xref: /linux/fs/bcachefs/super.c (revision 9e56ff53b4115875667760445b028357848b4748)
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_journal_iter.h"
17 #include "btree_key_cache.h"
18 #include "btree_update_interior.h"
19 #include "btree_io.h"
20 #include "btree_write_buffer.h"
21 #include "buckets_waiting_for_journal.h"
22 #include "chardev.h"
23 #include "checksum.h"
24 #include "clock.h"
25 #include "compress.h"
26 #include "debug.h"
27 #include "disk_groups.h"
28 #include "ec.h"
29 #include "errcode.h"
30 #include "error.h"
31 #include "fs.h"
32 #include "fs-io.h"
33 #include "fs-io-buffered.h"
34 #include "fs-io-direct.h"
35 #include "fsck.h"
36 #include "inode.h"
37 #include "io_read.h"
38 #include "io_write.h"
39 #include "journal.h"
40 #include "journal_reclaim.h"
41 #include "journal_seq_blacklist.h"
42 #include "move.h"
43 #include "migrate.h"
44 #include "movinggc.h"
45 #include "nocow_locking.h"
46 #include "quota.h"
47 #include "rebalance.h"
48 #include "recovery.h"
49 #include "replicas.h"
50 #include "sb-clean.h"
51 #include "sb-counters.h"
52 #include "sb-errors.h"
53 #include "sb-members.h"
54 #include "snapshot.h"
55 #include "subvolume.h"
56 #include "super.h"
57 #include "super-io.h"
58 #include "sysfs.h"
59 #include "trace.h"
60 
61 #include <linux/backing-dev.h>
62 #include <linux/blkdev.h>
63 #include <linux/debugfs.h>
64 #include <linux/device.h>
65 #include <linux/idr.h>
66 #include <linux/module.h>
67 #include <linux/percpu.h>
68 #include <linux/random.h>
69 #include <linux/sysfs.h>
70 #include <crypto/hash.h>
71 
72 MODULE_LICENSE("GPL");
73 MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
74 MODULE_DESCRIPTION("bcachefs filesystem");
75 MODULE_SOFTDEP("pre: crc32c");
76 MODULE_SOFTDEP("pre: crc64");
77 MODULE_SOFTDEP("pre: sha256");
78 MODULE_SOFTDEP("pre: chacha20");
79 MODULE_SOFTDEP("pre: poly1305");
80 MODULE_SOFTDEP("pre: xxhash");
81 
82 const char * const bch2_fs_flag_strs[] = {
83 #define x(n)		#n,
84 	BCH_FS_FLAGS()
85 #undef x
86 	NULL
87 };
88 
89 void __bch2_print(struct bch_fs *c, const char *fmt, ...)
90 {
91 	struct stdio_redirect *stdio = bch2_fs_stdio_redirect(c);
92 
93 	va_list args;
94 	va_start(args, fmt);
95 	if (likely(!stdio)) {
96 		vprintk(fmt, args);
97 	} else {
98 		unsigned long flags;
99 
100 		if (fmt[0] == KERN_SOH[0])
101 			fmt += 2;
102 
103 		spin_lock_irqsave(&stdio->output_lock, flags);
104 		prt_vprintf(&stdio->output_buf, fmt, args);
105 		spin_unlock_irqrestore(&stdio->output_lock, flags);
106 
107 		wake_up(&stdio->output_wait);
108 	}
109 	va_end(args);
110 }
111 
112 #define KTYPE(type)							\
113 static const struct attribute_group type ## _group = {			\
114 	.attrs = type ## _files						\
115 };									\
116 									\
117 static const struct attribute_group *type ## _groups[] = {		\
118 	&type ## _group,						\
119 	NULL								\
120 };									\
121 									\
122 static const struct kobj_type type ## _ktype = {			\
123 	.release	= type ## _release,				\
124 	.sysfs_ops	= &type ## _sysfs_ops,				\
125 	.default_groups = type ## _groups				\
126 }
127 
128 static void bch2_fs_release(struct kobject *);
129 static void bch2_dev_release(struct kobject *);
130 static void bch2_fs_counters_release(struct kobject *k)
131 {
132 }
133 
134 static void bch2_fs_internal_release(struct kobject *k)
135 {
136 }
137 
138 static void bch2_fs_opts_dir_release(struct kobject *k)
139 {
140 }
141 
142 static void bch2_fs_time_stats_release(struct kobject *k)
143 {
144 }
145 
146 KTYPE(bch2_fs);
147 KTYPE(bch2_fs_counters);
148 KTYPE(bch2_fs_internal);
149 KTYPE(bch2_fs_opts_dir);
150 KTYPE(bch2_fs_time_stats);
151 KTYPE(bch2_dev);
152 
153 static struct kset *bcachefs_kset;
154 static LIST_HEAD(bch_fs_list);
155 static DEFINE_MUTEX(bch_fs_list_lock);
156 
157 DECLARE_WAIT_QUEUE_HEAD(bch2_read_only_wait);
158 
159 static void bch2_dev_free(struct bch_dev *);
160 static int bch2_dev_alloc(struct bch_fs *, unsigned);
161 static int bch2_dev_sysfs_online(struct bch_fs *, struct bch_dev *);
162 static void __bch2_dev_read_only(struct bch_fs *, struct bch_dev *);
163 
164 struct bch_fs *bch2_dev_to_fs(dev_t dev)
165 {
166 	struct bch_fs *c;
167 
168 	mutex_lock(&bch_fs_list_lock);
169 	rcu_read_lock();
170 
171 	list_for_each_entry(c, &bch_fs_list, list)
172 		for_each_member_device_rcu(c, ca, NULL)
173 			if (ca->disk_sb.bdev && ca->disk_sb.bdev->bd_dev == dev) {
174 				closure_get(&c->cl);
175 				goto found;
176 			}
177 	c = NULL;
178 found:
179 	rcu_read_unlock();
180 	mutex_unlock(&bch_fs_list_lock);
181 
182 	return c;
183 }
184 
185 static struct bch_fs *__bch2_uuid_to_fs(__uuid_t uuid)
186 {
187 	struct bch_fs *c;
188 
189 	lockdep_assert_held(&bch_fs_list_lock);
190 
191 	list_for_each_entry(c, &bch_fs_list, list)
192 		if (!memcmp(&c->disk_sb.sb->uuid, &uuid, sizeof(uuid)))
193 			return c;
194 
195 	return NULL;
196 }
197 
198 struct bch_fs *bch2_uuid_to_fs(__uuid_t uuid)
199 {
200 	struct bch_fs *c;
201 
202 	mutex_lock(&bch_fs_list_lock);
203 	c = __bch2_uuid_to_fs(uuid);
204 	if (c)
205 		closure_get(&c->cl);
206 	mutex_unlock(&bch_fs_list_lock);
207 
208 	return c;
209 }
210 
211 static void bch2_dev_usage_journal_reserve(struct bch_fs *c)
212 {
213 	unsigned nr = 0, u64s =
214 		((sizeof(struct jset_entry_dev_usage) +
215 		  sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR)) /
216 		sizeof(u64);
217 
218 	rcu_read_lock();
219 	for_each_member_device_rcu(c, ca, NULL)
220 		nr++;
221 	rcu_read_unlock();
222 
223 	bch2_journal_entry_res_resize(&c->journal,
224 			&c->dev_usage_journal_res, u64s * nr);
225 }
226 
227 /* Filesystem RO/RW: */
228 
229 /*
230  * For startup/shutdown of RW stuff, the dependencies are:
231  *
232  * - foreground writes depend on copygc and rebalance (to free up space)
233  *
234  * - copygc and rebalance depend on mark and sweep gc (they actually probably
235  *   don't because they either reserve ahead of time or don't block if
236  *   allocations fail, but allocations can require mark and sweep gc to run
237  *   because of generation number wraparound)
238  *
239  * - all of the above depends on the allocator threads
240  *
241  * - allocator depends on the journal (when it rewrites prios and gens)
242  */
243 
244 static void __bch2_fs_read_only(struct bch_fs *c)
245 {
246 	unsigned clean_passes = 0;
247 	u64 seq = 0;
248 
249 	bch2_fs_ec_stop(c);
250 	bch2_open_buckets_stop(c, NULL, true);
251 	bch2_rebalance_stop(c);
252 	bch2_copygc_stop(c);
253 	bch2_gc_thread_stop(c);
254 	bch2_fs_ec_flush(c);
255 
256 	bch_verbose(c, "flushing journal and stopping allocators, journal seq %llu",
257 		    journal_cur_seq(&c->journal));
258 
259 	do {
260 		clean_passes++;
261 
262 		if (bch2_btree_interior_updates_flush(c) ||
263 		    bch2_journal_flush_all_pins(&c->journal) ||
264 		    bch2_btree_flush_all_writes(c) ||
265 		    seq != atomic64_read(&c->journal.seq)) {
266 			seq = atomic64_read(&c->journal.seq);
267 			clean_passes = 0;
268 		}
269 	} while (clean_passes < 2);
270 
271 	bch_verbose(c, "flushing journal and stopping allocators complete, journal seq %llu",
272 		    journal_cur_seq(&c->journal));
273 
274 	if (test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags) &&
275 	    !test_bit(BCH_FS_emergency_ro, &c->flags))
276 		set_bit(BCH_FS_clean_shutdown, &c->flags);
277 	bch2_fs_journal_stop(&c->journal);
278 
279 	/*
280 	 * After stopping journal:
281 	 */
282 	for_each_member_device(c, ca)
283 		bch2_dev_allocator_remove(c, ca);
284 }
285 
286 #ifndef BCH_WRITE_REF_DEBUG
287 static void bch2_writes_disabled(struct percpu_ref *writes)
288 {
289 	struct bch_fs *c = container_of(writes, struct bch_fs, writes);
290 
291 	set_bit(BCH_FS_write_disable_complete, &c->flags);
292 	wake_up(&bch2_read_only_wait);
293 }
294 #endif
295 
296 void bch2_fs_read_only(struct bch_fs *c)
297 {
298 	if (!test_bit(BCH_FS_rw, &c->flags)) {
299 		bch2_journal_reclaim_stop(&c->journal);
