return -EINVAL;
if (FLAGS_SET(flags, PROC_CMDLINE_VALUE_OPTIONAL) && !ret_value)
return -EINVAL;
r = proc_cmdline(&line);
if (r < 0)
return r;
assert(line);
assert(key);
p = line;
for (;;) {
@@ -226,6 +218,48 @@ int proc_cmdline_get_key(const char *key, ProcCmdlineFlags flags, char **ret_val
return found;
}
int proc_cmdline_get_key(const char *key, ProcCmdlineFlags flags, char **ret_value) {
_cleanup_free_ char *line = NULL;
int r;
/* Looks for a specific key on the kernel command line and (with lower priority) the EFI variable.
* Supports three modes:
*
* a) The "ret_value" parameter is used. In this case a parameter beginning with the "key" string followed by
* "=" is searched for, and the value following it is returned in "ret_value".
*
* b) as above, but the PROC_CMDLINE_VALUE_OPTIONAL flag is set. In this case if the key is found as a separate
* word (i.e. not followed by "=" but instead by whitespace or the end of the command line), then this is
* also accepted, and "value" is returned as NULL.
*
* c) The "ret_value" parameter is NULL. In this case a search for the exact "key" parameter is performed.
*
* In all three cases, > 0 is returned if the key is found, 0 if not. */
if (isempty(key))
return -EINVAL;
if (FLAGS_SET(flags, PROC_CMDLINE_VALUE_OPTIONAL) && !ret_value)
return -EINVAL;
r = proc_cmdline(&line);
if (r < 0)
return r;
r = cmdline_get_key(line, key, flags, ret_value);
if (r != 0) /* Either error or true if found. */
return r;
line = mfree(line);
r = efi_systemd_options_variable(&line);
if (r == -ENODATA)
return false; /* Not found */
if (r < 0)
return r;
return cmdline_get_key(line, key, flags, ret_value);
}
int proc_cmdline_get_bool(const char *key, bool *ret) {
_cleanup_free_ char *v = NULL;
int r;
@@ -306,58 +340,3 @@ int proc_cmdline_get_key_many_internal(ProcCmdlineFlags flags, ...) {
return ret;
}
int shall_restore_state(void) {
bool ret;
int r;
r = proc_cmdline_get_bool("systemd.restore_state", &ret);
if (r < 0)
return r;
return r > 0 ? ret : true;
}
static const char * const rlmap[] = {
"emergency", SPECIAL_EMERGENCY_TARGET,
"-b", SPECIAL_EMERGENCY_TARGET,
"rescue", SPECIAL_RESCUE_TARGET,
"single", SPECIAL_RESCUE_TARGET,
"-s", SPECIAL_RESCUE_TARGET,
"s", SPECIAL_RESCUE_TARGET,
"S", SPECIAL_RESCUE_TARGET,
"1", SPECIAL_RESCUE_TARGET,
"2", SPECIAL_MULTI_USER_TARGET,
"3", SPECIAL_MULTI_USER_TARGET,
"4", SPECIAL_MULTI_USER_TARGET,
"5", SPECIAL_GRAPHICAL_TARGET,
NULL
};
static const char * const rlmap_initrd[] = {
"emergency", SPECIAL_EMERGENCY_TARGET,
"rescue", SPECIAL_RESCUE_TARGET,
NULL
};
const char* runlevel_to_target(const char *word) {
const char * const *rlmap_ptr;
size_t i;
if (!word)
return NULL;
if (in_initrd()) {
word = startswith(word, "rd.");
if (!word)
return NULL;
}
rlmap_ptr = in_initrd() ? rlmap_initrd : rlmap;
for (i = 0; rlmap_ptr[i]; i += 2)
if (streq(word, rlmap_ptr[i]))
return rlmap_ptr[i+1];
return NULL;
}
src/basic/proc-cmdline.h
-3
View File
@@ -27,9 +27,6 @@ int proc_cmdline_get_key_many_internal(ProcCmdlineFlags flags, ...);
char *proc_cmdline_key_startswith(const char *s, const char *prefix);
bool proc_cmdline_key_streq(const char *x, const char *y);
int shall_restore_state(void);
const char* runlevel_to_target(const char *rl);
/* A little helper call, to be used in proc_cmdline_parse_t callbacks */
static inline bool proc_cmdline_value_missing(const char *key, const char *value) {
if (!value) {
src/basic/string-util.h
+16
View File
@@ -45,6 +45,22 @@ static inline const char *strna(const char *s) {
return s ?: "n/a";
}
static inline const char* yes_no(bool b) {
return b ? "yes" : "no";
}
static inline const char* true_false(bool b) {
return b ? "true" : "false";
}
static inline const char* one_zero(bool b) {
return b ? "1" : "0";
}
static inline const char* enable_disable(bool b) {
return b ? "enable" : "disable";
}
static inline bool isempty(const char *p) {
return !p || !p[0];
}
src/basic/util.h
-16
View File
@@ -5,22 +5,6 @@
#include "macro.h"
static inline const char* yes_no(bool b) {
return b ? "yes" : "no";
}
static inline const char* true_false(bool b) {
return b ? "true" : "false";
}
static inline const char* one_zero(bool b) {
return b ? "1" : "0";
}
static inline const char* enable_disable(bool b) {
return b ? "enable" : "disable";
}
extern int saved_argc;
extern char **saved_argv;