projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
bcachefs: Scale down number of writepoints when low on space
[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"
1c6fdbd8
KO		 13#include "btree_cache.h"
14#include "btree_gc.h"
15#include "btree_update_interior.h"
16#include "btree_io.h"
17#include "chardev.h"
18#include "checksum.h"
19#include "clock.h"
20#include "compress.h"
21#include "debug.h"
22#include "disk_groups.h"
23#include "error.h"
24#include "fs.h"
25#include "fs-io.h"
26#include "fsck.h"
27#include "inode.h"
28#include "io.h"
29#include "journal.h"
30#include "journal_reclaim.h"
31#include "move.h"
32#include "migrate.h"
33#include "movinggc.h"
34#include "quota.h"
35#include "rebalance.h"
36#include "recovery.h"
37#include "replicas.h"
38#include "super.h"
39#include "super-io.h"
40#include "sysfs.h"
41#include "trace.h"
42
43#include <linux/backing-dev.h>
44#include <linux/blkdev.h>
45#include <linux/debugfs.h>
46#include <linux/device.h>
47#include <linux/idr.h>
48#include <linux/kthread.h>
49#include <linux/module.h>
50#include <linux/percpu.h>
51#include <linux/random.h>
52#include <linux/sysfs.h>
53#include <crypto/hash.h>
54
55MODULE_LICENSE("GPL");
56MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
57
58#define KTYPE(type) \
59static const struct attribute_group type ## _group = { \
60 .attrs = type ## _files \
61}; \
62 \
63static const struct attribute_group *type ## _groups[] = { \
64 &type ## _group, \
65 NULL \
66}; \
67 \
68static const struct kobj_type type ## _ktype = { \
69 .release = type ## _release, \
70 .sysfs_ops = &type ## _sysfs_ops, \
71 .default_groups = type ## _groups \
72}
73
74static void bch2_fs_release(struct kobject *);
75static void bch2_dev_release(struct kobject *);
76
77static void bch2_fs_internal_release(struct kobject *k)
78{
79}
80
81static void bch2_fs_opts_dir_release(struct kobject *k)
82{
83}
84
85static void bch2_fs_time_stats_release(struct kobject *k)
86{
87}
88
89KTYPE(bch2_fs);
90KTYPE(bch2_fs_internal);
91KTYPE(bch2_fs_opts_dir);
92KTYPE(bch2_fs_time_stats);
93KTYPE(bch2_dev);
94
95static struct kset *bcachefs_kset;
96static LIST_HEAD(bch_fs_list);
97static DEFINE_MUTEX(bch_fs_list_lock);
98
99static DECLARE_WAIT_QUEUE_HEAD(bch_read_only_wait);
100
101static void bch2_dev_free(struct bch_dev *);
102static int bch2_dev_alloc(struct bch_fs *, unsigned);
103static int bch2_dev_sysfs_online(struct bch_fs *, struct bch_dev *);
104static void __bch2_dev_read_only(struct bch_fs *, struct bch_dev *);
105
106struct bch_fs *bch2_dev_to_fs(dev_t dev)
107{
108 struct bch_fs *c;
109 struct bch_dev *ca;
110 unsigned i;
111
112 mutex_lock(&bch_fs_list_lock);
113 rcu_read_lock();
114
115 list_for_each_entry(c, &bch_fs_list, list)
116 for_each_member_device_rcu(ca, c, i, NULL)
117 if (ca->disk_sb.bdev->bd_dev == dev) {
118 closure_get(&c->cl);
119 goto found;
120 }
121 c = NULL;
122found:
123 rcu_read_unlock();
124 mutex_unlock(&bch_fs_list_lock);
125
126 return c;
127}
128
129static struct bch_fs *__bch2_uuid_to_fs(__uuid_t uuid)
130{
131 struct bch_fs *c;
132
133 lockdep_assert_held(&bch_fs_list_lock);
134
135 list_for_each_entry(c, &bch_fs_list, list)
136 if (!memcmp(&c->disk_sb.sb->uuid, &uuid, sizeof(uuid)))
137 return c;
138
139 return NULL;
140}
141
142struct bch_fs *bch2_uuid_to_fs(__uuid_t uuid)
143{
144 struct bch_fs *c;
145
146 mutex_lock(&bch_fs_list_lock);
147 c = __bch2_uuid_to_fs(uuid);
148 if (c)
149 closure_get(&c->cl);
150 mutex_unlock(&bch_fs_list_lock);
151
152 return c;
153}
154
155/* Filesystem RO/RW: */
156
157/*
158 * For startup/shutdown of RW stuff, the dependencies are:
159 *
160 * - foreground writes depend on copygc and rebalance (to free up space)
161 *
162 * - copygc and rebalance depend on mark and sweep gc (they actually probably
163 * don't because they either reserve ahead of time or don't block if
164 * allocations fail, but allocations can require mark and sweep gc to run
165 * because of generation number wraparound)
166 *
167 * - all of the above depends on the allocator threads
168 *
169 * - allocator depends on the journal (when it rewrites prios and gens)
170 */
171
172static void __bch2_fs_read_only(struct bch_fs *c)
173{
174 struct bch_dev *ca;
175 unsigned i;
176
177 bch2_rebalance_stop(c);
178
179 for_each_member_device(ca, c, i)
180 bch2_copygc_stop(ca);
181
182 bch2_gc_thread_stop(c);
183
184 /*
185 * Flush journal before stopping allocators, because flushing journal
186 * blacklist entries involves allocating new btree nodes:
187 */
188 bch2_journal_flush_all_pins(&c->journal);
189
190 for_each_member_device(ca, c, i)
191 bch2_dev_allocator_stop(ca);
192
193 bch2_journal_flush_all_pins(&c->journal);
194
195 /*
196 * We need to explicitly wait on btree interior updates to complete
197 * before stopping the journal, flushing all journal pins isn't
198 * sufficient, because in the BTREE_INTERIOR_UPDATING_ROOT case btree
199 * interior updates have to drop their journal pin before they're
200 * fully complete:
201 */
202 closure_wait_event(&c->btree_interior_update_wait,
203 !bch2_btree_interior_updates_nr_pending(c));
204
205 bch2_fs_journal_stop(&c->journal);
206
207 /*
208 * the journal kicks off btree writes via reclaim - wait for in flight
209 * writes after stopping journal:
210 */
211 if (test_bit(BCH_FS_EMERGENCY_RO, &c->flags))
212 bch2_btree_flush_all_writes(c);
213 else
214 bch2_btree_verify_flushed(c);
215
216 /*
217 * After stopping journal:
218 */
219 for_each_member_device(ca, c, i)
220 bch2_dev_allocator_remove(c, ca);
221}
222
223static void bch2_writes_disabled(struct percpu_ref *writes)
224{
225 struct bch_fs *c = container_of(writes, struct bch_fs, writes);
226
227 set_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
228 wake_up(&bch_read_only_wait);
229}
230
231void bch2_fs_read_only(struct bch_fs *c)
232{
233 if (c->state == BCH_FS_RO)
234 return;
235
236 BUG_ON(test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
237
238 /*
239 * Block new foreground-end write operations from starting - any new
240 * writes will return -EROFS:
241 *
242 * (This is really blocking new _allocations_, writes to previously
243 * allocated space can still happen until stopping the allocator in
244 * bch2_dev_allocator_stop()).
