HALF ECHO LEARN / REFERENCE COLLECTION
v262.2src/home/homed-manager.c
Git blob: 6dc0d82d7dca55da6726e4e1cd82945faa02caf0
Lines 1001–2000 of 2,238
if (!m->scan_slash_home)
return 0;
d = opendir(get_home_root());
if (!d)
return log_full_errno(errno == ENOENT ? LOG_DEBUG : LOG_ERR, errno,
"Failed to open %s: %m", get_home_root());
FOREACH_DIRENT(de, d, return log_error_errno(errno, "Failed to read %s directory: %m", get_home_root()))
(void) manager_assess_image(m, dirfd(d), get_home_root(), de->d_name);
return 0;
}
static int manager_connect_bus(Manager *m) {
_cleanup_free_ char *b = NULL;
const char *suffix, *busname;
int r;
assert(m);
assert(!m->bus);
r = sd_bus_default_system(&m->bus);
if (r < 0)
return log_error_errno(r, "Failed to connect to system bus: %m");
r = bus_add_implementation(m->bus, &manager_object, m);
if (r < 0)
return r;
r = bus_log_control_api_register(m->bus);
if (r < 0)
return r;
suffix = getenv("SYSTEMD_HOME_DEBUG_SUFFIX");
if (suffix) {
b = strjoin("org.freedesktop.home1.", suffix);
if (!b)
return log_oom();
busname = b;
} else
busname = "org.freedesktop.home1";
r = sd_bus_request_name_async(m->bus, NULL, busname, 0, NULL, NULL);
if (r < 0)
return log_error_errno(r, "Failed to request name: %m");
r = sd_bus_attach_event(m->bus, m->event, SD_EVENT_PRIORITY_NORMAL);
if (r < 0)
return log_error_errno(r, "Failed to attach bus to event loop: %m");
(void) sd_bus_set_exit_on_disconnect(m->bus, true);
return 0;
}
static int manager_bind_varlink(Manager *m) {
_cleanup_free_ char *p = NULL;
const char *suffix, *socket_path;
int r;
assert(m);
assert(!m->varlink_server);
r = varlink_server_new(
&m->varlink_server,
SD_VARLINK_SERVER_ACCOUNT_UID|SD_VARLINK_SERVER_INHERIT_USERDATA|SD_VARLINK_SERVER_INPUT_SENSITIVE,
m);
if (r < 0)
return log_error_errno(r, "Failed to allocate varlink server: %m");
r = sd_varlink_server_add_interface_many(
m->varlink_server,
&vl_interface_io_systemd_UserDatabase,
&vl_interface_io_systemd_service);
if (r < 0)
return log_error_errno(r, "Failed to add UserDatabase interface to varlink server: %m");
r = sd_varlink_server_bind_method_many(
m->varlink_server,
"io.systemd.UserDatabase.GetUserRecord", vl_method_get_user_record,
"io.systemd.UserDatabase.GetGroupRecord", vl_method_get_group_record,
"io.systemd.UserDatabase.GetMemberships", vl_method_get_memberships,
"io.systemd.service.Ping", varlink_method_ping,
"io.systemd.service.SetLogLevel", varlink_method_set_log_level,
"io.systemd.service.GetLogLevel", varlink_method_get_log_level,
"io.systemd.service.GetEnvironment", varlink_method_get_environment);
if (r < 0)
return log_error_errno(r, "Failed to register varlink methods: %m");
/* To make things easier to debug, when working from a homed managed home directory, let's optionally
* use a different varlink socket name */
suffix = getenv("SYSTEMD_HOME_DEBUG_SUFFIX");
if (suffix) {
p = strjoin("/run/systemd/userdb/io.systemd.Home.", suffix);
if (!p)
return log_oom();
socket_path = p;
} else
socket_path = "/run/systemd/userdb/io.systemd.Home";
r = sd_varlink_server_listen_address(m->varlink_server, socket_path, 0666 | SD_VARLINK_SERVER_MODE_MKDIR_0755);
if (r < 0)
return log_error_errno(r, "Failed to bind to varlink socket '%s': %m", socket_path);
/* Reduce the noise routed to us: advertise that only look-ups for non-system users in the 16bit
* range are routed to us (excluding 'nobody', i.e. 0xfffe) */
char text[DECIMAL_STR_MAX(uid_t) + 1 + DECIMAL_STR_MAX(gid_t) + 1];
xsprintf(text, UID_FMT "-" UID_FMT, (uid_t) (SYSTEM_UID_MAX + 1), UID_NOBODY - 1);
