xref: /linux/fs/bcachefs/super.c (revision 2845f512232de9e436b9e3b5529e906e62414013)
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 	EBUG_ON(c->online_reserved && percpu_u64_get(c->online_reserved));
567 	free_percpu(c->online_reserved);
568 
569 	darray_exit(&c->btree_roots_extra);
570 	free_percpu(c->pcpu);
571 	mempool_exit(&c->large_bkey_pool);
572 	mempool_exit(&c->btree_bounce_pool);
573 	bioset_exit(&c->btree_bio);
574 	mempool_exit(&c->fill_iter);
575 #ifndef BCH_WRITE_REF_DEBUG
576 	percpu_ref_exit(&c->writes);
577 #endif
578 	kfree(rcu_dereference_protected(c->disk_groups, 1));
579 	kfree(c->journal_seq_blacklist_table);
580 	kfree(c->unused_inode_hints);
581 
582 	if (c->write_ref_wq)
583 		destroy_workqueue(c->write_ref_wq);
584 	if (c->btree_write_submit_wq)
585 		destroy_workqueue(c->btree_write_submit_wq);
586 	if (c->btree_read_complete_wq)
587 		destroy_workqueue(c->btree_read_complete_wq);
588 	if (c->copygc_wq)
589 		destroy_workqueue(c->copygc_wq);
590 	if (c->btree_io_complete_wq)
591 		destroy_workqueue(c->btree_io_complete_wq);
592 	if (c->btree_update_wq)
593 		destroy_workqueue(c->btree_update_wq);
594 
595 	bch2_free_super(&c->disk_sb);
596 	kvfree(c);
597 	module_put(THIS_MODULE);
598 }
599 
600 static void bch2_fs_release(struct kobject *kobj)
601 {
602 	struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
603 
604 	__bch2_fs_free(c);
605 }
606 
607 void __bch2_fs_stop(struct bch_fs *c)
608 {
609 	bch_verbose(c, "shutting down");
610 
611 	set_bit(BCH_FS_stopping, &c->flags);
612 
613 	down_write(&c->state_lock);
614 	bch2_fs_read_only(c);
615 	up_write(&c->state_lock);
616 
617 	for_each_member_device(c, ca)
618 		if (ca->kobj.state_in_sysfs &&
619 		    ca->disk_sb.bdev)
620 			sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
621 
622 	if (c->kobj.state_in_sysfs)
623 		kobject_del(&c->kobj);
624 
625 	bch2_fs_debug_exit(c);
626 	bch2_fs_chardev_exit(c);
627 
628 	bch2_ro_ref_put(c);
629 	wait_event(c->ro_ref_wait, !refcount_read(&c->ro_ref));
630 
631 	kobject_put(&c->counters_kobj);
632 	kobject_put(&c->time_stats);
633 	kobject_put(&c->opts_dir);
634 	kobject_put(&c->internal);
635 
636 	/* btree prefetch might have kicked off reads in the background: */
637 	bch2_btree_flush_all_reads(c);
638 
639 	for_each_member_device(c, ca)
640 		cancel_work_sync(&ca->io_error_work);
641 
642 	cancel_work_sync(&c->read_only_work);
643 }
644 
645 void bch2_fs_free(struct bch_fs *c)
646 {
647 	unsigned i;
648 
649 	mutex_lock(&bch_fs_list_lock);
650 	list_del(&c->list);
651 	mutex_unlock(&bch_fs_list_lock);
652 
653 	closure_sync(&c->cl);
654 	closure_debug_destroy(&c->cl);
655 
656 	for (i = 0; i < c->sb.nr_devices; i++) {
657 		struct bch_dev *ca = rcu_dereference_protected(c->devs[i], true);
658 
659 		if (ca) {
660 			EBUG_ON(atomic_long_read(&ca->ref) != 1);
661 			bch2_free_super(&ca->disk_sb);
662 			bch2_dev_free(ca);
663 		}
664 	}
665 
666 	bch_verbose(c, "shutdown complete");
667 
668 	kobject_put(&c->kobj);
669 }
670 
671 void bch2_fs_stop(struct bch_fs *c)
672 {
673 	__bch2_fs_stop(c);
674 	bch2_fs_free(c);
675 }
676 
677 static int bch2_fs_online(struct bch_fs *c)
678 {
679 	int ret = 0;
680 
681 	lockdep_assert_held(&bch_fs_list_lock);
682 
683 	if (__bch2_uuid_to_fs(c->sb.uuid)) {
684 		bch_err(c, "filesystem UUID already open");
685 		return -EINVAL;
686 	}
687 
688 	ret = bch2_fs_chardev_init(c);
689 	if (ret) {
690 		bch_err(c, "error creating character device");
691 		return ret;
692 	}
693 
694 	bch2_fs_debug_init(c);
695 
696 	ret = kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ?:
697 	    kobject_add(&c->internal, &c->kobj, "internal") ?:
698 	    kobject_add(&c->opts_dir, &c->kobj, "options") ?:
699 #ifndef CONFIG_BCACHEFS_NO_LATENCY_ACCT
700 	    kobject_add(&c->time_stats, &c->kobj, "time_stats") ?:
701 #endif
702 	    kobject_add(&c->counters_kobj, &c->kobj, "counters") ?:
703 	    bch2_opts_create_sysfs_files(&c->opts_dir);
704 	if (ret) {
705 		bch_err(c, "error creating sysfs objects");
706 		return ret;
707 	}
708 
709 	down_write(&c->state_lock);
710 
711 	for_each_member_device(c, ca) {
712 		ret = bch2_dev_sysfs_online(c, ca);
713 		if (ret) {
714 			bch_err(c, "error creating sysfs objects");
715 			bch2_dev_put(ca);
716 			goto err;
717 		}
718 	}
719 
720 	BUG_ON(!list_empty(&c->list));
721 	list_add(&c->list, &bch_fs_list);
722 err:
723 	up_write(&c->state_lock);
724 	return ret;
725 }
726 
727 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
728 {
729 	struct bch_fs *c;
730 	struct printbuf name = PRINTBUF;
731 	unsigned i, iter_size;
732 	int ret = 0;
733 
734 	c = kvmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
735 	if (!c) {
736 		c = ERR_PTR(-BCH_ERR_ENOMEM_fs_alloc);
737 		goto out;
738 	}
739 
740 	c->stdio = (void *)(unsigned long) opts.stdio;
741 
742 	__module_get(THIS_MODULE);
743 
744 	closure_init(&c->cl, NULL);
745 
746 	c->kobj.kset = bcachefs_kset;
747 	kobject_init(&c->kobj, &bch2_fs_ktype);
748 	kobject_init(&c->internal, &bch2_fs_internal_ktype);
749 	kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
750 	kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
751 	kobject_init(&c->counters_kobj, &bch2_fs_counters_ktype);
752 
753 	c->minor		= -1;
754 	c->disk_sb.fs_sb	= true;
755 
756 	init_rwsem(&c->state_lock);
757 	mutex_init(&c->sb_lock);
758 	mutex_init(&c->replicas_gc_lock);
759 	mutex_init(&c->btree_root_lock);
760 	INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
761 
762 	refcount_set(&c->ro_ref, 1);
763 	init_waitqueue_head(&c->ro_ref_wait);
764 	sema_init(&c->online_fsck_mutex, 1);
765 
766 	init_rwsem(&c->gc_lock);
