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 "btree_cache.h" 7 #include "btree_io.h" 8 #include "btree_key_cache.h" 9 #include "btree_update.h" 10 #include "btree_update_interior.h" 11 #include "btree_gc.h" 12 #include "btree_write_buffer.h" 13 #include "buckets.h" 14 #include "buckets_waiting_for_journal.h" 15 #include "clock.h" 16 #include "debug.h" 17 #include "ec.h" 18 #include "error.h" 19 #include "lru.h" 20 #include "recovery.h" 21 #include "trace.h" 22 #include "varint.h" 23 24 #include <linux/kthread.h> 25 #include <linux/math64.h> 26 #include <linux/random.h> 27 #include <linux/rculist.h> 28 #include <linux/rcupdate.h> 29 #include <linux/sched/task.h> 30 #include <linux/sort.h> 31 32 static void bch2_discard_one_bucket_fast(struct bch_dev *, u64); 33 34 /* Persistent alloc info: */ 35 36 static const unsigned BCH_ALLOC_V1_FIELD_BYTES[] = { 37 #define x(name, bits) [BCH_ALLOC_FIELD_V1_##name] = bits / 8, 38 BCH_ALLOC_FIELDS_V1() 39 #undef x 40 }; 41 42 struct bkey_alloc_unpacked { 43 u64 journal_seq; 44 u8 gen; 45 u8 oldest_gen; 46 u8 data_type; 47 bool need_discard:1; 48 bool need_inc_gen:1; 49 #define x(_name, _bits) u##_bits _name; 50 BCH_ALLOC_FIELDS_V2() 51 #undef x 52 }; 53 54 static inline u64 alloc_field_v1_get(const struct bch_alloc *a, 55 const void **p, unsigned field) 56 { 57 unsigned bytes = BCH_ALLOC_V1_FIELD_BYTES[field]; 58 u64 v; 59 60 if (!(a->fields & (1 << field))) 61 return 0; 62 63 switch (bytes) { 64 case 1: 65 v = *((const u8 *) *p); 66 break; 67 case 2: 68 v = le16_to_cpup(*p); 69 break; 70 case 4: 71 v = le32_to_cpup(*p); 72 break; 73 case 8: 74 v = le64_to_cpup(*p); 75 break; 76 default: 77 BUG(); 78 } 79 80 *p += bytes; 81 return v; 82 } 83 84 static void bch2_alloc_unpack_v1(struct bkey_alloc_unpacked *out, 85 struct bkey_s_c k) 86 { 87 const struct bch_alloc *in = bkey_s_c_to_alloc(k).v; 88 const void *d = in->data; 89 unsigned idx = 0; 90 91 out->gen = in->gen; 92 93 #define x(_name, _bits) out->_name = alloc_field_v1_get(in, &d, idx++); 94 BCH_ALLOC_FIELDS_V1() 95 #undef x 96 } 97 98 static int bch2_alloc_unpack_v2(struct bkey_alloc_unpacked *out, 99 struct bkey_s_c k) 100 { 101 struct bkey_s_c_alloc_v2 a = bkey_s_c_to_alloc_v2(k); 102 const u8 *in = a.v->data; 103 const u8 *end = bkey_val_end(a); 104 unsigned fieldnr = 0; 105 int ret; 106 u64 v; 107 108 out->gen = a.v->gen; 109 out->oldest_gen = a.v->oldest_gen; 110 out->data_type = a.v->data_type; 111 112 #define x(_name, _bits) \ 113 if (fieldnr < a.v->nr_fields) { \ 114 ret = bch2_varint_decode_fast(in, end, &v); \ 115 if (ret < 0) \ 116 return ret; \ 117 in += ret; \ 118 } else { \ 119 v = 0; \ 120 } \ 121 out->_name = v; \ 122 if (v != out->_name) \ 123 return -1; \ 124 fieldnr++; 125 126 BCH_ALLOC_FIELDS_V2() 127 #undef x 128 return 0; 129 } 130 131 static int bch2_alloc_unpack_v3(struct bkey_alloc_unpacked *out, 132 struct bkey_s_c k) 133 { 134 struct bkey_s_c_alloc_v3 a = bkey_s_c_to_alloc_v3(k); 135 const u8 *in = a.v->data; 136 const u8 *end = bkey_val_end(a); 137 unsigned fieldnr = 0; 138 int ret; 139 u64 v; 140 141 out->gen = a.v->gen; 142 out->oldest_gen = a.v->oldest_gen; 143 out->data_type = a.v->data_type; 144 out->need_discard = BCH_ALLOC_V3_NEED_DISCARD(a.v); 145 out->need_inc_gen = BCH_ALLOC_V3_NEED_INC_GEN(a.v); 146 out->journal_seq = le64_to_cpu(a.v->journal_seq); 147 148 #define x(_name, _bits) \ 149 if (fieldnr < a.v->nr_fields) { \ 150 ret = bch2_varint_decode_fast(in, end, &v); \ 151 if (ret < 0) \ 152 return ret; \ 153 in += ret; \ 154 } else { \ 155 v = 0; \ 156 } \ 157 out->_name = v; \ 158 if (v != out->_name) \ 159 return -1; \ 160 fieldnr++; 161 162 BCH_ALLOC_FIELDS_V2() 163 #undef x 164 return 0; 165 } 166 167 static struct bkey_alloc_unpacked bch2_alloc_unpack(struct bkey_s_c k) 168 { 169 struct bkey_alloc_unpacked ret = { .gen = 0 }; 170 171 switch (k.k->type) { 172 case KEY_TYPE_alloc: 173 bch2_alloc_unpack_v1(&ret, k); 174 break; 175 case KEY_TYPE_alloc_v2: 176 bch2_alloc_unpack_v2(&ret, k); 177 break; 178 case KEY_TYPE_alloc_v3: 179 bch2_alloc_unpack_v3(&ret, k); 180 break; 181 } 182 183 return ret; 184 } 185 186 static unsigned bch_alloc_v1_val_u64s(const struct bch_alloc *a) 187 { 188 unsigned i, bytes = offsetof(struct bch_alloc, data); 189 190 for (i = 0; i < ARRAY_SIZE(BCH_ALLOC_V1_FIELD_BYTES); i++) 191 if (a->fields & (1 << i)) 192 bytes += BCH_ALLOC_V1_FIELD_BYTES[i]; 193 194 return DIV_ROUND_UP(bytes, sizeof(u64)); 195 } 196 197 int bch2_alloc_v1_invalid(struct bch_fs *c, struct bkey_s_c k, 198 enum bch_validate_flags flags, 199 struct printbuf *err) 200 { 201 struct bkey_s_c_alloc a = bkey_s_c_to_alloc(k); 202 int ret = 0; 203 204 /* allow for unknown fields */ 205 bkey_fsck_err_on(bkey_val_u64s(a.k) < bch_alloc_v1_val_u64s(a.v), c, err, 206 alloc_v1_val_size_bad, 207 "incorrect value size (%zu < %u)", 208 bkey_val_u64s(a.k), bch_alloc_v1_val_u64s(a.v)); 209 fsck_err: 210 return ret; 211 } 212 213 int bch2_alloc_v2_invalid(struct bch_fs *c, struct bkey_s_c k, 214 enum bch_validate_flags flags, 215 struct printbuf *err) 216 { 217 struct bkey_alloc_unpacked u; 218 int ret = 0; 219 220 bkey_fsck_err_on(bch2_alloc_unpack_v2(&u, k), c, err, 221 alloc_v2_unpack_error, 222 "unpack error"); 223 fsck_err: 224 return ret; 225 } 226 227 int bch2_alloc_v3_invalid(struct bch_fs *c, struct bkey_s_c k, 228 enum bch_validate_flags flags, 229 struct printbuf *err) 230 { 231 struct bkey_alloc_unpacked u; 232 int ret = 0; 233 234 bkey_fsck_err_on(bch2_alloc_unpack_v3(&u, k), c, err, 235 alloc_v2_unpack_error, 236 "unpack error"); 237 fsck_err: 238 return ret; 239 } 240 241 int bch2_alloc_v4_invalid(struct bch_fs *c, struct bkey_s_c k, 242 enum bch_validate_flags flags, struct printbuf *err) 243 { 244 struct bkey_s_c_alloc_v4 a = bkey_s_c_to_alloc_v4(k); 245 int ret = 0; 246 247 bkey_fsck_err_on(alloc_v4_u64s_noerror(a.v) > bkey_val_u64s(k.k), c, err, 248 alloc_v4_val_size_bad, 249 "bad val size (%u > %zu)", 250 alloc_v4_u64s_noerror(a.v), bkey_val_u64s(k.k)); 251 252 bkey_fsck_err_on(!BCH_ALLOC_V4_BACKPOINTERS_START(a.v) && 253 BCH_ALLOC_V4_NR_BACKPOINTERS(a.v), c, err, 254 alloc_v4_backpointers_start_bad, 255 "invalid backpointers_start"); 256 257 bkey_fsck_err_on(alloc_data_type(*a.v, a.v->data_type) != a.v->data_type, c, err, 258 alloc_key_data_type_bad, 259 "invalid data type (got %u should be %u)", 260 a.v->data_type, alloc_data_type(*a.v, a.v->data_type)); 261 262 for (unsigned i = 0; i < 2; i++) 263 bkey_fsck_err_on(a.v->io_time[i] > LRU_TIME_MAX, 264 c, err, 265 alloc_key_io_time_bad, 266 "invalid io_time[%s]: %llu, max %llu", 267 i == READ ? "read" : "write", 268 a.v->io_time[i], LRU_TIME_MAX); 269 270 switch (a.v->data_type) { 271 case BCH_DATA_free: 272 case BCH_DATA_need_gc_gens: 273 case BCH_DATA_need_discard: 274 bkey_fsck_err_on(bch2_bucket_sectors_total(*a.v) || a.v->stripe, 275 c, err, alloc_key_empty_but_have_data, 276 "empty data type free but have data"); 277 break; 278 case BCH_DATA_sb: 279 case BCH_DATA_journal: 280 case BCH_DATA_btree: 281 case BCH_DATA_user: 282 case BCH_DATA_parity: 283 bkey_fsck_err_on(!bch2_bucket_sectors_dirty(*a.v), 284 c, err, alloc_key_dirty_sectors_0, 285 "data_type %s but dirty_sectors==0", 286 bch2_data_type_str(a.v->data_type)); 287 break; 288 case BCH_DATA_cached: 289 bkey_fsck_err_on(!a.v->cached_sectors || 290 bch2_bucket_sectors_dirty(*a.v) || 291 a.v->stripe, 292 c, err, alloc_key_cached_inconsistency, 293 "data type inconsistency"); 294 295 bkey_fsck_err_on(!a.v->io_time[READ] && 296 c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_to_lru_refs, 297 c, err, alloc_key_cached_but_read_time_zero, 298 "cached bucket with read_time == 0"); 299 break; 300 case BCH_DATA_stripe: 301 break; 302 } 303 fsck_err: 304 return ret; 305 } 306 307 void bch2_alloc_v4_swab(struct bkey_s k) 308 { 309 struct bch_alloc_v4 *a = bkey_s_to_alloc_v4(k).v; 310 struct bch_backpointer *bp, *bps; 311 312 a->journal_seq = swab64(a->journal_seq); 313 a->flags = swab32(a->flags); 314 a->dirty_sectors = swab32(a->dirty_sectors); 315 a->cached_sectors = swab32(a->cached_sectors); 316 a->io_time[0] = swab64(a->io_time[0]); 317 a->io_time[1] = swab64(a->io_time[1]); 318 a->stripe = swab32(a->stripe); 319 a->nr_external_backpointers = swab32(a->nr_external_backpointers); 320 a->fragmentation_lru = swab64(a->fragmentation_lru); 321 322 bps = alloc_v4_backpointers(a); 323 for (bp = bps; bp < bps + BCH_ALLOC_V4_NR_BACKPOINTERS(a); bp++) { 324 bp->bucket_offset = swab40(bp->bucket_offset); 325 bp->bucket_len = swab32(bp->bucket_len); 326 bch2_bpos_swab(&bp->pos); 327 } 328 } 329 330 void