245 */
246 percpu_ref_kill(&c->writes);
247
248 cancel_delayed_work(&c->pd_controllers_update);
249
250 /*
251 * If we're not doing an emergency shutdown, we want to wait on
252 * outstanding writes to complete so they don't see spurious errors due
253 * to shutting down the allocator:
254 *
255 * If we are doing an emergency shutdown outstanding writes may
256 * hang until we shutdown the allocator so we don't want to wait
257 * on outstanding writes before shutting everything down - but
258 * we do need to wait on them before returning and signalling
259 * that going RO is complete:
260 */
261 wait_event(bch_read_only_wait,
262 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags) ||
263 test_bit(BCH_FS_EMERGENCY_RO, &c->flags));
264
265 __bch2_fs_read_only(c);
266
267 wait_event(bch_read_only_wait,
268 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
269
270 clear_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
271
272 if (!bch2_journal_error(&c->journal) &&
273 !test_bit(BCH_FS_ERROR, &c->flags) &&
274 !test_bit(BCH_FS_EMERGENCY_RO, &c->flags))
275 bch2_fs_mark_clean(c, true);
276
277 if (c->state != BCH_FS_STOPPING)
278 c->state = BCH_FS_RO;
279}
280
281static void bch2_fs_read_only_work(struct work_struct *work)
282{
283 struct bch_fs *c =
284 container_of(work, struct bch_fs, read_only_work);
285
286 mutex_lock(&c->state_lock);
287 bch2_fs_read_only(c);
288 mutex_unlock(&c->state_lock);
289}
290
291static void bch2_fs_read_only_async(struct bch_fs *c)
292{
293 queue_work(system_long_wq, &c->read_only_work);
294}
295
296bool bch2_fs_emergency_read_only(struct bch_fs *c)
297{
298 bool ret = !test_and_set_bit(BCH_FS_EMERGENCY_RO, &c->flags);
299
300 bch2_fs_read_only_async(c);
301 bch2_journal_halt(&c->journal);
302
303 wake_up(&bch_read_only_wait);
304 return ret;
305}
306
307const char *bch2_fs_read_write(struct bch_fs *c)
308{
309 struct bch_dev *ca;
310 const char *err = NULL;
311 unsigned i;
312
313 if (c->state == BCH_FS_RW)
314 return NULL;
315
316 bch2_fs_mark_clean(c, false);
317
318 for_each_rw_member(ca, c, i)
319 bch2_dev_allocator_add(c, ca);
320 bch2_recalc_capacity(c);
321
322 err = "error starting allocator thread";
323 for_each_rw_member(ca, c, i)
324 if (bch2_dev_allocator_start(ca)) {
325 percpu_ref_put(&ca->io_ref);
326 goto err;
327 }
328
329 err = "error starting btree GC thread";
330 if (bch2_gc_thread_start(c))
331 goto err;
332
333 err = "error starting copygc thread";
334 for_each_rw_member(ca, c, i)
335 if (bch2_copygc_start(c, ca)) {
336 percpu_ref_put(&ca->io_ref);
337 goto err;
338 }
339
340 err = "error starting rebalance thread";
341 if (bch2_rebalance_start(c))
342 goto err;
343
344 schedule_delayed_work(&c->pd_controllers_update, 5 * HZ);
345
346 if (c->state != BCH_FS_STARTING)
347 percpu_ref_reinit(&c->writes);
348
349 c->state = BCH_FS_RW;
350 return NULL;
351err:
352 __bch2_fs_read_only(c);
353 return err;
354}
355
356/* Filesystem startup/shutdown: */
357
358static void bch2_fs_free(struct bch_fs *c)
359{
360 unsigned i;
361
362 for (i = 0; i < BCH_TIME_STAT_NR; i++)
363 bch2_time_stats_exit(&c->times[i]);
364
365 bch2_fs_quota_exit(c);
366 bch2_fs_fsio_exit(c);
367 bch2_fs_encryption_exit(c);
368 bch2_fs_io_exit(c);
369 bch2_fs_btree_cache_exit(c);
370 bch2_fs_journal_exit(&c->journal);
371 bch2_io_clock_exit(&c->io_clock[WRITE]);
372 bch2_io_clock_exit(&c->io_clock[READ]);
373 bch2_fs_compress_exit(c);
374 percpu_free_rwsem(&c->usage_lock);
375 free_percpu(c->usage_percpu);
581edb63 376 mempool_exit(&c->btree_iters_pool);
1c6fdbd8
KO		 377 mempool_exit(&c->btree_bounce_pool);
378 bioset_exit(&c->btree_bio);
379 mempool_exit(&c->btree_interior_update_pool);
380 mempool_exit(&c->btree_reserve_pool);
381 mempool_exit(&c->fill_iter);
382 percpu_ref_exit(&c->writes);
383 kfree(rcu_dereference_protected(c->replicas, 1));
384 kfree(rcu_dereference_protected(c->disk_groups, 1));
385
386 if (c->copygc_wq)
387 destroy_workqueue(c->copygc_wq);
388 if (c->wq)
389 destroy_workqueue(c->wq);
390
391 free_pages((unsigned long) c->disk_sb.sb,
392 c->disk_sb.page_order);
393 kvpfree(c, sizeof(*c));
394 module_put(THIS_MODULE);
395}
396
397static void bch2_fs_release(struct kobject *kobj)
398{
399 struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
400
401 bch2_fs_free(c);
402}
403
404void bch2_fs_stop(struct bch_fs *c)
405{
406 struct bch_dev *ca;
407 unsigned i;
408
af1c6871
KO		 409 bch_verbose(c, "shutting down");
410
1c6fdbd8
KO		 411 for_each_member_device(ca, c, i)
412 if (ca->kobj.state_in_sysfs &&
413 ca->disk_sb.bdev)
414 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
415
416 if (c->kobj.state_in_sysfs)
417 kobject_del(&c->kobj);
418
419 bch2_fs_debug_exit(c);
420 bch2_fs_chardev_exit(c);
421
422 kobject_put(&c->time_stats);
423 kobject_put(&c->opts_dir);
424 kobject_put(&c->internal);
425
426 mutex_lock(&bch_fs_list_lock);
427 list_del(&c->list);
428 mutex_unlock(&bch_fs_list_lock);
429
430 closure_sync(&c->cl);
431 closure_debug_destroy(&c->cl);
432
433 mutex_lock(&c->state_lock);
434 bch2_fs_read_only(c);
435 mutex_unlock(&c->state_lock);
436
437 /* btree prefetch might have kicked off reads in the background: */
438 bch2_btree_flush_all_reads(c);
439
440 for_each_member_device(ca, c, i)
441 cancel_work_sync(&ca->io_error_work);
442
443 cancel_work_sync(&c->btree_write_error_work);
444 cancel_delayed_work_sync(&c->pd_controllers_update);
445 cancel_work_sync(&c->read_only_work);
446
447 for (i = 0; i < c->sb.nr_devices; i++)
448 if (c->devs[i])
449 bch2_dev_free(rcu_dereference_protected(c->devs[i], 1));
450
af1c6871
KO		 451 bch_verbose(c, "shutdown complete");
452
1c6fdbd8
KO		 453 kobject_put(&c->kobj);
454}
455
456static const char *bch2_fs_online(struct bch_fs *c)
457{
458 struct bch_dev *ca;
459 const char *err = NULL;
460 unsigned i;
461 int ret;
462
463 lockdep_assert_held(&bch_fs_list_lock);
464
465 if (!list_empty(&c->list))
466 return NULL;
467
468 if (__bch2_uuid_to_fs(c->sb.uuid))
469 return "filesystem UUID already open";
470
471 ret = bch2_fs_chardev_init(c);
472 if (ret)
473 return "error creating character device";
