projects / linux-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
bcachefs: Fix a bug when shutting down before allocator started
[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"
33#include "move.h"
34#include "migrate.h"
35#include "movinggc.h"
36#include "quota.h"
37#include "rebalance.h"
38#include "recovery.h"
39#include "replicas.h"
40#include "super.h"
41#include "super-io.h"
42#include "sysfs.h"
43#include "trace.h"
44
45#include <linux/backing-dev.h>
46#include <linux/blkdev.h>
47#include <linux/debugfs.h>
48#include <linux/device.h>
49#include <linux/idr.h>
50#include <linux/kthread.h>
51#include <linux/module.h>
52#include <linux/percpu.h>
53#include <linux/random.h>
54#include <linux/sysfs.h>
55#include <crypto/hash.h>
56
57MODULE_LICENSE("GPL");
58MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
59
60#define KTYPE(type) \
61static const struct attribute_group type ## _group = { \
62 .attrs = type ## _files \
63}; \
64 \
65static const struct attribute_group *type ## _groups[] = { \
66 &type ## _group, \
67 NULL \
68}; \
69 \
70static const struct kobj_type type ## _ktype = { \
71 .release = type ## _release, \
72 .sysfs_ops = &type ## _sysfs_ops, \
73 .default_groups = type ## _groups \
74}
75
76static void bch2_fs_release(struct kobject *);
77static void bch2_dev_release(struct kobject *);
78
79static void bch2_fs_internal_release(struct kobject *k)
80{
81}
82
83static void bch2_fs_opts_dir_release(struct kobject *k)
84{
85}
86
87static void bch2_fs_time_stats_release(struct kobject *k)
88{
89}
90
91KTYPE(bch2_fs);
92KTYPE(bch2_fs_internal);
93KTYPE(bch2_fs_opts_dir);
94KTYPE(bch2_fs_time_stats);
95KTYPE(bch2_dev);
96
97static struct kset *bcachefs_kset;
98static LIST_HEAD(bch_fs_list);
99static DEFINE_MUTEX(bch_fs_list_lock);
100
101static DECLARE_WAIT_QUEUE_HEAD(bch_read_only_wait);
102
103static void bch2_dev_free(struct bch_dev *);
104static int bch2_dev_alloc(struct bch_fs *, unsigned);
105static int bch2_dev_sysfs_online(struct bch_fs *, struct bch_dev *);
106static void __bch2_dev_read_only(struct bch_fs *, struct bch_dev *);
107
108struct bch_fs *bch2_dev_to_fs(dev_t dev)
109{
110 struct bch_fs *c;
111 struct bch_dev *ca;
112 unsigned i;
113
114 mutex_lock(&bch_fs_list_lock);
115 rcu_read_lock();
116
117 list_for_each_entry(c, &bch_fs_list, list)
118 for_each_member_device_rcu(ca, c, i, NULL)
119 if (ca->disk_sb.bdev->bd_dev == dev) {
120 closure_get(&c->cl);
121 goto found;
122 }
123 c = NULL;
124found:
125 rcu_read_unlock();
126 mutex_unlock(&bch_fs_list_lock);
127
128 return c;
129}
130
131static struct bch_fs *__bch2_uuid_to_fs(__uuid_t uuid)
132{
133 struct bch_fs *c;
134
135 lockdep_assert_held(&bch_fs_list_lock);
136
137 list_for_each_entry(c, &bch_fs_list, list)
138 if (!memcmp(&c->disk_sb.sb->uuid, &uuid, sizeof(uuid)))
139 return c;
140
141 return NULL;
142}
143
144struct bch_fs *bch2_uuid_to_fs(__uuid_t uuid)
145{
146 struct bch_fs *c;
147
148 mutex_lock(&bch_fs_list_lock);
149 c = __bch2_uuid_to_fs(uuid);
150 if (c)
151 closure_get(&c->cl);
152 mutex_unlock(&bch_fs_list_lock);
153
154 return c;
155}
156
157/* Filesystem RO/RW: */
158
159/*
160 * For startup/shutdown of RW stuff, the dependencies are:
161 *
162 * - foreground writes depend on copygc and rebalance (to free up space)
163 *
164 * - copygc and rebalance depend on mark and sweep gc (they actually probably
165 * don't because they either reserve ahead of time or don't block if
166 * allocations fail, but allocations can require mark and sweep gc to run
167 * because of generation number wraparound)
168 *
169 * - all of the above depends on the allocator threads
170 *
171 * - allocator depends on the journal (when it rewrites prios and gens)
172 */
173
174static void __bch2_fs_read_only(struct bch_fs *c)
175{
176 struct bch_dev *ca;
430735cd 177 bool wrote;
1c6fdbd8 178 unsigned i;
430735cd 179 int ret;
1c6fdbd8
KO		 180
181 bch2_rebalance_stop(c);
182
183 for_each_member_device(ca, c, i)
184 bch2_copygc_stop(ca);
185
186 bch2_gc_thread_stop(c);
187
188 /*
189 * Flush journal before stopping allocators, because flushing journal
190 * blacklist entries involves allocating new btree nodes:
191 */
192 bch2_journal_flush_all_pins(&c->journal);
193
b935a8a6
KO		 194 if (!test_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags))
195 goto allocator_not_running;
196
430735cd
KO		 197 do {
198 ret = bch2_alloc_write(c, false, &wrote);
199 if (ret) {
200 bch2_fs_inconsistent(c, "error writing out alloc info %i", ret);
201 break;
202 }
203
61c8d7c8
KO		 204 ret = bch2_stripes_write(c, &wrote);
205 if (ret) {
206 bch2_fs_inconsistent(c, "error writing out stripes");
207 break;
208 }
209
430735cd
KO		 210 for_each_member_device(ca, c, i)
211 bch2_dev_allocator_quiesce(c, ca);
212
213 bch2_journal_flush_all_pins(&c->journal);
214
215 /*
216 * We need to explicitly wait on btree interior updates to complete
217 * before stopping the journal, flushing all journal pins isn't
218 * sufficient, because in the BTREE_INTERIOR_UPDATING_ROOT case btree
219 * interior updates have to drop their journal pin before they're
220 * fully complete:
221 */
222 closure_wait_event(&c->btree_interior_update_wait,
223 !bch2_btree_interior_updates_nr_pending(c));
224 } while (wrote);
b935a8a6 225allocator_not_running:
1c6fdbd8
KO		 226 for_each_member_device(ca, c, i)
227 bch2_dev_allocator_stop(ca);
228
b935a8a6
KO		 229 clear_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags);
230
1c6fdbd8
KO		 231 bch2_fs_journal_stop(&c->journal);
232
430735cd
KO		 233 /* XXX: mark super that alloc info is persistent */
234
1c6fdbd8
KO		 235 /*
236 * the journal kicks off btree writes via reclaim - wait for in flight
237 * writes after stopping journal:
238 */
239 if (test_bit(BCH_FS_EMERGENCY_RO, &c->flags))
240 bch2_btree_flush_all_writes(c);
241 else
242 bch2_btree_verify_flushed(c);
243
244 /*
245 * After stopping journal:
246 */
247 for_each_member_device(ca, c, i)
248 bch2_dev_allocator_remove(c, ca);
249}
250
251static void bch2_writes_disabled(struct percpu_ref *writes)
252{
253 struct bch_fs *c = container_of(writes, struct bch_fs, writes);
254
255 set_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
256 wake_up(&bch_read_only_wait);
257}
258
259void bch2_fs_read_only(struct bch_fs *c)
260{
261 if (c->state == BCH_FS_RO)
262 return;
263
264 BUG_ON(test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
265
266 /*
267 * Block new foreground-end write operations from starting - any new
268 * writes will return -EROFS:
269 *
270 * (This is really blocking new _allocations_, writes to previously
271 * allocated space can still happen until stopping the allocator in
272 * bch2_dev_allocator_stop()).
