projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
bcachefs: Pass flags arg to bch2_alloc_write()
[linux-block.git] / fs / bcachefs / super.c
CommitLineData
1c6fdbd8
KO		 1// SPDX-License-Identifier: GPL-2.0
2/*
3 * bcachefs setup/teardown code, and some metadata io - read a superblock and
4 * figure out what to do with it.
5 *
6 * Copyright 2010, 2011 Kent Overstreet <kent.overstreet@gmail.com>
7 * Copyright 2012 Google, Inc.
8 */
9
10#include "bcachefs.h"
7b3f84ea
KO		 11#include "alloc_background.h"
12#include "alloc_foreground.h"
5b8a9227 13#include "bkey_sort.h"
1c6fdbd8
KO		 14#include "btree_cache.h"
15#include "btree_gc.h"
16#include "btree_update_interior.h"
17#include "btree_io.h"
18#include "chardev.h"
19#include "checksum.h"
20#include "clock.h"
21#include "compress.h"
22#include "debug.h"
23#include "disk_groups.h"
cd575ddf 24#include "ec.h"
1c6fdbd8
KO		 25#include "error.h"
26#include "fs.h"
27#include "fs-io.h"
28#include "fsck.h"
29#include "inode.h"
30#include "io.h"
31#include "journal.h"
32#include "journal_reclaim.h"
1dd7f9d9 33#include "journal_seq_blacklist.h"
1c6fdbd8
KO		 34#include "move.h"
35#include "migrate.h"
36#include "movinggc.h"
37#include "quota.h"
38#include "rebalance.h"
39#include "recovery.h"
40#include "replicas.h"
41#include "super.h"
42#include "super-io.h"
43#include "sysfs.h"
44#include "trace.h"
45
46#include <linux/backing-dev.h>
47#include <linux/blkdev.h>
48#include <linux/debugfs.h>
49#include <linux/device.h>
50#include <linux/idr.h>
51#include <linux/kthread.h>
52#include <linux/module.h>
53#include <linux/percpu.h>
54#include <linux/random.h>
55#include <linux/sysfs.h>
56#include <crypto/hash.h>
57
58MODULE_LICENSE("GPL");
59MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
60
61#define KTYPE(type) \
62static const struct attribute_group type ## _group = { \
63 .attrs = type ## _files \
64}; \
65 \
66static const struct attribute_group *type ## _groups[] = { \
67 &type ## _group, \
68 NULL \
69}; \
70 \
71static const struct kobj_type type ## _ktype = { \
72 .release = type ## _release, \
73 .sysfs_ops = &type ## _sysfs_ops, \
74 .default_groups = type ## _groups \
75}
76
77static void bch2_fs_release(struct kobject *);
78static void bch2_dev_release(struct kobject *);
79
80static void bch2_fs_internal_release(struct kobject *k)
81{
82}
83
84static void bch2_fs_opts_dir_release(struct kobject *k)
85{
86}
87
88static void bch2_fs_time_stats_release(struct kobject *k)
89{
90}
91
92KTYPE(bch2_fs);
93KTYPE(bch2_fs_internal);
94KTYPE(bch2_fs_opts_dir);
95KTYPE(bch2_fs_time_stats);
96KTYPE(bch2_dev);
97
98static struct kset *bcachefs_kset;
99static LIST_HEAD(bch_fs_list);
100static DEFINE_MUTEX(bch_fs_list_lock);
101
102static DECLARE_WAIT_QUEUE_HEAD(bch_read_only_wait);
103
104static void bch2_dev_free(struct bch_dev *);
105static int bch2_dev_alloc(struct bch_fs *, unsigned);
106static int bch2_dev_sysfs_online(struct bch_fs *, struct bch_dev *);
107static void __bch2_dev_read_only(struct bch_fs *, struct bch_dev *);
108
109struct bch_fs *bch2_dev_to_fs(dev_t dev)
110{
111 struct bch_fs *c;
112 struct bch_dev *ca;
113 unsigned i;
114
115 mutex_lock(&bch_fs_list_lock);
116 rcu_read_lock();
117
118 list_for_each_entry(c, &bch_fs_list, list)
119 for_each_member_device_rcu(ca, c, i, NULL)
120 if (ca->disk_sb.bdev->bd_dev == dev) {
121 closure_get(&c->cl);
122 goto found;
123 }
124 c = NULL;
125found:
126 rcu_read_unlock();
127 mutex_unlock(&bch_fs_list_lock);
128
129 return c;
130}
131
132static struct bch_fs *__bch2_uuid_to_fs(__uuid_t uuid)
133{
134 struct bch_fs *c;
135
136 lockdep_assert_held(&bch_fs_list_lock);
137
138 list_for_each_entry(c, &bch_fs_list, list)
139 if (!memcmp(&c->disk_sb.sb->uuid, &uuid, sizeof(uuid)))
140 return c;
141
142 return NULL;
143}
144
145struct bch_fs *bch2_uuid_to_fs(__uuid_t uuid)
146{
147 struct bch_fs *c;
148
149 mutex_lock(&bch_fs_list_lock);
150 c = __bch2_uuid_to_fs(uuid);
151 if (c)
152 closure_get(&c->cl);
153 mutex_unlock(&bch_fs_list_lock);
154
155 return c;
156}
157
158/* Filesystem RO/RW: */
159
160/*
161 * For startup/shutdown of RW stuff, the dependencies are:
162 *
163 * - foreground writes depend on copygc and rebalance (to free up space)
164 *
165 * - copygc and rebalance depend on mark and sweep gc (they actually probably
166 * don't because they either reserve ahead of time or don't block if
167 * allocations fail, but allocations can require mark and sweep gc to run
168 * because of generation number wraparound)
169 *
170 * - all of the above depends on the allocator threads
171 *
172 * - allocator depends on the journal (when it rewrites prios and gens)
173 */
174
175static void __bch2_fs_read_only(struct bch_fs *c)
176{
177 struct bch_dev *ca;
430735cd 178 bool wrote;
d5f70c1f 179 unsigned i, clean_passes = 0;
430735cd 180 int ret;
1c6fdbd8
KO		 181
182 bch2_rebalance_stop(c);
183
184 for_each_member_device(ca, c, i)
185 bch2_copygc_stop(ca);
186
187 bch2_gc_thread_stop(c);
188
189 /*
190 * Flush journal before stopping allocators, because flushing journal
191 * blacklist entries involves allocating new btree nodes:
192 */
193 bch2_journal_flush_all_pins(&c->journal);
194
b935a8a6
KO		 195 if (!test_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags))
196 goto allocator_not_running;
197
430735cd 198 do {
a0e0bda1
KO		 199 wrote = false;
200
201 ret = bch2_stripes_write(c, BTREE_INSERT_NOCHECK_RW, &wrote);
430735cd 202 if (ret) {
d5f70c1f 203 bch2_fs_inconsistent(c, "error writing out stripes");
430735cd
KO		 204 break;
205 }
206
a0e0bda1 207 ret = bch2_alloc_write(c, BTREE_INSERT_NOCHECK_RW, &wrote);
61c8d7c8 208 if (ret) {
d5f70c1f 209 bch2_fs_inconsistent(c, "error writing out alloc info %i", ret);
61c8d7c8
KO		 210 break;
211 }
212
430735cd
KO		 213 for_each_member_device(ca, c, i)
214 bch2_dev_allocator_quiesce(c, ca);
215
216 bch2_journal_flush_all_pins(&c->journal);
217
218 /*
219 * We need to explicitly wait on btree interior updates to complete
220 * before stopping the journal, flushing all journal pins isn't
221 * sufficient, because in the BTREE_INTERIOR_UPDATING_ROOT case btree
222 * interior updates have to drop their journal pin before they're
223 * fully complete:
224 */
225 closure_wait_event(&c->btree_interior_update_wait,
226 !bch2_btree_interior_updates_nr_pending(c));
d5f70c1f
KO		 227
228 clean_passes = wrote ? 0 : clean_passes + 1;
229 } while (clean_passes < 2);
b935a8a6 230allocator_not_running:
1c6fdbd8
KO		 231 for_each_member_device(ca, c, i)
232 bch2_dev_allocator_stop(ca);
233
b935a8a6
KO		 234 clear_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags);
235
1c6fdbd8
KO		 236 bch2_fs_journal_stop(&c->journal);
237
430735cd
KO		 238 /* XXX: mark super that alloc info is persistent */
239
1c6fdbd8
KO		 240 /*
241 * the journal kicks off btree writes via reclaim - wait for in flight
242 * writes after stopping journal:
243 */
244 if (test_bit(BCH_FS_EMERGENCY_RO, &c->flags))
245 bch2_btree_flush_all_writes(c);
246 else
247 bch2_btree_verify_flushed(c);
248
249 /*
250 * After stopping journal:
251 */
252 for_each_member_device(ca, c, i)
253 bch2_dev_allocator_remove(c, ca);
254}
255
256static void bch2_writes_disabled(struct percpu_ref *writes)
257{
258 struct bch_fs *c = container_of(writes, struct bch_fs, writes);
259
260 set_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
261 wake_up(&bch_read_only_wait);
262}
263
264void bch2_fs_read_only(struct bch_fs *c)
265{
134915f3
KO		 266 if (!test_bit(BCH_FS_RW, &c->flags)) {
267 cancel_delayed_work_sync(&c->journal.reclaim_work);
1c6fdbd8 268 return;
134915f3 269 }
1c6fdbd8
KO		 270
271 BUG_ON(test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
272
273 /*
274 * Block new foreground-end write operations from starting - any new
275 * writes will return -EROFS:
276 *
277 * (This is really blocking new _allocations_, writes to previously
278 * allocated space can still happen until stopping the allocator in
279 * bch2_dev_allocator_stop()).
