]> git.ipfire.org Git - thirdparty/systemd.git/blame - src/home/homed-manager.c
projects / thirdparty / systemd.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Merge pull request #22759 from msekletar/issue-18077-long-sysfs-paths-hashing
[thirdparty/systemd.git] / src / home / homed-manager.c
CommitLineData
db9ecf05 1/* SPDX-License-Identifier: LGPL-2.1-or-later */
70a5db58
LP		 2
3#include <grp.h>
4#include <linux/fs.h>
5#include <linux/magic.h>
d357b80d 6#include <math.h>
70a5db58
LP		 7#include <openssl/pem.h>
8#include <pwd.h>
9#include <sys/ioctl.h>
10#include <sys/quota.h>
11#include <sys/stat.h>
12
13#include "btrfs-util.h"
14#include "bus-common-errors.h"
15#include "bus-error.h"
ac9f55ed 16#include "bus-log-control-api.h"
70a5db58
LP		 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"
22#include "fd-util.h"
23#include "fileio.h"
24#include "format-util.h"
25#include "fs-util.h"
26#include "gpt.h"
27#include "home-util.h"
c76dd733 28#include "homed-conf.h"
70a5db58
LP		 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"
34#include "io-util.h"
35#include "mkdir.h"
36#include "process-util.h"
37#include "quota-util.h"
38#include "random-util.h"
d357b80d 39#include "resize-fs.h"
70a5db58 40#include "socket-util.h"
d357b80d 41#include "sort-util.h"
70a5db58
LP		 42#include "stat-util.h"
43#include "strv.h"
bf819d3a 44#include "sync-util.h"
70a5db58
LP		 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"
51
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"
61
62static bool uid_is_home(uid_t uid) {
63 return uid >= HOME_UID_MIN && uid <= HOME_UID_MAX;
64}
65/* Takes a value generated randomly or by hashing and turns it into a UID in the right range */
66
67#define UID_CLAMP_INTO_HOME_RANGE(rnd) (((uid_t) (rnd) % (HOME_UID_MAX - HOME_UID_MIN + 1)) + HOME_UID_MIN)
68
69DEFINE_PRIVATE_HASH_OPS_WITH_VALUE_DESTRUCTOR(homes_by_uid_hash_ops, void, trivial_hash_func, trivial_compare_func, Home, home_free);
70DEFINE_PRIVATE_HASH_OPS_WITH_VALUE_DESTRUCTOR(homes_by_name_hash_ops, char, string_hash_func, string_compare_func, Home, home_free);
71DEFINE_PRIVATE_HASH_OPS_WITH_VALUE_DESTRUCTOR(homes_by_worker_pid_hash_ops, void, trivial_hash_func, trivial_compare_func, Home, home_free);
72DEFINE_PRIVATE_HASH_OPS_WITH_VALUE_DESTRUCTOR(homes_by_sysfs_hash_ops, char, path_hash_func, path_compare, Home, home_free);
73
74static int on_home_inotify(sd_event_source *s, const struct inotify_event *event, void *userdata);
75static int manager_gc_images(Manager *m);
76static int manager_enumerate_images(Manager *m);
77static int manager_assess_image(Manager *m, int dir_fd, const char *dir_path, const char *dentry_name);
78static void manager_revalidate_image(Manager *m, Home *h);
79
80static void manager_watch_home(Manager *m) {
81 struct statfs sfs;
82 int r;
83
84 assert(m);
85
cf536638 86 m->inotify_event_source = sd_event_source_disable_unref(m->inotify_event_source);
70a5db58
LP		 87 m->scan_slash_home = false;
88
2700fecd 89 if (statfs(get_home_root(), &sfs) < 0) {
70a5db58 90 log_full_errno(errno == ENOENT ? LOG_DEBUG : LOG_WARNING, errno,
2700fecd 91 "Failed to statfs() %s directory, disabling automatic scanning.", get_home_root());
70a5db58
LP		 92 return;
93 }
94
95 if (is_network_fs(&sfs)) {
2700fecd 96 log_info("%s is a network file system, disabling automatic scanning.", get_home_root());
70a5db58
LP		 97 return;
98 }
99
100 if (is_fs_type(&sfs, AUTOFS_SUPER_MAGIC)) {
2700fecd 101 log_info("%s is on autofs, disabling automatic scanning.", get_home_root());
70a5db58
LP		 102 return;
103 }
104
105 m->scan_slash_home = true;
106
2700fecd 107 r = sd_event_add_inotify(m->event, &m->inotify_event_source, get_home_root(),
23d24b76
ZJS		 108 IN_CREATE|IN_CLOSE_WRITE|IN_DELETE_SELF|IN_MOVE_SELF|IN_ONLYDIR|IN_MOVED_TO|IN_MOVED_FROM|IN_DELETE,
109 on_home_inotify, m);
70a5db58
LP		 110 if (r < 0)
111 log_full_errno(r == -ENOENT ? LOG_DEBUG : LOG_WARNING, r,
2700fecd 112 "Failed to create inotify watch on %s, ignoring.", get_home_root());
70a5db58
LP		 113
114 (void) sd_event_source_set_description(m->inotify_event_source, "home-inotify");
2700fecd
LP		 115
116 log_info("Watching %s.", get_home_root());
70a5db58
LP		 117}
118
119static int on_home_inotify(sd_event_source *s, const struct inotify_event *event, void *userdata) {
2700fecd 120 _cleanup_free_ char *j = NULL;
70a5db58
LP		 121 Manager *m = userdata;
122 const char *e, *n;
123
124 assert(m);
125 assert(event);
126
127 if ((event->mask & (IN_Q_OVERFLOW|IN_MOVE_SELF|IN_DELETE_SELF|IN_IGNORED|IN_UNMOUNT)) != 0) {
128
129 if (FLAGS_SET(event->mask, IN_Q_OVERFLOW))
2700fecd 130 log_debug("%s inotify queue overflow, rescanning.", get_home_root());
70a5db58 131 else if (FLAGS_SET(event->mask, IN_MOVE_SELF))
2700fecd 132 log_info("%s moved or renamed, recreating watch and rescanning.", get_home_root());
70a5db58 133 else if (FLAGS_SET(event->mask, IN_DELETE_SELF))
2700fecd 134 log_info("%s deleted, recreating watch and rescanning.", get_home_root());
70a5db58 135 else if (FLAGS_SET(event->mask, IN_UNMOUNT))
2700fecd 136 log_info("%s unmounted, recreating watch and rescanning.", get_home_root());
70a5db58 137 else if (FLAGS_SET(event->mask, IN_IGNORED))
2700fecd 138 log_info("%s watch invalidated, recreating watch and rescanning.", get_home_root());
70a5db58
LP		 139
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);
144 return 0;
145 }
146
147 /* For the other inotify events, let's ignore all events for file names that don't match our
148 * expectations */
149 if (isempty(event->name))
150 return 0;
151 e = endswith(event->name, FLAGS_SET(event->mask, IN_ISDIR) ? ".homedir" : ".home");
152 if (!e)
153 return 0;
154
2f82562b 155 n = strndupa_safe(event->name, e - event->name);
70a5db58
LP		 156 if (!suitable_user_name(n))
157 return 0;
158
2700fecd
LP		 159 j = path_join(get_home_root(), event->name);
160 if (!j)
161 return log_oom();
162
70a5db58
LP		 163 if ((event->mask & (IN_CREATE|IN_CLOSE_WRITE|IN_MOVED_TO)) != 0) {
164 if (FLAGS_SET(event->mask, IN_CREATE))
2700fecd 165 log_debug("%s has been created, having a look.", j);
70a5db58 166 else if (FLAGS_SET(event->mask, IN_CLOSE_WRITE))
2700fecd 167 log_debug("%s has been modified, having a look.", j);
70a5db58 168 else if (FLAGS_SET(event->mask, IN_MOVED_TO))
2700fecd 169 log_debug("%s has been moved in, having a look.", j);
70a5db58 170
2700fecd 171 (void) manager_assess_image(m, -1, get_home_root(), event->name);
70a5db58
LP		 172 (void) bus_manager_emit_auto_login_changed(m);
173 }
174
755b35b1 175 if ((event->mask & (IN_DELETE | IN_CLOSE_WRITE | IN_MOVED_FROM)) != 0) {
70a5db58
LP		 176 Home *h;
177
178 if (FLAGS_SET(event->mask, IN_DELETE))
2700fecd 179 log_debug("%s has been deleted, revalidating.", j);
70a5db58 180 else if (FLAGS_SET(event->mask, IN_CLOSE_WRITE))
2700fecd 181 log_debug("%s has been closed after writing, revalidating.", j);
70a5db58 182 else if (FLAGS_SET(event->mask, IN_MOVED_FROM))
2700fecd 183 log_debug("%s has been moved away, revalidating.", j);
70a5db58
LP		 184
185 h = hashmap_get(m->homes_by_name, n);
186 if (h) {
187 manager_revalidate_image(m, h);
188 (void) bus_manager_emit_auto_login_changed(m);
189 }
190 }
191
192 return 0;
193}
194
195int manager_new(Manager **ret) {
196 _cleanup_(manager_freep) Manager *m = NULL;
197 int r;
198
199 assert(ret);
200
c76dd733 201 m = new(Manager, 1);
70a5db58
LP		 202 if (!m)
203 return -ENOMEM;
204
c76dd733
LP		 205 *m = (Manager) {
206 .default_storage = _USER_STORAGE_INVALID,
d357b80d 207 .rebalance_interval_usec = 2 * USEC_PER_MINUTE, /* initially, rebalance every 2min */
c76dd733
LP		 208 };
209
210 r = manager_parse_config_file(m);
211 if (r < 0)
212 return r;
213
70a5db58
LP		 214 r = sd_event_default(&m->event);
215 if (r < 0)
216 return r;
217
218 r = sd_event_add_signal(m->event, NULL, SIGINT, NULL, NULL);
219 if (r < 0)
220 return r;
221
222 r = sd_event_add_signal(m->event, NULL, SIGTERM, NULL, NULL);
223 if (r < 0)
224 return r;
225
226 (void) sd_event_set_watchdog(m->event, true);
227
228 m->homes_by_uid = hashmap_new(&homes_by_uid_hash_ops);
229 if (!m->homes_by_uid)
230 return -ENOMEM;
231
232 m->homes_by_name = hashmap_new(&homes_by_name_hash_ops);
233 if (!m->homes_by_name)
234 return -ENOMEM;
235
236 m->homes_by_worker_pid = hashmap_new(&homes_by_worker_pid_hash_ops);
237 if (!m->homes_by_worker_pid)
238 return -ENOMEM;
239
240 m->homes_by_sysfs = hashmap_new(&homes_by_sysfs_hash_ops);
241 if (!m->homes_by_sysfs)
242 return -ENOMEM;
243
244 *ret = TAKE_PTR(m);
245 return 0;
246}
247
248Manager* manager_free(Manager *m) {
9796a9fb
LP		 249 Home *h;
250
70a5db58
LP		 251 assert(m);
252
9796a9fb
LP		 253 HASHMAP_FOREACH(h, m->homes_by_worker_pid)
254 (void) home_wait_for_worker(h);
255
76fc1577
YW		 256 m->bus = sd_bus_flush_close_unref(m->bus);
257 m->polkit_registry = bus_verify_polkit_async_registry_free(m->polkit_registry);