273 */
274 percpu_ref_kill(&c->writes);
275
276 cancel_delayed_work(&c->pd_controllers_update);
277
278 /*
279 * If we're not doing an emergency shutdown, we want to wait on
280 * outstanding writes to complete so they don't see spurious errors due
281 * to shutting down the allocator:
282 *
283 * If we are doing an emergency shutdown outstanding writes may
284 * hang until we shutdown the allocator so we don't want to wait
285 * on outstanding writes before shutting everything down - but
286 * we do need to wait on them before returning and signalling
287 * that going RO is complete:
288 */
289 wait_event(bch_read_only_wait,
290 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags) ||
291 test_bit(BCH_FS_EMERGENCY_RO, &c->flags));
292
293 __bch2_fs_read_only(c);
294
295 wait_event(bch_read_only_wait,
296 test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
297
298 clear_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
299
300 if (!bch2_journal_error(&c->journal) &&
301 !test_bit(BCH_FS_ERROR, &c->flags) &&
302 !test_bit(BCH_FS_EMERGENCY_RO, &c->flags))
303 bch2_fs_mark_clean(c, true);
304
305 if (c->state != BCH_FS_STOPPING)
306 c->state = BCH_FS_RO;
307}
308
309static void bch2_fs_read_only_work(struct work_struct *work)
310{
311 struct bch_fs *c =
312 container_of(work, struct bch_fs, read_only_work);
313
314 mutex_lock(&c->state_lock);
315 bch2_fs_read_only(c);
316 mutex_unlock(&c->state_lock);
317}
318
319static void bch2_fs_read_only_async(struct bch_fs *c)
320{
321 queue_work(system_long_wq, &c->read_only_work);
322}
323
324bool bch2_fs_emergency_read_only(struct bch_fs *c)
325{
326 bool ret = !test_and_set_bit(BCH_FS_EMERGENCY_RO, &c->flags);
327
328 bch2_fs_read_only_async(c);
329 bch2_journal_halt(&c->journal);
330
331 wake_up(&bch_read_only_wait);
332 return ret;
333}
334
335const char *bch2_fs_read_write(struct bch_fs *c)
336{
337 struct bch_dev *ca;
338 const char *err = NULL;
339 unsigned i;
340
341 if (c->state == BCH_FS_RW)
342 return NULL;
343
344 bch2_fs_mark_clean(c, false);
345
346 for_each_rw_member(ca, c, i)
347 bch2_dev_allocator_add(c, ca);
348 bch2_recalc_capacity(c);
349
350 err = "error starting allocator thread";
351 for_each_rw_member(ca, c, i)
352 if (bch2_dev_allocator_start(ca)) {
353 percpu_ref_put(&ca->io_ref);
354 goto err;
355 }
356
b935a8a6
KO		 357 set_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags);
358
1c6fdbd8
KO		 359 err = "error starting btree GC thread";
360 if (bch2_gc_thread_start(c))
361 goto err;
362
363 err = "error starting copygc thread";
364 for_each_rw_member(ca, c, i)
365 if (bch2_copygc_start(c, ca)) {
366 percpu_ref_put(&ca->io_ref);
367 goto err;
368 }
369
370 err = "error starting rebalance thread";
371 if (bch2_rebalance_start(c))
372 goto err;
373
374 schedule_delayed_work(&c->pd_controllers_update, 5 * HZ);
375
376 if (c->state != BCH_FS_STARTING)
377 percpu_ref_reinit(&c->writes);
378
379 c->state = BCH_FS_RW;
380 return NULL;
381err:
382 __bch2_fs_read_only(c);
383 return err;
384}
385
386/* Filesystem startup/shutdown: */
387
388static void bch2_fs_free(struct bch_fs *c)
389{
390 unsigned i;
391
392 for (i = 0; i < BCH_TIME_STAT_NR; i++)
393 bch2_time_stats_exit(&c->times[i]);
394
395 bch2_fs_quota_exit(c);
396 bch2_fs_fsio_exit(c);
cd575ddf 397 bch2_fs_ec_exit(c);
1c6fdbd8
KO		 398 bch2_fs_encryption_exit(c);
399 bch2_fs_io_exit(c);
400 bch2_fs_btree_cache_exit(c);
401 bch2_fs_journal_exit(&c->journal);
402 bch2_io_clock_exit(&c->io_clock[WRITE]);
403 bch2_io_clock_exit(&c->io_clock[READ]);
404 bch2_fs_compress_exit(c);
9166b41d 405 percpu_free_rwsem(&c->mark_lock);
7ef2a73a 406 free_percpu(c->usage_scratch);
9ca53b55 407 free_percpu(c->usage[0]);
5663a415 408 free_percpu(c->pcpu);
581edb63 409 mempool_exit(&c->btree_iters_pool);
1c6fdbd8
KO		 410 mempool_exit(&c->btree_bounce_pool);
411 bioset_exit(&c->btree_bio);
412 mempool_exit(&c->btree_interior_update_pool);
413 mempool_exit(&c->btree_reserve_pool);
414 mempool_exit(&c->fill_iter);
415 percpu_ref_exit(&c->writes);
73e6ab95
KO		 416 kfree(c->replicas.entries);
417 kfree(c->replicas_gc.entries);
1c6fdbd8
KO		 418 kfree(rcu_dereference_protected(c->disk_groups, 1));
419
0519b72d
KO		 420 if (c->journal_reclaim_wq)
421 destroy_workqueue(c->journal_reclaim_wq);
1c6fdbd8
KO		 422 if (c->copygc_wq)
423 destroy_workqueue(c->copygc_wq);
424 if (c->wq)
425 destroy_workqueue(c->wq);
426
427 free_pages((unsigned long) c->disk_sb.sb,
428 c->disk_sb.page_order);
429 kvpfree(c, sizeof(*c));
430 module_put(THIS_MODULE);
431}
432
433static void bch2_fs_release(struct kobject *kobj)
434{
435 struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
436
437 bch2_fs_free(c);
438}
439
440void bch2_fs_stop(struct bch_fs *c)
441{
442 struct bch_dev *ca;
443 unsigned i;
444
af1c6871
KO		 445 bch_verbose(c, "shutting down");
446
1c6fdbd8
KO		 447 for_each_member_device(ca, c, i)
448 if (ca->kobj.state_in_sysfs &&
449 ca->disk_sb.bdev)
450 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
451
452 if (c->kobj.state_in_sysfs)
453 kobject_del(&c->kobj);
454
455 bch2_fs_debug_exit(c);
456 bch2_fs_chardev_exit(c);
457
458 kobject_put(&c->time_stats);
459 kobject_put(&c->opts_dir);
460 kobject_put(&c->internal);
461
462 mutex_lock(&bch_fs_list_lock);
463 list_del(&c->list);
464 mutex_unlock(&bch_fs_list_lock);
465
466 closure_sync(&c->cl);
467 closure_debug_destroy(&c->cl);
468
469 mutex_lock(&c->state_lock);
470 bch2_fs_read_only(c);
471 mutex_unlock(&c->state_lock);
472
473 /* btree prefetch might have kicked off reads in the background: */
474 bch2_btree_flush_all_reads(c);
475
476 for_each_member_device(ca, c, i)
477 cancel_work_sync(&ca->io_error_work);
478
479 cancel_work_sync(&c->btree_write_error_work);