474
475 bch2_fs_debug_init(c);
476
477 if (kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ||
478 kobject_add(&c->internal, &c->kobj, "internal") ||
479 kobject_add(&c->opts_dir, &c->kobj, "options") ||
480 kobject_add(&c->time_stats, &c->kobj, "time_stats") ||
481 bch2_opts_create_sysfs_files(&c->opts_dir))
482 return "error creating sysfs objects";
483
484 mutex_lock(&c->state_lock);
485
486 err = "error creating sysfs objects";
487 __for_each_member_device(ca, c, i, NULL)
488 if (bch2_dev_sysfs_online(c, ca))
489 goto err;
490
491 list_add(&c->list, &bch_fs_list);
492 err = NULL;
493err:
494 mutex_unlock(&c->state_lock);
495 return err;
496}
497
498static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
499{
500 struct bch_sb_field_members *mi;
501 struct bch_fs *c;
502 unsigned i, iter_size;
503 const char *err;
504
505 pr_verbose_init(opts, "");
506
507 c = kvpmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
508 if (!c)
509 goto out;
510
511 __module_get(THIS_MODULE);
512
513 c->minor = -1;
514 c->disk_sb.fs_sb = true;
515
516 mutex_init(&c->state_lock);
517 mutex_init(&c->sb_lock);
518 mutex_init(&c->replicas_gc_lock);
519 mutex_init(&c->btree_root_lock);
520 INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
521
522 init_rwsem(&c->gc_lock);
523
524 for (i = 0; i < BCH_TIME_STAT_NR; i++)
525 bch2_time_stats_init(&c->times[i]);
526
b092dadd
KO		 527 bch2_fs_allocator_background_init(c);
528 bch2_fs_allocator_foreground_init(c);
1c6fdbd8
KO		 529 bch2_fs_rebalance_init(c);
530 bch2_fs_quota_init(c);
531
532 INIT_LIST_HEAD(&c->list);
533
534 INIT_LIST_HEAD(&c->btree_interior_update_list);
535 mutex_init(&c->btree_reserve_cache_lock);
536 mutex_init(&c->btree_interior_update_lock);
537
538 mutex_init(&c->bio_bounce_pages_lock);
539
540 bio_list_init(&c->btree_write_error_list);
541 spin_lock_init(&c->btree_write_error_lock);
542 INIT_WORK(&c->btree_write_error_work, bch2_btree_write_error_work);
543
544 INIT_LIST_HEAD(&c->fsck_errors);
545 mutex_init(&c->fsck_error_lock);
546
547 seqcount_init(&c->gc_pos_lock);
548
549 c->copy_gc_enabled = 1;
550 c->rebalance.enabled = 1;
551 c->promote_whole_extents = true;
552
553 c->journal.write_time = &c->times[BCH_TIME_journal_write];
554 c->journal.delay_time = &c->times[BCH_TIME_journal_delay];
555 c->journal.blocked_time = &c->times[BCH_TIME_journal_blocked];
556 c->journal.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq];
557
558 bch2_fs_btree_cache_init_early(&c->btree_cache);
559
560 mutex_lock(&c->sb_lock);
561
562 if (bch2_sb_to_fs(c, sb)) {
563 mutex_unlock(&c->sb_lock);
564 goto err;
565 }
566
567 mutex_unlock(&c->sb_lock);
568
569 scnprintf(c->name, sizeof(c->name), "%pU", &c->sb.user_uuid);
570
571 c->opts = bch2_opts_default;
572 bch2_opts_apply(&c->opts, bch2_opts_from_sb(sb));
573 bch2_opts_apply(&c->opts, opts);
574
575 c->block_bits = ilog2(c->opts.block_size);
576 c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
577
578 c->opts.nochanges |= c->opts.noreplay;
579 c->opts.read_only |= c->opts.nochanges;
580
581 if (bch2_fs_init_fault("fs_alloc"))
582 goto err;
583
584 iter_size = sizeof(struct btree_node_iter_large) +
585 (btree_blocks(c) + 1) * 2 *
586 sizeof(struct btree_node_iter_set);
587
588 if (!(c->wq = alloc_workqueue("bcachefs",
589 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_HIGHPRI, 1)) ||
590 !(c->copygc_wq = alloc_workqueue("bcache_copygc",
591 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_HIGHPRI, 1)) ||
592 percpu_ref_init(&c->writes, bch2_writes_disabled, 0, GFP_KERNEL) ||
593 mempool_init_kmalloc_pool(&c->btree_reserve_pool, 1,
594 sizeof(struct btree_reserve)) ||
595 mempool_init_kmalloc_pool(&c->btree_interior_update_pool, 1,
596 sizeof(struct btree_update)) ||
597 mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
598 bioset_init(&c->btree_bio, 1,
599 max(offsetof(struct btree_read_bio, bio),
600 offsetof(struct btree_write_bio, wbio.bio)),
601 BIOSET_NEED_BVECS) ||
602 !(c->usage_percpu = alloc_percpu(struct bch_fs_usage)) ||
603 percpu_init_rwsem(&c->usage_lock) ||
604 mempool_init_kvpmalloc_pool(&c->btree_bounce_pool, 1,
605 btree_bytes(c)) ||
581edb63
KO		 606 mempool_init_kmalloc_pool(&c->btree_iters_pool, 1,
607 sizeof(struct btree_iter) * BTREE_ITER_MAX) ||
1c6fdbd8
KO		 608 bch2_io_clock_init(&c->io_clock[READ]) ||
609 bch2_io_clock_init(&c->io_clock[WRITE]) ||
610 bch2_fs_journal_init(&c->journal) ||
611 bch2_fs_btree_cache_init(c) ||
612 bch2_fs_io_init(c) ||
613 bch2_fs_encryption_init(c) ||
614 bch2_fs_compress_init(c) ||
615 bch2_fs_fsio_init(c))
616 goto err;
617
618 mi = bch2_sb_get_members(c->disk_sb.sb);
619 for (i = 0; i < c->sb.nr_devices; i++)
620 if (bch2_dev_exists(c->disk_sb.sb, mi, i) &&
621 bch2_dev_alloc(c, i))
622 goto err;
623
624 /*
625 * Now that all allocations have succeeded, init various refcounty
626 * things that let us shutdown:
627 */
628 closure_init(&c->cl, NULL);
629
630 c->kobj.kset = bcachefs_kset;
631 kobject_init(&c->kobj, &bch2_fs_ktype);
632 kobject_init(&c->internal, &bch2_fs_internal_ktype);
633 kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
634 kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
635
636 mutex_lock(&bch_fs_list_lock);
637 err = bch2_fs_online(c);
638 mutex_unlock(&bch_fs_list_lock);
639 if (err) {
640 bch_err(c, "bch2_fs_online() error: %s", err);
641 goto err;
642 }
643out:
644 pr_verbose_init(opts, "ret %i", c ? 0 : -ENOMEM);
645 return c;
646err:
647 bch2_fs_free(c);
648 c = NULL;
649 goto out;
650}
651
652const char *bch2_fs_start(struct bch_fs *c)
653{
654 const char *err = "cannot allocate memory";
655 struct bch_sb_field_members *mi;
656 struct bch_dev *ca;
657 time64_t now = ktime_get_seconds();
658 unsigned i;
659 int ret = -EINVAL;
660
661 mutex_lock(&c->state_lock);
662
663 BUG_ON(c->state != BCH_FS_STARTING);
664
665 mutex_lock(&c->sb_lock);
666
667 for_each_online_member(ca, c, i)
668 bch2_sb_from_fs(c, ca);
669
670 mi = bch2_sb_get_members(c->disk_sb.sb);
671 for_each_online_member(ca, c, i)
672 mi->members[ca->dev_idx].last_mount = cpu_to_le64(now);
