xref: /linux/fs/bcachefs/super.c (revision 78c3925c048c752334873f56c3a3d1c9d53e0416)
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 "thread_with_file.h"
60 #include "trace.h"
61 
62 #include <linux/backing-dev.h>
63 #include <linux/blkdev.h>
64 #include <linux/debugfs.h>
65 #include <linux/device.h>
66 #include <linux/idr.h>
67 #include <linux/module.h>
68 #include <linux/percpu.h>
69 #include <linux/random.h>
70 #include <linux/sysfs.h>
71 #include <crypto/hash.h>
72 
73 MODULE_LICENSE("GPL");
74 MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
75 MODULE_DESCRIPTION("bcachefs filesystem");
76 MODULE_SOFTDEP("pre: crc32c");
77 MODULE_SOFTDEP("pre: crc64");
78 MODULE_SOFTDEP("pre: sha256");
79 MODULE_SOFTDEP("pre: chacha20");
80 MODULE_SOFTDEP("pre: poly1305");
81 MODULE_SOFTDEP("pre: xxhash");
82 
83 const char * const bch2_fs_flag_strs[] = {
84 #define x(n)		#n,
85 	BCH_FS_FLAGS()
86 #undef x
87 	NULL
88 };
89 
90 void bch2_print_opts(struct bch_opts *opts, const char *fmt, ...)
91 {
92 	struct stdio_redirect *stdio = (void *)(unsigned long)opts->stdio;
93 
94 	va_list args;
95 	va_start(args, fmt);
96 	if (likely(!stdio)) {
97 		vprintk(fmt, args);
98 	} else {
99 		if (fmt[0] == KERN_SOH[0])
100 			fmt += 2;
101 
102 		bch2_stdio_redirect_vprintf(stdio, true, fmt, args);
103 	}
104 	va_end(args);
105 }
106 
107 void __bch2_print(struct bch_fs *c, const char *fmt, ...)
108 {
109 	struct stdio_redirect *stdio = bch2_fs_stdio_redirect(c);
110 
111 	va_list args;
112 	va_start(args, fmt);
113 	if (likely(!stdio)) {
114 		vprintk(fmt, args);
115 	} else {
116 		if (fmt[0] == KERN_SOH[0])
117 			fmt += 2;
118 
119 		bch2_stdio_redirect_vprintf(stdio, true, fmt, args);
120 	}
121 	va_end(args);
122 }
123 
124 #define KTYPE(type)							\
125 static const struct attribute_group type ## _group = {			\
126 	.attrs = type ## _files						\
127 };									\
128 									\
129 static const struct attribute_group *type ## _groups[] = {		\
130 	&type ## _group,						\
131 	NULL								\
132 };									\
133 									\
134 static const struct kobj_type type ## _ktype = {			\
135 	.release	= type ## _release,				\
136 	.sysfs_ops	= &type ## _sysfs_ops,				\
137 	.default_groups = type ## _groups				\
138 }
139 
140 static void bch2_fs_release(struct kobject *);
141 static void bch2_dev_release(struct kobject *);
142 static void bch2_fs_counters_release(struct kobject *k)
143 {
144 }
145 
146 static void bch2_fs_internal_release(struct kobject *k)
147 {
148 }
149 
150 static void bch2_fs_opts_dir_release(struct kobject *k)
151 {
152 }
153 
154 static void bch2_fs_time_stats_release(struct kobject *k)
155 {
156 }
157 
158 KTYPE(bch2_fs);
159 KTYPE(bch2_fs_counters);
160 KTYPE(bch2_fs_internal);
161 KTYPE(bch2_fs_opts_dir);
162 KTYPE(bch2_fs_time_stats);
163 KTYPE(bch2_dev);
164 
165 static struct kset *bcachefs_kset;
166 static LIST_HEAD(bch_fs_list);
167 static DEFINE_MUTEX(bch_fs_list_lock);
168 
169 DECLARE_WAIT_QUEUE_HEAD(bch2_read_only_wait);
170 
171 static void bch2_dev_free(struct bch_dev *);
172 static int bch2_dev_alloc(struct bch_fs *, unsigned);
173 static int bch2_dev_sysfs_online(struct bch_fs *, struct bch_dev *);
174 static void __bch2_dev_read_only(struct bch_fs *, struct bch_dev *);
175 
176 struct bch_fs *bch2_dev_to_fs(dev_t dev)
177 {
178 	struct bch_fs *c;
179 
180 	mutex_lock(&bch_fs_list_lock);
181 	rcu_read_lock();
182 
183 	list_for_each_entry(c, &bch_fs_list, list)
184 		for_each_member_device_rcu(c, ca, NULL)
185 			if (ca->disk_sb.bdev && ca->disk_sb.bdev->bd_dev == dev) {
186 				closure_get(&c->cl);
187 				goto found;
188 			}
189 	c = NULL;
190 found:
191 	rcu_read_unlock();
192 	mutex_unlock(&bch_fs_list_lock);
193 
194 	return c;
195 }
196 
197 static struct bch_fs *__bch2_uuid_to_fs(__uuid_t uuid)
198 {
199 	struct bch_fs *c;
200 
201 	lockdep_assert_held(&bch_fs_list_lock);
202 
203 	list_for_each_entry(c, &bch_fs_list, list)
204 		if (!memcmp(&c->disk_sb.sb->uuid, &uuid, sizeof(uuid)))
205 			return c;
206 
207 	return NULL;
208 }
209 
210 struct bch_fs *bch2_uuid_to_fs(__uuid_t uuid)
211 {
212 	struct bch_fs *c;
213 
214 	mutex_lock(&bch_fs_list_lock);
215 	c = __bch2_uuid_to_fs(uuid);
216 	if (c)
217 		closure_get(&c->cl);
218 	mutex_unlock(&bch_fs_list_lock);
219 
220 	return c;
221 }
222 
223 static void bch2_dev_usage_journal_reserve(struct bch_fs *c)
224 {
225 	unsigned nr = 0, u64s =
226 		((sizeof(struct jset_entry_dev_usage) +
227 		  sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR)) /
228 		sizeof(u64);
229 
230 	rcu_read_lock();
231 	for_each_member_device_rcu(c, ca, NULL)
232 		nr++;
233 	rcu_read_unlock();
234 
235 	bch2_journal_entry_res_resize(&c->journal,
236 			&c->dev_usage_journal_res, u64s * nr);
237 }
238 
239 /* Filesystem RO/RW: */
240 
241 /*
242  * For startup/shutdown of RW stuff, the dependencies are:
243  *
244  * - foreground writes depend on copygc and rebalance (to free up space)
245  *
246  * - copygc and rebalance depend on mark and sweep gc (they actually probably
247  *   don't because they either reserve ahead of time or don't block if
248  *   allocations fail, but allocations can require mark and sweep gc to run
249  *   because of generation number wraparound)
250  *
251  * - all of the above depends on the allocator threads
252  *
253  * - allocator depends on the journal (when it rewrites prios and gens)
254  */
255 
256 static void __bch2_fs_read_only(struct bch_fs *c)
257 {
258 	unsigned clean_passes = 0;
259 	u64 seq = 0;
260 
261 	bch2_fs_ec_stop(c);
262 	bch2_open_buckets_stop(c, NULL, true);
263 	bch2_rebalance_stop(c);
264 	bch2_copygc_stop(c);
265 	bch2_gc_thread_stop(c);
266 	bch2_fs_ec_flush(c);
267 
268 	bch_verbose(c, "flushing journal and stopping allocators, journal seq %llu",
269 		    journal_cur_seq(&c->journal));
270 
271 	do {
272 		clean_passes++;
273 
274 		if (bch2_btree_interior_updates_flush(c) ||
275 		    bch2_journal_flush_all_pins(&c->journal) ||
276 		    bch2_btree_flush_all_writes(c) ||
277 		    seq != atomic64_read(&c->journal.seq)) {
278 			seq = atomic64_read(&c->journal.seq);
279 			clean_passes = 0;
280 		}
281 	} while (clean_passes < 2);
282 
283 	bch_verbose(c, "flushing journal and stopping allocators complete, journal seq %llu",
284 		    journal_cur_seq(&c->journal));
285 
286 	if (test_bit(JOURNAL_REPLAY_DONE, &c->journal.flags) &&
287 	    !test_bit(BCH_FS_emergency_ro, &c->flags))
288 		set_bit(BCH_FS_clean_shutdown, &c->flags);
289 	bch2_fs_journal_stop(&c->journal);
290 
291 	/*
292 	 * After stopping journal:
293 	 */
294 	for_each_member_device(c, ca)
295 		bch2_dev_allocator_remove(c, ca);
296 }
297 
298 #ifndef BCH_WRITE_REF_DEBUG
299 static void bch2_writes_disabled(struct percpu_ref *writes)
300 {
301 	struct bch_fs *c = container_of(writes, struct bch_fs, writes);
302 
303 	set_bit(BCH_FS_write_disable_complete, &c->flags);
304 	wake_up(&bch2_read_only_wait);
305 }
306 #endif
307 
308 void bch2_fs_read_only(struct bch_fs *c)
309 {
310 	if (!test_bit(BCH_FS_rw, &c->flags)) {
311 		bch2_journal_reclaim_stop(&c->journal);
312 		return;
313 	}
314 
315 	BUG_ON(test_bit(BCH_FS_write_disable_complete, &c->flags));
316 
317 	bch_verbose(c, "going read-only");
318 
319 	/*
320 	 * Block new foreground-end write operations from starting - any new
321 	 * writes will return -EROFS:
322 	 */
323 	set_bit(BCH_FS_going_ro, &c->flags);
324 #ifndef BCH_WRITE_REF_DEBUG
325 	percpu_ref_kill(&c->writes);
326 #else
327 	for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++)
328 		bch2_write_ref_put(c, i);
329 #endif
330 
331 	/*
332 	 * If we're not doing an emergency shutdown, we want to wait on
333 	 * outstanding writes to complete so they don't see spurious errors due
334 	 * to shutting down the allocator:
335 	 *
336 	 * If we are doing an emergency shutdown outstanding writes may
337 	 * hang until we shutdown the allocator so we don't want to wait
