1 // SPDX-License-Identifier: GPL-2.0
2 #include "bcachefs.h"
3 #include "alloc_background.h"
4 #include "alloc_foreground.h"
5 #include "backpointers.h"
6 #include "bkey_buf.h"
7 #include "btree_cache.h"
8 #include "btree_io.h"
9 #include "btree_key_cache.h"
10 #include "btree_update.h"
11 #include "btree_update_interior.h"
12 #include "btree_gc.h"
13 #include "btree_write_buffer.h"
14 #include "buckets.h"
15 #include "buckets_waiting_for_journal.h"
16 #include "clock.h"
17 #include "debug.h"
18 #include "disk_accounting.h"
19 #include "ec.h"
20 #include "error.h"
21 #include "lru.h"
22 #include "recovery.h"
23 #include "trace.h"
24 #include "varint.h"
25
26 #include <linux/kthread.h>
27 #include <linux/math64.h>
28 #include <linux/random.h>
29 #include <linux/rculist.h>
30 #include <linux/rcupdate.h>
31 #include <linux/sched/task.h>
32 #include <linux/sort.h>
33
34 static void bch2_discard_one_bucket_fast(struct bch_dev *, u64);
35
36 /* Persistent alloc info: */
37
38 static const unsigned BCH_ALLOC_V1_FIELD_BYTES[] = {
39 #define x(name, bits) [BCH_ALLOC_FIELD_V1_##name] = bits / 8,
40 BCH_ALLOC_FIELDS_V1()
41 #undef x
42 };
43
44 struct bkey_alloc_unpacked {
45 u64 journal_seq;
46 u8 gen;
47 u8 oldest_gen;
48 u8 data_type;
49 bool need_discard:1;
50 bool need_inc_gen:1;
51 #define x(_name, _bits) u##_bits _name;
52 BCH_ALLOC_FIELDS_V2()
53 #undef x
54 };
55
56 static inline u64 alloc_field_v1_get(const struct bch_alloc *a,
57 const void **p, unsigned field)
58 {
59 unsigned bytes = BCH_ALLOC_V1_FIELD_BYTES[field];
60 u64 v;
61
62 if (!(a->fields & (1 << field)))
63 return 0;
64
65 switch (bytes) {
66 case 1:
67 v = *((const u8 *) *p);
68 break;
69 case 2:
70 v = le16_to_cpup(*p);
71 break;
72 case 4:
73 v = le32_to_cpup(*p);
74 break;
75 case 8:
76 v = le64_to_cpup(*p);
77 break;
78 default:
79 BUG();
80 }
81
82 *p += bytes;
83 return v;
84 }
85
86 static void bch2_alloc_unpack_v1(struct bkey_alloc_unpacked *out,
87 struct bkey_s_c k)
88 {
89 const struct bch_alloc *in = bkey_s_c_to_alloc(k).v;
90 const void *d = in->data;
91 unsigned idx = 0;
92
93 out->gen = in->gen;
94
95 #define x(_name, _bits) out->_name = alloc_field_v1_get(in, &d, idx++);
96 BCH_ALLOC_FIELDS_V1()
97 #undef x
98 }
99
100 static int bch2_alloc_unpack_v2(struct bkey_alloc_unpacked *out,
101 struct bkey_s_c k)
102 {
103 struct bkey_s_c_alloc_v2 a = bkey_s_c_to_alloc_v2(k);
104 const u8 *in = a.v->data;
105 const u8 *end = bkey_val_end(a);
106 unsigned fieldnr = 0;
107 int ret;
108 u64 v;
109
110 out->gen = a.v->gen;
111 out->oldest_gen = a.v->oldest_gen;
112 out->data_type = a.v->data_type;
113
114 #define x(_name, _bits) \
115 if (fieldnr < a.v->nr_fields) { \
116 ret = bch2_varint_decode_fast(in, end, &v); \
117 if (ret < 0) \
118 return ret; \
119 in += ret; \
120 } else { \
121 v = 0; \
122 } \
123 out->_name = v; \
124 if (v != out->_name) \
125 return -1; \
126 fieldnr++;
127
128 BCH_ALLOC_FIELDS_V2()
129 #undef x
130 return 0;
131 }
132
133 static int bch2_alloc_unpack_v3(struct bkey_alloc_unpacked *out,
134 struct bkey_s_c k)
135 {
136 struct bkey_s_c_alloc_v3 a = bkey_s_c_to_alloc_v3(k);
137 const u8 *in = a.v->data;
138 const u8 *end = bkey_val_end(a);
139 unsigned fieldnr = 0;
140 int ret;
141 u64 v;
142
143 out->gen = a.v->gen;
144 out->oldest_gen = a.v->oldest_gen;
145 out->data_type = a.v->data_type;
146 out->need_discard = BCH_ALLOC_V3_NEED_DISCARD(a.v);
147 out->need_inc_gen = BCH_ALLOC_V3_NEED_INC_GEN(a.v);
148 out->journal_seq = le64_to_cpu(a.v->journal_seq);
149
150 #define x(_name, _bits) \
151 if (fieldnr < a.v->nr_fields) { \
152 ret = bch2_varint_decode_fast(in, end, &v); \
153 if (ret < 0) \
154 return ret; \
155 in += ret; \
156 } else { \
157 v = 0; \
158 } \
159 out->_name = v; \
160 if (v != out->_name) \
161 return -1; \
162 fieldnr++;
163
164 BCH_ALLOC_FIELDS_V2()
165 #undef x
166 return 0;
167 }
168
169 static struct bkey_alloc_unpacked bch2_alloc_unpack(struct bkey_s_c k)
170 {
171 struct bkey_alloc_unpacked ret = { .gen = 0 };
172
173 switch (k.k->type) {
174 case KEY_TYPE_alloc:
175 bch2_alloc_unpack_v1(&ret, k);
176 break;
177 case KEY_TYPE_alloc_v2:
178 bch2_alloc_unpack_v2(&ret, k);
179 break;
180 case KEY_TYPE_alloc_v3:
181 bch2_alloc_unpack_v3(&ret, k);
182 break;
183 }
184
185 return ret;
186 }
187
188 static unsigned bch_alloc_v1_val_u64s(const struct bch_alloc *a)
189 {
190 unsigned i, bytes = offsetof(struct bch_alloc, data);
191
192 for (i = 0; i < ARRAY_SIZE(BCH_ALLOC_V1_FIELD_BYTES); i++)
193 if (a->fields & (1 << i))
194 bytes += BCH_ALLOC_V1_FIELD_BYTES[i];
195
196 return DIV_ROUND_UP(bytes, sizeof(u64));
197 }
198
199 int bch2_alloc_v1_invalid(struct bch_fs *c, struct bkey_s_c k,
200 enum bch_validate_flags flags,
201 struct printbuf *err)
202 {
203 struct bkey_s_c_alloc a = bkey_s_c_to_alloc(k);
204 int ret = 0;
205
206 /* allow for unknown fields */
207 bkey_fsck_err_on(bkey_val_u64s(a.k) < bch_alloc_v1_val_u64s(a.v), c, err,
208 alloc_v1_val_size_bad,
209 "incorrect value size (%zu < %u)",
210 bkey_val_u64s(a.k), bch_alloc_v1_val_u64s(a.v));
211 fsck_err:
212 return ret;
213 }
214
215 int bch2_alloc_v2_invalid(struct bch_fs *c, struct bkey_s_c k,
216 enum bch_validate_flags flags,
217 struct printbuf *err)
218 {
219 struct bkey_alloc_unpacked u;
220 int ret = 0;
221
222 bkey_fsck_err_on(bch2_alloc_unpack_v2(&u, k), c, err,
223 alloc_v2_unpack_error,
224 "unpack error");
225 fsck_err:
226 return ret;
227 }
228
229 int bch2_alloc_v3_invalid(struct bch_fs *c, struct bkey_s_c k,
230 enum bch_validate_flags flags,
231 struct printbuf *err)
232 {
233 struct bkey_alloc_unpacked u;
234 int ret = 0;
235
236 bkey_fsck_err_on(bch2_alloc_unpack_v3(&u, k), c, err,
237 alloc_v2_unpack_error,
238 "unpack error");
239 fsck_err:
240 return ret;
241 }
242
243 int bch2_alloc_v4_invalid(struct bch_fs *c, struct bkey_s_c k,
244 enum bch_validate_flags flags, struct printbuf *err)
245 {
246 struct bkey_s_c_alloc_v4 a = bkey_s_c_to_alloc_v4(k);
247 int ret = 0;
248
249 bkey_fsck_err_on(alloc_v4_u64s_noerror(a.v) > bkey_val_u64s(k.k), c, err,
250 alloc_v4_val_size_bad,
251 "bad val size (%u > %zu)",
252 alloc_v4_u64s_noerror(a.v), bkey_val_u64s(k.k));
253
254 bkey_fsck_err_on(!BCH_ALLOC_V4_BACKPOINTERS_START(a.v) &&
255 BCH_ALLOC_V4_NR_BACKPOINTERS(a.v), c, err,
256 alloc_v4_backpointers_start_bad,
257 "invalid backpointers_start");
258
259 bkey_fsck_err_on(alloc_data_type(*a.v, a.v->data_type) != a.v->data_type, c, err,
260 alloc_key_data_type_bad,
261 "invalid data type (got %u should be %u)",
262 a.v->data_type, alloc_data_type(*a.v, a.v->data_type));
263
264 for (unsigned i = 0; i < 2; i++)
265 bkey_fsck_err_on(a.v->io_time[i] > LRU_TIME_MAX,
266 c, err,
267 alloc_key_io_time_bad,
268 "invalid io_time[%s]: %llu, max %llu",
269 i == READ ? "read" : "write",
270 a.v->io_time[i], LRU_TIME_MAX);
271
272 unsigned stripe_sectors = BCH_ALLOC_V4_BACKPOINTERS_START(a.v) * sizeof(u64) >
273 offsetof(struct bch_alloc_v4, stripe_sectors)
274 ? a.v->stripe_sectors
275 : 0;
276
277 switch (a.v->data_type) {
278 case BCH_DATA_free:
279 case BCH_DATA_need_gc_gens:
280 case BCH_DATA_need_discard:
281 bkey_fsck_err_on(stripe_sectors ||
282 a.v->dirty_sectors ||
283 a.v->cached_sectors ||
284 a.v->stripe,
285 c, err, alloc_key_empty_but_have_data,
286 "empty data type free but have data %u.%u.%u %u",
287 stripe_sectors,
288 a.v->dirty_sectors,
289 a.v->cached_sectors,
290 a.v->stripe);
291 break;
292 case BCH_DATA_sb:
293 case BCH_DATA_journal:
294 case BCH_DATA_btree:
295 case BCH_DATA_user:
296 case BCH_DATA_parity:
297 bkey_fsck_err_on(!a.v->dirty_sectors &&
298 !stripe_sectors,
299 c, err, alloc_key_dirty_sectors_0,
300 "data_type %s but dirty_sectors==0",
301 bch2_data_type_str(a.v->data_type));
302 break;
303 case BCH_DATA_cached:
304 bkey_fsck_err_on(!a.v->cached_sectors ||
305 a.v->dirty_sectors ||
306 stripe_sectors ||
307 a.v->stripe,
308 c, err, alloc_key_cached_inconsistency,
309 "data type inconsistency");
310
311 bkey_fsck_err_on(!a.v->io_time[READ] &&
312 c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_to_lru_refs,
313 c, err, alloc_key_cached_but_read_time_zero,
314 "cached bucket with read_time == 0");
315 break;
316 case BCH_DATA_stripe:
317 break;
318 }
319 fsck_err:
320 return ret;
321 }
322
323 void bch2_alloc_v4_swab(struct bkey_s k)
324 {
325 struct bch_alloc_v4 *a = bkey_s_to_alloc_v4(k).v;
326 struct bch_backpointer *bp, *bps;
327
328 a->journal_seq = swab64(a->journal_seq);
329 a->flags = swab32(a->flags);
330 a->dirty_sectors = swab32(a->dirty_sectors);
331 a->cached_sectors = swab32(a->cached_sectors);
332 a->io_time[0] = swab64(a->io_time[0]);
333 a->io_time[1] = swab64(a->io_time[1]);
334 a->stripe = swab32(a->stripe);
335 a->nr_external_backpointers = swab32(a->nr_external_backpointers);
336 a->fragmentation_lru = swab64(a->fragmentation_lru);
337 a->stripe_sectors = swab32(a->stripe_sectors);