280 */
281 percpu_ref_kill(&c->writes);
282
283 cancel_delayed_work(&c->pd_controllers_update);
284
285 /*
286 * If we're not doing an emergency shutdown, we want to wait on
287 * outstanding writes to complete so they don't see spurious errors due
288 * to shutting down the allocator:
289 *
290 * If we are doing an emergency shutdown outstanding writes may
291 * hang until we shutdown the allocator so we don't want to wait
292 * on outstanding writes before shutting everything down - but
293 * we do need to wait on them before returning and signalling
294 * that going RO is complete:
295 */
296 wait_event(bch_read_only_wait,
297 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags) ||
298 test_bit(BCH_FS_EMERGENCY_RO, &c->flags));
299
300 __bch2_fs_read_only(c);
301
302 wait_event(bch_read_only_wait,
303 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
304
305 clear_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
306
307 if (!bch2_journal_error(&c->journal) &&
308 !test_bit(BCH_FS_ERROR, &c->flags) &&
3aea4342 309 !test_bit(BCH_FS_EMERGENCY_RO, &c->flags) &&
a0e0bda1
KO		 310 test_bit(BCH_FS_STARTED, &c->flags) &&
311 !c->opts.noreplay &&
312 !c->opts.norecovery)
134915f3 313 bch2_fs_mark_clean(c);
1c6fdbd8 314
134915f3 315 clear_bit(BCH_FS_RW, &c->flags);
1c6fdbd8
KO		 316}
317
318static void bch2_fs_read_only_work(struct work_struct *work)
319{
320 struct bch_fs *c =
321 container_of(work, struct bch_fs, read_only_work);
322
323 mutex_lock(&c->state_lock);
324 bch2_fs_read_only(c);
325 mutex_unlock(&c->state_lock);
326}
327
328static void bch2_fs_read_only_async(struct bch_fs *c)
329{
330 queue_work(system_long_wq, &c->read_only_work);
331}
332
333bool bch2_fs_emergency_read_only(struct bch_fs *c)
334{
335 bool ret = !test_and_set_bit(BCH_FS_EMERGENCY_RO, &c->flags);
336
337 bch2_fs_read_only_async(c);
338 bch2_journal_halt(&c->journal);
339
340 wake_up(&bch_read_only_wait);
341 return ret;
342}
343
134915f3 344static int bch2_fs_read_write_late(struct bch_fs *c)
1c6fdbd8
KO		 345{
346 struct bch_dev *ca;
1c6fdbd8 347 unsigned i;
134915f3 348 int ret;
1c6fdbd8 349
134915f3
KO		 350 ret = bch2_gc_thread_start(c);
351 if (ret) {
352 bch_err(c, "error starting gc thread");
353 return ret;
354 }
355
356 for_each_rw_member(ca, c, i) {
357 ret = bch2_copygc_start(c, ca);
358 if (ret) {
359 bch_err(c, "error starting copygc threads");
360 percpu_ref_put(&ca->io_ref);
361 return ret;
362 }
363 }
364
365 ret = bch2_rebalance_start(c);
366 if (ret) {
367 bch_err(c, "error starting rebalance thread");
368 return ret;
369 }
370
371 schedule_delayed_work(&c->pd_controllers_update, 5 * HZ);
372
373 return 0;
374}
1c6fdbd8 375
03e183cb 376int __bch2_fs_read_write(struct bch_fs *c, bool early)
134915f3
KO		 377{
378 struct bch_dev *ca;
379 unsigned i;
380 int ret;
381
382 if (test_bit(BCH_FS_RW, &c->flags))
383 return 0;
384
330581f1
KO		 385 if (c->opts.nochanges ||
386 c->opts.noreplay)
387 return -EINVAL;
388
134915f3
KO		 389 ret = bch2_fs_mark_dirty(c);
390 if (ret)
391 goto err;
1c6fdbd8
KO		 392
393 for_each_rw_member(ca, c, i)
394 bch2_dev_allocator_add(c, ca);
395 bch2_recalc_capacity(c);
396
134915f3
KO		 397 if (!test_bit(BCH_FS_ALLOCATOR_STARTED, &c->flags)) {
398 ret = bch2_fs_allocator_start(c);
399 if (ret) {
400 bch_err(c, "error initializing allocator");
1c6fdbd8
KO		 401 goto err;
402 }
403
134915f3
KO		 404 set_bit(BCH_FS_ALLOCATOR_STARTED, &c->flags);
405 }
1c6fdbd8 406
134915f3
KO		 407 for_each_rw_member(ca, c, i) {
408 ret = bch2_dev_allocator_start(ca);
409 if (ret) {
410 bch_err(c, "error starting allocator threads");
1c6fdbd8
KO		 411 percpu_ref_put(&ca->io_ref);
412 goto err;
413 }
134915f3 414 }
1c6fdbd8 415
134915f3 416 set_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags);
1c6fdbd8 417
134915f3
KO		 418 if (!early) {
419 ret = bch2_fs_read_write_late(c);
420 if (ret)
421 goto err;
422 }
1c6fdbd8 423
134915f3
KO		 424 percpu_ref_reinit(&c->writes);
425 set_bit(BCH_FS_RW, &c->flags);
1c6fdbd8 426
134915f3
KO		 427 queue_delayed_work(c->journal_reclaim_wq,
428 &c->journal.reclaim_work, 0);
429 return 0;
1c6fdbd8
KO		 430err:
431 __bch2_fs_read_only(c);
134915f3
KO		 432 return ret;
433}
434
435int bch2_fs_read_write(struct bch_fs *c)
436{
437 return __bch2_fs_read_write(c, false);
438}
439
440int bch2_fs_read_write_early(struct bch_fs *c)
441{
442 lockdep_assert_held(&c->state_lock);
443
134915f3 444 return __bch2_fs_read_write(c, true);
1c6fdbd8
KO		 445}
446
447/* Filesystem startup/shutdown: */
448
449static void bch2_fs_free(struct bch_fs *c)
450{
451 unsigned i;
452
453 for (i = 0; i < BCH_TIME_STAT_NR; i++)
454 bch2_time_stats_exit(&c->times[i]);
455
456 bch2_fs_quota_exit(c);
457 bch2_fs_fsio_exit(c);
cd575ddf 458 bch2_fs_ec_exit(c);
1c6fdbd8
KO		 459 bch2_fs_encryption_exit(c);
460 bch2_fs_io_exit(c);
461 bch2_fs_btree_cache_exit(c);
462 bch2_fs_journal_exit(&c->journal);
463 bch2_io_clock_exit(&c->io_clock[WRITE]);
464 bch2_io_clock_exit(&c->io_clock[READ]);
465 bch2_fs_compress_exit(c);
9166b41d 466 percpu_free_rwsem(&c->mark_lock);
4d8100da 467 kfree(c->usage_scratch);
9ca53b55 468 free_percpu(c->usage[0]);
5663a415 469 free_percpu(c->pcpu);
581edb63 470 mempool_exit(&c->btree_iters_pool);
1c6fdbd8
KO		 471 mempool_exit(&c->btree_bounce_pool);
472 bioset_exit(&c->btree_bio);
473 mempool_exit(&c->btree_interior_update_pool);
474 mempool_exit(&c->btree_reserve_pool);
475 mempool_exit(&c->fill_iter);
476 percpu_ref_exit(&c->writes);
73e6ab95
KO		 477 kfree(c->replicas.entries);
478 kfree(c->replicas_gc.entries);
1c6fdbd8 479 kfree(rcu_dereference_protected(c->disk_groups, 1));
1dd7f9d9 480 kfree(c->journal_seq_blacklist_table);
1c6fdbd8 481
0519b72d
KO		 482 if (c->journal_reclaim_wq)
483 destroy_workqueue(c->journal_reclaim_wq);
1c6fdbd8
KO		 484 if (c->copygc_wq)
485 destroy_workqueue(c->copygc_wq);
486 if (c->wq)
487 destroy_workqueue(c->wq);
488
489 free_pages((unsigned long) c->disk_sb.sb,
490 c->disk_sb.page_order);
491 kvpfree(c, sizeof(*c));
492 module_put(THIS_MODULE);
493}
494
495static void bch2_fs_release(struct kobject *kobj)
496{
497 struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
498
499 bch2_fs_free(c);
500}
501
502void bch2_fs_stop(struct bch_fs *c)
503{
504 struct bch_dev *ca;
505 unsigned i;
506
af1c6871
KO		 507 bch_verbose(c, "shutting down");
