]> git.sesse.net Git - bcachefs-tools-debian/blob - libbcachefs/super.c
1401cb576fa2fbeb510b85107314491dbee58c5c
[bcachefs-tools-debian] / libbcachefs / super.c
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"
11 #include "alloc_background.h"
12 #include "alloc_foreground.h"
13 #include "bkey_sort.h"
14 #include "btree_cache.h"
15 #include "btree_gc.h"
16 #include "btree_key_cache.h"
17 #include "btree_update_interior.h"
18 #include "btree_io.h"
19 #include "buckets_waiting_for_journal.h"
20 #include "chardev.h"
21 #include "checksum.h"
22 #include "clock.h"
23 #include "compress.h"
24 #include "debug.h"
25 #include "disk_groups.h"
26 #include "ec.h"
27 #include "error.h"
28 #include "fs.h"
29 #include "fs-io.h"
30 #include "fsck.h"
31 #include "inode.h"
32 #include "io.h"
33 #include "journal.h"
34 #include "journal_reclaim.h"
35 #include "journal_seq_blacklist.h"
36 #include "move.h"
37 #include "migrate.h"
38 #include "movinggc.h"
39 #include "quota.h"
40 #include "rebalance.h"
41 #include "recovery.h"
42 #include "replicas.h"
43 #include "subvolume.h"
44 #include "super.h"
45 #include "super-io.h"
46 #include "sysfs.h"
47
48 #include <linux/backing-dev.h>
49 #include <linux/blkdev.h>
50 #include <linux/debugfs.h>
51 #include <linux/device.h>
52 #include <linux/genhd.h>
53 #include <linux/idr.h>
54 #include <linux/module.h>
55 #include <linux/percpu.h>
56 #include <linux/random.h>
57 #include <linux/sysfs.h>
58 #include <crypto/hash.h>
59
60 #include <trace/events/bcachefs.h>
61
62 MODULE_LICENSE("GPL");
63 MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
64
65 #define KTYPE(type)                                                     \
66 struct kobj_type type ## _ktype = {                                     \
67         .release        = type ## _release,                             \
68         .sysfs_ops      = &type ## _sysfs_ops,                          \
69         .default_attrs  = type ## _files                                \
70 }
71
72 static void bch2_fs_release(struct kobject *);
73 static void bch2_dev_release(struct kobject *);
74
75 static void bch2_fs_internal_release(struct kobject *k)
76 {
77 }
78
79 static void bch2_fs_opts_dir_release(struct kobject *k)
80 {
81 }
82
83 static void bch2_fs_time_stats_release(struct kobject *k)
84 {
85 }
86
87 static KTYPE(bch2_fs);
88 static KTYPE(bch2_fs_internal);
89 static KTYPE(bch2_fs_opts_dir);
90 static KTYPE(bch2_fs_time_stats);
91 static KTYPE(bch2_dev);
92
93 static struct kset *bcachefs_kset;
94 static LIST_HEAD(bch_fs_list);
95 static DEFINE_MUTEX(bch_fs_list_lock);
96
97 static DECLARE_WAIT_QUEUE_HEAD(bch_read_only_wait);
98
99 static void bch2_dev_free(struct bch_dev *);
100 static int bch2_dev_alloc(struct bch_fs *, unsigned);
101 static int bch2_dev_sysfs_online(struct bch_fs *, struct bch_dev *);
102 static void __bch2_dev_read_only(struct bch_fs *, struct bch_dev *);
103
104 struct bch_fs *bch2_dev_to_fs(dev_t dev)
105 {
106         struct bch_fs *c;
107         struct bch_dev *ca;
108         unsigned i;
109
110         mutex_lock(&bch_fs_list_lock);
111         rcu_read_lock();
112
113         list_for_each_entry(c, &bch_fs_list, list)
114                 for_each_member_device_rcu(ca, c, i, NULL)
115                         if (ca->disk_sb.bdev && ca->disk_sb.bdev->bd_dev == dev) {
116                                 closure_get(&c->cl);
117                                 goto found;
118                         }
119         c = NULL;
120 found:
121         rcu_read_unlock();
122         mutex_unlock(&bch_fs_list_lock);
123
124         return c;
125 }
126
127 static struct bch_fs *__bch2_uuid_to_fs(uuid_le uuid)
128 {
129         struct bch_fs *c;
130
131         lockdep_assert_held(&bch_fs_list_lock);
132
133         list_for_each_entry(c, &bch_fs_list, list)
134                 if (!memcmp(&c->disk_sb.sb->uuid, &uuid, sizeof(uuid_le)))
135                         return c;
136
137         return NULL;
138 }
139
140 struct bch_fs *bch2_uuid_to_fs(uuid_le uuid)
141 {
142         struct bch_fs *c;
143
144         mutex_lock(&bch_fs_list_lock);
145         c = __bch2_uuid_to_fs(uuid);
146         if (c)
147                 closure_get(&c->cl);
148         mutex_unlock(&bch_fs_list_lock);
149
150         return c;
151 }
152
153 static void bch2_dev_usage_journal_reserve(struct bch_fs *c)
154 {
155         struct bch_dev *ca;
156         unsigned i, nr = 0, u64s =
157                 ((sizeof(struct jset_entry_dev_usage) +
158                   sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR)) /
159                 sizeof(u64);
160
161         rcu_read_lock();
162         for_each_member_device_rcu(ca, c, i, NULL)
163                 nr++;
164         rcu_read_unlock();
165
166         bch2_journal_entry_res_resize(&c->journal,
167                         &c->dev_usage_journal_res, u64s * nr);
168 }
169
170 /* Filesystem RO/RW: */
171
172 /*
173  * For startup/shutdown of RW stuff, the dependencies are:
174  *
175  * - foreground writes depend on copygc and rebalance (to free up space)
176  *
177  * - copygc and rebalance depend on mark and sweep gc (they actually probably
178  *   don't because they either reserve ahead of time or don't block if
179  *   allocations fail, but allocations can require mark and sweep gc to run
180  *   because of generation number wraparound)
181  *
182  * - all of the above depends on the allocator threads
183  *
184  * - allocator depends on the journal (when it rewrites prios and gens)
185  */
186
187 static void __bch2_fs_read_only(struct bch_fs *c)
188 {
189         struct bch_dev *ca;
190         unsigned i, clean_passes = 0;
191
192         bch2_rebalance_stop(c);
193         bch2_copygc_stop(c);
194         bch2_gc_thread_stop(c);
195
196         /*
197          * Flush journal before stopping allocators, because flushing journal
198          * blacklist entries involves allocating new btree nodes:
199          */
200         bch2_journal_flush_all_pins(&c->journal);
201
202         bch_verbose(c, "flushing journal and stopping allocators");
203
204         bch2_journal_flush_all_pins(&c->journal);
205
206         do {
207                 clean_passes++;
208
209                 if (bch2_journal_flush_all_pins(&c->journal))
210                         clean_passes = 0;
211
212                 /*
213                  * In flight interior btree updates will generate more journal
214                  * updates and btree updates (alloc btree):
215                  */
216                 if (bch2_btree_interior_updates_nr_pending(c)) {
217                         closure_wait_event(&c->btree_interior_update_wait,
218                                            !bch2_btree_interior_updates_nr_pending(c));
219                         clean_passes = 0;
220                 }
221                 flush_work(&c->btree_interior_update_work);
222
223                 if (bch2_journal_flush_all_pins(&c->journal))
224                         clean_passes = 0;
225         } while (clean_passes < 2);
226         bch_verbose(c, "flushing journal and stopping allocators complete");
227
228         set_bit(BCH_FS_ALLOC_CLEAN, &c->flags);
229
230         closure_wait_event(&c->btree_interior_update_wait,
231                            !bch2_btree_interior_updates_nr_pending(c));
232         flush_work(&c->btree_interior_update_work);
233
234         bch2_fs_journal_stop(&c->journal);
235
236         /*
237          * the journal kicks off btree writes via reclaim - wait for in flight
238          * writes after stopping journal:
239          */
240         bch2_btree_flush_all_writes(c);
241
242         /*
243          * After stopping journal:
244          */
245         for_each_member_device(ca, c, i)
246                 bch2_dev_allocator_remove(c, ca);
247 }
248
249 static void bch2_writes_disabled(struct percpu_ref *writes)
250 {
251         struct bch_fs *c = container_of(writes, struct bch_fs, writes);
252
253         set_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
254         wake_up(&bch_read_only_wait);
255 }
256
257 void bch2_fs_read_only(struct bch_fs *c)
258 {
259         if (!test_bit(BCH_FS_RW, &c->flags)) {
260                 bch2_journal_reclaim_stop(&c->journal);
261                 return;
262         }
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         percpu_ref_kill(&c->writes);
271
272         cancel_work_sync(&c->ec_stripe_delete_work);