bch2_alloc_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c k) 331 { 332 struct bch_alloc_v4 _a; 333 const struct bch_alloc_v4 *a = bch2_alloc_to_v4(k, &_a); 334 335 prt_newline(out); 336 printbuf_indent_add(out, 2); 337 338 prt_printf(out, "gen %u oldest_gen %u data_type ", a->gen, a->oldest_gen); 339 bch2_prt_data_type(out, a->data_type); 340 prt_newline(out); 341 prt_printf(out, "journal_seq %llu\n", a->journal_seq); 342 prt_printf(out, "need_discard %llu\n", BCH_ALLOC_V4_NEED_DISCARD(a)); 343 prt_printf(out, "need_inc_gen %llu\n", BCH_ALLOC_V4_NEED_INC_GEN(a)); 344 prt_printf(out, "dirty_sectors %u\n", a->dirty_sectors); 345 prt_printf(out, "cached_sectors %u\n", a->cached_sectors); 346 prt_printf(out, "stripe %u\n", a->stripe); 347 prt_printf(out, "stripe_redundancy %u\n", a->stripe_redundancy); 348 prt_printf(out, "io_time[READ] %llu\n", a->io_time[READ]); 349 prt_printf(out, "io_time[WRITE] %llu\n", a->io_time[WRITE]); 350 prt_printf(out, "fragmentation %llu\n", a->fragmentation_lru); 351 prt_printf(out, "bp_start %llu\n", BCH_ALLOC_V4_BACKPOINTERS_START(a)); 352 printbuf_indent_sub(out, 2); 353 } 354 355 void __bch2_alloc_to_v4(struct bkey_s_c k, struct bch_alloc_v4 *out) 356 { 357 if (k.k->type == KEY_TYPE_alloc_v4) { 358 void *src, *dst; 359 360 *out = *bkey_s_c_to_alloc_v4(k).v; 361 362 src = alloc_v4_backpointers(out); 363 SET_BCH_ALLOC_V4_BACKPOINTERS_START(out, BCH_ALLOC_V4_U64s); 364 dst = alloc_v4_backpointers(out); 365 366 if (src < dst) 367 memset(src, 0, dst - src); 368 369 SET_BCH_ALLOC_V4_NR_BACKPOINTERS(out, 0); 370 } else { 371 struct bkey_alloc_unpacked u = bch2_alloc_unpack(k); 372 373 *out = (struct bch_alloc_v4) { 374 .journal_seq = u.journal_seq, 375 .flags = u.need_discard, 376 .gen = u.gen, 377 .oldest_gen = u.oldest_gen, 378 .data_type = u.data_type, 379 .stripe_redundancy = u.stripe_redundancy, 380 .dirty_sectors = u.dirty_sectors, 381 .cached_sectors = u.cached_sectors, 382 .io_time[READ] = u.read_time, 383 .io_time[WRITE] = u.write_time, 384 .stripe = u.stripe, 385 }; 386 387 SET_BCH_ALLOC_V4_BACKPOINTERS_START(out, BCH_ALLOC_V4_U64s); 388 } 389 } 390 391 static noinline struct bkey_i_alloc_v4 * 392 __bch2_alloc_to_v4_mut(struct btree_trans *trans, struct bkey_s_c k) 393 { 394 struct bkey_i_alloc_v4 *ret; 395 396 ret = bch2_trans_kmalloc(trans, max(bkey_bytes(k.k), sizeof(struct bkey_i_alloc_v4))); 397 if (IS_ERR(ret)) 398 return ret; 399 400 if (k.k->type == KEY_TYPE_alloc_v4) { 401 void *src, *dst; 402 403 bkey_reassemble(&ret->k_i, k); 404 405 src = alloc_v4_backpointers(&ret->v); 406 SET_BCH_ALLOC_V4_BACKPOINTERS_START(&ret->v, BCH_ALLOC_V4_U64s); 407 dst = alloc_v4_backpointers(&ret->v); 408 409 if (src < dst) 410 memset(src, 0, dst - src); 411 412 SET_BCH_ALLOC_V4_NR_BACKPOINTERS(&ret->v, 0); 413 set_alloc_v4_u64s(ret); 414 } else { 415 bkey_alloc_v4_init(&ret->k_i); 416 ret->k.p = k.k->p; 417 bch2_alloc_to_v4(k, &ret->v); 418 } 419 return ret; 420 } 421 422 static inline struct bkey_i_alloc_v4 *bch2_alloc_to_v4_mut_inlined(struct btree_trans *trans, struct bkey_s_c k) 423 { 424 struct bkey_s_c_alloc_v4 a; 425 426 if (likely(k.k->type == KEY_TYPE_alloc_v4) && 427 ((a = bkey_s_c_to_alloc_v4(k), true) && 428 BCH_ALLOC_V4_NR_BACKPOINTERS(a.v) == 0)) 429 return bch2_bkey_make_mut_noupdate_typed(trans, k, alloc_v4); 430 431 return __bch2_alloc_to_v4_mut(trans, k); 432 } 433 434 struct bkey_i_alloc_v4 *bch2_alloc_to_v4_mut(struct btree_trans *trans, struct bkey_s_c k) 435 { 436 return bch2_alloc_to_v4_mut_inlined(trans, k); 437 } 438 439 struct bkey_i_alloc_v4 * 440 bch2_trans_start_alloc_update_noupdate(struct btree_trans *trans, struct btree_iter *iter, 441 struct bpos pos) 442 { 443 struct bkey_s_c k = bch2_bkey_get_iter(trans, iter, BTREE_ID_alloc, pos, 444 BTREE_ITER_with_updates| 445 BTREE_ITER_cached| 446 BTREE_ITER_intent); 447 int ret = bkey_err(k); 448 if (unlikely(ret)) 449 return ERR_PTR(ret); 450 451 struct bkey_i_alloc_v4 *a = bch2_alloc_to_v4_mut_inlined(trans, k); 452 ret = PTR_ERR_OR_ZERO(a); 453 if (unlikely(ret)) 454 goto err; 455 return a; 456 err: 457 bch2_trans_iter_exit(trans, iter); 458 return ERR_PTR(ret); 459 } 460 461 __flatten 462 struct bkey_i_alloc_v4 *bch2_trans_start_alloc_update(struct btree_trans *trans, struct bpos pos) 463 { 464 struct btree_iter iter; 465 struct bkey_i_alloc_v4 *a = bch2_trans_start_alloc_update_noupdate(trans, &iter, pos); 466 int ret = PTR_ERR_OR_ZERO(a); 467 if (ret) 468 return ERR_PTR(ret); 469 470 ret = bch2_trans_update(trans, &iter, &a->k_i, 0); 471 bch2_trans_iter_exit(trans, &iter); 472 return unlikely(ret) ? ERR_PTR(ret) : a; 473 } 474 475 static struct bpos alloc_gens_pos(struct bpos pos, unsigned *offset) 476 { 477 *offset = pos.offset & KEY_TYPE_BUCKET_GENS_MASK; 478 479 pos.offset >>= KEY_TYPE_BUCKET_GENS_BITS; 480 return pos; 481 } 482 483 static struct bpos bucket_gens_pos_to_alloc(struct bpos pos, unsigned offset) 484 { 485 pos.offset <<= KEY_TYPE_BUCKET_GENS_BITS; 486 pos.offset += offset; 487 return pos; 488 } 489 490 static unsigned alloc_gen(struct bkey_s_c k, unsigned offset) 491 { 492 return k.k->type == KEY_TYPE_bucket_gens 493 ? bkey_s_c_to_bucket_gens(k).v->gens[offset] 494 : 0; 495 } 496 497 int bch2_bucket_gens_invalid(struct bch_fs *c, struct bkey_s_c k, 498 enum bch_validate_flags flags, 499 struct printbuf *err) 500 { 501 int ret = 0; 502 503 bkey_fsck_err_on(bkey_val_bytes(k.k) != sizeof(struct bch_bucket_gens), c, err, 504 bucket_gens_val_size_bad, 505 "bad val size (%zu != %zu)", 506 bkey_val_bytes(k.k), sizeof(struct bch_bucket_gens)); 507 fsck_err: 508 return ret; 509 } 510 511 void bch2_bucket_gens_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c k) 512 { 513 struct bkey_s_c_bucket_gens g = bkey_s_c_to_bucket_gens(k); 514 unsigned i; 515 516 for (i = 0; i < ARRAY_SIZE(g.v->gens); i++) { 517 if (i) 518 prt_char(out, ' '); 519 prt_printf(out, "%u", g.v->gens[i]); 520 } 521 } 522 523 int bch2_bucket_gens_init(struct bch_fs *c) 524 { 525 struct btree_trans *trans = bch2_trans_get(c); 526 struct bkey_i_bucket_gens g; 527 bool have_bucket_gens_key = false; 528 int ret; 529 530 ret = for_each_btree_key(trans, iter, BTREE_ID_alloc, POS_MIN, 531 BTREE_ITER_prefetch, k, ({ 532 /* 533 * Not a fsck error because this is checked/repaired by 534 * bch2_check_alloc_key() which runs later: 535 */ 536 if (!bch2_dev_bucket_exists(c, k.k->p)) 537 continue; 538 539 struct bch_alloc_v4 a; 540 u8 gen = bch2_alloc_to_v4(k, &a)->gen; 541 unsigned offset; 542 struct bpos pos = alloc_gens_pos(iter.pos, &offset); 543 int ret2 = 0; 544 545 if (have_bucket_gens_key && bkey_cmp(iter.pos, pos)) { 546 ret2 = bch2_btree_insert_trans(trans, BTREE_ID_bucket_gens, &g.k_i, 0) ?: 547 bch2_trans_commit(trans, NULL, NULL, BCH_TRANS_COMMIT_no_enospc); 548 if (ret2) 549 goto iter_err; 550 have_bucket_gens_key = false; 551 } 552 553 if (!have_bucket_gens_key) { 554 bkey_bucket_gens_init(&g.k_i); 555 g.k.p = pos; 556 have_bucket_gens_key = true; 557 } 558 559 g.v.gens[offset] = gen; 560 iter_err: 561 ret2; 562 })); 563 564 if (have_bucket_gens_key && !ret) 565 ret = commit_do(trans, NULL, NULL, 566 BCH_TRANS_COMMIT_no_enospc, 567 bch2_btree_insert_trans(trans, BTREE_ID_bucket_gens, &g.k_i, 0)); 568 569 bch2_trans_put(trans); 570 571 bch_err_fn(c, ret); 572 return ret; 573 } 574 575 int bch2_alloc_read(struct bch_fs *c) 576 { 577 struct btree_trans *trans = bch2_trans_get(c); 578 struct bch_dev *ca = NULL; 579 int ret; 580 581 down_read(&c->gc_lock); 582 583 if (c->sb.version_upgrade_complete >= bcachefs_metadata_version_bucket_gens) { 584 ret = for_each_btree_key(trans, iter, BTREE_ID_bucket_gens, POS_MIN, 585 BTREE_ITER_prefetch, k, ({ 586 u64 start = bucket_gens_pos_to_alloc(k.k->p, 0).offset; 587 u64 end = bucket_gens_pos_to_alloc(bpos_nosnap_successor(k.k->p), 0).offset; 588 589 if (k.k->type != KEY_TYPE_bucket_gens) 590 continue; 591 592 ca = bch2_dev_iterate(c, ca, k.k->p.inode); 593 /* 594 * Not a fsck error because this is checked/repaired by 595 * bch2_check_alloc_key() which runs later: 596 */ 597 if (!ca) { 598 bch2_btree_iter_set_pos(&iter, POS(k.k->p.inode + 1, 0)); 599 continue; 600 } 601 602 const struct bch_bucket_gens *g = bkey_s_c_to_bucket_gens(k).v; 603 604 for (u64 b = max_t(u64, ca->mi.first_bucket, start); 605 b < min_t(u64, ca->mi.nbuckets, end); 606 b++) 607 *bucket_gen(ca, b) = g->gens[b & KEY_TYPE_BUCKET_GENS_MASK]; 608 0; 609 })); 610 } else { 611 ret = for_each_btree_key(trans, iter, BTREE_ID_alloc, POS_MIN, 612 BTREE_ITER_prefetch, k, ({ 613 ca = bch2_dev_iterate(c, ca, k.k->p.inode); 614 /* 615 * Not a fsck error because this is checked/repaired by 616 * bch2_check_alloc_key() which runs later: 617 */ 618 if (!ca) { 619 bch2_btree_iter_set_pos(&iter, POS(k.k->p.inode + 1, 0)); 620 continue; 621 } 622 623 struct bch_alloc_v4 a; 624 *bucket_gen(ca, k.k->p.offset) = bch2_alloc_to_v4(k, &a)->gen; 625 0; 626 })); 627 } 628 629 bch2_dev_put(ca); 630 bch2_trans_put(trans); 631 up_read(&c->gc_lock); 632 633 bch_err_fn(c, ret); 634 return ret; 635 } 636 637 /* Free space/discard btree: */ 638 639 static int bch2_bucket_do_index(struct btree_trans *trans, 640 struct bch_dev *ca, 641 struct bkey_s_c alloc_k, 642 const struct bch_alloc_v4 *a, 643 bool set) 644 { 645 struct bch_fs *c = trans->c; 646 struct btree_iter iter; 647 struct bkey_s_c old; 648 struct bkey_i *k; 649 enum btree_id btree; 650 enum bch_bkey_type old_type = !set ? KEY_TYPE_set : KEY_TYPE_deleted; 651 enum bch_bkey_type new_type = set ? KEY_TYPE_set : KEY_TYPE_deleted; 652 struct printbuf buf = PRINTBUF; 653 int ret; 654 655 if (a->data_type != BCH_DATA_free && 656 a->data_type != BCH_DATA_need_discard) 657 return 0; 658 659 k = bch2_trans_kmalloc_nomemzero(trans, sizeof(*k)); 660 if (IS_ERR(k)) 661 return PTR_ERR(k); 662 663 bkey_init(&k->k); 664 k->k.type = new_type; 665 666 switch (a->data_type) { 667 case BCH_DATA_free: 668 btree = BTREE_ID_freespace; 669 k->k.p = alloc_freespace_pos(alloc_k.k->p, *a); 670 bch2_key_resize(&k->k, 1); 671 break; 672 case BCH_DATA_need_discard: 673 btree = BTREE_ID_need_discard; 674 k->k.p = alloc_k.k->p; 675 break; 676 default: 677 return 0; 678 } 679 680 old = bch2_bkey_get_iter(trans, &iter, btree, 681 bkey_start_pos(&k->k), 682 BTREE_ITER_intent); 683 ret = bkey_err(old); 684 if (ret) 685 return ret; 686 687 if (ca->mi.freespace_initialized && 688 c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_info && 689 bch2_trans_inconsistent_on(old.k->type != old_type, trans, 690 "incorrect key when %s %s:%llu:%llu:0 (got %s should be %s)\n" 691 " for %s", 692 set ? "setting" : "clearing", 693 bch2_btree_id_str(btree), 694 iter.pos.inode, 695 iter.pos.offset, 696 bch2_bkey_types[old.k->type], 697 bch2_bkey_types[old_type], 698 (bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf))) { 699 ret = -EIO; 700 goto err; 701 } 702 703 ret = bch2_trans_update(trans, &iter, k, 0); 704 err: 705 bch2_trans_iter_exit(trans, &iter); 706 printbuf_exit(&buf); 707 return ret; 708 } 709 710 static noinline int bch2_bucket_gen_update(struct btree_trans *trans, 711 struct bpos bucket, u8 gen) 712 { 713 struct btree_iter iter; 714 unsigned offset; 715 struct bpos pos = alloc_gens_pos(bucket, &offset); 716 struct bkey_i_bucket_gens *g; 717 struct bkey_s_c k; 718 int ret; 719 720 g = bch2_trans_kmalloc(trans, sizeof(*g)); 721 ret = PTR_ERR_OR_ZERO(g); 722 if (ret) 723 return ret; 724 725 k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_bucket_gens, pos, 726 BTREE_ITER_intent| 727 BTREE_ITER_with_updates); 728 ret = bkey_err(k); 729 if (ret) 730 return ret; 731 732 if (k.k->type != KEY_TYPE_bucket_gens) { 733 bkey_bucket_gens_init(&g->k_i); 734 g->k.p = iter.pos; 735 } else { 736 bkey_reassemble(&g->k_i, k); 737 } 738 739 g->v.gens[offset] = gen; 740 741 ret = bch2_trans_update(trans, &iter, &g->k_i, 0); 742 bch2_trans_iter_exit(trans, &iter); 743 return ret; 744 } 745 746 int bch2_trigger_alloc(struct btree_trans *trans, 747 enum btree_id btree, unsigned level, 748 struct bkey_s_c old, struct bkey_s new, 749 enum btree_iter_update_trigger_flags flags) 750 { 751 struct bch_fs *c = trans->c; 752 struct printbuf buf = PRINTBUF; 753 int ret = 0; 754 755 struct bch_dev *ca = bch2_dev_bucket_tryget(c, new.k->p); 756 if (!ca) 757 return -EIO; 758 759 struct bch_alloc_v4 old_a_convert; 760 const struct bch_alloc_v4 *old_a = bch2_alloc_to_v4(old, &old_a_convert); 761 762 if (flags & BTREE_TRIGGER_transactional) { 763 struct bch_alloc_v4 *new_a = bkey_s_to_alloc_v4(new).v; 764 765 alloc_data_type_set(new_a, new_a->data_type); 766 767 if (bch2_bucket_sectors_total(*new_a) > bch2_bucket_sectors_total(*old_a)) { 768 new_a->io_time[READ] = bch2_current_io_time(c, READ); 769 new_a->io_time[WRITE]= bch2_current_io_time(c, WRITE); 770 SET_BCH_ALLOC_V4_NEED_INC_GEN(new_a, true); 771 SET_BCH_ALLOC_V4_NEED_DISCARD(new_a, true); 772 } 773 774 if (data_type_is_empty(new_a->data_type) && 775 BCH_ALLOC_V4_NEED_INC_GEN(new_a) && 776 !bch2_bucket_is_open_safe(c, new.k->p.inode, new.k->p.offset)) { 777 new_a->gen++; 778 SET_BCH_ALLOC_V4_NEED_INC_GEN(new_a, false); 779 alloc_data_type_set(new_a, new_a->data_type); 780 } 781 782 if (old_a->data_type != new_a->data_type || 783 (new_a->data_type == BCH_DATA_free && 784 alloc_freespace_genbits(*old_a) != alloc_freespace_genbits(*new_a))) { 785 ret = bch2_bucket_do_index(trans, ca, old, old_a, false) ?: 786 bch2_bucket_do_index(trans, ca, new.s_c, new_a, true); 787 if (ret) 788 goto err; 789 } 790 791 if (new_a->data_type == BCH_DATA_cached && 792 !new_a->io_time[READ]) 793 new_a->io_time[READ] = bch2_current_io_time(c, READ); 794 795 u64 old_lru = alloc_lru_idx_read(*old_a); 796 u64 new_lru = alloc_lru_idx_read(*new_a); 797 if (old_lru != new_lru) { 798 ret = bch2_lru_change(trans, new.k->p.inode, 799 bucket_to_u64(new.k->p), 800 old_lru, new_lru); 801 if (ret) 802 goto err; 803 } 804 805 new_a->fragmentation_lru = alloc_lru_idx_fragmentation(*new_a, ca); 806 if (old_a->fragmentation_lru != new_a->fragmentation_lru) { 807 ret = bch2_lru_change(trans, 808 BCH_LRU_FRAGMENTATION_START, 809 bucket_to_u64(new.k->p), 810 old_a->fragmentation_lru, new_a->fragmentation_lru); 811 if (ret) 812 goto err; 813 } 814 815 if (old_a->gen != new_a->gen) { 816 ret = bch2_bucket_gen_update(trans, new.k->p, new_a->gen); 817 if (ret) 818 goto err; 819 } 820 821 /* 822 * need to know if we're getting called from the invalidate path or 823 * not: 824 */ 825 826 if ((flags & BTREE_TRIGGER_bucket_invalidate) && 827 old_a->cached_sectors) { 828 ret = bch2_update_cached_sectors_list(trans, new.k->p.inode, 829 -((s64) old_a->cached_sectors)); 830 if (ret) 831 goto err; 832 } 833 } 834 835 if ((flags & BTREE_TRIGGER_atomic) && (flags & BTREE_TRIGGER_insert)) { 836 struct bch_alloc_v4 *new_a = bkey_s_to_alloc_v4(new).v; 837 u64 journal_seq = trans->journal_res.seq; 838 u64 bucket_journal_seq = new_a->journal_seq; 839 840 if ((flags & BTREE_TRIGGER_insert) && 841 data_type_is_empty(old_a->data_type) != 842 data_type_is_empty(new_a->data_type) && 843 new.k->type == KEY_TYPE_alloc_v4) { 844 struct bch_alloc_v4 *v = bkey_s_to_alloc_v4(new).v; 845 846 /* 847 * If the btree updates referring to a bucket weren't flushed 848 * before the bucket became empty again, then the we don't have 849 * to wait on a journal flush before we can reuse the bucket: 850 */ 851 v->journal_seq = bucket_journal_seq = 852 data_type_is_empty(new_a->data_type) && 853 (journal_seq == v->journal_seq || 854 bch2_journal_noflush_seq(&c->journal, v->journal_seq)) 855 ? 0 : journal_seq; 856 } 857 858 if (!data_type_is_empty(old_a->data_type) && 859 data_type_is_empty(new_a->data_type) && 860 bucket_journal_seq) { 861 ret = bch2_set_bucket_needs_journal_commit(&c->buckets_waiting_for_journal, 862 c->journal.flushed_seq_ondisk, 863 new.k->p.inode, new.k->p.offset, 864 bucket_journal_seq); 865 if (ret) { 866 bch2_fs_fatal_error(c, 867 "setting bucket_needs_journal_commit: %s", bch2_err_str(ret)); 868 goto err; 869 } 870 } 871 872 percpu_down_read(&c->mark_lock); 873 if (new_a->gen != old_a->gen) { 874 u8 *gen = bucket_gen(ca, new.k->p.offset); 875 if (unlikely(!gen)) { 876 percpu_up_read(&c->mark_lock); 877 goto invalid_bucket; 878 } 879 *gen = new_a->gen; 880 } 881 882 bch2_dev_usage_update(c, ca, old_a, new_a, journal_seq, false); 883 percpu_up_read(&c->mark_lock); 884 885 #define eval_state(_a, expr) ({ const struct bch_alloc_v4 *a = _a; expr; }) 886 #define statechange(expr) !eval_state(old_a, expr) && eval_state(new_a, expr) 887 #define bucket_flushed(a) (!a->journal_seq || a->journal_seq <= c->journal.flushed_seq_ondisk) 888 889 if (statechange(a->data_type == BCH_DATA_free) && 890 bucket_flushed(new_a)) 891 closure_wake_up(&c->freelist_wait); 892 893 if (statechange(a->data_type == BCH_DATA_need_discard) && 894 !bch2_bucket_is_open_safe(c, new.k->p.inode, new.k->p.offset) && 895 bucket_flushed(new_a)) 896 bch2_discard_one_bucket_fast(ca, new.k->p.offset); 897 898 if (statechange(a->data_type == BCH_DATA_cached) && 899 !bch2_bucket_is_open(c, new.k->p.inode, new.k->p.offset) && 900 should_invalidate_buckets(ca, bch2_dev_usage_read(ca))) 901 bch2_dev_do_invalidates(ca); 902 903 if (statechange(a->data_type == BCH_DATA_need_gc_gens)) 904 bch2_gc_gens_async(c); 905 } 906 907 if ((flags & BTREE_TRIGGER_gc) && 908 (flags & BTREE_TRIGGER_bucket_invalidate)) { 909 struct bch_alloc_v4 new_a_convert; 910 const struct bch_alloc_v4 *new_a = bch2_alloc_to_v4(new.s_c, &new_a_convert); 911 912 percpu_down_read(&c->mark_lock); 913 struct bucket *g = gc_bucket(ca, new.k->p.offset); 914 if (unlikely(!g)) { 915 percpu_up_read(&c->mark_lock); 916 goto invalid_bucket; 917 } 918 g->gen_valid = 1; 919 920 bucket_lock(g); 921 922 g->gen_valid = 1; 923 g->gen = new_a->gen; 924 g->data_type = new_a->data_type; 925 g->stripe = new_a->stripe; 926 g->stripe_redundancy = new_a->stripe_redundancy; 927 g->dirty_sectors = new_a->dirty_sectors; 928 g->cached_sectors = new_a->cached_sectors; 929 930 bucket_unlock(g); 931 percpu_up_read(&c->mark_lock); 932 } 933 err: 934 printbuf_exit(&buf); 935 bch2_dev_put(ca); 936 return ret; 937 invalid_bucket: 938 bch2_fs_inconsistent(c, "reference to invalid bucket\n %s", 939 (bch2_bkey_val_to_text(&buf, c, new.s_c), buf.buf)); 940 ret = -EIO; 941 goto err; 942 } 943 944 /* 945 * This synthesizes deleted extents for holes, similar to BTREE_ITER_slots for 946 * extents style btrees, but works on non-extents btrees: 947 */ 948 static struct bkey_s_c bch2_get_key_or_hole(struct btree_iter *iter, struct bpos end, struct bkey *hole) 949 { 950 struct bkey_s_c k = bch2_btree_iter_peek_slot(iter); 951 952 if (bkey_err(k)) 953 return k; 954 955 if (k.k->type) { 956 return k; 957 } else { 958 struct btree_iter iter2; 959 struct bpos next; 960 961 bch2_trans_copy_iter(&iter2, iter); 962 963 struct btree_path *path = btree_iter_path(iter->trans, iter); 964 if (!bpos_eq(path->l[0].b->key.k.p, SPOS_MAX)) 965 end = bkey_min(end, bpos_nosnap_successor(path->l[0].b->key.k.p)); 966 967 end = bkey_min(end, POS(iter->pos.inode, iter->pos.offset + U32_MAX - 1)); 968 969 /* 970 * btree node min/max is a closed interval, upto takes a half 971 * open interval: 972 */ 973 k = bch2_btree_iter_peek_upto(&iter2, end); 974 next = iter2.pos; 975 bch2_trans_iter_exit(iter->trans, &iter2); 976 977 BUG_ON(next.offset >= iter->pos.offset + U32_MAX); 978 979 if (bkey_err(k)) 980 return k; 981 982 bkey_init(hole); 983 hole->p = iter->pos; 984 985 bch2_key_resize(hole, next.offset - iter->pos.offset); 986 return (struct bkey_s_c) { hole, NULL }; 987 } 988 } 989 990 static bool next_bucket(struct bch_fs *c, struct bch_dev **ca, struct bpos *bucket) 991 { 992 if (*ca) { 993 if (bucket->offset < (*ca)->mi.first_bucket) 994 bucket->offset = (*ca)->mi.first_bucket; 995 996 if (bucket->offset < (*ca)->mi.nbuckets) 997 return true; 998 999 bch2_dev_put(*ca); 1000 *ca = NULL; 1001 bucket->inode++; 1002 bucket->offset = 0; 1003 } 1004 1005 rcu_read_lock(); 1006 *ca = __bch2_next_dev_idx(c, bucket->inode, NULL); 1007 if (*ca) { 1008 *bucket = POS((*ca)->dev_idx, (*ca)->mi.first_bucket); 1009 bch2_dev_get(*ca); 1010 } 1011 rcu_read_unlock(); 1012 1013 return *ca != NULL; 1014 } 1015 1016 static struct bkey_s_c bch2_get_key_or_real_bucket_hole(struct btree_iter *iter, 1017 struct bch_dev **ca, struct bkey *hole) 1018 { 1019 struct bch_fs *c = iter->trans->c; 1020 struct bkey_s_c k; 1021 again: 1022 k = bch2_get_key_or_hole(iter, POS_MAX, hole); 1023 if (bkey_err(k)) 1024 return k; 1025 1026 *ca = bch2_dev_iterate_noerror(c, *ca, k.k->p.inode); 1027 1028 if (!k.k->type) { 1029 struct bpos hole_start = bkey_start_pos(k.k); 1030 1031 if (!*ca || !bucket_valid(*ca, hole_start.offset)) { 1032 if (!next_bucket(c, ca, &hole_start)) 1033 return bkey_s_c_null; 1034 1035 bch2_btree_iter_set_pos(iter, hole_start); 1036 goto again; 1037 } 1038 1039 if (k.k->p.offset > (*ca)->mi.nbuckets) 1040 bch2_key_resize(hole, (*ca)->mi.nbuckets - hole_start.offset); 1041 } 1042 1043 return k; 1044 } 1045 1046 static noinline_for_stack 1047 int bch2_check_alloc_key(struct btree_trans *trans, 1048 struct bkey_s_c alloc_k, 1049 struct btree_iter *alloc_iter, 1050 struct btree_iter *discard_iter, 1051 struct btree_iter *freespace_iter, 1052 struct btree_iter *bucket_gens_iter) 1053 { 1054 struct bch_fs *c = trans->c; 1055 struct bch_alloc_v4 a_convert; 1056 const struct bch_alloc_v4 *a; 1057 unsigned discard_key_type, freespace_key_type; 1058 unsigned gens_offset; 1059 struct bkey_s_c k; 1060 struct printbuf buf = PRINTBUF; 1061 int ret = 0; 1062 1063 struct bch_dev *ca = bch2_dev_bucket_tryget_noerror(c, alloc_k.k->p); 1064 if (fsck_err_on(!ca, 1065 c, alloc_key_to_missing_dev_bucket, 1066 "alloc key for invalid device:bucket %llu:%llu", 1067 alloc_k.k->p.inode, alloc_k.k->p.offset)) 1068 ret = bch2_btree_delete_at(trans, alloc_iter, 0); 1069 if (!ca) 1070 return ret; 1071 1072 if (!ca->mi.freespace_initialized) 1073 goto out; 1074 1075 a = bch2_alloc_to_v4(alloc_k, &a_convert); 1076 1077 discard_key_type = a->data_type == BCH_DATA_need_discard ? KEY_TYPE_set : 0; 1078 bch2_btree_iter_set_pos(discard_iter, alloc_k.k->p); 1079 k = bch2_btree_iter_peek_slot(discard_iter); 1080 ret = bkey_err(k); 1081 if (ret) 1082 goto err; 1083 1084 if (fsck_err_on(k.k->type != discard_key_type, 1085 c, need_discard_key_wrong, 1086 "incorrect key in need_discard btree (got %s should be %s)\n" 1087 " %s", 1088 bch2_bkey_types[k.k->type], 1089 bch2_bkey_types[discard_key_type], 1090 (bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf))) { 1091 struct bkey_i *update = 1092 bch2_trans_kmalloc(trans, sizeof(*update)); 1093 1094 ret = PTR_ERR_OR_ZERO(update); 1095 if (ret) 1096 goto err; 1097 1098 bkey_init(&update->k); 1099 update->k.type = discard_key_type; 1100 update->k.p = discard_iter->pos; 1101 1102 ret = bch2_trans_update(trans, discard_iter, update, 0); 1103 if (ret) 1104 goto err; 1105 } 1106 1107 freespace_key_type = a->data_type == BCH_DATA_free ? KEY_TYPE_set : 0; 1108 bch2_btree_iter_set_pos(freespace_iter, alloc_freespace_pos(alloc_k.k->p, *a)); 1109 k = bch2_btree_iter_peek_slot(freespace_iter); 1110 ret = bkey_err(k); 1111 if (ret) 1112 goto err; 1113 1114 if (fsck_err_on(k.k->type != freespace_key_type, 1115 c, freespace_key_wrong, 1116 "incorrect key in freespace btree (got %s should be %s)\n" 1117 " %s", 1118 bch2_bkey_types[k.k->type], 1119 bch2_bkey_types[freespace_key_type], 1120 (printbuf_reset(&buf), 1121 bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf))) { 1122 struct bkey_i *update = 1123 bch2_trans_kmalloc(trans, sizeof(*update)); 1124 1125 ret = PTR_ERR_OR_ZERO(update); 1126 if (ret) 1127 goto err; 1128 1129 bkey_init(&update->k); 1130 update->k.type = freespace_key_type; 1131 update->k.p = freespace_iter->pos; 1132 bch2_key_resize(&update->k, 1); 1133 1134 ret = bch2_trans_update(trans, freespace_iter, update, 0); 1135 if (ret) 1136 goto err; 1137 } 1138 1139 bch2_btree_iter_set_pos(bucket_gens_iter, alloc_gens_pos(alloc_k.k->p, &gens_offset)); 1140 k = bch2_btree_iter_peek_slot(bucket_gens_iter); 1141 ret = bkey_err(k); 1142 if (ret) 1143 goto err; 1144 1145 if (fsck_err_on(a->gen != alloc_gen(k, gens_offset), 1146 c, bucket_gens_key_wrong, 1147 "incorrect gen in bucket_gens btree (got %u should be %u)\n" 1148 " %s", 1149 alloc_gen(k, gens_offset), a->gen, 1150 (printbuf_reset(&buf), 1151 bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf))) { 1152 struct bkey_i_bucket_gens *g = 1153 bch2_trans_kmalloc(trans, sizeof(*g)); 1154 1155 ret = PTR_ERR_OR_ZERO(g); 1156 if (ret) 1157 goto err; 1158 1159 if (k.k->type == KEY_TYPE_bucket_gens) { 1160 bkey_reassemble(&g->k_i, k); 1161 } else { 1162 bkey_bucket_gens_init(&g->k_i); 1163 g->k.p = alloc_gens_pos(alloc_k.k->p, &gens_offset); 1164 } 1165 1166 g->v.gens[gens_offset] = a->gen; 1167 1168 ret = bch2_trans_update(trans, bucket_gens_iter, &g->k_i, 0); 1169 if (ret) 1170 goto err; 1171 } 1172 out: 1173 err: 1174 fsck_err: 1175 bch2_dev_put(ca); 1176 printbuf_exit(&buf); 1177 return ret; 1178 } 1179 1180 static noinline_for_stack 1181 int bch2_check_alloc_hole_freespace(struct btree_trans *trans, 1182 struct bch_dev *ca, 1183 struct bpos start, 1184 struct bpos *end, 1185 struct btree_iter *freespace_iter) 1186 { 1187 struct bch_fs *c = trans->c; 1188 struct bkey_s_c k; 1189 struct printbuf buf = PRINTBUF; 1190 int ret; 1191 1192 if (!ca->mi.freespace_initialized) 1193 return 0; 1194 1195 bch2_btree_iter_set_pos(freespace_iter, start); 1196 1197 k = bch2_btree_iter_peek_slot(freespace_iter); 1198 ret = bkey_err(k); 1199 if (ret) 1200 goto err; 1201 1202 *end = bkey_min(k.k->p, *end); 1203 1204 if (fsck_err_on(k.k->type != KEY_TYPE_set, 1205 c, freespace_hole_missing, 1206 "hole in alloc btree missing in freespace btree\n" 1207 " device %llu buckets %llu-%llu", 1208 freespace_iter->pos.inode, 1209 freespace_iter->pos.offset, 1210 end->offset)) { 