508
1dd7f9d9
KO		 509 set_bit(BCH_FS_STOPPING, &c->flags);
510
511 cancel_work_sync(&c->journal_seq_blacklist_gc_work);
512
1c6fdbd8
KO		 513 for_each_member_device(ca, c, i)
514 if (ca->kobj.state_in_sysfs &&
515 ca->disk_sb.bdev)
516 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
517
518 if (c->kobj.state_in_sysfs)
519 kobject_del(&c->kobj);
520
521 bch2_fs_debug_exit(c);
522 bch2_fs_chardev_exit(c);
523
524 kobject_put(&c->time_stats);
525 kobject_put(&c->opts_dir);
526 kobject_put(&c->internal);
527
528 mutex_lock(&bch_fs_list_lock);
529 list_del(&c->list);
530 mutex_unlock(&bch_fs_list_lock);
531
532 closure_sync(&c->cl);
533 closure_debug_destroy(&c->cl);
534
535 mutex_lock(&c->state_lock);
536 bch2_fs_read_only(c);
537 mutex_unlock(&c->state_lock);
538
539 /* btree prefetch might have kicked off reads in the background: */
540 bch2_btree_flush_all_reads(c);
541
542 for_each_member_device(ca, c, i)
543 cancel_work_sync(&ca->io_error_work);
544
545 cancel_work_sync(&c->btree_write_error_work);
546 cancel_delayed_work_sync(&c->pd_controllers_update);
547 cancel_work_sync(&c->read_only_work);
548
549 for (i = 0; i < c->sb.nr_devices; i++)
550 if (c->devs[i])
551 bch2_dev_free(rcu_dereference_protected(c->devs[i], 1));
552
af1c6871
KO		 553 bch_verbose(c, "shutdown complete");
554
1c6fdbd8
KO		 555 kobject_put(&c->kobj);
556}
557
558static const char *bch2_fs_online(struct bch_fs *c)
559{
560 struct bch_dev *ca;
561 const char *err = NULL;
562 unsigned i;
563 int ret;
564
565 lockdep_assert_held(&bch_fs_list_lock);
566
567 if (!list_empty(&c->list))
568 return NULL;
569
570 if (__bch2_uuid_to_fs(c->sb.uuid))
571 return "filesystem UUID already open";
572
573 ret = bch2_fs_chardev_init(c);
574 if (ret)
575 return "error creating character device";
576
577 bch2_fs_debug_init(c);
578
579 if (kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ||
580 kobject_add(&c->internal, &c->kobj, "internal") ||
581 kobject_add(&c->opts_dir, &c->kobj, "options") ||
582 kobject_add(&c->time_stats, &c->kobj, "time_stats") ||
583 bch2_opts_create_sysfs_files(&c->opts_dir))
584 return "error creating sysfs objects";
585
586 mutex_lock(&c->state_lock);
587
588 err = "error creating sysfs objects";
589 __for_each_member_device(ca, c, i, NULL)
590 if (bch2_dev_sysfs_online(c, ca))
591 goto err;
592
593 list_add(&c->list, &bch_fs_list);
594 err = NULL;
595err:
596 mutex_unlock(&c->state_lock);
597 return err;
598}
599
600static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
601{
602 struct bch_sb_field_members *mi;
603 struct bch_fs *c;
ecf37a4a 604 unsigned i, iter_size;
1c6fdbd8
KO		 605 const char *err;
606
607 pr_verbose_init(opts, "");
608
609 c = kvpmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
610 if (!c)
611 goto out;
612
613 __module_get(THIS_MODULE);
614
615 c->minor = -1;
616 c->disk_sb.fs_sb = true;
617
618 mutex_init(&c->state_lock);
619 mutex_init(&c->sb_lock);
620 mutex_init(&c->replicas_gc_lock);
621 mutex_init(&c->btree_root_lock);
622 INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
623
624 init_rwsem(&c->gc_lock);
625
626 for (i = 0; i < BCH_TIME_STAT_NR; i++)
627 bch2_time_stats_init(&c->times[i]);
628
b092dadd
KO		 629 bch2_fs_allocator_background_init(c);
630 bch2_fs_allocator_foreground_init(c);
1c6fdbd8
KO		 631 bch2_fs_rebalance_init(c);
632 bch2_fs_quota_init(c);
633
634 INIT_LIST_HEAD(&c->list);
635
636 INIT_LIST_HEAD(&c->btree_interior_update_list);
637 mutex_init(&c->btree_reserve_cache_lock);
638 mutex_init(&c->btree_interior_update_lock);
639
4d8100da
KO		 640 mutex_init(&c->usage_scratch_lock);
641
1c6fdbd8
KO		 642 mutex_init(&c->bio_bounce_pages_lock);
643
644 bio_list_init(&c->btree_write_error_list);
645 spin_lock_init(&c->btree_write_error_lock);
646 INIT_WORK(&c->btree_write_error_work, bch2_btree_write_error_work);
647
1dd7f9d9
KO		 648 INIT_WORK(&c->journal_seq_blacklist_gc_work,
649 bch2_blacklist_entries_gc);
650
1c6fdbd8
KO		 651 INIT_LIST_HEAD(&c->fsck_errors);
652 mutex_init(&c->fsck_error_lock);
653
cd575ddf
KO		 654 INIT_LIST_HEAD(&c->ec_new_stripe_list);
655 mutex_init(&c->ec_new_stripe_lock);
dfe9bfb3 656 mutex_init(&c->ec_stripe_create_lock);
cd575ddf
KO		 657 spin_lock_init(&c->ec_stripes_heap_lock);
658
1c6fdbd8
KO		 659 seqcount_init(&c->gc_pos_lock);
660
661 c->copy_gc_enabled = 1;
662 c->rebalance.enabled = 1;
663 c->promote_whole_extents = true;
664
665 c->journal.write_time = &c->times[BCH_TIME_journal_write];
666 c->journal.delay_time = &c->times[BCH_TIME_journal_delay];
49a67206 667 c->journal.blocked_time = &c->times[BCH_TIME_blocked_journal];
1c6fdbd8
KO		 668 c->journal.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq];
669
670 bch2_fs_btree_cache_init_early(&c->btree_cache);
671
73e6ab95
KO		 672 if (percpu_init_rwsem(&c->mark_lock))
673 goto err;
674
1c6fdbd8
KO		 675 mutex_lock(&c->sb_lock);
676
677 if (bch2_sb_to_fs(c, sb)) {
678 mutex_unlock(&c->sb_lock);
679 goto err;
680 }
681
682 mutex_unlock(&c->sb_lock);
683
684 scnprintf(c->name, sizeof(c->name), "%pU", &c->sb.user_uuid);
685
686 c->opts = bch2_opts_default;
687 bch2_opts_apply(&c->opts, bch2_opts_from_sb(sb));
688 bch2_opts_apply(&c->opts, opts);
689
690 c->block_bits = ilog2(c->opts.block_size);
691 c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
692
693 c->opts.nochanges |= c->opts.noreplay;
694 c->opts.read_only |= c->opts.nochanges;
330581f1 695 c->opts.read_only |= c->opts.noreplay;
1c6fdbd8
KO		 696
697 if (bch2_fs_init_fault("fs_alloc"))
698 goto err;
699
700 iter_size = sizeof(struct btree_node_iter_large) +
701 (btree_blocks(c) + 1) * 2 *
702 sizeof(struct btree_node_iter_set);
703
704 if (!(c->wq = alloc_workqueue("bcachefs",
705 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_HIGHPRI, 1)) ||
706 !(c->copygc_wq = alloc_workqueue("bcache_copygc",
707 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_HIGHPRI, 1)) ||
0519b72d
KO		 708 !(c->journal_reclaim_wq = alloc_workqueue("bcache_journal",
709 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_HIGHPRI, 1)) ||
134915f3
KO		 710 percpu_ref_init(&c->writes, bch2_writes_disabled,
711 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
1c6fdbd8
KO		 712 mempool_init_kmalloc_pool(&c->btree_reserve_pool, 1,
713 sizeof(struct btree_reserve)) ||
714 mempool_init_kmalloc_pool(&c->btree_interior_update_pool, 1,
