git.sesse.net Git - bcachefs-tools-debian/blob - libbcachefs/alloc_background.c

   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 /* Persistent alloc info: */
  33
  34 static const unsigned BCH_ALLOC_V1_FIELD_BYTES[] = {
  35 #define x(name, bits) [BCH_ALLOC_FIELD_V1_##name] = bits / 8,
  36         BCH_ALLOC_FIELDS_V1()
  37 #undef x
  38 };
  39
  40 struct bkey_alloc_unpacked {
  41         u64             journal_seq;
  42         u8              gen;
  43         u8              oldest_gen;
  44         u8              data_type;
  45         bool            need_discard:1;
  46         bool            need_inc_gen:1;
  47 #define x(_name, _bits) u##_bits _name;
  48         BCH_ALLOC_FIELDS_V2()
  49 #undef  x
  50 };
  51
  52 static inline u64 alloc_field_v1_get(const struct bch_alloc *a,
  53                                      const void **p, unsigned field)
  54 {
  55         unsigned bytes = BCH_ALLOC_V1_FIELD_BYTES[field];
  56         u64 v;
  57
  58         if (!(a->fields & (1 << field)))
  59                 return 0;
  60
  61         switch (bytes) {
  62         case 1:
  63                 v = *((const u8 *) *p);
  64                 break;
  65         case 2:
  66                 v = le16_to_cpup(*p);
  67                 break;
  68         case 4:
  69                 v = le32_to_cpup(*p);
  70                 break;
  71         case 8:
  72                 v = le64_to_cpup(*p);
  73                 break;
  74         default:
  75                 BUG();
  76         }
  77
  78         *p += bytes;
  79         return v;
  80 }
  81
  82 static void bch2_alloc_unpack_v1(struct bkey_alloc_unpacked *out,
  83                                  struct bkey_s_c k)
  84 {
  85         const struct bch_alloc *in = bkey_s_c_to_alloc(k).v;
  86         const void *d = in->data;
  87         unsigned idx = 0;
  88
  89         out->gen = in->gen;
  90
  91 #define x(_name, _bits) out->_name = alloc_field_v1_get(in, &d, idx++);
  92         BCH_ALLOC_FIELDS_V1()
  93 #undef  x
  94 }
  95
  96 static int bch2_alloc_unpack_v2(struct bkey_alloc_unpacked *out,
  97                                 struct bkey_s_c k)
  98 {
  99         struct bkey_s_c_alloc_v2 a = bkey_s_c_to_alloc_v2(k);
 100         const u8 *in = a.v->data;
 101         const u8 *end = bkey_val_end(a);
 102         unsigned fieldnr = 0;
 103         int ret;
 104         u64 v;
 105
 106         out->gen        = a.v->gen;
 107         out->oldest_gen = a.v->oldest_gen;
 108         out->data_type  = a.v->data_type;
 109
 110 #define x(_name, _bits)                                                 \
 111         if (fieldnr < a.v->nr_fields) {                                 \
 112                 ret = bch2_varint_decode_fast(in, end, &v);             \
 113                 if (ret < 0)                                            \
 114                         return ret;                                     \
 115                 in += ret;                                              \
 116         } else {                                                        \
 117                 v = 0;                                                  \
 118         }                                                               \
 119         out->_name = v;                                                 \
 120         if (v != out->_name)                                            \
 121                 return -1;                                              \
 122         fieldnr++;
 123
 124         BCH_ALLOC_FIELDS_V2()
 125 #undef  x
 126         return 0;
 127 }
 128
 129 static int bch2_alloc_unpack_v3(struct bkey_alloc_unpacked *out,
 130                                 struct bkey_s_c k)
 131 {
 132         struct bkey_s_c_alloc_v3 a = bkey_s_c_to_alloc_v3(k);
 133         const u8 *in = a.v->data;
 134         const u8 *end = bkey_val_end(a);
 135         unsigned fieldnr = 0;
 136         int ret;
 137         u64 v;
 138
 139         out->gen        = a.v->gen;
 140         out->oldest_gen = a.v->oldest_gen;
 141         out->data_type  = a.v->data_type;
 142         out->need_discard = BCH_ALLOC_V3_NEED_DISCARD(a.v);
 143         out->need_inc_gen = BCH_ALLOC_V3_NEED_INC_GEN(a.v);
 144         out->journal_seq = le64_to_cpu(a.v->journal_seq);
 145
 146 #define x(_name, _bits)                                                 \
 147         if (fieldnr < a.v->nr_fields) {                                 \
 148                 ret = bch2_varint_decode_fast(in, end, &v);             \
 149                 if (ret < 0)                                            \
 150                         return ret;                                     \
 151                 in += ret;                                              \
 152         } else {                                                        \
 153                 v = 0;                                                  \
 154         }                                                               \
 155         out->_name = v;                                                 \
 156         if (v != out->_name)                                            \
 157                 return -1;                                              \
 158         fieldnr++;
 159
 160         BCH_ALLOC_FIELDS_V2()
 161 #undef  x
 162         return 0;
 163 }
 164
 165 static struct bkey_alloc_unpacked bch2_alloc_unpack(struct bkey_s_c k)
 166 {
 167         struct bkey_alloc_unpacked ret = { .gen = 0 };
 168
 169         switch (k.k->type) {
 170         case KEY_TYPE_alloc:
 171                 bch2_alloc_unpack_v1(&ret, k);
 172                 break;
 173         case KEY_TYPE_alloc_v2:
 174                 bch2_alloc_unpack_v2(&ret, k);
 175                 break;
 176         case KEY_TYPE_alloc_v3:
 177                 bch2_alloc_unpack_v3(&ret, k);
 178                 break;
 179         }
 180
 181         return ret;
 182 }
 183
 184 static unsigned bch_alloc_v1_val_u64s(const struct bch_alloc *a)
 185 {
 186         unsigned i, bytes = offsetof(struct bch_alloc, data);
 187
 188         for (i = 0; i < ARRAY_SIZE(BCH_ALLOC_V1_FIELD_BYTES); i++)
 189                 if (a->fields & (1 << i))
 190                         bytes += BCH_ALLOC_V1_FIELD_BYTES[i];
 191
 192         return DIV_ROUND_UP(bytes, sizeof(u64));
 193 }
 194
 195 int bch2_alloc_v1_invalid(struct bch_fs *c, struct bkey_s_c k,
 196                           enum bkey_invalid_flags flags,
 197                           struct printbuf *err)
 198 {
 199         struct bkey_s_c_alloc a = bkey_s_c_to_alloc(k);
 200         int ret = 0;
 201
 202         /* allow for unknown fields */
 203         bkey_fsck_err_on(bkey_val_u64s(a.k) < bch_alloc_v1_val_u64s(a.v), c, err,
 204                          alloc_v1_val_size_bad,
 205                          "incorrect value size (%zu < %u)",
 206                          bkey_val_u64s(a.k), bch_alloc_v1_val_u64s(a.v));
 207 fsck_err:
 208         return ret;
 209 }
 210
 211 int bch2_alloc_v2_invalid(struct bch_fs *c, struct bkey_s_c k,
 212                           enum bkey_invalid_flags flags,
 213                           struct printbuf *err)
 214 {
 215         struct bkey_alloc_unpacked u;
 216         int ret = 0;
 217
 218         bkey_fsck_err_on(bch2_alloc_unpack_v2(&u, k), c, err,
 219                          alloc_v2_unpack_error,
 220                          "unpack error");
 221 fsck_err:
 222         return ret;
 223 }
 224
 225 int bch2_alloc_v3_invalid(struct bch_fs *c, struct bkey_s_c k,
 226                           enum bkey_invalid_flags flags,
 227                           struct printbuf *err)
 228 {
 229         struct bkey_alloc_unpacked u;
 230         int ret = 0;
 231
 232         bkey_fsck_err_on(bch2_alloc_unpack_v3(&u, k), c, err,
 233                          alloc_v2_unpack_error,
 234                          "unpack error");
 235 fsck_err:
 236         return ret;
 237 }
 238
 239 int bch2_alloc_v4_invalid(struct bch_fs *c, struct bkey_s_c k,
 240                           enum bkey_invalid_flags flags, struct printbuf *err)
 241 {
 242         struct bkey_s_c_alloc_v4 a = bkey_s_c_to_alloc_v4(k);
 243         int ret = 0;
 244
 245         bkey_fsck_err_on(alloc_v4_u64s(a.v) > bkey_val_u64s(k.k), c, err,
 246                          alloc_v4_val_size_bad,
 247                          "bad val size (%u > %zu)",
 248                          alloc_v4_u64s(a.v), bkey_val_u64s(k.k));
 249
 250         bkey_fsck_err_on(!BCH_ALLOC_V4_BACKPOINTERS_START(a.v) &&
 251                          BCH_ALLOC_V4_NR_BACKPOINTERS(a.v), c, err,
 252                          alloc_v4_backpointers_start_bad,
 253                          "invalid backpointers_start");
 254
 255         bkey_fsck_err_on(alloc_data_type(*a.v, a.v->data_type) != a.v->data_type, c, err,
 256                          alloc_key_data_type_bad,
 257                          "invalid data type (got %u should be %u)",
 258                          a.v->data_type, alloc_data_type(*a.v, a.v->data_type));
 259
 260         switch (a.v->data_type) {
 261         case BCH_DATA_free:
 262         case BCH_DATA_need_gc_gens:
 263         case BCH_DATA_need_discard:
 264                 bkey_fsck_err_on(bch2_bucket_sectors(*a.v) || a.v->stripe,
 265                                  c, err, alloc_key_empty_but_have_data,
 266                                  "empty data type free but have data");
 267                 break;
 268         case BCH_DATA_sb:
 269         case BCH_DATA_journal:
 270         case BCH_DATA_btree:
 271         case BCH_DATA_user:
 272         case BCH_DATA_parity:
 273                 bkey_fsck_err_on(!bch2_bucket_sectors_dirty(*a.v),
 274                                  c, err, alloc_key_dirty_sectors_0,
 275                                  "data_type %s but dirty_sectors==0",
 276                                  bch2_data_types[a.v->data_type]);
 277                 break;
 278         case BCH_DATA_cached:
 279                 bkey_fsck_err_on(!a.v->cached_sectors ||
 280                                  bch2_bucket_sectors_dirty(*a.v) ||
 281                                  a.v->stripe,