338 	 * on outstanding writes before shutting everything down - but
339 	 * we do need to wait on them before returning and signalling
340 	 * that going RO is complete:
341 	 */
342 	wait_event(bch2_read_only_wait,
343 		   test_bit(BCH_FS_write_disable_complete, &c->flags) ||
344 		   test_bit(BCH_FS_emergency_ro, &c->flags));
345 
346 	bool writes_disabled = test_bit(BCH_FS_write_disable_complete, &c->flags);
347 	if (writes_disabled)
348 		bch_verbose(c, "finished waiting for writes to stop");
349 
350 	__bch2_fs_read_only(c);
351 
352 	wait_event(bch2_read_only_wait,
353 		   test_bit(BCH_FS_write_disable_complete, &c->flags));
354 
355 	if (!writes_disabled)
356 		bch_verbose(c, "finished waiting for writes to stop");
357 
358 	clear_bit(BCH_FS_write_disable_complete, &c->flags);
359 	clear_bit(BCH_FS_going_ro, &c->flags);
360 	clear_bit(BCH_FS_rw, &c->flags);
361 
362 	if (!bch2_journal_error(&c->journal) &&
363 	    !test_bit(BCH_FS_error, &c->flags) &&
364 	    !test_bit(BCH_FS_emergency_ro, &c->flags) &&
365 	    test_bit(BCH_FS_started, &c->flags) &&
366 	    test_bit(BCH_FS_clean_shutdown, &c->flags) &&
367 	    !c->opts.norecovery) {
368 		BUG_ON(c->journal.last_empty_seq != journal_cur_seq(&c->journal));
369 		BUG_ON(atomic_read(&c->btree_cache.dirty));
370 		BUG_ON(atomic_long_read(&c->btree_key_cache.nr_dirty));
371 		BUG_ON(c->btree_write_buffer.inc.keys.nr);
372 		BUG_ON(c->btree_write_buffer.flushing.keys.nr);
373 
374 		bch_verbose(c, "marking filesystem clean");
375 		bch2_fs_mark_clean(c);
376 	} else {
377 		bch_verbose(c, "done going read-only, filesystem not clean");
378 	}
379 }
380 
381 static void bch2_fs_read_only_work(struct work_struct *work)
382 {
383 	struct bch_fs *c =
384 		container_of(work, struct bch_fs, read_only_work);
385 
386 	down_write(&c->state_lock);
387 	bch2_fs_read_only(c);
388 	up_write(&c->state_lock);
389 }
390 
391 static void bch2_fs_read_only_async(struct bch_fs *c)
392 {
393 	queue_work(system_long_wq, &c->read_only_work);
394 }
395 
396 bool bch2_fs_emergency_read_only(struct bch_fs *c)
397 {
398 	bool ret = !test_and_set_bit(BCH_FS_emergency_ro, &c->flags);
399 
400 	bch2_journal_halt(&c->journal);
401 	bch2_fs_read_only_async(c);
402 
403 	wake_up(&bch2_read_only_wait);
404 	return ret;
405 }
406 
407 static int bch2_fs_read_write_late(struct bch_fs *c)
408 {
409 	int ret;
410 
411 	/*
412 	 * Data move operations can't run until after check_snapshots has
413 	 * completed, and bch2_snapshot_is_ancestor() is available.
414 	 *
415 	 * Ideally we'd start copygc/rebalance earlier instead of waiting for
416 	 * all of recovery/fsck to complete:
417 	 */
418 	ret = bch2_copygc_start(c);
419 	if (ret) {
420 		bch_err(c, "error starting copygc thread");
421 		return ret;
422 	}
423 
424 	ret = bch2_rebalance_start(c);
425 	if (ret) {
426 		bch_err(c, "error starting rebalance thread");
427 		return ret;
428 	}
429 
430 	return 0;
431 }
432 
433 static int __bch2_fs_read_write(struct bch_fs *c, bool early)
434 {
435 	int ret;
436 
437 	if (test_bit(BCH_FS_initial_gc_unfixed, &c->flags)) {
438 		bch_err(c, "cannot go rw, unfixed btree errors");
439 		return -BCH_ERR_erofs_unfixed_errors;
440 	}
441 
442 	if (test_bit(BCH_FS_rw, &c->flags))
443 		return 0;
444 
445 	bch_info(c, "going read-write");
446 
447 	ret = bch2_sb_members_v2_init(c);
448 	if (ret)
449 		goto err;
450 
451 	ret = bch2_fs_mark_dirty(c);
452 	if (ret)
453 		goto err;
454 
455 	clear_bit(BCH_FS_clean_shutdown, &c->flags);
456 
457 	/*
458 	 * First journal write must be a flush write: after a clean shutdown we
459 	 * don't read the journal, so the first journal write may end up
460 	 * overwriting whatever was there previously, and there must always be
461 	 * at least one non-flush write in the journal or recovery will fail:
462 	 */
463 	set_bit(JOURNAL_NEED_FLUSH_WRITE, &c->journal.flags);
464 
465 	for_each_rw_member(c, ca)
466 		bch2_dev_allocator_add(c, ca);
467 	bch2_recalc_capacity(c);
468 
469 	set_bit(BCH_FS_rw, &c->flags);
470 	set_bit(BCH_FS_was_rw, &c->flags);
471 
472 #ifndef BCH_WRITE_REF_DEBUG
473 	percpu_ref_reinit(&c->writes);
474 #else
475 	for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++) {
476 		BUG_ON(atomic_long_read(&c->writes[i]));
477 		atomic_long_inc(&c->writes[i]);
478 	}
479 #endif
480 
481 	ret = bch2_gc_thread_start(c);
482 	if (ret) {
483 		bch_err(c, "error starting gc thread");
484 		return ret;
485 	}
486 
487 	ret = bch2_journal_reclaim_start(&c->journal);
488 	if (ret)
489 		goto err;
490 
491 	if (!early) {
492 		ret = bch2_fs_read_write_late(c);
493 		if (ret)
494 			goto err;
495 	}
496 
497 	bch2_do_discards(c);
498 	bch2_do_invalidates(c);
499 	bch2_do_stripe_deletes(c);
500 	bch2_do_pending_node_rewrites(c);
501 	return 0;
502 err:
503 	if (test_bit(BCH_FS_rw, &c->flags))
504 		bch2_fs_read_only(c);
505 	else
506 		__bch2_fs_read_only(c);
507 	return ret;
508 }
509 
510 int bch2_fs_read_write(struct bch_fs *c)
511 {
512 	if (c->opts.norecovery)
513 		return -BCH_ERR_erofs_norecovery;
514 
515 	if (c->opts.nochanges)
516 		return -BCH_ERR_erofs_nochanges;
517 
518 	return __bch2_fs_read_write(c, false);
519 }
520 
521 int bch2_fs_read_write_early(struct bch_fs *c)
522 {
523 	lockdep_assert_held(&c->state_lock);
524 
525 	return __bch2_fs_read_write(c, true);
526 }
527 
528 /* Filesystem startup/shutdown: */
529 
530 static void __bch2_fs_free(struct bch_fs *c)
531 {
532 	unsigned i;
533 
534 	for (i = 0; i < BCH_TIME_STAT_NR; i++)
535 		bch2_time_stats_exit(&c->times[i]);
536 
537 	bch2_free_pending_node_rewrites(c);
538 	bch2_fs_sb_errors_exit(c);
539 	bch2_fs_counters_exit(c);
540 	bch2_fs_snapshots_exit(c);
541 	bch2_fs_quota_exit(c);
542 	bch2_fs_fs_io_direct_exit(c);
543 	bch2_fs_fs_io_buffered_exit(c);
544 	bch2_fs_fsio_exit(c);
545 	bch2_fs_ec_exit(c);
546 	bch2_fs_encryption_exit(c);
547 	bch2_fs_nocow_locking_exit(c);
548 	bch2_fs_io_write_exit(c);
549 	bch2_fs_io_read_exit(c);
550 	bch2_fs_buckets_waiting_for_journal_exit(c);
551 	bch2_fs_btree_interior_update_exit(c);
552 	bch2_fs_btree_iter_exit(c);
553 	bch2_fs_btree_key_cache_exit(&c->btree_key_cache);
554 	bch2_fs_btree_cache_exit(c);
555 	bch2_fs_replicas_exit(c);
556 	bch2_fs_journal_exit(&c->journal);
557 	bch2_io_clock_exit(&c->io_clock[WRITE]);
558 	bch2_io_clock_exit(&c->io_clock[READ]);
559 	bch2_fs_compress_exit(c);
560 	bch2_journal_keys_put_initial(c);
561 	BUG_ON(atomic_read(&c->journal_keys.ref));
562 	bch2_fs_btree_write_buffer_exit(c);
563 	percpu_free_rwsem(&c->mark_lock);
564 	free_percpu(c->online_reserved);
565 
566 	darray_exit(&c->btree_roots_extra);
567 	free_percpu(c->pcpu);
568 	mempool_exit(&c->large_bkey_pool);
569 	mempool_exit(&c->btree_bounce_pool);
570 	bioset_exit(&c->btree_bio);
571 	mempool_exit(&c->fill_iter);
572 #ifndef BCH_WRITE_REF_DEBUG
573 	percpu_ref_exit(&c->writes);
574 #endif
575 	kfree(rcu_dereference_protected(c->disk_groups, 1));
576 	kfree(c->journal_seq_blacklist_table);
577 	kfree(c->unused_inode_hints);
578 
579 	if (c->write_ref_wq)
580 		destroy_workqueue(c->write_ref_wq);
581 	if (c->io_complete_wq)
582 		destroy_workqueue(c->io_complete_wq);
583 	if (c->copygc_wq)
584 		destroy_workqueue(c->copygc_wq);
585 	if (c->btree_io_complete_wq)
586 		destroy_workqueue(c->btree_io_complete_wq);
587 	if (c->btree_update_wq)
588 		destroy_workqueue(c->btree_update_wq);
589 
590 	bch2_free_super(&c->disk_sb);
591 	kvfree(c);
592 	module_put(THIS_MODULE);
593 }
594 
595 static void bch2_fs_release(struct kobject *kobj)
596 {
597 	struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
598 
599 	__bch2_fs_free(c);
600 }
601 
602 void __bch2_fs_stop(struct bch_fs *c)
603 {
604 	bch_verbose(c, "shutting down");
605 
606 	set_bit(BCH_FS_stopping, &c->flags);
607 
608 	cancel_work_sync(&c->journal_seq_blacklist_gc_work);
609 
610 	down_write(&c->state_lock);
611 	bch2_fs_read_only(c);
612 	up_write(&c->state_lock);