715 sizeof(struct btree_update)) ||
716 mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
717 bioset_init(&c->btree_bio, 1,
718 max(offsetof(struct btree_read_bio, bio),
719 offsetof(struct btree_write_bio, wbio.bio)),
720 BIOSET_NEED_BVECS) ||
5663a415 721 !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
1c6fdbd8
KO		 722 mempool_init_kvpmalloc_pool(&c->btree_bounce_pool, 1,
723 btree_bytes(c)) ||
581edb63 724 mempool_init_kmalloc_pool(&c->btree_iters_pool, 1,
a8e00bd4
KO		 725 sizeof(struct btree_iter) * BTREE_ITER_MAX +
726 sizeof(struct btree_insert_entry) *
727 (BTREE_ITER_MAX + 4)) ||
1c6fdbd8
KO		 728 bch2_io_clock_init(&c->io_clock[READ]) ||
729 bch2_io_clock_init(&c->io_clock[WRITE]) ||
730 bch2_fs_journal_init(&c->journal) ||
2c5af169 731 bch2_fs_replicas_init(c) ||
1c6fdbd8
KO		 732 bch2_fs_btree_cache_init(c) ||
733 bch2_fs_io_init(c) ||
734 bch2_fs_encryption_init(c) ||
735 bch2_fs_compress_init(c) ||
cd575ddf 736 bch2_fs_ec_init(c) ||
1c6fdbd8
KO		 737 bch2_fs_fsio_init(c))
738 goto err;
739
740 mi = bch2_sb_get_members(c->disk_sb.sb);
741 for (i = 0; i < c->sb.nr_devices; i++)
742 if (bch2_dev_exists(c->disk_sb.sb, mi, i) &&
743 bch2_dev_alloc(c, i))
744 goto err;
745
746 /*
747 * Now that all allocations have succeeded, init various refcounty
748 * things that let us shutdown:
749 */
750 closure_init(&c->cl, NULL);
751
752 c->kobj.kset = bcachefs_kset;
753 kobject_init(&c->kobj, &bch2_fs_ktype);
754 kobject_init(&c->internal, &bch2_fs_internal_ktype);
755 kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
756 kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
757
758 mutex_lock(&bch_fs_list_lock);
759 err = bch2_fs_online(c);
760 mutex_unlock(&bch_fs_list_lock);
761 if (err) {
762 bch_err(c, "bch2_fs_online() error: %s", err);
763 goto err;
764 }
765out:
766 pr_verbose_init(opts, "ret %i", c ? 0 : -ENOMEM);
767 return c;
768err:
769 bch2_fs_free(c);
770 c = NULL;
771 goto out;
772}
773
774const char *bch2_fs_start(struct bch_fs *c)
775{
776 const char *err = "cannot allocate memory";
777 struct bch_sb_field_members *mi;
778 struct bch_dev *ca;
a420eea6 779 time64_t now = ktime_get_real_seconds();
1c6fdbd8
KO		 780 unsigned i;
781 int ret = -EINVAL;
782
783 mutex_lock(&c->state_lock);
784
134915f3 785 BUG_ON(test_bit(BCH_FS_STARTED, &c->flags));
1c6fdbd8
KO		 786
787 mutex_lock(&c->sb_lock);
788
789 for_each_online_member(ca, c, i)
790 bch2_sb_from_fs(c, ca);
791
792 mi = bch2_sb_get_members(c->disk_sb.sb);
793 for_each_online_member(ca, c, i)
794 mi->members[ca->dev_idx].last_mount = cpu_to_le64(now);
795
796 mutex_unlock(&c->sb_lock);
797
798 for_each_rw_member(ca, c, i)
799 bch2_dev_allocator_add(c, ca);
800 bch2_recalc_capacity(c);
801
802 ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
803 ? bch2_fs_recovery(c)
804 : bch2_fs_initialize(c);
805 if (ret)
806 goto err;
cd575ddf
KO		 807
808 ret = bch2_opts_check_may_set(c);
809 if (ret)
810 goto err;
1c6fdbd8
KO		 811
812 err = "dynamic fault";
813 if (bch2_fs_init_fault("fs_start"))
814 goto err;
815
816 if (c->opts.read_only) {
817 bch2_fs_read_only(c);
818 } else {
134915f3
KO		 819 if (!test_bit(BCH_FS_RW, &c->flags)
820 ? bch2_fs_read_write(c)
821 : bch2_fs_read_write_late(c)) {
822 err = "error going read write";
1c6fdbd8 823 goto err;
134915f3 824 }
1c6fdbd8
KO		 825 }
826
827 set_bit(BCH_FS_STARTED, &c->flags);
828
829 err = NULL;
830out:
831 mutex_unlock(&c->state_lock);
832 return err;
833err:
834 switch (ret) {
835 case BCH_FSCK_ERRORS_NOT_FIXED:
836 bch_err(c, "filesystem contains errors: please report this to the developers");
837 pr_cont("mount with -o fix_errors to repair\n");
838 err = "fsck error";
839 break;
840 case BCH_FSCK_REPAIR_UNIMPLEMENTED:
841 bch_err(c, "filesystem contains errors: please report this to the developers");
842 pr_cont("repair unimplemented: inform the developers so that it can be added\n");
843 err = "fsck error";
844 break;
845 case BCH_FSCK_REPAIR_IMPOSSIBLE:
846 bch_err(c, "filesystem contains errors, but repair impossible");
847 err = "fsck error";
848 break;
849 case BCH_FSCK_UNKNOWN_VERSION:
850 err = "unknown metadata version";;
851 break;
852 case -ENOMEM:
853 err = "cannot allocate memory";
854 break;
855 case -EIO:
856 err = "IO error";
857 break;
858 }
859
860 BUG_ON(!err);
1c6fdbd8
KO		 861 goto out;
862}
863
864static const char *bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
865{
866 struct bch_sb_field_members *sb_mi;
867
868 sb_mi = bch2_sb_get_members(sb);
869 if (!sb_mi)
870 return "Invalid superblock: member info area missing";
871
872 if (le16_to_cpu(sb->block_size) != c->opts.block_size)
873 return "mismatched block size";
874
875 if (le16_to_cpu(sb_mi->members[sb->dev_idx].bucket_size) <
876 BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
877 return "new cache bucket size is too small";
878
879 return NULL;
880}
881
882static const char *bch2_dev_in_fs(struct bch_sb *fs, struct bch_sb *sb)
883{
884 struct bch_sb *newest =
885 le64_to_cpu(fs->seq) > le64_to_cpu(sb->seq) ? fs : sb;
886 struct bch_sb_field_members *mi = bch2_sb_get_members(newest);
887
888 if (!uuid_equal(&fs->uuid, &sb->uuid))
889 return "device not a member of filesystem";
890
891 if (!bch2_dev_exists(newest, mi, sb->dev_idx))
892 return "device has been removed";
893
894 if (fs->block_size != sb->block_size)
895 return "mismatched block size";
896
897 return NULL;
898}
899
900/* Device startup/shutdown: */
901
902static void bch2_dev_release(struct kobject *kobj)
903{
904 struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
905
906 kfree(ca);
907}
908
909static void bch2_dev_free(struct bch_dev *ca)
910{
911 cancel_work_sync(&ca->io_error_work);
912
913 if (ca->kobj.state_in_sysfs &&
914 ca->disk_sb.bdev)
915 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
916
917 if (ca->kobj.state_in_sysfs)
918 kobject_del(&ca->kobj);
919
920 bch2_free_super(&ca->disk_sb);
921 bch2_dev_journal_exit(ca);
922
923 free_percpu(ca->io_done);
924 bioset_exit(&ca->replica_set);
925 bch2_dev_buckets_free(ca);
d1170ce5 926 free_page((unsigned long) ca->sb_read_scratch);
1c6fdbd8
KO		 927
928 bch2_time_stats_exit(&ca->io_latency[WRITE]);
929 bch2_time_stats_exit(&ca->io_latency[READ]);
930
931 percpu_ref_exit(&ca->io_ref);
932 percpu_ref_exit(&ca->ref);