338
339 bps = alloc_v4_backpointers(a);
340 for (bp = bps; bp < bps + BCH_ALLOC_V4_NR_BACKPOINTERS(a); bp++) {
341 bp->bucket_offset = swab40(bp->bucket_offset);
342 bp->bucket_len = swab32(bp->bucket_len);
343 bch2_bpos_swab(&bp->pos);
344 }
345 }
346
347 void bch2_alloc_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c k)
348 {
349 struct bch_alloc_v4 _a;
350 const struct bch_alloc_v4 *a = bch2_alloc_to_v4(k, &_a);
351
352 prt_newline(out);
353 printbuf_indent_add(out, 2);
354
355 prt_printf(out, "gen %u oldest_gen %u data_type ", a->gen, a->oldest_gen);
356 bch2_prt_data_type(out, a->data_type);
357 prt_newline(out);
358 prt_printf(out, "journal_seq %llu\n", a->journal_seq);
359 prt_printf(out, "need_discard %llu\n", BCH_ALLOC_V4_NEED_DISCARD(a));
360 prt_printf(out, "need_inc_gen %llu\n", BCH_ALLOC_V4_NEED_INC_GEN(a));
361 prt_printf(out, "dirty_sectors %u\n", a->dirty_sectors);
362 prt_printf(out, "stripe_sectors %u\n", a->stripe_sectors);
363 prt_printf(out, "cached_sectors %u\n", a->cached_sectors);
364 prt_printf(out, "stripe %u\n", a->stripe);
365 prt_printf(out, "stripe_redundancy %u\n", a->stripe_redundancy);
366 prt_printf(out, "io_time[READ] %llu\n", a->io_time[READ]);
367 prt_printf(out, "io_time[WRITE] %llu\n", a->io_time[WRITE]);
368 prt_printf(out, "fragmentation %llu\n", a->fragmentation_lru);
369 prt_printf(out, "bp_start %llu\n", BCH_ALLOC_V4_BACKPOINTERS_START(a));
370 printbuf_indent_sub(out, 2);
371 }
372
373 void __bch2_alloc_to_v4(struct bkey_s_c k, struct bch_alloc_v4 *out)
374 {
375 if (k.k->type == KEY_TYPE_alloc_v4) {
376 void *src, *dst;
377
378 *out = *bkey_s_c_to_alloc_v4(k).v;
379
380 src = alloc_v4_backpointers(out);
381 SET_BCH_ALLOC_V4_BACKPOINTERS_START(out, BCH_ALLOC_V4_U64s);
382 dst = alloc_v4_backpointers(out);
383
384 if (src < dst)
385 memset(src, 0, dst - src);
386
387 SET_BCH_ALLOC_V4_NR_BACKPOINTERS(out, 0);
388 } else {
389 struct bkey_alloc_unpacked u = bch2_alloc_unpack(k);
390
391 *out = (struct bch_alloc_v4) {
392 .journal_seq = u.journal_seq,
393 .flags = u.need_discard,
394 .gen = u.gen,
395 .oldest_gen = u.oldest_gen,
396 .data_type = u.data_type,
397 .stripe_redundancy = u.stripe_redundancy,
398 .dirty_sectors = u.dirty_sectors,
399 .cached_sectors = u.cached_sectors,
400 .io_time[READ] = u.read_time,
401 .io_time[WRITE] = u.write_time,
402 .stripe = u.stripe,
403 };
404
405 SET_BCH_ALLOC_V4_BACKPOINTERS_START(out, BCH_ALLOC_V4_U64s);
406 }
407 }
408
409 static noinline struct bkey_i_alloc_v4 *
410 __bch2_alloc_to_v4_mut(struct btree_trans *trans, struct bkey_s_c k)
411 {
412 struct bkey_i_alloc_v4 *ret;
413
414 ret = bch2_trans_kmalloc(trans, max(bkey_bytes(k.k), sizeof(struct bkey_i_alloc_v4)));
415 if (IS_ERR(ret))
416 return ret;
417
418 if (k.k->type == KEY_TYPE_alloc_v4) {
419 void *src, *dst;
420
421 bkey_reassemble(&ret->k_i, k);
422
423 src = alloc_v4_backpointers(&ret->v);
424 SET_BCH_ALLOC_V4_BACKPOINTERS_START(&ret->v, BCH_ALLOC_V4_U64s);
425 dst = alloc_v4_backpointers(&ret->v);
426
427 if (src < dst)
428 memset(src, 0, dst - src);
429
430 SET_BCH_ALLOC_V4_NR_BACKPOINTERS(&ret->v, 0);
431 set_alloc_v4_u64s(ret);
432 } else {
433 bkey_alloc_v4_init(&ret->k_i);
434 ret->k.p = k.k->p;
435 bch2_alloc_to_v4(k, &ret->v);
436 }
437 return ret;
438 }
439
440 static inline struct bkey_i_alloc_v4 *bch2_alloc_to_v4_mut_inlined(struct btree_trans *trans, struct bkey_s_c k)
441 {
442 struct bkey_s_c_alloc_v4 a;
443
444 if (likely(k.k->type == KEY_TYPE_alloc_v4) &&
445 ((a = bkey_s_c_to_alloc_v4(k), true) &&
446 BCH_ALLOC_V4_NR_BACKPOINTERS(a.v) == 0))
447 return bch2_bkey_make_mut_noupdate_typed(trans, k, alloc_v4);
448
449 return __bch2_alloc_to_v4_mut(trans, k);
450 }
451
452 struct bkey_i_alloc_v4 *bch2_alloc_to_v4_mut(struct btree_trans *trans, struct bkey_s_c k)
453 {
454 return bch2_alloc_to_v4_mut_inlined(trans, k);
455 }
456
457 struct bkey_i_alloc_v4 *
458 bch2_trans_start_alloc_update_noupdate(struct btree_trans *trans, struct btree_iter *iter,
459 struct bpos pos)
460 {
461 struct bkey_s_c k = bch2_bkey_get_iter(trans, iter, BTREE_ID_alloc, pos,
462 BTREE_ITER_with_updates|
463 BTREE_ITER_cached|
464 BTREE_ITER_intent);
465 int ret = bkey_err(k);
466 if (unlikely(ret))
467 return ERR_PTR(ret);
468
469 struct bkey_i_alloc_v4 *a = bch2_alloc_to_v4_mut_inlined(trans, k);
470 ret = PTR_ERR_OR_ZERO(a);
471 if (unlikely(ret))
472 goto err;
473 return a;
474 err:
475 bch2_trans_iter_exit(trans, iter);
476 return ERR_PTR(ret);
477 }
478
479 __flatten
480 struct bkey_i_alloc_v4 *bch2_trans_start_alloc_update(struct btree_trans *trans, struct bpos pos,
481 enum btree_iter_update_trigger_flags flags)
482 {
483 struct btree_iter iter;
484 struct bkey_i_alloc_v4 *a = bch2_trans_start_alloc_update_noupdate(trans, &iter, pos);
485 int ret = PTR_ERR_OR_ZERO(a);
486 if (ret)
487 return ERR_PTR(ret);
488
489 ret = bch2_trans_update(trans, &iter, &a->k_i, flags);
490 bch2_trans_iter_exit(trans, &iter);
491 return unlikely(ret) ? ERR_PTR(ret) : a;
492 }
493
494 static struct bpos alloc_gens_pos(struct bpos pos, unsigned *offset)
495 {
496 *offset = pos.offset & KEY_TYPE_BUCKET_GENS_MASK;
497
498 pos.offset >>= KEY_TYPE_BUCKET_GENS_BITS;
499 return pos;
500 }
501
502 static struct bpos bucket_gens_pos_to_alloc(struct bpos pos, unsigned offset)
503 {
504 pos.offset <<= KEY_TYPE_BUCKET_GENS_BITS;
505 pos.offset += offset;
506 return pos;
507 }
508
509 static unsigned alloc_gen(struct bkey_s_c k, unsigned offset)
510 {
511 return k.k->type == KEY_TYPE_bucket_gens
512 ? bkey_s_c_to_bucket_gens(k).v->gens[offset]
513 : 0;
514 }
515
516 int bch2_bucket_gens_invalid(struct bch_fs *c, struct bkey_s_c k,
517 enum bch_validate_flags flags,
518 struct printbuf *err)
519 {
520 int ret = 0;
521
522 bkey_fsck_err_on(bkey_val_bytes(k.k) != sizeof(struct bch_bucket_gens), c, err,
523 bucket_gens_val_size_bad,
524 "bad val size (%zu != %zu)",
525 bkey_val_bytes(k.k), sizeof(struct bch_bucket_gens));
526 fsck_err:
527 return ret;
528 }
529
530 void bch2_bucket_gens_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c k)
531 {
532 struct bkey_s_c_bucket_gens g = bkey_s_c_to_bucket_gens(k);
533 unsigned i;
534
535 for (i = 0; i < ARRAY_SIZE(g.v->gens); i++) {
536 if (i)
537 prt_char(out, ' ');
538 prt_printf(out, "%u", g.v->gens[i]);
539 }
540 }
541
542 int bch2_bucket_gens_init(struct bch_fs *c)
543 {
544 struct btree_trans *trans = bch2_trans_get(c);
545 struct bkey_i_bucket_gens g;
546 bool have_bucket_gens_key = false;
547 int ret;
548
549 ret = for_each_btree_key(trans, iter, BTREE_ID_alloc, POS_MIN,
550 BTREE_ITER_prefetch, k, ({
551 /*
552 * Not a fsck error because this is checked/repaired by
553 * bch2_check_alloc_key() which runs later:
554 */
555 if (!bch2_dev_bucket_exists(c, k.k->p))
556 continue;
557
558 struct bch_alloc_v4 a;
559 u8 gen = bch2_alloc_to_v4(k, &a)->gen;
560 unsigned offset;
561 struct bpos pos = alloc_gens_pos(iter.pos, &offset);
562 int ret2 = 0;
563
564 if (have_bucket_gens_key && bkey_cmp(iter.pos, pos)) {
565 ret2 = bch2_btree_insert_trans(trans, BTREE_ID_bucket_gens, &g.k_i, 0) ?:
566 bch2_trans_commit(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc);
567 if (ret2)
568 goto iter_err;
569 have_bucket_gens_key = false;
570 }
571
572 if (!have_bucket_gens_key) {
573 bkey_bucket_gens_init(&g.k_i);
574 g.k.p = pos;
575 have_bucket_gens_key = true;
576 }
577
578 g.v.gens[offset] = gen;
579 iter_err:
580 ret2;
581 }));
582
583 if (have_bucket_gens_key && !ret)
584 ret = commit_do(trans, NULL, NULL,
585 BCH_TRANS_COMMIT_no_enospc,
586 bch2_btree_insert_trans(trans, BTREE_ID_bucket_gens, &g.k_i, 0));
587
588 bch2_trans_put(trans);
589
590 bch_err_fn(c, ret);
591 return ret;
592 }
593
594 int bch2_alloc_read(struct bch_fs *c)
595 {
596 struct btree_trans *trans = bch2_trans_get(c);
597 struct bch_dev *ca = NULL;
598 int ret;
599
600 if (c->sb.version_upgrade_complete >= bcachefs_metadata_version_bucket_gens) {
601 ret = for_each_btree_key(trans, iter, BTREE_ID_bucket_gens, POS_MIN,
602 BTREE_ITER_prefetch, k, ({
603 u64 start = bucket_gens_pos_to_alloc(k.k->p, 0).offset;
604 u64 end = bucket_gens_pos_to_alloc(bpos_nosnap_successor(k.k->p), 0).offset;
605
606 if (k.k->type != KEY_TYPE_bucket_gens)
607 continue;
608
609 ca = bch2_dev_iterate(c, ca, k.k->p.inode);
610 /*
611 * Not a fsck error because this is checked/repaired by
612 * bch2_check_alloc_key() which runs later:
613 */
614 if (!ca) {
615 bch2_btree_iter_set_pos(&iter, POS(k.k->p.inode + 1, 0));
616 continue;
617 }
618
619 const struct bch_bucket_gens *g = bkey_s_c_to_bucket_gens(k).v;
620
621 for (u64 b = max_t(u64, ca->mi.first_bucket, start);
622 b < min_t(u64, ca->mi.nbuckets, end);
623 b++)
624 *bucket_gen(ca, b) = g->gens[b & KEY_TYPE_BUCKET_GENS_MASK];
625 0;
626 }));
627 } else {
628 ret = for_each_btree_key(trans, iter, BTREE_ID_alloc, POS_MIN,