273
274         /*
275          * If we're not doing an emergency shutdown, we want to wait on
276          * outstanding writes to complete so they don't see spurious errors due
277          * to shutting down the allocator:
278          *
279          * If we are doing an emergency shutdown outstanding writes may
280          * hang until we shutdown the allocator so we don't want to wait
281          * on outstanding writes before shutting everything down - but
282          * we do need to wait on them before returning and signalling
283          * that going RO is complete:
284          */
285         wait_event(bch_read_only_wait,
286                    test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags) ||
287                    test_bit(BCH_FS_EMERGENCY_RO, &c->flags));
288
289         __bch2_fs_read_only(c);
290
291         wait_event(bch_read_only_wait,
292                    test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags));
293
294         clear_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags);
295
296         if (!bch2_journal_error(&c->journal) &&
297             !test_bit(BCH_FS_ERROR, &c->flags) &&
298             !test_bit(BCH_FS_EMERGENCY_RO, &c->flags) &&
299             test_bit(BCH_FS_STARTED, &c->flags) &&
300             test_bit(BCH_FS_ALLOC_CLEAN, &c->flags) &&
301             !c->opts.norecovery) {
302                 bch_verbose(c, "marking filesystem clean");
303                 bch2_fs_mark_clean(c);
304         }
305
306         clear_bit(BCH_FS_RW, &c->flags);
307 }
308
309 static 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         down_write(&c->state_lock);
315         bch2_fs_read_only(c);
316         up_write(&c->state_lock);
317 }
318
319 static void bch2_fs_read_only_async(struct bch_fs *c)
320 {
321         queue_work(system_long_wq, &c->read_only_work);
322 }
323
324 bool 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_journal_halt(&c->journal);
329         bch2_fs_read_only_async(c);
330
331         wake_up(&bch_read_only_wait);
332         return ret;
333 }
334
335 static int bch2_fs_read_write_late(struct bch_fs *c)
336 {
337         int ret;
338
339         ret = bch2_gc_thread_start(c);
340         if (ret) {
341                 bch_err(c, "error starting gc thread");
342                 return ret;
343         }
344
345         ret = bch2_copygc_start(c);
346         if (ret) {
347                 bch_err(c, "error starting copygc thread");
348                 return ret;
349         }
350
351         ret = bch2_rebalance_start(c);
352         if (ret) {
353                 bch_err(c, "error starting rebalance thread");
354                 return ret;
355         }
356
357         schedule_work(&c->ec_stripe_delete_work);
358
359         return 0;
360 }
361
362 static int __bch2_fs_read_write(struct bch_fs *c, bool early)
363 {
364         struct bch_dev *ca;
365         unsigned i;
366         int ret;
367
368         if (test_bit(BCH_FS_INITIAL_GC_UNFIXED, &c->flags)) {
369                 bch_err(c, "cannot go rw, unfixed btree errors");
370                 return -EROFS;
371         }
372
373         if (test_bit(BCH_FS_RW, &c->flags))
374                 return 0;
375
376         /*
377          * nochanges is used for fsck -n mode - we have to allow going rw
378          * during recovery for that to work:
379          */
380         if (c->opts.norecovery ||
381             (c->opts.nochanges &&
382              (!early || c->opts.read_only)))
383                 return -EROFS;
384
385         bch_info(c, "going read-write");
386
387         ret = bch2_fs_mark_dirty(c);
388         if (ret)
389                 goto err;
390
391         clear_bit(BCH_FS_ALLOC_CLEAN, &c->flags);
392
393         for_each_rw_member(ca, c, i)
394                 bch2_dev_allocator_add(c, ca);
395         bch2_recalc_capacity(c);
396
397         bch2_do_discards(c);
398
399         if (!early) {
400                 ret = bch2_fs_read_write_late(c);
401                 if (ret)
402                         goto err;
403         }
404
405         percpu_ref_reinit(&c->writes);
406         set_bit(BCH_FS_RW, &c->flags);
407         set_bit(BCH_FS_WAS_RW, &c->flags);
408         return 0;
409 err:
410         __bch2_fs_read_only(c);
411         return ret;
412 }
413
414 int bch2_fs_read_write(struct bch_fs *c)
415 {
416         return __bch2_fs_read_write(c, false);
417 }
418
419 int bch2_fs_read_write_early(struct bch_fs *c)
420 {
421         lockdep_assert_held(&c->state_lock);
422
423         return __bch2_fs_read_write(c, true);
424 }
425
426 /* Filesystem startup/shutdown: */
427
428 static void __bch2_fs_free(struct bch_fs *c)
429 {
430         unsigned i;
431         int cpu;
432
433         for (i = 0; i < BCH_TIME_STAT_NR; i++)
434                 bch2_time_stats_exit(&c->times[i]);
435
436         bch2_fs_snapshots_exit(c);
437         bch2_fs_quota_exit(c);
438         bch2_fs_fsio_exit(c);
439         bch2_fs_ec_exit(c);
440         bch2_fs_encryption_exit(c);
441         bch2_fs_io_exit(c);
442         bch2_fs_buckets_waiting_for_journal_exit(c);
443         bch2_fs_btree_interior_update_exit(c);
444         bch2_fs_btree_iter_exit(c);
445         bch2_fs_btree_key_cache_exit(&c->btree_key_cache);
446         bch2_fs_btree_cache_exit(c);
447         bch2_fs_replicas_exit(c);
448         bch2_fs_journal_exit(&c->journal);
449         bch2_io_clock_exit(&c->io_clock[WRITE]);
450         bch2_io_clock_exit(&c->io_clock[READ]);
451         bch2_fs_compress_exit(c);
452         bch2_journal_keys_free(&c->journal_keys);
453         bch2_journal_entries_free(c);
454         percpu_free_rwsem(&c->mark_lock);
455
456         if (c->btree_paths_bufs)
457                 for_each_possible_cpu(cpu)
458                         kfree(per_cpu_ptr(c->btree_paths_bufs, cpu)->path);
459
460         free_percpu(c->online_reserved);
461         free_percpu(c->btree_paths_bufs);
462         free_percpu(c->pcpu);
463         mempool_exit(&c->large_bkey_pool);
464         mempool_exit(&c->btree_bounce_pool);
465         bioset_exit(&c->btree_bio);
466         mempool_exit(&c->fill_iter);
467         percpu_ref_exit(&c->writes);
468         kfree(rcu_dereference_protected(c->disk_groups, 1));
469         kfree(c->journal_seq_blacklist_table);
470         kfree(c->unused_inode_hints);
471         free_heap(&c->copygc_heap);
472
473         if (c->io_complete_wq )
474                 destroy_workqueue(c->io_complete_wq );
475         if (c->copygc_wq)
476                 destroy_workqueue(c->copygc_wq);
477         if (c->btree_io_complete_wq)
478                 destroy_workqueue(c->btree_io_complete_wq);
479         if (c->btree_update_wq)
480                 destroy_workqueue(c->btree_update_wq);
481
482         bch2_free_super(&c->disk_sb);
483         kvpfree(c, sizeof(*c));
484         module_put(THIS_MODULE);
485 }
486
487 static void bch2_fs_release(struct kobject *kobj)
488 {
489         struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
490
491         __bch2_fs_free(c);
492 }
493
494 void __bch2_fs_stop(struct bch_fs *c)
495 {
496         struct bch_dev *ca;
497         unsigned i;
498
499         bch_verbose(c, "shutting down");
500
501         set_bit(BCH_FS_STOPPING, &c->flags);
502
503         cancel_work_sync(&c->journal_seq_blacklist_gc_work);
504
505         down_write(&c->state_lock);
506         bch2_fs_read_only(c);
507         up_write(&c->state_lock);
508
509         for_each_member_device(ca, c, i)
510                 if (ca->kobj.state_in_sysfs &&
511                     ca->disk_sb.bdev)
512                         sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
513
514         if (c->kobj.state_in_sysfs)
515                 kobject_del(&c->kobj);
516
517         bch2_fs_debug_exit(c);
518         bch2_fs_chardev_exit(c);
519
520         kobject_put(&c->time_stats);
521         kobject_put(&c->opts_dir);
522         kobject_put(&c->internal);
523
524         /* btree prefetch might have kicked off reads in the background: */
525         bch2_btree_flush_all_reads(c);
526
527         for_each_member_device(ca, c, i)
528                 cancel_work_sync(&ca->io_error_work);
529
530         cancel_work_sync(&c->read_only_work);
531
532         for (i = 0; i < c->sb.nr_devices; i++)
533                 if (c->devs[i])
534                         bch2_free_super(&c->devs[i]->disk_sb);
535 }
