xref: /linux/fs/bcachefs/super.c (revision d7f39aee79f04eeaa42085728423501b33ac5be5)
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_node_scan.h"
19 #include "btree_update_interior.h"
20 #include "btree_io.h"
21 #include "btree_write_buffer.h"
22 #include "buckets_waiting_for_journal.h"
23 #include "chardev.h"
24 #include "checksum.h"
25 #include "clock.h"
26 #include "compress.h"
27 #include "debug.h"
28 #include "disk_groups.h"
29 #include "ec.h"
30 #include "errcode.h"
31 #include "error.h"
32 #include "fs.h"
33 #include "fs-io.h"
34 #include "fs-io-buffered.h"
35 #include "fs-io-direct.h"
36 #include "fsck.h"
37 #include "inode.h"
38 #include "io_read.h"
39 #include "io_write.h"
40 #include "journal.h"
41 #include "journal_reclaim.h"
42 #include "journal_seq_blacklist.h"
43 #include "move.h"
44 #include "migrate.h"
45 #include "movinggc.h"
46 #include "nocow_locking.h"
47 #include "quota.h"
48 #include "rebalance.h"
49 #include "recovery.h"
50 #include "replicas.h"
51 #include "sb-clean.h"
52 #include "sb-counters.h"
53 #include "sb-errors.h"
54 #include "sb-members.h"
55 #include "snapshot.h"
56 #include "subvolume.h"
57 #include "super.h"
58 #include "super-io.h"
59 #include "sysfs.h"
60 #include "thread_with_file.h"
61 #include "trace.h"
62 
63 #include <linux/backing-dev.h>
64 #include <linux/blkdev.h>
65 #include <linux/debugfs.h>
66 #include <linux/device.h>
67 #include <linux/idr.h>
68 #include <linux/module.h>
69 #include <linux/percpu.h>
70 #include <linux/random.h>
71 #include <linux/sysfs.h>
72 #include <crypto/hash.h>
73 
74 MODULE_LICENSE("GPL");
75 MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
76 MODULE_DESCRIPTION("bcachefs filesystem");
77 MODULE_SOFTDEP("pre: crc32c");
78 MODULE_SOFTDEP("pre: crc64");
79 MODULE_SOFTDEP("pre: sha256");
80 MODULE_SOFTDEP("pre: chacha20");
81 MODULE_SOFTDEP("pre: poly1305");
82 MODULE_SOFTDEP("pre: xxhash");
83 
84 const char * const bch2_fs_flag_strs[] = {
85 #define x(n)		#n,
86 	BCH_FS_FLAGS()
87 #undef x
88 	NULL
89 };
90 
91 __printf(2, 0)
92 static void bch2_print_maybe_redirect(struct stdio_redirect *stdio, const char *fmt, va_list args)
93 {
94 #ifdef __KERNEL__
95 	if (unlikely(stdio)) {
96 		if (fmt[0] == KERN_SOH[0])
97 			fmt += 2;
98 
99 		bch2_stdio_redirect_vprintf(stdio, true, fmt, args);
100 		return;
101 	}
102 #endif
103 	vprintk(fmt, args);
104 }
105 
106 void bch2_print_opts(struct bch_opts *opts, const char *fmt, ...)
107 {
108 	struct stdio_redirect *stdio = (void *)(unsigned long)opts->stdio;
109 
110 	va_list args;
111 	va_start(args, fmt);
112 	bch2_print_maybe_redirect(stdio, fmt, args);
113 	va_end(args);
114 }
115 
116 void __bch2_print(struct bch_fs *c, const char *fmt, ...)
117 {
118 	struct stdio_redirect *stdio = bch2_fs_stdio_redirect(c);
119 
120 	va_list args;
121 	va_start(args, fmt);
122 	bch2_print_maybe_redirect(stdio, fmt, args);
123 	va_end(args);
124 }
125 
126 #define KTYPE(type)							\
127 static const struct attribute_group type ## _group = {			\
128 	.attrs = type ## _files						\
129 };									\
130 									\
131 static const struct attribute_group *type ## _groups[] = {		\
132 	&type ## _group,						\
133 	NULL								\
134 };									\
135 									\
136 static const struct kobj_type type ## _ktype = {			\
137 	.release	= type ## _release,				\
138 	.sysfs_ops	= &type ## _sysfs_ops,				\
139 	.default_groups = type ## _groups				\
140 }
141 
142 static void bch2_fs_release(struct kobject *);
143 static void bch2_dev_release(struct kobject *);
144 static void bch2_fs_counters_release(struct kobject *k)
145 {
146 }
147 
148 static void bch2_fs_internal_release(struct kobject *k)
149 {
150 }
151 
152 static void bch2_fs_opts_dir_release(struct kobject *k)
153 {
154 }
155 
156 static void bch2_fs_time_stats_release(struct kobject *k)
157 {
158 }
159 
160 KTYPE(bch2_fs);
161 KTYPE(bch2_fs_counters);
162 KTYPE(bch2_fs_internal);
163 KTYPE(bch2_fs_opts_dir);
164 KTYPE(bch2_fs_time_stats);
165 KTYPE(bch2_dev);
166 
167 static struct kset *bcachefs_kset;
168 static LIST_HEAD(bch_fs_list);
169 static DEFINE_MUTEX(bch_fs_list_lock);
170 
171 DECLARE_WAIT_QUEUE_HEAD(bch2_read_only_wait);
172 
173 static void bch2_dev_free(struct bch_dev *);
174 static int bch2_dev_alloc(struct bch_fs *, unsigned);
175 static int bch2_dev_sysfs_online(struct bch_fs *, struct bch_dev *);
176 static void __bch2_dev_read_only(struct bch_fs *, struct bch_dev *);
177 
178 struct bch_fs *bch2_dev_to_fs(dev_t dev)
179 {
180 	struct bch_fs *c;
181 
182 	mutex_lock(&bch_fs_list_lock);
183 	rcu_read_lock();
184 
185 	list_for_each_entry(c, &bch_fs_list, list)
186 		for_each_member_device_rcu(c, ca, NULL)
187 			if (ca->disk_sb.bdev && ca->disk_sb.bdev->bd_dev == dev) {
188 				closure_get(&c->cl);
189 				goto found;
190 			}
191 	c = NULL;
192 found:
193 	rcu_read_unlock();
194 	mutex_unlock(&bch_fs_list_lock);
195 
196 	return c;
197 }
198 
199 static struct bch_fs *__bch2_uuid_to_fs(__uuid_t uuid)
200 {
201 	struct bch_fs *c;
202 
203 	lockdep_assert_held(&bch_fs_list_lock);
204 
205 	list_for_each_entry(c, &bch_fs_list, list)
206 		if (!memcmp(&c->disk_sb.sb->uuid, &uuid, sizeof(uuid)))
207 			return c;
208 
209 	return NULL;
210 }
211 
212 struct bch_fs *bch2_uuid_to_fs(__uuid_t uuid)
213 {
214 	struct bch_fs *c;
215 
216 	mutex_lock(&bch_fs_list_lock);
217 	c = __bch2_uuid_to_fs(uuid);
218 	if (c)
219 		closure_get(&c->cl);
220 	mutex_unlock(&bch_fs_list_lock);
221 
222 	return c;
223 }
224 
225 static void bch2_dev_usage_journal_reserve(struct bch_fs *c)
226 {
227 	unsigned nr = 0, u64s =
228 		((sizeof(struct jset_entry_dev_usage) +
229 		  sizeof(struct jset_entry_dev_usage_type) * BCH_DATA_NR)) /
230 		sizeof(u64);
231 
232 	rcu_read_lock();
233 	for_each_member_device_rcu(c, ca, NULL)
234 		nr++;
235 	rcu_read_unlock();
236 
237 	bch2_journal_entry_res_resize(&c->journal,
238 			&c->dev_usage_journal_res, u64s * nr);
239 }
240 
241 /* Filesystem RO/RW: */
242 
243 /*
244  * For startup/shutdown of RW stuff, the dependencies are:
245  *
246  * - foreground writes depend on copygc and rebalance (to free up space)
247  *
248  * - copygc and rebalance depend on mark and sweep gc (they actually probably
249  *   don't because they either reserve ahead of time or don't block if
250  *   allocations fail, but allocations can require mark and sweep gc to run
251  *   because of generation number wraparound)
252  *
253  * - all of the above depends on the allocator threads
254  *
255  * - allocator depends on the journal (when it rewrites prios and gens)
256  */
257 
258 static void __bch2_fs_read_only(struct bch_fs *c)
259 {
260 	unsigned clean_passes = 0;
261 	u64 seq = 0;
262 
263 	bch2_fs_ec_stop(c);
264 	bch2_open_buckets_stop(c, NULL, true);
265 	bch2_rebalance_stop(c);
266 	bch2_copygc_stop(c);
267 	bch2_fs_ec_flush(c);
268 
269 	bch_verbose(c, "flushing journal and stopping allocators, journal seq %llu",
270 		    journal_cur_seq(&c->journal));
271 
272 	do {
273 		clean_passes++;
274 
275 		if (bch2_btree_interior_updates_flush(c) ||
276 		    bch2_journal_flush_all_pins(&c->journal) ||
277 		    bch2_btree_flush_all_writes(c) ||
278 		    seq != atomic64_read(&c->journal.seq)) {
279 			seq = atomic64_read(&c->journal.seq);
280 			clean_passes = 0;
281 		}
282 	} while (clean_passes < 2);
283 
284 	bch_verbose(c, "flushing journal and stopping allocators complete, journal seq %llu",
285 		    journal_cur_seq(&c->journal));
286 
287 	if (test_bit(JOURNAL_replay_done, &c->journal.flags) &&
288 	    !test_bit(BCH_FS_emergency_ro, &c->flags))
289 		set_bit(BCH_FS_clean_shutdown, &c->flags);
290 
291 	bch2_fs_journal_stop(&c->journal);
292 
293 	bch_info(c, "%sshutdown complete, journal seq %llu",
294 		 test_bit(BCH_FS_clean_shutdown, &c->flags) ? "" : "un",
295 		 c->journal.seq_ondisk);
296 
297 	/*
298 	 * After stopping journal:
299 	 */
300 	for_each_member_device(c, ca)
301 		bch2_dev_allocator_remove(c, ca);
302 }
303 
304 #ifndef BCH_WRITE_REF_DEBUG
305 static void bch2_writes_disabled(struct percpu_ref *writes)
306 {
307 	struct bch_fs *c = container_of(writes, struct bch_fs, writes);
308 
309 	set_bit(BCH_FS_write_disable_complete, &c->flags);
310 	wake_up(&bch2_read_only_wait);
311 }
312 #endif
313 
314 void bch2_fs_read_only(struct bch_fs *c)
315 {
316 	if (!test_bit(BCH_FS_rw, &c->flags)) {
