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         ret = for_each_btree_key(trans, iter, BTREE_ID_alloc, POS_MIN,
 548                                  BTREE_ITER_PREFETCH, k, ({
 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                 0;
 576         }));
 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         bch_err_fn(c, ret);
 586         return ret;
 587 }
 588
 589 int bch2_alloc_read(struct bch_fs *c)
 590 {
 591         struct btree_trans *trans = bch2_trans_get(c);
 592         struct btree_iter iter;
 593         struct bkey_s_c k;
 594         struct bch_dev *ca;
 595         int ret;
 596
 597         down_read(&c->gc_lock);
 598
 599         if (c->sb.version_upgrade_complete >= bcachefs_metadata_version_bucket_gens) {
 600                 const struct bch_bucket_gens *g;
 601                 u64 b;
 602
 603                 ret = for_each_btree_key(trans, iter, BTREE_ID_bucket_gens, POS_MIN,
 604                                          BTREE_ITER_PREFETCH, k, ({
 605                         u64 start = bucket_gens_pos_to_alloc(k.k->p, 0).offset;
 606                         u64 end = bucket_gens_pos_to_alloc(bpos_nosnap_successor(k.k->p), 0).offset;
 607
 608                         if (k.k->type != KEY_TYPE_bucket_gens)
 609                                 continue;
 610
 611                         g = bkey_s_c_to_bucket_gens(k).v;
 612
 613                         /*
 614                          * Not a fsck error because this is checked/repaired by
 615                          * bch2_check_alloc_key() which runs later:
 616                          */
 617                         if (!bch2_dev_exists2(c, k.k->p.inode))
 618                                 continue;
 619
 620                         ca = bch_dev_bkey_exists(c, k.k->p.inode);
 621
 622                         for (b = max_t(u64, ca->mi.first_bucket, start);
 623                              b < min_t(u64, ca->mi.nbuckets, end);
 624                              b++)
 625                                 *bucket_gen(ca, b) = g->gens[b & KEY_TYPE_BUCKET_GENS_MASK];
 626                         0;
 627                 }));
 628         } else {
 629                 struct bch_alloc_v4 a;
 630
 631                 ret = for_each_btree_key(trans, iter, BTREE_ID_alloc, POS_MIN,
 632                                          BTREE_ITER_PREFETCH, k, ({
 633                         /*
 634                          * Not a fsck error because this is checked/repaired by
 635                          * bch2_check_alloc_key() which runs later:
 636                          */
 637                         if (!bch2_dev_bucket_exists(c, k.k->p))
 638                                 continue;
 639
 640                         ca = bch_dev_bkey_exists(c, k.k->p.inode);
 641
 642                         *bucket_gen(ca, k.k->p.offset) = bch2_alloc_to_v4(k, &a)->gen;
 643                         0;
 644                 }));
 645         }
 646
 647         bch2_trans_put(trans);
 648         up_read(&c->gc_lock);
 649
 650         bch_err_fn(c, ret);
 651         return ret;
 652 }
 653
 654 /* Free space/discard btree: */
 655
 656 static int bch2_bucket_do_index(struct btree_trans *trans,
 657                                 struct bkey_s_c alloc_k,
 658                                 const struct bch_alloc_v4 *a,
 659                                 bool set)
 660 {
 661         struct bch_fs *c = trans->c;
 662         struct bch_dev *ca = bch_dev_bkey_exists(c, alloc_k.k->p.inode);
 663         struct btree_iter iter;
 664         struct bkey_s_c old;
 665         struct bkey_i *k;
 666         enum btree_id btree;
 667         enum bch_bkey_type old_type = !set ? KEY_TYPE_set : KEY_TYPE_deleted;
 668         enum bch_bkey_type new_type =  set ? KEY_TYPE_set : KEY_TYPE_deleted;
 669         struct printbuf buf = PRINTBUF;
 670         int ret;
 671
 672         if (a->data_type != BCH_DATA_free &&
 673             a->data_type != BCH_DATA_need_discard)
 674                 return 0;
 675
 676         k = bch2_trans_kmalloc_nomemzero(trans, sizeof(*k));
 677         if (IS_ERR(k))
 678                 return PTR_ERR(k);
 679
 680         bkey_init(&k->k);
 681         k->k.type = new_type;
 682
 683         switch (a->data_type) {
 684         case BCH_DATA_free:
 685                 btree = BTREE_ID_freespace;
 686                 k->k.p = alloc_freespace_pos(alloc_k.k->p, *a);
 687                 bch2_key_resize(&k->k, 1);
 688                 break;
 689         case BCH_DATA_need_discard:
 690                 btree = BTREE_ID_need_discard;
 691                 k->k.p = alloc_k.k->p;
 692                 break;
 693         default:
 694                 return 0;
 695         }
 696
 697         old = bch2_bkey_get_iter(trans, &iter, btree,
 698                              bkey_start_pos(&k->k),
 699                              BTREE_ITER_INTENT);
 700         ret = bkey_err(old);
 701         if (ret)
 702                 return ret;
 703
 704         if (ca->mi.freespace_initialized &&
 705             c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_info &&
 706             bch2_trans_inconsistent_on(old.k->type != old_type, trans,
 707                         "incorrect key when %s %s:%llu:%llu:0 (got %s should be %s)\n"
 708                         "  for %s",
 709                         set ? "setting" : "clearing",
 710                         bch2_btree_id_str(btree),
 711                         iter.pos.inode,
 712                         iter.pos.offset,
 713                         bch2_bkey_types[old.k->type],
 714                         bch2_bkey_types[old_type],
 715                         (bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf))) {
 716                 ret = -EIO;
 717                 goto err;
 718         }
 719
 720         ret = bch2_trans_update(trans, &iter, k, 0);
 721 err:
 722         bch2_trans_iter_exit(trans, &iter);
 723         printbuf_exit(&buf);
 724         return ret;
 725 }
 726
 727 static noinline int bch2_bucket_gen_update(struct btree_trans *trans,
 728                                            struct bpos bucket, u8 gen)
 729 {
 730         struct btree_iter iter;
 731         unsigned offset;
 732         struct bpos pos = alloc_gens_pos(bucket, &offset);
 733         struct bkey_i_bucket_gens *g;
 734         struct bkey_s_c k;
 735         int ret;
 736
 737         g = bch2_trans_kmalloc(trans, sizeof(*g));
 738         ret = PTR_ERR_OR_ZERO(g);
 739         if (ret)
 740                 return ret;
 741
 742         k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_bucket_gens, pos,
 743                                BTREE_ITER_INTENT|
 744                                BTREE_ITER_WITH_UPDATES);
 745         ret = bkey_err(k);
 746         if (ret)
 747                 return ret;
 748
 749         if (k.k->type != KEY_TYPE_bucket_gens) {
 750                 bkey_bucket_gens_init(&g->k_i);
 751                 g->k.p = iter.pos;
 752         } else {
 753                 bkey_reassemble(&g->k_i, k);
 754         }
 755
 756         g->v.gens[offset] = gen;
 757
 758         ret = bch2_trans_update(trans, &iter, &g->k_i, 0);
 759         bch2_trans_iter_exit(trans, &iter);
 760         return ret;
 761 }
 762
 763 int bch2_trans_mark_alloc(struct btree_trans *trans,
 764                           enum btree_id btree_id, unsigned level,
 765                           struct bkey_s_c old, struct bkey_i *new,
 766                           unsigned flags)
 767 {
 768         struct bch_fs *c = trans->c;
 769         struct bch_alloc_v4 old_a_convert, *new_a;
 770         const struct bch_alloc_v4 *old_a;
 771         u64 old_lru, new_lru;
 772         int ret = 0;
 773
 774         /*
 775          * Deletion only happens in the device removal path, with
 776          * BTREE_TRIGGER_NORUN:
 777          */
 778         BUG_ON(new->k.type != KEY_TYPE_alloc_v4);
 779
 780         old_a = bch2_alloc_to_v4(old, &old_a_convert);