FOREACH_STRING(xattr, "user.userdb.uid", "user.userdb.gid")
(void) xsetxattr(AT_FDCWD, socket_path, /* at_flags= */ 0, xattr, text);
r = sd_varlink_server_attach_event(m->varlink_server, m->event, SD_EVENT_PRIORITY_NORMAL);
if (r < 0)
return log_error_errno(r, "Failed to attach varlink connection to event loop: %m");
assert(!m->userdb_service);
r = path_extract_filename(socket_path, &m->userdb_service);
if (r < 0)
return log_error_errno(r, "Failed to extract filename from socket path '%s': %m", socket_path);
/* Avoid recursion */
if (setenv("SYSTEMD_BYPASS_USERDB", m->userdb_service, 1) < 0)
return log_error_errno(errno, "Failed to set $SYSTEMD_BYPASS_USERDB: %m");
return 0;
}
static int on_notify_socket(sd_event_source *s, int fd, uint32_t revents, void *userdata) {
Manager *m = ASSERT_PTR(userdata);
int r;
assert(s);
_cleanup_(fdset_free_asyncp) FDSet *passed_fds = NULL;
_cleanup_(pidref_done) PidRef sender = PIDREF_NULL;
_cleanup_strv_free_ char **l = NULL;
r = notify_recv_with_fds_strv(fd, &l, /* ret_ucred= */ NULL, &sender, &passed_fds);
if (r == -EAGAIN)
return 0;
if (r < 0)
return r;
if (fdset_size(passed_fds) > 1) {
log_warning("Received notify datagram with multiple fds, ignoring.");
return 0;
}
Home *h = hashmap_get(m->homes_by_worker_pid, &sender);
if (!h) {
log_warning("Received notify datagram of unknown process, ignoring.");
return 0;
}
home_process_notify(h, l, fdset_steal_first(passed_fds));
return 0;
}
static int manager_listen_notify(Manager *m) {
int r;
assert(m);
assert(!m->notify_socket_path);
r = notify_socket_prepare_full(
m->event,
SD_EVENT_PRIORITY_NORMAL - 5, /* Make sure we process sd_notify() before child exit for
* any worker, so that we always know the error number
* of a client before it exits. */
on_notify_socket,
m,
/* accept_fds= */ true,
&m->notify_socket_path,
/* ret_event_source= */ NULL);
if (r < 0)
return log_error_errno(r, "Failed to prepare notify socket: %m");
return 0;
}
static int manager_add_device(Manager *m, sd_device *d) {
_cleanup_free_ char *user_name = NULL, *realm = NULL, *node = NULL;
const char *tabletype, *parttype, *partname, *partuuid, *sysfs;
sd_id128_t id;
int r;
assert(m);
assert(d);
r = sd_device_get_syspath(d, &sysfs);
if (r < 0)
return log_error_errno(r, "Failed to acquire sysfs path of device: %m");
r = sd_device_get_property_value(d, "ID_PART_TABLE_TYPE", &tabletype);
if (r == -ENOENT)
return 0;
if (r < 0)
return log_error_errno(r, "Failed to acquire ID_PART_TABLE_TYPE device property, ignoring: %m");
if (!streq(tabletype, "gpt")) {
log_debug("Found partition (%s) on non-GPT table, ignoring.", sysfs);
return 0;
}
r = sd_device_get_property_value(d, "ID_PART_ENTRY_TYPE", &parttype);
if (r == -ENOENT)
return 0;
if (r < 0)
return log_error_errno(r, "Failed to acquire ID_PART_ENTRY_TYPE device property, ignoring: %m");
if (sd_id128_string_equal(parttype, SD_GPT_USER_HOME) <= 0) {
log_debug("Found partition (%s) we don't care about, ignoring.", sysfs);
return 0;
}
r = sd_device_get_property_value(d, "ID_PART_ENTRY_NAME", &partname);
if (r < 0)
return log_warning_errno(r, "Failed to acquire ID_PART_ENTRY_NAME device property, ignoring: %m");
r = split_user_name_realm(partname, &user_name, &realm);
if (r == -EINVAL)
return log_warning_errno(r, "Found partition with correct partition type but a non-parsable partition name '%s', ignoring.", partname);
if (r < 0)
return log_error_errno(r, "Failed to validate partition name '%s': %m", partname);