767 	mutex_init(&c->gc_gens_lock);
768 	atomic_set(&c->journal_keys.ref, 1);
769 	c->journal_keys.initial_ref_held = true;
770 
771 	for (i = 0; i < BCH_TIME_STAT_NR; i++)
772 		bch2_time_stats_init(&c->times[i]);
773 
774 	bch2_fs_gc_init(c);
775 	bch2_fs_copygc_init(c);
776 	bch2_fs_btree_key_cache_init_early(&c->btree_key_cache);
777 	bch2_fs_btree_iter_init_early(c);
778 	bch2_fs_btree_interior_update_init_early(c);
779 	bch2_fs_allocator_background_init(c);
780 	bch2_fs_allocator_foreground_init(c);
781 	bch2_fs_rebalance_init(c);
782 	bch2_fs_quota_init(c);
783 	bch2_fs_ec_init_early(c);
784 	bch2_fs_move_init(c);
785 	bch2_fs_sb_errors_init_early(c);
786 
787 	INIT_LIST_HEAD(&c->list);
788 
789 	mutex_init(&c->usage_scratch_lock);
790 
791 	mutex_init(&c->bio_bounce_pages_lock);
792 	mutex_init(&c->snapshot_table_lock);
793 	init_rwsem(&c->snapshot_create_lock);
794 
795 	spin_lock_init(&c->btree_write_error_lock);
796 
797 	INIT_LIST_HEAD(&c->journal_iters);
798 
799 	INIT_LIST_HEAD(&c->fsck_error_msgs);
800 	mutex_init(&c->fsck_error_msgs_lock);
801 
802 	seqcount_init(&c->usage_lock);
803 
804 	sema_init(&c->io_in_flight, 128);
805 
806 	INIT_LIST_HEAD(&c->vfs_inodes_list);
807 	mutex_init(&c->vfs_inodes_lock);
808 
809 	c->copy_gc_enabled		= 1;
810 	c->rebalance.enabled		= 1;
811 	c->promote_whole_extents	= true;
812 
813 	c->journal.flush_write_time	= &c->times[BCH_TIME_journal_flush_write];
814 	c->journal.noflush_write_time	= &c->times[BCH_TIME_journal_noflush_write];
815 	c->journal.flush_seq_time	= &c->times[BCH_TIME_journal_flush_seq];
816 
817 	bch2_fs_btree_cache_init_early(&c->btree_cache);
818 
819 	mutex_init(&c->sectors_available_lock);
820 
821 	ret = percpu_init_rwsem(&c->mark_lock);
822 	if (ret)
823 		goto err;
824 
825 	mutex_lock(&c->sb_lock);
826 	ret = bch2_sb_to_fs(c, sb);
827 	mutex_unlock(&c->sb_lock);
828 
829 	if (ret)
830 		goto err;
831 
832 	pr_uuid(&name, c->sb.user_uuid.b);
833 	ret = name.allocation_failure ? -BCH_ERR_ENOMEM_fs_name_alloc : 0;
834 	if (ret)
835 		goto err;
836 
837 	strscpy(c->name, name.buf, sizeof(c->name));
838 	printbuf_exit(&name);
839 
840 	/* Compat: */
841 	if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_inode_v2 &&
842 	    !BCH_SB_JOURNAL_FLUSH_DELAY(sb))
843 		SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
844 
845 	if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_inode_v2 &&
846 	    !BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
847 		SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 100);
848 
849 	c->opts = bch2_opts_default;
850 	ret = bch2_opts_from_sb(&c->opts, sb);
851 	if (ret)
852 		goto err;
853 
854 	bch2_opts_apply(&c->opts, opts);
855 
856 	c->btree_key_cache_btrees |= 1U << BTREE_ID_alloc;
857 	if (c->opts.inodes_use_key_cache)
858 		c->btree_key_cache_btrees |= 1U << BTREE_ID_inodes;
859 	c->btree_key_cache_btrees |= 1U << BTREE_ID_logged_ops;
860 
861 	c->block_bits		= ilog2(block_sectors(c));
862 	c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
863 
864 	if (bch2_fs_init_fault("fs_alloc")) {
865 		bch_err(c, "fs_alloc fault injected");
866 		ret = -EFAULT;
867 		goto err;
868 	}
869 
870 	iter_size = sizeof(struct sort_iter) +
871 		(btree_blocks(c) + 1) * 2 *
872 		sizeof(struct sort_iter_set);
873 
874 	c->inode_shard_bits = ilog2(roundup_pow_of_two(num_possible_cpus()));
875 
876 	if (!(c->btree_update_wq = alloc_workqueue("bcachefs",
877 				WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_UNBOUND, 512)) ||
878 	    !(c->btree_io_complete_wq = alloc_workqueue("bcachefs_btree_io",
879 				WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM, 1)) ||
880 	    !(c->copygc_wq = alloc_workqueue("bcachefs_copygc",
881 				WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
882 	    !(c->btree_read_complete_wq = alloc_workqueue("bcachefs_btree_read_complete",
883 				WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM, 512)) ||
884 	    !(c->btree_write_submit_wq = alloc_workqueue("bcachefs_btree_write_sumit",
885 				WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM, 1)) ||
886 	    !(c->write_ref_wq = alloc_workqueue("bcachefs_write_ref",
887 				WQ_FREEZABLE, 0)) ||
888 #ifndef BCH_WRITE_REF_DEBUG
889 	    percpu_ref_init(&c->writes, bch2_writes_disabled,
890 			    PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
891 #endif
892 	    mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
893 	    bioset_init(&c->btree_bio, 1,
894 			max(offsetof(struct btree_read_bio, bio),
895 			    offsetof(struct btree_write_bio, wbio.bio)),
896 			BIOSET_NEED_BVECS) ||
897 	    !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
898 	    !(c->online_reserved = alloc_percpu(u64)) ||
899 	    mempool_init_kvmalloc_pool(&c->btree_bounce_pool, 1,
900 				       c->opts.btree_node_size) ||
901 	    mempool_init_kmalloc_pool(&c->large_bkey_pool, 1, 2048) ||
902 	    !(c->unused_inode_hints = kcalloc(1U << c->inode_shard_bits,
903 					      sizeof(u64), GFP_KERNEL))) {
904 		ret = -BCH_ERR_ENOMEM_fs_other_alloc;
905 		goto err;
906 	}
907 
908 	ret = bch2_fs_counters_init(c) ?:
909 	    bch2_fs_sb_errors_init(c) ?:
910 	    bch2_io_clock_init(&c->io_clock[READ]) ?:
911 	    bch2_io_clock_init(&c->io_clock[WRITE]) ?:
912 	    bch2_fs_journal_init(&c->journal) ?:
913 	    bch2_fs_replicas_init(c) ?:
914 	    bch2_fs_btree_iter_init(c) ?:
915 	    bch2_fs_btree_cache_init(c) ?:
916 	    bch2_fs_btree_key_cache_init(&c->btree_key_cache) ?:
917 	    bch2_fs_btree_interior_update_init(c) ?:
918 	    bch2_fs_buckets_waiting_for_journal_init(c) ?:
919 	    bch2_fs_btree_write_buffer_init(c) ?:
920 	    bch2_fs_subvolumes_init(c) ?:
921 	    bch2_fs_io_read_init(c) ?:
922 	    bch2_fs_io_write_init(c) ?:
923 	    bch2_fs_nocow_locking_init(c) ?:
924 	    bch2_fs_encryption_init(c) ?:
925 	    bch2_fs_compress_init(c) ?:
926 	    bch2_fs_ec_init(c) ?:
927 	    bch2_fs_fsio_init(c) ?:
928 	    bch2_fs_fs_io_buffered_init(c) ?:
929 	    bch2_fs_fs_io_direct_init(c);
930 	if (ret)
931 		goto err;
932 
933 	for (i = 0; i < c->sb.nr_devices; i++) {
934 		if (!bch2_member_exists(c->disk_sb.sb, i))
935 			continue;
936 		ret = bch2_dev_alloc(c, i);
937 		if (ret)
938 			goto err;
939 	}
940 
941 	bch2_journal_entry_res_resize(&c->journal,
942 			&c->btree_root_journal_res,
943 			BTREE_ID_NR * (JSET_KEYS_U64s + BKEY_BTREE_PTR_U64s_MAX));
944 	bch2_dev_usage_journal_reserve(c);
945 	bch2_journal_entry_res_resize(&c->journal,
946 			&c->clock_journal_res,
947 			(sizeof(struct jset_entry_clock) / sizeof(u64)) * 2);
948 
949 	mutex_lock(&bch_fs_list_lock);
950 	ret = bch2_fs_online(c);
951 	mutex_unlock(&bch_fs_list_lock);
952 
953 	if (ret)
954 		goto err;
955 out:
956 	return c;
957 err:
958 	bch2_fs_free(c);
959 	c = ERR_PTR(ret);
960 	goto out;
961 }
962 
963 noinline_for_stack
964 static void print_mount_opts(struct bch_fs *c)
965 {
966 	enum bch_opt_id i;
967 	struct printbuf p = PRINTBUF;
968 	bool first = true;
969 
970 	prt_str(&p, "mounting version ");
971 	bch2_version_to_text(&p, c->sb.version);
972 
973 	if (c->opts.read_only) {
974 		prt_str(&p, " opts=");
975 		first = false;
976 		prt_printf(&p, "ro");
977 	}
978 
979 	for (i = 0; i < bch2_opts_nr; i++) {
980 		const struct bch_option *opt = &bch2_opt_table[i];
981 		u64 v = bch2_opt_get_by_id(&c->opts, i);
982 
983 		if (!(opt->flags & OPT_MOUNT))
984 			continue;
985 
986 		if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
987 			continue;
988 
989 		prt_str(&p, first ? " opts=" : ",");
990 		first = false;
991 		bch2_opt_to_text(&p, c, c->disk_sb.sb, opt, v, OPT_SHOW_MOUNT_STYLE);
992 	}
993 
994 	bch_info(c, "%s", p.buf);
995 	printbuf_exit(&p);
996 }
997 
998 int bch2_fs_start(struct bch_fs *c)
999 {
1000 	time64_t now = ktime_get_real_seconds();
1001 	int ret;
1002 
1003 	print_mount_opts(c);
1004 
1005 	down_write(&c->state_lock);
1006 
1007 	BUG_ON(test_bit(BCH_FS_started, &c->flags));
1008 
1009 	mutex_lock(&c->sb_lock);
1010 
1011 	ret = bch2_sb_members_v2_init(c);
1012 	if (ret) {
1013 		mutex_unlock(&c->sb_lock);
1014 		goto err;
1015 	}
1016 
1017 	for_each_online_member(c, ca)
1018 		bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx)->last_mount = cpu_to_le64(now);
1019 
1020 	struct bch_sb_field_ext *ext =
1021 		bch2_sb_field_get_minsize(&c->disk_sb, ext, sizeof(*ext) / sizeof(u64));
1022 	mutex_unlock(&c->sb_lock);
1023 
1024 	if (!ext) {
1025 		bch_err(c, "insufficient space in superblock for sb_field_ext");
1026 		ret = -BCH_ERR_ENOSPC_sb;
1027 		goto err;
1028 	}
1029 
1030 	for_each_rw_member(c, ca)
1031 		bch2_dev_allocator_add(c, ca);
1032 	bch2_recalc_capacity(c);
1033 
1034 	ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
1035 		? bch2_fs_recovery(c)
1036 		: bch2_fs_initialize(c);
1037 	if (ret)
1038 		goto err;
1039 
1040 	ret = bch2_opts_check_may_set(c);
1041 	if (ret)
1042 		goto err;
1043 
1044 	if (bch2_fs_init_fault("fs_start")) {
1045 		bch_err(c, "fs_start fault injected");
1046 		ret = -EINVAL;
1047 		goto err;
1048 	}
1049 
1050 	set_bit(BCH_FS_started, &c->flags);
1051 
1052 	if (c->opts.read_only) {
1053 		bch2_fs_read_only(c);
1054 	} else {
1055 		ret = !test_bit(BCH_FS_rw, &c->flags)
1056 			? bch2_fs_read_write(c)
1057 			: bch2_fs_read_write_late(c);
1058 		if (ret)
1059 			goto err;
1060 	}
1061 
1062 	ret = 0;
1063 err:
1064 	if (ret)
1065 		bch_err_msg(c, ret, "starting filesystem");
1066 	else
1067 		bch_verbose(c, "done starting filesystem");
1068 	up_write(&c->state_lock);
1069 	return ret;
1070 }
1071 
1072 static int bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
1073 {
1074 	struct bch_member m = bch2_sb_member_get(sb, sb->dev_idx);
1075 
1076 	if (le16_to_cpu(sb->block_size) != block_sectors(c))
1077 		return -BCH_ERR_mismatched_block_size;
1078 
1079 	if (le16_to_cpu(m.bucket_size) <
1080 	    BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
1081 		return -BCH_ERR_bucket_size_too_small;
1082 
1083 	return 0;
1084 }
1085 
1086 static int bch2_dev_in_fs(struct bch_sb_handle *fs,
1087 			  struct bch_sb_handle *sb,
1088 			  struct bch_opts *opts)
1089 {
1090 	if (fs == sb)
1091 		return 0;
1092 
1093 	if (!uuid_equal(&fs->sb->uuid, &sb->sb->uuid))
1094 		return -BCH_ERR_device_not_a_member_of_filesystem;
1095 
1096 	if (!bch2_member_exists(fs->sb, sb->sb->dev_idx))
1097 		return -BCH_ERR_device_has_been_removed;
1098 
1099 	if (fs->sb->block_size != sb->sb->block_size)
1100 		return -BCH_ERR_mismatched_block_size;
1101 
1102 	if (le16_to_cpu(fs->sb->version) < bcachefs_metadata_version_member_seq ||
1103 	    le16_to_cpu(sb->sb->version) < bcachefs_metadata_version_member_seq)
1104 		return 0;
1105 
1106 	if (fs->sb->seq == sb->sb->seq &&
1107 	    fs->sb->write_time != sb->sb->write_time) {
1108 		struct printbuf buf = PRINTBUF;
1109 
1110 		prt_str(&buf, "Split brain detected between ");
1111 		prt_bdevname(&buf, sb->bdev);
1112 		prt_str(&buf, " and ");
1113 		prt_bdevname(&buf, fs->bdev);
1114 		prt_char(&buf, ':');
1115 		prt_newline(&buf);
1116 		prt_printf(&buf, "seq=%llu but write_time different, got", le64_to_cpu(sb->sb->seq));
1117 		prt_newline(&buf);
1118 
1119 		prt_bdevname(&buf, fs->bdev);
1120 		prt_char(&buf, ' ');
1121 		bch2_prt_datetime(&buf, le64_to_cpu(fs->sb->write_time));;
1122 		prt_newline(&buf);
1123 
1124 		prt_bdevname(&buf, sb->bdev);