933 kobject_put(&ca->kobj);
934}
935
936static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
937{
938
939 lockdep_assert_held(&c->state_lock);
940
941 if (percpu_ref_is_zero(&ca->io_ref))
942 return;
943
944 __bch2_dev_read_only(c, ca);
945
946 reinit_completion(&ca->io_ref_completion);
947 percpu_ref_kill(&ca->io_ref);
948 wait_for_completion(&ca->io_ref_completion);
949
950 if (ca->kobj.state_in_sysfs) {
951 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
952 sysfs_remove_link(&ca->kobj, "block");
953 }
954
955 bch2_free_super(&ca->disk_sb);
956 bch2_dev_journal_exit(ca);
957}
958
959static void bch2_dev_ref_complete(struct percpu_ref *ref)
960{
961 struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
962
963 complete(&ca->ref_completion);
964}
965
966static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
967{
968 struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
969
970 complete(&ca->io_ref_completion);
971}
972
973static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
974{
975 int ret;
976
977 if (!c->kobj.state_in_sysfs)
978 return 0;
979
980 if (!ca->kobj.state_in_sysfs) {
981 ret = kobject_add(&ca->kobj, &c->kobj,
982 "dev-%u", ca->dev_idx);
983 if (ret)
984 return ret;
985 }
986
987 if (ca->disk_sb.bdev) {
988 struct kobject *block = bdev_kobj(ca->disk_sb.bdev);
989
990 ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
991 if (ret)
992 return ret;
993
994 ret = sysfs_create_link(&ca->kobj, block, "block");
995 if (ret)
996 return ret;
997 }
998
999 return 0;
1000}
1001
1002static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
1003 struct bch_member *member)
1004{
1005 struct bch_dev *ca;
1006
1007 ca = kzalloc(sizeof(*ca), GFP_KERNEL);
1008 if (!ca)
1009 return NULL;
1010
1011 kobject_init(&ca->kobj, &bch2_dev_ktype);
1012 init_completion(&ca->ref_completion);
1013 init_completion(&ca->io_ref_completion);
1014
1015 init_rwsem(&ca->bucket_lock);
1016
1017 writepoint_init(&ca->copygc_write_point, BCH_DATA_USER);
1018
1019 spin_lock_init(&ca->freelist_lock);
1020 bch2_dev_copygc_init(ca);
1021
1022 INIT_WORK(&ca->io_error_work, bch2_io_error_work);
1023
1024 bch2_time_stats_init(&ca->io_latency[READ]);
1025 bch2_time_stats_init(&ca->io_latency[WRITE]);
1026
1027 ca->mi = bch2_mi_to_cpu(member);
1028 ca->uuid = member->uuid;
1029
1030 if (opt_defined(c->opts, discard))
1031 ca->mi.discard = opt_get(c->opts, discard);
1032
1033 if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
1034 0, GFP_KERNEL) ||
1035 percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
1036 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
d1170ce5 1037 !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) ||
1c6fdbd8
KO		 1038 bch2_dev_buckets_alloc(c, ca) ||
1039 bioset_init(&ca->replica_set, 4,
1040 offsetof(struct bch_write_bio, bio), 0) ||
1041 !(ca->io_done = alloc_percpu(*ca->io_done)))
1042 goto err;
1043
1044 return ca;
1045err:
1046 bch2_dev_free(ca);
1047 return NULL;
1048}
1049
1050static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
1051 unsigned dev_idx)
1052{
1053 ca->dev_idx = dev_idx;
1054 __set_bit(ca->dev_idx, ca->self.d);
1055 scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
1056
1057 ca->fs = c;
1058 rcu_assign_pointer(c->devs[ca->dev_idx], ca);
1059
1060 if (bch2_dev_sysfs_online(c, ca))
1061 pr_warn("error creating sysfs objects");
1062}
1063
1064static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
1065{
1066 struct bch_member *member =
1067 bch2_sb_get_members(c->disk_sb.sb)->members + dev_idx;
1068 struct bch_dev *ca = NULL;
1069 int ret = 0;
1070
1071 pr_verbose_init(c->opts, "");
1072
1073 if (bch2_fs_init_fault("dev_alloc"))
1074 goto err;
1075
1076 ca = __bch2_dev_alloc(c, member);
1077 if (!ca)
1078 goto err;
1079
1080 bch2_dev_attach(c, ca, dev_idx);
1081out:
1082 pr_verbose_init(c->opts, "ret %i", ret);
1083 return ret;
1084err:
1085 if (ca)
1086 bch2_dev_free(ca);
1087 ret = -ENOMEM;
1088 goto out;
1089}
1090
1091static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1092{
1093 unsigned ret;
1094
1095 if (bch2_dev_is_online(ca)) {
1096 bch_err(ca, "already have device online in slot %u",
1097 sb->sb->dev_idx);
1098 return -EINVAL;
1099 }
1100
1101 if (get_capacity(sb->bdev->bd_disk) <
1102 ca->mi.bucket_size * ca->mi.nbuckets) {
1103 bch_err(ca, "cannot online: device too small");
1104 return -EINVAL;
1105 }
1106
1107 BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1108
1109 if (get_capacity(sb->bdev->bd_disk) <
1110 ca->mi.bucket_size * ca->mi.nbuckets) {
1111 bch_err(ca, "device too small");
1112 return -EINVAL;
1113 }
1114
1115 ret = bch2_dev_journal_init(ca, sb->sb);
1116 if (ret)
1117 return ret;
1118
1119 /* Commit: */
1120 ca->disk_sb = *sb;
1121 memset(sb, 0, sizeof(*sb));
1122
1c6fdbd8
KO		 1123 percpu_ref_reinit(&ca->io_ref);
1124
1125 return 0;
1126}
1127
1128static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1129{
1130 struct bch_dev *ca;
1131 int ret;
1132
1133 lockdep_assert_held(&c->state_lock);
1134
1135 if (le64_to_cpu(sb->sb->seq) >
1136 le64_to_cpu(c->disk_sb.sb->seq))
1137 bch2_sb_to_fs(c, sb->sb);
1138
1139 BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
1140 !c->devs[sb->sb->dev_idx]);
1141
1142 ca = bch_dev_locked(c, sb->sb->dev_idx);
1143
1144 ret = __bch2_dev_attach_bdev(ca, sb);
1145 if (ret)
1146 return ret;
1147
3e0745e2
KO		 1148 if (test_bit(BCH_FS_ALLOC_READ_DONE, &c->flags) &&
1149 !percpu_u64_get(&ca->usage[0]->buckets[BCH_DATA_SB])) {
1150 mutex_lock(&c->sb_lock);
1151 bch2_mark_dev_superblock(ca->fs, ca, 0);
1152 mutex_unlock(&c->sb_lock);
1153 }
6eac2c2e 1154
1c6fdbd8
KO		 1155 bch2_dev_sysfs_online(c, ca);
1156
1157 if (c->sb.nr_devices == 1)
1158 snprintf(c->name, sizeof(c->name), "%pg", ca->disk_sb.bdev);
1159 snprintf(ca->name, sizeof(ca->name), "%pg", ca->disk_sb.bdev);
1160
1161 rebalance_wakeup(c);
1162 return 0;
1163}
1164
1165/* Device management: */
1166
1167/*
1168 * Note: this function is also used by the error paths - when a particular
1169 * device sees an error, we call it to determine whether we can just set the
1170 * device RO, or - if this function returns false - we'll set the whole
1171 * filesystem RO:
1172 *
1173 * XXX: maybe we should be more explicit about whether we're changing state
1174 * because we got an error or what have you?