673
674 mutex_unlock(&c->sb_lock);
675
676 for_each_rw_member(ca, c, i)
677 bch2_dev_allocator_add(c, ca);
678 bch2_recalc_capacity(c);
679
680 ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
681 ? bch2_fs_recovery(c)
682 : bch2_fs_initialize(c);
683 if (ret)
684 goto err;
685
686 err = "dynamic fault";
687 if (bch2_fs_init_fault("fs_start"))
688 goto err;
689
690 if (c->opts.read_only) {
691 bch2_fs_read_only(c);
692 } else {
693 err = bch2_fs_read_write(c);
694 if (err)
695 goto err;
696 }
697
698 set_bit(BCH_FS_STARTED, &c->flags);
699
700 err = NULL;
701out:
702 mutex_unlock(&c->state_lock);
703 return err;
704err:
705 switch (ret) {
706 case BCH_FSCK_ERRORS_NOT_FIXED:
707 bch_err(c, "filesystem contains errors: please report this to the developers");
708 pr_cont("mount with -o fix_errors to repair\n");
709 err = "fsck error";
710 break;
711 case BCH_FSCK_REPAIR_UNIMPLEMENTED:
712 bch_err(c, "filesystem contains errors: please report this to the developers");
713 pr_cont("repair unimplemented: inform the developers so that it can be added\n");
714 err = "fsck error";
715 break;
716 case BCH_FSCK_REPAIR_IMPOSSIBLE:
717 bch_err(c, "filesystem contains errors, but repair impossible");
718 err = "fsck error";
719 break;
720 case BCH_FSCK_UNKNOWN_VERSION:
721 err = "unknown metadata version";;
722 break;
723 case -ENOMEM:
724 err = "cannot allocate memory";
725 break;
726 case -EIO:
727 err = "IO error";
728 break;
729 }
730
731 BUG_ON(!err);
732 set_bit(BCH_FS_ERROR, &c->flags);
733 goto out;
734}
735
736static const char *bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
737{
738 struct bch_sb_field_members *sb_mi;
739
740 sb_mi = bch2_sb_get_members(sb);
741 if (!sb_mi)
742 return "Invalid superblock: member info area missing";
743
744 if (le16_to_cpu(sb->block_size) != c->opts.block_size)
745 return "mismatched block size";
746
747 if (le16_to_cpu(sb_mi->members[sb->dev_idx].bucket_size) <
748 BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
749 return "new cache bucket size is too small";
750
751 return NULL;
752}
753
754static const char *bch2_dev_in_fs(struct bch_sb *fs, struct bch_sb *sb)
755{
756 struct bch_sb *newest =
757 le64_to_cpu(fs->seq) > le64_to_cpu(sb->seq) ? fs : sb;
758 struct bch_sb_field_members *mi = bch2_sb_get_members(newest);
759
760 if (!uuid_equal(&fs->uuid, &sb->uuid))
761 return "device not a member of filesystem";
762
763 if (!bch2_dev_exists(newest, mi, sb->dev_idx))
764 return "device has been removed";
765
766 if (fs->block_size != sb->block_size)
767 return "mismatched block size";
768
769 return NULL;
770}
771
772/* Device startup/shutdown: */
773
774static void bch2_dev_release(struct kobject *kobj)
775{
776 struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
777
778 kfree(ca);
779}
780
781static void bch2_dev_free(struct bch_dev *ca)
782{
783 cancel_work_sync(&ca->io_error_work);
784
785 if (ca->kobj.state_in_sysfs &&
786 ca->disk_sb.bdev)
787 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
788
789 if (ca->kobj.state_in_sysfs)
790 kobject_del(&ca->kobj);
791
792 bch2_free_super(&ca->disk_sb);
793 bch2_dev_journal_exit(ca);
794
795 free_percpu(ca->io_done);
796 bioset_exit(&ca->replica_set);
797 bch2_dev_buckets_free(ca);
798
799 bch2_time_stats_exit(&ca->io_latency[WRITE]);
800 bch2_time_stats_exit(&ca->io_latency[READ]);
801
802 percpu_ref_exit(&ca->io_ref);
803 percpu_ref_exit(&ca->ref);
804 kobject_put(&ca->kobj);
805}
806
807static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
808{
809
810 lockdep_assert_held(&c->state_lock);
811
812 if (percpu_ref_is_zero(&ca->io_ref))
813 return;
814
815 __bch2_dev_read_only(c, ca);
816
817 reinit_completion(&ca->io_ref_completion);
818 percpu_ref_kill(&ca->io_ref);
819 wait_for_completion(&ca->io_ref_completion);
820
821 if (ca->kobj.state_in_sysfs) {
822 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
823 sysfs_remove_link(&ca->kobj, "block");
824 }
825
826 bch2_free_super(&ca->disk_sb);
827 bch2_dev_journal_exit(ca);
828}
829
830static void bch2_dev_ref_complete(struct percpu_ref *ref)
831{
832 struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
833
834 complete(&ca->ref_completion);
835}
836
837static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
838{
839 struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
840
841 complete(&ca->io_ref_completion);
842}
843
844static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
845{
846 int ret;
847
848 if (!c->kobj.state_in_sysfs)
849 return 0;
850
851 if (!ca->kobj.state_in_sysfs) {
852 ret = kobject_add(&ca->kobj, &c->kobj,
853 "dev-%u", ca->dev_idx);
854 if (ret)
855 return ret;
856 }
857
858 if (ca->disk_sb.bdev) {
859 struct kobject *block = bdev_kobj(ca->disk_sb.bdev);
860
861 ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
862 if (ret)
863 return ret;
864
865 ret = sysfs_create_link(&ca->kobj, block, "block");
866 if (ret)
867 return ret;
868 }
869
870 return 0;
871}
872
873static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
874 struct bch_member *member)
875{
876 struct bch_dev *ca;
877
878 ca = kzalloc(sizeof(*ca), GFP_KERNEL);
879 if (!ca)
880 return NULL;
881
882 kobject_init(&ca->kobj, &bch2_dev_ktype);
883 init_completion(&ca->ref_completion);
884 init_completion(&ca->io_ref_completion);
885
886 init_rwsem(&ca->bucket_lock);
887
888 writepoint_init(&ca->copygc_write_point, BCH_DATA_USER);
889
890 spin_lock_init(&ca->freelist_lock);
891 bch2_dev_copygc_init(ca);
892
893 INIT_WORK(&ca->io_error_work, bch2_io_error_work);
894
895 bch2_time_stats_init(&ca->io_latency[READ]);
896 bch2_time_stats_init(&ca->io_latency[WRITE]);
897
898 ca->mi = bch2_mi_to_cpu(member);
899 ca->uuid = member->uuid;
900
901 if (opt_defined(c->opts, discard))
902 ca->mi.discard = opt_get(c->opts, discard);
903
904 if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
905 0, GFP_KERNEL) ||
906 percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
907 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
908 bch2_dev_buckets_alloc(c, ca) ||
909 bioset_init(&ca->replica_set, 4,
