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(a.v->dirty_sectors ||
 265                                  a.v->cached_sectors ||
 266                                  a.v->stripe, c, err,
 267                                  alloc_key_empty_but_have_data,
 268                                  "empty data type free but have data");
 269                 break;
 270         case BCH_DATA_sb:
 271         case BCH_DATA_journal:
 272         case BCH_DATA_btree:
 273         case BCH_DATA_user:
 274         case BCH_DATA_parity:
 275                 bkey_fsck_err_on(!a.v->dirty_sectors, c, err,
 276                                  alloc_key_dirty_sectors_0,
 277                                  "data_type %s but dirty_sectors==0",
 278                                  bch2_data_types[a.v->data_type]);
 279                 break;
 280         case BCH_DATA_cached:
 281                 bkey_fsck_err_on(!a.v->cached_sectors ||
 282                                  a.v->dirty_sectors ||
 283                                  a.v->stripe, c, err,
 284                                  alloc_key_cached_inconsistency,
 285                                  "data type inconsistency");
 286
 287                 bkey_fsck_err_on(!a.v->io_time[READ] &&
 288                                  c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_to_lru_refs,
 289                                  c, err,
 290                                  alloc_key_cached_but_read_time_zero,
 291                                  "cached bucket with read_time == 0");
 292                 break;
 293         case BCH_DATA_stripe:
 294                 break;
 295         }
 296 fsck_err:
 297         return ret;
 298 }
 299
 300 void bch2_alloc_v4_swab(struct bkey_s k)
 301 {
 302         struct bch_alloc_v4 *a = bkey_s_to_alloc_v4(k).v;
 303         struct bch_backpointer *bp, *bps;
 304
 305         a->journal_seq          = swab64(a->journal_seq);
 306         a->flags                = swab32(a->flags);
 307         a->dirty_sectors        = swab32(a->dirty_sectors);
 308         a->cached_sectors       = swab32(a->cached_sectors);
 309         a->io_time[0]           = swab64(a->io_time[0]);
 310         a->io_time[1]           = swab64(a->io_time[1]);
 311         a->stripe               = swab32(a->stripe);
 312         a->nr_external_backpointers = swab32(a->nr_external_backpointers);
 313         a->fragmentation_lru    = swab64(a->fragmentation_lru);
 314
 315         bps = alloc_v4_backpointers(a);
 316         for (bp = bps; bp < bps + BCH_ALLOC_V4_NR_BACKPOINTERS(a); bp++) {
 317                 bp->bucket_offset       = swab40(bp->bucket_offset);
 318                 bp->bucket_len          = swab32(bp->bucket_len);
 319                 bch2_bpos_swab(&bp->pos);
 320         }
 321 }
 322
 323 void bch2_alloc_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c k)
 324 {
 325         struct bch_alloc_v4 _a;
 326         const struct bch_alloc_v4 *a = bch2_alloc_to_v4(k, &_a);
 327         unsigned i;
 328
 329         prt_newline(out);
 330         printbuf_indent_add(out, 2);
 331
 332         prt_printf(out, "gen %u oldest_gen %u data_type %s",
 333                a->gen, a->oldest_gen,
 334                a->data_type < BCH_DATA_NR
 335                ? bch2_data_types[a->data_type]
 336                : "(invalid data type)");
 337         prt_newline(out);
 338         prt_printf(out, "journal_seq       %llu",       a->journal_seq);
 339         prt_newline(out);
 340         prt_printf(out, "need_discard      %llu",       BCH_ALLOC_V4_NEED_DISCARD(a));
 341         prt_newline(out);
 342         prt_printf(out, "need_inc_gen      %llu",       BCH_ALLOC_V4_NEED_INC_GEN(a));
 343         prt_newline(out);
 344         prt_printf(out, "dirty_sectors     %u", a->dirty_sectors);
 345         prt_newline(out);
 346         prt_printf(out, "cached_sectors    %u", a->cached_sectors);
 347         prt_newline(out);
 348         prt_printf(out, "stripe            %u", a->stripe);
 349         prt_newline(out);
 350         prt_printf(out, "stripe_redundancy %u", a->stripe_redundancy);
 351         prt_newline(out);
 352         prt_printf(out, "io_time[READ]     %llu",       a->io_time[READ]);
 353         prt_newline(out);
 354         prt_printf(out, "io_time[WRITE]    %llu",       a->io_time[WRITE]);
 355         prt_newline(out);
 356         prt_printf(out, "fragmentation     %llu",       a->fragmentation_lru);
 357         prt_newline(out);
 358         prt_printf(out, "bp_start          %llu", BCH_ALLOC_V4_BACKPOINTERS_START(a));
 359         prt_newline(out);
 360
 361         if (BCH_ALLOC_V4_NR_BACKPOINTERS(a)) {
 362                 struct bkey_s_c_alloc_v4 a_raw = bkey_s_c_to_alloc_v4(k);
 363                 const struct bch_backpointer *bps = alloc_v4_backpointers_c(a_raw.v);
 364
 365                 prt_printf(out, "backpointers:     %llu", BCH_ALLOC_V4_NR_BACKPOINTERS(a_raw.v));
 366                 printbuf_indent_add(out, 2);
 367
 368                 for (i = 0; i < BCH_ALLOC_V4_NR_BACKPOINTERS(a_raw.v); i++) {
 369                         prt_newline(out);
 370                         bch2_backpointer_to_text(out, &bps[i]);
 371                 }
 372
 373                 printbuf_indent_sub(out, 2);
 374         }
 375
 376         printbuf_indent_sub(out, 2);
 377 }
 378
 379 void __bch2_alloc_to_v4(struct bkey_s_c k, struct bch_alloc_v4 *out)
 380 {
 381         if (k.k->type == KEY_TYPE_alloc_v4) {
 382                 void *src, *dst;
 383
 384                 *out = *bkey_s_c_to_alloc_v4(k).v;
 385
 386                 src = alloc_v4_backpointers(out);
 387                 SET_BCH_ALLOC_V4_BACKPOINTERS_START(out, BCH_ALLOC_V4_U64s);
 388                 dst = alloc_v4_backpointers(out);
 389
 390                 if (src < dst)
 391                         memset(src, 0, dst - src);
 392
 393                 SET_BCH_ALLOC_V4_NR_BACKPOINTERS(out, 0);
 394         } else {
 395                 struct bkey_alloc_unpacked u = bch2_alloc_unpack(k);
 396
 397                 *out = (struct bch_alloc_v4) {
 398                         .journal_seq            = u.journal_seq,
 399                         .flags                  = u.need_discard,
 400                         .gen                    = u.gen,
 401                         .oldest_gen             = u.oldest_gen,
 402                         .data_type              = u.data_type,
 403                         .stripe_redundancy      = u.stripe_redundancy,
 404                         .dirty_sectors          = u.dirty_sectors,
 405                         .cached_sectors         = u.cached_sectors,
 406                         .io_time[READ]          = u.read_time,
 407                         .io_time[WRITE]         = u.write_time,
 408                         .stripe                 = u.stripe,
 409                 };
 410
 411                 SET_BCH_ALLOC_V4_BACKPOINTERS_START(out, BCH_ALLOC_V4_U64s);
 412         }
 413 }
 414
 415 static noinline struct bkey_i_alloc_v4 *
 416 __bch2_alloc_to_v4_mut(struct btree_trans *trans, struct bkey_s_c k)
 417 {
 418         struct bkey_i_alloc_v4 *ret;
 419
 420         ret = bch2_trans_kmalloc(trans, max(bkey_bytes(k.k), sizeof(struct bkey_i_alloc_v4)));
 421         if (IS_ERR(ret))
 422                 return ret;
 423
 424         if (k.k->type == KEY_TYPE_alloc_v4) {
 425                 void *src, *dst;
 426
 427                 bkey_reassemble(&ret->k_i, k);
 428
 429                 src = alloc_v4_backpointers(&ret->v);
 430                 SET_BCH_ALLOC_V4_BACKPOINTERS_START(&ret->v, BCH_ALLOC_V4_U64s);
 431                 dst = alloc_v4_backpointers(&ret->v);
 432
 433                 if (src < dst)
 434                         memset(src, 0, dst - src);
 435
 436                 SET_BCH_ALLOC_V4_NR_BACKPOINTERS(&ret->v, 0);
 437                 set_alloc_v4_u64s(ret);
 438         } else {
 439                 bkey_alloc_v4_init(&ret->k_i);
 440                 ret->k.p = k.k->p;
 441                 bch2_alloc_to_v4(k, &ret->v);
 442         }
 443         return ret;
 444 }
 445
 446 static inline struct bkey_i_alloc_v4 *bch2_alloc_to_v4_mut_inlined(struct btree_trans *trans, struct bkey_s_c k)
 447 {
 448         struct bkey_s_c_alloc_v4 a;
 449
 450         if (likely(k.k->type == KEY_TYPE_alloc_v4) &&
 451             ((a = bkey_s_c_to_alloc_v4(k), true) &&
 452              BCH_ALLOC_V4_NR_BACKPOINTERS(a.v) == 0))
 453                 return bch2_bkey_make_mut_noupdate_typed(trans, k, alloc_v4);
 454
 455         return __bch2_alloc_to_v4_mut(trans, k);
 456 }
 457
 458 struct bkey_i_alloc_v4 *bch2_alloc_to_v4_mut(struct btree_trans *trans, struct bkey_s_c k)
 459 {
 460         return bch2_alloc_to_v4_mut_inlined(trans, k);
 461 }
 462
 463 struct bkey_i_alloc_v4 *
 464 bch2_trans_start_alloc_update(struct btree_trans *trans, struct btree_iter *iter,
 465                               struct bpos pos)
 466 {
 467         struct bkey_s_c k;
 468         struct bkey_i_alloc_v4 *a;
 469         int ret;
 470
 471         k = bch2_bkey_get_iter(trans, iter, BTREE_ID_alloc, pos,
 472                              BTREE_ITER_WITH_UPDATES|
 473                              BTREE_ITER_CACHED|
 474                              BTREE_ITER_INTENT);
 475         ret = bkey_err(k);
 476         if (unlikely(ret))
 477                 return ERR_PTR(ret);
 478
 479         a = bch2_alloc_to_v4_mut_inlined(trans, k);
 480         ret = PTR_ERR_OR_ZERO(a);
 481         if (unlikely(ret))
 482                 goto err;
 483         return a;
 484 err:
 485         bch2_trans_iter_exit(trans, iter);
 486         return ERR_PTR(ret);
 487 }
 488