629 BTREE_ITER_prefetch, k, ({
630 ca = bch2_dev_iterate(c, ca, k.k->p.inode);
631 /*
632 * Not a fsck error because this is checked/repaired by
633 * bch2_check_alloc_key() which runs later:
634 */
635 if (!ca) {
636 bch2_btree_iter_set_pos(&iter, POS(k.k->p.inode + 1, 0));
637 continue;
638 }
639
640 struct bch_alloc_v4 a;
641 *bucket_gen(ca, k.k->p.offset) = bch2_alloc_to_v4(k, &a)->gen;
642 0;
643 }));
644 }
645
646 bch2_dev_put(ca);
647 bch2_trans_put(trans);
648
649 bch_err_fn(c, ret);
650 return ret;
651 }
652
653 /* Free space/discard btree: */
654
655 static int bch2_bucket_do_index(struct btree_trans *trans,
656 struct bch_dev *ca,
657 struct bkey_s_c alloc_k,
658 const struct bch_alloc_v4 *a,
659 bool set)
660 {
661 struct bch_fs *c = trans->c;
662 struct btree_iter iter;
663 struct bkey_s_c old;
664 struct bkey_i *k;
665 enum btree_id btree;
666 enum bch_bkey_type old_type = !set ? KEY_TYPE_set : KEY_TYPE_deleted;
667 enum bch_bkey_type new_type = set ? KEY_TYPE_set : KEY_TYPE_deleted;
668 struct printbuf buf = PRINTBUF;
669 int ret;
670
671 if (a->data_type != BCH_DATA_free &&
672 a->data_type != BCH_DATA_need_discard)
673 return 0;
674
675 k = bch2_trans_kmalloc_nomemzero(trans, sizeof(*k));
676 if (IS_ERR(k))
677 return PTR_ERR(k);
678
679 bkey_init(&k->k);
680 k->k.type = new_type;
681
682 switch (a->data_type) {
683 case BCH_DATA_free:
684 btree = BTREE_ID_freespace;
685 k->k.p = alloc_freespace_pos(alloc_k.k->p, *a);
686 bch2_key_resize(&k->k, 1);
687 break;
688 case BCH_DATA_need_discard:
689 btree = BTREE_ID_need_discard;
690 k->k.p = alloc_k.k->p;
691 break;
692 default:
693 return 0;
694 }
695
696 old = bch2_bkey_get_iter(trans, &iter, btree,
697 bkey_start_pos(&k->k),
698 BTREE_ITER_intent);
699 ret = bkey_err(old);
700 if (ret)
701 return ret;
702
703 if (ca->mi.freespace_initialized &&
704 c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_info &&
705 bch2_trans_inconsistent_on(old.k->type != old_type, trans,
706 "incorrect key when %s %s:%llu:%llu:0 (got %s should be %s)\n"
707 " for %s",
708 set ? "setting" : "clearing",
709 bch2_btree_id_str(btree),
710 iter.pos.inode,
711 iter.pos.offset,
712 bch2_bkey_types[old.k->type],
713 bch2_bkey_types[old_type],
714 (bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf))) {
715 ret = -EIO;
716 goto err;
717 }
718
719 ret = bch2_trans_update(trans, &iter, k, 0);
720 err:
721 bch2_trans_iter_exit(trans, &iter);
722 printbuf_exit(&buf);
723 return ret;
724 }
725
726 static noinline int bch2_bucket_gen_update(struct btree_trans *trans,
727 struct bpos bucket, u8 gen)
728 {
729 struct btree_iter iter;
730 unsigned offset;
731 struct bpos pos = alloc_gens_pos(bucket, &offset);
732 struct bkey_i_bucket_gens *g;
733 struct bkey_s_c k;
734 int ret;
735
736 g = bch2_trans_kmalloc(trans, sizeof(*g));
737 ret = PTR_ERR_OR_ZERO(g);
738 if (ret)
739 return ret;
740
741 k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_bucket_gens, pos,
742 BTREE_ITER_intent|
743 BTREE_ITER_with_updates);
744 ret = bkey_err(k);
745 if (ret)
746 return ret;
747
748 if (k.k->type != KEY_TYPE_bucket_gens) {
749 bkey_bucket_gens_init(&g->k_i);
750 g->k.p = iter.pos;
751 } else {
752 bkey_reassemble(&g->k_i, k);
753 }
754
755 g->v.gens[offset] = gen;
756
757 ret = bch2_trans_update(trans, &iter, &g->k_i, 0);
758 bch2_trans_iter_exit(trans, &iter);
759 return ret;
760 }
761
762 static inline int bch2_dev_data_type_accounting_mod(struct btree_trans *trans, struct bch_dev *ca,
763 enum bch_data_type data_type,
764 s64 delta_buckets,
765 s64 delta_sectors,
766 s64 delta_fragmented, unsigned flags)
767 {
768 struct disk_accounting_pos acc = {
769 .type = BCH_DISK_ACCOUNTING_dev_data_type,
770 .dev_data_type.dev = ca->dev_idx,
771 .dev_data_type.data_type = data_type,
772 };
773 s64 d[3] = { delta_buckets, delta_sectors, delta_fragmented };
774
775 return bch2_disk_accounting_mod(trans, &acc, d, 3, flags & BTREE_TRIGGER_gc);
776 }
777
778 int bch2_alloc_key_to_dev_counters(struct btree_trans *trans, struct bch_dev *ca,
779 const struct bch_alloc_v4 *old,
780 const struct bch_alloc_v4 *new,
781 unsigned flags)
782 {
783 s64 old_sectors = bch2_bucket_sectors(*old);
784 s64 new_sectors = bch2_bucket_sectors(*new);
785 if (old->data_type != new->data_type) {
786 int ret = bch2_dev_data_type_accounting_mod(trans, ca, new->data_type,
787 1, new_sectors, bch2_bucket_sectors_fragmented(ca, *new), flags) ?:
788 bch2_dev_data_type_accounting_mod(trans, ca, old->data_type,
789 -1, -old_sectors, -bch2_bucket_sectors_fragmented(ca, *old), flags);
790 if (ret)
791 return ret;
792 } else if (old_sectors != new_sectors) {
793 int ret = bch2_dev_data_type_accounting_mod(trans, ca, new->data_type,
794 0,
795 new_sectors - old_sectors,
796 bch2_bucket_sectors_fragmented(ca, *new) -
797 bch2_bucket_sectors_fragmented(ca, *old), flags);
798 if (ret)
799 return ret;
800 }
801
802 s64 old_unstriped = bch2_bucket_sectors_unstriped(*old);
803 s64 new_unstriped = bch2_bucket_sectors_unstriped(*new);
804 if (old_unstriped != new_unstriped) {
805 int ret = bch2_dev_data_type_accounting_mod(trans, ca, BCH_DATA_unstriped,
806 !!new_unstriped - !!old_unstriped,
807 new_unstriped - old_unstriped,
808 0,
809 flags);
810 if (ret)
811 return ret;
812 }
813
814 return 0;
815 }
816
817 int bch2_trigger_alloc(struct btree_trans *trans,
818 enum btree_id btree, unsigned level,
819 struct bkey_s_c old, struct bkey_s new,
820 enum btree_iter_update_trigger_flags flags)
821 {
822 struct bch_fs *c = trans->c;
823 struct printbuf buf = PRINTBUF;
824 int ret = 0;
825
826 struct bch_dev *ca = bch2_dev_bucket_tryget(c, new.k->p);
827 if (!ca)
828 return -EIO;
829
830 struct bch_alloc_v4 old_a_convert;
831 const struct bch_alloc_v4 *old_a = bch2_alloc_to_v4(old, &old_a_convert);
832 struct bch_alloc_v4 *new_a = bkey_s_to_alloc_v4(new).v;
833
834 if (flags & BTREE_TRIGGER_transactional) {
835 alloc_data_type_set(new_a, new_a->data_type);
836
837 if (bch2_bucket_sectors_total(*new_a) > bch2_bucket_sectors_total(*old_a)) {
838 new_a->io_time[READ] = bch2_current_io_time(c, READ);
839 new_a->io_time[WRITE]= bch2_current_io_time(c, WRITE);
840 SET_BCH_ALLOC_V4_NEED_INC_GEN(new_a, true);
841 SET_BCH_ALLOC_V4_NEED_DISCARD(new_a, true);
842 }
843
844 if (data_type_is_empty(new_a->data_type) &&
845 BCH_ALLOC_V4_NEED_INC_GEN(new_a) &&
846 !bch2_bucket_is_open_safe(c, new.k->p.inode, new.k->p.offset)) {
847 new_a->gen++;
848 SET_BCH_ALLOC_V4_NEED_INC_GEN(new_a, false);
849 alloc_data_type_set(new_a, new_a->data_type);
850 }
851
852 if (old_a->data_type != new_a->data_type ||
853 (new_a->data_type == BCH_DATA_free &&
854 alloc_freespace_genbits(*old_a) != alloc_freespace_genbits(*new_a))) {
855 ret = bch2_bucket_do_index(trans, ca, old, old_a, false) ?:
856 bch2_bucket_do_index(trans, ca, new.s_c, new_a, true);
857 if (ret)
858 goto err;
859 }
860
861 if (new_a->data_type == BCH_DATA_cached &&
862 !new_a->io_time[READ])
863 new_a->io_time[READ] = bch2_current_io_time(c, READ);
864
865 u64 old_lru = alloc_lru_idx_read(*old_a);
866 u64 new_lru = alloc_lru_idx_read(*new_a);
867 if (old_lru != new_lru) {
868 ret = bch2_lru_change(trans, new.k->p.inode,
869 bucket_to_u64(new.k->p),
870 old_lru, new_lru);
871 if (ret)
872 goto err;
873 }
874
875 new_a->fragmentation_lru = alloc_lru_idx_fragmentation(*new_a, ca);
876 if (old_a->fragmentation_lru != new_a->fragmentation_lru) {
877 ret = bch2_lru_change(trans,
878 BCH_LRU_FRAGMENTATION_START,
879 bucket_to_u64(new.k->p),
880 old_a->fragmentation_lru, new_a->fragmentation_lru);
881 if (ret)
882 goto err;
883 }
884
885 if (old_a->gen != new_a->gen) {
886 ret = bch2_bucket_gen_update(trans, new.k->p, new_a->gen);
887 if (ret)
888 goto err;
889 }
890
891 if ((flags & BTREE_TRIGGER_bucket_invalidate) &&
892 old_a->cached_sectors) {
893 ret = bch2_mod_dev_cached_sectors(trans, ca->dev_idx,
894 -((s64) old_a->cached_sectors),
895 flags & BTREE_TRIGGER_gc);
896 if (ret)
897 goto err;
898 }
899
900 ret = bch2_alloc_key_to_dev_counters(trans, ca, old_a, new_a, flags);
901 if (ret)
902 goto err;
903 }
904
905 if ((flags & BTREE_TRIGGER_atomic) && (flags & BTREE_TRIGGER_insert)) {
906 u64 journal_seq = trans->journal_res.seq;
907 u64 bucket_journal_seq = new_a->journal_seq;
908
909 if ((flags & BTREE_TRIGGER_insert) &&
910 data_type_is_empty(old_a->data_type) !=
911 data_type_is_empty(new_a->data_type) &&
912 new.k->type == KEY_TYPE_alloc_v4) {
913 struct bch_alloc_v4 *v = bkey_s_to_alloc_v4(new).v;
914
915 /*
916 * If the btree updates referring to a bucket weren't flushed
917 * before the bucket became empty again, then the we don't have
918 * to wait on a journal flush before we can reuse the bucket:
919 */
920 v->journal_seq = bucket_journal_seq =
921 data_type_is_empty(new_a->data_type) &&
922 (journal_seq == v->journal_seq ||
923 bch2_journal_noflush_seq(&c->journal, v->journal_seq))