613 
614 	for_each_member_device(c, ca)
615 		if (ca->kobj.state_in_sysfs &&
616 		    ca->disk_sb.bdev)
617 			sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
618 
619 	if (c->kobj.state_in_sysfs)
620 		kobject_del(&c->kobj);
621 
622 	bch2_fs_debug_exit(c);
623 	bch2_fs_chardev_exit(c);
624 
625 	bch2_ro_ref_put(c);
626 	wait_event(c->ro_ref_wait, !refcount_read(&c->ro_ref));
627 
628 	kobject_put(&c->counters_kobj);
629 	kobject_put(&c->time_stats);
630 	kobject_put(&c->opts_dir);
631 	kobject_put(&c->internal);
632 
633 	/* btree prefetch might have kicked off reads in the background: */
634 	bch2_btree_flush_all_reads(c);
635 
636 	for_each_member_device(c, ca)
637 		cancel_work_sync(&ca->io_error_work);
638 
639 	cancel_work_sync(&c->read_only_work);
640 }
641 
642 void bch2_fs_free(struct bch_fs *c)
643 {
644 	unsigned i;
645 
646 	mutex_lock(&bch_fs_list_lock);
647 	list_del(&c->list);
648 	mutex_unlock(&bch_fs_list_lock);
649 
650 	closure_sync(&c->cl);
651 	closure_debug_destroy(&c->cl);
652 
653 	for (i = 0; i < c->sb.nr_devices; i++) {
654 		struct bch_dev *ca = rcu_dereference_protected(c->devs[i], true);
655 
656 		if (ca) {
657 			bch2_free_super(&ca->disk_sb);
658 			bch2_dev_free(ca);
659 		}
660 	}
661 
662 	bch_verbose(c, "shutdown complete");
663 
664 	kobject_put(&c->kobj);
665 }
666 
667 void bch2_fs_stop(struct bch_fs *c)
668 {
669 	__bch2_fs_stop(c);
670 	bch2_fs_free(c);
671 }
672 
673 static int bch2_fs_online(struct bch_fs *c)
674 {
675 	int ret = 0;
676 
677 	lockdep_assert_held(&bch_fs_list_lock);
678 
679 	if (__bch2_uuid_to_fs(c->sb.uuid)) {
680 		bch_err(c, "filesystem UUID already open");
681 		return -EINVAL;
682 	}
683 
684 	ret = bch2_fs_chardev_init(c);
685 	if (ret) {
686 		bch_err(c, "error creating character device");
687 		return ret;
688 	}
689 
690 	bch2_fs_debug_init(c);
691 
692 	ret = kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ?:
693 	    kobject_add(&c->internal, &c->kobj, "internal") ?:
694 	    kobject_add(&c->opts_dir, &c->kobj, "options") ?:
695 #ifndef CONFIG_BCACHEFS_NO_LATENCY_ACCT
696 	    kobject_add(&c->time_stats, &c->kobj, "time_stats") ?:
697 #endif
698 	    kobject_add(&c->counters_kobj, &c->kobj, "counters") ?:
699 	    bch2_opts_create_sysfs_files(&c->opts_dir);
700 	if (ret) {
701 		bch_err(c, "error creating sysfs objects");
702 		return ret;
703 	}
704 
705 	down_write(&c->state_lock);
706 
707 	for_each_member_device(c, ca) {
708 		ret = bch2_dev_sysfs_online(c, ca);
709 		if (ret) {
710 			bch_err(c, "error creating sysfs objects");
711 			percpu_ref_put(&ca->ref);
712 			goto err;
713 		}
714 	}
715 
716 	BUG_ON(!list_empty(&c->list));
717 	list_add(&c->list, &bch_fs_list);
718 err:
719 	up_write(&c->state_lock);
720 	return ret;
721 }
722 
723 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
724 {
725 	struct bch_fs *c;
726 	struct printbuf name = PRINTBUF;
727 	unsigned i, iter_size;
728 	int ret = 0;
729 
730 	c = kvmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
731 	if (!c) {
732 		c = ERR_PTR(-BCH_ERR_ENOMEM_fs_alloc);
733 		goto out;
734 	}
735 
736 	c->stdio = (void *)(unsigned long) opts.stdio;
737 
738 	__module_get(THIS_MODULE);
739 
740 	closure_init(&c->cl, NULL);
741 
742 	c->kobj.kset = bcachefs_kset;
743 	kobject_init(&c->kobj, &bch2_fs_ktype);
744 	kobject_init(&c->internal, &bch2_fs_internal_ktype);
745 	kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
746 	kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
747 	kobject_init(&c->counters_kobj, &bch2_fs_counters_ktype);
748 
749 	c->minor		= -1;
750 	c->disk_sb.fs_sb	= true;
751 
752 	init_rwsem(&c->state_lock);
753 	mutex_init(&c->sb_lock);
754 	mutex_init(&c->replicas_gc_lock);
755 	mutex_init(&c->btree_root_lock);
756 	INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
757 
758 	refcount_set(&c->ro_ref, 1);
759 	init_waitqueue_head(&c->ro_ref_wait);
760 	sema_init(&c->online_fsck_mutex, 1);
761 
762 	init_rwsem(&c->gc_lock);
763 	mutex_init(&c->gc_gens_lock);
764 	atomic_set(&c->journal_keys.ref, 1);
765 	c->journal_keys.initial_ref_held = true;
766 
767 	for (i = 0; i < BCH_TIME_STAT_NR; i++)
768 		bch2_time_stats_init(&c->times[i]);
769 
770 	bch2_fs_copygc_init(c);
771 	bch2_fs_btree_key_cache_init_early(&c->btree_key_cache);
772 	bch2_fs_btree_iter_init_early(c);
773 	bch2_fs_btree_interior_update_init_early(c);
774 	bch2_fs_allocator_background_init(c);
775 	bch2_fs_allocator_foreground_init(c);
776 	bch2_fs_rebalance_init(c);
777 	bch2_fs_quota_init(c);
778 	bch2_fs_ec_init_early(c);
779 	bch2_fs_move_init(c);
780 	bch2_fs_sb_errors_init_early(c);
781 
782 	INIT_LIST_HEAD(&c->list);
783 
784 	mutex_init(&c->usage_scratch_lock);
785 
786 	mutex_init(&c->bio_bounce_pages_lock);
787 	mutex_init(&c->snapshot_table_lock);
788 	init_rwsem(&c->snapshot_create_lock);
789 
790 	spin_lock_init(&c->btree_write_error_lock);
791 
792 	INIT_WORK(&c->journal_seq_blacklist_gc_work,
793 		  bch2_blacklist_entries_gc);
794 
795 	INIT_LIST_HEAD(&c->journal_iters);
796 
797 	INIT_LIST_HEAD(&c->fsck_error_msgs);
798 	mutex_init(&c->fsck_error_msgs_lock);
799 
800 	seqcount_init(&c->gc_pos_lock);
801 
802 	seqcount_init(&c->usage_lock);
803 
804 	sema_init(&c->io_in_flight, 128);
805 
806 	INIT_LIST_HEAD(&c->vfs_inodes_list);
807 	mutex_init(&c->vfs_inodes_lock);
808 
809 	c->copy_gc_enabled		= 1;
810 	c->rebalance.enabled		= 1;
811 	c->promote_whole_extents	= true;
812 
813 	c->journal.flush_write_time	= &c->times[BCH_TIME_journal_flush_write];
814 	c->journal.noflush_write_time	= &c->times[BCH_TIME_journal_noflush_write];
815 	c->journal.flush_seq_time	= &c->times[BCH_TIME_journal_flush_seq];
816 
817 	bch2_fs_btree_cache_init_early(&c->btree_cache);
818 
819 	mutex_init(&c->sectors_available_lock);
820 
821 	ret = percpu_init_rwsem(&c->mark_lock);
822 	if (ret)
823 		goto err;
824 
825 	mutex_lock(&c->sb_lock);
826 	ret = bch2_sb_to_fs(c, sb);
827 	mutex_unlock(&c->sb_lock);
828 
829 	if (ret)
830 		goto err;
831 
832 	pr_uuid(&name, c->sb.user_uuid.b);
833 	ret = name.allocation_failure ? -BCH_ERR_ENOMEM_fs_name_alloc : 0;
834 	if (ret)
835 		goto err;
836 
837 	strscpy(c->name, name.buf, sizeof(c->name));
838 	printbuf_exit(&name);
839 
840 	/* Compat: */
841 	if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_inode_v2 &&
842 	    !BCH_SB_JOURNAL_FLUSH_DELAY(sb))
843 		SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
844 
845 	if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_inode_v2 &&
846 	    !BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
847 		SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 100);
848 
849 	c->opts = bch2_opts_default;
850 	ret = bch2_opts_from_sb(&c->opts, sb);
851 	if (ret)
852 		goto err;
853 
854 	bch2_opts_apply(&c->opts, opts);
855 
856 	c->btree_key_cache_btrees |= 1U << BTREE_ID_alloc;
857 	if (c->opts.inodes_use_key_cache)
858 		c->btree_key_cache_btrees |= 1U << BTREE_ID_inodes;
859 	c->btree_key_cache_btrees |= 1U << BTREE_ID_logged_ops;
860 
861 	c->block_bits		= ilog2(block_sectors(c));
862 	c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
863 
864 	if (bch2_fs_init_fault("fs_alloc")) {
865 		bch_err(c, "fs_alloc fault injected");
866 		ret = -EFAULT;
867 		goto err;
868 	}
869 
870 	iter_size = sizeof(struct sort_iter) +
871 		(btree_blocks(c) + 1) * 2 *
872 		sizeof(struct sort_iter_set);
873 
874 	c->inode_shard_bits = ilog2(roundup_pow_of_two(num_possible_cpus()));
875 
876 	if (!(c->btree_update_wq = alloc_workqueue("bcachefs",
877 				WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_UNBOUND, 512)) ||
878 	    !(c->btree_io_complete_wq = alloc_workqueue("bcachefs_btree_io",
879 				WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM, 1)) ||
880 	    !(c->copygc_wq = alloc_workqueue("bcachefs_copygc",
881 				WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
882 	    !(c->io_complete_wq = alloc_workqueue("bcachefs_io",