 489 static struct bpos alloc_gens_pos(struct bpos pos, unsigned *offset)
 490 {
 491         *offset = pos.offset & KEY_TYPE_BUCKET_GENS_MASK;
 492
 493         pos.offset >>= KEY_TYPE_BUCKET_GENS_BITS;
 494         return pos;
 495 }
 496
 497 static struct bpos bucket_gens_pos_to_alloc(struct bpos pos, unsigned offset)
 498 {
 499         pos.offset <<= KEY_TYPE_BUCKET_GENS_BITS;
 500         pos.offset += offset;
 501         return pos;
 502 }
 503
 504 static unsigned alloc_gen(struct bkey_s_c k, unsigned offset)
 505 {
 506         return k.k->type == KEY_TYPE_bucket_gens
 507                 ? bkey_s_c_to_bucket_gens(k).v->gens[offset]
 508                 : 0;
 509 }
 510
 511 int bch2_bucket_gens_invalid(struct bch_fs *c, struct bkey_s_c k,
 512                              enum bkey_invalid_flags flags,
 513                              struct printbuf *err)
 514 {
 515         int ret = 0;
 516
 517         bkey_fsck_err_on(bkey_val_bytes(k.k) != sizeof(struct bch_bucket_gens), c, err,
 518                          bucket_gens_val_size_bad,
 519                          "bad val size (%zu != %zu)",
 520                          bkey_val_bytes(k.k), sizeof(struct bch_bucket_gens));
 521 fsck_err:
 522         return ret;
 523 }
 524
 525 void bch2_bucket_gens_to_text(struct printbuf *out, struct bch_fs *c, struct bkey_s_c k)
 526 {
 527         struct bkey_s_c_bucket_gens g = bkey_s_c_to_bucket_gens(k);
 528         unsigned i;
 529
 530         for (i = 0; i < ARRAY_SIZE(g.v->gens); i++) {
 531                 if (i)
 532                         prt_char(out, ' ');
 533                 prt_printf(out, "%u", g.v->gens[i]);
 534         }
 535 }
 536
 537 int bch2_bucket_gens_init(struct bch_fs *c)
 538 {
 539         struct btree_trans *trans = bch2_trans_get(c);
 540         struct btree_iter iter;
 541         struct bkey_s_c k;
 542         struct bch_alloc_v4 a;
 543         struct bkey_i_bucket_gens g;
 544         bool have_bucket_gens_key = false;
 545         unsigned offset;
 546         struct bpos pos;
 547         u8 gen;
 548         int ret;
 549
 550         for_each_btree_key(trans, iter, BTREE_ID_alloc, POS_MIN,
 551                            BTREE_ITER_PREFETCH, k, ret) {
 552                 /*
 553                  * Not a fsck error because this is checked/repaired by
 554                  * bch2_check_alloc_key() which runs later:
 555                  */
 556                 if (!bch2_dev_bucket_exists(c, k.k->p))
 557                         continue;
 558
 559                 gen = bch2_alloc_to_v4(k, &a)->gen;
 560                 pos = alloc_gens_pos(iter.pos, &offset);
 561
 562                 if (have_bucket_gens_key && bkey_cmp(iter.pos, pos)) {
 563                         ret = commit_do(trans, NULL, NULL,
 564                                         BCH_TRANS_COMMIT_no_enospc|
 565                                         BCH_TRANS_COMMIT_lazy_rw,
 566                                 bch2_btree_insert_trans(trans, BTREE_ID_bucket_gens, &g.k_i, 0));
 567                         if (ret)
 568                                 break;
 569                         have_bucket_gens_key = false;
 570                 }
 571
 572                 if (!have_bucket_gens_key) {
 573                         bkey_bucket_gens_init(&g.k_i);
 574                         g.k.p = pos;
 575                         have_bucket_gens_key = true;
 576                 }
 577
 578                 g.v.gens[offset] = gen;
 579         }
 580         bch2_trans_iter_exit(trans, &iter);
 581
 582         if (have_bucket_gens_key && !ret)
 583                 ret = commit_do(trans, NULL, NULL,
 584                                 BCH_TRANS_COMMIT_no_enospc|
 585                                 BCH_TRANS_COMMIT_lazy_rw,
 586                         bch2_btree_insert_trans(trans, BTREE_ID_bucket_gens, &g.k_i, 0));
 587
 588         bch2_trans_put(trans);
 589
 590         if (ret)
 591                 bch_err_fn(c, ret);
 592         return ret;
 593 }
 594
 595 int bch2_alloc_read(struct bch_fs *c)
 596 {
 597         struct btree_trans *trans = bch2_trans_get(c);
 598         struct btree_iter iter;
 599         struct bkey_s_c k;
 600         struct bch_dev *ca;
 601         int ret;
 602
 603         down_read(&c->gc_lock);
 604
 605         if (c->sb.version_upgrade_complete >= bcachefs_metadata_version_bucket_gens) {
 606                 const struct bch_bucket_gens *g;
 607                 u64 b;
 608
 609                 for_each_btree_key(trans, iter, BTREE_ID_bucket_gens, POS_MIN,
 610                                    BTREE_ITER_PREFETCH, k, ret) {
 611                         u64 start = bucket_gens_pos_to_alloc(k.k->p, 0).offset;
 612                         u64 end = bucket_gens_pos_to_alloc(bpos_nosnap_successor(k.k->p), 0).offset;
 613
 614                         if (k.k->type != KEY_TYPE_bucket_gens)
 615                                 continue;
 616
 617                         g = bkey_s_c_to_bucket_gens(k).v;
 618
 619                         /*
 620                          * Not a fsck error because this is checked/repaired by
 621                          * bch2_check_alloc_key() which runs later:
 622                          */
 623                         if (!bch2_dev_exists2(c, k.k->p.inode))
 624                                 continue;
 625
 626                         ca = bch_dev_bkey_exists(c, k.k->p.inode);
 627
 628                         for (b = max_t(u64, ca->mi.first_bucket, start);
 629                              b < min_t(u64, ca->mi.nbuckets, end);
 630                              b++)
 631                                 *bucket_gen(ca, b) = g->gens[b & KEY_TYPE_BUCKET_GENS_MASK];
 632                 }
 633                 bch2_trans_iter_exit(trans, &iter);
 634         } else {
 635                 struct bch_alloc_v4 a;
 636
 637                 for_each_btree_key(trans, iter, BTREE_ID_alloc, POS_MIN,
 638                                    BTREE_ITER_PREFETCH, k, ret) {
 639                         /*
 640                          * Not a fsck error because this is checked/repaired by
 641                          * bch2_check_alloc_key() which runs later:
 642                          */
 643                         if (!bch2_dev_bucket_exists(c, k.k->p))
 644                                 continue;
 645
 646                         ca = bch_dev_bkey_exists(c, k.k->p.inode);
 647
 648                         *bucket_gen(ca, k.k->p.offset) = bch2_alloc_to_v4(k, &a)->gen;
 649                 }
 650                 bch2_trans_iter_exit(trans, &iter);
 651         }
 652
 653         bch2_trans_put(trans);
 654         up_read(&c->gc_lock);
 655
 656         if (ret)
 657                 bch_err_fn(c, ret);
 658
 659         return ret;
 660 }
 661
 662 /* Free space/discard btree: */
 663
 664 static int bch2_bucket_do_index(struct btree_trans *trans,
 665                                 struct bkey_s_c alloc_k,
 666                                 const struct bch_alloc_v4 *a,
 667                                 bool set)
 668 {
 669         struct bch_fs *c = trans->c;
 670         struct bch_dev *ca = bch_dev_bkey_exists(c, alloc_k.k->p.inode);
 671         struct btree_iter iter;
 672         struct bkey_s_c old;
 673         struct bkey_i *k;
 674         enum btree_id btree;
 675         enum bch_bkey_type old_type = !set ? KEY_TYPE_set : KEY_TYPE_deleted;
 676         enum bch_bkey_type new_type =  set ? KEY_TYPE_set : KEY_TYPE_deleted;
 677         struct printbuf buf = PRINTBUF;
 678         int ret;
 679
 680         if (a->data_type != BCH_DATA_free &&
 681             a->data_type != BCH_DATA_need_discard)
 682                 return 0;
 683
 684         k = bch2_trans_kmalloc_nomemzero(trans, sizeof(*k));
 685         if (IS_ERR(k))
 686                 return PTR_ERR(k);
 687
 688         bkey_init(&k->k);
 689         k->k.type = new_type;
 690
 691         switch (a->data_type) {
 692         case BCH_DATA_free:
 693                 btree = BTREE_ID_freespace;
 694                 k->k.p = alloc_freespace_pos(alloc_k.k->p, *a);
 695                 bch2_key_resize(&k->k, 1);
 696                 break;
 697         case BCH_DATA_need_discard:
 698                 btree = BTREE_ID_need_discard;
 699                 k->k.p = alloc_k.k->p;
 700                 break;
 701         default:
 702                 return 0;
 703         }
 704
 705         old = bch2_bkey_get_iter(trans, &iter, btree,
 706                              bkey_start_pos(&k->k),
 707                              BTREE_ITER_INTENT);
 708         ret = bkey_err(old);
 709         if (ret)
 710                 return ret;
 711
 712         if (ca->mi.freespace_initialized &&
 713             c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_info &&
 714             bch2_trans_inconsistent_on(old.k->type != old_type, trans,
 715                         "incorrect key when %s %s:%llu:%llu:0 (got %s should be %s)\n"
 716                         "  for %s",
 717                         set ? "setting" : "clearing",
 718                         bch2_btree_id_str(btree),
 719                         iter.pos.inode,
 720                         iter.pos.offset,
 721                         bch2_bkey_types[old.k->type],
 722                         bch2_bkey_types[old_type],
 723                         (bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf))) {
 724                 ret = -EIO;
 725                 goto err;
 726         }
 727
 728         ret = bch2_trans_update(trans, &iter, k, 0);
 729 err:
 730         bch2_trans_iter_exit(trans, &iter);
 731         printbuf_exit(&buf);
 732         return ret;
 733 }
 734