1125 		prt_char(&buf, ' ');
1126 		bch2_prt_datetime(&buf, le64_to_cpu(sb->sb->write_time));;
1127 		prt_newline(&buf);
1128 
1129 		if (!opts->no_splitbrain_check)
1130 			prt_printf(&buf, "Not using older sb");
1131 
1132 		pr_err("%s", buf.buf);
1133 		printbuf_exit(&buf);
1134 
1135 		if (!opts->no_splitbrain_check)
1136 			return -BCH_ERR_device_splitbrain;
1137 	}
1138 
1139 	struct bch_member m = bch2_sb_member_get(fs->sb, sb->sb->dev_idx);
1140 	u64 seq_from_fs		= le64_to_cpu(m.seq);
1141 	u64 seq_from_member	= le64_to_cpu(sb->sb->seq);
1142 
1143 	if (seq_from_fs && seq_from_fs < seq_from_member) {
1144 		struct printbuf buf = PRINTBUF;
1145 
1146 		prt_str(&buf, "Split brain detected between ");
1147 		prt_bdevname(&buf, sb->bdev);
1148 		prt_str(&buf, " and ");
1149 		prt_bdevname(&buf, fs->bdev);
1150 		prt_char(&buf, ':');
1151 		prt_newline(&buf);
1152 
1153 		prt_bdevname(&buf, fs->bdev);
1154 		prt_str(&buf, " believes seq of ");
1155 		prt_bdevname(&buf, sb->bdev);
1156 		prt_printf(&buf, " to be %llu, but ", seq_from_fs);
1157 		prt_bdevname(&buf, sb->bdev);
1158 		prt_printf(&buf, " has %llu\n", seq_from_member);
1159 
1160 		if (!opts->no_splitbrain_check) {
1161 			prt_str(&buf, "Not using ");
1162 			prt_bdevname(&buf, sb->bdev);
1163 		}
1164 
1165 		pr_err("%s", buf.buf);
1166 		printbuf_exit(&buf);
1167 
1168 		if (!opts->no_splitbrain_check)
1169 			return -BCH_ERR_device_splitbrain;
1170 	}
1171 
1172 	return 0;
1173 }
1174 
1175 /* Device startup/shutdown: */
1176 
1177 static void bch2_dev_release(struct kobject *kobj)
1178 {
1179 	struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
1180 
1181 	kfree(ca);
1182 }
1183 
1184 static void bch2_dev_free(struct bch_dev *ca)
1185 {
1186 	cancel_work_sync(&ca->io_error_work);
1187 
1188 	if (ca->kobj.state_in_sysfs &&
1189 	    ca->disk_sb.bdev)
1190 		sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1191 
1192 	if (ca->kobj.state_in_sysfs)
1193 		kobject_del(&ca->kobj);
1194 
1195 	kfree(ca->buckets_nouse);
1196 	bch2_free_super(&ca->disk_sb);
1197 	bch2_dev_allocator_background_exit(ca);
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 	bch2_dev_allocator_background_init(ca);
1321 
1322 	if (percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
1323 			    PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
1324 	    !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) ||
1325 	    bch2_dev_buckets_alloc(c, ca) ||
1326 	    !(ca->io_done	= alloc_percpu(*ca->io_done)))
1327 		goto err;
1328 
1329 	return ca;
1330 err:
1331 	bch2_dev_free(ca);
1332 	return NULL;
1333 }
1334 
1335 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
1336 			    unsigned dev_idx)
1337 {
1338 	ca->dev_idx = dev_idx;
1339 	__set_bit(ca->dev_idx, ca->self.d);
1340 	scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
1341 
1342 	ca->fs = c;
1343 	rcu_assign_pointer(c->devs[ca->dev_idx], ca);
1344 
1345 	if (bch2_dev_sysfs_online(c, ca))
1346 		pr_warn("error creating sysfs objects");
1347 }
1348 
1349 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
1350 {
1351 	struct bch_member member = bch2_sb_member_get(c->disk_sb.sb, dev_idx);
1352 	struct bch_dev *ca = NULL;
1353 	int ret = 0;
1354 
1355 	if (bch2_fs_init_fault("dev_alloc"))
1356 		goto err;
1357 
1358 	ca = __bch2_dev_alloc(c, &member);
1359 	if (!ca)
1360 		goto err;
1361 
1362 	ca->fs = c;
1363 
1364 	bch2_dev_attach(c, ca, dev_idx);
1365 	return ret;
1366 err:
1367 	if (ca)
1368 		bch2_dev_free(ca);
1369 	return -BCH_ERR_ENOMEM_dev_alloc;
1370 }
1371 
1372 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1373 {
1374 	unsigned ret;
1375 
1376 	if (bch2_dev_is_online(ca)) {
1377 		bch_err(ca, "already have device online in slot %u",
1378 			sb->sb->dev_idx);
1379 		return -BCH_ERR_device_already_online;
1380 	}
1381 
1382 	if (get_capacity(sb->bdev->bd_disk) <
1383 	    ca->mi.bucket_size * ca->mi.nbuckets) {
1384 		bch_err(ca, "cannot online: device too small");
1385 		return -BCH_ERR_device_size_too_small;
1386 	}
1387 
1388 	BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1389 
1390 	ret = bch2_dev_journal_init(ca, sb->sb);
1391 	if (ret)
1392 		return ret;
1393 
1394 	/* Commit: */
1395 	ca->disk_sb = *sb;
1396 	memset(sb, 0, sizeof(*sb));
1397 
1398 	ca->dev = ca->disk_sb.bdev->bd_dev;
1399 
1400 	percpu_ref_reinit(&ca->io_ref);
1401 
1402 	return 0;
1403 }
1404 
1405 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1406 {
1407 	struct bch_dev *ca;
1408 	int ret;
1409 
1410 	lockdep_assert_held(&c->state_lock);
1411 
1412 	if (le64_to_cpu(sb->sb->seq) >
1413 	    le64_to_cpu(c->disk_sb.sb->seq))
1414 		bch2_sb_to_fs(c, sb->sb);
1415 
1416 	BUG_ON(!bch2_dev_exists(c, sb->sb->dev_idx));
1417 
1418 	ca = bch2_dev_locked(c, sb->sb->dev_idx);
1419 
1420 	ret = __bch2_dev_attach_bdev(ca, sb);
1421 	if (ret)
1422 		return ret;
1423 
1424 	bch2_dev_sysfs_online(c, ca);
1425 
1426 	struct printbuf name = PRINTBUF;
1427 	prt_bdevname(&name, ca->disk_sb.bdev);
1428 
1429 	if (c->sb.nr_devices == 1)
1430 		strscpy(c->name, name.buf, sizeof(c->name));
1431 	strscpy(ca->name, name.buf, sizeof(ca->name));
1432 
1433 	printbuf_exit(&name);
1434 
1435 	rebalance_wakeup(c);
1436 	return 0;
1437 }
1438 
1439 /* Device management: */
1440 
1441 /*
1442  * Note: this function is also used by the error paths - when a particular
1443  * device sees an error, we call it to determine whether we can just set the
1444  * device RO, or - if this function returns false - we'll set the whole
1445  * filesystem RO:
1446  *
1447  * XXX: maybe we should be more explicit about whether we're changing state
1448  * because we got an error or what have you?