300 		return;
301 	}
302 
303 	BUG_ON(test_bit(BCH_FS_write_disable_complete, &c->flags));
304 
305 	bch_verbose(c, "going read-only");
306 
307 	/*
308 	 * Block new foreground-end write operations from starting - any new
309 	 * writes will return -EROFS:
310 	 */
311 	set_bit(BCH_FS_going_ro, &c->flags);
312 #ifndef BCH_WRITE_REF_DEBUG
313 	percpu_ref_kill(&c->writes);
314 #else
315 	for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++)
316 		bch2_write_ref_put(c, i);
317 #endif
318 
319 	/*
320 	 * If we're not doing an emergency shutdown, we want to wait on
321 	 * outstanding writes to complete so they don't see spurious errors due
322 	 * to shutting down the allocator:
323 	 *
324 	 * If we are doing an emergency shutdown outstanding writes may
325 	 * hang until we shutdown the allocator so we don't want to wait
326 	 * on outstanding writes before shutting everything down - but
327 	 * we do need to wait on them before returning and signalling
328 	 * that going RO is complete:
329 	 */
330 	wait_event(bch2_read_only_wait,
331 		   test_bit(BCH_FS_write_disable_complete, &c->flags) ||
332 		   test_bit(BCH_FS_emergency_ro, &c->flags));
333 
334 	bool writes_disabled = test_bit(BCH_FS_write_disable_complete, &c->flags);
335 	if (writes_disabled)
336 		bch_verbose(c, "finished waiting for writes to stop");
337 
338 	__bch2_fs_read_only(c);
339 
340 	wait_event(bch2_read_only_wait,
341 		   test_bit(BCH_FS_write_disable_complete, &c->flags));
342 
343 	if (!writes_disabled)
344 		bch_verbose(c, "finished waiting for writes to stop");
345 
346 	clear_bit(BCH_FS_write_disable_complete, &c->flags);
347 	clear_bit(BCH_FS_going_ro, &c->flags);
348 	clear_bit(BCH_FS_rw, &c->flags);
349 
350 	if (!bch2_journal_error(&c->journal) &&
351 	    !test_bit(BCH_FS_error, &c->flags) &&
352 	    !test_bit(BCH_FS_emergency_ro, &c->flags) &&
353 	    test_bit(BCH_FS_started, &c->flags) &&
354 	    test_bit(BCH_FS_clean_shutdown, &c->flags) &&
355 	    !c->opts.norecovery) {
356 		BUG_ON(c->journal.last_empty_seq != journal_cur_seq(&c->journal));
357 		BUG_ON(atomic_read(&c->btree_cache.dirty));
358 		BUG_ON(atomic_long_read(&c->btree_key_cache.nr_dirty));
359 		BUG_ON(c->btree_write_buffer.inc.keys.nr);
360 		BUG_ON(c->btree_write_buffer.flushing.keys.nr);
361 
362 		bch_verbose(c, "marking filesystem clean");
363 		bch2_fs_mark_clean(c);
364 	} else {
365 		bch_verbose(c, "done going read-only, filesystem not clean");
366 	}
367 }
368 
369 static void bch2_fs_read_only_work(struct work_struct *work)
370 {
371 	struct bch_fs *c =
372 		container_of(work, struct bch_fs, read_only_work);
373 
374 	down_write(&c->state_lock);
375 	bch2_fs_read_only(c);
376 	up_write(&c->state_lock);
377 }
378 
379 static void bch2_fs_read_only_async(struct bch_fs *c)
380 {
381 	queue_work(system_long_wq, &c->read_only_work);
382 }
383 
384 bool bch2_fs_emergency_read_only(struct bch_fs *c)
385 {
386 	bool ret = !test_and_set_bit(BCH_FS_emergency_ro, &c->flags);
387 
388 	bch2_journal_halt(&c->journal);
389 	bch2_fs_read_only_async(c);
390 
391 	wake_up(&bch2_read_only_wait);
392 	return ret;
393 }
394 
395 static int bch2_fs_read_write_late(struct bch_fs *c)
396 {
397 	int ret;
398 
399 	/*
400 	 * Data move operations can't run until after check_snapshots has
401 	 * completed, and bch2_snapshot_is_ancestor() is available.
402 	 *
403 	 * Ideally we'd start copygc/rebalance earlier instead of waiting for
404 	 * all of recovery/fsck to complete:
405 	 */
406 	ret = bch2_copygc_start(c);
407 	if (ret) {
408 		bch_err(c, "error starting copygc thread");
409 		return ret;
410 	}
411 
412 	ret = bch2_rebalance_start(c);
413 	if (ret) {
414 		bch_err(c, "error starting rebalance thread");
415 		return ret;
416 	}
417 
418 	return 0;
419 }
420 
421 static int __bch2_fs_read_write(struct bch_fs *c, bool early)
422 {
423 	int ret;
424 
425 	if (test_bit(BCH_FS_initial_gc_unfixed, &c->flags)) {
426 		bch_err(c, "cannot go rw, unfixed btree errors");
427 		return -BCH_ERR_erofs_unfixed_errors;
428 	}
429 
430 	if (test_bit(BCH_FS_rw, &c->flags))
431 		return 0;
432 
433 	bch_info(c, "going read-write");
434 
435 	ret = bch2_sb_members_v2_init(c);
436 	if (ret)
437 		goto err;
438 
439 	ret = bch2_fs_mark_dirty(c);
440 	if (ret)
441 		goto err;
442 
443 	clear_bit(BCH_FS_clean_shutdown, &c->flags);
444 
445 	/*
446 	 * First journal write must be a flush write: after a clean shutdown we
447 	 * don't read the journal, so the first journal write may end up
448 	 * overwriting whatever was there previously, and there must always be
449 	 * at least one non-flush write in the journal or recovery will fail:
450 	 */
451 	set_bit(JOURNAL_NEED_FLUSH_WRITE, &c->journal.flags);
452 
453 	for_each_rw_member(c, ca)
454 		bch2_dev_allocator_add(c, ca);
455 	bch2_recalc_capacity(c);
456 
457 	set_bit(BCH_FS_rw, &c->flags);
458 	set_bit(BCH_FS_was_rw, &c->flags);
459 
460 #ifndef BCH_WRITE_REF_DEBUG
461 	percpu_ref_reinit(&c->writes);
462 #else
463 	for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++) {
464 		BUG_ON(atomic_long_read(&c->writes[i]));
465 		atomic_long_inc(&c->writes[i]);
466 	}
467 #endif
468 
469 	ret = bch2_gc_thread_start(c);
470 	if (ret) {
471 		bch_err(c, "error starting gc thread");
472 		return ret;
473 	}
474 
475 	ret = bch2_journal_reclaim_start(&c->journal);
476 	if (ret)
477 		goto err;
478 
479 	if (!early) {
480 		ret = bch2_fs_read_write_late(c);
481 		if (ret)
482 			goto err;
483 	}
484 
485 	bch2_do_discards(c);
486 	bch2_do_invalidates(c);
487 	bch2_do_stripe_deletes(c);
488 	bch2_do_pending_node_rewrites(c);
489 	return 0;
490 err:
491 	if (test_bit(BCH_FS_rw, &c->flags))
492 		bch2_fs_read_only(c);
493 	else
494 		__bch2_fs_read_only(c);
495 	return ret;
496 }
497 
498 int bch2_fs_read_write(struct bch_fs *c)
499 {
500 	if (c->opts.norecovery)
501 		return -BCH_ERR_erofs_norecovery;
502 
503 	if (c->opts.nochanges)
504 		return -BCH_ERR_erofs_nochanges;
505 
506 	return __bch2_fs_read_write(c, false);
507 }
508 
509 int bch2_fs_read_write_early(struct bch_fs *c)
510 {
511 	lockdep_assert_held(&c->state_lock);
512 
513 	return __bch2_fs_read_write(c, true);
514 }
515 
516 /* Filesystem startup/shutdown: */
517 
518 static void __bch2_fs_free(struct bch_fs *c)
519 {
520 	unsigned i;
521 
522 	for (i = 0; i < BCH_TIME_STAT_NR; i++)
523 		bch2_time_stats_exit(&c->times[i]);
524 
525 	bch2_free_pending_node_rewrites(c);
526 	bch2_fs_sb_errors_exit(c);
527 	bch2_fs_counters_exit(c);
528 	bch2_fs_snapshots_exit(c);
529 	bch2_fs_quota_exit(c);
530 	bch2_fs_fs_io_direct_exit(c);
531 	bch2_fs_fs_io_buffered_exit(c);
532 	bch2_fs_fsio_exit(c);
533 	bch2_fs_ec_exit(c);
534 	bch2_fs_encryption_exit(c);
535 	bch2_fs_nocow_locking_exit(c);
536 	bch2_fs_io_write_exit(c);
537 	bch2_fs_io_read_exit(c);
538 	bch2_fs_buckets_waiting_for_journal_exit(c);
539 	bch2_fs_btree_interior_update_exit(c);
540 	bch2_fs_btree_iter_exit(c);
541 	bch2_fs_btree_key_cache_exit(&c->btree_key_cache);
542 	bch2_fs_btree_cache_exit(c);
543 	bch2_fs_replicas_exit(c);
544 	bch2_fs_journal_exit(&c->journal);
545 	bch2_io_clock_exit(&c->io_clock[WRITE]);
546 	bch2_io_clock_exit(&c->io_clock[READ]);
547 	bch2_fs_compress_exit(c);
548 	bch2_journal_keys_put_initial(c);
549 	BUG_ON(atomic_read(&c->journal_keys.ref));
550 	bch2_fs_btree_write_buffer_exit(c);
551 	percpu_free_rwsem(&c->mark_lock);
552 	free_percpu(c->online_reserved);
553 
554 	darray_exit(&c->btree_roots_extra);
555 	free_percpu(c->pcpu);
556 	mempool_exit(&c->large_bkey_pool);
557 	mempool_exit(&c->btree_bounce_pool);
558 	bioset_exit(&c->btree_bio);
559 	mempool_exit(&c->fill_iter);
560 #ifndef BCH_WRITE_REF_DEBUG
561 	percpu_ref_exit(&c->writes);
562 #endif
563 	kfree(rcu_dereference_protected(c->disk_groups, 1));
564 	kfree(c->journal_seq_blacklist_table);
565 	kfree(c->unused_inode_hints);
566 
567 	if (c->write_ref_wq)
568 		destroy_workqueue(c->write_ref_wq);
569 	if (c->io_complete_wq)
