Commit | Line | Data |
---|---|---|
1c6fdbd8 KO |
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" | |
7b3f84ea KO |
11 | #include "alloc_background.h" |
12 | #include "alloc_foreground.h" | |
5b8a9227 | 13 | #include "bkey_sort.h" |
1c6fdbd8 KO |
14 | #include "btree_cache.h" |
15 | #include "btree_gc.h" | |
16 | #include "btree_update_interior.h" | |
17 | #include "btree_io.h" | |
18 | #include "chardev.h" | |
19 | #include "checksum.h" | |
20 | #include "clock.h" | |
21 | #include "compress.h" | |
22 | #include "debug.h" | |
23 | #include "disk_groups.h" | |
cd575ddf | 24 | #include "ec.h" |
1c6fdbd8 KO |
25 | #include "error.h" |
26 | #include "fs.h" | |
27 | #include "fs-io.h" | |
28 | #include "fsck.h" | |
29 | #include "inode.h" | |
30 | #include "io.h" | |
31 | #include "journal.h" | |
32 | #include "journal_reclaim.h" | |
1dd7f9d9 | 33 | #include "journal_seq_blacklist.h" |
1c6fdbd8 KO |
34 | #include "move.h" |
35 | #include "migrate.h" | |
36 | #include "movinggc.h" | |
37 | #include "quota.h" | |
38 | #include "rebalance.h" | |
39 | #include "recovery.h" | |
40 | #include "replicas.h" | |
41 | #include "super.h" | |
42 | #include "super-io.h" | |
43 | #include "sysfs.h" | |
44 | #include "trace.h" | |
45 | ||
46 | #include <linux/backing-dev.h> | |
47 | #include <linux/blkdev.h> | |
48 | #include <linux/debugfs.h> | |
49 | #include <linux/device.h> | |
50 | #include <linux/idr.h> | |
51 | #include <linux/kthread.h> | |
52 | #include <linux/module.h> | |
53 | #include <linux/percpu.h> | |
54 | #include <linux/random.h> | |
55 | #include <linux/sysfs.h> | |
56 | #include <crypto/hash.h> | |
57 | ||
58 | MODULE_LICENSE("GPL"); | |
59 | MODULE_AUTHOR("Kent Overstreet <kent.overstreet@gmail.com>"); | |
60 | ||
61 | #define KTYPE(type) \ | |
62 | static const struct attribute_group type ## _group = { \ | |
63 | .attrs = type ## _files \ | |
64 | }; \ | |
65 | \ | |
66 | static const struct attribute_group *type ## _groups[] = { \ | |
67 | &type ## _group, \ | |
68 | NULL \ | |
69 | }; \ | |
70 | \ | |
71 | static const struct kobj_type type ## _ktype = { \ | |
72 | .release = type ## _release, \ | |
73 | .sysfs_ops = &type ## _sysfs_ops, \ | |
74 | .default_groups = type ## _groups \ | |
75 | } | |
76 | ||
77 | static void bch2_fs_release(struct kobject *); | |
78 | static void bch2_dev_release(struct kobject *); | |
79 | ||
80 | static void bch2_fs_internal_release(struct kobject *k) | |
81 | { | |
82 | } | |
83 | ||
84 | static void bch2_fs_opts_dir_release(struct kobject *k) | |
85 | { | |
86 | } | |
87 | ||
88 | static void bch2_fs_time_stats_release(struct kobject *k) | |
89 | { | |
90 | } | |
91 | ||
92 | KTYPE(bch2_fs); | |
93 | KTYPE(bch2_fs_internal); | |
94 | KTYPE(bch2_fs_opts_dir); | |
95 | KTYPE(bch2_fs_time_stats); | |
96 | KTYPE(bch2_dev); | |
97 | ||
98 | static struct kset *bcachefs_kset; | |
99 | static LIST_HEAD(bch_fs_list); | |
100 | static DEFINE_MUTEX(bch_fs_list_lock); | |
101 | ||
102 | static DECLARE_WAIT_QUEUE_HEAD(bch_read_only_wait); | |
103 | ||
104 | static void bch2_dev_free(struct bch_dev *); | |
105 | static int bch2_dev_alloc(struct bch_fs *, unsigned); | |
106 | static int bch2_dev_sysfs_online(struct bch_fs *, struct bch_dev *); | |
107 | static void __bch2_dev_read_only(struct bch_fs *, struct bch_dev *); | |
108 | ||
109 | struct bch_fs *bch2_dev_to_fs(dev_t dev) | |
110 | { | |
111 | struct bch_fs *c; | |
112 | struct bch_dev *ca; | |
113 | unsigned i; | |
114 | ||
115 | mutex_lock(&bch_fs_list_lock); | |
116 | rcu_read_lock(); | |
117 | ||
118 | list_for_each_entry(c, &bch_fs_list, list) | |
119 | for_each_member_device_rcu(ca, c, i, NULL) | |
120 | if (ca->disk_sb.bdev->bd_dev == dev) { | |
121 | closure_get(&c->cl); | |
122 | goto found; | |
123 | } | |
124 | c = NULL; | |
125 | found: | |
126 | rcu_read_unlock(); | |
127 | mutex_unlock(&bch_fs_list_lock); | |
128 | ||
129 | return c; | |
130 | } | |
131 | ||
132 | static struct bch_fs *__bch2_uuid_to_fs(__uuid_t uuid) | |
133 | { | |
134 | struct bch_fs *c; | |
135 | ||
136 | lockdep_assert_held(&bch_fs_list_lock); | |
137 | ||
138 | list_for_each_entry(c, &bch_fs_list, list) | |
139 | if (!memcmp(&c->disk_sb.sb->uuid, &uuid, sizeof(uuid))) | |
140 | return c; | |
141 | ||
142 | return NULL; | |
143 | } | |
144 | ||
145 | struct bch_fs *bch2_uuid_to_fs(__uuid_t uuid) | |
146 | { | |
147 | struct bch_fs *c; | |
148 | ||
149 | mutex_lock(&bch_fs_list_lock); | |
150 | c = __bch2_uuid_to_fs(uuid); | |
151 | if (c) | |
152 | closure_get(&c->cl); | |
153 | mutex_unlock(&bch_fs_list_lock); | |
154 | ||
155 | return c; | |
156 | } | |
157 | ||
158 | /* Filesystem RO/RW: */ | |
159 | ||
160 | /* | |
161 | * For startup/shutdown of RW stuff, the dependencies are: | |
162 | * | |
163 | * - foreground writes depend on copygc and rebalance (to free up space) | |
164 | * | |
165 | * - copygc and rebalance depend on mark and sweep gc (they actually probably | |
166 | * don't because they either reserve ahead of time or don't block if | |
167 | * allocations fail, but allocations can require mark and sweep gc to run | |
168 | * because of generation number wraparound) | |
169 | * | |
170 | * - all of the above depends on the allocator threads | |
171 | * | |
172 | * - allocator depends on the journal (when it rewrites prios and gens) | |
173 | */ | |
174 | ||
175 | static void __bch2_fs_read_only(struct bch_fs *c) | |
176 | { | |
177 | struct bch_dev *ca; | |
430735cd | 178 | bool wrote; |
d5f70c1f | 179 | unsigned i, clean_passes = 0; |
430735cd | 180 | int ret; |
1c6fdbd8 KO |
181 | |
182 | bch2_rebalance_stop(c); | |
183 | ||
184 | for_each_member_device(ca, c, i) | |
185 | bch2_copygc_stop(ca); | |
186 | ||
187 | bch2_gc_thread_stop(c); | |
188 | ||
189 | /* | |
190 | * Flush journal before stopping allocators, because flushing journal | |
191 | * blacklist entries involves allocating new btree nodes: | |
192 | */ | |
193 | bch2_journal_flush_all_pins(&c->journal); | |
194 | ||
b935a8a6 KO |
195 | if (!test_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags)) |
196 | goto allocator_not_running; | |
197 | ||
430735cd | 198 | do { |
a0e0bda1 KO |
199 | wrote = false; |
200 | ||
460651ee KO |
201 | ret = bch2_stripes_write(c, BTREE_INSERT_NOCHECK_RW, &wrote) ?: |
202 | bch2_alloc_write(c, BTREE_INSERT_NOCHECK_RW, &wrote); | |
430735cd | 203 | |
460651ee | 204 | if (ret && !test_bit(BCH_FS_EMERGENCY_RO, &c->flags)) |
d5f70c1f | 205 | bch2_fs_inconsistent(c, "error writing out alloc info %i", ret); |
460651ee KO |
206 | |
207 | if (ret) | |
61c8d7c8 | 208 | break; |
61c8d7c8 | 209 | |
430735cd KO |
210 | for_each_member_device(ca, c, i) |
211 | bch2_dev_allocator_quiesce(c, ca); | |
212 | ||
213 | bch2_journal_flush_all_pins(&c->journal); | |
214 | ||
215 | /* | |
216 | * We need to explicitly wait on btree interior updates to complete | |
217 | * before stopping the journal, flushing all journal pins isn't | |
218 | * sufficient, because in the BTREE_INTERIOR_UPDATING_ROOT case btree | |
219 | * interior updates have to drop their journal pin before they're | |
220 | * fully complete: | |
221 | */ | |
222 | closure_wait_event(&c->btree_interior_update_wait, | |
223 | !bch2_btree_interior_updates_nr_pending(c)); | |
d5f70c1f KO |
224 | |
225 | clean_passes = wrote ? 0 : clean_passes + 1; | |
226 | } while (clean_passes < 2); | |
b935a8a6 | 227 | allocator_not_running: |
1c6fdbd8 KO |
228 | for_each_member_device(ca, c, i) |
229 | bch2_dev_allocator_stop(ca); | |
230 | ||
b935a8a6 KO |
231 | clear_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags); |
232 | ||
1c6fdbd8 KO |
233 | bch2_fs_journal_stop(&c->journal); |
234 | ||
430735cd KO |
235 | /* XXX: mark super that alloc info is persistent */ |
236 | ||
1c6fdbd8 KO |
237 | /* |
238 | * the journal kicks off btree writes via reclaim - wait for in flight | |
239 | * writes after stopping journal: | |
240 | */ | |
241 | if (test_bit(BCH_FS_EMERGENCY_RO, &c->flags)) | |
242 | bch2_btree_flush_all_writes(c); | |
243 | else | |
244 | bch2_btree_verify_flushed(c); | |
245 | ||
246 | /* | |
247 | * After stopping journal: | |