1175 */
1176bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1177 enum bch_member_state new_state, int flags)
1178{
1179 struct bch_devs_mask new_online_devs;
1180 struct replicas_status s;
1181 struct bch_dev *ca2;
1182 int i, nr_rw = 0, required;
1183
1184 lockdep_assert_held(&c->state_lock);
1185
1186 switch (new_state) {
1187 case BCH_MEMBER_STATE_RW:
1188 return true;
1189 case BCH_MEMBER_STATE_RO:
1190 if (ca->mi.state != BCH_MEMBER_STATE_RW)
1191 return true;
1192
1193 /* do we have enough devices to write to? */
1194 for_each_member_device(ca2, c, i)
1195 if (ca2 != ca)
1196 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_RW;
1197
1198 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1199 ? c->opts.metadata_replicas
1200 : c->opts.metadata_replicas_required,
1201 !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1202 ? c->opts.data_replicas
1203 : c->opts.data_replicas_required);
1204
1205 return nr_rw >= required;
1206 case BCH_MEMBER_STATE_FAILED:
1207 case BCH_MEMBER_STATE_SPARE:
1208 if (ca->mi.state != BCH_MEMBER_STATE_RW &&
1209 ca->mi.state != BCH_MEMBER_STATE_RO)
1210 return true;
1211
1212 /* do we have enough devices to read from? */
1213 new_online_devs = bch2_online_devs(c);
1214 __clear_bit(ca->dev_idx, new_online_devs.d);
1215
1216 s = __bch2_replicas_status(c, new_online_devs);
1217
1218 return bch2_have_enough_devs(s, flags);
1219 default:
1220 BUG();
1221 }
1222}
1223
1224static bool bch2_fs_may_start(struct bch_fs *c)
1225{
1226 struct replicas_status s;
1227 struct bch_sb_field_members *mi;
1228 struct bch_dev *ca;
1229 unsigned i, flags = c->opts.degraded
1230 ? BCH_FORCE_IF_DEGRADED
1231 : 0;
1232
1233 if (!c->opts.degraded) {
1234 mutex_lock(&c->sb_lock);
1235 mi = bch2_sb_get_members(c->disk_sb.sb);
1236
1237 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1238 if (!bch2_dev_exists(c->disk_sb.sb, mi, i))
1239 continue;
1240
1241 ca = bch_dev_locked(c, i);
1242
1243 if (!bch2_dev_is_online(ca) &&
1244 (ca->mi.state == BCH_MEMBER_STATE_RW ||
1245 ca->mi.state == BCH_MEMBER_STATE_RO)) {
1246 mutex_unlock(&c->sb_lock);
1247 return false;
1248 }
1249 }
1250 mutex_unlock(&c->sb_lock);
1251 }
1252
1253 s = bch2_replicas_status(c);
1254
1255 return bch2_have_enough_devs(s, flags);
1256}
1257
1258static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1259{
1260 bch2_copygc_stop(ca);
1261
1262 /*
1263 * The allocator thread itself allocates btree nodes, so stop it first:
1264 */
1265 bch2_dev_allocator_stop(ca);
1266 bch2_dev_allocator_remove(c, ca);
1267 bch2_dev_journal_stop(&c->journal, ca);
1268}
1269
1270static const char *__bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1271{
1272 lockdep_assert_held(&c->state_lock);
1273
1274 BUG_ON(ca->mi.state != BCH_MEMBER_STATE_RW);
1275
1276 bch2_dev_allocator_add(c, ca);
1277 bch2_recalc_capacity(c);
1278
1279 if (bch2_dev_allocator_start(ca))
1280 return "error starting allocator thread";
1281
1282 if (bch2_copygc_start(c, ca))
1283 return "error starting copygc thread";
1284
1285 return NULL;
1286}
1287
1288int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1289 enum bch_member_state new_state, int flags)
1290{
1291 struct bch_sb_field_members *mi;
1292 int ret = 0;
1293
1294 if (ca->mi.state == new_state)
1295 return 0;
1296
1297 if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1298 return -EINVAL;
1299
1300 if (new_state != BCH_MEMBER_STATE_RW)
1301 __bch2_dev_read_only(c, ca);
1302
1303 bch_notice(ca, "%s", bch2_dev_state[new_state]);
1304
1305 mutex_lock(&c->sb_lock);
1306 mi = bch2_sb_get_members(c->disk_sb.sb);
1307 SET_BCH_MEMBER_STATE(&mi->members[ca->dev_idx], new_state);
1308 bch2_write_super(c);
1309 mutex_unlock(&c->sb_lock);
1310
1311 if (new_state == BCH_MEMBER_STATE_RW &&
1312 __bch2_dev_read_write(c, ca))
1313 ret = -ENOMEM;
1314
1315 rebalance_wakeup(c);
1316
1317 return ret;
1318}
1319
1320int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1321 enum bch_member_state new_state, int flags)
1322{
1323 int ret;
1324
1325 mutex_lock(&c->state_lock);
1326 ret = __bch2_dev_set_state(c, ca, new_state, flags);
1327 mutex_unlock(&c->state_lock);
1328
1329 return ret;
1330}
1331
1332/* Device add/removal: */
1333
1334int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1335{
1336 struct bch_sb_field_members *mi;
1337 unsigned dev_idx = ca->dev_idx, data;
1338 int ret = -EINVAL;
1339
1340 mutex_lock(&c->state_lock);
1341
1342 percpu_ref_put(&ca->ref); /* XXX */
1343
1344 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_FAILED, flags)) {
1345 bch_err(ca, "Cannot remove without losing data");
1346 goto err;
1347 }
1348
1349 __bch2_dev_read_only(c, ca);
1350
1351 /*
1352 * XXX: verify that dev_idx is really not in use anymore, anywhere
1353 *
1354 * flag_data_bad() does not check btree pointers
1355 */
1356 ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1357 if (ret) {
1358 bch_err(ca, "Remove failed: error %i dropping data", ret);
1359 goto err;
1360 }
1361
1362 ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1363 if (ret) {
1364 bch_err(ca, "Remove failed: error %i flushing journal", ret);
1365 goto err;
1366 }
1367
1368 data = bch2_dev_has_data(c, ca);
1369 if (data) {
1370 char data_has_str[100];
319f9ac3 1371
dbaee468
KO		 1372 bch2_flags_to_text(&PBUF(data_has_str),
1373 bch2_data_types, data);
1c6fdbd8
KO		 1374 bch_err(ca, "Remove failed, still has data (%s)", data_has_str);
1375 ret = -EBUSY;
1376 goto err;
1377 }
1378
1379 ret = bch2_btree_delete_range(c, BTREE_ID_ALLOC,
1380 POS(ca->dev_idx, 0),
1381 POS(ca->dev_idx + 1, 0),
fc3268c1 1382 NULL);
1c6fdbd8
KO		 1383 if (ret) {
1384 bch_err(ca, "Remove failed, error deleting alloc info");
1385 goto err;
1386 }
1387
1388 /*
1389 * must flush all existing journal entries, they might have
1390 * (overwritten) keys that point to the device we're removing:
1391 */
1392 bch2_journal_flush_all_pins(&c->journal);
1393 ret = bch2_journal_error(&c->journal);
1394 if (ret) {
1395 bch_err(ca, "Remove failed, journal error");
1396 goto err;
1397 }
1398
1399 __bch2_dev_offline(c, ca);
1400
1401 mutex_lock(&c->sb_lock);
1402 rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1403 mutex_unlock(&c->sb_lock);
1404
1405 percpu_ref_kill(&ca->ref);
1406 wait_for_completion(&ca->ref_completion);
1407
1408 bch2_dev_free(ca);
1409
1410 /*
1411 * Free this device's slot in the bch_member array - all pointers to