570 		destroy_workqueue(c->io_complete_wq);
571 	if (c->copygc_wq)
572 		destroy_workqueue(c->copygc_wq);
573 	if (c->btree_io_complete_wq)
574 		destroy_workqueue(c->btree_io_complete_wq);
575 	if (c->btree_update_wq)
576 		destroy_workqueue(c->btree_update_wq);
577 
578 	bch2_free_super(&c->disk_sb);
579 	kvpfree(c, sizeof(*c));
580 	module_put(THIS_MODULE);
581 }
582 
583 static void bch2_fs_release(struct kobject *kobj)
584 {
585 	struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
586 
587 	__bch2_fs_free(c);
588 }
589 
590 void __bch2_fs_stop(struct bch_fs *c)
591 {
592 	bch_verbose(c, "shutting down");
593 
594 	set_bit(BCH_FS_stopping, &c->flags);
595 
596 	cancel_work_sync(&c->journal_seq_blacklist_gc_work);
597 
598 	down_write(&c->state_lock);
599 	bch2_fs_read_only(c);
600 	up_write(&c->state_lock);
601 
602 	for_each_member_device(c, ca)
603 		if (ca->kobj.state_in_sysfs &&
604 		    ca->disk_sb.bdev)
605 			sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
606 
607 	if (c->kobj.state_in_sysfs)
608 		kobject_del(&c->kobj);
609 
610 	bch2_fs_debug_exit(c);
611 	bch2_fs_chardev_exit(c);
612 
613 	bch2_ro_ref_put(c);
614 	wait_event(c->ro_ref_wait, !refcount_read(&c->ro_ref));
615 
616 	kobject_put(&c->counters_kobj);
617 	kobject_put(&c->time_stats);
618 	kobject_put(&c->opts_dir);
619 	kobject_put(&c->internal);
620 
621 	/* btree prefetch might have kicked off reads in the background: */
622 	bch2_btree_flush_all_reads(c);
623 
624 	for_each_member_device(c, ca)
625 		cancel_work_sync(&ca->io_error_work);
626 
627 	cancel_work_sync(&c->read_only_work);
628 }
629 
630 void bch2_fs_free(struct bch_fs *c)
631 {
632 	unsigned i;
633 
634 	mutex_lock(&bch_fs_list_lock);
635 	list_del(&c->list);
636 	mutex_unlock(&bch_fs_list_lock);
637 
638 	closure_sync(&c->cl);
639 	closure_debug_destroy(&c->cl);
640 
641 	for (i = 0; i < c->sb.nr_devices; i++) {
642 		struct bch_dev *ca = rcu_dereference_protected(c->devs[i], true);
643 
644 		if (ca) {
645 			bch2_free_super(&ca->disk_sb);
646 			bch2_dev_free(ca);
647 		}
648 	}
649 
650 	bch_verbose(c, "shutdown complete");
651 
652 	kobject_put(&c->kobj);
653 }
654 
655 void bch2_fs_stop(struct bch_fs *c)
656 {
657 	__bch2_fs_stop(c);
658 	bch2_fs_free(c);
659 }
660 
661 static int bch2_fs_online(struct bch_fs *c)
662 {
663 	int ret = 0;
664 
665 	lockdep_assert_held(&bch_fs_list_lock);
666 
667 	if (__bch2_uuid_to_fs(c->sb.uuid)) {
668 		bch_err(c, "filesystem UUID already open");
669 		return -EINVAL;
670 	}
671 
672 	ret = bch2_fs_chardev_init(c);
673 	if (ret) {
674 		bch_err(c, "error creating character device");
675 		return ret;
676 	}
677 
678 	bch2_fs_debug_init(c);
679 
680 	ret = kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ?:
681 	    kobject_add(&c->internal, &c->kobj, "internal") ?:
682 	    kobject_add(&c->opts_dir, &c->kobj, "options") ?:
683 #ifndef CONFIG_BCACHEFS_NO_LATENCY_ACCT
684 	    kobject_add(&c->time_stats, &c->kobj, "time_stats") ?:
685 #endif
686 	    kobject_add(&c->counters_kobj, &c->kobj, "counters") ?:
687 	    bch2_opts_create_sysfs_files(&c->opts_dir);
688 	if (ret) {
689 		bch_err(c, "error creating sysfs objects");
690 		return ret;
691 	}
692 
693 	down_write(&c->state_lock);
694 
695 	for_each_member_device(c, ca) {
696 		ret = bch2_dev_sysfs_online(c, ca);
697 		if (ret) {
698 			bch_err(c, "error creating sysfs objects");
699 			percpu_ref_put(&ca->ref);
700 			goto err;
701 		}
702 	}
703 
704 	BUG_ON(!list_empty(&c->list));
705 	list_add(&c->list, &bch_fs_list);
706 err:
707 	up_write(&c->state_lock);
708 	return ret;
709 }
710 
711 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
712 {
713 	struct bch_fs *c;
714 	struct printbuf name = PRINTBUF;
715 	unsigned i, iter_size;
716 	int ret = 0;
717 
718 	c = kvpmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
719 	if (!c) {
720 		c = ERR_PTR(-BCH_ERR_ENOMEM_fs_alloc);
721 		goto out;
722 	}
723 
724 	c->stdio = (void *)(unsigned long) opts.stdio;
725 
726 	__module_get(THIS_MODULE);
727 
728 	closure_init(&c->cl, NULL);
729 
730 	c->kobj.kset = bcachefs_kset;
731 	kobject_init(&c->kobj, &bch2_fs_ktype);
732 	kobject_init(&c->internal, &bch2_fs_internal_ktype);
733 	kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
734 	kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
735 	kobject_init(&c->counters_kobj, &bch2_fs_counters_ktype);
736 
737 	c->minor		= -1;
738 	c->disk_sb.fs_sb	= true;
739 
740 	init_rwsem(&c->state_lock);
741 	mutex_init(&c->sb_lock);
742 	mutex_init(&c->replicas_gc_lock);
743 	mutex_init(&c->btree_root_lock);
744 	INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
745 
746 	refcount_set(&c->ro_ref, 1);
747 	init_waitqueue_head(&c->ro_ref_wait);
748 	sema_init(&c->online_fsck_mutex, 1);
749 
750 	init_rwsem(&c->gc_lock);
751 	mutex_init(&c->gc_gens_lock);
752 	atomic_set(&c->journal_keys.ref, 1);
753 	c->journal_keys.initial_ref_held = true;
754 
755 	for (i = 0; i < BCH_TIME_STAT_NR; i++)
756 		bch2_time_stats_init(&c->times[i]);
757 
758 	bch2_fs_copygc_init(c);
759 	bch2_fs_btree_key_cache_init_early(&c->btree_key_cache);
760 	bch2_fs_btree_iter_init_early(c);
761 	bch2_fs_btree_interior_update_init_early(c);
762 	bch2_fs_allocator_background_init(c);
763 	bch2_fs_allocator_foreground_init(c);
764 	bch2_fs_rebalance_init(c);
765 	bch2_fs_quota_init(c);
766 	bch2_fs_ec_init_early(c);
767 	bch2_fs_move_init(c);
768 	bch2_fs_sb_errors_init_early(c);
769 
770 	INIT_LIST_HEAD(&c->list);
771 
772 	mutex_init(&c->usage_scratch_lock);
773 
774 	mutex_init(&c->bio_bounce_pages_lock);
775 	mutex_init(&c->snapshot_table_lock);
776 	init_rwsem(&c->snapshot_create_lock);
777 
778 	spin_lock_init(&c->btree_write_error_lock);
779 
780 	INIT_WORK(&c->journal_seq_blacklist_gc_work,
781 		  bch2_blacklist_entries_gc);
782 
783 	INIT_LIST_HEAD(&c->journal_iters);
784 
785 	INIT_LIST_HEAD(&c->fsck_error_msgs);
786 	mutex_init(&c->fsck_error_msgs_lock);
787 
788 	seqcount_init(&c->gc_pos_lock);
789 
790 	seqcount_init(&c->usage_lock);
791 
792 	sema_init(&c->io_in_flight, 128);
793 
794 	INIT_LIST_HEAD(&c->vfs_inodes_list);
795 	mutex_init(&c->vfs_inodes_lock);
796 
797 	c->copy_gc_enabled		= 1;
798 	c->rebalance.enabled		= 1;
799 	c->promote_whole_extents	= true;
800 
801 	c->journal.flush_write_time	= &c->times[BCH_TIME_journal_flush_write];
802 	c->journal.noflush_write_time	= &c->times[BCH_TIME_journal_noflush_write];
803 	c->journal.flush_seq_time	= &c->times[BCH_TIME_journal_flush_seq];
804 
805 	bch2_fs_btree_cache_init_early(&c->btree_cache);
806 
807 	mutex_init(&c->sectors_available_lock);
808 
809 	ret = percpu_init_rwsem(&c->mark_lock);
810 	if (ret)
811 		goto err;
812 
813 	mutex_lock(&c->sb_lock);
814 	ret = bch2_sb_to_fs(c, sb);
815 	mutex_unlock(&c->sb_lock);
816 
817 	if (ret)
818 		goto err;
819 
820 	pr_uuid(&name, c->sb.user_uuid.b);
821 	strscpy(c->name, name.buf, sizeof(c->name));
822 	printbuf_exit(&name);
823 
824 	ret = name.allocation_failure ? -BCH_ERR_ENOMEM_fs_name_alloc : 0;
825 	if (ret)
826 		goto err;
827 
828 	/* Compat: */
829 	if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_inode_v2 &&
830 	    !BCH_SB_JOURNAL_FLUSH_DELAY(sb))
831 		SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
832 
833 	if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_inode_v2 &&
834 	    !BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
835 		SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 100);
836 
837 	c->opts = bch2_opts_default;
838 	ret = bch2_opts_from_sb(&c->opts, sb);
839 	if (ret)
840 		goto err;
841 
842 	bch2_opts_apply(&c->opts, opts);
843 
844 	c->btree_key_cache_btrees |= 1U << BTREE_ID_alloc;
845 	if (c->opts.inodes_use_key_cache)
846 		c->btree_key_cache_btrees |= 1U << BTREE_ID_inodes;
847 	c->btree_key_cache_btrees |= 1U << BTREE_ID_logged_ops;
848 
849 	c->block_bits		= ilog2(block_sectors(c));
850 	c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