248 | */ | |
249 | for_each_member_device(ca, c, i) | |
250 | bch2_dev_allocator_remove(c, ca); | |
251 | } | |
252 | ||
253 | static void bch2_writes_disabled(struct percpu_ref *writes) | |
254 | { | |
255 | struct bch_fs *c = container_of(writes, struct bch_fs, writes); | |
256 | ||
257 | set_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags); | |
258 | wake_up(&bch_read_only_wait); | |
259 | } | |
260 | ||
261 | void bch2_fs_read_only(struct bch_fs *c) | |
262 | { | |
134915f3 KO |
263 | if (!test_bit(BCH_FS_RW, &c->flags)) { |
264 | cancel_delayed_work_sync(&c->journal.reclaim_work); | |
1c6fdbd8 | 265 | return; |
134915f3 | 266 | } |
1c6fdbd8 KO |
267 | |
268 | BUG_ON(test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags)); | |
269 | ||
270 | /* | |
271 | * Block new foreground-end write operations from starting - any new | |
272 | * writes will return -EROFS: | |
273 | * | |
274 | * (This is really blocking new _allocations_, writes to previously | |
275 | * allocated space can still happen until stopping the allocator in | |
276 | * bch2_dev_allocator_stop()). | |
277 | */ | |
278 | percpu_ref_kill(&c->writes); | |
279 | ||
f516c872 | 280 | cancel_work_sync(&c->ec_stripe_delete_work); |
1c6fdbd8 KO |
281 | cancel_delayed_work(&c->pd_controllers_update); |
282 | ||
283 | /* | |
284 | * If we're not doing an emergency shutdown, we want to wait on | |
285 | * outstanding writes to complete so they don't see spurious errors due | |
286 | * to shutting down the allocator: | |
287 | * | |
288 | * If we are doing an emergency shutdown outstanding writes may | |
289 | * hang until we shutdown the allocator so we don't want to wait | |
290 | * on outstanding writes before shutting everything down - but | |
291 | * we do need to wait on them before returning and signalling | |
292 | * that going RO is complete: | |
293 | */ | |
294 | wait_event(bch_read_only_wait, | |
295 | test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags) || | |
296 | test_bit(BCH_FS_EMERGENCY_RO, &c->flags)); | |
297 | ||
298 | __bch2_fs_read_only(c); | |
299 | ||
300 | wait_event(bch_read_only_wait, | |
301 | test_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags)); | |
302 | ||
303 | clear_bit(BCH_FS_WRITE_DISABLE_COMPLETE, &c->flags); | |
304 | ||
305 | if (!bch2_journal_error(&c->journal) && | |
306 | !test_bit(BCH_FS_ERROR, &c->flags) && | |
3aea4342 | 307 | !test_bit(BCH_FS_EMERGENCY_RO, &c->flags) && |
a0e0bda1 | 308 | test_bit(BCH_FS_STARTED, &c->flags) && |
a0e0bda1 | 309 | !c->opts.norecovery) |
134915f3 | 310 | bch2_fs_mark_clean(c); |
1c6fdbd8 | 311 | |
134915f3 | 312 | clear_bit(BCH_FS_RW, &c->flags); |
1c6fdbd8 KO |
313 | } |
314 | ||
315 | static void bch2_fs_read_only_work(struct work_struct *work) | |
316 | { | |
317 | struct bch_fs *c = | |
318 | container_of(work, struct bch_fs, read_only_work); | |
319 | ||
320 | mutex_lock(&c->state_lock); | |
321 | bch2_fs_read_only(c); | |
322 | mutex_unlock(&c->state_lock); | |
323 | } | |
324 | ||
325 | static void bch2_fs_read_only_async(struct bch_fs *c) | |
326 | { | |
327 | queue_work(system_long_wq, &c->read_only_work); | |
328 | } | |
329 | ||
330 | bool bch2_fs_emergency_read_only(struct bch_fs *c) | |
331 | { | |
332 | bool ret = !test_and_set_bit(BCH_FS_EMERGENCY_RO, &c->flags); | |
333 | ||
334 | bch2_fs_read_only_async(c); | |
335 | bch2_journal_halt(&c->journal); | |
336 | ||
337 | wake_up(&bch_read_only_wait); | |
338 | return ret; | |
339 | } | |
340 | ||
134915f3 | 341 | static int bch2_fs_read_write_late(struct bch_fs *c) |
1c6fdbd8 KO |
342 | { |
343 | struct bch_dev *ca; | |
1c6fdbd8 | 344 | unsigned i; |
134915f3 | 345 | int ret; |
1c6fdbd8 | 346 | |
134915f3 KO |
347 | ret = bch2_gc_thread_start(c); |
348 | if (ret) { | |
349 | bch_err(c, "error starting gc thread"); | |
350 | return ret; | |
351 | } | |
352 | ||
353 | for_each_rw_member(ca, c, i) { | |
354 | ret = bch2_copygc_start(c, ca); | |
355 | if (ret) { | |
356 | bch_err(c, "error starting copygc threads"); | |
357 | percpu_ref_put(&ca->io_ref); | |
358 | return ret; | |
359 | } | |
360 | } | |
361 | ||
362 | ret = bch2_rebalance_start(c); | |
363 | if (ret) { | |
364 | bch_err(c, "error starting rebalance thread"); | |
365 | return ret; | |
366 | } | |
367 | ||
368 | schedule_delayed_work(&c->pd_controllers_update, 5 * HZ); | |
369 | ||
97fd13ad KO |
370 | schedule_work(&c->ec_stripe_delete_work); |
371 | ||
134915f3 KO |
372 | return 0; |
373 | } | |
1c6fdbd8 | 374 | |
03e183cb | 375 | int __bch2_fs_read_write(struct bch_fs *c, bool early) |
134915f3 KO |
376 | { |
377 | struct bch_dev *ca; | |
378 | unsigned i; | |
379 | int ret; | |
380 | ||
381 | if (test_bit(BCH_FS_RW, &c->flags)) | |
382 | return 0; | |
383 | ||
619f5bee KO |
384 | /* |
385 | * nochanges is used for fsck -n mode - we have to allow going rw | |
386 | * during recovery for that to work: | |
387 | */ | |
388 | if (c->opts.norecovery || | |
389 | (c->opts.nochanges && | |
390 | (!early || c->opts.read_only))) | |
391 | return -EROFS; | |
330581f1 | 392 | |
134915f3 KO |
393 | ret = bch2_fs_mark_dirty(c); |
394 | if (ret) | |
395 | goto err; | |
1c6fdbd8 KO |
396 | |
397 | for_each_rw_member(ca, c, i) | |
398 | bch2_dev_allocator_add(c, ca); | |
399 | bch2_recalc_capacity(c); | |
400 | ||
134915f3 KO |
401 | if (!test_bit(BCH_FS_ALLOCATOR_STARTED, &c->flags)) { |
402 | ret = bch2_fs_allocator_start(c); | |
403 | if (ret) { | |
404 | bch_err(c, "error initializing allocator"); | |
1c6fdbd8 KO |
405 | goto err; |
406 | } | |
407 | ||
134915f3 KO |
408 | set_bit(BCH_FS_ALLOCATOR_STARTED, &c->flags); |
409 | } | |
1c6fdbd8 | 410 | |
134915f3 KO |
411 | for_each_rw_member(ca, c, i) { |
412 | ret = bch2_dev_allocator_start(ca); | |
413 | if (ret) { | |
414 | bch_err(c, "error starting allocator threads"); | |
1c6fdbd8 KO |
415 | percpu_ref_put(&ca->io_ref); |
416 | goto err; | |
417 | } | |
134915f3 | 418 | } |
1c6fdbd8 | 419 | |
134915f3 | 420 | set_bit(BCH_FS_ALLOCATOR_RUNNING, &c->flags); |
1c6fdbd8 | 421 | |
134915f3 KO |
422 | if (!early) { |
423 | ret = bch2_fs_read_write_late(c); | |
424 | if (ret) | |
425 | goto err; | |
426 | } | |
1c6fdbd8 | 427 | |
134915f3 KO |
428 | percpu_ref_reinit(&c->writes); |
429 | set_bit(BCH_FS_RW, &c->flags); | |
1c6fdbd8 | 430 | |
134915f3 KO |
431 | queue_delayed_work(c->journal_reclaim_wq, |
432 | &c->journal.reclaim_work, 0); | |
433 | return 0; | |
1c6fdbd8 KO |
434 | err: |
435 | __bch2_fs_read_only(c); | |
134915f3 KO |
436 | return ret; |
437 | } | |
438 | ||
439 | int bch2_fs_read_write(struct bch_fs *c) | |
440 | { | |
441 | return __bch2_fs_read_write(c, false); | |
442 | } | |
443 | ||
444 | int bch2_fs_read_write_early(struct bch_fs *c) | |
445 | { | |
446 | lockdep_assert_held(&c->state_lock); | |
447 | ||
134915f3 | 448 | return __bch2_fs_read_write(c, true); |
1c6fdbd8 KO |
449 | } |
450 | ||
451 | /* Filesystem startup/shutdown: */ | |
452 | ||
453 | static void bch2_fs_free(struct bch_fs *c) | |
454 | { | |
455 | unsigned i; | |
456 | ||
457 | for (i = 0; i < BCH_TIME_STAT_NR; i++) | |
458 | bch2_time_stats_exit(&c->times[i]); | |
459 | ||
460 | bch2_fs_quota_exit(c); | |
461 | bch2_fs_fsio_exit(c); | |
cd575ddf | 462 | bch2_fs_ec_exit(c); |
1c6fdbd8 KO |
463 | bch2_fs_encryption_exit(c); |
464 | bch2_fs_io_exit(c); | |
36e9d698 | 465 | bch2_fs_btree_iter_exit(c); |
1c6fdbd8 KO |
466 | bch2_fs_btree_cache_exit(c); |
467 | bch2_fs_journal_exit(&c->journal); | |
468 | bch2_io_clock_exit(&c->io_clock[WRITE]); | |
469 | bch2_io_clock_exit(&c->io_clock[READ]); | |
470 | bch2_fs_compress_exit(c); | |
9166b41d | 471 | percpu_free_rwsem(&c->mark_lock); |
5e82a9a1 | 472 | free_percpu(c->online_reserved); |
4d8100da | 473 | kfree(c->usage_scratch); |
5e82a9a1 | 474 | free_percpu(c->usage[1]); |
9ca53b55 | 475 | free_percpu(c->usage[0]); |
5e82a9a1 | 476 | kfree(c->usage_base); |
5663a415 | 477 | free_percpu(c->pcpu); |
1c6fdbd8 KO |
478 | mempool_exit(&c->btree_bounce_pool); |
479 | bioset_exit(&c->btree_bio); | |
480 | mempool_exit(&c->btree_interior_update_pool); | |
481 | mempool_exit(&c->btree_reserve_pool); | |
