xref: /linux/fs/bcachefs/super.c (revision ba199dc909a20fe62270ae4e93f263987bb9d119)
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_accounting.h"
29 #include "disk_groups.h"
30 #include "ec.h"
31 #include "errcode.h"
32 #include "error.h"
33 #include "fs.h"
34 #include "fs-io.h"
35 #include "fs-io-buffered.h"
36 #include "fs-io-direct.h"
37 #include "fsck.h"
38 #include "inode.h"
39 #include "io_read.h"
40 #include "io_write.h"
41 #include "journal.h"
42 #include "journal_reclaim.h"
43 #include "journal_seq_blacklist.h"
44 #include "move.h"
45 #include "migrate.h"
46 #include "movinggc.h"
47 #include "nocow_locking.h"
48 #include "quota.h"
49 #include "rebalance.h"
50 #include "recovery.h"
51 #include "replicas.h"
52 #include "sb-clean.h"
53 #include "sb-counters.h"
54 #include "sb-errors.h"
55 #include "sb-members.h"
56 #include "snapshot.h"
57 #include "subvolume.h"
58 #include "super.h"
59 #include "super-io.h"
60 #include "sysfs.h"
61 #include "thread_with_file.h"
62 #include "trace.h"
63 
64 #include <linux/backing-dev.h>
65 #include <linux/blkdev.h>
66 #include <linux/debugfs.h>
67 #include <linux/device.h>
68 #include <linux/idr.h>
69 #include <linux/module.h>
70 #include <linux/percpu.h>
71 #include <linux/random.h>
72 #include <linux/sysfs.h>
73 #include <crypto/hash.h>
74 
75 MODULE_LICENSE("GPL");
76 MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>");
77 MODULE_DESCRIPTION("bcachefs filesystem");
78 MODULE_SOFTDEP("pre: crc32c");
79 MODULE_SOFTDEP("pre: crc64");
80 MODULE_SOFTDEP("pre: sha256");
81 MODULE_SOFTDEP("pre: chacha20");
82 MODULE_SOFTDEP("pre: poly1305");
83 MODULE_SOFTDEP("pre: xxhash");
84 
85 const char * const bch2_fs_flag_strs[] = {
86 #define x(n)		#n,
87 	BCH_FS_FLAGS()
88 #undef x
89 	NULL
90 };
91 
92 void bch2_print_str(struct bch_fs *c, const char *str)
93 {
94 #ifdef __KERNEL__
95 	struct stdio_redirect *stdio = bch2_fs_stdio_redirect(c);
96 
97 	if (unlikely(stdio)) {
98 		bch2_stdio_redirect_printf(stdio, true, "%s", str);
99 		return;
100 	}
101 #endif
102 	bch2_print_string_as_lines(KERN_ERR, str);
103 }
104 
105 __printf(2, 0)
106 static void bch2_print_maybe_redirect(struct stdio_redirect *stdio, const char *fmt, va_list args)
107 {
108 #ifdef __KERNEL__
109 	if (unlikely(stdio)) {
110 		if (fmt[0] == KERN_SOH[0])
111 			fmt += 2;
112 
113 		bch2_stdio_redirect_vprintf(stdio, true, fmt, args);
114 		return;
115 	}
116 #endif
117 	vprintk(fmt, args);
118 }
119 
120 void bch2_print_opts(struct bch_opts *opts, const char *fmt, ...)
121 {
122 	struct stdio_redirect *stdio = (void *)(unsigned long)opts->stdio;
123 
124 	va_list args;
125 	va_start(args, fmt);
126 	bch2_print_maybe_redirect(stdio, fmt, args);
127 	va_end(args);
128 }
129 
130 void __bch2_print(struct bch_fs *c, const char *fmt, ...)
131 {
132 	struct stdio_redirect *stdio = bch2_fs_stdio_redirect(c);
133 
134 	va_list args;
135 	va_start(args, fmt);
136 	bch2_print_maybe_redirect(stdio, fmt, args);
137 	va_end(args);
138 }
139 
140 #define KTYPE(type)							\
141 static const struct attribute_group type ## _group = {			\
142 	.attrs = type ## _files						\
143 };									\
144 									\
145 static const struct attribute_group *type ## _groups[] = {		\
146 	&type ## _group,						\
147 	NULL								\
148 };									\
149 									\
150 static const struct kobj_type type ## _ktype = {			\
151 	.release	= type ## _release,				\
152 	.sysfs_ops	= &type ## _sysfs_ops,				\
153 	.default_groups = type ## _groups				\
154 }
155 
156 static void bch2_fs_release(struct kobject *);
157 static void bch2_dev_release(struct kobject *);
158 static void bch2_fs_counters_release(struct kobject *k)
159 {
160 }
161 
162 static void bch2_fs_internal_release(struct kobject *k)
163 {
164 }
165 
166 static void bch2_fs_opts_dir_release(struct kobject *k)
167 {
168 }
169 
170 static void bch2_fs_time_stats_release(struct kobject *k)
171 {
172 }
173 
174 KTYPE(bch2_fs);
175 KTYPE(bch2_fs_counters);
176 KTYPE(bch2_fs_internal);
177 KTYPE(bch2_fs_opts_dir);
178 KTYPE(bch2_fs_time_stats);
179 KTYPE(bch2_dev);
180 
181 static struct kset *bcachefs_kset;
182 static LIST_HEAD(bch_fs_list);
183 static DEFINE_MUTEX(bch_fs_list_lock);
184 
185 DECLARE_WAIT_QUEUE_HEAD(bch2_read_only_wait);
186 
187 static void bch2_dev_free(struct bch_dev *);
188 static int bch2_dev_alloc(struct bch_fs *, unsigned);
189 static int bch2_dev_sysfs_online(struct bch_fs *, struct bch_dev *);
190 static void __bch2_dev_read_only(struct bch_fs *, struct bch_dev *);
191 
192 struct bch_fs *bch2_dev_to_fs(dev_t dev)
193 {
194 	struct bch_fs *c;
195 
196 	mutex_lock(&bch_fs_list_lock);
197 	rcu_read_lock();
198 
199 	list_for_each_entry(c, &bch_fs_list, list)
200 		for_each_member_device_rcu(c, ca, NULL)
201 			if (ca->disk_sb.bdev && ca->disk_sb.bdev->bd_dev == dev) {
202 				closure_get(&c->cl);
203 				goto found;
204 			}
205 	c = NULL;
206 found:
207 	rcu_read_unlock();
208 	mutex_unlock(&bch_fs_list_lock);
209 
210 	return c;
211 }
212 
213 static struct bch_fs *__bch2_uuid_to_fs(__uuid_t uuid)
214 {
215 	struct bch_fs *c;
216 
217 	lockdep_assert_held(&bch_fs_list_lock);
218 
219 	list_for_each_entry(c, &bch_fs_list, list)
220 		if (!memcmp(&c->disk_sb.sb->uuid, &uuid, sizeof(uuid)))
221 			return c;
222 
223 	return NULL;
224 }
225 
226 struct bch_fs *bch2_uuid_to_fs(__uuid_t uuid)
227 {
228 	struct bch_fs *c;
229 
230 	mutex_lock(&bch_fs_list_lock);
231 	c = __bch2_uuid_to_fs(uuid);
232 	if (c)
233 		closure_get(&c->cl);
234 	mutex_unlock(&bch_fs_list_lock);
235 
236 	return c;
237 }
238 
239 /* Filesystem RO/RW: */
240 
241 /*
242  * For startup/shutdown of RW stuff, the dependencies are:
243  *
244  * - foreground writes depend on copygc and rebalance (to free up space)
245  *
246  * - copygc and rebalance depend on mark and sweep gc (they actually probably
247  *   don't because they either reserve ahead of time or don't block if
248  *   allocations fail, but allocations can require mark and sweep gc to run
249  *   because of generation number wraparound)
250  *
251  * - all of the above depends on the allocator threads
252  *
253  * - allocator depends on the journal (when it rewrites prios and gens)
254  */
255 
256 static void __bch2_fs_read_only(struct bch_fs *c)
257 {
258 	unsigned clean_passes = 0;
259 	u64 seq = 0;
260 
261 	bch2_fs_ec_stop(c);
262 	bch2_open_buckets_stop(c, NULL, true);
263 	bch2_rebalance_stop(c);
264 	bch2_copygc_stop(c);
265 	bch2_fs_ec_flush(c);
266 
267 	bch_verbose(c, "flushing journal and stopping allocators, journal seq %llu",
268 		    journal_cur_seq(&c->journal));
269 
270 	do {
271 		clean_passes++;
272 
273 		if (bch2_btree_interior_updates_flush(c) ||
274 		    bch2_journal_flush_all_pins(&c->journal) ||
275 		    bch2_btree_flush_all_writes(c) ||
276 		    seq != atomic64_read(&c->journal.seq)) {
277 			seq = atomic64_read(&c->journal.seq);
278 			clean_passes = 0;
279 		}
280 	} while (clean_passes < 2);
281 
282 	bch_verbose(c, "flushing journal and stopping allocators complete, journal seq %llu",
283 		    journal_cur_seq(&c->journal));
284 
285 	if (test_bit(JOURNAL_replay_done, &c->journal.flags) &&
286 	    !test_bit(BCH_FS_emergency_ro, &c->flags))
287 		set_bit(BCH_FS_clean_shutdown, &c->flags);
288 
289 	bch2_fs_journal_stop(&c->journal);
290 
291 	bch_info(c, "%sshutdown complete, journal seq %llu",
292 		 test_bit(BCH_FS_clean_shutdown, &c->flags) ? "" : "un",
293 		 c->journal.seq_ondisk);
294 
295 	/*
296 	 * After stopping journal:
297 	 */
298 	for_each_member_device(c, ca)
299 		bch2_dev_allocator_remove(c, ca);
300 }
301 
302 #ifndef BCH_WRITE_REF_DEBUG
303 static void bch2_writes_disabled(struct percpu_ref *writes)
304 {
305 	struct bch_fs *c = container_of(writes, struct bch_fs, writes);
306 
307 	set_bit(BCH_FS_write_disable_complete, &c->flags);
308 	wake_up(&bch2_read_only_wait);
309 }
310 #endif
311 
312 void bch2_fs_read_only(struct bch_fs *c)
313 {