70a5db58 258
70a5db58
LP		 259 m->device_monitor = sd_device_monitor_unref(m->device_monitor);
260
f76e5644
ZJS		 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);
70a5db58
LP		 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);
d357b80d 266 m->rebalance_event_source = sd_event_source_unref(m->rebalance_event_source);
70a5db58 267
76fc1577 268 m->event = sd_event_unref(m->event);
f76e5644 269
76fc1577
YW		 270 m->homes_by_uid = hashmap_free(m->homes_by_uid);
271 m->homes_by_name = hashmap_free(m->homes_by_name);
272 m->homes_by_worker_pid = hashmap_free(m->homes_by_worker_pid);
273 m->homes_by_sysfs = hashmap_free(m->homes_by_sysfs);
f76e5644 274
70a5db58
LP		 275 if (m->private_key)
276 EVP_PKEY_free(m->private_key);
277
278 hashmap_free(m->public_keys);
279
280 varlink_server_unref(m->varlink_server);
cc9886bc 281 free(m->userdb_service);
70a5db58 282
c76dd733
LP		 283 free(m->default_file_system_type);
284
70a5db58
LP		 285 return mfree(m);
286}
287
288int manager_verify_user_record(Manager *m, UserRecord *hr) {
289 EVP_PKEY *pkey;
70a5db58
LP		 290 int r;
291
292 assert(m);
293 assert(hr);
294
295 if (!m->private_key && hashmap_isempty(m->public_keys)) {
296 r = user_record_has_signature(hr);
297 if (r < 0)
298 return r;
299
300 return r ? -ENOKEY : USER_RECORD_UNSIGNED;
301 }
302
303 /* Is it our own? */
304 if (m->private_key) {
305 r = user_record_verify(hr, m->private_key);
306 switch (r) {
307
308 case USER_RECORD_FOREIGN:
309 /* This record is not signed by this key, but let's see below */
310 break;
311
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 */
315 default:
316 return r;
317 }
318 }
319
90e74a66 320 HASHMAP_FOREACH(pkey, m->public_keys) {
70a5db58
LP		 321 r = user_record_verify(hr, pkey);
322 switch (r) {
323
324 case USER_RECORD_FOREIGN:
325 /* This record is not signed by this key, but let's see our other keys */
326 break;
327
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;
331
332 case USER_RECORD_UNSIGNED: /* It's not signed at all */
333 default:
334 return r;
335 }
336 }
337
338 return -ENOKEY;
339}
340
341static int manager_add_home_by_record(
342 Manager *m,
343 const char *name,
344 int dir_fd,
345 const char *fname) {
346
347 _cleanup_(json_variant_unrefp) JsonVariant *v = NULL;
852640f8 348 _cleanup_(user_record_unrefp) UserRecord *hr = NULL;
70a5db58
LP		 349 unsigned line, column;
350 int r, is_signed;
20f4a308 351 struct stat st;
70a5db58
LP		 352 Home *h;
353
354 assert(m);
355 assert(name);
356 assert(fname);
357
20f4a308
LP		 358 if (fstatat(dir_fd, fname, &st, 0) < 0)
359 return log_error_errno(errno, "Failed to stat identity record %s: %m", fname);
360
361 if (!S_ISREG(st.st_mode)) {
362 log_debug("Identity record file %s is not a regular file, ignoring.", fname);
363 return 0;
364 }
365
366 if (st.st_size == 0)
367 goto unlink_this_file;
368
70a5db58
LP		 369 r = json_parse_file_at(NULL, dir_fd, fname, JSON_PARSE_SENSITIVE, &v, &line, &column);
370 if (r < 0)
371 return log_error_errno(r, "Failed to parse identity record at %s:%u%u: %m", fname, line, column);
372
20f4a308
LP		 373 if (json_variant_is_blank_object(v))
374 goto unlink_this_file;
375
70a5db58
LP		 376 hr = user_record_new();
377 if (!hr)
378 return log_oom();
379
bfc0cc1a 380 r = user_record_load(hr, v, USER_RECORD_LOAD_REFUSE_SECRET|USER_RECORD_LOG|USER_RECORD_PERMISSIVE);
70a5db58
LP		 381 if (r < 0)
382 return r;
383
384 if (!streq_ptr(hr->user_name, name))
23d24b76
ZJS		 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);
70a5db58
LP		 388
389 is_signed = manager_verify_user_record(m, hr);
390 switch (is_signed) {
391
392 case -ENOKEY:
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);
398 break;
399 case USER_RECORD_SIGNED_EXCLUSIVE:
400 log_info("User record %s is signed only by us, accepting.", fname);
401 break;
402 case USER_RECORD_FOREIGN:
403 log_info("User record %s is signed by registered key from others, accepting.", fname);
404 break;
405 default:
406 assert(is_signed < 0);
407 return log_error_errno(is_signed, "Failed to verify signature of user record in %s: %m", fname);
408 }
409
410 h = hashmap_get(m->homes_by_name, name);
411 if (h) {
412 r = home_set_record(h, hr);
413 if (r < 0)
414 return log_error_errno(r, "Failed to update home record for %s: %m", name);
415
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
162392b7 418 * in order to detect to distinguish HOME_INACTIVE and HOME_ABSENT. */
70a5db58
LP		 419 if (h->state == HOME_UNFIXATED)
420 h->state = _HOME_STATE_INVALID;
421 } else {
422 r = home_new(m, hr, NULL, &h);
423 if (r < 0)
424 return log_error_errno(r, "Failed to allocate new home object: %m");
425
426 log_info("Added registered home for user %s.", hr->user_name);
427 }
428
429 /* Only entries we exclusively signed are writable to us, hence remember the result */
430 h->signed_locally = is_signed == USER_RECORD_SIGNED_EXCLUSIVE;
431
432 return 1;
20f4a308
LP		 433
434unlink_this_file:
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. */
440
441 if (unlinkat(dir_fd, fname, 0) < 0)
442 return log_error_errno(errno, "Failed to remove empty user record file %s: %m", fname);
443
005daeed 444 log_notice("Discovered empty user record file %s/%s, removed automatically.", home_record_dir(), fname);
20f4a308 445 return 0;
70a5db58
LP		 446}
447
448static int manager_enumerate_records(Manager *m) {
449 _cleanup_closedir_ DIR *d = NULL;
70a5db58
LP		 450
451 assert(m);
452
005daeed 453 d = opendir(home_record_dir());
70a5db58
LP		 454 if (!d)
455 return log_full_errno(errno == ENOENT ? LOG_DEBUG : LOG_ERR, errno,
005daeed 456 "Failed to open %s: %m", home_record_dir());
70a5db58
LP		 457
458 FOREACH_DIRENT(de, d, return log_error_errno(errno, "Failed to read record directory: %m")) {
459 _cleanup_free_ char *n = NULL;
460 const char *e;
461
462 if (!dirent_is_file(de))
463 continue;
464
465 e = endswith(de->d_name, ".identity");
466 if (!e)
467 continue;
468
469 n = strndup(de->d_name, e - de->d_name);
470 if (!n)
471 return log_oom();
472
473 if (!suitable_user_name(n))
474 continue;
475
476 (void) manager_add_home_by_record(m, n, dirfd(d), de->d_name);
477 }
478
479 return 0;
480}
481
482static int search_quota(uid_t uid, const char *exclude_quota_path) {
483 struct stat exclude_st = {};
484 dev_t previous_devno = 0;
70a5db58
LP		 485 int r;
486
487 /* Checks whether the specified UID owns any files on the files system, but ignore any file system
488 * backing the specified file. The file is used when operating on home directories, where it's OK if
489 * the UID of them already owns files. */
490
491 if (exclude_quota_path && stat(exclude_quota_path, &exclude_st) < 0) {
492 if (errno != ENOENT)
493 return log_warning_errno(errno, "Failed to stat %s, ignoring: %m", exclude_quota_path);
494 }
495
496 /* Check a few usual suspects where regular users might own files. Note that this is by no means
497 * comprehensive, but should cover most cases. Note that in an ideal world every user would be
498 * registered in NSS and avoid our own UID range, but for all other cases, it's a good idea to be
499 * paranoid and check quota if we can. */
2700fecd 500 FOREACH_STRING(where, get_home_root(), "/tmp/", "/var/", "/var/mail/", "/var/tmp/", "/var/spool/") {
70a5db58
LP		 501 struct dqblk req;
502 struct stat st;
503
504 if (stat(where, &st) < 0) {
505 log_full_errno(errno == ENOENT ? LOG_DEBUG : LOG_ERR, errno,
506 "Failed to stat %s, ignoring: %m", where);
507 continue;
508 }
509
510 if (major(st.st_dev) == 0) {
511 log_debug("Directory %s is not on a real block device, not checking quota for UID use.", where);
512 continue;
513 }
514
515 if (st.st_dev == exclude_st.st_dev) { /* If an exclude path is specified, then ignore quota
516 * reported on the same block device as that path. */
517 log_debug("Directory %s is where the home directory is located, not checking quota for UID use.", where);
518 continue;
519 }
520
521 if (st.st_dev == previous_devno) { /* Does this directory have the same devno as the previous
522 * one we tested? If so, there's no point in testing this
523 * again. */
524 log_debug("Directory %s is on same device as previous tested directory, not checking quota for UID use a second time.", where);
525 continue;
526 }
527
528 previous_devno = st.st_dev;
529
7176f06c 530 r = quotactl_devnum(QCMD_FIXED(Q_GETQUOTA, USRQUOTA), st.st_dev, uid, &req);
70a5db58
LP		 531 if (r < 0) {
532 if (ERRNO_IS_NOT_SUPPORTED(r))
533 log_debug_errno(r, "No UID quota support on %s, ignoring.", where);
5e5e11b8
LP		 534 else if (ERRNO_IS_PRIVILEGE(r))
535 log_debug_errno(r, "UID quota support for %s prohibited, ignoring.", where);
70a5db58 536 else
1a53adb3 537 log_warning_errno(r, "Failed to query quota on %s, ignoring: %m", where);
70a5db58
LP		 538
539 continue;
540 }
541
542 if ((FLAGS_SET(req.dqb_valid, QIF_SPACE) && req.dqb_curspace > 0) ||