r = sd_device_get_property_value(d, "ID_FS_UUID", &partuuid);
if (r < 0)
return log_warning_errno(r, "Failed to acquire ID_FS_UUID device property, ignoring: %m");
r = sd_id128_from_string(partuuid, &id);
if (r < 0)
return log_warning_errno(r, "Failed to parse ID_FS_UUID field '%s', ignoring: %m", partuuid);
if (asprintf(&node, "/dev/disk/by-uuid/" SD_ID128_UUID_FORMAT_STR, SD_ID128_FORMAT_VAL(id)) < 0)
return log_oom();
return manager_add_home_by_image(m, user_name, realm, node, sysfs, USER_LUKS, UID_INVALID);
}
static int manager_on_device(sd_device_monitor *monitor, sd_device *d, void *userdata) {
Manager *m = ASSERT_PTR(userdata);
int r;
assert(d);
if (device_for_action(d, SD_DEVICE_REMOVE)) {
const char *sysfs;
Home *h;
r = sd_device_get_syspath(d, &sysfs);
if (r < 0) {
log_warning_errno(r, "Failed to acquire sysfs path from device: %m");
return 0;
}
log_info("block device %s has been removed.", sysfs);
/* Let's see if we previously synthesized a home record from this device, if so, let's just
* revalidate that. Otherwise let's revalidate them all, but asynchronously. */
h = hashmap_get(m->homes_by_sysfs, sysfs);
if (h)
manager_revalidate_image(m, h);
else
manager_enqueue_gc(m, NULL);
} else
(void) manager_add_device(m, d);
(void) bus_manager_emit_auto_login_changed(m);
return 0;
}
static int manager_watch_devices(Manager *m) {
int r;
assert(m);
assert(!m->device_monitor);
r = sd_device_monitor_new(&m->device_monitor);
if (r < 0)
return log_error_errno(r, "Failed to allocate device monitor: %m");
r = sd_device_monitor_filter_add_match_subsystem_devtype(m->device_monitor, "block", NULL);
if (r < 0)
return log_error_errno(r, "Failed to configure device monitor match: %m");
r = sd_device_monitor_attach_event(m->device_monitor, m->event);
if (r < 0)
return log_error_errno(r, "Failed to attach device monitor to event loop: %m");
r = sd_device_monitor_start(m->device_monitor, manager_on_device, m);
if (r < 0)
return log_error_errno(r, "Failed to start device monitor: %m");
return 0;
}
static int manager_enumerate_devices(Manager *m) {
_cleanup_(sd_device_enumerator_unrefp) sd_device_enumerator *e = NULL;
int r;
assert(m);
r = sd_device_enumerator_new(&e);
if (r < 0)
return r;
r = sd_device_enumerator_add_match_subsystem(e, "block", true);
if (r < 0)
return r;
FOREACH_DEVICE(e, d) {
if (device_is_processed(d) <= 0)
continue;
(void) manager_add_device(m, d);
}
return 0;
}
static int manager_load_key_pair(Manager *m) {
_cleanup_fclose_ FILE *f = NULL;
struct stat st;
int r;
assert(m);
if (m->private_key) {
sym_EVP_PKEY_free(m->private_key);
m->private_key = NULL;
}
r = search_and_fopen_nulstr("local.private", "re", NULL, KEY_PATHS_NULSTR, &f, NULL);
if (r == -ENOENT)
return 0;
if (r < 0)
return log_error_errno(r, "Failed to read private key file: %m");
if (fstat(fileno(f), &st) < 0)
return log_error_errno(errno, "Failed to stat private key file: %m");
r = stat_verify_regular(&st);
if (r < 0)
return log_error_errno(r, "Private key file is not regular: %m");
if (st.st_uid != 0 || (st.st_mode & 0077) != 0)
return log_error_errno(SYNTHETIC_ERRNO(EPERM), "Private key file is readable by more than the root user");
m->private_key = sym_PEM_read_PrivateKey(f, NULL, NULL, NULL);
if (!m->private_key)
return log_openssl_errors(LOG_ERR, "Failed to load private key pair");
log_info("Successfully loaded private key pair.");
return 1;
}
static int manager_generate_key_pair(Manager *m) {
_cleanup_(EVP_PKEY_CTX_freep) EVP_PKEY_CTX *ctx = NULL;