924 ? 0 : journal_seq;
925 }
926
927 if (!data_type_is_empty(old_a->data_type) &&
928 data_type_is_empty(new_a->data_type) &&
929 bucket_journal_seq) {
930 ret = bch2_set_bucket_needs_journal_commit(&c->buckets_waiting_for_journal,
931 c->journal.flushed_seq_ondisk,
932 new.k->p.inode, new.k->p.offset,
933 bucket_journal_seq);
934 if (bch2_fs_fatal_err_on(ret, c,
935 "setting bucket_needs_journal_commit: %s", bch2_err_str(ret)))
936 goto err;
937 }
938
939 if (new_a->gen != old_a->gen) {
940 rcu_read_lock();
941 u8 *gen = bucket_gen(ca, new.k->p.offset);
942 if (unlikely(!gen)) {
943 rcu_read_unlock();
944 goto invalid_bucket;
945 }
946 *gen = new_a->gen;
947 rcu_read_unlock();
948 }
949
950 #define eval_state(_a, expr) ({ const struct bch_alloc_v4 *a = _a; expr; })
951 #define statechange(expr) !eval_state(old_a, expr) && eval_state(new_a, expr)
952 #define bucket_flushed(a) (!a->journal_seq || a->journal_seq <= c->journal.flushed_seq_ondisk)
953
954 if (statechange(a->data_type == BCH_DATA_free) &&
955 bucket_flushed(new_a))
956 closure_wake_up(&c->freelist_wait);
957
958 if (statechange(a->data_type == BCH_DATA_need_discard) &&
959 !bch2_bucket_is_open_safe(c, new.k->p.inode, new.k->p.offset) &&
960 bucket_flushed(new_a))
961 bch2_discard_one_bucket_fast(ca, new.k->p.offset);
962
963 if (statechange(a->data_type == BCH_DATA_cached) &&
964 !bch2_bucket_is_open(c, new.k->p.inode, new.k->p.offset) &&
965 should_invalidate_buckets(ca, bch2_dev_usage_read(ca)))
966 bch2_dev_do_invalidates(ca);
967
968 if (statechange(a->data_type == BCH_DATA_need_gc_gens))
969 bch2_gc_gens_async(c);
970 }
971
972 if ((flags & BTREE_TRIGGER_gc) && (flags & BTREE_TRIGGER_insert)) {
973 rcu_read_lock();
974 struct bucket *g = gc_bucket(ca, new.k->p.offset);
975 if (unlikely(!g)) {
976 rcu_read_unlock();
977 goto invalid_bucket;
978 }
979 g->gen_valid = 1;
980 g->gen = new_a->gen;
981 rcu_read_unlock();
982 }
983 err:
984 printbuf_exit(&buf);
985 bch2_dev_put(ca);
986 return ret;
987 invalid_bucket:
988 bch2_fs_inconsistent(c, "reference to invalid bucket\n %s",
989 (bch2_bkey_val_to_text(&buf, c, new.s_c), buf.buf));
990 ret = -EIO;
991 goto err;
992 }
993
994 /*
995 * This synthesizes deleted extents for holes, similar to BTREE_ITER_slots for
996 * extents style btrees, but works on non-extents btrees:
997 */
998 static struct bkey_s_c bch2_get_key_or_hole(struct btree_iter *iter, struct bpos end, struct bkey *hole)
999 {
1000 struct bkey_s_c k = bch2_btree_iter_peek_slot(iter);
1001
1002 if (bkey_err(k))
1003 return k;
1004
1005 if (k.k->type) {
1006 return k;
1007 } else {
1008 struct btree_iter iter2;
1009 struct bpos next;
1010
1011 bch2_trans_copy_iter(&iter2, iter);
1012
1013 struct btree_path *path = btree_iter_path(iter->trans, iter);
1014 if (!bpos_eq(path->l[0].b->key.k.p, SPOS_MAX))
1015 end = bkey_min(end, bpos_nosnap_successor(path->l[0].b->key.k.p));
1016
1017 end = bkey_min(end, POS(iter->pos.inode, iter->pos.offset + U32_MAX - 1));
1018
1019 /*
1020 * btree node min/max is a closed interval, upto takes a half
1021 * open interval:
1022 */
1023 k = bch2_btree_iter_peek_upto(&iter2, end);
1024 next = iter2.pos;
1025 bch2_trans_iter_exit(iter->trans, &iter2);
1026
1027 BUG_ON(next.offset >= iter->pos.offset + U32_MAX);
1028
1029 if (bkey_err(k))
1030 return k;
1031
1032 bkey_init(hole);
1033 hole->p = iter->pos;
1034
1035 bch2_key_resize(hole, next.offset - iter->pos.offset);
1036 return (struct bkey_s_c) { hole, NULL };
1037 }
1038 }
1039
1040 static bool next_bucket(struct bch_fs *c, struct bch_dev **ca, struct bpos *bucket)
1041 {
1042 if (*ca) {
1043 if (bucket->offset < (*ca)->mi.first_bucket)
1044 bucket->offset = (*ca)->mi.first_bucket;
1045
1046 if (bucket->offset < (*ca)->mi.nbuckets)
1047 return true;
1048
1049 bch2_dev_put(*ca);
1050 *ca = NULL;
1051 bucket->inode++;
1052 bucket->offset = 0;
1053 }
1054
1055 rcu_read_lock();
1056 *ca = __bch2_next_dev_idx(c, bucket->inode, NULL);
1057 if (*ca) {
1058 *bucket = POS((*ca)->dev_idx, (*ca)->mi.first_bucket);
1059 bch2_dev_get(*ca);
1060 }
1061 rcu_read_unlock();
1062
1063 return *ca != NULL;
1064 }
1065
1066 static struct bkey_s_c bch2_get_key_or_real_bucket_hole(struct btree_iter *iter,
1067 struct bch_dev **ca, struct bkey *hole)
1068 {
1069 struct bch_fs *c = iter->trans->c;
1070 struct bkey_s_c k;
1071 again:
1072 k = bch2_get_key_or_hole(iter, POS_MAX, hole);
1073 if (bkey_err(k))
1074 return k;
1075
1076 *ca = bch2_dev_iterate_noerror(c, *ca, k.k->p.inode);
1077
1078 if (!k.k->type) {
1079 struct bpos hole_start = bkey_start_pos(k.k);
1080
1081 if (!*ca || !bucket_valid(*ca, hole_start.offset)) {
1082 if (!next_bucket(c, ca, &hole_start))
1083 return bkey_s_c_null;
1084
1085 bch2_btree_iter_set_pos(iter, hole_start);
1086 goto again;
1087 }
1088
1089 if (k.k->p.offset > (*ca)->mi.nbuckets)
1090 bch2_key_resize(hole, (*ca)->mi.nbuckets - hole_start.offset);
1091 }
1092
1093 return k;
1094 }
1095
1096 static noinline_for_stack
1097 int bch2_check_alloc_key(struct btree_trans *trans,
1098 struct bkey_s_c alloc_k,
1099 struct btree_iter *alloc_iter,
1100 struct btree_iter *discard_iter,
1101 struct btree_iter *freespace_iter,
1102 struct btree_iter *bucket_gens_iter)
1103 {
1104 struct bch_fs *c = trans->c;
1105 struct bch_alloc_v4 a_convert;
1106 const struct bch_alloc_v4 *a;
1107 unsigned discard_key_type, freespace_key_type;
1108 unsigned gens_offset;
1109 struct bkey_s_c k;
1110 struct printbuf buf = PRINTBUF;
1111 int ret = 0;
1112
1113 struct bch_dev *ca = bch2_dev_bucket_tryget_noerror(c, alloc_k.k->p);
1114 if (fsck_err_on(!ca,
1115 trans, alloc_key_to_missing_dev_bucket,
1116 "alloc key for invalid device:bucket %llu:%llu",
1117 alloc_k.k->p.inode, alloc_k.k->p.offset))
1118 ret = bch2_btree_delete_at(trans, alloc_iter, 0);
1119 if (!ca)
1120 return ret;
1121
1122 if (!ca->mi.freespace_initialized)
1123 goto out;
1124
1125 a = bch2_alloc_to_v4(alloc_k, &a_convert);
1126
1127 discard_key_type = a->data_type == BCH_DATA_need_discard ? KEY_TYPE_set : 0;
1128 bch2_btree_iter_set_pos(discard_iter, alloc_k.k->p);
1129 k = bch2_btree_iter_peek_slot(discard_iter);
1130 ret = bkey_err(k);
1131 if (ret)
1132 goto err;
1133
1134 if (fsck_err_on(k.k->type != discard_key_type,
1135 trans, need_discard_key_wrong,
1136 "incorrect key in need_discard btree (got %s should be %s)\n"
1137 " %s",
1138 bch2_bkey_types[k.k->type],
1139 bch2_bkey_types[discard_key_type],
1140 (bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf))) {
1141 struct bkey_i *update =
1142 bch2_trans_kmalloc(trans, sizeof(*update));
1143
1144 ret = PTR_ERR_OR_ZERO(update);
1145 if (ret)
1146 goto err;
1147
1148 bkey_init(&update->k);
1149 update->k.type = discard_key_type;
1150 update->k.p = discard_iter->pos;
1151
1152 ret = bch2_trans_update(trans, discard_iter, update, 0);
1153 if (ret)
1154 goto err;
1155 }
1156
1157 freespace_key_type = a->data_type == BCH_DATA_free ? KEY_TYPE_set : 0;
1158 bch2_btree_iter_set_pos(freespace_iter, alloc_freespace_pos(alloc_k.k->p, *a));
1159 k = bch2_btree_iter_peek_slot(freespace_iter);
1160 ret = bkey_err(k);
1161 if (ret)
1162 goto err;
1163
1164 if (fsck_err_on(k.k->type != freespace_key_type,
1165 trans, freespace_key_wrong,
1166 "incorrect key in freespace btree (got %s should be %s)\n"
1167 " %s",
1168 bch2_bkey_types[k.k->type],
1169 bch2_bkey_types[freespace_key_type],
1170 (printbuf_reset(&buf),
1171 bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf))) {
1172 struct bkey_i *update =
1173 bch2_trans_kmalloc(trans, sizeof(*update));
1174
1175 ret = PTR_ERR_OR_ZERO(update);
1176 if (ret)
1177 goto err;
1178
1179 bkey_init(&update->k);
1180 update->k.type = freespace_key_type;
1181 update->k.p = freespace_iter->pos;
1182 bch2_key_resize(&update->k, 1);
1183
1184 ret = bch2_trans_update(trans, freespace_iter, update, 0);
1185 if (ret)
1186 goto err;
1187 }
1188
1189 bch2_btree_iter_set_pos(bucket_gens_iter, alloc_gens_pos(alloc_k.k->p, &gens_offset));
1190 k = bch2_btree_iter_peek_slot(bucket_gens_iter);
1191 ret = bkey_err(k);
1192 if (ret)
1193 goto err;
1194
1195 if (fsck_err_on(a->gen != alloc_gen(k, gens_offset),
1196 trans, bucket_gens_key_wrong,
1197 "incorrect gen in bucket_gens btree (got %u should be %u)\n"
1198 " %s",
1199 alloc_gen(k, gens_offset), a->gen,
1200 (printbuf_reset(&buf),
1201 bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf))) {
1202 struct bkey_i_bucket_gens *g =
1203 bch2_trans_kmalloc(trans, sizeof(*g));
1204
1205 ret = PTR_ERR_OR_ZERO(g);
1206 if (ret)
1207 goto err;
1208
1209 if (k.k->type == KEY_TYPE_bucket_gens) {
1210 bkey_reassemble(&g->k_i, k);
1211 } else {
1212 bkey_bucket_gens_init(&g->k_i);
1213 g->k.p = alloc_gens_pos(alloc_k.k->p, &gens_offset);
1214 }
1215
1216 g->v.gens[gens_offset] = a->gen;
1217
1218 ret = bch2_trans_update(trans, bucket_gens_iter, &g->k_i, 0);
1219 if (ret)
1220 goto err;