482 | mempool_exit(&c->fill_iter); | |
483 | percpu_ref_exit(&c->writes); | |
73e6ab95 KO |
484 | kfree(c->replicas.entries); |
485 | kfree(c->replicas_gc.entries); | |
1c6fdbd8 | 486 | kfree(rcu_dereference_protected(c->disk_groups, 1)); |
1dd7f9d9 | 487 | kfree(c->journal_seq_blacklist_table); |
1c6fdbd8 | 488 | |
0519b72d KO |
489 | if (c->journal_reclaim_wq) |
490 | destroy_workqueue(c->journal_reclaim_wq); | |
1c6fdbd8 KO |
491 | if (c->copygc_wq) |
492 | destroy_workqueue(c->copygc_wq); | |
493 | if (c->wq) | |
494 | destroy_workqueue(c->wq); | |
495 | ||
496 | free_pages((unsigned long) c->disk_sb.sb, | |
497 | c->disk_sb.page_order); | |
498 | kvpfree(c, sizeof(*c)); | |
499 | module_put(THIS_MODULE); | |
500 | } | |
501 | ||
502 | static void bch2_fs_release(struct kobject *kobj) | |
503 | { | |
504 | struct bch_fs *c = container_of(kobj, struct bch_fs, kobj); | |
505 | ||
506 | bch2_fs_free(c); | |
507 | } | |
508 | ||
509 | void bch2_fs_stop(struct bch_fs *c) | |
510 | { | |
511 | struct bch_dev *ca; | |
512 | unsigned i; | |
513 | ||
af1c6871 KO |
514 | bch_verbose(c, "shutting down"); |
515 | ||
1dd7f9d9 KO |
516 | set_bit(BCH_FS_STOPPING, &c->flags); |
517 | ||
518 | cancel_work_sync(&c->journal_seq_blacklist_gc_work); | |
519 | ||
1c6fdbd8 KO |
520 | for_each_member_device(ca, c, i) |
521 | if (ca->kobj.state_in_sysfs && | |
522 | ca->disk_sb.bdev) | |
523 | sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs"); | |
524 | ||
525 | if (c->kobj.state_in_sysfs) | |
526 | kobject_del(&c->kobj); | |
527 | ||
528 | bch2_fs_debug_exit(c); | |
529 | bch2_fs_chardev_exit(c); | |
530 | ||
531 | kobject_put(&c->time_stats); | |
532 | kobject_put(&c->opts_dir); | |
533 | kobject_put(&c->internal); | |
534 | ||
535 | mutex_lock(&bch_fs_list_lock); | |
536 | list_del(&c->list); | |
537 | mutex_unlock(&bch_fs_list_lock); | |
538 | ||
539 | closure_sync(&c->cl); | |
540 | closure_debug_destroy(&c->cl); | |
541 | ||
542 | mutex_lock(&c->state_lock); | |
543 | bch2_fs_read_only(c); | |
544 | mutex_unlock(&c->state_lock); | |
545 | ||
546 | /* btree prefetch might have kicked off reads in the background: */ | |
547 | bch2_btree_flush_all_reads(c); | |
548 | ||
549 | for_each_member_device(ca, c, i) | |
550 | cancel_work_sync(&ca->io_error_work); | |
551 | ||
552 | cancel_work_sync(&c->btree_write_error_work); | |
553 | cancel_delayed_work_sync(&c->pd_controllers_update); | |
554 | cancel_work_sync(&c->read_only_work); | |
555 | ||
556 | for (i = 0; i < c->sb.nr_devices; i++) | |
557 | if (c->devs[i]) | |
558 | bch2_dev_free(rcu_dereference_protected(c->devs[i], 1)); | |
559 | ||
af1c6871 KO |
560 | bch_verbose(c, "shutdown complete"); |
561 | ||
1c6fdbd8 KO |
562 | kobject_put(&c->kobj); |
563 | } | |
564 | ||
565 | static const char *bch2_fs_online(struct bch_fs *c) | |
566 | { | |
567 | struct bch_dev *ca; | |
568 | const char *err = NULL; | |
569 | unsigned i; | |
570 | int ret; | |
571 | ||
572 | lockdep_assert_held(&bch_fs_list_lock); | |
573 | ||
574 | if (!list_empty(&c->list)) | |
575 | return NULL; | |
576 | ||
577 | if (__bch2_uuid_to_fs(c->sb.uuid)) | |
578 | return "filesystem UUID already open"; | |
579 | ||
580 | ret = bch2_fs_chardev_init(c); | |
581 | if (ret) | |
582 | return "error creating character device"; | |
583 | ||
584 | bch2_fs_debug_init(c); | |
585 | ||
586 | if (kobject_add(&c->kobj, NULL, "%pU", c->sb.user_uuid.b) || | |
587 | kobject_add(&c->internal, &c->kobj, "internal") || | |
588 | kobject_add(&c->opts_dir, &c->kobj, "options") || | |
589 | kobject_add(&c->time_stats, &c->kobj, "time_stats") || | |
590 | bch2_opts_create_sysfs_files(&c->opts_dir)) | |
591 | return "error creating sysfs objects"; | |
592 | ||
593 | mutex_lock(&c->state_lock); | |
594 | ||
595 | err = "error creating sysfs objects"; | |
596 | __for_each_member_device(ca, c, i, NULL) | |
597 | if (bch2_dev_sysfs_online(c, ca)) | |
598 | goto err; | |
599 | ||
600 | list_add(&c->list, &bch_fs_list); | |
601 | err = NULL; | |
602 | err: | |
603 | mutex_unlock(&c->state_lock); | |
604 | return err; | |
605 | } | |
606 | ||
607 | static struct bch_fs *bch2_fs_alloc(struct bch_sb *sb, struct bch_opts opts) | |
608 | { | |
609 | struct bch_sb_field_members *mi; | |
610 | struct bch_fs *c; | |
ecf37a4a | 611 | unsigned i, iter_size; |
1c6fdbd8 KO |
612 | const char *err; |
613 | ||
614 | pr_verbose_init(opts, ""); | |
615 | ||
616 | c = kvpmalloc(sizeof(struct bch_fs), GFP_KERNEL|__GFP_ZERO); | |
617 | if (!c) | |
618 | goto out; | |
619 | ||
620 | __module_get(THIS_MODULE); | |
621 | ||
622 | c->minor = -1; | |
623 | c->disk_sb.fs_sb = true; | |
624 | ||
625 | mutex_init(&c->state_lock); | |
626 | mutex_init(&c->sb_lock); | |
627 | mutex_init(&c->replicas_gc_lock); | |
628 | mutex_init(&c->btree_root_lock); | |
629 | INIT_WORK(&c->read_only_work, bch2_fs_read_only_work); | |
630 | ||
631 | init_rwsem(&c->gc_lock); | |
632 | ||
633 | for (i = 0; i < BCH_TIME_STAT_NR; i++) | |
634 | bch2_time_stats_init(&c->times[i]); | |
635 | ||
b092dadd KO |
636 | bch2_fs_allocator_background_init(c); |
637 | bch2_fs_allocator_foreground_init(c); | |
1c6fdbd8 KO |
638 | bch2_fs_rebalance_init(c); |
639 | bch2_fs_quota_init(c); | |
640 | ||
641 | INIT_LIST_HEAD(&c->list); | |
642 | ||
643 | INIT_LIST_HEAD(&c->btree_interior_update_list); | |
644 | mutex_init(&c->btree_reserve_cache_lock); | |
645 | mutex_init(&c->btree_interior_update_lock); | |
646 | ||
4d8100da KO |
647 | mutex_init(&c->usage_scratch_lock); |
648 | ||
1c6fdbd8 KO |
649 | mutex_init(&c->bio_bounce_pages_lock); |
650 | ||
651 | bio_list_init(&c->btree_write_error_list); | |
652 | spin_lock_init(&c->btree_write_error_lock); | |
653 | INIT_WORK(&c->btree_write_error_work, bch2_btree_write_error_work); | |
654 | ||
1dd7f9d9 KO |
655 | INIT_WORK(&c->journal_seq_blacklist_gc_work, |
656 | bch2_blacklist_entries_gc); | |
657 | ||
1c6fdbd8 KO |
658 | INIT_LIST_HEAD(&c->fsck_errors); |
659 | mutex_init(&c->fsck_error_lock); | |
660 | ||
cd575ddf KO |
661 | INIT_LIST_HEAD(&c->ec_new_stripe_list); |
662 | mutex_init(&c->ec_new_stripe_lock); | |
dfe9bfb3 | 663 | mutex_init(&c->ec_stripe_create_lock); |
cd575ddf KO |
664 | spin_lock_init(&c->ec_stripes_heap_lock); |
665 | ||
1c6fdbd8 KO |
666 | seqcount_init(&c->gc_pos_lock); |
667 | ||
5e82a9a1 KO |
668 | seqcount_init(&c->usage_lock); |
669 | ||
1c6fdbd8 KO |
670 | c->copy_gc_enabled = 1; |
671 | c->rebalance.enabled = 1; | |
672 | c->promote_whole_extents = true; | |
673 | ||
674 | c->journal.write_time = &c->times[BCH_TIME_journal_write]; | |
675 | c->journal.delay_time = &c->times[BCH_TIME_journal_delay]; | |
49a67206 | 676 | c->journal.blocked_time = &c->times[BCH_TIME_blocked_journal]; |
1c6fdbd8 KO |
677 | c->journal.flush_seq_time = &c->times[BCH_TIME_journal_flush_seq]; |
678 | ||
679 | bch2_fs_btree_cache_init_early(&c->btree_cache); | |
680 | ||
fca1223c KO |
681 | mutex_init(&c->sectors_available_lock); |
682 | ||
73e6ab95 KO |
683 | if (percpu_init_rwsem(&c->mark_lock)) |
684 | goto err; | |
685 | ||
1c6fdbd8 KO |
686 | mutex_lock(&c->sb_lock); |
687 | ||
688 | if (bch2_sb_to_fs(c, sb)) { | |
689 | mutex_unlock(&c->sb_lock); | |
690 | goto err; | |
691 | } | |
692 | ||
693 | mutex_unlock(&c->sb_lock); | |
694 | ||
695 | scnprintf(c->name, sizeof(c->name), "%pU", &c->sb.user_uuid); | |
696 | ||
697 | c->opts = bch2_opts_default; | |
698 | bch2_opts_apply(&c->opts, bch2_opts_from_sb(sb)); | |
699 | bch2_opts_apply(&c->opts, opts); | |
700 | ||
701 | c->block_bits = ilog2(c->opts.block_size); | |
702 | c->btree_foreground_merge_threshold = BTREE_FOREGROUND_MERGE_THRESHOLD(c); | |
703 | ||
1c6fdbd8 KO |
704 | if (bch2_fs_init_fault("fs_alloc")) |
705 | goto err; | |
706 | ||
707 | iter_size = sizeof(struct btree_node_iter_large) + | |
708 | (btree_blocks(c) + 1) * 2 * | |
709 | sizeof(struct btree_node_iter_set); | |
710 | ||
711 | if (!(c->wq = alloc_workqueue("bcachefs", | |
712 | WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_HIGHPRI, 1)) || | |
713 | !(c->copygc_wq = alloc_workqueue("bcache_copygc", | |
714 | WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_HIGHPRI, 1)) || | |