543 (FLAGS_SET(req.dqb_valid, QIF_INODES) && req.dqb_curinodes > 0)) {
544 log_debug_errno(errno, "Quota reports UID " UID_FMT " occupies disk space on %s.", uid, where);
545 return 1;
546 }
547 }
548
549 return 0;
550}
551
552static int manager_acquire_uid(
553 Manager *m,
554 uid_t start_uid,
555 const char *user_name,
556 const char *exclude_quota_path,
557 uid_t *ret) {
558
559 static const uint8_t hash_key[] = {
560 0xa3, 0xb8, 0x82, 0x69, 0x9a, 0x71, 0xf7, 0xa9,
561 0xe0, 0x7c, 0xf6, 0xf1, 0x21, 0x69, 0xd2, 0x1e
562 };
563
564 enum {
565 PHASE_SUGGESTED,
566 PHASE_HASHED,
567 PHASE_RANDOM
568 } phase = PHASE_SUGGESTED;
569
570 unsigned n_tries = 100;
571 int r;
572
573 assert(m);
574 assert(ret);
575
576 for (;;) {
577 struct passwd *pw;
578 struct group *gr;
579 uid_t candidate;
580 Home *other;
581
582 if (--n_tries <= 0)
583 return -EBUSY;
584
585 switch (phase) {
586
587 case PHASE_SUGGESTED:
588 phase = PHASE_HASHED;
589
590 if (!uid_is_home(start_uid))
591 continue;
592
593 candidate = start_uid;
594 break;
595
596 case PHASE_HASHED:
597 phase = PHASE_RANDOM;
598
599 if (!user_name)
600 continue;
601
602 candidate = UID_CLAMP_INTO_HOME_RANGE(siphash24(user_name, strlen(user_name), hash_key));
603 break;
604
605 case PHASE_RANDOM:
606 random_bytes(&candidate, sizeof(candidate));
607 candidate = UID_CLAMP_INTO_HOME_RANGE(candidate);
608 break;
609
610 default:
04499a70 611 assert_not_reached();
70a5db58
LP		 612 }
613
614 other = hashmap_get(m->homes_by_uid, UID_TO_PTR(candidate));
615 if (other) {
23d24b76
ZJS		 616 log_debug("Candidate UID " UID_FMT " already used by another home directory (%s), let's try another.",
617 candidate, other->user_name);
70a5db58
LP		 618 continue;
619 }
620
621 pw = getpwuid(candidate);
622 if (pw) {
23d24b76
ZJS		 623 log_debug("Candidate UID " UID_FMT " already registered by another user in NSS (%s), let's try another.",
624 candidate, pw->pw_name);
70a5db58
LP		 625 continue;
626 }
627
628 gr = getgrgid((gid_t) candidate);
629 if (gr) {
23d24b76
ZJS		 630 log_debug("Candidate UID " UID_FMT " already registered by another group in NSS (%s), let's try another.",
631 candidate, gr->gr_name);
70a5db58
LP		 632 continue;
633 }
634
635 r = search_ipc(candidate, (gid_t) candidate);
636 if (r < 0)
637 continue;
638 if (r > 0) {
23d24b76
ZJS		 639 log_debug_errno(r, "Candidate UID " UID_FMT " already owns IPC objects, let's try another: %m",
640 candidate);
70a5db58
LP		 641 continue;
642 }
643
644 r = search_quota(candidate, exclude_quota_path);
645 if (r != 0)
646 continue;
647
648 *ret = candidate;
649 return 0;
650 }
651}
652
653static int manager_add_home_by_image(
654 Manager *m,
655 const char *user_name,
656 const char *realm,
657 const char *image_path,
658 const char *sysfs,
659 UserStorage storage,
660 uid_t start_uid) {
661
662 _cleanup_(user_record_unrefp) UserRecord *hr = NULL;
663 uid_t uid;
664 Home *h;
665 int r;
666
667 assert(m);
668
669 assert(m);
670 assert(user_name);
671 assert(image_path);
672 assert(storage >= 0);
673 assert(storage < _USER_STORAGE_MAX);
674
675 h = hashmap_get(m->homes_by_name, user_name);
676 if (h) {
677 bool same;
678
679 if (h->state != HOME_UNFIXATED) {
680 log_debug("Found an image for user %s which already has a record, skipping.", user_name);
681 return 0; /* ignore images that synthesize a user we already have a record for */
682 }
683
684 same = user_record_storage(h->record) == storage;
685 if (same) {
686 if (h->sysfs && sysfs)
687 same = path_equal(h->sysfs, sysfs);
688 else if (!!h->sysfs != !!sysfs)
689 same = false;
690 else {
691 const char *p;
692
693 p = user_record_image_path(h->record);
694 same = p && path_equal(p, image_path);
695 }
696 }
697
698 if (!same) {
80ace4f2 699 log_debug("Found multiple images for user '%s', ignoring image '%s'.", user_name, image_path);
70a5db58
LP		 700 return 0;
701 }
702 } else {
703 /* Check NSS, in case there's another user or group by this name */
704 if (getpwnam(user_name) || getgrnam(user_name)) {
705 log_debug("Found an existing user or group by name '%s', ignoring image '%s'.", user_name, image_path);
706 return 0;
707 }
708 }
709
710 if (h && uid_is_valid(h->uid))
711 uid = h->uid;
712 else {
23d24b76
ZJS		 713 r = manager_acquire_uid(m, start_uid, user_name,
714 IN_SET(storage, USER_SUBVOLUME, USER_DIRECTORY, USER_FSCRYPT) ? image_path : NULL,
715 &uid);
70a5db58
LP		 716 if (r < 0)
717 return log_warning_errno(r, "Failed to acquire unused UID for %s: %m", user_name);
718 }
719
720 hr = user_record_new();
721 if (!hr)
722 return log_oom();
723
724 r = user_record_synthesize(hr, user_name, realm, image_path, storage, uid, (gid_t) uid);
725 if (r < 0)
726 return log_error_errno(r, "Failed to synthesize home record for %s (image %s): %m", user_name, image_path);
727
728 if (h) {
729 r = home_set_record(h, hr);
730 if (r < 0)
731 return log_error_errno(r, "Failed to update home record for %s: %m", user_name);
732 } else {
733 r = home_new(m, hr, sysfs, &h);
734 if (r < 0)
735 return log_error_errno(r, "Failed to allocate new home object: %m");
736
737 h->state = HOME_UNFIXATED;
738
739 log_info("Discovered new home for user %s through image %s.", user_name, image_path);
740 }
741
742 return 1;
743}
744
745int manager_augment_record_with_uid(
746 Manager *m,
747 UserRecord *hr) {
748
749 const char *exclude_quota_path = NULL;
750 uid_t start_uid = UID_INVALID, uid;
751 int r;
752
753 assert(m);
754 assert(hr);
755
756 if (uid_is_valid(hr->uid))
757 return 0;
758
759 if (IN_SET(hr->storage, USER_CLASSIC, USER_SUBVOLUME, USER_DIRECTORY, USER_FSCRYPT)) {
760 const char * ip;
761
762 ip = user_record_image_path(hr);
763 if (ip) {
764 struct stat st;
765
766 if (stat(ip, &st) < 0) {
767 if (errno != ENOENT)
768 log_warning_errno(errno, "Failed to stat(%s): %m", ip);
769 } else if (uid_is_home(st.st_uid)) {
770 start_uid = st.st_uid;
771 exclude_quota_path = ip;
772 }
773 }
774 }
775
776 r = manager_acquire_uid(m, start_uid, hr->user_name, exclude_quota_path, &uid);
777 if (r < 0)
778 return r;
779
780 log_debug("Acquired new UID " UID_FMT " for %s.", uid, hr->user_name);
781
782 r = user_record_add_binding(
783 hr,
784 _USER_STORAGE_INVALID,
785 NULL,
786 SD_ID128_NULL,
787 SD_ID128_NULL,
788 SD_ID128_NULL,
789 NULL,
790 NULL,
791 UINT64_MAX,
792 NULL,
793 NULL,
794 uid,
795 (gid_t) uid);
796 if (r < 0)
797 return r;
798
799 return 1;
800}
801
802static int manager_assess_image(
803 Manager *m,
804 int dir_fd,
805 const char *dir_path,
806 const char *dentry_name) {
807
808 char *luks_suffix, *directory_suffix;
809 _cleanup_free_ char *path = NULL;
810 struct stat st;
811 int r;
812
813 assert(m);
814 assert(dir_path);
815 assert(dentry_name);
816
817 luks_suffix = endswith(dentry_name, ".home");
818 if (luks_suffix)
819 directory_suffix = NULL;
820 else
821 directory_suffix = endswith(dentry_name, ".homedir");
822
823 /* Early filter out: by name */
824 if (!luks_suffix && !directory_suffix)
825 return 0;
826
827 path = path_join(dir_path, dentry_name);
828 if (!path)
829 return log_oom();
830
831 /* Follow symlinks here, to allow people to link in stuff to make them available locally. */
832 if (dir_fd >= 0)
833 r = fstatat(dir_fd, dentry_name, &st, 0);
834 else
835 r = stat(path, &st);
836 if (r < 0)
837 return log_full_errno(errno == ENOENT ? LOG_DEBUG : LOG_WARNING, errno,
80ace4f2 838 "Failed to stat() directory entry '%s', ignoring: %m", dentry_name);
70a5db58
LP		 839
840 if (S_ISREG(st.st_mode)) {
841 _cleanup_free_ char *n = NULL, *user_name = NULL, *realm = NULL;
842
843 if (!luks_suffix)
844 return 0;
845
846 n = strndup(dentry_name, luks_suffix - dentry_name);
847 if (!n)
848 return log_oom();
849
850 r = split_user_name_realm(n, &user_name, &realm);
851 if (r == -EINVAL) /* Not the right format: ignore */
852 return 0;
853 if (r < 0)
854 return log_error_errno(r, "Failed to split image name into user name/realm: %m");
855
856 return manager_add_home_by_image(m, user_name, realm, path, NULL, USER_LUKS, UID_INVALID);
857 }
858
859 if (S_ISDIR(st.st_mode)) {
860 _cleanup_free_ char *n = NULL, *user_name = NULL, *realm = NULL;
861 _cleanup_close_ int fd = -1;
862 UserStorage storage;
863
864 if (!directory_suffix)
865 return 0;
866
867 n = strndup(dentry_name, directory_suffix - dentry_name);
868 if (!n)
869 return log_oom();
870
871 r = split_user_name_realm(n, &user_name, &realm);
872 if (r == -EINVAL) /* Not the right format: ignore */
873 return 0;
874 if (r < 0)
875 return log_error_errno(r, "Failed to split image name into user name/realm: %m");
876
877 if (dir_fd >= 0)
878 fd = openat(dir_fd, dentry_name, O_DIRECTORY|O_RDONLY|O_CLOEXEC);
879 else
880 fd = open(path, O_DIRECTORY|O_RDONLY|O_CLOEXEC);
881 if (fd < 0)
882 return log_full_errno(errno == ENOENT ? LOG_DEBUG : LOG_WARNING, errno,
883 "Failed to open directory '%s', ignoring: %m", path);