1221 }
1222 out:
1223 err:
1224 fsck_err:
1225 bch2_dev_put(ca);
1226 printbuf_exit(&buf);
1227 return ret;
1228 }
1229
1230 static noinline_for_stack
1231 int bch2_check_alloc_hole_freespace(struct btree_trans *trans,
1232 struct bch_dev *ca,
1233 struct bpos start,
1234 struct bpos *end,
1235 struct btree_iter *freespace_iter)
1236 {
1237 struct bkey_s_c k;
1238 struct printbuf buf = PRINTBUF;
1239 int ret;
1240
1241 if (!ca->mi.freespace_initialized)
1242 return 0;
1243
1244 bch2_btree_iter_set_pos(freespace_iter, start);
1245
1246 k = bch2_btree_iter_peek_slot(freespace_iter);
1247 ret = bkey_err(k);
1248 if (ret)
1249 goto err;
1250
1251 *end = bkey_min(k.k->p, *end);
1252
1253 if (fsck_err_on(k.k->type != KEY_TYPE_set,
1254 trans, freespace_hole_missing,
1255 "hole in alloc btree missing in freespace btree\n"
1256 " device %llu buckets %llu-%llu",
1257 freespace_iter->pos.inode,
1258 freespace_iter->pos.offset,
1259 end->offset)) {
1260 struct bkey_i *update =
1261 bch2_trans_kmalloc(trans, sizeof(*update));
1262
1263 ret = PTR_ERR_OR_ZERO(update);
1264 if (ret)
1265 goto err;
1266
1267 bkey_init(&update->k);
1268 update->k.type = KEY_TYPE_set;
1269 update->k.p = freespace_iter->pos;
1270 bch2_key_resize(&update->k,
1271 min_t(u64, U32_MAX, end->offset -
1272 freespace_iter->pos.offset));
1273
1274 ret = bch2_trans_update(trans, freespace_iter, update, 0);
1275 if (ret)
1276 goto err;
1277 }
1278 err:
1279 fsck_err:
1280 printbuf_exit(&buf);
1281 return ret;
1282 }
1283
1284 static noinline_for_stack
1285 int bch2_check_alloc_hole_bucket_gens(struct btree_trans *trans,
1286 struct bpos start,
1287 struct bpos *end,
1288 struct btree_iter *bucket_gens_iter)
1289 {
1290 struct bkey_s_c k;
1291 struct printbuf buf = PRINTBUF;
1292 unsigned i, gens_offset, gens_end_offset;
1293 int ret;
1294
1295 bch2_btree_iter_set_pos(bucket_gens_iter, alloc_gens_pos(start, &gens_offset));
1296
1297 k = bch2_btree_iter_peek_slot(bucket_gens_iter);
1298 ret = bkey_err(k);
1299 if (ret)
1300 goto err;
1301
1302 if (bkey_cmp(alloc_gens_pos(start, &gens_offset),
1303 alloc_gens_pos(*end, &gens_end_offset)))
1304 gens_end_offset = KEY_TYPE_BUCKET_GENS_NR;
1305
1306 if (k.k->type == KEY_TYPE_bucket_gens) {
1307 struct bkey_i_bucket_gens g;
1308 bool need_update = false;
1309
1310 bkey_reassemble(&g.k_i, k);
1311
1312 for (i = gens_offset; i < gens_end_offset; i++) {
1313 if (fsck_err_on(g.v.gens[i], trans,
1314 bucket_gens_hole_wrong,
1315 "hole in alloc btree at %llu:%llu with nonzero gen in bucket_gens btree (%u)",
1316 bucket_gens_pos_to_alloc(k.k->p, i).inode,
1317 bucket_gens_pos_to_alloc(k.k->p, i).offset,
1318 g.v.gens[i])) {
1319 g.v.gens[i] = 0;
1320 need_update = true;
1321 }
1322 }
1323
1324 if (need_update) {
1325 struct bkey_i *u = bch2_trans_kmalloc(trans, sizeof(g));
1326
1327 ret = PTR_ERR_OR_ZERO(u);
1328 if (ret)
1329 goto err;
1330
1331 memcpy(u, &g, sizeof(g));
1332
1333 ret = bch2_trans_update(trans, bucket_gens_iter, u, 0);
1334 if (ret)
1335 goto err;
1336 }
1337 }
1338
1339 *end = bkey_min(*end, bucket_gens_pos_to_alloc(bpos_nosnap_successor(k.k->p), 0));
1340 err:
1341 fsck_err:
1342 printbuf_exit(&buf);
1343 return ret;
1344 }
1345
1346 static noinline_for_stack int bch2_check_discard_freespace_key(struct btree_trans *trans,
1347 struct btree_iter *iter)
1348 {
1349 struct bch_fs *c = trans->c;
1350 struct btree_iter alloc_iter;
1351 struct bkey_s_c alloc_k;
1352 struct bch_alloc_v4 a_convert;
1353 const struct bch_alloc_v4 *a;
1354 u64 genbits;
1355 struct bpos pos;
1356 enum bch_data_type state = iter->btree_id == BTREE_ID_need_discard
1357 ? BCH_DATA_need_discard
1358 : BCH_DATA_free;
1359 struct printbuf buf = PRINTBUF;
1360 int ret;
1361
1362 pos = iter->pos;
1363 pos.offset &= ~(~0ULL << 56);
1364 genbits = iter->pos.offset & (~0ULL << 56);
1365
1366 alloc_k = bch2_bkey_get_iter(trans, &alloc_iter, BTREE_ID_alloc, pos, 0);
1367 ret = bkey_err(alloc_k);
1368 if (ret)
1369 return ret;
1370
1371 if (fsck_err_on(!bch2_dev_bucket_exists(c, pos),
1372 trans, need_discard_freespace_key_to_invalid_dev_bucket,
1373 "entry in %s btree for nonexistant dev:bucket %llu:%llu",
1374 bch2_btree_id_str(iter->btree_id), pos.inode, pos.offset))
1375 goto delete;
1376
1377 a = bch2_alloc_to_v4(alloc_k, &a_convert);
1378
1379 if (fsck_err_on(a->data_type != state ||
1380 (state == BCH_DATA_free &&
1381 genbits != alloc_freespace_genbits(*a)),
1382 trans, need_discard_freespace_key_bad,
1383 "%s\n incorrectly set at %s:%llu:%llu:0 (free %u, genbits %llu should be %llu)",
1384 (bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf),
1385 bch2_btree_id_str(iter->btree_id),
1386 iter->pos.inode,
1387 iter->pos.offset,
1388 a->data_type == state,
1389 genbits >> 56, alloc_freespace_genbits(*a) >> 56))
1390 goto delete;
1391 out:
1392 fsck_err:
1393 bch2_set_btree_iter_dontneed(&alloc_iter);
1394 bch2_trans_iter_exit(trans, &alloc_iter);
1395 printbuf_exit(&buf);
1396 return ret;
1397 delete:
1398 ret = bch2_btree_delete_extent_at(trans, iter,
1399 iter->btree_id == BTREE_ID_freespace ? 1 : 0, 0) ?:
1400 bch2_trans_commit(trans, NULL, NULL,
1401 BCH_TRANS_COMMIT_no_enospc);
1402 goto out;
1403 }
1404
1405 /*
1406 * We've already checked that generation numbers in the bucket_gens btree are
1407 * valid for buckets that exist; this just checks for keys for nonexistent
1408 * buckets.
1409 */
1410 static noinline_for_stack
1411 int bch2_check_bucket_gens_key(struct btree_trans *trans,
1412 struct btree_iter *iter,
1413 struct bkey_s_c k)
1414 {
1415 struct bch_fs *c = trans->c;
1416 struct bkey_i_bucket_gens g;
1417 u64 start = bucket_gens_pos_to_alloc(k.k->p, 0).offset;
1418 u64 end = bucket_gens_pos_to_alloc(bpos_nosnap_successor(k.k->p), 0).offset;
1419 u64 b;
1420 bool need_update = false;
1421 struct printbuf buf = PRINTBUF;
1422 int ret = 0;
1423
1424 BUG_ON(k.k->type != KEY_TYPE_bucket_gens);
1425 bkey_reassemble(&g.k_i, k);
1426
1427 struct bch_dev *ca = bch2_dev_tryget_noerror(c, k.k->p.inode);
1428 if (!ca) {
1429 if (fsck_err(trans, bucket_gens_to_invalid_dev,
1430 "bucket_gens key for invalid device:\n %s",
1431 (bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
1432 ret = bch2_btree_delete_at(trans, iter, 0);
1433 goto out;
1434 }
1435
1436 if (fsck_err_on(end <= ca->mi.first_bucket ||
1437 start >= ca->mi.nbuckets,
1438 trans, bucket_gens_to_invalid_buckets,
1439 "bucket_gens key for invalid buckets:\n %s",
1440 (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1441 ret = bch2_btree_delete_at(trans, iter, 0);
1442 goto out;
1443 }
1444
1445 for (b = start; b < ca->mi.first_bucket; b++)
1446 if (fsck_err_on(g.v.gens[b & KEY_TYPE_BUCKET_GENS_MASK],
1447 trans, bucket_gens_nonzero_for_invalid_buckets,
1448 "bucket_gens key has nonzero gen for invalid bucket")) {
1449 g.v.gens[b & KEY_TYPE_BUCKET_GENS_MASK] = 0;
1450 need_update = true;
1451 }
1452
1453 for (b = ca->mi.nbuckets; b < end; b++)
1454 if (fsck_err_on(g.v.gens[b & KEY_TYPE_BUCKET_GENS_MASK],
1455 trans, bucket_gens_nonzero_for_invalid_buckets,
1456 "bucket_gens key has nonzero gen for invalid bucket")) {
1457 g.v.gens[b & KEY_TYPE_BUCKET_GENS_MASK] = 0;
1458 need_update = true;
1459 }
1460
1461 if (need_update) {
1462 struct bkey_i *u = bch2_trans_kmalloc(trans, sizeof(g));
1463
1464 ret = PTR_ERR_OR_ZERO(u);
1465 if (ret)
1466 goto out;
1467
1468 memcpy(u, &g, sizeof(g));
1469 ret = bch2_trans_update(trans, iter, u, 0);
1470 }
1471 out:
1472 fsck_err:
1473 bch2_dev_put(ca);
1474 printbuf_exit(&buf);
1475 return ret;
1476 }
1477
1478 int bch2_check_alloc_info(struct bch_fs *c)
1479 {
1480 struct btree_trans *trans = bch2_trans_get(c);
1481 struct btree_iter iter, discard_iter, freespace_iter, bucket_gens_iter;
1482 struct bch_dev *ca = NULL;
1483 struct bkey hole;
1484 struct bkey_s_c k;
1485 int ret = 0;
1486
1487 bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc, POS_MIN,
1488 BTREE_ITER_prefetch);
1489 bch2_trans_iter_init(trans, &discard_iter, BTREE_ID_need_discard, POS_MIN,
1490 BTREE_ITER_prefetch);
1491 bch2_trans_iter_init(trans, &freespace_iter, BTREE_ID_freespace, POS_MIN,
1492 BTREE_ITER_prefetch);
1493 bch2_trans_iter_init(trans, &bucket_gens_iter, BTREE_ID_bucket_gens, POS_MIN,
1494 BTREE_ITER_prefetch);
1495
1496 while (1) {
1497 struct bpos next;
1498
1499 bch2_trans_begin(trans);
1500
1501 k = bch2_get_key_or_real_bucket_hole(&iter, &ca, &hole);
1502 ret = bkey_err(k);
1503 if (ret)
1504 goto bkey_err;
1505
1506 if (!k.k)
1507 break;
1508
1509 if (k.k->type) {
1510 next = bpos_nosnap_successor(k.k->p);
1511
1512 ret = bch2_check_alloc_key(trans,
1513 k, &iter,
1514 &discard_iter,
1515 &freespace_iter,
1516 &bucket_gens_iter);
1517 if (ret)
1518 goto bkey_err;
1519 } else {
1520 next = k.k->p;
1521
1522 ret = bch2_check_alloc_hole_freespace(trans, ca,
1523 bkey_start_pos(k.k),
1524 &next,
1525 &freespace_iter) ?:
1526 bch2_check_alloc_hole_bucket_gens(trans,
1527 bkey_start_pos(k.k),
1528 &next,
1529 &bucket_gens_iter);
1530 if (ret)
1531 goto bkey_err;
1532 }
1533
1534 ret = bch2_trans_commit(trans, NULL, NULL,
1535 BCH_TRANS_COMMIT_no_enospc);
1536 if (ret)
1537 goto bkey_err;
1538
1539 bch2_btree_iter_set_pos(&iter, next);
1540 bkey_err:
1541 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
1542 continue;
1543 if (ret)
1544 break;
1545 }
1546 bch2_trans_iter_exit(trans, &bucket_gens_iter);
1547 bch2_trans_iter_exit(trans, &freespace_iter);
1548 bch2_trans_iter_exit(trans, &discard_iter);
1549 bch2_trans_iter_exit(trans, &iter);
1550 bch2_dev_put(ca);
1551 ca = NULL;
1552
1553 if (ret < 0)
1554 goto err;
1555
1556 ret = for_each_btree_key(trans, iter,
1557 BTREE_ID_need_discard, POS_MIN,
1558 BTREE_ITER_prefetch, k,
1559 bch2_check_discard_freespace_key(trans, &iter));
1560 if (ret)
1561 goto err;
1562
1563 bch2_trans_iter_init(trans, &iter, BTREE_ID_freespace, POS_MIN,
1564 BTREE_ITER_prefetch);
1565 while (1) {
1566 bch2_trans_begin(trans);
1567 k = bch2_btree_iter_peek(&iter);
1568 if (!k.k)
1569 break;
1570
1571 ret = bkey_err(k) ?:
1572 bch2_check_discard_freespace_key(trans, &iter);
1573 if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) {
1574 ret = 0;
1575 continue;
1576 }
1577 if (ret) {
1578 struct printbuf buf = PRINTBUF;
1579 bch2_bkey_val_to_text(&buf, c, k);
1580
1581 bch_err(c, "while checking %s", buf.buf);
1582 printbuf_exit(&buf);
1583 break;
1584 }
1585
1586 bch2_btree_iter_set_pos(&iter, bpos_nosnap_successor(iter.pos));
1587 }
1588 bch2_trans_iter_exit(trans, &iter);
1589 if (ret)
1590 goto err;
1591
1592 ret = for_each_btree_key_commit(trans, iter,
1593 BTREE_ID_bucket_gens, POS_MIN,
1594 BTREE_ITER_prefetch, k,
1595 NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
1596 bch2_check_bucket_gens_key(trans, &iter, k));
1597 err:
1598 bch2_trans_put(trans);
1599 bch_err_fn(c, ret);
1600 return ret;
1601 }
1602
1603 static int bch2_check_alloc_to_lru_ref(struct btree_trans *trans,
1604 struct btree_iter *alloc_iter,
1605 struct bkey_buf *last_flushed)
1606 {
1607 struct bch_fs *c = trans->c;
1608 struct bch_alloc_v4 a_convert;
1609 const struct bch_alloc_v4 *a;
1610 struct bkey_s_c alloc_k;
1611 struct printbuf buf = PRINTBUF;
1612 int ret;
1613
1614 alloc_k = bch2_btree_iter_peek(alloc_iter);
1615 if (!alloc_k.k)
1616 return 0;
1617
1618 ret = bkey_err(alloc_k);
1619 if (ret)
1620 return ret;
1621
1622 a = bch2_alloc_to_v4(alloc_k, &a_convert);
1623
1624 if (a->fragmentation_lru) {
1625 ret = bch2_lru_check_set(trans, BCH_LRU_FRAGMENTATION_START,
1626 a->fragmentation_lru,
1627 alloc_k, last_flushed);
1628 if (ret)
1629 return ret;
1630 }
1631
1632 if (a->data_type != BCH_DATA_cached)
1633 return 0;
1634
1635 if (fsck_err_on(!a->io_time[READ],
1636 trans, alloc_key_cached_but_read_time_zero,
1637 "cached bucket with read_time 0\n"
1638 " %s",
1639 (printbuf_reset(&buf),
1640 bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf))) {
1641 struct bkey_i_alloc_v4 *a_mut =
1642 bch2_alloc_to_v4_mut(trans, alloc_k);
1643 ret = PTR_ERR_OR_ZERO(a_mut);
1644 if (ret)
1645 goto err;
1646
1647 a_mut->v.io_time[READ] = bch2_current_io_time(c, READ);
1648 ret = bch2_trans_update(trans, alloc_iter,
1649 &a_mut->k_i, BTREE_TRIGGER_norun);
1650 if (ret)
1651 goto err;
1652
1653 a = &a_mut->v;
1654 }
1655
1656 ret = bch2_lru_check_set(trans, alloc_k.k->p.inode, a->io_time[READ],
1657 alloc_k, last_flushed);
1658 if (ret)
1659 goto err;
1660 err:
1661 fsck_err:
1662 printbuf_exit(&buf);
1663 return ret;
1664 }
1665
1666 int bch2_check_alloc_to_lru_refs(struct bch_fs *c)
1667 {
1668 struct bkey_buf last_flushed;
1669
1670 bch2_bkey_buf_init(&last_flushed);
1671 bkey_init(&last_flushed.k->k);
1672
1673 int ret = bch2_trans_run(c,
1674 for_each_btree_key_commit(trans, iter, BTREE_ID_alloc,
1675 POS_MIN, BTREE_ITER_prefetch, k,
1676 NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
1677 bch2_check_alloc_to_lru_ref(trans, &iter, &last_flushed)));
1678
1679 bch2_bkey_buf_exit(&last_flushed, c);
1680 bch_err_fn(c, ret);
1681 return ret;
1682 }
1683
1684 static int discard_in_flight_add(struct bch_dev *ca, u64 bucket, bool in_progress)
1685 {
1686 int ret;
1687
1688 mutex_lock(&ca->discard_buckets_in_flight_lock);
1689 darray_for_each(ca->discard_buckets_in_flight, i)
1690 if (i->bucket == bucket) {
1691 ret = -BCH_ERR_EEXIST_discard_in_flight_add;
1692 goto out;
1693 }
1694
1695 ret = darray_push(&ca->discard_buckets_in_flight, ((struct discard_in_flight) {
1696 .in_progress = in_progress,
1697 .bucket = bucket,
1698 }));
1699 out:
1700 mutex_unlock(&ca->discard_buckets_in_flight_lock);
1701 return ret;
1702 }
1703
1704 static void discard_in_flight_remove(struct bch_dev *ca, u64 bucket)
1705 {
1706 mutex_lock(&ca->discard_buckets_in_flight_lock);
1707 darray_for_each(ca->discard_buckets_in_flight, i)
1708 if (i->bucket == bucket) {
1709 BUG_ON(!i->in_progress);
1710 darray_remove_item(&ca->discard_buckets_in_flight, i);
1711 goto found;
1712 }
1713 BUG();
1714 found:
1715 mutex_unlock(&ca->discard_buckets_in_flight_lock);
1716 }
1717
1718 struct discard_buckets_state {
1719 u64 seen;
1720 u64 open;
1721 u64 need_journal_commit;
1722 u64 discarded;
1723 u64 need_journal_commit_this_dev;
1724 };
1725
1726 static int bch2_discard_one_bucket(struct btree_trans *trans,
1727 struct bch_dev *ca,
1728 struct btree_iter *need_discard_iter,
1729 struct bpos *discard_pos_done,
1730 struct discard_buckets_state *s)
1731 {
1732 struct bch_fs *c = trans->c;
1733 struct bpos pos = need_discard_iter->pos;
1734 struct btree_iter iter = { NULL };
1735 struct bkey_s_c k;
1736 struct bkey_i_alloc_v4 *a;
1737 struct printbuf buf = PRINTBUF;
1738 bool discard_locked = false;
1739 int ret = 0;
1740
1741 if (bch2_bucket_is_open_safe(c, pos.inode, pos.offset)) {
1742 s->open++;
1743 goto out;
1744 }
1745
1746 if (bch2_bucket_needs_journal_commit(&c->buckets_waiting_for_journal,
1747 c->journal.flushed_seq_ondisk,
1748 pos.inode, pos.offset)) {
1749 s->need_journal_commit++;
1750 s->need_journal_commit_this_dev++;
1751 goto out;
1752 }
1753
1754 k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_alloc,
1755 need_discard_iter->pos,
1756 BTREE_ITER_cached);
1757 ret = bkey_err(k);
1758 if (ret)
1759 goto out;
1760
1761 a = bch2_alloc_to_v4_mut(trans, k);
1762 ret = PTR_ERR_OR_ZERO(a);
1763 if (ret)
1764 goto out;
1765
1766 if (bch2_bucket_sectors_total(a->v)) {
1767 if (bch2_trans_inconsistent_on(c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_info,
1768 trans, "attempting to discard bucket with dirty data\n%s",
1769 (bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
1770 ret = -EIO;
1771 goto out;
1772 }
1773
1774 if (a->v.data_type != BCH_DATA_need_discard) {
1775 if (data_type_is_empty(a->v.data_type) &&
1776 BCH_ALLOC_V4_NEED_INC_GEN(&a->v)) {
1777 a->v.gen++;
1778 SET_BCH_ALLOC_V4_NEED_INC_GEN(&a->v, false);
1779 goto write;
1780 }
1781
1782 if (bch2_trans_inconsistent_on(c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_info,
1783 trans, "bucket incorrectly set in need_discard btree\n"
1784 "%s",
1785 (bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
1786 ret = -EIO;
1787 goto out;
1788 }
1789
1790 if (a->v.journal_seq > c->journal.flushed_seq_ondisk) {
1791 if (bch2_trans_inconsistent_on(c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_info,
1792 trans, "clearing need_discard but journal_seq %llu > flushed_seq %llu\n%s",
1793 a->v.journal_seq,
1794 c->journal.flushed_seq_ondisk,
1795 (bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
1796 ret = -EIO;
1797 goto out;
1798 }
1799
1800 if (discard_in_flight_add(ca, iter.pos.offset, true))
1801 goto out;
1802
1803 discard_locked = true;
1804
1805 if (!bkey_eq(*discard_pos_done, iter.pos) &&
1806 ca->mi.discard && !c->opts.nochanges) {
1807 /*
1808 * This works without any other locks because this is the only
1809 * thread that removes items from the need_discard tree
1810 */
1811 bch2_trans_unlock_long(trans);
1812 blkdev_issue_discard(ca->disk_sb.bdev,
1813 k.k->p.offset * ca->mi.bucket_size,
1814 ca->mi.bucket_size,
1815 GFP_KERNEL);
1816 *discard_pos_done = iter.pos;
1817
1818 ret = bch2_trans_relock_notrace(trans);
1819 if (ret)
1820 goto out;
1821 }
1822
1823 SET_BCH_ALLOC_V4_NEED_DISCARD(&a->v, false);
1824 write:
1825 alloc_data_type_set(&a->v, a->v.data_type);
1826
1827 ret = bch2_trans_update(trans, &iter, &a->k_i, 0) ?:
1828 bch2_trans_commit(trans, NULL, NULL,
1829 BCH_WATERMARK_btree|
1830 BCH_TRANS_COMMIT_no_enospc);
1831 if (ret)
1832 goto out;
1833
1834 count_event(c, bucket_discard);
1835 s->discarded++;
1836 out:
1837 if (discard_locked)
1838 discard_in_flight_remove(ca, iter.pos.offset);
1839 s->seen++;
1840 bch2_trans_iter_exit(trans, &iter);
1841 printbuf_exit(&buf);
1842 return ret;
1843 }
1844