0519b72d KO |
715 | !(c->journal_reclaim_wq = alloc_workqueue("bcache_journal", |
716 | WQ_FREEZABLE|WQ_MEM_RECLAIM|WQ_HIGHPRI, 1)) || | |
134915f3 KO |
717 | percpu_ref_init(&c->writes, bch2_writes_disabled, |
718 | PERCPU_REF_INIT_DEAD, GFP_KERNEL) || | |
1c6fdbd8 KO |
719 | mempool_init_kmalloc_pool(&c->btree_reserve_pool, 1, |
720 | sizeof(struct btree_reserve)) || | |
721 | mempool_init_kmalloc_pool(&c->btree_interior_update_pool, 1, | |
722 | sizeof(struct btree_update)) || | |
723 | mempool_init_kmalloc_pool(&c->fill_iter, 1, iter_size) || | |
724 | bioset_init(&c->btree_bio, 1, | |
725 | max(offsetof(struct btree_read_bio, bio), | |
726 | offsetof(struct btree_write_bio, wbio.bio)), | |
727 | BIOSET_NEED_BVECS) || | |
5663a415 | 728 | !(c->pcpu = alloc_percpu(struct bch_fs_pcpu)) || |
5e82a9a1 | 729 | !(c->online_reserved = alloc_percpu(u64)) || |
1c6fdbd8 KO |
730 | mempool_init_kvpmalloc_pool(&c->btree_bounce_pool, 1, |
731 | btree_bytes(c)) || | |
732 | bch2_io_clock_init(&c->io_clock[READ]) || | |
733 | bch2_io_clock_init(&c->io_clock[WRITE]) || | |
734 | bch2_fs_journal_init(&c->journal) || | |
2c5af169 | 735 | bch2_fs_replicas_init(c) || |
1c6fdbd8 | 736 | bch2_fs_btree_cache_init(c) || |
36e9d698 | 737 | bch2_fs_btree_iter_init(c) || |
1c6fdbd8 KO |
738 | bch2_fs_io_init(c) || |
739 | bch2_fs_encryption_init(c) || | |
740 | bch2_fs_compress_init(c) || | |
cd575ddf | 741 | bch2_fs_ec_init(c) || |
1c6fdbd8 KO |
742 | bch2_fs_fsio_init(c)) |
743 | goto err; | |
744 | ||
745 | mi = bch2_sb_get_members(c->disk_sb.sb); | |
746 | for (i = 0; i < c->sb.nr_devices; i++) | |
747 | if (bch2_dev_exists(c->disk_sb.sb, mi, i) && | |
748 | bch2_dev_alloc(c, i)) | |
749 | goto err; | |
750 | ||
751 | /* | |
752 | * Now that all allocations have succeeded, init various refcounty | |
753 | * things that let us shutdown: | |
754 | */ | |
755 | closure_init(&c->cl, NULL); | |
756 | ||
757 | c->kobj.kset = bcachefs_kset; | |
758 | kobject_init(&c->kobj, &bch2_fs_ktype); | |
759 | kobject_init(&c->internal, &bch2_fs_internal_ktype); | |
760 | kobject_init(&c->opts_dir, &bch2_fs_opts_dir_ktype); | |
761 | kobject_init(&c->time_stats, &bch2_fs_time_stats_ktype); | |
762 | ||
763 | mutex_lock(&bch_fs_list_lock); | |
764 | err = bch2_fs_online(c); | |
765 | mutex_unlock(&bch_fs_list_lock); | |
766 | if (err) { | |
767 | bch_err(c, "bch2_fs_online() error: %s", err); | |
768 | goto err; | |
769 | } | |
770 | out: | |
771 | pr_verbose_init(opts, "ret %i", c ? 0 : -ENOMEM); | |
772 | return c; | |
773 | err: | |
774 | bch2_fs_free(c); | |
775 | c = NULL; | |
776 | goto out; | |
777 | } | |
778 | ||
619f5bee KO |
779 | noinline_for_stack |
780 | static void print_mount_opts(struct bch_fs *c) | |
781 | { | |
782 | enum bch_opt_id i; | |
783 | char buf[512]; | |
784 | struct printbuf p = PBUF(buf); | |
785 | bool first = true; | |
786 | ||
787 | strcpy(buf, "(null)"); | |
788 | ||
789 | if (c->opts.read_only) { | |
790 | pr_buf(&p, "ro"); | |
791 | first = false; | |
792 | } | |
793 | ||
794 | for (i = 0; i < bch2_opts_nr; i++) { | |
795 | const struct bch_option *opt = &bch2_opt_table[i]; | |
796 | u64 v = bch2_opt_get_by_id(&c->opts, i); | |
797 | ||
798 | if (!(opt->mode & OPT_MOUNT)) | |
799 | continue; | |
800 | ||
801 | if (v == bch2_opt_get_by_id(&bch2_opts_default, i)) | |
802 | continue; | |
803 | ||
804 | if (!first) | |
805 | pr_buf(&p, ","); | |
806 | first = false; | |
807 | bch2_opt_to_text(&p, c, opt, v, OPT_SHOW_MOUNT_STYLE); | |
808 | } | |
809 | ||
810 | bch_info(c, "mounted with opts: %s", buf); | |
811 | } | |
812 | ||
813 | int bch2_fs_start(struct bch_fs *c) | |
1c6fdbd8 KO |
814 | { |
815 | const char *err = "cannot allocate memory"; | |
816 | struct bch_sb_field_members *mi; | |
817 | struct bch_dev *ca; | |
a420eea6 | 818 | time64_t now = ktime_get_real_seconds(); |
1c6fdbd8 KO |
819 | unsigned i; |
820 | int ret = -EINVAL; | |
821 | ||
822 | mutex_lock(&c->state_lock); | |
823 | ||
134915f3 | 824 | BUG_ON(test_bit(BCH_FS_STARTED, &c->flags)); |
1c6fdbd8 KO |
825 | |
826 | mutex_lock(&c->sb_lock); | |
827 | ||
828 | for_each_online_member(ca, c, i) | |
829 | bch2_sb_from_fs(c, ca); | |
830 | ||
831 | mi = bch2_sb_get_members(c->disk_sb.sb); | |
832 | for_each_online_member(ca, c, i) | |
833 | mi->members[ca->dev_idx].last_mount = cpu_to_le64(now); | |
834 | ||
835 | mutex_unlock(&c->sb_lock); | |
836 | ||
837 | for_each_rw_member(ca, c, i) | |
838 | bch2_dev_allocator_add(c, ca); | |
839 | bch2_recalc_capacity(c); | |
840 | ||
841 | ret = BCH_SB_INITIALIZED(c->disk_sb.sb) | |
842 | ? bch2_fs_recovery(c) | |
843 | : bch2_fs_initialize(c); | |
844 | if (ret) | |
845 | goto err; | |
cd575ddf KO |
846 | |
847 | ret = bch2_opts_check_may_set(c); | |
848 | if (ret) | |
849 | goto err; | |
1c6fdbd8 KO |
850 | |
851 | err = "dynamic fault"; | |
619f5bee | 852 | ret = -EINVAL; |
1c6fdbd8 KO |
853 | if (bch2_fs_init_fault("fs_start")) |
854 | goto err; | |
855 | ||
619f5bee | 856 | if (c->opts.read_only || c->opts.nochanges) { |
1c6fdbd8 KO |
857 | bch2_fs_read_only(c); |
858 | } else { | |
619f5bee KO |
859 | err = "error going read write"; |
860 | ret = !test_bit(BCH_FS_RW, &c->flags) | |
861 | ? bch2_fs_read_write(c) | |
862 | : bch2_fs_read_write_late(c); | |
863 | if (ret) | |
1c6fdbd8 KO |
864 | goto err; |
865 | } | |
866 | ||
867 | set_bit(BCH_FS_STARTED, &c->flags); | |
619f5bee KO |
868 | print_mount_opts(c); |
869 | ret = 0; | |
1c6fdbd8 KO |
870 | out: |
871 | mutex_unlock(&c->state_lock); | |
619f5bee | 872 | return ret; |
1c6fdbd8 KO |
873 | err: |
874 | switch (ret) { | |
875 | case BCH_FSCK_ERRORS_NOT_FIXED: | |
876 | bch_err(c, "filesystem contains errors: please report this to the developers"); | |
877 | pr_cont("mount with -o fix_errors to repair\n"); | |
878 | err = "fsck error"; | |
879 | break; | |
880 | case BCH_FSCK_REPAIR_UNIMPLEMENTED: | |
881 | bch_err(c, "filesystem contains errors: please report this to the developers"); | |
882 | pr_cont("repair unimplemented: inform the developers so that it can be added\n"); | |
883 | err = "fsck error"; | |
884 | break; | |
885 | case BCH_FSCK_REPAIR_IMPOSSIBLE: | |
886 | bch_err(c, "filesystem contains errors, but repair impossible"); | |
887 | err = "fsck error"; | |
888 | break; | |
889 | case BCH_FSCK_UNKNOWN_VERSION: | |
890 | err = "unknown metadata version";; | |
891 | break; | |
892 | case -ENOMEM: | |
893 | err = "cannot allocate memory"; | |
894 | break; | |
895 | case -EIO: | |
896 | err = "IO error"; | |
897 | break; | |
898 | } | |
899 | ||
9516950c KO |
900 | if (ret >= 0) |
901 | ret = -EIO; | |
1c6fdbd8 KO |
902 | goto out; |
903 | } | |
904 | ||
905 | static const char *bch2_dev_may_add(struct bch_sb *sb, struct bch_fs *c) | |
906 | { | |
907 | struct bch_sb_field_members *sb_mi; | |
908 | ||
909 | sb_mi = bch2_sb_get_members(sb); | |
910 | if (!sb_mi) | |
911 | return "Invalid superblock: member info area missing"; | |
912 | ||
913 | if (le16_to_cpu(sb->block_size) != c->opts.block_size) | |
914 | return "mismatched block size"; | |
915 | ||
916 | if (le16_to_cpu(sb_mi->members[sb->dev_idx].bucket_size) < | |
917 | BCH_SB_BTREE_NODE_SIZE(c->disk_sb.sb)) | |
918 | return "new cache bucket size is too small"; | |
919 | ||
920 | return NULL; | |
921 | } | |
922 | ||
923 | static const char *bch2_dev_in_fs(struct bch_sb *fs, struct bch_sb *sb) | |
924 | { | |
925 | struct bch_sb *newest = | |
926 | le64_to_cpu(fs->seq) > le64_to_cpu(sb->seq) ? fs : sb; | |
927 | struct bch_sb_field_members *mi = bch2_sb_get_members(newest); | |
928 | ||
929 | if (!uuid_equal(&fs->uuid, &sb->uuid)) | |
930 | return "device not a member of filesystem"; | |
931 | ||
932 | if (!bch2_dev_exists(newest, mi, sb->dev_idx)) | |
933 | return "device has been removed"; | |
934 | ||
935 | if (fs->block_size != sb->block_size) | |
936 | return "mismatched block size"; | |
937 | ||
938 | return NULL; | |
939 | } | |
940 | ||
941 | /* Device startup/shutdown: */ | |
942 | ||
943 | static void bch2_dev_release(struct kobject *kobj) | |
944 | { | |
945 | struct bch_dev *ca = container_of(kobj, struct bch_dev, kobj); | |