_cleanup_(unlink_and_freep) char *temp_public = NULL, *temp_private = NULL;
_cleanup_fclose_ FILE *fpublic = NULL, *fprivate = NULL;
int r;
if (m->private_key) {
sym_EVP_PKEY_free(m->private_key);
m->private_key = NULL;
}
ctx = sym_EVP_PKEY_CTX_new_id(EVP_PKEY_ED25519, NULL);
if (!ctx)
return log_openssl_errors(LOG_ERR, "Failed to allocate Ed25519 key generation context.");
if (sym_EVP_PKEY_keygen_init(ctx) <= 0)
return log_openssl_errors(LOG_ERR, "Failed to initialize Ed25519 key generation context.");
log_info("Generating key pair for signing local user identity records.");
if (sym_EVP_PKEY_keygen(ctx, &m->private_key) <= 0)
return log_openssl_errors(LOG_ERR, "Failed to generate Ed25519 key pair");
log_info("Successfully created Ed25519 key pair.");
(void) mkdir_p("/var/lib/systemd/home", 0755);
/* Write out public key (note that we only do that as a help to the user, we don't make use of this ever */
r = fopen_temporary("/var/lib/systemd/home/local.public", &fpublic, &temp_public);
if (r < 0)
return log_error_errno(r, "Failed to open key file for writing: %m");
if (sym_PEM_write_PUBKEY(fpublic, m->private_key) <= 0)
return log_openssl_errors(LOG_ERR, "Failed to write public key.");
(void) fchmod(fileno(fpublic), 0444); /* Make public key world readable */
r = fflush_sync_and_check(fpublic);
if (r < 0)
return log_error_errno(r, "Failed to write private key: %m");
fpublic = safe_fclose(fpublic);
/* Write out the private key (this actually writes out both private and public, OpenSSL is confusing) */
r = fopen_temporary("/var/lib/systemd/home/local.private", &fprivate, &temp_private);
if (r < 0)
return log_error_errno(r, "Failed to open key file for writing: %m");
if (sym_PEM_write_PrivateKey(fprivate, m->private_key, NULL, NULL, 0, NULL, NULL) <= 0)
return log_openssl_errors(LOG_ERR, "Failed to write private key pair.");
(void) fchmod(fileno(fprivate), 0400); /* Make private key root readable */
r = fflush_sync_and_check(fprivate);
if (r < 0)
return log_error_errno(r, "Failed to write private key: %m");
fprivate = safe_fclose(fprivate);
/* Both are written now, move them into place */
if (rename(temp_public, "/var/lib/systemd/home/local.public") < 0)
return log_error_errno(errno, "Failed to move public key file into place: %m");
temp_public = mfree(temp_public);
r = RET_NERRNO(rename(temp_private, "/var/lib/systemd/home/local.private"));
if (r < 0) {
(void) unlink("/var/lib/systemd/home/local.public"); /* try to remove the file we already created */
return log_error_errno(r, "Failed to move private key file into place: %m");
}
temp_private = mfree(temp_private);
r = fsync_path_at(AT_FDCWD, "/var/lib/systemd/home/");
if (r < 0)
log_warning_errno(r, "Failed to sync /var/lib/systemd/home/, ignoring: %m");
return 1;
}
int manager_acquire_key_pair(Manager *m) {
int r;
assert(m);
/* Already there? */
if (m->private_key)
return 1;
/* First try to load key off disk */
r = manager_load_key_pair(m);
if (r != 0)
return r;
/* Didn't work, generate a new one */
return manager_generate_key_pair(m);
}
int manager_sign_user_record(Manager *m, UserRecord *u, UserRecord **ret, sd_bus_error *error) {
int r;
assert(m);
assert(u);
assert(ret);
r = manager_acquire_key_pair(m);
if (r < 0)
return r;
if (r == 0)
return sd_bus_error_set(error, BUS_ERROR_NO_PRIVATE_KEY, "Can't sign without local key.");
return user_record_sign(u, m->private_key, ret);
}
/* dlopen_libcrypto() must have been called before populating this hashmap. */