910 offsetof(struct bch_write_bio, bio), 0) ||
911 !(ca->io_done = alloc_percpu(*ca->io_done)))
912 goto err;
913
914 return ca;
915err:
916 bch2_dev_free(ca);
917 return NULL;
918}
919
920static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
921 unsigned dev_idx)
922{
923 ca->dev_idx = dev_idx;
924 __set_bit(ca->dev_idx, ca->self.d);
925 scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
926
927 ca->fs = c;
928 rcu_assign_pointer(c->devs[ca->dev_idx], ca);
929
930 if (bch2_dev_sysfs_online(c, ca))
931 pr_warn("error creating sysfs objects");
932}
933
934static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
935{
936 struct bch_member *member =
937 bch2_sb_get_members(c->disk_sb.sb)->members + dev_idx;
938 struct bch_dev *ca = NULL;
939 int ret = 0;
940
941 pr_verbose_init(c->opts, "");
942
943 if (bch2_fs_init_fault("dev_alloc"))
944 goto err;
945
946 ca = __bch2_dev_alloc(c, member);
947 if (!ca)
948 goto err;
949
950 bch2_dev_attach(c, ca, dev_idx);
951out:
952 pr_verbose_init(c->opts, "ret %i", ret);
953 return ret;
954err:
955 if (ca)
956 bch2_dev_free(ca);
957 ret = -ENOMEM;
958 goto out;
959}
960
961static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
962{
963 unsigned ret;
964
965 if (bch2_dev_is_online(ca)) {
966 bch_err(ca, "already have device online in slot %u",
967 sb->sb->dev_idx);
968 return -EINVAL;
969 }
970
971 if (get_capacity(sb->bdev->bd_disk) <
972 ca->mi.bucket_size * ca->mi.nbuckets) {
973 bch_err(ca, "cannot online: device too small");
974 return -EINVAL;
975 }
976
977 BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
978
979 if (get_capacity(sb->bdev->bd_disk) <
980 ca->mi.bucket_size * ca->mi.nbuckets) {
981 bch_err(ca, "device too small");
982 return -EINVAL;
983 }
984
985 ret = bch2_dev_journal_init(ca, sb->sb);
986 if (ret)
987 return ret;
988
989 /* Commit: */
990 ca->disk_sb = *sb;
991 memset(sb, 0, sizeof(*sb));
992
1c6fdbd8
KO		 993 percpu_ref_reinit(&ca->io_ref);
994
995 return 0;
996}
997
998static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
999{
1000 struct bch_dev *ca;
1001 int ret;
1002
1003 lockdep_assert_held(&c->state_lock);
1004
1005 if (le64_to_cpu(sb->sb->seq) >
1006 le64_to_cpu(c->disk_sb.sb->seq))
1007 bch2_sb_to_fs(c, sb->sb);
1008
1009 BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
1010 !c->devs[sb->sb->dev_idx]);
1011
1012 ca = bch_dev_locked(c, sb->sb->dev_idx);
1013
1014 ret = __bch2_dev_attach_bdev(ca, sb);
1015 if (ret)
1016 return ret;
1017
6eac2c2e
KO		 1018 mutex_lock(&c->sb_lock);
1019 bch2_mark_dev_superblock(ca->fs, ca,
1020 BCH_BUCKET_MARK_MAY_MAKE_UNAVAILABLE);
1021 mutex_unlock(&c->sb_lock);
1022
1c6fdbd8
KO		 1023 bch2_dev_sysfs_online(c, ca);
1024
1025 if (c->sb.nr_devices == 1)
1026 snprintf(c->name, sizeof(c->name), "%pg", ca->disk_sb.bdev);
1027 snprintf(ca->name, sizeof(ca->name), "%pg", ca->disk_sb.bdev);
1028
1029 rebalance_wakeup(c);
1030 return 0;
1031}
1032
1033/* Device management: */
1034
1035/*
1036 * Note: this function is also used by the error paths - when a particular
1037 * device sees an error, we call it to determine whether we can just set the
1038 * device RO, or - if this function returns false - we'll set the whole
1039 * filesystem RO:
1040 *
1041 * XXX: maybe we should be more explicit about whether we're changing state
1042 * because we got an error or what have you?
1043 */
1044bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1045 enum bch_member_state new_state, int flags)
1046{
1047 struct bch_devs_mask new_online_devs;
1048 struct replicas_status s;
1049 struct bch_dev *ca2;
1050 int i, nr_rw = 0, required;
1051
1052 lockdep_assert_held(&c->state_lock);
1053
1054 switch (new_state) {
1055 case BCH_MEMBER_STATE_RW:
1056 return true;
1057 case BCH_MEMBER_STATE_RO:
1058 if (ca->mi.state != BCH_MEMBER_STATE_RW)
1059 return true;
1060
1061 /* do we have enough devices to write to? */
1062 for_each_member_device(ca2, c, i)
1063 if (ca2 != ca)
1064 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_RW;
1065
1066 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1067 ? c->opts.metadata_replicas
1068 : c->opts.metadata_replicas_required,
1069 !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1070 ? c->opts.data_replicas
1071 : c->opts.data_replicas_required);
1072
1073 return nr_rw >= required;
1074 case BCH_MEMBER_STATE_FAILED:
1075 case BCH_MEMBER_STATE_SPARE:
1076 if (ca->mi.state != BCH_MEMBER_STATE_RW &&
1077 ca->mi.state != BCH_MEMBER_STATE_RO)
1078 return true;
1079
1080 /* do we have enough devices to read from? */
1081 new_online_devs = bch2_online_devs(c);
1082 __clear_bit(ca->dev_idx, new_online_devs.d);
1083
1084 s = __bch2_replicas_status(c, new_online_devs);
1085
1086 return bch2_have_enough_devs(s, flags);
1087 default:
1088 BUG();
1089 }
1090}
1091
1092static bool bch2_fs_may_start(struct bch_fs *c)
1093{
1094 struct replicas_status s;
1095 struct bch_sb_field_members *mi;
1096 struct bch_dev *ca;
1097 unsigned i, flags = c->opts.degraded
1098 ? BCH_FORCE_IF_DEGRADED
1099 : 0;
1100
1101 if (!c->opts.degraded) {
1102 mutex_lock(&c->sb_lock);
1103 mi = bch2_sb_get_members(c->disk_sb.sb);
1104
1105 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1106 if (!bch2_dev_exists(c->disk_sb.sb, mi, i))
1107 continue;
1108
1109 ca = bch_dev_locked(c, i);
1110
1111 if (!bch2_dev_is_online(ca) &&
1112 (ca->mi.state == BCH_MEMBER_STATE_RW ||
1113 ca->mi.state == BCH_MEMBER_STATE_RO)) {
1114 mutex_unlock(&c->sb_lock);
1115 return false;
1116 }
1117 }
1118 mutex_unlock(&c->sb_lock);
1119 }
1120
1121 s = bch2_replicas_status(c);
1122
1123 return bch2_have_enough_devs(s, flags);
1124}
1125
1126static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1127{
1128 bch2_copygc_stop(ca);
1129
1130 /*
1131 * The allocator thread itself allocates btree nodes, so stop it first:
1132 */
1133 bch2_dev_allocator_stop(ca);
1134 bch2_dev_allocator_remove(c, ca);
1135 bch2_dev_journal_stop(&c->journal, ca);
1136}
1137
1138static const char *__bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1139{
1140 lockdep_assert_held(&c->state_lock);
1141
