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                                         BCH_TRANS_COMMIT_lazy_rw,
 563                                 bch2_btree_insert_trans(trans, BTREE_ID_bucket_gens, &g.k_i, 0));
 564                         if (ret)
 565                                 break;
 566                         have_bucket_gens_key = false;
 567                 }
 568
 569                 if (!have_bucket_gens_key) {
 570                         bkey_bucket_gens_init(&g.k_i);
 571                         g.k.p = pos;
 572                         have_bucket_gens_key = true;
 573                 }
 574
 575                 g.v.gens[offset] = gen;
 576         }
 577         bch2_trans_iter_exit(trans, &iter);
 578
 579         if (have_bucket_gens_key && !ret)
 580                 ret = commit_do(trans, NULL, NULL,
 581                                 BCH_TRANS_COMMIT_no_enospc|
 582                                 BCH_TRANS_COMMIT_lazy_rw,
 583                         bch2_btree_insert_trans(trans, BTREE_ID_bucket_gens, &g.k_i, 0));
 584
 585         bch2_trans_put(trans);
 586
 587         if (ret)
 588                 bch_err_fn(c, ret);
 589         return ret;
 590 }
 591
 592 int bch2_alloc_read(struct bch_fs *c)
 593 {
 594         struct btree_trans *trans = bch2_trans_get(c);
 595         struct btree_iter iter;
 596         struct bkey_s_c k;
 597         struct bch_dev *ca;
 598         int ret;
 599
 600         down_read(&c->gc_lock);
 601
 602         if (c->sb.version_upgrade_complete >= bcachefs_metadata_version_bucket_gens) {
 603                 const struct bch_bucket_gens *g;
 604                 u64 b;
 605
 606                 for_each_btree_key(trans, iter, BTREE_ID_bucket_gens, POS_MIN,
 607                                    BTREE_ITER_PREFETCH, k, ret) {
 608                         u64 start = bucket_gens_pos_to_alloc(k.k->p, 0).offset;
 609                         u64 end = bucket_gens_pos_to_alloc(bpos_nosnap_successor(k.k->p), 0).offset;
 610
 611                         if (k.k->type != KEY_TYPE_bucket_gens)
 612                                 continue;
 613
 614                         g = bkey_s_c_to_bucket_gens(k).v;
 615
 616                         /*
 617                          * Not a fsck error because this is checked/repaired by
 618                          * bch2_check_alloc_key() which runs later:
 619                          */
 620                         if (!bch2_dev_exists2(c, k.k->p.inode))
 621                                 continue;
 622
 623                         ca = bch_dev_bkey_exists(c, k.k->p.inode);
 624
 625                         for (b = max_t(u64, ca->mi.first_bucket, start);
 626                              b < min_t(u64, ca->mi.nbuckets, end);
 627                              b++)
 628                                 *bucket_gen(ca, b) = g->gens[b & KEY_TYPE_BUCKET_GENS_MASK];
 629                 }
 630                 bch2_trans_iter_exit(trans, &iter);
 631         } else {
 632                 struct bch_alloc_v4 a;
 633
 634                 for_each_btree_key(trans, iter, BTREE_ID_alloc, POS_MIN,
 635                                    BTREE_ITER_PREFETCH, k, ret) {
 636                         /*
 637                          * Not a fsck error because this is checked/repaired by
 638                          * bch2_check_alloc_key() which runs later:
 639                          */
 640                         if (!bch2_dev_bucket_exists(c, k.k->p))
 641                                 continue;
 642
 643                         ca = bch_dev_bkey_exists(c, k.k->p.inode);
 644
 645                         *bucket_gen(ca, k.k->p.offset) = bch2_alloc_to_v4(k, &a)->gen;
 646                 }
 647                 bch2_trans_iter_exit(trans, &iter);
 648         }
 649
 650         bch2_trans_put(trans);
 651         up_read(&c->gc_lock);
 652
 653         if (ret)
 654                 bch_err_fn(c, ret);
 655
 656         return ret;
 657 }
 658
 659 /* Free space/discard btree: */
 660
 661 static int bch2_bucket_do_index(struct btree_trans *trans,
 662                                 struct bkey_s_c alloc_k,
 663                                 const struct bch_alloc_v4 *a,
 664                                 bool set)
 665 {
 666         struct bch_fs *c = trans->c;
 667         struct bch_dev *ca = bch_dev_bkey_exists(c, alloc_k.k->p.inode);
 668         struct btree_iter iter;
 669         struct bkey_s_c old;
 670         struct bkey_i *k;
 671         enum btree_id btree;
 672         enum bch_bkey_type old_type = !set ? KEY_TYPE_set : KEY_TYPE_deleted;
 673         enum bch_bkey_type new_type =  set ? KEY_TYPE_set : KEY_TYPE_deleted;
 674         struct printbuf buf = PRINTBUF;
 675         int ret;
 676
 677         if (a->data_type != BCH_DATA_free &&
 678             a->data_type != BCH_DATA_need_discard)
 679                 return 0;
 680
 681         k = bch2_trans_kmalloc_nomemzero(trans, sizeof(*k));
 682         if (IS_ERR(k))
 683                 return PTR_ERR(k);
 684
 685         bkey_init(&k->k);
 686         k->k.type = new_type;
 687
 688         switch (a->data_type) {
 689         case BCH_DATA_free:
 690                 btree = BTREE_ID_freespace;
 691                 k->k.p = alloc_freespace_pos(alloc_k.k->p, *a);
 692                 bch2_key_resize(&k->k, 1);
 693                 break;
 694         case BCH_DATA_need_discard:
 695                 btree = BTREE_ID_need_discard;
 696                 k->k.p = alloc_k.k->p;
 697                 break;
 698         default:
 699                 return 0;
 700         }
 701
 702         old = bch2_bkey_get_iter(trans, &iter, btree,
 703                              bkey_start_pos(&k->k),
 704                              BTREE_ITER_INTENT);
 705         ret = bkey_err(old);
 706         if (ret)
 707                 return ret;
 708
 709         if (ca->mi.freespace_initialized &&
 710             c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_info &&
 711             bch2_trans_inconsistent_on(old.k->type != old_type, trans,
 712                         "incorrect key when %s %s:%llu:%llu:0 (got %s should be %s)\n"
 713                         "  for %s",
 714                         set ? "setting" : "clearing",
 715                         bch2_btree_id_str(btree),
 716                         iter.pos.inode,
 717                         iter.pos.offset,
 718                         bch2_bkey_types[old.k->type],
 719                         bch2_bkey_types[old_type],
 720                         (bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf))) {
 721                 ret = -EIO;
 722                 goto err;
 723         }
 724
 725         ret = bch2_trans_update(trans, &iter, k, 0);
 726 err:
 727         bch2_trans_iter_exit(trans, &iter);
 728         printbuf_exit(&buf);
 729         return ret;
 730 }
 731
 732 static noinline int bch2_bucket_gen_update(struct btree_trans *trans,
 733                                            struct bpos bucket, u8 gen)
 734 {
 735         struct btree_iter iter;
 736         unsigned offset;
 737         struct bpos pos = alloc_gens_pos(bucket, &offset);
 738         struct bkey_i_bucket_gens *g;
 739         struct bkey_s_c k;
 740         int ret;
 741
 742         g = bch2_trans_kmalloc(trans, sizeof(*g));
 743         ret = PTR_ERR_OR_ZERO(g);