DEFINE_HASH_OPS_FULL(public_key_hash_ops, char, string_hash_func, string_compare_func, free, EVP_PKEY, sym_EVP_PKEY_free);
static int manager_load_public_key_one(Manager *m, const char *path) {
_cleanup_(EVP_PKEY_freep) EVP_PKEY *pkey = NULL;
_cleanup_fclose_ FILE *f = NULL;
_cleanup_free_ char *fn = NULL;
struct stat st;
int r;
assert(m);
r = path_extract_filename(path, &fn);
if (r < 0)
return log_error_errno(r, "Failed to extract filename of path '%s': %m", path);
if (streq(fn, "local.public")) /* we already loaded the private key, which includes the public one */
return 0;
f = fopen(path, "re");
if (!f) {
if (errno == ENOENT)
return 0;
return log_error_errno(errno, "Failed to open public key %s: %m", path);
}
if (fstat(fileno(f), &st) < 0)
return log_error_errno(errno, "Failed to stat public key %s: %m", path);
r = stat_verify_regular(&st);
if (r < 0)
return log_error_errno(r, "Public key file %s is not a regular file: %m", path);
if (st.st_uid != 0 || (st.st_mode & 0022) != 0)
return log_error_errno(SYNTHETIC_ERRNO(EPERM), "Public key file %s is writable by more than the root user, refusing.", path);
pkey = sym_PEM_read_PUBKEY(f, &pkey, NULL, NULL);
if (!pkey)
return log_openssl_errors(LOG_ERR, "Failed to parse public key file %s.", path);
r = hashmap_ensure_put(&m->public_keys, &public_key_hash_ops, fn, pkey);
if (r < 0)
return log_error_errno(r, "Failed to add public key to set: %m");
TAKE_PTR(fn);
TAKE_PTR(pkey);
return 0;
}
static int manager_load_public_keys(Manager *m) {
_cleanup_strv_free_ char **files = NULL;
int r;
assert(m);
m->public_keys = hashmap_free(m->public_keys);
r = conf_files_list_nulstr(
&files,
".public",
NULL,
CONF_FILES_REGULAR|CONF_FILES_FILTER_MASKED|CONF_FILES_WARN,
KEY_PATHS_NULSTR);
if (r < 0)
return log_error_errno(r, "Failed to assemble list of public key directories: %m");
STRV_FOREACH(i, files)
(void) manager_load_public_key_one(m, *i);
return 0;
}
int manager_startup(Manager *m) {
int r;
assert(m);
r = dlopen_libcrypto(LOG_ERR);
if (r < 0)
return r;
r = manager_listen_notify(m);
if (r < 0)
return r;
r = manager_connect_bus(m);
if (r < 0)
return r;
r = manager_bind_varlink(m);
if (r < 0)
return r;
r = manager_load_key_pair(m); /* only try to load it, don't generate any */
if (r < 0)
return r;
r = manager_load_public_keys(m);
if (r < 0)
return r;
manager_watch_home(m);
(void) manager_watch_devices(m);
(void) manager_enumerate_records(m);
(void) manager_enumerate_images(m);
(void) manager_enumerate_devices(m);
/* Let's clean up home directories whose devices got removed while we were not running */
(void) manager_enqueue_gc(m, NULL);
/* Let's clean up blob directories for home dirs that no longer exist */
(void) manager_gc_blob(m);
return 0;
}
void manager_revalidate_image(Manager *m, Home *h) {
int r;
assert(m);
assert(h);
/* Frees an automatically discovered image, if it's synthetic and its image disappeared. Unmounts any
* image if it's mounted but its image vanished. */
if (h->current_operation || !ordered_set_isempty(h->pending_operations))
return;
if (h->state == HOME_UNFIXATED) {
r = user_record_test_image_path(h->record);
if (r < 0)
log_warning_errno(r, "Can't determine if image of %s exists, freeing unfixated user: %m", h->user_name);
else if (r == USER_TEST_ABSENT)
log_info("Image for %s disappeared, freeing unfixated user.", h->user_name);
else
return;
home_free(h);
} else if (h->state < 0) {
r = user_record_test_home_directory(h->record);
if (r < 0) {
log_warning_errno(r, "Unable to determine state of home directory, ignoring: %m");
return;
}