1142 BUG_ON(ca->mi.state != BCH_MEMBER_STATE_RW);
1143
1144 bch2_dev_allocator_add(c, ca);
1145 bch2_recalc_capacity(c);
1146
1147 if (bch2_dev_allocator_start(ca))
1148 return "error starting allocator thread";
1149
1150 if (bch2_copygc_start(c, ca))
1151 return "error starting copygc thread";
1152
1153 return NULL;
1154}
1155
1156int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1157 enum bch_member_state new_state, int flags)
1158{
1159 struct bch_sb_field_members *mi;
1160 int ret = 0;
1161
1162 if (ca->mi.state == new_state)
1163 return 0;
1164
1165 if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1166 return -EINVAL;
1167
1168 if (new_state != BCH_MEMBER_STATE_RW)
1169 __bch2_dev_read_only(c, ca);
1170
1171 bch_notice(ca, "%s", bch2_dev_state[new_state]);
1172
1173 mutex_lock(&c->sb_lock);
1174 mi = bch2_sb_get_members(c->disk_sb.sb);
1175 SET_BCH_MEMBER_STATE(&mi->members[ca->dev_idx], new_state);
1176 bch2_write_super(c);
1177 mutex_unlock(&c->sb_lock);
1178
1179 if (new_state == BCH_MEMBER_STATE_RW &&
1180 __bch2_dev_read_write(c, ca))
1181 ret = -ENOMEM;
1182
1183 rebalance_wakeup(c);
1184
1185 return ret;
1186}
1187
1188int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1189 enum bch_member_state new_state, int flags)
1190{
1191 int ret;
1192
1193 mutex_lock(&c->state_lock);
1194 ret = __bch2_dev_set_state(c, ca, new_state, flags);
1195 mutex_unlock(&c->state_lock);
1196
1197 return ret;
1198}
1199
1200/* Device add/removal: */
1201
1202int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1203{
1204 struct bch_sb_field_members *mi;
1205 unsigned dev_idx = ca->dev_idx, data;
1206 int ret = -EINVAL;
1207
1208 mutex_lock(&c->state_lock);
1209
1210 percpu_ref_put(&ca->ref); /* XXX */
1211
1212 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_FAILED, flags)) {
1213 bch_err(ca, "Cannot remove without losing data");
1214 goto err;
1215 }
1216
1217 __bch2_dev_read_only(c, ca);
1218
1219 /*
1220 * XXX: verify that dev_idx is really not in use anymore, anywhere
1221 *
1222 * flag_data_bad() does not check btree pointers
1223 */
1224 ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1225 if (ret) {
1226 bch_err(ca, "Remove failed: error %i dropping data", ret);
1227 goto err;
1228 }
1229
1230 ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1231 if (ret) {
1232 bch_err(ca, "Remove failed: error %i flushing journal", ret);
1233 goto err;
1234 }
1235
1236 data = bch2_dev_has_data(c, ca);
1237 if (data) {
1238 char data_has_str[100];
1239 bch2_scnprint_flag_list(data_has_str,
1240 sizeof(data_has_str),
1241 bch2_data_types,
1242 data);
1243 bch_err(ca, "Remove failed, still has data (%s)", data_has_str);
1244 ret = -EBUSY;
1245 goto err;
1246 }
1247
1248 ret = bch2_btree_delete_range(c, BTREE_ID_ALLOC,
1249 POS(ca->dev_idx, 0),
1250 POS(ca->dev_idx + 1, 0),
fc3268c1 1251 NULL);
1c6fdbd8
KO		 1252 if (ret) {
1253 bch_err(ca, "Remove failed, error deleting alloc info");
1254 goto err;
1255 }
1256
1257 /*
1258 * must flush all existing journal entries, they might have
1259 * (overwritten) keys that point to the device we're removing:
1260 */
1261 bch2_journal_flush_all_pins(&c->journal);
1262 ret = bch2_journal_error(&c->journal);
1263 if (ret) {
1264 bch_err(ca, "Remove failed, journal error");
1265 goto err;
1266 }
1267
1268 __bch2_dev_offline(c, ca);
1269
1270 mutex_lock(&c->sb_lock);
1271 rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1272 mutex_unlock(&c->sb_lock);
1273
1274 percpu_ref_kill(&ca->ref);
1275 wait_for_completion(&ca->ref_completion);
1276
1277 bch2_dev_free(ca);
1278
1279 /*
1280 * Free this device's slot in the bch_member array - all pointers to
1281 * this device must be gone:
1282 */
1283 mutex_lock(&c->sb_lock);
1284 mi = bch2_sb_get_members(c->disk_sb.sb);
1285 memset(&mi->members[dev_idx].uuid, 0, sizeof(mi->members[dev_idx].uuid));
1286
1287 bch2_write_super(c);
1288
1289 mutex_unlock(&c->sb_lock);
1290 mutex_unlock(&c->state_lock);
1291 return 0;
1292err:
1293 if (ca->mi.state == BCH_MEMBER_STATE_RW)
1294 __bch2_dev_read_write(c, ca);
1295 mutex_unlock(&c->state_lock);
1296 return ret;
1297}
1298
6eac2c2e
KO		 1299static void dev_usage_clear(struct bch_dev *ca)
1300{
1301 struct bucket_array *buckets;
1302 int cpu;
1303
1304 for_each_possible_cpu(cpu) {
1305 struct bch_dev_usage *p =
1306 per_cpu_ptr(ca->usage_percpu, cpu);
1307 memset(p, 0, sizeof(*p));
1308 }
1309
1310 down_read(&ca->bucket_lock);
1311 buckets = bucket_array(ca);
1312
1313 memset(buckets->b, 0, sizeof(buckets->b[0]) * buckets->nbuckets);
1314 up_read(&ca->bucket_lock);
1315}
1316
1c6fdbd8
KO		 1317/* Add new device to running filesystem: */
1318int bch2_dev_add(struct bch_fs *c, const char *path)
1319{
1320 struct bch_opts opts = bch2_opts_empty();
1321 struct bch_sb_handle sb;
1322 const char *err;
1323 struct bch_dev *ca = NULL;
1324 struct bch_sb_field_members *mi;
1325 struct bch_member dev_mi;
1326 unsigned dev_idx, nr_devices, u64s;
1327 int ret;
1328
1329 ret = bch2_read_super(path, &opts, &sb);
1330 if (ret)
1331 return ret;
1332
1333 err = bch2_sb_validate(&sb);
1334 if (err)
1335 return -EINVAL;
1336
1337 dev_mi = bch2_sb_get_members(sb.sb)->members[sb.sb->dev_idx];
1338
1339 err = bch2_dev_may_add(sb.sb, c);
1340 if (err)
1341 return -EINVAL;
1342
1343 ca = __bch2_dev_alloc(c, &dev_mi);
1344 if (!ca) {
1345 bch2_free_super(&sb);
1346 return -ENOMEM;
1347 }
1348
1349 ret = __bch2_dev_attach_bdev(ca, &sb);
1350 if (ret) {
1351 bch2_dev_free(ca);
1352 return ret;
1353 }
1354
6eac2c2e
KO		 1355 /*
1356 * We want to allocate journal on the new device before adding the new
1357 * device to the filesystem because allocating after we attach requires
1358 * spinning up the allocator thread, and the allocator thread requires
1359 * doing btree writes, which if the existing devices are RO isn't going
1360 * to work
1361 *
1362 * So we have to mark where the superblocks are, but marking allocated
1363 * data normally updates the filesystem usage too, so we have to mark,
1364 * allocate the journal, reset all the marks, then remark after we
1365 * attach...