 735 static noinline int bch2_bucket_gen_update(struct btree_trans *trans,
 736                                            struct bpos bucket, u8 gen)
 737 {
 738         struct btree_iter iter;
 739         unsigned offset;
 740         struct bpos pos = alloc_gens_pos(bucket, &offset);
 741         struct bkey_i_bucket_gens *g;
 742         struct bkey_s_c k;
 743         int ret;
 744
 745         g = bch2_trans_kmalloc(trans, sizeof(*g));
 746         ret = PTR_ERR_OR_ZERO(g);
 747         if (ret)
 748                 return ret;
 749
 750         k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_bucket_gens, pos,
 751                                BTREE_ITER_INTENT|
 752                                BTREE_ITER_WITH_UPDATES);
 753         ret = bkey_err(k);
 754         if (ret)
 755                 return ret;
 756
 757         if (k.k->type != KEY_TYPE_bucket_gens) {
 758                 bkey_bucket_gens_init(&g->k_i);
 759                 g->k.p = iter.pos;
 760         } else {
 761                 bkey_reassemble(&g->k_i, k);
 762         }
 763
 764         g->v.gens[offset] = gen;
 765
 766         ret = bch2_trans_update(trans, &iter, &g->k_i, 0);
 767         bch2_trans_iter_exit(trans, &iter);
 768         return ret;
 769 }
 770
 771 int bch2_trans_mark_alloc(struct btree_trans *trans,
 772                           enum btree_id btree_id, unsigned level,
 773                           struct bkey_s_c old, struct bkey_i *new,
 774                           unsigned flags)
 775 {
 776         struct bch_fs *c = trans->c;
 777         struct bch_alloc_v4 old_a_convert, *new_a;
 778         const struct bch_alloc_v4 *old_a;
 779         u64 old_lru, new_lru;
 780         int ret = 0;
 781
 782         /*
 783          * Deletion only happens in the device removal path, with
 784          * BTREE_TRIGGER_NORUN:
 785          */
 786         BUG_ON(new->k.type != KEY_TYPE_alloc_v4);
 787
 788         old_a = bch2_alloc_to_v4(old, &old_a_convert);
 789         new_a = &bkey_i_to_alloc_v4(new)->v;
 790
 791         new_a->data_type = alloc_data_type(*new_a, new_a->data_type);
 792
 793         if (new_a->dirty_sectors > old_a->dirty_sectors ||
 794             new_a->cached_sectors > old_a->cached_sectors) {
 795                 new_a->io_time[READ] = max_t(u64, 1, atomic64_read(&c->io_clock[READ].now));
 796                 new_a->io_time[WRITE]= max_t(u64, 1, atomic64_read(&c->io_clock[WRITE].now));
 797                 SET_BCH_ALLOC_V4_NEED_INC_GEN(new_a, true);
 798                 SET_BCH_ALLOC_V4_NEED_DISCARD(new_a, true);
 799         }
 800
 801         if (data_type_is_empty(new_a->data_type) &&
 802             BCH_ALLOC_V4_NEED_INC_GEN(new_a) &&
 803             !bch2_bucket_is_open_safe(c, new->k.p.inode, new->k.p.offset)) {
 804                 new_a->gen++;
 805                 SET_BCH_ALLOC_V4_NEED_INC_GEN(new_a, false);
 806         }
 807
 808         if (old_a->data_type != new_a->data_type ||
 809             (new_a->data_type == BCH_DATA_free &&
 810              alloc_freespace_genbits(*old_a) != alloc_freespace_genbits(*new_a))) {
 811                 ret =   bch2_bucket_do_index(trans, old, old_a, false) ?:
 812                         bch2_bucket_do_index(trans, bkey_i_to_s_c(new), new_a, true);
 813                 if (ret)
 814                         return ret;
 815         }
 816
 817         if (new_a->data_type == BCH_DATA_cached &&
 818             !new_a->io_time[READ])
 819                 new_a->io_time[READ] = max_t(u64, 1, atomic64_read(&c->io_clock[READ].now));
 820
 821         old_lru = alloc_lru_idx_read(*old_a);
 822         new_lru = alloc_lru_idx_read(*new_a);
 823
 824         if (old_lru != new_lru) {
 825                 ret = bch2_lru_change(trans, new->k.p.inode,
 826                                       bucket_to_u64(new->k.p),
 827                                       old_lru, new_lru);
 828                 if (ret)
 829                         return ret;
 830         }
 831
 832         new_a->fragmentation_lru = alloc_lru_idx_fragmentation(*new_a,
 833                                         bch_dev_bkey_exists(c, new->k.p.inode));
 834
 835         if (old_a->fragmentation_lru != new_a->fragmentation_lru) {
 836                 ret = bch2_lru_change(trans,
 837                                 BCH_LRU_FRAGMENTATION_START,
 838                                 bucket_to_u64(new->k.p),
 839                                 old_a->fragmentation_lru, new_a->fragmentation_lru);
 840                 if (ret)
 841                         return ret;
 842         }
 843
 844         if (old_a->gen != new_a->gen) {
 845                 ret = bch2_bucket_gen_update(trans, new->k.p, new_a->gen);
 846                 if (ret)
 847                         return ret;
 848         }
 849
 850         /*
 851          * need to know if we're getting called from the invalidate path or
 852          * not:
 853          */
 854
 855         if ((flags & BTREE_TRIGGER_BUCKET_INVALIDATE) &&
 856             old_a->cached_sectors) {
 857                 ret = bch2_update_cached_sectors_list(trans, new->k.p.inode,
 858                                                       -((s64) old_a->cached_sectors));
 859                 if (ret)
 860                         return ret;
 861         }
 862
 863         return 0;
 864 }
 865
 866 /*
 867  * This synthesizes deleted extents for holes, similar to BTREE_ITER_SLOTS for
 868  * extents style btrees, but works on non-extents btrees:
 869  */
 870 static struct bkey_s_c bch2_get_key_or_hole(struct btree_iter *iter, struct bpos end, struct bkey *hole)
 871 {
 872         struct bkey_s_c k = bch2_btree_iter_peek_slot(iter);
 873
 874         if (bkey_err(k))
 875                 return k;
 876
 877         if (k.k->type) {
 878                 return k;
 879         } else {
 880                 struct btree_iter iter2;
 881                 struct bpos next;
 882
 883                 bch2_trans_copy_iter(&iter2, iter);
 884
 885                 if (!bpos_eq(iter->path->l[0].b->key.k.p, SPOS_MAX))
 886                         end = bkey_min(end, bpos_nosnap_successor(iter->path->l[0].b->key.k.p));
 887
 888                 end = bkey_min(end, POS(iter->pos.inode, iter->pos.offset + U32_MAX - 1));
 889
 890                 /*
 891                  * btree node min/max is a closed interval, upto takes a half
 892                  * open interval:
 893                  */
 894                 k = bch2_btree_iter_peek_upto(&iter2, end);
 895                 next = iter2.pos;
 896                 bch2_trans_iter_exit(iter->trans, &iter2);
 897
 898                 BUG_ON(next.offset >= iter->pos.offset + U32_MAX);
 899
 900                 if (bkey_err(k))
 901                         return k;
 902
 903                 bkey_init(hole);
 904                 hole->p = iter->pos;
 905
 906                 bch2_key_resize(hole, next.offset - iter->pos.offset);
 907                 return (struct bkey_s_c) { hole, NULL };
 908         }
 909 }
 910
 911 static bool next_bucket(struct bch_fs *c, struct bpos *bucket)
 912 {
 913         struct bch_dev *ca;
 914         unsigned iter;
 915
 916         if (bch2_dev_bucket_exists(c, *bucket))
 917                 return true;
 918
 919         if (bch2_dev_exists2(c, bucket->inode)) {
 920                 ca = bch_dev_bkey_exists(c, bucket->inode);
 921
 922                 if (bucket->offset < ca->mi.first_bucket) {
 923                         bucket->offset = ca->mi.first_bucket;
 924                         return true;
 925                 }
 926
 927                 bucket->inode++;
 928                 bucket->offset = 0;
 929         }
 930
 931         rcu_read_lock();
 932         iter = bucket->inode;
 933         ca = __bch2_next_dev(c, &iter, NULL);
 934         if (ca)
 935                 *bucket = POS(ca->dev_idx, ca->mi.first_bucket);
 936         rcu_read_unlock();
 937
 938         return ca != NULL;
 939 }
 940
 941 static struct bkey_s_c bch2_get_key_or_real_bucket_hole(struct btree_iter *iter, struct bkey *hole)
 942 {
 943         struct bch_fs *c = iter->trans->c;
 944         struct bkey_s_c k;
 945 again:
 946         k = bch2_get_key_or_hole(iter, POS_MAX, hole);
 947         if (bkey_err(k))
 948                 return k;
 949
 950         if (!k.k->type) {
 951                 struct bpos bucket = bkey_start_pos(k.k);
 952
 953                 if (!bch2_dev_bucket_exists(c, bucket)) {
 954                         if (!next_bucket(c, &bucket))
 955                                 return bkey_s_c_null;
 956
 957                         bch2_btree_iter_set_pos(iter, bucket);
 958                         goto again;
 959                 }
 960
 961                 if (!bch2_dev_bucket_exists(c, k.k->p)) {
 962                         struct bch_dev *ca = bch_dev_bkey_exists(c, bucket.inode);
 963
 964                         bch2_key_resize(hole, ca->mi.nbuckets - bucket.offset);
 965                 }
 966         }
 967
 968         return k;
 969 }
 970
 971 static noinline_for_stack
 972 int bch2_check_alloc_key(struct btree_trans *trans,
 973                          struct bkey_s_c alloc_k,
 974                          struct btree_iter *alloc_iter,
 975                          struct btree_iter *discard_iter,
 976                          struct btree_iter *freespace_iter,
 977                          struct btree_iter *bucket_gens_iter)
 978 {
 979         struct bch_fs *c = trans->c;
 980         struct bch_dev *ca;
 981         struct bch_alloc_v4 a_convert;
 982         const struct bch_alloc_v4 *a;
 983         unsigned discard_key_type, freespace_key_type;
 984         unsigned gens_offset;
