xref: /linux/fs/bcachefs/super.c (revision 3e7819886281e077e82006fe4804b0d6b0f5643b)
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_sb_errors_exit(c);
540 	bch2_fs_counters_exit(c);
541 	bch2_fs_snapshots_exit(c);
542 	bch2_fs_quota_exit(c);
543 	bch2_fs_fs_io_direct_exit(c);
544 	bch2_fs_fs_io_buffered_exit(c);
545 	bch2_fs_fsio_exit(c);
546 	bch2_fs_ec_exit(c);
547 	bch2_fs_encryption_exit(c);
548 	bch2_fs_nocow_locking_exit(c);
549 	bch2_fs_io_write_exit(c);
550 	bch2_fs_io_read_exit(c);
551 	bch2_fs_buckets_waiting_for_journal_exit(c);
552 	bch2_fs_btree_interior_update_exit(c);
553 	bch2_fs_btree_key_cache_exit(&c->btree_key_cache);
554 	bch2_fs_btree_cache_exit(c);
555 	bch2_fs_btree_iter_exit(c);
556 	bch2_fs_replicas_exit(c);
557 	bch2_fs_journal_exit(&c->journal);
558 	bch2_io_clock_exit(&c->io_clock[WRITE]);
559 	bch2_io_clock_exit(&c->io_clock[READ]);
560 	bch2_fs_compress_exit(c);
561 	bch2_journal_keys_put_initial(c);
562 	bch2_find_btree_nodes_exit(&c->found_btree_nodes);
563 	BUG_ON(atomic_read(&c->journal_keys.ref));
564 	bch2_fs_btree_write_buffer_exit(c);
565 	percpu_free_rwsem(&c->mark_lock);
566 	if (c->online_reserved) {
567 		u64 v = percpu_u64_get(c->online_reserved);
568 		WARN(v, "online_reserved not 0 at shutdown: %lli", v);
569 		free_percpu(c->online_reserved);
570 	}
571 
572 	darray_exit(&c->btree_roots_extra);
573 	free_percpu(c->pcpu);
574 	mempool_exit(&c->large_bkey_pool);
575 	mempool_exit(&c->btree_bounce_pool);
576 	bioset_exit(&c->btree_bio);
577 	mempool_exit(&c->fill_iter);
578 #ifndef BCH_WRITE_REF_DEBUG
579 	percpu_ref_exit(&c->writes);
580 #endif
581 	kfree(rcu_dereference_protected(c->disk_groups, 1));
582 	kfree(c->journal_seq_blacklist_table);
583 	kfree(c->unused_inode_hints);
584 
585 	if (c->write_ref_wq)
586 		destroy_workqueue(c->write_ref_wq);
587 	if (c->btree_write_submit_wq)
588 		destroy_workqueue(c->btree_write_submit_wq);
589 	if (c->btree_read_complete_wq)
590 		destroy_workqueue(c->btree_read_complete_wq);
591 	if (c->copygc_wq)
592 		destroy_workqueue(c->copygc_wq);
593 	if (c->btree_io_complete_wq)
594 		destroy_workqueue(c->btree_io_complete_wq);
595 	if (c->btree_update_wq)
596 		destroy_workqueue(c->btree_update_wq);
597 
598 	bch2_free_super(&c->disk_sb);
599 	kvfree(c);
600 	module_put(THIS_MODULE);
601 }
602 
603 static void bch2_fs_release(struct kobject *kobj)
604 {
605 	struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
606 
607 	__bch2_fs_free(c);
608 }
609 
610 void __bch2_fs_stop(struct bch_fs *c)
611 {
612 	bch_verbose(c, "shutting down");
613 
614 	set_bit(BCH_FS_stopping, &c->flags);
615 
616 	down_write(&c->state_lock);
617 	bch2_fs_read_only(c);
618 	up_write(&c->state_lock);
619 
620 	for_each_member_device(c, ca)
621 		if (ca->kobj.state_in_sysfs &&
622 		    ca->disk_sb.bdev)
623 			sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
624 
625 	if (c->kobj.state_in_sysfs)
626 		kobject_del(&c->kobj);
627 
628 	bch2_fs_debug_exit(c);
629 	bch2_fs_chardev_exit(c);
630 
631 	bch2_ro_ref_put(c);
632 	wait_event(c->ro_ref_wait, !refcount_read(&c->ro_ref));
633 
634 	kobject_put(&c->counters_kobj);
635 	kobject_put(&c->time_stats);
636 	kobject_put(&c->opts_dir);
637 	kobject_put(&c->internal);
638 
639 	/* btree prefetch might have kicked off reads in the background: */
640 	bch2_btree_flush_all_reads(c);
641 
642 	for_each_member_device(c, ca)
643 		cancel_work_sync(&ca->io_error_work);
644 
645 	cancel_work_sync(&c->read_only_work);
646 }
647 
648 void bch2_fs_free(struct bch_fs *c)
649 {
650 	unsigned i;
651 
652 	mutex_lock(&bch_fs_list_lock);
653 	list_del(&c->list);
654 	mutex_unlock(&bch_fs_list_lock);
655 
656 	closure_sync(&c->cl);
657 	closure_debug_destroy(&c->cl);
658 
659 	for (i = 0; i < c->sb.nr_devices; i++) {
660 		struct bch_dev *ca = rcu_dereference_protected(c->devs[i], true);
661 
662 		if (ca) {
663 			EBUG_ON(atomic_long_read(&ca->ref) != 1);
664 			bch2_free_super(&ca->disk_sb);
665 			bch2_dev_free(ca);
666 		}
667 	}
668 
669 	bch_verbose(c, "shutdown complete");
670 
671 	kobject_put(&c->kobj);
672 }
673 
674 void bch2_fs_stop(struct bch_fs *c)
675 {
676 	__bch2_fs_stop(c);
677 	bch2_fs_free(c);
678 }
679 
680 static int bch2_fs_online(struct bch_fs *c)
681 {
682 	int ret = 0;
683 
684 	lockdep_assert_held(&bch_fs_list_lock);
685 
686 	if (__bch2_uuid_to_fs(c->sb.uuid)) {
687 		bch_err(c, "filesystem UUID already open");
688 		return -EINVAL;
689 	}
690 
691 	ret = bch2_fs_chardev_init(c);
692 	if (ret) {
693 		bch_err(c, "error creating character device");
694 		return ret;
695 	}
696 
697 	bch2_fs_debug_init(c);
698 
699 	ret = kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ?:
700 	    kobject_add(&c->internal, &c->kobj, "internal") ?:
701 	    kobject_add(&c->opts_dir, &c->kobj, "options") ?:
702 #ifndef CONFIG_BCACHEFS_NO_LATENCY_ACCT
703 	    kobject_add(&c->time_stats, &c->kobj, "time_stats") ?:
704 #endif
705 	    kobject_add(&c->counters_kobj, &c->kobj, "counters") ?:
706 	    bch2_opts_create_sysfs_files(&c->opts_dir);
707 	if (ret) {
708 		bch_err(c, "error creating sysfs objects");
709 		return ret;
710 	}
711 
712 	down_write(&c->state_lock);
713 
714 	for_each_member_device(c, ca) {
715 		ret = bch2_dev_sysfs_online(c, ca);
716 		if (ret) {
717 			bch_err(c, "error creating sysfs objects");
718 			bch2_dev_put(ca);
719 			goto err;
720 		}
721 	}
722 
723 	BUG_ON(!list_empty(&c->list));
724 	list_add(&c->list, &bch_fs_list);
725 err:
726 	up_write(&c->state_lock);
727 	return ret;
728 }
729 
730 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
731 {
732 	struct bch_fs *c;
733 	struct printbuf name = PRINTBUF;
734 	unsigned i, iter_size;
735 	int ret = 0;
736 
737 	c = kvmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
738 	if (!c) {
739 		c = ERR_PTR(-BCH_ERR_ENOMEM_fs_alloc);
740 		goto out;
741 	}
742 
743 	c->stdio = (void *)(unsigned long) opts.stdio;
744 
745 	__module_get(THIS_MODULE);
746 
747 	closure_init(&c->cl, NULL);
748 