1412 * this device must be gone:
1413 */
1414 mutex_lock(&c->sb_lock);
1415 mi = bch2_sb_get_members(c->disk_sb.sb);
1416 memset(&mi->members[dev_idx].uuid, 0, sizeof(mi->members[dev_idx].uuid));
1417
1418 bch2_write_super(c);
1419
1420 mutex_unlock(&c->sb_lock);
1421 mutex_unlock(&c->state_lock);
1422 return 0;
1423err:
d3bb629d
KO		 1424 if (ca->mi.state == BCH_MEMBER_STATE_RW &&
1425 !percpu_ref_is_zero(&ca->io_ref))
1c6fdbd8
KO		 1426 __bch2_dev_read_write(c, ca);
1427 mutex_unlock(&c->state_lock);
1428 return ret;
1429}
1430
6eac2c2e
KO		 1431static void dev_usage_clear(struct bch_dev *ca)
1432{
1433 struct bucket_array *buckets;
1434 int cpu;
1435
1436 for_each_possible_cpu(cpu) {
1437 struct bch_dev_usage *p =
9ca53b55 1438 per_cpu_ptr(ca->usage[0], cpu);
6eac2c2e
KO		 1439 memset(p, 0, sizeof(*p));
1440 }
1441
1442 down_read(&ca->bucket_lock);
1443 buckets = bucket_array(ca);
1444
1445 memset(buckets->b, 0, sizeof(buckets->b[0]) * buckets->nbuckets);
1446 up_read(&ca->bucket_lock);
1447}
1448
1c6fdbd8
KO		 1449/* Add new device to running filesystem: */
1450int bch2_dev_add(struct bch_fs *c, const char *path)
1451{
1452 struct bch_opts opts = bch2_opts_empty();
1453 struct bch_sb_handle sb;
1454 const char *err;
1455 struct bch_dev *ca = NULL;
1456 struct bch_sb_field_members *mi;
1457 struct bch_member dev_mi;
1458 unsigned dev_idx, nr_devices, u64s;
1459 int ret;
1460
1461 ret = bch2_read_super(path, &opts, &sb);
1462 if (ret)
1463 return ret;
1464
1465 err = bch2_sb_validate(&sb);
1466 if (err)
1467 return -EINVAL;
1468
1469 dev_mi = bch2_sb_get_members(sb.sb)->members[sb.sb->dev_idx];
1470
1471 err = bch2_dev_may_add(sb.sb, c);
1472 if (err)
1473 return -EINVAL;
1474
1475 ca = __bch2_dev_alloc(c, &dev_mi);
1476 if (!ca) {
1477 bch2_free_super(&sb);
1478 return -ENOMEM;
1479 }
1480
1481 ret = __bch2_dev_attach_bdev(ca, &sb);
1482 if (ret) {
1483 bch2_dev_free(ca);
1484 return ret;
1485 }
1486
6eac2c2e
KO		 1487 /*
1488 * We want to allocate journal on the new device before adding the new
1489 * device to the filesystem because allocating after we attach requires
1490 * spinning up the allocator thread, and the allocator thread requires
1491 * doing btree writes, which if the existing devices are RO isn't going
1492 * to work
1493 *
1494 * So we have to mark where the superblocks are, but marking allocated
1495 * data normally updates the filesystem usage too, so we have to mark,
1496 * allocate the journal, reset all the marks, then remark after we
1497 * attach...
1498 */
9ca53b55 1499 bch2_mark_dev_superblock(ca->fs, ca, 0);
6eac2c2e 1500
1c6fdbd8
KO		 1501 err = "journal alloc failed";
1502 ret = bch2_dev_journal_alloc(ca);
1503 if (ret)
1504 goto err;
1505
6eac2c2e
KO		 1506 dev_usage_clear(ca);
1507
1c6fdbd8
KO		 1508 mutex_lock(&c->state_lock);
1509 mutex_lock(&c->sb_lock);
1510
1511 err = "insufficient space in new superblock";
1512 ret = bch2_sb_from_fs(c, ca);
1513 if (ret)
1514 goto err_unlock;
1515
1516 mi = bch2_sb_get_members(ca->disk_sb.sb);
1517
1518 if (!bch2_sb_resize_members(&ca->disk_sb,
1519 le32_to_cpu(mi->field.u64s) +
1520 sizeof(dev_mi) / sizeof(u64))) {
1521 ret = -ENOSPC;
1522 goto err_unlock;
1523 }
1524
1525 if (dynamic_fault("bcachefs:add:no_slot"))
1526 goto no_slot;
1527
1528 mi = bch2_sb_get_members(c->disk_sb.sb);
1529 for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++)
1530 if (!bch2_dev_exists(c->disk_sb.sb, mi, dev_idx))
1531 goto have_slot;
1532no_slot:
1533 err = "no slots available in superblock";
1534 ret = -ENOSPC;
1535 goto err_unlock;
1536
1537have_slot:
1538 nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1539 u64s = (sizeof(struct bch_sb_field_members) +
1540 sizeof(struct bch_member) * nr_devices) / sizeof(u64);
1541
1542 err = "no space in superblock for member info";
1543 ret = -ENOSPC;
1544
1545 mi = bch2_sb_resize_members(&c->disk_sb, u64s);
1546 if (!mi)
1547 goto err_unlock;
1548
1549 /* success: */
1550
1551 mi->members[dev_idx] = dev_mi;
a420eea6 1552 mi->members[dev_idx].last_mount = cpu_to_le64(ktime_get_real_seconds());
1c6fdbd8
KO		 1553 c->disk_sb.sb->nr_devices = nr_devices;
1554
1555 ca->disk_sb.sb->dev_idx = dev_idx;
1556 bch2_dev_attach(c, ca, dev_idx);
1557
9ca53b55 1558 bch2_mark_dev_superblock(c, ca, 0);
6eac2c2e 1559
1c6fdbd8
KO		 1560 bch2_write_super(c);
1561 mutex_unlock(&c->sb_lock);
1562
1563 if (ca->mi.state == BCH_MEMBER_STATE_RW) {
1564 err = __bch2_dev_read_write(c, ca);
1565 if (err)
1566 goto err_late;
1567 }
1568
1569 mutex_unlock(&c->state_lock);
1570 return 0;
1571
1572err_unlock:
1573 mutex_unlock(&c->sb_lock);
1574 mutex_unlock(&c->state_lock);
1575err:
1576 if (ca)
1577 bch2_dev_free(ca);
1578 bch2_free_super(&sb);
1579 bch_err(c, "Unable to add device: %s", err);
1580 return ret;
1581err_late:
1582 bch_err(c, "Error going rw after adding device: %s", err);
1583 return -EINVAL;
1584}
1585
1586/* Hot add existing device to running filesystem: */
1587int bch2_dev_online(struct bch_fs *c, const char *path)
1588{
1589 struct bch_opts opts = bch2_opts_empty();
1590 struct bch_sb_handle sb = { NULL };
1591 struct bch_sb_field_members *mi;
1592 struct bch_dev *ca;
1593 unsigned dev_idx;
1594 const char *err;
1595 int ret;
1596
1597 mutex_lock(&c->state_lock);
1598
1599 ret = bch2_read_super(path, &opts, &sb);
1600 if (ret) {
1601 mutex_unlock(&c->state_lock);
1602 return ret;
1603 }
1604
1605 dev_idx = sb.sb->dev_idx;
1606
1607 err = bch2_dev_in_fs(c->disk_sb.sb, sb.sb);
1608 if (err)
1609 goto err;
1610
1611 if (bch2_dev_attach_bdev(c, &sb)) {
1612 err = "bch2_dev_attach_bdev() error";
1613 goto err;
1614 }
1615
1616 ca = bch_dev_locked(c, dev_idx);
1617 if (ca->mi.state == BCH_MEMBER_STATE_RW) {
1618 err = __bch2_dev_read_write(c, ca);
1619 if (err)
1620 goto err;
1621 }
1622
1623 mutex_lock(&c->sb_lock);
1624 mi = bch2_sb_get_members(c->disk_sb.sb);
1625
1626 mi->members[ca->dev_idx].last_mount =
a420eea6 1627 cpu_to_le64(ktime_get_real_seconds());
1c6fdbd8
KO		 1628
1629 bch2_write_super(c);
1630 mutex_unlock(&c->sb_lock);
1631
1632 mutex_unlock(&c->state_lock);
1633 return 0;
1634err:
1635 mutex_unlock(&c->state_lock);
1636 bch2_free_super(&sb);
1637 bch_err(c, "error bringing %s online: %s", path, err);
1638 return -EINVAL;
1639}
1640
1641int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1642{
1643 mutex_lock(&c->state_lock);
1644
1645 if (!bch2_dev_is_online(ca)) {
1646 bch_err(ca, "Already offline");
