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

   1 // SPDX-License-Identifier: GPL-2.0
   2
   3 #include "bcachefs.h"
   4 #include "backpointers.h"
   5 #include "bkey_methods.h"
   6 #include "btree_types.h"
   7 #include "alloc_background.h"
   8 #include "dirent.h"
   9 #include "ec.h"
  10 #include "error.h"
  11 #include "extents.h"
  12 #include "inode.h"
  13 #include "lru.h"
  14 #include "quota.h"
  15 #include "reflink.h"
  16 #include "subvolume.h"
  17 #include "xattr.h"
  18
  19 const char * const bch2_bkey_types[] = {
  20 #define x(name, nr) #name,
  21         BCH_BKEY_TYPES()
  22 #undef x
  23         NULL
  24 };
  25
  26 static int deleted_key_invalid(const struct bch_fs *c, struct bkey_s_c k,
  27                                int rw, struct printbuf *err)
  28 {
  29         return 0;
  30 }
  31
  32 #define bch2_bkey_ops_deleted (struct bkey_ops) {       \
  33         .key_invalid = deleted_key_invalid,             \
  34 }
  35
  36 #define bch2_bkey_ops_whiteout (struct bkey_ops) {      \
  37         .key_invalid = deleted_key_invalid,             \
  38 }
  39
  40 static int empty_val_key_invalid(const struct bch_fs *c, struct bkey_s_c k,
  41                                  int rw, struct printbuf *err)
  42 {
  43         if (bkey_val_bytes(k.k)) {
  44                 prt_printf(err, "incorrect value size (%zu != 0)",
  45                        bkey_val_bytes(k.k));
  46                 return -EINVAL;
  47         }
  48
  49         return 0;
  50 }
  51
  52 #define bch2_bkey_ops_error (struct bkey_ops) {         \
  53         .key_invalid = empty_val_key_invalid,           \
  54 }
  55
  56 static int key_type_cookie_invalid(const struct bch_fs *c, struct bkey_s_c k,
  57                                    int rw, struct printbuf *err)
  58 {
  59         if (bkey_val_bytes(k.k) != sizeof(struct bch_cookie)) {
  60                 prt_printf(err, "incorrect value size (%zu != %zu)",
  61                        bkey_val_bytes(k.k), sizeof(struct bch_cookie));
  62                 return -EINVAL;
  63         }
  64
  65         return 0;
  66 }
  67
  68 #define bch2_bkey_ops_cookie (struct bkey_ops) {        \
  69         .key_invalid = key_type_cookie_invalid,         \
  70 }
  71
  72 #define bch2_bkey_ops_hash_whiteout (struct bkey_ops) { \
  73         .key_invalid = empty_val_key_invalid,           \
  74 }
  75
  76 static int key_type_inline_data_invalid(const struct bch_fs *c, struct bkey_s_c k,
  77                                         int rw, struct printbuf *err)
  78 {
  79         return 0;
  80 }
  81
  82 static void key_type_inline_data_to_text(struct printbuf *out, struct bch_fs *c,
  83                                          struct bkey_s_c k)
  84 {
  85         struct bkey_s_c_inline_data d = bkey_s_c_to_inline_data(k);
  86         unsigned datalen = bkey_inline_data_bytes(k.k);
  87
  88         prt_printf(out, "datalen %u: %*phN",
  89                datalen, min(datalen, 32U), d.v->data);
  90 }
  91
  92 #define bch2_bkey_ops_inline_data (struct bkey_ops) {   \
  93         .key_invalid    = key_type_inline_data_invalid, \
  94         .val_to_text    = key_type_inline_data_to_text, \
  95 }
  96
  97 static int key_type_set_invalid(const struct bch_fs *c, struct bkey_s_c k,
  98                                 int rw, struct printbuf *err)
  99 {
 100         if (bkey_val_bytes(k.k)) {
 101                 prt_printf(err, "incorrect value size (%zu != %zu)",
 102                        bkey_val_bytes(k.k), sizeof(struct bch_cookie));
 103                 return -EINVAL;
 104         }
 105
 106         return 0;
 107 }
 108
 109 static bool key_type_set_merge(struct bch_fs *c, struct bkey_s l, struct bkey_s_c r)
 110 {
 111         bch2_key_resize(l.k, l.k->size + r.k->size);
 112         return true;
 113 }
 114
 115 #define bch2_bkey_ops_set (struct bkey_ops) {           \