317 		bch2_journal_reclaim_stop(&c->journal);
318 		return;
319 	}
320 
321 	BUG_ON(test_bit(BCH_FS_write_disable_complete, &c->flags));
322 
323 	bch_verbose(c, "going read-only");
324 
325 	/*
326 	 * Block new foreground-end write operations from starting - any new
327 	 * writes will return -EROFS:
328 	 */
329 	set_bit(BCH_FS_going_ro, &c->flags);
330 #ifndef BCH_WRITE_REF_DEBUG
331 	percpu_ref_kill(&c->writes);
332 #else
333 	for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++)
334 		bch2_write_ref_put(c, i);
335 #endif
336 
337 	/*
338 	 * If we're not doing an emergency shutdown, we want to wait on
339 	 * outstanding writes to complete so they don't see spurious errors due
340 	 * to shutting down the allocator:
341 	 *
342 	 * If we are doing an emergency shutdown outstanding writes may
343 	 * hang until we shutdown the allocator so we don't want to wait
344 	 * on outstanding writes before shutting everything down - but
345 	 * we do need to wait on them before returning and signalling
346 	 * that going RO is complete:
347 	 */
348 	wait_event(bch2_read_only_wait,
349 		   test_bit(BCH_FS_write_disable_complete, &c->flags) ||
350 		   test_bit(BCH_FS_emergency_ro, &c->flags));
351 
352 	bool writes_disabled = test_bit(BCH_FS_write_disable_complete, &c->flags);
353 	if (writes_disabled)
354 		bch_verbose(c, "finished waiting for writes to stop");
355 
356 	__bch2_fs_read_only(c);
357 
358 	wait_event(bch2_read_only_wait,
359 		   test_bit(BCH_FS_write_disable_complete, &c->flags));
360 
361 	if (!writes_disabled)
362 		bch_verbose(c, "finished waiting for writes to stop");
363 
364 	clear_bit(BCH_FS_write_disable_complete, &c->flags);
365 	clear_bit(BCH_FS_going_ro, &c->flags);
366 	clear_bit(BCH_FS_rw, &c->flags);
367 
368 	if (!bch2_journal_error(&c->journal) &&
369 	    !test_bit(BCH_FS_error, &c->flags) &&
370 	    !test_bit(BCH_FS_emergency_ro, &c->flags) &&
371 	    test_bit(BCH_FS_started, &c->flags) &&
372 	    test_bit(BCH_FS_clean_shutdown, &c->flags) &&
373 	    c->recovery_pass_done >= BCH_RECOVERY_PASS_journal_replay) {
374 		BUG_ON(c->journal.last_empty_seq != journal_cur_seq(&c->journal));
375 		BUG_ON(atomic_read(&c->btree_cache.dirty));
376 		BUG_ON(atomic_long_read(&c->btree_key_cache.nr_dirty));
377 		BUG_ON(c->btree_write_buffer.inc.keys.nr);
378 		BUG_ON(c->btree_write_buffer.flushing.keys.nr);
379 
380 		bch_verbose(c, "marking filesystem clean");
381 		bch2_fs_mark_clean(c);
382 	} else {
383 		bch_verbose(c, "done going read-only, filesystem not clean");
384 	}
385 }
386 
387 static void bch2_fs_read_only_work(struct work_struct *work)
388 {
389 	struct bch_fs *c =
390 		container_of(work, struct bch_fs, read_only_work);
391 
392 	down_write(&c->state_lock);
393 	bch2_fs_read_only(c);
394 	up_write(&c->state_lock);
395 }
396 
397 static void bch2_fs_read_only_async(struct bch_fs *c)
398 {
399 	queue_work(system_long_wq, &c->read_only_work);
400 }
401 
402 bool bch2_fs_emergency_read_only(struct bch_fs *c)
403 {
404 	bool ret = !test_and_set_bit(BCH_FS_emergency_ro, &c->flags);
405 
406 	bch2_journal_halt(&c->journal);
407 	bch2_fs_read_only_async(c);
408 
409 	wake_up(&bch2_read_only_wait);
410 	return ret;
411 }
412 
413 static int bch2_fs_read_write_late(struct bch_fs *c)
414 {
415 	int ret;
416 
417 	/*
418 	 * Data move operations can't run until after check_snapshots has
419 	 * completed, and bch2_snapshot_is_ancestor() is available.
420 	 *
421 	 * Ideally we'd start copygc/rebalance earlier instead of waiting for
422 	 * all of recovery/fsck to complete:
423 	 */
424 	ret = bch2_copygc_start(c);
425 	if (ret) {
426 		bch_err(c, "error starting copygc thread");
427 		return ret;
428 	}
429 
430 	ret = bch2_rebalance_start(c);
431 	if (ret) {
432 		bch_err(c, "error starting rebalance thread");
433 		return ret;
434 	}
435 
436 	return 0;
437 }
438 
439 static int __bch2_fs_read_write(struct bch_fs *c, bool early)
440 {
441 	int ret;
442 
443 	if (test_bit(BCH_FS_initial_gc_unfixed, &c->flags)) {
444 		bch_err(c, "cannot go rw, unfixed btree errors");
445 		return -BCH_ERR_erofs_unfixed_errors;
446 	}
447 
448 	if (test_bit(BCH_FS_rw, &c->flags))
449 		return 0;
450 
451 	bch_info(c, "going read-write");
452 
453 	ret = bch2_sb_members_v2_init(c);
454 	if (ret)
455 		goto err;
456 
457 	ret = bch2_fs_mark_dirty(c);
458 	if (ret)
459 		goto err;
460 
461 	clear_bit(BCH_FS_clean_shutdown, &c->flags);
462 
463 	/*
464 	 * First journal write must be a flush write: after a clean shutdown we
465 	 * don't read the journal, so the first journal write may end up
466 	 * overwriting whatever was there previously, and there must always be
467 	 * at least one non-flush write in the journal or recovery will fail:
468 	 */
469 	set_bit(JOURNAL_need_flush_write, &c->journal.flags);
470 	set_bit(JOURNAL_running, &c->journal.flags);
471 
472 	for_each_rw_member(c, ca)
473 		bch2_dev_allocator_add(c, ca);
474 	bch2_recalc_capacity(c);
475 
476 	set_bit(BCH_FS_rw, &c->flags);
477 	set_bit(BCH_FS_was_rw, &c->flags);
478 
479 #ifndef BCH_WRITE_REF_DEBUG
480 	percpu_ref_reinit(&c->writes);
481 #else
482 	for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++) {
483 		BUG_ON(atomic_long_read(&c->writes[i]));
484 		atomic_long_inc(&c->writes[i]);
485 	}
486 #endif
487 
488 	ret = bch2_journal_reclaim_start(&c->journal);
489 	if (ret)
490 		goto err;
491 
492 	if (!early) {
493 		ret = bch2_fs_read_write_late(c);
494 		if (ret)
495 			goto err;
496 	}
497 
498 	bch2_do_discards(c);
499 	bch2_do_invalidates(c);
500 	bch2_do_stripe_deletes(c);
501 	bch2_do_pending_node_rewrites(c);
502 	return 0;
503 err:
504 	if (test_bit(BCH_FS_rw, &c->flags))
505 		bch2_fs_read_only(c);
506 	else
507 		__bch2_fs_read_only(c);
508 	return ret;
509 }
510 
511 int bch2_fs_read_write(struct bch_fs *c)
512 {
513 	if (c->opts.recovery_pass_last &&
514 	    c->opts.recovery_pass_last < BCH_RECOVERY_PASS_journal_replay)
515 		return -BCH_ERR_erofs_norecovery;
516 
517 	if (c->opts.nochanges)
518 		return -BCH_ERR_erofs_nochanges;
519 
520 	return __bch2_fs_read_write(c, false);
521 }
522 
523 int bch2_fs_read_write_early(struct bch_fs *c)
524 {
525 	lockdep_assert_held(&c->state_lock);
526 
527 	return __bch2_fs_read_write(c, true);
528 }
529 
530 /* Filesystem startup/shutdown: */
531 
532 static void __bch2_fs_free(struct bch_fs *c)
533 {
534 	for (unsigned i = 0; i < BCH_TIME_STAT_NR; i++)
535 		bch2_time_stats_exit(&c->times[i]);
536 
537 	bch2_find_btree_nodes_exit(&c->found_btree_nodes);
538 	bch2_free_pending_node_rewrites(c);
539 	bch2_fs_allocator_background_exit(c);
540 	bch2_fs_sb_errors_exit(c);
541 	bch2_fs_counters_exit(c);
542 	bch2_fs_snapshots_exit(c);
543 	bch2_fs_quota_exit(c);
544 	bch2_fs_fs_io_direct_exit(c);
545 	bch2_fs_fs_io_buffered_exit(c);
546 	bch2_fs_fsio_exit(c);
547 	bch2_fs_ec_exit(c);
548 	bch2_fs_encryption_exit(c);
549 	bch2_fs_nocow_locking_exit(c);
550 	bch2_fs_io_write_exit(c);
551 	bch2_fs_io_read_exit(c);
552 	bch2_fs_buckets_waiting_for_journal_exit(c);
553 	bch2_fs_btree_interior_update_exit(c);
554 	bch2_fs_btree_key_cache_exit(&c->btree_key_cache);
555 	bch2_fs_btree_cache_exit(c);
556 	bch2_fs_btree_iter_exit(c);
557 	bch2_fs_replicas_exit(c);
558 	bch2_fs_journal_exit(&c->journal);
559 	bch2_io_clock_exit(&c->io_clock[WRITE]);
560 	bch2_io_clock_exit(&c->io_clock[READ]);
561 	bch2_fs_compress_exit(c);
562 	bch2_journal_keys_put_initial(c);
563 	bch2_find_btree_nodes_exit(&c->found_btree_nodes);
564 	BUG_ON(atomic_read(&c->journal_keys.ref));
565 	bch2_fs_btree_write_buffer_exit(c);
566 	percpu_free_rwsem(&c->mark_lock);
567 	EBUG_ON(c->online_reserved && percpu_u64_get(c->online_reserved));
568 	free_percpu(c->online_reserved);
569 
570 	darray_exit(&c->btree_roots_extra);
571 	free_percpu(c->pcpu);
572 	mempool_exit(&c->large_bkey_pool);
573 	mempool_exit(&c->btree_bounce_pool);
574 	bioset_exit(&c->btree_bio);
575 	mempool_exit(&c->fill_iter);
576 #ifndef BCH_WRITE_REF_DEBUG
577 	percpu_ref_exit(&c->writes);
578 #endif
579 	kfree(rcu_dereference_protected(c->disk_groups, 1));
580 	kfree(c->journal_seq_blacklist_table);
581 	kfree(c->unused_inode_hints);
582 
583 	if (c->write_ref_wq)
584 		destroy_workqueue(c->write_ref_wq);
585 	if (c->btree_write_submit_wq)
586 		destroy_workqueue(c->btree_write_submit_wq);