749 	c->kobj.kset = bcachefs_kset;
750 	kobject_init(&c->kobj, &bch2_fs_ktype);
751 	kobject_init(&c->internal, &bch2_fs_internal_ktype);
752 	kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
753 	kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
754 	kobject_init(&c->counters_kobj, &bch2_fs_counters_ktype);
755 
756 	c->minor		= -1;
757 	c->disk_sb.fs_sb	= true;
758 
759 	init_rwsem(&c->state_lock);
760 	mutex_init(&c->sb_lock);
761 	mutex_init(&c->replicas_gc_lock);
762 	mutex_init(&c->btree_root_lock);
763 	INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
764 
765 	refcount_set(&c->ro_ref, 1);
766 	init_waitqueue_head(&c->ro_ref_wait);
767 	sema_init(&c->online_fsck_mutex, 1);
768 
769 	init_rwsem(&c->gc_lock);
770 	mutex_init(&c->gc_gens_lock);
771 	atomic_set(&c->journal_keys.ref, 1);
772 	c->journal_keys.initial_ref_held = true;
773 
774 	for (i = 0; i < BCH_TIME_STAT_NR; i++)
775 		bch2_time_stats_init(&c->times[i]);
776 
777 	bch2_fs_gc_init(c);
778 	bch2_fs_copygc_init(c);
779 	bch2_fs_btree_key_cache_init_early(&c->btree_key_cache);
780 	bch2_fs_btree_iter_init_early(c);
781 	bch2_fs_btree_interior_update_init_early(c);
782 	bch2_fs_allocator_background_init(c);
783 	bch2_fs_allocator_foreground_init(c);
784 	bch2_fs_rebalance_init(c);
785 	bch2_fs_quota_init(c);
786 	bch2_fs_ec_init_early(c);
787 	bch2_fs_move_init(c);
788 	bch2_fs_sb_errors_init_early(c);
789 
790 	INIT_LIST_HEAD(&c->list);
791 
792 	mutex_init(&c->usage_scratch_lock);
793 
794 	mutex_init(&c->bio_bounce_pages_lock);
795 	mutex_init(&c->snapshot_table_lock);
796 	init_rwsem(&c->snapshot_create_lock);
797 
798 	spin_lock_init(&c->btree_write_error_lock);
799 
800 	INIT_LIST_HEAD(&c->journal_iters);
801 
802 	INIT_LIST_HEAD(&c->fsck_error_msgs);
803 	mutex_init(&c->fsck_error_msgs_lock);
804 
805 	seqcount_init(&c->usage_lock);
806 
807 	sema_init(&c->io_in_flight, 128);
808 
809 	INIT_LIST_HEAD(&c->vfs_inodes_list);
810 	mutex_init(&c->vfs_inodes_lock);
811 
812 	c->copy_gc_enabled		= 1;
813 	c->rebalance.enabled		= 1;
814 	c->promote_whole_extents	= true;
815 
816 	c->journal.flush_write_time	= &c->times[BCH_TIME_journal_flush_write];
817 	c->journal.noflush_write_time	= &c->times[BCH_TIME_journal_noflush_write];
818 	c->journal.flush_seq_time	= &c->times[BCH_TIME_journal_flush_seq];
819 
820 	bch2_fs_btree_cache_init_early(&c->btree_cache);
821 
822 	mutex_init(&c->sectors_available_lock);
823 
824 	ret = percpu_init_rwsem(&c->mark_lock);
825 	if (ret)
826 		goto err;
827 
828 	mutex_lock(&c->sb_lock);
829 	ret = bch2_sb_to_fs(c, sb);
830 	mutex_unlock(&c->sb_lock);
831 
832 	if (ret)
833 		goto err;
834 
835 	pr_uuid(&name, c->sb.user_uuid.b);
836 	ret = name.allocation_failure ? -BCH_ERR_ENOMEM_fs_name_alloc : 0;
837 	if (ret)
838 		goto err;
839 
840 	strscpy(c->name, name.buf, sizeof(c->name));
841 	printbuf_exit(&name);
842 
843 	/* Compat: */
844 	if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_inode_v2 &&
845 	    !BCH_SB_JOURNAL_FLUSH_DELAY(sb))
846 		SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
847 
848 	if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_inode_v2 &&
849 	    !BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
850 		SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 100);
851 
852 	c->opts = bch2_opts_default;
853 	ret = bch2_opts_from_sb(&c->opts, sb);
854 	if (ret)
855 		goto err;
856 
857 	bch2_opts_apply(&c->opts, opts);
858 
859 	c->btree_key_cache_btrees |= 1U << BTREE_ID_alloc;
860 	if (c->opts.inodes_use_key_cache)
861 		c->btree_key_cache_btrees |= 1U << BTREE_ID_inodes;
862 	c->btree_key_cache_btrees |= 1U << BTREE_ID_logged_ops;
863 
864 	c->block_bits		= ilog2(block_sectors(c));
865 	c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
866 
867 	if (bch2_fs_init_fault("fs_alloc")) {
868 		bch_err(c, "fs_alloc fault injected");
869 		ret = -EFAULT;
870 		goto err;
871 	}
872 
873 	iter_size = sizeof(struct sort_iter) +
874 		(btree_blocks(c) + 1) * 2 *
875 		sizeof(struct sort_iter_set);
876 
877 	c->inode_shard_bits = ilog2(roundup_pow_of_two(num_possible_cpus()));
878 
879 	if (!(c->btree_update_wq = alloc_workqueue("bcachefs",
880 				WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_UNBOUND, 512)) ||
881 	    !(c->btree_io_complete_wq = alloc_workqueue("bcachefs_btree_io",
882 				WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM, 1)) ||
883 	    !(c->copygc_wq = alloc_workqueue("bcachefs_copygc",
884 				WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
885 	    !(c->btree_read_complete_wq = alloc_workqueue("bcachefs_btree_read_complete",
886 				WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM, 512)) ||
887 	    !(c->btree_write_submit_wq = alloc_workqueue("bcachefs_btree_write_sumit",
888 				WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM, 1)) ||
889 	    !(c->write_ref_wq = alloc_workqueue("bcachefs_write_ref",
890 				WQ_FREEZABLE, 0)) ||
891 #ifndef BCH_WRITE_REF_DEBUG
892 	    percpu_ref_init(&c->writes, bch2_writes_disabled,
893 			    PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
894 #endif
895 	    mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
896 	    bioset_init(&c->btree_bio, 1,
897 			max(offsetof(struct btree_read_bio, bio),
898 			    offsetof(struct btree_write_bio, wbio.bio)),
899 			BIOSET_NEED_BVECS) ||
900 	    !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
901 	    !(c->online_reserved = alloc_percpu(u64)) ||
902 	    mempool_init_kvmalloc_pool(&c->btree_bounce_pool, 1,
903 				       c->opts.btree_node_size) ||
904 	    mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048) ||
905 	    !(c->unused_inode_hints = kcalloc(1U << c->inode_shard_bits,
906 					      sizeof(u64), GFP_KERNEL))) {
907 		ret = -BCH_ERR_ENOMEM_fs_other_alloc;
908 		goto err;
909 	}
910 
911 	ret = bch2_fs_counters_init(c) ?:
912 	    bch2_fs_sb_errors_init(c) ?:
913 	    bch2_io_clock_init(&c->io_clock[READ]) ?:
914 	    bch2_io_clock_init(&c->io_clock[WRITE]) ?:
915 	    bch2_fs_journal_init(&c->journal) ?:
916 	    bch2_fs_replicas_init(c) ?:
917 	    bch2_fs_btree_iter_init(c) ?:
918 	    bch2_fs_btree_cache_init(c) ?:
919 	    bch2_fs_btree_key_cache_init(&c->btree_key_cache) ?:
920 	    bch2_fs_btree_interior_update_init(c) ?:
921 	    bch2_fs_buckets_waiting_for_journal_init(c) ?:
922 	    bch2_fs_btree_write_buffer_init(c) ?:
923 	    bch2_fs_subvolumes_init(c) ?:
924 	    bch2_fs_io_read_init(c) ?:
925 	    bch2_fs_io_write_init(c) ?:
926 	    bch2_fs_nocow_locking_init(c) ?:
927 	    bch2_fs_encryption_init(c) ?:
928 	    bch2_fs_compress_init(c) ?:
929 	    bch2_fs_ec_init(c) ?:
930 	    bch2_fs_fsio_init(c) ?:
931 	    bch2_fs_fs_io_buffered_init(c) ?:
932 	    bch2_fs_fs_io_direct_init(c);
933 	if (ret)
934 		goto err;
935 
936 	for (i = 0; i < c->sb.nr_devices; i++) {
937 		if (!bch2_member_exists(c->disk_sb.sb, i))
938 			continue;
939 		ret = bch2_dev_alloc(c, i);
940 		if (ret)
941 			goto err;
942 	}
943 
944 	bch2_journal_entry_res_resize(&c->journal,
945 			&c->btree_root_journal_res,
946 			BTREE_ID_NR * (JSET_KEYS_U64s + BKEY_BTREE_PTR_U64s_MAX));
947 	bch2_dev_usage_journal_reserve(c);
948 	bch2_journal_entry_res_resize(&c->journal,
949 			&c->clock_journal_res,
950 			(sizeof(struct jset_entry_clock) / sizeof(u64)) * 2);
951 
952 	mutex_lock(&bch_fs_list_lock);
953 	ret = bch2_fs_online(c);
954 	mutex_unlock(&bch_fs_list_lock);
955 
956 	if (ret)
957 		goto err;
958 out:
959 	return c;
960 err:
961 	bch2_fs_free(c);
962 	c = ERR_PTR(ret);
963 	goto out;
964 }
965 
966 noinline_for_stack
967 static void print_mount_opts(struct bch_fs *c)
968 {
969 	enum bch_opt_id i;
970 	struct printbuf p = PRINTBUF;
971 	bool first = true;
972 
973 	prt_str(&p, "mounting version ");
974 	bch2_version_to_text(&p, c->sb.version);
975 
976 	if (c->opts.read_only) {
977 		prt_str(&p, " opts=");
978 		first = false;
979 		prt_printf(&p, "ro");
980 	}
981 
982 	for (i = 0; i < bch2_opts_nr; i++) {
983 		const struct bch_option *opt = &bch2_opt_table[i];
984 		u64 v = bch2_opt_get_by_id(&c->opts, i);
985 
986 		if (!(opt->flags & OPT_MOUNT))
987 			continue;
988 
989 		if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
990 			continue;
991 
992 		prt_str(&p, first ? " opts=" : ",");
993 		first = false;
994 		bch2_opt_to_text(&p, c, c->disk_sb.sb, opt, v, OPT_SHOW_MOUNT_STYLE);
995 	}
996 
997 	bch_info(c, "%s", p.buf);
998 	printbuf_exit(&p);
999 }
1000 
1001 int bch2_fs_start(struct bch_fs *c)
1002 {
1003 	time64_t now = ktime_get_real_seconds();
1004 	int ret;
1005 
1006 	print_mount_opts(c);
1007 
1008 	down_write(&c->state_lock);
1009 
1010 	BUG_ON(test_bit(BCH_FS_started, &c->flags));
1011 
1012 	mutex_lock(&c->sb_lock);
1013 
1014 	ret = bch2_sb_members_v2_init(c);
1015 	if (ret) {
1016 		mutex_unlock(&c->sb_lock);
1017 		goto err;
1018 	}
1019 
1020 	for_each_online_member(c, ca)
1021 		bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx)->last_mount = cpu_to_le64(now);
1022 
1023 	struct bch_sb_field_ext *ext =
1024 		bch2_sb_field_get_minsize(&c->disk_sb, ext, sizeof(*ext) / sizeof(u64));
1025 	mutex_unlock(&c->sb_lock);
1026 
1027 	if (!ext) {
1028 		bch_err(c, "insufficient space in superblock for sb_field_ext");
1029 		ret = -BCH_ERR_ENOSPC_sb;
1030 		goto err;
1031 	}
1032 
1033 	for_each_rw_member(c, ca)
1034 		bch2_dev_allocator_add(c, ca);
1035 	bch2_recalc_capacity(c);
1036 
1037 	ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
1038 		? bch2_fs_recovery(c)
1039 		: bch2_fs_initialize(c);
1040 	if (ret)
1041 		goto err;
1042 
1043 	ret = bch2_opts_check_may_set(c);
1044 	if (ret)
1045 		goto err;
1046 
1047 	if (bch2_fs_init_fault("fs_start")) {
1048 		bch_err(c, "fs_start fault injected");
1049 		ret = -EINVAL;
1050 		goto err;
1051 	}
1052 
1053 	set_bit(BCH_FS_started, &c->flags);
1054 
1055 	if (c->opts.read_only) {
1056 		bch2_fs_read_only(c);
1057 	} else {
1058 		ret = !test_bit(BCH_FS_rw, &c->flags)
1059 			? bch2_fs_read_write(c)
1060 			: bch2_fs_read_write_late(c);
1061 		if (ret)
1062 			goto err;
1063 	}
1064 
1065 	ret = 0;
1066 err:
1067 	if (ret)
1068 		bch_err_msg(c, ret, "starting filesystem");
1069 	else
1070 		bch_verbose(c, "done starting filesystem");
1071 	up_write(&c->state_lock);
1072 	return ret;
1073 }
1074 
1075 static int bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
1076 {
1077 	struct bch_member m = bch2_sb_member_get(sb, sb->dev_idx);
1078 
1079 	if (le16_to_cpu(sb->block_size) != block_sectors(c))
1080 		return -BCH_ERR_mismatched_block_size;
1081 
1082 	if (le16_to_cpu(m.bucket_size) <
1083 	    BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
1084 		return -BCH_ERR_bucket_size_too_small;
1085 
1086 	return 0;
1087 }
1088 
1089 static int bch2_dev_in_fs(struct bch_sb_handle *fs,
1090 			  struct bch_sb_handle *sb,
1091 			  struct bch_opts *opts)
1092 {
1093 	if (fs == sb)
1094 		return 0;
1095 
1096 	if (!uuid_equal(&fs->sb->uuid, &sb->sb->uuid))
1097 		return -BCH_ERR_device_not_a_member_of_filesystem;
1098 
1099 	if (!bch2_member_exists(fs->sb, sb->sb->dev_idx))
1100 		return -BCH_ERR_device_has_been_removed;
1101 
1102 	if (fs->sb->block_size != sb->sb->block_size)
1103 		return -BCH_ERR_mismatched_block_size;
1104 
1105 	if (le16_to_cpu(fs->sb->version) < bcachefs_metadata_version_member_seq ||
1106 	    le16_to_cpu(sb->sb->version) < bcachefs_metadata_version_member_seq)
1107 		return 0;
1108 
1109 	if (fs->sb->seq == sb->sb->seq &&
1110 	    fs->sb->write_time != sb->sb->write_time) {
1111 		struct printbuf buf = PRINTBUF;
1112 
1113 		prt_str(&buf, "Split brain detected between ");
1114 		prt_bdevname(&buf, sb->bdev);
1115 		prt_str(&buf, " and ");
1116 		prt_bdevname(&buf, fs->bdev);
1117 		prt_char(&buf, ':');
1118 		prt_newline(&buf);
1119 		prt_printf(&buf, "seq=%llu but write_time different, got", le64_to_cpu(sb->sb->seq));
1120 		prt_newline(&buf);
1121 
1122 		prt_bdevname(&buf, fs->bdev);
1123 		prt_char(&buf, ' ');
1124 		bch2_prt_datetime(&buf, le64_to_cpu(fs->sb->write_time));;