 116         .key_invalid    = key_type_set_invalid,         \
 117         .key_merge      = key_type_set_merge,           \
 118 }
 119
 120 const struct bkey_ops bch2_bkey_ops[] = {
 121 #define x(name, nr) [KEY_TYPE_##name]   = bch2_bkey_ops_##name,
 122         BCH_BKEY_TYPES()
 123 #undef x
 124 };
 125
 126 int bch2_bkey_val_invalid(struct bch_fs *c, struct bkey_s_c k,
 127                           int rw, struct printbuf *err)
 128 {
 129         if (k.k->type >= KEY_TYPE_MAX) {
 130                 prt_printf(err, "invalid type (%u >= %u)", k.k->type, KEY_TYPE_MAX);
 131                 return -EINVAL;
 132         }
 133
 134         return bch2_bkey_ops[k.k->type].key_invalid(c, k, rw, err);
 135 }
 136
 137 static unsigned bch2_key_types_allowed[] = {
 138         [BKEY_TYPE_extents] =
 139                 (1U << KEY_TYPE_deleted)|
 140                 (1U << KEY_TYPE_whiteout)|
 141                 (1U << KEY_TYPE_error)|
 142                 (1U << KEY_TYPE_cookie)|
 143                 (1U << KEY_TYPE_extent)|
 144                 (1U << KEY_TYPE_reservation)|
 145                 (1U << KEY_TYPE_reflink_p)|
 146                 (1U << KEY_TYPE_inline_data),
 147         [BKEY_TYPE_inodes] =
 148                 (1U << KEY_TYPE_deleted)|
 149                 (1U << KEY_TYPE_whiteout)|
 150                 (1U << KEY_TYPE_inode)|
 151                 (1U << KEY_TYPE_inode_v2)|
 152                 (1U << KEY_TYPE_inode_generation),
 153         [BKEY_TYPE_dirents] =
 154                 (1U << KEY_TYPE_deleted)|
 155                 (1U << KEY_TYPE_whiteout)|
 156                 (1U << KEY_TYPE_hash_whiteout)|
 157                 (1U << KEY_TYPE_dirent),
 158         [BKEY_TYPE_xattrs] =
 159                 (1U << KEY_TYPE_deleted)|
 160                 (1U << KEY_TYPE_whiteout)|
 161                 (1U << KEY_TYPE_cookie)|
 162                 (1U << KEY_TYPE_hash_whiteout)|
 163                 (1U << KEY_TYPE_xattr),
 164         [BKEY_TYPE_alloc] =
 165                 (1U << KEY_TYPE_deleted)|
 166                 (1U << KEY_TYPE_alloc)|
 167                 (1U << KEY_TYPE_alloc_v2)|
 168                 (1U << KEY_TYPE_alloc_v3)|
 169                 (1U << KEY_TYPE_alloc_v4),
 170         [BKEY_TYPE_quotas] =
 171                 (1U << KEY_TYPE_deleted)|
 172                 (1U << KEY_TYPE_quota),
 173         [BKEY_TYPE_stripes] =
 174                 (1U << KEY_TYPE_deleted)|
 175                 (1U << KEY_TYPE_stripe),
 176         [BKEY_TYPE_reflink] =
 177                 (1U << KEY_TYPE_deleted)|
 178                 (1U << KEY_TYPE_reflink_v)|
 179                 (1U << KEY_TYPE_indirect_inline_data),
 180         [BKEY_TYPE_subvolumes] =
 181                 (1U << KEY_TYPE_deleted)|
 182                 (1U << KEY_TYPE_subvolume),
 183         [BKEY_TYPE_snapshots] =
 184                 (1U << KEY_TYPE_deleted)|
 185                 (1U << KEY_TYPE_snapshot),
 186         [BKEY_TYPE_lru] =
 187                 (1U << KEY_TYPE_deleted)|
 188                 (1U << KEY_TYPE_lru),
 189         [BKEY_TYPE_freespace] =
 190                 (1U << KEY_TYPE_deleted)|
 191                 (1U << KEY_TYPE_set),
 192         [BKEY_TYPE_need_discard] =
 193                 (1U << KEY_TYPE_deleted)|
 194                 (1U << KEY_TYPE_set),
 195         [BKEY_TYPE_backpointers] =
 196                 (1U << KEY_TYPE_deleted)|
 197                 (1U << KEY_TYPE_backpointer),
 198         [BKEY_TYPE_btree] =
 199                 (1U << KEY_TYPE_deleted)|
 200                 (1U << KEY_TYPE_btree_ptr)|
 201                 (1U << KEY_TYPE_btree_ptr_v2),
 202 };
 203
 204 int __bch2_bkey_invalid(struct bch_fs *c, struct bkey_s_c k,
 205                         enum btree_node_type type,
 206                         int rw, struct printbuf *err)
 207 {
 208         if (k.k->u64s < BKEY_U64s) {
 209                 prt_printf(err, "u64s too small (%u < %zu)", k.k->u64s, BKEY_U64s);