1449  */
1450 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1451 			    enum bch_member_state new_state, int flags)
1452 {
1453 	struct bch_devs_mask new_online_devs;
1454 	int nr_rw = 0, required;
1455 
1456 	lockdep_assert_held(&c->state_lock);
1457 
1458 	switch (new_state) {
1459 	case BCH_MEMBER_STATE_rw:
1460 		return true;
1461 	case BCH_MEMBER_STATE_ro:
1462 		if (ca->mi.state != BCH_MEMBER_STATE_rw)
1463 			return true;
1464 
1465 		/* do we have enough devices to write to?  */
1466 		for_each_member_device(c, ca2)
1467 			if (ca2 != ca)
1468 				nr_rw += ca2->mi.state == BCH_MEMBER_STATE_rw;
1469 
1470 		required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1471 			       ? c->opts.metadata_replicas
1472 			       : metadata_replicas_required(c),
1473 			       !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1474 			       ? c->opts.data_replicas
1475 			       : data_replicas_required(c));
1476 
1477 		return nr_rw >= required;
1478 	case BCH_MEMBER_STATE_failed:
1479 	case BCH_MEMBER_STATE_spare:
1480 		if (ca->mi.state != BCH_MEMBER_STATE_rw &&
1481 		    ca->mi.state != BCH_MEMBER_STATE_ro)
1482 			return true;
1483 
1484 		/* do we have enough devices to read from?  */
1485 		new_online_devs = bch2_online_devs(c);
1486 		__clear_bit(ca->dev_idx, new_online_devs.d);
1487 
1488 		return bch2_have_enough_devs(c, new_online_devs, flags, false);
1489 	default:
1490 		BUG();
1491 	}
1492 }
1493 
1494 static bool bch2_fs_may_start(struct bch_fs *c)
1495 {
1496 	struct bch_dev *ca;
1497 	unsigned i, flags = 0;
1498 
1499 	if (c->opts.very_degraded)
1500 		flags |= BCH_FORCE_IF_DEGRADED|BCH_FORCE_IF_LOST;
1501 
1502 	if (c->opts.degraded)
1503 		flags |= BCH_FORCE_IF_DEGRADED;
1504 
1505 	if (!c->opts.degraded &&
1506 	    !c->opts.very_degraded) {
1507 		mutex_lock(&c->sb_lock);
1508 
1509 		for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1510 			if (!bch2_member_exists(c->disk_sb.sb, i))
1511 				continue;
1512 
1513 			ca = bch2_dev_locked(c, i);
1514 
1515 			if (!bch2_dev_is_online(ca) &&
1516 			    (ca->mi.state == BCH_MEMBER_STATE_rw ||
1517 			     ca->mi.state == BCH_MEMBER_STATE_ro)) {
1518 				mutex_unlock(&c->sb_lock);
1519 				return false;
1520 			}
1521 		}
1522 		mutex_unlock(&c->sb_lock);
1523 	}
1524 
1525 	return bch2_have_enough_devs(c, bch2_online_devs(c), flags, true);
1526 }
1527 
1528 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1529 {
1530 	/*
1531 	 * The allocator thread itself allocates btree nodes, so stop it first:
1532 	 */
1533 	bch2_dev_allocator_remove(c, ca);
1534 	bch2_recalc_capacity(c);
1535 	bch2_dev_journal_stop(&c->journal, ca);
1536 }
1537 
1538 static void __bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1539 {
1540 	lockdep_assert_held(&c->state_lock);
1541 
1542 	BUG_ON(ca->mi.state != BCH_MEMBER_STATE_rw);
1543 
1544 	bch2_dev_allocator_add(c, ca);
1545 	bch2_recalc_capacity(c);
1546 	bch2_dev_do_discards(ca);
1547 }
1548 
1549 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1550 			 enum bch_member_state new_state, int flags)
1551 {
1552 	struct bch_member *m;
1553 	int ret = 0;
1554 
1555 	if (ca->mi.state == new_state)
1556 		return 0;
1557 
1558 	if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1559 		return -BCH_ERR_device_state_not_allowed;
1560 
1561 	if (new_state != BCH_MEMBER_STATE_rw)
1562 		__bch2_dev_read_only(c, ca);
1563 
1564 	bch_notice(ca, "%s", bch2_member_states[new_state]);
1565 
1566 	mutex_lock(&c->sb_lock);
1567 	m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
1568 	SET_BCH_MEMBER_STATE(m, new_state);
1569 	bch2_write_super(c);
1570 	mutex_unlock(&c->sb_lock);
1571 
1572 	if (new_state == BCH_MEMBER_STATE_rw)
1573 		__bch2_dev_read_write(c, ca);
1574 
1575 	rebalance_wakeup(c);
1576 
1577 	return ret;
1578 }
1579 
1580 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1581 		       enum bch_member_state new_state, int flags)
1582 {
1583 	int ret;
1584 
1585 	down_write(&c->state_lock);
1586 	ret = __bch2_dev_set_state(c, ca, new_state, flags);
1587 	up_write(&c->state_lock);
1588 
1589 	return ret;
1590 }
1591 
1592 /* Device add/removal: */
1593 
1594 static int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
1595 {
1596 	struct bpos start	= POS(ca->dev_idx, 0);
1597 	struct bpos end		= POS(ca->dev_idx, U64_MAX);
1598 	int ret;
1599 
1600 	/*
1601 	 * We clear the LRU and need_discard btrees first so that we don't race
1602 	 * with bch2_do_invalidates() and bch2_do_discards()
1603 	 */
1604 	ret =   bch2_btree_delete_range(c, BTREE_ID_lru, start, end,
1605 					BTREE_TRIGGER_norun, NULL) ?:
1606 		bch2_btree_delete_range(c, BTREE_ID_need_discard, start, end,
1607 					BTREE_TRIGGER_norun, NULL) ?:
1608 		bch2_btree_delete_range(c, BTREE_ID_freespace, start, end,
1609 					BTREE_TRIGGER_norun, NULL) ?:
1610 		bch2_btree_delete_range(c, BTREE_ID_backpointers, start, end,
1611 					BTREE_TRIGGER_norun, NULL) ?:
1612 		bch2_btree_delete_range(c, BTREE_ID_alloc, start, end,
1613 					BTREE_TRIGGER_norun, NULL) ?:
1614 		bch2_btree_delete_range(c, BTREE_ID_bucket_gens, start, end,
1615 					BTREE_TRIGGER_norun, NULL);
1616 	bch_err_msg(c, ret, "removing dev alloc info");
1617 	return ret;
1618 }
1619 