 282                                  c, err, alloc_key_cached_inconsistency,
 283                                  "data type inconsistency");
 284
 285                 bkey_fsck_err_on(!a.v->io_time[READ] &&
 286                                  c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_to_lru_refs,
 287                                  c, err, alloc_key_cached_but_read_time_zero,
 288                                  "cached bucket with read_time == 0");
 289                 break;
 290         case BCH_DATA_stripe:
 291                 break;
 292         }
 293 fsck_err:
 294         return ret;
 295 }
 296
 297 void bch2_alloc_v4_swab(struct bkey_s k)
 298 {
 299         struct bch_alloc_v4 *a = bkey_s_to_alloc_v4(k).v;
 300         struct bch_backpointer *bp, *bps;
 301
 302         a->journal_seq          = swab64(a->journal_seq);
 303         a->flags                = swab32(a->flags);
 304         a->dirty_sectors        = swab32(a->dirty_sectors);
 305         a->cached_sectors       = swab32(a->cached_sectors);
 306         a->io_time[0]           = swab64(a->io_time[0]);
 307         a->io_time[1]           = swab64(a->io_time[1]);
 308         a->stripe               = swab32(a->stripe);
 309         a->nr_external_backpointers = swab32(a->nr_external_backpointers);
 310         a->fragmentation_lru    = swab64(a->fragmentation_lru);
 311
 312         bps = alloc_v4_backpointers(a);
 313         for (bp = bps; bp < bps + BCH_ALLOC_V4_NR_BACKPOINTERS(a); bp++) {
 314                 bp->bucket_offset       = swab40(bp->bucket_offset);
 315                 bp->bucket_len          = swab32(bp->bucket_len);
 316                 bch2_bpos_swab(&bp->pos);
 317         }
 318 }
 319
 320 void bch2_alloc_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c k)
 321 {
 322         struct bch_alloc_v4 _a;
 323         const struct bch_alloc_v4 *a = bch2_alloc_to_v4(k, &_a);
 324         unsigned i;
 325
 326         prt_newline(out);
 327         printbuf_indent_add(out, 2);
 328
 329         prt_printf(out, "gen %u oldest_gen %u data_type %s",
 330                a->gen, a->oldest_gen,
 331                a->data_type < BCH_DATA_NR
 332                ? bch2_data_types[a->data_type]
 333                : "(invalid data type)");
 334         prt_newline(out);
 335         prt_printf(out, "journal_seq       %llu",       a->journal_seq);
 336         prt_newline(out);
 337         prt_printf(out, "need_discard      %llu",       BCH_ALLOC_V4_NEED_DISCARD(a));
 338         prt_newline(out);
 339         prt_printf(out, "need_inc_gen      %llu",       BCH_ALLOC_V4_NEED_INC_GEN(a));
 340         prt_newline(out);
 341         prt_printf(out, "dirty_sectors     %u", a->dirty_sectors);
 342         prt_newline(out);
 343         prt_printf(out, "cached_sectors    %u", a->cached_sectors);
 344         prt_newline(out);
 345         prt_printf(out, "stripe            %u", a->stripe);
 346         prt_newline(out);
 347         prt_printf(out, "stripe_redundancy %u", a->stripe_redundancy);
 348         prt_newline(out);
 349         prt_printf(out, "io_time[READ]     %llu",       a->io_time[READ]);
 350         prt_newline(out);
 351         prt_printf(out, "io_time[WRITE]    %llu",       a->io_time[WRITE]);
 352         prt_newline(out);
 353         prt_printf(out, "fragmentation     %llu",       a->fragmentation_lru);
 354         prt_newline(out);
 355         prt_printf(out, "bp_start          %llu", BCH_ALLOC_V4_BACKPOINTERS_START(a));
 356         prt_newline(out);
 357
 358         if (BCH_ALLOC_V4_NR_BACKPOINTERS(a)) {
 359                 struct bkey_s_c_alloc_v4 a_raw = bkey_s_c_to_alloc_v4(k);
 360                 const struct bch_backpointer *bps = alloc_v4_backpointers_c(a_raw.v);
 361
 362                 prt_printf(out, "backpointers:     %llu", BCH_ALLOC_V4_NR_BACKPOINTERS(a_raw.v));
 363                 printbuf_indent_add(out, 2);
 364
 365                 for (i = 0; i < BCH_ALLOC_V4_NR_BACKPOINTERS(a_raw.v); i++) {
 366                         prt_newline(out);
 367                         bch2_backpointer_to_text(out, &bps[i]);
 368                 }
 369
 370                 printbuf_indent_sub(out, 2);
 371         }
 372
 373         printbuf_indent_sub(out, 2);
 374 }
 375
 376 void __bch2_alloc_to_v4(struct bkey_s_c k, struct bch_alloc_v4 *out)
 377 {
 378         if (k.k->type == KEY_TYPE_alloc_v4) {
 379                 void *src, *dst;
 380
 381                 *out = *bkey_s_c_to_alloc_v4(k).v;
 382
 383                 src = alloc_v4_backpointers(out);
 384                 SET_BCH_ALLOC_V4_BACKPOINTERS_START(out, BCH_ALLOC_V4_U64s);
 385                 dst = alloc_v4_backpointers(out);
 386
 387                 if (src < dst)
 388                         memset(src, 0, dst - src);
 389
 390                 SET_BCH_ALLOC_V4_NR_BACKPOINTERS(out, 0);
 391         } else {
 392                 struct bkey_alloc_unpacked u = bch2_alloc_unpack(k);
 393
 394                 *out = (struct bch_alloc_v4) {
 395                         .journal_seq            = u.journal_seq,
 396                         .flags                  = u.need_discard,
 397                         .gen                    = u.gen,
 398                         .oldest_gen             = u.oldest_gen,
 399                         .data_type              = u.data_type,
 400                         .stripe_redundancy      = u.stripe_redundancy,
 401                         .dirty_sectors          = u.dirty_sectors,
 402                         .cached_sectors         = u.cached_sectors,
 403                         .io_time[READ]          = u.read_time,
 404                         .io_time[WRITE]         = u.write_time,
 405                         .stripe                 = u.stripe,
 406                 };
 407
 408                 SET_BCH_ALLOC_V4_BACKPOINTERS_START(out, BCH_ALLOC_V4_U64s);
 409         }
 410 }
 411
 412 static noinline struct bkey_i_alloc_v4 *
 413 __bch2_alloc_to_v4_mut(struct btree_trans *trans, struct bkey_s_c k)
 414 {
 415         struct bkey_i_alloc_v4 *ret;
 416
 417         ret = bch2_trans_kmalloc(trans, max(bkey_bytes(k.k), sizeof(struct bkey_i_alloc_v4)));
 418         if (IS_ERR(ret))
 419                 return ret;
 420
 421         if (k.k->type == KEY_TYPE_alloc_v4) {
 422                 void *src, *dst;
 423
 424                 bkey_reassemble(&ret->k_i, k);
 425
 426                 src = alloc_v4_backpointers(&ret->v);
 427                 SET_BCH_ALLOC_V4_BACKPOINTERS_START(&ret->v, BCH_ALLOC_V4_U64s);
 428                 dst = alloc_v4_backpointers(&ret->v);
 429
 430                 if (src < dst)
 431                         memset(src, 0, dst - src);
 432
 433                 SET_BCH_ALLOC_V4_NR_BACKPOINTERS(&ret->v, 0);
 434                 set_alloc_v4_u64s(ret);
 435         } else {
 436                 bkey_alloc_v4_init(&ret->k_i);
 437                 ret->k.p = k.k->p;
 438                 bch2_alloc_to_v4(k, &ret->v);
 439         }
 440         return ret;
 441 }
 442
 443 static inline struct bkey_i_alloc_v4 *bch2_alloc_to_v4_mut_inlined(struct btree_trans *trans, struct bkey_s_c k)
 444 {
 445         struct bkey_s_c_alloc_v4 a;
 446
 447         if (likely(k.k->type == KEY_TYPE_alloc_v4) &&
 448             ((a = bkey_s_c_to_alloc_v4(k), true) &&
 449              BCH_ALLOC_V4_NR_BACKPOINTERS(a.v) == 0))
 450                 return bch2_bkey_make_mut_noupdate_typed(trans, k, alloc_v4);
 451
 452         return __bch2_alloc_to_v4_mut(trans, k);
 453 }
 454
 455 struct bkey_i_alloc_v4 *bch2_alloc_to_v4_mut(struct btree_trans *trans, struct bkey_s_c k)
 456 {
 457         return bch2_alloc_to_v4_mut_inlined(trans, k);
 458 }
 459
 460 struct bkey_i_alloc_v4 *
 461 bch2_trans_start_alloc_update(struct btree_trans *trans, struct btree_iter *iter,
 462                               struct bpos pos)
 463 {
 464         struct bkey_s_c k;
 465         struct bkey_i_alloc_v4 *a;
 466         int ret;
 467
 468         k = bch2_bkey_get_iter(trans, iter, BTREE_ID_alloc, pos,
 469                              BTREE_ITER_WITH_UPDATES|
 470                              BTREE_ITER_CACHED|
 471                              BTREE_ITER_INTENT);
 472         ret = bkey_err(k);
 473         if (unlikely(ret))
 474                 return ERR_PTR(ret);
 475
 476         a = bch2_alloc_to_v4_mut_inlined(trans, k);
 477         ret = PTR_ERR_OR_ZERO(a);
 478         if (unlikely(ret))
 479                 goto err;
 480         return a;
 481 err:
 482         bch2_trans_iter_exit(trans, iter);
 483         return ERR_PTR(ret);
 484 }
 485
 486 static struct bpos alloc_gens_pos(struct bpos pos, unsigned *offset)
 487 {
 488         *offset = pos.offset & KEY_TYPE_BUCKET_GENS_MASK;
 489
 490         pos.offset >>= KEY_TYPE_BUCKET_GENS_BITS;
 491         return pos;
 492 }
 493
 494 static struct bpos bucket_gens_pos_to_alloc(struct bpos pos, unsigned offset)
 495 {
 496         pos.offset <<= KEY_TYPE_BUCKET_GENS_BITS;
 497         pos.offset += offset;
 498         return pos;
 499 }
 500
 501 static unsigned alloc_gen(struct bkey_s_c k, unsigned offset)
 502 {
 503         return k.k->type == KEY_TYPE_bucket_gens
 504                 ? bkey_s_c_to_bucket_gens(k).v->gens[offset]
 505                 : 0;
 506 }
 507
 508 int bch2_bucket_gens_invalid(struct bch_fs *c, struct bkey_s_c k,
 509                              enum bkey_invalid_flags flags,
 510                              struct printbuf *err)
 511 {
 512         int ret = 0;
 513
 514         bkey_fsck_err_on(bkey_val_bytes(k.k) != sizeof(struct bch_bucket_gens), c, err,
 515                          bucket_gens_val_size_bad,
 516                          "bad val size (%zu != %zu)",
 517                          bkey_val_bytes(k.k), sizeof(struct bch_bucket_gens));
 518 fsck_err:
 519         return ret;
 520 }
 521
 522 void bch2_bucket_gens_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c k)
 523 {
 524         struct bkey_s_c_bucket_gens g = bkey_s_c_to_bucket_gens(k);
 525         unsigned i;
 526
 527         for (i = 0; i < ARRAY_SIZE(g.v->gens); i++) {
 528                 if (i)
 529                         prt_char(out, ' ');