587 	if (c->btree_read_complete_wq)
588 		destroy_workqueue(c->btree_read_complete_wq);
589 	if (c->copygc_wq)
590 		destroy_workqueue(c->copygc_wq);
591 	if (c->btree_io_complete_wq)
592 		destroy_workqueue(c->btree_io_complete_wq);
593 	if (c->btree_update_wq)
594 		destroy_workqueue(c->btree_update_wq);
595 
596 	bch2_free_super(&c->disk_sb);
597 	kvfree(c);
598 	module_put(THIS_MODULE);
599 }
600 
601 static void bch2_fs_release(struct kobject *kobj)
602 {
603 	struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
604 
605 	__bch2_fs_free(c);
606 }
607 
608 void __bch2_fs_stop(struct bch_fs *c)
609 {
610 	bch_verbose(c, "shutting down");
611 
612 	set_bit(BCH_FS_stopping, &c->flags);
613 
614 	down_write(&c->state_lock);
615 	bch2_fs_read_only(c);
616 	up_write(&c->state_lock);
617 
618 	for_each_member_device(c, ca)
619 		if (ca->kobj.state_in_sysfs &&
620 		    ca->disk_sb.bdev)
621 			sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
622 
623 	if (c->kobj.state_in_sysfs)
624 		kobject_del(&c->kobj);
625 
626 	bch2_fs_debug_exit(c);
627 	bch2_fs_chardev_exit(c);
628 
629 	bch2_ro_ref_put(c);
630 	wait_event(c->ro_ref_wait, !refcount_read(&c->ro_ref));
631 
632 	kobject_put(&c->counters_kobj);
633 	kobject_put(&c->time_stats);
634 	kobject_put(&c->opts_dir);
635 	kobject_put(&c->internal);
636 
637 	/* btree prefetch might have kicked off reads in the background: */
638 	bch2_btree_flush_all_reads(c);
639 
640 	for_each_member_device(c, ca)
641 		cancel_work_sync(&ca->io_error_work);
642 
643 	cancel_work_sync(&c->read_only_work);
644 }
645 
646 void bch2_fs_free(struct bch_fs *c)
647 {
648 	unsigned i;
649 
650 	mutex_lock(&bch_fs_list_lock);
651 	list_del(&c->list);
652 	mutex_unlock(&bch_fs_list_lock);
653 
654 	closure_sync(&c->cl);
655 	closure_debug_destroy(&c->cl);
656 
657 	for (i = 0; i < c->sb.nr_devices; i++) {
658 		struct bch_dev *ca = rcu_dereference_protected(c->devs[i], true);
659 
660 		if (ca) {
661 			EBUG_ON(atomic_long_read(&ca->ref) != 1);
662 			bch2_free_super(&ca->disk_sb);
663 			bch2_dev_free(ca);
664 		}
665 	}
666 
667 	bch_verbose(c, "shutdown complete");
668 
669 	kobject_put(&c->kobj);
670 }
671 
672 void bch2_fs_stop(struct bch_fs *c)
673 {
674 	__bch2_fs_stop(c);
675 	bch2_fs_free(c);
676 }
677 
678 static int bch2_fs_online(struct bch_fs *c)
679 {
680 	int ret = 0;
681 
682 	lockdep_assert_held(&bch_fs_list_lock);
683 
684 	if (__bch2_uuid_to_fs(c->sb.uuid)) {
685 		bch_err(c, "filesystem UUID already open");
686 		return -EINVAL;
687 	}
688 
689 	ret = bch2_fs_chardev_init(c);
690 	if (ret) {
691 		bch_err(c, "error creating character device");
692 		return ret;
693 	}
694 
695 	bch2_fs_debug_init(c);
696 
697 	ret = kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ?:
698 	    kobject_add(&c->internal, &c->kobj, "internal") ?:
699 	    kobject_add(&c->opts_dir, &c->kobj, "options") ?:
700 #ifndef CONFIG_BCACHEFS_NO_LATENCY_ACCT
701 	    kobject_add(&c->time_stats, &c->kobj, "time_stats") ?:
702 #endif
703 	    kobject_add(&c->counters_kobj, &c->kobj, "counters") ?:
704 	    bch2_opts_create_sysfs_files(&c->opts_dir);
705 	if (ret) {
706 		bch_err(c, "error creating sysfs objects");
707 		return ret;
708 	}
709 
710 	down_write(&c->state_lock);
711 
712 	for_each_member_device(c, ca) {
713 		ret = bch2_dev_sysfs_online(c, ca);
714 		if (ret) {
715 			bch_err(c, "error creating sysfs objects");
716 			bch2_dev_put(ca);
717 			goto err;
718 		}
719 	}
720 
721 	BUG_ON(!list_empty(&c->list));
722 	list_add(&c->list, &bch_fs_list);
723 err:
724 	up_write(&c->state_lock);
725 	return ret;
726 }
727 
728 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
729 {
730 	struct bch_fs *c;
731 	struct printbuf name = PRINTBUF;
732 	unsigned i, iter_size;
733 	int ret = 0;
734 
735 	c = kvmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
736 	if (!c) {
737 		c = ERR_PTR(-BCH_ERR_ENOMEM_fs_alloc);
738 		goto out;
739 	}
740 
741 	c->stdio = (void *)(unsigned long) opts.stdio;
742 
743 	__module_get(THIS_MODULE);
744 
745 	closure_init(&c->cl, NULL);
746 
747 	c->kobj.kset = bcachefs_kset;
748 	kobject_init(&c->kobj, &bch2_fs_ktype);
749 	kobject_init(&c->internal, &bch2_fs_internal_ktype);
750 	kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
751 	kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
752 	kobject_init(&c->counters_kobj, &bch2_fs_counters_ktype);
753 
754 	c->minor		= -1;
755 	c->disk_sb.fs_sb	= true;
756 
757 	init_rwsem(&c->state_lock);
758 	mutex_init(&c->sb_lock);
759 	mutex_init(&c->replicas_gc_lock);
760 	mutex_init(&c->btree_root_lock);
761 	INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
762 
763 	refcount_set(&c->ro_ref, 1);
764 	init_waitqueue_head(&c->ro_ref_wait);
765 	sema_init(&c->online_fsck_mutex, 1);
766 
767 	init_rwsem(&c->gc_lock);
768 	mutex_init(&c->gc_gens_lock);
769 	atomic_set(&c->journal_keys.ref, 1);
770 	c->journal_keys.initial_ref_held = true;
771 
772 	for (i = 0; i < BCH_TIME_STAT_NR; i++)
773 		bch2_time_stats_init(&c->times[i]);
774 
775 	bch2_fs_gc_init(c);
776 	bch2_fs_copygc_init(c);
777 	bch2_fs_btree_key_cache_init_early(&c->btree_key_cache);
778 	bch2_fs_btree_iter_init_early(c);
779 	bch2_fs_btree_interior_update_init_early(c);
780 	bch2_fs_allocator_background_init(c);
781 	bch2_fs_allocator_foreground_init(c);
782 	bch2_fs_rebalance_init(c);
783 	bch2_fs_quota_init(c);
784 	bch2_fs_ec_init_early(c);
785 	bch2_fs_move_init(c);
786 	bch2_fs_sb_errors_init_early(c);
787 
788 	INIT_LIST_HEAD(&c->list);
789 
790 	mutex_init(&c->usage_scratch_lock);
791 
792 	mutex_init(&c->bio_bounce_pages_lock);
793 	mutex_init(&c->snapshot_table_lock);
794 	init_rwsem(&c->snapshot_create_lock);
795 
796 	spin_lock_init(&c->btree_write_error_lock);
797 
798 	INIT_LIST_HEAD(&c->journal_iters);
799 
800 	INIT_LIST_HEAD(&c->fsck_error_msgs);
801 	mutex_init(&c->fsck_error_msgs_lock);
802 
803 	seqcount_init(&c->usage_lock);
804 
805 	sema_init(&c->io_in_flight, 128);
806 
807 	INIT_LIST_HEAD(&c->vfs_inodes_list);
808 	mutex_init(&c->vfs_inodes_lock);
809 
810 	c->copy_gc_enabled		= 1;
811 	c->rebalance.enabled		= 1;
812 	c->promote_whole_extents	= true;
813 
814 	c->journal.flush_write_time	= &c->times[BCH_TIME_journal_flush_write];
815 	c->journal.noflush_write_time	= &c->times[BCH_TIME_journal_noflush_write];
816 	c->journal.flush_seq_time	= &c->times[BCH_TIME_journal_flush_seq];
817 
818 	bch2_fs_btree_cache_init_early(&c->btree_cache);
819 
820 	mutex_init(&c->sectors_available_lock);
821 
822 	ret = percpu_init_rwsem(&c->mark_lock);
823 	if (ret)
824 		goto err;
825 
826 	mutex_lock(&c->sb_lock);
827 	ret = bch2_sb_to_fs(c, sb);
828 	mutex_unlock(&c->sb_lock);
829 
830 	if (ret)
831 		goto err;
832 
833 	pr_uuid(&name, c->sb.user_uuid.b);
834 	ret = name.allocation_failure ? -BCH_ERR_ENOMEM_fs_name_alloc : 0;
835 	if (ret)
836 		goto err;
837 
838 	strscpy(c->name, name.buf, sizeof(c->name));
839 	printbuf_exit(&name);
840 
841 	/* Compat: */
842 	if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_inode_v2 &&
843 	    !BCH_SB_JOURNAL_FLUSH_DELAY(sb))
844 		SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
845 
846 	if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_inode_v2 &&
847 	    !BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
848 		SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 100);
849 
850 	c->opts = bch2_opts_default;
851 	ret = bch2_opts_from_sb(&c->opts, sb);
852 	if (ret)
853 		goto err;
854 
855 	bch2_opts_apply(&c->opts, opts);
856 
857 	c->btree_key_cache_btrees |= 1U << BTREE_ID_alloc;
858 	if (c->opts.inodes_use_key_cache)
859 		c->btree_key_cache_btrees |= 1U << BTREE_ID_inodes;
860 	c->btree_key_cache_btrees |= 1U << BTREE_ID_logged_ops;
861 
862 	c->block_bits		= ilog2(block_sectors(c));
863 	c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
864 
865 	if (bch2_fs_init_fault("fs_alloc")) {
866 		bch_err(c, "fs_alloc fault injected");
867 		ret = -EFAULT;
868 		goto err;
869 	}
870 
871 	iter_size = sizeof(struct sort_iter) +
872 		(btree_blocks(c) + 1) * 2 *
873 		sizeof(struct sort_iter_set);
874 
875 	c->inode_shard_bits = ilog2(roundup_pow_of_two(num_possible_cpus()));
876 
877 	if (!(c->btree_update_wq = alloc_workqueue("bcachefs",
878 				WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_UNBOUND, 512)) ||