if (r == USER_TEST_MOUNTED) {
r = user_record_test_image_path(h->record);
if (r < 0) {
log_warning_errno(r, "Unable to determine state of image path, ignoring: %m");
return;
}
if (r == USER_TEST_ABSENT) {
_cleanup_(operation_unrefp) Operation *o = NULL;
log_notice("Backing image disappeared while home directory %s was mounted, unmounting it forcibly.", h->user_name);
/* Wowza, the thing is mounted, but the device is gone? Act on it. */
r = home_killall(h);
if (r < 0)
log_warning_errno(r, "Failed to kill processes of user %s, ignoring: %m", h->user_name);
/* We enqueue the operation here, after all the home directory might
* currently already run some operation, and we can deactivate it only after
* that's complete. */
o = operation_new(OPERATION_DEACTIVATE_FORCE, NULL);
if (!o) {
log_oom();
return;
}
r = home_schedule_operation(h, o, NULL);
if (r < 0)
log_warning_errno(r, "Failed to enqueue forced home directory %s deactivation, ignoring: %m", h->user_name);
}
}
}
}
int manager_gc_images(Manager *m) {
Home *h;
assert_se(m);
if (m->gc_focus) {
/* Focus on a specific home */
h = TAKE_PTR(m->gc_focus);
manager_revalidate_image(m, h);
} else {
/* Gc all */
HASHMAP_FOREACH(h, m->homes_by_uid)
manager_revalidate_image(m, h);
}
return 0;
}
static int manager_gc_blob(Manager *m) {
_cleanup_closedir_ DIR *d = NULL;
int r;
assert(m);
d = opendir(home_system_blob_dir());
if (!d) {
if (errno == ENOENT)
return 0;
return log_error_errno(errno, "Failed to open %s: %m", home_system_blob_dir());
}
FOREACH_DIRENT(de, d, return log_error_errno(errno, "Failed to read system blob directory: %m")) {
Home *found = hashmap_get(m->homes_by_name, de->d_name);
if (!found || !streq(found->user_name, de->d_name)) {
r = rm_rf_at(dirfd(d), de->d_name, REMOVE_ROOT|REMOVE_PHYSICAL|REMOVE_SUBVOLUME);
if (r < 0)
log_warning_errno(r, "Failed to delete blob dir for missing user '%s', ignoring: %m", de->d_name);
}
}
return 0;
}
static int on_deferred_rescan(sd_event_source *s, void *userdata) {
Manager *m = ASSERT_PTR(userdata);
m->deferred_rescan_event_source = sd_event_source_disable_unref(m->deferred_rescan_event_source);
manager_enumerate_devices(m);
manager_enumerate_images(m);
return 0;
}
int manager_enqueue_rescan(Manager *m) {
int r;
assert(m);
if (m->deferred_rescan_event_source)
return 0;
if (!m->event)
return 0;
if (IN_SET(sd_event_get_state(m->event), SD_EVENT_FINISHED, SD_EVENT_EXITING))
return 0;
r = sd_event_add_defer(m->event, &m->deferred_rescan_event_source, on_deferred_rescan, m);
if (r < 0)
return log_error_errno(r, "Failed to allocate rescan event source: %m");
r = sd_event_source_set_priority(m->deferred_rescan_event_source, SD_EVENT_PRIORITY_IDLE+1);
if (r < 0)
log_warning_errno(r, "Failed to tweak priority of event source, ignoring: %m");
(void) sd_event_source_set_description(m->deferred_rescan_event_source, "deferred-rescan");
return 1;
}
static int on_deferred_gc(sd_event_source *s, void *userdata) {
Manager *m = ASSERT_PTR(userdata);
m->deferred_gc_event_source = sd_event_source_disable_unref(m->deferred_gc_event_source);
manager_gc_images(m);
return 0;
}
int manager_enqueue_gc(Manager *m, Home *focus) {
int r;
assert(m);
/* This enqueues a request to GC dead homes. It may be called with focus=NULL in which case all homes
* will be scanned, or with the parameter set, in which case only that home is checked. */
if (!m->event)
return 0;
if (IN_SET(sd_event_get_state(m->event), SD_EVENT_FINISHED, SD_EVENT_EXITING))
return 0;