 530                 prt_printf(out, "%u", g.v->gens[i]);
 531         }
 532 }
 533
 534 int bch2_bucket_gens_init(struct bch_fs *c)
 535 {
 536         struct btree_trans *trans = bch2_trans_get(c);
 537         struct btree_iter iter;
 538         struct bkey_s_c k;
 539         struct bch_alloc_v4 a;
 540         struct bkey_i_bucket_gens g;
 541         bool have_bucket_gens_key = false;
 542         unsigned offset;
 543         struct bpos pos;
 544         u8 gen;
 545         int ret;
 546
 547         for_each_btree_key(trans, iter, BTREE_ID_alloc, POS_MIN,
 548                            BTREE_ITER_PREFETCH, k, ret) {
 549                 /*
 550                  * Not a fsck error because this is checked/repaired by
 551                  * bch2_check_alloc_key() which runs later:
 552                  */
 553                 if (!bch2_dev_bucket_exists(c, k.k->p))
 554                         continue;
 555
 556                 gen = bch2_alloc_to_v4(k, &a)->gen;
 557                 pos = alloc_gens_pos(iter.pos, &offset);
 558
 559                 if (have_bucket_gens_key && bkey_cmp(iter.pos, pos)) {
 560                         ret = commit_do(trans, NULL, NULL,
 561                                         BCH_TRANS_COMMIT_no_enospc,
 562                                 bch2_btree_insert_trans(trans, BTREE_ID_bucket_gens, &g.k_i, 0));
 563                         if (ret)
 564                                 break;
 565                         have_bucket_gens_key = false;
 566                 }
 567
 568                 if (!have_bucket_gens_key) {
 569                         bkey_bucket_gens_init(&g.k_i);
 570                         g.k.p = pos;
 571                         have_bucket_gens_key = true;
 572                 }
 573
 574                 g.v.gens[offset] = gen;
 575         }
 576         bch2_trans_iter_exit(trans, &iter);
 577
 578         if (have_bucket_gens_key && !ret)
 579                 ret = commit_do(trans, NULL, NULL,
 580                                 BCH_TRANS_COMMIT_no_enospc,
 581                         bch2_btree_insert_trans(trans, BTREE_ID_bucket_gens, &g.k_i, 0));
 582
 583         bch2_trans_put(trans);
 584
 585         if (ret)
 586                 bch_err_fn(c, ret);
 587         return ret;
 588 }
 589
 590 int bch2_alloc_read(struct bch_fs *c)
 591 {
 592         struct btree_trans *trans = bch2_trans_get(c);
 593         struct btree_iter iter;
 594         struct bkey_s_c k;
 595         struct bch_dev *ca;
 596         int ret;
 597
 598         down_read(&c->gc_lock);
 599
 600         if (c->sb.version_upgrade_complete >= bcachefs_metadata_version_bucket_gens) {
 601                 const struct bch_bucket_gens *g;
 602                 u64 b;
 603
 604                 for_each_btree_key(trans, iter, BTREE_ID_bucket_gens, POS_MIN,
 605                                    BTREE_ITER_PREFETCH, k, ret) {
 606                         u64 start = bucket_gens_pos_to_alloc(k.k->p, 0).offset;
 607                         u64 end = bucket_gens_pos_to_alloc(bpos_nosnap_successor(k.k->p), 0).offset;
 608
 609                         if (k.k->type != KEY_TYPE_bucket_gens)
 610                                 continue;
 611
 612                         g = bkey_s_c_to_bucket_gens(k).v;
 613
 614                         /*
 615                          * Not a fsck error because this is checked/repaired by
 616                          * bch2_check_alloc_key() which runs later:
 617                          */
 618                         if (!bch2_dev_exists2(c, k.k->p.inode))
 619                                 continue;
 620
 621                         ca = bch_dev_bkey_exists(c, k.k->p.inode);
 622
 623                         for (b = max_t(u64, ca->mi.first_bucket, start);
 624                              b < min_t(u64, ca->mi.nbuckets, end);
 625                              b++)
 626                                 *bucket_gen(ca, b) = g->gens[b & KEY_TYPE_BUCKET_GENS_MASK];
 627                 }
 628                 bch2_trans_iter_exit(trans, &iter);
 629         } else {
 630                 struct bch_alloc_v4 a;
 631
 632                 for_each_btree_key(trans, iter, BTREE_ID_alloc, POS_MIN,
 633                                    BTREE_ITER_PREFETCH, k, ret) {
 634                         /*
 635                          * Not a fsck error because this is checked/repaired by
 636                          * bch2_check_alloc_key() which runs later:
 637                          */
 638                         if (!bch2_dev_bucket_exists(c, k.k->p))
 639                                 continue;
 640
 641                         ca = bch_dev_bkey_exists(c, k.k->p.inode);
 642
 643                         *bucket_gen(ca, k.k->p.offset) = bch2_alloc_to_v4(k, &a)->gen;
 644                 }
 645                 bch2_trans_iter_exit(trans, &iter);
 646         }
 647
 648         bch2_trans_put(trans);
 649         up_read(&c->gc_lock);
 650
 651         if (ret)
 652                 bch_err_fn(c, ret);
 653
 654         return ret;
 655 }
 656
 657 /* Free space/discard btree: */
 658
 659 static int bch2_bucket_do_index(struct btree_trans *trans,
 660                                 struct bkey_s_c alloc_k,
 661                                 const struct bch_alloc_v4 *a,
 662                                 bool set)
 663 {
 664         struct bch_fs *c = trans->c;
 665         struct bch_dev *ca = bch_dev_bkey_exists(c, alloc_k.k->p.inode);
 666         struct btree_iter iter;
 667         struct bkey_s_c old;
 668         struct bkey_i *k;
 669         enum btree_id btree;
 670         enum bch_bkey_type old_type = !set ? KEY_TYPE_set : KEY_TYPE_deleted;
 671         enum bch_bkey_type new_type =  set ? KEY_TYPE_set : KEY_TYPE_deleted;
 672         struct printbuf buf = PRINTBUF;
 673         int ret;
 674
 675         if (a->data_type != BCH_DATA_free &&
 676             a->data_type != BCH_DATA_need_discard)
 677                 return 0;
 678
 679         k = bch2_trans_kmalloc_nomemzero(trans, sizeof(*k));
 680         if (IS_ERR(k))
 681                 return PTR_ERR(k);
 682
 683         bkey_init(&k->k);
 684         k->k.type = new_type;
 685
 686         switch (a->data_type) {
 687         case BCH_DATA_free:
 688                 btree = BTREE_ID_freespace;
 689                 k->k.p = alloc_freespace_pos(alloc_k.k->p, *a);
 690                 bch2_key_resize(&k->k, 1);
 691                 break;
 692         case BCH_DATA_need_discard:
 693                 btree = BTREE_ID_need_discard;
 694                 k->k.p = alloc_k.k->p;
 695                 break;
 696         default:
 697                 return 0;
 698         }
 699
 700         old = bch2_bkey_get_iter(trans, &iter, btree,
 701                              bkey_start_pos(&k->k),
 702                              BTREE_ITER_INTENT);
 703         ret = bkey_err(old);
 704         if (ret)
 705                 return ret;
 706
 707         if (ca->mi.freespace_initialized &&
 708             c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_info &&
 709             bch2_trans_inconsistent_on(old.k->type != old_type, trans,
 710                         "incorrect key when %s %s:%llu:%llu:0 (got %s should be %s)\n"
 711                         "  for %s",
 712                         set ? "setting" : "clearing",
 713                         bch2_btree_id_str(btree),
 714                         iter.pos.inode,
 715                         iter.pos.offset,
 716                         bch2_bkey_types[old.k->type],
 717                         bch2_bkey_types[old_type],
 718                         (bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf))) {
 719                 ret = -EIO;
 720                 goto err;
 721         }
 722
 723         ret = bch2_trans_update(trans, &iter, k, 0);
 724 err:
 725         bch2_trans_iter_exit(trans, &iter);
 726         printbuf_exit(&buf);
 727         return ret;
 728 }
 729
 730 static noinline int bch2_bucket_gen_update(struct btree_trans *trans,
 731                                            struct bpos bucket, u8 gen)
 732 {
 733         struct btree_iter iter;
 734         unsigned offset;
 735         struct bpos pos = alloc_gens_pos(bucket, &offset);
 736         struct bkey_i_bucket_gens *g;
 737         struct bkey_s_c k;
 738         int ret;
 739
 740         g = bch2_trans_kmalloc(trans, sizeof(*g));
 741         ret = PTR_ERR_OR_ZERO(g);
 742         if (ret)
 743                 return ret;
 744
 745         k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_bucket_gens, pos,
 746                                BTREE_ITER_INTENT|
 747                                BTREE_ITER_WITH_UPDATES);
 748         ret = bkey_err(k);
 749         if (ret)
 750                 return ret;
 751
 752         if (k.k->type != KEY_TYPE_bucket_gens) {
 753                 bkey_bucket_gens_init(&g->k_i);
 754                 g->k.p = iter.pos;
 755         } else {
 756                 bkey_reassemble(&g->k_i, k);
 757         }
 758
 759         g->v.gens[offset] = gen;
 760
 761         ret = bch2_trans_update(trans, &iter, &g->k_i, 0);
 762         bch2_trans_iter_exit(trans, &iter);
 763         return ret;
 764 }
 765
 766 int bch2_trans_mark_alloc(struct btree_trans *trans,
 767                           enum btree_id btree_id, unsigned level,
 768                           struct bkey_s_c old, struct bkey_i *new,
 769                           unsigned flags)
 770 {
 771         struct bch_fs *c = trans->c;
 772         struct bch_alloc_v4 old_a_convert, *new_a;
 773         const struct bch_alloc_v4 *old_a;
 774         u64 old_lru, new_lru;
 775         int ret = 0;
 776
 777         /*
 778          * Deletion only happens in the device removal path, with
 779          * BTREE_TRIGGER_NORUN:
 780          */
 781         BUG_ON(new->k.type != KEY_TYPE_alloc_v4);
 782
 783         old_a = bch2_alloc_to_v4(old, &old_a_convert);
 784         new_a = &bkey_i_to_alloc_v4(new)->v;
 785
 786         new_a->data_type = alloc_data_type(*new_a, new_a->data_type);
 787
 788         if (bch2_bucket_sectors(*new_a) > bch2_bucket_sectors(*old_a)) {