884
885 if (fstat(fd, &st) < 0)
886 return log_warning_errno(errno, "Failed to fstat() %s, ignoring: %m", path);
887
888 assert(S_ISDIR(st.st_mode)); /* Must hold, we used O_DIRECTORY above */
889
890 r = btrfs_is_subvol_fd(fd);
891 if (r < 0)
892 return log_warning_errno(errno, "Failed to determine whether %s is a btrfs subvolume: %m", path);
893 if (r > 0)
894 storage = USER_SUBVOLUME;
895 else {
896 struct fscrypt_policy policy;
897
898 if (ioctl(fd, FS_IOC_GET_ENCRYPTION_POLICY, &policy) < 0) {
899
900 if (errno == ENODATA)
901 log_debug_errno(errno, "Determined %s is not fscrypt encrypted.", path);
902 else if (ERRNO_IS_NOT_SUPPORTED(errno))
80ace4f2 903 log_debug_errno(errno, "Determined %s is not fscrypt encrypted because kernel or file system doesn't support it.", path);
70a5db58
LP		 904 else
905 log_debug_errno(errno, "FS_IOC_GET_ENCRYPTION_POLICY failed with unexpected error code on %s, ignoring: %m", path);
906
907 storage = USER_DIRECTORY;
908 } else
909 storage = USER_FSCRYPT;
910 }
911
912 return manager_add_home_by_image(m, user_name, realm, path, NULL, storage, st.st_uid);
913 }
914
915 return 0;
916}
917
918int manager_enumerate_images(Manager *m) {
919 _cleanup_closedir_ DIR *d = NULL;
70a5db58
LP		 920
921 assert(m);
922
923 if (!m->scan_slash_home)
924 return 0;
925
2700fecd 926 d = opendir(get_home_root());
70a5db58
LP		 927 if (!d)
928 return log_full_errno(errno == ENOENT ? LOG_DEBUG : LOG_ERR, errno,
2700fecd 929 "Failed to open %s: %m", get_home_root());
70a5db58 930
2700fecd
LP		 931 FOREACH_DIRENT(de, d, return log_error_errno(errno, "Failed to read %s directory: %m", get_home_root()))
932 (void) manager_assess_image(m, dirfd(d), get_home_root(), de->d_name);
70a5db58
LP		 933
934 return 0;
935}
936
937static int manager_connect_bus(Manager *m) {
1d3b68f6 938 _cleanup_free_ char *b = NULL;
cc9886bc 939 const char *suffix, *busname;
70a5db58
LP		 940 int r;
941
942 assert(m);
943 assert(!m->bus);
944
945 r = sd_bus_default_system(&m->bus);
946 if (r < 0)
947 return log_error_errno(r, "Failed to connect to system bus: %m");
948
cfd508a9 949 r = bus_add_implementation(m->bus, &manager_object, m);
ac9f55ed
LP		 950 if (r < 0)
951 return r;
952
c42234ab
LP		 953 r = bus_log_control_api_register(m->bus);
954 if (r < 0)
955 return r;
956
cc9886bc 957 suffix = getenv("SYSTEMD_HOME_DEBUG_SUFFIX");
1d3b68f6
AZ		 958 if (suffix) {
959 b = strjoin("org.freedesktop.home1.", suffix);
960 if (!b)
961 return log_oom();
962 busname = b;
963 } else
cc9886bc
LP		 964 busname = "org.freedesktop.home1";
965
966 r = sd_bus_request_name_async(m->bus, NULL, busname, 0, NULL, NULL);
70a5db58
LP		 967 if (r < 0)
968 return log_error_errno(r, "Failed to request name: %m");
969
970 r = sd_bus_attach_event(m->bus, m->event, 0);
971 if (r < 0)
972 return log_error_errno(r, "Failed to attach bus to event loop: %m");
973
974 (void) sd_bus_set_exit_on_disconnect(m->bus, true);
975
976 return 0;
977}
978
979static int manager_bind_varlink(Manager *m) {
1d3b68f6 980 _cleanup_free_ char *p = NULL;
cc9886bc 981 const char *suffix, *socket_path;
70a5db58
LP		 982 int r;
983
984 assert(m);
985 assert(!m->varlink_server);
986
9807fdc1 987 r = varlink_server_new(&m->varlink_server, VARLINK_SERVER_ACCOUNT_UID|VARLINK_SERVER_INHERIT_USERDATA);
70a5db58
LP		 988 if (r < 0)
989 return log_error_errno(r, "Failed to allocate varlink server object: %m");
990
991 varlink_server_set_userdata(m->varlink_server, m);
992
993 r = varlink_server_bind_method_many(
994 m->varlink_server,
995 "io.systemd.UserDatabase.GetUserRecord", vl_method_get_user_record,
996 "io.systemd.UserDatabase.GetGroupRecord", vl_method_get_group_record,
997 "io.systemd.UserDatabase.GetMemberships", vl_method_get_memberships);
998 if (r < 0)
999 return log_error_errno(r, "Failed to register varlink methods: %m");
1000
1001 (void) mkdir_p("/run/systemd/userdb", 0755);
1002
cc9886bc
LP		 1003 /* To make things easier to debug, when working from a homed managed home directory, let's optionally
1004 * use a different varlink socket name */
1005 suffix = getenv("SYSTEMD_HOME_DEBUG_SUFFIX");
1d3b68f6
AZ		 1006 if (suffix) {
1007 p = strjoin("/run/systemd/userdb/io.systemd.Home.", suffix);
1008 if (!p)
1009 return log_oom();
1010 socket_path = p;
1011 } else
cc9886bc
LP		 1012 socket_path = "/run/systemd/userdb/io.systemd.Home";
1013
1014 r = varlink_server_listen_address(m->varlink_server, socket_path, 0666);
70a5db58
LP		 1015 if (r < 0)
1016 return log_error_errno(r, "Failed to bind to varlink socket: %m");
1017
1018 r = varlink_server_attach_event(m->varlink_server, m->event, SD_EVENT_PRIORITY_NORMAL);
1019 if (r < 0)
1020 return log_error_errno(r, "Failed to attach varlink connection to event loop: %m");
1021
cc9886bc
LP		 1022 assert(!m->userdb_service);
1023 m->userdb_service = strdup(basename(socket_path));
1024 if (!m->userdb_service)
1025 return log_oom();
1026
1027 /* Avoid recursion */
1028 if (setenv("SYSTEMD_BYPASS_USERDB", m->userdb_service, 1) < 0)
1029 return log_error_errno(SYNTHETIC_ERRNO(EINVAL), "Failed to set $SYSTEMD_BYPASS_USERDB: %m");
1030
70a5db58
LP		 1031 return 0;
1032}
1033
2aaf565a
LP		 1034static ssize_t read_datagram(
1035 int fd,
1036 struct ucred *ret_sender,
1037 void **ret,
1038 int *ret_passed_fd) {
1039
1040 CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(struct ucred)) + CMSG_SPACE(sizeof(int))) control;
70a5db58 1041 _cleanup_free_ void *buffer = NULL;
2aaf565a
LP		 1042 _cleanup_close_ int passed_fd = -1;
1043 struct ucred *sender = NULL;
1044 struct cmsghdr *cmsg;
1045 struct msghdr mh;
1046 struct iovec iov;
70a5db58
LP		 1047 ssize_t n, m;
1048
1049 assert(fd >= 0);
1050 assert(ret_sender);
1051 assert(ret);
2aaf565a 1052 assert(ret_passed_fd);
70a5db58
LP		 1053
1054 n = next_datagram_size_fd(fd);
1055 if (n < 0)
1056 return n;
1057
1058 buffer = malloc(n + 2);
1059 if (!buffer)
1060 return -ENOMEM;
1061
2aaf565a
LP		 1062 /* Pass one extra byte, as a size check */
1063 iov = IOVEC_MAKE(buffer, n + 1);
70a5db58 1064
2aaf565a
LP		 1065 mh = (struct msghdr) {
1066 .msg_iov = &iov,
1067 .msg_iovlen = 1,
1068 .msg_control = &control,
1069 .msg_controllen = sizeof(control),
1070 };
70a5db58 1071
2aaf565a
LP		 1072 m = recvmsg_safe(fd, &mh, MSG_DONTWAIT|MSG_CMSG_CLOEXEC);
1073 if (m < 0)
1074 return m;
70a5db58 1075
2aaf565a
LP		 1076 /* Ensure the size matches what we determined before */
1077 if (m != n) {
70a5db58 1078 cmsg_close_all(&mh);
2aaf565a
LP		 1079 return -EMSGSIZE;
1080 }
70a5db58 1081
2aaf565a
LP		 1082 CMSG_FOREACH(cmsg, &mh) {
1083 if (cmsg->cmsg_level == SOL_SOCKET &&
1084 cmsg->cmsg_type == SCM_CREDENTIALS &&
1085 cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred))) {
1086 assert(!sender);
1087 sender = (struct ucred*) CMSG_DATA(cmsg);
1088 }
70a5db58 1089
2aaf565a
LP		 1090 if (cmsg->cmsg_level == SOL_SOCKET &&
1091 cmsg->cmsg_type == SCM_RIGHTS) {
70a5db58 1092
2aaf565a
LP		 1093 if (cmsg->cmsg_len != CMSG_LEN(sizeof(int))) {
1094 cmsg_close_all(&mh);
1095 return -EMSGSIZE;
70a5db58
LP		 1096 }
1097
2aaf565a
LP		 1098 assert(passed_fd < 0);
1099 passed_fd = *(int*) CMSG_DATA(cmsg);
1100 }
70a5db58
LP		 1101 }
1102
2aaf565a
LP		 1103 if (sender)
1104 *ret_sender = *sender;
1105 else
1106 *ret_sender = (struct ucred) UCRED_INVALID;
1107
1108 *ret_passed_fd = TAKE_FD(passed_fd);
1109
70a5db58
LP		 1110 /* For safety reasons: let's always NUL terminate. */
1111 ((char*) buffer)[n] = 0;
1112 *ret = TAKE_PTR(buffer);
1113
1114 return 0;
1115}
1116
1117static int on_notify_socket(sd_event_source *s, int fd, uint32_t revents, void *userdata) {
1118 _cleanup_strv_free_ char **l = NULL;
1119 _cleanup_free_ void *datagram = NULL;
2aaf565a
LP		 1120 _cleanup_close_ int passed_fd = -1;
1121 struct ucred sender = UCRED_INVALID;
70a5db58
LP		 1122 Manager *m = userdata;
1123 ssize_t n;
1124 Home *h;
1125
1126 assert(s);
1127 assert(m);
1128
2aaf565a 1129 n = read_datagram(fd, &sender, &datagram, &passed_fd);
8add30a0
YW		 1130 if (n < 0) {
1131 if (ERRNO_IS_TRANSIENT(n))
1132 return 0;
70a5db58 1133 return log_error_errno(n, "Failed to read notify datagram: %m");
8add30a0 1134 }
70a5db58
LP		 1135
1136 if (sender.pid <= 0) {
1137 log_warning("Received notify datagram without valid sender PID, ignoring.");
1138 return 0;
1139 }
1140
1141 h = hashmap_get(m->homes_by_worker_pid, PID_TO_PTR(sender.pid));
1142 if (!h) {
162392b7 1143 log_warning("Received notify datagram of unknown process, ignoring.");
70a5db58
LP		 1144 return 0;
1145 }
1146
1147 l = strv_split(datagram, "\n");
1148 if (!l)
1149 return log_oom();
1150
2aaf565a 1151 home_process_notify(h, l, TAKE_FD(passed_fd));
70a5db58
LP		 1152 return 0;
1153}
1154
1155static int manager_listen_notify(Manager *m) {
1156 _cleanup_close_ int fd = -1;
425d925f
ZJS		 1157 union sockaddr_union sa = {
1158 .un.sun_family = AF_UNIX,