1125 		prt_newline(&buf);
1126 
1127 		prt_bdevname(&buf, sb->bdev);
1128 		prt_char(&buf, ' ');
1129 		bch2_prt_datetime(&buf, le64_to_cpu(sb->sb->write_time));;
1130 		prt_newline(&buf);
1131 
1132 		if (!opts->no_splitbrain_check)
1133 			prt_printf(&buf, "Not using older sb");
1134 
1135 		pr_err("%s", buf.buf);
1136 		printbuf_exit(&buf);
1137 
1138 		if (!opts->no_splitbrain_check)
1139 			return -BCH_ERR_device_splitbrain;
1140 	}
1141 
1142 	struct bch_member m = bch2_sb_member_get(fs->sb, sb->sb->dev_idx);
1143 	u64 seq_from_fs		= le64_to_cpu(m.seq);
1144 	u64 seq_from_member	= le64_to_cpu(sb->sb->seq);
1145 
1146 	if (seq_from_fs && seq_from_fs < seq_from_member) {
1147 		struct printbuf buf = PRINTBUF;
1148 
1149 		prt_str(&buf, "Split brain detected between ");
1150 		prt_bdevname(&buf, sb->bdev);
1151 		prt_str(&buf, " and ");
1152 		prt_bdevname(&buf, fs->bdev);
1153 		prt_char(&buf, ':');
1154 		prt_newline(&buf);
1155 
1156 		prt_bdevname(&buf, fs->bdev);
1157 		prt_str(&buf, " believes seq of ");
1158 		prt_bdevname(&buf, sb->bdev);
1159 		prt_printf(&buf, " to be %llu, but ", seq_from_fs);
1160 		prt_bdevname(&buf, sb->bdev);
1161 		prt_printf(&buf, " has %llu\n", seq_from_member);
1162 
1163 		if (!opts->no_splitbrain_check) {
1164 			prt_str(&buf, "Not using ");
1165 			prt_bdevname(&buf, sb->bdev);
1166 		}
1167 
1168 		pr_err("%s", buf.buf);
1169 		printbuf_exit(&buf);
1170 
1171 		if (!opts->no_splitbrain_check)
1172 			return -BCH_ERR_device_splitbrain;
1173 	}
1174 
1175 	return 0;
1176 }
1177 
1178 /* Device startup/shutdown: */
1179 
1180 static void bch2_dev_release(struct kobject *kobj)
1181 {
1182 	struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
1183 
1184 	kfree(ca);
1185 }
1186 
1187 static void bch2_dev_free(struct bch_dev *ca)
1188 {
1189 	cancel_work_sync(&ca->io_error_work);
1190 
1191 	if (ca->kobj.state_in_sysfs &&
1192 	    ca->disk_sb.bdev)
1193 		sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1194 
1195 	if (ca->kobj.state_in_sysfs)
1196 		kobject_del(&ca->kobj);
1197 
1198 	kfree(ca->buckets_nouse);
1199 	bch2_free_super(&ca->disk_sb);
1200 	bch2_dev_allocator_background_exit(ca);
1201 	bch2_dev_journal_exit(ca);
1202 
1203 	free_percpu(ca->io_done);
1204 	bch2_dev_buckets_free(ca);
1205 	free_page((unsigned long) ca->sb_read_scratch);
1206 
1207 	bch2_time_stats_quantiles_exit(&ca->io_latency[WRITE]);
1208 	bch2_time_stats_quantiles_exit(&ca->io_latency[READ]);
1209 
1210 	percpu_ref_exit(&ca->io_ref);
1211 #ifndef CONFIG_BCACHEFS_DEBUG
1212 	percpu_ref_exit(&ca->ref);
1213 #endif
1214 	kobject_put(&ca->kobj);
1215 }
1216 
1217 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
1218 {
1219 
1220 	lockdep_assert_held(&c->state_lock);
1221 
1222 	if (percpu_ref_is_zero(&ca->io_ref))
1223 		return;
1224 
1225 	__bch2_dev_read_only(c, ca);
1226 
1227 	reinit_completion(&ca->io_ref_completion);
1228 	percpu_ref_kill(&ca->io_ref);
1229 	wait_for_completion(&ca->io_ref_completion);
1230 
1231 	if (ca->kobj.state_in_sysfs) {
1232 		sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1233 		sysfs_remove_link(&ca->kobj, "block");
1234 	}
1235 
1236 	bch2_free_super(&ca->disk_sb);
1237 	bch2_dev_journal_exit(ca);
1238 }
1239 
1240 #ifndef CONFIG_BCACHEFS_DEBUG
1241 static void bch2_dev_ref_complete(struct percpu_ref *ref)
1242 {
1243 	struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
1244 
1245 	complete(&ca->ref_completion);
1246 }
1247 #endif
1248 
1249 static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
1250 {
1251 	struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
1252 
1253 	complete(&ca->io_ref_completion);
1254 }
1255 
1256 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
1257 {
1258 	int ret;
1259 
1260 	if (!c->kobj.state_in_sysfs)
1261 		return 0;
1262 
1263 	if (!ca->kobj.state_in_sysfs) {
1264 		ret = kobject_add(&ca->kobj, &c->kobj,
1265 				  "dev-%u", ca->dev_idx);
1266 		if (ret)
1267 			return ret;
1268 	}
1269 
1270 	if (ca->disk_sb.bdev) {
1271 		struct kobject *block = bdev_kobj(ca->disk_sb.bdev);
1272 
1273 		ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
1274 		if (ret)
1275 			return ret;
1276 
1277 		ret = sysfs_create_link(&ca->kobj, block, "block");
1278 		if (ret)
1279 			return ret;
1280 	}
1281 
1282 	return 0;
1283 }
1284 
1285 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
1286 					struct bch_member *member)
1287 {
1288 	struct bch_dev *ca;
1289 	unsigned i;
1290 
1291 	ca = kzalloc(sizeof(*ca), GFP_KERNEL);
1292 	if (!ca)
1293 		return NULL;
1294 
1295 	kobject_init(&ca->kobj, &bch2_dev_ktype);
1296 	init_completion(&ca->ref_completion);
1297 	init_completion(&ca->io_ref_completion);
1298 
1299 	init_rwsem(&ca->bucket_lock);
1300 
1301 	INIT_WORK(&ca->io_error_work, bch2_io_error_work);
1302 
1303 	bch2_time_stats_quantiles_init(&ca->io_latency[READ]);
1304 	bch2_time_stats_quantiles_init(&ca->io_latency[WRITE]);
1305 
1306 	ca->mi = bch2_mi_to_cpu(member);
1307 
1308 	for (i = 0; i < ARRAY_SIZE(member->errors); i++)
1309 		atomic64_set(&ca->errors[i], le64_to_cpu(member->errors[i]));
1310 
1311 	ca->uuid = member->uuid;
1312 
1313 	ca->nr_btree_reserve = DIV_ROUND_UP(BTREE_NODE_RESERVE,
1314 			     ca->mi.bucket_size / btree_sectors(c));
1315 
1316 #ifndef CONFIG_BCACHEFS_DEBUG
1317 	if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete, 0, GFP_KERNEL))
1318 		goto err;
1319 #else
1320 	atomic_long_set(&ca->ref, 1);
1321 #endif
1322 
1323 	bch2_dev_allocator_background_init(ca);
1324 
1325 	if (percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
1326 			    PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
1327 	    !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) ||
1328 	    bch2_dev_buckets_alloc(c, ca) ||
1329 	    !(ca->io_done	= alloc_percpu(*ca->io_done)))
1330 		goto err;
1331 
1332 	return ca;
1333 err:
1334 	bch2_dev_free(ca);
1335 	return NULL;
1336 }
1337 
1338 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