 789                 new_a->io_time[READ] = max_t(u64, 1, atomic64_read(&c->io_clock[READ].now));
 790                 new_a->io_time[WRITE]= max_t(u64, 1, atomic64_read(&c->io_clock[WRITE].now));
 791                 SET_BCH_ALLOC_V4_NEED_INC_GEN(new_a, true);
 792                 SET_BCH_ALLOC_V4_NEED_DISCARD(new_a, true);
 793         }
 794
 795         if (data_type_is_empty(new_a->data_type) &&
 796             BCH_ALLOC_V4_NEED_INC_GEN(new_a) &&
 797             !bch2_bucket_is_open_safe(c, new->k.p.inode, new->k.p.offset)) {
 798                 new_a->gen++;
 799                 SET_BCH_ALLOC_V4_NEED_INC_GEN(new_a, false);
 800         }
 801
 802         if (old_a->data_type != new_a->data_type ||
 803             (new_a->data_type == BCH_DATA_free &&
 804              alloc_freespace_genbits(*old_a) != alloc_freespace_genbits(*new_a))) {
 805                 ret =   bch2_bucket_do_index(trans, old, old_a, false) ?:
 806                         bch2_bucket_do_index(trans, bkey_i_to_s_c(new), new_a, true);
 807                 if (ret)
 808                         return ret;
 809         }
 810
 811         if (new_a->data_type == BCH_DATA_cached &&
 812             !new_a->io_time[READ])
 813                 new_a->io_time[READ] = max_t(u64, 1, atomic64_read(&c->io_clock[READ].now));
 814
 815         old_lru = alloc_lru_idx_read(*old_a);
 816         new_lru = alloc_lru_idx_read(*new_a);
 817
 818         if (old_lru != new_lru) {
 819                 ret = bch2_lru_change(trans, new->k.p.inode,
 820                                       bucket_to_u64(new->k.p),
 821                                       old_lru, new_lru);
 822                 if (ret)
 823                         return ret;
 824         }
 825
 826         new_a->fragmentation_lru = alloc_lru_idx_fragmentation(*new_a,
 827                                         bch_dev_bkey_exists(c, new->k.p.inode));
 828
 829         if (old_a->fragmentation_lru != new_a->fragmentation_lru) {
 830                 ret = bch2_lru_change(trans,
 831                                 BCH_LRU_FRAGMENTATION_START,
 832                                 bucket_to_u64(new->k.p),
 833                                 old_a->fragmentation_lru, new_a->fragmentation_lru);
 834                 if (ret)
 835                         return ret;
 836         }
 837
 838         if (old_a->gen != new_a->gen) {
 839                 ret = bch2_bucket_gen_update(trans, new->k.p, new_a->gen);
 840                 if (ret)
 841                         return ret;
 842         }
 843
 844         /*
 845          * need to know if we're getting called from the invalidate path or
 846          * not:
 847          */
 848
 849         if ((flags & BTREE_TRIGGER_BUCKET_INVALIDATE) &&
 850             old_a->cached_sectors) {
 851                 ret = bch2_update_cached_sectors_list(trans, new->k.p.inode,
 852                                                       -((s64) old_a->cached_sectors));
 853                 if (ret)
 854                         return ret;
 855         }
 856
 857         return 0;
 858 }
 859
 860 /*
 861  * This synthesizes deleted extents for holes, similar to BTREE_ITER_SLOTS for
 862  * extents style btrees, but works on non-extents btrees:
 863  */
 864 static struct bkey_s_c bch2_get_key_or_hole(struct btree_iter *iter, struct bpos end, struct bkey *hole)
 865 {
 866         struct bkey_s_c k = bch2_btree_iter_peek_slot(iter);
 867
 868         if (bkey_err(k))
 869                 return k;
 870
 871         if (k.k->type) {
 872                 return k;
 873         } else {
 874                 struct btree_iter iter2;
 875                 struct bpos next;
 876
 877                 bch2_trans_copy_iter(&iter2, iter);
 878
 879                 if (!bpos_eq(iter->path->l[0].b->key.k.p, SPOS_MAX))
 880                         end = bkey_min(end, bpos_nosnap_successor(iter->path->l[0].b->key.k.p));
 881
 882                 end = bkey_min(end, POS(iter->pos.inode, iter->pos.offset + U32_MAX - 1));
 883
 884                 /*
 885                  * btree node min/max is a closed interval, upto takes a half
 886                  * open interval:
 887                  */
 888                 k = bch2_btree_iter_peek_upto(&iter2, end);
 889                 next = iter2.pos;
 890                 bch2_trans_iter_exit(iter->trans, &iter2);
 891
 892                 BUG_ON(next.offset >= iter->pos.offset + U32_MAX);
 893
 894                 if (bkey_err(k))
 895                         return k;
 896
 897                 bkey_init(hole);
 898                 hole->p = iter->pos;
 899
 900                 bch2_key_resize(hole, next.offset - iter->pos.offset);
 901                 return (struct bkey_s_c) { hole, NULL };
 902         }
 903 }
 904
 905 static bool next_bucket(struct bch_fs *c, struct bpos *bucket)
 906 {
 907         struct bch_dev *ca;
 908         unsigned iter;
 909
 910         if (bch2_dev_bucket_exists(c, *bucket))
 911                 return true;
 912
 913         if (bch2_dev_exists2(c, bucket->inode)) {
 914                 ca = bch_dev_bkey_exists(c, bucket->inode);
 915
 916                 if (bucket->offset < ca->mi.first_bucket) {
 917                         bucket->offset = ca->mi.first_bucket;
 918                         return true;
 919                 }
 920
 921                 bucket->inode++;
 922                 bucket->offset = 0;
 923         }
 924
 925         rcu_read_lock();
 926         iter = bucket->inode;
 927         ca = __bch2_next_dev(c, &iter, NULL);
 928         if (ca)
 929                 *bucket = POS(ca->dev_idx, ca->mi.first_bucket);
 930         rcu_read_unlock();
 931
 932         return ca != NULL;
 933 }
 934
 935 static struct bkey_s_c bch2_get_key_or_real_bucket_hole(struct btree_iter *iter, struct bkey *hole)
 936 {
 937         struct bch_fs *c = iter->trans->c;
 938         struct bkey_s_c k;
 939 again:
 940         k = bch2_get_key_or_hole(iter, POS_MAX, hole);
 941         if (bkey_err(k))
 942                 return k;
 943
 944         if (!k.k->type) {
 945                 struct bpos bucket = bkey_start_pos(k.k);
 946
 947                 if (!bch2_dev_bucket_exists(c, bucket)) {
 948                         if (!next_bucket(c, &bucket))
 949                                 return bkey_s_c_null;
 950
 951                         bch2_btree_iter_set_pos(iter, bucket);
 952                         goto again;
 953                 }
 954
 955                 if (!bch2_dev_bucket_exists(c, k.k->p)) {
 956                         struct bch_dev *ca = bch_dev_bkey_exists(c, bucket.inode);
 957
 958                         bch2_key_resize(hole, ca->mi.nbuckets - bucket.offset);
 959                 }
 960         }
 961
 962         return k;
 963 }
 964
 965 static noinline_for_stack
 966 int bch2_check_alloc_key(struct btree_trans *trans,
 967                          struct bkey_s_c alloc_k,
 968                          struct btree_iter *alloc_iter,
 969                          struct btree_iter *discard_iter,
 970                          struct btree_iter *freespace_iter,
 971                          struct btree_iter *bucket_gens_iter)
 972 {
 973         struct bch_fs *c = trans->c;
 974         struct bch_dev *ca;
 975         struct bch_alloc_v4 a_convert;
 976         const struct bch_alloc_v4 *a;
 977         unsigned discard_key_type, freespace_key_type;
 978         unsigned gens_offset;
 979         struct bkey_s_c k;
 980         struct printbuf buf = PRINTBUF;
 981         int ret;
 982
 983         if (fsck_err_on(!bch2_dev_bucket_exists(c, alloc_k.k->p), c,
 984                         alloc_key_to_missing_dev_bucket,
 985                         "alloc key for invalid device:bucket %llu:%llu",
 986                         alloc_k.k->p.inode, alloc_k.k->p.offset))
 987                 return bch2_btree_delete_at(trans, alloc_iter, 0);
 988
 989         ca = bch_dev_bkey_exists(c, alloc_k.k->p.inode);
 990         if (!ca->mi.freespace_initialized)
 991                 return 0;
 992
 993         a = bch2_alloc_to_v4(alloc_k, &a_convert);
 994
 995         discard_key_type = a->data_type == BCH_DATA_need_discard ? KEY_TYPE_set : 0;
 996         bch2_btree_iter_set_pos(discard_iter, alloc_k.k->p);
 997         k = bch2_btree_iter_peek_slot(discard_iter);
 998         ret = bkey_err(k);
 999         if (ret)
1000                 goto err;
1001
1002         if (k.k->type != discard_key_type &&
1003             (c->opts.reconstruct_alloc ||
1004              fsck_err(c, need_discard_key_wrong,
1005                       "incorrect key in need_discard btree (got %s should be %s)\n"
1006                       "  %s",
1007                       bch2_bkey_types[k.k->type],
1008                       bch2_bkey_types[discard_key_type],
1009                       (bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf)))) {
1010                 struct bkey_i *update =
1011                         bch2_trans_kmalloc(trans, sizeof(*update));
1012
1013                 ret = PTR_ERR_OR_ZERO(update);
1014                 if (ret)
1015                         goto err;
1016
1017                 bkey_init(&update->k);
1018                 update->k.type  = discard_key_type;
1019                 update->k.p     = discard_iter->pos;
1020
1021                 ret = bch2_trans_update(trans, discard_iter, update, 0);
1022                 if (ret)
1023                         goto err;
1024         }
1025
1026         freespace_key_type = a->data_type == BCH_DATA_free ? KEY_TYPE_set : 0;
1027         bch2_btree_iter_set_pos(freespace_iter, alloc_freespace_pos(alloc_k.k->p, *a));
1028         k = bch2_btree_iter_peek_slot(freespace_iter);
1029         ret = bkey_err(k);
1030         if (ret)
1031                 goto err;
1032
1033         if (k.k->type != freespace_key_type &&
1034             (c->opts.reconstruct_alloc ||
1035              fsck_err(c, freespace_key_wrong,
1036                       "incorrect key in freespace btree (got %s should be %s)\n"
1037                       "  %s",
1038                       bch2_bkey_types[k.k->type],
1039                       bch2_bkey_types[freespace_key_type],