 781         new_a = &bkey_i_to_alloc_v4(new)->v;
 782
 783         new_a->data_type = alloc_data_type(*new_a, new_a->data_type);
 784
 785         if (bch2_bucket_sectors(*new_a) > bch2_bucket_sectors(*old_a)) {
 786                 new_a->io_time[READ] = max_t(u64, 1, atomic64_read(&c->io_clock[READ].now));
 787                 new_a->io_time[WRITE]= max_t(u64, 1, atomic64_read(&c->io_clock[WRITE].now));
 788                 SET_BCH_ALLOC_V4_NEED_INC_GEN(new_a, true);
 789                 SET_BCH_ALLOC_V4_NEED_DISCARD(new_a, true);
 790         }
 791
 792         if (data_type_is_empty(new_a->data_type) &&
 793             BCH_ALLOC_V4_NEED_INC_GEN(new_a) &&
 794             !bch2_bucket_is_open_safe(c, new->k.p.inode, new->k.p.offset)) {
 795                 new_a->gen++;
 796                 SET_BCH_ALLOC_V4_NEED_INC_GEN(new_a, false);
 797         }
 798
 799         if (old_a->data_type != new_a->data_type ||
 800             (new_a->data_type == BCH_DATA_free &&
 801              alloc_freespace_genbits(*old_a) != alloc_freespace_genbits(*new_a))) {
 802                 ret =   bch2_bucket_do_index(trans, old, old_a, false) ?:
 803                         bch2_bucket_do_index(trans, bkey_i_to_s_c(new), new_a, true);
 804                 if (ret)
 805                         return ret;
 806         }
 807
 808         if (new_a->data_type == BCH_DATA_cached &&
 809             !new_a->io_time[READ])
 810                 new_a->io_time[READ] = max_t(u64, 1, atomic64_read(&c->io_clock[READ].now));
 811
 812         old_lru = alloc_lru_idx_read(*old_a);
 813         new_lru = alloc_lru_idx_read(*new_a);
 814
 815         if (old_lru != new_lru) {
 816                 ret = bch2_lru_change(trans, new->k.p.inode,
 817                                       bucket_to_u64(new->k.p),
 818                                       old_lru, new_lru);
 819                 if (ret)
 820                         return ret;
 821         }
 822
 823         new_a->fragmentation_lru = alloc_lru_idx_fragmentation(*new_a,
 824                                         bch_dev_bkey_exists(c, new->k.p.inode));
 825
 826         if (old_a->fragmentation_lru != new_a->fragmentation_lru) {
 827                 ret = bch2_lru_change(trans,
 828                                 BCH_LRU_FRAGMENTATION_START,
 829                                 bucket_to_u64(new->k.p),
 830                                 old_a->fragmentation_lru, new_a->fragmentation_lru);
 831                 if (ret)
 832                         return ret;
 833         }
 834
 835         if (old_a->gen != new_a->gen) {
 836                 ret = bch2_bucket_gen_update(trans, new->k.p, new_a->gen);
 837                 if (ret)
 838                         return ret;
 839         }
 840
 841         /*
 842          * need to know if we're getting called from the invalidate path or
 843          * not:
 844          */
 845
 846         if ((flags & BTREE_TRIGGER_BUCKET_INVALIDATE) &&
 847             old_a->cached_sectors) {
 848                 ret = bch2_update_cached_sectors_list(trans, new->k.p.inode,
 849                                                       -((s64) old_a->cached_sectors));
 850                 if (ret)
 851                         return ret;
 852         }
 853
 854         return 0;
 855 }
 856
 857 /*
 858  * This synthesizes deleted extents for holes, similar to BTREE_ITER_SLOTS for
 859  * extents style btrees, but works on non-extents btrees:
 860  */
 861 static struct bkey_s_c bch2_get_key_or_hole(struct btree_iter *iter, struct bpos end, struct bkey *hole)
 862 {
 863         struct bkey_s_c k = bch2_btree_iter_peek_slot(iter);
 864
 865         if (bkey_err(k))
 866                 return k;
 867
 868         if (k.k->type) {
 869                 return k;
 870         } else {
 871                 struct btree_iter iter2;
 872                 struct bpos next;
 873
 874                 bch2_trans_copy_iter(&iter2, iter);
 875
 876                 struct btree_path *path = btree_iter_path(iter->trans, iter);
 877                 if (!bpos_eq(path->l[0].b->key.k.p, SPOS_MAX))
 878                         end = bkey_min(end, bpos_nosnap_successor(path->l[0].b->key.k.p));
 879
 880                 end = bkey_min(end, POS(iter->pos.inode, iter->pos.offset + U32_MAX - 1));
 881
 882                 /*
 883                  * btree node min/max is a closed interval, upto takes a half
 884                  * open interval:
 885                  */
 886                 k = bch2_btree_iter_peek_upto(&iter2, end);
 887                 next = iter2.pos;
 888                 bch2_trans_iter_exit(iter->trans, &iter2);
 889
 890                 BUG_ON(next.offset >= iter->pos.offset + U32_MAX);
 891
 892                 if (bkey_err(k))
 893                         return k;
 894
 895                 bkey_init(hole);
 896                 hole->p = iter->pos;
 897
 898                 bch2_key_resize(hole, next.offset - iter->pos.offset);
 899                 return (struct bkey_s_c) { hole, NULL };
 900         }
 901 }
 902
 903 static bool next_bucket(struct bch_fs *c, struct bpos *bucket)
 904 {
 905         struct bch_dev *ca;
 906         unsigned iter;
 907
 908         if (bch2_dev_bucket_exists(c, *bucket))
 909                 return true;
 910
 911         if (bch2_dev_exists2(c, bucket->inode)) {
 912                 ca = bch_dev_bkey_exists(c, bucket->inode);
 913
 914                 if (bucket->offset < ca->mi.first_bucket) {
 915                         bucket->offset = ca->mi.first_bucket;
 916                         return true;
 917                 }
 918
 919                 bucket->inode++;
 920                 bucket->offset = 0;
 921         }
 922
 923         rcu_read_lock();
 924         iter = bucket->inode;
 925         ca = __bch2_next_dev(c, &iter, NULL);
 926         if (ca)
 927                 *bucket = POS(ca->dev_idx, ca->mi.first_bucket);
 928         rcu_read_unlock();
 929
 930         return ca != NULL;
 931 }
 932
 933 static struct bkey_s_c bch2_get_key_or_real_bucket_hole(struct btree_iter *iter, struct bkey *hole)
 934 {
 935         struct bch_fs *c = iter->trans->c;
 936         struct bkey_s_c k;
 937 again:
 938         k = bch2_get_key_or_hole(iter, POS_MAX, hole);
 939         if (bkey_err(k))
 940                 return k;
 941
 942         if (!k.k->type) {
 943                 struct bpos bucket = bkey_start_pos(k.k);
 944
 945                 if (!bch2_dev_bucket_exists(c, bucket)) {
 946                         if (!next_bucket(c, &bucket))
 947                                 return bkey_s_c_null;
 948
 949                         bch2_btree_iter_set_pos(iter, bucket);
 950                         goto again;
 951                 }
 952
 953                 if (!bch2_dev_bucket_exists(c, k.k->p)) {
 954                         struct bch_dev *ca = bch_dev_bkey_exists(c, bucket.inode);
 955
 956                         bch2_key_resize(hole, ca->mi.nbuckets - bucket.offset);
 957                 }
 958         }
 959
 960         return k;
 961 }
 962
 963 static noinline_for_stack
 964 int bch2_check_alloc_key(struct btree_trans *trans,
 965                          struct bkey_s_c alloc_k,
 966                          struct btree_iter *alloc_iter,
 967                          struct btree_iter *discard_iter,
 968                          struct btree_iter *freespace_iter,
 969                          struct btree_iter *bucket_gens_iter)
 970 {
 971         struct bch_fs *c = trans->c;
 972         struct bch_dev *ca;
 973         struct bch_alloc_v4 a_convert;
 974         const struct bch_alloc_v4 *a;
 975         unsigned discard_key_type, freespace_key_type;
 976         unsigned gens_offset;
 977         struct bkey_s_c k;
 978         struct printbuf buf = PRINTBUF;
 979         int ret;
 980
 981         if (fsck_err_on(!bch2_dev_bucket_exists(c, alloc_k.k->p), c,
 982                         alloc_key_to_missing_dev_bucket,
