1 /* SPDX-License-Identifier: LGPL-2.1-or-later */
5 #include <linux/magic.h>
7 #include <openssl/pem.h>
10 #include <sys/quota.h>
13 #include "btrfs-util.h"
14 #include "bus-common-errors.h"
15 #include "bus-error.h"
16 #include "bus-log-control-api.h"
17 #include "bus-polkit.h"
18 #include "clean-ipc.h"
19 #include "conf-files.h"
20 #include "device-util.h"
21 #include "dirent-util.h"
24 #include "format-util.h"
27 #include "home-util.h"
28 #include "homed-conf.h"
29 #include "homed-home-bus.h"
30 #include "homed-home.h"
31 #include "homed-manager-bus.h"
32 #include "homed-manager.h"
33 #include "homed-varlink.h"
36 #include "process-util.h"
37 #include "quota-util.h"
38 #include "random-util.h"
39 #include "resize-fs.h"
40 #include "socket-util.h"
41 #include "sort-util.h"
42 #include "stat-util.h"
44 #include "sync-util.h"
45 #include "tmpfile-util.h"
46 #include "udev-util.h"
47 #include "user-record-sign.h"
48 #include "user-record-util.h"
49 #include "user-record.h"
50 #include "user-util.h"
52 /* Where to look for private/public keys that are used to sign the user records. We are not using
53 * CONF_PATHS_NULSTR() here since we want to insert /var/lib/systemd/home/ in the middle. And we insert that
54 * since we want to auto-generate a persistent private/public key pair if we need to. */
55 #define KEY_PATHS_NULSTR \
56 "/etc/systemd/home/\0" \
57 "/run/systemd/home/\0" \
58 "/var/lib/systemd/home/\0" \
59 "/usr/local/lib/systemd/home/\0" \
60 "/usr/lib/systemd/home/\0"
62 static bool uid_is_home(uid_t uid
) {
63 return uid
>= HOME_UID_MIN
&& uid
<= HOME_UID_MAX
;
65 /* Takes a value generated randomly or by hashing and turns it into a UID in the right range */
67 #define UID_CLAMP_INTO_HOME_RANGE(rnd) (((uid_t) (rnd) % (HOME_UID_MAX - HOME_UID_MIN + 1)) + HOME_UID_MIN)
69 DEFINE_PRIVATE_HASH_OPS_WITH_VALUE_DESTRUCTOR(homes_by_uid_hash_ops
, void, trivial_hash_func
, trivial_compare_func
, Home
, home_free
);
70 DEFINE_PRIVATE_HASH_OPS_WITH_VALUE_DESTRUCTOR(homes_by_name_hash_ops
, char, string_hash_func
, string_compare_func
, Home
, home_free
);
71 DEFINE_PRIVATE_HASH_OPS_WITH_VALUE_DESTRUCTOR(homes_by_worker_pid_hash_ops
, void, trivial_hash_func
, trivial_compare_func
, Home
, home_free
);
72 DEFINE_PRIVATE_HASH_OPS_WITH_VALUE_DESTRUCTOR(homes_by_sysfs_hash_ops
, char, path_hash_func
, path_compare
, Home
, home_free
);
74 static int on_home_inotify(sd_event_source
*s
, const struct inotify_event
*event
, void *userdata
);
75 static int manager_gc_images(Manager
*m
);
76 static int manager_enumerate_images(Manager
*m
);
77 static int manager_assess_image(Manager
*m
, int dir_fd
, const char *dir_path
, const char *dentry_name
);
78 static void manager_revalidate_image(Manager
*m
, Home
*h
);
80 static void manager_watch_home(Manager
*m
) {
86 m
->inotify_event_source
= sd_event_source_disable_unref(m
->inotify_event_source
);
87 m
->scan_slash_home
= false;
89 if (statfs(get_home_root(), &sfs
) < 0) {
90 log_full_errno(errno
== ENOENT
? LOG_DEBUG
: LOG_WARNING
, errno
,
91 "Failed to statfs() %s directory, disabling automatic scanning.", get_home_root());
95 if (is_network_fs(&sfs
)) {
96 log_info("%s is a network file system, disabling automatic scanning.", get_home_root());
100 if (is_fs_type(&sfs
, AUTOFS_SUPER_MAGIC
)) {
101 log_info("%s is on autofs, disabling automatic scanning.", get_home_root());
105 m
->scan_slash_home
= true;
107 r
= sd_event_add_inotify(m
->event
, &m
->inotify_event_source
, get_home_root(),
108 IN_CREATE
|IN_CLOSE_WRITE
|IN_DELETE_SELF
|IN_MOVE_SELF
|IN_ONLYDIR
|IN_MOVED_TO
|IN_MOVED_FROM
|IN_DELETE
,
111 log_full_errno(r
== -ENOENT
? LOG_DEBUG
: LOG_WARNING
, r
,
112 "Failed to create inotify watch on %s, ignoring.", get_home_root());
114 (void) sd_event_source_set_description(m
->inotify_event_source
, "home-inotify");
116 log_info("Watching %s.", get_home_root());
119 static int on_home_inotify(sd_event_source
*s
, const struct inotify_event
*event
, void *userdata
) {
120 _cleanup_free_
char *j
= NULL
;
121 Manager
*m
= userdata
;
127 if ((event
->mask
& (IN_Q_OVERFLOW
|IN_MOVE_SELF
|IN_DELETE_SELF
|IN_IGNORED
|IN_UNMOUNT
)) != 0) {
129 if (FLAGS_SET(event
->mask
, IN_Q_OVERFLOW
))
130 log_debug("%s inotify queue overflow, rescanning.", get_home_root());
131 else if (FLAGS_SET(event
->mask
, IN_MOVE_SELF
))
132 log_info("%s moved or renamed, recreating watch and rescanning.", get_home_root());
133 else if (FLAGS_SET(event
->mask
, IN_DELETE_SELF
))
134 log_info("%s deleted, recreating watch and rescanning.", get_home_root());
135 else if (FLAGS_SET(event
->mask
, IN_UNMOUNT
))
136 log_info("%s unmounted, recreating watch and rescanning.", get_home_root());
137 else if (FLAGS_SET(event
->mask
, IN_IGNORED
))
138 log_info("%s watch invalidated, recreating watch and rescanning.", get_home_root());
140 manager_watch_home(m
);
141 (void) manager_gc_images(m
);
142 (void) manager_enumerate_images(m
);
143 (void) bus_manager_emit_auto_login_changed(m
);
147 /* For the other inotify events, let's ignore all events for file names that don't match our
149 if (isempty(event
->name
))
151 e
= endswith(event
->name
, FLAGS_SET(event
->mask
, IN_ISDIR
) ? ".homedir" : ".home");
155 n
= strndupa_safe(event
->name
, e
- event
->name
);
156 if (!suitable_user_name(n
))
159 j
= path_join(get_home_root(), event
->name
);
163 if ((event
->mask
& (IN_CREATE
|IN_CLOSE_WRITE
|IN_MOVED_TO
)) != 0) {
164 if (FLAGS_SET(event
->mask
, IN_CREATE
))
165 log_debug("%s has been created, having a look.", j
);
166 else if (FLAGS_SET(event
->mask
, IN_CLOSE_WRITE
))
167 log_debug("%s has been modified, having a look.", j
);
168 else if (FLAGS_SET(event
->mask
, IN_MOVED_TO
))
169 log_debug("%s has been moved in, having a look.", j
);
171 (void) manager_assess_image(m
, -1, get_home_root(), event
->name
);
172 (void) bus_manager_emit_auto_login_changed(m
);
175 if ((event
->mask
& (IN_DELETE
| IN_CLOSE_WRITE
| IN_MOVED_FROM
)) != 0) {
178 if (FLAGS_SET(event
->mask
, IN_DELETE
))
179 log_debug("%s has been deleted, revalidating.", j
);
180 else if (FLAGS_SET(event
->mask
, IN_CLOSE_WRITE
))
181 log_debug("%s has been closed after writing, revalidating.", j
);
182 else if (FLAGS_SET(event
->mask
, IN_MOVED_FROM
))
183 log_debug("%s has been moved away, revalidating.", j
);
185 h
= hashmap_get(m
->homes_by_name
, n
);
187 manager_revalidate_image(m
, h
);
188 (void) bus_manager_emit_auto_login_changed(m
);
195 int manager_new(Manager
**ret
) {
196 _cleanup_(manager_freep
) Manager
*m
= NULL
;
206 .default_storage
= _USER_STORAGE_INVALID
,
207 .rebalance_interval_usec
= 2 * USEC_PER_MINUTE
, /* initially, rebalance every 2min */
210 r
= manager_parse_config_file(m
);
214 r
= sd_event_default(&m
->event
);
218 r
= sd_event_add_signal(m
->event
, NULL
, SIGINT
, NULL
, NULL
);
222 r
= sd_event_add_signal(m
->event
, NULL
, SIGTERM
, NULL
, NULL
);
226 (void) sd_event_set_watchdog(m
->event
, true);
228 m
->homes_by_uid
= hashmap_new(&homes_by_uid_hash_ops
);
229 if (!m
->homes_by_uid
)
232 m
->homes_by_name
= hashmap_new(&homes_by_name_hash_ops
);
233 if (!m
->homes_by_name
)
236 m
->homes_by_worker_pid
= hashmap_new(&homes_by_worker_pid_hash_ops
);
237 if (!m
->homes_by_worker_pid
)
240 m
->homes_by_sysfs
= hashmap_new(&homes_by_sysfs_hash_ops