879 	    !(c->btree_io_complete_wq = alloc_workqueue("bcachefs_btree_io",
880 				WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM, 1)) ||
881 	    !(c->copygc_wq = alloc_workqueue("bcachefs_copygc",
882 				WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
883 	    !(c->btree_read_complete_wq = alloc_workqueue("bcachefs_btree_read_complete",
884 				WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM, 512)) ||
885 	    !(c->btree_write_submit_wq = alloc_workqueue("bcachefs_btree_write_sumit",
886 				WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM, 1)) ||
887 	    !(c->write_ref_wq = alloc_workqueue("bcachefs_write_ref",
888 				WQ_FREEZABLE, 0)) ||
889 #ifndef BCH_WRITE_REF_DEBUG
890 	    percpu_ref_init(&c->writes, bch2_writes_disabled,
891 			    PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
892 #endif
893 	    mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
894 	    bioset_init(&c->btree_bio, 1,
895 			max(offsetof(struct btree_read_bio, bio),
896 			    offsetof(struct btree_write_bio, wbio.bio)),
897 			BIOSET_NEED_BVECS) ||
898 	    !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
899 	    !(c->online_reserved = alloc_percpu(u64)) ||
900 	    mempool_init_kvmalloc_pool(&c->btree_bounce_pool, 1,
901 				       c->opts.btree_node_size) ||
902 	    mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048) ||
903 	    !(c->unused_inode_hints = kcalloc(1U << c->inode_shard_bits,
904 					      sizeof(u64), GFP_KERNEL))) {
905 		ret = -BCH_ERR_ENOMEM_fs_other_alloc;
906 		goto err;
907 	}
908 
909 	ret = bch2_fs_counters_init(c) ?:
910 	    bch2_fs_sb_errors_init(c) ?:
911 	    bch2_io_clock_init(&c->io_clock[READ]) ?:
912 	    bch2_io_clock_init(&c->io_clock[WRITE]) ?:
913 	    bch2_fs_journal_init(&c->journal) ?:
914 	    bch2_fs_replicas_init(c) ?:
915 	    bch2_fs_btree_iter_init(c) ?:
916 	    bch2_fs_btree_cache_init(c) ?:
917 	    bch2_fs_btree_key_cache_init(&c->btree_key_cache) ?:
918 	    bch2_fs_btree_interior_update_init(c) ?:
919 	    bch2_fs_buckets_waiting_for_journal_init(c) ?:
920 	    bch2_fs_btree_write_buffer_init(c) ?:
921 	    bch2_fs_subvolumes_init(c) ?:
922 	    bch2_fs_io_read_init(c) ?:
923 	    bch2_fs_io_write_init(c) ?:
924 	    bch2_fs_nocow_locking_init(c) ?:
925 	    bch2_fs_encryption_init(c) ?:
926 	    bch2_fs_compress_init(c) ?:
927 	    bch2_fs_ec_init(c) ?:
928 	    bch2_fs_fsio_init(c) ?:
929 	    bch2_fs_fs_io_buffered_init(c) ?:
930 	    bch2_fs_fs_io_direct_init(c);
931 	if (ret)
932 		goto err;
933 
934 	for (i = 0; i < c->sb.nr_devices; i++) {
935 		if (!bch2_member_exists(c->disk_sb.sb, i))
936 			continue;
937 		ret = bch2_dev_alloc(c, i);
938 		if (ret)
939 			goto err;
940 	}
941 
942 	bch2_journal_entry_res_resize(&c->journal,
943 			&c->btree_root_journal_res,
944 			BTREE_ID_NR * (JSET_KEYS_U64s + BKEY_BTREE_PTR_U64s_MAX));
945 	bch2_dev_usage_journal_reserve(c);
946 	bch2_journal_entry_res_resize(&c->journal,
947 			&c->clock_journal_res,
948 			(sizeof(struct jset_entry_clock) / sizeof(u64)) * 2);
949 
950 	mutex_lock(&bch_fs_list_lock);
951 	ret = bch2_fs_online(c);
952 	mutex_unlock(&bch_fs_list_lock);
953 
954 	if (ret)
955 		goto err;
956 out:
957 	return c;
958 err:
959 	bch2_fs_free(c);
960 	c = ERR_PTR(ret);
961 	goto out;
962 }
963 
964 noinline_for_stack
965 static void print_mount_opts(struct bch_fs *c)
966 {
967 	enum bch_opt_id i;
968 	struct printbuf p = PRINTBUF;
969 	bool first = true;
970 
971 	prt_str(&p, "mounting version ");
972 	bch2_version_to_text(&p, c->sb.version);
973 
974 	if (c->opts.read_only) {
975 		prt_str(&p, " opts=");
976 		first = false;
977 		prt_printf(&p, "ro");
978 	}
979 
980 	for (i = 0; i < bch2_opts_nr; i++) {
981 		const struct bch_option *opt = &bch2_opt_table[i];
982 		u64 v = bch2_opt_get_by_id(&c->opts, i);
983 
984 		if (!(opt->flags & OPT_MOUNT))
985 			continue;
986 
987 		if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
988 			continue;
989 
990 		prt_str(&p, first ? " opts=" : ",");
991 		first = false;
992 		bch2_opt_to_text(&p, c, c->disk_sb.sb, opt, v, OPT_SHOW_MOUNT_STYLE);
993 	}
994 
995 	bch_info(c, "%s", p.buf);
996 	printbuf_exit(&p);
997 }
998 
999 int bch2_fs_start(struct bch_fs *c)
1000 {
1001 	time64_t now = ktime_get_real_seconds();
1002 	int ret;
1003 
1004 	print_mount_opts(c);
1005 
1006 	down_write(&c->state_lock);
1007 
1008 	BUG_ON(test_bit(BCH_FS_started, &c->flags));
1009 
1010 	mutex_lock(&c->sb_lock);
1011 
1012 	ret = bch2_sb_members_v2_init(c);
1013 	if (ret) {
1014 		mutex_unlock(&c->sb_lock);
1015 		goto err;
1016 	}
1017 
1018 	for_each_online_member(c, ca)
1019 		bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx)->last_mount = cpu_to_le64(now);
1020 
1021 	struct bch_sb_field_ext *ext =
1022 		bch2_sb_field_get_minsize(&c->disk_sb, ext, sizeof(*ext) / sizeof(u64));
1023 	mutex_unlock(&c->sb_lock);
1024 
1025 	if (!ext) {
1026 		bch_err(c, "insufficient space in superblock for sb_field_ext");
1027 		ret = -BCH_ERR_ENOSPC_sb;
1028 		goto err;
1029 	}
1030 
1031 	for_each_rw_member(c, ca)
1032 		bch2_dev_allocator_add(c, ca);
1033 	bch2_recalc_capacity(c);
1034 
1035 	ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
1036 		? bch2_fs_recovery(c)
1037 		: bch2_fs_initialize(c);
1038 	if (ret)
1039 		goto err;
1040 
1041 	ret = bch2_opts_check_may_set(c);
1042 	if (ret)
1043 		goto err;
1044 
1045 	if (bch2_fs_init_fault("fs_start")) {
1046 		bch_err(c, "fs_start fault injected");
1047 		ret = -EINVAL;
1048 		goto err;
1049 	}
1050 
1051 	set_bit(BCH_FS_started, &c->flags);
1052 
1053 	if (c->opts.read_only) {
1054 		bch2_fs_read_only(c);
1055 	} else {
1056 		ret = !test_bit(BCH_FS_rw, &c->flags)
1057 			? bch2_fs_read_write(c)
1058 			: bch2_fs_read_write_late(c);
1059 		if (ret)
1060 			goto err;
1061 	}
1062 
1063 	ret = 0;
1064 err:
1065 	if (ret)
1066 		bch_err_msg(c, ret, "starting filesystem");
1067 	else
1068 		bch_verbose(c, "done starting filesystem");
1069 	up_write(&c->state_lock);
1070 	return ret;
1071 }
1072 
1073 static int bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
1074 {
1075 	struct bch_member m = bch2_sb_member_get(sb, sb->dev_idx);
1076 
1077 	if (le16_to_cpu(sb->block_size) != block_sectors(c))
1078 		return -BCH_ERR_mismatched_block_size;
1079 
1080 	if (le16_to_cpu(m.bucket_size) <
1081 	    BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
1082 		return -BCH_ERR_bucket_size_too_small;
1083 
1084 	return 0;
1085 }
1086 
1087 static int bch2_dev_in_fs(struct bch_sb_handle *fs,
1088 			  struct bch_sb_handle *sb,
1089 			  struct bch_opts *opts)
1090 {
1091 	if (fs == sb)
1092 		return 0;
1093 
1094 	if (!uuid_equal(&fs->sb->uuid, &sb->sb->uuid))
1095 		return -BCH_ERR_device_not_a_member_of_filesystem;
1096 
1097 	if (!bch2_member_exists(fs->sb, sb->sb->dev_idx))
1098 		return -BCH_ERR_device_has_been_removed;
1099 
1100 	if (fs->sb->block_size != sb->sb->block_size)
1101 		return -BCH_ERR_mismatched_block_size;
1102 
1103 	if (le16_to_cpu(fs->sb->version) < bcachefs_metadata_version_member_seq ||
1104 	    le16_to_cpu(sb->sb->version) < bcachefs_metadata_version_member_seq)
1105 		return 0;
1106 
1107 	if (fs->sb->seq == sb->sb->seq &&
1108 	    fs->sb->write_time != sb->sb->write_time) {
1109 		struct printbuf buf = PRINTBUF;
1110 
1111 		prt_str(&buf, "Split brain detected between ");
1112 		prt_bdevname(&buf, sb->bdev);
1113 		prt_str(&buf, " and ");
1114 		prt_bdevname(&buf, fs->bdev);
1115 		prt_char(&buf, ':');
1116 		prt_newline(&buf);
1117 		prt_printf(&buf, "seq=%llu but write_time different, got", le64_to_cpu(sb->sb->seq));
1118 		prt_newline(&buf);
1119 
1120 		prt_bdevname(&buf, fs->bdev);
1121 		prt_char(&buf, ' ');
1122 		bch2_prt_datetime(&buf, le64_to_cpu(fs->sb->write_time));;
1123 		prt_newline(&buf);
1124 
1125 		prt_bdevname(&buf, sb->bdev);
1126 		prt_char(&buf, ' ');
1127 		bch2_prt_datetime(&buf, le64_to_cpu(sb->sb->write_time));;
1128 		prt_newline(&buf);
1129 
1130 		if (!opts->no_splitbrain_check)
1131 			prt_printf(&buf, "Not using older sb");
1132 
1133 		pr_err("%s", buf.buf);
1134 		printbuf_exit(&buf);
1135 
1136 		if (!opts->no_splitbrain_check)
1137 			return -BCH_ERR_device_splitbrain;
1138 	}
1139 
1140 	struct bch_member m = bch2_sb_member_get(fs->sb, sb->sb->dev_idx);
1141 	u64 seq_from_fs		= le64_to_cpu(m.seq);