 983                         "alloc key for invalid device:bucket %llu:%llu",
 984                         alloc_k.k->p.inode, alloc_k.k->p.offset))
 985                 return bch2_btree_delete_at(trans, alloc_iter, 0);
 986
 987         ca = bch_dev_bkey_exists(c, alloc_k.k->p.inode);
 988         if (!ca->mi.freespace_initialized)
 989                 return 0;
 990
 991         a = bch2_alloc_to_v4(alloc_k, &a_convert);
 992
 993         discard_key_type = a->data_type == BCH_DATA_need_discard ? KEY_TYPE_set : 0;
 994         bch2_btree_iter_set_pos(discard_iter, alloc_k.k->p);
 995         k = bch2_btree_iter_peek_slot(discard_iter);
 996         ret = bkey_err(k);
 997         if (ret)
 998                 goto err;
 999
1000         if (k.k->type != discard_key_type &&
1001             (c->opts.reconstruct_alloc ||
1002              fsck_err(c, need_discard_key_wrong,
1003                       "incorrect key in need_discard btree (got %s should be %s)\n"
1004                       "  %s",
1005                       bch2_bkey_types[k.k->type],
1006                       bch2_bkey_types[discard_key_type],
1007                       (bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf)))) {
1008                 struct bkey_i *update =
1009                         bch2_trans_kmalloc(trans, sizeof(*update));
1010
1011                 ret = PTR_ERR_OR_ZERO(update);
1012                 if (ret)
1013                         goto err;
1014
1015                 bkey_init(&update->k);
1016                 update->k.type  = discard_key_type;
1017                 update->k.p     = discard_iter->pos;
1018
1019                 ret = bch2_trans_update(trans, discard_iter, update, 0);
1020                 if (ret)
1021                         goto err;
1022         }
1023
1024         freespace_key_type = a->data_type == BCH_DATA_free ? KEY_TYPE_set : 0;
1025         bch2_btree_iter_set_pos(freespace_iter, alloc_freespace_pos(alloc_k.k->p, *a));
1026         k = bch2_btree_iter_peek_slot(freespace_iter);
1027         ret = bkey_err(k);
1028         if (ret)
1029                 goto err;
1030
1031         if (k.k->type != freespace_key_type &&
1032             (c->opts.reconstruct_alloc ||
1033              fsck_err(c, freespace_key_wrong,
1034                       "incorrect key in freespace btree (got %s should be %s)\n"
1035                       "  %s",
1036                       bch2_bkey_types[k.k->type],
1037                       bch2_bkey_types[freespace_key_type],
1038                       (printbuf_reset(&buf),
1039                        bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf)))) {
1040                 struct bkey_i *update =
1041                         bch2_trans_kmalloc(trans, sizeof(*update));
1042
1043                 ret = PTR_ERR_OR_ZERO(update);
1044                 if (ret)
1045                         goto err;
1046
1047                 bkey_init(&update->k);
1048                 update->k.type  = freespace_key_type;
1049                 update->k.p     = freespace_iter->pos;
1050                 bch2_key_resize(&update->k, 1);
1051
1052                 ret = bch2_trans_update(trans, freespace_iter, update, 0);
1053                 if (ret)
1054                         goto err;
1055         }
1056
1057         bch2_btree_iter_set_pos(bucket_gens_iter, alloc_gens_pos(alloc_k.k->p, &gens_offset));
1058         k = bch2_btree_iter_peek_slot(bucket_gens_iter);
1059         ret = bkey_err(k);
1060         if (ret)
1061                 goto err;
1062
1063         if (a->gen != alloc_gen(k, gens_offset) &&
1064             (c->opts.reconstruct_alloc ||
1065              fsck_err(c, bucket_gens_key_wrong,
1066                       "incorrect gen in bucket_gens btree (got %u should be %u)\n"
1067                       "  %s",
1068                       alloc_gen(k, gens_offset), a->gen,
1069                       (printbuf_reset(&buf),
1070                        bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf)))) {
1071                 struct bkey_i_bucket_gens *g =
1072                         bch2_trans_kmalloc(trans, sizeof(*g));
1073
1074                 ret = PTR_ERR_OR_ZERO(g);
1075                 if (ret)
1076                         goto err;
1077
1078                 if (k.k->type == KEY_TYPE_bucket_gens) {
1079                         bkey_reassemble(&g->k_i, k);
1080                 } else {
1081                         bkey_bucket_gens_init(&g->k_i);
1082                         g->k.p = alloc_gens_pos(alloc_k.k->p, &gens_offset);
1083                 }
1084
1085                 g->v.gens[gens_offset] = a->gen;
1086
1087                 ret = bch2_trans_update(trans, bucket_gens_iter, &g->k_i, 0);
1088                 if (ret)
1089                         goto err;
1090         }
1091 err:
1092 fsck_err:
1093         printbuf_exit(&buf);
1094         return ret;
1095 }
1096
1097 static noinline_for_stack
1098 int bch2_check_alloc_hole_freespace(struct btree_trans *trans,
1099                                     struct bpos start,
1100                                     struct bpos *end,
1101                                     struct btree_iter *freespace_iter)
1102 {
1103         struct bch_fs *c = trans->c;
1104         struct bch_dev *ca;
1105         struct bkey_s_c k;
1106         struct printbuf buf = PRINTBUF;
1107         int ret;
1108
1109         ca = bch_dev_bkey_exists(c, start.inode);
1110         if (!ca->mi.freespace_initialized)
1111                 return 0;
1112
1113         bch2_btree_iter_set_pos(freespace_iter, start);
1114
1115         k = bch2_btree_iter_peek_slot(freespace_iter);
1116         ret = bkey_err(k);
1117         if (ret)
1118                 goto err;
1119
1120         *end = bkey_min(k.k->p, *end);
1121
1122         if (k.k->type != KEY_TYPE_set &&
1123             (c->opts.reconstruct_alloc ||
1124              fsck_err(c, freespace_hole_missing,
1125                       "hole in alloc btree missing in freespace btree\n"
1126                       "  device %llu buckets %llu-%llu",
1127                       freespace_iter->pos.inode,
1128                       freespace_iter->pos.offset,
1129                       end->offset))) {
1130                 struct bkey_i *update =
1131                         bch2_trans_kmalloc(trans, sizeof(*update));
1132
1133                 ret = PTR_ERR_OR_ZERO(update);
1134                 if (ret)
1135                         goto err;
1136
1137                 bkey_init(&update->k);
1138                 update->k.type  = KEY_TYPE_set;
1139                 update->k.p     = freespace_iter->pos;
1140                 bch2_key_resize(&update->k,
1141                                 min_t(u64, U32_MAX, end->offset -
1142                                       freespace_iter->pos.offset));
1143
1144                 ret = bch2_trans_update(trans, freespace_iter, update, 0);
1145                 if (ret)
1146                         goto err;
1147         }
1148 err:
1149 fsck_err:
1150         printbuf_exit(&buf);
1151         return ret;
1152 }
1153
1154 static noinline_for_stack
1155 int bch2_check_alloc_hole_bucket_gens(struct btree_trans *trans,
1156                                       struct bpos start,
1157                                       struct bpos *end,
1158                                       struct btree_iter *bucket_gens_iter)
1159 {
1160         struct bch_fs *c = trans->c;
1161         struct bkey_s_c k;
1162         struct printbuf buf = PRINTBUF;
1163         unsigned i, gens_offset, gens_end_offset;
1164         int ret;
1165
1166         if (c->sb.version < bcachefs_metadata_version_bucket_gens)
1167                 return 0;
1168
1169         bch2_btree_iter_set_pos(bucket_gens_iter, alloc_gens_pos(start, &gens_offset));
1170
1171         k = bch2_btree_iter_peek_slot(bucket_gens_iter);
1172         ret = bkey_err(k);
1173         if (ret)
1174                 goto err;
1175
1176         if (bkey_cmp(alloc_gens_pos(start, &gens_offset),
1177                      alloc_gens_pos(*end,  &gens_end_offset)))