 210                 return -EINVAL;
 211         }
 212
 213         if (!(bch2_key_types_allowed[type] & (1U << k.k->type))) {
 214                 prt_printf(err, "invalid key type for this btree (%s)",
 215                        bch2_bkey_types[type]);
 216                 return -EINVAL;
 217         }
 218
 219         if (btree_node_type_is_extents(type) && !bkey_whiteout(k.k)) {
 220                 if (k.k->size == 0) {
 221                         prt_printf(err, "size == 0");
 222                         return -EINVAL;
 223                 }
 224
 225                 if (k.k->size > k.k->p.offset) {
 226                         prt_printf(err, "size greater than offset (%u > %llu)",
 227                                k.k->size, k.k->p.offset);
 228                         return -EINVAL;
 229                 }
 230         } else {
 231                 if (k.k->size) {
 232                         prt_printf(err, "size != 0");
 233                         return -EINVAL;
 234                 }
 235         }
 236
 237         if (type != BKEY_TYPE_btree &&
 238             !btree_type_has_snapshots(type) &&
 239             k.k->p.snapshot) {
 240                 prt_printf(err, "nonzero snapshot");
 241                 return -EINVAL;
 242         }
 243
 244         if (type != BKEY_TYPE_btree &&
 245             btree_type_has_snapshots(type) &&
 246             !k.k->p.snapshot) {
 247                 prt_printf(err, "snapshot == 0");
 248                 return -EINVAL;
 249         }
 250
 251         if (type != BKEY_TYPE_btree &&
 252             !bkey_cmp(k.k->p, POS_MAX)) {
 253                 prt_printf(err, "key at POS_MAX");
 254                 return -EINVAL;
 255         }
 256
 257         return 0;
 258 }
 259
 260 int bch2_bkey_invalid(struct bch_fs *c, struct bkey_s_c k,
 261                       enum btree_node_type type,
 262                       int rw, struct printbuf *err)
 263 {
 264         return __bch2_bkey_invalid(c, k, type, rw, err) ?:
 265                 bch2_bkey_val_invalid(c, k, rw, err);
 266 }
 267
 268 int bch2_bkey_in_btree_node(struct btree *b, struct bkey_s_c k,
 269                             struct printbuf *err)
 270 {
 271         if (bpos_cmp(k.k->p, b->data->min_key) < 0) {
 272                 prt_printf(err, "key before start of btree node");
 273                 return -EINVAL;
 274         }
 275
 276         if (bpos_cmp(k.k->p, b->data->max_key) > 0) {
 277                 prt_printf(err, "key past end of btree node");
 278                 return -EINVAL;
 279         }
 280
 281         return 0;
 282 }
 283
 284 void bch2_bpos_to_text(struct printbuf *out, struct bpos pos)
 285 {
 286         if (!bpos_cmp(pos, POS_MIN))
 287                 prt_printf(out, "POS_MIN");
 288         else if (!bpos_cmp(pos, POS_MAX))
 289                 prt_printf(out, "POS_MAX");
 290         else if (!bpos_cmp(pos, SPOS_MAX))
 291                 prt_printf(out, "SPOS_MAX");
 292         else {
 293                 if (pos.inode == U64_MAX)
 294                         prt_printf(out, "U64_MAX");
 295                 else
 296                         prt_printf(out, "%llu", pos.inode);
 297                 prt_printf(out, ":");
 298                 if (pos.offset == U64_MAX)
 299                         prt_printf(out, "U64_MAX");
 300                 else
 301                         prt_printf(out, "%llu", pos.offset);
 302                 prt_printf(out, ":");
 303                 if (pos.snapshot == U32_MAX)
 304                         prt_printf(out, "U32_MAX");
 305                 else
 306                         prt_printf(out, "%u", pos.snapshot);
 307         }
 308 }
 309
 310 void bch2_bkey_to_text(struct printbuf *out, const struct bkey *k)
 311 {
 312         if (k) {
 313                 prt_printf(out, "u64s %u type ", k->u64s);
 314
 315                 if (k->type < KEY_TYPE_MAX)