314 	if (!test_bit(BCH_FS_rw, &c->flags)) {
315 		bch2_journal_reclaim_stop(&c->journal);
316 		return;
317 	}
318 
319 	BUG_ON(test_bit(BCH_FS_write_disable_complete, &c->flags));
320 
321 	bch_verbose(c, "going read-only");
322 
323 	/*
324 	 * Block new foreground-end write operations from starting - any new
325 	 * writes will return -EROFS:
326 	 */
327 	set_bit(BCH_FS_going_ro, &c->flags);
328 #ifndef BCH_WRITE_REF_DEBUG
329 	percpu_ref_kill(&c->writes);
330 #else
331 	for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++)
332 		bch2_write_ref_put(c, i);
333 #endif
334 
335 	/*
336 	 * If we're not doing an emergency shutdown, we want to wait on
337 	 * outstanding writes to complete so they don't see spurious errors due
338 	 * to shutting down the allocator:
339 	 *
340 	 * If we are doing an emergency shutdown outstanding writes may
341 	 * hang until we shutdown the allocator so we don't want to wait
342 	 * on outstanding writes before shutting everything down - but
343 	 * we do need to wait on them before returning and signalling
344 	 * that going RO is complete:
345 	 */
346 	wait_event(bch2_read_only_wait,
347 		   test_bit(BCH_FS_write_disable_complete, &c->flags) ||
348 		   test_bit(BCH_FS_emergency_ro, &c->flags));
349 
350 	bool writes_disabled = test_bit(BCH_FS_write_disable_complete, &c->flags);
351 	if (writes_disabled)
352 		bch_verbose(c, "finished waiting for writes to stop");
353 
354 	__bch2_fs_read_only(c);
355 
356 	wait_event(bch2_read_only_wait,
357 		   test_bit(BCH_FS_write_disable_complete, &c->flags));
358 
359 	if (!writes_disabled)
360 		bch_verbose(c, "finished waiting for writes to stop");
361 
362 	clear_bit(BCH_FS_write_disable_complete, &c->flags);
363 	clear_bit(BCH_FS_going_ro, &c->flags);
364 	clear_bit(BCH_FS_rw, &c->flags);
365 
366 	if (!bch2_journal_error(&c->journal) &&
367 	    !test_bit(BCH_FS_error, &c->flags) &&
368 	    !test_bit(BCH_FS_emergency_ro, &c->flags) &&
369 	    test_bit(BCH_FS_started, &c->flags) &&
370 	    test_bit(BCH_FS_clean_shutdown, &c->flags) &&
371 	    c->recovery_pass_done >= BCH_RECOVERY_PASS_journal_replay) {
372 		BUG_ON(c->journal.last_empty_seq != journal_cur_seq(&c->journal));
373 		BUG_ON(atomic_read(&c->btree_cache.dirty));
374 		BUG_ON(atomic_long_read(&c->btree_key_cache.nr_dirty));
375 		BUG_ON(c->btree_write_buffer.inc.keys.nr);
376 		BUG_ON(c->btree_write_buffer.flushing.keys.nr);
377 		bch2_verify_accounting_clean(c);
378 
379 		bch_verbose(c, "marking filesystem clean");
380 		bch2_fs_mark_clean(c);
381 	} else {
382 		bch_verbose(c, "done going read-only, filesystem not clean");
383 	}
384 }
385 
386 static void bch2_fs_read_only_work(struct work_struct *work)
387 {
388 	struct bch_fs *c =
389 		container_of(work, struct bch_fs, read_only_work);
390 
391 	down_write(&c->state_lock);
392 	bch2_fs_read_only(c);
393 	up_write(&c->state_lock);
394 }
395 
396 static void bch2_fs_read_only_async(struct bch_fs *c)
397 {
398 	queue_work(system_long_wq, &c->read_only_work);
399 }
400 
401 bool bch2_fs_emergency_read_only(struct bch_fs *c)
402 {
403 	bool ret = !test_and_set_bit(BCH_FS_emergency_ro, &c->flags);
404 
405 	bch2_journal_halt(&c->journal);
406 	bch2_fs_read_only_async(c);
407 
408 	wake_up(&bch2_read_only_wait);
409 	return ret;
410 }
411 
412 static int bch2_fs_read_write_late(struct bch_fs *c)
413 {
414 	int ret;
415 
416 	/*
417 	 * Data move operations can't run until after check_snapshots has
418 	 * completed, and bch2_snapshot_is_ancestor() is available.
419 	 *
420 	 * Ideally we'd start copygc/rebalance earlier instead of waiting for
421 	 * all of recovery/fsck to complete:
422 	 */
423 	ret = bch2_copygc_start(c);
424 	if (ret) {
425 		bch_err(c, "error starting copygc thread");
426 		return ret;
427 	}
428 
429 	ret = bch2_rebalance_start(c);
430 	if (ret) {
431 		bch_err(c, "error starting rebalance thread");
432 		return ret;
433 	}
434 
435 	return 0;
436 }
437 
438 static int __bch2_fs_read_write(struct bch_fs *c, bool early)
439 {
440 	int ret;
441 
442 	if (test_bit(BCH_FS_initial_gc_unfixed, &c->flags)) {
443 		bch_err(c, "cannot go rw, unfixed btree errors");
444 		return -BCH_ERR_erofs_unfixed_errors;
445 	}
446 
447 	if (test_bit(BCH_FS_rw, &c->flags))
448 		return 0;
449 
450 	bch_info(c, "going read-write");
451 
452 	ret = bch2_sb_members_v2_init(c);
453 	if (ret)
454 		goto err;
455 
456 	ret = bch2_fs_mark_dirty(c);
457 	if (ret)
458 		goto err;
459 
460 	clear_bit(BCH_FS_clean_shutdown, &c->flags);
461 
462 	/*
463 	 * First journal write must be a flush write: after a clean shutdown we
464 	 * don't read the journal, so the first journal write may end up
465 	 * overwriting whatever was there previously, and there must always be
466 	 * at least one non-flush write in the journal or recovery will fail:
467 	 */
468 	set_bit(JOURNAL_need_flush_write, &c->journal.flags);
469 	set_bit(JOURNAL_running, &c->journal.flags);
470 
471 	for_each_rw_member(c, ca)
472 		bch2_dev_allocator_add(c, ca);
473 	bch2_recalc_capacity(c);
474 
475 	set_bit(BCH_FS_rw, &c->flags);
476 	set_bit(BCH_FS_was_rw, &c->flags);
477 
478 #ifndef BCH_WRITE_REF_DEBUG
479 	percpu_ref_reinit(&c->writes);
480 #else
481 	for (unsigned i = 0; i < BCH_WRITE_REF_NR; i++) {
482 		BUG_ON(atomic_long_read(&c->writes[i]));
483 		atomic_long_inc(&c->writes[i]);
484 	}
485 #endif
486 
487 	ret = bch2_journal_reclaim_start(&c->journal);
488 	if (ret)
489 		goto err;
490 
491 	if (!early) {
492 		ret = bch2_fs_read_write_late(c);
493 		if (ret)
494 			goto err;
495 	}
496 
497 	bch2_do_discards(c);
498 	bch2_do_invalidates(c);
499 	bch2_do_stripe_deletes(c);
500 	bch2_do_pending_node_rewrites(c);
501 	return 0;
502 err:
503 	if (test_bit(BCH_FS_rw, &c->flags))
504 		bch2_fs_read_only(c);
505 	else
506 		__bch2_fs_read_only(c);
507 	return ret;
508 }
509 
510 int bch2_fs_read_write(struct bch_fs *c)
511 {
512 	if (c->opts.recovery_pass_last &&
513 	    c->opts.recovery_pass_last < BCH_RECOVERY_PASS_journal_replay)
514 		return -BCH_ERR_erofs_norecovery;
515 
516 	if (c->opts.nochanges)
517 		return -BCH_ERR_erofs_nochanges;
518 
519 	return __bch2_fs_read_write(c, false);
520 }
521 
522 int bch2_fs_read_write_early(struct bch_fs *c)
523 {
524 	lockdep_assert_held(&c->state_lock);
525 
526 	return __bch2_fs_read_write(c, true);
527 }
528 
529 /* Filesystem startup/shutdown: */
530 
531 static void __bch2_fs_free(struct bch_fs *c)
532 {
533 	for (unsigned i = 0; i < BCH_TIME_STAT_NR; i++)
534 		bch2_time_stats_exit(&c->times[i]);
535 
536 	bch2_find_btree_nodes_exit(&c->found_btree_nodes);
537 	bch2_free_pending_node_rewrites(c);
538 	bch2_fs_accounting_exit(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 	free_percpu(c->usage);
575 	mempool_exit(&c->large_bkey_pool);
576 	mempool_exit(&c->btree_bounce_pool);
577 	bioset_exit(&c->btree_bio);
578 	mempool_exit(&c->fill_iter);
579 #ifndef BCH_WRITE_REF_DEBUG
580 	percpu_ref_exit(&c->writes);
581 #endif
582 	kfree(rcu_dereference_protected(c->disk_groups, 1));
583 	kfree(c->journal_seq_blacklist_table);
584 	kfree(c->unused_inode_hints);
585 
586 	if (c->write_ref_wq)
587 		destroy_workqueue(c->write_ref_wq);
588 	if (c->btree_write_submit_wq)
589 		destroy_workqueue(c->btree_write_submit_wq);
590 	if (c->btree_read_complete_wq)
591 		destroy_workqueue(c->btree_read_complete_wq);
592 	if (c->copygc_wq)
593 		destroy_workqueue(c->copygc_wq);
594 	if (c->btree_io_complete_wq)
595 		destroy_workqueue(c->btree_io_complete_wq);
596 	if (c->btree_update_wq)
597 		destroy_workqueue(c->btree_update_wq);
598 
599 	bch2_free_super(&c->disk_sb);
600 	kvfree(c);
601 	module_put(THIS_MODULE);
602 }
603 
604 static void bch2_fs_release(struct kobject *kobj)
605 {
606 	struct bch_fs *c = container_of(kobj, struct bch_fs, kobj);
607 
608 	__bch2_fs_free(c);
609 }
610 
611 void __bch2_fs_stop(struct bch_fs *c)
612 {
613 	bch_verbose(c, "shutting down");
614 