480 cancel_delayed_work_sync(&c->pd_controllers_update);
481 cancel_work_sync(&c->read_only_work);
482
483 for (i = 0; i < c->sb.nr_devices; i++)
484 if (c->devs[i])
485 bch2_dev_free(rcu_dereference_protected(c->devs[i], 1));
486
af1c6871
KO		 487 bch_verbose(c, "shutdown complete");
488
1c6fdbd8
KO		 489 kobject_put(&c->kobj);
490}
491
492static const char *bch2_fs_online(struct bch_fs *c)
493{
494 struct bch_dev *ca;
495 const char *err = NULL;
496 unsigned i;
497 int ret;
498
499 lockdep_assert_held(&bch_fs_list_lock);
500
501 if (!list_empty(&c->list))
502 return NULL;
503
504 if (__bch2_uuid_to_fs(c->sb.uuid))
505 return "filesystem UUID already open";
506
507 ret = bch2_fs_chardev_init(c);
508 if (ret)
509 return "error creating character device";
510
511 bch2_fs_debug_init(c);
512
513 if (kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ||
514 kobject_add(&c->internal, &c->kobj, "internal") ||
515 kobject_add(&c->opts_dir, &c->kobj, "options") ||
516 kobject_add(&c->time_stats, &c->kobj, "time_stats") ||
517 bch2_opts_create_sysfs_files(&c->opts_dir))
518 return "error creating sysfs objects";
519
520 mutex_lock(&c->state_lock);
521
522 err = "error creating sysfs objects";
523 __for_each_member_device(ca, c, i, NULL)
524 if (bch2_dev_sysfs_online(c, ca))
525 goto err;
526
527 list_add(&c->list, &bch_fs_list);
528 err = NULL;
529err:
530 mutex_unlock(&c->state_lock);
531 return err;
532}
533
534static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
535{
536 struct bch_sb_field_members *mi;
537 struct bch_fs *c;
7ef2a73a 538 unsigned i, iter_size, fs_usage_size;
1c6fdbd8
KO		 539 const char *err;
540
541 pr_verbose_init(opts, "");
542
543 c = kvpmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
544 if (!c)
545 goto out;
546
547 __module_get(THIS_MODULE);
548
549 c->minor = -1;
550 c->disk_sb.fs_sb = true;
551
552 mutex_init(&c->state_lock);
553 mutex_init(&c->sb_lock);
554 mutex_init(&c->replicas_gc_lock);
555 mutex_init(&c->btree_root_lock);
556 INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
557
558 init_rwsem(&c->gc_lock);
559
560 for (i = 0; i < BCH_TIME_STAT_NR; i++)
561 bch2_time_stats_init(&c->times[i]);
562
b092dadd
KO		 563 bch2_fs_allocator_background_init(c);
564 bch2_fs_allocator_foreground_init(c);
1c6fdbd8
KO		 565 bch2_fs_rebalance_init(c);
566 bch2_fs_quota_init(c);
567
568 INIT_LIST_HEAD(&c->list);
569
570 INIT_LIST_HEAD(&c->btree_interior_update_list);
571 mutex_init(&c->btree_reserve_cache_lock);
572 mutex_init(&c->btree_interior_update_lock);
573
574 mutex_init(&c->bio_bounce_pages_lock);
575
576 bio_list_init(&c->btree_write_error_list);
577 spin_lock_init(&c->btree_write_error_lock);
578 INIT_WORK(&c->btree_write_error_work, bch2_btree_write_error_work);
579
580 INIT_LIST_HEAD(&c->fsck_errors);
581 mutex_init(&c->fsck_error_lock);
582
cd575ddf
KO		 583 INIT_LIST_HEAD(&c->ec_new_stripe_list);
584 mutex_init(&c->ec_new_stripe_lock);
dfe9bfb3 585 mutex_init(&c->ec_stripe_create_lock);
cd575ddf
KO		 586 spin_lock_init(&c->ec_stripes_heap_lock);
587
1c6fdbd8
KO		 588 seqcount_init(&c->gc_pos_lock);
589
590 c->copy_gc_enabled = 1;
591 c->rebalance.enabled = 1;
592 c->promote_whole_extents = true;
593
594 c->journal.write_time = &c->times[BCH_TIME_journal_write];
595 c->journal.delay_time = &c->times[BCH_TIME_journal_delay];
596 c->journal.blocked_time = &c->times[BCH_TIME_journal_blocked];
597 c->journal.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq];
598
599 bch2_fs_btree_cache_init_early(&c->btree_cache);
600
73e6ab95
KO		 601 if (percpu_init_rwsem(&c->mark_lock))
602 goto err;
603
1c6fdbd8
KO		 604 mutex_lock(&c->sb_lock);
605
606 if (bch2_sb_to_fs(c, sb)) {
607 mutex_unlock(&c->sb_lock);
608 goto err;
609 }
610
611 mutex_unlock(&c->sb_lock);
612
613 scnprintf(c->name, sizeof(c->name), "%pU", &c->sb.user_uuid);
614
615 c->opts = bch2_opts_default;
616 bch2_opts_apply(&c->opts, bch2_opts_from_sb(sb));
617 bch2_opts_apply(&c->opts, opts);
618
619 c->block_bits = ilog2(c->opts.block_size);
620 c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
621
622 c->opts.nochanges |= c->opts.noreplay;
623 c->opts.read_only |= c->opts.nochanges;
624
625 if (bch2_fs_init_fault("fs_alloc"))
626 goto err;
627
628 iter_size = sizeof(struct btree_node_iter_large) +
629 (btree_blocks(c) + 1) * 2 *
630 sizeof(struct btree_node_iter_set);
631
7ef2a73a
KO		 632 fs_usage_size = sizeof(struct bch_fs_usage) +
633 sizeof(u64) * c->replicas.nr;
634
1c6fdbd8
KO		 635 if (!(c->wq = alloc_workqueue("bcachefs",
636 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_HIGHPRI, 1)) ||
637 !(c->copygc_wq = alloc_workqueue("bcache_copygc",
638 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_HIGHPRI, 1)) ||
0519b72d
KO		 639 !(c->journal_reclaim_wq = alloc_workqueue("bcache_journal",
640 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_HIGHPRI, 1)) ||
1c6fdbd8
KO		 641 percpu_ref_init(&c->writes, bch2_writes_disabled, 0, GFP_KERNEL) ||
642 mempool_init_kmalloc_pool(&c->btree_reserve_pool, 1,
643 sizeof(struct btree_reserve)) ||
644 mempool_init_kmalloc_pool(&c->btree_interior_update_pool, 1,
645 sizeof(struct btree_update)) ||
646 mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
647 bioset_init(&c->btree_bio, 1,
648 max(offsetof(struct btree_read_bio, bio),
649 offsetof(struct btree_write_bio, wbio.bio)),
650 BIOSET_NEED_BVECS) ||
7ef2a73a
KO		 651 !(c->usage[0] = __alloc_percpu(fs_usage_size, sizeof(u64))) ||
652 !(c->usage_scratch = __alloc_percpu(fs_usage_size, sizeof(u64))) ||
5663a415 653 !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
1c6fdbd8
KO		 654 mempool_init_kvpmalloc_pool(&c->btree_bounce_pool, 1,
655 btree_bytes(c)) ||
581edb63
KO		 656 mempool_init_kmalloc_pool(&c->btree_iters_pool, 1,