1647 mutex_unlock(&c->state_lock);
1648 return 0;
1649 }
1650
1651 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_FAILED, flags)) {
1652 bch_err(ca, "Cannot offline required disk");
1653 mutex_unlock(&c->state_lock);
1654 return -EINVAL;
1655 }
1656
1657 __bch2_dev_offline(c, ca);
1658
1659 mutex_unlock(&c->state_lock);
1660 return 0;
1661}
1662
1663int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1664{
1665 struct bch_member *mi;
1666 int ret = 0;
1667
1668 mutex_lock(&c->state_lock);
1669
1670 if (nbuckets < ca->mi.nbuckets) {
1671 bch_err(ca, "Cannot shrink yet");
1672 ret = -EINVAL;
1673 goto err;
1674 }
1675
1676 if (bch2_dev_is_online(ca) &&
1677 get_capacity(ca->disk_sb.bdev->bd_disk) <
1678 ca->mi.bucket_size * nbuckets) {
1679 bch_err(ca, "New size larger than device");
1680 ret = -EINVAL;
1681 goto err;
1682 }
1683
1684 ret = bch2_dev_buckets_resize(c, ca, nbuckets);
1685 if (ret) {
1686 bch_err(ca, "Resize error: %i", ret);
1687 goto err;
1688 }
1689
1690 mutex_lock(&c->sb_lock);
1691 mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
1692 mi->nbuckets = cpu_to_le64(nbuckets);
1693
1694 bch2_write_super(c);
1695 mutex_unlock(&c->sb_lock);
1696
1697 bch2_recalc_capacity(c);
1698err:
1699 mutex_unlock(&c->state_lock);
1700 return ret;
1701}
1702
1703/* return with ref on ca->ref: */
1704struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *path)
1705{
1706
1707 struct bch_dev *ca;
1708 dev_t dev;
1709 unsigned i;
1710 int ret;
1711
1712 ret = lookup_bdev(path, &dev);
1713 if (ret)
1714 return ERR_PTR(ret);
1715
1716 for_each_member_device(ca, c, i)
1717 if (ca->disk_sb.bdev->bd_dev == dev)
1718 goto found;
1719
1720 ca = ERR_PTR(-ENOENT);
1721found:
1722 return ca;
1723}
1724
1725/* Filesystem open: */
1726
1727struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
1728 struct bch_opts opts)
1729{
1730 struct bch_sb_handle *sb = NULL;
1731 struct bch_fs *c = NULL;
1732 unsigned i, best_sb = 0;
1733 const char *err;
1734 int ret = -ENOMEM;
1735
1736 pr_verbose_init(opts, "");
1737
1738 if (!nr_devices) {
1739 c = ERR_PTR(-EINVAL);
1740 goto out2;
1741 }
1742
1743 if (!try_module_get(THIS_MODULE)) {
1744 c = ERR_PTR(-ENODEV);
1745 goto out2;
1746 }
1747
1748 sb = kcalloc(nr_devices, sizeof(*sb), GFP_KERNEL);
1749 if (!sb)
1750 goto err;
1751
1752 for (i = 0; i < nr_devices; i++) {
1753 ret = bch2_read_super(devices[i], &opts, &sb[i]);
1754 if (ret)
1755 goto err;
1756
1757 err = bch2_sb_validate(&sb[i]);
1758 if (err)
1759 goto err_print;
1760 }
1761
1762 for (i = 1; i < nr_devices; i++)
1763 if (le64_to_cpu(sb[i].sb->seq) >
1764 le64_to_cpu(sb[best_sb].sb->seq))
1765 best_sb = i;
1766
1767 for (i = 0; i < nr_devices; i++) {
1768 err = bch2_dev_in_fs(sb[best_sb].sb, sb[i].sb);
1769 if (err)
1770 goto err_print;
1771 }
1772
1773 ret = -ENOMEM;
1774 c = bch2_fs_alloc(sb[best_sb].sb, opts);
1775 if (!c)
1776 goto err;
1777
1778 err = "bch2_dev_online() error";
1779 mutex_lock(&c->state_lock);
1780 for (i = 0; i < nr_devices; i++)
1781 if (bch2_dev_attach_bdev(c, &sb[i])) {
1782 mutex_unlock(&c->state_lock);
1783 goto err_print;
1784 }
1785 mutex_unlock(&c->state_lock);
1786
1787 err = "insufficient devices";
1788 if (!bch2_fs_may_start(c))
1789 goto err_print;
1790
1791 if (!c->opts.nostart) {
1792 err = bch2_fs_start(c);
1793 if (err)
1794 goto err_print;
1795 }
1796out:
1797 kfree(sb);
1798 module_put(THIS_MODULE);
1799out2:
1800 pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
1801 return c;
1802err_print:
1803 pr_err("bch_fs_open err opening %s: %s",
1804 devices[0], err);
1805 ret = -EINVAL;
1806err:
1807 if (c)
1808 bch2_fs_stop(c);
1809 for (i = 0; i < nr_devices; i++)
1810 bch2_free_super(&sb[i]);
1811 c = ERR_PTR(ret);
1812 goto out;
1813}
1814
1815static const char *__bch2_fs_open_incremental(struct bch_sb_handle *sb,
1816 struct bch_opts opts)
1817{
1818 const char *err;
1819 struct bch_fs *c;
1820 bool allocated_fs = false;
1821
1822 err = bch2_sb_validate(sb);
1823 if (err)
1824 return err;
1825
1826 mutex_lock(&bch_fs_list_lock);
1827 c = __bch2_uuid_to_fs(sb->sb->uuid);
1828 if (c) {
1829 closure_get(&c->cl);
1830
1831 err = bch2_dev_in_fs(c->disk_sb.sb, sb->sb);
1832 if (err)
1833 goto err;
1834 } else {
1835 c = bch2_fs_alloc(sb->sb, opts);
1836 err = "cannot allocate memory";
1837 if (!c)
1838 goto err;
1839
1840 allocated_fs = true;
1841 }
1842
1843 err = "bch2_dev_online() error";
1844
1845 mutex_lock(&c->sb_lock);
1846 if (bch2_dev_attach_bdev(c, sb)) {
1847 mutex_unlock(&c->sb_lock);
1848 goto err;
1849 }
1850 mutex_unlock(&c->sb_lock);
1851
1852 if (!c->opts.nostart && bch2_fs_may_start(c)) {
1853 err = bch2_fs_start(c);
1854 if (err)
1855 goto err;
1856 }
1857
1858 closure_put(&c->cl);
1859 mutex_unlock(&bch_fs_list_lock);
1860
1861 return NULL;
1862err:
1863 mutex_unlock(&bch_fs_list_lock);
1864
1865 if (allocated_fs)
1866 bch2_fs_stop(c);
1867 else if (c)
1868 closure_put(&c->cl);
1869
1870 return err;
1871}
1872
1873const char *bch2_fs_open_incremental(const char *path)
1874{
1875 struct bch_sb_handle sb;
1876 struct bch_opts opts = bch2_opts_empty();
1877 const char *err;
1878
1879 if (bch2_read_super(path, &opts, &sb))
1880 return "error reading superblock";
1881
1882 err = __bch2_fs_open_incremental(&sb, opts);
1883 bch2_free_super(&sb);
1884
1885 return err;
1886}
1887
1888/* Global interfaces/init */
1889
1890static void bcachefs_exit(void)
1891{
1892 bch2_debug_exit();
1893 bch2_vfs_exit();
1894 bch2_chardev_exit();
1895 if (bcachefs_kset)
1896 kset_unregister(bcachefs_kset);
1897}
1898
1899static int __init bcachefs_init(void)
1900{
1901 bch2_bkey_pack_test();
1902 bch2_inode_pack_test();
1903
1904 if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
1905 bch2_chardev_init() ||
1906 bch2_vfs_init() ||
1907 bch2_debug_init())
1908 goto err;
1909
1910 return 0;
1911err:
1912 bcachefs_exit();
1913 return -ENOMEM;
1914}
1915
1916#define BCH_DEBUG_PARAM(name, description) \
1917 bool bch2_##name; \
1918 module_param_named(name, bch2_##name, bool, 0644); \
1919 MODULE_PARM_DESC(name, description);
1920BCH_DEBUG_PARAMS()
1921#undef BCH_DEBUG_PARAM
1922
26609b61 1923unsigned bch2_metadata_version = bcachefs_metadata_version_current;
1c6fdbd8
KO		 1924module_param_named(version, bch2_metadata_version, uint, 0400);
1925
1926module_exit(bcachefs_exit);
1927module_init(bcachefs_init);
Linux 6.x block layer and io_uring tree(s)