 744         if (ret)
 745                 return ret;
 746
 747         k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_bucket_gens, pos,
 748                                BTREE_ITER_INTENT|
 749                                BTREE_ITER_WITH_UPDATES);
 750         ret = bkey_err(k);
 751         if (ret)
 752                 return ret;
 753
 754         if (k.k->type != KEY_TYPE_bucket_gens) {
 755                 bkey_bucket_gens_init(&g->k_i);
 756                 g->k.p = iter.pos;
 757         } else {
 758                 bkey_reassemble(&g->k_i, k);
 759         }
 760
 761         g->v.gens[offset] = gen;
 762
 763         ret = bch2_trans_update(trans, &iter, &g->k_i, 0);
 764         bch2_trans_iter_exit(trans, &iter);
 765         return ret;
 766 }
 767
 768 int bch2_trans_mark_alloc(struct btree_trans *trans,
 769                           enum btree_id btree_id, unsigned level,
 770                           struct bkey_s_c old, struct bkey_i *new,
 771                           unsigned flags)
 772 {
 773         struct bch_fs *c = trans->c;
 774         struct bch_alloc_v4 old_a_convert, *new_a;
 775         const struct bch_alloc_v4 *old_a;
 776         u64 old_lru, new_lru;
 777         int ret = 0;
 778
 779         /*
 780          * Deletion only happens in the device removal path, with
 781          * BTREE_TRIGGER_NORUN:
 782          */
 783         BUG_ON(new->k.type != KEY_TYPE_alloc_v4);
 784
 785         old_a = bch2_alloc_to_v4(old, &old_a_convert);
 786         new_a = &bkey_i_to_alloc_v4(new)->v;
 787
 788         new_a->data_type = alloc_data_type(*new_a, new_a->data_type);
 789
 790         if (bch2_bucket_sectors(*new_a) > bch2_bucket_sectors(*old_a)) {
 791                 new_a->io_time[READ] = max_t(u64, 1, atomic64_read(&c->io_clock[READ].now));
 792                 new_a->io_time[WRITE]= max_t(u64, 1, atomic64_read(&c->io_clock[WRITE].now));
 793                 SET_BCH_ALLOC_V4_NEED_INC_GEN(new_a, true);
 794                 SET_BCH_ALLOC_V4_NEED_DISCARD(new_a, true);
 795         }
 796
 797         if (data_type_is_empty(new_a->data_type) &&
 798             BCH_ALLOC_V4_NEED_INC_GEN(new_a) &&
 799             !bch2_bucket_is_open_safe(c, new->k.p.inode, new->k.p.offset)) {
 800                 new_a->gen++;
 801                 SET_BCH_ALLOC_V4_NEED_INC_GEN(new_a, false);
 802         }
 803
 804         if (old_a->data_type != new_a->data_type ||
 805             (new_a->data_type == BCH_DATA_free &&
 806              alloc_freespace_genbits(*old_a) != alloc_freespace_genbits(*new_a))) {
 807                 ret =   bch2_bucket_do_index(trans, old, old_a, false) ?:
 808                         bch2_bucket_do_index(trans, bkey_i_to_s_c(new), new_a, true);
 809                 if (ret)
 810                         return ret;
 811         }
 812
 813         if (new_a->data_type == BCH_DATA_cached &&
 814             !new_a->io_time[READ])
 815                 new_a->io_time[READ] = max_t(u64, 1, atomic64_read(&c->io_clock[READ].now));
 816
 817         old_lru = alloc_lru_idx_read(*old_a);
 818         new_lru = alloc_lru_idx_read(*new_a);
 819
 820         if (old_lru != new_lru) {
 821                 ret = bch2_lru_change(trans, new->k.p.inode,
 822                                       bucket_to_u64(new->k.p),
 823                                       old_lru, new_lru);
 824                 if (ret)
 825                         return ret;
 826         }
 827
 828         new_a->fragmentation_lru = alloc_lru_idx_fragmentation(*new_a,
 829                                         bch_dev_bkey_exists(c, new->k.p.inode));
 830
 831         if (old_a->fragmentation_lru != new_a->fragmentation_lru) {
 832                 ret = bch2_lru_change(trans,
 833                                 BCH_LRU_FRAGMENTATION_START,
 834                                 bucket_to_u64(new->k.p),
 835                                 old_a->fragmentation_lru, new_a->fragmentation_lru);
 836                 if (ret)
 837                         return ret;
 838         }
 839
 840         if (old_a->gen != new_a->gen) {
 841                 ret = bch2_bucket_gen_update(trans, new->k.p, new_a->gen);
 842                 if (ret)
 843                         return ret;
 844         }
 845
 846         /*
 847          * need to know if we're getting called from the invalidate path or
 848          * not:
 849          */
 850
 851         if ((flags & BTREE_TRIGGER_BUCKET_INVALIDATE) &&
 852             old_a->cached_sectors) {
 853                 ret = bch2_update_cached_sectors_list(trans, new->k.p.inode,
 854                                                       -((s64) old_a->cached_sectors));
 855                 if (ret)
 856                         return ret;
 857         }
 858
 859         return 0;
 860 }
 861
 862 /*
 863  * This synthesizes deleted extents for holes, similar to BTREE_ITER_SLOTS for
 864  * extents style btrees, but works on non-extents btrees:
 865  */
 866 static struct bkey_s_c bch2_get_key_or_hole(struct btree_iter *iter, struct bpos end, struct bkey *hole)
 867 {
 868         struct bkey_s_c k = bch2_btree_iter_peek_slot(iter);
 869
 870         if (bkey_err(k))
 871                 return k;
 872
 873         if (k.k->type) {
 874                 return k;
 875         } else {
 876                 struct btree_iter iter2;
 877                 struct bpos next;
 878
 879                 bch2_trans_copy_iter(&iter2, iter);
 880
 881                 if (!bpos_eq(iter->path->l[0].b->key.k.p, SPOS_MAX))
 882                         end = bkey_min(end, bpos_nosnap_successor(iter->path->l[0].b->key.k.p));
 883
 884                 end = bkey_min(end, POS(iter->pos.inode, iter->pos.offset + U32_MAX - 1));
 885
 886                 /*
 887                  * btree node min/max is a closed interval, upto takes a half
 888                  * open interval:
 889                  */
 890                 k = bch2_btree_iter_peek_upto(&iter2, end);
 891                 next = iter2.pos;
 892                 bch2_trans_iter_exit(iter->trans, &iter2);
 893
 894                 BUG_ON(next.offset >= iter->pos.offset + U32_MAX);
 895
 896                 if (bkey_err(k))
 897                         return k;
 898
 899                 bkey_init(hole);
 900                 hole->p = iter->pos;
 901
 902                 bch2_key_resize(hole, next.offset - iter->pos.offset);
 903                 return (struct bkey_s_c) { hole, NULL };
 904         }
 905 }
 906
 907 static bool next_bucket(struct bch_fs *c, struct bpos *bucket)
 908 {
 909         struct bch_dev *ca;
 910         unsigned iter;
 911
 912         if (bch2_dev_bucket_exists(c, *bucket))
 913                 return true;
 914
 915         if (bch2_dev_exists2(c, bucket->inode)) {
 916                 ca = bch_dev_bkey_exists(c, bucket->inode);
 917
 918                 if (bucket->offset < ca->mi.first_bucket) {
 919                         bucket->offset = ca->mi.first_bucket;
 920                         return true;
 921                 }
 922
 923                 bucket->inode++;
 924                 bucket->offset = 0;
 925         }
 926
 927         rcu_read_lock();
 928         iter = bucket->inode;
 929         ca = __bch2_next_dev(c, &iter, NULL);
 930         if (ca)
 931                 *bucket = POS(ca->dev_idx, ca->mi.first_bucket);