657 sizeof(struct btree_iter) * BTREE_ITER_MAX) ||
1c6fdbd8
KO		 658 bch2_io_clock_init(&c->io_clock[READ]) ||
659 bch2_io_clock_init(&c->io_clock[WRITE]) ||
660 bch2_fs_journal_init(&c->journal) ||
661 bch2_fs_btree_cache_init(c) ||
662 bch2_fs_io_init(c) ||
663 bch2_fs_encryption_init(c) ||
664 bch2_fs_compress_init(c) ||
cd575ddf 665 bch2_fs_ec_init(c) ||
1c6fdbd8
KO		 666 bch2_fs_fsio_init(c))
667 goto err;
668
669 mi = bch2_sb_get_members(c->disk_sb.sb);
670 for (i = 0; i < c->sb.nr_devices; i++)
671 if (bch2_dev_exists(c->disk_sb.sb, mi, i) &&
672 bch2_dev_alloc(c, i))
673 goto err;
674
675 /*
676 * Now that all allocations have succeeded, init various refcounty
677 * things that let us shutdown:
678 */
679 closure_init(&c->cl, NULL);
680
681 c->kobj.kset = bcachefs_kset;
682 kobject_init(&c->kobj, &bch2_fs_ktype);
683 kobject_init(&c->internal, &bch2_fs_internal_ktype);
684 kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
685 kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
686
687 mutex_lock(&bch_fs_list_lock);
688 err = bch2_fs_online(c);
689 mutex_unlock(&bch_fs_list_lock);
690 if (err) {
691 bch_err(c, "bch2_fs_online() error: %s", err);
692 goto err;
693 }
694out:
695 pr_verbose_init(opts, "ret %i", c ? 0 : -ENOMEM);
696 return c;
697err:
698 bch2_fs_free(c);
699 c = NULL;
700 goto out;
701}
702
703const char *bch2_fs_start(struct bch_fs *c)
704{
705 const char *err = "cannot allocate memory";
706 struct bch_sb_field_members *mi;
707 struct bch_dev *ca;
a420eea6 708 time64_t now = ktime_get_real_seconds();
1c6fdbd8
KO		 709 unsigned i;
710 int ret = -EINVAL;
711
712 mutex_lock(&c->state_lock);
713
714 BUG_ON(c->state != BCH_FS_STARTING);
715
716 mutex_lock(&c->sb_lock);
717
718 for_each_online_member(ca, c, i)
719 bch2_sb_from_fs(c, ca);
720
721 mi = bch2_sb_get_members(c->disk_sb.sb);
722 for_each_online_member(ca, c, i)
723 mi->members[ca->dev_idx].last_mount = cpu_to_le64(now);
724
725 mutex_unlock(&c->sb_lock);
726
727 for_each_rw_member(ca, c, i)
728 bch2_dev_allocator_add(c, ca);
729 bch2_recalc_capacity(c);
730
731 ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
732 ? bch2_fs_recovery(c)
733 : bch2_fs_initialize(c);
734 if (ret)
735 goto err;
cd575ddf
KO		 736
737 ret = bch2_opts_check_may_set(c);
738 if (ret)
739 goto err;
1c6fdbd8
KO		 740
741 err = "dynamic fault";
742 if (bch2_fs_init_fault("fs_start"))
743 goto err;
744
745 if (c->opts.read_only) {
746 bch2_fs_read_only(c);
747 } else {
748 err = bch2_fs_read_write(c);
749 if (err)
750 goto err;
751 }
752
753 set_bit(BCH_FS_STARTED, &c->flags);
754
755 err = NULL;
756out:
757 mutex_unlock(&c->state_lock);
758 return err;
759err:
760 switch (ret) {
761 case BCH_FSCK_ERRORS_NOT_FIXED:
762 bch_err(c, "filesystem contains errors: please report this to the developers");
763 pr_cont("mount with -o fix_errors to repair\n");
764 err = "fsck error";
765 break;
766 case BCH_FSCK_REPAIR_UNIMPLEMENTED:
767 bch_err(c, "filesystem contains errors: please report this to the developers");
768 pr_cont("repair unimplemented: inform the developers so that it can be added\n");
769 err = "fsck error";
770 break;
771 case BCH_FSCK_REPAIR_IMPOSSIBLE:
772 bch_err(c, "filesystem contains errors, but repair impossible");
773 err = "fsck error";
774 break;
775 case BCH_FSCK_UNKNOWN_VERSION:
776 err = "unknown metadata version";;
777 break;
778 case -ENOMEM:
779 err = "cannot allocate memory";
780 break;
781 case -EIO:
782 err = "IO error";
783 break;
784 }
785
786 BUG_ON(!err);
787 set_bit(BCH_FS_ERROR, &c->flags);
788 goto out;
789}
790
791static const char *bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
792{
793 struct bch_sb_field_members *sb_mi;
794
795 sb_mi = bch2_sb_get_members(sb);
796 if (!sb_mi)
797 return "Invalid superblock: member info area missing";
798
799 if (le16_to_cpu(sb->block_size) != c->opts.block_size)
800 return "mismatched block size";
801
802 if (le16_to_cpu(sb_mi->members[sb->dev_idx].bucket_size) <
803 BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
804 return "new cache bucket size is too small";
805
806 return NULL;
807}
808
809static const char *bch2_dev_in_fs(struct bch_sb *fs, struct bch_sb *sb)
810{
811 struct bch_sb *newest =
812 le64_to_cpu(fs->seq) > le64_to_cpu(sb->seq) ? fs : sb;
813 struct bch_sb_field_members *mi = bch2_sb_get_members(newest);
814
815 if (!uuid_equal(&fs->uuid, &sb->uuid))
816 return "device not a member of filesystem";
817
818 if (!bch2_dev_exists(newest, mi, sb->dev_idx))
819 return "device has been removed";
820
821 if (fs->block_size != sb->block_size)
822 return "mismatched block size";
823
824 return NULL;
825}
826
827/* Device startup/shutdown: */
828
829static void bch2_dev_release(struct kobject *kobj)
830{
831 struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
832
833 kfree(ca);
834}
835
836static void bch2_dev_free(struct bch_dev *ca)
837{
838 cancel_work_sync(&ca->io_error_work);
839
840 if (ca->kobj.state_in_sysfs &&
841 ca->disk_sb.bdev)
842 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
843
844 if (ca->kobj.state_in_sysfs)
845 kobject_del(&ca->kobj);
846
847 bch2_free_super(&ca->disk_sb);
848 bch2_dev_journal_exit(ca);
849
850 free_percpu(ca->io_done);
851 bioset_exit(&ca->replica_set);
852 bch2_dev_buckets_free(ca);
853
854 bch2_time_stats_exit(&ca->io_latency[WRITE]);
855 bch2_time_stats_exit(&ca->io_latency[READ]);
856
857 percpu_ref_exit(&ca->io_ref);
858 percpu_ref_exit(&ca->ref);
859 kobject_put(&ca->kobj);
860}
861
862static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
863{
864
865 lockdep_assert_held(&c->state_lock);
866
867 if (percpu_ref_is_zero(&ca->io_ref))
868 return;
869
870 __bch2_dev_read_only(c, ca);
871
872 reinit_completion(&ca->io_ref_completion);