1211 struct bkey_i *update = 1212 bch2_trans_kmalloc(trans, sizeof(*update)); 1213 1214 ret = PTR_ERR_OR_ZERO(update); 1215 if (ret) 1216 goto err; 1217 1218 bkey_init(&update->k); 1219 update->k.type = KEY_TYPE_set; 1220 update->k.p = freespace_iter->pos; 1221 bch2_key_resize(&update->k, 1222 min_t(u64, U32_MAX, end->offset - 1223 freespace_iter->pos.offset)); 1224 1225 ret = bch2_trans_update(trans, freespace_iter, update, 0); 1226 if (ret) 1227 goto err; 1228 } 1229 err: 1230 fsck_err: 1231 printbuf_exit(&buf); 1232 return ret; 1233 } 1234 1235 static noinline_for_stack 1236 int bch2_check_alloc_hole_bucket_gens(struct btree_trans *trans, 1237 struct bpos start, 1238 struct bpos *end, 1239 struct btree_iter *bucket_gens_iter) 1240 { 1241 struct bch_fs *c = trans->c; 1242 struct bkey_s_c k; 1243 struct printbuf buf = PRINTBUF; 1244 unsigned i, gens_offset, gens_end_offset; 1245 int ret; 1246 1247 bch2_btree_iter_set_pos(bucket_gens_iter, alloc_gens_pos(start, &gens_offset)); 1248 1249 k = bch2_btree_iter_peek_slot(bucket_gens_iter); 1250 ret = bkey_err(k); 1251 if (ret) 1252 goto err; 1253 1254 if (bkey_cmp(alloc_gens_pos(start, &gens_offset), 1255 alloc_gens_pos(*end, &gens_end_offset))) 1256 gens_end_offset = KEY_TYPE_BUCKET_GENS_NR; 1257 1258 if (k.k->type == KEY_TYPE_bucket_gens) { 1259 struct bkey_i_bucket_gens g; 1260 bool need_update = false; 1261 1262 bkey_reassemble(&g.k_i, k); 1263 1264 for (i = gens_offset; i < gens_end_offset; i++) { 1265 if (fsck_err_on(g.v.gens[i], c, 1266 bucket_gens_hole_wrong, 1267 "hole in alloc btree at %llu:%llu with nonzero gen in bucket_gens btree (%u)", 1268 bucket_gens_pos_to_alloc(k.k->p, i).inode, 1269 bucket_gens_pos_to_alloc(k.k->p, i).offset, 1270 g.v.gens[i])) { 1271 g.v.gens[i] = 0; 1272 need_update = true; 1273 } 1274 } 1275 1276 if (need_update) { 1277 struct bkey_i *u = bch2_trans_kmalloc(trans, sizeof(g)); 1278 1279 ret = PTR_ERR_OR_ZERO(u); 1280 if (ret) 1281 goto err; 1282 1283 memcpy(u, &g, sizeof(g)); 1284 1285 ret = bch2_trans_update(trans, bucket_gens_iter, u, 0); 1286 if (ret) 1287 goto err; 1288 } 1289 } 1290 1291 *end = bkey_min(*end, bucket_gens_pos_to_alloc(bpos_nosnap_successor(k.k->p), 0)); 1292 err: 1293 fsck_err: 1294 printbuf_exit(&buf); 1295 return ret; 1296 } 1297 1298 static noinline_for_stack int bch2_check_discard_freespace_key(struct btree_trans *trans, 1299 struct btree_iter *iter) 1300 { 1301 struct bch_fs *c = trans->c; 1302 struct btree_iter alloc_iter; 1303 struct bkey_s_c alloc_k; 1304 struct bch_alloc_v4 a_convert; 1305 const struct bch_alloc_v4 *a; 1306 u64 genbits; 1307 struct bpos pos; 1308 enum bch_data_type state = iter->btree_id == BTREE_ID_need_discard 1309 ? BCH_DATA_need_discard 1310 : BCH_DATA_free; 1311 struct printbuf buf = PRINTBUF; 1312 int ret; 1313 1314 pos = iter->pos; 1315 pos.offset &= ~(~0ULL << 56); 1316 genbits = iter->pos.offset & (~0ULL << 56); 1317 1318 alloc_k = bch2_bkey_get_iter(trans, &alloc_iter, BTREE_ID_alloc, pos, 0); 1319 ret = bkey_err(alloc_k); 1320 if (ret) 1321 return ret; 1322 1323 if (fsck_err_on(!bch2_dev_bucket_exists(c, pos), c, 1324 need_discard_freespace_key_to_invalid_dev_bucket, 1325 "entry in %s btree for nonexistant dev:bucket %llu:%llu", 1326 bch2_btree_id_str(iter->btree_id), pos.inode, pos.offset)) 1327 goto delete; 1328 1329 a = bch2_alloc_to_v4(alloc_k, &a_convert); 1330 1331 if (fsck_err_on(a->data_type != state || 1332 (state == BCH_DATA_free && 1333 genbits != alloc_freespace_genbits(*a)), c, 1334 need_discard_freespace_key_bad, 1335 "%s\n incorrectly set at %s:%llu:%llu:0 (free %u, genbits %llu should be %llu)", 1336 (bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf), 1337 bch2_btree_id_str(iter->btree_id), 1338 iter->pos.inode, 1339 iter->pos.offset, 1340 a->data_type == state, 1341 genbits >> 56, alloc_freespace_genbits(*a) >> 56)) 1342 goto delete; 1343 out: 1344 fsck_err: 1345 bch2_set_btree_iter_dontneed(&alloc_iter); 1346 bch2_trans_iter_exit(trans, &alloc_iter); 1347 printbuf_exit(&buf); 1348 return ret; 1349 delete: 1350 ret = bch2_btree_delete_extent_at(trans, iter, 1351 iter->btree_id == BTREE_ID_freespace ? 1 : 0, 0) ?: 1352 bch2_trans_commit(trans, NULL, NULL, 1353 BCH_TRANS_COMMIT_no_enospc); 1354 goto out; 1355 } 1356 1357 /* 1358 * We've already checked that generation numbers in the bucket_gens btree are 1359 * valid for buckets that exist; this just checks for keys for nonexistent 1360 * buckets. 1361 */ 1362 static noinline_for_stack 1363 int bch2_check_bucket_gens_key(struct btree_trans *trans, 1364 struct btree_iter *iter, 1365 struct bkey_s_c k) 1366 { 1367 struct bch_fs *c = trans->c; 1368 struct bkey_i_bucket_gens g; 1369 u64 start = bucket_gens_pos_to_alloc(k.k->p, 0).offset; 1370 u64 end = bucket_gens_pos_to_alloc(bpos_nosnap_successor(k.k->p), 0).offset; 1371 u64 b; 1372 bool need_update = false; 1373 struct printbuf buf = PRINTBUF; 1374 int ret = 0; 1375 1376 BUG_ON(k.k->type != KEY_TYPE_bucket_gens); 1377 bkey_reassemble(&g.k_i, k); 1378 1379 struct bch_dev *ca = bch2_dev_tryget_noerror(c, k.k->p.inode); 1380 if (!ca) { 1381 if (fsck_err(c, bucket_gens_to_invalid_dev, 1382 "bucket_gens key for invalid device:\n %s", 1383 (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) 1384 ret = bch2_btree_delete_at(trans, iter, 0); 1385 goto out; 1386 } 1387 1388 if (fsck_err_on(end <= ca->mi.first_bucket || 1389 start >= ca->mi.nbuckets, c, 1390 bucket_gens_to_invalid_buckets, 1391 "bucket_gens key for invalid buckets:\n %s", 1392 (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) { 1393 ret = bch2_btree_delete_at(trans, iter, 0); 1394 goto out; 1395 } 1396 1397 for (b = start; b < ca->mi.first_bucket; b++) 1398 if (fsck_err_on(g.v.gens[b & KEY_TYPE_BUCKET_GENS_MASK], c, 1399 bucket_gens_nonzero_for_invalid_buckets, 1400 "bucket_gens key has nonzero gen for invalid bucket")) { 1401 g.v.gens[b & KEY_TYPE_BUCKET_GENS_MASK] = 0; 1402 need_update = true; 1403 } 1404 1405 for (b = ca->mi.nbuckets; b < end; b++) 1406 if (fsck_err_on(g.v.gens[b & KEY_TYPE_BUCKET_GENS_MASK], c, 1407 bucket_gens_nonzero_for_invalid_buckets, 1408 "bucket_gens key has nonzero gen for invalid bucket")) { 1409 g.v.gens[b & KEY_TYPE_BUCKET_GENS_MASK] = 0; 1410 need_update = true; 1411 } 1412 1413 if (need_update) { 1414 struct bkey_i *u = bch2_trans_kmalloc(trans, sizeof(g)); 1415 1416 ret = PTR_ERR_OR_ZERO(u); 1417 if (ret) 1418 goto out; 1419 1420 memcpy(u, &g, sizeof(g)); 1421 ret = bch2_trans_update(trans, iter, u, 0); 1422 } 1423 out: 1424 fsck_err: 1425 bch2_dev_put(ca); 1426 printbuf_exit(&buf); 1427 return ret; 1428 } 1429 1430 int bch2_check_alloc_info(struct bch_fs *c) 1431 { 1432 struct btree_trans *trans = bch2_trans_get(c); 1433 struct btree_iter iter, discard_iter, freespace_iter, bucket_gens_iter; 1434 struct bch_dev *ca = NULL; 1435 struct bkey hole; 1436 struct bkey_s_c k; 1437 int ret = 0; 1438 1439 bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc, POS_MIN, 1440 BTREE_ITER_prefetch); 1441 bch2_trans_iter_init(trans, &discard_iter, BTREE_ID_need_discard, POS_MIN, 1442 BTREE_ITER_prefetch); 1443 bch2_trans_iter_init(trans, &freespace_iter, BTREE_ID_freespace, POS_MIN, 1444 BTREE_ITER_prefetch); 1445 bch2_trans_iter_init(trans, &bucket_gens_iter, BTREE_ID_bucket_gens, POS_MIN, 1446 BTREE_ITER_prefetch); 1447 1448 while (1) { 1449 struct bpos next; 1450 1451 bch2_trans_begin(trans); 1452 1453 k = bch2_get_key_or_real_bucket_hole(&iter, &ca, &hole); 1454 ret = bkey_err(k); 1455 if (ret) 1456 goto bkey_err; 1457 1458 if (!k.k) 1459 break; 1460 1461 if (k.k->type) { 1462 next = bpos_nosnap_successor(k.k->p); 1463 1464 ret = bch2_check_alloc_key(trans, 1465 k, &iter, 1466 &discard_iter, 1467 &freespace_iter, 1468 &bucket_gens_iter); 1469 if (ret) 1470 goto bkey_err; 1471 } else { 1472 next = k.k->p; 1473 1474 ret = bch2_check_alloc_hole_freespace(trans, ca, 1475 bkey_start_pos(k.k), 1476 &next, 1477 &freespace_iter) ?: 1478 bch2_check_alloc_hole_bucket_gens(trans, 1479 bkey_start_pos(k.k), 1480 &next, 1481 &bucket_gens_iter); 1482 if (ret) 1483 goto bkey_err; 1484 } 1485 1486 ret = bch2_trans_commit(trans, NULL, NULL, 1487 BCH_TRANS_COMMIT_no_enospc); 1488 if (ret) 1489 goto bkey_err; 1490 1491 bch2_btree_iter_set_pos(&iter, next); 1492 bkey_err: 1493 if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) 1494 continue; 1495 if (ret) 1496 break; 1497 } 1498 bch2_trans_iter_exit(trans, &bucket_gens_iter); 1499 bch2_trans_iter_exit(trans, &freespace_iter); 1500 bch2_trans_iter_exit(trans, &discard_iter); 1501 bch2_trans_iter_exit(trans, &iter); 1502 bch2_dev_put(ca); 1503 ca = NULL; 1504 1505 if (ret < 0) 1506 goto err; 1507 1508 ret = for_each_btree_key(trans, iter, 1509 BTREE_ID_need_discard, POS_MIN, 1510 BTREE_ITER_prefetch, k, 1511 bch2_check_discard_freespace_key(trans, &iter)); 1512 if (ret) 1513 goto err; 1514 1515 bch2_trans_iter_init(trans, &iter, BTREE_ID_freespace, POS_MIN, 1516 BTREE_ITER_prefetch); 1517 while (1) { 1518 bch2_trans_begin(trans); 1519 k = bch2_btree_iter_peek(&iter); 1520 if (!k.k) 1521 break; 1522 1523 ret = bkey_err(k) ?: 1524 bch2_check_discard_freespace_key(trans, &iter); 1525 if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) { 1526 ret = 0; 1527 continue; 1528 } 1529 if (ret) { 1530 struct printbuf buf = PRINTBUF; 1531 bch2_bkey_val_to_text(&buf, c, k); 1532 1533 bch_err(c, "while checking %s", buf.buf); 1534 printbuf_exit(&buf); 1535 break; 1536 } 1537 1538 bch2_btree_iter_set_pos(&iter, bpos_nosnap_successor(iter.pos)); 1539 } 1540 bch2_trans_iter_exit(trans, &iter); 1541 if (ret) 1542 goto err; 1543 1544 ret = for_each_btree_key_commit(trans, iter, 1545 BTREE_ID_bucket_gens, POS_MIN, 1546 BTREE_ITER_prefetch, k, 1547 NULL, NULL, BCH_TRANS_COMMIT_no_enospc, 1548 bch2_check_bucket_gens_key(trans, &iter, k)); 1549 err: 1550 bch2_trans_put(trans); 1551 bch_err_fn(c, ret); 1552 return ret; 1553 } 1554 1555 static int bch2_check_alloc_to_lru_ref(struct btree_trans *trans, 1556 struct btree_iter *alloc_iter) 1557 { 1558 struct bch_fs *c = trans->c; 1559 struct btree_iter lru_iter; 1560 struct bch_alloc_v4 a_convert; 1561 const struct bch_alloc_v4 *a; 1562 struct bkey_s_c alloc_k, lru_k; 1563 struct printbuf buf = PRINTBUF; 1564 int ret; 1565 1566 alloc_k = bch2_btree_iter_peek(alloc_iter); 1567 if (!alloc_k.k) 1568 return 0; 1569 1570 ret = bkey_err(alloc_k); 1571 if (ret) 1572 return ret; 1573 1574 a = bch2_alloc_to_v4(alloc_k, &a_convert); 1575 1576 if (a->data_type != BCH_DATA_cached) 1577 return 0; 1578 1579 if (fsck_err_on(!a->io_time[READ], c, 1580 alloc_key_cached_but_read_time_zero, 1581 "cached bucket with read_time 0\n" 1582 " %s", 1583 (printbuf_reset(&buf), 1584 bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf))) { 1585 struct bkey_i_alloc_v4 *a_mut = 1586 bch2_alloc_to_v4_mut(trans, alloc_k); 1587 ret = PTR_ERR_OR_ZERO(a_mut); 1588 if (ret) 1589 goto err; 1590 1591 a_mut->v.io_time[READ] = bch2_current_io_time(c, READ); 1592 ret = bch2_trans_update(trans, alloc_iter, 1593 &a_mut->k_i, BTREE_TRIGGER_norun); 1594 if (ret) 1595 goto err; 1596 1597 a = &a_mut->v; 1598 } 1599 1600 lru_k = bch2_bkey_get_iter(trans, &lru_iter, BTREE_ID_lru, 1601 lru_pos(alloc_k.k->p.inode, 1602 bucket_to_u64(alloc_k.k->p), 1603 a->io_time[READ]), 0); 1604 ret = bkey_err(lru_k); 1605 if (ret) 1606 return ret; 1607 1608 if (fsck_err_on(lru_k.k->type != KEY_TYPE_set, c, 1609 alloc_key_to_missing_lru_entry, 1610 "missing lru entry\n" 1611 " %s", 1612 (printbuf_reset(&buf), 1613 bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf))) { 1614 ret = bch2_lru_set(trans, 1615 alloc_k.k->p.inode, 1616 bucket_to_u64(alloc_k.k->p), 1617 a->io_time[READ]); 1618 if (ret) 1619 goto err; 1620 } 1621 err: 1622 fsck_err: 1623 bch2_trans_iter_exit(trans, &lru_iter); 1624 printbuf_exit(&buf); 1625 return ret; 1626 } 1627 1628 int bch2_check_alloc_to_lru_refs(struct bch_fs *c) 1629 { 1630 int ret = bch2_trans_run(c, 1631 for_each_btree_key_commit(trans, iter, BTREE_ID_alloc, 1632 POS_MIN, BTREE_ITER_prefetch, k, 1633 NULL, NULL, BCH_TRANS_COMMIT_no_enospc, 1634 bch2_check_alloc_to_lru_ref(trans, &iter))); 1635 bch_err_fn(c, ret); 1636 return ret; 1637 } 1638 1639 static int discard_in_flight_add(struct bch_dev *ca, u64 bucket, bool in_progress) 1640 { 1641 int ret; 1642 1643 mutex_lock(&ca->discard_buckets_in_flight_lock); 1644 darray_for_each(ca->discard_buckets_in_flight, i) 1645 if (i->bucket == bucket) { 1646 ret = -BCH_ERR_EEXIST_discard_in_flight_add; 1647 goto out; 1648 } 1649 1650 ret = darray_push(&ca->discard_buckets_in_flight, ((struct discard_in_flight) { 1651 .in_progress = in_progress, 1652 .bucket = bucket, 1653 })); 1654 out: 1655 mutex_unlock(&ca->discard_buckets_in_flight_lock); 1656 return ret; 1657 } 1658 1659 static void discard_in_flight_remove(struct bch_dev *ca, u64 bucket) 1660 { 1661 mutex_lock(&ca->discard_buckets_in_flight_lock); 1662 darray_for_each(ca->discard_buckets_in_flight, i) 1663 if (i->bucket == bucket) { 1664 BUG_ON(!i->in_progress); 1665 darray_remove_item(&ca->discard_buckets_in_flight, i); 1666 goto found; 1667 } 1668 BUG(); 1669 found: 1670 mutex_unlock(&ca->discard_buckets_in_flight_lock); 1671 } 1672 1673 struct discard_buckets_state { 1674 u64 seen; 1675 u64 open; 1676 u64 need_journal_commit; 1677 u64 discarded; 1678 u64 need_journal_commit_this_dev; 1679 }; 1680 1681 static int bch2_discard_one_bucket(struct btree_trans *trans, 1682 struct bch_dev *ca, 1683 struct btree_iter *need_discard_iter, 1684 struct bpos *discard_pos_done, 1685 struct discard_buckets_state *s) 1686 { 1687 struct bch_fs *c = trans->c; 1688 struct bpos pos = need_discard_iter->pos; 1689 struct btree_iter iter = { NULL }; 1690 struct bkey_s_c k; 1691 struct bkey_i_alloc_v4 *a; 1692 struct printbuf buf = PRINTBUF; 1693 bool discard_locked = false; 1694 int ret = 0; 1695 1696 if (bch2_bucket_is_open_safe(c, pos.inode, pos.offset)) { 1697 s->open++; 1698 goto out; 1699 } 1700 1701 if (bch2_bucket_needs_journal_commit(&c->buckets_waiting_for_journal, 1702 c->journal.flushed_seq_ondisk, 1703 pos.inode, pos.offset)) { 1704 s->need_journal_commit++; 1705 s->need_journal_commit_this_dev++; 1706 goto out; 1707 } 1708 1709 k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_alloc, 1710 need_discard_iter->pos, 1711 BTREE_ITER_cached); 1712 ret = bkey_err(k); 1713 if (ret) 1714 goto out; 1715 1716 a = bch2_alloc_to_v4_mut(trans, k); 1717 ret = PTR_ERR_OR_ZERO(a); 1718 if (ret) 1719 goto out; 1720 1721 if (bch2_bucket_sectors_total(a->v)) { 1722 if (bch2_trans_inconsistent_on(c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_info, 1723 trans, "attempting to discard bucket with dirty data\n%s", 1724 (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) 1725 ret = -EIO; 1726 goto out; 1727 } 1728 1729 if (a->v.data_type != BCH_DATA_need_discard) { 1730 if (data_type_is_empty(a->v.data_type) && 1731 BCH_ALLOC_V4_NEED_INC_GEN(&a->v)) { 1732 a->v.gen++; 1733 SET_BCH_ALLOC_V4_NEED_INC_GEN(&a->v, false); 1734 goto write; 1735 } 1736 1737 if (bch2_trans_inconsistent_on(c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_info, 1738 trans, "bucket incorrectly set in need_discard btree\n" 1739 "%s", 1740 (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) 1741 ret = -EIO; 1742 goto out; 1743 } 1744 1745 if (a->v.journal_seq > c->journal.flushed_seq_ondisk) { 1746 if (bch2_trans_inconsistent_on(c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_info, 1747 trans, "clearing need_discard but journal_seq %llu > flushed_seq %llu\n%s", 1748 a->v.journal_seq, 1749 c->journal.flushed_seq_ondisk, 1750 (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) 1751 ret = -EIO; 1752 goto out; 1753 } 1754 1755 if (discard_in_flight_add(ca, iter.pos.offset, true)) 1756 goto out; 1757 1758 discard_locked = true; 1759 1760 if (!bkey_eq(*discard_pos_done, iter.pos) && 1761 ca->mi.discard && !c->opts.nochanges) { 1762 /* 1763 * This works without any other locks because this is the only 1764 * thread that removes items from the need_discard tree 1765 */ 1766 bch2_trans_unlock_long(trans); 1767 blkdev_issue_discard(ca->disk_sb.bdev, 1768 k.k->p.offset * ca->mi.bucket_size, 1769 ca->mi.bucket_size, 1770 GFP_KERNEL); 1771 *discard_pos_done = iter.pos; 1772 1773 ret = bch2_trans_relock_notrace(trans); 1774 if (ret) 1775 goto out; 1776 } 1777 1778 SET_BCH_ALLOC_V4_NEED_DISCARD(&a->v, false); 1779 write: 1780 alloc_data_type_set(&a->v, a->v.data_type); 1781 1782 ret = bch2_trans_update(trans, &iter, &a->k_i, 0) ?: 1783 bch2_trans_commit(trans, NULL, NULL, 1784 BCH_WATERMARK_btree| 1785 BCH_TRANS_COMMIT_no_enospc); 1786 if (ret) 1787 goto out; 1788 1789 count_event(c, bucket_discard); 1790 s->discarded++; 1791 out: 1792 if (discard_locked) 1793 discard_in_flight_remove(ca, iter.pos.offset); 1794 s->seen++; 1795 bch2_trans_iter_exit(trans, &iter); 1796 printbuf_exit(&buf); 1797 return ret; 1798 } 1799 1800 static void bch2_do_discards_work(struct work_struct *work) 1801 { 1802 struct bch_dev *ca = container_of(work, struct bch_dev, discard_work); 1803 struct bch_fs *c = ca->fs; 1804 struct discard_buckets_state s = {}; 1805 struct bpos discard_pos_done = POS_MAX; 1806 int ret; 1807 1808 /* 1809 * We're doing the commit in bch2_discard_one_bucket instead