1040                       (printbuf_reset(&buf),
1041                        bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf)))) {
1042                 struct bkey_i *update =
1043                         bch2_trans_kmalloc(trans, sizeof(*update));
1044
1045                 ret = PTR_ERR_OR_ZERO(update);
1046                 if (ret)
1047                         goto err;
1048
1049                 bkey_init(&update->k);
1050                 update->k.type  = freespace_key_type;
1051                 update->k.p     = freespace_iter->pos;
1052                 bch2_key_resize(&update->k, 1);
1053
1054                 ret = bch2_trans_update(trans, freespace_iter, update, 0);
1055                 if (ret)
1056                         goto err;
1057         }
1058
1059         bch2_btree_iter_set_pos(bucket_gens_iter, alloc_gens_pos(alloc_k.k->p, &gens_offset));
1060         k = bch2_btree_iter_peek_slot(bucket_gens_iter);
1061         ret = bkey_err(k);
1062         if (ret)
1063                 goto err;
1064
1065         if (a->gen != alloc_gen(k, gens_offset) &&
1066             (c->opts.reconstruct_alloc ||
1067              fsck_err(c, bucket_gens_key_wrong,
1068                       "incorrect gen in bucket_gens btree (got %u should be %u)\n"
1069                       "  %s",
1070                       alloc_gen(k, gens_offset), a->gen,
1071                       (printbuf_reset(&buf),
1072                        bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf)))) {
1073                 struct bkey_i_bucket_gens *g =
1074                         bch2_trans_kmalloc(trans, sizeof(*g));
1075
1076                 ret = PTR_ERR_OR_ZERO(g);
1077                 if (ret)
1078                         goto err;
1079
1080                 if (k.k->type == KEY_TYPE_bucket_gens) {
1081                         bkey_reassemble(&g->k_i, k);
1082                 } else {
1083                         bkey_bucket_gens_init(&g->k_i);
1084                         g->k.p = alloc_gens_pos(alloc_k.k->p, &gens_offset);
1085                 }
1086
1087                 g->v.gens[gens_offset] = a->gen;
1088
1089                 ret = bch2_trans_update(trans, bucket_gens_iter, &g->k_i, 0);
1090                 if (ret)
1091                         goto err;
1092         }
1093 err:
1094 fsck_err:
1095         printbuf_exit(&buf);
1096         return ret;
1097 }
1098
1099 static noinline_for_stack
1100 int bch2_check_alloc_hole_freespace(struct btree_trans *trans,
1101                                     struct bpos start,
1102                                     struct bpos *end,
1103                                     struct btree_iter *freespace_iter)
1104 {
1105         struct bch_fs *c = trans->c;
1106         struct bch_dev *ca;
1107         struct bkey_s_c k;
1108         struct printbuf buf = PRINTBUF;
1109         int ret;
1110
1111         ca = bch_dev_bkey_exists(c, start.inode);
1112         if (!ca->mi.freespace_initialized)
1113                 return 0;
1114
1115         bch2_btree_iter_set_pos(freespace_iter, start);
1116
1117         k = bch2_btree_iter_peek_slot(freespace_iter);
1118         ret = bkey_err(k);
1119         if (ret)
1120                 goto err;
1121
1122         *end = bkey_min(k.k->p, *end);
1123
1124         if (k.k->type != KEY_TYPE_set &&
1125             (c->opts.reconstruct_alloc ||
1126              fsck_err(c, freespace_hole_missing,
1127                       "hole in alloc btree missing in freespace btree\n"
1128                       "  device %llu buckets %llu-%llu",
1129                       freespace_iter->pos.inode,
1130                       freespace_iter->pos.offset,
1131                       end->offset))) {
1132                 struct bkey_i *update =
1133                         bch2_trans_kmalloc(trans, sizeof(*update));
1134
1135                 ret = PTR_ERR_OR_ZERO(update);
1136                 if (ret)
1137                         goto err;
1138
1139                 bkey_init(&update->k);
1140                 update->k.type  = KEY_TYPE_set;
1141                 update->k.p     = freespace_iter->pos;
1142                 bch2_key_resize(&update->k,
1143                                 min_t(u64, U32_MAX, end->offset -
1144                                       freespace_iter->pos.offset));
1145
1146                 ret = bch2_trans_update(trans, freespace_iter, update, 0);
1147                 if (ret)
1148                         goto err;
1149         }
1150 err:
1151 fsck_err:
1152         printbuf_exit(&buf);
1153         return ret;
1154 }
1155
1156 static noinline_for_stack
1157 int bch2_check_alloc_hole_bucket_gens(struct btree_trans *trans,
1158                                       struct bpos start,
1159                                       struct bpos *end,
1160                                       struct btree_iter *bucket_gens_iter)
1161 {
1162         struct bch_fs *c = trans->c;
1163         struct bkey_s_c k;
1164         struct printbuf buf = PRINTBUF;
1165         unsigned i, gens_offset, gens_end_offset;
1166         int ret;
1167
1168         if (c->sb.version < bcachefs_metadata_version_bucket_gens)
1169                 return 0;
1170
1171         bch2_btree_iter_set_pos(bucket_gens_iter, alloc_gens_pos(start, &gens_offset));
1172
1173         k = bch2_btree_iter_peek_slot(bucket_gens_iter);
1174         ret = bkey_err(k);
1175         if (ret)
1176                 goto err;
1177
1178         if (bkey_cmp(alloc_gens_pos(start, &gens_offset),
1179                      alloc_gens_pos(*end,  &gens_end_offset)))
1180                 gens_end_offset = KEY_TYPE_BUCKET_GENS_NR;
1181
1182         if (k.k->type == KEY_TYPE_bucket_gens) {
1183                 struct bkey_i_bucket_gens g;
1184                 bool need_update = false;
1185
1186                 bkey_reassemble(&g.k_i, k);
1187
1188                 for (i = gens_offset; i < gens_end_offset; i++) {
1189                         if (fsck_err_on(g.v.gens[i], c,
1190                                         bucket_gens_hole_wrong,
1191                                         "hole in alloc btree at %llu:%llu with nonzero gen in bucket_gens btree (%u)",
1192                                         bucket_gens_pos_to_alloc(k.k->p, i).inode,
1193                                         bucket_gens_pos_to_alloc(k.k->p, i).offset,
1194                                         g.v.gens[i])) {
1195                                 g.v.gens[i] = 0;
1196                                 need_update = true;
1197                         }
1198                 }
1199
1200                 if (need_update) {
1201                         struct bkey_i *u = bch2_trans_kmalloc(trans, sizeof(g));
1202
1203                         ret = PTR_ERR_OR_ZERO(u);
1204                         if (ret)
1205                                 goto err;
1206
1207                         memcpy(u, &g, sizeof(g));
1208
1209                         ret = bch2_trans_update(trans, bucket_gens_iter, u, 0);
1210                         if (ret)
1211                                 goto err;
1212                 }
1213         }
1214
1215         *end = bkey_min(*end, bucket_gens_pos_to_alloc(bpos_nosnap_successor(k.k->p), 0));
1216 err:
1217 fsck_err:
1218         printbuf_exit(&buf);
1219         return ret;
1220 }
1221
1222 static noinline_for_stack int bch2_check_discard_freespace_key(struct btree_trans *trans,
1223                                               struct btree_iter *iter)
1224 {
1225         struct bch_fs *c = trans->c;
1226         struct btree_iter alloc_iter;
1227         struct bkey_s_c alloc_k;
1228         struct bch_alloc_v4 a_convert;
1229         const struct bch_alloc_v4 *a;
1230         u64 genbits;
1231         struct bpos pos;
1232         enum bch_data_type state = iter->btree_id == BTREE_ID_need_discard
1233                 ? BCH_DATA_need_discard
1234                 : BCH_DATA_free;
1235         struct printbuf buf = PRINTBUF;
1236         int ret;
1237
1238         pos = iter->pos;
1239         pos.offset &= ~(~0ULL << 56);
1240         genbits = iter->pos.offset & (~0ULL << 56);
1241
1242         alloc_k = bch2_bkey_get_iter(trans, &alloc_iter, BTREE_ID_alloc, pos, 0);
1243         ret = bkey_err(alloc_k);
1244         if (ret)
1245                 return ret;
1246
1247         if (fsck_err_on(!bch2_dev_bucket_exists(c, pos), c,
1248                         need_discard_freespace_key_to_invalid_dev_bucket,
1249                         "entry in %s btree for nonexistant dev:bucket %llu:%llu",
1250                         bch2_btree_id_str(iter->btree_id), pos.inode, pos.offset))
1251                 goto delete;
1252
1253         a = bch2_alloc_to_v4(alloc_k, &a_convert);
1254
1255         if (fsck_err_on(a->data_type != state ||
1256                         (state == BCH_DATA_free &&
1257                          genbits != alloc_freespace_genbits(*a)), c,
1258                         need_discard_freespace_key_bad,
1259                         "%s\n  incorrectly set at %s:%llu:%llu:0 (free %u, genbits %llu should be %llu)",
1260                         (bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf),
1261                         bch2_btree_id_str(iter->btree_id),
1262                         iter->pos.inode,
1263                         iter->pos.offset,
1264                         a->data_type == state,
1265                         genbits >> 56, alloc_freespace_genbits(*a) >> 56))
1266                 goto delete;
1267 out:
1268 fsck_err:
1269         set_btree_iter_dontneed(&alloc_iter);
1270         bch2_trans_iter_exit(trans, &alloc_iter);
1271         printbuf_exit(&buf);
1272         return ret;
1273 delete:
1274         ret =   bch2_btree_delete_extent_at(trans, iter,
1275                         iter->btree_id == BTREE_ID_freespace ? 1 : 0, 0) ?:
1276                 bch2_trans_commit(trans, NULL, NULL,
1277                         BCH_TRANS_COMMIT_no_enospc);
1278         goto out;
1279 }
1280
1281 /*
1282  * We've already checked that generation numbers in the bucket_gens btree are
1283  * valid for buckets that exist; this just checks for keys for nonexistent
1284  * buckets.