946 | ||
947 | kfree(ca); | |
948 | } | |
949 | ||
950 | static void bch2_dev_free(struct bch_dev *ca) | |
951 | { | |
952 | cancel_work_sync(&ca->io_error_work); | |
953 | ||
954 | if (ca->kobj.state_in_sysfs && | |
955 | ca->disk_sb.bdev) | |
956 | sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs"); | |
957 | ||
958 | if (ca->kobj.state_in_sysfs) | |
959 | kobject_del(&ca->kobj); | |
960 | ||
961 | bch2_free_super(&ca->disk_sb); | |
962 | bch2_dev_journal_exit(ca); | |
963 | ||
964 | free_percpu(ca->io_done); | |
965 | bioset_exit(&ca->replica_set); | |
966 | bch2_dev_buckets_free(ca); | |
d1170ce5 | 967 | free_page((unsigned long) ca->sb_read_scratch); |
1c6fdbd8 KO |
968 | |
969 | bch2_time_stats_exit(&ca->io_latency[WRITE]); | |
970 | bch2_time_stats_exit(&ca->io_latency[READ]); | |
971 | ||
972 | percpu_ref_exit(&ca->io_ref); | |
973 | percpu_ref_exit(&ca->ref); | |
974 | kobject_put(&ca->kobj); | |
975 | } | |
976 | ||
977 | static void __bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca) | |
978 | { | |
979 | ||
980 | lockdep_assert_held(&c->state_lock); | |
981 | ||
982 | if (percpu_ref_is_zero(&ca->io_ref)) | |
983 | return; | |
984 | ||
985 | __bch2_dev_read_only(c, ca); | |
986 | ||
987 | reinit_completion(&ca->io_ref_completion); | |
988 | percpu_ref_kill(&ca->io_ref); | |
989 | wait_for_completion(&ca->io_ref_completion); | |
990 | ||
991 | if (ca->kobj.state_in_sysfs) { | |
992 | sysfs_remove_link(bdev_kobj(ca->disk_sb.bdev), "bcachefs"); | |
993 | sysfs_remove_link(&ca->kobj, "block"); | |
994 | } | |
995 | ||
996 | bch2_free_super(&ca->disk_sb); | |
997 | bch2_dev_journal_exit(ca); | |
998 | } | |
999 | ||
1000 | static void bch2_dev_ref_complete(struct percpu_ref *ref) | |
1001 | { | |
1002 | struct bch_dev *ca = container_of(ref, struct bch_dev, ref); | |
1003 | ||
1004 | complete(&ca->ref_completion); | |
1005 | } | |
1006 | ||
1007 | static void bch2_dev_io_ref_complete(struct percpu_ref *ref) | |
1008 | { | |
1009 | struct bch_dev *ca = container_of(ref, struct bch_dev, io_ref); | |
1010 | ||
1011 | complete(&ca->io_ref_completion); | |
1012 | } | |
1013 | ||
1014 | static int bch2_dev_sysfs_online(struct bch_fs *c, struct bch_dev *ca) | |
1015 | { | |
1016 | int ret; | |
1017 | ||
1018 | if (!c->kobj.state_in_sysfs) | |
1019 | return 0; | |
1020 | ||
1021 | if (!ca->kobj.state_in_sysfs) { | |
1022 | ret = kobject_add(&ca->kobj, &c->kobj, | |
1023 | "dev-%u", ca->dev_idx); | |
1024 | if (ret) | |
1025 | return ret; | |
1026 | } | |
1027 | ||
1028 | if (ca->disk_sb.bdev) { | |
1029 | struct kobject *block = bdev_kobj(ca->disk_sb.bdev); | |
1030 | ||
1031 | ret = sysfs_create_link(block, &ca->kobj, "bcachefs"); | |
1032 | if (ret) | |
1033 | return ret; | |
1034 | ||
1035 | ret = sysfs_create_link(&ca->kobj, block, "block"); | |
1036 | if (ret) | |
1037 | return ret; | |
1038 | } | |
1039 | ||
1040 | return 0; | |
1041 | } | |
1042 | ||
1043 | static struct bch_dev *__bch2_dev_alloc(struct bch_fs *c, | |
1044 | struct bch_member *member) | |
1045 | { | |
1046 | struct bch_dev *ca; | |
1047 | ||
1048 | ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
1049 | if (!ca) | |
1050 | return NULL; | |
1051 | ||
1052 | kobject_init(&ca->kobj, &bch2_dev_ktype); | |
1053 | init_completion(&ca->ref_completion); | |
1054 | init_completion(&ca->io_ref_completion); | |
1055 | ||
1056 | init_rwsem(&ca->bucket_lock); | |
1057 | ||
1058 | writepoint_init(&ca->copygc_write_point, BCH_DATA_USER); | |
1059 | ||
1060 | spin_lock_init(&ca->freelist_lock); | |
1061 | bch2_dev_copygc_init(ca); | |
1062 | ||
1063 | INIT_WORK(&ca->io_error_work, bch2_io_error_work); | |
1064 | ||
1065 | bch2_time_stats_init(&ca->io_latency[READ]); | |
1066 | bch2_time_stats_init(&ca->io_latency[WRITE]); | |
1067 | ||
1068 | ca->mi = bch2_mi_to_cpu(member); | |
1069 | ca->uuid = member->uuid; | |
1070 | ||
1071 | if (opt_defined(c->opts, discard)) | |
1072 | ca->mi.discard = opt_get(c->opts, discard); | |
1073 | ||
1074 | if (percpu_ref_init(&ca->ref, bch2_dev_ref_complete, | |
1075 | 0, GFP_KERNEL) || | |
1076 | percpu_ref_init(&ca->io_ref, bch2_dev_io_ref_complete, | |
1077 | PERCPU_REF_INIT_DEAD, GFP_KERNEL) || | |
d1170ce5 | 1078 | !(ca->sb_read_scratch = (void *) __get_free_page(GFP_KERNEL)) || |
1c6fdbd8 KO |
1079 | bch2_dev_buckets_alloc(c, ca) || |
1080 | bioset_init(&ca->replica_set, 4, | |
1081 | offsetof(struct bch_write_bio, bio), 0) || | |
1082 | !(ca->io_done = alloc_percpu(*ca->io_done))) | |
1083 | goto err; | |
1084 | ||
1085 | return ca; | |
1086 | err: | |
1087 | bch2_dev_free(ca); | |
1088 | return NULL; | |
1089 | } | |
1090 | ||
1091 | static void bch2_dev_attach(struct bch_fs *c, struct bch_dev *ca, | |
1092 | unsigned dev_idx) | |
1093 | { | |
1094 | ca->dev_idx = dev_idx; | |
1095 | __set_bit(ca->dev_idx, ca->self.d); | |
1096 | scnprintf(ca->name, sizeof(ca->name), "dev-%u", dev_idx); | |
1097 | ||
1098 | ca->fs = c; | |
1099 | rcu_assign_pointer(c->devs[ca->dev_idx], ca); | |
1100 | ||
1101 | if (bch2_dev_sysfs_online(c, ca)) | |
1102 | pr_warn("error creating sysfs objects"); | |
1103 | } | |
1104 | ||
1105 | static int bch2_dev_alloc(struct bch_fs *c, unsigned dev_idx) | |
1106 | { | |
1107 | struct bch_member *member = | |
1108 | bch2_sb_get_members(c->disk_sb.sb)->members + dev_idx; | |
1109 | struct bch_dev *ca = NULL; | |
1110 | int ret = 0; | |
1111 | ||
1112 | pr_verbose_init(c->opts, ""); | |
1113 | ||
1114 | if (bch2_fs_init_fault("dev_alloc")) | |
1115 | goto err; | |
1116 | ||
1117 | ca = __bch2_dev_alloc(c, member); | |
1118 | if (!ca) | |
1119 | goto err; | |
1120 | ||
1121 | bch2_dev_attach(c, ca, dev_idx); | |
1122 | out: | |
1123 | pr_verbose_init(c->opts, "ret %i", ret); | |
1124 | return ret; | |
1125 | err: | |
1126 | if (ca) | |
1127 | bch2_dev_free(ca); | |
1128 | ret = -ENOMEM; | |
1129 | goto out; | |
1130 | } | |
1131 | ||
1132 | static int __bch2_dev_attach_bdev(struct bch_dev *ca, struct bch_sb_handle *sb) | |
1133 | { | |
1134 | unsigned ret; | |
1135 | ||
1136 | if (bch2_dev_is_online(ca)) { | |
1137 | bch_err(ca, "already have device online in slot %u", | |
1138 | sb->sb->dev_idx); | |
1139 | return -EINVAL; | |
1140 | } | |
1141 | ||
1142 | if (get_capacity(sb->bdev->bd_disk) < | |
1143 | ca->mi.bucket_size * ca->mi.nbuckets) { | |
1144 | bch_err(ca, "cannot online: device too small"); | |
1145 | return -EINVAL; | |
1146 | } | |
1147 | ||
1148 | BUG_ON(!percpu_ref_is_zero(&ca->io_ref)); | |
1149 | ||
1150 | if (get_capacity(sb->bdev->bd_disk) < | |
1151 | ca->mi.bucket_size * ca->mi.nbuckets) { | |
1152 | bch_err(ca, "device too small"); | |
1153 | return -EINVAL; | |
1154 | } | |
1155 | ||
1156 | ret = bch2_dev_journal_init(ca, sb->sb); | |
1157 | if (ret) | |
1158 | return ret; | |
1159 | ||
1160 | /* Commit: */ | |
1161 | ca->disk_sb = *sb; | |
1162 | memset(sb, 0, sizeof(*sb)); | |
1163 | ||
1c6fdbd8 KO |
1164 | percpu_ref_reinit(&ca->io_ref); |
1165 | ||
1166 | return 0; | |
1167 | } | |
1168 | ||
1169 | static int bch2_dev_attach_bdev(struct bch_fs *c, struct bch_sb_handle *sb) | |
1170 | { | |
1171 | struct bch_dev *ca; | |
1172 | int ret; | |
1173 | ||
1174 | lockdep_assert_held(&c->state_lock); | |
1175 | ||
1176 | if (le64_to_cpu(sb->sb->seq) > | |
1177 | le64_to_cpu(c->disk_sb.sb->seq)) | |
1178 | bch2_sb_to_fs(c, sb->sb); | |
1179 | ||
1180 | BUG_ON(sb->sb->dev_idx >= c->sb.nr_devices || | |
1181 | !c->devs[sb->sb->dev_idx]); | |
1182 | ||
1183 | ca = bch_dev_locked(c, sb->sb->dev_idx); | |
1184 | ||
1185 | ret = __bch2_dev_attach_bdev(ca, sb); | |
1186 | if (ret) | |
1187 | return ret; | |
1188 | ||
3e0745e2 KO |
1189 | if (test_bit(BCH_FS_ALLOC_READ_DONE, &c->flags) && |
1190 | !percpu_u64_get(&ca->usage[0]->buckets[BCH_DATA_SB])) { | |
1191 | mutex_lock(&c->sb_lock); | |
1192 | bch2_mark_dev_superblock(ca->fs, ca, 0); | |
1193 | mutex_unlock(&c->sb_lock); | |
1194 | } | |
6eac2c2e | 1195 | |
1c6fdbd8 KO |
1196 | bch2_dev_sysfs_online(c, ca); |
1197 | ||
1198 | if (c->sb.nr_devices == 1) | |
1199 | snprintf(c->name, sizeof(c->name), "%pg", ca->disk_sb.bdev); | |
1200 | snprintf(ca->name, sizeof(ca->name), "%pg", ca->disk_sb.bdev); | |