);
241 if (!m
->homes_by_sysfs
)
248 Manager
* manager_free(Manager
*m
) {
253 HASHMAP_FOREACH(h
, m
->homes_by_worker_pid
)
254 (void) home_wait_for_worker(h
);
256 sd_bus_flush_close_unref(m
->bus
);
257 bus_verify_polkit_async_registry_free(m
->polkit_registry
);
259 m
->device_monitor
= sd_device_monitor_unref(m
->device_monitor
);
261 m
->inotify_event_source
= sd_event_source_unref(m
->inotify_event_source
);
262 m
->notify_socket_event_source
= sd_event_source_unref(m
->notify_socket_event_source
);
263 m
->deferred_rescan_event_source
= sd_event_source_unref(m
->deferred_rescan_event_source
);
264 m
->deferred_gc_event_source
= sd_event_source_unref(m
->deferred_gc_event_source
);
265 m
->deferred_auto_login_event_source
= sd_event_source_unref(m
->deferred_auto_login_event_source
);
266 m
->rebalance_event_source
= sd_event_source_unref(m
->rebalance_event_source
);
268 sd_event_unref(m
->event
);
270 hashmap_free(m
->homes_by_uid
);
271 hashmap_free(m
->homes_by_name
);
272 hashmap_free(m
->homes_by_worker_pid
);
273 hashmap_free(m
->homes_by_sysfs
);
276 EVP_PKEY_free(m
->private_key
);
278 hashmap_free(m
->public_keys
);
280 varlink_server_unref(m
->varlink_server
);
281 free(m
->userdb_service
);
283 free(m
->default_file_system_type
);
288 int manager_verify_user_record(Manager
*m
, UserRecord
*hr
) {
295 if (!m
->private_key
&& hashmap_isempty(m
->public_keys
)) {
296 r
= user_record_has_signature(hr
);
300 return r
? -ENOKEY
: USER_RECORD_UNSIGNED
;
304 if (m
->private_key
) {
305 r
= user_record_verify(hr
, m
->private_key
);
308 case USER_RECORD_FOREIGN
:
309 /* This record is not signed by this key, but let's see below */
312 case USER_RECORD_SIGNED
: /* Signed by us, but also by others, let's propagate that */
313 case USER_RECORD_SIGNED_EXCLUSIVE
: /* Signed by us, and nothing else, ditto */
314 case USER_RECORD_UNSIGNED
: /* Not signed at all, ditto */
320 HASHMAP_FOREACH(pkey
, m
->public_keys
) {
321 r
= user_record_verify(hr
, pkey
);
324 case USER_RECORD_FOREIGN
:
325 /* This record is not signed by this key, but let's see our other keys */
328 case USER_RECORD_SIGNED
: /* It's signed by this key we are happy with, but which is not our own. */
329 case USER_RECORD_SIGNED_EXCLUSIVE
:
330 return USER_RECORD_FOREIGN
;
332 case USER_RECORD_UNSIGNED
: /* It's not signed at all */
341 static int manager_add_home_by_record(
347 _cleanup_(json_variant_unrefp
) JsonVariant
*v
= NULL
;
348 _cleanup_(user_record_unrefp
) UserRecord
*hr
= NULL
;
349 unsigned line
, column
;
358 if (fstatat(dir_fd
, fname
, &st
, 0) < 0)
359 return log_error_errno(errno
, "Failed to stat identity record %s: %m", fname
);
361 if (!S_ISREG(st
.st_mode
)) {
362 log_debug("Identity record file %s is not a regular file, ignoring.", fname
);
367 goto unlink_this_file
;
369 r
= json_parse_file_at(NULL
, dir_fd
, fname
, JSON_PARSE_SENSITIVE
, &v
, &line
, &column
);
371 return log_error_errno(r
, "Failed to parse identity record at %s:%u%u: %m", fname
, line
, column
);
373 if (json_variant_is_blank_object(v
))
374 goto unlink_this_file
;
376 hr
= user_record_new();
380 r
= user_record_load(hr
, v
, USER_RECORD_LOAD_REFUSE_SECRET
|USER_RECORD_LOG
|USER_RECORD_PERMISSIVE
);
384 if (!streq_ptr(hr
->user_name
, name
))
385 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
),
386 "Identity's user name %s does not match file name %s, refusing.",
387 hr
->user_name
, name
);
389 is_signed
= manager_verify_user_record(m
, hr
);
393 return log_warning_errno(is_signed
, "User record %s is not signed by any accepted key, ignoring.", fname
);
394 case USER_RECORD_UNSIGNED
:
395 return log_warning_errno(SYNTHETIC_ERRNO(EPERM
), "User record %s is not signed at all, ignoring.", fname
);
396 case USER_RECORD_SIGNED
:
397 log_info("User record %s is signed by us (and others), accepting.", fname
);
399 case USER_RECORD_SIGNED_EXCLUSIVE
:
400 log_info("User record %s is signed only by us, accepting.", fname
);
402 case USER_RECORD_FOREIGN
:
403 log_info("User record %s is signed by registered key from others, accepting.", fname
);
406 assert(is_signed
< 0);
407 return log_error_errno(is_signed
, "Failed to verify signature of user record in %s: %m", fname
);
410 h
= hashmap_get(m
->homes_by_name
, name
);
412 r
= home_set_record(h
, hr
);
414 return log_error_errno(r
, "Failed to update home record for %s: %m", name
);
416 /* If we acquired a record now for a previously unallocated entry, then reset the state. This
417 * makes sure home_get_state() will check for the availability of the image file dynamically
418 * in order to detect to distinguish HOME_INACTIVE and HOME_ABSENT. */
419 if (h
->state
== HOME_UNFIXATED
)
420 h
->state
= _HOME_STATE_INVALID
;
422 r
= home_new(m
, hr
, NULL
, &h
);
424 return log_error_errno(r
, "Failed to allocate new home object: %m");
426 log_info("Added registered home for user %s.", hr
->user_name
);
429 /* Only entries we exclusively signed are writable to us, hence remember the result */
430 h
->signed_locally
= is_signed
== USER_RECORD_SIGNED_EXCLUSIVE
;
435 /* If this is an empty file, then let's just remove it. An empty file is not useful in any case, and
436 * apparently xfs likes to leave empty files around when not unmounted cleanly (see
437 * https://github.com/systemd/systemd/issues/15178 for example). Note that we don't delete non-empty
438 * files even if they are invalid, because that's just too risky, we might delete data the user still
439 * needs. But empty files are never useful, hence let's just remove them. */
441 if (unlinkat(dir_fd
, fname
, 0) < 0)
442 return log_error_errno(errno
, "Failed to remove empty user record file %s: %m", fname
);
444 log_notice("Discovered empty user record file %s/%s, removed automatically.", home_record_dir(), fname
);
448 static int manager_enumerate_records(Manager
*m
) {
449 _cleanup_closedir_
DIR *d
= NULL
;
454 d
= opendir(home_record_dir());
456 return log_full_errno(errno
== ENOENT
? LOG_DEBUG
: LOG_ERR
, errno
,
457 "Failed to open %s: %m", home_record_dir());
459 FOREACH_DIRENT(de
, d
, return log_error_errno(errno
, "Failed to read record directory: %m")) {
460 _cleanup_free_
char *n
= NULL
;
463 if (!dirent_is_file(de
))
466 e
= endswith(de
->d_name
, ".identity");
470 n
= strndup(de
->d_name
, e
- de
->d_name
);
474 if (!suitable_user_name(n
))
477 (void) manager_add_home_by_record(m
, n
, dirfd(d
), de
->d_name
);
483 static int search_quota(uid_t uid
, const char *exclude_quota_path
) {
484 struct stat exclude_st
= {};
485 dev_t previous_devno
= 0;
489 /* Checks whether the specified UID owns any files on the files system, but ignore any file system
490 * backing the specified file. The file is used when operating on home directories, where it's OK if
491 * the UID of them already owns files. */
493 if (exclude_quota_path
&& stat(exclude_quota_path
, &exclude_st
) < 0) {
495 return log_warning_errno(errno
, "Failed to stat %s, ignoring: %m", exclude_quota_path
);
498 /* Check a few usual suspects where regular users might own files. Note that this is by no means
499 * comprehensive, but should cover most cases. Note that in an ideal world every user would be
500 * registered in NSS and avoid our own UID range, but for all other cases, it's a good idea to be