/* If a focus home is specified, then remember to focus just on this home. Otherwise invalidate any
* focus that might be set to look at all homes. */
if (m->deferred_gc_event_source) {
if (m->gc_focus != focus) /* not the same focus, then look at everything */
m->gc_focus = NULL;
return 0;
} else
m->gc_focus = focus; /* start focused */
r = sd_event_add_defer(m->event, &m->deferred_gc_event_source, on_deferred_gc, m);
if (r < 0)
return log_error_errno(r, "Failed to allocate GC event source: %m");
r = sd_event_source_set_priority(m->deferred_gc_event_source, SD_EVENT_PRIORITY_IDLE);
if (r < 0)
log_warning_errno(r, "Failed to tweak priority of event source, ignoring: %m");
(void) sd_event_source_set_description(m->deferred_gc_event_source, "deferred-gc");
return 1;
}
static bool manager_shall_rebalance(Manager *m) {
Home *h;
assert(m);
if (IN_SET(m->rebalance_state, REBALANCE_PENDING, REBALANCE_SHRINKING, REBALANCE_GROWING))
return true;
HASHMAP_FOREACH(h, m->homes_by_uid)
if (home_shall_rebalance(h))
return true;
return false;
}
static int home_cmp(Home *const*a, Home *const*b) {
int r;
assert(a);
assert(*a);
assert(b);
assert(*b);
/* Order user records by their weight (and by their name, to make things stable). We put the records
* with the highest weight last, since we distribute space from the beginning and round down, hence
* later entries tend to get slightly more than earlier entries. */
r = CMP(user_record_rebalance_weight((*a)->record), user_record_rebalance_weight((*b)->record));
if (r != 0)
return r;
return strcmp((*a)->user_name, (*b)->user_name);
}
static int manager_rebalance_calculate(Manager *m) {
uint64_t weight_sum, free_sum, usage_sum = 0, min_free = UINT64_MAX;
_cleanup_free_ Home **array = NULL;
bool relevant = false;
struct statfs sfs;
int c = 0, r;
Home *h;
assert(m);
if (statfs(get_home_root(), &sfs) < 0)
return log_error_errno(errno, "Failed to statfs() /home: %m");
free_sum = (uint64_t) sfs.f_bsize * sfs.f_bavail; /* This much free space is available on the
* underlying pool directory */
weight_sum = REBALANCE_WEIGHT_BACKING; /* Grant the underlying pool directory a fixed weight of 20
* (home dirs get 100 by default, i.e. 5x more). This weight
* is not configurable, the per-home weights are. */
HASHMAP_FOREACH(h, m->homes_by_uid) {
statfs_f_type_t fstype;
h->rebalance_pending = false; /* First, reset the flag, we only want it to be true for the
* homes that qualify for rebalancing */
if (!home_shall_rebalance(h)) /* Only look at actual candidates */
continue;
if (home_is_busy(h))
return -EBUSY; /* Let's not rebalance if there's a busy home directory. */
r = home_get_disk_status(
h,
&h->rebalance_size,
&h->rebalance_usage,
&h->rebalance_free,
NULL,
NULL,
&fstype,
NULL);
if (r < 0) {
log_warning_errno(r, "Failed to get free space of home '%s', ignoring.", h->user_name);
continue;
}
if (h->rebalance_free > UINT64_MAX - free_sum)
return log_error_errno(SYNTHETIC_ERRNO(EOVERFLOW), "Rebalance free overflow");
free_sum += h->rebalance_free;
if (h->rebalance_usage > UINT64_MAX - usage_sum)
return log_error_errno(SYNTHETIC_ERRNO(EOVERFLOW), "Rebalance usage overflow");
usage_sum += h->rebalance_usage;
h->rebalance_weight = user_record_rebalance_weight(h->record);
if (h->rebalance_weight > UINT64_MAX - weight_sum)
return log_error_errno(SYNTHETIC_ERRNO(EOVERFLOW), "Rebalance weight overflow");
weight_sum += h->rebalance_weight;
h->rebalance_min = minimal_size_by_fs_magic(fstype);
if (!GREEDY_REALLOC(array, c+1))
return log_oom();
array[c++] = h;
}
if (c == 0) {