873 percpu_ref_kill(&ca->io_ref);
874 wait_for_completion(&ca->io_ref_completion);
875
876 if (ca->kobj.state_in_sysfs) {
877 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
878 sysfs_remove_link(&ca->kobj, "block");
879 }
880
881 bch2_free_super(&ca->disk_sb);
882 bch2_dev_journal_exit(ca);
883}
884
885static void bch2_dev_ref_complete(struct percpu_ref *ref)
886{
887 struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
888
889 complete(&ca->ref_completion);
890}
891
892static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
893{
894 struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
895
896 complete(&ca->io_ref_completion);
897}
898
899static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
900{
901 int ret;
902
903 if (!c->kobj.state_in_sysfs)
904 return 0;
905
906 if (!ca->kobj.state_in_sysfs) {
907 ret = kobject_add(&ca->kobj, &c->kobj,
908 "dev-%u", ca->dev_idx);
909 if (ret)
910 return ret;
911 }
912
913 if (ca->disk_sb.bdev) {
914 struct kobject *block = bdev_kobj(ca->disk_sb.bdev);
915
916 ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
917 if (ret)
918 return ret;
919
920 ret = sysfs_create_link(&ca->kobj, block, "block");
921 if (ret)
922 return ret;
923 }
924
925 return 0;
926}
927
928static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
929 struct bch_member *member)
930{
931 struct bch_dev *ca;
932
933 ca = kzalloc(sizeof(*ca), GFP_KERNEL);
934 if (!ca)
935 return NULL;
936
937 kobject_init(&ca->kobj, &bch2_dev_ktype);
938 init_completion(&ca->ref_completion);
939 init_completion(&ca->io_ref_completion);
940
941 init_rwsem(&ca->bucket_lock);
942
943 writepoint_init(&ca->copygc_write_point, BCH_DATA_USER);
944
945 spin_lock_init(&ca->freelist_lock);
946 bch2_dev_copygc_init(ca);
947
948 INIT_WORK(&ca->io_error_work, bch2_io_error_work);
949
950 bch2_time_stats_init(&ca->io_latency[READ]);
951 bch2_time_stats_init(&ca->io_latency[WRITE]);
952
953 ca->mi = bch2_mi_to_cpu(member);
954 ca->uuid = member->uuid;
955
956 if (opt_defined(c->opts, discard))
957 ca->mi.discard = opt_get(c->opts, discard);
958
959 if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
960 0, GFP_KERNEL) ||
961 percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
962 PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
963 bch2_dev_buckets_alloc(c, ca) ||
964 bioset_init(&ca->replica_set, 4,
965 offsetof(struct bch_write_bio, bio), 0) ||
966 !(ca->io_done = alloc_percpu(*ca->io_done)))
967 goto err;
968
969 return ca;
970err:
971 bch2_dev_free(ca);
972 return NULL;
973}
974
975static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
976 unsigned dev_idx)
977{
978 ca->dev_idx = dev_idx;
979 __set_bit(ca->dev_idx, ca->self.d);
980 scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
981
982 ca->fs = c;
983 rcu_assign_pointer(c->devs[ca->dev_idx], ca);
984
985 if (bch2_dev_sysfs_online(c, ca))
986 pr_warn("error creating sysfs objects");
987}
988
989static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
990{
991 struct bch_member *member =
992 bch2_sb_get_members(c->disk_sb.sb)->members + dev_idx;
993 struct bch_dev *ca = NULL;
994 int ret = 0;
995
996 pr_verbose_init(c->opts, "");
997
998 if (bch2_fs_init_fault("dev_alloc"))
999 goto err;
1000
1001 ca = __bch2_dev_alloc(c, member);
1002 if (!ca)
1003 goto err;
1004
1005 bch2_dev_attach(c, ca, dev_idx);
1006out:
1007 pr_verbose_init(c->opts, "ret %i", ret);
1008 return ret;
1009err:
1010 if (ca)
1011 bch2_dev_free(ca);
1012 ret = -ENOMEM;
1013 goto out;
1014}
1015
1016static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1017{
1018 unsigned ret;
1019
1020 if (bch2_dev_is_online(ca)) {
1021 bch_err(ca, "already have device online in slot %u",
1022 sb->sb->dev_idx);
1023 return -EINVAL;
1024 }
1025
1026 if (get_capacity(sb->bdev->bd_disk) <
1027 ca->mi.bucket_size * ca->mi.nbuckets) {
1028 bch_err(ca, "cannot online: device too small");
1029 return -EINVAL;
1030 }
1031
1032 BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1033
1034 if (get_capacity(sb->bdev->bd_disk) <
1035 ca->mi.bucket_size * ca->mi.nbuckets) {
1036 bch_err(ca, "device too small");
1037 return -EINVAL;
1038 }
1039
1040 ret = bch2_dev_journal_init(ca, sb->sb);
1041 if (ret)
1042 return ret;
1043
1044 /* Commit: */
1045 ca->disk_sb = *sb;
1046 memset(sb, 0, sizeof(*sb));
1047
1c6fdbd8
KO		 1048 percpu_ref_reinit(&ca->io_ref);
1049
1050 return 0;
1051}
1052
1053static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1054{
1055 struct bch_dev *ca;
1056 int ret;
1057
1058 lockdep_assert_held(&c->state_lock);
1059
1060 if (le64_to_cpu(sb->sb->seq) >
1061 le64_to_cpu(c->disk_sb.sb->seq))
1062 bch2_sb_to_fs(c, sb->sb);
1063
1064 BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
1065 !c->devs[sb->sb->dev_idx]);
1066
1067 ca = bch_dev_locked(c, sb->sb->dev_idx);
1068
1069 ret = __bch2_dev_attach_bdev(ca, sb);
1070 if (ret)
1071 return ret;
1072
6eac2c2e 1073 mutex_lock(&c->sb_lock);
9ca53b55 1074 bch2_mark_dev_superblock(ca->fs, ca, 0);
6eac2c2e
KO		 1075 mutex_unlock(&c->sb_lock);
1076
1c6fdbd8
KO		 1077 bch2_dev_sysfs_online(c, ca);
1078
1079 if (c->sb.nr_devices == 1)
1080 snprintf(c->name, sizeof(c->name), "%pg", ca->disk_sb.bdev);
1081 snprintf(ca->name, sizeof(ca->name), "%pg", ca->disk_sb.bdev);
1082
1083 rebalance_wakeup(c);
1084 return 0;
1085}
1086
1087/* Device management: */
1088
1089/*
1090 * Note: this function is also used by the error paths - when a particular
1091 * device sees an error, we call it to determine whether we can just set the
1092 * device RO, or - if this function returns false - we'll set the whole
1093 * filesystem RO:
1094 *
1095 * XXX: maybe we should be more explicit about whether we're changing state
1096 * because we got an error or what have you?