1142 	u64 seq_from_member	= le64_to_cpu(sb->sb->seq);
1143 
1144 	if (seq_from_fs && seq_from_fs < seq_from_member) {
1145 		struct printbuf buf = PRINTBUF;
1146 
1147 		prt_str(&buf, "Split brain detected between ");
1148 		prt_bdevname(&buf, sb->bdev);
1149 		prt_str(&buf, " and ");
1150 		prt_bdevname(&buf, fs->bdev);
1151 		prt_char(&buf, ':');
1152 		prt_newline(&buf);
1153 
1154 		prt_bdevname(&buf, fs->bdev);
1155 		prt_str(&buf, " believes seq of ");
1156 		prt_bdevname(&buf, sb->bdev);
1157 		prt_printf(&buf, " to be %llu, but ", seq_from_fs);
1158 		prt_bdevname(&buf, sb->bdev);
1159 		prt_printf(&buf, " has %llu\n", seq_from_member);
1160 
1161 		if (!opts->no_splitbrain_check) {
1162 			prt_str(&buf, "Not using ");
1163 			prt_bdevname(&buf, sb->bdev);
1164 		}
1165 
1166 		pr_err("%s", buf.buf);
1167 		printbuf_exit(&buf);
1168 
1169 		if (!opts->no_splitbrain_check)
1170 			return -BCH_ERR_device_splitbrain;
1171 	}
1172 
1173 	return 0;
1174 }
1175 
1176 /* Device startup/shutdown: */
1177 
1178 static void bch2_dev_release(struct kobject *kobj)
1179 {
1180 	struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
1181 
1182 	kfree(ca);
1183 }
1184 
1185 static void bch2_dev_free(struct bch_dev *ca)
1186 {
1187 	cancel_work_sync(&ca->io_error_work);
1188 
1189 	if (ca->kobj.state_in_sysfs &&
1190 	    ca->disk_sb.bdev)
1191 		sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1192 
1193 	if (ca->kobj.state_in_sysfs)
1194 		kobject_del(&ca->kobj);
1195 
1196 	kfree(ca->buckets_nouse);
1197 	bch2_free_super(&ca->disk_sb);
1198 	bch2_dev_journal_exit(ca);
1199 
1200 	free_percpu(ca->io_done);
1201 	bch2_dev_buckets_free(ca);
1202 	free_page((unsigned long) ca->sb_read_scratch);
1203 
1204 	bch2_time_stats_quantiles_exit(&ca->io_latency[WRITE]);
1205 	bch2_time_stats_quantiles_exit(&ca->io_latency[READ]);
1206 
1207 	percpu_ref_exit(&ca->io_ref);
1208 #ifndef CONFIG_BCACHEFS_DEBUG
1209 	percpu_ref_exit(&ca->ref);
1210 #endif
1211 	kobject_put(&ca->kobj);
1212 }
1213 
1214 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
1215 {
1216 
1217 	lockdep_assert_held(&c->state_lock);
1218 
1219 	if (percpu_ref_is_zero(&ca->io_ref))
1220 		return;
1221 
1222 	__bch2_dev_read_only(c, ca);
1223 
1224 	reinit_completion(&ca->io_ref_completion);
1225 	percpu_ref_kill(&ca->io_ref);
1226 	wait_for_completion(&ca->io_ref_completion);
1227 
1228 	if (ca->kobj.state_in_sysfs) {
1229 		sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1230 		sysfs_remove_link(&ca->kobj, "block");
1231 	}
1232 
1233 	bch2_free_super(&ca->disk_sb);
1234 	bch2_dev_journal_exit(ca);
1235 }
1236 
1237 #ifndef CONFIG_BCACHEFS_DEBUG
1238 static void bch2_dev_ref_complete(struct percpu_ref *ref)
1239 {
1240 	struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
1241 
1242 	complete(&ca->ref_completion);
1243 }
1244 #endif
1245 
1246 static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
1247 {
1248 	struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
1249 
1250 	complete(&ca->io_ref_completion);
1251 }
1252 
1253 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
1254 {
1255 	int ret;
1256 
1257 	if (!c->kobj.state_in_sysfs)
1258 		return 0;
1259 
1260 	if (!ca->kobj.state_in_sysfs) {
1261 		ret = kobject_add(&ca->kobj, &c->kobj,
1262 				  "dev-%u", ca->dev_idx);
1263 		if (ret)
1264 			return ret;
1265 	}
1266 
1267 	if (ca->disk_sb.bdev) {
1268 		struct kobject *block = bdev_kobj(ca->disk_sb.bdev);
1269 
1270 		ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
1271 		if (ret)
1272 			return ret;
1273 
1274 		ret = sysfs_create_link(&ca->kobj, block, "block");
1275 		if (ret)
1276 			return ret;
1277 	}
1278 
1279 	return 0;
1280 }
1281 
1282 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
1283 					struct bch_member *member)
1284 {
1285 	struct bch_dev *ca;
1286 	unsigned i;
1287 
1288 	ca = kzalloc(sizeof(*ca), GFP_KERNEL);
1289 	if (!ca)
1290 		return NULL;
1291 
1292 	kobject_init(&ca->kobj, &bch2_dev_ktype);
1293 	init_completion(&ca->ref_completion);
1294 	init_completion(&ca->io_ref_completion);
1295 
1296 	init_rwsem(&ca->bucket_lock);
1297 
1298 	INIT_WORK(&ca->io_error_work, bch2_io_error_work);
1299 
1300 	bch2_time_stats_quantiles_init(&ca->io_latency[READ]);
1301 	bch2_time_stats_quantiles_init(&ca->io_latency[WRITE]);
1302 
1303 	ca->mi = bch2_mi_to_cpu(member);
1304 
1305 	for (i = 0; i < ARRAY_SIZE(member->errors); i++)
1306 		atomic64_set(&ca->errors[i], le64_to_cpu(member->errors[i]));
1307 
1308 	ca->uuid = member->uuid;
1309 
1310 	ca->nr_btree_reserve = DIV_ROUND_UP(BTREE_NODE_RESERVE,
1311 			     ca->mi.bucket_size / btree_sectors(c));
1312 
1313 #ifndef CONFIG_BCACHEFS_DEBUG
1314 	if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete, 0, GFP_KERNEL))
1315 		goto err;
1316 #else
1317 	atomic_long_set(&ca->ref, 1);
1318 #endif
1319 
1320 	if (percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
1321 			    PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
1322 	    !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) ||
1323 	    bch2_dev_buckets_alloc(c, ca) ||
1324 	    !(ca->io_done	= alloc_percpu(*ca->io_done)))
1325 		goto err;
1326 
1327 	return ca;
1328 err:
1329 	bch2_dev_free(ca);
1330 	return NULL;
1331 }
1332 
1333 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
1334 			    unsigned dev_idx)
1335 {
1336 	ca->dev_idx = dev_idx;
1337 	__set_bit(ca->dev_idx, ca->self.d);
1338 	scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
1339 
1340 	ca->fs = c;
1341 	rcu_assign_pointer(c->devs[ca->dev_idx], ca);
1342 
1343 	if (bch2_dev_sysfs_online(c, ca))
1344 		pr_warn("error creating sysfs objects");
1345 }
1346 
1347 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
1348 {
1349 	struct bch_member member = bch2_sb_member_get(c->disk_sb.sb, dev_idx);
1350 	struct bch_dev *ca = NULL;
1351 	int ret = 0;
1352 
1353 	if (bch2_fs_init_fault("dev_alloc"))
1354 		goto err;
1355 
1356 	ca = __bch2_dev_alloc(c, &member);
1357 	if (!ca)
1358 		goto err;
1359 
1360 	ca->fs = c;
1361 
1362 	bch2_dev_attach(c, ca, dev_idx);
1363 	return ret;
1364 err:
1365 	if (ca)
1366 		bch2_dev_free(ca);
1367 	return -BCH_ERR_ENOMEM_dev_alloc;
1368 }
1369 
1370 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1371 {
1372 	unsigned ret;
1373 
1374 	if (bch2_dev_is_online(ca)) {
1375 		bch_err(ca, "already have device online in slot %u",
1376 			sb->sb->dev_idx);
1377 		return -BCH_ERR_device_already_online;
1378 	}
1379 
1380 	if (get_capacity(sb->bdev->bd_disk) <
1381 	    ca->mi.bucket_size * ca->mi.nbuckets) {
1382 		bch_err(ca, "cannot online: device too small");
1383 		return -BCH_ERR_device_size_too_small;
1384 	}
1385 
1386 	BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1387 
1388 	ret = bch2_dev_journal_init(ca, sb->sb);
1389 	if (ret)
1390 		return ret;
1391 
1392 	/* Commit: */
1393 	ca->disk_sb = *sb;
1394 	memset(sb, 0, sizeof(*sb));
1395 
1396 	ca->dev = ca->disk_sb.bdev->bd_dev;
1397 
1398 	percpu_ref_reinit(&ca->io_ref);
1399 
1400 	return 0;
1401 }
1402 
1403 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1404 {
1405 	struct bch_dev *ca;
1406 	int ret;
1407 
1408 	lockdep_assert_held(&c->state_lock);
1409 
1410 	if (le64_to_cpu(sb->sb->seq) >
1411 	    le64_to_cpu(c->disk_sb.sb->seq))
1412 		bch2_sb_to_fs(c, sb->sb);
1413 
1414 	BUG_ON(!bch2_dev_exists(c, sb->sb->dev_idx));
1415 
1416 	ca = bch2_dev_locked(c, sb->sb->dev_idx);
1417 
1418 	ret = __bch2_dev_attach_bdev(ca, sb);
1419 	if (ret)
1420 		return ret;
1421 
1422 	bch2_dev_sysfs_online(c, ca);
1423 
1424 	struct printbuf name = PRINTBUF;
1425 	prt_bdevname(&name, ca->disk_sb.bdev);
1426 
1427 	if (c->sb.nr_devices == 1)
1428 		strscpy(c->name, name.buf, sizeof(c->name));
1429 	strscpy(ca->name, name.buf, sizeof(ca->name));
1430 
1431 	printbuf_exit(&name);
1432 
1433 	rebalance_wakeup(c);
1434 	return 0;
1435 }
1436 
1437 /* Device management: */
1438 
1439 /*
1440  * Note: this function is also used by the error paths - when a particular
1441  * device sees an error, we call it to determine whether we can just set the
1442  * device RO, or - if this function returns false - we'll set the whole
1443  * filesystem RO:
1444  *
1445  * XXX: maybe we should be more explicit about whether we're changing state
1446  * because we got an error or what have you?