536
537 void bch2_fs_free(struct bch_fs *c)
538 {
539         unsigned i;
540
541         mutex_lock(&bch_fs_list_lock);
542         list_del(&c->list);
543         mutex_unlock(&bch_fs_list_lock);
544
545         closure_sync(&c->cl);
546         closure_debug_destroy(&c->cl);
547
548         for (i = 0; i < c->sb.nr_devices; i++)
549                 if (c->devs[i])
550                         bch2_dev_free(rcu_dereference_protected(c->devs[i], 1));
551
552         bch_verbose(c, "shutdown complete");
553
554         kobject_put(&c->kobj);
555 }
556
557 void bch2_fs_stop(struct bch_fs *c)
558 {
559         __bch2_fs_stop(c);
560         bch2_fs_free(c);
561 }
562
563 static int bch2_fs_online(struct bch_fs *c)
564 {
565         struct bch_dev *ca;
566         unsigned i;
567         int ret = 0;
568
569         lockdep_assert_held(&bch_fs_list_lock);
570
571         if (__bch2_uuid_to_fs(c->sb.uuid)) {
572                 bch_err(c, "filesystem UUID already open");
573                 return -EINVAL;
574         }
575
576         ret = bch2_fs_chardev_init(c);
577         if (ret) {
578                 bch_err(c, "error creating character device");
579                 return ret;
580         }
581
582         bch2_fs_debug_init(c);
583
584         ret = kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ?:
585             kobject_add(&c->internal, &c->kobj, "internal") ?:
586             kobject_add(&c->opts_dir, &c->kobj, "options") ?:
587             kobject_add(&c->time_stats, &c->kobj, "time_stats") ?:
588             bch2_opts_create_sysfs_files(&c->opts_dir);
589         if (ret) {
590                 bch_err(c, "error creating sysfs objects");
591                 return ret;
592         }
593
594         down_write(&c->state_lock);
595
596         for_each_member_device(ca, c, i) {
597                 ret = bch2_dev_sysfs_online(c, ca);
598                 if (ret) {
599                         bch_err(c, "error creating sysfs objects");
600                         percpu_ref_put(&ca->ref);
601                         goto err;
602                 }
603         }
604
605         BUG_ON(!list_empty(&c->list));
606         list_add(&c->list, &bch_fs_list);
607 err:
608         up_write(&c->state_lock);
609         return ret;
610 }
611
612 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
613 {
614         struct bch_sb_field_members *mi;
615         struct bch_fs *c;
616         unsigned i, iter_size;
617         int ret = 0;
618
619         pr_verbose_init(opts, "");
620
621         c = kvpmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
622         if (!c) {
623                 c = ERR_PTR(-ENOMEM);
624                 goto out;
625         }
626
627         __module_get(THIS_MODULE);
628
629         closure_init(&c->cl, NULL);
630
631         c->kobj.kset = bcachefs_kset;
632         kobject_init(&c->kobj, &bch2_fs_ktype);
633         kobject_init(&c->internal, &bch2_fs_internal_ktype);
634         kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
635         kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
636
637         c->minor                = -1;
638         c->disk_sb.fs_sb        = true;
639
640         init_rwsem(&c->state_lock);
641         mutex_init(&c->sb_lock);
642         mutex_init(&c->replicas_gc_lock);
643         mutex_init(&c->btree_root_lock);
644         INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
645
646         init_rwsem(&c->gc_lock);
647         mutex_init(&c->gc_gens_lock);
648
649         for (i = 0; i < BCH_TIME_STAT_NR; i++)
650                 bch2_time_stats_init(&c->times[i]);
651
652         bch2_fs_copygc_init(c);
653         bch2_fs_btree_key_cache_init_early(&c->btree_key_cache);
654         bch2_fs_allocator_background_init(c);
655         bch2_fs_allocator_foreground_init(c);
656         bch2_fs_rebalance_init(c);
657         bch2_fs_quota_init(c);
658         bch2_fs_ec_init_early(c);
659
660         INIT_LIST_HEAD(&c->list);
661
662         mutex_init(&c->usage_scratch_lock);
663
664         mutex_init(&c->bio_bounce_pages_lock);
665         mutex_init(&c->snapshot_table_lock);
666
667         spin_lock_init(&c->btree_write_error_lock);
668
669         INIT_WORK(&c->journal_seq_blacklist_gc_work,
670                   bch2_blacklist_entries_gc);
671
672         INIT_LIST_HEAD(&c->journal_iters);
673
674         INIT_LIST_HEAD(&c->fsck_errors);
675         mutex_init(&c->fsck_error_lock);
676
677         INIT_LIST_HEAD(&c->ec_stripe_head_list);
678         mutex_init(&c->ec_stripe_head_lock);
679
680         INIT_LIST_HEAD(&c->ec_stripe_new_list);
681         mutex_init(&c->ec_stripe_new_lock);
682
683         INIT_LIST_HEAD(&c->data_progress_list);
684         mutex_init(&c->data_progress_lock);
685
686         spin_lock_init(&c->ec_stripes_heap_lock);
687
688         seqcount_init(&c->gc_pos_lock);
689
690         seqcount_init(&c->usage_lock);
691
692         sema_init(&c->io_in_flight, 64);
693
694         c->copy_gc_enabled              = 1;
695         c->rebalance.enabled            = 1;
696         c->promote_whole_extents        = true;
697
698         c->journal.flush_write_time     = &c->times[BCH_TIME_journal_flush_write];
699         c->journal.noflush_write_time   = &c->times[BCH_TIME_journal_noflush_write];
700         c->journal.blocked_time         = &c->times[BCH_TIME_blocked_journal];
701         c->journal.flush_seq_time       = &c->times[BCH_TIME_journal_flush_seq];
702
703         bch2_fs_btree_cache_init_early(&c->btree_cache);
704
705         mutex_init(&c->sectors_available_lock);
706
707         ret = percpu_init_rwsem(&c->mark_lock);
708         if (ret)
709                 goto err;
710
711         mutex_lock(&c->sb_lock);
712         ret = bch2_sb_to_fs(c, sb);
713         mutex_unlock(&c->sb_lock);
714
715         if (ret)
716                 goto err;
717
718         uuid_unparse_lower(c->sb.user_uuid.b, c->name);
719
720         /* Compat: */
721         if (sb->version <= bcachefs_metadata_version_inode_v2 &&
722             !BCH_SB_JOURNAL_FLUSH_DELAY(sb))
723                 SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
724
725         if (sb->version <= bcachefs_metadata_version_inode_v2 &&
726             !BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
727                 SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 100);
728
729         c->opts = bch2_opts_default;
730         ret = bch2_opts_from_sb(&c->opts, sb);
731         if (ret)
732                 goto err;
733
734         bch2_opts_apply(&c->opts, opts);
735
736         /* key cache currently disabled for inodes, because of snapshots: */
737         c->opts.inodes_use_key_cache = 0;
738
739         c->btree_key_cache_btrees |= 1U << BTREE_ID_alloc;
740         if (c->opts.inodes_use_key_cache)
741                 c->btree_key_cache_btrees |= 1U << BTREE_ID_inodes;
742
743         c->block_bits           = ilog2(block_sectors(c));
744         c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
745
746         if (bch2_fs_init_fault("fs_alloc")) {
747                 bch_err(c, "fs_alloc fault injected");
748                 ret = -EFAULT;
749                 goto err;
750         }
751
752         iter_size = sizeof(struct sort_iter) +
753                 (btree_blocks(c) + 1) * 2 *
754                 sizeof(struct sort_iter_set);
755
756         c->inode_shard_bits = ilog2(roundup_pow_of_two(num_possible_cpus()));
757
758         if (!(c->btree_update_wq = alloc_workqueue("bcachefs",
759                                 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
760             !(c->btree_io_complete_wq = alloc_workqueue("bcachefs_btree_io",
761                                 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
762             !(c->copygc_wq = alloc_workqueue("bcachefs_copygc",
763                                 WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
764             !(c->io_complete_wq = alloc_workqueue("bcachefs_io",
765                                 WQ_FREEZABLE|WQ_HIGHPRI|WQ_MEM_RECLAIM, 1)) ||
766             percpu_ref_init(&c->writes, bch2_writes_disabled,
767                             PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
768             mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
769             bioset_init(&c->btree_bio, 1,
770                         max(offsetof(struct btree_read_bio, bio),