883 				WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM, 512)) ||
884 	    !(c->write_ref_wq = alloc_workqueue("bcachefs_write_ref",
885 				WQ_FREEZABLE, 0)) ||
886 #ifndef BCH_WRITE_REF_DEBUG
887 	    percpu_ref_init(&c->writes, bch2_writes_disabled,
888 			    PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
889 #endif
890 	    mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
891 	    bioset_init(&c->btree_bio, 1,
892 			max(offsetof(struct btree_read_bio, bio),
893 			    offsetof(struct btree_write_bio, wbio.bio)),
894 			BIOSET_NEED_BVECS) ||
895 	    !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
896 	    !(c->online_reserved = alloc_percpu(u64)) ||
897 	    mempool_init_kvmalloc_pool(&c->btree_bounce_pool, 1,
898 				       c->opts.btree_node_size) ||
899 	    mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048) ||
900 	    !(c->unused_inode_hints = kcalloc(1U << c->inode_shard_bits,
901 					      sizeof(u64), GFP_KERNEL))) {
902 		ret = -BCH_ERR_ENOMEM_fs_other_alloc;
903 		goto err;
904 	}
905 
906 	ret = bch2_fs_counters_init(c) ?:
907 	    bch2_fs_sb_errors_init(c) ?:
908 	    bch2_io_clock_init(&c->io_clock[READ]) ?:
909 	    bch2_io_clock_init(&c->io_clock[WRITE]) ?:
910 	    bch2_fs_journal_init(&c->journal) ?:
911 	    bch2_fs_replicas_init(c) ?:
912 	    bch2_fs_btree_cache_init(c) ?:
913 	    bch2_fs_btree_key_cache_init(&c->btree_key_cache) ?:
914 	    bch2_fs_btree_iter_init(c) ?:
915 	    bch2_fs_btree_interior_update_init(c) ?:
916 	    bch2_fs_buckets_waiting_for_journal_init(c) ?:
917 	    bch2_fs_btree_write_buffer_init(c) ?:
918 	    bch2_fs_subvolumes_init(c) ?:
919 	    bch2_fs_io_read_init(c) ?:
920 	    bch2_fs_io_write_init(c) ?:
921 	    bch2_fs_nocow_locking_init(c) ?:
922 	    bch2_fs_encryption_init(c) ?:
923 	    bch2_fs_compress_init(c) ?:
924 	    bch2_fs_ec_init(c) ?:
925 	    bch2_fs_fsio_init(c) ?:
926 	    bch2_fs_fs_io_buffered_init(c) ?:
927 	    bch2_fs_fs_io_direct_init(c);
928 	if (ret)
929 		goto err;
930 
931 	for (i = 0; i < c->sb.nr_devices; i++)
932 		if (bch2_dev_exists(c->disk_sb.sb, i) &&
933 		    bch2_dev_alloc(c, i)) {
934 			ret = -EEXIST;
935 			goto err;
936 		}
937 
938 	bch2_journal_entry_res_resize(&c->journal,
939 			&c->btree_root_journal_res,
940 			BTREE_ID_NR * (JSET_KEYS_U64s + BKEY_BTREE_PTR_U64s_MAX));
941 	bch2_dev_usage_journal_reserve(c);
942 	bch2_journal_entry_res_resize(&c->journal,
943 			&c->clock_journal_res,
944 			(sizeof(struct jset_entry_clock) / sizeof(u64)) * 2);
945 
946 	mutex_lock(&bch_fs_list_lock);
947 	ret = bch2_fs_online(c);
948 	mutex_unlock(&bch_fs_list_lock);
949 
950 	if (ret)
951 		goto err;
952 out:
953 	return c;
954 err:
955 	bch2_fs_free(c);
956 	c = ERR_PTR(ret);
957 	goto out;
958 }
959 
960 noinline_for_stack
961 static void print_mount_opts(struct bch_fs *c)
962 {
963 	enum bch_opt_id i;
964 	struct printbuf p = PRINTBUF;
965 	bool first = true;
966 
967 	prt_str(&p, "mounting version ");
968 	bch2_version_to_text(&p, c->sb.version);
969 
970 	if (c->opts.read_only) {
971 		prt_str(&p, " opts=");
972 		first = false;
973 		prt_printf(&p, "ro");
974 	}
975 
976 	for (i = 0; i < bch2_opts_nr; i++) {
977 		const struct bch_option *opt = &bch2_opt_table[i];
978 		u64 v = bch2_opt_get_by_id(&c->opts, i);
979 
980 		if (!(opt->flags & OPT_MOUNT))
981 			continue;
982 
983 		if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
984 			continue;
985 
986 		prt_str(&p, first ? " opts=" : ",");
987 		first = false;
988 		bch2_opt_to_text(&p, c, c->disk_sb.sb, opt, v, OPT_SHOW_MOUNT_STYLE);
989 	}
990 
991 	bch_info(c, "%s", p.buf);
992 	printbuf_exit(&p);
993 }
994 
995 int bch2_fs_start(struct bch_fs *c)
996 {
997 	time64_t now = ktime_get_real_seconds();
998 	int ret;
999 
1000 	print_mount_opts(c);
1001 
1002 	down_write(&c->state_lock);
1003 
1004 	BUG_ON(test_bit(BCH_FS_started, &c->flags));
1005 
1006 	mutex_lock(&c->sb_lock);
1007 
1008 	ret = bch2_sb_members_v2_init(c);
1009 	if (ret) {
1010 		mutex_unlock(&c->sb_lock);
1011 		goto err;
1012 	}
1013 
1014 	for_each_online_member(c, ca)
1015 		bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx)->last_mount = cpu_to_le64(now);
1016 
1017 	mutex_unlock(&c->sb_lock);
1018 
1019 	for_each_rw_member(c, ca)
1020 		bch2_dev_allocator_add(c, ca);
1021 	bch2_recalc_capacity(c);
1022 
1023 	ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
1024 		? bch2_fs_recovery(c)
1025 		: bch2_fs_initialize(c);
1026 	if (ret)
1027 		goto err;
1028 
1029 	ret = bch2_opts_check_may_set(c);
1030 	if (ret)
1031 		goto err;
1032 
1033 	if (bch2_fs_init_fault("fs_start")) {
1034 		bch_err(c, "fs_start fault injected");
1035 		ret = -EINVAL;
1036 		goto err;
1037 	}
1038 
1039 	set_bit(BCH_FS_started, &c->flags);
1040 
1041 	if (c->opts.read_only) {
1042 		bch2_fs_read_only(c);
1043 	} else {
1044 		ret = !test_bit(BCH_FS_rw, &c->flags)
1045 			? bch2_fs_read_write(c)
1046 			: bch2_fs_read_write_late(c);
1047 		if (ret)
1048 			goto err;
1049 	}
1050 
1051 	ret = 0;
1052 err:
1053 	if (ret)
1054 		bch_err_msg(c, ret, "starting filesystem");
1055 	else
1056 		bch_verbose(c, "done starting filesystem");
1057 	up_write(&c->state_lock);
1058 	return ret;
1059 }
1060 
1061 static int bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
1062 {
1063 	struct bch_member m = bch2_sb_member_get(sb, sb->dev_idx);
1064 
1065 	if (le16_to_cpu(sb->block_size) != block_sectors(c))
1066 		return -BCH_ERR_mismatched_block_size;
1067 
1068 	if (le16_to_cpu(m.bucket_size) <
1069 	    BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
1070 		return -BCH_ERR_bucket_size_too_small;
1071 
1072 	return 0;
1073 }
1074 
1075 static int bch2_dev_in_fs(struct bch_sb_handle *fs,
1076 			  struct bch_sb_handle *sb,
1077 			  struct bch_opts *opts)
1078 {
1079 	if (fs == sb)
1080 		return 0;
1081 
1082 	if (!uuid_equal(&fs->sb->uuid, &sb->sb->uuid))
1083 		return -BCH_ERR_device_not_a_member_of_filesystem;
1084 
1085 	if (!bch2_dev_exists(fs->sb, sb->sb->dev_idx))
1086 		return -BCH_ERR_device_has_been_removed;
1087 
1088 	if (fs->sb->block_size != sb->sb->block_size)
1089 		return -BCH_ERR_mismatched_block_size;
1090 
1091 	if (le16_to_cpu(fs->sb->version) < bcachefs_metadata_version_member_seq ||
1092 	    le16_to_cpu(sb->sb->version) < bcachefs_metadata_version_member_seq)
1093 		return 0;
1094 
1095 	if (fs->sb->seq == sb->sb->seq &&
1096 	    fs->sb->write_time != sb->sb->write_time) {
1097 		struct printbuf buf = PRINTBUF;
1098 
1099 		prt_str(&buf, "Split brain detected between ");
1100 		prt_bdevname(&buf, sb->bdev);
1101 		prt_str(&buf, " and ");
1102 		prt_bdevname(&buf, fs->bdev);
1103 		prt_char(&buf, ':');
1104 		prt_newline(&buf);
1105 		prt_printf(&buf, "seq=%llu but write_time different, got", le64_to_cpu(sb->sb->seq));
1106 		prt_newline(&buf);
1107 
1108 		prt_bdevname(&buf, fs->bdev);
1109 		prt_char(&buf, ' ');
1110 		bch2_prt_datetime(&buf, le64_to_cpu(fs->sb->write_time));;
1111 		prt_newline(&buf);
1112 
1113 		prt_bdevname(&buf, sb->bdev);
1114 		prt_char(&buf, ' ');
1115 		bch2_prt_datetime(&buf, le64_to_cpu(sb->sb->write_time));;
1116 		prt_newline(&buf);
1117 
1118 		if (!opts->no_splitbrain_check)
1119 			prt_printf(&buf, "Not using older sb");
1120 
1121 		pr_err("%s", buf.buf);
1122 		printbuf_exit(&buf);
1123 
1124 		if (!opts->no_splitbrain_check)
1125 			return -BCH_ERR_device_splitbrain;
1126 	}
1127 
1128 	struct bch_member m = bch2_sb_member_get(fs->sb, sb->sb->dev_idx);
1129 	u64 seq_from_fs		= le64_to_cpu(m.seq);
1130 	u64 seq_from_member	= le64_to_cpu(sb->sb->seq);
1131 
1132 	if (seq_from_fs && seq_from_fs < seq_from_member) {
1133 		struct printbuf buf = PRINTBUF;
1134 
1135 		prt_str(&buf, "Split brain detected between ");
1136 		prt_bdevname(&buf, sb->bdev);
1137 		prt_str(&buf, " and ");
1138 		prt_bdevname(&buf, fs->bdev);
1139 		prt_char(&buf, ':');
1140 		prt_newline(&buf);