1285  */
1286 static noinline_for_stack
1287 int bch2_check_bucket_gens_key(struct btree_trans *trans,
1288                                struct btree_iter *iter,
1289                                struct bkey_s_c k)
1290 {
1291         struct bch_fs *c = trans->c;
1292         struct bkey_i_bucket_gens g;
1293         struct bch_dev *ca;
1294         u64 start = bucket_gens_pos_to_alloc(k.k->p, 0).offset;
1295         u64 end = bucket_gens_pos_to_alloc(bpos_nosnap_successor(k.k->p), 0).offset;
1296         u64 b;
1297         bool need_update = false, dev_exists;
1298         struct printbuf buf = PRINTBUF;
1299         int ret = 0;
1300
1301         BUG_ON(k.k->type != KEY_TYPE_bucket_gens);
1302         bkey_reassemble(&g.k_i, k);
1303
1304         /* if no bch_dev, skip out whether we repair or not */
1305         dev_exists = bch2_dev_exists2(c, k.k->p.inode);
1306         if (!dev_exists) {
1307                 if (fsck_err_on(!dev_exists, c,
1308                                 bucket_gens_to_invalid_dev,
1309                                 "bucket_gens key for invalid device:\n  %s",
1310                                 (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1311                         ret = bch2_btree_delete_at(trans, iter, 0);
1312                 }
1313                 goto out;
1314         }
1315
1316         ca = bch_dev_bkey_exists(c, k.k->p.inode);
1317         if (fsck_err_on(end <= ca->mi.first_bucket ||
1318                         start >= ca->mi.nbuckets, c,
1319                         bucket_gens_to_invalid_buckets,
1320                         "bucket_gens key for invalid buckets:\n  %s",
1321                         (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1322                 ret = bch2_btree_delete_at(trans, iter, 0);
1323                 goto out;
1324         }
1325
1326         for (b = start; b < ca->mi.first_bucket; b++)
1327                 if (fsck_err_on(g.v.gens[b & KEY_TYPE_BUCKET_GENS_MASK], c,
1328                                 bucket_gens_nonzero_for_invalid_buckets,
1329                                 "bucket_gens key has nonzero gen for invalid bucket")) {
1330                         g.v.gens[b & KEY_TYPE_BUCKET_GENS_MASK] = 0;
1331                         need_update = true;
1332                 }
1333
1334         for (b = ca->mi.nbuckets; b < end; b++)
1335                 if (fsck_err_on(g.v.gens[b & KEY_TYPE_BUCKET_GENS_MASK], c,
1336                                 bucket_gens_nonzero_for_invalid_buckets,
1337                                 "bucket_gens key has nonzero gen for invalid bucket")) {
1338                         g.v.gens[b & KEY_TYPE_BUCKET_GENS_MASK] = 0;
1339                         need_update = true;
1340                 }
1341
1342         if (need_update) {
1343                 struct bkey_i *u = bch2_trans_kmalloc(trans, sizeof(g));
1344
1345                 ret = PTR_ERR_OR_ZERO(u);
1346                 if (ret)
1347                         goto out;
1348
1349                 memcpy(u, &g, sizeof(g));
1350                 ret = bch2_trans_update(trans, iter, u, 0);
1351         }
1352 out:
1353 fsck_err:
1354         printbuf_exit(&buf);
1355         return ret;
1356 }
1357
1358 int bch2_check_alloc_info(struct bch_fs *c)
1359 {
1360         struct btree_trans *trans = bch2_trans_get(c);
1361         struct btree_iter iter, discard_iter, freespace_iter, bucket_gens_iter;
1362         struct bkey hole;
1363         struct bkey_s_c k;
1364         int ret = 0;
1365
1366         bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc, POS_MIN,
1367                              BTREE_ITER_PREFETCH);
1368         bch2_trans_iter_init(trans, &discard_iter, BTREE_ID_need_discard, POS_MIN,
1369                              BTREE_ITER_PREFETCH);
1370         bch2_trans_iter_init(trans, &freespace_iter, BTREE_ID_freespace, POS_MIN,
1371                              BTREE_ITER_PREFETCH);
1372         bch2_trans_iter_init(trans, &bucket_gens_iter, BTREE_ID_bucket_gens, POS_MIN,
1373                              BTREE_ITER_PREFETCH);
1374
1375         while (1) {
1376                 struct bpos next;
1377
1378                 bch2_trans_begin(trans);
1379
1380                 k = bch2_get_key_or_real_bucket_hole(&iter, &hole);
1381                 ret = bkey_err(k);
1382                 if (ret)
1383                         goto bkey_err;
1384
1385                 if (!k.k)
1386                         break;
1387
1388                 if (k.k->type) {
1389                         next = bpos_nosnap_successor(k.k->p);
1390
1391                         ret = bch2_check_alloc_key(trans,
1392                                                    k, &iter,
1393                                                    &discard_iter,
1394                                                    &freespace_iter,
1395                                                    &bucket_gens_iter);
1396                         if (ret)
1397                                 goto bkey_err;
1398                 } else {
1399                         next = k.k->p;
1400
1401                         ret = bch2_check_alloc_hole_freespace(trans,
1402                                                     bkey_start_pos(k.k),
1403                                                     &next,
1404                                                     &freespace_iter) ?:
1405                                 bch2_check_alloc_hole_bucket_gens(trans,
1406                                                     bkey_start_pos(k.k),
1407                                                     &next,
1408                                                     &bucket_gens_iter);
1409                         if (ret)
1410                                 goto bkey_err;
1411                 }
1412
1413                 ret = bch2_trans_commit(trans, NULL, NULL,
1414                                         BCH_TRANS_COMMIT_no_enospc);
1415                 if (ret)
1416                         goto bkey_err;
1417
1418                 bch2_btree_iter_set_pos(&iter, next);
1419 bkey_err:
1420                 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
1421                         continue;
1422                 if (ret)
1423                         break;
1424         }
1425         bch2_trans_iter_exit(trans, &bucket_gens_iter);
1426         bch2_trans_iter_exit(trans, &freespace_iter);
1427         bch2_trans_iter_exit(trans, &discard_iter);
1428         bch2_trans_iter_exit(trans, &iter);
1429
1430         if (ret < 0)
1431                 goto err;
1432
1433         ret = for_each_btree_key2(trans, iter,
1434                         BTREE_ID_need_discard, POS_MIN,
1435                         BTREE_ITER_PREFETCH, k,
1436                 bch2_check_discard_freespace_key(trans, &iter));
1437         if (ret)
1438                 goto err;
1439
1440         bch2_trans_iter_init(trans, &iter, BTREE_ID_freespace, POS_MIN,
1441                              BTREE_ITER_PREFETCH);
1442         while (1) {
1443                 bch2_trans_begin(trans);
1444                 k = bch2_btree_iter_peek(&iter);
1445                 if (!k.k)
1446                         break;
1447
1448                 ret = bkey_err(k) ?:
1449                         bch2_check_discard_freespace_key(trans, &iter);
1450                 if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) {
1451                         ret = 0;
1452                         continue;
1453                 }
1454                 if (ret) {
1455                         struct printbuf buf = PRINTBUF;
1456                         bch2_bkey_val_to_text(&buf, c, k);
1457
1458                         bch_err(c, "while checking %s", buf.buf);
1459                         printbuf_exit(&buf);
1460                         break;
1461                 }
1462
1463                 bch2_btree_iter_set_pos(&iter, bpos_nosnap_successor(iter.pos));
1464         }
1465         bch2_trans_iter_exit(trans, &iter);
1466         if (ret)
1467                 goto err;
1468
1469         ret = for_each_btree_key_commit(trans, iter,
1470                         BTREE_ID_bucket_gens, POS_MIN,
1471                         BTREE_ITER_PREFETCH, k,
1472                         NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
1473                 bch2_check_bucket_gens_key(trans, &iter, k));
1474 err:
1475         bch2_trans_put(trans);
1476         if (ret)
1477                 bch_err_fn(c, ret);
1478         return ret;
1479 }
1480
1481 static int bch2_check_alloc_to_lru_ref(struct btree_trans *trans,
1482                                        struct btree_iter *alloc_iter)
1483 {
1484         struct bch_fs *c = trans->c;
1485         struct btree_iter lru_iter;
1486         struct bch_alloc_v4 a_convert;
1487         const struct bch_alloc_v4 *a;
1488         struct bkey_s_c alloc_k, lru_k;
1489         struct printbuf buf = PRINTBUF;
1490         int ret;
1491
1492         alloc_k = bch2_btree_iter_peek(alloc_iter);
1493         if (!alloc_k.k)
1494                 return 0;
1495
1496         ret = bkey_err(alloc_k);
1497         if (ret)
1498                 return ret;
1499
1500         a = bch2_alloc_to_v4(alloc_k, &a_convert);
1501
1502         if (a->data_type != BCH_DATA_cached)
1503                 return 0;
1504
1505         if (fsck_err_on(!a->io_time[READ], c,
1506                         alloc_key_cached_but_read_time_zero,
1507                         "cached bucket with read_time 0\n"
1508                         "  %s",
1509                 (printbuf_reset(&buf),
1510                  bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf))) {
1511                 struct bkey_i_alloc_v4 *a_mut =
1512                         bch2_alloc_to_v4_mut(trans, alloc_k);
1513                 ret = PTR_ERR_OR_ZERO(a_mut);
1514                 if (ret)
1515                         goto err;
1516
1517                 a_mut->v.io_time[READ] = atomic64_read(&c->io_clock[READ].now);
1518                 ret = bch2_trans_update(trans, alloc_iter,
1519                                         &a_mut->k_i, BTREE_TRIGGER_NORUN);
1520                 if (ret)
1521                         goto err;
1522
1523                 a = &a_mut->v;
1524         }
1525
1526         lru_k = bch2_bkey_get_iter(trans, &lru_iter, BTREE_ID_lru,
1527                              lru_pos(alloc_k.k->p.inode,
1528                                      bucket_to_u64(alloc_k.k->p),
1529                                      a->io_time[READ]), 0);
1530         ret = bkey_err(lru_k);
1531         if (ret)
1532                 return ret;
1533
1534         if (fsck_err_on(lru_k.k->type != KEY_TYPE_set, c,
1535                         alloc_key_to_missing_lru_entry,
1536                         "missing lru entry\n"
1537                         "  %s",
1538                         (printbuf_reset(&buf),
1539                          bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf))) {
1540                 ret = bch2_lru_set(trans,
1541                                    alloc_k.k->p.inode,
1542                                    bucket_to_u64(alloc_k.k->p),
1543                                    a->io_time[READ]);
1544                 if (ret)
1545                         goto err;
1546         }
1547 err:
1548 fsck_err:
1549         bch2_trans_iter_exit(trans, &lru_iter);
1550         printbuf_exit(&buf);
1551         return ret;
1552 }
1553
1554 int bch2_check_alloc_to_lru_refs(struct bch_fs *c)
1555 {
1556         struct btree_iter iter;
1557         struct bkey_s_c k;
1558
1559         int ret = bch2_trans_run(c,
1560                 for_each_btree_key_commit(trans, iter, BTREE_ID_alloc,
1561                                 POS_MIN, BTREE_ITER_PREFETCH, k,
1562                                 NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