771                             offsetof(struct btree_write_bio, wbio.bio)),
772                         BIOSET_NEED_BVECS) ||
773             !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
774             !(c->btree_paths_bufs = alloc_percpu(struct btree_path_buf)) ||
775             !(c->online_reserved = alloc_percpu(u64)) ||
776             mempool_init_kvpmalloc_pool(&c->btree_bounce_pool, 1,
777                                         btree_bytes(c)) ||
778             mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048) ||
779             !(c->unused_inode_hints = kcalloc(1U << c->inode_shard_bits,
780                                               sizeof(u64), GFP_KERNEL))) {
781                 ret = -ENOMEM;
782                 goto err;
783         }
784
785         ret = bch2_io_clock_init(&c->io_clock[READ]) ?:
786             bch2_io_clock_init(&c->io_clock[WRITE]) ?:
787             bch2_fs_journal_init(&c->journal) ?:
788             bch2_fs_replicas_init(c) ?:
789             bch2_fs_btree_cache_init(c) ?:
790             bch2_fs_btree_key_cache_init(&c->btree_key_cache) ?:
791             bch2_fs_btree_iter_init(c) ?:
792             bch2_fs_btree_interior_update_init(c) ?:
793             bch2_fs_buckets_waiting_for_journal_init(c);
794             bch2_fs_subvolumes_init(c) ?:
795             bch2_fs_io_init(c) ?:
796             bch2_fs_encryption_init(c) ?:
797             bch2_fs_compress_init(c) ?:
798             bch2_fs_ec_init(c) ?:
799             bch2_fs_fsio_init(c);
800         if (ret)
801                 goto err;
802
803         mi = bch2_sb_get_members(c->disk_sb.sb);
804         for (i = 0; i < c->sb.nr_devices; i++)
805                 if (bch2_dev_exists(c->disk_sb.sb, mi, i) &&
806                     bch2_dev_alloc(c, i)) {
807                         ret = -EEXIST;
808                         goto err;
809                 }
810
811         bch2_journal_entry_res_resize(&c->journal,
812                         &c->btree_root_journal_res,
813                         BTREE_ID_NR * (JSET_KEYS_U64s + BKEY_BTREE_PTR_U64s_MAX));
814         bch2_dev_usage_journal_reserve(c);
815         bch2_journal_entry_res_resize(&c->journal,
816                         &c->clock_journal_res,
817                         (sizeof(struct jset_entry_clock) / sizeof(u64)) * 2);
818
819         mutex_lock(&bch_fs_list_lock);
820         ret = bch2_fs_online(c);
821         mutex_unlock(&bch_fs_list_lock);
822
823         if (ret)
824                 goto err;
825 out:
826         pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
827         return c;
828 err:
829         bch2_fs_free(c);
830         c = ERR_PTR(ret);
831         goto out;
832 }
833
834 noinline_for_stack
835 static void print_mount_opts(struct bch_fs *c)
836 {
837         enum bch_opt_id i;
838         struct printbuf p = PRINTBUF;
839         bool first = true;
840
841         if (c->opts.read_only) {
842                 pr_buf(&p, "ro");
843                 first = false;
844         }
845
846         for (i = 0; i < bch2_opts_nr; i++) {
847                 const struct bch_option *opt = &bch2_opt_table[i];
848                 u64 v = bch2_opt_get_by_id(&c->opts, i);
849
850                 if (!(opt->flags & OPT_MOUNT))
851                         continue;
852
853                 if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
854                         continue;
855
856                 if (!first)
857                         pr_buf(&p, ",");
858                 first = false;
859                 bch2_opt_to_text(&p, c, c->disk_sb.sb, opt, v, OPT_SHOW_MOUNT_STYLE);
860         }
861
862         if (!p.pos)
863                 pr_buf(&p, "(null)");
864
865         bch_info(c, "mounted version=%s opts=%s", bch2_metadata_versions[c->sb.version], p.buf);
866         printbuf_exit(&p);
867 }
868
869 int bch2_fs_start(struct bch_fs *c)
870 {
871         struct bch_sb_field_members *mi;
872         struct bch_dev *ca;
873         time64_t now = ktime_get_real_seconds();
874         unsigned i;
875         int ret = -EINVAL;
876
877         down_write(&c->state_lock);
878
879         BUG_ON(test_bit(BCH_FS_STARTED, &c->flags));
880
881         mutex_lock(&c->sb_lock);
882
883         for_each_online_member(ca, c, i)
884                 bch2_sb_from_fs(c, ca);
885
886         mi = bch2_sb_get_members(c->disk_sb.sb);
887         for_each_online_member(ca, c, i)
888                 mi->members[ca->dev_idx].last_mount = cpu_to_le64(now);
889
890         mutex_unlock(&c->sb_lock);
891
892         for_each_rw_member(ca, c, i)
893                 bch2_dev_allocator_add(c, ca);
894         bch2_recalc_capacity(c);
895
896         ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
897                 ? bch2_fs_recovery(c)
898                 : bch2_fs_initialize(c);
899         if (ret)
900                 goto err;
901
902         ret = bch2_opts_check_may_set(c);
903         if (ret)
904                 goto err;
905
906         ret = -EINVAL;
907         if (bch2_fs_init_fault("fs_start")) {
908                 bch_err(c, "fs_start fault injected");
909                 goto err;
910         }
911
912         set_bit(BCH_FS_STARTED, &c->flags);
913
914         if (c->opts.read_only || c->opts.nochanges) {
915                 bch2_fs_read_only(c);
916         } else {
917                 ret = !test_bit(BCH_FS_RW, &c->flags)
918                         ? bch2_fs_read_write(c)
919                         : bch2_fs_read_write_late(c);
920                 if (ret)
921                         goto err;
922         }
923
924         print_mount_opts(c);
925         ret = 0;
926 out:
927         up_write(&c->state_lock);
928         return ret;
929 err:
930         switch (ret) {
931         case BCH_FSCK_ERRORS_NOT_FIXED:
932                 bch_err(c, "filesystem contains errors: please report this to the developers");
933                 pr_cont("mount with -o fix_errors to repair\n");
934                 break;
935         case BCH_FSCK_REPAIR_UNIMPLEMENTED:
936                 bch_err(c, "filesystem contains errors: please report this to the developers");
937                 pr_cont("repair unimplemented: inform the developers so that it can be added\n");
938                 break;
939         case BCH_FSCK_REPAIR_IMPOSSIBLE:
940                 bch_err(c, "filesystem contains errors, but repair impossible");
941                 break;
942         case BCH_FSCK_UNKNOWN_VERSION:
943                 bch_err(c, "unknown metadata version");
944                 break;
945         case -ENOMEM:
946                 bch_err(c, "cannot allocate memory");
947                 break;
948         case -EIO:
949                 bch_err(c, "IO error");
950                 break;
951         }
952
953         if (ret >= 0)
954                 ret = -EIO;
955         goto out;
956 }
957
958 static const char *bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
959 {
960         struct bch_sb_field_members *sb_mi;
961
962         sb_mi = bch2_sb_get_members(sb);
963         if (!sb_mi)
964                 return "Invalid superblock: member info area missing";
965
966         if (le16_to_cpu(sb->block_size) != block_sectors(c))
967                 return "mismatched block size";
968
969         if (le16_to_cpu(sb_mi->members[sb->dev_idx].bucket_size) <
970             BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
971                 return "new cache bucket size is too small";
972
973         return NULL;
974 }
975
976 static const char *bch2_dev_in_fs(struct bch_sb *fs, struct bch_sb *sb)
977 {
978         struct bch_sb *newest =
979                 le64_to_cpu(fs->seq) > le64_to_cpu(sb->seq) ? fs : sb;
980         struct bch_sb_field_members *mi = bch2_sb_get_members(newest);
981
982         if (uuid_le_cmp(fs->uuid, sb->uuid))
983                 return "device not a member of filesystem";
984
985         if (!bch2_dev_exists(newest, mi, sb->dev_idx))
986                 return "device has been removed";
987
988         if (fs->block_size != sb->block_size)
989                 return "mismatched block size";
990
991         return NULL;
992 }
993
994 /* Device startup/shutdown: */
995
996 static void bch2_dev_release(struct kobject *kobj)
997 {
998         struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
999
1000         kfree(ca);
1001 }
1002
1003 static void bch2_dev_free(struct bch_dev *ca)