1141 
1142 		prt_bdevname(&buf, fs->bdev);
1143 		prt_str(&buf, " believes seq of ");
1144 		prt_bdevname(&buf, sb->bdev);
1145 		prt_printf(&buf, " to be %llu, but ", seq_from_fs);
1146 		prt_bdevname(&buf, sb->bdev);
1147 		prt_printf(&buf, " has %llu\n", seq_from_member);
1148 
1149 		if (!opts->no_splitbrain_check) {
1150 			prt_str(&buf, "Not using ");
1151 			prt_bdevname(&buf, sb->bdev);
1152 		}
1153 
1154 		pr_err("%s", buf.buf);
1155 		printbuf_exit(&buf);
1156 
1157 		if (!opts->no_splitbrain_check)
1158 			return -BCH_ERR_device_splitbrain;
1159 	}
1160 
1161 	return 0;
1162 }
1163 
1164 /* Device startup/shutdown: */
1165 
1166 static void bch2_dev_release(struct kobject *kobj)
1167 {
1168 	struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
1169 
1170 	kfree(ca);
1171 }
1172 
1173 static void bch2_dev_free(struct bch_dev *ca)
1174 {
1175 	cancel_work_sync(&ca->io_error_work);
1176 
1177 	if (ca->kobj.state_in_sysfs &&
1178 	    ca->disk_sb.bdev)
1179 		sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1180 
1181 	if (ca->kobj.state_in_sysfs)
1182 		kobject_del(&ca->kobj);
1183 
1184 	bch2_free_super(&ca->disk_sb);
1185 	bch2_dev_journal_exit(ca);
1186 
1187 	free_percpu(ca->io_done);
1188 	bioset_exit(&ca->replica_set);
1189 	bch2_dev_buckets_free(ca);
1190 	free_page((unsigned long) ca->sb_read_scratch);
1191 
1192 	bch2_time_stats_quantiles_exit(&ca->io_latency[WRITE]);
1193 	bch2_time_stats_quantiles_exit(&ca->io_latency[READ]);
1194 
1195 	percpu_ref_exit(&ca->io_ref);
1196 	percpu_ref_exit(&ca->ref);
1197 	kobject_put(&ca->kobj);
1198 }
1199 
1200 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
1201 {
1202 
1203 	lockdep_assert_held(&c->state_lock);
1204 
1205 	if (percpu_ref_is_zero(&ca->io_ref))
1206 		return;
1207 
1208 	__bch2_dev_read_only(c, ca);
1209 
1210 	reinit_completion(&ca->io_ref_completion);
1211 	percpu_ref_kill(&ca->io_ref);
1212 	wait_for_completion(&ca->io_ref_completion);
1213 
1214 	if (ca->kobj.state_in_sysfs) {
1215 		sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1216 		sysfs_remove_link(&ca->kobj, "block");
1217 	}
1218 
1219 	bch2_free_super(&ca->disk_sb);
1220 	bch2_dev_journal_exit(ca);
1221 }
1222 
1223 static void bch2_dev_ref_complete(struct percpu_ref *ref)
1224 {
1225 	struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
1226 
1227 	complete(&ca->ref_completion);
1228 }
1229 
1230 static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
1231 {
1232 	struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
1233 
1234 	complete(&ca->io_ref_completion);
1235 }
1236 
1237 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
1238 {
1239 	int ret;
1240 
1241 	if (!c->kobj.state_in_sysfs)
1242 		return 0;
1243 
1244 	if (!ca->kobj.state_in_sysfs) {
1245 		ret = kobject_add(&ca->kobj, &c->kobj,
1246 				  "dev-%u", ca->dev_idx);
1247 		if (ret)
1248 			return ret;
1249 	}
1250 
1251 	if (ca->disk_sb.bdev) {
1252 		struct kobject *block = bdev_kobj(ca->disk_sb.bdev);
1253 
1254 		ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
1255 		if (ret)
1256 			return ret;
1257 
1258 		ret = sysfs_create_link(&ca->kobj, block, "block");
1259 		if (ret)
1260 			return ret;
1261 	}
1262 
1263 	return 0;
1264 }
1265 
1266 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
1267 					struct bch_member *member)
1268 {
1269 	struct bch_dev *ca;
1270 	unsigned i;
1271 
1272 	ca = kzalloc(sizeof(*ca), GFP_KERNEL);
1273 	if (!ca)
1274 		return NULL;
1275 
1276 	kobject_init(&ca->kobj, &bch2_dev_ktype);
1277 	init_completion(&ca->ref_completion);
1278 	init_completion(&ca->io_ref_completion);
1279 
1280 	init_rwsem(&ca->bucket_lock);
1281 
1282 	INIT_WORK(&ca->io_error_work, bch2_io_error_work);
1283 
1284 	bch2_time_stats_quantiles_init(&ca->io_latency[READ]);
1285 	bch2_time_stats_quantiles_init(&ca->io_latency[WRITE]);
1286 
1287 	ca->mi = bch2_mi_to_cpu(member);
1288 
1289 	for (i = 0; i < ARRAY_SIZE(member->errors); i++)
1290 		atomic64_set(&ca->errors[i], le64_to_cpu(member->errors[i]));
1291 
1292 	ca->uuid = member->uuid;
1293 
1294 	ca->nr_btree_reserve = DIV_ROUND_UP(BTREE_NODE_RESERVE,
1295 			     ca->mi.bucket_size / btree_sectors(c));
1296 
1297 	if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete,
1298 			    0, GFP_KERNEL) ||
1299 	    percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
1300 			    PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
1301 	    !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) ||
1302 	    bch2_dev_buckets_alloc(c, ca) ||
1303 	    bioset_init(&ca->replica_set, 4,
1304 			offsetof(struct bch_write_bio, bio), 0) ||
1305 	    !(ca->io_done	= alloc_percpu(*ca->io_done)))
1306 		goto err;
1307 
1308 	return ca;
1309 err:
1310 	bch2_dev_free(ca);
1311 	return NULL;
1312 }
1313 
1314 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
1315 			    unsigned dev_idx)
1316 {
1317 	ca->dev_idx = dev_idx;
1318 	__set_bit(ca->dev_idx, ca->self.d);
1319 	scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
1320 
1321 	ca->fs = c;
1322 	rcu_assign_pointer(c->devs[ca->dev_idx], ca);
1323 
1324 	if (bch2_dev_sysfs_online(c, ca))
1325 		pr_warn("error creating sysfs objects");
1326 }
1327 
1328 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
1329 {
1330 	struct bch_member member = bch2_sb_member_get(c->disk_sb.sb, dev_idx);
1331 	struct bch_dev *ca = NULL;
1332 	int ret = 0;
1333 
1334 	if (bch2_fs_init_fault("dev_alloc"))
1335 		goto err;
1336 
1337 	ca = __bch2_dev_alloc(c, &member);
1338 	if (!ca)
1339 		goto err;
1340 
1341 	ca->fs = c;
1342 
1343 	bch2_dev_attach(c, ca, dev_idx);
1344 	return ret;
1345 err:
1346 	if (ca)
1347 		bch2_dev_free(ca);
1348 	return -BCH_ERR_ENOMEM_dev_alloc;
1349 }
1350 
1351 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1352 {
1353 	unsigned ret;
1354 
1355 	if (bch2_dev_is_online(ca)) {
1356 		bch_err(ca, "already have device online in slot %u",
1357 			sb->sb->dev_idx);
1358 		return -BCH_ERR_device_already_online;
1359 	}
1360 
1361 	if (get_capacity(sb->bdev->bd_disk) <
1362 	    ca->mi.bucket_size * ca->mi.nbuckets) {
1363 		bch_err(ca, "cannot online: device too small");
1364 		return -BCH_ERR_device_size_too_small;
1365 	}
1366 
1367 	BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1368 
1369 	ret = bch2_dev_journal_init(ca, sb->sb);
1370 	if (ret)
1371 		return ret;
1372 
1373 	/* Commit: */
1374 	ca->disk_sb = *sb;
1375 	memset(sb, 0, sizeof(*sb));
1376 
1377 	ca->dev = ca->disk_sb.bdev->bd_dev;
1378 
1379 	percpu_ref_reinit(&ca->io_ref);
1380 
1381 	return 0;
1382 }
1383 
1384 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1385 {
1386 	struct bch_dev *ca;
1387 	int ret;
1388 
1389 	lockdep_assert_held(&c->state_lock);
1390 
1391 	if (le64_to_cpu(sb->sb->seq) >
1392 	    le64_to_cpu(c->disk_sb.sb->seq))
1393 		bch2_sb_to_fs(c, sb->sb);
1394 
1395 	BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices ||
1396 	       !c->devs[sb->sb->dev_idx]);
1397 
1398 	ca = bch_dev_locked(c, sb->sb->dev_idx);
1399 
1400 	ret = __bch2_dev_attach_bdev(ca, sb);
1401 	if (ret)
1402 		return ret;
1403 
1404 	bch2_dev_sysfs_online(c, ca);
1405 
1406 	struct printbuf name = PRINTBUF;
1407 	prt_bdevname(&name, ca->disk_sb.bdev);
1408 
1409 	if (c->sb.nr_devices == 1)
1410 		strscpy(c->name, name.buf, sizeof(c->name));
1411 	strscpy(ca->name, name.buf, sizeof(ca->name));
1412 
1413 	printbuf_exit(&name);
1414 
1415 	rebalance_wakeup(c);
1416 	return 0;
1417 }
1418 
1419 /* Device management: */
1420 
1421 /*
1422  * Note: this function is also used by the error paths - when a particular
1423  * device sees an error, we call it to determine whether we can just set the
1424  * device RO, or - if this function returns false - we'll set the whole
1425  * filesystem RO:
1426  *
1427  * XXX: maybe we should be more explicit about whether we're changing state
1428  * because we got an error or what have you?