1366 */
1367 bch2_mark_dev_superblock(ca->fs, ca,
1368 BCH_BUCKET_MARK_MAY_MAKE_UNAVAILABLE);
1369
1c6fdbd8
KO		 1370 err = "journal alloc failed";
1371 ret = bch2_dev_journal_alloc(ca);
1372 if (ret)
1373 goto err;
1374
6eac2c2e
KO		 1375 dev_usage_clear(ca);
1376
1c6fdbd8
KO		 1377 mutex_lock(&c->state_lock);
1378 mutex_lock(&c->sb_lock);
1379
1380 err = "insufficient space in new superblock";
1381 ret = bch2_sb_from_fs(c, ca);
1382 if (ret)
1383 goto err_unlock;
1384
1385 mi = bch2_sb_get_members(ca->disk_sb.sb);
1386
1387 if (!bch2_sb_resize_members(&ca->disk_sb,
1388 le32_to_cpu(mi->field.u64s) +
1389 sizeof(dev_mi) / sizeof(u64))) {
1390 ret = -ENOSPC;
1391 goto err_unlock;
1392 }
1393
1394 if (dynamic_fault("bcachefs:add:no_slot"))
1395 goto no_slot;
1396
1397 mi = bch2_sb_get_members(c->disk_sb.sb);
1398 for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++)
1399 if (!bch2_dev_exists(c->disk_sb.sb, mi, dev_idx))
1400 goto have_slot;
1401no_slot:
1402 err = "no slots available in superblock";
1403 ret = -ENOSPC;
1404 goto err_unlock;
1405
1406have_slot:
1407 nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1408 u64s = (sizeof(struct bch_sb_field_members) +
1409 sizeof(struct bch_member) * nr_devices) / sizeof(u64);
1410
1411 err = "no space in superblock for member info";
1412 ret = -ENOSPC;
1413
1414 mi = bch2_sb_resize_members(&c->disk_sb, u64s);
1415 if (!mi)
1416 goto err_unlock;
1417
1418 /* success: */
1419
1420 mi->members[dev_idx] = dev_mi;
1421 mi->members[dev_idx].last_mount = cpu_to_le64(ktime_get_seconds());
1422 c->disk_sb.sb->nr_devices = nr_devices;
1423
1424 ca->disk_sb.sb->dev_idx = dev_idx;
1425 bch2_dev_attach(c, ca, dev_idx);
1426
6eac2c2e
KO		 1427 bch2_mark_dev_superblock(c, ca,
1428 BCH_BUCKET_MARK_MAY_MAKE_UNAVAILABLE);
1429
1c6fdbd8
KO		 1430 bch2_write_super(c);
1431 mutex_unlock(&c->sb_lock);
1432
1433 if (ca->mi.state == BCH_MEMBER_STATE_RW) {
1434 err = __bch2_dev_read_write(c, ca);
1435 if (err)
1436 goto err_late;
1437 }
1438
1439 mutex_unlock(&c->state_lock);
1440 return 0;
1441
1442err_unlock:
1443 mutex_unlock(&c->sb_lock);
1444 mutex_unlock(&c->state_lock);
1445err:
1446 if (ca)
1447 bch2_dev_free(ca);
1448 bch2_free_super(&sb);
1449 bch_err(c, "Unable to add device: %s", err);
1450 return ret;
1451err_late:
1452 bch_err(c, "Error going rw after adding device: %s", err);
1453 return -EINVAL;
1454}
1455
1456/* Hot add existing device to running filesystem: */
1457int bch2_dev_online(struct bch_fs *c, const char *path)
1458{
1459 struct bch_opts opts = bch2_opts_empty();
1460 struct bch_sb_handle sb = { NULL };
1461 struct bch_sb_field_members *mi;
1462 struct bch_dev *ca;
1463 unsigned dev_idx;
1464 const char *err;
1465 int ret;
1466
1467 mutex_lock(&c->state_lock);
1468
1469 ret = bch2_read_super(path, &opts, &sb);
1470 if (ret) {
1471 mutex_unlock(&c->state_lock);
1472 return ret;
1473 }
1474
1475 dev_idx = sb.sb->dev_idx;
1476
1477 err = bch2_dev_in_fs(c->disk_sb.sb, sb.sb);
1478 if (err)
1479 goto err;
1480
1481 if (bch2_dev_attach_bdev(c, &sb)) {
1482 err = "bch2_dev_attach_bdev() error";
1483 goto err;
1484 }
1485
1486 ca = bch_dev_locked(c, dev_idx);
1487 if (ca->mi.state == BCH_MEMBER_STATE_RW) {
1488 err = __bch2_dev_read_write(c, ca);
1489 if (err)
1490 goto err;
1491 }
1492
1493 mutex_lock(&c->sb_lock);
1494 mi = bch2_sb_get_members(c->disk_sb.sb);
1495
1496 mi->members[ca->dev_idx].last_mount =
1497 cpu_to_le64(ktime_get_seconds());
1498
1499 bch2_write_super(c);
1500 mutex_unlock(&c->sb_lock);
1501
1502 mutex_unlock(&c->state_lock);
1503 return 0;
1504err:
1505 mutex_unlock(&c->state_lock);
1506 bch2_free_super(&sb);
1507 bch_err(c, "error bringing %s online: %s", path, err);
1508 return -EINVAL;
1509}
1510
1511int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1512{
1513 mutex_lock(&c->state_lock);
1514
1515 if (!bch2_dev_is_online(ca)) {
1516 bch_err(ca, "Already offline");
1517 mutex_unlock(&c->state_lock);
1518 return 0;
1519 }
1520
1521 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_FAILED, flags)) {
1522 bch_err(ca, "Cannot offline required disk");
1523 mutex_unlock(&c->state_lock);
1524 return -EINVAL;
1525 }
1526
1527 __bch2_dev_offline(c, ca);
1528
1529 mutex_unlock(&c->state_lock);
1530 return 0;
1531}
1532
1533int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1534{
1535 struct bch_member *mi;
1536 int ret = 0;
1537
1538 mutex_lock(&c->state_lock);
1539
1540 if (nbuckets < ca->mi.nbuckets) {
1541 bch_err(ca, "Cannot shrink yet");
1542 ret = -EINVAL;
1543 goto err;
1544 }
1545
1546 if (bch2_dev_is_online(ca) &&
1547 get_capacity(ca->disk_sb.bdev->bd_disk) <
1548 ca->mi.bucket_size * nbuckets) {
1549 bch_err(ca, "New size larger than device");
1550 ret = -EINVAL;
1551 goto err;
1552 }
1553
1554 ret = bch2_dev_buckets_resize(c, ca, nbuckets);
1555 if (ret) {
1556 bch_err(ca, "Resize error: %i", ret);
1557 goto err;
1558 }
1559
1560 mutex_lock(&c->sb_lock);
1561 mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
1562 mi->nbuckets = cpu_to_le64(nbuckets);
1563
1564 bch2_write_super(c);
1565 mutex_unlock(&c->sb_lock);
1566
1567 bch2_recalc_capacity(c);