1339 			    unsigned dev_idx)
1340 {
1341 	ca->dev_idx = dev_idx;
1342 	__set_bit(ca->dev_idx, ca->self.d);
1343 	scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
1344 
1345 	ca->fs = c;
1346 	rcu_assign_pointer(c->devs[ca->dev_idx], ca);
1347 
1348 	if (bch2_dev_sysfs_online(c, ca))
1349 		pr_warn("error creating sysfs objects");
1350 }
1351 
1352 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
1353 {
1354 	struct bch_member member = bch2_sb_member_get(c->disk_sb.sb, dev_idx);
1355 	struct bch_dev *ca = NULL;
1356 	int ret = 0;
1357 
1358 	if (bch2_fs_init_fault("dev_alloc"))
1359 		goto err;
1360 
1361 	ca = __bch2_dev_alloc(c, &member);
1362 	if (!ca)
1363 		goto err;
1364 
1365 	ca->fs = c;
1366 
1367 	bch2_dev_attach(c, ca, dev_idx);
1368 	return ret;
1369 err:
1370 	if (ca)
1371 		bch2_dev_free(ca);
1372 	return -BCH_ERR_ENOMEM_dev_alloc;
1373 }
1374 
1375 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1376 {
1377 	unsigned ret;
1378 
1379 	if (bch2_dev_is_online(ca)) {
1380 		bch_err(ca, "already have device online in slot %u",
1381 			sb->sb->dev_idx);
1382 		return -BCH_ERR_device_already_online;
1383 	}
1384 
1385 	if (get_capacity(sb->bdev->bd_disk) <
1386 	    ca->mi.bucket_size * ca->mi.nbuckets) {
1387 		bch_err(ca, "cannot online: device too small");
1388 		return -BCH_ERR_device_size_too_small;
1389 	}
1390 
1391 	BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1392 
1393 	ret = bch2_dev_journal_init(ca, sb->sb);
1394 	if (ret)
1395 		return ret;
1396 
1397 	/* Commit: */
1398 	ca->disk_sb = *sb;
1399 	memset(sb, 0, sizeof(*sb));
1400 
1401 	ca->dev = ca->disk_sb.bdev->bd_dev;
1402 
1403 	percpu_ref_reinit(&ca->io_ref);
1404 
1405 	return 0;
1406 }
1407 
1408 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1409 {
1410 	struct bch_dev *ca;
1411 	int ret;
1412 
1413 	lockdep_assert_held(&c->state_lock);
1414 
1415 	if (le64_to_cpu(sb->sb->seq) >
1416 	    le64_to_cpu(c->disk_sb.sb->seq))
1417 		bch2_sb_to_fs(c, sb->sb);
1418 
1419 	BUG_ON(!bch2_dev_exists(c, sb->sb->dev_idx));
1420 
1421 	ca = bch2_dev_locked(c, sb->sb->dev_idx);
1422 
1423 	ret = __bch2_dev_attach_bdev(ca, sb);
1424 	if (ret)
1425 		return ret;
1426 
1427 	bch2_dev_sysfs_online(c, ca);
1428 
1429 	struct printbuf name = PRINTBUF;
1430 	prt_bdevname(&name, ca->disk_sb.bdev);
1431 
1432 	if (c->sb.nr_devices == 1)
1433 		strscpy(c->name, name.buf, sizeof(c->name));
1434 	strscpy(ca->name, name.buf, sizeof(ca->name));
1435 
1436 	printbuf_exit(&name);
1437 
1438 	rebalance_wakeup(c);
1439 	return 0;
1440 }
1441 
1442 /* Device management: */
1443 
1444 /*
1445  * Note: this function is also used by the error paths - when a particular
1446  * device sees an error, we call it to determine whether we can just set the
1447  * device RO, or - if this function returns false - we'll set the whole
1448  * filesystem RO:
1449  *
1450  * XXX: maybe we should be more explicit about whether we're changing state
1451  * because we got an error or what have you?
1452  */
1453 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1454 			    enum bch_member_state new_state, int flags)
1455 {
1456 	struct bch_devs_mask new_online_devs;
1457 	int nr_rw = 0, required;
1458 
1459 	lockdep_assert_held(&c->state_lock);
1460 
1461 	switch (new_state) {
1462 	case BCH_MEMBER_STATE_rw:
1463 		return true;
1464 	case BCH_MEMBER_STATE_ro:
1465 		if (ca->mi.state != BCH_MEMBER_STATE_rw)
1466 			return true;
1467 
1468 		/* do we have enough devices to write to?  */
1469 		for_each_member_device(c, ca2)
1470 			if (ca2 != ca)
1471 				nr_rw += ca2->mi.state == BCH_MEMBER_STATE_rw;
1472 
1473 		required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1474 			       ? c->opts.metadata_replicas
1475 			       : metadata_replicas_required(c),
1476 			       !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1477 			       ? c->opts.data_replicas
1478 			       : data_replicas_required(c));
1479 
1480 		return nr_rw >= required;
1481 	case BCH_MEMBER_STATE_failed:
1482 	case BCH_MEMBER_STATE_spare:
1483 		if (ca->mi.state != BCH_MEMBER_STATE_rw &&
1484 		    ca->mi.state != BCH_MEMBER_STATE_ro)
1485 			return true;
1486 
1487 		/* do we have enough devices to read from?  */
1488 		new_online_devs = bch2_online_devs(c);
1489 		__clear_bit(ca->dev_idx, new_online_devs.d);
1490 
1491 		return bch2_have_enough_devs(c, new_online_devs, flags, false);
1492 	default:
1493 		BUG();
1494 	}
1495 }
1496 
1497 static bool bch2_fs_may_start(struct bch_fs *c)
1498 {
1499 	struct bch_dev *ca;
1500 	unsigned i, flags = 0;
1501 
1502 	if (c->opts.very_degraded)
1503 		flags |= BCH_FORCE_IF_DEGRADED|BCH_FORCE_IF_LOST;
1504 
1505 	if (c->opts.degraded)
1506 		flags |= BCH_FORCE_IF_DEGRADED;
1507 
1508 	if (!c->opts.degraded &&
1509 	    !c->opts.very_degraded) {
1510 		mutex_lock(&c->sb_lock);
1511 
1512 		for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1513 			if (!bch2_member_exists(c->disk_sb.sb, i))
1514 				continue;
1515 
1516 			ca = bch2_dev_locked(c, i);
1517 
1518 			if (!bch2_dev_is_online(ca) &&
1519 			    (ca->mi.state == BCH_MEMBER_STATE_rw ||
1520 			     ca->mi.state == BCH_MEMBER_STATE_ro)) {
1521 				mutex_unlock(&c->sb_lock);
1522 				return false;
1523 			}
1524 		}
1525 		mutex_unlock(&c->sb_lock);
1526 	}
1527 
1528 	return bch2_have_enough_devs(c, bch2_online_devs(c), flags, true);
1529 }
1530 
1531 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1532 {
1533 	/*
1534 	 * The allocator thread itself allocates btree nodes, so stop it first:
1535 	 */
1536 	bch2_dev_allocator_remove(c, ca);
1537 	bch2_recalc_capacity(c);
1538 	bch2_dev_journal_stop(&c->journal, ca);
1539 }
1540 
1541 static void __bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1542 {
1543 	lockdep_assert_held(&c->state_lock);
1544 
1545 	BUG_ON(ca->mi.state != BCH_MEMBER_STATE_rw);
1546 
1547 	bch2_dev_allocator_add(c, ca);
1548 	bch2_recalc_capacity(c);
1549 	bch2_dev_do_discards(ca);
1550 }
1551 