1429  */
1430 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1431 			    enum bch_member_state new_state, int flags)
1432 {
1433 	struct bch_devs_mask new_online_devs;
1434 	int nr_rw = 0, required;
1435 
1436 	lockdep_assert_held(&c->state_lock);
1437 
1438 	switch (new_state) {
1439 	case BCH_MEMBER_STATE_rw:
1440 		return true;
1441 	case BCH_MEMBER_STATE_ro:
1442 		if (ca->mi.state != BCH_MEMBER_STATE_rw)
1443 			return true;
1444 
1445 		/* do we have enough devices to write to?  */
1446 		for_each_member_device(c, ca2)
1447 			if (ca2 != ca)
1448 				nr_rw += ca2->mi.state == BCH_MEMBER_STATE_rw;
1449 
1450 		required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1451 			       ? c->opts.metadata_replicas
1452 			       : metadata_replicas_required(c),
1453 			       !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1454 			       ? c->opts.data_replicas
1455 			       : data_replicas_required(c));
1456 
1457 		return nr_rw >= required;
1458 	case BCH_MEMBER_STATE_failed:
1459 	case BCH_MEMBER_STATE_spare:
1460 		if (ca->mi.state != BCH_MEMBER_STATE_rw &&
1461 		    ca->mi.state != BCH_MEMBER_STATE_ro)
1462 			return true;
1463 
1464 		/* do we have enough devices to read from?  */
1465 		new_online_devs = bch2_online_devs(c);
1466 		__clear_bit(ca->dev_idx, new_online_devs.d);
1467 
1468 		return bch2_have_enough_devs(c, new_online_devs, flags, false);
1469 	default:
1470 		BUG();
1471 	}
1472 }
1473 
1474 static bool bch2_fs_may_start(struct bch_fs *c)
1475 {
1476 	struct bch_dev *ca;
1477 	unsigned i, flags = 0;
1478 
1479 	if (c->opts.very_degraded)
1480 		flags |= BCH_FORCE_IF_DEGRADED|BCH_FORCE_IF_LOST;
1481 
1482 	if (c->opts.degraded)
1483 		flags |= BCH_FORCE_IF_DEGRADED;
1484 
1485 	if (!c->opts.degraded &&
1486 	    !c->opts.very_degraded) {
1487 		mutex_lock(&c->sb_lock);
1488 
1489 		for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1490 			if (!bch2_dev_exists(c->disk_sb.sb, i))
1491 				continue;
1492 
1493 			ca = bch_dev_locked(c, i);
1494 
1495 			if (!bch2_dev_is_online(ca) &&
1496 			    (ca->mi.state == BCH_MEMBER_STATE_rw ||
1497 			     ca->mi.state == BCH_MEMBER_STATE_ro)) {
1498 				mutex_unlock(&c->sb_lock);
1499 				return false;
1500 			}
1501 		}
1502 		mutex_unlock(&c->sb_lock);
1503 	}
1504 
1505 	return bch2_have_enough_devs(c, bch2_online_devs(c), flags, true);
1506 }
1507 
1508 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1509 {
1510 	/*
1511 	 * The allocator thread itself allocates btree nodes, so stop it first:
1512 	 */
1513 	bch2_dev_allocator_remove(c, ca);
1514 	bch2_dev_journal_stop(&c->journal, ca);
1515 }
1516 
1517 static void __bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1518 {
1519 	lockdep_assert_held(&c->state_lock);
1520 
1521 	BUG_ON(ca->mi.state != BCH_MEMBER_STATE_rw);
1522 
1523 	bch2_dev_allocator_add(c, ca);
1524 	bch2_recalc_capacity(c);
1525 }
1526 
1527 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1528 			 enum bch_member_state new_state, int flags)
1529 {
1530 	struct bch_member *m;
1531 	int ret = 0;
1532 
1533 	if (ca->mi.state == new_state)
1534 		return 0;
1535 
1536 	if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1537 		return -BCH_ERR_device_state_not_allowed;
1538 
1539 	if (new_state != BCH_MEMBER_STATE_rw)
1540 		__bch2_dev_read_only(c, ca);
1541 
1542 	bch_notice(ca, "%s", bch2_member_states[new_state]);
1543 
1544 	mutex_lock(&c->sb_lock);
1545 	m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
1546 	SET_BCH_MEMBER_STATE(m, new_state);
1547 	bch2_write_super(c);
1548 	mutex_unlock(&c->sb_lock);
1549 
1550 	if (new_state == BCH_MEMBER_STATE_rw)
1551 		__bch2_dev_read_write(c, ca);
1552 
1553 	rebalance_wakeup(c);
1554 
1555 	return ret;
1556 }
1557 
1558 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1559 		       enum bch_member_state new_state, int flags)
1560 {
1561 	int ret;
1562 
1563 	down_write(&c->state_lock);
1564 	ret = __bch2_dev_set_state(c, ca, new_state, flags);
1565 	up_write(&c->state_lock);
1566 
1567 	return ret;
1568 }
1569 
1570 /* Device add/removal: */
1571 
1572 static int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
1573 {
1574 	struct bpos start	= POS(ca->dev_idx, 0);
1575 	struct bpos end		= POS(ca->dev_idx, U64_MAX);
1576 	int ret;
1577 
1578 	/*
1579 	 * We clear the LRU and need_discard btrees first so that we don't race
1580 	 * with bch2_do_invalidates() and bch2_do_discards()
1581 	 */
1582 	ret =   bch2_btree_delete_range(c, BTREE_ID_lru, start, end,
1583 					BTREE_TRIGGER_NORUN, NULL) ?:
1584 		bch2_btree_delete_range(c, BTREE_ID_need_discard, start, end,
1585 					BTREE_TRIGGER_NORUN, NULL) ?:
1586 		bch2_btree_delete_range(c, BTREE_ID_freespace, start, end,
1587 					BTREE_TRIGGER_NORUN, NULL) ?:
1588 		bch2_btree_delete_range(c, BTREE_ID_backpointers, start, end,
1589 					BTREE_TRIGGER_NORUN, NULL) ?:
1590 		bch2_btree_delete_range(c, BTREE_ID_alloc, start, end,
1591 					BTREE_TRIGGER_NORUN, NULL) ?:
1592 		bch2_btree_delete_range(c, BTREE_ID_bucket_gens, start, end,
1593 					BTREE_TRIGGER_NORUN, NULL);
1594 	bch_err_msg(c, ret, "removing dev alloc info");
1595 	return ret;
1596 }
1597 
1598 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1599 {
1600 	struct bch_member *m;
1601 	unsigned dev_idx = ca->dev_idx, data;
1602 	int ret;
1603 
1604 	down_write(&c->state_lock);
1605 
1606 	/*
1607 	 * We consume a reference to ca->ref, regardless of whether we succeed
1608 	 * or fail:
1609 	 */
1610 	percpu_ref_put(&ca->ref);
1611 
1612 	if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1613 		bch_err(ca, "Cannot remove without losing data");
1614 		ret = -BCH_ERR_device_state_not_allowed;
1615 		goto err;
1616 	}
1617 
1618 	__bch2_dev_read_only(c, ca);
1619 
1620 	ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1621 	bch_err_msg(ca, ret, "bch2_dev_data_drop()");
1622 	if (ret)
1623 		goto err;
1624 
1625 	ret = bch2_dev_remove_alloc(c, ca);
1626 	bch_err_msg(ca, ret, "bch2_dev_remove_alloc()");
1627 	if (ret)
1628 		goto err;
1629 
1630 	ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1631 	bch_err_msg(ca, ret, "bch2_journal_flush_device_pins()");
1632 	if (ret)
1633 		goto err;
1634 
1635 	ret = bch2_journal_flush(&c->journal);
1636 	bch_err_msg(ca, ret, "bch2_journal_flush()");
1637 	if (ret)
1638 		goto err;
1639 
1640 	ret = bch2_replicas_gc2(c);
1641 	bch_err_msg(ca, ret, "bch2_replicas_gc2()");
1642 	if (ret)
1643 		goto err;
1644 
1645 	data = bch2_dev_has_data(c, ca);
1646 	if (data) {
1647 		struct printbuf data_has = PRINTBUF;
1648 
1649 		prt_bitflags(&data_has, __bch2_data_types, data);
1650 		bch_err(ca, "Remove failed, still has data (%s)", data_has.buf);
1651 		printbuf_exit(&data_has);
1652 		ret = -EBUSY;
1653 		goto err;
1654 	}
1655 
1656 	__bch2_dev_offline(c, ca);
1657 
1658 	mutex_lock(&c->sb_lock);
1659 	rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1660 	mutex_unlock(&c->sb_lock);
1661 
1662 	percpu_ref_kill(&ca->ref);
1663 	wait_for_completion(&ca->ref_completion);
1664 
1665 	bch2_dev_free(ca);
1666 
1667 	/*
1668 	 * At this point the device object has been removed in-core, but the
1669 	 * on-disk journal might still refer to the device index via sb device
1670 	 * usage entries. Recovery fails if it sees usage information for an
1671 	 * invalid device. Flush journal pins to push the back of the journal
1672 	 * past now invalid device index references before we update the
1673 	 * superblock, but after the device object has been removed so any
1674 	 * further journal writes elide usage info for the device.