1004 {
1005         cancel_work_sync(&ca->io_error_work);
1006
1007         if (ca->kobj.state_in_sysfs &&
1008             ca->disk_sb.bdev)
1009                 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1010
1011         if (ca->kobj.state_in_sysfs)
1012                 kobject_del(&ca->kobj);
1013
1014         bch2_free_super(&ca->disk_sb);
1015         bch2_dev_journal_exit(ca);
1016
1017         free_percpu(ca->io_done);
1018         bioset_exit(&ca->replica_set);
1019         bch2_dev_buckets_free(ca);
1020         free_page((unsigned long) ca->sb_read_scratch);
1021
1022         bch2_time_stats_exit(&ca->io_latency[WRITE]);
1023         bch2_time_stats_exit(&ca->io_latency[READ]);
1024
1025         percpu_ref_exit(&ca->io_ref);
1026         percpu_ref_exit(&ca->ref);
1027         kobject_put(&ca->kobj);
1028 }
1029
1030 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
1031 {
1032
1033         lockdep_assert_held(&c->state_lock);
1034
1035         if (percpu_ref_is_zero(&ca->io_ref))
1036                 return;
1037
1038         __bch2_dev_read_only(c, ca);
1039
1040         reinit_completion(&ca->io_ref_completion);
1041         percpu_ref_kill(&ca->io_ref);
1042         wait_for_completion(&ca->io_ref_completion);
1043
1044         if (ca->kobj.state_in_sysfs) {
1045                 sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1046                 sysfs_remove_link(&ca->kobj, "block");
1047         }
1048
1049         bch2_free_super(&ca->disk_sb);
1050         bch2_dev_journal_exit(ca);
1051 }
1052
1053 static void bch2_dev_ref_complete(struct percpu_ref *ref)
1054 {
1055         struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
1056
1057         complete(&ca->ref_completion);
1058 }
1059
1060 static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
1061 {
1062         struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
1063
1064         complete(&ca->io_ref_completion);
1065 }
1066
1067 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
1068 {
1069         int ret;
1070
1071         if (!c->kobj.state_in_sysfs)
1072                 return 0;
1073
1074         if (!ca->kobj.state_in_sysfs) {
1075                 ret = kobject_add(&ca->kobj, &c->kobj,
1076                                   "dev-%u", ca->dev_idx);
1077                 if (ret)
1078                         return ret;
1079         }
1080
1081         if (ca->disk_sb.bdev) {
1082                 struct kobject *block = bdev_kobj(ca->disk_sb.bdev);
1083
1084                 ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
1085                 if (ret)
1086                         return ret;
1087
1088                 ret = sysfs_create_link(&ca->kobj, block, "block");
1089                 if (ret)
1090                         return ret;
1091         }
1092
1093         return 0;
1094 }
1095
1096 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
1097                                         struct bch_member *member)
1098 {
1099         struct bch_dev *ca;
1100
1101         ca = kzalloc(sizeof(*ca), GFP_KERNEL);
1102         if (!ca)
1103                 return NULL;
1104
1105         kobject_init(&ca->kobj, &bch2_dev_ktype);
1106         init_completion(&ca->ref_completion);
1107         init_completion(&ca->io_ref_completion);
1108
1109         init_rwsem(&ca->bucket_lock);
1110
1111         INIT_WORK(&ca->io_error_work, bch2_io_error_work);
1112
1113         bch2_time_stats_init(&ca->io_latency[READ]);
1114         bch2_time_stats_init(&ca->io_latency[WRITE]);
1115
1116         ca->mi = bch2_mi_to_cpu(member);
1117         ca->uuid = member->uuid;
1118
1119         ca->nr_btree_reserve = DIV_ROUND_UP(BTREE_NODE_RESERVE,
1120                              ca->mi.bucket_size / btree_sectors(c));
1121
1122         if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
1123                             0, GFP_KERNEL) ||
1124             percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
1125                             PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
1126             !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) ||
1127             bch2_dev_buckets_alloc(c, ca) ||
1128             bioset_init(&ca->replica_set, 4,
1129                         offsetof(struct bch_write_bio, bio), 0) ||
1130             !(ca->io_done       = alloc_percpu(*ca->io_done)))
1131                 goto err;
1132
1133         return ca;
1134 err:
1135         bch2_dev_free(ca);
1136         return NULL;
1137 }
1138
1139 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
1140                             unsigned dev_idx)
1141 {
1142         ca->dev_idx = dev_idx;
1143         __set_bit(ca->dev_idx, ca->self.d);
1144         scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
1145
1146         ca->fs = c;
1147         rcu_assign_pointer(c->devs[ca->dev_idx], ca);
1148
1149         if (bch2_dev_sysfs_online(c, ca))
1150                 pr_warn("error creating sysfs objects");
1151 }
1152
1153 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
1154 {
1155         struct bch_member *member =
1156                 bch2_sb_get_members(c->disk_sb.sb)->members + dev_idx;
1157         struct bch_dev *ca = NULL;
1158         int ret = 0;
1159
1160         pr_verbose_init(c->opts, "");
1161
1162         if (bch2_fs_init_fault("dev_alloc"))
1163                 goto err;
1164
1165         ca = __bch2_dev_alloc(c, member);
1166         if (!ca)
1167                 goto err;
1168
1169         ca->fs = c;
1170
1171         bch2_dev_attach(c, ca, dev_idx);
1172 out:
1173         pr_verbose_init(c->opts, "ret %i", ret);
1174         return ret;
1175 err:
1176         if (ca)
1177                 bch2_dev_free(ca);
1178         ret = -ENOMEM;
1179         goto out;
1180 }
1181
1182 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1183 {
1184         unsigned ret;
1185
1186         if (bch2_dev_is_online(ca)) {
1187                 bch_err(ca, "already have device online in slot %u",
1188                         sb->sb->dev_idx);
1189                 return -EINVAL;
1190         }
1191
1192         if (get_capacity(sb->bdev->bd_disk) <
1193             ca->mi.bucket_size * ca->mi.nbuckets) {
1194                 bch_err(ca, "cannot online: device too small");
1195                 return -EINVAL;
1196         }
1197
1198         BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1199
1200         if (get_capacity(sb->bdev->bd_disk) <
1201             ca->mi.bucket_size * ca->mi.nbuckets) {
1202                 bch_err(ca, "device too small");
1203                 return -EINVAL;
1204         }
1205
1206         ret = bch2_dev_journal_init(ca, sb->sb);
1207         if (ret)
1208                 return ret;
1209
1210         /* Commit: */
1211         ca->disk_sb = *sb;
1212         if (sb->mode & FMODE_EXCL)
1213                 ca->disk_sb.bdev->bd_holder = ca;
1214         memset(sb, 0, sizeof(*sb));
1215
1216         ca->dev = ca->disk_sb.bdev->bd_dev;
1217
1218         percpu_ref_reinit(&ca->io_ref);
1219
1220         return 0;
1221 }
1222
1223 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1224 {
1225         struct bch_dev *ca;
1226         int ret;
1227
1228         lockdep_assert_held(&c->state_lock);
1229
1230         if (le64_to_cpu(sb->sb->seq) >
1231             le64_to_cpu(c->disk_sb.sb->seq))
1232                 bch2_sb_to_fs(c, sb->sb);
1233
1234         BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
1235                !c->devs[sb->sb->dev_idx]);
1236
1237         ca = bch_dev_locked(c, sb->sb->dev_idx);
1238
1239         ret = __bch2_dev_attach_bdev(ca, sb);
1240         if (ret)
1241                 return ret;
1242
1243         bch2_dev_sysfs_online(c, ca);
1244
1245         if (c->sb.nr_devices == 1)
1246                 bdevname(ca->disk_sb.bdev, c->name);
1247         bdevname(ca->disk_sb.bdev, ca->name);
1248
1249         rebalance_wakeup(c);
1250         return 0;
1251 }
1252
1253 /* Device management: */
1254
1255 /*
1256  * Note: this function is also used by the error paths - when a particular
1257  * device sees an error, we call it to determine whether we can just set the
1258  * device RO, or - if this function returns false - we'll set the whole
1259  * filesystem RO:
1260  *
1261  * XXX: maybe we should be more explicit about whether we're changing state
1262  * because we got an error or what have you?