615 	set_bit(BCH_FS_stopping, &c->flags);
616 
617 	down_write(&c->state_lock);
618 	bch2_fs_read_only(c);
619 	up_write(&c->state_lock);
620 
621 	for_each_member_device(c, ca)
622 		if (ca->kobj.state_in_sysfs &&
623 		    ca->disk_sb.bdev)
624 			sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
625 
626 	if (c->kobj.state_in_sysfs)
627 		kobject_del(&c->kobj);
628 
629 	bch2_fs_debug_exit(c);
630 	bch2_fs_chardev_exit(c);
631 
632 	bch2_ro_ref_put(c);
633 	wait_event(c->ro_ref_wait, !refcount_read(&c->ro_ref));
634 
635 	kobject_put(&c->counters_kobj);
636 	kobject_put(&c->time_stats);
637 	kobject_put(&c->opts_dir);
638 	kobject_put(&c->internal);
639 
640 	/* btree prefetch might have kicked off reads in the background: */
641 	bch2_btree_flush_all_reads(c);
642 
643 	for_each_member_device(c, ca)
644 		cancel_work_sync(&ca->io_error_work);
645 
646 	cancel_work_sync(&c->read_only_work);
647 }
648 
649 void bch2_fs_free(struct bch_fs *c)
650 {
651 	unsigned i;
652 
653 	mutex_lock(&bch_fs_list_lock);
654 	list_del(&c->list);
655 	mutex_unlock(&bch_fs_list_lock);
656 
657 	closure_sync(&c->cl);
658 	closure_debug_destroy(&c->cl);
659 
660 	for (i = 0; i < c->sb.nr_devices; i++) {
661 		struct bch_dev *ca = rcu_dereference_protected(c->devs[i], true);
662 
663 		if (ca) {
664 			EBUG_ON(atomic_long_read(&ca->ref) != 1);
665 			bch2_free_super(&ca->disk_sb);
666 			bch2_dev_free(ca);
667 		}
668 	}
669 
670 	bch_verbose(c, "shutdown complete");
671 
672 	kobject_put(&c->kobj);
673 }
674 
675 void bch2_fs_stop(struct bch_fs *c)
676 {
677 	__bch2_fs_stop(c);
678 	bch2_fs_free(c);
679 }
680 
681 static int bch2_fs_online(struct bch_fs *c)
682 {
683 	int ret = 0;
684 
685 	lockdep_assert_held(&bch_fs_list_lock);
686 
687 	if (__bch2_uuid_to_fs(c->sb.uuid)) {
688 		bch_err(c, "filesystem UUID already open");
689 		return -EINVAL;
690 	}
691 
692 	ret = bch2_fs_chardev_init(c);
693 	if (ret) {
694 		bch_err(c, "error creating character device");
695 		return ret;
696 	}
697 
698 	bch2_fs_debug_init(c);
699 
700 	ret = kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) ?:
701 	    kobject_add(&c->internal, &c->kobj, "internal") ?:
702 	    kobject_add(&c->opts_dir, &c->kobj, "options") ?:
703 #ifndef CONFIG_BCACHEFS_NO_LATENCY_ACCT
704 	    kobject_add(&c->time_stats, &c->kobj, "time_stats") ?:
705 #endif
706 	    kobject_add(&c->counters_kobj, &c->kobj, "counters") ?:
707 	    bch2_opts_create_sysfs_files(&c->opts_dir);
708 	if (ret) {
709 		bch_err(c, "error creating sysfs objects");
710 		return ret;
711 	}
712 
713 	down_write(&c->state_lock);
714 
715 	for_each_member_device(c, ca) {
716 		ret = bch2_dev_sysfs_online(c, ca);
717 		if (ret) {
718 			bch_err(c, "error creating sysfs objects");
719 			bch2_dev_put(ca);
720 			goto err;
721 		}
722 	}
723 
724 	BUG_ON(!list_empty(&c->list));
725 	list_add(&c->list, &bch_fs_list);
726 err:
727 	up_write(&c->state_lock);
728 	return ret;
729 }
730 
731 static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts)
732 {
733 	struct bch_fs *c;
734 	struct printbuf name = PRINTBUF;
735 	unsigned i, iter_size;
736 	int ret = 0;
737 
738 	c = kvmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO);
739 	if (!c) {
740 		c = ERR_PTR(-BCH_ERR_ENOMEM_fs_alloc);
741 		goto out;
742 	}
743 
744 	c->stdio = (void *)(unsigned long) opts.stdio;
745 
746 	__module_get(THIS_MODULE);
747 
748 	closure_init(&c->cl, NULL);
749 
750 	c->kobj.kset = bcachefs_kset;
751 	kobject_init(&c->kobj, &bch2_fs_ktype);
752 	kobject_init(&c->internal, &bch2_fs_internal_ktype);
753 	kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype);
754 	kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype);
755 	kobject_init(&c->counters_kobj, &bch2_fs_counters_ktype);
756 
757 	c->minor		= -1;
758 	c->disk_sb.fs_sb	= true;
759 
760 	init_rwsem(&c->state_lock);
761 	mutex_init(&c->sb_lock);
762 	mutex_init(&c->replicas_gc_lock);
763 	mutex_init(&c->btree_root_lock);
764 	INIT_WORK(&c->read_only_work, bch2_fs_read_only_work);
765 
766 	refcount_set(&c->ro_ref, 1);
767 	init_waitqueue_head(&c->ro_ref_wait);
768 	sema_init(&c->online_fsck_mutex, 1);
769 
770 	init_rwsem(&c->gc_lock);
771 	mutex_init(&c->gc_gens_lock);
772 	atomic_set(&c->journal_keys.ref, 1);
773 	c->journal_keys.initial_ref_held = true;
774 
775 	for (i = 0; i < BCH_TIME_STAT_NR; i++)
776 		bch2_time_stats_init(&c->times[i]);
777 
778 	bch2_fs_gc_init(c);
779 	bch2_fs_copygc_init(c);
780 	bch2_fs_btree_key_cache_init_early(&c->btree_key_cache);
781 	bch2_fs_btree_iter_init_early(c);
782 	bch2_fs_btree_interior_update_init_early(c);
783 	bch2_fs_allocator_background_init(c);
784 	bch2_fs_allocator_foreground_init(c);
785 	bch2_fs_rebalance_init(c);
786 	bch2_fs_quota_init(c);
787 	bch2_fs_ec_init_early(c);
788 	bch2_fs_move_init(c);
789 	bch2_fs_sb_errors_init_early(c);
790 
791 	INIT_LIST_HEAD(&c->list);
792 
793 	mutex_init(&c->bio_bounce_pages_lock);
794 	mutex_init(&c->snapshot_table_lock);
795 	init_rwsem(&c->snapshot_create_lock);
796 
797 	spin_lock_init(&c->btree_write_error_lock);
798 
799 	INIT_LIST_HEAD(&c->journal_iters);
800 
801 	INIT_LIST_HEAD(&c->fsck_error_msgs);
802 	mutex_init(&c->fsck_error_msgs_lock);
803 
804 	seqcount_init(&c->usage_lock);
805 
806 	sema_init(&c->io_in_flight, 128);
807 
808 	INIT_LIST_HEAD(&c->vfs_inodes_list);
809 	mutex_init(&c->vfs_inodes_lock);
810 
811 	c->copy_gc_enabled		= 1;
812 	c->rebalance.enabled		= 1;
813 	c->promote_whole_extents	= true;
814 
815 	c->journal.flush_write_time	= &c->times[BCH_TIME_journal_flush_write];
816 	c->journal.noflush_write_time	= &c->times[BCH_TIME_journal_noflush_write];
817 	c->journal.flush_seq_time	= &c->times[BCH_TIME_journal_flush_seq];
818 
819 	bch2_fs_btree_cache_init_early(&c->btree_cache);
820 
821 	mutex_init(&c->sectors_available_lock);
822 
823 	ret = percpu_init_rwsem(&c->mark_lock);
824 	if (ret)
825 		goto err;
826 
827 	mutex_lock(&c->sb_lock);
828 	ret = bch2_sb_to_fs(c, sb);
829 	mutex_unlock(&c->sb_lock);
830 
831 	if (ret)
832 		goto err;
833 
834 	pr_uuid(&name, c->sb.user_uuid.b);
835 	ret = name.allocation_failure ? -BCH_ERR_ENOMEM_fs_name_alloc : 0;
836 	if (ret)
837 		goto err;
838 
839 	strscpy(c->name, name.buf, sizeof(c->name));
840 	printbuf_exit(&name);
841 
842 	/* Compat: */
843 	if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_inode_v2 &&
844 	    !BCH_SB_JOURNAL_FLUSH_DELAY(sb))
845 		SET_BCH_SB_JOURNAL_FLUSH_DELAY(sb, 1000);
846 
847 	if (le16_to_cpu(sb->version) <= bcachefs_metadata_version_inode_v2 &&
848 	    !BCH_SB_JOURNAL_RECLAIM_DELAY(sb))
849 		SET_BCH_SB_JOURNAL_RECLAIM_DELAY(sb, 100);
850 
851 	c->opts = bch2_opts_default;
852 	ret = bch2_opts_from_sb(&c->opts, sb);
853 	if (ret)
854 		goto err;
855 
856 	bch2_opts_apply(&c->opts, opts);
857 
858 	c->btree_key_cache_btrees |= 1U << BTREE_ID_alloc;
859 	if (c->opts.inodes_use_key_cache)
860 		c->btree_key_cache_btrees |= 1U << BTREE_ID_inodes;
861 	c->btree_key_cache_btrees |= 1U << BTREE_ID_logged_ops;
862 
863 	c->block_bits		= ilog2(block_sectors(c));
864 	c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c);
865 
866 	if (bch2_fs_init_fault("fs_alloc")) {
867 		bch_err(c, "fs_alloc fault injected");
868 		ret = -EFAULT;
869 		goto err;
870 	}
871 
872 	iter_size = sizeof(struct sort_iter) +
873 		(btree_blocks(c) + 1) * 2 *
874 		sizeof(struct sort_iter_set);
875 
876 	c->inode_shard_bits = ilog2(roundup_pow_of_two(num_possible_cpus()));
877 
878 	if (!(c->btree_update_wq = alloc_workqueue("bcachefs",
879 				WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_UNBOUND, 512)) ||
880 	    !(c->btree_io_complete_wq = alloc_workqueue("bcachefs_btree_io",
881 				WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM, 1)) ||
882 	    !(c->copygc_wq = alloc_workqueue("bcachefs_copygc",
883 				WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_CPU_INTENSIVE, 1)) ||
884 	    !(c->btree_read_complete_wq = alloc_workqueue("bcachefs_btree_read_complete",