851 
852 	if (bch2_fs_init_fault("fs_alloc")) {
853 		bch_err(c, "fs_alloc fault injected");
854 		ret = -EFAULT;
855 		goto err;
856 	}
857 
858 	iter_size = sizeof(struct sort_iter) +
859 		(btree_blocks(c) + 1) * 2 *
860 		sizeof(struct sort_iter_set);
861 
862 	c->inode_shard_bits = ilog2(roundup_pow_of_two(num_possible_cpus()));
863 
864 	if (!(c->btree_update_wq = alloc_workqueue("bcachefs",
865 				WQ_FREEZABLE|WQ_UNBOUND|WQ_MEM_RECLAIM, 512)) ||
866 	    !(c->btree_io_complete_wq = alloc_workqueue("bcachefs_btree_io",
867 				WQ_FREEZABLE|WQ_MEM_RECLAIM, 1)) ||
868 	    !(c->copygc_wq = alloc_workqueue("bcachefs_copygc",
869 				WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
870 	    !(c->io_complete_wq = alloc_workqueue("bcachefs_io",
871 				WQ_FREEZABLE|WQ_HIGHPRI|WQ_MEM_RECLAIM, 512)) ||
872 	    !(c->write_ref_wq = alloc_workqueue("bcachefs_write_ref",
873 				WQ_FREEZABLE, 0)) ||
874 #ifndef BCH_WRITE_REF_DEBUG
875 	    percpu_ref_init(&c->writes, bch2_writes_disabled,
876 			    PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
877 #endif
878 	    mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
879 	    bioset_init(&c->btree_bio, 1,
880 			max(offsetof(struct btree_read_bio, bio),
881 			    offsetof(struct btree_write_bio, wbio.bio)),
882 			BIOSET_NEED_BVECS) ||
883 	    !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
884 	    !(c->online_reserved = alloc_percpu(u64)) ||
885 	    mempool_init_kvpmalloc_pool(&c->btree_bounce_pool, 1,
886 					c->opts.btree_node_size) ||
887 	    mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048) ||
888 	    !(c->unused_inode_hints = kcalloc(1U << c->inode_shard_bits,
889 					      sizeof(u64), GFP_KERNEL))) {
890 		ret = -BCH_ERR_ENOMEM_fs_other_alloc;
891 		goto err;
892 	}
893 
894 	ret = bch2_fs_counters_init(c) ?:
895 	    bch2_fs_sb_errors_init(c) ?:
896 	    bch2_io_clock_init(&c->io_clock[READ]) ?:
897 	    bch2_io_clock_init(&c->io_clock[WRITE]) ?:
898 	    bch2_fs_journal_init(&c->journal) ?:
899 	    bch2_fs_replicas_init(c) ?:
900 	    bch2_fs_btree_cache_init(c) ?:
901 	    bch2_fs_btree_key_cache_init(&c->btree_key_cache) ?:
902 	    bch2_fs_btree_iter_init(c) ?:
903 	    bch2_fs_btree_interior_update_init(c) ?:
904 	    bch2_fs_buckets_waiting_for_journal_init(c) ?:
905 	    bch2_fs_btree_write_buffer_init(c) ?:
906 	    bch2_fs_subvolumes_init(c) ?:
907 	    bch2_fs_io_read_init(c) ?:
908 	    bch2_fs_io_write_init(c) ?:
909 	    bch2_fs_nocow_locking_init(c) ?:
910 	    bch2_fs_encryption_init(c) ?:
911 	    bch2_fs_compress_init(c) ?:
912 	    bch2_fs_ec_init(c) ?:
913 	    bch2_fs_fsio_init(c) ?:
914 	    bch2_fs_fs_io_buffered_init(c) ?:
915 	    bch2_fs_fs_io_direct_init(c);
916 	if (ret)
917 		goto err;
918 
919 	for (i = 0; i < c->sb.nr_devices; i++)
920 		if (bch2_dev_exists(c->disk_sb.sb, i) &&
921 		    bch2_dev_alloc(c, i)) {
922 			ret = -EEXIST;
923 			goto err;
924 		}
925 
926 	bch2_journal_entry_res_resize(&c->journal,
927 			&c->btree_root_journal_res,
928 			BTREE_ID_NR * (JSET_KEYS_U64s + BKEY_BTREE_PTR_U64s_MAX));
929 	bch2_dev_usage_journal_reserve(c);
930 	bch2_journal_entry_res_resize(&c->journal,
931 			&c->clock_journal_res,
932 			(sizeof(struct jset_entry_clock) / sizeof(u64)) * 2);
933 
934 	mutex_lock(&bch_fs_list_lock);
935 	ret = bch2_fs_online(c);
936 	mutex_unlock(&bch_fs_list_lock);
937 
938 	if (ret)
939 		goto err;
940 out:
941 	return c;
942 err:
943 	bch2_fs_free(c);
944 	c = ERR_PTR(ret);
945 	goto out;
946 }
947 
948 noinline_for_stack
949 static void print_mount_opts(struct bch_fs *c)
950 {
951 	enum bch_opt_id i;
952 	struct printbuf p = PRINTBUF;
953 	bool first = true;
954 
955 	prt_str(&p, "mounting version ");
956 	bch2_version_to_text(&p, c->sb.version);
957 
958 	if (c->opts.read_only) {
959 		prt_str(&p, " opts=");
960 		first = false;
961 		prt_printf(&p, "ro");
962 	}
963 
964 	for (i = 0; i < bch2_opts_nr; i++) {
965 		const struct bch_option *opt = &bch2_opt_table[i];
966 		u64 v = bch2_opt_get_by_id(&c->opts, i);
967 
968 		if (!(opt->flags & OPT_MOUNT))
969 			continue;
970 
971 		if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
972 			continue;
973 
974 		prt_str(&p, first ? " opts=" : ",");
975 		first = false;
976 		bch2_opt_to_text(&p, c, c->disk_sb.sb, opt, v, OPT_SHOW_MOUNT_STYLE);
977 	}
978 
979 	bch_info(c, "%s", p.buf);
980 	printbuf_exit(&p);
981 }
982 
983 int bch2_fs_start(struct bch_fs *c)
984 {
985 	time64_t now = ktime_get_real_seconds();
986 	int ret;
987 
988 	print_mount_opts(c);
989 
990 	down_write(&c->state_lock);
991 
992 	BUG_ON(test_bit(BCH_FS_started, &c->flags));
993 
994 	mutex_lock(&c->sb_lock);
995 
996 	ret = bch2_sb_members_v2_init(c);
997 	if (ret) {
998 		mutex_unlock(&c->sb_lock);
999 		goto err;
1000 	}
1001 
1002 	for_each_online_member(c, ca)
1003 		bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx)->last_mount = cpu_to_le64(now);
1004 
1005 	mutex_unlock(&c->sb_lock);
1006 
1007 	for_each_rw_member(c, ca)
1008 		bch2_dev_allocator_add(c, ca);
1009 	bch2_recalc_capacity(c);
1010 
1011 	ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
1012 		? bch2_fs_recovery(c)
1013 		: bch2_fs_initialize(c);
1014 	if (ret)
1015 		goto err;
1016 
1017 	ret = bch2_opts_check_may_set(c);
1018 	if (ret)
1019 		goto err;
1020 
1021 	if (bch2_fs_init_fault("fs_start")) {
1022 		bch_err(c, "fs_start fault injected");
1023 		ret = -EINVAL;
1024 		goto err;
1025 	}
1026 
1027 	set_bit(BCH_FS_started, &c->flags);
1028 
1029 	if (c->opts.read_only) {
1030 		bch2_fs_read_only(c);
1031 	} else {
1032 		ret = !test_bit(BCH_FS_rw, &c->flags)
1033 			? bch2_fs_read_write(c)
1034 			: bch2_fs_read_write_late(c);
1035 		if (ret)
1036 			goto err;
1037 	}
1038 
1039 	ret = 0;
1040 err:
1041 	if (ret)
1042 		bch_err_msg(c, ret, "starting filesystem");
1043 	else
1044 		bch_verbose(c, "done starting filesystem");
1045 	up_write(&c->state_lock);
1046 	return ret;
1047 }
1048 
1049 static int bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
1050 {
1051 	struct bch_member m = bch2_sb_member_get(sb, sb->dev_idx);
1052 
1053 	if (le16_to_cpu(sb->block_size) != block_sectors(c))
1054 		return -BCH_ERR_mismatched_block_size;
1055 
1056 	if (le16_to_cpu(m.bucket_size) <
1057 	    BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
1058 		return -BCH_ERR_bucket_size_too_small;
1059 
1060 	return 0;
1061 }
1062 
1063 static int bch2_dev_in_fs(struct bch_sb_handle *fs,
1064 			  struct bch_sb_handle *sb)
1065 {
1066 	if (fs == sb)
1067 		return 0;
1068 
1069 	if (!uuid_equal(&fs->sb->uuid, &sb->sb->uuid))
1070 		return -BCH_ERR_device_not_a_member_of_filesystem;
1071 
1072 	if (!bch2_dev_exists(fs->sb, sb->sb->dev_idx))
1073 		return -BCH_ERR_device_has_been_removed;
1074 
1075 	if (fs->sb->block_size != sb->sb->block_size)
1076 		return -BCH_ERR_mismatched_block_size;
1077 
1078 	if (le16_to_cpu(fs->sb->version) < bcachefs_metadata_version_member_seq ||
1079 	    le16_to_cpu(sb->sb->version) < bcachefs_metadata_version_member_seq)
1080 		return 0;
1081 
1082 	if (fs->sb->seq == sb->sb->seq &&
1083 	    fs->sb->write_time != sb->sb->write_time) {
1084 		struct printbuf buf = PRINTBUF;
1085 
1086 		prt_str(&buf, "Split brain detected between ");
1087 		prt_bdevname(&buf, sb->bdev);
1088 		prt_str(&buf, " and ");
1089 		prt_bdevname(&buf, fs->bdev);
1090 		prt_char(&buf, ':');
1091 		prt_newline(&buf);
1092 		prt_printf(&buf, "seq=%llu but write_time different, got", le64_to_cpu(sb->sb->seq));
1093 		prt_newline(&buf);
1094 
1095 		prt_bdevname(&buf, fs->bdev);
1096 		prt_char(&buf, ' ');
1097 		bch2_prt_datetime(&buf, le64_to_cpu(fs->sb->write_time));;
1098 		prt_newline(&buf);
1099 
1100 		prt_bdevname(&buf, sb->bdev);
1101 		prt_char(&buf, ' ');
1102 		bch2_prt_datetime(&buf, le64_to_cpu(sb->sb->write_time));;
1103 		prt_newline(&buf);
1104 
1105 		prt_printf(&buf, "Not using older sb");
1106 
1107 		pr_err("%s", buf.buf);