1620 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1621 {
1622 	struct bch_member *m;
1623 	unsigned dev_idx = ca->dev_idx, data;
1624 	int ret;
1625 
1626 	down_write(&c->state_lock);
1627 
1628 	/*
1629 	 * We consume a reference to ca->ref, regardless of whether we succeed
1630 	 * or fail:
1631 	 */
1632 	bch2_dev_put(ca);
1633 
1634 	if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1635 		bch_err(ca, "Cannot remove without losing data");
1636 		ret = -BCH_ERR_device_state_not_allowed;
1637 		goto err;
1638 	}
1639 
1640 	__bch2_dev_read_only(c, ca);
1641 
1642 	ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1643 	bch_err_msg(ca, ret, "bch2_dev_data_drop()");
1644 	if (ret)
1645 		goto err;
1646 
1647 	ret = bch2_dev_remove_alloc(c, ca);
1648 	bch_err_msg(ca, ret, "bch2_dev_remove_alloc()");
1649 	if (ret)
1650 		goto err;
1651 
1652 	ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1653 	bch_err_msg(ca, ret, "bch2_journal_flush_device_pins()");
1654 	if (ret)
1655 		goto err;
1656 
1657 	ret = bch2_journal_flush(&c->journal);
1658 	bch_err_msg(ca, ret, "bch2_journal_flush()");
1659 	if (ret)
1660 		goto err;
1661 
1662 	ret = bch2_replicas_gc2(c);
1663 	bch_err_msg(ca, ret, "bch2_replicas_gc2()");
1664 	if (ret)
1665 		goto err;
1666 
1667 	data = bch2_dev_has_data(c, ca);
1668 	if (data) {
1669 		struct printbuf data_has = PRINTBUF;
1670 
1671 		prt_bitflags(&data_has, __bch2_data_types, data);
1672 		bch_err(ca, "Remove failed, still has data (%s)", data_has.buf);
1673 		printbuf_exit(&data_has);
1674 		ret = -EBUSY;
1675 		goto err;
1676 	}
1677 
1678 	__bch2_dev_offline(c, ca);
1679 
1680 	mutex_lock(&c->sb_lock);
1681 	rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1682 	mutex_unlock(&c->sb_lock);
1683 
1684 #ifndef CONFIG_BCACHEFS_DEBUG
1685 	percpu_ref_kill(&ca->ref);
1686 #else
1687 	ca->dying = true;
1688 	bch2_dev_put(ca);
1689 #endif
1690 	wait_for_completion(&ca->ref_completion);
1691 
1692 	bch2_dev_free(ca);
1693 
1694 	/*
1695 	 * At this point the device object has been removed in-core, but the
1696 	 * on-disk journal might still refer to the device index via sb device
1697 	 * usage entries. Recovery fails if it sees usage information for an
1698 	 * invalid device. Flush journal pins to push the back of the journal
1699 	 * past now invalid device index references before we update the
1700 	 * superblock, but after the device object has been removed so any
1701 	 * further journal writes elide usage info for the device.
1702 	 */
1703 	bch2_journal_flush_all_pins(&c->journal);
1704 
1705 	/*
1706 	 * Free this device's slot in the bch_member array - all pointers to
1707 	 * this device must be gone:
1708 	 */
1709 	mutex_lock(&c->sb_lock);
1710 	m = bch2_members_v2_get_mut(c->disk_sb.sb, dev_idx);
1711 	memset(&m->uuid, 0, sizeof(m->uuid));
1712 
1713 	bch2_write_super(c);
1714 
1715 	mutex_unlock(&c->sb_lock);
1716 	up_write(&c->state_lock);
1717 
1718 	bch2_dev_usage_journal_reserve(c);
1719 	return 0;
1720 err:
1721 	if (ca->mi.state == BCH_MEMBER_STATE_rw &&
1722 	    !percpu_ref_is_zero(&ca->io_ref))
1723 		__bch2_dev_read_write(c, ca);
1724 	up_write(&c->state_lock);
1725 	return ret;
1726 }
1727 
1728 /* Add new device to running filesystem: */
1729 int bch2_dev_add(struct bch_fs *c, const char *path)
1730 {
1731 	struct bch_opts opts = bch2_opts_empty();
1732 	struct bch_sb_handle sb;
1733 	struct bch_dev *ca = NULL;
1734 	struct bch_sb_field_members_v2 *mi;
1735 	struct bch_member dev_mi;
1736 	unsigned dev_idx, nr_devices, u64s;
1737 	struct printbuf errbuf = PRINTBUF;
1738 	struct printbuf label = PRINTBUF;
1739 	int ret;
1740 
1741 	ret = bch2_read_super(path, &opts, &sb);
1742 	bch_err_msg(c, ret, "reading super");
1743 	if (ret)
1744 		goto err;
1745 
1746 	dev_mi = bch2_sb_member_get(sb.sb, sb.sb->dev_idx);
1747 
1748 	if (BCH_MEMBER_GROUP(&dev_mi)) {
1749 		bch2_disk_path_to_text_sb(&label, sb.sb, BCH_MEMBER_GROUP(&dev_mi) - 1);
1750 		if (label.allocation_failure) {
1751 			ret = -ENOMEM;
1752 			goto err;
1753 		}
1754 	}
1755 
1756 	ret = bch2_dev_may_add(sb.sb, c);
1757 	if (ret)
1758 		goto err;
1759 
1760 	ca = __bch2_dev_alloc(c, &dev_mi);
1761 	if (!ca) {
1762 		ret = -ENOMEM;
1763 		goto err;
1764 	}
1765 
1766 	bch2_dev_usage_init(ca);
1767 
1768 	ret = __bch2_dev_attach_bdev(ca, &sb);
1769 	if (ret)
1770 		goto err;
1771 
1772 	ret = bch2_dev_journal_alloc(ca);
1773 	bch_err_msg(c, ret, "allocating journal");
1774 	if (ret)
1775 		goto err;
1776 
1777 	down_write(&c->state_lock);
1778 	mutex_lock(&c->sb_lock);
1779 
1780 	ret = bch2_sb_from_fs(c, ca);
1781 	bch_err_msg(c, ret, "setting up new superblock");
1782 	if (ret)
1783 		goto err_unlock;
1784 
1785 	if (dynamic_fault("bcachefs:add:no_slot"))
1786 		goto no_slot;
1787 
1788 	if (c->sb.nr_devices < BCH_SB_MEMBERS_MAX) {
1789 		dev_idx = c->sb.nr_devices;
1790 		goto have_slot;
1791 	}
1792 
1793 	int best = -1;
1794 	u64 best_last_mount = 0;
1795 	for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++) {
1796 		struct bch_member m = bch2_sb_member_get(c->disk_sb.sb, dev_idx);
1797 		if (bch2_member_alive(&m))
1798 			continue;
1799 
1800 		u64 last_mount = le64_to_cpu(m.last_mount);
1801 		if (best < 0 || last_mount < best_last_mount) {