1178                 gens_end_offset = KEY_TYPE_BUCKET_GENS_NR;
1179
1180         if (k.k->type == KEY_TYPE_bucket_gens) {
1181                 struct bkey_i_bucket_gens g;
1182                 bool need_update = false;
1183
1184                 bkey_reassemble(&g.k_i, k);
1185
1186                 for (i = gens_offset; i < gens_end_offset; i++) {
1187                         if (fsck_err_on(g.v.gens[i], c,
1188                                         bucket_gens_hole_wrong,
1189                                         "hole in alloc btree at %llu:%llu with nonzero gen in bucket_gens btree (%u)",
1190                                         bucket_gens_pos_to_alloc(k.k->p, i).inode,
1191                                         bucket_gens_pos_to_alloc(k.k->p, i).offset,
1192                                         g.v.gens[i])) {
1193                                 g.v.gens[i] = 0;
1194                                 need_update = true;
1195                         }
1196                 }
1197
1198                 if (need_update) {
1199                         struct bkey_i *u = bch2_trans_kmalloc(trans, sizeof(g));
1200
1201                         ret = PTR_ERR_OR_ZERO(u);
1202                         if (ret)
1203                                 goto err;
1204
1205                         memcpy(u, &g, sizeof(g));
1206
1207                         ret = bch2_trans_update(trans, bucket_gens_iter, u, 0);
1208                         if (ret)
1209                                 goto err;
1210                 }
1211         }
1212
1213         *end = bkey_min(*end, bucket_gens_pos_to_alloc(bpos_nosnap_successor(k.k->p), 0));
1214 err:
1215 fsck_err:
1216         printbuf_exit(&buf);
1217         return ret;
1218 }
1219
1220 static noinline_for_stack int bch2_check_discard_freespace_key(struct btree_trans *trans,
1221                                               struct btree_iter *iter)
1222 {
1223         struct bch_fs *c = trans->c;
1224         struct btree_iter alloc_iter;
1225         struct bkey_s_c alloc_k;
1226         struct bch_alloc_v4 a_convert;
1227         const struct bch_alloc_v4 *a;
1228         u64 genbits;
1229         struct bpos pos;
1230         enum bch_data_type state = iter->btree_id == BTREE_ID_need_discard
1231                 ? BCH_DATA_need_discard
1232                 : BCH_DATA_free;
1233         struct printbuf buf = PRINTBUF;
1234         int ret;
1235
1236         pos = iter->pos;
1237         pos.offset &= ~(~0ULL << 56);
1238         genbits = iter->pos.offset & (~0ULL << 56);
1239
1240         alloc_k = bch2_bkey_get_iter(trans, &alloc_iter, BTREE_ID_alloc, pos, 0);
1241         ret = bkey_err(alloc_k);
1242         if (ret)
1243                 return ret;
1244
1245         if (fsck_err_on(!bch2_dev_bucket_exists(c, pos), c,
1246                         need_discard_freespace_key_to_invalid_dev_bucket,
1247                         "entry in %s btree for nonexistant dev:bucket %llu:%llu",
1248                         bch2_btree_id_str(iter->btree_id), pos.inode, pos.offset))
1249                 goto delete;
1250
1251         a = bch2_alloc_to_v4(alloc_k, &a_convert);
1252
1253         if (fsck_err_on(a->data_type != state ||
1254                         (state == BCH_DATA_free &&
1255                          genbits != alloc_freespace_genbits(*a)), c,
1256                         need_discard_freespace_key_bad,
1257                         "%s\n  incorrectly set at %s:%llu:%llu:0 (free %u, genbits %llu should be %llu)",
1258                         (bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf),
1259                         bch2_btree_id_str(iter->btree_id),
1260                         iter->pos.inode,
1261                         iter->pos.offset,
1262                         a->data_type == state,
1263                         genbits >> 56, alloc_freespace_genbits(*a) >> 56))
1264                 goto delete;
1265 out:
1266 fsck_err:
1267         set_btree_iter_dontneed(&alloc_iter);
1268         bch2_trans_iter_exit(trans, &alloc_iter);
1269         printbuf_exit(&buf);
1270         return ret;
1271 delete:
1272         ret =   bch2_btree_delete_extent_at(trans, iter,
1273                         iter->btree_id == BTREE_ID_freespace ? 1 : 0, 0) ?:
1274                 bch2_trans_commit(trans, NULL, NULL,
1275                         BCH_TRANS_COMMIT_no_enospc);
1276         goto out;
1277 }
1278
1279 /*
1280  * We've already checked that generation numbers in the bucket_gens btree are
1281  * valid for buckets that exist; this just checks for keys for nonexistent
1282  * buckets.
1283  */
1284 static noinline_for_stack
1285 int bch2_check_bucket_gens_key(struct btree_trans *trans,
1286                                struct btree_iter *iter,
1287                                struct bkey_s_c k)
1288 {
1289         struct bch_fs *c = trans->c;
1290         struct bkey_i_bucket_gens g;
1291         struct bch_dev *ca;
1292         u64 start = bucket_gens_pos_to_alloc(k.k->p, 0).offset;
1293         u64 end = bucket_gens_pos_to_alloc(bpos_nosnap_successor(k.k->p), 0).offset;
1294         u64 b;
1295         bool need_update = false, dev_exists;
1296         struct printbuf buf = PRINTBUF;
1297         int ret = 0;
1298
1299         BUG_ON(k.k->type != KEY_TYPE_bucket_gens);
1300         bkey_reassemble(&g.k_i, k);
1301
1302         /* if no bch_dev, skip out whether we repair or not */
1303         dev_exists = bch2_dev_exists2(c, k.k->p.inode);
1304         if (!dev_exists) {
1305                 if (fsck_err_on(!dev_exists, c,
1306                                 bucket_gens_to_invalid_dev,
1307                                 "bucket_gens key for invalid device:\n  %s",
1308                                 (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1309                         ret = bch2_btree_delete_at(trans, iter, 0);
1310                 }
1311                 goto out;
1312         }
1313
1314         ca = bch_dev_bkey_exists(c, k.k->p.inode);
1315         if (fsck_err_on(end <= ca->mi.first_bucket ||
1316                         start >= ca->mi.nbuckets, c,
1317                         bucket_gens_to_invalid_buckets,
1318                         "bucket_gens key for invalid buckets:\n  %s",
1319                         (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1320                 ret = bch2_btree_delete_at(trans, iter, 0);
1321                 goto out;
1322         }
1323
1324         for (b = start; b < ca->mi.first_bucket; b++)
1325                 if (fsck_err_on(g.v.gens[b & KEY_TYPE_BUCKET_GENS_MASK], c,
1326                                 bucket_gens_nonzero_for_invalid_buckets,
1327                                 "bucket_gens key has nonzero gen for invalid bucket")) {
1328                         g.v.gens[b & KEY_TYPE_BUCKET_GENS_MASK] = 0;
1329                         need_update = true;
1330                 }
1331
1332         for (b = ca->mi.nbuckets; b < end; b++)
1333                 if (fsck_err_on(g.v.gens[b & KEY_TYPE_BUCKET_GENS_MASK], c,
1334                                 bucket_gens_nonzero_for_invalid_buckets,
1335                                 "bucket_gens key has nonzero gen for invalid bucket")) {
1336                         g.v.gens[b & KEY_TYPE_BUCKET_GENS_MASK] = 0;
1337                         need_update = true;
1338                 }
1339
1340         if (need_update) {
1341                 struct bkey_i *u = bch2_trans_kmalloc(trans, sizeof(g));
1342
1343                 ret = PTR_ERR_OR_ZERO(u);
1344                 if (ret)
1345                         goto out;
1346
1347                 memcpy(u, &g, sizeof(g));
1348                 ret = bch2_trans_update(trans, iter, u, 0);
1349         }
1350 out:
1351 fsck_err:
1352         printbuf_exit(&buf);
1353         return ret;
1354 }
1355
1356 int bch2_check_alloc_info(struct bch_fs *c)
1357 {
1358         struct btree_trans *trans = bch2_trans_get(c);
1359         struct btree_iter iter, discard_iter, freespace_iter, bucket_gens_iter;