1263  */
1264 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1265                             enum bch_member_state new_state, int flags)
1266 {
1267         struct bch_devs_mask new_online_devs;
1268         struct bch_dev *ca2;
1269         int i, nr_rw = 0, required;
1270
1271         lockdep_assert_held(&c->state_lock);
1272
1273         switch (new_state) {
1274         case BCH_MEMBER_STATE_rw:
1275                 return true;
1276         case BCH_MEMBER_STATE_ro:
1277                 if (ca->mi.state != BCH_MEMBER_STATE_rw)
1278                         return true;
1279
1280                 /* do we have enough devices to write to?  */
1281                 for_each_member_device(ca2, c, i)
1282                         if (ca2 != ca)
1283                                 nr_rw += ca2->mi.state == BCH_MEMBER_STATE_rw;
1284
1285                 required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1286                                ? c->opts.metadata_replicas
1287                                : c->opts.metadata_replicas_required,
1288                                !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1289                                ? c->opts.data_replicas
1290                                : c->opts.data_replicas_required);
1291
1292                 return nr_rw >= required;
1293         case BCH_MEMBER_STATE_failed:
1294         case BCH_MEMBER_STATE_spare:
1295                 if (ca->mi.state != BCH_MEMBER_STATE_rw &&
1296                     ca->mi.state != BCH_MEMBER_STATE_ro)
1297                         return true;
1298
1299                 /* do we have enough devices to read from?  */
1300                 new_online_devs = bch2_online_devs(c);
1301                 __clear_bit(ca->dev_idx, new_online_devs.d);
1302
1303                 return bch2_have_enough_devs(c, new_online_devs, flags, false);
1304         default:
1305                 BUG();
1306         }
1307 }
1308
1309 static bool bch2_fs_may_start(struct bch_fs *c)
1310 {
1311         struct bch_sb_field_members *mi;
1312         struct bch_dev *ca;
1313         unsigned i, flags = 0;
1314
1315         if (c->opts.very_degraded)
1316                 flags |= BCH_FORCE_IF_DEGRADED|BCH_FORCE_IF_LOST;
1317
1318         if (c->opts.degraded)
1319                 flags |= BCH_FORCE_IF_DEGRADED;
1320
1321         if (!c->opts.degraded &&
1322             !c->opts.very_degraded) {
1323                 mutex_lock(&c->sb_lock);
1324                 mi = bch2_sb_get_members(c->disk_sb.sb);
1325
1326                 for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1327                         if (!bch2_dev_exists(c->disk_sb.sb, mi, i))
1328                                 continue;
1329
1330                         ca = bch_dev_locked(c, i);
1331
1332                         if (!bch2_dev_is_online(ca) &&
1333                             (ca->mi.state == BCH_MEMBER_STATE_rw ||
1334                              ca->mi.state == BCH_MEMBER_STATE_ro)) {
1335                                 mutex_unlock(&c->sb_lock);
1336                                 return false;
1337                         }
1338                 }
1339                 mutex_unlock(&c->sb_lock);
1340         }
1341
1342         return bch2_have_enough_devs(c, bch2_online_devs(c), flags, true);
1343 }
1344
1345 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1346 {
1347         /*
1348          * Device going read only means the copygc reserve get smaller, so we
1349          * don't want that happening while copygc is in progress:
1350          */
1351         bch2_copygc_stop(c);
1352
1353         /*
1354          * The allocator thread itself allocates btree nodes, so stop it first:
1355          */
1356         bch2_dev_allocator_remove(c, ca);
1357         bch2_dev_journal_stop(&c->journal, ca);
1358
1359         bch2_copygc_start(c);
1360 }
1361
1362 static void __bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1363 {
1364         lockdep_assert_held(&c->state_lock);
1365
1366         BUG_ON(ca->mi.state != BCH_MEMBER_STATE_rw);
1367
1368         bch2_dev_allocator_add(c, ca);
1369         bch2_recalc_capacity(c);
1370 }
1371
1372 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1373                          enum bch_member_state new_state, int flags)
1374 {
1375         struct bch_sb_field_members *mi;
1376         int ret = 0;
1377
1378         if (ca->mi.state == new_state)
1379                 return 0;
1380
1381         if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1382                 return -EINVAL;
1383
1384         if (new_state != BCH_MEMBER_STATE_rw)
1385                 __bch2_dev_read_only(c, ca);
1386
1387         bch_notice(ca, "%s", bch2_member_states[new_state]);
1388
1389         mutex_lock(&c->sb_lock);
1390         mi = bch2_sb_get_members(c->disk_sb.sb);
1391         SET_BCH_MEMBER_STATE(&mi->members[ca->dev_idx], new_state);
1392         bch2_write_super(c);
1393         mutex_unlock(&c->sb_lock);
1394
1395         if (new_state == BCH_MEMBER_STATE_rw)
1396                 __bch2_dev_read_write(c, ca);
1397
1398         rebalance_wakeup(c);
1399
1400         return ret;
1401 }
1402
1403 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1404                        enum bch_member_state new_state, int flags)
1405 {
1406         int ret;
1407
1408         down_write(&c->state_lock);
1409         ret = __bch2_dev_set_state(c, ca, new_state, flags);
1410         up_write(&c->state_lock);
1411
1412         return ret;
1413 }
1414
1415 /* Device add/removal: */
1416
1417 static int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
1418 {
1419         struct bpos start       = POS(ca->dev_idx, 0);
1420         struct bpos end         = POS(ca->dev_idx, U64_MAX);
1421         int ret;
1422
1423         /*
1424          * We clear the LRU and need_discard btrees first so that we don't race
1425          * with bch2_do_invalidates() and bch2_do_discards()
1426          */
1427         ret =   bch2_btree_delete_range(c, BTREE_ID_lru, start, end,
1428                                         BTREE_TRIGGER_NORUN, NULL) ?:
1429                 bch2_btree_delete_range(c, BTREE_ID_need_discard, start, end,
1430                                         BTREE_TRIGGER_NORUN, NULL) ?:
1431                 bch2_btree_delete_range(c, BTREE_ID_freespace, start, end,
1432                                         BTREE_TRIGGER_NORUN, NULL) ?:
1433                 bch2_btree_delete_range(c, BTREE_ID_alloc, start, end,
1434                                         BTREE_TRIGGER_NORUN, NULL);
1435         if (ret)
1436                 bch_err(c, "error %i removing dev alloc info", ret);
1437
1438         return ret;
1439 }
1440
1441 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1442 {
1443         struct bch_sb_field_members *mi;
1444         unsigned dev_idx = ca->dev_idx, data;
1445         int ret = -EINVAL;
1446
1447         down_write(&c->state_lock);
1448
1449         /*
1450          * We consume a reference to ca->ref, regardless of whether we succeed
1451          * or fail:
1452          */
1453         percpu_ref_put(&ca->ref);
1454
1455         if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1456                 bch_err(ca, "Cannot remove without losing data");
1457                 goto err;
1458         }
1459
1460         __bch2_dev_read_only(c, ca);
1461
1462         ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1463         if (ret) {
1464                 bch_err(ca, "Remove failed: error %i dropping data", ret);
1465                 goto err;
1466         }
1467
1468         ret = bch2_dev_remove_alloc(c, ca);
1469         if (ret) {
1470                 bch_err(ca, "Remove failed, error deleting alloc info");
1471                 goto err;
1472         }
1473
1474         ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1475         if (ret) {