1097 */
1098bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1099 enum bch_member_state new_state, int flags)
1100{
1101 struct bch_devs_mask new_online_devs;
1102 struct replicas_status s;
1103 struct bch_dev *ca2;
1104 int i, nr_rw = 0, required;
1105
1106 lockdep_assert_held(&c->state_lock);
1107
1108 switch (new_state) {
1109 case BCH_MEMBER_STATE_RW:
1110 return true;
1111 case BCH_MEMBER_STATE_RO:
1112 if (ca->mi.state != BCH_MEMBER_STATE_RW)
1113 return true;
1114
1115 /* do we have enough devices to write to? */
1116 for_each_member_device(ca2, c, i)
1117 if (ca2 != ca)
1118 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_RW;
1119
1120 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1121 ? c->opts.metadata_replicas
1122 : c->opts.metadata_replicas_required,
1123 !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1124 ? c->opts.data_replicas
1125 : c->opts.data_replicas_required);
1126
1127 return nr_rw >= required;
1128 case BCH_MEMBER_STATE_FAILED:
1129 case BCH_MEMBER_STATE_SPARE:
1130 if (ca->mi.state != BCH_MEMBER_STATE_RW &&
1131 ca->mi.state != BCH_MEMBER_STATE_RO)
1132 return true;
1133
1134 /* do we have enough devices to read from? */
1135 new_online_devs = bch2_online_devs(c);
1136 __clear_bit(ca->dev_idx, new_online_devs.d);
1137
1138 s = __bch2_replicas_status(c, new_online_devs);
1139
1140 return bch2_have_enough_devs(s, flags);
1141 default:
1142 BUG();
1143 }
1144}
1145
1146static bool bch2_fs_may_start(struct bch_fs *c)
1147{
1148 struct replicas_status s;
1149 struct bch_sb_field_members *mi;
1150 struct bch_dev *ca;
1151 unsigned i, flags = c->opts.degraded
1152 ? BCH_FORCE_IF_DEGRADED
1153 : 0;
1154
1155 if (!c->opts.degraded) {
1156 mutex_lock(&c->sb_lock);
1157 mi = bch2_sb_get_members(c->disk_sb.sb);
1158
1159 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1160 if (!bch2_dev_exists(c->disk_sb.sb, mi, i))
1161 continue;
1162
1163 ca = bch_dev_locked(c, i);
1164
1165 if (!bch2_dev_is_online(ca) &&
1166 (ca->mi.state == BCH_MEMBER_STATE_RW ||
1167 ca->mi.state == BCH_MEMBER_STATE_RO)) {
1168 mutex_unlock(&c->sb_lock);
1169 return false;
1170 }
1171 }
1172 mutex_unlock(&c->sb_lock);
1173 }
1174
1175 s = bch2_replicas_status(c);
1176
1177 return bch2_have_enough_devs(s, flags);
1178}
1179
1180static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1181{
1182 bch2_copygc_stop(ca);
1183
1184 /*
1185 * The allocator thread itself allocates btree nodes, so stop it first:
1186 */
1187 bch2_dev_allocator_stop(ca);
1188 bch2_dev_allocator_remove(c, ca);
1189 bch2_dev_journal_stop(&c->journal, ca);
1190}
1191
1192static const char *__bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1193{
1194 lockdep_assert_held(&c->state_lock);
1195
1196 BUG_ON(ca->mi.state != BCH_MEMBER_STATE_RW);
1197
1198 bch2_dev_allocator_add(c, ca);
1199 bch2_recalc_capacity(c);
1200
1201 if (bch2_dev_allocator_start(ca))
1202 return "error starting allocator thread";
1203
1204 if (bch2_copygc_start(c, ca))
1205 return "error starting copygc thread";
1206
1207 return NULL;
1208}
1209
1210int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1211 enum bch_member_state new_state, int flags)
1212{
1213 struct bch_sb_field_members *mi;
1214 int ret = 0;
1215
1216 if (ca->mi.state == new_state)
1217 return 0;
1218
1219 if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1220 return -EINVAL;
1221
1222 if (new_state != BCH_MEMBER_STATE_RW)
1223 __bch2_dev_read_only(c, ca);
1224
1225 bch_notice(ca, "%s", bch2_dev_state[new_state]);
1226
1227 mutex_lock(&c->sb_lock);
1228 mi = bch2_sb_get_members(c->disk_sb.sb);
1229 SET_BCH_MEMBER_STATE(&mi->members[ca->dev_idx], new_state);
1230 bch2_write_super(c);
1231 mutex_unlock(&c->sb_lock);
1232
1233 if (new_state == BCH_MEMBER_STATE_RW &&
1234 __bch2_dev_read_write(c, ca))
1235 ret = -ENOMEM;
1236
1237 rebalance_wakeup(c);
1238
1239 return ret;
1240}
1241
1242int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1243 enum bch_member_state new_state, int flags)
1244{
1245 int ret;
1246
1247 mutex_lock(&c->state_lock);
1248 ret = __bch2_dev_set_state(c, ca, new_state, flags);
1249 mutex_unlock(&c->state_lock);
1250
1251 return ret;
1252}
1253
1254/* Device add/removal: */
1255
1256int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1257{
1258 struct bch_sb_field_members *mi;
1259 unsigned dev_idx = ca->dev_idx, data;
1260 int ret = -EINVAL;
1261
1262 mutex_lock(&c->state_lock);
1263
1264 percpu_ref_put(&ca->ref); /* XXX */
1265
1266 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_FAILED, flags)) {
1267 bch_err(ca, "Cannot remove without losing data");
1268 goto err;
1269 }
1270
1271 __bch2_dev_read_only(c, ca);
1272
1273 /*
1274 * XXX: verify that dev_idx is really not in use anymore, anywhere
1275 *
1276 * flag_data_bad() does not check btree pointers
1277 */
1278 ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1279 if (ret) {
1280 bch_err(ca, "Remove failed: error %i dropping data", ret);
1281 goto err;
1282 }
1283
1284 ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1285 if (ret) {
1286 bch_err(ca, "Remove failed: error %i flushing journal", ret);
1287 goto err;
1288 }
1289
1290 data = bch2_dev_has_data(c, ca);
1291 if (data) {
1292 char data_has_str[100];
319f9ac3 1293
dbaee468
KO		 1294 bch2_flags_to_text(&PBUF(data_has_str),
1295 bch2_data_types, data);
1c6fdbd8
KO		 1296 bch_err(ca, "Remove failed, still has data (%s)", data_has_str);
1297 ret = -EBUSY;
1298 goto err;
1299 }
1300
1301 ret = bch2_btree_delete_range(c, BTREE_ID_ALLOC,
1302 POS(ca->dev_idx, 0),
1303 POS(ca->dev_idx + 1, 0),
fc3268c1 1304 NULL);
1c6fdbd8
KO		 1305 if (ret) {
1306 bch_err(ca, "Remove failed, error deleting alloc info");
1307 goto err;
1308 }
1309
1310 /*
1311 * must flush all existing journal entries, they might have
1312 * (overwritten) keys that point to the device we're removing:
1313 */
1314 bch2_journal_flush_all_pins(&c->journal);
1315 ret = bch2_journal_error(&c->journal);
1316 if (ret) {
1317 bch_err(ca, "Remove failed, journal error");
1318 goto err;
1319 }
1320
1321 __bch2_dev_offline(c, ca);
1322
1323 mutex_lock(&c->sb_lock);
1324 rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1325 mutex_unlock(&c->sb_lock);
1326
1327 percpu_ref_kill(&ca->ref);
1328 wait_for_completion(&ca->ref_completion);
1329
1330 bch2_dev_free(ca);
1331
1332 /*
1333 * Free this device's slot in the bch_member array - all pointers to
1334 * this device must be gone:
1335 */
1336 mutex_lock(&c->sb_lock);
1337 mi = bch2_sb_get_members(c->disk_sb.sb);
1338 memset(&mi->members[dev_idx].uuid, 0, sizeof(mi->members[dev_idx].uuid));
1339
1340 bch2_write_super(c);
1341
1342 mutex_unlock(&c->sb_lock);
1343 mutex_unlock(&c->state_lock);
1344 return 0;
1345err:
d3bb629d
KO		 1346 if (ca->mi.state == BCH_MEMBER_STATE_RW &&
1347 !percpu_ref_is_zero(&ca->io_ref))
1c6fdbd8
KO		 1348 __bch2_dev_read_write(c, ca);
1349 mutex_unlock(&c->state_lock);
1350 return ret;
1351}
1352
6eac2c2e
KO		 1353static void dev_usage_clear(struct bch_dev *ca)
1354{
1355 struct bucket_array *buckets;
1356 int cpu;
1357
1358 for_each_possible_cpu(cpu) {
1359 struct bch_dev_usage *p =
9ca53b55 1360 per_cpu_ptr(ca->usage[0], cpu);
6eac2c2e
KO		 1361 memset(p, 0, sizeof(*p));
1362 }
1363
1364 down_read(&ca->bucket_lock);
1365 buckets = bucket_array(ca);
1366
1367 memset(buckets->b, 0, sizeof(buckets->b[0]) * buckets->nbuckets);
1368 up_read(&ca->bucket_lock);
1369}
1370
1c6fdbd8
KO		 1371/* Add new device to running filesystem: */
1372int bch2_dev_add(struct bch_fs *c, const char *path)
1373{
1374 struct bch_opts opts = bch2_opts_empty();
1375 struct bch_sb_handle sb;
1376 const char *err;
1377 struct bch_dev *ca = NULL;
1378 struct bch_sb_field_members *mi;
1379 struct bch_member dev_mi;
1380 unsigned dev_idx, nr_devices, u64s;
1381 int ret;
1382
1383 ret = bch2_read_super(path, &opts, &sb);
1384 if (ret)
1385 return ret;
1386
1387 err = bch2_sb_validate(&sb);
1388 if (err)
1389 return -EINVAL;
1390
1391 dev_mi = bch2_sb_get_members(sb.sb)->members[sb.sb->dev_idx];
1392
1393 err = bch2_dev_may_add(sb.sb, c);
1394 if (err)
1395 return -EINVAL;
1396
1397 ca = __bch2_dev_alloc(c, &dev_mi);
1398 if (!ca) {
1399 bch2_free_super(&sb);
1400 return -ENOMEM;
1401 }
1402
1403 ret = __bch2_dev_attach_bdev(ca, &sb);
1404 if (ret) {
1405 bch2_dev_free(ca);
1406 return ret;
1407 }
1408
6eac2c2e
KO		 1409 /*
1410 * We want to allocate journal on the new device before adding the new
1411 * device to the filesystem because allocating after we attach requires
1412 * spinning up the allocator thread, and the allocator thread requires
1413 * doing btree writes, which if the existing devices are RO isn't going
1414 * to work
1415 *
1416 * So we have to mark where the superblocks are, but marking allocated
1417 * data normally updates the filesystem usage too, so we have to mark,
1418 * allocate the journal, reset all the marks, then remark after we
1419 * attach...