501 * paranoid and check quota if we can. */
502 FOREACH_STRING(where
, get_home_root(), "/tmp/", "/var/", "/var/mail/", "/var/tmp/", "/var/spool/") {
506 if (stat(where
, &st
) < 0) {
507 log_full_errno(errno
== ENOENT
? LOG_DEBUG
: LOG_ERR
, errno
,
508 "Failed to stat %s, ignoring: %m", where
);
512 if (major(st
.st_dev
) == 0) {
513 log_debug("Directory %s is not on a real block device, not checking quota for UID use.", where
);
517 if (st
.st_dev
== exclude_st
.st_dev
) { /* If an exclude path is specified, then ignore quota
518 * reported on the same block device as that path. */
519 log_debug("Directory %s is where the home directory is located, not checking quota for UID use.", where
);
523 if (st
.st_dev
== previous_devno
) { /* Does this directory have the same devno as the previous
524 * one we tested? If so, there's no point in testing this
526 log_debug("Directory %s is on same device as previous tested directory, not checking quota for UID use a second time.", where
);
530 previous_devno
= st
.st_dev
;
532 r
= quotactl_devno(QCMD_FIXED(Q_GETQUOTA
, USRQUOTA
), st
.st_dev
, uid
, &req
);
534 if (ERRNO_IS_NOT_SUPPORTED(r
))
535 log_debug_errno(r
, "No UID quota support on %s, ignoring.", where
);
536 else if (ERRNO_IS_PRIVILEGE(r
))
537 log_debug_errno(r
, "UID quota support for %s prohibited, ignoring.", where
);
539 log_warning_errno(r
, "Failed to query quota on %s, ignoring: %m", where
);
544 if ((FLAGS_SET(req
.dqb_valid
, QIF_SPACE
) && req
.dqb_curspace
> 0) ||
545 (FLAGS_SET(req
.dqb_valid
, QIF_INODES
) && req
.dqb_curinodes
> 0)) {
546 log_debug_errno(errno
, "Quota reports UID " UID_FMT
" occupies disk space on %s.", uid
, where
);
554 static int manager_acquire_uid(
557 const char *user_name
,
558 const char *exclude_quota_path
,
561 static const uint8_t hash_key
[] = {
562 0xa3, 0xb8, 0x82, 0x69, 0x9a, 0x71, 0xf7, 0xa9,
563 0xe0, 0x7c, 0xf6, 0xf1, 0x21, 0x69, 0xd2, 0x1e
570 } phase
= PHASE_SUGGESTED
;
572 unsigned n_tries
= 100;
589 case PHASE_SUGGESTED
:
590 phase
= PHASE_HASHED
;
592 if (!uid_is_home(start_uid
))
595 candidate
= start_uid
;
599 phase
= PHASE_RANDOM
;
604 candidate
= UID_CLAMP_INTO_HOME_RANGE(siphash24(user_name
, strlen(user_name
), hash_key
));
608 random_bytes(&candidate
, sizeof(candidate
));
609 candidate
= UID_CLAMP_INTO_HOME_RANGE(candidate
);
613 assert_not_reached();
616 other
= hashmap_get(m
->homes_by_uid
, UID_TO_PTR(candidate
));
618 log_debug("Candidate UID " UID_FMT
" already used by another home directory (%s), let's try another.",
619 candidate
, other
->user_name
);
623 pw
= getpwuid(candidate
);
625 log_debug("Candidate UID " UID_FMT
" already registered by another user in NSS (%s), let's try another.",
626 candidate
, pw
->pw_name
);
630 gr
= getgrgid((gid_t
) candidate
);
632 log_debug("Candidate UID " UID_FMT
" already registered by another group in NSS (%s), let's try another.",
633 candidate
, gr
->gr_name
);
637 r
= search_ipc(candidate
, (gid_t
) candidate
);
641 log_debug_errno(r
, "Candidate UID " UID_FMT
" already owns IPC objects, let's try another: %m",
646 r
= search_quota(candidate
, exclude_quota_path
);
655 static int manager_add_home_by_image(
657 const char *user_name
,
659 const char *image_path
,
664 _cleanup_(user_record_unrefp
) UserRecord
*hr
= NULL
;
674 assert(storage
>= 0);
675 assert(storage
< _USER_STORAGE_MAX
);
677 h
= hashmap_get(m
->homes_by_name
, user_name
);
681 if (h
->state
!= HOME_UNFIXATED
) {
682 log_debug("Found an image for user %s which already has a record, skipping.", user_name
);
683 return 0; /* ignore images that synthesize a user we already have a record for */
686 same
= user_record_storage(h
->record
) == storage
;
688 if (h
->sysfs
&& sysfs
)
689 same
= path_equal(h
->sysfs
, sysfs
);
690 else if (!!h
->sysfs
!= !!sysfs
)
695 p
= user_record_image_path(h
->record
);
696 same
= p
&& path_equal(p
, image_path
);
701 log_debug("Found multiple images for user '%s', ignoring image '%s'.", user_name
, image_path
);
705 /* Check NSS, in case there's another user or group by this name */
706 if (getpwnam(user_name
) || getgrnam(user_name
)) {
707 log_debug("Found an existing user or group by name '%s', ignoring image '%s'.", user_name
, image_path
);
712 if (h
&& uid_is_valid(h
->uid
))
715 r
= manager_acquire_uid(m
, start_uid
, user_name
,
716 IN_SET(storage
, USER_SUBVOLUME
, USER_DIRECTORY
, USER_FSCRYPT
) ? image_path
: NULL
,
719 return log_warning_errno(r
, "Failed to acquire unused UID for %s: %m", user_name
);
722 hr
= user_record_new();
726 r
= user_record_synthesize(hr
, user_name
, realm
, image_path
, storage
, uid
, (gid_t
) uid
);
728 return log_error_errno(r
, "Failed to synthesize home record for %s (image %s): %m", user_name
, image_path
);
731 r
= home_set_record(h
, hr
);
733 return log_error_errno(r
, "Failed to update home record for %s: %m", user_name
);
735 r
= home_new(m
, hr
, sysfs
, &h
);
737 return log_error_errno(r
, "Failed to allocate new home object: %m");
739 h
->state
= HOME_UNFIXATED
;
741 log_info("Discovered new home for user %s through image %s.", user_name
, image_path
);
747 int manager_augment_record_with_uid(
751 const char *exclude_quota_path
= NULL
;
752 uid_t start_uid
= UID_INVALID
, uid
;
758 if (uid_is_valid(hr
->uid
))
761 if (IN_SET(hr
->storage
, USER_CLASSIC
, USER_SUBVOLUME
, USER_DIRECTORY
, USER_FSCRYPT
)) {
764 ip
= user_record_image_path(hr
);
768 if (stat(ip
, &st
) < 0) {
770 log_warning_errno(errno
, "Failed to stat(%s): %m", ip
);
771 } else if (uid_is_home(st
.st_uid
)) {
772 start_uid
= st
.st_uid
;
773 exclude_quota_path
= ip
;
778 r
= manager_acquire_uid(m
, start_uid
, hr
->user_name
, exclude_quota_path
, &uid
);
782 log_debug("Acquired new UID " UID_FMT
" for %s.", uid
, hr
->user_name
);
784 r
= user_record_add_binding(
786 _USER_STORAGE_INVALID
,
804 static int manager_assess_image(
807 const char *dir_path
,
808 const char *dentry_name
) {
810 char *luks_suffix
, *directory_suffix
;
811 _cleanup_free_
char *path
= NULL
;
819 luks_suffix
= endswith(dentry_name
, ".home");
821 directory_suffix
= NULL
;
823 directory_suffix
= endswith(dentry_name
, ".homedir");
825 /* Early filter out: by name */
826 if (!luks_suffix
&& !directory_suffix
)
829 path
= path_join(dir_path
, dentry_name
);
833 /* Follow symlinks here, to allow people to link in stuff to make them available locally. */
835 r
= fstatat(dir_fd
, dentry_name
, &st
, 0);
839 return log_full_errno(errno
== ENOENT
? LOG_DEBUG
: LOG_WARNING
, errno
,
840 "Failed to stat() directory entry '%s', ignoring: %m", dentry_name
);
842 if (S_ISREG(st
.st_mode
)) {
843 _cleanup_free_
char *n
= NULL
, *user_name
= NULL
, *realm
= NULL
;
848 n
= strndup(dentry_name
, luks_suffix
- dentry_name
);
852 r
= split_user_name_realm(n
, &user_name
, &realm
);
853 if (r
== -EINVAL
) /* Not the right format: ignore */
856 return log_error_errno(r
, "Failed to split image name into user name/realm: %m");
858 return manager_add_home_by_image(m
, user_name
, realm
, path
, NULL
, USER_LUKS
, UID_INVALID
);
861 if (S_ISDIR(st
.st_mode
)) {
862 _cleanup_free_
char *n
= NULL
, *user_name
= NULL
, *realm
= NULL
;
863 _cleanup_close_
int fd
= -1;
866 if (!directory_suffix
)
869 n
= strndup(dentry_name
, directory_suffix