885 				WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM, 512)) ||
886 	    !(c->btree_write_submit_wq = alloc_workqueue("bcachefs_btree_write_sumit",
887 				WQ_HIGHPRI|WQ_FREEZABLE|WQ_MEM_RECLAIM, 1)) ||
888 	    !(c->write_ref_wq = alloc_workqueue("bcachefs_write_ref",
889 				WQ_FREEZABLE, 0)) ||
890 #ifndef BCH_WRITE_REF_DEBUG
891 	    percpu_ref_init(&c->writes, bch2_writes_disabled,
892 			    PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
893 #endif
894 	    mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) ||
895 	    bioset_init(&c->btree_bio, 1,
896 			max(offsetof(struct btree_read_bio, bio),
897 			    offsetof(struct btree_write_bio, wbio.bio)),
898 			BIOSET_NEED_BVECS) ||
899 	    !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) ||
900 	    !(c->usage = alloc_percpu(struct bch_fs_usage_base)) ||
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_btree_iter_init(c) ?:
917 	    bch2_fs_btree_cache_init(c) ?:
918 	    bch2_fs_btree_key_cache_init(&c->btree_key_cache) ?:
919 	    bch2_fs_btree_interior_update_init(c) ?:
920 	    bch2_fs_buckets_waiting_for_journal_init(c) ?:
921 	    bch2_fs_btree_write_buffer_init(c) ?:
922 	    bch2_fs_subvolumes_init(c) ?:
923 	    bch2_fs_io_read_init(c) ?:
924 	    bch2_fs_io_write_init(c) ?:
925 	    bch2_fs_nocow_locking_init(c) ?:
926 	    bch2_fs_encryption_init(c) ?:
927 	    bch2_fs_compress_init(c) ?:
928 	    bch2_fs_ec_init(c) ?:
929 	    bch2_fs_fsio_init(c) ?:
930 	    bch2_fs_fs_io_buffered_init(c) ?:
931 	    bch2_fs_fs_io_direct_init(c);
932 	if (ret)
933 		goto err;
934 
935 	for (i = 0; i < c->sb.nr_devices; i++) {
936 		if (!bch2_member_exists(c->disk_sb.sb, i))
937 			continue;
938 		ret = bch2_dev_alloc(c, i);
939 		if (ret)
940 			goto err;
941 	}
942 
943 	bch2_journal_entry_res_resize(&c->journal,
944 			&c->btree_root_journal_res,
945 			BTREE_ID_NR * (JSET_KEYS_U64s + BKEY_BTREE_PTR_U64s_MAX));
946 	bch2_journal_entry_res_resize(&c->journal,
947 			&c->clock_journal_res,
948 			(sizeof(struct jset_entry_clock) / sizeof(u64)) * 2);
949 
950 	mutex_lock(&bch_fs_list_lock);
951 	ret = bch2_fs_online(c);
952 	mutex_unlock(&bch_fs_list_lock);
953 
954 	if (ret)
955 		goto err;
956 out:
957 	return c;
958 err:
959 	bch2_fs_free(c);
960 	c = ERR_PTR(ret);
961 	goto out;
962 }
963 
964 noinline_for_stack
965 static void print_mount_opts(struct bch_fs *c)
966 {
967 	enum bch_opt_id i;
968 	struct printbuf p = PRINTBUF;
969 	bool first = true;
970 
971 	prt_str(&p, "starting version ");
972 	bch2_version_to_text(&p, c->sb.version);
973 
974 	if (c->opts.read_only) {
975 		prt_str(&p, " opts=");
976 		first = false;
977 		prt_printf(&p, "ro");
978 	}
979 
980 	for (i = 0; i < bch2_opts_nr; i++) {
981 		const struct bch_option *opt = &bch2_opt_table[i];
982 		u64 v = bch2_opt_get_by_id(&c->opts, i);
983 
984 		if (!(opt->flags & OPT_MOUNT))
985 			continue;
986 
987 		if (v == bch2_opt_get_by_id(&bch2_opts_default, i))
988 			continue;
989 
990 		prt_str(&p, first ? " opts=" : ",");
991 		first = false;
992 		bch2_opt_to_text(&p, c, c->disk_sb.sb, opt, v, OPT_SHOW_MOUNT_STYLE);
993 	}
994 
995 	bch_info(c, "%s", p.buf);
996 	printbuf_exit(&p);
997 }
998 
999 int bch2_fs_start(struct bch_fs *c)
1000 {
1001 	time64_t now = ktime_get_real_seconds();
1002 	int ret;
1003 
1004 	print_mount_opts(c);
1005 
1006 	down_write(&c->state_lock);
1007 
1008 	BUG_ON(test_bit(BCH_FS_started, &c->flags));
1009 
1010 	mutex_lock(&c->sb_lock);
1011 
1012 	ret = bch2_sb_members_v2_init(c);
1013 	if (ret) {
1014 		mutex_unlock(&c->sb_lock);
1015 		goto err;
1016 	}
1017 
1018 	for_each_online_member(c, ca)
1019 		bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx)->last_mount = cpu_to_le64(now);
1020 
1021 	struct bch_sb_field_ext *ext =
1022 		bch2_sb_field_get_minsize(&c->disk_sb, ext, sizeof(*ext) / sizeof(u64));
1023 	mutex_unlock(&c->sb_lock);
1024 
1025 	if (!ext) {
1026 		bch_err(c, "insufficient space in superblock for sb_field_ext");
1027 		ret = -BCH_ERR_ENOSPC_sb;
1028 		goto err;
1029 	}
1030 
1031 	for_each_rw_member(c, ca)
1032 		bch2_dev_allocator_add(c, ca);
1033 	bch2_recalc_capacity(c);
1034 
1035 	ret = BCH_SB_INITIALIZED(c->disk_sb.sb)
1036 		? bch2_fs_recovery(c)
1037 		: bch2_fs_initialize(c);
1038 	if (ret)
1039 		goto err;
1040 
1041 	ret = bch2_opts_check_may_set(c);
1042 	if (ret)
1043 		goto err;
1044 
1045 	if (bch2_fs_init_fault("fs_start")) {
1046 		bch_err(c, "fs_start fault injected");
1047 		ret = -EINVAL;
1048 		goto err;
1049 	}
1050 
1051 	set_bit(BCH_FS_started, &c->flags);
1052 
1053 	if (c->opts.read_only) {
1054 		bch2_fs_read_only(c);
1055 	} else {
1056 		ret = !test_bit(BCH_FS_rw, &c->flags)
1057 			? bch2_fs_read_write(c)
1058 			: bch2_fs_read_write_late(c);
1059 		if (ret)
1060 			goto err;
1061 	}
1062 
1063 	ret = 0;
1064 err:
1065 	if (ret)
1066 		bch_err_msg(c, ret, "starting filesystem");
1067 	else
1068 		bch_verbose(c, "done starting filesystem");
1069 	up_write(&c->state_lock);
1070 	return ret;
1071 }
1072 
1073 static int bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c)
1074 {
1075 	struct bch_member m = bch2_sb_member_get(sb, sb->dev_idx);
1076 
1077 	if (le16_to_cpu(sb->block_size) != block_sectors(c))
1078 		return -BCH_ERR_mismatched_block_size;
1079 
1080 	if (le16_to_cpu(m.bucket_size) <
1081 	    BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb))
1082 		return -BCH_ERR_bucket_size_too_small;
1083 
1084 	return 0;
1085 }
1086 
1087 static int bch2_dev_in_fs(struct bch_sb_handle *fs,
1088 			  struct bch_sb_handle *sb,
1089 			  struct bch_opts *opts)
1090 {
1091 	if (fs == sb)
1092 		return 0;
1093 
1094 	if (!uuid_equal(&fs->sb->uuid, &sb->sb->uuid))
1095 		return -BCH_ERR_device_not_a_member_of_filesystem;
1096 
1097 	if (!bch2_member_exists(fs->sb, sb->sb->dev_idx))
1098 		return -BCH_ERR_device_has_been_removed;
1099 
1100 	if (fs->sb->block_size != sb->sb->block_size)
1101 		return -BCH_ERR_mismatched_block_size;
1102 
1103 	if (le16_to_cpu(fs->sb->version) < bcachefs_metadata_version_member_seq ||
1104 	    le16_to_cpu(sb->sb->version) < bcachefs_metadata_version_member_seq)
1105 		return 0;
1106 
1107 	if (fs->sb->seq == sb->sb->seq &&
1108 	    fs->sb->write_time != sb->sb->write_time) {
1109 		struct printbuf buf = PRINTBUF;
1110 
1111 		prt_str(&buf, "Split brain detected between ");
1112 		prt_bdevname(&buf, sb->bdev);
1113 		prt_str(&buf, " and ");
1114 		prt_bdevname(&buf, fs->bdev);
1115 		prt_char(&buf, ':');
1116 		prt_newline(&buf);
1117 		prt_printf(&buf, "seq=%llu but write_time different, got", le64_to_cpu(sb->sb->seq));
1118 		prt_newline(&buf);
1119 
1120 		prt_bdevname(&buf, fs->bdev);
1121 		prt_char(&buf, ' ');
1122 		bch2_prt_datetime(&buf, le64_to_cpu(fs->sb->write_time));;
1123 		prt_newline(&buf);
1124 
1125 		prt_bdevname(&buf, sb->bdev);
1126 		prt_char(&buf, ' ');
1127 		bch2_prt_datetime(&buf, le64_to_cpu(sb->sb->write_time));;
1128 		prt_newline(&buf);
1129 
1130 		if (!opts->no_splitbrain_check)
1131 			prt_printf(&buf, "Not using older sb");
1132 
1133 		pr_err("%s", buf.buf);
1134 		printbuf_exit(&buf);
1135 
1136 		if (!opts->no_splitbrain_check)
1137 			return -BCH_ERR_device_splitbrain;
1138 	}
1139 
1140 	struct bch_member m = bch2_sb_member_get(fs->sb, sb->sb->dev_idx);
1141 	u64 seq_from_fs		= le64_to_cpu(m.seq);
1142 	u64 seq_from_member	= le64_to_cpu(sb->sb->seq);
1143 
1144 	if (seq_from_fs && seq_from_fs < seq_from_member) {
1145 		struct printbuf buf = PRINTBUF;
1146 
1147 		prt_str(&buf, "Split brain detected between ");
1148 		prt_bdevname(&buf, sb->bdev);
1149 		prt_str(&buf, " and ");
1150 		prt_bdevname(&buf, fs->bdev);
1151 		prt_char(&buf, ':');
1152 		prt_newline(&buf);
1153 
1154 		prt_bdevname(&buf, fs->bdev);
1155 		prt_str(&buf, " believes seq of ");
1156 		prt_bdevname(&buf, sb->bdev);
1157 		prt_printf(&buf, " to be %llu, but ", seq_from_fs);