1159 .un.sun_path = "/run/systemd/home/notify",
1160 };
cc9886bc 1161 const char *suffix;
70a5db58
LP		 1162 int r;
1163
1164 assert(m);
1165 assert(!m->notify_socket_event_source);
1166
cc9886bc
LP		 1167 suffix = getenv("SYSTEMD_HOME_DEBUG_SUFFIX");
1168 if (suffix) {
1d3b68f6 1169 _cleanup_free_ char *unix_path = NULL;
cc9886bc 1170
1d3b68f6
AZ		 1171 unix_path = strjoin("/run/systemd/home/notify.", suffix);
1172 if (!unix_path)
1173 return log_oom();
cc9886bc
LP		 1174 r = sockaddr_un_set_path(&sa.un, unix_path);
1175 if (r < 0)
1176 return log_error_errno(r, "Socket path %s does not fit in sockaddr_un: %m", unix_path);
1177 }
1178
70a5db58
LP		 1179 fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
1180 if (fd < 0)
1181 return log_error_errno(errno, "Failed to create listening socket: %m");
1182
70a5db58
LP		 1183 (void) mkdir_parents(sa.un.sun_path, 0755);
1184 (void) sockaddr_un_unlink(&sa.un);
1185
1186 if (bind(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un)) < 0)
1187 return log_error_errno(errno, "Failed to bind to socket: %m");
1188
1189 r = setsockopt_int(fd, SOL_SOCKET, SO_PASSCRED, true);
1190 if (r < 0)
1191 return r;
1192
1193 r = sd_event_add_io(m->event, &m->notify_socket_event_source, fd, EPOLLIN, on_notify_socket, m);
1194 if (r < 0)
1195 return log_error_errno(r, "Failed to allocate event source for notify socket: %m");
1196
1197 (void) sd_event_source_set_description(m->notify_socket_event_source, "notify-socket");
1198
1199 /* Make sure we process sd_notify() before SIGCHLD for any worker, so that we always know the error
1200 * number of a client before it exits. */
1201 r = sd_event_source_set_priority(m->notify_socket_event_source, SD_EVENT_PRIORITY_NORMAL - 5);
1202 if (r < 0)
1203 return log_error_errno(r, "Failed to alter priority of NOTIFY_SOCKET event source: %m");
1204
1205 r = sd_event_source_set_io_fd_own(m->notify_socket_event_source, true);
1206 if (r < 0)
1207 return log_error_errno(r, "Failed to pass ownership of notify socket: %m");
1208
1209 return TAKE_FD(fd);
1210}
1211
1212static int manager_add_device(Manager *m, sd_device *d) {
1213 _cleanup_free_ char *user_name = NULL, *realm = NULL, *node = NULL;
1214 const char *tabletype, *parttype, *partname, *partuuid, *sysfs;
1215 sd_id128_t id;
1216 int r;
1217
1218 assert(m);
1219 assert(d);
1220
1221 r = sd_device_get_syspath(d, &sysfs);
1222 if (r < 0)
1223 return log_error_errno(r, "Failed to acquire sysfs path of device: %m");
1224
1225 r = sd_device_get_property_value(d, "ID_PART_TABLE_TYPE", &tabletype);
1226 if (r == -ENOENT)
1227 return 0;
1228 if (r < 0)
1229 return log_error_errno(r, "Failed to acquire ID_PART_TABLE_TYPE device property, ignoring: %m");
1230
1231 if (!streq(tabletype, "gpt")) {
1232 log_debug("Found partition (%s) on non-GPT table, ignoring.", sysfs);
1233 return 0;
1234 }
1235
1236 r = sd_device_get_property_value(d, "ID_PART_ENTRY_TYPE", &parttype);
1237 if (r == -ENOENT)
1238 return 0;
1239 if (r < 0)
1240 return log_error_errno(r, "Failed to acquire ID_PART_ENTRY_TYPE device property, ignoring: %m");
e46433bb 1241 if (id128_equal_string(parttype, GPT_USER_HOME) <= 0) {
70a5db58
LP		 1242 log_debug("Found partition (%s) we don't care about, ignoring.", sysfs);
1243 return 0;
1244 }
1245
1246 r = sd_device_get_property_value(d, "ID_PART_ENTRY_NAME", &partname);
1247 if (r < 0)
1248 return log_warning_errno(r, "Failed to acquire ID_PART_ENTRY_NAME device property, ignoring: %m");
1249
1250 r = split_user_name_realm(partname, &user_name, &realm);
1251 if (r == -EINVAL)
1252 return log_warning_errno(r, "Found partition with correct partition type but a non-parsable partition name '%s', ignoring.", partname);
1253 if (r < 0)
1254 return log_error_errno(r, "Failed to validate partition name '%s': %m", partname);
1255
1256 r = sd_device_get_property_value(d, "ID_FS_UUID", &partuuid);
1257 if (r < 0)
1258 return log_warning_errno(r, "Failed to acquire ID_FS_UUID device property, ignoring: %m");
1259
1260 r = sd_id128_from_string(partuuid, &id);
1261 if (r < 0)
1262 return log_warning_errno(r, "Failed to parse ID_FS_UUID field '%s', ignoring: %m", partuuid);
1263
1264 if (asprintf(&node, "/dev/disk/by-uuid/" SD_ID128_UUID_FORMAT_STR, SD_ID128_FORMAT_VAL(id)) < 0)
1265 return log_oom();
1266
1267 return manager_add_home_by_image(m, user_name, realm, node, sysfs, USER_LUKS, UID_INVALID);
1268}
1269
1270static int manager_on_device(sd_device_monitor *monitor, sd_device *d, void *userdata) {
1271 Manager *m = userdata;
1272 int r;
1273
1274 assert(m);
1275 assert(d);
1276
a1130022 1277 if (device_for_action(d, SD_DEVICE_REMOVE)) {
70a5db58
LP		 1278 const char *sysfs;
1279 Home *h;
1280
1281 r = sd_device_get_syspath(d, &sysfs);
1282 if (r < 0) {
1283 log_warning_errno(r, "Failed to acquire sysfs path from device: %m");
1284 return 0;
1285 }
1286
1287 log_info("block device %s has been removed.", sysfs);
1288
1289 /* Let's see if we previously synthesized a home record from this device, if so, let's just
1290 * revalidate that. Otherwise let's revalidate them all, but asynchronously. */
1291 h = hashmap_get(m->homes_by_sysfs, sysfs);
1292 if (h)
1293 manager_revalidate_image(m, h);
1294 else
1295 manager_enqueue_gc(m, NULL);
1296 } else
1297 (void) manager_add_device(m, d);
1298
1299 (void) bus_manager_emit_auto_login_changed(m);
1300 return 0;
1301}
1302
1303static int manager_watch_devices(Manager *m) {
1304 int r;
1305
1306 assert(m);
1307 assert(!m->device_monitor);
1308
1309 r = sd_device_monitor_new(&m->device_monitor);
1310 if (r < 0)
1311 return log_error_errno(r, "Failed to allocate device monitor: %m");
1312
1313 r = sd_device_monitor_filter_add_match_subsystem_devtype(m->device_monitor, "block", NULL);
1314 if (r < 0)
1315 return log_error_errno(r, "Failed to configure device monitor match: %m");
1316
1317 r = sd_device_monitor_attach_event(m->device_monitor, m->event);
1318 if (r < 0)
1319 return log_error_errno(r, "Failed to attach device monitor to event loop: %m");
1320
1321 r = sd_device_monitor_start(m->device_monitor, manager_on_device, m);
1322 if (r < 0)
1323 return log_error_errno(r, "Failed to start device monitor: %m");
1324
1325 return 0;
1326}
1327
1328static int manager_enumerate_devices(Manager *m) {
1329 _cleanup_(sd_device_enumerator_unrefp) sd_device_enumerator *e = NULL;
1330 sd_device *d;
1331 int r;
1332
1333 assert(m);
1334
1335 r = sd_device_enumerator_new(&e);
1336 if (r < 0)
1337 return r;
1338
1339 r = sd_device_enumerator_add_match_subsystem(e, "block", true);
1340 if (r < 0)
1341 return r;
1342
1343 FOREACH_DEVICE(e, d)
1344 (void) manager_add_device(m, d);
1345
1346 return 0;
1347}
1348
1349static int manager_load_key_pair(Manager *m) {
1350 _cleanup_(fclosep) FILE *f = NULL;
1351 struct stat st;
1352 int r;
1353
1354 assert(m);
1355
1356 if (m->private_key) {
1357 EVP_PKEY_free(m->private_key);
1358 m->private_key = NULL;
1359 }
1360
2708160c 1361 r = search_and_fopen_nulstr("local.private", "re", NULL, KEY_PATHS_NULSTR, &f, NULL);
70a5db58
LP		 1362 if (r == -ENOENT)
1363 return 0;
1364 if (r < 0)
1365 return log_error_errno(r, "Failed to read private key file: %m");
1366
1367 if (fstat(fileno(f), &st) < 0)
1368 return log_error_errno(errno, "Failed to stat private key file: %m");
1369
1370 r = stat_verify_regular(&st);
1371 if (r < 0)
1372 return log_error_errno(r, "Private key file is not regular: %m");
1373
1374 if (st.st_uid != 0 || (st.st_mode & 0077) != 0)
1375 return log_error_errno(SYNTHETIC_ERRNO(EPERM), "Private key file is readable by more than the root user");
1376
1377 m->private_key = PEM_read_PrivateKey(f, NULL, NULL, NULL);
1378 if (!m->private_key)
1379 return log_error_errno(SYNTHETIC_ERRNO(EIO), "Failed to load private key pair");
1380
1381 log_info("Successfully loaded private key pair.");
1382
1383 return 1;
1384}
1385
fd421c4a 1386DEFINE_TRIVIAL_CLEANUP_FUNC_FULL(EVP_PKEY_CTX*, EVP_PKEY_CTX_free, NULL);
70a5db58
LP		 1387
1388static int manager_generate_key_pair(Manager *m) {
1389 _cleanup_(EVP_PKEY_CTX_freep) EVP_PKEY_CTX *ctx = NULL;
1390 _cleanup_(unlink_and_freep) char *temp_public = NULL, *temp_private = NULL;
1391 _cleanup_fclose_ FILE *fpublic = NULL, *fprivate = NULL;
1392 int r;
1393
1394 if (m->private_key) {
1395 EVP_PKEY_free(m->private_key);
1396 m->private_key = NULL;
1397 }
1398
1399 ctx = EVP_PKEY_CTX_new_id(EVP_PKEY_ED25519, NULL);
1400 if (!ctx)
1401 return log_error_errno(SYNTHETIC_ERRNO(EIO), "Failed to allocate Ed25519 key generation context.");
1402
1403 if (EVP_PKEY_keygen_init(ctx) <= 0)
1404 return log_error_errno(SYNTHETIC_ERRNO(EIO), "Failed to initialize Ed25519 key generation context.");
1405
1406 log_info("Generating key pair for signing local user identity records.");
1407
1408 if (EVP_PKEY_keygen(ctx, &m->private_key) <= 0)
1409 return log_error_errno(SYNTHETIC_ERRNO(EIO), "Failed to generate Ed25519 key pair");