of using 1810 * for_each_btree_key_commit() so that we can increment counters after 1811 * successful commit: 1812 */ 1813 ret = bch2_trans_run(c, 1814 for_each_btree_key_upto(trans, iter, 1815 BTREE_ID_need_discard, 1816 POS(ca->dev_idx, 0), 1817 POS(ca->dev_idx, U64_MAX), 0, k, 1818 bch2_discard_one_bucket(trans, ca, &iter, &discard_pos_done, &s))); 1819 1820 trace_discard_buckets(c, s.seen, s.open, s.need_journal_commit, s.discarded, 1821 bch2_err_str(ret)); 1822 1823 bch2_write_ref_put(c, BCH_WRITE_REF_discard); 1824 percpu_ref_put(&ca->io_ref); 1825 } 1826 1827 void bch2_dev_do_discards(struct bch_dev *ca) 1828 { 1829 struct bch_fs *c = ca->fs; 1830 1831 if (!bch2_dev_get_ioref(c, ca->dev_idx, WRITE)) 1832 return; 1833 1834 if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_discard)) 1835 goto put_ioref; 1836 1837 if (queue_work(c->write_ref_wq, &ca->discard_work)) 1838 return; 1839 1840 bch2_write_ref_put(c, BCH_WRITE_REF_discard); 1841 put_ioref: 1842 percpu_ref_put(&ca->io_ref); 1843 } 1844 1845 void bch2_do_discards(struct bch_fs *c) 1846 { 1847 for_each_member_device(c, ca) 1848 bch2_dev_do_discards(ca); 1849 } 1850 1851 static int bch2_clear_bucket_needs_discard(struct btree_trans *trans, struct bpos bucket) 1852 { 1853 struct btree_iter iter; 1854 bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc, bucket, BTREE_ITER_intent); 1855 struct bkey_s_c k = bch2_btree_iter_peek_slot(&iter); 1856 int ret = bkey_err(k); 1857 if (ret) 1858 goto err; 1859 1860 struct bkey_i_alloc_v4 *a = bch2_alloc_to_v4_mut(trans, k); 1861 ret = PTR_ERR_OR_ZERO(a); 1862 if (ret) 1863 goto err; 1864 1865 BUG_ON(a->v.dirty_sectors); 1866 SET_BCH_ALLOC_V4_NEED_DISCARD(&a->v, false); 1867 alloc_data_type_set(&a->v, a->v.data_type); 1868 1869 ret = bch2_trans_update(trans, &iter, &a->k_i, 0); 1870 err: 1871 bch2_trans_iter_exit(trans, &iter); 1872 return ret; 1873 } 1874 1875 static void bch2_do_discards_fast_work(struct work_struct *work) 1876 { 1877 struct bch_dev *ca = container_of(work, struct bch_dev, discard_fast_work); 1878 struct bch_fs *c = ca->fs; 1879 1880 while (1) { 1881 bool got_bucket = false; 1882 u64 bucket; 1883 1884 mutex_lock(&ca->discard_buckets_in_flight_lock); 1885 darray_for_each(ca->discard_buckets_in_flight, i) { 1886 if (i->in_progress) 1887 continue; 1888 1889 got_bucket = true; 1890 bucket = i->bucket; 1891 i->in_progress = true; 1892 break; 1893 } 1894 mutex_unlock(&ca->discard_buckets_in_flight_lock); 1895 1896 if (!got_bucket) 1897 break; 1898 1899 if (ca->mi.discard && !c->opts.nochanges) 1900 blkdev_issue_discard(ca->disk_sb.bdev, 1901 bucket_to_sector(ca, bucket), 1902 ca->mi.bucket_size, 1903 GFP_KERNEL); 1904 1905 int ret = bch2_trans_do(c, NULL, NULL, 1906 BCH_WATERMARK_btree| 1907 BCH_TRANS_COMMIT_no_enospc, 1908 bch2_clear_bucket_needs_discard(trans, POS(ca->dev_idx, bucket))); 1909 bch_err_fn(c, ret); 1910 1911 discard_in_flight_remove(ca, bucket); 1912 1913 if (ret) 1914 break; 1915 } 1916 1917 bch2_write_ref_put(c, BCH_WRITE_REF_discard_fast); 1918 percpu_ref_put(&ca->io_ref); 1919 } 1920 1921 static void bch2_discard_one_bucket_fast(struct bch_dev *ca, u64 bucket) 1922 { 1923 struct bch_fs *c = ca->fs; 1924 1925 if (discard_in_flight_add(ca, bucket, false)) 1926 return; 1927 1928 if (!bch2_dev_get_ioref(c, ca->dev_idx, WRITE)) 1929 return; 1930 1931 if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_discard_fast)) 1932 goto put_ioref; 1933 1934 if (queue_work(c->write_ref_wq, &ca->discard_fast_work)) 1935 return; 1936 1937 bch2_write_ref_put(c, BCH_WRITE_REF_discard_fast); 1938 put_ioref: 1939 percpu_ref_put(&ca->io_ref); 1940 } 1941 1942 static int invalidate_one_bucket(struct btree_trans *trans, 1943 struct btree_iter *lru_iter, 1944 struct bkey_s_c lru_k, 1945 s64 *nr_to_invalidate) 1946 { 1947 struct bch_fs *c = trans->c; 1948 struct bkey_i_alloc_v4 *a = NULL; 1949 struct printbuf buf = PRINTBUF; 1950 struct bpos bucket = u64_to_bucket(lru_k.k->p.offset); 1951 unsigned cached_sectors; 1952 int ret = 0; 1953 1954 if (*nr_to_invalidate <= 0) 1955 return 1; 1956 1957 if (!bch2_dev_bucket_exists(c, bucket)) { 1958 prt_str(&buf, "lru entry points to invalid bucket"); 1959 goto err; 1960 } 1961 1962 if (bch2_bucket_is_open_safe(c, bucket.inode, bucket.offset)) 1963 return 0; 1964 1965 a = bch2_trans_start_alloc_update(trans, bucket); 1966 ret = PTR_ERR_OR_ZERO(a); 1967 if (ret) 1968 goto out; 1969 1970 /* We expect harmless races here due to the btree write buffer: */ 1971 if (lru_pos_time(lru_iter->pos) != alloc_lru_idx_read(a->v)) 1972 goto out; 1973 1974 BUG_ON(a->v.data_type != BCH_DATA_cached); 1975 BUG_ON(a->v.dirty_sectors); 1976 1977 if (!a->v.cached_sectors) 1978 bch_err(c, "invalidating empty bucket, confused"); 1979 1980 cached_sectors = a->v.cached_sectors; 1981 1982 SET_BCH_ALLOC_V4_NEED_INC_GEN(&a->v, false); 1983 a->v.gen++; 1984 a->v.data_type = 0; 1985 a->v.dirty_sectors = 0; 1986 a->v.cached_sectors = 0; 1987 a->v.io_time[READ] = bch2_current_io_time(c, READ); 1988 a->v.io_time[WRITE] = bch2_current_io_time(c, WRITE); 1989 1990 ret = bch2_trans_commit(trans, NULL, NULL, 1991 BCH_WATERMARK_btree| 1992 BCH_TRANS_COMMIT_no_enospc); 1993 if (ret) 1994 goto out; 1995 1996 trace_and_count(c, bucket_invalidate, c, bucket.inode, bucket.offset, cached_sectors); 1997 --*nr_to_invalidate; 1998 out: 1999 printbuf_exit(&buf); 2000 return ret; 2001 err: 2002 prt_str(&buf, "\n lru key: "); 2003 bch2_bkey_val_to_text(&buf, c, lru_k); 2004 2005 prt_str(&buf, "\n lru entry: "); 2006 bch2_lru_pos_to_text(&buf, lru_iter->pos); 2007 2008 prt_str(&buf, "\n alloc key: "); 2009 if (!a) 2010 bch2_bpos_to_text(&buf, bucket); 2011 else 2012 bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&a->k_i)); 2013 2014 bch_err(c, "%s", buf.buf); 2015 if (c->curr_recovery_pass > BCH_RECOVERY_PASS_check_lrus) { 2016 bch2_inconsistent_error(c); 2017 ret = -EINVAL; 2018 } 2019 2020 goto out; 2021 } 2022 2023 static struct bkey_s_c next_lru_key(struct btree_trans *trans, struct btree_iter *iter, 2024 struct bch_dev *ca, bool *wrapped) 2025 { 2026 struct bkey_s_c k; 2027 again: 2028 k = bch2_btree_iter_peek_upto(iter, lru_pos(ca->dev_idx, U64_MAX, LRU_TIME_MAX)); 2029 if (!k.k && !*wrapped) { 2030 bch2_btree_iter_set_pos(iter, lru_pos(ca->dev_idx, 0, 0)); 2031 *wrapped = true; 2032 goto again; 2033 } 2034 2035 return k; 2036 } 2037 2038 static void bch2_do_invalidates_work(struct work_struct *work) 2039 { 2040 struct bch_dev *ca = container_of(work, struct bch_dev, invalidate_work); 2041 struct bch_fs *c = ca->fs; 2042 struct btree_trans *trans = bch2_trans_get(c); 2043 int ret = 0; 2044 2045 ret = bch2_btree_write_buffer_tryflush(trans); 2046 if (ret) 2047 goto err; 2048 2049 s64 nr_to_invalidate = 2050 should_invalidate_buckets(ca, bch2_dev_usage_read(ca)); 2051 struct btree_iter iter; 2052 bool wrapped = false; 2053 2054 bch2_trans_iter_init(trans, &iter, BTREE_ID_lru, 2055 lru_pos(ca->dev_idx, 0, 2056 ((bch2_current_io_time(c, READ) + U32_MAX) & 2057 LRU_TIME_MAX)), 0); 2058 2059 while (true) { 2060 bch2_trans_begin(trans); 2061 2062 struct bkey_s_c k = next_lru_key(trans, &iter, ca, &wrapped); 2063 ret = bkey_err(k); 2064 if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) 2065 continue; 2066 if (ret) 2067 break; 2068 if (!k.k) 2069 break; 2070 2071 ret = invalidate_one_bucket(trans, &iter, k, &nr_to_invalidate); 2072 if (ret) 2073 break; 2074 2075 bch2_btree_iter_advance(&iter); 2076 } 2077 bch2_trans_iter_exit(trans, &iter); 2078 err: 2079 bch2_trans_put(trans); 2080 bch2_write_ref_put(c, BCH_WRITE_REF_invalidate); 2081 percpu_ref_put(&ca->io_ref); 2082 } 2083 2084 void bch2_dev_do_invalidates(struct bch_dev *ca) 2085 { 2086 struct bch_fs *c = ca->fs; 2087 2088 if (!bch2_dev_get_ioref(c, ca->dev_idx, WRITE)) 2089 return; 2090 2091 if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_invalidate)) 2092 goto put_ioref; 2093 2094 if (queue_work(c->write_ref_wq, &ca->invalidate_work)) 2095 return; 2096 2097 bch2_write_ref_put(c, BCH_WRITE_REF_invalidate); 2098 put_ioref: 2099 percpu_ref_put(&ca->io_ref); 2100 } 2101 2102 void bch2_do_invalidates(struct bch_fs *c) 2103 { 2104 for_each_member_device(c, ca) 2105 bch2_dev_do_invalidates(ca); 2106 } 2107 2108 int bch2_dev_freespace_init(struct bch_fs *c, struct bch_dev *ca, 2109 u64 bucket_start, u64 bucket_end) 2110 { 2111 struct btree_trans *trans = bch2_trans_get(c); 2112 struct btree_iter iter; 2113 struct bkey_s_c k; 2114 struct bkey hole; 2115 struct bpos end = POS(ca->dev_idx, bucket_end); 2116 struct bch_member *m; 2117 unsigned long last_updated = jiffies; 2118 int ret; 2119 2120 