1552 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1553 			 enum bch_member_state new_state, int flags)
1554 {
1555 	struct bch_member *m;
1556 	int ret = 0;
1557 
1558 	if (ca->mi.state == new_state)
1559 		return 0;
1560 
1561 	if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1562 		return -BCH_ERR_device_state_not_allowed;
1563 
1564 	if (new_state != BCH_MEMBER_STATE_rw)
1565 		__bch2_dev_read_only(c, ca);
1566 
1567 	bch_notice(ca, "%s", bch2_member_states[new_state]);
1568 
1569 	mutex_lock(&c->sb_lock);
1570 	m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
1571 	SET_BCH_MEMBER_STATE(m, new_state);
1572 	bch2_write_super(c);
1573 	mutex_unlock(&c->sb_lock);
1574 
1575 	if (new_state == BCH_MEMBER_STATE_rw)
1576 		__bch2_dev_read_write(c, ca);
1577 
1578 	rebalance_wakeup(c);
1579 
1580 	return ret;
1581 }
1582 
1583 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1584 		       enum bch_member_state new_state, int flags)
1585 {
1586 	int ret;
1587 
1588 	down_write(&c->state_lock);
1589 	ret = __bch2_dev_set_state(c, ca, new_state, flags);
1590 	up_write(&c->state_lock);
1591 
1592 	return ret;
1593 }
1594 
1595 /* Device add/removal: */
1596 
1597 static int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
1598 {
1599 	struct bpos start	= POS(ca->dev_idx, 0);
1600 	struct bpos end		= POS(ca->dev_idx, U64_MAX);
1601 	int ret;
1602 
1603 	/*
1604 	 * We clear the LRU and need_discard btrees first so that we don't race
1605 	 * with bch2_do_invalidates() and bch2_do_discards()
1606 	 */
1607 	ret =   bch2_btree_delete_range(c, BTREE_ID_lru, start, end,
1608 					BTREE_TRIGGER_norun, NULL) ?:
1609 		bch2_btree_delete_range(c, BTREE_ID_need_discard, start, end,
1610 					BTREE_TRIGGER_norun, NULL) ?:
1611 		bch2_btree_delete_range(c, BTREE_ID_freespace, start, end,
1612 					BTREE_TRIGGER_norun, NULL) ?:
1613 		bch2_btree_delete_range(c, BTREE_ID_backpointers, start, end,
1614 					BTREE_TRIGGER_norun, NULL) ?:
1615 		bch2_btree_delete_range(c, BTREE_ID_alloc, start, end,
1616 					BTREE_TRIGGER_norun, NULL) ?:
1617 		bch2_btree_delete_range(c, BTREE_ID_bucket_gens, start, end,
1618 					BTREE_TRIGGER_norun, NULL);
1619 	bch_err_msg(c, ret, "removing dev alloc info");
1620 	return ret;
1621 }
1622 
1623 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1624 {
1625 	struct bch_member *m;
1626 	unsigned dev_idx = ca->dev_idx, data;
1627 	int ret;
1628 
1629 	down_write(&c->state_lock);
1630 
1631 	/*
1632 	 * We consume a reference to ca->ref, regardless of whether we succeed
1633 	 * or fail:
1634 	 */
1635 	bch2_dev_put(ca);
1636 
1637 	if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1638 		bch_err(ca, "Cannot remove without losing data");
1639 		ret = -BCH_ERR_device_state_not_allowed;
1640 		goto err;
1641 	}
1642 
1643 	__bch2_dev_read_only(c, ca);
1644 
1645 	ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1646 	bch_err_msg(ca, ret, "bch2_dev_data_drop()");
1647 	if (ret)
1648 		goto err;
1649 
1650 	ret = bch2_dev_remove_alloc(c, ca);
1651 	bch_err_msg(ca, ret, "bch2_dev_remove_alloc()");
1652 	if (ret)
1653 		goto err;
1654 
1655 	ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1656 	bch_err_msg(ca, ret, "bch2_journal_flush_device_pins()");
1657 	if (ret)
1658 		goto err;
1659 
1660 	ret = bch2_journal_flush(&c->journal);
1661 	bch_err_msg(ca, ret, "bch2_journal_flush()");
1662 	if (ret)
1663 		goto err;
1664 
1665 	ret = bch2_replicas_gc2(c);
1666 	bch_err_msg(ca, ret, "bch2_replicas_gc2()");
1667 	if (ret)
1668 		goto err;
1669 
1670 	data = bch2_dev_has_data(c, ca);
1671 	if (data) {
1672 		struct printbuf data_has = PRINTBUF;
1673 
1674 		prt_bitflags(&data_has, __bch2_data_types, data);
1675 		bch_err(ca, "Remove failed, still has data (%s)", data_has.buf);
1676 		printbuf_exit(&data_has);
1677 		ret = -EBUSY;
1678 		goto err;
1679 	}
1680 
1681 	__bch2_dev_offline(c, ca);
1682 
1683 	mutex_lock(&c->sb_lock);
1684 	rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1685 	mutex_unlock(&c->sb_lock);
1686 
1687 #ifndef CONFIG_BCACHEFS_DEBUG
1688 	percpu_ref_kill(&ca->ref);
1689 #else
1690 	ca->dying = true;
1691 	bch2_dev_put(ca);
1692 #endif
1693 	wait_for_completion(&ca->ref_completion);
1694 
1695 	bch2_dev_free(ca);
1696 
1697 	/*
1698 	 * At this point the device object has been removed in-core, but the
1699 	 * on-disk journal might still refer to the device index via sb device
1700 	 * usage entries. Recovery fails if it sees usage information for an
1701 	 * invalid device. Flush journal pins to push the back of the journal
1702 	 * past now invalid device index references before we update the
1703 	 * superblock, but after the device object has been removed so any
1704 	 * further journal writes elide usage info for the device.
1705 	 */
1706 	bch2_journal_flush_all_pins(&c->journal);
1707 
1708 	/*
1709 	 * Free this device's slot in the bch_member array - all pointers to
1710 	 * this device must be gone:
1711 	 */
1712 	mutex_lock(&c->sb_lock);
1713 	m = bch2_members_v2_get_mut(c->disk_sb.sb, dev_idx);
1714 	memset(&m->uuid, 0, sizeof(m->uuid));
1715 
1716 	bch2_write_super(c);
1717 
1718 	mutex_unlock(&c->sb_lock);
1719 	up_write(&c->state_lock);
1720 
1721 	bch2_dev_usage_journal_reserve(c);
1722 	return 0;
1723 err:
1724 	if (ca->mi.state == BCH_MEMBER_STATE_rw &&
1725 	    !percpu_ref_is_zero(&ca->io_ref))
1726 		__bch2_dev_read_write(c, ca);
1727 	up_write(&c->state_lock);
1728 	return ret;
1729 }
1730 
1731 /* Add new device to running filesystem: */
1732 int bch2_dev_add(struct bch_fs *c, const char *path)
1733 {
1734 	struct bch_opts opts = bch2_opts_empty();
1735 	struct bch_sb_handle sb;
1736 	struct bch_dev *ca = NULL;
1737 	struct bch_sb_field_members_v2 *mi;
1738 	struct bch_member dev_mi;
1739 	unsigned dev_idx, nr_devices, u64s;
1740 	struct printbuf errbuf = PRINTBUF;
1741 	struct printbuf label = PRINTBUF;
1742 	int ret;
1743 
1744 	ret = bch2_read_super(path, &opts, &sb);
1745 	bch_err_msg(c, ret, "reading super");
1746 	if (ret)