 316                         prt_printf(out, "%s ", bch2_bkey_types[k->type]);
 317                 else
 318                         prt_printf(out, "%u ", k->type);
 319
 320                 bch2_bpos_to_text(out, k->p);
 321
 322                 prt_printf(out, " len %u ver %llu", k->size, k->version.lo);
 323         } else {
 324                 prt_printf(out, "(null)");
 325         }
 326 }
 327
 328 void bch2_val_to_text(struct printbuf *out, struct bch_fs *c,
 329                       struct bkey_s_c k)
 330 {
 331         if (k.k->type < KEY_TYPE_MAX) {
 332                 const struct bkey_ops *ops = &bch2_bkey_ops[k.k->type];
 333
 334                 if (likely(ops->val_to_text))
 335                         ops->val_to_text(out, c, k);
 336         } else {
 337                 prt_printf(out, "(invalid type %u)", k.k->type);
 338         }
 339 }
 340
 341 void bch2_bkey_val_to_text(struct printbuf *out, struct bch_fs *c,
 342                            struct bkey_s_c k)
 343 {
 344         bch2_bkey_to_text(out, k.k);
 345
 346         if (bkey_val_bytes(k.k)) {
 347                 prt_printf(out, ": ");
 348                 bch2_val_to_text(out, c, k);
 349         }
 350 }
 351
 352 void bch2_bkey_swab_val(struct bkey_s k)
 353 {
 354         const struct bkey_ops *ops = &bch2_bkey_ops[k.k->type];
 355
 356         if (ops->swab)
 357                 ops->swab(k);
 358 }
 359
 360 bool bch2_bkey_normalize(struct bch_fs *c, struct bkey_s k)
 361 {
 362         const struct bkey_ops *ops = &bch2_bkey_ops[k.k->type];
 363
 364         return ops->key_normalize
 365                 ? ops->key_normalize(c, k)
 366                 : false;
 367 }
 368
 369 bool bch2_bkey_merge(struct bch_fs *c, struct bkey_s l, struct bkey_s_c r)
 370 {
 371         const struct bkey_ops *ops = &bch2_bkey_ops[l.k->type];
 372
 373         return bch2_bkey_maybe_mergable(l.k, r.k) && ops->key_merge(c, l, r);
 374 }
 375
 376 static const struct old_bkey_type {
 377         u8              btree_node_type;
 378         u8              old;
 379         u8              new;
 380 } bkey_renumber_table[] = {
 381         {BKEY_TYPE_btree,       128, KEY_TYPE_btree_ptr         },
 382         {BKEY_TYPE_extents,     128, KEY_TYPE_extent            },
 383         {BKEY_TYPE_extents,     129, KEY_TYPE_extent            },
 384         {BKEY_TYPE_extents,     130, KEY_TYPE_reservation       },
 385         {BKEY_TYPE_inodes,      128, KEY_TYPE_inode             },
 386         {BKEY_TYPE_inodes,      130, KEY_TYPE_inode_generation  },
 387         {BKEY_TYPE_dirents,     128, KEY_TYPE_dirent            },
 388         {BKEY_TYPE_dirents,     129, KEY_TYPE_hash_whiteout     },
 389         {BKEY_TYPE_xattrs,      128, KEY_TYPE_xattr             },
 390         {BKEY_TYPE_xattrs,      129, KEY_TYPE_hash_whiteout     },
 391         {BKEY_TYPE_alloc,       128, KEY_TYPE_alloc             },
 392         {BKEY_TYPE_quotas,      128, KEY_TYPE_quota             },
 393 };
 394
 395 void bch2_bkey_renumber(enum btree_node_type btree_node_type,
 396                         struct bkey_packed *k,
 397                         int write)
 398 {
 399         const struct old_bkey_type *i;
 400
 401         for (i = bkey_renumber_table;
 402              i < bkey_renumber_table + ARRAY_SIZE(bkey_renumber_table);
 403              i++)
 404                 if (btree_node_type == i->btree_node_type &&
 405                     k->type == (write ? i->new : i->old)) {
 406                         k->type = write ? i->old : i->new;
 407                         break;
 408                 }
 409 }
 410
 411 void __bch2_bkey_compat(unsigned level, enum btree_id btree_id,
 412                         unsigned version, unsigned big_endian,
 413                         int write,
 414                         struct bkey_format *f,
 415                         struct bkey_packed *k)