 932         rcu_read_unlock();
 933
 934         return ca != NULL;
 935 }
 936
 937 static struct bkey_s_c bch2_get_key_or_real_bucket_hole(struct btree_iter *iter, struct bkey *hole)
 938 {
 939         struct bch_fs *c = iter->trans->c;
 940         struct bkey_s_c k;
 941 again:
 942         k = bch2_get_key_or_hole(iter, POS_MAX, hole);
 943         if (bkey_err(k))
 944                 return k;
 945
 946         if (!k.k->type) {
 947                 struct bpos bucket = bkey_start_pos(k.k);
 948
 949                 if (!bch2_dev_bucket_exists(c, bucket)) {
 950                         if (!next_bucket(c, &bucket))
 951                                 return bkey_s_c_null;
 952
 953                         bch2_btree_iter_set_pos(iter, bucket);
 954                         goto again;
 955                 }
 956
 957                 if (!bch2_dev_bucket_exists(c, k.k->p)) {
 958                         struct bch_dev *ca = bch_dev_bkey_exists(c, bucket.inode);
 959
 960                         bch2_key_resize(hole, ca->mi.nbuckets - bucket.offset);
 961                 }
 962         }
 963
 964         return k;
 965 }
 966
 967 static noinline_for_stack
 968 int bch2_check_alloc_key(struct btree_trans *trans,
 969                          struct bkey_s_c alloc_k,
 970                          struct btree_iter *alloc_iter,
 971                          struct btree_iter *discard_iter,
 972                          struct btree_iter *freespace_iter,
 973                          struct btree_iter *bucket_gens_iter)
 974 {
 975         struct bch_fs *c = trans->c;
 976         struct bch_dev *ca;
 977         struct bch_alloc_v4 a_convert;
 978         const struct bch_alloc_v4 *a;
 979         unsigned discard_key_type, freespace_key_type;
 980         unsigned gens_offset;
 981         struct bkey_s_c k;
 982         struct printbuf buf = PRINTBUF;
 983         int ret;
 984
 985         if (fsck_err_on(!bch2_dev_bucket_exists(c, alloc_k.k->p), c,
 986                         alloc_key_to_missing_dev_bucket,
 987                         "alloc key for invalid device:bucket %llu:%llu",
 988                         alloc_k.k->p.inode, alloc_k.k->p.offset))
 989                 return bch2_btree_delete_at(trans, alloc_iter, 0);
 990
 991         ca = bch_dev_bkey_exists(c, alloc_k.k->p.inode);
 992         if (!ca->mi.freespace_initialized)
 993                 return 0;
 994
 995         a = bch2_alloc_to_v4(alloc_k, &a_convert);
 996
 997         discard_key_type = a->data_type == BCH_DATA_need_discard ? KEY_TYPE_set : 0;
 998         bch2_btree_iter_set_pos(discard_iter, alloc_k.k->p);
 999         k = bch2_btree_iter_peek_slot(discard_iter);
1000         ret = bkey_err(k);
1001         if (ret)
1002                 goto err;
1003
1004         if (k.k->type != discard_key_type &&
1005             (c->opts.reconstruct_alloc ||
1006              fsck_err(c, need_discard_key_wrong,
1007                       "incorrect key in need_discard btree (got %s should be %s)\n"
1008                       "  %s",
1009                       bch2_bkey_types[k.k->type],
1010                       bch2_bkey_types[discard_key_type],
1011                       (bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf)))) {
1012                 struct bkey_i *update =
1013                         bch2_trans_kmalloc(trans, sizeof(*update));
1014
1015                 ret = PTR_ERR_OR_ZERO(update);
1016                 if (ret)
1017                         goto err;
1018
1019                 bkey_init(&update->k);
1020                 update->k.type  = discard_key_type;
1021                 update->k.p     = discard_iter->pos;
1022
1023                 ret = bch2_trans_update(trans, discard_iter, update, 0);
1024                 if (ret)
1025                         goto err;
1026         }
1027
1028         freespace_key_type = a->data_type == BCH_DATA_free ? KEY_TYPE_set : 0;
1029         bch2_btree_iter_set_pos(freespace_iter, alloc_freespace_pos(alloc_k.k->p, *a));
1030         k = bch2_btree_iter_peek_slot(freespace_iter);
1031         ret = bkey_err(k);
1032         if (ret)
1033                 goto err;
1034
1035         if (k.k->type != freespace_key_type &&
1036             (c->opts.reconstruct_alloc ||
1037              fsck_err(c, freespace_key_wrong,
1038                       "incorrect key in freespace btree (got %s should be %s)\n"
1039                       "  %s",
1040                       bch2_bkey_types[k.k->type],
1041                       bch2_bkey_types[freespace_key_type],
1042                       (printbuf_reset(&buf),
1043                        bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf)))) {
1044                 struct bkey_i *update =
1045                         bch2_trans_kmalloc(trans, sizeof(*update));
1046
1047                 ret = PTR_ERR_OR_ZERO(update);
1048                 if (ret)
1049                         goto err;
1050
1051                 bkey_init(&update->k);
1052                 update->k.type  = freespace_key_type;
1053                 update->k.p     = freespace_iter->pos;
1054                 bch2_key_resize(&update->k, 1);
1055
1056                 ret = bch2_trans_update(trans, freespace_iter, update, 0);
1057                 if (ret)
1058                         goto err;
1059         }
1060
1061         bch2_btree_iter_set_pos(bucket_gens_iter, alloc_gens_pos(alloc_k.k->p, &gens_offset));
1062         k = bch2_btree_iter_peek_slot(bucket_gens_iter);
1063         ret = bkey_err(k);
1064         if (ret)
1065                 goto err;
1066
1067         if (a->gen != alloc_gen(k, gens_offset) &&
1068             (c->opts.reconstruct_alloc ||
1069              fsck_err(c, bucket_gens_key_wrong,
1070                       "incorrect gen in bucket_gens btree (got %u should be %u)\n"
1071                       "  %s",
1072                       alloc_gen(k, gens_offset), a->gen,
1073                       (printbuf_reset(&buf),
1074                        bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf)))) {
1075                 struct bkey_i_bucket_gens *g =
1076                         bch2_trans_kmalloc(trans, sizeof(*g));
1077
1078                 ret = PTR_ERR_OR_ZERO(g);
1079                 if (ret)
1080                         goto err;
1081
1082                 if (k.k->type == KEY_TYPE_bucket_gens) {
1083                         bkey_reassemble(&g->k_i, k);
1084                 } else {
1085                         bkey_bucket_gens_init(&g->k_i);
1086                         g->k.p = alloc_gens_pos(alloc_k.k->p, &gens_offset);
1087                 }
1088
1089                 g->v.gens[gens_offset] = a->gen;
1090
1091                 ret = bch2_trans_update(trans, bucket_gens_iter, &g->k_i, 0);
1092                 if (ret)
1093                         goto err;
1094         }
1095 err:
1096 fsck_err:
1097         printbuf_exit(&buf);
1098         return ret;
1099 }
1100
1101 static noinline_for_stack
1102 int bch2_check_alloc_hole_freespace(struct btree_trans *trans,
1103                                     struct bpos start,
1104                                     struct bpos *end,
1105                                     struct btree_iter *freespace_iter)
1106 {
1107         struct bch_fs *c = trans->c;
1108         struct bch_dev *ca;
1109         struct bkey_s_c k;