1420 */
9ca53b55 1421 bch2_mark_dev_superblock(ca->fs, ca, 0);
6eac2c2e 1422
1c6fdbd8
KO		 1423 err = "journal alloc failed";
1424 ret = bch2_dev_journal_alloc(ca);
1425 if (ret)
1426 goto err;
1427
6eac2c2e
KO		 1428 dev_usage_clear(ca);
1429
1c6fdbd8
KO		 1430 mutex_lock(&c->state_lock);
1431 mutex_lock(&c->sb_lock);
1432
1433 err = "insufficient space in new superblock";
1434 ret = bch2_sb_from_fs(c, ca);
1435 if (ret)
1436 goto err_unlock;
1437
1438 mi = bch2_sb_get_members(ca->disk_sb.sb);
1439
1440 if (!bch2_sb_resize_members(&ca->disk_sb,
1441 le32_to_cpu(mi->field.u64s) +
1442 sizeof(dev_mi) / sizeof(u64))) {
1443 ret = -ENOSPC;
1444 goto err_unlock;
1445 }
1446
1447 if (dynamic_fault("bcachefs:add:no_slot"))
1448 goto no_slot;
1449
1450 mi = bch2_sb_get_members(c->disk_sb.sb);
1451 for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++)
1452 if (!bch2_dev_exists(c->disk_sb.sb, mi, dev_idx))
1453 goto have_slot;
1454no_slot:
1455 err = "no slots available in superblock";
1456 ret = -ENOSPC;
1457 goto err_unlock;
1458
1459have_slot:
1460 nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1461 u64s = (sizeof(struct bch_sb_field_members) +
1462 sizeof(struct bch_member) * nr_devices) / sizeof(u64);
1463
1464 err = "no space in superblock for member info";
1465 ret = -ENOSPC;
1466
1467 mi = bch2_sb_resize_members(&c->disk_sb, u64s);
1468 if (!mi)
1469 goto err_unlock;
1470
1471 /* success: */
1472
1473 mi->members[dev_idx] = dev_mi;
a420eea6 1474 mi->members[dev_idx].last_mount = cpu_to_le64(ktime_get_real_seconds());
1c6fdbd8
KO		 1475 c->disk_sb.sb->nr_devices = nr_devices;
1476
1477 ca->disk_sb.sb->dev_idx = dev_idx;
1478 bch2_dev_attach(c, ca, dev_idx);
1479
9ca53b55 1480 bch2_mark_dev_superblock(c, ca, 0);
6eac2c2e 1481
1c6fdbd8
KO		 1482 bch2_write_super(c);
1483 mutex_unlock(&c->sb_lock);
1484
1485 if (ca->mi.state == BCH_MEMBER_STATE_RW) {
1486 err = __bch2_dev_read_write(c, ca);
1487 if (err)
1488 goto err_late;
1489 }
1490
1491 mutex_unlock(&c->state_lock);
1492 return 0;
1493
1494err_unlock:
1495 mutex_unlock(&c->sb_lock);
1496 mutex_unlock(&c->state_lock);
1497err:
1498 if (ca)
1499 bch2_dev_free(ca);
1500 bch2_free_super(&sb);
1501 bch_err(c, "Unable to add device: %s", err);
1502 return ret;
1503err_late:
1504 bch_err(c, "Error going rw after adding device: %s", err);
1505 return -EINVAL;
1506}
1507
1508/* Hot add existing device to running filesystem: */
1509int bch2_dev_online(struct bch_fs *c, const char *path)
1510{
1511 struct bch_opts opts = bch2_opts_empty();
1512 struct bch_sb_handle sb = { NULL };
1513 struct bch_sb_field_members *mi;
1514 struct bch_dev *ca;
1515 unsigned dev_idx;
1516 const char *err;
1517 int ret;
1518
1519 mutex_lock(&c->state_lock);
1520
1521 ret = bch2_read_super(path, &opts, &sb);
1522 if (ret) {
1523 mutex_unlock(&c->state_lock);
1524 return ret;
1525 }
1526
1527 dev_idx = sb.sb->dev_idx;
1528
1529 err = bch2_dev_in_fs(c->disk_sb.sb, sb.sb);
1530 if (err)
1531 goto err;
1532
1533 if (bch2_dev_attach_bdev(c, &sb)) {
1534 err = "bch2_dev_attach_bdev() error";
1535 goto err;
1536 }
1537
1538 ca = bch_dev_locked(c, dev_idx);
1539 if (ca->mi.state == BCH_MEMBER_STATE_RW) {
1540 err = __bch2_dev_read_write(c, ca);
1541 if (err)
1542 goto err;
1543 }
1544
1545 mutex_lock(&c->sb_lock);
1546 mi = bch2_sb_get_members(c->disk_sb.sb);
1547
1548 mi->members[ca->dev_idx].last_mount =
a420eea6 1549 cpu_to_le64(ktime_get_real_seconds());
1c6fdbd8
KO		 1550
1551 bch2_write_super(c);
1552 mutex_unlock(&c->sb_lock);
1553
1554 mutex_unlock(&c->state_lock);
1555 return 0;
1556err:
1557 mutex_unlock(&c->state_lock);
1558 bch2_free_super(&sb);
1559 bch_err(c, "error bringing %s online: %s", path, err);
1560 return -EINVAL;
1561}
1562
1563int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1564{
1565 mutex_lock(&c->state_lock);
1566
1567 if (!bch2_dev_is_online(ca)) {
1568 bch_err(ca, "Already offline");
1569 mutex_unlock(&c->state_lock);
1570 return 0;
1571 }
1572
1573 if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_FAILED, flags)) {
1574 bch_err(ca, "Cannot offline required disk");
1575 mutex_unlock(&c->state_lock);
1576 return -EINVAL;
1577 }
1578
1579 __bch2_dev_offline(c, ca);
1580
1581 mutex_unlock(&c->state_lock);
1582 return 0;
1583}
1584
1585int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1586{
1587 struct bch_member *mi;
1588 int ret = 0;
1589
1590 mutex_lock(&c->state_lock);
1591
1592 if (nbuckets < ca->mi.nbuckets) {
1593 bch_err(ca, "Cannot shrink yet");
1594 ret = -EINVAL;
1595 goto err;
1596 }
1597
1598 if (bch2_dev_is_online(ca) &&
1599 get_capacity(ca->disk_sb.bdev->bd_disk) <
1600 ca->mi.bucket_size * nbuckets) {
1601 bch_err(ca, "New size larger than device");
1602 ret = -EINVAL;
1603 goto err;
1604 }
1605
1606 ret = bch2_dev_buckets_resize(c, ca, nbuckets);
1607 if (ret) {
1608 bch_err(ca, "Resize error: %i", ret);
1609 goto err;
1610 }
1611
1612 mutex_lock(&c->sb_lock);
1613 mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
1614 mi->nbuckets = cpu_to_le64(nbuckets);
1615
1616 bch2_write_super(c);
1617 mutex_unlock(&c->sb_lock);
1618
1619 bch2_recalc_capacity(c);
1620err:
1621 mutex_unlock(&c->state_lock);
1622 return ret;
1623}
1624
1625/* return with ref on ca->ref: */
1626struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *path)
1627{
1628
1629 struct bch_dev *ca;
1630 dev_t dev;
1631 unsigned i;
1632 int ret;
1633
1634 ret = lookup_bdev(path, &dev);
1635 if (ret)
1636 return ERR_PTR(ret);
1637
1638 for_each_member_device(ca, c, i)
1639 if (ca->disk_sb.bdev->bd_dev == dev)
1640 goto found;
1641
1642 ca = ERR_PTR(-ENOENT);
1643found:
1644 return ca;
1645}
1646
1647/* Filesystem open: */
1648
1649struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
1650 struct bch_opts opts)
1651{
1652 struct bch_sb_handle *sb = NULL;
1653 struct bch_fs *c = NULL;
1654 unsigned i, best_sb = 0;
1655 const char *err;
1656 int ret = -ENOMEM;
1657
1658 pr_verbose_init(opts, "");
1659
1660 if (!nr_devices) {
1661 c = ERR_PTR(-EINVAL);
1662 goto out2;
1663 }
1664
1665 if (!try_module_get(THIS_MODULE)) {
1666 c = ERR_PTR(-ENODEV);
1667 goto out2;
1668 }
1669
1670 sb = kcalloc(nr_devices, sizeof(*sb), GFP_KERNEL);
1671 if (!sb)
1672 goto err;
1673
1674 for (i = 0; i < nr_devices; i++) {
1675 ret = bch2_read_super(devices[i], &opts, &sb[i]);
1676 if (ret)
1677 goto err;
1678
1679 err = bch2_sb_validate(&sb[i]);
1680 if (err)
1681 goto err_print;
1682 }
1683
1684 for (i = 1; i < nr_devices; i++)
1685 if (le64_to_cpu(sb[i].sb->seq) >
1686 le64_to_cpu(sb[best_sb].sb->seq))
1687 best_sb = i;
1688
1689 for (i = 0; i < nr_devices; i++) {
1690 err = bch2_dev_in_fs(sb[best_sb].sb, sb[i].sb);
1691 if (err)
1692 goto err_print;
1693 }
1694
1695 ret = -ENOMEM;
1696 c = bch2_fs_alloc(sb[best_sb].sb, opts);
1697 if (!c)
1698 goto err;
1699
1700 err = "bch2_dev_online() error";
1701 mutex_lock(&c->state_lock);
1702 for (i = 0; i < nr_devices; i++)
1703 if (bch2_dev_attach_bdev(c, &sb[i])) {
1704 mutex_unlock(&c->state_lock);
1705 goto err_print;
1706 }
1707 mutex_unlock(&c->state_lock);
1708
1709 err = "insufficient devices";
1710 if (!bch2_fs_may_start(c))
1711 goto err_print;
1712
1713 if (!c->opts.nostart) {
1714 err = bch2_fs_start(c);
1715 if (err)
1716 goto err_print;
1717 }
1718out:
1719 kfree(sb);
1720 module_put(THIS_MODULE);
1721out2:
1722 pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
1723 return c;
1724err_print:
1725 pr_err("bch_fs_open err opening %s: %s",
1726 devices[0], err);
1727 ret = -EINVAL;
1728err:
1729 if (c)
1730 bch2_fs_stop(c);
1731 for (i = 0; i < nr_devices; i++)
1732 bch2_free_super(&sb[i]);
1733 c = ERR_PTR(ret);
1734 goto out;
1735}
1736
1737static const char *__bch2_fs_open_incremental(struct bch_sb_handle *sb,
1738 struct bch_opts opts)
1739{
1740 const char *err;
1741 struct bch_fs *c;
1742 bool allocated_fs = false;
1743
1744 err = bch2_sb_validate(sb);
1745 if (err)
1746 return err;
1747
1748 mutex_lock(&bch_fs_list_lock);
1749 c = __bch2_uuid_to_fs(sb->sb->uuid);
1750 if (c) {
1751 closure_get(&c->cl);
1752
1753 err = bch2_dev_in_fs(c->disk_sb.sb, sb->sb);
1754 if (err)
1755 goto err;
1756 } else {
1757 c = bch2_fs_alloc(sb->sb, opts);
1758 err = "cannot allocate memory";
1759 if (!c)
1760 goto err;
1761
1762 allocated_fs = true;
1763 }
1764
1765 err = "bch2_dev_online() error";
1766
1767 mutex_lock(&c->sb_lock);
1768 if (bch2_dev_attach_bdev(c, sb)) {
1769 mutex_unlock(&c->sb_lock);
1770 goto err;
1771 }
1772 mutex_unlock(&c->sb_lock);
1773
1774 if (!c->opts.nostart && bch2_fs_may_start(c)) {
1775 err = bch2_fs_start(c);
1776 if (err)
1777 goto err;
1778 }
1779
1780 closure_put(&c->cl);
1781 mutex_unlock(&bch_fs_list_lock);
1782
1783 return NULL;
1784err:
1785 mutex_unlock(&bch_fs_list_lock);
1786
1787 if (allocated_fs)
1788 bch2_fs_stop(c);
1789 else if (c)
1790 closure_put(&c->cl);
1791
1792 return err;
1793}
1794
1795const char *bch2_fs_open_incremental(const char *path)
1796{
1797 struct bch_sb_handle sb;
1798 struct bch_opts opts = bch2_opts_empty();
1799 const char *err;
1800
1801 if (bch2_read_super(path, &opts, &sb))
1802 return "error reading superblock";
1803
1804 err = __bch2_fs_open_incremental(&sb, opts);
1805 bch2_free_super(&sb);
1806
1807 return err;
1808}
1809
1810/* Global interfaces/init */
1811
1812static void bcachefs_exit(void)
1813{
1814 bch2_debug_exit();
1815 bch2_vfs_exit();
1816 bch2_chardev_exit();
1817 if (bcachefs_kset)
1818 kset_unregister(bcachefs_kset);
1819}
1820
1821static int __init bcachefs_init(void)
1822{
1823 bch2_bkey_pack_test();
1824 bch2_inode_pack_test();
1825
1826 if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
1827 bch2_chardev_init() ||
1828 bch2_vfs_init() ||
1829 bch2_debug_init())
1830 goto err;
1831
1832 return 0;
1833err:
1834 bcachefs_exit();
1835 return -ENOMEM;
1836}
1837
1838#define BCH_DEBUG_PARAM(name, description) \
1839 bool bch2_##name; \
1840 module_param_named(name, bch2_##name, bool, 0644); \
1841 MODULE_PARM_DESC(name, description);
1842BCH_DEBUG_PARAMS()
1843#undef BCH_DEBUG_PARAM
1844
26609b61 1845unsigned bch2_metadata_version = bcachefs_metadata_version_current;
1c6fdbd8
KO		 1846module_param_named(version, bch2_metadata_version, uint, 0400);
1847
1848module_exit(bcachefs_exit);
1849module_init(bcachefs_init);
Linux 6.x block layer and io_uring tree(s)