1201 | ||
1202 | rebalance_wakeup(c); | |
1203 | return 0; | |
1204 | } | |
1205 | ||
1206 | /* Device management: */ | |
1207 | ||
1208 | /* | |
1209 | * Note: this function is also used by the error paths - when a particular | |
1210 | * device sees an error, we call it to determine whether we can just set the | |
1211 | * device RO, or - if this function returns false - we'll set the whole | |
1212 | * filesystem RO: | |
1213 | * | |
1214 | * XXX: maybe we should be more explicit about whether we're changing state | |
1215 | * because we got an error or what have you? | |
1216 | */ | |
1217 | bool bch2_dev_state_allowed(struct bch_fs *c, struct bch_dev *ca, | |
1218 | enum bch_member_state new_state, int flags) | |
1219 | { | |
1220 | struct bch_devs_mask new_online_devs; | |
1221 | struct replicas_status s; | |
1222 | struct bch_dev *ca2; | |
1223 | int i, nr_rw = 0, required; | |
1224 | ||
1225 | lockdep_assert_held(&c->state_lock); | |
1226 | ||
1227 | switch (new_state) { | |
1228 | case BCH_MEMBER_STATE_RW: | |
1229 | return true; | |
1230 | case BCH_MEMBER_STATE_RO: | |
1231 | if (ca->mi.state != BCH_MEMBER_STATE_RW) | |
1232 | return true; | |
1233 | ||
1234 | /* do we have enough devices to write to? */ | |
1235 | for_each_member_device(ca2, c, i) | |
1236 | if (ca2 != ca) | |
1237 | nr_rw += ca2->mi.state == BCH_MEMBER_STATE_RW; | |
1238 | ||
1239 | required = max(!(flags & BCH_FORCE_IF_METADATA_DEGRADED) | |
1240 | ? c->opts.metadata_replicas | |
1241 | : c->opts.metadata_replicas_required, | |
1242 | !(flags & BCH_FORCE_IF_DATA_DEGRADED) | |
1243 | ? c->opts.data_replicas | |
1244 | : c->opts.data_replicas_required); | |
1245 | ||
1246 | return nr_rw >= required; | |
1247 | case BCH_MEMBER_STATE_FAILED: | |
1248 | case BCH_MEMBER_STATE_SPARE: | |
1249 | if (ca->mi.state != BCH_MEMBER_STATE_RW && | |
1250 | ca->mi.state != BCH_MEMBER_STATE_RO) | |
1251 | return true; | |
1252 | ||
1253 | /* do we have enough devices to read from? */ | |
1254 | new_online_devs = bch2_online_devs(c); | |
1255 | __clear_bit(ca->dev_idx, new_online_devs.d); | |
1256 | ||
1257 | s = __bch2_replicas_status(c, new_online_devs); | |
1258 | ||
1259 | return bch2_have_enough_devs(s, flags); | |
1260 | default: | |
1261 | BUG(); | |
1262 | } | |
1263 | } | |
1264 | ||
1265 | static bool bch2_fs_may_start(struct bch_fs *c) | |
1266 | { | |
1267 | struct replicas_status s; | |
1268 | struct bch_sb_field_members *mi; | |
1269 | struct bch_dev *ca; | |
1270 | unsigned i, flags = c->opts.degraded | |
1271 | ? BCH_FORCE_IF_DEGRADED | |
1272 | : 0; | |
1273 | ||
1274 | if (!c->opts.degraded) { | |
1275 | mutex_lock(&c->sb_lock); | |
1276 | mi = bch2_sb_get_members(c->disk_sb.sb); | |
1277 | ||
1278 | for (i = 0; i < c->disk_sb.sb->nr_devices; i++) { | |
1279 | if (!bch2_dev_exists(c->disk_sb.sb, mi, i)) | |
1280 | continue; | |
1281 | ||
1282 | ca = bch_dev_locked(c, i); | |
1283 | ||
1284 | if (!bch2_dev_is_online(ca) && | |
1285 | (ca->mi.state == BCH_MEMBER_STATE_RW || | |
1286 | ca->mi.state == BCH_MEMBER_STATE_RO)) { | |
1287 | mutex_unlock(&c->sb_lock); | |
1288 | return false; | |
1289 | } | |
1290 | } | |
1291 | mutex_unlock(&c->sb_lock); | |
1292 | } | |
1293 | ||
1294 | s = bch2_replicas_status(c); | |
1295 | ||
1296 | return bch2_have_enough_devs(s, flags); | |
1297 | } | |
1298 | ||
1299 | static void __bch2_dev_read_only(struct bch_fs *c, struct bch_dev *ca) | |
1300 | { | |
1301 | bch2_copygc_stop(ca); | |
1302 | ||
1303 | /* | |
1304 | * The allocator thread itself allocates btree nodes, so stop it first: | |
1305 | */ | |
1306 | bch2_dev_allocator_stop(ca); | |
1307 | bch2_dev_allocator_remove(c, ca); | |
1308 | bch2_dev_journal_stop(&c->journal, ca); | |
1309 | } | |
1310 | ||
1311 | static const char *__bch2_dev_read_write(struct bch_fs *c, struct bch_dev *ca) | |
1312 | { | |
1313 | lockdep_assert_held(&c->state_lock); | |
1314 | ||
1315 | BUG_ON(ca->mi.state != BCH_MEMBER_STATE_RW); | |
1316 | ||
1317 | bch2_dev_allocator_add(c, ca); | |
1318 | bch2_recalc_capacity(c); | |
1319 | ||
1320 | if (bch2_dev_allocator_start(ca)) | |
1321 | return "error starting allocator thread"; | |
1322 | ||
1323 | if (bch2_copygc_start(c, ca)) | |
1324 | return "error starting copygc thread"; | |
1325 | ||
1326 | return NULL; | |
1327 | } | |
1328 | ||
1329 | int __bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca, | |
1330 | enum bch_member_state new_state, int flags) | |
1331 | { | |
1332 | struct bch_sb_field_members *mi; | |
1333 | int ret = 0; | |
1334 | ||
1335 | if (ca->mi.state == new_state) | |
1336 | return 0; | |
1337 | ||
1338 | if (!bch2_dev_state_allowed(c, ca, new_state, flags)) | |
1339 | return -EINVAL; | |
1340 | ||
1341 | if (new_state != BCH_MEMBER_STATE_RW) | |
1342 | __bch2_dev_read_only(c, ca); | |
1343 | ||
1344 | bch_notice(ca, "%s", bch2_dev_state[new_state]); | |
1345 | ||
1346 | mutex_lock(&c->sb_lock); | |
1347 | mi = bch2_sb_get_members(c->disk_sb.sb); | |
1348 | SET_BCH_MEMBER_STATE(&mi->members[ca->dev_idx], new_state); | |
1349 | bch2_write_super(c); | |
1350 | mutex_unlock(&c->sb_lock); | |
1351 | ||
1352 | if (new_state == BCH_MEMBER_STATE_RW && | |
1353 | __bch2_dev_read_write(c, ca)) | |
1354 | ret = -ENOMEM; | |
1355 | ||
1356 | rebalance_wakeup(c); | |
1357 | ||
1358 | return ret; | |
1359 | } | |
1360 | ||
1361 | int bch2_dev_set_state(struct bch_fs *c, struct bch_dev *ca, | |
1362 | enum bch_member_state new_state, int flags) | |
1363 | { | |
1364 | int ret; | |
1365 | ||
1366 | mutex_lock(&c->state_lock); | |
1367 | ret = __bch2_dev_set_state(c, ca, new_state, flags); | |
1368 | mutex_unlock(&c->state_lock); | |
1369 | ||
1370 | return ret; | |
1371 | } | |
1372 | ||
1373 | /* Device add/removal: */ | |
1374 | ||
1375 | int bch2_dev_remove(struct bch_fs *c, struct bch_dev *ca, int flags) | |
1376 | { | |
1377 | struct bch_sb_field_members *mi; | |
1378 | unsigned dev_idx = ca->dev_idx, data; | |
1379 | int ret = -EINVAL; | |
1380 | ||
1381 | mutex_lock(&c->state_lock); | |
1382 | ||
1383 | percpu_ref_put(&ca->ref); /* XXX */ | |
1384 | ||
1385 | if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_FAILED, flags)) { | |
1386 | bch_err(ca, "Cannot remove without losing data"); | |
1387 | goto err; | |
1388 | } | |
1389 | ||
1390 | __bch2_dev_read_only(c, ca); | |
1391 | ||
1392 | /* | |
1393 | * XXX: verify that dev_idx is really not in use anymore, anywhere | |
1394 | * | |
1395 | * flag_data_bad() does not check btree pointers | |
1396 | */ | |
1397 | ret = bch2_dev_data_drop(c, ca->dev_idx, flags); | |
1398 | if (ret) { | |
1399 | bch_err(ca, "Remove failed: error %i dropping data", ret); | |
1400 | goto err; | |
1401 | } | |
1402 | ||
1403 | ret = bch2_journal_flush_device_pins(&c->journal, ca->dev_idx); | |
1404 | if (ret) { | |
1405 | bch_err(ca, "Remove failed: error %i flushing journal", ret); | |
1406 | goto err; | |
1407 | } | |
1408 | ||
1409 | data = bch2_dev_has_data(c, ca); | |
1410 | if (data) { | |
1411 | char data_has_str[100]; | |
319f9ac3 | 1412 | |
dbaee468 KO |
1413 | bch2_flags_to_text(&PBUF(data_has_str), |
1414 | bch2_data_types, data); | |
1c6fdbd8 KO |
1415 | bch_err(ca, "Remove failed, still has data (%s)", data_has_str); |
1416 | ret = -EBUSY; | |
1417 | goto err; | |
1418 | } | |
1419 | ||
1420 | ret = bch2_btree_delete_range(c, BTREE_ID_ALLOC, | |
1421 | POS(ca->dev_idx, 0), | |
1422 | POS(ca->dev_idx + 1, 0), | |
fc3268c1 | 1423 | NULL); |
1c6fdbd8 KO |
1424 | if (ret) { |
1425 | bch_err(ca, "Remove failed, error deleting alloc info"); | |
1426 | goto err; | |
1427 | } | |
1428 | ||
1429 | /* | |
1430 | * must flush all existing journal entries, they might have | |
1431 | * (overwritten) keys that point to the device we're removing: | |
1432 | */ | |
1433 | bch2_journal_flush_all_pins(&c->journal); | |
1434 | ret = bch2_journal_error(&c->journal); | |
1435 | if (ret) { | |
1436 | bch_err(ca, "Remove failed, journal error"); | |
1437 | goto err; | |
1438 | } | |
1439 | ||
1440 | __bch2_dev_offline(c, ca); | |
1441 | ||
1442 | mutex_lock(&c->sb_lock); | |