1108 		printbuf_exit(&buf);
1109 		return -BCH_ERR_device_splitbrain;
1110 	}
1111 
1112 	struct bch_member m = bch2_sb_member_get(fs->sb, sb->sb->dev_idx);
1113 	u64 seq_from_fs		= le64_to_cpu(m.seq);
1114 	u64 seq_from_member	= le64_to_cpu(sb->sb->seq);
1115 
1116 	if (seq_from_fs && seq_from_fs < seq_from_member) {
1117 		struct printbuf buf = PRINTBUF;
1118 
1119 		prt_str(&buf, "Split brain detected between ");
1120 		prt_bdevname(&buf, sb->bdev);
1121 		prt_str(&buf, " and ");
1122 		prt_bdevname(&buf, fs->bdev);
1123 		prt_char(&buf, ':');
1124 		prt_newline(&buf);
1125 
1126 		prt_bdevname(&buf, fs->bdev);
1127 		prt_str(&buf, "believes seq of ");
1128 		prt_bdevname(&buf, sb->bdev);
1129 		prt_printf(&buf, " to be %llu, but ", seq_from_fs);
1130 		prt_bdevname(&buf, sb->bdev);
1131 		prt_printf(&buf, " has %llu\n", seq_from_member);
1132 		prt_str(&buf, "Not using ");
1133 		prt_bdevname(&buf, sb->bdev);
1134 
1135 		pr_err("%s", buf.buf);
1136 		printbuf_exit(&buf);
1137 		return -BCH_ERR_device_splitbrain;
1138 	}
1139 
1140 	return 0;
1141 }
1142 
1143 /* Device startup/shutdown: */
1144 
1145 static void bch2_dev_release(struct kobject *kobj)
1146 {
1147 	struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
1148 
1149 	kfree(ca);
1150 }
1151 
1152 static void bch2_dev_free(struct bch_dev *ca)
1153 {
1154 	cancel_work_sync(&ca->io_error_work);
1155 
1156 	if (ca->kobj.state_in_sysfs &&
1157 	    ca->disk_sb.bdev)
1158 		sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1159 
1160 	if (ca->kobj.state_in_sysfs)
1161 		kobject_del(&ca->kobj);
1162 
1163 	bch2_free_super(&ca->disk_sb);
1164 	bch2_dev_journal_exit(ca);
1165 
1166 	free_percpu(ca->io_done);
1167 	bioset_exit(&ca->replica_set);
1168 	bch2_dev_buckets_free(ca);
1169 	free_page((unsigned long) ca->sb_read_scratch);
1170 
1171 	bch2_time_stats_exit(&ca->io_latency[WRITE]);
1172 	bch2_time_stats_exit(&ca->io_latency[READ]);
1173 
1174 	percpu_ref_exit(&ca->io_ref);
1175 	percpu_ref_exit(&ca->ref);
1176 	kobject_put(&ca->kobj);
1177 }
1178 
1179 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
1180 {
1181 
1182 	lockdep_assert_held(&c->state_lock);
1183 
1184 	if (percpu_ref_is_zero(&ca->io_ref))
1185 		return;
1186 
1187 	__bch2_dev_read_only(c, ca);
1188 
1189 	reinit_completion(&ca->io_ref_completion);
1190 	percpu_ref_kill(&ca->io_ref);
1191 	wait_for_completion(&ca->io_ref_completion);
1192 
1193 	if (ca->kobj.state_in_sysfs) {
1194 		sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1195 		sysfs_remove_link(&ca->kobj, "block");
1196 	}
1197 
1198 	bch2_free_super(&ca->disk_sb);
1199 	bch2_dev_journal_exit(ca);
1200 }
1201 
1202 static void bch2_dev_ref_complete(struct percpu_ref *ref)
1203 {
1204 	struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
1205 
1206 	complete(&ca->ref_completion);
1207 }
1208 
1209 static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
1210 {
1211 	struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
1212 
1213 	complete(&ca->io_ref_completion);
1214 }
1215 
1216 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
1217 {
1218 	int ret;
1219 
1220 	if (!c->kobj.state_in_sysfs)
1221 		return 0;
1222 
1223 	if (!ca->kobj.state_in_sysfs) {
1224 		ret = kobject_add(&ca->kobj, &c->kobj,
1225 				  "dev-%u", ca->dev_idx);
1226 		if (ret)
1227 			return ret;
1228 	}
1229 
1230 	if (ca->disk_sb.bdev) {
1231 		struct kobject *block = bdev_kobj(ca->disk_sb.bdev);
1232 
1233 		ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
1234 		if (ret)
1235 			return ret;
1236 
1237 		ret = sysfs_create_link(&ca->kobj, block, "block");
1238 		if (ret)
1239 			return ret;
1240 	}
1241 
1242 	return 0;
1243 }
1244 
1245 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
1246 					struct bch_member *member)
1247 {
1248 	struct bch_dev *ca;
1249 	unsigned i;
1250 
1251 	ca = kzalloc(sizeof(*ca), GFP_KERNEL);
1252 	if (!ca)
1253 		return NULL;
1254 
1255 	kobject_init(&ca->kobj, &bch2_dev_ktype);
1256 	init_completion(&ca->ref_completion);
1257 	init_completion(&ca->io_ref_completion);
1258 
1259 	init_rwsem(&ca->bucket_lock);
1260 
1261 	INIT_WORK(&ca->io_error_work, bch2_io_error_work);
1262 
1263 	bch2_time_stats_init(&ca->io_latency[READ]);
1264 	bch2_time_stats_init(&ca->io_latency[WRITE]);
1265 
1266 	ca->mi = bch2_mi_to_cpu(member);
1267 
1268 	for (i = 0; i < ARRAY_SIZE(member->errors); i++)
1269 		atomic64_set(&ca->errors[i], le64_to_cpu(member->errors[i]));
1270 
1271 	ca->uuid = member->uuid;
1272 
1273 	ca->nr_btree_reserve = DIV_ROUND_UP(BTREE_NODE_RESERVE,
1274 			     ca->mi.bucket_size / btree_sectors(c));
1275 
1276 	if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
1277 			    0, GFP_KERNEL) ||
1278 	    percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
1279 			    PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
1280 	    !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) ||
1281 	    bch2_dev_buckets_alloc(c, ca) ||
1282 	    bioset_init(&ca->replica_set, 4,
1283 			offsetof(struct bch_write_bio, bio), 0) ||
1284 	    !(ca->io_done	= alloc_percpu(*ca->io_done)))
1285 		goto err;
1286 
1287 	return ca;
1288 err:
1289 	bch2_dev_free(ca);
1290 	return NULL;
1291 }
1292 
1293 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
1294 			    unsigned dev_idx)
1295 {
1296 	ca->dev_idx = dev_idx;
1297 	__set_bit(ca->dev_idx, ca->self.d);
1298 	scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
1299 
1300 	ca->fs = c;
1301 	rcu_assign_pointer(c->devs[ca->dev_idx], ca);
1302 
1303 	if (bch2_dev_sysfs_online(c, ca))
1304 		pr_warn("error creating sysfs objects");
1305 }
1306 
1307 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
1308 {
1309 	struct bch_member member = bch2_sb_member_get(c->disk_sb.sb, dev_idx);
1310 	struct bch_dev *ca = NULL;
1311 	int ret = 0;
1312 
1313 	if (bch2_fs_init_fault("dev_alloc"))
1314 		goto err;
1315 
1316 	ca = __bch2_dev_alloc(c, &member);
1317 	if (!ca)
1318 		goto err;
1319 
1320 	ca->fs = c;
1321 
1322 	bch2_dev_attach(c, ca, dev_idx);
1323 	return ret;
1324 err:
1325 	if (ca)
1326 		bch2_dev_free(ca);
1327 	return -BCH_ERR_ENOMEM_dev_alloc;
1328 }
1329 
1330 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1331 {
1332 	unsigned ret;
1333 
1334 	if (bch2_dev_is_online(ca)) {
1335 		bch_err(ca, "already have device online in slot %u",
1336 			sb->sb->dev_idx);
1337 		return -BCH_ERR_device_already_online;
1338 	}
1339 
1340 	if (get_capacity(sb->bdev->bd_disk) <
1341 	    ca->mi.bucket_size * ca->mi.nbuckets) {
1342 		bch_err(ca, "cannot online: device too small");
1343 		return -BCH_ERR_device_size_too_small;
1344 	}
1345 
1346 	BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1347 
1348 	ret = bch2_dev_journal_init(ca, sb->sb);
1349 	if (ret)
1350 		return ret;
1351 
1352 	/* Commit: */
1353 	ca->disk_sb = *sb;
1354 	memset(sb, 0, sizeof(*sb));
1355 
1356 	ca->dev = ca->disk_sb.bdev->bd_dev;
1357 
1358 	percpu_ref_reinit(&ca->io_ref);
1359 
1360 	return 0;
1361 }
1362 
1363 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1364 {
1365 	struct bch_dev *ca;
1366 	int ret;
1367 
1368 	lockdep_assert_held(&c->state_lock);
1369 
1370 	if (le64_to_cpu(sb->sb->seq) >
1371 	    le64_to_cpu(c->disk_sb.sb->seq))
1372 		bch2_sb_to_fs(c, sb->sb);
1373 
1374 	BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
1375 	       !c->devs[sb->sb->dev_idx]);
1376 
1377 	ca = bch_dev_locked(c, sb->sb->dev_idx);
1378 
1379 	ret = __bch2_dev_attach_bdev(ca, sb);
1380 	if (ret)
1381 		return ret;
1382 
1383 	bch2_dev_sysfs_online(c, ca);
1384 
1385 	struct printbuf name = PRINTBUF;
1386 	prt_bdevname(&name, ca->disk_sb.bdev);
1387 
1388 	if (c->sb.nr_devices == 1)
1389 		strscpy(c->name, name.buf, sizeof(c->name));
1390 	strscpy(ca->name, name.buf, sizeof(ca->name));
1391 
1392 	printbuf_exit(&name);
1393 
1394 	rebalance_wakeup(c);
1395 	return 0;
1396 }
1397 
1398 /* Device management: */
1399 
1400 /*
