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 const char *deleted_key_invalid(const struct bch_fs *c,
  26                                         struct bkey_s_c k)
  27 {
  28         return NULL;
  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 const char *empty_val_key_invalid(const struct bch_fs *c, struct bkey_s_c k)
  40 {
  41         if (bkey_val_bytes(k.k))
  42                 return "value size should be zero";
  43
  44         return NULL;
  45 }
  46
  47 #define bch2_bkey_ops_error (struct bkey_ops) {         \
  48         .key_invalid = empty_val_key_invalid,           \
  49 }
  50
  51 static const char *key_type_cookie_invalid(const struct bch_fs *c,
  52                                            struct bkey_s_c k)
  53 {
  54         if (bkey_val_bytes(k.k) != sizeof(struct bch_cookie))
  55                 return "incorrect value size";
  56
  57         return NULL;
  58 }
  59
  60 #define bch2_bkey_ops_cookie (struct bkey_ops) {        \
  61         .key_invalid = key_type_cookie_invalid,         \
  62 }
  63
  64 #define bch2_bkey_ops_hash_whiteout (struct bkey_ops) { \
  65         .key_invalid = empty_val_key_invalid,           \
  66 }
  67
  68 static const char *key_type_inline_data_invalid(const struct bch_fs *c,
  69                                            struct bkey_s_c k)
  70 {
  71         return NULL;
  72 }
  73
  74 static void key_type_inline_data_to_text(struct printbuf *out, struct bch_fs *c,
  75                                          struct bkey_s_c k)
  76 {
  77         struct bkey_s_c_inline_data d = bkey_s_c_to_inline_data(k);
  78         unsigned datalen = bkey_inline_data_bytes(k.k);
  79
  80         pr_buf(out, "datalen %u: %*phN",
  81                datalen, min(datalen, 32U), d.v->data);
  82 }
  83
  84 #define bch2_bkey_ops_inline_data (struct bkey_ops) {   \
  85         .key_invalid    = key_type_inline_data_invalid, \
  86         .val_to_text    = key_type_inline_data_to_text, \
  87 }
  88
  89 static const char *key_type_set_invalid(const struct bch_fs *c, struct bkey_s_c k)
  90 {
  91         if (bkey_val_bytes(k.k))
  92                 return "nonempty value";
  93         return NULL;
  94 }
  95
  96 static bool key_type_set_merge(struct bch_fs *c, struct bkey_s l, struct bkey_s_c r)
  97 {
  98         bch2_key_resize(l.k, l.k->size + r.k->size);
  99         return true;
 100 }
 101
 102 #define bch2_bkey_ops_set (struct bkey_ops) {           \
 103         .key_invalid    = key_type_set_invalid,         \
 104         .key_merge      = key_type_set_merge,           \
 105 }
 106
 107 const struct bkey_ops bch2_bkey_ops[] = {
 108 #define x(name, nr) [KEY_TYPE_##name]   = bch2_bkey_ops_##name,
 109         BCH_BKEY_TYPES()
 110 #undef x
 111 };
 112
 113 const char *bch2_bkey_val_invalid(struct bch_fs *c, struct bkey_s_c k)
 114 {
 115         if (k.k->type >= KEY_TYPE_MAX)
 116                 return "invalid type";
 117
 118         return bch2_bkey_ops[k.k->type].key_invalid(c, k);
 119 }
 120
 121 static unsigned bch2_key_types_allowed[] = {
 122         [BKEY_TYPE_extents] =
 123                 (1U << KEY_TYPE_deleted)|
 124                 (1U << KEY_TYPE_whiteout)|
 125                 (1U << KEY_TYPE_error)|
 126                 (1U << KEY_TYPE_cookie)|
 127                 (1U << KEY_TYPE_extent)|
 128                 (1U << KEY_TYPE_reservation)|
 129                 (1U << KEY_TYPE_reflink_p)|
 130                 (1U << KEY_TYPE_inline_data),
 131         [BKEY_TYPE_inodes] =
 132                 (1U << KEY_TYPE_deleted)|