log_debug("No homes to rebalance.");
return 0;
}
assert(weight_sum > 0);
log_debug("Disk space usage by all home directories to rebalance: %s — available disk space: %s",
FORMAT_BYTES(usage_sum), FORMAT_BYTES(free_sum));
/* Bring the home directories in a well-defined order, so that we distribute space in a reproducible
* way for the same parameters. */
typesafe_qsort(array, c, home_cmp);
for (int i = 0; i < c; i++) {
uint64_t new_free;
double d;
h = array[i];
assert(h->rebalance_free <= free_sum);
assert(h->rebalance_usage <= usage_sum);
assert(h->rebalance_weight <= weight_sum);
d = free_sum / 4096.0 * h->rebalance_weight / weight_sum; /* Calculate new space for this home in units of 4K */
/* Convert from units of 4K back to bytes */
if (d >= UINT64_MAX / 4096)
new_free = UINT64_MAX;
else
new_free = (uint64_t) d * 4096;
/* Subtract the weight and assigned space from the sums now, to distribute the rounding noise
* to the remaining home dirs */
free_sum = LESS_BY(free_sum, new_free);
weight_sum = LESS_BY(weight_sum, h->rebalance_weight);
/* Keep track of home directory with the least amount of space left: we want to schedule the
* next rebalance more quickly if this is low */
if (new_free < min_free)
min_free = new_free;
if (new_free > UINT64_MAX - h->rebalance_usage)
h->rebalance_goal = UINT64_MAX-1; /* maximum size */
else {
h->rebalance_goal = h->rebalance_usage + new_free;
if (h->rebalance_min != UINT64_MAX && h->rebalance_goal < h->rebalance_min)
h->rebalance_goal = h->rebalance_min;
}
/* Skip over this home if the state doesn't match the operation */
if ((m->rebalance_state == REBALANCE_SHRINKING && h->rebalance_goal > h->rebalance_size) ||
(m->rebalance_state == REBALANCE_GROWING && h->rebalance_goal < h->rebalance_size))
h->rebalance_pending = false;
else {
log_debug("Rebalancing home directory '%s' %s %s %s.", h->user_name,
FORMAT_BYTES(h->rebalance_size),
glyph(GLYPH_ARROW_RIGHT),
FORMAT_BYTES(h->rebalance_goal));
h->rebalance_pending = true;
}
if (ABS_DIFF(h->rebalance_size, h->rebalance_goal) * 100.0 / h->rebalance_size >= 5.0)
relevant = true;
}
/* Scale next rebalancing interval based on the least amount of space of any of the home
* directories. We pick a time in the range 1min … 15min, scaled by floor(log2(min_free)),
* so that: 10M → ~0.6min, 100M → ~2.3min, 1G → ~4.0min, 10G → ~6.3min, 100G → ~8.1min */
m->rebalance_interval_usec = (usec_t) CLAMP((LESS_BY(log2u64(min_free), 22u)*15*USEC_PER_MINUTE)/26,
1 * USEC_PER_MINUTE,
15 * USEC_PER_MINUTE);
log_debug("Rebalancing interval set to %s.", FORMAT_TIMESPAN(m->rebalance_interval_usec, USEC_PER_MSEC));
/* Let's suppress small resizes, growing/shrinking file systems isn't free after all */
if (!relevant) {
log_debug("Skipping rebalancing, since all calculated size changes are below ±5%%.");
return 0;
}
return c;
}
static int manager_rebalance_apply(Manager *m) {
int c = 0, r;
Home *h;
assert(m);
HASHMAP_FOREACH(h, m->homes_by_uid) {
_cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
if (!h->rebalance_pending)
continue;
h->rebalance_pending = false;
r = home_resize(h, h->rebalance_goal, /* secret= */ NULL, &error);
if (r < 0)
log_warning_errno(r, "Failed to resize home '%s' for rebalancing, ignoring: %s",
h->user_name, bus_error_message(&error, r));
else
c++;
}
return c;
}
static void manager_rebalance_reply_messages(Manager *m) {
int r;
assert(m);
for (;;) {
_cleanup_(sd_bus_message_unrefp) sd_bus_message *msg =
set_steal_first(m->rebalance_pending_method_calls);
if (!msg)
break;