1410
1411 log_info("Successfully created Ed25519 key pair.");
1412
1413 (void) mkdir_p("/var/lib/systemd/home", 0755);
1414
1415 /* Write out public key (note that we only do that as a help to the user, we don't make use of this ever */
1416 r = fopen_temporary("/var/lib/systemd/home/local.public", &fpublic, &temp_public);
1417 if (r < 0)
80ace4f2 1418 return log_error_errno(errno, "Failed to open key file for writing: %m");
70a5db58
LP		 1419
1420 if (PEM_write_PUBKEY(fpublic, m->private_key) <= 0)
1421 return log_error_errno(SYNTHETIC_ERRNO(EIO), "Failed to write public key.");
1422
fa3709c5 1423 r = fflush_sync_and_check(fpublic);
70a5db58
LP		 1424 if (r < 0)
1425 return log_error_errno(r, "Failed to write private key: %m");
1426
1427 fpublic = safe_fclose(fpublic);
1428
1429 /* Write out the private key (this actually writes out both private and public, OpenSSL is confusing) */
1430 r = fopen_temporary("/var/lib/systemd/home/local.private", &fprivate, &temp_private);
1431 if (r < 0)
80ace4f2 1432 return log_error_errno(errno, "Failed to open key file for writing: %m");
70a5db58
LP		 1433
1434 if (PEM_write_PrivateKey(fprivate, m->private_key, NULL, NULL, 0, NULL, 0) <= 0)
1435 return log_error_errno(SYNTHETIC_ERRNO(EIO), "Failed to write private key pair.");
1436
fa3709c5 1437 r = fflush_sync_and_check(fprivate);
70a5db58
LP		 1438 if (r < 0)
1439 return log_error_errno(r, "Failed to write private key: %m");
1440
1441 fprivate = safe_fclose(fprivate);
1442
1443 /* Both are written now, move them into place */
1444
1445 if (rename(temp_public, "/var/lib/systemd/home/local.public") < 0)
1446 return log_error_errno(errno, "Failed to move public key file into place: %m");
1447 temp_public = mfree(temp_public);
1448
1449 if (rename(temp_private, "/var/lib/systemd/home/local.private") < 0) {
1450 (void) unlink_noerrno("/var/lib/systemd/home/local.public"); /* try to remove the file we already created */
e8dd54ab 1451 return log_error_errno(errno, "Failed to move private key file into place: %m");
70a5db58
LP		 1452 }
1453 temp_private = mfree(temp_private);
1454
fa3709c5
LP		 1455 r = fsync_path_at(AT_FDCWD, "/var/lib/systemd/home/");
1456 if (r < 0)
1457 log_warning_errno(r, "Failed to sync /var/lib/systemd/home/, ignoring: %m");
1458
70a5db58
LP		 1459 return 1;
1460}
1461
1462int manager_acquire_key_pair(Manager *m) {
1463 int r;
1464
1465 assert(m);
1466
1467 /* Already there? */
1468 if (m->private_key)
1469 return 1;
1470
1471 /* First try to load key off disk */
1472 r = manager_load_key_pair(m);
1473 if (r != 0)
1474 return r;
1475
1476 /* Didn't work, generate a new one */
1477 return manager_generate_key_pair(m);
1478}
1479
1480int manager_sign_user_record(Manager *m, UserRecord *u, UserRecord **ret, sd_bus_error *error) {
1481 int r;
1482
1483 assert(m);
1484 assert(u);
1485 assert(ret);
1486
1487 r = manager_acquire_key_pair(m);
1488 if (r < 0)
1489 return r;
1490 if (r == 0)
1b09b81c 1491 return sd_bus_error_set(error, BUS_ERROR_NO_PRIVATE_KEY, "Can't sign without local key.");
70a5db58
LP		 1492
1493 return user_record_sign(u, m->private_key, ret);
1494}
1495
1496DEFINE_PRIVATE_HASH_OPS_FULL(public_key_hash_ops, char, string_hash_func, string_compare_func, free, EVP_PKEY, EVP_PKEY_free);
fd421c4a 1497DEFINE_TRIVIAL_CLEANUP_FUNC_FULL(EVP_PKEY*, EVP_PKEY_free, NULL);
70a5db58
LP		 1498
1499static int manager_load_public_key_one(Manager *m, const char *path) {
1500 _cleanup_(EVP_PKEY_freep) EVP_PKEY *pkey = NULL;
1501 _cleanup_fclose_ FILE *f = NULL;
1502 _cleanup_free_ char *fn = NULL;
1503 struct stat st;
1504 int r;
1505
1506 assert(m);
1507
1508 if (streq(basename(path), "local.public")) /* we already loaded the private key, which includes the public one */
1509 return 0;
1510
1511 f = fopen(path, "re");
1512 if (!f) {
1513 if (errno == ENOENT)
1514 return 0;
1515
1516 return log_error_errno(errno, "Failed to open public key %s: %m", path);
1517 }
1518
1519 if (fstat(fileno(f), &st) < 0)
1520 return log_error_errno(errno, "Failed to stat public key %s: %m", path);
1521
1522 r = stat_verify_regular(&st);
1523 if (r < 0)
1524 return log_error_errno(r, "Public key file %s is not a regular file: %m", path);
1525
1526 if (st.st_uid != 0 || (st.st_mode & 0022) != 0)
1527 return log_error_errno(SYNTHETIC_ERRNO(EPERM), "Public key file %s is writable by more than the root user, refusing.", path);
1528
1529 r = hashmap_ensure_allocated(&m->public_keys, &public_key_hash_ops);
1530 if (r < 0)
1531 return log_oom();
1532
1533 pkey = PEM_read_PUBKEY(f, &pkey, NULL, NULL);
1534 if (!pkey)
1535 return log_error_errno(SYNTHETIC_ERRNO(EIO), "Failed to parse public key file %s.", path);
1536
1537 fn = strdup(basename(path));
1538 if (!fn)
1539 return log_oom();
1540
1541 r = hashmap_put(m->public_keys, fn, pkey);
1542 if (r < 0)
1543 return log_error_errno(r, "Failed to add public key to set: %m");
1544
1545 TAKE_PTR(fn);
1546 TAKE_PTR(pkey);
1547
1548 return 0;
1549}
1550
1551static int manager_load_public_keys(Manager *m) {
1552 _cleanup_strv_free_ char **files = NULL;
70a5db58
LP		 1553 int r;
1554
1555 assert(m);
1556
1557 m->public_keys = hashmap_free(m->public_keys);
1558
1559 r = conf_files_list_nulstr(
1560 &files,
1561 ".public",
1562 NULL,
1563 CONF_FILES_REGULAR|CONF_FILES_FILTER_MASKED,
1564 KEY_PATHS_NULSTR);
1565 if (r < 0)
1566 return log_error_errno(r, "Failed to assemble list of public key directories: %m");
1567
1568 STRV_FOREACH(i, files)
1569 (void) manager_load_public_key_one(m, *i);
1570
1571 return 0;
1572}
1573
1574int manager_startup(Manager *m) {
1575 int r;
1576
1577 assert(m);
1578
1579 r = manager_listen_notify(m);
1580 if (r < 0)
1581 return r;
1582
1583 r = manager_connect_bus(m);
1584 if (r < 0)
1585 return r;
1586
1587 r = manager_bind_varlink(m);
1588 if (r < 0)
1589 return r;
1590
1591 r = manager_load_key_pair(m); /* only try to load it, don't generate any */
1592 if (r < 0)
1593 return r;
1594
1595 r = manager_load_public_keys(m);
1596 if (r < 0)
1597 return r;
1598
1599 manager_watch_home(m);
1600 (void) manager_watch_devices(m);
1601
1602 (void) manager_enumerate_records(m);
1603 (void) manager_enumerate_images(m);
1604 (void) manager_enumerate_devices(m);
1605
1606 /* Let's clean up home directories whose devices got removed while we were not running */
1607 (void) manager_enqueue_gc(m, NULL);
1608
1609 return 0;
1610}
1611
1612void manager_revalidate_image(Manager *m, Home *h) {
1613 int r;
1614
1615 assert(m);
1616 assert(h);
1617
1618 /* Frees an automatically discovered image, if it's synthetic and its image disappeared. Unmounts any
1619 * image if it's mounted but it's image vanished. */
1620
1621 if (h->current_operation || !ordered_set_isempty(h->pending_operations))
1622 return;
1623
1624 if (h->state == HOME_UNFIXATED) {
1625 r = user_record_test_image_path(h->record);
1626 if (r < 0)
1627 log_warning_errno(r, "Can't determine if image of %s exists, freeing unfixated user: %m", h->user_name);
1628 else if (r == USER_TEST_ABSENT)
1629 log_info("Image for %s disappeared, freeing unfixated user.", h->user_name);
1630 else
1631 return;
1632
1633 home_free(h);
1634
1635 } else if (h->state < 0) {
1636
1637 r = user_record_test_home_directory(h->record);
1638 if (r < 0) {
1639 log_warning_errno(r, "Unable to determine state of home directory, ignoring: %m");
1640 return;
1641 }
1642
1643 if (r == USER_TEST_MOUNTED) {
1644 r = user_record_test_image_path(h->record);
1645 if (r < 0) {
1646 log_warning_errno(r, "Unable to determine state of image path, ignoring: %m");
1647 return;
1648 }
1649
1650 if (r == USER_TEST_ABSENT) {
1651 _cleanup_(operation_unrefp) Operation *o = NULL;
1652
1653 log_notice("Backing image disappeared while home directory %s was mounted, unmounting it forcibly.", h->user_name);
1654 /* Wowza, the thing is mounted, but the device is gone? Act on it. */
1655
1656 r = home_killall(h);
1657 if (r < 0)
1658 log_warning_errno(r, "Failed to kill processes of user %s, ignoring: %m", h->user_name);
1659
1660 /* We enqueue the operation here, after all the home directory might
1661 * currently already run some operation, and we can deactivate it only after
1662 * that's complete. */
1663 o = operation_new(OPERATION_DEACTIVATE_FORCE, NULL);
1664 if (!o) {
1665 log_oom();
1666 return;
1667 }
1668
1669 r = home_schedule_operation(h, o, NULL);
1670 if (r < 0)
1671 log_warning_errno(r, "Failed to enqueue forced home directory %s deactivation, ignoring: %m", h->user_name);
1672 }
1673 }
1674 }
1675}
1676
1677int manager_gc_images(Manager *m) {
1678 Home *h;
1679
1680 assert_se(m);
1681
1682 if (m->gc_focus) {
1683 /* Focus on a specific home */
1684
1685 h = TAKE_PTR(m->gc_focus);
1686 manager_revalidate_image(m, h);
1687 } else {
1688 /* Gc all */
70a5db58 1689
90e74a66 1690 HASHMAP_FOREACH(h, m->homes_by_name)
70a5db58
LP		 1691 manager_revalidate_image(m, h);