1360         struct bkey hole;
1361         struct bkey_s_c k;
1362         int ret = 0;
1363
1364         bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc, POS_MIN,
1365                              BTREE_ITER_PREFETCH);
1366         bch2_trans_iter_init(trans, &discard_iter, BTREE_ID_need_discard, POS_MIN,
1367                              BTREE_ITER_PREFETCH);
1368         bch2_trans_iter_init(trans, &freespace_iter, BTREE_ID_freespace, POS_MIN,
1369                              BTREE_ITER_PREFETCH);
1370         bch2_trans_iter_init(trans, &bucket_gens_iter, BTREE_ID_bucket_gens, POS_MIN,
1371                              BTREE_ITER_PREFETCH);
1372
1373         while (1) {
1374                 struct bpos next;
1375
1376                 bch2_trans_begin(trans);
1377
1378                 k = bch2_get_key_or_real_bucket_hole(&iter, &hole);
1379                 ret = bkey_err(k);
1380                 if (ret)
1381                         goto bkey_err;
1382
1383                 if (!k.k)
1384                         break;
1385
1386                 if (k.k->type) {
1387                         next = bpos_nosnap_successor(k.k->p);
1388
1389                         ret = bch2_check_alloc_key(trans,
1390                                                    k, &iter,
1391                                                    &discard_iter,
1392                                                    &freespace_iter,
1393                                                    &bucket_gens_iter);
1394                         if (ret)
1395                                 goto bkey_err;
1396                 } else {
1397                         next = k.k->p;
1398
1399                         ret = bch2_check_alloc_hole_freespace(trans,
1400                                                     bkey_start_pos(k.k),
1401                                                     &next,
1402                                                     &freespace_iter) ?:
1403                                 bch2_check_alloc_hole_bucket_gens(trans,
1404                                                     bkey_start_pos(k.k),
1405                                                     &next,
1406                                                     &bucket_gens_iter);
1407                         if (ret)
1408                                 goto bkey_err;
1409                 }
1410
1411                 ret = bch2_trans_commit(trans, NULL, NULL,
1412                                         BCH_TRANS_COMMIT_no_enospc);
1413                 if (ret)
1414                         goto bkey_err;
1415
1416                 bch2_btree_iter_set_pos(&iter, next);
1417 bkey_err:
1418                 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
1419                         continue;
1420                 if (ret)
1421                         break;
1422         }
1423         bch2_trans_iter_exit(trans, &bucket_gens_iter);
1424         bch2_trans_iter_exit(trans, &freespace_iter);
1425         bch2_trans_iter_exit(trans, &discard_iter);
1426         bch2_trans_iter_exit(trans, &iter);
1427
1428         if (ret < 0)
1429                 goto err;
1430
1431         ret = for_each_btree_key(trans, iter,
1432                         BTREE_ID_need_discard, POS_MIN,
1433                         BTREE_ITER_PREFETCH, k,
1434                 bch2_check_discard_freespace_key(trans, &iter));
1435         if (ret)
1436                 goto err;
1437
1438         bch2_trans_iter_init(trans, &iter, BTREE_ID_freespace, POS_MIN,
1439                              BTREE_ITER_PREFETCH);
1440         while (1) {
1441                 bch2_trans_begin(trans);
1442                 k = bch2_btree_iter_peek(&iter);
1443                 if (!k.k)
1444                         break;
1445
1446                 ret = bkey_err(k) ?:
1447                         bch2_check_discard_freespace_key(trans, &iter);
1448                 if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) {
1449                         ret = 0;
1450                         continue;
1451                 }
1452                 if (ret) {
1453                         struct printbuf buf = PRINTBUF;
1454                         bch2_bkey_val_to_text(&buf, c, k);
1455
1456                         bch_err(c, "while checking %s", buf.buf);
1457                         printbuf_exit(&buf);
1458                         break;
1459                 }
1460
1461                 bch2_btree_iter_set_pos(&iter, bpos_nosnap_successor(iter.pos));
1462         }
1463         bch2_trans_iter_exit(trans, &iter);
1464         if (ret)
1465                 goto err;
1466
1467         ret = for_each_btree_key_commit(trans, iter,
1468                         BTREE_ID_bucket_gens, POS_MIN,
1469                         BTREE_ITER_PREFETCH, k,
1470                         NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
1471                 bch2_check_bucket_gens_key(trans, &iter, k));
1472 err:
1473         bch2_trans_put(trans);
1474         if (ret)
1475                 bch_err_fn(c, ret);
1476         return ret;
1477 }
1478
1479 static int bch2_check_alloc_to_lru_ref(struct btree_trans *trans,
1480                                        struct btree_iter *alloc_iter)
1481 {
1482         struct bch_fs *c = trans->c;
1483         struct btree_iter lru_iter;
1484         struct bch_alloc_v4 a_convert;
1485         const struct bch_alloc_v4 *a;
1486         struct bkey_s_c alloc_k, lru_k;
1487         struct printbuf buf = PRINTBUF;
1488         int ret;
1489
1490         alloc_k = bch2_btree_iter_peek(alloc_iter);
1491         if (!alloc_k.k)
1492                 return 0;
1493
1494         ret = bkey_err(alloc_k);
1495         if (ret)
1496                 return ret;
1497
1498         a = bch2_alloc_to_v4(alloc_k, &a_convert);
1499
1500         if (a->data_type != BCH_DATA_cached)
1501                 return 0;
1502
1503         if (fsck_err_on(!a->io_time[READ], c,
1504                         alloc_key_cached_but_read_time_zero,
1505                         "cached bucket with read_time 0\n"
1506                         "  %s",
1507                 (printbuf_reset(&buf),
1508                  bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf))) {
1509                 struct bkey_i_alloc_v4 *a_mut =
1510                         bch2_alloc_to_v4_mut(trans, alloc_k);
1511                 ret = PTR_ERR_OR_ZERO(a_mut);
1512                 if (ret)
1513                         goto err;
1514
1515                 a_mut->v.io_time[READ] = atomic64_read(&c->io_clock[READ].now);
1516                 ret = bch2_trans_update(trans, alloc_iter,
1517                                         &a_mut->k_i, BTREE_TRIGGER_NORUN);
1518                 if (ret)
1519                         goto err;
1520
1521                 a = &a_mut->v;
1522         }
1523
1524         lru_k = bch2_bkey_get_iter(trans, &lru_iter, BTREE_ID_lru,
1525                              lru_pos(alloc_k.k->p.inode,
1526                                      bucket_to_u64(alloc_k.k->p),
1527                                      a->io_time[READ]), 0);
1528         ret = bkey_err(lru_k);
1529         if (ret)
1530                 return ret;
1531
1532         if (fsck_err_on(lru_k.k->type != KEY_TYPE_set, c,
1533                         alloc_key_to_missing_lru_entry,
1534                         "missing lru entry\n"
1535                         "  %s",
1536                         (printbuf_reset(&buf),
1537                          bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf))) {
1538                 ret = bch2_lru_set(trans,
1539                                    alloc_k.k->p.inode,
1540                                    bucket_to_u64(alloc_k.k->p),
1541                                    a->io_time[READ]);
1542                 if (ret)
1543                         goto err;
1544         }
1545 err:
1546 fsck_err:
1547         bch2_trans_iter_exit(trans, &lru_iter);
1548         printbuf_exit(&buf);