1476                 bch_err(ca, "Remove failed: error %i flushing journal", ret);
1477                 goto err;
1478         }
1479
1480         ret = bch2_journal_flush(&c->journal);
1481         if (ret) {
1482                 bch_err(ca, "Remove failed, journal error");
1483                 goto err;
1484         }
1485
1486         ret = bch2_replicas_gc2(c);
1487         if (ret) {
1488                 bch_err(ca, "Remove failed: error %i from replicas gc", ret);
1489                 goto err;
1490         }
1491
1492         data = bch2_dev_has_data(c, ca);
1493         if (data) {
1494                 struct printbuf data_has = PRINTBUF;
1495
1496                 bch2_flags_to_text(&data_has, bch2_data_types, data);
1497                 bch_err(ca, "Remove failed, still has data (%s)", data_has.buf);
1498                 printbuf_exit(&data_has);
1499                 ret = -EBUSY;
1500                 goto err;
1501         }
1502
1503         __bch2_dev_offline(c, ca);
1504
1505         mutex_lock(&c->sb_lock);
1506         rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1507         mutex_unlock(&c->sb_lock);
1508
1509         percpu_ref_kill(&ca->ref);
1510         wait_for_completion(&ca->ref_completion);
1511
1512         bch2_dev_free(ca);
1513
1514         /*
1515          * Free this device's slot in the bch_member array - all pointers to
1516          * this device must be gone:
1517          */
1518         mutex_lock(&c->sb_lock);
1519         mi = bch2_sb_get_members(c->disk_sb.sb);
1520         memset(&mi->members[dev_idx].uuid, 0, sizeof(mi->members[dev_idx].uuid));
1521
1522         bch2_write_super(c);
1523
1524         mutex_unlock(&c->sb_lock);
1525         up_write(&c->state_lock);
1526
1527         bch2_dev_usage_journal_reserve(c);
1528         return 0;
1529 err:
1530         if (ca->mi.state == BCH_MEMBER_STATE_rw &&
1531             !percpu_ref_is_zero(&ca->io_ref))
1532                 __bch2_dev_read_write(c, ca);
1533         up_write(&c->state_lock);
1534         return ret;
1535 }
1536
1537 /* Add new device to running filesystem: */
1538 int bch2_dev_add(struct bch_fs *c, const char *path)
1539 {
1540         struct bch_opts opts = bch2_opts_empty();
1541         struct bch_sb_handle sb;
1542         const char *err;
1543         struct bch_dev *ca = NULL;
1544         struct bch_sb_field_members *mi;
1545         struct bch_member dev_mi;
1546         unsigned dev_idx, nr_devices, u64s;
1547         struct printbuf errbuf = PRINTBUF;
1548         int ret;
1549
1550         ret = bch2_read_super(path, &opts, &sb);
1551         if (ret) {
1552                 bch_err(c, "device add error: error reading super: %i", ret);
1553                 goto err;
1554         }
1555
1556         dev_mi = bch2_sb_get_members(sb.sb)->members[sb.sb->dev_idx];
1557
1558         err = bch2_dev_may_add(sb.sb, c);
1559         if (err) {
1560                 bch_err(c, "device add error: %s", err);
1561                 ret = -EINVAL;
1562                 goto err;
1563         }
1564
1565         ca = __bch2_dev_alloc(c, &dev_mi);
1566         if (!ca) {
1567                 bch2_free_super(&sb);
1568                 ret = -ENOMEM;
1569                 goto err;
1570         }
1571
1572         bch2_dev_usage_init(ca);
1573
1574         ret = __bch2_dev_attach_bdev(ca, &sb);
1575         if (ret) {
1576                 bch2_dev_free(ca);
1577                 goto err;
1578         }
1579
1580         ret = bch2_dev_journal_alloc(ca);
1581         if (ret) {
1582                 bch_err(c, "device add error: journal alloc failed");
1583                 goto err;
1584         }
1585
1586         down_write(&c->state_lock);
1587         mutex_lock(&c->sb_lock);
1588
1589         ret = bch2_sb_from_fs(c, ca);
1590         if (ret) {
1591                 bch_err(c, "device add error: new device superblock too small");
1592                 goto err_unlock;
1593         }
1594
1595         mi = bch2_sb_get_members(ca->disk_sb.sb);
1596
1597         if (!bch2_sb_resize_members(&ca->disk_sb,
1598                                 le32_to_cpu(mi->field.u64s) +
1599                                 sizeof(dev_mi) / sizeof(u64))) {
1600                 bch_err(c, "device add error: new device superblock too small");
1601                 ret = -ENOSPC;
1602                 goto err_unlock;
1603         }
1604
1605         if (dynamic_fault("bcachefs:add:no_slot"))
1606                 goto no_slot;
1607
1608         mi = bch2_sb_get_members(c->disk_sb.sb);
1609         for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++)
1610                 if (!bch2_dev_exists(c->disk_sb.sb, mi, dev_idx))
1611                         goto have_slot;
1612 no_slot:
1613         bch_err(c, "device add error: already have maximum number of devices");
1614         ret = -ENOSPC;
1615         goto err_unlock;
1616
1617 have_slot:
1618         nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1619         u64s = (sizeof(struct bch_sb_field_members) +
1620                 sizeof(struct bch_member) * nr_devices) / sizeof(u64);
1621
1622         mi = bch2_sb_resize_members(&c->disk_sb, u64s);
1623         if (!mi) {
1624                 bch_err(c, "device add error: no room in superblock for member info");
1625                 ret = -ENOSPC;
1626                 goto err_unlock;
1627         }
1628
1629         /* success: */
1630
1631         mi->members[dev_idx] = dev_mi;
1632         mi->members[dev_idx].last_mount = cpu_to_le64(ktime_get_real_seconds());
1633         c->disk_sb.sb->nr_devices       = nr_devices;
1634
1635         ca->disk_sb.sb->dev_idx = dev_idx;
1636         bch2_dev_attach(c, ca, dev_idx);
1637
1638         bch2_write_super(c);
1639         mutex_unlock(&c->sb_lock);
1640
1641         bch2_dev_usage_journal_reserve(c);
1642
1643         ret = bch2_trans_mark_dev_sb(c, ca);
1644         if (ret) {
1645                 bch_err(c, "device add error: error marking new superblock: %i", ret);
1646                 goto err_late;
1647         }
1648
1649         ret = bch2_fs_freespace_init(c);
1650         if (ret) {
1651                 bch_err(c, "device add error: error initializing free space: %i", ret);
1652                 goto err_late;
1653         }
1654
1655         ca->new_fs_bucket_idx = 0;
1656
1657         if (ca->mi.state == BCH_MEMBER_STATE_rw)
1658                 __bch2_dev_read_write(c, ca);
1659
1660         up_write(&c->state_lock);
1661         return 0;
1662
1663 err_unlock:
1664         mutex_unlock(&c->sb_lock);
1665         up_write(&c->state_lock);
1666 err:
1667         if (ca)
1668                 bch2_dev_free(ca);
1669         bch2_free_super(&sb);
1670         printbuf_exit(&errbuf);
1671         return ret;
1672 err_late:
1673         up_write(&c->state_lock);
1674         ca = NULL;
1675         goto err;
1676 }
1677
1678 /* Hot add existing device to running filesystem: */
1679 int bch2_dev_online(struct bch_fs *c, const char *path)
1680 {
1681         struct bch_opts opts = bch2_opts_empty();
1682         struct bch_sb_handle sb = { NULL };
1683         struct bch_sb_field_members *mi;
1684         struct bch_dev *ca;
1685         unsigned dev_idx;
1686         const char *err;
1687         int ret;
1688
1689         down_write(&c->state_lock);
1690
1691         ret = bch2_read_super(path, &opts, &sb);
1692         if (ret) {
1693                 up_write(&c->state_lock);
1694                 return ret;
1695         }
1696
1697         dev_idx = sb.sb->dev_idx;
1698
1699         err = bch2_dev_in_fs(c->disk_sb.sb, sb.sb);
1700         if (err) {
1701                 bch_err(c, "error bringing %s online: %s", path, err);
1702                 goto err;
1703         }
1704
1705         ret = bch2_dev_attach_bdev(c, &sb);
1706         if (ret)
1707                 goto err;
1708
1709         ca = bch_dev_locked(c, dev_idx);
1710
1711         ret = bch2_trans_mark_dev_sb(c, ca);
1712         if (ret) {