1563                         bch2_check_alloc_to_lru_ref(trans, &iter)));
1564         bch_err_fn(c, ret);
1565         return ret;
1566 }
1567
1568 static int bch2_discard_one_bucket(struct btree_trans *trans,
1569                                    struct btree_iter *need_discard_iter,
1570                                    struct bpos *discard_pos_done,
1571                                    u64 *seen,
1572                                    u64 *open,
1573                                    u64 *need_journal_commit,
1574                                    u64 *discarded)
1575 {
1576         struct bch_fs *c = trans->c;
1577         struct bpos pos = need_discard_iter->pos;
1578         struct btree_iter iter = { NULL };
1579         struct bkey_s_c k;
1580         struct bch_dev *ca;
1581         struct bkey_i_alloc_v4 *a;
1582         struct printbuf buf = PRINTBUF;
1583         int ret = 0;
1584
1585         ca = bch_dev_bkey_exists(c, pos.inode);
1586         if (!percpu_ref_tryget(&ca->io_ref)) {
1587                 bch2_btree_iter_set_pos(need_discard_iter, POS(pos.inode + 1, 0));
1588                 return 0;
1589         }
1590
1591         if (bch2_bucket_is_open_safe(c, pos.inode, pos.offset)) {
1592                 (*open)++;
1593                 goto out;
1594         }
1595
1596         if (bch2_bucket_needs_journal_commit(&c->buckets_waiting_for_journal,
1597                         c->journal.flushed_seq_ondisk,
1598                         pos.inode, pos.offset)) {
1599                 (*need_journal_commit)++;
1600                 goto out;
1601         }
1602
1603         k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_alloc,
1604                                need_discard_iter->pos,
1605                                BTREE_ITER_CACHED);
1606         ret = bkey_err(k);
1607         if (ret)
1608                 goto out;
1609
1610         a = bch2_alloc_to_v4_mut(trans, k);
1611         ret = PTR_ERR_OR_ZERO(a);
1612         if (ret)
1613                 goto out;
1614
1615         if (BCH_ALLOC_V4_NEED_INC_GEN(&a->v)) {
1616                 a->v.gen++;
1617                 SET_BCH_ALLOC_V4_NEED_INC_GEN(&a->v, false);
1618                 goto write;
1619         }
1620
1621         if (a->v.journal_seq > c->journal.flushed_seq_ondisk) {
1622                 if (c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_info) {
1623                         bch2_trans_inconsistent(trans,
1624                                 "clearing need_discard but journal_seq %llu > flushed_seq %llu\n"
1625                                 "%s",
1626                                 a->v.journal_seq,
1627                                 c->journal.flushed_seq_ondisk,
1628                                 (bch2_bkey_val_to_text(&buf, c, k), buf.buf));
1629                         ret = -EIO;
1630                 }
1631                 goto out;
1632         }
1633
1634         if (a->v.data_type != BCH_DATA_need_discard) {
1635                 if (c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_info) {
1636                         bch2_trans_inconsistent(trans,
1637                                 "bucket incorrectly set in need_discard btree\n"
1638                                 "%s",
1639                                 (bch2_bkey_val_to_text(&buf, c, k), buf.buf));
1640                         ret = -EIO;
1641                 }
1642
1643                 goto out;
1644         }
1645
1646         if (!bkey_eq(*discard_pos_done, iter.pos) &&
1647             ca->mi.discard && !c->opts.nochanges) {
1648                 /*
1649                  * This works without any other locks because this is the only
1650                  * thread that removes items from the need_discard tree
1651                  */
1652                 bch2_trans_unlock(trans);
1653                 blkdev_issue_discard(ca->disk_sb.bdev,
1654                                      k.k->p.offset * ca->mi.bucket_size,
1655                                      ca->mi.bucket_size,
1656                                      GFP_KERNEL);
1657                 *discard_pos_done = iter.pos;
1658
1659                 ret = bch2_trans_relock_notrace(trans);
1660                 if (ret)
1661                         goto out;
1662         }
1663
1664         SET_BCH_ALLOC_V4_NEED_DISCARD(&a->v, false);
1665         a->v.data_type = alloc_data_type(a->v, a->v.data_type);
1666 write:
1667         ret =   bch2_trans_update(trans, &iter, &a->k_i, 0) ?:
1668                 bch2_trans_commit(trans, NULL, NULL,
1669                                   BCH_WATERMARK_btree|
1670                                   BCH_TRANS_COMMIT_no_enospc);
1671         if (ret)
1672                 goto out;
1673
1674         count_event(c, bucket_discard);
1675         (*discarded)++;
1676 out:
1677         (*seen)++;
1678         bch2_trans_iter_exit(trans, &iter);
1679         percpu_ref_put(&ca->io_ref);
1680         printbuf_exit(&buf);
1681         return ret;
1682 }
1683
1684 static void bch2_do_discards_work(struct work_struct *work)
1685 {
1686         struct bch_fs *c = container_of(work, struct bch_fs, discard_work);
1687         struct btree_iter iter;
1688         struct bkey_s_c k;
1689         u64 seen = 0, open = 0, need_journal_commit = 0, discarded = 0;
1690         struct bpos discard_pos_done = POS_MAX;
1691         int ret;
1692
1693         /*
1694          * We're doing the commit in bch2_discard_one_bucket instead of using
1695          * for_each_btree_key_commit() so that we can increment counters after
1696          * successful commit:
1697          */
1698         ret = bch2_trans_run(c,
1699                 for_each_btree_key2(trans, iter,
1700                                 BTREE_ID_need_discard, POS_MIN, 0, k,
1701                         bch2_discard_one_bucket(trans, &iter, &discard_pos_done,
1702                                                 &seen,
1703                                                 &open,
1704                                                 &need_journal_commit,
1705                                                 &discarded)));
1706
1707         if (need_journal_commit * 2 > seen)
1708                 bch2_journal_flush_async(&c->journal, NULL);
1709
1710         bch2_write_ref_put(c, BCH_WRITE_REF_discard);
1711
1712         trace_discard_buckets(c, seen, open, need_journal_commit, discarded,
1713                               bch2_err_str(ret));
1714 }
1715
1716 void bch2_do_discards(struct bch_fs *c)
1717 {
1718         if (bch2_write_ref_tryget(c, BCH_WRITE_REF_discard) &&
1719             !queue_work(c->write_ref_wq, &c->discard_work))
1720                 bch2_write_ref_put(c, BCH_WRITE_REF_discard);
1721 }
1722
1723 static int invalidate_one_bucket(struct btree_trans *trans,
1724                                  struct btree_iter *lru_iter,
1725                                  struct bkey_s_c lru_k,
1726                                  s64 *nr_to_invalidate)
1727 {
1728         struct bch_fs *c = trans->c;
1729         struct btree_iter alloc_iter = { NULL };
1730         struct bkey_i_alloc_v4 *a = NULL;
1731         struct printbuf buf = PRINTBUF;
1732         struct bpos bucket = u64_to_bucket(lru_k.k->p.offset);
1733         unsigned cached_sectors;
1734         int ret = 0;
1735
1736         if (*nr_to_invalidate <= 0)
1737                 return 1;
1738
1739         if (!bch2_dev_bucket_exists(c, bucket)) {
1740                 prt_str(&buf, "lru entry points to invalid bucket");
1741                 goto err;
1742         }
1743
1744         if (bch2_bucket_is_open_safe(c, bucket.inode, bucket.offset))
1745                 return 0;
1746
1747         a = bch2_trans_start_alloc_update(trans, &alloc_iter, bucket);
1748         ret = PTR_ERR_OR_ZERO(a);
1749         if (ret)
1750                 goto out;
1751
1752         /* We expect harmless races here due to the btree write buffer: */
1753         if (lru_pos_time(lru_iter->pos) != alloc_lru_idx_read(a->v))
1754                 goto out;
1755
1756         BUG_ON(a->v.data_type != BCH_DATA_cached);
1757
1758         if (!a->v.cached_sectors)
1759                 bch_err(c, "invalidating empty bucket, confused");
1760
1761         cached_sectors = a->v.cached_sectors;
1762
1763         SET_BCH_ALLOC_V4_NEED_INC_GEN(&a->v, false);
1764         a->v.gen++;
1765         a->v.data_type          = 0;
1766         a->v.dirty_sectors      = 0;
1767         a->v.cached_sectors     = 0;
1768         a->v.io_time[READ]      = atomic64_read(&c->io_clock[READ].now);
1769         a->v.io_time[WRITE]     = atomic64_read(&c->io_clock[WRITE].now);
1770
1771         ret =   bch2_trans_update(trans, &alloc_iter, &a->k_i,
1772                                 BTREE_TRIGGER_BUCKET_INVALIDATE) ?:
1773                 bch2_trans_commit(trans, NULL, NULL,
1774                                   BCH_WATERMARK_btree|
1775                                   BCH_TRANS_COMMIT_no_enospc);
1776         if (ret)
1777                 goto out;
1778
1779         trace_and_count(c, bucket_invalidate, c, bucket.inode, bucket.offset, cached_sectors);
1780         --*nr_to_invalidate;
1781 out:
1782         bch2_trans_iter_exit(trans, &alloc_iter);
1783         printbuf_exit(&buf);
1784         return ret;
1785 err:
1786         prt_str(&buf, "\n  lru key: ");
1787         bch2_bkey_val_to_text(&buf, c, lru_k);
1788
1789         prt_str(&buf, "\n  lru entry: ");
1790         bch2_lru_pos_to_text(&buf, lru_iter->pos);
1791
1792         prt_str(&buf, "\n  alloc key: ");
1793         if (!a)
1794                 bch2_bpos_to_text(&buf, bucket);
1795         else
1796                 bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&a->k_i));
1797
1798         bch_err(c, "%s", buf.buf);
1799         if (c->curr_recovery_pass > BCH_RECOVERY_PASS_check_lrus) {
1800                 bch2_inconsistent_error(c);
1801                 ret = -EINVAL;
1802         }
1803
1804         goto out;
1805 }
1806
1807 static void bch2_do_invalidates_work(struct work_struct *work)
1808 {
1809         struct bch_fs *c = container_of(work, struct bch_fs, invalidate_work);
1810         struct bch_dev *ca;
1811         struct btree_trans *trans = bch2_trans_get(c);
1812         struct btree_iter iter;
1813         struct bkey_s_c k;
1814         unsigned i;
1815         int ret = 0;
1816
1817         ret = bch2_btree_write_buffer_tryflush(trans);
1818         if (ret)
1819                 goto err;
1820
1821         for_each_member_device(ca, c, i) {
1822                 s64 nr_to_invalidate =
1823                         should_invalidate_buckets(ca, bch2_dev_usage_read(ca));
1824
1825                 ret = for_each_btree_key2_upto(trans, iter, BTREE_ID_lru,
1826                                 lru_pos(ca->dev_idx, 0, 0),
1827                                 lru_pos(ca->dev_idx, U64_MAX, LRU_TIME_MAX),
1828                                 BTREE_ITER_INTENT, k,
1829                         invalidate_one_bucket(trans, &iter, k, &nr_to_invalidate));
1830
1831                 if (ret < 0) {
1832                         percpu_ref_put(&ca->ref);
1833                         break;
1834                 }
1835         }
1836 err:
1837         bch2_trans_put(trans);
1838         bch2_write_ref_put(c, BCH_WRITE_REF_invalidate);
1839 }
1840
1841 void bch2_do_invalidates(struct bch_fs *c)
1842 {
1843         if (bch2_write_ref_tryget(c, BCH_WRITE_REF_invalidate) &&
1844             !queue_work(c->write_ref_wq, &c->invalidate_work))
1845                 bch2_write_ref_put(c, BCH_WRITE_REF_invalidate);
1846 }
1847
1848 int bch2_dev_freespace_init(struct bch_fs *c, struct bch_dev *ca,
1849                             u64 bucket_start, u64 bucket_end)
1850 {
1851         struct btree_trans *trans = bch2_trans_get(c);