1158 		prt_bdevname(&buf, sb->bdev);
1159 		prt_printf(&buf, " has %llu\n", seq_from_member);
1160 
1161 		if (!opts->no_splitbrain_check) {
1162 			prt_str(&buf, "Not using ");
1163 			prt_bdevname(&buf, sb->bdev);
1164 		}
1165 
1166 		pr_err("%s", buf.buf);
1167 		printbuf_exit(&buf);
1168 
1169 		if (!opts->no_splitbrain_check)
1170 			return -BCH_ERR_device_splitbrain;
1171 	}
1172 
1173 	return 0;
1174 }
1175 
1176 /* Device startup/shutdown: */
1177 
1178 static void bch2_dev_release(struct kobject *kobj)
1179 {
1180 	struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj);
1181 
1182 	kfree(ca);
1183 }
1184 
1185 static void bch2_dev_free(struct bch_dev *ca)
1186 {
1187 	cancel_work_sync(&ca->io_error_work);
1188 
1189 	if (ca->kobj.state_in_sysfs &&
1190 	    ca->disk_sb.bdev)
1191 		sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1192 
1193 	if (ca->kobj.state_in_sysfs)
1194 		kobject_del(&ca->kobj);
1195 
1196 	kfree(ca->buckets_nouse);
1197 	bch2_free_super(&ca->disk_sb);
1198 	bch2_dev_allocator_background_exit(ca);
1199 	bch2_dev_journal_exit(ca);
1200 
1201 	free_percpu(ca->io_done);
1202 	bch2_dev_buckets_free(ca);
1203 	free_page((unsigned long) ca->sb_read_scratch);
1204 
1205 	bch2_time_stats_quantiles_exit(&ca->io_latency[WRITE]);
1206 	bch2_time_stats_quantiles_exit(&ca->io_latency[READ]);
1207 
1208 	percpu_ref_exit(&ca->io_ref);
1209 #ifndef CONFIG_BCACHEFS_DEBUG
1210 	percpu_ref_exit(&ca->ref);
1211 #endif
1212 	kobject_put(&ca->kobj);
1213 }
1214 
1215 static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca)
1216 {
1217 
1218 	lockdep_assert_held(&c->state_lock);
1219 
1220 	if (percpu_ref_is_zero(&ca->io_ref))
1221 		return;
1222 
1223 	__bch2_dev_read_only(c, ca);
1224 
1225 	reinit_completion(&ca->io_ref_completion);
1226 	percpu_ref_kill(&ca->io_ref);
1227 	wait_for_completion(&ca->io_ref_completion);
1228 
1229 	if (ca->kobj.state_in_sysfs) {
1230 		sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs");
1231 		sysfs_remove_link(&ca->kobj, "block");
1232 	}
1233 
1234 	bch2_free_super(&ca->disk_sb);
1235 	bch2_dev_journal_exit(ca);
1236 }
1237 
1238 #ifndef CONFIG_BCACHEFS_DEBUG
1239 static void bch2_dev_ref_complete(struct percpu_ref *ref)
1240 {
1241 	struct bch_dev *ca = container_of(ref, struct bch_dev, ref);
1242 
1243 	complete(&ca->ref_completion);
1244 }
1245 #endif
1246 
1247 static void bch2_dev_io_ref_complete(struct percpu_ref *ref)
1248 {
1249 	struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref);
1250 
1251 	complete(&ca->io_ref_completion);
1252 }
1253 
1254 static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca)
1255 {
1256 	int ret;
1257 
1258 	if (!c->kobj.state_in_sysfs)
1259 		return 0;
1260 
1261 	if (!ca->kobj.state_in_sysfs) {
1262 		ret = kobject_add(&ca->kobj, &c->kobj,
1263 				  "dev-%u", ca->dev_idx);
1264 		if (ret)
1265 			return ret;
1266 	}
1267 
1268 	if (ca->disk_sb.bdev) {
1269 		struct kobject *block = bdev_kobj(ca->disk_sb.bdev);
1270 
1271 		ret = sysfs_create_link(block, &ca->kobj, "bcachefs");
1272 		if (ret)
1273 			return ret;
1274 
1275 		ret = sysfs_create_link(&ca->kobj, block, "block");
1276 		if (ret)
1277 			return ret;
1278 	}
1279 
1280 	return 0;
1281 }
1282 
1283 static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c,
1284 					struct bch_member *member)
1285 {
1286 	struct bch_dev *ca;
1287 	unsigned i;
1288 
1289 	ca = kzalloc(sizeof(*ca), GFP_KERNEL);
1290 	if (!ca)
1291 		return NULL;
1292 
1293 	kobject_init(&ca->kobj, &bch2_dev_ktype);
1294 	init_completion(&ca->ref_completion);
1295 	init_completion(&ca->io_ref_completion);
1296 
1297 	init_rwsem(&ca->bucket_lock);
1298 
1299 	INIT_WORK(&ca->io_error_work, bch2_io_error_work);
1300 
1301 	bch2_time_stats_quantiles_init(&ca->io_latency[READ]);
1302 	bch2_time_stats_quantiles_init(&ca->io_latency[WRITE]);
1303 
1304 	ca->mi = bch2_mi_to_cpu(member);
1305 
1306 	for (i = 0; i < ARRAY_SIZE(member->errors); i++)
1307 		atomic64_set(&ca->errors[i], le64_to_cpu(member->errors[i]));
1308 
1309 	ca->uuid = member->uuid;
1310 
1311 	ca->nr_btree_reserve = DIV_ROUND_UP(BTREE_NODE_RESERVE,
1312 			     ca->mi.bucket_size / btree_sectors(c));
1313 
1314 #ifndef CONFIG_BCACHEFS_DEBUG
1315 	if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete, 0, GFP_KERNEL))
1316 		goto err;
1317 #else
1318 	atomic_long_set(&ca->ref, 1);
1319 #endif
1320 
1321 	bch2_dev_allocator_background_init(ca);
1322 
1323 	if (percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete,
1324 			    PERCPU_REF_INIT_DEAD, GFP_KERNEL) ||
1325 	    !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) ||
1326 	    bch2_dev_buckets_alloc(c, ca) ||
1327 	    !(ca->io_done	= alloc_percpu(*ca->io_done)))
1328 		goto err;
1329 
1330 	return ca;
1331 err:
1332 	bch2_dev_free(ca);
1333 	return NULL;
1334 }
1335 
1336 static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca,
1337 			    unsigned dev_idx)
1338 {
1339 	ca->dev_idx = dev_idx;
1340 	__set_bit(ca->dev_idx, ca->self.d);
1341 	scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx);
1342 
1343 	ca->fs = c;
1344 	rcu_assign_pointer(c->devs[ca->dev_idx], ca);
1345 
1346 	if (bch2_dev_sysfs_online(c, ca))
1347 		pr_warn("error creating sysfs objects");
1348 }
1349 
1350 static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx)
1351 {
1352 	struct bch_member member = bch2_sb_member_get(c->disk_sb.sb, dev_idx);
1353 	struct bch_dev *ca = NULL;
1354 	int ret = 0;
1355 
1356 	if (bch2_fs_init_fault("dev_alloc"))
1357 		goto err;
1358 
1359 	ca = __bch2_dev_alloc(c, &member);
1360 	if (!ca)
1361 		goto err;
1362 
1363 	ca->fs = c;
1364 
1365 	bch2_dev_attach(c, ca, dev_idx);
1366 	return ret;
1367 err:
1368 	if (ca)
1369 		bch2_dev_free(ca);
1370 	return -BCH_ERR_ENOMEM_dev_alloc;
1371 }
1372 
1373 static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb)
1374 {
1375 	unsigned ret;
1376 
1377 	if (bch2_dev_is_online(ca)) {
1378 		bch_err(ca, "already have device online in slot %u",
1379 			sb->sb->dev_idx);
1380 		return -BCH_ERR_device_already_online;
1381 	}
1382 
1383 	if (get_capacity(sb->bdev->bd_disk) <
1384 	    ca->mi.bucket_size * ca->mi.nbuckets) {
1385 		bch_err(ca, "cannot online: device too small");
1386 		return -BCH_ERR_device_size_too_small;
1387 	}
1388 
1389 	BUG_ON(!percpu_ref_is_zero(&ca->io_ref));
1390 
1391 	ret = bch2_dev_journal_init(ca, sb->sb);
1392 	if (ret)
1393 		return ret;
1394 
1395 	/* Commit: */
1396 	ca->disk_sb = *sb;
1397 	memset(sb, 0, sizeof(*sb));
1398 
1399 	ca->dev = ca->disk_sb.bdev->bd_dev;
1400 
1401 	percpu_ref_reinit(&ca->io_ref);
1402 
1403 	return 0;
1404 }
1405 
1406 static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb)
1407 {
1408 	struct bch_dev *ca;
1409 	int ret;
1410 
1411 	lockdep_assert_held(&c->state_lock);
1412 
1413 	if (le64_to_cpu(sb->sb->seq) >
1414 	    le64_to_cpu(c->disk_sb.sb->seq))
1415 		bch2_sb_to_fs(c, sb->sb);
1416 
1417 	BUG_ON(!bch2_dev_exists(c, sb->sb->dev_idx));
1418 
1419 	ca = bch2_dev_locked(c, sb->sb->dev_idx);
1420 
1421 	ret = __bch2_dev_attach_bdev(ca, sb);
1422 	if (ret)
1423 		return ret;
1424 
1425 	bch2_dev_sysfs_online(c, ca);
1426 
1427 	struct printbuf name = PRINTBUF;
1428 	prt_bdevname(&name, ca->disk_sb.bdev);
1429 
1430 	if (c->sb.nr_devices == 1)
1431 		strscpy(c->name, name.buf, sizeof(c->name));
1432 	strscpy(ca->name, name.buf, sizeof(ca->name));
1433 
1434 	printbuf_exit(&name);
1435 
1436 	rebalance_wakeup(c);
1437 	return 0;
1438 }
1439 
1440 /* Device management: */
1441 
1442 /*
1443  * Note: this function is also used by the error paths - when a particular
1444  * device sees an error, we call it to determine whether we can just set the
1445  * device RO, or - if this function returns false - we'll set the whole
1446  * filesystem RO:
1447  *
1448  * XXX: maybe we should be more explicit about whether we're changing state
1449  * because we got an error or what have you?