 133                 (1U << KEY_TYPE_whiteout)|
 134                 (1U << KEY_TYPE_inode)|
 135                 (1U << KEY_TYPE_inode_v2)|
 136                 (1U << KEY_TYPE_inode_generation),
 137         [BKEY_TYPE_dirents] =
 138                 (1U << KEY_TYPE_deleted)|
 139                 (1U << KEY_TYPE_whiteout)|
 140                 (1U << KEY_TYPE_hash_whiteout)|
 141                 (1U << KEY_TYPE_dirent),
 142         [BKEY_TYPE_xattrs] =
 143                 (1U << KEY_TYPE_deleted)|
 144                 (1U << KEY_TYPE_whiteout)|
 145                 (1U << KEY_TYPE_cookie)|
 146                 (1U << KEY_TYPE_hash_whiteout)|
 147                 (1U << KEY_TYPE_xattr),
 148         [BKEY_TYPE_alloc] =
 149                 (1U << KEY_TYPE_deleted)|
 150                 (1U << KEY_TYPE_alloc)|
 151                 (1U << KEY_TYPE_alloc_v2)|
 152                 (1U << KEY_TYPE_alloc_v3),
 153         [BKEY_TYPE_quotas] =
 154                 (1U << KEY_TYPE_deleted)|
 155                 (1U << KEY_TYPE_quota),
 156         [BKEY_TYPE_stripes] =
 157                 (1U << KEY_TYPE_deleted)|
 158                 (1U << KEY_TYPE_stripe),
 159         [BKEY_TYPE_reflink] =
 160                 (1U << KEY_TYPE_deleted)|
 161                 (1U << KEY_TYPE_reflink_v)|
 162                 (1U << KEY_TYPE_indirect_inline_data),
 163         [BKEY_TYPE_subvolumes] =
 164                 (1U << KEY_TYPE_deleted)|
 165                 (1U << KEY_TYPE_subvolume),
 166         [BKEY_TYPE_snapshots] =
 167                 (1U << KEY_TYPE_deleted)|
 168                 (1U << KEY_TYPE_snapshot),
 169         [BKEY_TYPE_lru] =
 170                 (1U << KEY_TYPE_deleted)|
 171                 (1U << KEY_TYPE_lru),
 172         [BKEY_TYPE_freespace] =
 173                 (1U << KEY_TYPE_deleted)|
 174                 (1U << KEY_TYPE_set),
 175         [BKEY_TYPE_need_discard] =
 176                 (1U << KEY_TYPE_deleted)|
 177                 (1U << KEY_TYPE_set),
 178         [BKEY_TYPE_btree] =
 179                 (1U << KEY_TYPE_deleted)|
 180                 (1U << KEY_TYPE_btree_ptr)|
 181                 (1U << KEY_TYPE_btree_ptr_v2),
 182 };
 183
 184 const char *__bch2_bkey_invalid(struct bch_fs *c, struct bkey_s_c k,
 185                                 enum btree_node_type type)
 186 {
 187         if (k.k->u64s < BKEY_U64s)
 188                 return "u64s too small";
 189
 190         if (!(bch2_key_types_allowed[type] & (1U << k.k->type)))
 191                 return "invalid key type for this btree";
 192
 193         if (type == BKEY_TYPE_btree &&
 194             bkey_val_u64s(k.k) > BKEY_BTREE_PTR_VAL_U64s_MAX)
 195                 return "value too big";
 196
 197         if (btree_node_type_is_extents(type) && !bkey_whiteout(k.k)) {
 198                 if (k.k->size == 0)
 199                         return "bad size field";
 200
 201                 if (k.k->size > k.k->p.offset)
 202                         return "size greater than offset";
 203         } else {
 204                 if (k.k->size)
 205                         return "nonzero size field";
 206         }
 207
 208         if (type != BKEY_TYPE_btree &&
 209             !btree_type_has_snapshots(type) &&
 210             k.k->p.snapshot)
 211                 return "nonzero snapshot";
 212
 213         if (type != BKEY_TYPE_btree &&
 214             btree_type_has_snapshots(type) &&
 215             !k.k->p.snapshot)
 216                 return "invalid snapshot field";
 217
 218         if (type != BKEY_TYPE_btree &&