1401  * Note: this function is also used by the error paths - when a particular
1402  * device sees an error, we call it to determine whether we can just set the
1403  * device RO, or - if this function returns false - we'll set the whole
1404  * filesystem RO:
1405  *
1406  * XXX: maybe we should be more explicit about whether we're changing state
1407  * because we got an error or what have you?
1408  */
1409 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1410 			    enum bch_member_state new_state, int flags)
1411 {
1412 	struct bch_devs_mask new_online_devs;
1413 	int nr_rw = 0, required;
1414 
1415 	lockdep_assert_held(&c->state_lock);
1416 
1417 	switch (new_state) {
1418 	case BCH_MEMBER_STATE_rw:
1419 		return true;
1420 	case BCH_MEMBER_STATE_ro:
1421 		if (ca->mi.state != BCH_MEMBER_STATE_rw)
1422 			return true;
1423 
1424 		/* do we have enough devices to write to?  */
1425 		for_each_member_device(c, ca2)
1426 			if (ca2 != ca)
1427 				nr_rw += ca2->mi.state == BCH_MEMBER_STATE_rw;
1428 
1429 		required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1430 			       ? c->opts.metadata_replicas
1431 			       : c->opts.metadata_replicas_required,
1432 			       !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1433 			       ? c->opts.data_replicas
1434 			       : c->opts.data_replicas_required);
1435 
1436 		return nr_rw >= required;
1437 	case BCH_MEMBER_STATE_failed:
1438 	case BCH_MEMBER_STATE_spare:
1439 		if (ca->mi.state != BCH_MEMBER_STATE_rw &&
1440 		    ca->mi.state != BCH_MEMBER_STATE_ro)
1441 			return true;
1442 
1443 		/* do we have enough devices to read from?  */
1444 		new_online_devs = bch2_online_devs(c);
1445 		__clear_bit(ca->dev_idx, new_online_devs.d);
1446 
1447 		return bch2_have_enough_devs(c, new_online_devs, flags, false);
1448 	default:
1449 		BUG();
1450 	}
1451 }
1452 
1453 static bool bch2_fs_may_start(struct bch_fs *c)
1454 {
1455 	struct bch_dev *ca;
1456 	unsigned i, flags = 0;
1457 
1458 	if (c->opts.very_degraded)
1459 		flags |= BCH_FORCE_IF_DEGRADED|BCH_FORCE_IF_LOST;
1460 
1461 	if (c->opts.degraded)
1462 		flags |= BCH_FORCE_IF_DEGRADED;
1463 
1464 	if (!c->opts.degraded &&
1465 	    !c->opts.very_degraded) {
1466 		mutex_lock(&c->sb_lock);
1467 
1468 		for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1469 			if (!bch2_dev_exists(c->disk_sb.sb, i))
1470 				continue;
1471 
1472 			ca = bch_dev_locked(c, i);
1473 
1474 			if (!bch2_dev_is_online(ca) &&
1475 			    (ca->mi.state == BCH_MEMBER_STATE_rw ||
1476 			     ca->mi.state == BCH_MEMBER_STATE_ro)) {
1477 				mutex_unlock(&c->sb_lock);
1478 				return false;
1479 			}
1480 		}
1481 		mutex_unlock(&c->sb_lock);
1482 	}
1483 
1484 	return bch2_have_enough_devs(c, bch2_online_devs(c), flags, true);
1485 }
1486 
1487 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1488 {
1489 	/*
1490 	 * The allocator thread itself allocates btree nodes, so stop it first:
1491 	 */
1492 	bch2_dev_allocator_remove(c, ca);
1493 	bch2_dev_journal_stop(&c->journal, ca);
1494 }
1495 
1496 static void __bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1497 {
1498 	lockdep_assert_held(&c->state_lock);
1499 
1500 	BUG_ON(ca->mi.state != BCH_MEMBER_STATE_rw);
1501 
1502 	bch2_dev_allocator_add(c, ca);
1503 	bch2_recalc_capacity(c);
1504 }
1505 
1506 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1507 			 enum bch_member_state new_state, int flags)
1508 {
1509 	struct bch_member *m;
1510 	int ret = 0;
1511 
1512 	if (ca->mi.state == new_state)
1513 		return 0;
1514 
1515 	if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1516 		return -BCH_ERR_device_state_not_allowed;
1517 
1518 	if (new_state != BCH_MEMBER_STATE_rw)
1519 		__bch2_dev_read_only(c, ca);
1520 
1521 	bch_notice(ca, "%s", bch2_member_states[new_state]);
1522 
1523 	mutex_lock(&c->sb_lock);
1524 	m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
1525 	SET_BCH_MEMBER_STATE(m, new_state);
1526 	bch2_write_super(c);
1527 	mutex_unlock(&c->sb_lock);
1528 
1529 	if (new_state == BCH_MEMBER_STATE_rw)
1530 		__bch2_dev_read_write(c, ca);
1531 
1532 	rebalance_wakeup(c);
1533 
1534 	return ret;
1535 }
1536 
1537 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1538 		       enum bch_member_state new_state, int flags)
1539 {
1540 	int ret;
1541 
1542 	down_write(&c->state_lock);
1543 	ret = __bch2_dev_set_state(c, ca, new_state, flags);
1544 	up_write(&c->state_lock);
1545 
1546 	return ret;
1547 }
1548 
1549 /* Device add/removal: */
1550 
1551 static int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
1552 {
1553 	struct bpos start	= POS(ca->dev_idx, 0);
1554 	struct bpos end		= POS(ca->dev_idx, U64_MAX);
1555 	int ret;
1556 
1557 	/*
1558 	 * We clear the LRU and need_discard btrees first so that we don't race
1559 	 * with bch2_do_invalidates() and bch2_do_discards()
1560 	 */
1561 	ret =   bch2_btree_delete_range(c, BTREE_ID_lru, start, end,
1562 					BTREE_TRIGGER_NORUN, NULL) ?:
1563 		bch2_btree_delete_range(c, BTREE_ID_need_discard, start, end,
1564 					BTREE_TRIGGER_NORUN, NULL) ?:
1565 		bch2_btree_delete_range(c, BTREE_ID_freespace, start, end,
1566 					BTREE_TRIGGER_NORUN, NULL) ?:
1567 		bch2_btree_delete_range(c, BTREE_ID_backpointers, start, end,
1568 					BTREE_TRIGGER_NORUN, NULL) ?:
1569 		bch2_btree_delete_range(c, BTREE_ID_alloc, start, end,
1570 					BTREE_TRIGGER_NORUN, NULL) ?:
1571 		bch2_btree_delete_range(c, BTREE_ID_bucket_gens, start, end,
1572 					BTREE_TRIGGER_NORUN, NULL);
1573 	bch_err_msg(c, ret, "removing dev alloc info");
1574 	return ret;
1575 }
1576 
1577 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1578 {
1579 	struct bch_member *m;
1580 	unsigned dev_idx = ca->dev_idx, data;
1581 	int ret;
1582 
1583 	down_write(&c->state_lock);
1584 
1585 	/*
1586 	 * We consume a reference to ca->ref, regardless of whether we succeed
1587 	 * or fail:
1588 	 */
1589 	percpu_ref_put(&ca->ref);
1590 
1591 	if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1592 		bch_err(ca, "Cannot remove without losing data");
1593 		ret = -BCH_ERR_device_state_not_allowed;
1594 		goto err;
1595 	}
1596 
1597 	__bch2_dev_read_only(c, ca);
1598 
1599 	ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1600 	bch_err_msg(ca, ret, "dropping data");
1601 	if (ret)
1602 		goto err;
1603 
1604 	ret = bch2_dev_remove_alloc(c, ca);
1605 	bch_err_msg(ca, ret, "deleting alloc info");
1606 	if (ret)
1607 		goto err;
1608 
1609 	ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1610 	bch_err_msg(ca, ret, "flushing journal");
1611 	if (ret)
1612 		goto err;
1613 
1614 	ret = bch2_journal_flush(&c->journal);
1615 	bch_err(ca, "journal error");
1616 	if (ret)
1617 		goto err;
1618 
1619 	ret = bch2_replicas_gc2(c);
1620 	bch_err_msg(ca, ret, "in replicas_gc2()");
1621 	if (ret)
1622 		goto err;
1623 
1624 	data = bch2_dev_has_data(c, ca);
1625 	if (data) {
1626 		struct printbuf data_has = PRINTBUF;
1627 
1628 		prt_bitflags(&data_has, __bch2_data_types, data);
1629 		bch_err(ca, "Remove failed, still has data (%s)", data_has.buf);
1630 		printbuf_exit(&data_has);
1631 		ret = -EBUSY;
1632 		goto err;
1633 	}
1634 
1635 	__bch2_dev_offline(c, ca);
1636 
1637 	mutex_lock(&c->sb_lock);
1638 	rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1639 	mutex_unlock(&c->sb_lock);
1640 
1641 	percpu_ref_kill(&ca->ref);
1642 	wait_for_completion(&ca->ref_completion);
1643 
1644 	bch2_dev_free(ca);
1645 
1646 	/*
1647 	 * At this point the device object has been removed in-core, but the
1648 	 * on-disk journal might still refer to the device index via sb device
1649 	 * usage entries. Recovery fails if it sees usage information for an
1650 	 * invalid device. Flush journal pins to push the back of the journal
1651 	 * past now invalid device index references before we update the
1652 	 * superblock, but after the device object has been removed so any
1653 	 * further journal writes elide usage info for the device.