1549         return ret;
1550 }
1551
1552 int bch2_check_alloc_to_lru_refs(struct bch_fs *c)
1553 {
1554         struct btree_iter iter;
1555         struct bkey_s_c k;
1556
1557         int ret = bch2_trans_run(c,
1558                 for_each_btree_key_commit(trans, iter, BTREE_ID_alloc,
1559                                 POS_MIN, BTREE_ITER_PREFETCH, k,
1560                                 NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
1561                         bch2_check_alloc_to_lru_ref(trans, &iter)));
1562         bch_err_fn(c, ret);
1563         return ret;
1564 }
1565
1566 static int bch2_discard_one_bucket(struct btree_trans *trans,
1567                                    struct btree_iter *need_discard_iter,
1568                                    struct bpos *discard_pos_done,
1569                                    u64 *seen,
1570                                    u64 *open,
1571                                    u64 *need_journal_commit,
1572                                    u64 *discarded)
1573 {
1574         struct bch_fs *c = trans->c;
1575         struct bpos pos = need_discard_iter->pos;
1576         struct btree_iter iter = { NULL };
1577         struct bkey_s_c k;
1578         struct bch_dev *ca;
1579         struct bkey_i_alloc_v4 *a;
1580         struct printbuf buf = PRINTBUF;
1581         int ret = 0;
1582
1583         ca = bch_dev_bkey_exists(c, pos.inode);
1584         if (!percpu_ref_tryget(&ca->io_ref)) {
1585                 bch2_btree_iter_set_pos(need_discard_iter, POS(pos.inode + 1, 0));
1586                 return 0;
1587         }
1588
1589         if (bch2_bucket_is_open_safe(c, pos.inode, pos.offset)) {
1590                 (*open)++;
1591                 goto out;
1592         }
1593
1594         if (bch2_bucket_needs_journal_commit(&c->buckets_waiting_for_journal,
1595                         c->journal.flushed_seq_ondisk,
1596                         pos.inode, pos.offset)) {
1597                 (*need_journal_commit)++;
1598                 goto out;
1599         }
1600
1601         k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_alloc,
1602                                need_discard_iter->pos,
1603                                BTREE_ITER_CACHED);
1604         ret = bkey_err(k);
1605         if (ret)
1606                 goto out;
1607
1608         a = bch2_alloc_to_v4_mut(trans, k);
1609         ret = PTR_ERR_OR_ZERO(a);
1610         if (ret)
1611                 goto out;
1612
1613         if (BCH_ALLOC_V4_NEED_INC_GEN(&a->v)) {
1614                 a->v.gen++;
1615                 SET_BCH_ALLOC_V4_NEED_INC_GEN(&a->v, false);
1616                 goto write;
1617         }
1618
1619         if (a->v.journal_seq > c->journal.flushed_seq_ondisk) {
1620                 if (c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_info) {
1621                         bch2_trans_inconsistent(trans,
1622                                 "clearing need_discard but journal_seq %llu > flushed_seq %llu\n"
1623                                 "%s",
1624                                 a->v.journal_seq,
1625                                 c->journal.flushed_seq_ondisk,
1626                                 (bch2_bkey_val_to_text(&buf, c, k), buf.buf));
1627                         ret = -EIO;
1628                 }
1629                 goto out;
1630         }
1631
1632         if (a->v.data_type != BCH_DATA_need_discard) {
1633                 if (c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_info) {
1634                         bch2_trans_inconsistent(trans,
1635                                 "bucket incorrectly set in need_discard btree\n"
1636                                 "%s",
1637                                 (bch2_bkey_val_to_text(&buf, c, k), buf.buf));
1638                         ret = -EIO;
1639                 }
1640
1641                 goto out;
1642         }
1643
1644         if (!bkey_eq(*discard_pos_done, iter.pos) &&
1645             ca->mi.discard && !c->opts.nochanges) {
1646                 /*
1647                  * This works without any other locks because this is the only
1648                  * thread that removes items from the need_discard tree
1649                  */
1650                 bch2_trans_unlock(trans);
1651                 blkdev_issue_discard(ca->disk_sb.bdev,
1652                                      k.k->p.offset * ca->mi.bucket_size,
1653                                      ca->mi.bucket_size,
1654                                      GFP_KERNEL);
1655                 *discard_pos_done = iter.pos;
1656
1657                 ret = bch2_trans_relock_notrace(trans);
1658                 if (ret)
1659                         goto out;
1660         }
1661
1662         SET_BCH_ALLOC_V4_NEED_DISCARD(&a->v, false);
1663         a->v.data_type = alloc_data_type(a->v, a->v.data_type);
1664 write:
1665         ret =   bch2_trans_update(trans, &iter, &a->k_i, 0) ?:
1666                 bch2_trans_commit(trans, NULL, NULL,
1667                                   BCH_WATERMARK_btree|
1668                                   BCH_TRANS_COMMIT_no_enospc);
1669         if (ret)
1670                 goto out;
1671
1672         count_event(c, bucket_discard);
1673         (*discarded)++;
1674 out:
1675         (*seen)++;
1676         bch2_trans_iter_exit(trans, &iter);
1677         percpu_ref_put(&ca->io_ref);
1678         printbuf_exit(&buf);
1679         return ret;
1680 }
1681
1682 static void bch2_do_discards_work(struct work_struct *work)
1683 {
1684         struct bch_fs *c = container_of(work, struct bch_fs, discard_work);
1685         struct btree_iter iter;
1686         struct bkey_s_c k;
1687         u64 seen = 0, open = 0, need_journal_commit = 0, discarded = 0;
1688         struct bpos discard_pos_done = POS_MAX;
1689         int ret;
1690
1691         /*
1692          * We're doing the commit in bch2_discard_one_bucket instead of using
1693          * for_each_btree_key_commit() so that we can increment counters after
1694          * successful commit:
1695          */
1696         ret = bch2_trans_run(c,
1697                 for_each_btree_key(trans, iter,
1698                                    BTREE_ID_need_discard, POS_MIN, 0, k,
1699                         bch2_discard_one_bucket(trans, &iter, &discard_pos_done,
1700                                                 &seen,
1701                                                 &open,
1702                                                 &need_journal_commit,
1703                                                 &discarded)));
1704
1705         if (need_journal_commit * 2 > seen)
1706                 bch2_journal_flush_async(&c->journal, NULL);
1707
1708         bch2_write_ref_put(c, BCH_WRITE_REF_discard);
1709
1710         trace_discard_buckets(c, seen, open, need_journal_commit, discarded,
1711                               bch2_err_str(ret));
1712 }
1713
1714 void bch2_do_discards(struct bch_fs *c)
1715 {
1716         if (bch2_write_ref_tryget(c, BCH_WRITE_REF_discard) &&
1717             !queue_work(c->write_ref_wq, &c->discard_work))
1718                 bch2_write_ref_put(c, BCH_WRITE_REF_discard);
1719 }
1720
1721 static int invalidate_one_bucket(struct btree_trans *trans,
1722                                  struct btree_iter *lru_iter,
1723                                  struct bkey_s_c lru_k,
1724                                  s64 *nr_to_invalidate)
1725 {
1726         struct bch_fs *c = trans->c;
1727         struct btree_iter alloc_iter = { NULL };
1728         struct bkey_i_alloc_v4 *a = NULL;
1729         struct printbuf buf = PRINTBUF;
1730         struct bpos bucket = u64_to_bucket(lru_k.k->p.offset);
1731         unsigned cached_sectors;
1732         int ret = 0;
1733
1734         if (*nr_to_invalidate <= 0)
1735                 return 1;
1736
1737         if (!bch2_dev_bucket_exists(c, bucket)) {
1738                 prt_str(&buf, "lru entry points to invalid bucket");
1739                 goto err;
1740         }
1741