1110         struct printbuf buf = PRINTBUF;
1111         int ret;
1112
1113         ca = bch_dev_bkey_exists(c, start.inode);
1114         if (!ca->mi.freespace_initialized)
1115                 return 0;
1116
1117         bch2_btree_iter_set_pos(freespace_iter, start);
1118
1119         k = bch2_btree_iter_peek_slot(freespace_iter);
1120         ret = bkey_err(k);
1121         if (ret)
1122                 goto err;
1123
1124         *end = bkey_min(k.k->p, *end);
1125
1126         if (k.k->type != KEY_TYPE_set &&
1127             (c->opts.reconstruct_alloc ||
1128              fsck_err(c, freespace_hole_missing,
1129                       "hole in alloc btree missing in freespace btree\n"
1130                       "  device %llu buckets %llu-%llu",
1131                       freespace_iter->pos.inode,
1132                       freespace_iter->pos.offset,
1133                       end->offset))) {
1134                 struct bkey_i *update =
1135                         bch2_trans_kmalloc(trans, sizeof(*update));
1136
1137                 ret = PTR_ERR_OR_ZERO(update);
1138                 if (ret)
1139                         goto err;
1140
1141                 bkey_init(&update->k);
1142                 update->k.type  = KEY_TYPE_set;
1143                 update->k.p     = freespace_iter->pos;
1144                 bch2_key_resize(&update->k,
1145                                 min_t(u64, U32_MAX, end->offset -
1146                                       freespace_iter->pos.offset));
1147
1148                 ret = bch2_trans_update(trans, freespace_iter, update, 0);
1149                 if (ret)
1150                         goto err;
1151         }
1152 err:
1153 fsck_err:
1154         printbuf_exit(&buf);
1155         return ret;
1156 }
1157
1158 static noinline_for_stack
1159 int bch2_check_alloc_hole_bucket_gens(struct btree_trans *trans,
1160                                       struct bpos start,
1161                                       struct bpos *end,
1162                                       struct btree_iter *bucket_gens_iter)
1163 {
1164         struct bch_fs *c = trans->c;
1165         struct bkey_s_c k;
1166         struct printbuf buf = PRINTBUF;
1167         unsigned i, gens_offset, gens_end_offset;
1168         int ret;
1169
1170         if (c->sb.version < bcachefs_metadata_version_bucket_gens)
1171                 return 0;
1172
1173         bch2_btree_iter_set_pos(bucket_gens_iter, alloc_gens_pos(start, &gens_offset));
1174
1175         k = bch2_btree_iter_peek_slot(bucket_gens_iter);
1176         ret = bkey_err(k);
1177         if (ret)
1178                 goto err;
1179
1180         if (bkey_cmp(alloc_gens_pos(start, &gens_offset),
1181                      alloc_gens_pos(*end,  &gens_end_offset)))
1182                 gens_end_offset = KEY_TYPE_BUCKET_GENS_NR;
1183
1184         if (k.k->type == KEY_TYPE_bucket_gens) {
1185                 struct bkey_i_bucket_gens g;
1186                 bool need_update = false;
1187
1188                 bkey_reassemble(&g.k_i, k);
1189
1190                 for (i = gens_offset; i < gens_end_offset; i++) {
1191                         if (fsck_err_on(g.v.gens[i], c,
1192                                         bucket_gens_hole_wrong,
1193                                         "hole in alloc btree at %llu:%llu with nonzero gen in bucket_gens btree (%u)",
1194                                         bucket_gens_pos_to_alloc(k.k->p, i).inode,
1195                                         bucket_gens_pos_to_alloc(k.k->p, i).offset,
1196                                         g.v.gens[i])) {
1197                                 g.v.gens[i] = 0;
1198                                 need_update = true;
1199                         }
1200                 }
1201
1202                 if (need_update) {
1203                         struct bkey_i *u = bch2_trans_kmalloc(trans, sizeof(g));
1204
1205                         ret = PTR_ERR_OR_ZERO(u);
1206                         if (ret)
1207                                 goto err;
1208
1209                         memcpy(u, &g, sizeof(g));
1210
1211                         ret = bch2_trans_update(trans, bucket_gens_iter, u, 0);
1212                         if (ret)
1213                                 goto err;
1214                 }
1215         }
1216
1217         *end = bkey_min(*end, bucket_gens_pos_to_alloc(bpos_nosnap_successor(k.k->p), 0));
1218 err:
1219 fsck_err:
1220         printbuf_exit(&buf);
1221         return ret;
1222 }
1223
1224 static noinline_for_stack int bch2_check_discard_freespace_key(struct btree_trans *trans,
1225                                               struct btree_iter *iter)
1226 {
1227         struct bch_fs *c = trans->c;
1228         struct btree_iter alloc_iter;
1229         struct bkey_s_c alloc_k;
1230         struct bch_alloc_v4 a_convert;
1231         const struct bch_alloc_v4 *a;
1232         u64 genbits;
1233         struct bpos pos;
1234         enum bch_data_type state = iter->btree_id == BTREE_ID_need_discard
1235                 ? BCH_DATA_need_discard
1236                 : BCH_DATA_free;
1237         struct printbuf buf = PRINTBUF;
1238         int ret;
1239
1240         pos = iter->pos;
1241         pos.offset &= ~(~0ULL << 56);
1242         genbits = iter->pos.offset & (~0ULL << 56);
1243
1244         alloc_k = bch2_bkey_get_iter(trans, &alloc_iter, BTREE_ID_alloc, pos, 0);
1245         ret = bkey_err(alloc_k);
1246         if (ret)
1247                 return ret;
1248
1249         if (fsck_err_on(!bch2_dev_bucket_exists(c, pos), c,
1250                         need_discard_freespace_key_to_invalid_dev_bucket,
1251                         "entry in %s btree for nonexistant dev:bucket %llu:%llu",
1252                         bch2_btree_id_str(iter->btree_id), pos.inode, pos.offset))
1253                 goto delete;
1254
1255         a = bch2_alloc_to_v4(alloc_k, &a_convert);
1256
1257         if (fsck_err_on(a->data_type != state ||
1258                         (state == BCH_DATA_free &&
1259                          genbits != alloc_freespace_genbits(*a)), c,
1260                         need_discard_freespace_key_bad,
1261                         "%s\n  incorrectly set at %s:%llu:%llu:0 (free %u, genbits %llu should be %llu)",
1262                         (bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf),
1263                         bch2_btree_id_str(iter->btree_id),
1264                         iter->pos.inode,
1265                         iter->pos.offset,
1266                         a->data_type == state,
1267                         genbits >> 56, alloc_freespace_genbits(*a) >> 56))
1268                 goto delete;
1269 out:
1270 fsck_err:
1271         set_btree_iter_dontneed(&alloc_iter);
1272         bch2_trans_iter_exit(trans, &alloc_iter);
1273         printbuf_exit(&buf);
1274         return ret;
1275 delete:
1276         ret =   bch2_btree_delete_extent_at(trans, iter,
1277                         iter->btree_id == BTREE_ID_freespace ? 1 : 0, 0) ?:
1278                 bch2_trans_commit(trans, NULL, NULL,
1279                         BCH_TRANS_COMMIT_no_enospc|BCH_TRANS_COMMIT_lazy_rw);
1280         goto out;
1281 }
1282
1283 /*
1284  * We've already checked that generation numbers in the bucket_gens btree are
1285  * valid for buckets that exist; this just checks for keys for nonexistent
1286  * buckets.