1447  */
1448 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1449 			    enum bch_member_state new_state, int flags)
1450 {
1451 	struct bch_devs_mask new_online_devs;
1452 	int nr_rw = 0, required;
1453 
1454 	lockdep_assert_held(&c->state_lock);
1455 
1456 	switch (new_state) {
1457 	case BCH_MEMBER_STATE_rw:
1458 		return true;
1459 	case BCH_MEMBER_STATE_ro:
1460 		if (ca->mi.state != BCH_MEMBER_STATE_rw)
1461 			return true;
1462 
1463 		/* do we have enough devices to write to?  */
1464 		for_each_member_device(c, ca2)
1465 			if (ca2 != ca)
1466 				nr_rw += ca2->mi.state == BCH_MEMBER_STATE_rw;
1467 
1468 		required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1469 			       ? c->opts.metadata_replicas
1470 			       : metadata_replicas_required(c),
1471 			       !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1472 			       ? c->opts.data_replicas
1473 			       : data_replicas_required(c));
1474 
1475 		return nr_rw >= required;
1476 	case BCH_MEMBER_STATE_failed:
1477 	case BCH_MEMBER_STATE_spare:
1478 		if (ca->mi.state != BCH_MEMBER_STATE_rw &&
1479 		    ca->mi.state != BCH_MEMBER_STATE_ro)
1480 			return true;
1481 
1482 		/* do we have enough devices to read from?  */
1483 		new_online_devs = bch2_online_devs(c);
1484 		__clear_bit(ca->dev_idx, new_online_devs.d);
1485 
1486 		return bch2_have_enough_devs(c, new_online_devs, flags, false);
1487 	default:
1488 		BUG();
1489 	}
1490 }
1491 
1492 static bool bch2_fs_may_start(struct bch_fs *c)
1493 {
1494 	struct bch_dev *ca;
1495 	unsigned i, flags = 0;
1496 
1497 	if (c->opts.very_degraded)
1498 		flags |= BCH_FORCE_IF_DEGRADED|BCH_FORCE_IF_LOST;
1499 
1500 	if (c->opts.degraded)
1501 		flags |= BCH_FORCE_IF_DEGRADED;
1502 
1503 	if (!c->opts.degraded &&
1504 	    !c->opts.very_degraded) {
1505 		mutex_lock(&c->sb_lock);
1506 
1507 		for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1508 			if (!bch2_member_exists(c->disk_sb.sb, i))
1509 				continue;
1510 
1511 			ca = bch2_dev_locked(c, i);
1512 
1513 			if (!bch2_dev_is_online(ca) &&
1514 			    (ca->mi.state == BCH_MEMBER_STATE_rw ||
1515 			     ca->mi.state == BCH_MEMBER_STATE_ro)) {
1516 				mutex_unlock(&c->sb_lock);
1517 				return false;
1518 			}
1519 		}
1520 		mutex_unlock(&c->sb_lock);
1521 	}
1522 
1523 	return bch2_have_enough_devs(c, bch2_online_devs(c), flags, true);
1524 }
1525 
1526 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1527 {
1528 	/*
1529 	 * The allocator thread itself allocates btree nodes, so stop it first:
1530 	 */
1531 	bch2_dev_allocator_remove(c, ca);
1532 	bch2_dev_journal_stop(&c->journal, ca);
1533 }
1534 
1535 static void __bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1536 {
1537 	lockdep_assert_held(&c->state_lock);
1538 
1539 	BUG_ON(ca->mi.state != BCH_MEMBER_STATE_rw);
1540 
1541 	bch2_dev_allocator_add(c, ca);
1542 	bch2_recalc_capacity(c);
1543 }
1544 
1545 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1546 			 enum bch_member_state new_state, int flags)
1547 {
1548 	struct bch_member *m;
1549 	int ret = 0;
1550 
1551 	if (ca->mi.state == new_state)
1552 		return 0;
1553 
1554 	if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1555 		return -BCH_ERR_device_state_not_allowed;
1556 
1557 	if (new_state != BCH_MEMBER_STATE_rw)
1558 		__bch2_dev_read_only(c, ca);
1559 
1560 	bch_notice(ca, "%s", bch2_member_states[new_state]);
1561 
1562 	mutex_lock(&c->sb_lock);
1563 	m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
1564 	SET_BCH_MEMBER_STATE(m, new_state);
1565 	bch2_write_super(c);
1566 	mutex_unlock(&c->sb_lock);
1567 
1568 	if (new_state == BCH_MEMBER_STATE_rw)
1569 		__bch2_dev_read_write(c, ca);
1570 
1571 	rebalance_wakeup(c);
1572 
1573 	return ret;
1574 }
1575 
1576 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1577 		       enum bch_member_state new_state, int flags)
1578 {
1579 	int ret;
1580 
1581 	down_write(&c->state_lock);
1582 	ret = __bch2_dev_set_state(c, ca, new_state, flags);
1583 	up_write(&c->state_lock);
1584 
1585 	return ret;
1586 }
1587 
1588 /* Device add/removal: */
1589 
1590 static int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
1591 {
1592 	struct bpos start	= POS(ca->dev_idx, 0);
1593 	struct bpos end		= POS(ca->dev_idx, U64_MAX);
1594 	int ret;
1595 
1596 	/*
1597 	 * We clear the LRU and need_discard btrees first so that we don't race
1598 	 * with bch2_do_invalidates() and bch2_do_discards()
1599 	 */
1600 	ret =   bch2_btree_delete_range(c, BTREE_ID_lru, start, end,
1601 					BTREE_TRIGGER_norun, NULL) ?:
1602 		bch2_btree_delete_range(c, BTREE_ID_need_discard, start, end,
1603 					BTREE_TRIGGER_norun, NULL) ?:
1604 		bch2_btree_delete_range(c, BTREE_ID_freespace, start, end,
1605 					BTREE_TRIGGER_norun, NULL) ?:
1606 		bch2_btree_delete_range(c, BTREE_ID_backpointers, start, end,
1607 					BTREE_TRIGGER_norun, NULL) ?:
1608 		bch2_btree_delete_range(c, BTREE_ID_alloc, start, end,
1609 					BTREE_TRIGGER_norun, NULL) ?:
1610 		bch2_btree_delete_range(c, BTREE_ID_bucket_gens, start, end,
1611 					BTREE_TRIGGER_norun, NULL);
1612 	bch_err_msg(c, ret, "removing dev alloc info");
1613 	return ret;
1614 }
1615 
1616 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1617 {
1618 	struct bch_member *m;
1619 	unsigned dev_idx = ca->dev_idx, data;
1620 	int ret;
1621 
1622 	down_write(&c->state_lock);
1623 
1624 	/*
1625 	 * We consume a reference to ca->ref, regardless of whether we succeed
1626 	 * or fail:
1627 	 */
1628 	bch2_dev_put(ca);
1629 
1630 	if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1631 		bch_err(ca, "Cannot remove without losing data");
1632 		ret = -BCH_ERR_device_state_not_allowed;
1633 		goto err;
1634 	}
1635 
1636 	__bch2_dev_read_only(c, ca);
1637 
1638 	ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1639 	bch_err_msg(ca, ret, "bch2_dev_data_drop()");
1640 	if (ret)
1641 		goto err;
1642 
1643 	ret = bch2_dev_remove_alloc(c, ca);
1644 	bch_err_msg(ca, ret, "bch2_dev_remove_alloc()");
1645 	if (ret)
1646 		goto err;
1647 
1648 	ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1649 	bch_err_msg(ca, ret, "bch2_journal_flush_device_pins()");
1650 	if (ret)
1651 		goto err;
1652 
1653 	ret = bch2_journal_flush(&c->journal);
1654 	bch_err_msg(ca, ret, "bch2_journal_flush()");
1655 	if (ret)
1656 		goto err;
1657 
1658 	ret = bch2_replicas_gc2(c);
1659 	bch_err_msg(ca, ret, "bch2_replicas_gc2()");
1660 	if (ret)
1661 		goto err;
1662 
1663 	data = bch2_dev_has_data(c, ca);
1664 	if (data) {
1665 		struct printbuf data_has = PRINTBUF;
1666 
1667 		prt_bitflags(&data_has, __bch2_data_types, data);
1668 		bch_err(ca, "Remove failed, still has data (%s)", data_has.buf);
1669 		printbuf_exit(&data_has);
1670 		ret = -EBUSY;
1671 		goto err;
1672 	}
1673 
1674 	__bch2_dev_offline(c, ca);
1675 
1676 	mutex_lock(&c->sb_lock);
1677 	rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1678 	mutex_unlock(&c->sb_lock);
1679 
1680 #ifndef CONFIG_BCACHEFS_DEBUG
1681 	percpu_ref_kill(&ca->ref);
1682 #else
1683 	ca->dying = true;
1684 	bch2_dev_put(ca);
1685 #endif
1686 	wait_for_completion(&ca->ref_completion);
1687 
1688 	bch2_dev_free(ca);
1689 
1690 	/*
1691 	 * At this point the device object has been removed in-core, but the
1692 	 * on-disk journal might still refer to the device index via sb device
1693 	 * usage entries. Recovery fails if it sees usage information for an
1694 	 * invalid device. Flush journal pins to push the back of the journal
1695 	 * past now invalid device index references before we update the
1696 	 * superblock, but after the device object has been removed so any
1697 	 * further journal writes elide usage info for the device.