1450  */
1451 bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca,
1452 			    enum bch_member_state new_state, int flags)
1453 {
1454 	struct bch_devs_mask new_online_devs;
1455 	int nr_rw = 0, required;
1456 
1457 	lockdep_assert_held(&c->state_lock);
1458 
1459 	switch (new_state) {
1460 	case BCH_MEMBER_STATE_rw:
1461 		return true;
1462 	case BCH_MEMBER_STATE_ro:
1463 		if (ca->mi.state != BCH_MEMBER_STATE_rw)
1464 			return true;
1465 
1466 		/* do we have enough devices to write to?  */
1467 		for_each_member_device(c, ca2)
1468 			if (ca2 != ca)
1469 				nr_rw += ca2->mi.state == BCH_MEMBER_STATE_rw;
1470 
1471 		required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED)
1472 			       ? c->opts.metadata_replicas
1473 			       : metadata_replicas_required(c),
1474 			       !(flags & BCH_FORCE_IF_DATA_DEGRADED)
1475 			       ? c->opts.data_replicas
1476 			       : data_replicas_required(c));
1477 
1478 		return nr_rw >= required;
1479 	case BCH_MEMBER_STATE_failed:
1480 	case BCH_MEMBER_STATE_spare:
1481 		if (ca->mi.state != BCH_MEMBER_STATE_rw &&
1482 		    ca->mi.state != BCH_MEMBER_STATE_ro)
1483 			return true;
1484 
1485 		/* do we have enough devices to read from?  */
1486 		new_online_devs = bch2_online_devs(c);
1487 		__clear_bit(ca->dev_idx, new_online_devs.d);
1488 
1489 		return bch2_have_enough_devs(c, new_online_devs, flags, false);
1490 	default:
1491 		BUG();
1492 	}
1493 }
1494 
1495 static bool bch2_fs_may_start(struct bch_fs *c)
1496 {
1497 	struct bch_dev *ca;
1498 	unsigned i, flags = 0;
1499 
1500 	if (c->opts.very_degraded)
1501 		flags |= BCH_FORCE_IF_DEGRADED|BCH_FORCE_IF_LOST;
1502 
1503 	if (c->opts.degraded)
1504 		flags |= BCH_FORCE_IF_DEGRADED;
1505 
1506 	if (!c->opts.degraded &&
1507 	    !c->opts.very_degraded) {
1508 		mutex_lock(&c->sb_lock);
1509 
1510 		for (i = 0; i < c->disk_sb.sb->nr_devices; i++) {
1511 			if (!bch2_member_exists(c->disk_sb.sb, i))
1512 				continue;
1513 
1514 			ca = bch2_dev_locked(c, i);
1515 
1516 			if (!bch2_dev_is_online(ca) &&
1517 			    (ca->mi.state == BCH_MEMBER_STATE_rw ||
1518 			     ca->mi.state == BCH_MEMBER_STATE_ro)) {
1519 				mutex_unlock(&c->sb_lock);
1520 				return false;
1521 			}
1522 		}
1523 		mutex_unlock(&c->sb_lock);
1524 	}
1525 
1526 	return bch2_have_enough_devs(c, bch2_online_devs(c), flags, true);
1527 }
1528 
1529 static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca)
1530 {
1531 	/*
1532 	 * The allocator thread itself allocates btree nodes, so stop it first:
1533 	 */
1534 	bch2_dev_allocator_remove(c, ca);
1535 	bch2_recalc_capacity(c);
1536 	bch2_dev_journal_stop(&c->journal, ca);
1537 }
1538 
1539 static void __bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca)
1540 {
1541 	lockdep_assert_held(&c->state_lock);
1542 
1543 	BUG_ON(ca->mi.state != BCH_MEMBER_STATE_rw);
1544 
1545 	bch2_dev_allocator_add(c, ca);
1546 	bch2_recalc_capacity(c);
1547 	bch2_dev_do_discards(ca);
1548 }
1549 
1550 int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1551 			 enum bch_member_state new_state, int flags)
1552 {
1553 	struct bch_member *m;
1554 	int ret = 0;
1555 
1556 	if (ca->mi.state == new_state)
1557 		return 0;
1558 
1559 	if (!bch2_dev_state_allowed(c, ca, new_state, flags))
1560 		return -BCH_ERR_device_state_not_allowed;
1561 
1562 	if (new_state != BCH_MEMBER_STATE_rw)
1563 		__bch2_dev_read_only(c, ca);
1564 
1565 	bch_notice(ca, "%s", bch2_member_states[new_state]);
1566 
1567 	mutex_lock(&c->sb_lock);
1568 	m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
1569 	SET_BCH_MEMBER_STATE(m, new_state);
1570 	bch2_write_super(c);
1571 	mutex_unlock(&c->sb_lock);
1572 
1573 	if (new_state == BCH_MEMBER_STATE_rw)
1574 		__bch2_dev_read_write(c, ca);
1575 
1576 	rebalance_wakeup(c);
1577 
1578 	return ret;
1579 }
1580 
1581 int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca,
1582 		       enum bch_member_state new_state, int flags)
1583 {
1584 	int ret;
1585 
1586 	down_write(&c->state_lock);
1587 	ret = __bch2_dev_set_state(c, ca, new_state, flags);
1588 	up_write(&c->state_lock);
1589 
1590 	return ret;
1591 }
1592 
1593 /* Device add/removal: */
1594 
1595 static int bch2_dev_remove_alloc(struct bch_fs *c, struct bch_dev *ca)
1596 {
1597 	struct bpos start	= POS(ca->dev_idx, 0);
1598 	struct bpos end		= POS(ca->dev_idx, U64_MAX);
1599 	int ret;
1600 
1601 	/*
1602 	 * We clear the LRU and need_discard btrees first so that we don't race
1603 	 * with bch2_do_invalidates() and bch2_do_discards()
1604 	 */
1605 	ret =   bch2_btree_delete_range(c, BTREE_ID_lru, start, end,
1606 					BTREE_TRIGGER_norun, NULL) ?:
1607 		bch2_btree_delete_range(c, BTREE_ID_need_discard, start, end,
1608 					BTREE_TRIGGER_norun, NULL) ?:
1609 		bch2_btree_delete_range(c, BTREE_ID_freespace, start, end,
1610 					BTREE_TRIGGER_norun, NULL) ?:
1611 		bch2_btree_delete_range(c, BTREE_ID_backpointers, start, end,
1612 					BTREE_TRIGGER_norun, NULL) ?:
1613 		bch2_btree_delete_range(c, BTREE_ID_alloc, start, end,
1614 					BTREE_TRIGGER_norun, NULL) ?:
1615 		bch2_btree_delete_range(c, BTREE_ID_bucket_gens, start, end,
1616 					BTREE_TRIGGER_norun, NULL) ?:
1617 		bch2_dev_usage_remove(c, ca->dev_idx);
1618 	bch_err_msg(c, ret, "removing dev alloc info");
1619 	return ret;
1620 }
1621 
1622 int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags)
1623 {
1624 	struct bch_member *m;
1625 	unsigned dev_idx = ca->dev_idx, data;
1626 	int ret;
1627 
1628 	down_write(&c->state_lock);
1629 
1630 	/*
1631 	 * We consume a reference to ca->ref, regardless of whether we succeed
1632 	 * or fail:
1633 	 */
1634 	bch2_dev_put(ca);
1635 
1636 	if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1637 		bch_err(ca, "Cannot remove without losing data");
1638 		ret = -BCH_ERR_device_state_not_allowed;
1639 		goto err;
1640 	}
1641 
1642 	__bch2_dev_read_only(c, ca);
1643 
1644 	ret = bch2_dev_data_drop(c, ca->dev_idx, flags);
1645 	bch_err_msg(ca, ret, "bch2_dev_data_drop()");
1646 	if (ret)
1647 		goto err;
1648 
1649 	ret = bch2_dev_remove_alloc(c, ca);
1650 	bch_err_msg(ca, ret, "bch2_dev_remove_alloc()");
1651 	if (ret)
1652 		goto err;
1653 
1654 	/*
1655 	 * We need to flush the entire journal to get rid of keys that reference
1656 	 * the device being removed before removing the superblock entry
1657 	 */
1658 	bch2_journal_flush_all_pins(&c->journal);
1659 
1660 	/*
1661 	 * this is really just needed for the bch2_replicas_gc_(start|end)
1662 	 * calls, and could be cleaned up:
1663 	 */
1664 	ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx);
1665 	bch_err_msg(ca, ret, "bch2_journal_flush_device_pins()");
1666 	if (ret)
1667 		goto err;
1668 
1669 	ret = bch2_journal_flush(&c->journal);
1670 	bch_err_msg(ca, ret, "bch2_journal_flush()");
1671 	if (ret)
1672 		goto err;
1673 
1674 	ret = bch2_replicas_gc2(c);
1675 	bch_err_msg(ca, ret, "bch2_replicas_gc2()");
1676 	if (ret)
1677 		goto err;
1678 
1679 	data = bch2_dev_has_data(c, ca);
1680 	if (data) {