1287  */
1288 static noinline_for_stack
1289 int bch2_check_bucket_gens_key(struct btree_trans *trans,
1290                                struct btree_iter *iter,
1291                                struct bkey_s_c k)
1292 {
1293         struct bch_fs *c = trans->c;
1294         struct bkey_i_bucket_gens g;
1295         struct bch_dev *ca;
1296         u64 start = bucket_gens_pos_to_alloc(k.k->p, 0).offset;
1297         u64 end = bucket_gens_pos_to_alloc(bpos_nosnap_successor(k.k->p), 0).offset;
1298         u64 b;
1299         bool need_update = false, dev_exists;
1300         struct printbuf buf = PRINTBUF;
1301         int ret = 0;
1302
1303         BUG_ON(k.k->type != KEY_TYPE_bucket_gens);
1304         bkey_reassemble(&g.k_i, k);
1305
1306         /* if no bch_dev, skip out whether we repair or not */
1307         dev_exists = bch2_dev_exists2(c, k.k->p.inode);
1308         if (!dev_exists) {
1309                 if (fsck_err_on(!dev_exists, c,
1310                                 bucket_gens_to_invalid_dev,
1311                                 "bucket_gens key for invalid device:\n  %s",
1312                                 (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1313                         ret = bch2_btree_delete_at(trans, iter, 0);
1314                 }
1315                 goto out;
1316         }
1317
1318         ca = bch_dev_bkey_exists(c, k.k->p.inode);
1319         if (fsck_err_on(end <= ca->mi.first_bucket ||
1320                         start >= ca->mi.nbuckets, c,
1321                         bucket_gens_to_invalid_buckets,
1322                         "bucket_gens key for invalid buckets:\n  %s",
1323                         (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1324                 ret = bch2_btree_delete_at(trans, iter, 0);
1325                 goto out;
1326         }
1327
1328         for (b = start; b < ca->mi.first_bucket; b++)
1329                 if (fsck_err_on(g.v.gens[b & KEY_TYPE_BUCKET_GENS_MASK], c,
1330                                 bucket_gens_nonzero_for_invalid_buckets,
1331                                 "bucket_gens key has nonzero gen for invalid bucket")) {
1332                         g.v.gens[b & KEY_TYPE_BUCKET_GENS_MASK] = 0;
1333                         need_update = true;
1334                 }
1335
1336         for (b = ca->mi.nbuckets; b < end; b++)
1337                 if (fsck_err_on(g.v.gens[b & KEY_TYPE_BUCKET_GENS_MASK], c,
1338                                 bucket_gens_nonzero_for_invalid_buckets,
1339                                 "bucket_gens key has nonzero gen for invalid bucket")) {
1340                         g.v.gens[b & KEY_TYPE_BUCKET_GENS_MASK] = 0;
1341                         need_update = true;
1342                 }
1343
1344         if (need_update) {
1345                 struct bkey_i *u = bch2_trans_kmalloc(trans, sizeof(g));
1346
1347                 ret = PTR_ERR_OR_ZERO(u);
1348                 if (ret)
1349                         goto out;
1350
1351                 memcpy(u, &g, sizeof(g));
1352                 ret = bch2_trans_update(trans, iter, u, 0);
1353         }
1354 out:
1355 fsck_err:
1356         printbuf_exit(&buf);
1357         return ret;
1358 }
1359
1360 int bch2_check_alloc_info(struct bch_fs *c)
1361 {
1362         struct btree_trans *trans = bch2_trans_get(c);
1363         struct btree_iter iter, discard_iter, freespace_iter, bucket_gens_iter;
1364         struct bkey hole;
1365         struct bkey_s_c k;
1366         int ret = 0;
1367
1368         bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc, POS_MIN,
1369                              BTREE_ITER_PREFETCH);
1370         bch2_trans_iter_init(trans, &discard_iter, BTREE_ID_need_discard, POS_MIN,
1371                              BTREE_ITER_PREFETCH);
1372         bch2_trans_iter_init(trans, &freespace_iter, BTREE_ID_freespace, POS_MIN,
1373                              BTREE_ITER_PREFETCH);
1374         bch2_trans_iter_init(trans, &bucket_gens_iter, BTREE_ID_bucket_gens, POS_MIN,
1375                              BTREE_ITER_PREFETCH);
1376
1377         while (1) {
1378                 struct bpos next;
1379
1380                 bch2_trans_begin(trans);
1381
1382                 k = bch2_get_key_or_real_bucket_hole(&iter, &hole);
1383                 ret = bkey_err(k);
1384                 if (ret)
1385                         goto bkey_err;
1386
1387                 if (!k.k)
1388                         break;
1389
1390                 if (k.k->type) {
1391                         next = bpos_nosnap_successor(k.k->p);
1392
1393                         ret = bch2_check_alloc_key(trans,
1394                                                    k, &iter,
1395                                                    &discard_iter,
1396                                                    &freespace_iter,
1397                                                    &bucket_gens_iter);
1398                         if (ret)
1399                                 goto bkey_err;
1400                 } else {
1401                         next = k.k->p;
1402
1403                         ret = bch2_check_alloc_hole_freespace(trans,
1404                                                     bkey_start_pos(k.k),
1405                                                     &next,
1406                                                     &freespace_iter) ?:
1407                                 bch2_check_alloc_hole_bucket_gens(trans,
1408                                                     bkey_start_pos(k.k),
1409                                                     &next,
1410                                                     &bucket_gens_iter);
1411                         if (ret)
1412                                 goto bkey_err;
1413                 }
1414
1415                 ret = bch2_trans_commit(trans, NULL, NULL,
1416                                         BCH_TRANS_COMMIT_no_enospc|
1417                                         BCH_TRANS_COMMIT_lazy_rw);
1418                 if (ret)
1419                         goto bkey_err;
1420
1421                 bch2_btree_iter_set_pos(&iter, next);
1422 bkey_err:
1423                 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
1424                         continue;
1425                 if (ret)
1426                         break;
1427         }
1428         bch2_trans_iter_exit(trans, &bucket_gens_iter);
1429         bch2_trans_iter_exit(trans, &freespace_iter);
1430         bch2_trans_iter_exit(trans, &discard_iter);
1431         bch2_trans_iter_exit(trans, &iter);
1432
1433         if (ret < 0)
1434                 goto err;
1435
1436         ret = for_each_btree_key2(trans, iter,
1437                         BTREE_ID_need_discard, POS_MIN,
1438                         BTREE_ITER_PREFETCH, k,
1439                 bch2_check_discard_freespace_key(trans, &iter));
1440         if (ret)
1441                 goto err;
1442
1443         bch2_trans_iter_init(trans, &iter, BTREE_ID_freespace, POS_MIN,
1444                              BTREE_ITER_PREFETCH);
1445         while (1) {
1446                 bch2_trans_begin(trans);
1447                 k = bch2_btree_iter_peek(&iter);
1448                 if (!k.k)
1449                         break;
1450
1451                 ret = bkey_err(k) ?:
1452                         bch2_check_discard_freespace_key(trans, &iter);
1453                 if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) {
1454                         ret = 0;
1455                         continue;
1456                 }
1457                 if (ret) {
1458                         struct printbuf buf = PRINTBUF;
1459                         bch2_bkey_val_to_text(&buf, c, k);
1460
1461                         bch_err(c, "while checking %s", buf.buf);
1462                         printbuf_exit(&buf);
1463                         break;
1464                 }
1465
1466                 bch2_btree_iter_set_pos(&iter, bpos_nosnap_successor(iter.pos));
1467         }
1468         bch2_trans_iter_exit(trans, &iter);
1469         if (ret)
1470                 goto err;
1471
1472         ret = for_each_btree_key_commit(trans, iter,
1473                         BTREE_ID_bucket_gens, POS_MIN,
1474                         BTREE_ITER_PREFETCH, k,
1475                         NULL, NULL, BCH_TRANS_COMMIT_no_enospc|BCH_TRANS_COMMIT_lazy_rw,