- dentry_name
);
873 r
= split_user_name_realm(n
, &user_name
, &realm
);
874 if (r
== -EINVAL
) /* Not the right format: ignore */
877 return log_error_errno(r
, "Failed to split image name into user name/realm: %m");
880 fd
= openat(dir_fd
, dentry_name
, O_DIRECTORY
|O_RDONLY
|O_CLOEXEC
);
882 fd
= open(path
, O_DIRECTORY
|O_RDONLY
|O_CLOEXEC
);
884 return log_full_errno(errno
== ENOENT
? LOG_DEBUG
: LOG_WARNING
, errno
,
885 "Failed to open directory '%s', ignoring: %m", path
);
887 if (fstat(fd
, &st
) < 0)
888 return log_warning_errno(errno
, "Failed to fstat() %s, ignoring: %m", path
);
890 assert(S_ISDIR(st
.st_mode
)); /* Must hold, we used O_DIRECTORY above */
892 r
= btrfs_is_subvol_fd(fd
);
894 return log_warning_errno(errno
, "Failed to determine whether %s is a btrfs subvolume: %m", path
);
896 storage
= USER_SUBVOLUME
;
898 struct fscrypt_policy policy
;
900 if (ioctl(fd
, FS_IOC_GET_ENCRYPTION_POLICY
, &policy
) < 0) {
902 if (errno
== ENODATA
)
903 log_debug_errno(errno
, "Determined %s is not fscrypt encrypted.", path
);
904 else if (ERRNO_IS_NOT_SUPPORTED(errno
))
905 log_debug_errno(errno
, "Determined %s is not fscrypt encrypted because kernel or file system doesn't support it.", path
);
907 log_debug_errno(errno
, "FS_IOC_GET_ENCRYPTION_POLICY failed with unexpected error code on %s, ignoring: %m", path
);
909 storage
= USER_DIRECTORY
;
911 storage
= USER_FSCRYPT
;
914 return manager_add_home_by_image(m
, user_name
, realm
, path
, NULL
, storage
, st
.st_uid
);
920 int manager_enumerate_images(Manager
*m
) {
921 _cleanup_closedir_
DIR *d
= NULL
;
926 if (!m
->scan_slash_home
)
929 d
= opendir(get_home_root());
931 return log_full_errno(errno
== ENOENT
? LOG_DEBUG
: LOG_ERR
, errno
,
932 "Failed to open %s: %m", get_home_root());
934 FOREACH_DIRENT(de
, d
, return log_error_errno(errno
, "Failed to read %s directory: %m", get_home_root()))
935 (void) manager_assess_image(m
, dirfd(d
), get_home_root(), de
->d_name
);
940 static int manager_connect_bus(Manager
*m
) {
941 const char *suffix
, *busname
;
947 r
= sd_bus_default_system(&m
->bus
);
949 return log_error_errno(r
, "Failed to connect to system bus: %m");
951 r
= bus_add_implementation(m
->bus
, &manager_object
, m
);
955 r
= bus_log_control_api_register(m
->bus
);
959 suffix
= getenv("SYSTEMD_HOME_DEBUG_SUFFIX");
961 busname
= strjoina("org.freedesktop.home1.", suffix
);
963 busname
= "org.freedesktop.home1";
965 r
= sd_bus_request_name_async(m
->bus
, NULL
, busname
, 0, NULL
, NULL
);
967 return log_error_errno(r
, "Failed to request name: %m");
969 r
= sd_bus_attach_event(m
->bus
, m
->event
, 0);
971 return log_error_errno(r
, "Failed to attach bus to event loop: %m");
973 (void) sd_bus_set_exit_on_disconnect(m
->bus
, true);
978 static int manager_bind_varlink(Manager
*m
) {
979 const char *suffix
, *socket_path
;
983 assert(!m
->varlink_server
);
985 r
= varlink_server_new(&m
->varlink_server
, VARLINK_SERVER_ACCOUNT_UID
|VARLINK_SERVER_INHERIT_USERDATA
);
987 return log_error_errno(r
, "Failed to allocate varlink server object: %m");
989 varlink_server_set_userdata(m
->varlink_server
, m
);
991 r
= varlink_server_bind_method_many(
993 "io.systemd.UserDatabase.GetUserRecord", vl_method_get_user_record
,
994 "io.systemd.UserDatabase.GetGroupRecord", vl_method_get_group_record
,
995 "io.systemd.UserDatabase.GetMemberships", vl_method_get_memberships
);
997 return log_error_errno(r
, "Failed to register varlink methods: %m");
999 (void) mkdir_p("/run/systemd/userdb", 0755);
1001 /* To make things easier to debug, when working from a homed managed home directory, let's optionally
1002 * use a different varlink socket name */
1003 suffix
= getenv("SYSTEMD_HOME_DEBUG_SUFFIX");
1005 socket_path
= strjoina("/run/systemd/userdb/io.systemd.Home.", suffix
);
1007 socket_path
= "/run/systemd/userdb/io.systemd.Home";
1009 r
= varlink_server_listen_address(m
->varlink_server
, socket_path
, 0666);
1011 return log_error_errno(r
, "Failed to bind to varlink socket: %m");
1013 r
= varlink_server_attach_event(m
->varlink_server
, m
->event
, SD_EVENT_PRIORITY_NORMAL
);
1015 return log_error_errno(r
, "Failed to attach varlink connection to event loop: %m");
1017 assert(!m
->userdb_service
);
1018 m
->userdb_service
= strdup(basename(socket_path
));
1019 if (!m
->userdb_service
)
1022 /* Avoid recursion */
1023 if (setenv("SYSTEMD_BYPASS_USERDB", m
->userdb_service
, 1) < 0)
1024 return log_error_errno(SYNTHETIC_ERRNO(EINVAL
), "Failed to set $SYSTEMD_BYPASS_USERDB: %m");
1029 static ssize_t
read_datagram(
1031 struct ucred
*ret_sender
,
1033 int *ret_passed_fd
) {
1035 CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(struct ucred
)) + CMSG_SPACE(sizeof(int))) control
;
1036 _cleanup_free_
void *buffer
= NULL
;
1037 _cleanup_close_
int passed_fd
= -1;
1038 struct ucred
*sender
= NULL
;
1039 struct cmsghdr
*cmsg
;
1047 assert(ret_passed_fd
);
1049 n
= next_datagram_size_fd(fd
);
1053 buffer
= malloc(n
+ 2);
1057 /* Pass one extra byte, as a size check */
1058 iov
= IOVEC_MAKE(buffer
, n
+ 1);
1060 mh
= (struct msghdr
) {
1063 .msg_control
= &control
,
1064 .msg_controllen
= sizeof(control
),
1067 m
= recvmsg_safe(fd
, &mh
, MSG_DONTWAIT
|MSG_CMSG_CLOEXEC
);
1071 /* Ensure the size matches what we determined before */
1073 cmsg_close_all(&mh
);
1077 CMSG_FOREACH(cmsg
, &mh
) {
1078 if (cmsg
->cmsg_level
== SOL_SOCKET
&&
1079 cmsg
->cmsg_type
== SCM_CREDENTIALS
&&
1080 cmsg
->cmsg_len
== CMSG_LEN(sizeof(struct ucred
))) {
1082 sender
= (struct ucred
*) CMSG_DATA(cmsg
);
1085 if (cmsg
->cmsg_level
== SOL_SOCKET
&&
1086 cmsg
->cmsg_type
== SCM_RIGHTS
) {
1088 if (cmsg
->cmsg_len
!= CMSG_LEN(sizeof(int))) {
1089 cmsg_close_all(&mh
);
1093 assert(passed_fd
< 0);
1094 passed_fd
= *(int*) CMSG_DATA(cmsg
);
1099 *ret_sender
= *sender
;
1101 *ret_sender
= (struct ucred
) UCRED_INVALID
;
1103 *ret_passed_fd
= TAKE_FD(passed_fd
);
1105 /* For safety reasons: let's always NUL terminate. */
1106 ((char*) buffer
)[n
] = 0;
1107 *ret
= TAKE_PTR(buffer
);
1112 static int on_notify_socket(sd_event_source
*s
, int fd
, uint32_t revents
, void *userdata
) {
1113 _cleanup_strv_free_
char **l
= NULL
;
1114 _cleanup_free_
void *datagram
= NULL
;
1115 _cleanup_close_
int passed_fd
= -1;
1116 struct ucred sender
= UCRED_INVALID
;
1117 Manager
*m
= userdata
;
1124 n
= read_datagram(fd
, &sender
, &datagram
, &passed_fd
);
1126 if (ERRNO_IS_TRANSIENT(n
))
1128 return log_error_errno(n
, "Failed to read notify datagram: %m");
1131 if (sender
.pid
<= 0) {
1132 log_warning("Received notify datagram without valid sender PID, ignoring.");
1136 h
= hashmap_get(m
->homes_by_worker_pid
, PID_TO_PTR(sender
.pid
));
1138 log_warning("Received notify datagram of unknown process, ignoring.");
1142 l
= strv_split(datagram
, "\n");
1146 home_process_notify(h
, l
, TAKE_FD(passed_fd
));
1150 static int manager_listen_notify(Manager
*m
) {
1151 _cleanup_close_
int fd
= -1;
1152 union sockaddr_union sa
= {
1153 .un
.sun_family
= AF_UNIX
,
1154 .un
.sun_path
= "/run/systemd/home/notify",
1160 assert(!m
->notify_socket_event_source
);
1162 suffix
= getenv("SYSTEMD_HOME_DEBUG_SUFFIX");
1164 const char *unix_path
;
1166 unix_path
= strjoina("/run/systemd/home/notify.", suffix
);
1167 r
= sockaddr_un_set_path(&sa