1747 		goto err;
1748 
1749 	dev_mi = bch2_sb_member_get(sb.sb, sb.sb->dev_idx);
1750 
1751 	if (BCH_MEMBER_GROUP(&dev_mi)) {
1752 		bch2_disk_path_to_text_sb(&label, sb.sb, BCH_MEMBER_GROUP(&dev_mi) - 1);
1753 		if (label.allocation_failure) {
1754 			ret = -ENOMEM;
1755 			goto err;
1756 		}
1757 	}
1758 
1759 	ret = bch2_dev_may_add(sb.sb, c);
1760 	if (ret)
1761 		goto err;
1762 
1763 	ca = __bch2_dev_alloc(c, &dev_mi);
1764 	if (!ca) {
1765 		ret = -ENOMEM;
1766 		goto err;
1767 	}
1768 
1769 	bch2_dev_usage_init(ca);
1770 
1771 	ret = __bch2_dev_attach_bdev(ca, &sb);
1772 	if (ret)
1773 		goto err;
1774 
1775 	ret = bch2_dev_journal_alloc(ca, true);
1776 	bch_err_msg(c, ret, "allocating journal");
1777 	if (ret)
1778 		goto err;
1779 
1780 	down_write(&c->state_lock);
1781 	mutex_lock(&c->sb_lock);
1782 
1783 	ret = bch2_sb_from_fs(c, ca);
1784 	bch_err_msg(c, ret, "setting up new superblock");
1785 	if (ret)
1786 		goto err_unlock;
1787 
1788 	if (dynamic_fault("bcachefs:add:no_slot"))
1789 		goto no_slot;
1790 
1791 	if (c->sb.nr_devices < BCH_SB_MEMBERS_MAX) {
1792 		dev_idx = c->sb.nr_devices;
1793 		goto have_slot;
1794 	}
1795 
1796 	int best = -1;
1797 	u64 best_last_mount = 0;
1798 	for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++) {
1799 		struct bch_member m = bch2_sb_member_get(c->disk_sb.sb, dev_idx);
1800 		if (bch2_member_alive(&m))
1801 			continue;
1802 
1803 		u64 last_mount = le64_to_cpu(m.last_mount);
1804 		if (best < 0 || last_mount < best_last_mount) {
1805 			best = dev_idx;
1806 			best_last_mount = last_mount;
1807 		}
1808 	}
1809 	if (best >= 0) {
1810 		dev_idx = best;
1811 		goto have_slot;
1812 	}
1813 no_slot:
1814 	ret = -BCH_ERR_ENOSPC_sb_members;
1815 	bch_err_msg(c, ret, "setting up new superblock");
1816 	goto err_unlock;
1817 
1818 have_slot:
1819 	nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1820 
1821 	mi = bch2_sb_field_get(c->disk_sb.sb, members_v2);
1822 	u64s = DIV_ROUND_UP(sizeof(struct bch_sb_field_members_v2) +
1823 			    le16_to_cpu(mi->member_bytes) * nr_devices, sizeof(u64));
1824 
1825 	mi = bch2_sb_field_resize(&c->disk_sb, members_v2, u64s);
1826 	if (!mi) {
1827 		ret = -BCH_ERR_ENOSPC_sb_members;
1828 		bch_err_msg(c, ret, "setting up new superblock");
1829 		goto err_unlock;
1830 	}
1831 	struct bch_member *m = bch2_members_v2_get_mut(c->disk_sb.sb, dev_idx);
1832 
1833 	/* success: */
1834 
1835 	*m = dev_mi;
1836 	m->last_mount = cpu_to_le64(ktime_get_real_seconds());
1837 	c->disk_sb.sb->nr_devices	= nr_devices;
1838 
1839 	ca->disk_sb.sb->dev_idx	= dev_idx;
1840 	bch2_dev_attach(c, ca, dev_idx);
1841 
1842 	if (BCH_MEMBER_GROUP(&dev_mi)) {
1843 		ret = __bch2_dev_group_set(c, ca, label.buf);
1844 		bch_err_msg(c, ret, "creating new label");
1845 		if (ret)
1846 			goto err_unlock;
1847 	}
1848 
1849 	bch2_write_super(c);
1850 	mutex_unlock(&c->sb_lock);
1851 
1852 	bch2_dev_usage_journal_reserve(c);
1853 
1854 	ret = bch2_trans_mark_dev_sb(c, ca, BTREE_TRIGGER_transactional);
1855 	bch_err_msg(ca, ret, "marking new superblock");
1856 	if (ret)
1857 		goto err_late;
1858 
1859 	ret = bch2_fs_freespace_init(c);
1860 	bch_err_msg(ca, ret, "initializing free space");
1861 	if (ret)
1862 		goto err_late;
1863 
1864 	ca->new_fs_bucket_idx = 0;
1865 
1866 	if (ca->mi.state == BCH_MEMBER_STATE_rw)
1867 		__bch2_dev_read_write(c, ca);
1868 
1869 	up_write(&c->state_lock);
1870 	return 0;
1871 
1872 err_unlock:
1873 	mutex_unlock(&c->sb_lock);
1874 	up_write(&c->state_lock);
1875 err:
1876 	if (ca)
1877 		bch2_dev_free(ca);
1878 	bch2_free_super(&sb);
1879 	printbuf_exit(&label);
1880 	printbuf_exit(&errbuf);
1881 	bch_err_fn(c, ret);
1882 	return ret;
1883 err_late:
1884 	up_write(&c->state_lock);
1885 	ca = NULL;
1886 	goto err;
1887 }
1888 
1889 /* Hot add existing device to running filesystem: */
1890 int bch2_dev_online(struct bch_fs *c, const char *path)
1891 {
1892 	struct bch_opts opts = bch2_opts_empty();
1893 	struct bch_sb_handle sb = { NULL };
1894 	struct bch_dev *ca;
1895 	unsigned dev_idx;
1896 	int ret;
1897 
1898 	down_write(&c->state_lock);
1899 
1900 	ret = bch2_read_super(path, &opts, &sb);
1901 	if (ret) {
1902 		up_write(&c->state_lock);
1903 		return ret;
1904 	}
1905 
1906 	dev_idx = sb.sb->dev_idx;
1907 
1908 	ret = bch2_dev_in_fs(&c->disk_sb, &sb, &c->opts);
1909 	bch_err_msg(c, ret, "bringing %s online", path);
1910 	if (ret)
1911 		goto err;
1912 
1913 	ret = bch2_dev_attach_bdev(c, &sb);
1914 	if (ret)
1915 		goto err;
1916 
1917 	ca = bch2_dev_locked(c, dev_idx);
1918 
1919 	ret = bch2_trans_mark_dev_sb(c, ca, BTREE_TRIGGER_transactional);
1920 	bch_err_msg(c, ret, "bringing %s online: error from bch2_trans_mark_dev_sb", path);
1921 	if (ret)
1922 		goto err;
1923 
1924 	if (ca->mi.state == BCH_MEMBER_STATE_rw)
1925 		__bch2_dev_read_write(c, ca);
1926 
1927 	if (!ca->mi.freespace_initialized) {
1928 		ret = bch2_dev_freespace_init(c, ca, 0, ca->mi.nbuckets);
1929 		bch_err_msg(ca, ret, "initializing free space");
1930 		if (ret)
1931 			goto err;
1932 	}
1933 
1934 	if (!ca->journal.nr) {
1935 		ret = bch2_dev_journal_alloc(ca, false);
1936 		bch_err_msg(ca, ret, "allocating journal");
1937 		if (ret)
1938 			goto err;
1939 	}
1940 
1941 	mutex_lock(&c->sb_lock);
1942 	bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx)->last_mount =
1943 		cpu_to_le64(ktime_get_real_seconds());
1944 	bch2_write_super(c);
1945 	mutex_unlock(&c->sb_lock);
1946 
1947 	up_write(&c->state_lock);
1948 	return 0;
1949 err:
1950 	up_write(&c->state_lock);
1951 	bch2_free_super(&sb);
1952 	return ret;
1953 }
1954 
1955 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1956 {
1957 	down_write(&c->state_lock);
1958 
1959 	if (!bch2_dev_is_online(ca)) {
1960 		bch_err(ca, "Already offline");
1961 		up_write(&c->state_lock);
1962 		return 0;
1963 	}
1964 
1965 	if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1966 		bch_err(ca, "Cannot offline required disk");
1967 		up_write(&c->state_lock);