1802 			best = dev_idx;
1803 			best_last_mount = last_mount;
1804 		}
1805 	}
1806 	if (best >= 0) {
1807 		dev_idx = best;
1808 		goto have_slot;
1809 	}
1810 no_slot:
1811 	ret = -BCH_ERR_ENOSPC_sb_members;
1812 	bch_err_msg(c, ret, "setting up new superblock");
1813 	goto err_unlock;
1814 
1815 have_slot:
1816 	nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1817 
1818 	mi = bch2_sb_field_get(c->disk_sb.sb, members_v2);
1819 	u64s = DIV_ROUND_UP(sizeof(struct bch_sb_field_members_v2) +
1820 			    le16_to_cpu(mi->member_bytes) * nr_devices, sizeof(u64));
1821 
1822 	mi = bch2_sb_field_resize(&c->disk_sb, members_v2, u64s);
1823 	if (!mi) {
1824 		ret = -BCH_ERR_ENOSPC_sb_members;
1825 		bch_err_msg(c, ret, "setting up new superblock");
1826 		goto err_unlock;
1827 	}
1828 	struct bch_member *m = bch2_members_v2_get_mut(c->disk_sb.sb, dev_idx);
1829 
1830 	/* success: */
1831 
1832 	*m = dev_mi;
1833 	m->last_mount = cpu_to_le64(ktime_get_real_seconds());
1834 	c->disk_sb.sb->nr_devices	= nr_devices;
1835 
1836 	ca->disk_sb.sb->dev_idx	= dev_idx;
1837 	bch2_dev_attach(c, ca, dev_idx);
1838 
1839 	if (BCH_MEMBER_GROUP(&dev_mi)) {
1840 		ret = __bch2_dev_group_set(c, ca, label.buf);
1841 		bch_err_msg(c, ret, "creating new label");
1842 		if (ret)
1843 			goto err_unlock;
1844 	}
1845 
1846 	bch2_write_super(c);
1847 	mutex_unlock(&c->sb_lock);
1848 
1849 	bch2_dev_usage_journal_reserve(c);
1850 
1851 	ret = bch2_trans_mark_dev_sb(c, ca, BTREE_TRIGGER_transactional);
1852 	bch_err_msg(ca, ret, "marking new superblock");
1853 	if (ret)
1854 		goto err_late;
1855 
1856 	ret = bch2_fs_freespace_init(c);
1857 	bch_err_msg(ca, ret, "initializing free space");
1858 	if (ret)
1859 		goto err_late;
1860 
1861 	ca->new_fs_bucket_idx = 0;
1862 
1863 	if (ca->mi.state == BCH_MEMBER_STATE_rw)
1864 		__bch2_dev_read_write(c, ca);
1865 
1866 	up_write(&c->state_lock);
1867 	return 0;
1868 
1869 err_unlock:
1870 	mutex_unlock(&c->sb_lock);
1871 	up_write(&c->state_lock);
1872 err:
1873 	if (ca)
1874 		bch2_dev_free(ca);
1875 	bch2_free_super(&sb);
1876 	printbuf_exit(&label);
1877 	printbuf_exit(&errbuf);
1878 	bch_err_fn(c, ret);
1879 	return ret;
1880 err_late:
1881 	up_write(&c->state_lock);
1882 	ca = NULL;
1883 	goto err;
1884 }
1885 
1886 /* Hot add existing device to running filesystem: */
1887 int bch2_dev_online(struct bch_fs *c, const char *path)
1888 {
1889 	struct bch_opts opts = bch2_opts_empty();
1890 	struct bch_sb_handle sb = { NULL };
1891 	struct bch_dev *ca;
1892 	unsigned dev_idx;
1893 	int ret;
1894 
1895 	down_write(&c->state_lock);
1896 
1897 	ret = bch2_read_super(path, &opts, &sb);
1898 	if (ret) {
1899 		up_write(&c->state_lock);
1900 		return ret;
1901 	}
1902 
1903 	dev_idx = sb.sb->dev_idx;
1904 
1905 	ret = bch2_dev_in_fs(&c->disk_sb, &sb, &c->opts);
1906 	bch_err_msg(c, ret, "bringing %s online", path);
1907 	if (ret)
1908 		goto err;
1909 
1910 	ret = bch2_dev_attach_bdev(c, &sb);
1911 	if (ret)
1912 		goto err;
1913 
1914 	ca = bch2_dev_locked(c, dev_idx);
1915 
1916 	ret = bch2_trans_mark_dev_sb(c, ca, BTREE_TRIGGER_transactional);
1917 	bch_err_msg(c, ret, "bringing %s online: error from bch2_trans_mark_dev_sb", path);
1918 	if (ret)
1919 		goto err;
1920 
1921 	if (ca->mi.state == BCH_MEMBER_STATE_rw)
1922 		__bch2_dev_read_write(c, ca);
1923 
1924 	if (!ca->mi.freespace_initialized) {
1925 		ret = bch2_dev_freespace_init(c, ca, 0, ca->mi.nbuckets);
1926 		bch_err_msg(ca, ret, "initializing free space");
1927 		if (ret)
1928 			goto err;
1929 	}
1930 
1931 	if (!ca->journal.nr) {
1932 		ret = bch2_dev_journal_alloc(ca);
1933 		bch_err_msg(ca, ret, "allocating journal");
1934 		if (ret)
1935 			goto err;
1936 	}
1937 
1938 	mutex_lock(&c->sb_lock);
1939 	bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx)->last_mount =
1940 		cpu_to_le64(ktime_get_real_seconds());
1941 	bch2_write_super(c);
1942 	mutex_unlock(&c->sb_lock);
1943 
1944 	up_write(&c->state_lock);
1945 	return 0;
1946 err:
1947 	up_write(&c->state_lock);
1948 	bch2_free_super(&sb);
1949 	return ret;
1950 }
1951 
1952 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1953 {
1954 	down_write(&c->state_lock);
1955 
1956 	if (!bch2_dev_is_online(ca)) {
1957 		bch_err(ca, "Already offline");
1958 		up_write(&c->state_lock);
1959 		return 0;
1960 	}
1961 
1962 	if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1963 		bch_err(ca, "Cannot offline required disk");
1964 		up_write(&c->state_lock);
1965 		return -BCH_ERR_device_state_not_allowed;
1966 	}
1967 
1968 	__bch2_dev_offline(c, ca);
1969 
1970 	up_write(&c->state_lock);
1971 	return 0;
1972 }
1973 
1974 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1975 {
1976 	struct bch_member *m;
1977 	u64 old_nbuckets;
1978 	int ret = 0;
1979 
1980 	down_write(&c->state_lock);
1981 	old_nbuckets = ca->mi.nbuckets;
1982 
1983 	if (nbuckets < ca->mi.nbuckets) {
1984 		bch_err(ca, "Cannot shrink yet");
1985 		ret = -EINVAL;
1986 		goto err;
1987 	}
1988 
1989 	if (nbuckets > BCH_MEMBER_NBUCKETS_MAX) {
1990 		bch_err(ca, "New device size too big (%llu greater than max %u)",
1991 			nbuckets, BCH_MEMBER_NBUCKETS_MAX);
1992 		ret = -BCH_ERR_device_size_too_big;
1993 		goto err;
1994 	}
1995 