1476                 bch2_check_bucket_gens_key(trans, &iter, k));
1477 err:
1478         bch2_trans_put(trans);
1479         if (ret)
1480                 bch_err_fn(c, ret);
1481         return ret;
1482 }
1483
1484 static int bch2_check_alloc_to_lru_ref(struct btree_trans *trans,
1485                                        struct btree_iter *alloc_iter)
1486 {
1487         struct bch_fs *c = trans->c;
1488         struct btree_iter lru_iter;
1489         struct bch_alloc_v4 a_convert;
1490         const struct bch_alloc_v4 *a;
1491         struct bkey_s_c alloc_k, lru_k;
1492         struct printbuf buf = PRINTBUF;
1493         int ret;
1494
1495         alloc_k = bch2_btree_iter_peek(alloc_iter);
1496         if (!alloc_k.k)
1497                 return 0;
1498
1499         ret = bkey_err(alloc_k);
1500         if (ret)
1501                 return ret;
1502
1503         a = bch2_alloc_to_v4(alloc_k, &a_convert);
1504
1505         if (a->data_type != BCH_DATA_cached)
1506                 return 0;
1507
1508         if (fsck_err_on(!a->io_time[READ], c,
1509                         alloc_key_cached_but_read_time_zero,
1510                         "cached bucket with read_time 0\n"
1511                         "  %s",
1512                 (printbuf_reset(&buf),
1513                  bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf))) {
1514                 struct bkey_i_alloc_v4 *a_mut =
1515                         bch2_alloc_to_v4_mut(trans, alloc_k);
1516                 ret = PTR_ERR_OR_ZERO(a_mut);
1517                 if (ret)
1518                         goto err;
1519
1520                 a_mut->v.io_time[READ] = atomic64_read(&c->io_clock[READ].now);
1521                 ret = bch2_trans_update(trans, alloc_iter,
1522                                         &a_mut->k_i, BTREE_TRIGGER_NORUN);
1523                 if (ret)
1524                         goto err;
1525
1526                 a = &a_mut->v;
1527         }
1528
1529         lru_k = bch2_bkey_get_iter(trans, &lru_iter, BTREE_ID_lru,
1530                              lru_pos(alloc_k.k->p.inode,
1531                                      bucket_to_u64(alloc_k.k->p),
1532                                      a->io_time[READ]), 0);
1533         ret = bkey_err(lru_k);
1534         if (ret)
1535                 return ret;
1536
1537         if (fsck_err_on(lru_k.k->type != KEY_TYPE_set, c,
1538                         alloc_key_to_missing_lru_entry,
1539                         "missing lru entry\n"
1540                         "  %s",
1541                         (printbuf_reset(&buf),
1542                          bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf))) {
1543                 ret = bch2_lru_set(trans,
1544                                    alloc_k.k->p.inode,
1545                                    bucket_to_u64(alloc_k.k->p),
1546                                    a->io_time[READ]);
1547                 if (ret)
1548                         goto err;
1549         }
1550 err:
1551 fsck_err:
1552         bch2_trans_iter_exit(trans, &lru_iter);
1553         printbuf_exit(&buf);
1554         return ret;
1555 }
1556
1557 int bch2_check_alloc_to_lru_refs(struct bch_fs *c)
1558 {
1559         struct btree_iter iter;
1560         struct bkey_s_c k;
1561
1562         int ret = bch2_trans_run(c,
1563                 for_each_btree_key_commit(trans, iter, BTREE_ID_alloc,
1564                                 POS_MIN, BTREE_ITER_PREFETCH, k,
1565                                 NULL, NULL, BCH_TRANS_COMMIT_no_enospc|BCH_TRANS_COMMIT_lazy_rw,
1566                         bch2_check_alloc_to_lru_ref(trans, &iter)));
1567         bch_err_fn(c, ret);
1568         return ret;
1569 }
1570
1571 static int bch2_discard_one_bucket(struct btree_trans *trans,
1572                                    struct btree_iter *need_discard_iter,
1573                                    struct bpos *discard_pos_done,
1574                                    u64 *seen,
1575                                    u64 *open,
1576                                    u64 *need_journal_commit,
1577                                    u64 *discarded)
1578 {
1579         struct bch_fs *c = trans->c;
1580         struct bpos pos = need_discard_iter->pos;
1581         struct btree_iter iter = { NULL };
1582         struct bkey_s_c k;
1583         struct bch_dev *ca;
1584         struct bkey_i_alloc_v4 *a;
1585         struct printbuf buf = PRINTBUF;
1586         int ret = 0;
1587
1588         ca = bch_dev_bkey_exists(c, pos.inode);
1589         if (!percpu_ref_tryget(&ca->io_ref)) {
1590                 bch2_btree_iter_set_pos(need_discard_iter, POS(pos.inode + 1, 0));
1591                 return 0;
1592         }
1593
1594         if (bch2_bucket_is_open_safe(c, pos.inode, pos.offset)) {
1595                 (*open)++;
1596                 goto out;
1597         }
1598
1599         if (bch2_bucket_needs_journal_commit(&c->buckets_waiting_for_journal,
1600                         c->journal.flushed_seq_ondisk,
1601                         pos.inode, pos.offset)) {
1602                 (*need_journal_commit)++;
1603                 goto out;
1604         }
1605
1606         k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_alloc,
1607                                need_discard_iter->pos,
1608                                BTREE_ITER_CACHED);
1609         ret = bkey_err(k);
1610         if (ret)
1611                 goto out;
1612
1613         a = bch2_alloc_to_v4_mut(trans, k);
1614         ret = PTR_ERR_OR_ZERO(a);
1615         if (ret)
1616                 goto out;
1617
1618         if (BCH_ALLOC_V4_NEED_INC_GEN(&a->v)) {
1619                 a->v.gen++;
1620                 SET_BCH_ALLOC_V4_NEED_INC_GEN(&a->v, false);
1621                 goto write;
1622         }
1623
1624         if (a->v.journal_seq > c->journal.flushed_seq_ondisk) {
1625                 if (c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_info) {
1626                         bch2_trans_inconsistent(trans,
1627                                 "clearing need_discard but journal_seq %llu > flushed_seq %llu\n"
1628                                 "%s",
1629                                 a->v.journal_seq,
1630                                 c->journal.flushed_seq_ondisk,
1631                                 (bch2_bkey_val_to_text(&buf, c, k), buf.buf));
1632                         ret = -EIO;
1633                 }
1634                 goto out;
1635         }
1636
1637         if (a->v.data_type != BCH_DATA_need_discard) {
1638                 if (c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_info) {
1639                         bch2_trans_inconsistent(trans,
1640                                 "bucket incorrectly set in need_discard btree\n"
1641                                 "%s",
1642                                 (bch2_bkey_val_to_text(&buf, c, k), buf.buf));
1643                         ret = -EIO;
1644                 }
1645
1646                 goto out;
1647         }
1648
1649         if (!bkey_eq(*discard_pos_done, iter.pos) &&
1650             ca->mi.discard && !c->opts.nochanges) {
1651                 /*
1652                  * This works without any other locks because this is the only
1653                  * thread that removes items from the need_discard tree
1654                  */
1655                 bch2_trans_unlock(trans);
1656                 blkdev_issue_discard(ca->disk_sb.bdev,
1657                                      k.k->p.offset * ca->mi.bucket_size,
1658                                      ca->mi.bucket_size,
1659                                      GFP_KERNEL);
1660                 *discard_pos_done = iter.pos;
1661
1662                 ret = bch2_trans_relock_notrace(trans);
1663                 if (ret)
1664                         goto out;
1665         }
1666
1667         SET_BCH_ALLOC_V4_NEED_DISCARD(&a->v, false);
1668         a->v.data_type = alloc_data_type(a->v, a->v.data_type);
1669 write:
1670         ret =   bch2_trans_update(trans, &iter, &a->k_i, 0) ?:
1671                 bch2_trans_commit(trans, NULL, NULL,
1672                                   BCH_WATERMARK_btree|
1673                                   BCH_TRANS_COMMIT_no_enospc);
1674         if (ret)
1675                 goto out;
1676
1677         this_cpu_inc(c->counters[BCH_COUNTER_bucket_discard]);
1678         (*discarded)++;
1679 out:
1680         (*seen)++;