1443 | rcu_assign_pointer(c->devs[ca->dev_idx], NULL); | |
1444 | mutex_unlock(&c->sb_lock); | |
1445 | ||
1446 | percpu_ref_kill(&ca->ref); | |
1447 | wait_for_completion(&ca->ref_completion); | |
1448 | ||
1449 | bch2_dev_free(ca); | |
1450 | ||
1451 | /* | |
1452 | * Free this device's slot in the bch_member array - all pointers to | |
1453 | * this device must be gone: | |
1454 | */ | |
1455 | mutex_lock(&c->sb_lock); | |
1456 | mi = bch2_sb_get_members(c->disk_sb.sb); | |
1457 | memset(&mi->members[dev_idx].uuid, 0, sizeof(mi->members[dev_idx].uuid)); | |
1458 | ||
1459 | bch2_write_super(c); | |
1460 | ||
1461 | mutex_unlock(&c->sb_lock); | |
1462 | mutex_unlock(&c->state_lock); | |
1463 | return 0; | |
1464 | err: | |
d3bb629d KO |
1465 | if (ca->mi.state == BCH_MEMBER_STATE_RW && |
1466 | !percpu_ref_is_zero(&ca->io_ref)) | |
1c6fdbd8 KO |
1467 | __bch2_dev_read_write(c, ca); |
1468 | mutex_unlock(&c->state_lock); | |
1469 | return ret; | |
1470 | } | |
1471 | ||
6eac2c2e KO |
1472 | static void dev_usage_clear(struct bch_dev *ca) |
1473 | { | |
1474 | struct bucket_array *buckets; | |
6eac2c2e | 1475 | |
5e82a9a1 | 1476 | percpu_memset(ca->usage[0], 0, sizeof(*ca->usage[0])); |
6eac2c2e KO |
1477 | |
1478 | down_read(&ca->bucket_lock); | |
1479 | buckets = bucket_array(ca); | |
1480 | ||
1481 | memset(buckets->b, 0, sizeof(buckets->b[0]) * buckets->nbuckets); | |
1482 | up_read(&ca->bucket_lock); | |
1483 | } | |
1484 | ||
1c6fdbd8 KO |
1485 | /* Add new device to running filesystem: */ |
1486 | int bch2_dev_add(struct bch_fs *c, const char *path) | |
1487 | { | |
1488 | struct bch_opts opts = bch2_opts_empty(); | |
1489 | struct bch_sb_handle sb; | |
1490 | const char *err; | |
1491 | struct bch_dev *ca = NULL; | |
1492 | struct bch_sb_field_members *mi; | |
1493 | struct bch_member dev_mi; | |
1494 | unsigned dev_idx, nr_devices, u64s; | |
1495 | int ret; | |
1496 | ||
1497 | ret = bch2_read_super(path, &opts, &sb); | |
1498 | if (ret) | |
1499 | return ret; | |
1500 | ||
1501 | err = bch2_sb_validate(&sb); | |
1502 | if (err) | |
1503 | return -EINVAL; | |
1504 | ||
1505 | dev_mi = bch2_sb_get_members(sb.sb)->members[sb.sb->dev_idx]; | |
1506 | ||
1507 | err = bch2_dev_may_add(sb.sb, c); | |
1508 | if (err) | |
1509 | return -EINVAL; | |
1510 | ||
1511 | ca = __bch2_dev_alloc(c, &dev_mi); | |
1512 | if (!ca) { | |
1513 | bch2_free_super(&sb); | |
1514 | return -ENOMEM; | |
1515 | } | |
1516 | ||
1517 | ret = __bch2_dev_attach_bdev(ca, &sb); | |
1518 | if (ret) { | |
1519 | bch2_dev_free(ca); | |
1520 | return ret; | |
1521 | } | |
1522 | ||
6eac2c2e KO |
1523 | /* |
1524 | * We want to allocate journal on the new device before adding the new | |
1525 | * device to the filesystem because allocating after we attach requires | |
1526 | * spinning up the allocator thread, and the allocator thread requires | |
1527 | * doing btree writes, which if the existing devices are RO isn't going | |
1528 | * to work | |
1529 | * | |
1530 | * So we have to mark where the superblocks are, but marking allocated | |
1531 | * data normally updates the filesystem usage too, so we have to mark, | |
1532 | * allocate the journal, reset all the marks, then remark after we | |
1533 | * attach... | |
1534 | */ | |
9ca53b55 | 1535 | bch2_mark_dev_superblock(ca->fs, ca, 0); |
6eac2c2e | 1536 | |
1c6fdbd8 KO |
1537 | err = "journal alloc failed"; |
1538 | ret = bch2_dev_journal_alloc(ca); | |
1539 | if (ret) | |
1540 | goto err; | |
1541 | ||
6eac2c2e KO |
1542 | dev_usage_clear(ca); |
1543 | ||
1c6fdbd8 KO |
1544 | mutex_lock(&c->state_lock); |
1545 | mutex_lock(&c->sb_lock); | |
1546 | ||
1547 | err = "insufficient space in new superblock"; | |
1548 | ret = bch2_sb_from_fs(c, ca); | |
1549 | if (ret) | |
1550 | goto err_unlock; | |
1551 | ||
1552 | mi = bch2_sb_get_members(ca->disk_sb.sb); | |
1553 | ||
1554 | if (!bch2_sb_resize_members(&ca->disk_sb, | |
1555 | le32_to_cpu(mi->field.u64s) + | |
1556 | sizeof(dev_mi) / sizeof(u64))) { | |
1557 | ret = -ENOSPC; | |
1558 | goto err_unlock; | |
1559 | } | |
1560 | ||
1561 | if (dynamic_fault("bcachefs:add:no_slot")) | |
1562 | goto no_slot; | |
1563 | ||
1564 | mi = bch2_sb_get_members(c->disk_sb.sb); | |
1565 | for (dev_idx = 0; dev_idx < BCH_SB_MEMBERS_MAX; dev_idx++) | |
1566 | if (!bch2_dev_exists(c->disk_sb.sb, mi, dev_idx)) | |
1567 | goto have_slot; | |
1568 | no_slot: | |
1569 | err = "no slots available in superblock"; | |
1570 | ret = -ENOSPC; | |
1571 | goto err_unlock; | |
1572 | ||
1573 | have_slot: | |
1574 | nr_devices = max_t(unsigned, dev_idx + 1, c->sb.nr_devices); | |
1575 | u64s = (sizeof(struct bch_sb_field_members) + | |
1576 | sizeof(struct bch_member) * nr_devices) / sizeof(u64); | |
1577 | ||
1578 | err = "no space in superblock for member info"; | |
1579 | ret = -ENOSPC; | |
1580 | ||
1581 | mi = bch2_sb_resize_members(&c->disk_sb, u64s); | |
1582 | if (!mi) | |
1583 | goto err_unlock; | |
1584 | ||
1585 | /* success: */ | |
1586 | ||
1587 | mi->members[dev_idx] = dev_mi; | |
a420eea6 | 1588 | mi->members[dev_idx].last_mount = cpu_to_le64(ktime_get_real_seconds()); |
1c6fdbd8 KO |
1589 | c->disk_sb.sb->nr_devices = nr_devices; |
1590 | ||
1591 | ca->disk_sb.sb->dev_idx = dev_idx; | |
1592 | bch2_dev_attach(c, ca, dev_idx); | |
1593 | ||
9ca53b55 | 1594 | bch2_mark_dev_superblock(c, ca, 0); |
6eac2c2e | 1595 | |
1c6fdbd8 KO |
1596 | bch2_write_super(c); |
1597 | mutex_unlock(&c->sb_lock); | |
1598 | ||
1599 | if (ca->mi.state == BCH_MEMBER_STATE_RW) { | |
1600 | err = __bch2_dev_read_write(c, ca); | |
1601 | if (err) | |
1602 | goto err_late; | |
1603 | } | |
1604 | ||
1605 | mutex_unlock(&c->state_lock); | |
1606 | return 0; | |
1607 | ||
1608 | err_unlock: | |
1609 | mutex_unlock(&c->sb_lock); | |
1610 | mutex_unlock(&c->state_lock); | |
1611 | err: | |
1612 | if (ca) | |
1613 | bch2_dev_free(ca); | |
1614 | bch2_free_super(&sb); | |
1615 | bch_err(c, "Unable to add device: %s", err); | |
1616 | return ret; | |
1617 | err_late: | |
1618 | bch_err(c, "Error going rw after adding device: %s", err); | |
1619 | return -EINVAL; | |
1620 | } | |
1621 | ||
1622 | /* Hot add existing device to running filesystem: */ | |
1623 | int bch2_dev_online(struct bch_fs *c, const char *path) | |
1624 | { | |
1625 | struct bch_opts opts = bch2_opts_empty(); | |
1626 | struct bch_sb_handle sb = { NULL }; | |
1627 | struct bch_sb_field_members *mi; | |
1628 | struct bch_dev *ca; | |
1629 | unsigned dev_idx; | |
1630 | const char *err; | |
1631 | int ret; | |
1632 | ||
1633 | mutex_lock(&c->state_lock); | |
1634 | ||
1635 | ret = bch2_read_super(path, &opts, &sb); | |
1636 | if (ret) { | |
1637 | mutex_unlock(&c->state_lock); | |
1638 | return ret; | |
1639 | } | |
1640 | ||
1641 | dev_idx = sb.sb->dev_idx; | |
1642 | ||
1643 | err = bch2_dev_in_fs(c->disk_sb.sb, sb.sb); | |
1644 | if (err) | |
1645 | goto err; | |
1646 | ||
1647 | if (bch2_dev_attach_bdev(c, &sb)) { | |
1648 | err = "bch2_dev_attach_bdev() error"; | |
1649 | goto err; | |
1650 | } | |
1651 | ||
1652 | ca = bch_dev_locked(c, dev_idx); | |
1653 | if (ca->mi.state == BCH_MEMBER_STATE_RW) { | |
1654 | err = __bch2_dev_read_write(c, ca); | |
1655 | if (err) | |
1656 | goto err; | |
1657 | } | |
1658 | ||
1659 | mutex_lock(&c->sb_lock); | |
1660 | mi = bch2_sb_get_members(c->disk_sb.sb); | |
1661 | ||
1662 | mi->members[ca->dev_idx].last_mount = | |
a420eea6 | 1663 | cpu_to_le64(ktime_get_real_seconds()); |
1c6fdbd8 KO |
1664 | |
1665 | bch2_write_super(c); | |
1666 | mutex_unlock(&c->sb_lock); | |
1667 | ||
1668 | mutex_unlock(&c->state_lock); | |
1669 | return 0; | |
1670 | err: | |
1671 | mutex_unlock(&c->state_lock); | |
1672 | bch2_free_super(&sb); | |
1673 | bch_err(c, "error bringing %s online: %s", path, err); | |
1674 | return -EINVAL; | |
1675 | } | |
1676 | ||
1677 | int bch2_dev_offline(struct bch_fs *c, struct bch_dev *ca, int flags) | |
1678 | { | |
1679 | mutex_lock(&c->state_lock); | |
1680 | ||
1681 | if (!bch2_dev_is_online(ca)) { | |
1682 | bch_err(ca, "Already offline"); | |
1683 | mutex_unlock(&c->state_lock); | |
1684 | return 0; | |
1685 | } | |
1686 | ||