 985         struct bkey_s_c k;
 986         struct printbuf buf = PRINTBUF;
 987         int ret;
 988
 989         if (fsck_err_on(!bch2_dev_bucket_exists(c, alloc_k.k->p), c,
 990                         alloc_key_to_missing_dev_bucket,
 991                         "alloc key for invalid device:bucket %llu:%llu",
 992                         alloc_k.k->p.inode, alloc_k.k->p.offset))
 993                 return bch2_btree_delete_at(trans, alloc_iter, 0);
 994
 995         ca = bch_dev_bkey_exists(c, alloc_k.k->p.inode);
 996         if (!ca->mi.freespace_initialized)
 997                 return 0;
 998
 999         a = bch2_alloc_to_v4(alloc_k, &a_convert);
1000
1001         discard_key_type = a->data_type == BCH_DATA_need_discard ? KEY_TYPE_set : 0;
1002         bch2_btree_iter_set_pos(discard_iter, alloc_k.k->p);
1003         k = bch2_btree_iter_peek_slot(discard_iter);
1004         ret = bkey_err(k);
1005         if (ret)
1006                 goto err;
1007
1008         if (k.k->type != discard_key_type &&
1009             (c->opts.reconstruct_alloc ||
1010              fsck_err(c, need_discard_key_wrong,
1011                       "incorrect key in need_discard btree (got %s should be %s)\n"
1012                       "  %s",
1013                       bch2_bkey_types[k.k->type],
1014                       bch2_bkey_types[discard_key_type],
1015                       (bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf)))) {
1016                 struct bkey_i *update =
1017                         bch2_trans_kmalloc(trans, sizeof(*update));
1018
1019                 ret = PTR_ERR_OR_ZERO(update);
1020                 if (ret)
1021                         goto err;
1022
1023                 bkey_init(&update->k);
1024                 update->k.type  = discard_key_type;
1025                 update->k.p     = discard_iter->pos;
1026
1027                 ret = bch2_trans_update(trans, discard_iter, update, 0);
1028                 if (ret)
1029                         goto err;
1030         }
1031
1032         freespace_key_type = a->data_type == BCH_DATA_free ? KEY_TYPE_set : 0;
1033         bch2_btree_iter_set_pos(freespace_iter, alloc_freespace_pos(alloc_k.k->p, *a));
1034         k = bch2_btree_iter_peek_slot(freespace_iter);
1035         ret = bkey_err(k);
1036         if (ret)
1037                 goto err;
1038
1039         if (k.k->type != freespace_key_type &&
1040             (c->opts.reconstruct_alloc ||
1041              fsck_err(c, freespace_key_wrong,
1042                       "incorrect key in freespace btree (got %s should be %s)\n"
1043                       "  %s",
1044                       bch2_bkey_types[k.k->type],
1045                       bch2_bkey_types[freespace_key_type],
1046                       (printbuf_reset(&buf),
1047                        bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf)))) {
1048                 struct bkey_i *update =
1049                         bch2_trans_kmalloc(trans, sizeof(*update));
1050
1051                 ret = PTR_ERR_OR_ZERO(update);
1052                 if (ret)
1053                         goto err;
1054
1055                 bkey_init(&update->k);
1056                 update->k.type  = freespace_key_type;
1057                 update->k.p     = freespace_iter->pos;
1058                 bch2_key_resize(&update->k, 1);
1059
1060                 ret = bch2_trans_update(trans, freespace_iter, update, 0);
1061                 if (ret)
1062                         goto err;
1063         }
1064
1065         bch2_btree_iter_set_pos(bucket_gens_iter, alloc_gens_pos(alloc_k.k->p, &gens_offset));
1066         k = bch2_btree_iter_peek_slot(bucket_gens_iter);
1067         ret = bkey_err(k);
1068         if (ret)
1069                 goto err;
1070
1071         if (a->gen != alloc_gen(k, gens_offset) &&
1072             (c->opts.reconstruct_alloc ||
1073              fsck_err(c, bucket_gens_key_wrong,
1074                       "incorrect gen in bucket_gens btree (got %u should be %u)\n"
1075                       "  %s",
1076                       alloc_gen(k, gens_offset), a->gen,
1077                       (printbuf_reset(&buf),
1078                        bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf)))) {
1079                 struct bkey_i_bucket_gens *g =
1080                         bch2_trans_kmalloc(trans, sizeof(*g));
1081
1082                 ret = PTR_ERR_OR_ZERO(g);
1083                 if (ret)
1084                         goto err;
1085
1086                 if (k.k->type == KEY_TYPE_bucket_gens) {
1087                         bkey_reassemble(&g->k_i, k);
1088                 } else {
1089                         bkey_bucket_gens_init(&g->k_i);
1090                         g->k.p = alloc_gens_pos(alloc_k.k->p, &gens_offset);
1091                 }
1092
1093                 g->v.gens[gens_offset] = a->gen;
1094
1095                 ret = bch2_trans_update(trans, bucket_gens_iter, &g->k_i, 0);
1096                 if (ret)
1097                         goto err;
1098         }
1099 err:
1100 fsck_err:
1101         printbuf_exit(&buf);
1102         return ret;
1103 }
1104
1105 static noinline_for_stack
1106 int bch2_check_alloc_hole_freespace(struct btree_trans *trans,
1107                                     struct bpos start,
1108                                     struct bpos *end,
1109                                     struct btree_iter *freespace_iter)
1110 {
1111         struct bch_fs *c = trans->c;
1112         struct bch_dev *ca;
1113         struct bkey_s_c k;
1114         struct printbuf buf = PRINTBUF;
1115         int ret;
1116
1117         ca = bch_dev_bkey_exists(c, start.inode);
1118         if (!ca->mi.freespace_initialized)
1119                 return 0;
1120
1121         bch2_btree_iter_set_pos(freespace_iter, start);
1122
1123         k = bch2_btree_iter_peek_slot(freespace_iter);
1124         ret = bkey_err(k);
1125         if (ret)
1126                 goto err;
1127
1128         *end = bkey_min(k.k->p, *end);
1129
1130         if (k.k->type != KEY_TYPE_set &&
1131             (c->opts.reconstruct_alloc ||
1132              fsck_err(c, freespace_hole_missing,
1133                       "hole in alloc btree missing in freespace btree\n"
1134                       "  device %llu buckets %llu-%llu",
1135                       freespace_iter->pos.inode,
1136                       freespace_iter->pos.offset,
1137                       end->offset))) {
1138                 struct bkey_i *update =
1139                         bch2_trans_kmalloc(trans, sizeof(*update));
1140
1141                 ret = PTR_ERR_OR_ZERO(update);
1142                 if (ret)
1143                         goto err;
1144
1145                 bkey_init(&update->k);
1146                 update->k.type  = KEY_TYPE_set;
1147                 update->k.p     = freespace_iter->pos;
1148                 bch2_key_resize(&update->k,
1149                                 min_t(u64, U32_MAX, end->offset -
1150                                       freespace_iter->pos.offset));
1151
1152                 ret = bch2_trans_update(trans, freespace_iter, update, 0);
1153                 if (ret)
1154                         goto err;
1155         }
1156 err:
1157 fsck_err:
1158         printbuf_exit(&buf);
1159         return ret;
1160 }
1161
1162 static noinline_for_stack
1163 int bch2_check_alloc_hole_bucket_gens(struct btree_trans *trans,
1164                                       struct bpos start,
1165                                       struct bpos *end,
1166                                       struct btree_iter *bucket_gens_iter)
1167 {
1168         struct bch_fs *c = trans->c;
1169         struct bkey_s_c k;
1170         struct printbuf buf = PRINTBUF;
1171         unsigned i, gens_offset, gens_end_offset;
1172         int ret;
1173
1174         if (c->sb.version < bcachefs_metadata_version_bucket_gens)
1175                 return 0;
1176
1177         bch2_btree_iter_set_pos(bucket_gens_iter, alloc_gens_pos(start, &gens_offset));
1178
1179         k = bch2_btree_iter_peek_slot(bucket_gens_iter);
1180         ret = bkey_err(k);
1181         if (ret)
1182                 goto err;
1183
1184         if (bkey_cmp(alloc_gens_pos(start, &gens_offset),
1185                      alloc_gens_pos(*end,  &gens_end_offset)))
1186                 gens_end_offset = KEY_TYPE_BUCKET_GENS_NR;
1187
1188         if (k.k->type == KEY_TYPE_bucket_gens) {
1189                 struct bkey_i_bucket_gens g;
1190                 bool need_update = false;
1191
1192                 bkey_reassemble(&g.k_i, k);
1193
1194                 for (i = gens_offset; i < gens_end_offset; i++) {
1195                         if (fsck_err_on(g.v.gens[i], c,
1196                                         bucket_gens_hole_wrong,
1197                                         "hole in alloc btree at %llu:%llu with nonzero gen in bucket_gens btree (%u)",
1198                                         bucket_gens_pos_to_alloc(k.k->p, i).inode,
1199                                         bucket_gens_pos_to_alloc(k.k->p, i).offset,
1200                                         g.v.gens[i])) {
1201                                 g.v.gens[i] = 0;
1202                                 need_update = true;
1203                         }
1204                 }
1205
1206                 if (need_update) {
1207                         struct bkey_i *u = bch2_trans_kmalloc(trans, sizeof(g));
1208
1209                         ret = PTR_ERR_OR_ZERO(u);
1210                         if (ret)
1211                                 goto err;
1212
1213                         memcpy(u, &g, sizeof(g));
1214
1215                         ret = bch2_trans_update(trans, bucket_gens_iter, u, 0);