1681         bch2_trans_iter_exit(trans, &iter);
1682         percpu_ref_put(&ca->io_ref);
1683         printbuf_exit(&buf);
1684         return ret;
1685 }
1686
1687 static void bch2_do_discards_work(struct work_struct *work)
1688 {
1689         struct bch_fs *c = container_of(work, struct bch_fs, discard_work);
1690         struct btree_iter iter;
1691         struct bkey_s_c k;
1692         u64 seen = 0, open = 0, need_journal_commit = 0, discarded = 0;
1693         struct bpos discard_pos_done = POS_MAX;
1694         int ret;
1695
1696         /*
1697          * We're doing the commit in bch2_discard_one_bucket instead of using
1698          * for_each_btree_key_commit() so that we can increment counters after
1699          * successful commit:
1700          */
1701         ret = bch2_trans_run(c,
1702                 for_each_btree_key2(trans, iter,
1703                                 BTREE_ID_need_discard, POS_MIN, 0, k,
1704                         bch2_discard_one_bucket(trans, &iter, &discard_pos_done,
1705                                                 &seen,
1706                                                 &open,
1707                                                 &need_journal_commit,
1708                                                 &discarded)));
1709
1710         if (need_journal_commit * 2 > seen)
1711                 bch2_journal_flush_async(&c->journal, NULL);
1712
1713         bch2_write_ref_put(c, BCH_WRITE_REF_discard);
1714
1715         trace_discard_buckets(c, seen, open, need_journal_commit, discarded,
1716                               bch2_err_str(ret));
1717 }
1718
1719 void bch2_do_discards(struct bch_fs *c)
1720 {
1721         if (bch2_write_ref_tryget(c, BCH_WRITE_REF_discard) &&
1722             !queue_work(c->write_ref_wq, &c->discard_work))
1723                 bch2_write_ref_put(c, BCH_WRITE_REF_discard);
1724 }
1725
1726 static int invalidate_one_bucket(struct btree_trans *trans,
1727                                  struct btree_iter *lru_iter,
1728                                  struct bkey_s_c lru_k,
1729                                  struct bpos *last_flushed_pos,
1730                                  s64 *nr_to_invalidate)
1731 {
1732         struct bch_fs *c = trans->c;
1733         int ret = 0;
1734
1735         if (*nr_to_invalidate <= 0)
1736                 return 1;
1737
1738         ret = bch2_check_lru_key(trans, lru_iter, lru_k, last_flushed_pos);
1739         if (ret)
1740                 return ret < 0 ? ret : 0;
1741
1742         struct bpos bucket = u64_to_bucket(lru_k.k->p.offset);
1743         if (bch2_bucket_is_open_safe(c, bucket.inode, bucket.offset))
1744                 return 0;
1745
1746         struct btree_iter alloc_iter;
1747         struct bkey_i_alloc_v4 *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         unsigned 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         return ret;
1784 }
1785
1786 static void bch2_do_invalidates_work(struct work_struct *work)
1787 {
1788         struct bch_fs *c = container_of(work, struct bch_fs, invalidate_work);
1789         struct bch_dev *ca;
1790         struct btree_trans *trans = bch2_trans_get(c);
1791         struct btree_iter iter;
1792         struct bkey_s_c k;
1793         struct bpos last_flushed_pos = POS_MIN;
1794         unsigned i;
1795         int ret = 0;
1796
1797         ret = bch2_btree_write_buffer_tryflush(trans);
1798         if (ret)
1799                 goto err;
1800
1801         for_each_member_device(ca, c, i) {
1802                 s64 nr_to_invalidate =
1803                         should_invalidate_buckets(ca, bch2_dev_usage_read(ca));
1804
1805                 ret = for_each_btree_key2_upto(trans, iter, BTREE_ID_lru,
1806                                 lru_pos(ca->dev_idx, 0, 0),
1807                                 lru_pos(ca->dev_idx, U64_MAX, LRU_TIME_MAX),
1808                                 BTREE_ITER_INTENT, k,
1809                         invalidate_one_bucket(trans, &iter, k, &last_flushed_pos,
1810                                               &nr_to_invalidate));
1811
1812                 if (ret < 0) {
1813                         percpu_ref_put(&ca->ref);
1814                         break;
1815                 }
1816         }
1817 err:
1818         bch2_trans_put(trans);
1819         bch2_write_ref_put(c, BCH_WRITE_REF_invalidate);
1820 }
1821
1822 void bch2_do_invalidates(struct bch_fs *c)
1823 {
1824         if (bch2_write_ref_tryget(c, BCH_WRITE_REF_invalidate) &&
1825             !queue_work(c->write_ref_wq, &c->invalidate_work))
1826                 bch2_write_ref_put(c, BCH_WRITE_REF_invalidate);
1827 }
1828
1829 int bch2_dev_freespace_init(struct bch_fs *c, struct bch_dev *ca,
1830                             u64 bucket_start, u64 bucket_end)
1831 {
1832         struct btree_trans *trans = bch2_trans_get(c);
1833         struct btree_iter iter;
1834         struct bkey_s_c k;
1835         struct bkey hole;
1836         struct bpos end = POS(ca->dev_idx, bucket_end);
1837         struct bch_member *m;
1838         unsigned long last_updated = jiffies;
1839         int ret;
1840
1841         BUG_ON(bucket_start > bucket_end);
1842         BUG_ON(bucket_end > ca->mi.nbuckets);
1843
1844         bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc,
1845                 POS(ca->dev_idx, max_t(u64, ca->mi.first_bucket, bucket_start)),
1846                 BTREE_ITER_PREFETCH);
1847         /*
1848          * Scan the alloc btree for every bucket on @ca, and add buckets to the
1849          * freespace/need_discard/need_gc_gens btrees as needed:
1850          */
1851         while (1) {
1852                 if (last_updated + HZ * 10 < jiffies) {
1853                         bch_info(ca, "%s: currently at %llu/%llu",
1854                                  __func__, iter.pos.offset, ca->mi.nbuckets);
1855                         last_updated = jiffies;
1856                 }
1857
1858                 bch2_trans_begin(trans);
1859
1860                 if (bkey_ge(iter.pos, end)) {
1861                         ret = 0;
1862                         break;
1863                 }
1864
1865                 k = bch2_get_key_or_hole(&iter, end, &hole);
1866                 ret = bkey_err(k);
1867                 if (ret)
1868                         goto bkey_err;
1869
1870                 if (k.k->type) {
1871                         /*
1872                          * We process live keys in the alloc btree one at a
1873                          * time:
1874                          */
1875                         struct bch_alloc_v4 a_convert;
1876                         const struct bch_alloc_v4 *a = bch2_alloc_to_v4(k, &a_convert);
1877
1878                         ret =   bch2_bucket_do_index(trans, k, a, true) ?:
1879                                 bch2_trans_commit(trans, NULL, NULL,
1880                                                   BCH_TRANS_COMMIT_lazy_rw|
1881                                                   BCH_TRANS_COMMIT_no_enospc);
1882                         if (ret)
1883                                 goto bkey_err;
1884
1885                         bch2_btree_iter_advance(&iter);
1886                 } else {
1887                         struct bkey_i *freespace;
1888
1889                         freespace = bch2_trans_kmalloc(trans, sizeof(*freespace));
1890                         ret = PTR_ERR_OR_ZERO(freespace);
1891                         if (ret)
1892                                 goto bkey_err;
1893
1894                         bkey_init(&freespace->k);
1895                         freespace->k.type       = KEY_TYPE_set;
1896                         freespace->k.p          = k.k->p;
1897                         freespace->k.size       = k.k->size;
1898
1899                         ret = bch2_btree_insert_trans(trans, BTREE_ID_freespace, freespace, 0) ?:
1900                                 bch2_trans_commit(trans, NULL, NULL,
1901                                                   BCH_TRANS_COMMIT_lazy_rw|
1902                                                   BCH_TRANS_COMMIT_no_enospc);
1903                         if (ret)
1904                                 goto bkey_err;
1905
1906                         bch2_btree_iter_set_pos(&iter, k.k->p);
1907                 }
1908 bkey_err:
1909                 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
1910                         continue;