1681 		struct printbuf data_has = PRINTBUF;
1682 
1683 		prt_bitflags(&data_has, __bch2_data_types, data);
1684 		bch_err(ca, "Remove failed, still has data (%s)", data_has.buf);
1685 		printbuf_exit(&data_has);
1686 		ret = -EBUSY;
1687 		goto err;
1688 	}
1689 
1690 	__bch2_dev_offline(c, ca);
1691 
1692 	mutex_lock(&c->sb_lock);
1693 	rcu_assign_pointer(c->devs[ca->dev_idx], NULL);
1694 	mutex_unlock(&c->sb_lock);
1695 
1696 #ifndef CONFIG_BCACHEFS_DEBUG
1697 	percpu_ref_kill(&ca->ref);
1698 #else
1699 	ca->dying = true;
1700 	bch2_dev_put(ca);
1701 #endif
1702 	wait_for_completion(&ca->ref_completion);
1703 
1704 	bch2_dev_free(ca);
1705 
1706 	/*
1707 	 * Free this device's slot in the bch_member array - all pointers to
1708 	 * this device must be gone:
1709 	 */
1710 	mutex_lock(&c->sb_lock);
1711 	m = bch2_members_v2_get_mut(c->disk_sb.sb, dev_idx);
1712 	memset(&m->uuid, 0, sizeof(m->uuid));
1713 
1714 	bch2_write_super(c);
1715 
1716 	mutex_unlock(&c->sb_lock);
1717 	up_write(&c->state_lock);
1718 	return 0;
1719 err:
1720 	if (ca->mi.state == BCH_MEMBER_STATE_rw &&
1721 	    !percpu_ref_is_zero(&ca->io_ref))
1722 		__bch2_dev_read_write(c, ca);
1723 	up_write(&c->state_lock);
1724 	return ret;
1725 }
1726 
1727 /* Add new device to running filesystem: */
1728 int bch2_dev_add(struct bch_fs *c, const char *path)
1729 {
1730 	struct bch_opts opts = bch2_opts_empty();
1731 	struct bch_sb_handle sb;
1732 	struct bch_dev *ca = NULL;
1733 	struct bch_sb_field_members_v2 *mi;
1734 	struct bch_member dev_mi;
1735 	unsigned dev_idx, nr_devices, u64s;
1736 	struct printbuf errbuf = PRINTBUF;
1737 	struct printbuf label = PRINTBUF;
1738 	int ret;
1739 
1740 	ret = bch2_read_super(path, &opts, &sb);
1741 	bch_err_msg(c, ret, "reading super");
1742 	if (ret)
1743 		goto err;
1744 
1745 	dev_mi = bch2_sb_member_get(sb.sb, sb.sb->dev_idx);
1746 
1747 	if (BCH_MEMBER_GROUP(&dev_mi)) {
1748 		bch2_disk_path_to_text_sb(&label, sb.sb, BCH_MEMBER_GROUP(&dev_mi) - 1);
1749 		if (label.allocation_failure) {
1750 			ret = -ENOMEM;
1751 			goto err;
1752 		}
1753 	}
1754 
1755 	ret = bch2_dev_may_add(sb.sb, c);
1756 	if (ret)
1757 		goto err;
1758 
1759 	ca = __bch2_dev_alloc(c, &dev_mi);
1760 	if (!ca) {
1761 		ret = -ENOMEM;
1762 		goto err;
1763 	}
1764 
1765 	ret = __bch2_dev_attach_bdev(ca, &sb);
1766 	if (ret)
1767 		goto err;
1768 
1769 	ret = bch2_dev_journal_alloc(ca, true);
1770 	bch_err_msg(c, ret, "allocating journal");
1771 	if (ret)
1772 		goto err;
1773 
1774 	down_write(&c->state_lock);
1775 	mutex_lock(&c->sb_lock);
1776 
1777 	ret = bch2_sb_from_fs(c, ca);
1778 	bch_err_msg(c, ret, "setting up new superblock");
1779 	if (ret)
1780 		goto err_unlock;
1781 
1782 	if (dynamic_fault("bcachefs:add:no_slot"))
1783 		goto no_slot;
1784 
1785 	if (c->sb.nr_devices < BCH_SB_MEMBERS_MAX) {
1786 		dev_idx = c->sb.nr_devices;
1787 		goto have_slot;
1788 	}
1789 
1790 	int best = -1;
1791 	u64 best_last_mount = 0;
1792 	for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++) {
1793 		struct bch_member m = bch2_sb_member_get(c->disk_sb.sb, dev_idx);
1794 		if (bch2_member_alive(&m))
1795 			continue;
1796 
1797 		u64 last_mount = le64_to_cpu(m.last_mount);
1798 		if (best < 0 || last_mount < best_last_mount) {
1799 			best = dev_idx;
1800 			best_last_mount = last_mount;
1801 		}
1802 	}
1803 	if (best >= 0) {
1804 		dev_idx = best;
1805 		goto have_slot;
1806 	}
1807 no_slot:
1808 	ret = -BCH_ERR_ENOSPC_sb_members;
1809 	bch_err_msg(c, ret, "setting up new superblock");
1810 	goto err_unlock;
1811 
1812 have_slot:
1813 	nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices);
1814 
1815 	mi = bch2_sb_field_get(c->disk_sb.sb, members_v2);
1816 	u64s = DIV_ROUND_UP(sizeof(struct bch_sb_field_members_v2) +
1817 			    le16_to_cpu(mi->member_bytes) * nr_devices, sizeof(u64));
1818 
1819 	mi = bch2_sb_field_resize(&c->disk_sb, members_v2, u64s);
1820 	if (!mi) {
1821 		ret = -BCH_ERR_ENOSPC_sb_members;
1822 		bch_err_msg(c, ret, "setting up new superblock");
1823 		goto err_unlock;
1824 	}
1825 	struct bch_member *m = bch2_members_v2_get_mut(c->disk_sb.sb, dev_idx);
1826 
1827 	/* success: */
1828 
1829 	*m = dev_mi;
1830 	m->last_mount = cpu_to_le64(ktime_get_real_seconds());
1831 	c->disk_sb.sb->nr_devices	= nr_devices;
1832 
1833 	ca->disk_sb.sb->dev_idx	= dev_idx;
1834 	bch2_dev_attach(c, ca, dev_idx);
1835 
1836 	if (BCH_MEMBER_GROUP(&dev_mi)) {
1837 		ret = __bch2_dev_group_set(c, ca, label.buf);
1838 		bch_err_msg(c, ret, "creating new label");
1839 		if (ret)
1840 			goto err_unlock;
1841 	}
1842 
1843 	bch2_write_super(c);
1844 	mutex_unlock(&c->sb_lock);
1845 
1846 	ret = bch2_dev_usage_init(ca, false);
1847 	if (ret)
1848 		goto err_late;
1849 
1850 	ret = bch2_trans_mark_dev_sb(c, ca, BTREE_TRIGGER_transactional);
1851 	bch_err_msg(ca, ret, "marking new superblock");
1852 	if (ret)
1853 		goto err_late;
1854 
1855 	ret = bch2_fs_freespace_init(c);
1856 	bch_err_msg(ca, ret, "initializing free space");
1857 	if (ret)
1858 		goto err_late;
1859 
1860 	ca->new_fs_bucket_idx = 0;
1861 
1862 	if (ca->mi.state == BCH_MEMBER_STATE_rw)
1863 		__bch2_dev_read_write(c, ca);
1864 
1865 	up_write(&c->state_lock);
1866 	return 0;
1867 
1868 err_unlock:
1869 	mutex_unlock(&c->sb_lock);
1870 	up_write(&c->state_lock);
1871 err:
1872 	if (ca)
1873 		bch2_dev_free(ca);
1874 	bch2_free_super(&sb);
1875 	printbuf_exit(&label);
1876 	printbuf_exit(&errbuf);
1877 	bch_err_fn(c, ret);
1878 	return ret;
1879 err_late:
1880 	up_write(&c->state_lock);
1881 	ca = NULL;
1882 	goto err;
1883 }
1884 
1885 /* Hot add existing device to running filesystem: */
1886 int bch2_dev_online(struct bch_fs *c, const char *path)
1887 {
1888 	struct bch_opts opts = bch2_opts_empty();
1889 	struct bch_sb_handle sb = { NULL };
1890 	struct bch_dev *ca;
1891 	unsigned dev_idx;
1892 	int ret;
1893 
1894 	down_write(&c->state_lock);
1895 
1896 	ret = bch2_read_super(path, &opts, &sb);
1897 	if (ret) {
1898 		up_write(&c->state_lock);
1899 		return ret;
1900 	}
1901 
1902 	dev_idx = sb.sb->dev_idx;
1903 
1904 	ret = bch2_dev_in_fs(&c->disk_sb, &sb, &c->opts);
1905 	bch_err_msg(c, ret, "bringing %s online", path);
1906 	if (ret)
1907 		goto err;
1908 
1909 	ret = bch2_dev_attach_bdev(c, &sb);
1910 	if (ret)
1911 		goto err;
1912 
1913 	ca = bch2_dev_locked(c, dev_idx);
1914 
1915 	ret = bch2_trans_mark_dev_sb(c, ca, BTREE_TRIGGER_transactional);
1916 	bch_err_msg(c, ret, "bringing %s online: error from bch2_trans_mark_dev_sb", path);
1917 	if (ret)
1918 		goto err;
1919 
1920 	if (ca->mi.state == BCH_MEMBER_STATE_rw)
1921 		__bch2_dev_read_write(c, ca);
1922 
1923 	if (!ca->mi.freespace_initialized) {
1924 		ret = bch2_dev_freespace_init(c, ca, 0, ca->mi.nbuckets);
1925 		bch_err_msg(ca, ret, "initializing free space");
1926 		if (ret)
1927 			goto err;
1928 	}
1929 
1930 	if (!ca->journal.nr) {
1931 		ret = bch2_dev_journal_alloc(ca, false);
1932 		bch_err_msg(ca, ret, "allocating journal");
1933 		if (ret)
1934 			goto err;
1935 	}
1936 
1937 	mutex_lock(&c->sb_lock);