1216                         if (ret)
1217                                 goto err;
1218                 }
1219         }
1220
1221         *end = bkey_min(*end, bucket_gens_pos_to_alloc(bpos_nosnap_successor(k.k->p), 0));
1222 err:
1223 fsck_err:
1224         printbuf_exit(&buf);
1225         return ret;
1226 }
1227
1228 static noinline_for_stack int bch2_check_discard_freespace_key(struct btree_trans *trans,
1229                                               struct btree_iter *iter)
1230 {
1231         struct bch_fs *c = trans->c;
1232         struct btree_iter alloc_iter;
1233         struct bkey_s_c alloc_k;
1234         struct bch_alloc_v4 a_convert;
1235         const struct bch_alloc_v4 *a;
1236         u64 genbits;
1237         struct bpos pos;
1238         enum bch_data_type state = iter->btree_id == BTREE_ID_need_discard
1239                 ? BCH_DATA_need_discard
1240                 : BCH_DATA_free;
1241         struct printbuf buf = PRINTBUF;
1242         int ret;
1243
1244         pos = iter->pos;
1245         pos.offset &= ~(~0ULL << 56);
1246         genbits = iter->pos.offset & (~0ULL << 56);
1247
1248         alloc_k = bch2_bkey_get_iter(trans, &alloc_iter, BTREE_ID_alloc, pos, 0);
1249         ret = bkey_err(alloc_k);
1250         if (ret)
1251                 return ret;
1252
1253         if (fsck_err_on(!bch2_dev_bucket_exists(c, pos), c,
1254                         need_discard_freespace_key_to_invalid_dev_bucket,
1255                         "entry in %s btree for nonexistant dev:bucket %llu:%llu",
1256                         bch2_btree_id_str(iter->btree_id), pos.inode, pos.offset))
1257                 goto delete;
1258
1259         a = bch2_alloc_to_v4(alloc_k, &a_convert);
1260
1261         if (fsck_err_on(a->data_type != state ||
1262                         (state == BCH_DATA_free &&
1263                          genbits != alloc_freespace_genbits(*a)), c,
1264                         need_discard_freespace_key_bad,
1265                         "%s\n  incorrectly set at %s:%llu:%llu:0 (free %u, genbits %llu should be %llu)",
1266                         (bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf),
1267                         bch2_btree_id_str(iter->btree_id),
1268                         iter->pos.inode,
1269                         iter->pos.offset,
1270                         a->data_type == state,
1271                         genbits >> 56, alloc_freespace_genbits(*a) >> 56))
1272                 goto delete;
1273 out:
1274 fsck_err:
1275         set_btree_iter_dontneed(&alloc_iter);
1276         bch2_trans_iter_exit(trans, &alloc_iter);
1277         printbuf_exit(&buf);
1278         return ret;
1279 delete:
1280         ret =   bch2_btree_delete_extent_at(trans, iter,
1281                         iter->btree_id == BTREE_ID_freespace ? 1 : 0, 0) ?:
1282                 bch2_trans_commit(trans, NULL, NULL,
1283                         BCH_TRANS_COMMIT_no_enospc|BCH_TRANS_COMMIT_lazy_rw);
1284         goto out;
1285 }
1286
1287 /*
1288  * We've already checked that generation numbers in the bucket_gens btree are
1289  * valid for buckets that exist; this just checks for keys for nonexistent
1290  * buckets.
1291  */
1292 static noinline_for_stack
1293 int bch2_check_bucket_gens_key(struct btree_trans *trans,
1294                                struct btree_iter *iter,
1295                                struct bkey_s_c k)
1296 {
1297         struct bch_fs *c = trans->c;
1298         struct bkey_i_bucket_gens g;
1299         struct bch_dev *ca;
1300         u64 start = bucket_gens_pos_to_alloc(k.k->p, 0).offset;
1301         u64 end = bucket_gens_pos_to_alloc(bpos_nosnap_successor(k.k->p), 0).offset;
1302         u64 b;
1303         bool need_update = false, dev_exists;
1304         struct printbuf buf = PRINTBUF;
1305         int ret = 0;
1306
1307         BUG_ON(k.k->type != KEY_TYPE_bucket_gens);
1308         bkey_reassemble(&g.k_i, k);
1309
1310         /* if no bch_dev, skip out whether we repair or not */
1311         dev_exists = bch2_dev_exists2(c, k.k->p.inode);
1312         if (!dev_exists) {
1313                 if (fsck_err_on(!dev_exists, c,
1314                                 bucket_gens_to_invalid_dev,
1315                                 "bucket_gens key for invalid device:\n  %s",
1316                                 (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1317                         ret = bch2_btree_delete_at(trans, iter, 0);
1318                 }
1319                 goto out;
1320         }
1321
1322         ca = bch_dev_bkey_exists(c, k.k->p.inode);
1323         if (fsck_err_on(end <= ca->mi.first_bucket ||
1324                         start >= ca->mi.nbuckets, c,
1325                         bucket_gens_to_invalid_buckets,
1326                         "bucket_gens key for invalid buckets:\n  %s",
1327                         (bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
1328                 ret = bch2_btree_delete_at(trans, iter, 0);
1329                 goto out;
1330         }
1331
1332         for (b = start; b < ca->mi.first_bucket; 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         for (b = ca->mi.nbuckets; b < end; b++)
1341                 if (fsck_err_on(g.v.gens[b & KEY_TYPE_BUCKET_GENS_MASK], c,
1342                                 bucket_gens_nonzero_for_invalid_buckets,
1343                                 "bucket_gens key has nonzero gen for invalid bucket")) {
1344                         g.v.gens[b & KEY_TYPE_BUCKET_GENS_MASK] = 0;
1345                         need_update = true;
1346                 }
1347
1348         if (need_update) {
1349                 struct bkey_i *u = bch2_trans_kmalloc(trans, sizeof(g));
1350
1351                 ret = PTR_ERR_OR_ZERO(u);
1352                 if (ret)
1353                         goto out;
1354
1355                 memcpy(u, &g, sizeof(g));
1356                 ret = bch2_trans_update(trans, iter, u, 0);
1357         }
1358 out:
1359 fsck_err:
1360         printbuf_exit(&buf);
1361         return ret;
1362 }
1363
1364 int bch2_check_alloc_info(struct bch_fs *c)
1365 {
1366         struct btree_trans *trans = bch2_trans_get(c);
1367         struct btree_iter iter, discard_iter, freespace_iter, bucket_gens_iter;
1368         struct bkey hole;
1369         struct bkey_s_c k;
1370         int ret = 0;
1371
1372         bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc, POS_MIN,
1373                              BTREE_ITER_PREFETCH);
1374         bch2_trans_iter_init(trans, &discard_iter, BTREE_ID_need_discard, POS_MIN,
1375                              BTREE_ITER_PREFETCH);
1376         bch2_trans_iter_init(trans, &freespace_iter, BTREE_ID_freespace, POS_MIN,
1377                              BTREE_ITER_PREFETCH);
1378         bch2_trans_iter_init(trans, &bucket_gens_iter, BTREE_ID_bucket_gens, POS_MIN,
1379                              BTREE_ITER_PREFETCH);
1380
1381         while (1) {
1382                 struct bpos next;
1383
1384                 bch2_trans_begin(trans);
1385
1386                 k = bch2_get_key_or_real_bucket_hole(&iter, &hole);
1387                 ret = bkey_err(k);
1388                 if (ret)
1389                         goto bkey_err;
1390
1391                 if (!k.k)
1392                         break;
1393
1394                 if (k.k->type) {
1395                         next = bpos_nosnap_successor(k.k->p);
1396
1397                         ret = bch2_check_alloc_key(trans,
1398                                                    k, &iter,
1399                                                    &discard_iter,
1400                                                    &freespace_iter,
1401                                                    &bucket_gens_iter);
1402                         if (ret)
1403                                 goto bkey_err;
1404                 } else {
1405                         next = k.k->p;
1406
1407                         ret = bch2_check_alloc_hole_freespace(trans,
1408                                                     bkey_start_pos(k.k),
1409                                                     &next,
1410                                                     &freespace_iter) ?:
1411                                 bch2_check_alloc_hole_bucket_gens(trans,
1412                                                     bkey_start_pos(k.k),
1413                                                     &next,
1414                                                     &bucket_gens_iter);
1415                         if (ret)
1416                                 goto bkey_err;
1417                 }
1418
1419                 ret = bch2_trans_commit(trans, NULL, NULL,
1420                                         BCH_TRANS_COMMIT_no_enospc|
1421                                         BCH_TRANS_COMMIT_lazy_rw);
1422                 if (ret)
1423                         goto bkey_err;
1424
1425                 bch2_btree_iter_set_pos(&iter, next);
1426 bkey_err:
1427                 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
1428                         continue;
1429                 if (ret)
1430                         break;
1431         }
1432         bch2_trans_iter_exit(trans, &bucket_gens_iter);
1433         bch2_trans_iter_exit(trans, &freespace_iter);
1434         bch2_trans_iter_exit(trans, &discard_iter);
1435         bch2_trans_iter_exit(trans, &iter);
1436
1437         if (ret < 0)
1438                 goto err;
1439
1440         ret = for_each_btree_key2(trans, iter,
1441                         BTREE_ID_need_discard, POS_MIN,
1442                         BTREE_ITER_PREFETCH, k,