1654 	 */
1655 	bch2_journal_flush_all_pins(&c->journal);
1656 
1657 	/*
1658 	 * Free this device's slot in the bch_member array - all pointers to
1659 	 * this device must be gone:
1660 	 */
1661 	mutex_lock(&c->sb_lock);
1662 	m = bch2_members_v2_get_mut(c->disk_sb.sb, dev_idx);
1663 	memset(&m->uuid, 0, sizeof(m->uuid));
1664 
1665 	bch2_write_super(c);
1666 
1667 	mutex_unlock(&c->sb_lock);
1668 	up_write(&c->state_lock);
1669 
1670 	bch2_dev_usage_journal_reserve(c);
1671 	return 0;
1672 err:
1673 	if (ca->mi.state == BCH_MEMBER_STATE_rw &&
1674 	    !percpu_ref_is_zero(&ca->io_ref))
1675 		__bch2_dev_read_write(c, ca);
1676 	up_write(&c->state_lock);
1677 	return ret;
1678 }
1679 
1680 /* Add new device to running filesystem: */
1681 int bch2_dev_add(struct bch_fs *c, const char *path)
1682 {
1683 	struct bch_opts opts = bch2_opts_empty();
1684 	struct bch_sb_handle sb;
1685 	struct bch_dev *ca = NULL;
1686 	struct bch_sb_field_members_v2 *mi;
1687 	struct bch_member dev_mi;
1688 	unsigned dev_idx, nr_devices, u64s;
1689 	struct printbuf errbuf = PRINTBUF;
1690 	struct printbuf label = PRINTBUF;
1691 	int ret;
1692 
1693 	ret = bch2_read_super(path, &opts, &sb);
1694 	bch_err_msg(c, ret, "reading super");
1695 	if (ret)
1696 		goto err;
1697 
1698 	dev_mi = bch2_sb_member_get(sb.sb, sb.sb->dev_idx);
1699 
1700 	if (BCH_MEMBER_GROUP(&dev_mi)) {
1701 		bch2_disk_path_to_text_sb(&label, sb.sb, BCH_MEMBER_GROUP(&dev_mi) - 1);
1702 		if (label.allocation_failure) {
1703 			ret = -ENOMEM;
1704 			goto err;
1705 		}
1706 	}
1707 
1708 	ret = bch2_dev_may_add(sb.sb, c);
1709 	if (ret)
1710 		goto err;
1711 
1712 	ca = __bch2_dev_alloc(c, &dev_mi);
1713 	if (!ca) {
1714 		ret = -ENOMEM;
1715 		goto err;
1716 	}
1717 
1718 	bch2_dev_usage_init(ca);
1719 
1720 	ret = __bch2_dev_attach_bdev(ca, &sb);
1721 	if (ret)
1722 		goto err;
1723 
1724 	ret = bch2_dev_journal_alloc(ca);
1725 	bch_err_msg(c, ret, "allocating journal");
1726 	if (ret)
1727 		goto err;
1728 
1729 	down_write(&c->state_lock);
1730 	mutex_lock(&c->sb_lock);
1731 
1732 	ret = bch2_sb_from_fs(c, ca);
1733 	bch_err_msg(c, ret, "setting up new superblock");
1734 	if (ret)
1735 		goto err_unlock;
1736 
1737 	if (dynamic_fault("bcachefs:add:no_slot"))
1738 		goto no_slot;
1739 
1740 	for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++)
1741 		if (!bch2_dev_exists(c->disk_sb.sb, dev_idx))
1742 			goto have_slot;
1743 no_slot:
1744 	ret = -BCH_ERR_ENOSPC_sb_members;
1745 	bch_err_msg(c, ret, "setting up new superblock");
1746 	goto err_unlock;
1747 
1748 have_slot:
1749 	nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1750 
1751 	mi = bch2_sb_field_get(c->disk_sb.sb, members_v2);
1752 	u64s = DIV_ROUND_UP(sizeof(struct bch_sb_field_members_v2) +
1753 			    le16_to_cpu(mi->member_bytes) * nr_devices, sizeof(u64));
1754 
1755 	mi = bch2_sb_field_resize(&c->disk_sb, members_v2, u64s);
1756 	if (!mi) {
1757 		ret = -BCH_ERR_ENOSPC_sb_members;
1758 		bch_err_msg(c, ret, "setting up new superblock");
1759 		goto err_unlock;
1760 	}
1761 	struct bch_member *m = bch2_members_v2_get_mut(c->disk_sb.sb, dev_idx);
1762 
1763 	/* success: */
1764 
1765 	*m = dev_mi;
1766 	m->last_mount = cpu_to_le64(ktime_get_real_seconds());
1767 	c->disk_sb.sb->nr_devices	= nr_devices;
1768 
1769 	ca->disk_sb.sb->dev_idx	= dev_idx;
1770 	bch2_dev_attach(c, ca, dev_idx);
1771 
1772 	if (BCH_MEMBER_GROUP(&dev_mi)) {
1773 		ret = __bch2_dev_group_set(c, ca, label.buf);
1774 		bch_err_msg(c, ret, "creating new label");
1775 		if (ret)
1776 			goto err_unlock;
1777 	}
1778 
1779 	bch2_write_super(c);
1780 	mutex_unlock(&c->sb_lock);
1781 
1782 	bch2_dev_usage_journal_reserve(c);
1783 
1784 	ret = bch2_trans_mark_dev_sb(c, ca);
1785 	bch_err_msg(ca, ret, "marking new superblock");
1786 	if (ret)
1787 		goto err_late;
1788 
1789 	ret = bch2_fs_freespace_init(c);
1790 	bch_err_msg(ca, ret, "initializing free space");
1791 	if (ret)
1792 		goto err_late;
1793 
1794 	ca->new_fs_bucket_idx = 0;
1795 
1796 	if (ca->mi.state == BCH_MEMBER_STATE_rw)
1797 		__bch2_dev_read_write(c, ca);
1798 
1799 	up_write(&c->state_lock);
1800 	return 0;
1801 
1802 err_unlock:
1803 	mutex_unlock(&c->sb_lock);
1804 	up_write(&c->state_lock);
1805 err:
1806 	if (ca)
1807 		bch2_dev_free(ca);
1808 	bch2_free_super(&sb);
1809 	printbuf_exit(&label);
1810 	printbuf_exit(&errbuf);
1811 	bch_err_fn(c, ret);
1812 	return ret;
1813 err_late:
1814 	up_write(&c->state_lock);
1815 	ca = NULL;
1816 	goto err;
1817 }
1818 
1819 /* Hot add existing device to running filesystem: */
1820 int bch2_dev_online(struct bch_fs *c, const char *path)
1821 {
1822 	struct bch_opts opts = bch2_opts_empty();
1823 	struct bch_sb_handle sb = { NULL };
1824 	struct bch_dev *ca;
1825 	unsigned dev_idx;
1826 	int ret;
1827 
1828 	down_write(&c->state_lock);
1829 
1830 	ret = bch2_read_super(path, &opts, &sb);
1831 	if (ret) {
1832 		up_write(&c->state_lock);
1833 		return ret;
1834 	}
1835 
1836 	dev_idx = sb.sb->dev_idx;
1837 
1838 	ret = bch2_dev_in_fs(&c->disk_sb, &sb);
1839 	bch_err_msg(c, ret, "bringing %s online", path);
1840 	if (ret)
1841 		goto err;
1842 
1843 	ret = bch2_dev_attach_bdev(c, &sb);
1844 	if (ret)
1845 		goto err;
1846 
1847 	ca = bch_dev_locked(c, dev_idx);
1848 
1849 	ret = bch2_trans_mark_dev_sb(c, ca);
1850 	bch_err_msg(c, ret, "bringing %s online: error from bch2_trans_mark_dev_sb", path);
1851 	if (ret)
1852 		goto err;
1853 
1854 	if (ca->mi.state == BCH_MEMBER_STATE_rw)
1855 		__bch2_dev_read_write(c, ca);
1856 
1857 	if (!ca->mi.freespace_initialized) {
1858 		ret = bch2_dev_freespace_init(c, ca, 0, ca->mi.nbuckets);
1859 		bch_err_msg(ca, ret, "initializing free space");
1860 		if (ret)
1861 			goto err;
1862 	}
1863 
1864 	if (!ca->journal.nr) {
1865 		ret = bch2_dev_journal_alloc(ca);
1866 		bch_err_msg(ca, ret, "allocating journal");
1867 		if (ret)
1868 			goto err;
1869 	}
1870 
1871 	mutex_lock(&c->sb_lock);
1872 	bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx)->last_mount =
1873 		cpu_to_le64(ktime_get_real_seconds());
1874 	bch2_write_super(c);
1875 	mutex_unlock(&c->sb_lock);
1876 
1877 	up_write(&c->state_lock);
1878 	return 0;
1879 err:
1880 	up_write(&c->state_lock);
1881 	bch2_free_super(&sb);