1675 	 */
1676 	bch2_journal_flush_all_pins(&c->journal);
1677 
1678 	/*
1679 	 * Free this device's slot in the bch_member array - all pointers to
1680 	 * this device must be gone:
1681 	 */
1682 	mutex_lock(&c->sb_lock);
1683 	m = bch2_members_v2_get_mut(c->disk_sb.sb, dev_idx);
1684 	memset(&m->uuid, 0, sizeof(m->uuid));
1685 
1686 	bch2_write_super(c);
1687 
1688 	mutex_unlock(&c->sb_lock);
1689 	up_write(&c->state_lock);
1690 
1691 	bch2_dev_usage_journal_reserve(c);
1692 	return 0;
1693 err:
1694 	if (ca->mi.state == BCH_MEMBER_STATE_rw &&
1695 	    !percpu_ref_is_zero(&ca->io_ref))
1696 		__bch2_dev_read_write(c, ca);
1697 	up_write(&c->state_lock);
1698 	return ret;
1699 }
1700 
1701 /* Add new device to running filesystem: */
1702 int bch2_dev_add(struct bch_fs *c, const char *path)
1703 {
1704 	struct bch_opts opts = bch2_opts_empty();
1705 	struct bch_sb_handle sb;
1706 	struct bch_dev *ca = NULL;
1707 	struct bch_sb_field_members_v2 *mi;
1708 	struct bch_member dev_mi;
1709 	unsigned dev_idx, nr_devices, u64s;
1710 	struct printbuf errbuf = PRINTBUF;
1711 	struct printbuf label = PRINTBUF;
1712 	int ret;
1713 
1714 	ret = bch2_read_super(path, &opts, &sb);
1715 	bch_err_msg(c, ret, "reading super");
1716 	if (ret)
1717 		goto err;
1718 
1719 	dev_mi = bch2_sb_member_get(sb.sb, sb.sb->dev_idx);
1720 
1721 	if (BCH_MEMBER_GROUP(&dev_mi)) {
1722 		bch2_disk_path_to_text_sb(&label, sb.sb, BCH_MEMBER_GROUP(&dev_mi) - 1);
1723 		if (label.allocation_failure) {
1724 			ret = -ENOMEM;
1725 			goto err;
1726 		}
1727 	}
1728 
1729 	ret = bch2_dev_may_add(sb.sb, c);
1730 	if (ret)
1731 		goto err;
1732 
1733 	ca = __bch2_dev_alloc(c, &dev_mi);
1734 	if (!ca) {
1735 		ret = -ENOMEM;
1736 		goto err;
1737 	}
1738 
1739 	bch2_dev_usage_init(ca);
1740 
1741 	ret = __bch2_dev_attach_bdev(ca, &sb);
1742 	if (ret)
1743 		goto err;
1744 
1745 	ret = bch2_dev_journal_alloc(ca);
1746 	bch_err_msg(c, ret, "allocating journal");
1747 	if (ret)
1748 		goto err;
1749 
1750 	down_write(&c->state_lock);
1751 	mutex_lock(&c->sb_lock);
1752 
1753 	ret = bch2_sb_from_fs(c, ca);
1754 	bch_err_msg(c, ret, "setting up new superblock");
1755 	if (ret)
1756 		goto err_unlock;
1757 
1758 	if (dynamic_fault("bcachefs:add:no_slot"))
1759 		goto no_slot;
1760 
1761 	for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++)
1762 		if (!bch2_dev_exists(c->disk_sb.sb, dev_idx))
1763 			goto have_slot;
1764 no_slot:
1765 	ret = -BCH_ERR_ENOSPC_sb_members;
1766 	bch_err_msg(c, ret, "setting up new superblock");
1767 	goto err_unlock;
1768 
1769 have_slot:
1770 	nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1771 
1772 	mi = bch2_sb_field_get(c->disk_sb.sb, members_v2);
1773 	u64s = DIV_ROUND_UP(sizeof(struct bch_sb_field_members_v2) +
1774 			    le16_to_cpu(mi->member_bytes) * nr_devices, sizeof(u64));
1775 
1776 	mi = bch2_sb_field_resize(&c->disk_sb, members_v2, u64s);
1777 	if (!mi) {
1778 		ret = -BCH_ERR_ENOSPC_sb_members;
1779 		bch_err_msg(c, ret, "setting up new superblock");
1780 		goto err_unlock;
1781 	}
1782 	struct bch_member *m = bch2_members_v2_get_mut(c->disk_sb.sb, dev_idx);
1783 
1784 	/* success: */
1785 
1786 	*m = dev_mi;
1787 	m->last_mount = cpu_to_le64(ktime_get_real_seconds());
1788 	c->disk_sb.sb->nr_devices	= nr_devices;
1789 
1790 	ca->disk_sb.sb->dev_idx	= dev_idx;
1791 	bch2_dev_attach(c, ca, dev_idx);
1792 
1793 	if (BCH_MEMBER_GROUP(&dev_mi)) {
1794 		ret = __bch2_dev_group_set(c, ca, label.buf);
1795 		bch_err_msg(c, ret, "creating new label");
1796 		if (ret)
1797 			goto err_unlock;
1798 	}
1799 
1800 	bch2_write_super(c);
1801 	mutex_unlock(&c->sb_lock);
1802 
1803 	bch2_dev_usage_journal_reserve(c);
1804 
1805 	ret = bch2_trans_mark_dev_sb(c, ca);
1806 	bch_err_msg(ca, ret, "marking new superblock");
1807 	if (ret)
1808 		goto err_late;
1809 
1810 	ret = bch2_fs_freespace_init(c);
1811 	bch_err_msg(ca, ret, "initializing free space");
1812 	if (ret)
1813 		goto err_late;
1814 
1815 	ca->new_fs_bucket_idx = 0;
1816 
1817 	if (ca->mi.state == BCH_MEMBER_STATE_rw)
1818 		__bch2_dev_read_write(c, ca);
1819 
1820 	up_write(&c->state_lock);
1821 	return 0;
1822 
1823 err_unlock:
1824 	mutex_unlock(&c->sb_lock);
1825 	up_write(&c->state_lock);
1826 err:
1827 	if (ca)
1828 		bch2_dev_free(ca);
1829 	bch2_free_super(&sb);
1830 	printbuf_exit(&label);
1831 	printbuf_exit(&errbuf);
1832 	bch_err_fn(c, ret);
1833 	return ret;
1834 err_late:
1835 	up_write(&c->state_lock);
1836 	ca = NULL;
1837 	goto err;
1838 }
1839 
1840 /* Hot add existing device to running filesystem: */
1841 int bch2_dev_online(struct bch_fs *c, const char *path)
1842 {
1843 	struct bch_opts opts = bch2_opts_empty();
1844 	struct bch_sb_handle sb = { NULL };
1845 	struct bch_dev *ca;
1846 	unsigned dev_idx;
1847 	int ret;
1848 
1849 	down_write(&c->state_lock);
1850 
1851 	ret = bch2_read_super(path, &opts, &sb);
1852 	if (ret) {
1853 		up_write(&c->state_lock);
1854 		return ret;
1855 	}
1856 
1857 	dev_idx = sb.sb->dev_idx;
1858 
1859 	ret = bch2_dev_in_fs(&c->disk_sb, &sb, &c->opts);
1860 	bch_err_msg(c, ret, "bringing %s online", path);
1861 	if (ret)
1862 		goto err;
1863 
1864 	ret = bch2_dev_attach_bdev(c, &sb);
1865 	if (ret)
1866 		goto err;
1867 
1868 	ca = bch_dev_locked(c, dev_idx);
1869 
1870 	ret = bch2_trans_mark_dev_sb(c, ca);
1871 	bch_err_msg(c, ret, "bringing %s online: error from bch2_trans_mark_dev_sb", path);
1872 	if (ret)
1873 		goto err;
1874 
1875 	if (ca->mi.state == BCH_MEMBER_STATE_rw)
1876 		__bch2_dev_read_write(c, ca);
1877 
1878 	if (!ca->mi.freespace_initialized) {
1879 		ret = bch2_dev_freespace_init(c, ca, 0, ca->mi.nbuckets);
1880 		bch_err_msg(ca, ret, "initializing free space");
1881 		if (ret)
1882 			goto err;
1883 	}
1884 
1885 	if (!ca->journal.nr) {
1886 		ret = bch2_dev_journal_alloc(ca);
1887 		bch_err_msg(ca, ret, "allocating journal");
1888 		if (ret)
1889 			goto err;
1890 	}
1891 
1892 	mutex_lock(&c->sb_lock);
1893 	bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx)->last_mount =
1894 		cpu_to_le64(ktime_get_real_seconds());
1895 	bch2_write_super(c);
1896 	mutex_unlock(&c->sb_lock);
1897 
1898 	up_write(&c->state_lock);
1899 	return 0;
1900 err:
1901 	up_write(&c->state_lock);
1902 	bch2_free_super(&sb);