1443                 bch2_check_discard_freespace_key(trans, &iter));
1444         if (ret)
1445                 goto err;
1446
1447         bch2_trans_iter_init(trans, &iter, BTREE_ID_freespace, POS_MIN,
1448                              BTREE_ITER_PREFETCH);
1449         while (1) {
1450                 bch2_trans_begin(trans);
1451                 k = bch2_btree_iter_peek(&iter);
1452                 if (!k.k)
1453                         break;
1454
1455                 ret = bkey_err(k) ?:
1456                         bch2_check_discard_freespace_key(trans, &iter);
1457                 if (bch2_err_matches(ret, BCH_ERR_transaction_restart)) {
1458                         ret = 0;
1459                         continue;
1460                 }
1461                 if (ret) {
1462                         struct printbuf buf = PRINTBUF;
1463                         bch2_bkey_val_to_text(&buf, c, k);
1464
1465                         bch_err(c, "while checking %s", buf.buf);
1466                         printbuf_exit(&buf);
1467                         break;
1468                 }
1469
1470                 bch2_btree_iter_set_pos(&iter, bpos_nosnap_successor(iter.pos));
1471         }
1472         bch2_trans_iter_exit(trans, &iter);
1473         if (ret)
1474                 goto err;
1475
1476         ret = for_each_btree_key_commit(trans, iter,
1477                         BTREE_ID_bucket_gens, POS_MIN,
1478                         BTREE_ITER_PREFETCH, k,
1479                         NULL, NULL, BCH_TRANS_COMMIT_no_enospc|BCH_TRANS_COMMIT_lazy_rw,
1480                 bch2_check_bucket_gens_key(trans, &iter, k));
1481 err:
1482         bch2_trans_put(trans);
1483         if (ret)
1484                 bch_err_fn(c, ret);
1485         return ret;
1486 }
1487
1488 static int bch2_check_alloc_to_lru_ref(struct btree_trans *trans,
1489                                        struct btree_iter *alloc_iter)
1490 {
1491         struct bch_fs *c = trans->c;
1492         struct btree_iter lru_iter;
1493         struct bch_alloc_v4 a_convert;
1494         const struct bch_alloc_v4 *a;
1495         struct bkey_s_c alloc_k, lru_k;
1496         struct printbuf buf = PRINTBUF;
1497         int ret;
1498
1499         alloc_k = bch2_btree_iter_peek(alloc_iter);
1500         if (!alloc_k.k)
1501                 return 0;
1502
1503         ret = bkey_err(alloc_k);
1504         if (ret)
1505                 return ret;
1506
1507         a = bch2_alloc_to_v4(alloc_k, &a_convert);
1508
1509         if (a->data_type != BCH_DATA_cached)
1510                 return 0;
1511
1512         lru_k = bch2_bkey_get_iter(trans, &lru_iter, BTREE_ID_lru,
1513                              lru_pos(alloc_k.k->p.inode,
1514                                      bucket_to_u64(alloc_k.k->p),
1515                                      a->io_time[READ]), 0);
1516         ret = bkey_err(lru_k);
1517         if (ret)
1518                 return ret;
1519
1520         if (fsck_err_on(!a->io_time[READ], c,
1521                         alloc_key_cached_but_read_time_zero,
1522                         "cached bucket with read_time 0\n"
1523                         "  %s",
1524                 (printbuf_reset(&buf),
1525                  bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf)) ||
1526             fsck_err_on(lru_k.k->type != KEY_TYPE_set, c,
1527                         alloc_key_to_missing_lru_entry,
1528                         "missing lru entry\n"
1529                         "  %s",
1530                         (printbuf_reset(&buf),
1531                          bch2_bkey_val_to_text(&buf, c, alloc_k), buf.buf))) {
1532                 u64 read_time = a->io_time[READ] ?:
1533                         atomic64_read(&c->io_clock[READ].now);
1534
1535                 ret = bch2_lru_set(trans,
1536                                    alloc_k.k->p.inode,
1537                                    bucket_to_u64(alloc_k.k->p),
1538                                    read_time);
1539                 if (ret)
1540                         goto err;
1541
1542                 if (a->io_time[READ] != read_time) {
1543                         struct bkey_i_alloc_v4 *a_mut =
1544                                 bch2_alloc_to_v4_mut(trans, alloc_k);
1545                         ret = PTR_ERR_OR_ZERO(a_mut);
1546                         if (ret)
1547                                 goto err;
1548
1549                         a_mut->v.io_time[READ] = read_time;
1550                         ret = bch2_trans_update(trans, alloc_iter,
1551                                                 &a_mut->k_i, BTREE_TRIGGER_NORUN);
1552                         if (ret)
1553                                 goto err;
1554                 }
1555         }
1556 err:
1557 fsck_err:
1558         bch2_trans_iter_exit(trans, &lru_iter);
1559         printbuf_exit(&buf);
1560         return ret;
1561 }
1562
1563 int bch2_check_alloc_to_lru_refs(struct bch_fs *c)
1564 {
1565         struct btree_iter iter;
1566         struct bkey_s_c k;
1567         int ret = 0;
1568
1569         ret = bch2_trans_run(c,
1570                 for_each_btree_key_commit(trans, iter, BTREE_ID_alloc,
1571                                 POS_MIN, BTREE_ITER_PREFETCH, k,
1572                                 NULL, NULL, BCH_TRANS_COMMIT_no_enospc|BCH_TRANS_COMMIT_lazy_rw,
1573                         bch2_check_alloc_to_lru_ref(trans, &iter)));
1574         if (ret)
1575                 bch_err_fn(c, ret);
1576         return ret;
1577 }
1578
1579 static int bch2_discard_one_bucket(struct btree_trans *trans,
1580                                    struct btree_iter *need_discard_iter,
1581                                    struct bpos *discard_pos_done,
1582                                    u64 *seen,
1583                                    u64 *open,
1584                                    u64 *need_journal_commit,
1585                                    u64 *discarded)
1586 {
1587         struct bch_fs *c = trans->c;
1588         struct bpos pos = need_discard_iter->pos;
1589         struct btree_iter iter = { NULL };
1590         struct bkey_s_c k;
1591         struct bch_dev *ca;
1592         struct bkey_i_alloc_v4 *a;
1593         struct printbuf buf = PRINTBUF;
1594         int ret = 0;
1595
1596         ca = bch_dev_bkey_exists(c, pos.inode);
1597         if (!percpu_ref_tryget(&ca->io_ref)) {
1598                 bch2_btree_iter_set_pos(need_discard_iter, POS(pos.inode + 1, 0));
1599                 return 0;
1600         }
1601
1602         if (bch2_bucket_is_open_safe(c, pos.inode, pos.offset)) {
1603                 (*open)++;
1604                 goto out;
1605         }
1606
1607         if (bch2_bucket_needs_journal_commit(&c->buckets_waiting_for_journal,
1608                         c->journal.flushed_seq_ondisk,
1609                         pos.inode, pos.offset)) {
1610                 (*need_journal_commit)++;
1611                 goto out;
1612         }
1613
1614         k = bch2_bkey_get_iter(trans, &iter, BTREE_ID_alloc,
1615                                need_discard_iter->pos,
1616                                BTREE_ITER_CACHED);
1617         ret = bkey_err(k);
1618         if (ret)
1619                 goto out;
1620
1621         a = bch2_alloc_to_v4_mut(trans, k);
1622         ret = PTR_ERR_OR_ZERO(a);
1623         if (ret)
1624                 goto out;
1625
1626         if (BCH_ALLOC_V4_NEED_INC_GEN(&a->v)) {
1627                 a->v.gen++;
1628                 SET_BCH_ALLOC_V4_NEED_INC_GEN(&a->v, false);
1629                 goto write;
1630         }
1631
1632         if (a->v.journal_seq > c->journal.flushed_seq_ondisk) {
1633                 if (c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_info) {
1634                         bch2_trans_inconsistent(trans,
1635                                 "clearing need_discard but journal_seq %llu > flushed_seq %llu\n"
1636                                 "%s",
1637                                 a->v.journal_seq,
1638                                 c->journal.flushed_seq_ondisk,
1639                                 (bch2_bkey_val_to_text(&buf, c, k), buf.buf));
1640                         ret = -EIO;
1641                 }
1642                 goto out;
1643         }
1644
1645         if (a->v.data_type != BCH_DATA_need_discard) {
1646                 if (c->curr_recovery_pass > BCH_RECOVERY_PASS_check_alloc_info) {
1647                         bch2_trans_inconsistent(trans,
1648                                 "bucket incorrectly set in need_discard btree\n"
1649                                 "%s",
1650                                 (bch2_bkey_val_to_text(&buf, c, k), buf.buf));
1651                         ret = -EIO;
1652                 }
1653
1654                 goto out;
1655         }
1656
1657         if (!bkey_eq(*discard_pos_done, iter.pos) &&
1658             ca->mi.discard && !c->opts.nochanges) {
1659                 /*
1660                  * This works without any other locks because this is the only
1661                  * thread that removes items from the need_discard tree
1662                  */
1663                 bch2_trans_unlock(trans);
1664                 blkdev_issue_discard(ca->disk_sb.bdev,
1665                                      k.k->p.offset * ca->mi.bucket_size,
1666                                      ca->mi.bucket_size,
1667                                      GFP_KERNEL);
1668                 *discard_pos_done = iter.pos;
1669
1670                 ret = bch2_trans_relock_notrace(trans);
1671                 if (ret)
1672                         goto out;
1673         }
1674
1675         SET_BCH_ALLOC_V4_NEED_DISCARD(&a->v, false);