1698 	 */
1699 	bch2_journal_flush_all_pins(&c->journal);
1700 
1701 	/*
1702 	 * Free this device's slot in the bch_member array - all pointers to
1703 	 * this device must be gone:
1704 	 */
1705 	mutex_lock(&c->sb_lock);
1706 	m = bch2_members_v2_get_mut(c->disk_sb.sb, dev_idx);
1707 	memset(&m->uuid, 0, sizeof(m->uuid));
1708 
1709 	bch2_write_super(c);
1710 
1711 	mutex_unlock(&c->sb_lock);
1712 	up_write(&c->state_lock);
1713 
1714 	bch2_dev_usage_journal_reserve(c);
1715 	return 0;
1716 err:
1717 	if (ca->mi.state == BCH_MEMBER_STATE_rw &&
1718 	    !percpu_ref_is_zero(&ca->io_ref))
1719 		__bch2_dev_read_write(c, ca);
1720 	up_write(&c->state_lock);
1721 	return ret;
1722 }
1723 
1724 /* Add new device to running filesystem: */
1725 int bch2_dev_add(struct bch_fs *c, const char *path)
1726 {
1727 	struct bch_opts opts = bch2_opts_empty();
1728 	struct bch_sb_handle sb;
1729 	struct bch_dev *ca = NULL;
1730 	struct bch_sb_field_members_v2 *mi;
1731 	struct bch_member dev_mi;
1732 	unsigned dev_idx, nr_devices, u64s;
1733 	struct printbuf errbuf = PRINTBUF;
1734 	struct printbuf label = PRINTBUF;
1735 	int ret;
1736 
1737 	ret = bch2_read_super(path, &opts, &sb);
1738 	bch_err_msg(c, ret, "reading super");
1739 	if (ret)
1740 		goto err;
1741 
1742 	dev_mi = bch2_sb_member_get(sb.sb, sb.sb->dev_idx);
1743 
1744 	if (BCH_MEMBER_GROUP(&dev_mi)) {
1745 		bch2_disk_path_to_text_sb(&label, sb.sb, BCH_MEMBER_GROUP(&dev_mi) - 1);
1746 		if (label.allocation_failure) {
1747 			ret = -ENOMEM;
1748 			goto err;
1749 		}
1750 	}
1751 
1752 	ret = bch2_dev_may_add(sb.sb, c);
1753 	if (ret)
1754 		goto err;
1755 
1756 	ca = __bch2_dev_alloc(c, &dev_mi);
1757 	if (!ca) {
1758 		ret = -ENOMEM;
1759 		goto err;
1760 	}
1761 
1762 	bch2_dev_usage_init(ca);
1763 
1764 	ret = __bch2_dev_attach_bdev(ca, &sb);
1765 	if (ret)
1766 		goto err;
1767 
1768 	ret = bch2_dev_journal_alloc(ca);
1769 	bch_err_msg(c, ret, "allocating journal");
1770 	if (ret)
1771 		goto err;
1772 
1773 	down_write(&c->state_lock);
1774 	mutex_lock(&c->sb_lock);
1775 
1776 	ret = bch2_sb_from_fs(c, ca);
1777 	bch_err_msg(c, ret, "setting up new superblock");
1778 	if (ret)
1779 		goto err_unlock;
1780 
1781 	if (dynamic_fault("bcachefs:add:no_slot"))
1782 		goto no_slot;
1783 
1784 	if (c->sb.nr_devices < BCH_SB_MEMBERS_MAX) {
1785 		dev_idx = c->sb.nr_devices;
1786 		goto have_slot;
1787 	}
1788 
1789 	int best = -1;
1790 	u64 best_last_mount = 0;
1791 	for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++) {
1792 		struct bch_member m = bch2_sb_member_get(c->disk_sb.sb, dev_idx);
1793 		if (bch2_member_alive(&m))
1794 			continue;
1795 
1796 		u64 last_mount = le64_to_cpu(m.last_mount);
1797 		if (best < 0 || last_mount < best_last_mount) {
1798 			best = dev_idx;
1799 			best_last_mount = last_mount;
1800 		}
1801 	}
1802 	if (best >= 0) {
1803 		dev_idx = best;
1804 		goto have_slot;
1805 	}
1806 no_slot:
1807 	ret = -BCH_ERR_ENOSPC_sb_members;
1808 	bch_err_msg(c, ret, "setting up new superblock");
1809 	goto err_unlock;
1810 
1811 have_slot:
1812 	nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1813 
1814 	mi = bch2_sb_field_get(c->disk_sb.sb, members_v2);
1815 	u64s = DIV_ROUND_UP(sizeof(struct bch_sb_field_members_v2) +
1816 			    le16_to_cpu(mi->member_bytes) * nr_devices, sizeof(u64));
1817 
1818 	mi = bch2_sb_field_resize(&c->disk_sb, members_v2, u64s);
1819 	if (!mi) {
1820 		ret = -BCH_ERR_ENOSPC_sb_members;
1821 		bch_err_msg(c, ret, "setting up new superblock");
1822 		goto err_unlock;
1823 	}
1824 	struct bch_member *m = bch2_members_v2_get_mut(c->disk_sb.sb, dev_idx);
1825 
1826 	/* success: */
1827 
1828 	*m = dev_mi;
1829 	m->last_mount = cpu_to_le64(ktime_get_real_seconds());
1830 	c->disk_sb.sb->nr_devices	= nr_devices;
1831 
1832 	ca->disk_sb.sb->dev_idx	= dev_idx;
1833 	bch2_dev_attach(c, ca, dev_idx);
1834 
1835 	if (BCH_MEMBER_GROUP(&dev_mi)) {
1836 		ret = __bch2_dev_group_set(c, ca, label.buf);
1837 		bch_err_msg(c, ret, "creating new label");
1838 		if (ret)
1839 			goto err_unlock;
1840 	}
1841 
1842 	bch2_write_super(c);
1843 	mutex_unlock(&c->sb_lock);
1844 
1845 	bch2_dev_usage_journal_reserve(c);
1846 
1847 	ret = bch2_trans_mark_dev_sb(c, ca, BTREE_TRIGGER_transactional);
1848 	bch_err_msg(ca, ret, "marking new superblock");
1849 	if (ret)
1850 		goto err_late;
1851 
1852 	ret = bch2_fs_freespace_init(c);
1853 	bch_err_msg(ca, ret, "initializing free space");
1854 	if (ret)
1855 		goto err_late;
1856 
1857 	ca->new_fs_bucket_idx = 0;
1858 
1859 	if (ca->mi.state == BCH_MEMBER_STATE_rw)
1860 		__bch2_dev_read_write(c, ca);
1861 
1862 	up_write(&c->state_lock);
1863 	return 0;
1864 
1865 err_unlock:
1866 	mutex_unlock(&c->sb_lock);
1867 	up_write(&c->state_lock);
1868 err:
1869 	if (ca)
1870 		bch2_dev_free(ca);
1871 	bch2_free_super(&sb);
1872 	printbuf_exit(&label);
1873 	printbuf_exit(&errbuf);
1874 	bch_err_fn(c, ret);
1875 	return ret;
1876 err_late:
1877 	up_write(&c->state_lock);
1878 	ca = NULL;
1879 	goto err;
1880 }
1881 
1882 /* Hot add existing device to running filesystem: */
1883 int bch2_dev_online(struct bch_fs *c, const char *path)
1884 {
1885 	struct bch_opts opts = bch2_opts_empty();
1886 	struct bch_sb_handle sb = { NULL };
1887 	struct bch_dev *ca;
1888 	unsigned dev_idx;
1889 	int ret;
1890 
1891 	down_write(&c->state_lock);
1892 
1893 	ret = bch2_read_super(path, &opts, &sb);
1894 	if (ret) {
1895 		up_write(&c->state_lock);
1896 		return ret;
1897 	}
1898 
1899 	dev_idx = sb.sb->dev_idx;
1900 
1901 	ret = bch2_dev_in_fs(&c->disk_sb, &sb, &c->opts);
1902 	bch_err_msg(c, ret, "bringing %s online", path);
1903 	if (ret)
1904 		goto err;
1905 
1906 	ret = bch2_dev_attach_bdev(c, &sb);
1907 	if (ret)
1908 		goto err;
1909 
1910 	ca = bch2_dev_locked(c, dev_idx);
1911 
1912 	ret = bch2_trans_mark_dev_sb(c, ca, BTREE_TRIGGER_transactional);
1913 	bch_err_msg(c, ret, "bringing %s online: error from bch2_trans_mark_dev_sb", path);
1914 	if (ret)
1915 		goto err;
1916 
1917 	if (ca->mi.state == BCH_MEMBER_STATE_rw)
1918 		__bch2_dev_read_write(c, ca);
1919 
1920 	if (!ca->mi.freespace_initialized) {
1921 		ret = bch2_dev_freespace_init(c, ca, 0, ca->mi.nbuckets);
1922 		bch_err_msg(ca, ret, "initializing free space");
1923 		if (ret)
1924 			goto err;
1925 	}
1926 
1927 	if (!ca->journal.nr) {
1928 		ret = bch2_dev_journal_alloc(ca);
1929 		bch_err_msg(ca, ret, "allocating journal");
1930 		if (ret)
1931 			goto err;
1932 	}
1933 
1934 	mutex_lock(&c->sb_lock);
1935 	bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx)->last_mount =
1936 		cpu_to_le64(ktime_get_real_seconds());