1903 	return ret;
1904 }
1905 
1906 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1907 {
1908 	down_write(&c->state_lock);
1909 
1910 	if (!bch2_dev_is_online(ca)) {
1911 		bch_err(ca, "Already offline");
1912 		up_write(&c->state_lock);
1913 		return 0;
1914 	}
1915 
1916 	if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1917 		bch_err(ca, "Cannot offline required disk");
1918 		up_write(&c->state_lock);
1919 		return -BCH_ERR_device_state_not_allowed;
1920 	}
1921 
1922 	__bch2_dev_offline(c, ca);
1923 
1924 	up_write(&c->state_lock);
1925 	return 0;
1926 }
1927 
1928 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1929 {
1930 	struct bch_member *m;
1931 	u64 old_nbuckets;
1932 	int ret = 0;
1933 
1934 	down_write(&c->state_lock);
1935 	old_nbuckets = ca->mi.nbuckets;
1936 
1937 	if (nbuckets < ca->mi.nbuckets) {
1938 		bch_err(ca, "Cannot shrink yet");
1939 		ret = -EINVAL;
1940 		goto err;
1941 	}
1942 
1943 	if (bch2_dev_is_online(ca) &&
1944 	    get_capacity(ca->disk_sb.bdev->bd_disk) <
1945 	    ca->mi.bucket_size * nbuckets) {
1946 		bch_err(ca, "New size larger than device");
1947 		ret = -BCH_ERR_device_size_too_small;
1948 		goto err;
1949 	}
1950 
1951 	ret = bch2_dev_buckets_resize(c, ca, nbuckets);
1952 	bch_err_msg(ca, ret, "resizing buckets");
1953 	if (ret)
1954 		goto err;
1955 
1956 	ret = bch2_trans_mark_dev_sb(c, ca);
1957 	if (ret)
1958 		goto err;
1959 
1960 	mutex_lock(&c->sb_lock);
1961 	m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
1962 	m->nbuckets = cpu_to_le64(nbuckets);
1963 
1964 	bch2_write_super(c);
1965 	mutex_unlock(&c->sb_lock);
1966 
1967 	if (ca->mi.freespace_initialized) {
1968 		ret = bch2_dev_freespace_init(c, ca, old_nbuckets, nbuckets);
1969 		if (ret)
1970 			goto err;
1971 
1972 		/*
1973 		 * XXX: this is all wrong transactionally - we'll be able to do
1974 		 * this correctly after the disk space accounting rewrite
1975 		 */
1976 		ca->usage_base->d[BCH_DATA_free].buckets += nbuckets - old_nbuckets;
1977 	}
1978 
1979 	bch2_recalc_capacity(c);
1980 err:
1981 	up_write(&c->state_lock);
1982 	return ret;
1983 }
1984 
1985 /* return with ref on ca->ref: */
1986 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *name)
1987 {
1988 	rcu_read_lock();
1989 	for_each_member_device_rcu(c, ca, NULL)
1990 		if (!strcmp(name, ca->name)) {
1991 			rcu_read_unlock();
1992 			return ca;
1993 		}
1994 	rcu_read_unlock();
1995 	return ERR_PTR(-BCH_ERR_ENOENT_dev_not_found);
1996 }
1997 
1998 /* Filesystem open: */
1999 
2000 static inline int sb_cmp(struct bch_sb *l, struct bch_sb *r)
2001 {
2002 	return  cmp_int(le64_to_cpu(l->seq), le64_to_cpu(r->seq)) ?:
2003 		cmp_int(le64_to_cpu(l->write_time), le64_to_cpu(r->write_time));
2004 }
2005 
2006 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
2007 			    struct bch_opts opts)
2008 {
2009 	DARRAY(struct bch_sb_handle) sbs = { 0 };
2010 	struct bch_fs *c = NULL;
2011 	struct bch_sb_handle *best = NULL;
2012 	struct printbuf errbuf = PRINTBUF;
2013 	int ret = 0;
2014 
2015 	if (!try_module_get(THIS_MODULE))
2016 		return ERR_PTR(-ENODEV);
2017 
2018 	if (!nr_devices) {
2019 		ret = -EINVAL;
2020 		goto err;
2021 	}
2022 
2023 	ret = darray_make_room(&sbs, nr_devices);
2024 	if (ret)
2025 		goto err;
2026 
2027 	for (unsigned i = 0; i < nr_devices; i++) {
2028 		struct bch_sb_handle sb = { NULL };
2029 
2030 		ret = bch2_read_super(devices[i], &opts, &sb);
2031 		if (ret)
2032 			goto err;
2033 
2034 		BUG_ON(darray_push(&sbs, sb));
2035 	}
2036 
2037 	if (opts.nochanges && !opts.read_only) {
2038 		ret = -BCH_ERR_erofs_nochanges;
2039 		goto err_print;
2040 	}
2041 
2042 	darray_for_each(sbs, sb)
2043 		if (!best || sb_cmp(sb->sb, best->sb) > 0)
2044 			best = sb;
2045 
2046 	darray_for_each_reverse(sbs, sb) {
2047 		ret = bch2_dev_in_fs(best, sb, &opts);
2048 
2049 		if (ret == -BCH_ERR_device_has_been_removed ||
2050 		    ret == -BCH_ERR_device_splitbrain) {
2051 			bch2_free_super(sb);
2052 			darray_remove_item(&sbs, sb);
2053 			best -= best > sb;
2054 			ret = 0;
2055 			continue;
2056 		}
2057 
2058 		if (ret)
2059 			goto err_print;
2060 	}
2061 
2062 	c = bch2_fs_alloc(best->sb, opts);
2063 	ret = PTR_ERR_OR_ZERO(c);
2064 	if (ret)
2065 		goto err;
2066 
2067 	down_write(&c->state_lock);
2068 	darray_for_each(sbs, sb) {
2069 		ret = bch2_dev_attach_bdev(c, sb);
2070 		if (ret) {
2071 			up_write(&c->state_lock);
2072 			goto err;
2073 		}
2074 	}
2075 	up_write(&c->state_lock);
2076 
2077 	if (!bch2_fs_may_start(c)) {
2078 		ret = -BCH_ERR_insufficient_devices_to_start;
2079 		goto err_print;
2080 	}
2081 
2082 	if (!c->opts.nostart) {
2083 		ret = bch2_fs_start(c);
2084 		if (ret)
2085 			goto err;
2086 	}
2087 out:
2088 	darray_for_each(sbs, sb)
2089 		bch2_free_super(sb);
2090 	darray_exit(&sbs);
2091 	printbuf_exit(&errbuf);
2092 	module_put(THIS_MODULE);
2093 	return c;
2094 err_print:
2095 	pr_err("bch_fs_open err opening %s: %s",
2096 	       devices[0], bch2_err_str(ret));
2097 err:
2098 	if (!IS_ERR_OR_NULL(c))
2099 		bch2_fs_stop(c);
2100 	c = ERR_PTR(ret);
2101 	goto out;
2102 }
2103 
2104 /* Global interfaces/init */
2105 
2106 static void bcachefs_exit(void)
2107 {
2108 	bch2_debug_exit();
2109 	bch2_vfs_exit();
2110 	bch2_chardev_exit();
2111 	bch2_btree_key_cache_exit();
2112 	if (bcachefs_kset)
2113 		kset_unregister(bcachefs_kset);
2114 }
2115 
2116 static int __init bcachefs_init(void)
2117 {
2118 	bch2_bkey_pack_test();
2119 
2120 	if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
2121 	    bch2_btree_key_cache_init() ||
2122 	    bch2_chardev_init() ||
2123 	    bch2_vfs_init() ||
2124 	    bch2_debug_init())
2125 		goto err;
2126 
2127 	return 0;
2128 err:
2129 	bcachefs_exit();
2130 	return -ENOMEM;
2131 }
2132 
2133 #define BCH_DEBUG_PARAM(name, description)			\
2134 	bool bch2_##name;					\
2135 	module_param_named(name, bch2_##name, bool, 0644);	\
2136 	MODULE_PARM_DESC(name, description);
2137 BCH_DEBUG_PARAMS()
2138 #undef BCH_DEBUG_PARAM
2139 
2140 __maybe_unused
2141 static unsigned bch2_metadata_version = bcachefs_metadata_version_current;
2142 module_param_named(version, bch2_metadata_version, uint, 0400);
2143 
2144 module_exit(bcachefs_exit);
2145 module_init(bcachefs_init);
2146