1845 static void bch2_do_discards_work(struct work_struct *work)
1846 {
1847 struct bch_dev *ca = container_of(work, struct bch_dev, discard_work);
1848 struct bch_fs *c = ca->fs;
1849 struct discard_buckets_state s = {};
1850 struct bpos discard_pos_done = POS_MAX;
1851 int ret;
1852
1853 /*
1854 * We're doing the commit in bch2_discard_one_bucket instead of using
1855 * for_each_btree_key_commit() so that we can increment counters after
1856 * successful commit:
1857 */
1858 ret = bch2_trans_run(c,
1859 for_each_btree_key_upto(trans, iter,
1860 BTREE_ID_need_discard,
1861 POS(ca->dev_idx, 0),
1862 POS(ca->dev_idx, U64_MAX), 0, k,
1863 bch2_discard_one_bucket(trans, ca, &iter, &discard_pos_done, &s)));
1864
1865 trace_discard_buckets(c, s.seen, s.open, s.need_journal_commit, s.discarded,
1866 bch2_err_str(ret));
1867
1868 bch2_write_ref_put(c, BCH_WRITE_REF_discard);
1869 percpu_ref_put(&ca->io_ref);
1870 }
1871
1872 void bch2_dev_do_discards(struct bch_dev *ca)
1873 {
1874 struct bch_fs *c = ca->fs;
1875
1876 if (!bch2_dev_get_ioref(c, ca->dev_idx, WRITE))
1877 return;
1878
1879 if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_discard))
1880 goto put_ioref;
1881
1882 if (queue_work(c->write_ref_wq, &ca->discard_work))
1883 return;
1884
1885 bch2_write_ref_put(c, BCH_WRITE_REF_discard);
1886 put_ioref:
1887 percpu_ref_put(&ca->io_ref);
1888 }
1889
1890 void bch2_do_discards(struct bch_fs *c)
1891 {
1892 for_each_member_device(c, ca)
1893 bch2_dev_do_discards(ca);
1894 }
1895
1896 static int bch2_clear_bucket_needs_discard(struct btree_trans *trans, struct bpos bucket)
1897 {
1898 struct btree_iter iter;
1899 bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc, bucket, BTREE_ITER_intent);
1900 struct bkey_s_c k = bch2_btree_iter_peek_slot(&iter);
1901 int ret = bkey_err(k);
1902 if (ret)
1903 goto err;
1904
1905 struct bkey_i_alloc_v4 *a = bch2_alloc_to_v4_mut(trans, k);
1906 ret = PTR_ERR_OR_ZERO(a);
1907 if (ret)
1908 goto err;
1909
1910 BUG_ON(a->v.dirty_sectors);
1911 SET_BCH_ALLOC_V4_NEED_DISCARD(&a->v, false);
1912 alloc_data_type_set(&a->v, a->v.data_type);
1913
1914 ret = bch2_trans_update(trans, &iter, &a->k_i, 0);
1915 err:
1916 bch2_trans_iter_exit(trans, &iter);
1917 return ret;
1918 }
1919
1920 static void bch2_do_discards_fast_work(struct work_struct *work)
1921 {
1922 struct bch_dev *ca = container_of(work, struct bch_dev, discard_fast_work);
1923 struct bch_fs *c = ca->fs;
1924
1925 while (1) {
1926 bool got_bucket = false;
1927 u64 bucket;
1928
1929 mutex_lock(&ca->discard_buckets_in_flight_lock);
1930 darray_for_each(ca->discard_buckets_in_flight, i) {
1931 if (i->in_progress)
1932 continue;
1933
1934 got_bucket = true;
1935 bucket = i->bucket;
1936 i->in_progress = true;
1937 break;
1938 }
1939 mutex_unlock(&ca->discard_buckets_in_flight_lock);
1940
1941 if (!got_bucket)
1942 break;
1943
1944 if (ca->mi.discard && !c->opts.nochanges)
1945 blkdev_issue_discard(ca->disk_sb.bdev,
1946 bucket_to_sector(ca, bucket),
1947 ca->mi.bucket_size,
1948 GFP_KERNEL);
1949
1950 int ret = bch2_trans_do(c, NULL, NULL,
1951 BCH_WATERMARK_btree|
1952 BCH_TRANS_COMMIT_no_enospc,
1953 bch2_clear_bucket_needs_discard(trans, POS(ca->dev_idx, bucket)));
1954 bch_err_fn(c, ret);
1955
1956 discard_in_flight_remove(ca, bucket);
1957
1958 if (ret)
1959 break;
1960 }
1961
1962 bch2_write_ref_put(c, BCH_WRITE_REF_discard_fast);
1963 percpu_ref_put(&ca->io_ref);
1964 }
1965
1966 static void bch2_discard_one_bucket_fast(struct bch_dev *ca, u64 bucket)
1967 {
1968 struct bch_fs *c = ca->fs;
1969
1970 if (discard_in_flight_add(ca, bucket, false))
1971 return;
1972
1973 if (!bch2_dev_get_ioref(c, ca->dev_idx, WRITE))
1974 return;
1975
1976 if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_discard_fast))
1977 goto put_ioref;
1978
1979 if (queue_work(c->write_ref_wq, &ca->discard_fast_work))
1980 return;
1981
1982 bch2_write_ref_put(c, BCH_WRITE_REF_discard_fast);
1983 put_ioref:
1984 percpu_ref_put(&ca->io_ref);
1985 }
1986
1987 static int invalidate_one_bucket(struct btree_trans *trans,
1988 struct btree_iter *lru_iter,
1989 struct bkey_s_c lru_k,
1990 s64 *nr_to_invalidate)
1991 {
1992 struct bch_fs *c = trans->c;
1993 struct bkey_i_alloc_v4 *a = NULL;
1994 struct printbuf buf = PRINTBUF;
1995 struct bpos bucket = u64_to_bucket(lru_k.k->p.offset);
1996 unsigned cached_sectors;
1997 int ret = 0;
1998
1999 if (*nr_to_invalidate <= 0)
2000 return 1;
2001
2002 if (!bch2_dev_bucket_exists(c, bucket)) {
2003 prt_str(&buf, "lru entry points to invalid bucket");
2004 goto err;
2005 }
2006
2007 if (bch2_bucket_is_open_safe(c, bucket.inode, bucket.offset))
2008 return 0;
2009
2010 a = bch2_trans_start_alloc_update(trans, bucket, BTREE_TRIGGER_bucket_invalidate);
2011 ret = PTR_ERR_OR_ZERO(a);
2012 if (ret)
2013 goto out;
2014
2015 /* We expect harmless races here due to the btree write buffer: */
2016 if (lru_pos_time(lru_iter->pos) != alloc_lru_idx_read(a->v))
2017 goto out;
2018
2019 BUG_ON(a->v.data_type != BCH_DATA_cached);
2020 BUG_ON(a->v.dirty_sectors);
2021
2022 if (!a->v.cached_sectors)
2023 bch_err(c, "invalidating empty bucket, confused");
2024
2025 cached_sectors = a->v.cached_sectors;
2026
2027 SET_BCH_ALLOC_V4_NEED_INC_GEN(&a->v, false);
2028 a->v.gen++;
2029 a->v.data_type = 0;
2030 a->v.dirty_sectors = 0;
2031 a->v.stripe_sectors = 0;
2032 a->v.cached_sectors = 0;
2033 a->v.io_time[READ] = bch2_current_io_time(c, READ);
2034 a->v.io_time[WRITE] = bch2_current_io_time(c, WRITE);
2035
2036 ret = bch2_trans_commit(trans, NULL, NULL,
2037 BCH_WATERMARK_btree|
2038 BCH_TRANS_COMMIT_no_enospc);
2039 if (ret)
2040 goto out;
2041
2042 trace_and_count(c, bucket_invalidate, c, bucket.inode, bucket.offset, cached_sectors);
2043 --*nr_to_invalidate;
2044 out:
2045 printbuf_exit(&buf);
2046 return ret;
2047 err:
2048 prt_str(&buf, "\n lru key: ");
2049 bch2_bkey_val_to_text(&buf, c, lru_k);
2050
2051 prt_str(&buf, "\n lru entry: ");
2052 bch2_lru_pos_to_text(&buf, lru_iter->pos);
2053
2054 prt_str(&buf, "\n alloc key: ");
2055 if (!a)
2056 bch2_bpos_to_text(&buf, bucket);
2057 else
2058 bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&a->k_i));
2059
2060 bch_err(c, "%s", buf.buf);
2061 if (c->curr_recovery_pass > BCH_RECOVERY_PASS_check_lrus) {
2062 bch2_inconsistent_error(c);
2063 ret = -EINVAL;
2064 }
2065
2066 goto out;
2067 }
2068
2069 static struct bkey_s_c next_lru_key(struct btree_trans *trans, struct btree_iter *iter,
2070 struct bch_dev *ca, bool *wrapped)
2071 {
2072 struct bkey_s_c k;
2073 again:
2074 k = bch2_btree_iter_peek_upto(iter, lru_pos(ca->dev_idx, U64_MAX, LRU_TIME_MAX));
2075 if (!k.k && !*wrapped) {
2076 bch2_btree_iter_set_pos(iter, lru_pos(ca->dev_idx, 0, 0));
2077 *wrapped = true;
2078 goto again;
2079 }
2080
2081 return k;
2082 }
2083
2084 static void bch2_do_invalidates_work(struct work_struct *work)
2085 {
2086 struct bch_dev *ca = container_of(work, struct bch_dev, invalidate_work);
2087 struct bch_fs *c = ca->fs;
2088 struct btree_trans *trans = bch2_trans_get(c);
2089 int ret = 0;
2090
2091 ret = bch2_btree_write_buffer_tryflush(trans);
2092 if (ret)
2093 goto err;
2094
2095 s64 nr_to_invalidate =
2096 should_invalidate_buckets(ca, bch2_dev_usage_read(ca));
2097 struct btree_iter iter;
2098 bool wrapped = false;
2099
2100 bch2_trans_iter_init(trans, &iter, BTREE_ID_lru,
2101 lru_pos(ca->dev_idx, 0,
2102 ((bch2_current_io_time(c, READ) + U32_MAX) &
2103 LRU_TIME_MAX)), 0);
2104
2105 while (true) {
2106 bch2_trans_begin(trans);
2107
2108 struct bkey_s_c k = next_lru_key(trans, &iter, ca, &wrapped);
2109 ret = bkey_err(k);
2110 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
2111 continue;
2112 if (ret)
2113 break;
2114 if (!k.k)
2115 break;
2116
2117 ret = invalidate_one_bucket(trans, &iter, k, &nr_to_invalidate);
2118 if (ret)
2119 break;
2120
2121 bch2_btree_iter_advance(&iter);
2122 }
2123 bch2_trans_iter_exit(trans, &iter);
2124 err:
2125 bch2_trans_put(trans);
2126 bch2_write_ref_put(c, BCH_WRITE_REF_invalidate);
2127 percpu_ref_put(&ca->io_ref);
2128 }
2129
2130 void bch2_dev_do_invalidates(struct bch_dev *ca)
2131 {
2132 struct bch_fs *c = ca->fs;
2133
2134 if (!bch2_dev_get_ioref(c, ca->dev_idx, WRITE))
2135 return;
2136
2137 if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_invalidate))
2138 goto put_ioref;
2139
2140 if (queue_work(c->write_ref_wq, &ca->invalidate_work))
2141 return;
2142
2143 bch2_write_ref_put(c, BCH_WRITE_REF_invalidate);
2144 put_ioref:
2145 percpu_ref_put(&ca->io_ref);
2146 }
2147
2148 void bch2_do_invalidates(struct bch_fs *c)
2149 {
2150 for_each_member_device(c, ca)
2151 bch2_dev_do_invalidates(ca);
2152 }
2153
2154 int bch2_dev_freespace_init(struct bch_fs *c, struct bch_dev *ca,
2155 u64 bucket_start, u64 bucket_end)
2156 {
2157 struct btree_trans *trans = bch2_trans_get(c);
2158 struct btree_iter iter;
2159 struct bkey_s_c k;