.un
, unix_path
);
1169 return log_error_errno(r
, "Socket path %s does not fit in sockaddr_un: %m", unix_path
);
1172 fd
= socket(AF_UNIX
, SOCK_DGRAM
|SOCK_CLOEXEC
|SOCK_NONBLOCK
, 0);
1174 return log_error_errno(errno
, "Failed to create listening socket: %m");
1176 (void) mkdir_parents(sa
.un
.sun_path
, 0755);
1177 (void) sockaddr_un_unlink(&sa
.un
);
1179 if (bind(fd
, &sa
.sa
, SOCKADDR_UN_LEN(sa
.un
)) < 0)
1180 return log_error_errno(errno
, "Failed to bind to socket: %m");
1182 r
= setsockopt_int(fd
, SOL_SOCKET
, SO_PASSCRED
, true);
1186 r
= sd_event_add_io(m
->event
, &m
->notify_socket_event_source
, fd
, EPOLLIN
, on_notify_socket
, m
);
1188 return log_error_errno(r
, "Failed to allocate event source for notify socket: %m");
1190 (void) sd_event_source_set_description(m
->notify_socket_event_source
, "notify-socket");
1192 /* Make sure we process sd_notify() before SIGCHLD for any worker, so that we always know the error
1193 * number of a client before it exits. */
1194 r
= sd_event_source_set_priority(m
->notify_socket_event_source
, SD_EVENT_PRIORITY_NORMAL
- 5);
1196 return log_error_errno(r
, "Failed to alter priority of NOTIFY_SOCKET event source: %m");
1198 r
= sd_event_source_set_io_fd_own(m
->notify_socket_event_source
, true);
1200 return log_error_errno(r
, "Failed to pass ownership of notify socket: %m");
1205 static int manager_add_device(Manager
*m
, sd_device
*d
) {
1206 _cleanup_free_
char *user_name
= NULL
, *realm
= NULL
, *node
= NULL
;
1207 const char *tabletype
, *parttype
, *partname
, *partuuid
, *sysfs
;
1214 r
= sd_device_get_syspath(d
, &sysfs
);
1216 return log_error_errno(r
, "Failed to acquire sysfs path of device: %m");
1218 r
= sd_device_get_property_value(d
, "ID_PART_TABLE_TYPE", &tabletype
);
1222 return log_error_errno(r
, "Failed to acquire ID_PART_TABLE_TYPE device property, ignoring: %m");
1224 if (!streq(tabletype
, "gpt")) {
1225 log_debug("Found partition (%s) on non-GPT table, ignoring.", sysfs
);
1229 r
= sd_device_get_property_value(d
, "ID_PART_ENTRY_TYPE", &parttype
);
1233 return log_error_errno(r
, "Failed to acquire ID_PART_ENTRY_TYPE device property, ignoring: %m");
1234 r
= sd_id128_from_string(parttype
, &id
);
1236 return log_debug_errno(r
, "Failed to parse ID_PART_ENTRY_TYPE field '%s', ignoring: %m", parttype
);
1237 if (!sd_id128_equal(id
, GPT_USER_HOME
)) {
1238 log_debug("Found partition (%s) we don't care about, ignoring.", sysfs
);
1242 r
= sd_device_get_property_value(d
, "ID_PART_ENTRY_NAME", &partname
);
1244 return log_warning_errno(r
, "Failed to acquire ID_PART_ENTRY_NAME device property, ignoring: %m");
1246 r
= split_user_name_realm(partname
, &user_name
, &realm
);
1248 return log_warning_errno(r
, "Found partition with correct partition type but a non-parsable partition name '%s', ignoring.", partname
);
1250 return log_error_errno(r
, "Failed to validate partition name '%s': %m", partname
);
1252 r
= sd_device_get_property_value(d
, "ID_FS_UUID", &partuuid
);
1254 return log_warning_errno(r
, "Failed to acquire ID_FS_UUID device property, ignoring: %m");
1256 r
= sd_id128_from_string(partuuid
, &id
);
1258 return log_warning_errno(r
, "Failed to parse ID_FS_UUID field '%s', ignoring: %m", partuuid
);
1260 if (asprintf(&node
, "/dev/disk/by-uuid/" SD_ID128_UUID_FORMAT_STR
, SD_ID128_FORMAT_VAL(id
)) < 0)
1263 return manager_add_home_by_image(m
, user_name
, realm
, node
, sysfs
, USER_LUKS
, UID_INVALID
);
1266 static int manager_on_device(sd_device_monitor
*monitor
, sd_device
*d
, void *userdata
) {
1267 Manager
*m
= userdata
;
1273 if (device_for_action(d
, SD_DEVICE_REMOVE
)) {
1277 r
= sd_device_get_syspath(d
, &sysfs
);
1279 log_warning_errno(r
, "Failed to acquire sysfs path from device: %m");
1283 log_info("block device %s has been removed.", sysfs
);
1285 /* Let's see if we previously synthesized a home record from this device, if so, let's just
1286 * revalidate that. Otherwise let's revalidate them all, but asynchronously. */
1287 h
= hashmap_get(m
->homes_by_sysfs
, sysfs
);
1289 manager_revalidate_image(m
, h
);
1291 manager_enqueue_gc(m
, NULL
);
1293 (void) manager_add_device(m
, d
);
1295 (void) bus_manager_emit_auto_login_changed(m
);
1299 static int manager_watch_devices(Manager
*m
) {
1303 assert(!m
->device_monitor
);
1305 r
= sd_device_monitor_new(&m
->device_monitor
);
1307 return log_error_errno(r
, "Failed to allocate device monitor: %m");
1309 r
= sd_device_monitor_filter_add_match_subsystem_devtype(m
->device_monitor
, "block", NULL
);
1311 return log_error_errno(r
, "Failed to configure device monitor match: %m");
1313 r
= sd_device_monitor_attach_event(m
->device_monitor
, m
->event
);
1315 return log_error_errno(r
, "Failed to attach device monitor to event loop: %m");
1317 r
= sd_device_monitor_start(m
->device_monitor
, manager_on_device
, m
);
1319 return log_error_errno(r
, "Failed to start device monitor: %m");
1324 static int manager_enumerate_devices(Manager
*m
) {
1325 _cleanup_(sd_device_enumerator_unrefp
) sd_device_enumerator
*e
= NULL
;
1331 r
= sd_device_enumerator_new(&e
);
1335 r
= sd_device_enumerator_add_match_subsystem(e
, "block", true);
1339 FOREACH_DEVICE(e
, d
)
1340 (void) manager_add_device(m
, d
);
1345 static int manager_load_key_pair(Manager
*m
) {
1346 _cleanup_(fclosep
) FILE *f
= NULL
;
1352 if (m
->private_key
) {
1353 EVP_PKEY_free(m
->private_key
);
1354 m
->private_key
= NULL
;
1357 r
= search_and_fopen_nulstr("local.private", "re", NULL
, KEY_PATHS_NULSTR
, &f
, NULL
);
1361 return log_error_errno(r
, "Failed to read private key file: %m");
1363 if (fstat(fileno(f
), &st
) < 0)
1364 return log_error_errno(errno
, "Failed to stat private key file: %m");
1366 r
= stat_verify_regular(&st
);
1368 return log_error_errno(r
, "Private key file is not regular: %m");
1370 if (st
.st_uid
!= 0 || (st
.st_mode
& 0077) != 0)
1371 return log_error_errno(SYNTHETIC_ERRNO(EPERM
), "Private key file is readable by more than the root user");
1373 m
->private_key
= PEM_read_PrivateKey(f
, NULL
, NULL
, NULL
);
1374 if (!m
->private_key
)
1375 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Failed to load private key pair");
1377 log_info("Successfully loaded private key pair.");
1382 DEFINE_TRIVIAL_CLEANUP_FUNC_FULL(EVP_PKEY_CTX
*, EVP_PKEY_CTX_free
, NULL
);
1384 static int manager_generate_key_pair(Manager
*m
) {
1385 _cleanup_(EVP_PKEY_CTX_freep
) EVP_PKEY_CTX
*ctx
= NULL
;
1386 _cleanup_(unlink_and_freep
) char *temp_public
= NULL
, *temp_private
= NULL
;
1387 _cleanup_fclose_
FILE *fpublic
= NULL
, *fprivate
= NULL
;
1390 if (m
->private_key
) {
1391 EVP_PKEY_free(m
->private_key
);
1392 m
->private_key
= NULL
;
1395 ctx
= EVP_PKEY_CTX_new_id(EVP_PKEY_ED25519
, NULL
);
1397 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Failed to allocate Ed25519 key generation context.");
1399 if (EVP_PKEY_keygen_init(ctx
) <= 0)
1400 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Failed to initialize Ed25519 key generation context.");
1402 log_info("Generating key pair for signing local user identity records.");
1404 if (EVP_PKEY_keygen(ctx
, &m
->private_key
) <= 0)
1405 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Failed to generate Ed25519 key pair");
1407 log_info("Successfully created Ed25519 key pair.");