1742         if (bch2_bucket_is_open_safe(c, bucket.inode, bucket.offset))
1743                 return 0;
1744
1745         a = bch2_trans_start_alloc_update(trans, &alloc_iter, bucket);
1746         ret = PTR_ERR_OR_ZERO(a);
1747         if (ret)
1748                 goto out;
1749
1750         /* We expect harmless races here due to the btree write buffer: */
1751         if (lru_pos_time(lru_iter->pos) != alloc_lru_idx_read(a->v))
1752                 goto out;
1753
1754         BUG_ON(a->v.data_type != BCH_DATA_cached);
1755
1756         if (!a->v.cached_sectors)
1757                 bch_err(c, "invalidating empty bucket, confused");
1758
1759         cached_sectors = a->v.cached_sectors;
1760
1761         SET_BCH_ALLOC_V4_NEED_INC_GEN(&a->v, false);
1762         a->v.gen++;
1763         a->v.data_type          = 0;
1764         a->v.dirty_sectors      = 0;
1765         a->v.cached_sectors     = 0;
1766         a->v.io_time[READ]      = atomic64_read(&c->io_clock[READ].now);
1767         a->v.io_time[WRITE]     = atomic64_read(&c->io_clock[WRITE].now);
1768
1769         ret =   bch2_trans_update(trans, &alloc_iter, &a->k_i,
1770                                 BTREE_TRIGGER_BUCKET_INVALIDATE) ?:
1771                 bch2_trans_commit(trans, NULL, NULL,
1772                                   BCH_WATERMARK_btree|
1773                                   BCH_TRANS_COMMIT_no_enospc);
1774         if (ret)
1775                 goto out;
1776
1777         trace_and_count(c, bucket_invalidate, c, bucket.inode, bucket.offset, cached_sectors);
1778         --*nr_to_invalidate;
1779 out:
1780         bch2_trans_iter_exit(trans, &alloc_iter);
1781         printbuf_exit(&buf);
1782         return ret;
1783 err:
1784         prt_str(&buf, "\n  lru key: ");
1785         bch2_bkey_val_to_text(&buf, c, lru_k);
1786
1787         prt_str(&buf, "\n  lru entry: ");
1788         bch2_lru_pos_to_text(&buf, lru_iter->pos);
1789
1790         prt_str(&buf, "\n  alloc key: ");
1791         if (!a)
1792                 bch2_bpos_to_text(&buf, bucket);
1793         else
1794                 bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&a->k_i));
1795
1796         bch_err(c, "%s", buf.buf);
1797         if (c->curr_recovery_pass > BCH_RECOVERY_PASS_check_lrus) {
1798                 bch2_inconsistent_error(c);
1799                 ret = -EINVAL;
1800         }
1801
1802         goto out;
1803 }
1804
1805 static void bch2_do_invalidates_work(struct work_struct *work)
1806 {
1807         struct bch_fs *c = container_of(work, struct bch_fs, invalidate_work);
1808         struct bch_dev *ca;
1809         struct btree_trans *trans = bch2_trans_get(c);
1810         struct btree_iter iter;
1811         struct bkey_s_c k;
1812         unsigned i;
1813         int ret = 0;
1814
1815         ret = bch2_btree_write_buffer_tryflush(trans);
1816         if (ret)
1817                 goto err;
1818
1819         for_each_member_device(ca, c, i) {
1820                 s64 nr_to_invalidate =
1821                         should_invalidate_buckets(ca, bch2_dev_usage_read(ca));
1822
1823                 ret = for_each_btree_key2_upto(trans, iter, BTREE_ID_lru,
1824                                 lru_pos(ca->dev_idx, 0, 0),
1825                                 lru_pos(ca->dev_idx, U64_MAX, LRU_TIME_MAX),
1826                                 BTREE_ITER_INTENT, k,
1827                         invalidate_one_bucket(trans, &iter, k, &nr_to_invalidate));
1828
1829                 if (ret < 0) {
1830                         percpu_ref_put(&ca->ref);
1831                         break;
1832                 }
1833         }
1834 err:
1835         bch2_trans_put(trans);
1836         bch2_write_ref_put(c, BCH_WRITE_REF_invalidate);
1837 }
1838
1839 void bch2_do_invalidates(struct bch_fs *c)
1840 {
1841         if (bch2_write_ref_tryget(c, BCH_WRITE_REF_invalidate) &&
1842             !queue_work(c->write_ref_wq, &c->invalidate_work))
1843                 bch2_write_ref_put(c, BCH_WRITE_REF_invalidate);
1844 }
1845
1846 int bch2_dev_freespace_init(struct bch_fs *c, struct bch_dev *ca,
1847                             u64 bucket_start, u64 bucket_end)
1848 {
1849         struct btree_trans *trans = bch2_trans_get(c);
1850         struct btree_iter iter;
1851         struct bkey_s_c k;
1852         struct bkey hole;
1853         struct bpos end = POS(ca->dev_idx, bucket_end);
1854         struct bch_member *m;
1855         unsigned long last_updated = jiffies;
1856         int ret;
1857
1858         BUG_ON(bucket_start > bucket_end);
1859         BUG_ON(bucket_end > ca->mi.nbuckets);
1860
1861         bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc,
1862                 POS(ca->dev_idx, max_t(u64, ca->mi.first_bucket, bucket_start)),
1863                 BTREE_ITER_PREFETCH);
1864         /*
1865          * Scan the alloc btree for every bucket on @ca, and add buckets to the
1866          * freespace/need_discard/need_gc_gens btrees as needed:
1867          */
1868         while (1) {
1869                 if (last_updated + HZ * 10 < jiffies) {
1870                         bch_info(ca, "%s: currently at %llu/%llu",
1871                                  __func__, iter.pos.offset, ca->mi.nbuckets);
1872                         last_updated = jiffies;
1873                 }
1874
1875                 bch2_trans_begin(trans);
1876
1877                 if (bkey_ge(iter.pos, end)) {
1878                         ret = 0;
1879                         break;
1880                 }
1881
1882                 k = bch2_get_key_or_hole(&iter, end, &hole);
1883                 ret = bkey_err(k);
1884                 if (ret)
1885                         goto bkey_err;
1886
1887                 if (k.k->type) {
1888                         /*
1889                          * We process live keys in the alloc btree one at a
1890                          * time:
1891                          */
1892                         struct bch_alloc_v4 a_convert;
1893                         const struct bch_alloc_v4 *a = bch2_alloc_to_v4(k, &a_convert);
1894
1895                         ret =   bch2_bucket_do_index(trans, k, a, true) ?:
1896                                 bch2_trans_commit(trans, NULL, NULL,
1897                                                   BCH_TRANS_COMMIT_no_enospc);
1898                         if (ret)
1899                                 goto bkey_err;
1900
1901                         bch2_btree_iter_advance(&iter);
1902                 } else {
1903                         struct bkey_i *freespace;
1904
1905                         freespace = bch2_trans_kmalloc(trans, sizeof(*freespace));
1906                         ret = PTR_ERR_OR_ZERO(freespace);
1907                         if (ret)
1908                                 goto bkey_err;
1909
1910                         bkey_init(&freespace->k);
1911                         freespace->k.type       = KEY_TYPE_set;
1912                         freespace->k.p          = k.k->p;
1913                         freespace->k.size       = k.k->size;
1914
1915                         ret = bch2_btree_insert_trans(trans, BTREE_ID_freespace, freespace, 0) ?:
1916                                 bch2_trans_commit(trans, NULL, NULL,
1917                                                   BCH_TRANS_COMMIT_no_enospc);
1918                         if (ret)
1919                                 goto bkey_err;
1920
1921                         bch2_btree_iter_set_pos(&iter, k.k->p);
1922                 }
1923 bkey_err:
1924                 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
1925                         continue;
1926                 if (ret)
1927                         break;
1928         }
1929
1930         bch2_trans_iter_exit(trans, &iter);
1931         bch2_trans_put(trans);
1932
1933         if (ret < 0) {
1934                 bch_err_msg(ca, ret, "initializing free space");
1935                 return ret;
1936         }
1937
1938         mutex_lock(&c->sb_lock);