1911                 if (ret)
1912                         break;
1913         }
1914
1915         bch2_trans_iter_exit(trans, &iter);
1916         bch2_trans_put(trans);
1917
1918         if (ret < 0) {
1919                 bch_err_msg(ca, ret, "initializing free space");
1920                 return ret;
1921         }
1922
1923         mutex_lock(&c->sb_lock);
1924         m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
1925         SET_BCH_MEMBER_FREESPACE_INITIALIZED(m, true);
1926         mutex_unlock(&c->sb_lock);
1927
1928         return 0;
1929 }
1930
1931 int bch2_fs_freespace_init(struct bch_fs *c)
1932 {
1933         struct bch_dev *ca;
1934         unsigned i;
1935         int ret = 0;
1936         bool doing_init = false;
1937
1938         /*
1939          * We can crash during the device add path, so we need to check this on
1940          * every mount:
1941          */
1942
1943         for_each_member_device(ca, c, i) {
1944                 if (ca->mi.freespace_initialized)
1945                         continue;
1946
1947                 if (!doing_init) {
1948                         bch_info(c, "initializing freespace");
1949                         doing_init = true;
1950                 }
1951
1952                 ret = bch2_dev_freespace_init(c, ca, 0, ca->mi.nbuckets);
1953                 if (ret) {
1954                         percpu_ref_put(&ca->ref);
1955                         bch_err_fn(c, ret);
1956                         return ret;
1957                 }
1958         }
1959
1960         if (doing_init) {
1961                 mutex_lock(&c->sb_lock);
1962                 bch2_write_super(c);
1963                 mutex_unlock(&c->sb_lock);
1964                 bch_verbose(c, "done initializing freespace");
1965         }
1966
1967         return 0;
1968 }
1969
1970 /* Bucket IO clocks: */
1971
1972 int bch2_bucket_io_time_reset(struct btree_trans *trans, unsigned dev,
1973                               size_t bucket_nr, int rw)
1974 {
1975         struct bch_fs *c = trans->c;
1976         struct btree_iter iter;
1977         struct bkey_i_alloc_v4 *a;
1978         u64 now;
1979         int ret = 0;
1980
1981         a = bch2_trans_start_alloc_update(trans, &iter,  POS(dev, bucket_nr));
1982         ret = PTR_ERR_OR_ZERO(a);
1983         if (ret)
1984                 return ret;
1985
1986         now = atomic64_read(&c->io_clock[rw].now);
1987         if (a->v.io_time[rw] == now)
1988                 goto out;
1989
1990         a->v.io_time[rw] = now;
1991
1992         ret   = bch2_trans_update(trans, &iter, &a->k_i, 0) ?:
1993                 bch2_trans_commit(trans, NULL, NULL, 0);
1994 out:
1995         bch2_trans_iter_exit(trans, &iter);
1996         return ret;
1997 }
1998
1999 /* Startup/shutdown (ro/rw): */
2000
2001 void bch2_recalc_capacity(struct bch_fs *c)
2002 {
2003         struct bch_dev *ca;
2004         u64 capacity = 0, reserved_sectors = 0, gc_reserve;
2005         unsigned bucket_size_max = 0;
2006         unsigned long ra_pages = 0;
2007         unsigned i;
2008
2009         lockdep_assert_held(&c->state_lock);
2010
2011         for_each_online_member(ca, c, i) {
2012                 struct backing_dev_info *bdi = ca->disk_sb.bdev->bd_disk->bdi;
2013
2014                 ra_pages += bdi->ra_pages;
2015         }
2016
2017         bch2_set_ra_pages(c, ra_pages);
2018
2019         for_each_rw_member(ca, c, i) {
2020                 u64 dev_reserve = 0;
2021
2022                 /*
2023                  * We need to reserve buckets (from the number
2024                  * of currently available buckets) against
2025                  * foreground writes so that mainly copygc can
2026                  * make forward progress.
2027                  *
2028                  * We need enough to refill the various reserves
2029                  * from scratch - copygc will use its entire
2030                  * reserve all at once, then run against when
2031                  * its reserve is refilled (from the formerly
2032                  * available buckets).
2033                  *
2034                  * This reserve is just used when considering if
2035                  * allocations for foreground writes must wait -
2036                  * not -ENOSPC calculations.
2037                  */
2038
2039                 dev_reserve += ca->nr_btree_reserve * 2;
2040                 dev_reserve += ca->mi.nbuckets >> 6; /* copygc reserve */
2041
2042                 dev_reserve += 1;       /* btree write point */
2043                 dev_reserve += 1;       /* copygc write point */
2044                 dev_reserve += 1;       /* rebalance write point */
2045
2046                 dev_reserve *= ca->mi.bucket_size;
2047
2048                 capacity += bucket_to_sector(ca, ca->mi.nbuckets -
2049                                              ca->mi.first_bucket);
2050
2051                 reserved_sectors += dev_reserve * 2;
2052
2053                 bucket_size_max = max_t(unsigned, bucket_size_max,
2054                                         ca->mi.bucket_size);
2055         }
2056
2057         gc_reserve = c->opts.gc_reserve_bytes
2058                 ? c->opts.gc_reserve_bytes >> 9
2059                 : div64_u64(capacity * c->opts.gc_reserve_percent, 100);
2060
2061         reserved_sectors = max(gc_reserve, reserved_sectors);
2062
2063         reserved_sectors = min(reserved_sectors, capacity);
2064
2065         c->capacity = capacity - reserved_sectors;
2066
2067         c->bucket_size_max = bucket_size_max;
2068
2069         /* Wake up case someone was waiting for buckets */
2070         closure_wake_up(&c->freelist_wait);
2071 }
2072
2073 u64 bch2_min_rw_member_capacity(struct bch_fs *c)
2074 {
2075         struct bch_dev *ca;
2076         unsigned i;
2077         u64 ret = U64_MAX;
2078
2079         for_each_rw_member(ca, c, i)
2080                 ret = min(ret, ca->mi.nbuckets * ca->mi.bucket_size);
2081         return ret;
2082 }
2083
2084 static bool bch2_dev_has_open_write_point(struct bch_fs *c, struct bch_dev *ca)
2085 {
2086         struct open_bucket *ob;
2087         bool ret = false;
2088
2089         for (ob = c->open_buckets;
2090              ob < c->open_buckets + ARRAY_SIZE(c->open_buckets);
2091              ob++) {
2092                 spin_lock(&ob->lock);
2093                 if (ob->valid && !ob->on_partial_list &&
2094                     ob->dev == ca->dev_idx)
2095                         ret = true;
2096                 spin_unlock(&ob->lock);
2097         }
2098
2099         return ret;
2100 }
2101
2102 /* device goes ro: */
2103 void bch2_dev_allocator_remove(struct bch_fs *c, struct bch_dev *ca)
2104 {
2105         unsigned i;
2106
2107         /* First, remove device from allocation groups: */
2108
2109         for (i = 0; i < ARRAY_SIZE(c->rw_devs); i++)
2110                 clear_bit(ca->dev_idx, c->rw_devs[i].d);
2111
2112         /*
2113          * Capacity is calculated based off of devices in allocation groups:
2114          */
2115         bch2_recalc_capacity(c);
2116
2117         bch2_open_buckets_stop(c, ca, false);
2118
2119         /*
2120          * Wake up threads that were blocked on allocation, so they can notice
2121          * the device can no longer be removed and the capacity has changed:
2122          */
2123         closure_wake_up(&c->freelist_wait);
2124
2125         /*
2126          * journal_res_get() can block waiting for free space in the journal -
2127          * it needs to notice there may not be devices to allocate from anymore:
2128          */
2129         wake_up(&c->journal.wait);
2130
2131         /* Now wait for any in flight writes: */
2132
2133         closure_wait_event(&c->open_buckets_wait,
2134                            !bch2_dev_has_open_write_point(c, ca));
2135 }
2136
2137 /* device goes rw: */
2138 void bch2_dev_allocator_add(struct bch_fs *c, struct bch_dev *ca)
2139 {
2140         unsigned i;
2141
2142         for (i = 0; i < ARRAY_SIZE(c->rw_devs); i++)
2143                 if (ca->mi.data_allowed & (1 << i))
2144                         set_bit(ca->dev_idx, c->rw_devs[i].d);
2145 }
2146
2147 void bch2_fs_allocator_background_init(struct bch_fs *c)
2148 {
2149         spin_lock_init(&c->freelist_lock);
2150         INIT_WORK(&c->discard_work, bch2_do_discards_work);
2151         INIT_WORK(&c->invalidate_work, bch2_do_invalidates_work);
2152 }