1852         struct btree_iter iter;
1853         struct bkey_s_c k;
1854         struct bkey hole;
1855         struct bpos end = POS(ca->dev_idx, bucket_end);
1856         struct bch_member *m;
1857         unsigned long last_updated = jiffies;
1858         int ret;
1859
1860         BUG_ON(bucket_start > bucket_end);
1861         BUG_ON(bucket_end > ca->mi.nbuckets);
1862
1863         bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc,
1864                 POS(ca->dev_idx, max_t(u64, ca->mi.first_bucket, bucket_start)),
1865                 BTREE_ITER_PREFETCH);
1866         /*
1867          * Scan the alloc btree for every bucket on @ca, and add buckets to the
1868          * freespace/need_discard/need_gc_gens btrees as needed:
1869          */
1870         while (1) {
1871                 if (last_updated + HZ * 10 < jiffies) {
1872                         bch_info(ca, "%s: currently at %llu/%llu",
1873                                  __func__, iter.pos.offset, ca->mi.nbuckets);
1874                         last_updated = jiffies;
1875                 }
1876
1877                 bch2_trans_begin(trans);
1878
1879                 if (bkey_ge(iter.pos, end)) {
1880                         ret = 0;
1881                         break;
1882                 }
1883
1884                 k = bch2_get_key_or_hole(&iter, end, &hole);
1885                 ret = bkey_err(k);
1886                 if (ret)
1887                         goto bkey_err;
1888
1889                 if (k.k->type) {
1890                         /*
1891                          * We process live keys in the alloc btree one at a
1892                          * time:
1893                          */
1894                         struct bch_alloc_v4 a_convert;
1895                         const struct bch_alloc_v4 *a = bch2_alloc_to_v4(k, &a_convert);
1896
1897                         ret =   bch2_bucket_do_index(trans, k, a, true) ?:
1898                                 bch2_trans_commit(trans, NULL, NULL,
1899                                                   BCH_TRANS_COMMIT_no_enospc);
1900                         if (ret)
1901                                 goto bkey_err;
1902
1903                         bch2_btree_iter_advance(&iter);
1904                 } else {
1905                         struct bkey_i *freespace;
1906
1907                         freespace = bch2_trans_kmalloc(trans, sizeof(*freespace));
1908                         ret = PTR_ERR_OR_ZERO(freespace);
1909                         if (ret)
1910                                 goto bkey_err;
1911
1912                         bkey_init(&freespace->k);
1913                         freespace->k.type       = KEY_TYPE_set;
1914                         freespace->k.p          = k.k->p;
1915                         freespace->k.size       = k.k->size;
1916
1917                         ret = bch2_btree_insert_trans(trans, BTREE_ID_freespace, freespace, 0) ?:
1918                                 bch2_trans_commit(trans, NULL, NULL,
1919                                                   BCH_TRANS_COMMIT_no_enospc);
1920                         if (ret)
1921                                 goto bkey_err;
1922
1923                         bch2_btree_iter_set_pos(&iter, k.k->p);
1924                 }
1925 bkey_err:
1926                 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
1927                         continue;
1928                 if (ret)
1929                         break;
1930         }
1931
1932         bch2_trans_iter_exit(trans, &iter);
1933         bch2_trans_put(trans);
1934
1935         if (ret < 0) {
1936                 bch_err_msg(ca, ret, "initializing free space");
1937                 return ret;
1938         }
1939
1940         mutex_lock(&c->sb_lock);
1941         m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
1942         SET_BCH_MEMBER_FREESPACE_INITIALIZED(m, true);
1943         mutex_unlock(&c->sb_lock);
1944
1945         return 0;
1946 }
1947
1948 int bch2_fs_freespace_init(struct bch_fs *c)
1949 {
1950         struct bch_dev *ca;
1951         unsigned i;
1952         int ret = 0;
1953         bool doing_init = false;
1954
1955         /*
1956          * We can crash during the device add path, so we need to check this on
1957          * every mount:
1958          */
1959
1960         for_each_member_device(ca, c, i) {
1961                 if (ca->mi.freespace_initialized)
1962                         continue;
1963
1964                 if (!doing_init) {
1965                         bch_info(c, "initializing freespace");
1966                         doing_init = true;
1967                 }
1968
1969                 ret = bch2_dev_freespace_init(c, ca, 0, ca->mi.nbuckets);
1970                 if (ret) {
1971                         percpu_ref_put(&ca->ref);
1972                         bch_err_fn(c, ret);
1973                         return ret;
1974                 }
1975         }
1976
1977         if (doing_init) {
1978                 mutex_lock(&c->sb_lock);
1979                 bch2_write_super(c);
1980                 mutex_unlock(&c->sb_lock);
1981                 bch_verbose(c, "done initializing freespace");
1982         }
1983
1984         return 0;
1985 }
1986
1987 /* Bucket IO clocks: */
1988
1989 int bch2_bucket_io_time_reset(struct btree_trans *trans, unsigned dev,
1990                               size_t bucket_nr, int rw)
1991 {
1992         struct bch_fs *c = trans->c;
1993         struct btree_iter iter;
1994         struct bkey_i_alloc_v4 *a;
1995         u64 now;
1996         int ret = 0;
1997
1998         a = bch2_trans_start_alloc_update(trans, &iter,  POS(dev, bucket_nr));
1999         ret = PTR_ERR_OR_ZERO(a);
2000         if (ret)
2001                 return ret;
2002
2003         now = atomic64_read(&c->io_clock[rw].now);
2004         if (a->v.io_time[rw] == now)
2005                 goto out;
2006
2007         a->v.io_time[rw] = now;
2008
2009         ret   = bch2_trans_update(trans, &iter, &a->k_i, 0) ?:
2010                 bch2_trans_commit(trans, NULL, NULL, 0);
2011 out:
2012         bch2_trans_iter_exit(trans, &iter);
2013         return ret;
2014 }
2015
2016 /* Startup/shutdown (ro/rw): */
2017
2018 void bch2_recalc_capacity(struct bch_fs *c)
2019 {
2020         struct bch_dev *ca;
2021         u64 capacity = 0, reserved_sectors = 0, gc_reserve;
2022         unsigned bucket_size_max = 0;
2023         unsigned long ra_pages = 0;
2024         unsigned i;
2025
2026         lockdep_assert_held(&c->state_lock);
2027
2028         for_each_online_member(ca, c, i) {
2029                 struct backing_dev_info *bdi = ca->disk_sb.bdev->bd_disk->bdi;
2030
2031                 ra_pages += bdi->ra_pages;
2032         }
2033
2034         bch2_set_ra_pages(c, ra_pages);
2035
2036         for_each_rw_member(ca, c, i) {
2037                 u64 dev_reserve = 0;
2038
2039                 /*
2040                  * We need to reserve buckets (from the number
2041                  * of currently available buckets) against
2042                  * foreground writes so that mainly copygc can
2043                  * make forward progress.
2044                  *
2045                  * We need enough to refill the various reserves
2046                  * from scratch - copygc will use its entire
2047                  * reserve all at once, then run against when
2048                  * its reserve is refilled (from the formerly
2049                  * available buckets).
2050                  *
2051                  * This reserve is just used when considering if
2052                  * allocations for foreground writes must wait -
2053                  * not -ENOSPC calculations.
2054                  */
2055
2056                 dev_reserve += ca->nr_btree_reserve * 2;
2057                 dev_reserve += ca->mi.nbuckets >> 6; /* copygc reserve */
2058
2059                 dev_reserve += 1;       /* btree write point */
2060                 dev_reserve += 1;       /* copygc write point */
2061                 dev_reserve += 1;       /* rebalance write point */
2062
2063                 dev_reserve *= ca->mi.bucket_size;
2064
2065                 capacity += bucket_to_sector(ca, ca->mi.nbuckets -
2066                                              ca->mi.first_bucket);
2067
2068                 reserved_sectors += dev_reserve * 2;
2069
2070                 bucket_size_max = max_t(unsigned, bucket_size_max,
2071                                         ca->mi.bucket_size);
2072         }
2073
2074         gc_reserve = c->opts.gc_reserve_bytes
2075                 ? c->opts.gc_reserve_bytes >> 9
2076                 : div64_u64(capacity * c->opts.gc_reserve_percent, 100);
2077
2078         reserved_sectors = max(gc_reserve, reserved_sectors);
2079
2080         reserved_sectors = min(reserved_sectors, capacity);
2081
2082         c->capacity = capacity - reserved_sectors;
2083
2084         c->bucket_size_max = bucket_size_max;
2085
2086         /* Wake up case someone was waiting for buckets */
2087         closure_wake_up(&c->freelist_wait);
2088 }
2089
2090 u64 bch2_min_rw_member_capacity(struct bch_fs *c)
2091 {
2092         struct bch_dev *ca;
2093         unsigned i;
2094         u64 ret = U64_MAX;
2095
2096         for_each_rw_member(ca, c, i)
2097                 ret = min(ret, ca->mi.nbuckets * ca->mi.bucket_size);
2098         return ret;
2099 }
2100
2101 static bool bch2_dev_has_open_write_point(struct bch_fs *c, struct bch_dev *ca)
2102 {
2103         struct open_bucket *ob;
2104         bool ret = false;
2105
2106         for (ob = c->open_buckets;
2107              ob < c->open_buckets + ARRAY_SIZE(c->open_buckets);
2108              ob++) {
2109                 spin_lock(&ob->lock);
2110                 if (ob->valid && !ob->on_partial_list &&
2111                     ob->dev == ca->dev_idx)
2112                         ret = true;
2113                 spin_unlock(&ob->lock);
2114         }
2115
2116         return ret;
2117 }
2118
2119 /* device goes ro: */
2120 void bch2_dev_allocator_remove(struct bch_fs *c, struct bch_dev *ca)
2121 {
2122         unsigned i;
2123
2124         /* First, remove device from allocation groups: */
2125
2126         for (i = 0; i < ARRAY_SIZE(c->rw_devs); i++)
2127                 clear_bit(ca->dev_idx, c->rw_devs[i].d);
2128
2129         /*
2130          * Capacity is calculated based off of devices in allocation groups:
2131          */
2132         bch2_recalc_capacity(c);
2133
2134         bch2_open_buckets_stop(c, ca, false);
2135
2136         /*
2137          * Wake up threads that were blocked on allocation, so they can notice
2138          * the device can no longer be removed and the capacity has changed:
2139          */
2140         closure_wake_up(&c->freelist_wait);
2141
2142         /*
2143          * journal_res_get() can block waiting for free space in the journal -
2144          * it needs to notice there may not be devices to allocate from anymore:
2145          */
2146         wake_up(&c->journal.wait);
2147
2148         /* Now wait for any in flight writes: */
2149
2150         closure_wait_event(&c->open_buckets_wait,
2151                            !bch2_dev_has_open_write_point(c, ca));
2152 }
2153
2154 /* device goes rw: */
2155 void bch2_dev_allocator_add(struct bch_fs *c, struct bch_dev *ca)
2156 {
2157         unsigned i;
2158
2159         for (i = 0; i < ARRAY_SIZE(c->rw_devs); i++)
2160                 if (ca->mi.data_allowed & (1 << i))
2161                         set_bit(ca->dev_idx, c->rw_devs[i].d);
2162 }
2163
2164 void bch2_fs_allocator_background_init(struct bch_fs *c)
2165 {
2166         spin_lock_init(&c->freelist_lock);
2167         INIT_WORK(&c->discard_work, bch2_do_discards_work);
2168         INIT_WORK(&c->invalidate_work, bch2_do_invalidates_work);
2169 }