1939         m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
1940         SET_BCH_MEMBER_FREESPACE_INITIALIZED(m, true);
1941         mutex_unlock(&c->sb_lock);
1942
1943         return 0;
1944 }
1945
1946 int bch2_fs_freespace_init(struct bch_fs *c)
1947 {
1948         struct bch_dev *ca;
1949         unsigned i;
1950         int ret = 0;
1951         bool doing_init = false;
1952
1953         /*
1954          * We can crash during the device add path, so we need to check this on
1955          * every mount:
1956          */
1957
1958         for_each_member_device(ca, c, i) {
1959                 if (ca->mi.freespace_initialized)
1960                         continue;
1961
1962                 if (!doing_init) {
1963                         bch_info(c, "initializing freespace");
1964                         doing_init = true;
1965                 }
1966
1967                 ret = bch2_dev_freespace_init(c, ca, 0, ca->mi.nbuckets);
1968                 if (ret) {
1969                         percpu_ref_put(&ca->ref);
1970                         bch_err_fn(c, ret);
1971                         return ret;
1972                 }
1973         }
1974
1975         if (doing_init) {
1976                 mutex_lock(&c->sb_lock);
1977                 bch2_write_super(c);
1978                 mutex_unlock(&c->sb_lock);
1979                 bch_verbose(c, "done initializing freespace");
1980         }
1981
1982         return 0;
1983 }
1984
1985 /* Bucket IO clocks: */
1986
1987 int bch2_bucket_io_time_reset(struct btree_trans *trans, unsigned dev,
1988                               size_t bucket_nr, int rw)
1989 {
1990         struct bch_fs *c = trans->c;
1991         struct btree_iter iter;
1992         struct bkey_i_alloc_v4 *a;
1993         u64 now;
1994         int ret = 0;
1995
1996         a = bch2_trans_start_alloc_update(trans, &iter,  POS(dev, bucket_nr));
1997         ret = PTR_ERR_OR_ZERO(a);
1998         if (ret)
1999                 return ret;
2000
2001         now = atomic64_read(&c->io_clock[rw].now);
2002         if (a->v.io_time[rw] == now)
2003                 goto out;
2004
2005         a->v.io_time[rw] = now;
2006
2007         ret   = bch2_trans_update(trans, &iter, &a->k_i, 0) ?:
2008                 bch2_trans_commit(trans, NULL, NULL, 0);
2009 out:
2010         bch2_trans_iter_exit(trans, &iter);
2011         return ret;
2012 }
2013
2014 /* Startup/shutdown (ro/rw): */
2015
2016 void bch2_recalc_capacity(struct bch_fs *c)
2017 {
2018         struct bch_dev *ca;
2019         u64 capacity = 0, reserved_sectors = 0, gc_reserve;
2020         unsigned bucket_size_max = 0;
2021         unsigned long ra_pages = 0;
2022         unsigned i;
2023
2024         lockdep_assert_held(&c->state_lock);
2025
2026         for_each_online_member(ca, c, i) {
2027                 struct backing_dev_info *bdi = ca->disk_sb.bdev->bd_disk->bdi;
2028
2029                 ra_pages += bdi->ra_pages;
2030         }
2031
2032         bch2_set_ra_pages(c, ra_pages);
2033
2034         for_each_rw_member(ca, c, i) {
2035                 u64 dev_reserve = 0;
2036
2037                 /*
2038                  * We need to reserve buckets (from the number
2039                  * of currently available buckets) against
2040                  * foreground writes so that mainly copygc can
2041                  * make forward progress.
2042                  *
2043                  * We need enough to refill the various reserves
2044                  * from scratch - copygc will use its entire
2045                  * reserve all at once, then run against when
2046                  * its reserve is refilled (from the formerly
2047                  * available buckets).
2048                  *
2049                  * This reserve is just used when considering if
2050                  * allocations for foreground writes must wait -
2051                  * not -ENOSPC calculations.
2052                  */
2053
2054                 dev_reserve += ca->nr_btree_reserve * 2;
2055                 dev_reserve += ca->mi.nbuckets >> 6; /* copygc reserve */
2056
2057                 dev_reserve += 1;       /* btree write point */
2058                 dev_reserve += 1;       /* copygc write point */
2059                 dev_reserve += 1;       /* rebalance write point */
2060
2061                 dev_reserve *= ca->mi.bucket_size;
2062
2063                 capacity += bucket_to_sector(ca, ca->mi.nbuckets -
2064                                              ca->mi.first_bucket);
2065
2066                 reserved_sectors += dev_reserve * 2;
2067
2068                 bucket_size_max = max_t(unsigned, bucket_size_max,
2069                                         ca->mi.bucket_size);
2070         }
2071
2072         gc_reserve = c->opts.gc_reserve_bytes
2073                 ? c->opts.gc_reserve_bytes >> 9
2074                 : div64_u64(capacity * c->opts.gc_reserve_percent, 100);
2075
2076         reserved_sectors = max(gc_reserve, reserved_sectors);
2077
2078         reserved_sectors = min(reserved_sectors, capacity);
2079
2080         c->capacity = capacity - reserved_sectors;
2081
2082         c->bucket_size_max = bucket_size_max;
2083
2084         /* Wake up case someone was waiting for buckets */
2085         closure_wake_up(&c->freelist_wait);
2086 }
2087
2088 u64 bch2_min_rw_member_capacity(struct bch_fs *c)
2089 {
2090         struct bch_dev *ca;
2091         unsigned i;
2092         u64 ret = U64_MAX;
2093
2094         for_each_rw_member(ca, c, i)
2095                 ret = min(ret, ca->mi.nbuckets * ca->mi.bucket_size);
2096         return ret;
2097 }
2098
2099 static bool bch2_dev_has_open_write_point(struct bch_fs *c, struct bch_dev *ca)
2100 {
2101         struct open_bucket *ob;
2102         bool ret = false;
2103
2104         for (ob = c->open_buckets;
2105              ob < c->open_buckets + ARRAY_SIZE(c->open_buckets);
2106              ob++) {
2107                 spin_lock(&ob->lock);
2108                 if (ob->valid && !ob->on_partial_list &&
2109                     ob->dev == ca->dev_idx)
2110                         ret = true;
2111                 spin_unlock(&ob->lock);
2112         }
2113
2114         return ret;
2115 }
2116
2117 /* device goes ro: */
2118 void bch2_dev_allocator_remove(struct bch_fs *c, struct bch_dev *ca)
2119 {
2120         unsigned i;
2121
2122         /* First, remove device from allocation groups: */
2123
2124         for (i = 0; i < ARRAY_SIZE(c->rw_devs); i++)
2125                 clear_bit(ca->dev_idx, c->rw_devs[i].d);
2126
2127         /*
2128          * Capacity is calculated based off of devices in allocation groups:
2129          */
2130         bch2_recalc_capacity(c);
2131
2132         bch2_open_buckets_stop(c, ca, false);
2133
2134         /*
2135          * Wake up threads that were blocked on allocation, so they can notice
2136          * the device can no longer be removed and the capacity has changed:
2137          */
2138         closure_wake_up(&c->freelist_wait);
2139
2140         /*
2141          * journal_res_get() can block waiting for free space in the journal -
2142          * it needs to notice there may not be devices to allocate from anymore:
2143          */
2144         wake_up(&c->journal.wait);
2145
2146         /* Now wait for any in flight writes: */
2147
2148         closure_wait_event(&c->open_buckets_wait,
2149                            !bch2_dev_has_open_write_point(c, ca));
2150 }
2151
2152 /* device goes rw: */
2153 void bch2_dev_allocator_add(struct bch_fs *c, struct bch_dev *ca)
2154 {
2155         unsigned i;
2156
2157         for (i = 0; i < ARRAY_SIZE(c->rw_devs); i++)
2158                 if (ca->mi.data_allowed & (1 << i))
2159                         set_bit(ca->dev_idx, c->rw_devs[i].d);
2160 }
2161
2162 void bch2_fs_allocator_background_init(struct bch_fs *c)
2163 {
2164         spin_lock_init(&c->freelist_lock);
2165         INIT_WORK(&c->discard_work, bch2_do_discards_work);
2166         INIT_WORK(&c->invalidate_work, bch2_do_invalidates_work);
2167 }