1938 	bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx)->last_mount =
1939 		cpu_to_le64(ktime_get_real_seconds());
1940 	bch2_write_super(c);
1941 	mutex_unlock(&c->sb_lock);
1942 
1943 	up_write(&c->state_lock);
1944 	return 0;
1945 err:
1946 	up_write(&c->state_lock);
1947 	bch2_free_super(&sb);
1948 	return ret;
1949 }
1950 
1951 int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags)
1952 {
1953 	down_write(&c->state_lock);
1954 
1955 	if (!bch2_dev_is_online(ca)) {
1956 		bch_err(ca, "Already offline");
1957 		up_write(&c->state_lock);
1958 		return 0;
1959 	}
1960 
1961 	if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_failed, flags)) {
1962 		bch_err(ca, "Cannot offline required disk");
1963 		up_write(&c->state_lock);
1964 		return -BCH_ERR_device_state_not_allowed;
1965 	}
1966 
1967 	__bch2_dev_offline(c, ca);
1968 
1969 	up_write(&c->state_lock);
1970 	return 0;
1971 }
1972 
1973 int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets)
1974 {
1975 	struct bch_member *m;
1976 	u64 old_nbuckets;
1977 	int ret = 0;
1978 
1979 	down_write(&c->state_lock);
1980 	old_nbuckets = ca->mi.nbuckets;
1981 
1982 	if (nbuckets < ca->mi.nbuckets) {
1983 		bch_err(ca, "Cannot shrink yet");
1984 		ret = -EINVAL;
1985 		goto err;
1986 	}
1987 
1988 	if (nbuckets > BCH_MEMBER_NBUCKETS_MAX) {
1989 		bch_err(ca, "New device size too big (%llu greater than max %u)",
1990 			nbuckets, BCH_MEMBER_NBUCKETS_MAX);
1991 		ret = -BCH_ERR_device_size_too_big;
1992 		goto err;
1993 	}
1994 
1995 	if (bch2_dev_is_online(ca) &&
1996 	    get_capacity(ca->disk_sb.bdev->bd_disk) <
1997 	    ca->mi.bucket_size * nbuckets) {
1998 		bch_err(ca, "New size larger than device");
1999 		ret = -BCH_ERR_device_size_too_small;
2000 		goto err;
2001 	}
2002 
2003 	ret = bch2_dev_buckets_resize(c, ca, nbuckets);
2004 	bch_err_msg(ca, ret, "resizing buckets");
2005 	if (ret)
2006 		goto err;
2007 
2008 	ret = bch2_trans_mark_dev_sb(c, ca, BTREE_TRIGGER_transactional);
2009 	if (ret)
2010 		goto err;
2011 
2012 	mutex_lock(&c->sb_lock);
2013 	m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
2014 	m->nbuckets = cpu_to_le64(nbuckets);
2015 
2016 	bch2_write_super(c);
2017 	mutex_unlock(&c->sb_lock);
2018 
2019 	if (ca->mi.freespace_initialized) {
2020 		struct disk_accounting_pos acc = {
2021 			.type = BCH_DISK_ACCOUNTING_dev_data_type,
2022 			.dev_data_type.dev = ca->dev_idx,
2023 			.dev_data_type.data_type = BCH_DATA_free,
2024 		};
2025 		u64 v[3] = { nbuckets - old_nbuckets, 0, 0 };
2026 
2027 		ret   = bch2_trans_do(ca->fs, NULL, NULL, 0,
2028 				bch2_disk_accounting_mod(trans, &acc, v, ARRAY_SIZE(v), false)) ?:
2029 			bch2_dev_freespace_init(c, ca, old_nbuckets, nbuckets);
2030 		if (ret)
2031 			goto err;
2032 	}
2033 
2034 	bch2_recalc_capacity(c);
2035 err:
2036 	up_write(&c->state_lock);
2037 	return ret;
2038 }
2039 
2040 /* return with ref on ca->ref: */
2041 struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *name)
2042 {
2043 	if (!strncmp(name, "/dev/", strlen("/dev/")))
2044 		name += strlen("/dev/");
2045 
2046 	for_each_member_device(c, ca)
2047 		if (!strcmp(name, ca->name))
2048 			return ca;
2049 	return ERR_PTR(-BCH_ERR_ENOENT_dev_not_found);
2050 }
2051 
2052 /* Filesystem open: */
2053 
2054 static inline int sb_cmp(struct bch_sb *l, struct bch_sb *r)
2055 {
2056 	return  cmp_int(le64_to_cpu(l->seq), le64_to_cpu(r->seq)) ?:
2057 		cmp_int(le64_to_cpu(l->write_time), le64_to_cpu(r->write_time));
2058 }
2059 
2060 struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices,
2061 			    struct bch_opts opts)
2062 {
2063 	DARRAY(struct bch_sb_handle) sbs = { 0 };
2064 	struct bch_fs *c = NULL;
2065 	struct bch_sb_handle *best = NULL;
2066 	struct printbuf errbuf = PRINTBUF;
2067 	int ret = 0;
2068 
2069 	if (!try_module_get(THIS_MODULE))
2070 		return ERR_PTR(-ENODEV);
2071 
2072 	if (!nr_devices) {
2073 		ret = -EINVAL;
2074 		goto err;
2075 	}
2076 
2077 	ret = darray_make_room(&sbs, nr_devices);
2078 	if (ret)
2079 		goto err;
2080 
2081 	for (unsigned i = 0; i < nr_devices; i++) {
2082 		struct bch_sb_handle sb = { NULL };
2083 
2084 		ret = bch2_read_super(devices[i], &opts, &sb);
2085 		if (ret)
2086 			goto err;
2087 
2088 		BUG_ON(darray_push(&sbs, sb));
2089 	}
2090 
2091 	if (opts.nochanges && !opts.read_only) {
2092 		ret = -BCH_ERR_erofs_nochanges;
2093 		goto err_print;
2094 	}
2095 
2096 	darray_for_each(sbs, sb)
2097 		if (!best || sb_cmp(sb->sb, best->sb) > 0)
2098 			best = sb;
2099 
2100 	darray_for_each_reverse(sbs, sb) {
2101 		ret = bch2_dev_in_fs(best, sb, &opts);
2102 
2103 		if (ret == -BCH_ERR_device_has_been_removed ||
2104 		    ret == -BCH_ERR_device_splitbrain) {
2105 			bch2_free_super(sb);
2106 			darray_remove_item(&sbs, sb);
2107 			best -= best > sb;
2108 			ret = 0;
2109 			continue;
2110 		}
2111 
2112 		if (ret)
2113 			goto err_print;
2114 	}
2115 
2116 	c = bch2_fs_alloc(best->sb, opts);
2117 	ret = PTR_ERR_OR_ZERO(c);
2118 	if (ret)
2119 		goto err;
2120 
2121 	down_write(&c->state_lock);
2122 	darray_for_each(sbs, sb) {
2123 		ret = bch2_dev_attach_bdev(c, sb);
2124 		if (ret) {
2125 			up_write(&c->state_lock);
2126 			goto err;
2127 		}
2128 	}
2129 	up_write(&c->state_lock);
2130 
2131 	if (!bch2_fs_may_start(c)) {
2132 		ret = -BCH_ERR_insufficient_devices_to_start;
2133 		goto err_print;
2134 	}
2135 
2136 	if (!c->opts.nostart) {
2137 		ret = bch2_fs_start(c);
2138 		if (ret)
2139 			goto err;
2140 	}
2141 out:
2142 	darray_for_each(sbs, sb)
2143 		bch2_free_super(sb);
2144 	darray_exit(&sbs);
2145 	printbuf_exit(&errbuf);
2146 	module_put(THIS_MODULE);
2147 	return c;
2148 err_print:
2149 	pr_err("bch_fs_open err opening %s: %s",
2150 	       devices[0], bch2_err_str(ret));
2151 err:
2152 	if (!IS_ERR_OR_NULL(c))
2153 		bch2_fs_stop(c);
2154 	c = ERR_PTR(ret);
2155 	goto out;
2156 }
2157 
2158 /* Global interfaces/init */
2159 
2160 static void bcachefs_exit(void)
2161 {
2162 	bch2_debug_exit();
2163 	bch2_vfs_exit();
2164 	bch2_chardev_exit();
2165 	bch2_btree_key_cache_exit();
2166 	if (bcachefs_kset)
2167 		kset_unregister(bcachefs_kset);
2168 }
2169 
2170 static int __init bcachefs_init(void)
2171 {
2172 	bch2_bkey_pack_test();
2173 
2174 	if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) ||
2175 	    bch2_btree_key_cache_init() ||
2176 	    bch2_chardev_init() ||
2177 	    bch2_vfs_init() ||
2178 	    bch2_debug_init())
2179 		goto err;
2180 
2181 	return 0;
2182 err:
2183 	bcachefs_exit();
2184 	return -ENOMEM;
2185 }
2186 
2187 #define BCH_DEBUG_PARAM(name, description)			\
2188 	bool bch2_##name;					\
2189 	module_param_named(name, bch2_##name, bool, 0644);	\
2190 	MODULE_PARM_DESC(name, description);
2191 BCH_DEBUG_PARAMS()
2192 #undef BCH_DEBUG_PARAM
2193 
2194 __maybe_unused
2195 static unsigned bch2_metadata_version = bcachefs_metadata_version_current;
2196 module_param_named(version, bch2_metadata_version, uint, 0400);
2197 
2198 module_exit(bcachefs_exit);
2199 module_init(bcachefs_init);
2200