1409 (void) mkdir_p("/var/lib/systemd/home", 0755);
1411 /* Write out public key (note that we only do that as a help to the user, we don't make use of this ever */
1412 r
= fopen_temporary("/var/lib/systemd/home/local.public", &fpublic
, &temp_public
);
1414 return log_error_errno(errno
, "Failed to open key file for writing: %m");
1416 if (PEM_write_PUBKEY(fpublic
, m
->private_key
) <= 0)
1417 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Failed to write public key.");
1419 r
= fflush_sync_and_check(fpublic
);
1421 return log_error_errno(r
, "Failed to write private key: %m");
1423 fpublic
= safe_fclose(fpublic
);
1425 /* Write out the private key (this actually writes out both private and public, OpenSSL is confusing) */
1426 r
= fopen_temporary("/var/lib/systemd/home/local.private", &fprivate
, &temp_private
);
1428 return log_error_errno(errno
, "Failed to open key file for writing: %m");
1430 if (PEM_write_PrivateKey(fprivate
, m
->private_key
, NULL
, NULL
, 0, NULL
, 0) <= 0)
1431 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Failed to write private key pair.");
1433 r
= fflush_sync_and_check(fprivate
);
1435 return log_error_errno(r
, "Failed to write private key: %m");
1437 fprivate
= safe_fclose(fprivate
);
1439 /* Both are written now, move them into place */
1441 if (rename(temp_public
, "/var/lib/systemd/home/local.public") < 0)
1442 return log_error_errno(errno
, "Failed to move public key file into place: %m");
1443 temp_public
= mfree(temp_public
);
1445 if (rename(temp_private
, "/var/lib/systemd/home/local.private") < 0) {
1446 (void) unlink_noerrno("/var/lib/systemd/home/local.public"); /* try to remove the file we already created */
1447 return log_error_errno(errno
, "Failed to move private key file into place: %m");
1449 temp_private
= mfree(temp_private
);
1451 r
= fsync_path_at(AT_FDCWD
, "/var/lib/systemd/home/");
1453 log_warning_errno(r
, "Failed to sync /var/lib/systemd/home/, ignoring: %m");
1458 int manager_acquire_key_pair(Manager
*m
) {
1463 /* Already there? */
1467 /* First try to load key off disk */
1468 r
= manager_load_key_pair(m
);
1472 /* Didn't work, generate a new one */
1473 return manager_generate_key_pair(m
);
1476 int manager_sign_user_record(Manager
*m
, UserRecord
*u
, UserRecord
**ret
, sd_bus_error
*error
) {
1483 r
= manager_acquire_key_pair(m
);
1487 return sd_bus_error_set(error
, BUS_ERROR_NO_PRIVATE_KEY
, "Can't sign without local key.");
1489 return user_record_sign(u
, m
->private_key
, ret
);
1492 DEFINE_PRIVATE_HASH_OPS_FULL(public_key_hash_ops
, char, string_hash_func
, string_compare_func
, free
, EVP_PKEY
, EVP_PKEY_free
);
1493 DEFINE_TRIVIAL_CLEANUP_FUNC_FULL(EVP_PKEY
*, EVP_PKEY_free
, NULL
);
1495 static int manager_load_public_key_one(Manager
*m
, const char *path
) {
1496 _cleanup_(EVP_PKEY_freep
) EVP_PKEY
*pkey
= NULL
;
1497 _cleanup_fclose_
FILE *f
= NULL
;
1498 _cleanup_free_
char *fn
= NULL
;
1504 if (streq(basename(path
), "local.public")) /* we already loaded the private key, which includes the public one */
1507 f
= fopen(path
, "re");
1509 if (errno
== ENOENT
)
1512 return log_error_errno(errno
, "Failed to open public key %s: %m", path
);
1515 if (fstat(fileno(f
), &st
) < 0)
1516 return log_error_errno(errno
, "Failed to stat public key %s: %m", path
);
1518 r
= stat_verify_regular(&st
);
1520 return log_error_errno(r
, "Public key file %s is not a regular file: %m", path
);
1522 if (st
.st_uid
!= 0 || (st
.st_mode
& 0022) != 0)
1523 return log_error_errno(SYNTHETIC_ERRNO(EPERM
), "Public key file %s is writable by more than the root user, refusing.", path
);
1525 r
= hashmap_ensure_allocated(&m
->public_keys
, &public_key_hash_ops
);
1529 pkey
= PEM_read_PUBKEY(f
, &pkey
, NULL
, NULL
);
1531 return log_error_errno(SYNTHETIC_ERRNO(EIO
), "Failed to parse public key file %s.", path
);
1533 fn
= strdup(basename(path
));
1537 r
= hashmap_put(m
->public_keys
, fn
, pkey
);
1539 return log_error_errno(r
, "Failed to add public key to set: %m");
1547 static int manager_load_public_keys(Manager
*m
) {
1548 _cleanup_strv_free_
char **files
= NULL
;
1554 m
->public_keys
= hashmap_free(m
->public_keys
);
1556 r
= conf_files_list_nulstr(
1560 CONF_FILES_REGULAR
|CONF_FILES_FILTER_MASKED
,
1563 return log_error_errno(r
, "Failed to assemble list of public key directories: %m");
1565 STRV_FOREACH(i
, files
)
1566 (void) manager_load_public_key_one(m
, *i
);
1571 int manager_startup(Manager
*m
) {
1576 r
= manager_listen_notify(m
);
1580 r
= manager_connect_bus(m
);
1584 r
= manager_bind_varlink(m
);
1588 r
= manager_load_key_pair(m
); /* only try to load it, don't generate any */
1592 r
= manager_load_public_keys(m
);
1596 manager_watch_home(m
);
1597 (void) manager_watch_devices(m
);
1599 (void) manager_enumerate_records(m
);
1600 (void) manager_enumerate_images(m
);
1601 (void) manager_enumerate_devices(m
);
1603 /* Let's clean up home directories whose devices got removed while we were not running */
1604 (void) manager_enqueue_gc(m
, NULL
);
1609 void manager_revalidate_image(Manager
*m
, Home
*h
) {
1615 /* Frees an automatically discovered image, if it's synthetic and its image disappeared. Unmounts any
1616 * image if it's mounted but it's image vanished. */
1618 if (h
->current_operation
|| !ordered_set_isempty(h
->pending_operations
))
1621 if (h
->state
== HOME_UNFIXATED
) {
1622 r
= user_record_test_image_path(h
->record
);
1624 log_warning_errno(r
, "Can't determine if image of %s exists, freeing unfixated user: %m", h
->user_name
);
1625 else if (r
== USER_TEST_ABSENT
)
1626 log_info("Image for %s disappeared, freeing unfixated user.", h
->user_name
);
1632 } else if (h
->state
< 0) {
1634 r
= user_record_test_home_directory(h
->record
);
1636 log_warning_errno(r
, "Unable to determine state of home directory, ignoring: %m");
1640 if (r
== USER_TEST_MOUNTED
) {
1641 r
= user_record_test_image_path(h
->record
);
1643 log_warning_errno(r
, "Unable to determine state of image path, ignoring: %m");
1647 if (r
== USER_TEST_ABSENT
) {
1648 _cleanup_(operation_unrefp
) Operation
*o
= NULL
;
1650 log_notice("Backing image disappeared while home directory %s was mounted, unmounting it forcibly.", h
->user_name
);
1651 /* Wowza, the thing is mounted, but the device is gone? Act on it. */
1653 r
= home_killall(h
);
1655 log_warning_errno(r
, "Failed to kill processes of user %s, ignoring: %m", h
->user_name
);
1657 /* We enqueue the operation here, after all the home directory might
1658 * currently already run some operation, and we can deactivate it only after
1659 * that's complete. */
1660 o
= operation_new(OPERATION_DEACTIVATE_FORCE
, NULL
);
1666 r
= home_schedule_operation(h
, o
, NULL
);
1668 log_warning_errno(r
, "Failed to enqueue forced home directory %s deactivation, ignoring: %m", h
->user_name
);
1674 int manager_gc_images(Manager
*m
) {
1680 /* Focus on a specific home */
1682 h
= TAKE_PTR(m
->gc_focus
);
1683 manager_revalidate_image(m
, h
);
1687 HASHMAP_FOREACH(h
, m
->homes_by_name
)
1688 manager_revalidate_image(m
, h
);
1694 static int on_deferred_rescan(sd_event_source
*s
, void *userdata
) {
1695 Manager
*m
= userdata
;
1699 m
->deferred_rescan_event_source
= sd_event_source_disable_unref(m