1882 	return ret;
1883 }
1884 
1885 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1886 {
1887 	down_write(&c->state_lock);
1888 
1889 	if (!bch2_dev_is_online(ca)) {
1890 		bch_err(ca, "Already offline");
1891 		up_write(&c->state_lock);
1892 		return 0;
1893 	}
1894 
1895 	if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1896 		bch_err(ca, "Cannot offline required disk");
1897 		up_write(&c->state_lock);
1898 		return -BCH_ERR_device_state_not_allowed;
1899 	}
1900 
1901 	__bch2_dev_offline(c, ca);
1902 
1903 	up_write(&c->state_lock);
1904 	return 0;
1905 }
1906 
1907 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1908 {
1909 	struct bch_member *m;
1910 	u64 old_nbuckets;
1911 	int ret = 0;
1912 
1913 	down_write(&c->state_lock);
1914 	old_nbuckets = ca->mi.nbuckets;
1915 
1916 	if (nbuckets < ca->mi.nbuckets) {
1917 		bch_err(ca, "Cannot shrink yet");
1918 		ret = -EINVAL;
1919 		goto err;
1920 	}
1921 
1922 	if (bch2_dev_is_online(ca) &&
1923 	    get_capacity(ca->disk_sb.bdev->bd_disk) <
1924 	    ca->mi.bucket_size * nbuckets) {
1925 		bch_err(ca, "New size larger than device");
1926 		ret = -BCH_ERR_device_size_too_small;
1927 		goto err;
1928 	}
1929 
1930 	ret = bch2_dev_buckets_resize(c, ca, nbuckets);
1931 	bch_err_msg(ca, ret, "resizing buckets");
1932 	if (ret)
1933 		goto err;
1934 
1935 	ret = bch2_trans_mark_dev_sb(c, ca);
1936 	if (ret)
1937 		goto err;
1938 
1939 	mutex_lock(&c->sb_lock);
1940 	m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
1941 	m->nbuckets = cpu_to_le64(nbuckets);
1942 
1943 	bch2_write_super(c);
1944 	mutex_unlock(&c->sb_lock);
1945 
1946 	if (ca->mi.freespace_initialized) {
1947 		ret = bch2_dev_freespace_init(c, ca, old_nbuckets, nbuckets);
1948 		if (ret)
1949 			goto err;
1950 
1951 		/*
1952 		 * XXX: this is all wrong transactionally - we'll be able to do
1953 		 * this correctly after the disk space accounting rewrite
1954 		 */
1955 		ca->usage_base->d[BCH_DATA_free].buckets += nbuckets - old_nbuckets;
1956 	}
1957 
1958 	bch2_recalc_capacity(c);
1959 err:
1960 	up_write(&c->state_lock);
1961 	return ret;
1962 }
1963 
1964 /* return with ref on ca->ref: */
1965 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *name)
1966 {
1967 	rcu_read_lock();
1968 	for_each_member_device_rcu(c, ca, NULL)
1969 		if (!strcmp(name, ca->name)) {
1970 			rcu_read_unlock();
1971 			return ca;
1972 		}
1973 	rcu_read_unlock();
1974 	return ERR_PTR(-BCH_ERR_ENOENT_dev_not_found);
1975 }
1976 
1977 /* Filesystem open: */
1978 
1979 static inline int sb_cmp(struct bch_sb *l, struct bch_sb *r)
1980 {
1981 	return  cmp_int(le64_to_cpu(l->seq), le64_to_cpu(r->seq)) ?:
1982 		cmp_int(le64_to_cpu(l->write_time), le64_to_cpu(r->write_time));
1983 }
1984 
1985 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
1986 			    struct bch_opts opts)
1987 {
1988 	DARRAY(struct bch_sb_handle) sbs = { 0 };
1989 	struct bch_fs *c = NULL;
1990 	struct bch_sb_handle *best = NULL;
1991 	struct printbuf errbuf = PRINTBUF;
1992 	int ret = 0;
1993 
1994 	if (!try_module_get(THIS_MODULE))
1995 		return ERR_PTR(-ENODEV);
1996 
1997 	if (!nr_devices) {
1998 		ret = -EINVAL;
1999 		goto err;
2000 	}
2001 
2002 	ret = darray_make_room(&sbs, nr_devices);
2003 	if (ret)
2004 		goto err;
2005 
2006 	for (unsigned i = 0; i < nr_devices; i++) {
2007 		struct bch_sb_handle sb = { NULL };
2008 
2009 		ret = bch2_read_super(devices[i], &opts, &sb);
2010 		if (ret)
2011 			goto err;
2012 
2013 		BUG_ON(darray_push(&sbs, sb));
2014 	}
2015 
2016 	if (opts.nochanges && !opts.read_only) {
2017 		ret = -BCH_ERR_erofs_nochanges;
2018 		goto err_print;
2019 	}
2020 
2021 	darray_for_each(sbs, sb)
2022 		if (!best || sb_cmp(sb->sb, best->sb) > 0)
2023 			best = sb;
2024 
2025 	darray_for_each_reverse(sbs, sb) {
2026 		ret = bch2_dev_in_fs(best, sb);
2027 
2028 		if (ret == -BCH_ERR_device_has_been_removed ||
2029 		    ret == -BCH_ERR_device_splitbrain) {
2030 			bch2_free_super(sb);
2031 			darray_remove_item(&sbs, sb);
2032 			best -= best > sb;
2033 			ret = 0;
2034 			continue;
2035 		}
2036 
2037 		if (ret)
2038 			goto err_print;
2039 	}
2040 
2041 	c = bch2_fs_alloc(best->sb, opts);
2042 	ret = PTR_ERR_OR_ZERO(c);
2043 	if (ret)
2044 		goto err;
2045 
2046 	down_write(&c->state_lock);
2047 	darray_for_each(sbs, sb) {
2048 		ret = bch2_dev_attach_bdev(c, sb);
2049 		if (ret) {
2050 			up_write(&c->state_lock);
2051 			goto err;
2052 		}
2053 	}
2054 	up_write(&c->state_lock);
2055 
2056 	if (!bch2_fs_may_start(c)) {
2057 		ret = -BCH_ERR_insufficient_devices_to_start;
2058 		goto err_print;
2059 	}
2060 
2061 	if (!c->opts.nostart) {
2062 		ret = bch2_fs_start(c);
2063 		if (ret)
2064 			goto err;
2065 	}
2066 out:
2067 	darray_for_each(sbs, sb)
2068 		bch2_free_super(sb);
2069 	darray_exit(&sbs);
2070 	printbuf_exit(&errbuf);
2071 	module_put(THIS_MODULE);
2072 	return c;
2073 err_print:
2074 	pr_err("bch_fs_open err opening %s: %s",
2075 	       devices[0], bch2_err_str(ret));
2076 err:
2077 	if (!IS_ERR_OR_NULL(c))
2078 		bch2_fs_stop(c);
2079 	c = ERR_PTR(ret);
2080 	goto out;
2081 }
2082 
2083 /* Global interfaces/init */
2084 
2085 static void bcachefs_exit(void)
2086 {
2087 	bch2_debug_exit();
2088 	bch2_vfs_exit();
2089 	bch2_chardev_exit();
2090 	bch2_btree_key_cache_exit();
2091 	if (bcachefs_kset)
2092 		kset_unregister(bcachefs_kset);
2093 }
2094 
2095 static int __init bcachefs_init(void)
2096 {
2097 	bch2_bkey_pack_test();
2098 
2099 	if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
2100 	    bch2_btree_key_cache_init() ||
2101 	    bch2_chardev_init() ||
2102 	    bch2_vfs_init() ||
2103 	    bch2_debug_init())
2104 		goto err;
2105 
2106 	return 0;
2107 err:
2108 	bcachefs_exit();
2109 	return -ENOMEM;
2110 }
2111 
2112 #define BCH_DEBUG_PARAM(name, description)			\
2113 	bool bch2_##name;					\
2114 	module_param_named(name, bch2_##name, bool, 0644);	\
2115 	MODULE_PARM_DESC(name, description);
2116 BCH_DEBUG_PARAMS()
2117 #undef BCH_DEBUG_PARAM
2118 
2119 __maybe_unused
2120 static unsigned bch2_metadata_version = bcachefs_metadata_version_current;
2121 module_param_named(version, bch2_metadata_version, uint, 0400);
2122 
2123 module_exit(bcachefs_exit);
2124 module_init(bcachefs_init);
2125