1713                 bch_err(c, "error bringing %s online: error %i from bch2_trans_mark_dev_sb",
1714                         path, ret);
1715                 goto err;
1716         }
1717
1718         if (ca->mi.state == BCH_MEMBER_STATE_rw)
1719                 __bch2_dev_read_write(c, ca);
1720
1721         mutex_lock(&c->sb_lock);
1722         mi = bch2_sb_get_members(c->disk_sb.sb);
1723
1724         mi->members[ca->dev_idx].last_mount =
1725                 cpu_to_le64(ktime_get_real_seconds());
1726
1727         bch2_write_super(c);
1728         mutex_unlock(&c->sb_lock);
1729
1730         up_write(&c->state_lock);
1731         return 0;
1732 err:
1733         up_write(&c->state_lock);
1734         bch2_free_super(&sb);
1735         return -EINVAL;
1736 }
1737
1738 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1739 {
1740         down_write(&c->state_lock);
1741
1742         if (!bch2_dev_is_online(ca)) {
1743                 bch_err(ca, "Already offline");
1744                 up_write(&c->state_lock);
1745                 return 0;
1746         }
1747
1748         if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1749                 bch_err(ca, "Cannot offline required disk");
1750                 up_write(&c->state_lock);
1751                 return -EINVAL;
1752         }
1753
1754         __bch2_dev_offline(c, ca);
1755
1756         up_write(&c->state_lock);
1757         return 0;
1758 }
1759
1760 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1761 {
1762         struct bch_member *mi;
1763         int ret = 0;
1764
1765         down_write(&c->state_lock);
1766
1767         if (nbuckets < ca->mi.nbuckets) {
1768                 bch_err(ca, "Cannot shrink yet");
1769                 ret = -EINVAL;
1770                 goto err;
1771         }
1772
1773         if (bch2_dev_is_online(ca) &&
1774             get_capacity(ca->disk_sb.bdev->bd_disk) <
1775             ca->mi.bucket_size * nbuckets) {
1776                 bch_err(ca, "New size larger than device");
1777                 ret = -EINVAL;
1778                 goto err;
1779         }
1780
1781         ret = bch2_dev_buckets_resize(c, ca, nbuckets);
1782         if (ret) {
1783                 bch_err(ca, "Resize error: %i", ret);
1784                 goto err;
1785         }
1786
1787         ret = bch2_trans_mark_dev_sb(c, ca);
1788         if (ret) {
1789                 goto err;
1790         }
1791
1792         mutex_lock(&c->sb_lock);
1793         mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx];
1794         mi->nbuckets = cpu_to_le64(nbuckets);
1795
1796         bch2_write_super(c);
1797         mutex_unlock(&c->sb_lock);
1798
1799         bch2_recalc_capacity(c);
1800 err:
1801         up_write(&c->state_lock);
1802         return ret;
1803 }
1804
1805 /* return with ref on ca->ref: */
1806 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *name)
1807 {
1808         struct bch_dev *ca;
1809         unsigned i;
1810
1811         rcu_read_lock();
1812         for_each_member_device_rcu(ca, c, i, NULL)
1813                 if (!strcmp(name, ca->name))
1814                         goto found;
1815         ca = ERR_PTR(-ENOENT);
1816 found:
1817         rcu_read_unlock();
1818
1819         return ca;
1820 }
1821
1822 /* Filesystem open: */
1823
1824 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
1825                             struct bch_opts opts)
1826 {
1827         struct bch_sb_handle *sb = NULL;
1828         struct bch_fs *c = NULL;
1829         struct bch_sb_field_members *mi;
1830         unsigned i, best_sb = 0;
1831         const char *err;
1832         struct printbuf errbuf = PRINTBUF;
1833         int ret = 0;
1834
1835         if (!try_module_get(THIS_MODULE))
1836                 return ERR_PTR(-ENODEV);
1837
1838         pr_verbose_init(opts, "");
1839
1840         if (!nr_devices) {
1841                 ret = -EINVAL;
1842                 goto err;
1843         }
1844
1845         sb = kcalloc(nr_devices, sizeof(*sb), GFP_KERNEL);
1846         if (!sb) {
1847                 ret = -ENOMEM;
1848                 goto err;
1849         }
1850
1851         for (i = 0; i < nr_devices; i++) {
1852                 ret = bch2_read_super(devices[i], &opts, &sb[i]);
1853                 if (ret)
1854                         goto err;
1855
1856         }
1857
1858         for (i = 1; i < nr_devices; i++)
1859                 if (le64_to_cpu(sb[i].sb->seq) >
1860                     le64_to_cpu(sb[best_sb].sb->seq))
1861                         best_sb = i;
1862
1863         mi = bch2_sb_get_members(sb[best_sb].sb);
1864
1865         i = 0;
1866         while (i < nr_devices) {
1867                 if (i != best_sb &&
1868                     !bch2_dev_exists(sb[best_sb].sb, mi, sb[i].sb->dev_idx)) {
1869                         char buf[BDEVNAME_SIZE];
1870                         pr_info("%s has been removed, skipping",
1871                                 bdevname(sb[i].bdev, buf));
1872                         bch2_free_super(&sb[i]);
1873                         array_remove_item(sb, nr_devices, i);
1874                         continue;
1875                 }
1876
1877                 err = bch2_dev_in_fs(sb[best_sb].sb, sb[i].sb);
1878                 if (err)
1879                         goto err_print;
1880                 i++;
1881         }
1882
1883         c = bch2_fs_alloc(sb[best_sb].sb, opts);
1884         if (IS_ERR(c)) {
1885                 ret = PTR_ERR(c);
1886                 goto err;
1887         }
1888
1889         down_write(&c->state_lock);
1890         for (i = 0; i < nr_devices; i++) {
1891                 ret = bch2_dev_attach_bdev(c, &sb[i]);
1892                 if (ret) {
1893                         up_write(&c->state_lock);
1894                         goto err;
1895                 }
1896         }
1897         up_write(&c->state_lock);
1898
1899         err = "insufficient devices";
1900         if (!bch2_fs_may_start(c))
1901                 goto err_print;
1902
1903         if (!c->opts.nostart) {
1904                 ret = bch2_fs_start(c);
1905                 if (ret)
1906                         goto err;
1907         }
1908 out:
1909         kfree(sb);
1910         printbuf_exit(&errbuf);
1911         module_put(THIS_MODULE);
1912         pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c));
1913         return c;
1914 err_print:
1915         pr_err("bch_fs_open err opening %s: %s",
1916                devices[0], err);
1917         ret = -EINVAL;
1918 err:
1919         if (!IS_ERR_OR_NULL(c))
1920                 bch2_fs_stop(c);
1921         if (sb)
1922                 for (i = 0; i < nr_devices; i++)
1923                         bch2_free_super(&sb[i]);
1924         c = ERR_PTR(ret);
1925         goto out;
1926 }
1927
1928 /* Global interfaces/init */
1929
1930 static void bcachefs_exit(void)
1931 {
1932         bch2_debug_exit();
1933         bch2_vfs_exit();
1934         bch2_chardev_exit();
1935         bch2_btree_key_cache_exit();
1936         if (bcachefs_kset)
1937                 kset_unregister(bcachefs_kset);
1938 }
1939
1940 static int __init bcachefs_init(void)
1941 {
1942         bch2_bkey_pack_test();
1943
1944         if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
1945             bch2_btree_key_cache_init() ||
1946             bch2_chardev_init() ||
1947             bch2_vfs_init() ||
1948             bch2_debug_init())
1949                 goto err;
1950
1951         return 0;
1952 err:
1953         bcachefs_exit();
1954         return -ENOMEM;
1955 }
1956
1957 #define BCH_DEBUG_PARAM(name, description)                      \
1958         bool bch2_##name;                                       \
1959         module_param_named(name, bch2_##name, bool, 0644);      \
1960         MODULE_PARM_DESC(name, description);
1961 BCH_DEBUG_PARAMS()
1962 #undef BCH_DEBUG_PARAM
1963
1964 module_exit(bcachefs_exit);
1965 module_init(bcachefs_init);