BUG_ON(bucket_start > bucket_end); 2121 BUG_ON(bucket_end > ca->mi.nbuckets); 2122 2123 bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc, 2124 POS(ca->dev_idx, max_t(u64, ca->mi.first_bucket, bucket_start)), 2125 BTREE_ITER_prefetch); 2126 /* 2127 * Scan the alloc btree for every bucket on @ca, and add buckets to the 2128 * freespace/need_discard/need_gc_gens btrees as needed: 2129 */ 2130 while (1) { 2131 if (last_updated + HZ * 10 < jiffies) { 2132 bch_info(ca, "%s: currently at %llu/%llu", 2133 __func__, iter.pos.offset, ca->mi.nbuckets); 2134 last_updated = jiffies; 2135 } 2136 2137 bch2_trans_begin(trans); 2138 2139 if (bkey_ge(iter.pos, end)) { 2140 ret = 0; 2141 break; 2142 } 2143 2144 k = bch2_get_key_or_hole(&iter, end, &hole); 2145 ret = bkey_err(k); 2146 if (ret) 2147 goto bkey_err; 2148 2149 if (k.k->type) { 2150 /* 2151 * We process live keys in the alloc btree one at a 2152 * time: 2153 */ 2154 struct bch_alloc_v4 a_convert; 2155 const struct bch_alloc_v4 *a = bch2_alloc_to_v4(k, &a_convert); 2156 2157 ret = bch2_bucket_do_index(trans, ca, k, a, true) ?: 2158 bch2_trans_commit(trans, NULL, NULL, 2159 BCH_TRANS_COMMIT_no_enospc); 2160 if (ret) 2161 goto bkey_err; 2162 2163 bch2_btree_iter_advance(&iter); 2164 } else { 2165 struct bkey_i *freespace; 2166 2167 freespace = bch2_trans_kmalloc(trans, sizeof(*freespace)); 2168 ret = PTR_ERR_OR_ZERO(freespace); 2169 if (ret) 2170 goto bkey_err; 2171 2172 bkey_init(&freespace->k); 2173 freespace->k.type = KEY_TYPE_set; 2174 freespace->k.p = k.k->p; 2175 freespace->k.size = k.k->size; 2176 2177 ret = bch2_btree_insert_trans(trans, BTREE_ID_freespace, freespace, 0) ?: 2178 bch2_trans_commit(trans, NULL, NULL, 2179 BCH_TRANS_COMMIT_no_enospc); 2180 if (ret) 2181 goto bkey_err; 2182 2183 bch2_btree_iter_set_pos(&iter, k.k->p); 2184 } 2185 bkey_err: 2186 if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) 2187 continue; 2188 if (ret) 2189 break; 2190 } 2191 2192 bch2_trans_iter_exit(trans, &iter); 2193 bch2_trans_put(trans); 2194 2195 if (ret < 0) { 2196 bch_err_msg(ca, ret, "initializing free space"); 2197 return ret; 2198 } 2199 2200 mutex_lock(&c->sb_lock); 2201 m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx); 2202 SET_BCH_MEMBER_FREESPACE_INITIALIZED(m, true); 2203 mutex_unlock(&c->sb_lock); 2204 2205 return 0; 2206 } 2207 2208 int bch2_fs_freespace_init(struct bch_fs *c) 2209 { 2210 int ret = 0; 2211 bool doing_init = false; 2212 2213 /* 2214 * We can crash during the device add path, so we need to check this on 2215 * every mount: 2216 */ 2217 2218 for_each_member_device(c, ca) { 2219 if (ca->mi.freespace_initialized) 2220 continue; 2221 2222 if (!doing_init) { 2223 bch_info(c, "initializing freespace"); 2224 doing_init = true; 2225 } 2226 2227 ret = bch2_dev_freespace_init(c, ca, 0, ca->mi.nbuckets); 2228 if (ret) { 2229 bch2_dev_put(ca); 2230 bch_err_fn(c, ret); 2231 return ret; 2232 } 2233 } 2234 2235 if (doing_init) { 2236 mutex_lock(&c->sb_lock); 2237 bch2_write_super(c); 2238 mutex_unlock(&c->sb_lock); 2239 bch_verbose(c, "done initializing freespace"); 2240 } 2241 2242 return 0; 2243 } 2244 2245 /* Bucket IO clocks: */ 2246 2247 int bch2_bucket_io_time_reset(struct btree_trans *trans, unsigned dev, 2248 size_t bucket_nr, int rw) 2249 { 2250 struct bch_fs *c = trans->c; 2251 struct btree_iter iter; 2252 struct bkey_i_alloc_v4 *a; 2253 u64 now; 2254 int ret = 0; 2255 2256 if (bch2_trans_relock(trans)) 2257 bch2_trans_begin(trans); 2258 2259 a = bch2_trans_start_alloc_update_noupdate(trans, &iter, POS(dev, bucket_nr)); 2260 ret = PTR_ERR_OR_ZERO(a); 2261 if (ret) 2262 return ret; 2263 2264 now = bch2_current_io_time(c, rw); 2265 if (a->v.io_time[rw] == now) 2266 goto out; 2267 2268 a->v.io_time[rw] = now; 2269 2270 ret = bch2_trans_update(trans, &iter, &a->k_i, 0) ?: 2271 bch2_trans_commit(trans, NULL, NULL, 0); 2272 out: 2273 bch2_trans_iter_exit(trans, &iter); 2274 return ret; 2275 } 2276 2277 /* Startup/shutdown (ro/rw): */ 2278 2279 void bch2_recalc_capacity(struct bch_fs *c) 2280 { 2281 u64 capacity = 0, reserved_sectors = 0, gc_reserve; 2282 unsigned bucket_size_max = 0; 2283 unsigned long ra_pages = 0; 2284 2285 lockdep_assert_held(&c->state_lock); 2286 2287 for_each_online_member(c, ca) { 2288 struct backing_dev_info *bdi = ca->disk_sb.bdev->bd_disk->bdi; 2289 2290 ra_pages += bdi->ra_pages; 2291 } 2292 2293 bch2_set_ra_pages(c, ra_pages); 2294 2295 for_each_rw_member(c, ca) { 2296 u64 dev_reserve = 0; 2297 2298 /* 2299 * We need to reserve buckets (from the number 2300 * of currently available buckets) against 2301 * foreground writes so that mainly copygc can 2302 * make forward progress. 2303 * 2304 * We need enough to refill the various reserves 2305 * from scratch - copygc will use its entire 2306 * reserve all at once, then run against when 2307 * its reserve is refilled (from the formerly 2308 * available buckets). 2309 * 2310 * This reserve is just used when considering if 2311 * allocations for foreground writes must wait - 2312 * not -ENOSPC calculations. 2313 */ 2314 2315 dev_reserve += ca->nr_btree_reserve * 2; 2316 dev_reserve += ca->mi.nbuckets >> 6; /* copygc reserve */ 2317 2318 dev_reserve += 1; /* btree write point */ 2319 dev_reserve += 1; /* copygc write point */ 2320 dev_reserve += 1; /* rebalance write point */ 2321 2322 dev_reserve *= ca->mi.bucket_size; 2323 2324 capacity += bucket_to_sector(ca, ca->mi.nbuckets - 2325 ca->mi.first_bucket); 2326 2327 reserved_sectors += dev_reserve * 2; 2328 2329 bucket_size_max = max_t(unsigned, bucket_size_max, 2330 ca->mi.bucket_size); 2331 } 2332 2333 gc_reserve = c->opts.gc_reserve_bytes 2334 ? c->opts.gc_reserve_bytes >> 9 2335 : div64_u64(capacity * c->opts.gc_reserve_percent, 100); 2336 2337 reserved_sectors = max(gc_reserve, reserved_sectors); 2338 2339 reserved_sectors = min(reserved_sectors, capacity); 2340 2341 c->capacity = capacity - reserved_sectors; 2342 2343 c->bucket_size_max = bucket_size_max; 2344 2345 /* Wake up case someone was waiting for buckets */ 2346 closure_wake_up(&c->freelist_wait); 2347 } 2348 2349 u64 bch2_min_rw_member_capacity(struct bch_fs *c) 2350 { 2351 u64 ret = U64_MAX; 2352 2353 for_each_rw_member(c, ca) 2354 ret = min(ret, ca->mi.nbuckets * ca->mi.bucket_size); 2355 return ret; 2356 } 2357 2358 static bool bch2_dev_has_open_write_point(struct bch_fs *c, struct bch_dev *ca) 2359 { 2360 struct open_bucket *ob; 2361 bool ret = false; 2362 2363 for (ob = c->open_buckets; 2364 ob < c->open_buckets + ARRAY_SIZE(c->open_buckets); 2365 ob++) { 2366 spin_lock(&ob->lock); 2367 if (ob->valid && !ob->on_partial_list && 2368 ob->dev == ca->dev_idx) 2369 ret = true; 2370 spin_unlock(&ob->lock); 2371 } 2372 2373 return ret; 2374 } 2375 2376 /* device goes ro: */ 2377 void bch2_dev_allocator_remove(struct bch_fs *c, struct bch_dev *ca) 2378 { 2379 unsigned i; 2380 2381 /* First, remove device from allocation groups: */ 2382 2383 for (i = 0; i < ARRAY_SIZE(c->rw_devs); i++) 2384 clear_bit(ca->dev_idx, c->rw_devs[i].d); 2385 2386 /* 2387 * Capacity is calculated based off of devices in allocation groups: 2388 */ 2389 bch2_recalc_capacity(c); 2390 2391 bch2_open_buckets_stop(c, ca, false); 2392 2393 /* 2394 * Wake up threads that were blocked on allocation, so they can notice 2395 * the device can no longer be removed and the capacity has changed: 2396 */ 2397 closure_wake_up(&c->freelist_wait); 2398 2399 /* 2400 * journal_res_get() can block waiting for free space in the journal - 2401 * it needs to notice there may not be devices to allocate from anymore: 2402 */ 2403 wake_up(&c->journal.wait); 2404 2405 /* Now wait for any in flight writes: */ 2406 2407 closure_wait_event(&c->open_buckets_wait, 2408 !bch2_dev_has_open_write_point(c, ca)); 2409 } 2410 2411 /* device goes rw: */ 2412 void bch2_dev_allocator_add(struct bch_fs *c, struct bch_dev *ca) 2413 { 2414 unsigned i; 2415 2416 for (i = 0; i < ARRAY_SIZE(c->rw_devs); i++) 2417 if (ca->mi.data_allowed & (1 << i)) 2418 set_bit(ca->dev_idx, c->rw_devs[i].d); 2419 } 2420 2421 void bch2_dev_allocator_background_exit(struct bch_dev *ca) 2422 { 2423 darray_exit(&ca->discard_buckets_in_flight); 2424 } 2425 2426 void bch2_dev_allocator_background_init(struct bch_dev *ca) 2427 { 2428 mutex_init(&ca->discard_buckets_in_flight_lock); 2429 INIT_WORK(&ca->discard_work, bch2_do_discards_work); 2430 INIT_WORK(&ca->discard_fast_work, bch2_do_discards_fast_work); 2431 INIT_WORK(&ca->invalidate_work, bch2_do_invalidates_work); 2432 } 2433 2434 void bch2_fs_allocator_background_init(struct bch_fs *c) 2435 { 2436 spin_lock_init(&c->freelist_lock); 2437 } 2438