2160 struct bkey hole;
2161 struct bpos end = POS(ca->dev_idx, bucket_end);
2162 struct bch_member *m;
2163 unsigned long last_updated = jiffies;
2164 int ret;
2165
2166 BUG_ON(bucket_start > bucket_end);
2167 BUG_ON(bucket_end > ca->mi.nbuckets);
2168
2169 bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc,
2170 POS(ca->dev_idx, max_t(u64, ca->mi.first_bucket, bucket_start)),
2171 BTREE_ITER_prefetch);
2172 /*
2173 * Scan the alloc btree for every bucket on @ca, and add buckets to the
2174 * freespace/need_discard/need_gc_gens btrees as needed:
2175 */
2176 while (1) {
2177 if (last_updated + HZ * 10 < jiffies) {
2178 bch_info(ca, "%s: currently at %llu/%llu",
2179 __func__, iter.pos.offset, ca->mi.nbuckets);
2180 last_updated = jiffies;
2181 }
2182
2183 bch2_trans_begin(trans);
2184
2185 if (bkey_ge(iter.pos, end)) {
2186 ret = 0;
2187 break;
2188 }
2189
2190 k = bch2_get_key_or_hole(&iter, end, &hole);
2191 ret = bkey_err(k);
2192 if (ret)
2193 goto bkey_err;
2194
2195 if (k.k->type) {
2196 /*
2197 * We process live keys in the alloc btree one at a
2198 * time:
2199 */
2200 struct bch_alloc_v4 a_convert;
2201 const struct bch_alloc_v4 *a = bch2_alloc_to_v4(k, &a_convert);
2202
2203 ret = bch2_bucket_do_index(trans, ca, k, a, true) ?:
2204 bch2_trans_commit(trans, NULL, NULL,
2205 BCH_TRANS_COMMIT_no_enospc);
2206 if (ret)
2207 goto bkey_err;
2208
2209 bch2_btree_iter_advance(&iter);
2210 } else {
2211 struct bkey_i *freespace;
2212
2213 freespace = bch2_trans_kmalloc(trans, sizeof(*freespace));
2214 ret = PTR_ERR_OR_ZERO(freespace);
2215 if (ret)
2216 goto bkey_err;
2217
2218 bkey_init(&freespace->k);
2219 freespace->k.type = KEY_TYPE_set;
2220 freespace->k.p = k.k->p;
2221 freespace->k.size = k.k->size;
2222
2223 ret = bch2_btree_insert_trans(trans, BTREE_ID_freespace, freespace, 0) ?:
2224 bch2_trans_commit(trans, NULL, NULL,
2225 BCH_TRANS_COMMIT_no_enospc);
2226 if (ret)
2227 goto bkey_err;
2228
2229 bch2_btree_iter_set_pos(&iter, k.k->p);
2230 }
2231 bkey_err:
2232 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
2233 continue;
2234 if (ret)
2235 break;
2236 }
2237
2238 bch2_trans_iter_exit(trans, &iter);
2239 bch2_trans_put(trans);
2240
2241 if (ret < 0) {
2242 bch_err_msg(ca, ret, "initializing free space");
2243 return ret;
2244 }
2245
2246 mutex_lock(&c->sb_lock);
2247 m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
2248 SET_BCH_MEMBER_FREESPACE_INITIALIZED(m, true);
2249 mutex_unlock(&c->sb_lock);
2250
2251 return 0;
2252 }
2253
2254 int bch2_fs_freespace_init(struct bch_fs *c)
2255 {
2256 int ret = 0;
2257 bool doing_init = false;
2258
2259 /*
2260 * We can crash during the device add path, so we need to check this on
2261 * every mount:
2262 */
2263
2264 for_each_member_device(c, ca) {
2265 if (ca->mi.freespace_initialized)
2266 continue;
2267
2268 if (!doing_init) {
2269 bch_info(c, "initializing freespace");
2270 doing_init = true;
2271 }
2272
2273 ret = bch2_dev_freespace_init(c, ca, 0, ca->mi.nbuckets);
2274 if (ret) {
2275 bch2_dev_put(ca);
2276 bch_err_fn(c, ret);
2277 return ret;
2278 }
2279 }
2280
2281 if (doing_init) {
2282 mutex_lock(&c->sb_lock);
2283 bch2_write_super(c);
2284 mutex_unlock(&c->sb_lock);
2285 bch_verbose(c, "done initializing freespace");
2286 }
2287
2288 return 0;
2289 }
2290
2291 /* Bucket IO clocks: */
2292
2293 int bch2_bucket_io_time_reset(struct btree_trans *trans, unsigned dev,
2294 size_t bucket_nr, int rw)
2295 {
2296 struct bch_fs *c = trans->c;
2297 struct btree_iter iter;
2298 struct bkey_i_alloc_v4 *a;
2299 u64 now;
2300 int ret = 0;
2301
2302 if (bch2_trans_relock(trans))
2303 bch2_trans_begin(trans);
2304
2305 a = bch2_trans_start_alloc_update_noupdate(trans, &iter, POS(dev, bucket_nr));
2306 ret = PTR_ERR_OR_ZERO(a);
2307 if (ret)
2308 return ret;
2309
2310 now = bch2_current_io_time(c, rw);
2311 if (a->v.io_time[rw] == now)
2312 goto out;
2313
2314 a->v.io_time[rw] = now;
2315
2316 ret = bch2_trans_update(trans, &iter, &a->k_i, 0) ?:
2317 bch2_trans_commit(trans, NULL, NULL, 0);
2318 out:
2319 bch2_trans_iter_exit(trans, &iter);
2320 return ret;
2321 }
2322
2323 /* Startup/shutdown (ro/rw): */
2324
2325 void bch2_recalc_capacity(struct bch_fs *c)
2326 {
2327 u64 capacity = 0, reserved_sectors = 0, gc_reserve;
2328 unsigned bucket_size_max = 0;
2329 unsigned long ra_pages = 0;
2330
2331 lockdep_assert_held(&c->state_lock);
2332
2333 for_each_online_member(c, ca) {
2334 struct backing_dev_info *bdi = ca->disk_sb.bdev->bd_disk->bdi;
2335
2336 ra_pages += bdi->ra_pages;
2337 }
2338
2339 bch2_set_ra_pages(c, ra_pages);
2340
2341 for_each_rw_member(c, ca) {
2342 u64 dev_reserve = 0;
2343
2344 /*
2345 * We need to reserve buckets (from the number
2346 * of currently available buckets) against
2347 * foreground writes so that mainly copygc can
2348 * make forward progress.
2349 *
2350 * We need enough to refill the various reserves
2351 * from scratch - copygc will use its entire
2352 * reserve all at once, then run against when
2353 * its reserve is refilled (from the formerly
2354 * available buckets).
2355 *
2356 * This reserve is just used when considering if
2357 * allocations for foreground writes must wait -
2358 * not -ENOSPC calculations.
2359 */
2360
2361 dev_reserve += ca->nr_btree_reserve * 2;
2362 dev_reserve += ca->mi.nbuckets >> 6; /* copygc reserve */
2363
2364 dev_reserve += 1; /* btree write point */
2365 dev_reserve += 1; /* copygc write point */
2366 dev_reserve += 1; /* rebalance write point */
2367
2368 dev_reserve *= ca->mi.bucket_size;
2369
2370 capacity += bucket_to_sector(ca, ca->mi.nbuckets -
2371 ca->mi.first_bucket);
2372
2373 reserved_sectors += dev_reserve * 2;
2374
2375 bucket_size_max = max_t(unsigned, bucket_size_max,
2376 ca->mi.bucket_size);
2377 }
2378
2379 gc_reserve = c->opts.gc_reserve_bytes
2380 ? c->opts.gc_reserve_bytes >> 9
2381 : div64_u64(capacity * c->opts.gc_reserve_percent, 100);
2382
2383 reserved_sectors = max(gc_reserve, reserved_sectors);
2384
2385 reserved_sectors = min(reserved_sectors, capacity);
2386
2387 c->reserved = reserved_sectors;
2388 c->capacity = capacity - reserved_sectors;
2389
2390 c->bucket_size_max = bucket_size_max;
2391
2392 /* Wake up case someone was waiting for buckets */
2393 closure_wake_up(&c->freelist_wait);
2394 }
2395
2396 u64 bch2_min_rw_member_capacity(struct bch_fs *c)
2397 {
2398 u64 ret = U64_MAX;
2399
2400 for_each_rw_member(c, ca)
2401 ret = min(ret, ca->mi.nbuckets * ca->mi.bucket_size);
2402 return ret;
2403 }
2404
2405 static bool bch2_dev_has_open_write_point(struct bch_fs *c, struct bch_dev *ca)
2406 {
2407 struct open_bucket *ob;
2408 bool ret = false;
2409
2410 for (ob = c->open_buckets;
2411 ob < c->open_buckets + ARRAY_SIZE(c->open_buckets);
2412 ob++) {
2413 spin_lock(&ob->lock);
2414 if (ob->valid && !ob->on_partial_list &&
2415 ob->dev == ca->dev_idx)
2416 ret = true;
2417 spin_unlock(&ob->lock);
2418 }
2419
2420 return ret;
2421 }
2422
2423 /* device goes ro: */
2424 void bch2_dev_allocator_remove(struct bch_fs *c, struct bch_dev *ca)
2425 {
2426 unsigned i;
2427
2428 /* First, remove device from allocation groups: */
2429
2430 for (i = 0; i < ARRAY_SIZE(c->rw_devs); i++)
2431 clear_bit(ca->dev_idx, c->rw_devs[i].d);
2432
2433 /*
2434 * Capacity is calculated based off of devices in allocation groups:
2435 */
2436 bch2_recalc_capacity(c);
2437
2438 bch2_open_buckets_stop(c, ca, false);
2439
2440 /*
2441 * Wake up threads that were blocked on allocation, so they can notice
2442 * the device can no longer be removed and the capacity has changed:
2443 */
2444 closure_wake_up(&c->freelist_wait);
2445
2446 /*
2447 * journal_res_get() can block waiting for free space in the journal -
2448 * it needs to notice there may not be devices to allocate from anymore:
2449 */
2450 wake_up(&c->journal.wait);
2451
2452 /* Now wait for any in flight writes: */
2453
2454 closure_wait_event(&c->open_buckets_wait,
2455 !bch2_dev_has_open_write_point(c, ca));
2456 }
2457
2458 /* device goes rw: */
2459 void bch2_dev_allocator_add(struct bch_fs *c, struct bch_dev *ca)
2460 {
2461 unsigned i;
2462
2463 for (i = 0; i < ARRAY_SIZE(c->rw_devs); i++)
2464 if (ca->mi.data_allowed & (1 << i))
2465 set_bit(ca->dev_idx, c->rw_devs[i].d);
2466 }
2467
2468 void bch2_dev_allocator_background_exit(struct bch_dev *ca)
2469 {
2470 darray_exit(&ca->discard_buckets_in_flight);
2471 }
2472
2473 void bch2_dev_allocator_background_init(struct bch_dev *ca)
2474 {
2475 mutex_init(&ca->discard_buckets_in_flight_lock);
2476 INIT_WORK(&ca->discard_work, bch2_do_discards_work);
2477 INIT_WORK(&ca->discard_fast_work, bch2_do_discards_fast_work);
2478 INIT_WORK(&ca->invalidate_work, bch2_do_invalidates_work);
2479 }
2480
2481 void bch2_fs_allocator_background_init(struct bch_fs *c)
2482 {
2483 spin_lock_init(&c->freelist_lock);
2484 }
2485
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.