1996 	if (bch2_dev_is_online(ca) &&
1997 	    get_capacity(ca->disk_sb.bdev->bd_disk) <
1998 	    ca->mi.bucket_size * nbuckets) {
1999 		bch_err(ca, "New size larger than device");
2000 		ret = -BCH_ERR_device_size_too_small;
2001 		goto err;
2002 	}
2003 
2004 	ret = bch2_dev_buckets_resize(c, ca, nbuckets);
2005 	bch_err_msg(ca, ret, "resizing buckets");
2006 	if (ret)
2007 		goto err;
2008 
2009 	ret = bch2_trans_mark_dev_sb(c, ca, BTREE_TRIGGER_transactional);
2010 	if (ret)
2011 		goto err;
2012 
2013 	mutex_lock(&c->sb_lock);
2014 	m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
2015 	m->nbuckets = cpu_to_le64(nbuckets);
2016 
2017 	bch2_write_super(c);
2018 	mutex_unlock(&c->sb_lock);
2019 
2020 	if (ca->mi.freespace_initialized) {
2021 		ret = bch2_dev_freespace_init(c, ca, old_nbuckets, nbuckets);
2022 		if (ret)
2023 			goto err;
2024 
2025 		/*
2026 		 * XXX: this is all wrong transactionally - we'll be able to do
2027 		 * this correctly after the disk space accounting rewrite
2028 		 */
2029 		ca->usage_base->d[BCH_DATA_free].buckets += nbuckets - old_nbuckets;
2030 	}
2031 
2032 	bch2_recalc_capacity(c);
2033 err:
2034 	up_write(&c->state_lock);
2035 	return ret;
2036 }
2037 
2038 /* return with ref on ca->ref: */
2039 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *name)
2040 {
2041 	for_each_member_device(c, ca)
2042 		if (!strcmp(name, ca->name))
2043 			return ca;
2044 	return ERR_PTR(-BCH_ERR_ENOENT_dev_not_found);
2045 }
2046 
2047 /* Filesystem open: */
2048 
2049 static inline int sb_cmp(struct bch_sb *l, struct bch_sb *r)
2050 {
2051 	return  cmp_int(le64_to_cpu(l->seq), le64_to_cpu(r->seq)) ?:
2052 		cmp_int(le64_to_cpu(l->write_time), le64_to_cpu(r->write_time));
2053 }
2054 
2055 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
2056 			    struct bch_opts opts)
2057 {
2058 	DARRAY(struct bch_sb_handle) sbs = { 0 };
2059 	struct bch_fs *c = NULL;
2060 	struct bch_sb_handle *best = NULL;
2061 	struct printbuf errbuf = PRINTBUF;
2062 	int ret = 0;
2063 
2064 	if (!try_module_get(THIS_MODULE))
2065 		return ERR_PTR(-ENODEV);
2066 
2067 	if (!nr_devices) {
2068 		ret = -EINVAL;
2069 		goto err;
2070 	}
2071 
2072 	ret = darray_make_room(&sbs, nr_devices);
2073 	if (ret)
2074 		goto err;
2075 
2076 	for (unsigned i = 0; i < nr_devices; i++) {
2077 		struct bch_sb_handle sb = { NULL };
2078 
2079 		ret = bch2_read_super(devices[i], &opts, &sb);
2080 		if (ret)
2081 			goto err;
2082 
2083 		BUG_ON(darray_push(&sbs, sb));
2084 	}
2085 
2086 	if (opts.nochanges && !opts.read_only) {
2087 		ret = -BCH_ERR_erofs_nochanges;
2088 		goto err_print;
2089 	}
2090 
2091 	darray_for_each(sbs, sb)
2092 		if (!best || sb_cmp(sb->sb, best->sb) > 0)
2093 			best = sb;
2094 
2095 	darray_for_each_reverse(sbs, sb) {
2096 		ret = bch2_dev_in_fs(best, sb, &opts);
2097 
2098 		if (ret == -BCH_ERR_device_has_been_removed ||
2099 		    ret == -BCH_ERR_device_splitbrain) {
2100 			bch2_free_super(sb);
2101 			darray_remove_item(&sbs, sb);
2102 			best -= best > sb;
2103 			ret = 0;
2104 			continue;
2105 		}
2106 
2107 		if (ret)
2108 			goto err_print;
2109 	}
2110 
2111 	c = bch2_fs_alloc(best->sb, opts);
2112 	ret = PTR_ERR_OR_ZERO(c);
2113 	if (ret)
2114 		goto err;
2115 
2116 	down_write(&c->state_lock);
2117 	darray_for_each(sbs, sb) {
2118 		ret = bch2_dev_attach_bdev(c, sb);
2119 		if (ret) {
2120 			up_write(&c->state_lock);
2121 			goto err;
2122 		}
2123 	}
2124 	up_write(&c->state_lock);
2125 
2126 	if (!bch2_fs_may_start(c)) {
2127 		ret = -BCH_ERR_insufficient_devices_to_start;
2128 		goto err_print;
2129 	}
2130 
2131 	if (!c->opts.nostart) {
2132 		ret = bch2_fs_start(c);
2133 		if (ret)
2134 			goto err;
2135 	}
2136 out:
2137 	darray_for_each(sbs, sb)
2138 		bch2_free_super(sb);
2139 	darray_exit(&sbs);
2140 	printbuf_exit(&errbuf);
2141 	module_put(THIS_MODULE);
2142 	return c;
2143 err_print:
2144 	pr_err("bch_fs_open err opening %s: %s",
2145 	       devices[0], bch2_err_str(ret));
2146 err:
2147 	if (!IS_ERR_OR_NULL(c))
2148 		bch2_fs_stop(c);
2149 	c = ERR_PTR(ret);
2150 	goto out;
2151 }
2152 
2153 /* Global interfaces/init */
2154 
2155 static void bcachefs_exit(void)
2156 {
2157 	bch2_debug_exit();
2158 	bch2_vfs_exit();
2159 	bch2_chardev_exit();
2160 	bch2_btree_key_cache_exit();
2161 	if (bcachefs_kset)
2162 		kset_unregister(bcachefs_kset);
2163 }
2164 
2165 static int __init bcachefs_init(void)
2166 {
2167 	bch2_bkey_pack_test();
2168 
2169 	if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
2170 	    bch2_btree_key_cache_init() ||
2171 	    bch2_chardev_init() ||
2172 	    bch2_vfs_init() ||
2173 	    bch2_debug_init())
2174 		goto err;
2175 
2176 	return 0;
2177 err:
2178 	bcachefs_exit();
2179 	return -ENOMEM;
2180 }
2181 
2182 #define BCH_DEBUG_PARAM(name, description)			\
2183 	bool bch2_##name;					\
2184 	module_param_named(name, bch2_##name, bool, 0644);	\
2185 	MODULE_PARM_DESC(name, description);
2186 BCH_DEBUG_PARAMS()
2187 #undef BCH_DEBUG_PARAM
2188 
2189 __maybe_unused
2190 static unsigned bch2_metadata_version = bcachefs_metadata_version_current;
2191 module_param_named(version, bch2_metadata_version, uint, 0400);
2192 
2193 module_exit(bcachefs_exit);
2194 module_init(bcachefs_init);
2195