1568err:
1569 mutex_unlock(&c->state_lock);
1570 return ret;
1571}
1572
1573/* return with ref on ca->ref: */
1574struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *path)
1575{
1576
1577 struct bch_dev *ca;
1578 dev_t dev;
1579 unsigned i;
1580 int ret;
1581
1582 ret = lookup_bdev(path, &dev);
1583 if (ret)
1584 return ERR_PTR(ret);
1585
1586 for_each_member_device(ca, c, i)
1587 if (ca->disk_sb.bdev->bd_dev == dev)
1588 goto found;
1589
1590 ca = ERR_PTR(-ENOENT);
1591found:
1592 return ca;
1593}
1594
1595/* Filesystem open: */
1596
1597struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
1598 struct bch_opts opts)
1599{
1600 struct bch_sb_handle *sb = NULL;
1601 struct bch_fs *c = NULL;
1602 unsigned i, best_sb = 0;
1603 const char *err;
1604 int ret = -ENOMEM;
1605
1606 pr_verbose_init(opts, "");
1607
1608 if (!nr_devices) {
1609 c = ERR_PTR(-EINVAL);
1610 goto out2;
1611 }
1612
1613 if (!try_module_get(THIS_MODULE)) {
1614 c = ERR_PTR(-ENODEV);
1615 goto out2;
1616 }
1617
1618 sb = kcalloc(nr_devices, sizeof(*sb), GFP_KERNEL);
1619 if (!sb)
1620 goto err;
1621
1622 for (i = 0; i < nr_devices; i++) {
1623 ret = bch2_read_super(devices[i], &opts, &sb[i]);
1624 if (ret)
1625 goto err;
1626
1627 err = bch2_sb_validate(&sb[i]);
1628 if (err)
1629 goto err_print;
1630 }
1631
1632 for (i = 1; i < nr_devices; i++)
1633 if (le64_to_cpu(sb[i].sb->seq) >
1634 le64_to_cpu(sb[best_sb].sb->seq))
1635 best_sb = i;
1636
1637 for (i = 0; i < nr_devices; i++) {
1638 err = bch2_dev_in_fs(sb[best_sb].sb, sb[i].sb);
1639 if (err)
1640 goto err_print;
1641 }
1642
1643 ret = -ENOMEM;
1644 c = bch2_fs_alloc(sb[best_sb].sb, opts);
1645 if (!c)
1646 goto err;
1647
1648 err = "bch2_dev_online() error";
1649 mutex_lock(&c->state_lock);
1650 for (i = 0; i < nr_devices; i++)
1651 if (bch2_dev_attach_bdev(c, &sb[i])) {
1652 mutex_unlock(&c->state_lock);
1653 goto err_print;
1654 }
1655 mutex_unlock(&c->state_lock);
1656
1657 err = "insufficient devices";
1658 if (!bch2_fs_may_start(c))
1659 goto err_print;
1660
1661 if (!c->opts.nostart) {
1662 err = bch2_fs_start(c);
1663 if (err)
1664 goto err_print;
1665 }
1666out:
1667 kfree(sb);
1668 module_put(THIS_MODULE);
1669out2:
1670 pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
1671 return c;
1672err_print:
1673 pr_err("bch_fs_open err opening %s: %s",
1674 devices[0], err);
1675 ret = -EINVAL;
1676err:
1677 if (c)
1678 bch2_fs_stop(c);
1679 for (i = 0; i < nr_devices; i++)
1680 bch2_free_super(&sb[i]);
1681 c = ERR_PTR(ret);
1682 goto out;
1683}
1684
1685static const char *__bch2_fs_open_incremental(struct bch_sb_handle *sb,
1686 struct bch_opts opts)
1687{
1688 const char *err;
1689 struct bch_fs *c;
1690 bool allocated_fs = false;
1691
1692 err = bch2_sb_validate(sb);
1693 if (err)
1694 return err;
1695
1696 mutex_lock(&bch_fs_list_lock);
1697 c = __bch2_uuid_to_fs(sb->sb->uuid);
1698 if (c) {
1699 closure_get(&c->cl);
1700
1701 err = bch2_dev_in_fs(c->disk_sb.sb, sb->sb);
1702 if (err)
1703 goto err;
1704 } else {
1705 c = bch2_fs_alloc(sb->sb, opts);
1706 err = "cannot allocate memory";
1707 if (!c)
1708 goto err;
1709
1710 allocated_fs = true;
1711 }
1712
1713 err = "bch2_dev_online() error";
1714
1715 mutex_lock(&c->sb_lock);
1716 if (bch2_dev_attach_bdev(c, sb)) {
1717 mutex_unlock(&c->sb_lock);
1718 goto err;
1719 }
1720 mutex_unlock(&c->sb_lock);
1721
1722 if (!c->opts.nostart && bch2_fs_may_start(c)) {
1723 err = bch2_fs_start(c);
1724 if (err)
1725 goto err;
1726 }
1727
1728 closure_put(&c->cl);
1729 mutex_unlock(&bch_fs_list_lock);
1730
1731 return NULL;
1732err:
1733 mutex_unlock(&bch_fs_list_lock);
1734
1735 if (allocated_fs)
1736 bch2_fs_stop(c);
1737 else if (c)
1738 closure_put(&c->cl);
1739
1740 return err;
1741}
1742
1743const char *bch2_fs_open_incremental(const char *path)
1744{
1745 struct bch_sb_handle sb;
1746 struct bch_opts opts = bch2_opts_empty();
1747 const char *err;
1748
1749 if (bch2_read_super(path, &opts, &sb))
1750 return "error reading superblock";
1751
1752 err = __bch2_fs_open_incremental(&sb, opts);
1753 bch2_free_super(&sb);
1754
1755 return err;
1756}
1757
1758/* Global interfaces/init */
1759
1760static void bcachefs_exit(void)
1761{
1762 bch2_debug_exit();
1763 bch2_vfs_exit();
1764 bch2_chardev_exit();
1765 if (bcachefs_kset)
1766 kset_unregister(bcachefs_kset);
1767}
1768
1769static int __init bcachefs_init(void)
1770{
1771 bch2_bkey_pack_test();
1772 bch2_inode_pack_test();
1773
1774 if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
1775 bch2_chardev_init() ||
1776 bch2_vfs_init() ||
1777 bch2_debug_init())
1778 goto err;
1779
1780 return 0;
1781err:
1782 bcachefs_exit();
1783 return -ENOMEM;
1784}
1785
1786#define BCH_DEBUG_PARAM(name, description) \
1787 bool bch2_##name; \
1788 module_param_named(name, bch2_##name, bool, 0644); \
1789 MODULE_PARM_DESC(name, description);
1790BCH_DEBUG_PARAMS()
1791#undef BCH_DEBUG_PARAM
1792
1793unsigned bch2_metadata_version = BCH_SB_VERSION_MAX;
1794module_param_named(version, bch2_metadata_version, uint, 0400);
1795
1796module_exit(bcachefs_exit);
1797module_init(bcachefs_init);
Linux 6.x block layer and io_uring tree(s)