1687 | if (!bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_FAILED, flags)) { | |
1688 | bch_err(ca, "Cannot offline required disk"); | |
1689 | mutex_unlock(&c->state_lock); | |
1690 | return -EINVAL; | |
1691 | } | |
1692 | ||
1693 | __bch2_dev_offline(c, ca); | |
1694 | ||
1695 | mutex_unlock(&c->state_lock); | |
1696 | return 0; | |
1697 | } | |
1698 | ||
1699 | int bch2_dev_resize(struct bch_fs *c, struct bch_dev *ca, u64 nbuckets) | |
1700 | { | |
1701 | struct bch_member *mi; | |
1702 | int ret = 0; | |
1703 | ||
1704 | mutex_lock(&c->state_lock); | |
1705 | ||
1706 | if (nbuckets < ca->mi.nbuckets) { | |
1707 | bch_err(ca, "Cannot shrink yet"); | |
1708 | ret = -EINVAL; | |
1709 | goto err; | |
1710 | } | |
1711 | ||
1712 | if (bch2_dev_is_online(ca) && | |
1713 | get_capacity(ca->disk_sb.bdev->bd_disk) < | |
1714 | ca->mi.bucket_size * nbuckets) { | |
1715 | bch_err(ca, "New size larger than device"); | |
1716 | ret = -EINVAL; | |
1717 | goto err; | |
1718 | } | |
1719 | ||
1720 | ret = bch2_dev_buckets_resize(c, ca, nbuckets); | |
1721 | if (ret) { | |
1722 | bch_err(ca, "Resize error: %i", ret); | |
1723 | goto err; | |
1724 | } | |
1725 | ||
1726 | mutex_lock(&c->sb_lock); | |
1727 | mi = &bch2_sb_get_members(c->disk_sb.sb)->members[ca->dev_idx]; | |
1728 | mi->nbuckets = cpu_to_le64(nbuckets); | |
1729 | ||
1730 | bch2_write_super(c); | |
1731 | mutex_unlock(&c->sb_lock); | |
1732 | ||
1733 | bch2_recalc_capacity(c); | |
1734 | err: | |
1735 | mutex_unlock(&c->state_lock); | |
1736 | return ret; | |
1737 | } | |
1738 | ||
1739 | /* return with ref on ca->ref: */ | |
1740 | struct bch_dev *bch2_dev_lookup(struct bch_fs *c, const char *path) | |
1741 | { | |
1742 | ||
1743 | struct bch_dev *ca; | |
1744 | dev_t dev; | |
1745 | unsigned i; | |
1746 | int ret; | |
1747 | ||
1748 | ret = lookup_bdev(path, &dev); | |
1749 | if (ret) | |
1750 | return ERR_PTR(ret); | |
1751 | ||
1752 | for_each_member_device(ca, c, i) | |
1753 | if (ca->disk_sb.bdev->bd_dev == dev) | |
1754 | goto found; | |
1755 | ||
1756 | ca = ERR_PTR(-ENOENT); | |
1757 | found: | |
1758 | return ca; | |
1759 | } | |
1760 | ||
1761 | /* Filesystem open: */ | |
1762 | ||
1763 | struct bch_fs *bch2_fs_open(char * const *devices, unsigned nr_devices, | |
1764 | struct bch_opts opts) | |
1765 | { | |
1766 | struct bch_sb_handle *sb = NULL; | |
1767 | struct bch_fs *c = NULL; | |
1768 | unsigned i, best_sb = 0; | |
1769 | const char *err; | |
1770 | int ret = -ENOMEM; | |
1771 | ||
1772 | pr_verbose_init(opts, ""); | |
1773 | ||
1774 | if (!nr_devices) { | |
1775 | c = ERR_PTR(-EINVAL); | |
1776 | goto out2; | |
1777 | } | |
1778 | ||
1779 | if (!try_module_get(THIS_MODULE)) { | |
1780 | c = ERR_PTR(-ENODEV); | |
1781 | goto out2; | |
1782 | } | |
1783 | ||
1784 | sb = kcalloc(nr_devices, sizeof(*sb), GFP_KERNEL); | |
1785 | if (!sb) | |
1786 | goto err; | |
1787 | ||
1788 | for (i = 0; i < nr_devices; i++) { | |
1789 | ret = bch2_read_super(devices[i], &opts, &sb[i]); | |
1790 | if (ret) | |
1791 | goto err; | |
1792 | ||
1793 | err = bch2_sb_validate(&sb[i]); | |
1794 | if (err) | |
1795 | goto err_print; | |
1796 | } | |
1797 | ||
1798 | for (i = 1; i < nr_devices; i++) | |
1799 | if (le64_to_cpu(sb[i].sb->seq) > | |
1800 | le64_to_cpu(sb[best_sb].sb->seq)) | |
1801 | best_sb = i; | |
1802 | ||
1803 | for (i = 0; i < nr_devices; i++) { | |
1804 | err = bch2_dev_in_fs(sb[best_sb].sb, sb[i].sb); | |
1805 | if (err) | |
1806 | goto err_print; | |
1807 | } | |
1808 | ||
1809 | ret = -ENOMEM; | |
1810 | c = bch2_fs_alloc(sb[best_sb].sb, opts); | |
1811 | if (!c) | |
1812 | goto err; | |
1813 | ||
1814 | err = "bch2_dev_online() error"; | |
1815 | mutex_lock(&c->state_lock); | |
1816 | for (i = 0; i < nr_devices; i++) | |
1817 | if (bch2_dev_attach_bdev(c, &sb[i])) { | |
1818 | mutex_unlock(&c->state_lock); | |
1819 | goto err_print; | |
1820 | } | |
1821 | mutex_unlock(&c->state_lock); | |
1822 | ||
1823 | err = "insufficient devices"; | |
1824 | if (!bch2_fs_may_start(c)) | |
1825 | goto err_print; | |
1826 | ||
1827 | if (!c->opts.nostart) { | |
619f5bee KO |
1828 | ret = bch2_fs_start(c); |
1829 | if (ret) | |
1830 | goto err; | |
1c6fdbd8 KO |
1831 | } |
1832 | out: | |
1833 | kfree(sb); | |
1834 | module_put(THIS_MODULE); | |
1835 | out2: | |
1836 | pr_verbose_init(opts, "ret %i", PTR_ERR_OR_ZERO(c)); | |
1837 | return c; | |
1838 | err_print: | |
1839 | pr_err("bch_fs_open err opening %s: %s", | |
1840 | devices[0], err); | |
1841 | ret = -EINVAL; | |
1842 | err: | |
1843 | if (c) | |
1844 | bch2_fs_stop(c); | |
1845 | for (i = 0; i < nr_devices; i++) | |
1846 | bch2_free_super(&sb[i]); | |
1847 | c = ERR_PTR(ret); | |
1848 | goto out; | |
1849 | } | |
1850 | ||
1851 | static const char *__bch2_fs_open_incremental(struct bch_sb_handle *sb, | |
1852 | struct bch_opts opts) | |
1853 | { | |
1854 | const char *err; | |
1855 | struct bch_fs *c; | |
1856 | bool allocated_fs = false; | |
619f5bee | 1857 | int ret; |
1c6fdbd8 KO |
1858 | |
1859 | err = bch2_sb_validate(sb); | |
1860 | if (err) | |
1861 | return err; | |
1862 | ||
1863 | mutex_lock(&bch_fs_list_lock); | |
1864 | c = __bch2_uuid_to_fs(sb->sb->uuid); | |
1865 | if (c) { | |
1866 | closure_get(&c->cl); | |
1867 | ||
1868 | err = bch2_dev_in_fs(c->disk_sb.sb, sb->sb); | |
1869 | if (err) | |
1870 | goto err; | |
1871 | } else { | |
1872 | c = bch2_fs_alloc(sb->sb, opts); | |
1873 | err = "cannot allocate memory"; | |
1874 | if (!c) | |
1875 | goto err; | |
1876 | ||
1877 | allocated_fs = true; | |
1878 | } | |
1879 | ||
1880 | err = "bch2_dev_online() error"; | |
1881 | ||
1882 | mutex_lock(&c->sb_lock); | |
1883 | if (bch2_dev_attach_bdev(c, sb)) { | |
1884 | mutex_unlock(&c->sb_lock); | |
1885 | goto err; | |
1886 | } | |
1887 | mutex_unlock(&c->sb_lock); | |
1888 | ||
1889 | if (!c->opts.nostart && bch2_fs_may_start(c)) { | |
619f5bee KO |
1890 | err = "error starting filesystem"; |
1891 | ret = bch2_fs_start(c); | |
1892 | if (ret) | |
1c6fdbd8 KO |
1893 | goto err; |
1894 | } | |
1895 | ||
1896 | closure_put(&c->cl); | |
1897 | mutex_unlock(&bch_fs_list_lock); | |
1898 | ||
1899 | return NULL; | |
1900 | err: | |
1901 | mutex_unlock(&bch_fs_list_lock); | |
1902 | ||
1903 | if (allocated_fs) | |
1904 | bch2_fs_stop(c); | |
1905 | else if (c) | |
1906 | closure_put(&c->cl); | |
1907 | ||
1908 | return err; | |
1909 | } | |
1910 | ||
1911 | const char *bch2_fs_open_incremental(const char *path) | |
1912 | { | |
1913 | struct bch_sb_handle sb; | |
1914 | struct bch_opts opts = bch2_opts_empty(); | |
1915 | const char *err; | |
1916 | ||
1917 | if (bch2_read_super(path, &opts, &sb)) | |
1918 | return "error reading superblock"; | |
1919 | ||
1920 | err = __bch2_fs_open_incremental(&sb, opts); | |
1921 | bch2_free_super(&sb); | |
1922 | ||
1923 | return err; | |
1924 | } | |
1925 | ||
1926 | /* Global interfaces/init */ | |
1927 | ||
1928 | static void bcachefs_exit(void) | |
1929 | { | |
1930 | bch2_debug_exit(); | |
1931 | bch2_vfs_exit(); | |
1932 | bch2_chardev_exit(); | |
1933 | if (bcachefs_kset) | |
1934 | kset_unregister(bcachefs_kset); | |
1935 | } | |
1936 | ||
1937 | static int __init bcachefs_init(void) | |
1938 | { | |
1939 | bch2_bkey_pack_test(); | |
1940 | bch2_inode_pack_test(); | |
1941 | ||
1942 | if (!(bcachefs_kset = kset_create_and_add("bcachefs", NULL, fs_kobj)) || | |
1943 | bch2_chardev_init() || | |
1944 | bch2_vfs_init() || | |
1945 | bch2_debug_init()) | |
1946 | goto err; | |
1947 | ||
1948 | return 0; | |
1949 | err: | |
1950 | bcachefs_exit(); | |
1951 | return -ENOMEM; | |
1952 | } | |
1953 | ||
1954 | #define BCH_DEBUG_PARAM(name, description) \ | |
1955 | bool bch2_##name; \ | |
1956 | module_param_named(name, bch2_##name, bool, 0644); \ | |
1957 | MODULE_PARM_DESC(name, description); | |
1958 | BCH_DEBUG_PARAMS() | |
1959 | #undef BCH_DEBUG_PARAM | |
1960 | ||
26609b61 | 1961 | unsigned bch2_metadata_version = bcachefs_metadata_version_current; |
1c6fdbd8 KO |
1962 | module_param_named(version, bch2_metadata_version, uint, 0400); |
1963 | ||
1964 | module_exit(bcachefs_exit); | |
1965 | module_init(bcachefs_init); |