->deferred_rescan_event_source
);
1701 manager_enumerate_devices(m
);
1702 manager_enumerate_images(m
);
1706 int manager_enqueue_rescan(Manager
*m
) {
1711 if (m
->deferred_rescan_event_source
)
1717 if (IN_SET(sd_event_get_state(m
->event
), SD_EVENT_FINISHED
, SD_EVENT_EXITING
))
1720 r
= sd_event_add_defer(m
->event
, &m
->deferred_rescan_event_source
, on_deferred_rescan
, m
);
1722 return log_error_errno(r
, "Failed to allocate rescan event source: %m");
1724 r
= sd_event_source_set_priority(m
->deferred_rescan_event_source
, SD_EVENT_PRIORITY_IDLE
+1);
1726 log_warning_errno(r
, "Failed to tweak priority of event source, ignoring: %m");
1728 (void) sd_event_source_set_description(m
->deferred_rescan_event_source
, "deferred-rescan");
1732 static int on_deferred_gc(sd_event_source
*s
, void *userdata
) {
1733 Manager
*m
= userdata
;
1737 m
->deferred_gc_event_source
= sd_event_source_disable_unref(m
->deferred_gc_event_source
);
1739 manager_gc_images(m
);
1743 int manager_enqueue_gc(Manager
*m
, Home
*focus
) {
1748 /* This enqueues a request to GC dead homes. It may be called with focus=NULL in which case all homes
1749 * will be scanned, or with the parameter set, in which case only that home is checked. */
1754 if (IN_SET(sd_event_get_state(m
->event
), SD_EVENT_FINISHED
, SD_EVENT_EXITING
))
1757 /* If a focus home is specified, then remember to focus just on this home. Otherwise invalidate any
1758 * focus that might be set to look at all homes. */
1760 if (m
->deferred_gc_event_source
) {
1761 if (m
->gc_focus
!= focus
) /* not the same focus, then look at everything */
1766 m
->gc_focus
= focus
; /* start focused */
1768 r
= sd_event_add_defer(m
->event
, &m
->deferred_gc_event_source
, on_deferred_gc
, m
);
1770 return log_error_errno(r
, "Failed to allocate GC event source: %m");
1772 r
= sd_event_source_set_priority(m
->deferred_gc_event_source
, SD_EVENT_PRIORITY_IDLE
);
1774 log_warning_errno(r
, "Failed to tweak priority of event source, ignoring: %m");
1776 (void) sd_event_source_set_description(m
->deferred_gc_event_source
, "deferred-gc");
1780 static bool manager_shall_rebalance(Manager
*m
) {
1785 if (IN_SET(m
->rebalance_state
, REBALANCE_PENDING
, REBALANCE_SHRINKING
, REBALANCE_GROWING
))
1788 HASHMAP_FOREACH(h
, m
->homes_by_name
)
1789 if (home_shall_rebalance(h
))
1795 static int home_cmp(Home
*const*a
, Home
*const*b
) {
1803 /* Order user records by their weight (and by their name, to make things stable). We put the records
1804 * with the highest weight last, since we distribute space from the beginning and round down, hence
1805 * later entries tend to get slightly more than earlier entries. */
1807 r
= CMP(user_record_rebalance_weight((*a
)->record
), user_record_rebalance_weight((*b
)->record
));
1811 return strcmp((*a
)->user_name
, (*b
)->user_name
);
1814 static int manager_rebalance_calculate(Manager
*m
) {
1815 uint64_t weight_sum
, free_sum
, usage_sum
= 0, min_free
= UINT64_MAX
;
1816 _cleanup_free_ Home
**array
= NULL
;
1817 bool relevant
= false;
1824 if (statfs(get_home_root(), &sfs
) < 0)
1825 return log_error_errno(errno
, "Failed to statfs() /home: %m");
1827 free_sum
= (uint64_t) sfs
.f_bsize
* sfs
.f_bavail
; /* This much free space is available on the
1828 * underlying pool directory */
1830 weight_sum
= REBALANCE_WEIGHT_BACKING
; /* Grant the underlying pool directory a fixed weight of 20
1831 * (home dirs get 100 by default, i.e. 5x more). This weight
1832 * is not configurable, the per-home weights are. */
1834 HASHMAP_FOREACH(h
, m
->homes_by_name
) {
1835 statfs_f_type_t fstype
;
1836 h
->rebalance_pending
= false; /* First, reset the flag, we only want it to be true for the
1837 * homes that qualify for rebalancing */
1839 if (!home_shall_rebalance(h
)) /* Only look at actual candidates */
1842 if (home_is_busy(h
))
1843 return -EBUSY
; /* Let's not rebalance if there's a busy home directory. */
1845 r
= home_get_disk_status(
1848 &h
->rebalance_usage
,
1855 log_warning_errno(r
, "Failed to get free space of home '%s', ignoring.", h
->user_name
);
1859 if (h
->rebalance_free
> UINT64_MAX
- free_sum
)
1860 return log_error_errno(SYNTHETIC_ERRNO(EOVERFLOW
), "Rebalance free overflow");
1861 free_sum
+= h
->rebalance_free
;
1863 if (h
->rebalance_usage
> UINT64_MAX
- usage_sum
)
1864 return log_error_errno(SYNTHETIC_ERRNO(EOVERFLOW
), "Rebalance usage overflow");
1865 usage_sum
+= h
->rebalance_usage
;
1867 h
->rebalance_weight
= user_record_rebalance_weight(h
->record
);
1868 if (h
->rebalance_weight
> UINT64_MAX
- weight_sum
)
1869 return log_error_errno(SYNTHETIC_ERRNO(EOVERFLOW
), "Rebalance weight overflow");
1870 weight_sum
+= h
->rebalance_weight
;
1872 h
->rebalance_min
= minimal_size_by_fs_magic(fstype
);
1874 if (!GREEDY_REALLOC(array
, c
+1))
1881 log_debug("No homes to rebalance.");
1885 assert(weight_sum
> 0);
1887 log_debug("Disk space usage by all home directories to rebalance: %s — available disk space: %s",
1888 FORMAT_BYTES(usage_sum
), FORMAT_BYTES(free_sum
));
1890 /* Bring the home directories in a well-defined order, so that we distribute space in a reproducible
1891 * way for the same parameters. */
1892 typesafe_qsort(array
, c
, home_cmp
);
1894 for (int i
= 0; i
< c
; i
++) {
1900 assert(h
->rebalance_free
<= free_sum
);
1901 assert(h
->rebalance_usage
<= usage_sum
);
1902 assert(h
->rebalance_weight
<= weight_sum
);
1904 d
= ((double) (free_sum
/ 4096) * (double) h
->rebalance_weight
) / (double) weight_sum
; /* Calculate new space for this home in units of 4K */
1906 /* Convert from units of 4K back to bytes */
1907 if (d
>= (double) (UINT64_MAX
/4096))
1908 new_free
= UINT64_MAX
;
1910 new_free
= (uint64_t) d
* 4096;
1912 /* Subtract the weight and assigned space from the sums now, to distribute the rounding noise
1913 * to the remaining home dirs */
1914 free_sum
= LESS_BY(free_sum
, new_free
);
1915 weight_sum
= LESS_BY(weight_sum
, h
->rebalance_weight
);
1917 /* Keep track of home directory with the least amount of space left: we want to schedule the
1918 * next rebalance more quickly if this is low */
1919 if (new_free
< min_free
)
1920 min_free
= h
->rebalance_size
;
1922 if (new_free
> UINT64_MAX
- h
->rebalance_usage
)
1923 h
->rebalance_goal
= UINT64_MAX
-1; /* maximum size */
1925 h
->rebalance_goal
= h
->rebalance_usage
+ new_free
;
1927 if (h
->rebalance_min
!= UINT64_MAX
&& h
->rebalance_goal
< h
->rebalance_min
)
1928 h
->rebalance_goal
= h
->rebalance_min
;
1931 /* Skip over this home if the state doesn't match the operation */
1932 if ((m
->rebalance_state
== REBALANCE_SHRINKING
&& h
->rebalance_goal
> h
->rebalance_size
) ||
1933 (m
->rebalance_state
== REBALANCE_GROWING
&& h
->rebalance_goal
< h
->rebalance_size
))
1934 h
->rebalance_pending
= false;
1936 log_debug("Rebalancing home directory '%s' %s → %s.", h
->user_name
,
1937 FORMAT_BYTES(h
->rebalance_size
), FORMAT_BYTES(h
->rebalance_goal
));
1938 h
->rebalance_pending
= true;
1941 if ((fabs((double) h
->rebalance_size
- (double) h
->rebalance_goal
) * 100 / (double) h
->rebalance_size
) >= 5.0)
1945 /* Scale next rebalancing interval based on the least amount of space of any of the home
1946 * directories. We pick a time in the range 1min … 15min, scaled by log2(min_free), so that:
1947 * 10M → ~0.7min, 100M → ~2.7min, 1G → ~4.6min, 10G → ~6.5min, 100G ~8.4 */
1948 m
->rebalance_interval_usec
= (usec_t
) CLAMP((LESS_BY(log2(min_free
), 22)*15*USEC_PER_MINUTE
)/26,
1949 1 * USEC_PER_MINUTE
,
1950 15 * USEC_PER_MINUTE
);
1953 log_debug("Rebalancing interval set to %s.", FORMAT_TIMESPAN(m
->rebalance_interval_usec
, USEC_PER_MSEC
));
1955 /* Let's suppress small resizes, growing/shrinking file systems isn't free after all */
1957 log_debug("Skipping rebalancing, since all calculated size changes are below ±5%%.");
1964 static int manager_rebalance_apply(Manager
*m
) {
1970 HASHMAP_FOREACH(h
, m
->homes_by_name
) {
1971 _cleanup_(sd_bus_error_free
) sd_bus_error error
= SD_BUS_ERROR_NULL
;
1973 if (!h
->rebalance_pending
)
1976 h
->rebalance_pending
= false;
1978 r
= home_resize(h
, h
->rebalance_goal
, /* secret= */ NULL
, /* automatic= */ true, &error
);
1980 log_warning_errno(r
, "Failed to resize home '%s' for rebalancing, ignoring: %s",
1981 h
->user_name
, bus_error_message(&error
, r
));
1989 static void manager_rebalance_reply_messages(Manager
*m
) {
1995 _cleanup_(sd_bus_message_unrefp
) sd_bus_message
*msg
=
1996 set_steal_first(m
->rebalance_pending_method_calls
);
2001 r
= sd_bus_reply_method_return(msg
, NULL
);
2003 log_debug_errno(r
, "Failed to reply to rebalance method call, ignoring: %m");
2007 static int manager_rebalance_now(Manager
*m
) {
2008 RebalanceState busy_state
; /* the state to revert to when operation fails if busy */
2013 log_debug("Rebalancing now...");
2015 /* We maintain a simple state engine here to keep track of what we are doing. We'll first shrink all
2016 * homes that shall be shrunk and then grow all homes that shall be grown, so that they can take up
2017 * the space now freed. */
2020 switch (m
->rebalance_state
) {
2022 case REBALANCE_IDLE
:
2023 case REBALANCE_PENDING
:
2024 case REBALANCE_WAITING
:
2025 /* First shrink large home dirs */
2026 m
->rebalance_state
= REBALANCE_SHRINKING
;
2027 busy_state
= REBALANCE_PENDING
;
2029 /* We are initiating the next rebalancing cycle now, let's make the queued methods
2030 * calls the pending ones, and flush out any pending ones (which shouldn't exist at
2031 * this time anyway) */
2032 set_clear(m
->rebalance_pending_method_calls
);
2033 SWAP_TWO(m
->rebalance_pending_method_calls
, m
->rebalance_queued_method_calls
);
2035 log_debug("Shrinking phase..");
2038 case REBALANCE_SHRINKING
:
2039 /* Then grow small home dirs */
2040 m
->rebalance_state
= REBALANCE_GROWING
;
2041 busy_state
= REBALANCE_SHRINKING
;
2042 log_debug("Growing phase..");
2045 case REBALANCE_GROWING
:
2046 /* Finally, we are done */
2047 log_info("Rebalancing complete.");
2048 m
->rebalance_state
= REBALANCE_IDLE
;
2054 assert_not_reached();
2057 r
= manager_rebalance_calculate(m
);
2059 /* Calculations failed because one home directory is currently busy. Revert to a state that
2060 * tells us what to do next. */
2061 log_debug("Can't enter phase, busy.");
2062 m
->rebalance_state
= busy_state
;
2068 continue; /* got to next step immediately, if there's nothing to do */
2070 r
= manager_rebalance_apply(m
);
2074 break; /* At least one resize operation is now pending, we are done for now */
2076 /* If there was nothing to apply, go for next state right-away */
2082 /* Reset state and schedule next rebalance */
2083 m
->rebalance_state
= REBALANCE_IDLE
;
2084 manager_rebalance_reply_messages(m
);
2085 (void) manager_schedule_rebalance(m
, /* immediately= */ false);
2089 static int on_rebalance_timer(sd_event_source
*s
, usec_t t
, void *userdata
) {
2090 Manager
*m
= userdata
;
2094 assert(IN_SET(m
->rebalance_state
, REBALANCE_WAITING
, REBALANCE_PENDING
, REBALANCE_SHRINKING
, REBALANCE_GROWING
));
2096 (void) manager_rebalance_now(m
);
2100 int manager_schedule_rebalance(Manager
*m
, bool immediately
) {
2105 /* Check if there are any records where rebalancing is requested */
2106 if (!manager_shall_rebalance(m
)) {
2107 log_debug("Not scheduling rebalancing, not needed.");
2108 r
= 0; /* report that we didn't schedule anything because nothing needed it */
2113 /* If we are told to rebalance immediately, then mark a rebalance as pending (even if we area
2114 * already running one) */
2116 if (m
->rebalance_event_source
) {
2117 r
= sd_event_source_set_time(m
->rebalance_event_source
, 0);
2119 log_error_errno(r
, "Failed to schedule immediate rebalancing: %m");
2123 r
= sd_event_source_set_enabled(m
->rebalance_event_source
, SD_EVENT_ONESHOT
);
2125 log_error_errno(r
, "Failed to enable rebalancing event source: %m");
2129 r
= sd_event_add_time(m
->event
, &m
->rebalance_event_source
, CLOCK_MONOTONIC
, 0, USEC_PER_SEC
, on_rebalance_timer
, m
);
2131 log_error_errno(r
, "Failed to allocate rebalance event source: %m");
2135 r
= sd_event_source_set_priority(m
->rebalance_event_source
, SD_EVENT_PRIORITY_IDLE
+ 10);
2137 log_error_errno(r
, "Failed to set rebalance event source priority: %m");
2141 (void) sd_event_source_set_description(m
->rebalance_event_source
, "rebalance");
2145 if (!IN_SET(m
->rebalance_state
, REBALANCE_PENDING
, REBALANCE_SHRINKING
, REBALANCE_GROWING
))
2146 m
->rebalance_state
= REBALANCE_PENDING
;
2148 log_debug("Scheduled immediate rebalancing...");
2149 return 1; /* report that we scheduled something */
2152 /* If we are told to schedule a rebalancing eventually, then do so only if we are not executing
2153 * anything yet. Also if we have something scheduled already, leave it in place */
2154 if (!IN_SET(m
->rebalance_state
, REBALANCE_OFF
, REBALANCE_IDLE
))
2155 return 1; /* report that there's already something scheduled */
2157 if (m
->rebalance_event_source
) {
2158 r
= sd_event_source_set_time_relative(m
->rebalance_event_source
, m
->rebalance_interval_usec
);
2160 log_error_errno(r
, "Failed to schedule immediate rebalancing: %m");
2164 r
= sd_event_source_set_enabled(m
->rebalance_event_source
, SD_EVENT_ONESHOT
);
2166 log_error_errno(r
, "Failed to enable rebalancing event source: %m");
2170 r
= sd_event_add_time_relative(m
->event
, &m
->rebalance_event_source
, CLOCK_MONOTONIC
, m
->rebalance_interval_usec
, USEC_PER_SEC
, on_rebalance_timer
, m
);
2172 log_error_errno(r
, "Failed to allocate rebalance event source: %m");
2176 r
= sd_event_source_set_priority(m
->rebalance_event_source
, SD_EVENT_PRIORITY_IDLE
+ 10);
2178 log_error_errno(r
, "Failed to set rebalance event source priority: %m");
2182 (void) sd_event_source_set_description(m
->rebalance_event_source
, "rebalance");
2185 m
->rebalance_state
= REBALANCE_WAITING
; /* We managed to enqueue a timer event, we now wait until it fires */
2186 log_debug("Scheduled rebalancing in %s...", FORMAT_TIMESPAN(m
->rebalance_interval_usec
, 0));
2187 return 1; /* report that we scheduled something */
2190 m
->rebalance_event_source
= sd_event_source_disable_unref(m
->rebalance_event_source
);
2191 m
->rebalance_state
= REBALANCE_OFF
;
2192 manager_rebalance_reply_messages(m
);
2196 int manager_reschedule_rebalance(Manager
*m
) {
2201 /* If a rebalance is pending reschedules it so it gets executed immediately */
2203 if (!IN_SET(m
->rebalance_state
, REBALANCE_PENDING
, REBALANCE_SHRINKING
, REBALANCE_GROWING
))
2206 r
= manager_schedule_rebalance(m
, /* immediately= */ true);