1676         a->v.data_type = alloc_data_type(a->v, a->v.data_type);
1677 write:
1678         ret =   bch2_trans_update(trans, &iter, &a->k_i, 0) ?:
1679                 bch2_trans_commit(trans, NULL, NULL,
1680                                   BCH_WATERMARK_btree|
1681                                   BCH_TRANS_COMMIT_no_enospc);
1682         if (ret)
1683                 goto out;
1684
1685         this_cpu_inc(c->counters[BCH_COUNTER_bucket_discard]);
1686         (*discarded)++;
1687 out:
1688         (*seen)++;
1689         bch2_trans_iter_exit(trans, &iter);
1690         percpu_ref_put(&ca->io_ref);
1691         printbuf_exit(&buf);
1692         return ret;
1693 }
1694
1695 static void bch2_do_discards_work(struct work_struct *work)
1696 {
1697         struct bch_fs *c = container_of(work, struct bch_fs, discard_work);
1698         struct btree_iter iter;
1699         struct bkey_s_c k;
1700         u64 seen = 0, open = 0, need_journal_commit = 0, discarded = 0;
1701         struct bpos discard_pos_done = POS_MAX;
1702         int ret;
1703
1704         /*
1705          * We're doing the commit in bch2_discard_one_bucket instead of using
1706          * for_each_btree_key_commit() so that we can increment counters after
1707          * successful commit:
1708          */
1709         ret = bch2_trans_run(c,
1710                 for_each_btree_key2(trans, iter,
1711                                 BTREE_ID_need_discard, POS_MIN, 0, k,
1712                         bch2_discard_one_bucket(trans, &iter, &discard_pos_done,
1713                                                 &seen,
1714                                                 &open,
1715                                                 &need_journal_commit,
1716                                                 &discarded)));
1717
1718         if (need_journal_commit * 2 > seen)
1719                 bch2_journal_flush_async(&c->journal, NULL);
1720
1721         bch2_write_ref_put(c, BCH_WRITE_REF_discard);
1722
1723         trace_discard_buckets(c, seen, open, need_journal_commit, discarded,
1724                               bch2_err_str(ret));
1725 }
1726
1727 void bch2_do_discards(struct bch_fs *c)
1728 {
1729         if (bch2_write_ref_tryget(c, BCH_WRITE_REF_discard) &&
1730             !queue_work(c->write_ref_wq, &c->discard_work))
1731                 bch2_write_ref_put(c, BCH_WRITE_REF_discard);
1732 }
1733
1734 static int invalidate_one_bucket(struct btree_trans *trans,
1735                                  struct btree_iter *lru_iter,
1736                                  struct bkey_s_c lru_k,
1737                                  s64 *nr_to_invalidate)
1738 {
1739         struct bch_fs *c = trans->c;
1740         struct btree_iter alloc_iter = { NULL };
1741         struct bkey_i_alloc_v4 *a = NULL;
1742         struct printbuf buf = PRINTBUF;
1743         struct bpos bucket = u64_to_bucket(lru_k.k->p.offset);
1744         unsigned cached_sectors;
1745         int ret = 0;
1746
1747         if (*nr_to_invalidate <= 0)
1748                 return 1;
1749
1750         if (!bch2_dev_bucket_exists(c, bucket)) {
1751                 prt_str(&buf, "lru entry points to invalid bucket");
1752                 goto err;
1753         }
1754
1755         if (bch2_bucket_is_open_safe(c, bucket.inode, bucket.offset))
1756                 return 0;
1757
1758         a = bch2_trans_start_alloc_update(trans, &alloc_iter, bucket);
1759         ret = PTR_ERR_OR_ZERO(a);
1760         if (ret)
1761                 goto out;
1762
1763         /* We expect harmless races here due to the btree write buffer: */
1764         if (lru_pos_time(lru_iter->pos) != alloc_lru_idx_read(a->v))
1765                 goto out;
1766
1767         BUG_ON(a->v.data_type != BCH_DATA_cached);
1768
1769         if (!a->v.cached_sectors)
1770                 bch_err(c, "invalidating empty bucket, confused");
1771
1772         cached_sectors = a->v.cached_sectors;
1773
1774         SET_BCH_ALLOC_V4_NEED_INC_GEN(&a->v, false);
1775         a->v.gen++;
1776         a->v.data_type          = 0;
1777         a->v.dirty_sectors      = 0;
1778         a->v.cached_sectors     = 0;
1779         a->v.io_time[READ]      = atomic64_read(&c->io_clock[READ].now);
1780         a->v.io_time[WRITE]     = atomic64_read(&c->io_clock[WRITE].now);
1781
1782         ret =   bch2_trans_update(trans, &alloc_iter, &a->k_i,
1783                                 BTREE_TRIGGER_BUCKET_INVALIDATE) ?:
1784                 bch2_trans_commit(trans, NULL, NULL,
1785                                   BCH_WATERMARK_btree|
1786                                   BCH_TRANS_COMMIT_no_enospc);
1787         if (ret)
1788                 goto out;
1789
1790         trace_and_count(c, bucket_invalidate, c, bucket.inode, bucket.offset, cached_sectors);
1791         --*nr_to_invalidate;
1792 out:
1793         bch2_trans_iter_exit(trans, &alloc_iter);
1794         printbuf_exit(&buf);
1795         return ret;
1796 err:
1797         prt_str(&buf, "\n  lru key: ");
1798         bch2_bkey_val_to_text(&buf, c, lru_k);
1799
1800         prt_str(&buf, "\n  lru entry: ");
1801         bch2_lru_pos_to_text(&buf, lru_iter->pos);
1802
1803         prt_str(&buf, "\n  alloc key: ");
1804         if (!a)
1805                 bch2_bpos_to_text(&buf, bucket);
1806         else
1807                 bch2_bkey_val_to_text(&buf, c, bkey_i_to_s_c(&a->k_i));
1808
1809         bch_err(c, "%s", buf.buf);
1810         if (c->curr_recovery_pass > BCH_RECOVERY_PASS_check_lrus) {
1811                 bch2_inconsistent_error(c);
1812                 ret = -EINVAL;
1813         }
1814
1815         goto out;
1816 }
1817
1818 static void bch2_do_invalidates_work(struct work_struct *work)
1819 {
1820         struct bch_fs *c = container_of(work, struct bch_fs, invalidate_work);
1821         struct bch_dev *ca;
1822         struct btree_trans *trans = bch2_trans_get(c);
1823         struct btree_iter iter;
1824         struct bkey_s_c k;
1825         unsigned i;
1826         int ret = 0;
1827
1828         ret = bch2_btree_write_buffer_tryflush(trans);
1829         if (ret)
1830                 goto err;
1831
1832         for_each_member_device(ca, c, i) {
1833                 s64 nr_to_invalidate =
1834                         should_invalidate_buckets(ca, bch2_dev_usage_read(ca));
1835
1836                 ret = for_each_btree_key2_upto(trans, iter, BTREE_ID_lru,
1837                                 lru_pos(ca->dev_idx, 0, 0),
1838                                 lru_pos(ca->dev_idx, U64_MAX, LRU_TIME_MAX),
1839                                 BTREE_ITER_INTENT, k,
1840                         invalidate_one_bucket(trans, &iter, k, &nr_to_invalidate));
1841
1842                 if (ret < 0) {
1843                         percpu_ref_put(&ca->ref);
1844                         break;
1845                 }
1846         }
1847 err:
1848         bch2_trans_put(trans);
1849         bch2_write_ref_put(c, BCH_WRITE_REF_invalidate);
1850 }
1851
1852 void bch2_do_invalidates(struct bch_fs *c)
1853 {
1854         if (bch2_write_ref_tryget(c, BCH_WRITE_REF_invalidate) &&
1855             !queue_work(c->write_ref_wq, &c->invalidate_work))
1856                 bch2_write_ref_put(c, BCH_WRITE_REF_invalidate);
1857 }
1858
1859 int bch2_dev_freespace_init(struct bch_fs *c, struct bch_dev *ca,
1860                             u64 bucket_start, u64 bucket_end)
1861 {
1862         struct btree_trans *trans = bch2_trans_get(c);
1863         struct btree_iter iter;
1864         struct bkey_s_c k;
1865         struct bkey hole;
1866         struct bpos end = POS(ca->dev_idx, bucket_end);
1867         struct bch_member *m;
1868         unsigned long last_updated = jiffies;
1869         int ret;
1870
1871         BUG_ON(bucket_start > bucket_end);
1872         BUG_ON(bucket_end > ca->mi.nbuckets);
1873
1874         bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc,
1875                 POS(ca->dev_idx, max_t(u64, ca->mi.first_bucket, bucket_start)),
1876                 BTREE_ITER_PREFETCH);
1877         /*
1878          * Scan the alloc btree for every bucket on @ca, and add buckets to the
1879          * freespace/need_discard/need_gc_gens btrees as needed:
1880          */
1881         while (1) {
1882                 if (last_updated + HZ * 10 < jiffies) {
1883                         bch_info(ca, "%s: currently at %llu/%llu",
1884                                  __func__, iter.pos.offset, ca->mi.nbuckets);
1885                         last_updated = jiffies;
1886                 }
1887
1888                 bch2_trans_begin(trans);
1889
1890                 if (bkey_ge(iter.pos, end)) {
1891                         ret = 0;
1892                         break;
1893                 }
1894
1895                 k = bch2_get_key_or_hole(&iter, end, &hole);
1896                 ret = bkey_err(k);
1897                 if (ret)
1898                         goto bkey_err;
1899
1900                 if (k.k->type) {
1901                         /*
1902                          * We process live keys in the alloc btree one at a
1903                          * time:
1904                          */
1905                         struct bch_alloc_v4 a_convert;
1906                         const struct bch_alloc_v4 *a = bch2_alloc_to_v4(k, &a_convert);
1907
1908                         ret =   bch2_bucket_do_index(trans, k, a, true) ?:
1909                                 bch2_trans_commit(trans, NULL, NULL,
1910                                                   BCH_TRANS_COMMIT_lazy_rw|
1911                                                   BCH_TRANS_COMMIT_no_enospc);
1912                         if (ret)
1913                                 goto bkey_err;
1914