1692 }
1693
1694 return 0;
1695}
1696
1697static int on_deferred_rescan(sd_event_source *s, void *userdata) {
1698 Manager *m = userdata;
1699
1700 assert(m);
1701
cf536638 1702 m->deferred_rescan_event_source = sd_event_source_disable_unref(m->deferred_rescan_event_source);
70a5db58
LP		 1703
1704 manager_enumerate_devices(m);
1705 manager_enumerate_images(m);
1706 return 0;
1707}
1708
1709int manager_enqueue_rescan(Manager *m) {
1710 int r;
1711
1712 assert(m);
1713
1714 if (m->deferred_rescan_event_source)
1715 return 0;
1716
1717 if (!m->event)
1718 return 0;
1719
1720 if (IN_SET(sd_event_get_state(m->event), SD_EVENT_FINISHED, SD_EVENT_EXITING))
1721 return 0;
1722
1723 r = sd_event_add_defer(m->event, &m->deferred_rescan_event_source, on_deferred_rescan, m);
1724 if (r < 0)
1725 return log_error_errno(r, "Failed to allocate rescan event source: %m");
1726
1727 r = sd_event_source_set_priority(m->deferred_rescan_event_source, SD_EVENT_PRIORITY_IDLE+1);
1728 if (r < 0)
1729 log_warning_errno(r, "Failed to tweak priority of event source, ignoring: %m");
1730
1731 (void) sd_event_source_set_description(m->deferred_rescan_event_source, "deferred-rescan");
1732 return 1;
1733}
1734
1735static int on_deferred_gc(sd_event_source *s, void *userdata) {
1736 Manager *m = userdata;
1737
1738 assert(m);
1739
cf536638 1740 m->deferred_gc_event_source = sd_event_source_disable_unref(m->deferred_gc_event_source);
70a5db58
LP		 1741
1742 manager_gc_images(m);
1743 return 0;
1744}
1745
1746int manager_enqueue_gc(Manager *m, Home *focus) {
1747 int r;
1748
1749 assert(m);
1750
1751 /* This enqueues a request to GC dead homes. It may be called with focus=NULL in which case all homes
1752 * will be scanned, or with the parameter set, in which case only that home is checked. */
1753
1754 if (!m->event)
1755 return 0;
1756
1757 if (IN_SET(sd_event_get_state(m->event), SD_EVENT_FINISHED, SD_EVENT_EXITING))
1758 return 0;
1759
1760 /* If a focus home is specified, then remember to focus just on this home. Otherwise invalidate any
1761 * focus that might be set to look at all homes. */
1762
1763 if (m->deferred_gc_event_source) {
1764 if (m->gc_focus != focus) /* not the same focus, then look at everything */
1765 m->gc_focus = NULL;
1766
1767 return 0;
1768 } else
162392b7 1769 m->gc_focus = focus; /* start focused */
70a5db58
LP		 1770
1771 r = sd_event_add_defer(m->event, &m->deferred_gc_event_source, on_deferred_gc, m);
1772 if (r < 0)
80ace4f2 1773 return log_error_errno(r, "Failed to allocate GC event source: %m");
70a5db58
LP		 1774
1775 r = sd_event_source_set_priority(m->deferred_gc_event_source, SD_EVENT_PRIORITY_IDLE);
1776 if (r < 0)
1777 log_warning_errno(r, "Failed to tweak priority of event source, ignoring: %m");
1778
1779 (void) sd_event_source_set_description(m->deferred_gc_event_source, "deferred-gc");
1780 return 1;
1781}
d357b80d
LP		 1782
1783static bool manager_shall_rebalance(Manager *m) {
1784 Home *h;
1785
1786 assert(m);
1787
1788 if (IN_SET(m->rebalance_state, REBALANCE_PENDING, REBALANCE_SHRINKING, REBALANCE_GROWING))
1789 return true;
1790
1791 HASHMAP_FOREACH(h, m->homes_by_name)
1792 if (home_shall_rebalance(h))
1793 return true;
1794
1795 return false;
1796}
1797
1798static int home_cmp(Home *const*a, Home *const*b) {
1799 int r;
1800
1801 assert(a);
1802 assert(*a);
1803 assert(b);
1804 assert(*b);
1805
1806 /* Order user records by their weight (and by their name, to make things stable). We put the records
a6f44d61 1807 * with the highest weight last, since we distribute space from the beginning and round down, hence
d357b80d
LP		 1808 * later entries tend to get slightly more than earlier entries. */
1809
1810 r = CMP(user_record_rebalance_weight((*a)->record), user_record_rebalance_weight((*b)->record));
1811 if (r != 0)
1812 return r;
1813
1814 return strcmp((*a)->user_name, (*b)->user_name);
1815}
1816
1817static int manager_rebalance_calculate(Manager *m) {
1818 uint64_t weight_sum, free_sum, usage_sum = 0, min_free = UINT64_MAX;
1819 _cleanup_free_ Home **array = NULL;
1820 bool relevant = false;
1821 struct statfs sfs;
1822 int c = 0, r;
1823 Home *h;
1824
1825 assert(m);
1826
1827 if (statfs(get_home_root(), &sfs) < 0)
1828 return log_error_errno(errno, "Failed to statfs() /home: %m");
1829
1830 free_sum = (uint64_t) sfs.f_bsize * sfs.f_bavail; /* This much free space is available on the
1831 * underlying pool directory */
1832
1833 weight_sum = REBALANCE_WEIGHT_BACKING; /* Grant the underlying pool directory a fixed weight of 20
1834 * (home dirs get 100 by default, i.e. 5x more). This weight
1835 * is not configurable, the per-home weights are. */
1836
1837 HASHMAP_FOREACH(h, m->homes_by_name) {
1838 statfs_f_type_t fstype;
1839 h->rebalance_pending = false; /* First, reset the flag, we only want it to be true for the
1840 * homes that qualify for rebalancing */
1841
1842 if (!home_shall_rebalance(h)) /* Only look at actual candidates */
1843 continue;
1844
1845 if (home_is_busy(h))
1846 return -EBUSY; /* Let's not rebalance if there's a busy home directory. */
1847
1848 r = home_get_disk_status(
1849 h,
1850 &h->rebalance_size,
1851 &h->rebalance_usage,
1852 &h->rebalance_free,
1853 NULL,
1854 NULL,
1855 &fstype,
1856 NULL);
1857 if (r < 0) {
1858 log_warning_errno(r, "Failed to get free space of home '%s', ignoring.", h->user_name);
1859 continue;
1860 }
1861
1862 if (h->rebalance_free > UINT64_MAX - free_sum)
1863 return log_error_errno(SYNTHETIC_ERRNO(EOVERFLOW), "Rebalance free overflow");
1864 free_sum += h->rebalance_free;
1865
1866 if (h->rebalance_usage > UINT64_MAX - usage_sum)
1867 return log_error_errno(SYNTHETIC_ERRNO(EOVERFLOW), "Rebalance usage overflow");
1868 usage_sum += h->rebalance_usage;
1869
1870 h->rebalance_weight = user_record_rebalance_weight(h->record);
1871 if (h->rebalance_weight > UINT64_MAX - weight_sum)
1872 return log_error_errno(SYNTHETIC_ERRNO(EOVERFLOW), "Rebalance weight overflow");
1873 weight_sum += h->rebalance_weight;
1874
1875 h->rebalance_min = minimal_size_by_fs_magic(fstype);
1876
1877 if (!GREEDY_REALLOC(array, c+1))
1878 return log_oom();
1879
1880 array[c++] = h;
1881 }
1882
1883 if (c == 0) {
1884 log_debug("No homes to rebalance.");
1885 return 0;
1886 }
1887
1888 assert(weight_sum > 0);
1889
1890 log_debug("Disk space usage by all home directories to rebalance: %s — available disk space: %s",
1891 FORMAT_BYTES(usage_sum), FORMAT_BYTES(free_sum));
1892
1893 /* Bring the home directories in a well-defined order, so that we distribute space in a reproducible
1894 * way for the same parameters. */
1895 typesafe_qsort(array, c, home_cmp);
1896
1897 for (int i = 0; i < c; i++) {
1898 uint64_t new_free;
1899 double d;
1900
1901 h = array[i];
1902
1903 assert(h->rebalance_free <= free_sum);
1904 assert(h->rebalance_usage <= usage_sum);
1905 assert(h->rebalance_weight <= weight_sum);
1906
1907 d = ((double) (free_sum / 4096) * (double) h->rebalance_weight) / (double) weight_sum; /* Calculate new space for this home in units of 4K */
1908
1909 /* Convert from units of 4K back to bytes */
1910 if (d >= (double) (UINT64_MAX/4096))
1911 new_free = UINT64_MAX;
1912 else
1913 new_free = (uint64_t) d * 4096;
1914
1915 /* Subtract the weight and assigned space from the sums now, to distribute the rounding noise
1916 * to the remaining home dirs */
1917 free_sum = LESS_BY(free_sum, new_free);
1918 weight_sum = LESS_BY(weight_sum, h->rebalance_weight);
1919
1920 /* Keep track of home directory with the least amount of space left: we want to schedule the
1921 * next rebalance more quickly if this is low */
1922 if (new_free < min_free)
1923 min_free = h->rebalance_size;
1924
1925 if (new_free > UINT64_MAX - h->rebalance_usage)
1926 h->rebalance_goal = UINT64_MAX-1; /* maximum size */
1927 else {
1928 h->rebalance_goal = h->rebalance_usage + new_free;
1929
1930 if (h->rebalance_min != UINT64_MAX && h->rebalance_goal < h->rebalance_min)
1931 h->rebalance_goal = h->rebalance_min;
1932 }
1933
1934 /* Skip over this home if the state doesn't match the operation */
1935 if ((m->rebalance_state == REBALANCE_SHRINKING && h->rebalance_goal > h->rebalance_size) ||
1936 (m->rebalance_state == REBALANCE_GROWING && h->rebalance_goal < h->rebalance_size))
1937 h->rebalance_pending = false;
1938 else {
1939 log_debug("Rebalancing home directory '%s' %s → %s.", h->user_name,
1940 FORMAT_BYTES(h->rebalance_size), FORMAT_BYTES(h->rebalance_goal));
1941 h->rebalance_pending = true;
1942 }
1943
1944 if ((fabs((double) h->rebalance_size - (double) h->rebalance_goal) * 100 / (double) h->rebalance_size) >= 5.0)
1945 relevant = true;
1946 }
1947
1948 /* Scale next rebalancing interval based on the least amount of space of any of the home
1949 * directories. We pick a time in the range 1min … 15min, scaled by log2(min_free), so that:
1950 * 10M → ~0.7min, 100M → ~2.7min, 1G → ~4.6min, 10G → ~6.5min, 100G ~8.4 */
1951 m->rebalance_interval_usec = (usec_t) CLAMP((LESS_BY(log2(min_free), 22)*15*USEC_PER_MINUTE)/26,
1952 1 * USEC_PER_MINUTE,
1953 15 * USEC_PER_MINUTE);
1954
1955
1956 log_debug("Rebalancing interval set to %s.", FORMAT_TIMESPAN(m->rebalance_interval_usec, USEC_PER_MSEC));
1957
1958 /* Let's suppress small resizes, growing/shrinking file systems isn't free after all */
1959 if (!relevant) {
1960 log_debug("Skipping rebalancing, since all calculated size changes are below ±5%%.");
1961 return 0;
1962 }
1963
1964 return c;
1965}
1966
1967static int manager_rebalance_apply(Manager *m) {
1968 int c = 0, r;
1969 Home *h;
1970
1971 assert(m);
1972
1973 HASHMAP_FOREACH(h, m->homes_by_name) {
1974 _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1975
1976 if (!h->rebalance_pending)
1977 continue;
1978
1979 h->rebalance_pending = false;
1980
1981 r = home_resize(h, h->rebalance_goal, /* secret= */ NULL, /* automatic= */ true, &error);
1982 if (r < 0)
1983 log_warning_errno(r, "Failed to resize home '%s' for rebalancing, ignoring: %s",
1984 h->user_name, bus_error_message(&error, r));
1985 else
1986 c++;
1987 }
1988
1989 return c;
1990}
1991
49505916
LP		 1992static void manager_rebalance_reply_messages(Manager *m) {
1993 int r;
1994
1995 assert(m);
1996
1997 for (;;) {
1998 _cleanup_(sd_bus_message_unrefp) sd_bus_message *msg =
1999 set_steal_first(m->rebalance_pending_method_calls);
2000
2001 if (!msg)
2002 break;
2003
2004 r = sd_bus_reply_method_return(msg, NULL);
2005 if (r < 0)
2006 log_debug_errno(r, "Failed to reply to rebalance method call, ignoring: %m");
2007 }
2008}
2009
d357b80d
LP		 2010static int manager_rebalance_now(Manager *m) {
2011 RebalanceState busy_state; /* the state to revert to when operation fails if busy */
2012 int r;
2013
2014 assert(m);
2015
2016 log_debug("Rebalancing now...");
2017
2018 /* We maintain a simple state engine here to keep track of what we are doing. We'll first shrink all
a6f44d61 2019 * homes that shall be shrunk and then grow all homes that shall be grown, so that they can take up
d357b80d
LP		 2020 * the space now freed. */
2021
2022 for (;;) {
2023 switch (m->rebalance_state) {
2024
2025 case REBALANCE_IDLE:
2026 case REBALANCE_PENDING:
2027 case REBALANCE_WAITING:
2028 /* First shrink large home dirs */
2029 m->rebalance_state = REBALANCE_SHRINKING;
2030 busy_state = REBALANCE_PENDING;
49505916
LP		 2031
2032 /* We are initiating the next rebalancing cycle now, let's make the queued methods
2033 * calls the pending ones, and flush out any pending ones (which shouldn't exist at
2034 * this time anyway) */
2035 set_clear(m->rebalance_pending_method_calls);
2036 SWAP_TWO(m->rebalance_pending_method_calls, m->rebalance_queued_method_calls);
2037
d357b80d
LP		 2038 log_debug("Shrinking phase..");
2039 break;
2040
2041 case REBALANCE_SHRINKING:
2042 /* Then grow small home dirs */
2043 m->rebalance_state = REBALANCE_GROWING;
2044 busy_state = REBALANCE_SHRINKING;
2045 log_debug("Growing phase..");
2046 break;
2047
2048 case REBALANCE_GROWING:
2049 /* Finally, we are done */
2050 log_info("Rebalancing complete.");
2051 m->rebalance_state = REBALANCE_IDLE;
2052 r = 0;
2053 goto finish;
2054
2055 case REBALANCE_OFF:
2056 default:
2057 assert_not_reached();
2058 }
2059
2060 r = manager_rebalance_calculate(m);
2061 if (r == -EBUSY) {
2062 /* Calculations failed because one home directory is currently busy. Revert to a state that
2063 * tells us what to do next. */
2064 log_debug("Can't enter phase, busy.");
2065 m->rebalance_state = busy_state;
2066 return r;
2067 }
2068 if (r < 0)
2069 goto finish;
2070 if (r == 0)
2071 continue; /* got to next step immediately, if there's nothing to do */
2072
2073 r = manager_rebalance_apply(m);
2074 if (r < 0)
2075 goto finish;
2076 if (r > 0)
2077 break; /* At least one resize operation is now pending, we are done for now */
2078
2079 /* If there was nothing to apply, go for next state right-away */
2080 }
2081
2082 return 0;
2083
2084finish:
2085 /* Reset state and schedule next rebalance */
2086 m->rebalance_state = REBALANCE_IDLE;
49505916 2087 manager_rebalance_reply_messages(m);
d357b80d
LP		 2088 (void) manager_schedule_rebalance(m, /* immediately= */ false);
2089 return r;
2090}
2091
2092static int on_rebalance_timer(sd_event_source *s, usec_t t, void *userdata) {
2093 Manager *m = userdata;
2094
2095 assert(s);
2096 assert(m);
2097 assert(IN_SET(m->rebalance_state, REBALANCE_WAITING, REBALANCE_PENDING, REBALANCE_SHRINKING, REBALANCE_GROWING));
2098
2099 (void) manager_rebalance_now(m);
2100 return 0;
2101}
2102
2103int manager_schedule_rebalance(Manager *m, bool immediately) {
2104 int r;
2105
2106 assert(m);
2107
2108 /* Check if there are any records where rebalancing is requested */
2109 if (!manager_shall_rebalance(m)) {
2110 log_debug("Not scheduling rebalancing, not needed.");
49505916 2111 r = 0; /* report that we didn't schedule anything because nothing needed it */
d357b80d
LP		 2112 goto turn_off;
2113 }
2114
2115 if (immediately) {
2116 /* If we are told to rebalance immediately, then mark a rebalance as pending (even if we area
2117 * already running one) */
2118
2119 if (m->rebalance_event_source) {
2120 r = sd_event_source_set_time(m->rebalance_event_source, 0);
2121 if (r < 0) {
2122 log_error_errno(r, "Failed to schedule immediate rebalancing: %m");
2123 goto turn_off;
2124 }
2125
2126 r = sd_event_source_set_enabled(m->rebalance_event_source, SD_EVENT_ONESHOT);
2127 if (r < 0) {
2128 log_error_errno(r, "Failed to enable rebalancing event source: %m");
2129 goto turn_off;
2130 }
2131 } else {
2132 r = sd_event_add_time(m->event, &m->rebalance_event_source, CLOCK_MONOTONIC, 0, USEC_PER_SEC, on_rebalance_timer, m);
2133 if (r < 0) {
2134 log_error_errno(r, "Failed to allocate rebalance event source: %m");
2135 goto turn_off;
2136 }
2137
2138 r = sd_event_source_set_priority(m->rebalance_event_source, SD_EVENT_PRIORITY_IDLE + 10);
2139 if (r < 0) {
2140 log_error_errno(r, "Failed to set rebalance event source priority: %m");
2141 goto turn_off;
2142 }
2143
2144 (void) sd_event_source_set_description(m->rebalance_event_source, "rebalance");
2145
2146 }
2147
2148 if (!IN_SET(m->rebalance_state, REBALANCE_PENDING, REBALANCE_SHRINKING, REBALANCE_GROWING))
2149 m->rebalance_state = REBALANCE_PENDING;
2150
2151 log_debug("Scheduled immediate rebalancing...");
49505916 2152 return 1; /* report that we scheduled something */
d357b80d
LP		 2153 }
2154
2155 /* If we are told to schedule a rebalancing eventually, then do so only if we are not executing
2156 * anything yet. Also if we have something scheduled already, leave it in place */
2157 if (!IN_SET(m->rebalance_state, REBALANCE_OFF, REBALANCE_IDLE))
49505916 2158 return 1; /* report that there's already something scheduled */
d357b80d
LP		 2159
2160 if (m->rebalance_event_source) {
2161 r = sd_event_source_set_time_relative(m->rebalance_event_source, m->rebalance_interval_usec);
2162 if (r < 0) {
2163 log_error_errno(r, "Failed to schedule immediate rebalancing: %m");
2164 goto turn_off;
2165 }
2166
2167 r = sd_event_source_set_enabled(m->rebalance_event_source, SD_EVENT_ONESHOT);
2168 if (r < 0) {
2169 log_error_errno(r, "Failed to enable rebalancing event source: %m");
2170 goto turn_off;
2171 }
2172 } else {
2173 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);
2174 if (r < 0) {
2175 log_error_errno(r, "Failed to allocate rebalance event source: %m");
2176 goto turn_off;
2177 }
2178
2179 r = sd_event_source_set_priority(m->rebalance_event_source, SD_EVENT_PRIORITY_IDLE + 10);
2180 if (r < 0) {
2181 log_error_errno(r, "Failed to set rebalance event source priority: %m");
2182 goto turn_off;
2183 }
2184
2185 (void) sd_event_source_set_description(m->rebalance_event_source, "rebalance");
2186 }
2187
2188 m->rebalance_state = REBALANCE_WAITING; /* We managed to enqueue a timer event, we now wait until it fires */
2189 log_debug("Scheduled rebalancing in %s...", FORMAT_TIMESPAN(m->rebalance_interval_usec, 0));
49505916 2190 return 1; /* report that we scheduled something */
d357b80d
LP		 2191
2192turn_off:
2193 m->rebalance_event_source = sd_event_source_disable_unref(m->rebalance_event_source);
2194 m->rebalance_state = REBALANCE_OFF;
49505916 2195 manager_rebalance_reply_messages(m);
d357b80d
LP		 2196 return r;
2197}
2198
2199int manager_reschedule_rebalance(Manager *m) {
2200 int r;
2201
2202 assert(m);
2203
2204 /* If a rebalance is pending reschedules it so it gets executed immediately */
2205
2206 if (!IN_SET(m->rebalance_state, REBALANCE_PENDING, REBALANCE_SHRINKING, REBALANCE_GROWING))
2207 return 0;
2208
2209 r = manager_schedule_rebalance(m, /* immediately= */ true);
2210 if (r < 0)
2211 return r;
2212
2213 return 1;
2214}
Unnamed repository; edit this file 'description' to name the repository.