1937 	bch2_write_super(c);
1938 	mutex_unlock(&c->sb_lock);
1939 
1940 	up_write(&c->state_lock);
1941 	return 0;
1942 err:
1943 	up_write(&c->state_lock);
1944 	bch2_free_super(&sb);
1945 	return ret;
1946 }
1947 
1948 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1949 {
1950 	down_write(&c->state_lock);
1951 
1952 	if (!bch2_dev_is_online(ca)) {
1953 		bch_err(ca, "Already offline");
1954 		up_write(&c->state_lock);
1955 		return 0;
1956 	}
1957 
1958 	if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1959 		bch_err(ca, "Cannot offline required disk");
1960 		up_write(&c->state_lock);
1961 		return -BCH_ERR_device_state_not_allowed;
1962 	}
1963 
1964 	__bch2_dev_offline(c, ca);
1965 
1966 	up_write(&c->state_lock);
1967 	return 0;
1968 }
1969 
1970 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1971 {
1972 	struct bch_member *m;
1973 	u64 old_nbuckets;
1974 	int ret = 0;
1975 
1976 	down_write(&c->state_lock);
1977 	old_nbuckets = ca->mi.nbuckets;
1978 
1979 	if (nbuckets < ca->mi.nbuckets) {
1980 		bch_err(ca, "Cannot shrink yet");
1981 		ret = -EINVAL;
1982 		goto err;
1983 	}
1984 
1985 	if (nbuckets > BCH_MEMBER_NBUCKETS_MAX) {
1986 		bch_err(ca, "New device size too big (%llu greater than max %u)",
1987 			nbuckets, BCH_MEMBER_NBUCKETS_MAX);
1988 		ret = -BCH_ERR_device_size_too_big;
1989 		goto err;
1990 	}
1991 
1992 	if (bch2_dev_is_online(ca) &&
1993 	    get_capacity(ca->disk_sb.bdev->bd_disk) <
1994 	    ca->mi.bucket_size * nbuckets) {
1995 		bch_err(ca, "New size larger than device");
1996 		ret = -BCH_ERR_device_size_too_small;
1997 		goto err;
1998 	}
1999 
2000 	ret = bch2_dev_buckets_resize(c, ca, nbuckets);
2001 	bch_err_msg(ca, ret, "resizing buckets");
2002 	if (ret)
2003 		goto err;
2004 
2005 	ret = bch2_trans_mark_dev_sb(c, ca, BTREE_TRIGGER_transactional);
2006 	if (ret)
2007 		goto err;
2008 
2009 	mutex_lock(&c->sb_lock);
2010 	m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
2011 	m->nbuckets = cpu_to_le64(nbuckets);
2012 
2013 	bch2_write_super(c);
2014 	mutex_unlock(&c->sb_lock);
2015 
2016 	if (ca->mi.freespace_initialized) {
2017 		ret = bch2_dev_freespace_init(c, ca, old_nbuckets, nbuckets);
2018 		if (ret)
2019 			goto err;
2020 
2021 		/*
2022 		 * XXX: this is all wrong transactionally - we'll be able to do
2023 		 * this correctly after the disk space accounting rewrite
2024 		 */
2025 		ca->usage_base->d[BCH_DATA_free].buckets += nbuckets - old_nbuckets;
2026 	}
2027 
2028 	bch2_recalc_capacity(c);
2029 err:
2030 	up_write(&c->state_lock);
2031 	return ret;
2032 }
2033 
2034 /* return with ref on ca->ref: */
2035 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *name)
2036 {
2037 	for_each_member_device(c, ca)
2038 		if (!strcmp(name, ca->name))
2039 			return ca;
2040 	return ERR_PTR(-BCH_ERR_ENOENT_dev_not_found);
2041 }
2042 
2043 /* Filesystem open: */
2044 
2045 static inline int sb_cmp(struct bch_sb *l, struct bch_sb *r)
2046 {
2047 	return  cmp_int(le64_to_cpu(l->seq), le64_to_cpu(r->seq)) ?:
2048 		cmp_int(le64_to_cpu(l->write_time), le64_to_cpu(r->write_time));
2049 }
2050 
2051 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
2052 			    struct bch_opts opts)
2053 {
2054 	DARRAY(struct bch_sb_handle) sbs = { 0 };
2055 	struct bch_fs *c = NULL;
2056 	struct bch_sb_handle *best = NULL;
2057 	struct printbuf errbuf = PRINTBUF;
2058 	int ret = 0;
2059 
2060 	if (!try_module_get(THIS_MODULE))
2061 		return ERR_PTR(-ENODEV);
2062 
2063 	if (!nr_devices) {
2064 		ret = -EINVAL;
2065 		goto err;
2066 	}
2067 
2068 	ret = darray_make_room(&sbs, nr_devices);
2069 	if (ret)
2070 		goto err;
2071 
2072 	for (unsigned i = 0; i < nr_devices; i++) {
2073 		struct bch_sb_handle sb = { NULL };
2074 
2075 		ret = bch2_read_super(devices[i], &opts, &sb);
2076 		if (ret)
2077 			goto err;
2078 
2079 		BUG_ON(darray_push(&sbs, sb));
2080 	}
2081 
2082 	if (opts.nochanges && !opts.read_only) {
2083 		ret = -BCH_ERR_erofs_nochanges;
2084 		goto err_print;
2085 	}
2086 
2087 	darray_for_each(sbs, sb)
2088 		if (!best || sb_cmp(sb->sb, best->sb) > 0)
2089 			best = sb;
2090 
2091 	darray_for_each_reverse(sbs, sb) {
2092 		ret = bch2_dev_in_fs(best, sb, &opts);
2093 
2094 		if (ret == -BCH_ERR_device_has_been_removed ||
2095 		    ret == -BCH_ERR_device_splitbrain) {
2096 			bch2_free_super(sb);
2097 			darray_remove_item(&sbs, sb);
2098 			best -= best > sb;
2099 			ret = 0;
2100 			continue;
2101 		}
2102 
2103 		if (ret)
2104 			goto err_print;
2105 	}
2106 
2107 	c = bch2_fs_alloc(best->sb, opts);
2108 	ret = PTR_ERR_OR_ZERO(c);
2109 	if (ret)
2110 		goto err;
2111 
2112 	down_write(&c->state_lock);
2113 	darray_for_each(sbs, sb) {
2114 		ret = bch2_dev_attach_bdev(c, sb);
2115 		if (ret) {
2116 			up_write(&c->state_lock);
2117 			goto err;
2118 		}
2119 	}
2120 	up_write(&c->state_lock);
2121 
2122 	if (!bch2_fs_may_start(c)) {
2123 		ret = -BCH_ERR_insufficient_devices_to_start;
2124 		goto err_print;
2125 	}
2126 
2127 	if (!c->opts.nostart) {
2128 		ret = bch2_fs_start(c);
2129 		if (ret)
2130 			goto err;
2131 	}
2132 out:
2133 	darray_for_each(sbs, sb)
2134 		bch2_free_super(sb);
2135 	darray_exit(&sbs);
2136 	printbuf_exit(&errbuf);
2137 	module_put(THIS_MODULE);
2138 	return c;
2139 err_print:
2140 	pr_err("bch_fs_open err opening %s: %s",
2141 	       devices[0], bch2_err_str(ret));
2142 err:
2143 	if (!IS_ERR_OR_NULL(c))
2144 		bch2_fs_stop(c);
2145 	c = ERR_PTR(ret);
2146 	goto out;
2147 }
2148 
2149 /* Global interfaces/init */
2150 
2151 static void bcachefs_exit(void)
2152 {
2153 	bch2_debug_exit();
2154 	bch2_vfs_exit();
2155 	bch2_chardev_exit();
2156 	bch2_btree_key_cache_exit();
2157 	if (bcachefs_kset)
2158 		kset_unregister(bcachefs_kset);
2159 }
2160 
2161 static int __init bcachefs_init(void)
2162 {
2163 	bch2_bkey_pack_test();
2164 
2165 	if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
2166 	    bch2_btree_key_cache_init() ||
2167 	    bch2_chardev_init() ||
2168 	    bch2_vfs_init() ||
2169 	    bch2_debug_init())
2170 		goto err;
2171 
2172 	return 0;
2173 err:
2174 	bcachefs_exit();
2175 	return -ENOMEM;
2176 }
2177 
2178 #define BCH_DEBUG_PARAM(name, description)			\
2179 	bool bch2_##name;					\
2180 	module_param_named(name, bch2_##name, bool, 0644);	\
2181 	MODULE_PARM_DESC(name, description);
2182 BCH_DEBUG_PARAMS()
2183 #undef BCH_DEBUG_PARAM
2184 
2185 __maybe_unused
2186 static unsigned bch2_metadata_version = bcachefs_metadata_version_current;
2187 module_param_named(version, bch2_metadata_version, uint, 0400);
2188 
2189 module_exit(bcachefs_exit);
2190 module_init(bcachefs_init);
2191