1968 		return -BCH_ERR_device_state_not_allowed;
1969 	}
1970 
1971 	__bch2_dev_offline(c, ca);
1972 
1973 	up_write(&c->state_lock);
1974 	return 0;
1975 }
1976 
1977 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1978 {
1979 	struct bch_member *m;
1980 	u64 old_nbuckets;
1981 	int ret = 0;
1982 
1983 	down_write(&c->state_lock);
1984 	old_nbuckets = ca->mi.nbuckets;
1985 
1986 	if (nbuckets < ca->mi.nbuckets) {
1987 		bch_err(ca, "Cannot shrink yet");
1988 		ret = -EINVAL;
1989 		goto err;
1990 	}
1991 
1992 	if (nbuckets > BCH_MEMBER_NBUCKETS_MAX) {
1993 		bch_err(ca, "New device size too big (%llu greater than max %u)",
1994 			nbuckets, BCH_MEMBER_NBUCKETS_MAX);
1995 		ret = -BCH_ERR_device_size_too_big;
1996 		goto err;
1997 	}
1998 
1999 	if (bch2_dev_is_online(ca) &&
2000 	    get_capacity(ca->disk_sb.bdev->bd_disk) <
2001 	    ca->mi.bucket_size * nbuckets) {
2002 		bch_err(ca, "New size larger than device");
2003 		ret = -BCH_ERR_device_size_too_small;
2004 		goto err;
2005 	}
2006 
2007 	ret = bch2_dev_buckets_resize(c, ca, nbuckets);
2008 	bch_err_msg(ca, ret, "resizing buckets");
2009 	if (ret)
2010 		goto err;
2011 
2012 	ret = bch2_trans_mark_dev_sb(c, ca, BTREE_TRIGGER_transactional);
2013 	if (ret)
2014 		goto err;
2015 
2016 	mutex_lock(&c->sb_lock);
2017 	m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
2018 	m->nbuckets = cpu_to_le64(nbuckets);
2019 
2020 	bch2_write_super(c);
2021 	mutex_unlock(&c->sb_lock);
2022 
2023 	if (ca->mi.freespace_initialized) {
2024 		ret = bch2_dev_freespace_init(c, ca, old_nbuckets, nbuckets);
2025 		if (ret)
2026 			goto err;
2027 
2028 		/*
2029 		 * XXX: this is all wrong transactionally - we'll be able to do
2030 		 * this correctly after the disk space accounting rewrite
2031 		 */
2032 		ca->usage_base->d[BCH_DATA_free].buckets += nbuckets - old_nbuckets;
2033 	}
2034 
2035 	bch2_recalc_capacity(c);
2036 err:
2037 	up_write(&c->state_lock);
2038 	return ret;
2039 }
2040 
2041 /* return with ref on ca->ref: */
2042 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *name)
2043 {
2044 	for_each_member_device(c, ca)
2045 		if (!strcmp(name, ca->name))
2046 			return ca;
2047 	return ERR_PTR(-BCH_ERR_ENOENT_dev_not_found);
2048 }
2049 
2050 /* Filesystem open: */
2051 
2052 static inline int sb_cmp(struct bch_sb *l, struct bch_sb *r)
2053 {
2054 	return  cmp_int(le64_to_cpu(l->seq), le64_to_cpu(r->seq)) ?:
2055 		cmp_int(le64_to_cpu(l->write_time), le64_to_cpu(r->write_time));
2056 }
2057 
2058 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
2059 			    struct bch_opts opts)
2060 {
2061 	DARRAY(struct bch_sb_handle) sbs = { 0 };
2062 	struct bch_fs *c = NULL;
2063 	struct bch_sb_handle *best = NULL;
2064 	struct printbuf errbuf = PRINTBUF;
2065 	int ret = 0;
2066 
2067 	if (!try_module_get(THIS_MODULE))
2068 		return ERR_PTR(-ENODEV);
2069 
2070 	if (!nr_devices) {
2071 		ret = -EINVAL;
2072 		goto err;
2073 	}
2074 
2075 	ret = darray_make_room(&sbs, nr_devices);
2076 	if (ret)
2077 		goto err;
2078 
2079 	for (unsigned i = 0; i < nr_devices; i++) {
2080 		struct bch_sb_handle sb = { NULL };
2081 
2082 		ret = bch2_read_super(devices[i], &opts, &sb);
2083 		if (ret)
2084 			goto err;
2085 
2086 		BUG_ON(darray_push(&sbs, sb));
2087 	}
2088 
2089 	if (opts.nochanges && !opts.read_only) {
2090 		ret = -BCH_ERR_erofs_nochanges;
2091 		goto err_print;
2092 	}
2093 
2094 	darray_for_each(sbs, sb)
2095 		if (!best || sb_cmp(sb->sb, best->sb) > 0)
2096 			best = sb;
2097 
2098 	darray_for_each_reverse(sbs, sb) {
2099 		ret = bch2_dev_in_fs(best, sb, &opts);
2100 
2101 		if (ret == -BCH_ERR_device_has_been_removed ||
2102 		    ret == -BCH_ERR_device_splitbrain) {
2103 			bch2_free_super(sb);
2104 			darray_remove_item(&sbs, sb);
2105 			best -= best > sb;
2106 			ret = 0;
2107 			continue;
2108 		}
2109 
2110 		if (ret)
2111 			goto err_print;
2112 	}
2113 
2114 	c = bch2_fs_alloc(best->sb, opts);
2115 	ret = PTR_ERR_OR_ZERO(c);
2116 	if (ret)
2117 		goto err;
2118 
2119 	down_write(&c->state_lock);
2120 	darray_for_each(sbs, sb) {
2121 		ret = bch2_dev_attach_bdev(c, sb);
2122 		if (ret) {
2123 			up_write(&c->state_lock);
2124 			goto err;
2125 		}
2126 	}
2127 	up_write(&c->state_lock);
2128 
2129 	if (!bch2_fs_may_start(c)) {
2130 		ret = -BCH_ERR_insufficient_devices_to_start;
2131 		goto err_print;
2132 	}
2133 
2134 	if (!c->opts.nostart) {
2135 		ret = bch2_fs_start(c);
2136 		if (ret)
2137 			goto err;
2138 	}
2139 out:
2140 	darray_for_each(sbs, sb)
2141 		bch2_free_super(sb);
2142 	darray_exit(&sbs);
2143 	printbuf_exit(&errbuf);
2144 	module_put(THIS_MODULE);
2145 	return c;
2146 err_print:
2147 	pr_err("bch_fs_open err opening %s: %s",
2148 	       devices[0], bch2_err_str(ret));
2149 err:
2150 	if (!IS_ERR_OR_NULL(c))
2151 		bch2_fs_stop(c);
2152 	c = ERR_PTR(ret);
2153 	goto out;
2154 }
2155 
2156 /* Global interfaces/init */
2157 
2158 static void bcachefs_exit(void)
2159 {
2160 	bch2_debug_exit();
2161 	bch2_vfs_exit();
2162 	bch2_chardev_exit();
2163 	bch2_btree_key_cache_exit();
2164 	if (bcachefs_kset)
2165 		kset_unregister(bcachefs_kset);
2166 }
2167 
2168 static int __init bcachefs_init(void)
2169 {
2170 	bch2_bkey_pack_test();
2171 
2172 	if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
2173 	    bch2_btree_key_cache_init() ||
2174 	    bch2_chardev_init() ||
2175 	    bch2_vfs_init() ||
2176 	    bch2_debug_init())
2177 		goto err;
2178 
2179 	return 0;
2180 err:
2181 	bcachefs_exit();
2182 	return -ENOMEM;
2183 }
2184 
2185 #define BCH_DEBUG_PARAM(name, description)			\
2186 	bool bch2_##name;					\
2187 	module_param_named(name, bch2_##name, bool, 0644);	\
2188 	MODULE_PARM_DESC(name, description);
2189 BCH_DEBUG_PARAMS()
2190 #undef BCH_DEBUG_PARAM
2191 
2192 __maybe_unused
2193 static unsigned bch2_metadata_version = bcachefs_metadata_version_current;
2194 module_param_named(version, bch2_metadata_version, uint, 0400);
2195 
2196 module_exit(bcachefs_exit);
2197 module_init(bcachefs_init);
2198