 416 {
 417         const struct bkey_ops *ops;
 418         struct bkey uk;
 419         struct bkey_s u;
 420         unsigned nr_compat = 5;
 421         int i;
 422
 423         /*
 424          * Do these operations in reverse order in the write path:
 425          */
 426
 427         for (i = 0; i < nr_compat; i++)
 428         switch (!write ? i : nr_compat - 1 - i) {
 429         case 0:
 430                 if (big_endian != CPU_BIG_ENDIAN)
 431                         bch2_bkey_swab_key(f, k);
 432                 break;
 433         case 1:
 434                 if (version < bcachefs_metadata_version_bkey_renumber)
 435                         bch2_bkey_renumber(__btree_node_type(level, btree_id), k, write);
 436                 break;
 437         case 2:
 438                 if (version < bcachefs_metadata_version_inode_btree_change &&
 439                     btree_id == BTREE_ID_inodes) {
 440                         if (!bkey_packed(k)) {
 441                                 struct bkey_i *u = packed_to_bkey(k);
 442                                 swap(u->k.p.inode, u->k.p.offset);
 443                         } else if (f->bits_per_field[BKEY_FIELD_INODE] &&
 444                                    f->bits_per_field[BKEY_FIELD_OFFSET]) {
 445                                 struct bkey_format tmp = *f, *in = f, *out = &tmp;
 446
 447                                 swap(tmp.bits_per_field[BKEY_FIELD_INODE],
 448                                      tmp.bits_per_field[BKEY_FIELD_OFFSET]);
 449                                 swap(tmp.field_offset[BKEY_FIELD_INODE],
 450                                      tmp.field_offset[BKEY_FIELD_OFFSET]);
 451
 452                                 if (!write)
 453                                         swap(in, out);
 454
 455                                 uk = __bch2_bkey_unpack_key(in, k);
 456                                 swap(uk.p.inode, uk.p.offset);
 457                                 BUG_ON(!bch2_bkey_pack_key(k, &uk, out));
 458                         }
 459                 }
 460                 break;
 461         case 3:
 462                 if (version < bcachefs_metadata_version_snapshot &&
 463                     (level || btree_type_has_snapshots(btree_id))) {
 464                         struct bkey_i *u = packed_to_bkey(k);
 465
 466                         if (u) {
 467                                 u->k.p.snapshot = write
 468                                         ? 0 : U32_MAX;
 469                         } else {
 470                                 u64 min_packed = f->field_offset[BKEY_FIELD_SNAPSHOT];
 471                                 u64 max_packed = min_packed +
 472                                         ~(~0ULL << f->bits_per_field[BKEY_FIELD_SNAPSHOT]);
 473
 474                                 uk = __bch2_bkey_unpack_key(f, k);
 475                                 uk.p.snapshot = write
 476                                         ? min_packed : min_t(u64, U32_MAX, max_packed);
 477
 478                                 BUG_ON(!bch2_bkey_pack_key(k, &uk, f));
 479                         }
 480                 }
 481
 482                 break;
 483         case 4:
 484                 if (!bkey_packed(k)) {
 485                         u = bkey_i_to_s(packed_to_bkey(k));
 486                 } else {
 487                         uk = __bch2_bkey_unpack_key(f, k);
 488                         u.k = &uk;
 489                         u.v = bkeyp_val(f, k);
 490                 }
 491
 492                 if (big_endian != CPU_BIG_ENDIAN)
 493                         bch2_bkey_swab_val(u);
 494
 495                 ops = &bch2_bkey_ops[k->type];
 496
 497                 if (ops->compat)
 498                         ops->compat(btree_id, version, big_endian, write, u);
 499                 break;
 500         default:
 501                 BUG();
 502         }
 503 }