1915                         bch2_btree_iter_advance(&iter);
1916                 } else {
1917                         struct bkey_i *freespace;
1918
1919                         freespace = bch2_trans_kmalloc(trans, sizeof(*freespace));
1920                         ret = PTR_ERR_OR_ZERO(freespace);
1921                         if (ret)
1922                                 goto bkey_err;
1923
1924                         bkey_init(&freespace->k);
1925                         freespace->k.type       = KEY_TYPE_set;
1926                         freespace->k.p          = k.k->p;
1927                         freespace->k.size       = k.k->size;
1928
1929                         ret = bch2_btree_insert_trans(trans, BTREE_ID_freespace, freespace, 0) ?:
1930                                 bch2_trans_commit(trans, NULL, NULL,
1931                                                   BCH_TRANS_COMMIT_lazy_rw|
1932                                                   BCH_TRANS_COMMIT_no_enospc);
1933                         if (ret)
1934                                 goto bkey_err;
1935
1936                         bch2_btree_iter_set_pos(&iter, k.k->p);
1937                 }
1938 bkey_err:
1939                 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
1940                         continue;
1941                 if (ret)
1942                         break;
1943         }
1944
1945         bch2_trans_iter_exit(trans, &iter);
1946         bch2_trans_put(trans);
1947
1948         if (ret < 0) {
1949                 bch_err_msg(ca, ret, "initializing free space");
1950                 return ret;
1951         }
1952
1953         mutex_lock(&c->sb_lock);
1954         m = bch2_members_v2_get_mut(c->disk_sb.sb, ca->dev_idx);
1955         SET_BCH_MEMBER_FREESPACE_INITIALIZED(m, true);
1956         mutex_unlock(&c->sb_lock);
1957
1958         return 0;
1959 }
1960
1961 int bch2_fs_freespace_init(struct bch_fs *c)
1962 {
1963         struct bch_dev *ca;
1964         unsigned i;
1965         int ret = 0;
1966         bool doing_init = false;
1967
1968         /*
1969          * We can crash during the device add path, so we need to check this on
1970          * every mount:
1971          */
1972
1973         for_each_member_device(ca, c, i) {
1974                 if (ca->mi.freespace_initialized)
1975                         continue;
1976
1977                 if (!doing_init) {
1978                         bch_info(c, "initializing freespace");
1979                         doing_init = true;
1980                 }
1981
1982                 ret = bch2_dev_freespace_init(c, ca, 0, ca->mi.nbuckets);
1983                 if (ret) {
1984                         percpu_ref_put(&ca->ref);
1985                         bch_err_fn(c, ret);
1986                         return ret;
1987                 }
1988         }
1989
1990         if (doing_init) {
1991                 mutex_lock(&c->sb_lock);
1992                 bch2_write_super(c);
1993                 mutex_unlock(&c->sb_lock);
1994                 bch_verbose(c, "done initializing freespace");
1995         }
1996
1997         return 0;
1998 }
1999
2000 /* Bucket IO clocks: */
2001
2002 int bch2_bucket_io_time_reset(struct btree_trans *trans, unsigned dev,
2003                               size_t bucket_nr, int rw)
2004 {
2005         struct bch_fs *c = trans->c;
2006         struct btree_iter iter;
2007         struct bkey_i_alloc_v4 *a;
2008         u64 now;
2009         int ret = 0;
2010
2011         a = bch2_trans_start_alloc_update(trans, &iter,  POS(dev, bucket_nr));
2012         ret = PTR_ERR_OR_ZERO(a);
2013         if (ret)
2014                 return ret;
2015
2016         now = atomic64_read(&c->io_clock[rw].now);
2017         if (a->v.io_time[rw] == now)
2018                 goto out;
2019
2020         a->v.io_time[rw] = now;
2021
2022         ret   = bch2_trans_update(trans, &iter, &a->k_i, 0) ?:
2023                 bch2_trans_commit(trans, NULL, NULL, 0);
2024 out:
2025         bch2_trans_iter_exit(trans, &iter);
2026         return ret;
2027 }
2028
2029 /* Startup/shutdown (ro/rw): */
2030
2031 void bch2_recalc_capacity(struct bch_fs *c)
2032 {
2033         struct bch_dev *ca;
2034         u64 capacity = 0, reserved_sectors = 0, gc_reserve;
2035         unsigned bucket_size_max = 0;
2036         unsigned long ra_pages = 0;
2037         unsigned i;
2038
2039         lockdep_assert_held(&c->state_lock);
2040
2041         for_each_online_member(ca, c, i) {
2042                 struct backing_dev_info *bdi = ca->disk_sb.bdev->bd_disk->bdi;
2043
2044                 ra_pages += bdi->ra_pages;
2045         }
2046
2047         bch2_set_ra_pages(c, ra_pages);
2048
2049         for_each_rw_member(ca, c, i) {
2050                 u64 dev_reserve = 0;
2051
2052                 /*
2053                  * We need to reserve buckets (from the number
2054                  * of currently available buckets) against
2055                  * foreground writes so that mainly copygc can
2056                  * make forward progress.
2057                  *
2058                  * We need enough to refill the various reserves
2059                  * from scratch - copygc will use its entire
2060                  * reserve all at once, then run against when
2061                  * its reserve is refilled (from the formerly
2062                  * available buckets).
2063                  *
2064                  * This reserve is just used when considering if
2065                  * allocations for foreground writes must wait -
2066                  * not -ENOSPC calculations.
2067                  */
2068
2069                 dev_reserve += ca->nr_btree_reserve * 2;
2070                 dev_reserve += ca->mi.nbuckets >> 6; /* copygc reserve */
2071
2072                 dev_reserve += 1;       /* btree write point */
2073                 dev_reserve += 1;       /* copygc write point */
2074                 dev_reserve += 1;       /* rebalance write point */
2075
2076                 dev_reserve *= ca->mi.bucket_size;
2077
2078                 capacity += bucket_to_sector(ca, ca->mi.nbuckets -
2079                                              ca->mi.first_bucket);
2080
2081                 reserved_sectors += dev_reserve * 2;
2082
2083                 bucket_size_max = max_t(unsigned, bucket_size_max,
2084                                         ca->mi.bucket_size);
2085         }
2086
2087         gc_reserve = c->opts.gc_reserve_bytes
2088                 ? c->opts.gc_reserve_bytes >> 9
2089                 : div64_u64(capacity * c->opts.gc_reserve_percent, 100);
2090
2091         reserved_sectors = max(gc_reserve, reserved_sectors);
2092
2093         reserved_sectors = min(reserved_sectors, capacity);
2094
2095         c->capacity = capacity - reserved_sectors;
2096
2097         c->bucket_size_max = bucket_size_max;
2098
2099         /* Wake up case someone was waiting for buckets */
2100         closure_wake_up(&c->freelist_wait);
2101 }
2102
2103 u64 bch2_min_rw_member_capacity(struct bch_fs *c)
2104 {
2105         struct bch_dev *ca;
2106         unsigned i;
2107         u64 ret = U64_MAX;
2108
2109         for_each_rw_member(ca, c, i)
2110                 ret = min(ret, ca->mi.nbuckets * ca->mi.bucket_size);
2111         return ret;
2112 }
2113
2114 static bool bch2_dev_has_open_write_point(struct bch_fs *c, struct bch_dev *ca)
2115 {
2116         struct open_bucket *ob;
2117         bool ret = false;
2118
2119         for (ob = c->open_buckets;
2120              ob < c->open_buckets + ARRAY_SIZE(c->open_buckets);
2121              ob++) {
2122                 spin_lock(&ob->lock);
2123                 if (ob->valid && !ob->on_partial_list &&
2124                     ob->dev == ca->dev_idx)
2125                         ret = true;
2126                 spin_unlock(&ob->lock);
2127         }
2128
2129         return ret;
2130 }
2131
2132 /* device goes ro: */
2133 void bch2_dev_allocator_remove(struct bch_fs *c, struct bch_dev *ca)
2134 {
2135         unsigned i;
2136
2137         /* First, remove device from allocation groups: */
2138
2139         for (i = 0; i < ARRAY_SIZE(c->rw_devs); i++)
2140                 clear_bit(ca->dev_idx, c->rw_devs[i].d);
2141
2142         /*
2143          * Capacity is calculated based off of devices in allocation groups:
2144          */
2145         bch2_recalc_capacity(c);
2146
2147         bch2_open_buckets_stop(c, ca, false);
2148
2149         /*
2150          * Wake up threads that were blocked on allocation, so they can notice
2151          * the device can no longer be removed and the capacity has changed:
2152          */
2153         closure_wake_up(&c->freelist_wait);
2154
2155         /*
2156          * journal_res_get() can block waiting for free space in the journal -
2157          * it needs to notice there may not be devices to allocate from anymore:
2158          */
2159         wake_up(&c->journal.wait);
2160
2161         /* Now wait for any in flight writes: */
2162
2163         closure_wait_event(&c->open_buckets_wait,
2164                            !bch2_dev_has_open_write_point(c, ca));
2165 }
2166
2167 /* device goes rw: */
2168 void bch2_dev_allocator_add(struct bch_fs *c, struct bch_dev *ca)
2169 {
2170         unsigned i;
2171
2172         for (i = 0; i < ARRAY_SIZE(c->rw_devs); i++)
2173                 if (ca->mi.data_allowed & (1 << i))
2174                         set_bit(ca->dev_idx, c->rw_devs[i].d);
2175 }
2176
2177 void bch2_fs_allocator_background_init(struct bch_fs *c)
2178 {
2179         spin_lock_init(&c->freelist_lock);
2180         INIT_WORK(&c->discard_work, bch2_do_discards_work);
2181         INIT_WORK(&c->invalidate_work, bch2_do_invalidates_work);
2182 }