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