 219             !bkey_cmp(k.k->p, POS_MAX))
 220                 return "POS_MAX key";
 221
 222         return NULL;
 223 }
 224
 225 const char *bch2_bkey_invalid(struct bch_fs *c, struct bkey_s_c k,
 226                               enum btree_node_type type)
 227 {
 228         return __bch2_bkey_invalid(c, k, type) ?:
 229                 bch2_bkey_val_invalid(c, k);
 230 }
 231
 232 const char *bch2_bkey_in_btree_node(struct btree *b, struct bkey_s_c k)
 233 {
 234         if (bpos_cmp(k.k->p, b->data->min_key) < 0)
 235                 return "key before start of btree node";
 236
 237         if (bpos_cmp(k.k->p, b->data->max_key) > 0)
 238                 return "key past end of btree node";
 239
 240         return NULL;
 241 }
 242
 243 void bch2_bpos_to_text(struct printbuf *out, struct bpos pos)
 244 {
 245         if (!bpos_cmp(pos, POS_MIN))
 246                 pr_buf(out, "POS_MIN");
 247         else if (!bpos_cmp(pos, POS_MAX))
 248                 pr_buf(out, "POS_MAX");
 249         else if (!bpos_cmp(pos, SPOS_MAX))
 250                 pr_buf(out, "SPOS_MAX");
 251         else {
 252                 if (pos.inode == U64_MAX)
 253                         pr_buf(out, "U64_MAX");
 254                 else
 255                         pr_buf(out, "%llu", pos.inode);
 256                 pr_buf(out, ":");
 257                 if (pos.offset == U64_MAX)
 258                         pr_buf(out, "U64_MAX");
 259                 else
 260                         pr_buf(out, "%llu", pos.offset);
 261                 pr_buf(out, ":");
 262                 if (pos.snapshot == U32_MAX)
 263                         pr_buf(out, "U32_MAX");
 264                 else
 265                         pr_buf(out, "%u", pos.snapshot);
 266         }
 267 }
 268
 269 void bch2_bkey_to_text(struct printbuf *out, const struct bkey *k)
 270 {
 271         if (k) {
 272                 pr_buf(out, "u64s %u type ", k->u64s);
 273
 274                 if (k->type < KEY_TYPE_MAX)
 275                         pr_buf(out, "%s ", bch2_bkey_types[k->type]);
 276                 else
 277                         pr_buf(out, "%u ", k->type);
 278
 279                 bch2_bpos_to_text(out, k->p);
 280
 281                 pr_buf(out, " len %u ver %llu", k->size, k->version.lo);
 282         } else {
 283                 pr_buf(out, "(null)");
 284         }
 285 }
 286
 287 void bch2_val_to_text(struct printbuf *out, struct bch_fs *c,
 288                       struct bkey_s_c k)
 289 {
 290         if (k.k->type < KEY_TYPE_MAX) {
 291                 const struct bkey_ops *ops = &bch2_bkey_ops[k.k->type];
 292
 293                 if (likely(ops->val_to_text))
 294                         ops->val_to_text(out, c, k);
 295         } else {
 296                 pr_buf(out, "(invalid type %u)", k.k->type);
 297         }
 298 }
 299
 300 void bch2_bkey_val_to_text(struct printbuf *out, struct bch_fs *c,
 301                            struct bkey_s_c k)
 302 {
 303         bch2_bkey_to_text(out, k.k);
 304
 305         if (bkey_val_bytes(k.k)) {
 306                 pr_buf(out, ": ");
 307                 bch2_val_to_text(out, c, k);
 308         }
 309 }
 310
 311 void bch2_bkey_swab_val(struct bkey_s k)
 312 {
 313         const struct bkey_ops *ops = &bch2_bkey_ops[k.k->type];
 314
 315         if (ops->swab)
 316                 ops->swab(k);
 317 }
 318
 319 bool bch2_bkey_normalize(struct bch_fs *c, struct bkey_s k)
 320 {
 321         const struct bkey_ops *ops = &bch2_bkey_ops[k.k->type];
 322
 323         return ops->key_normalize
 324                 ? ops->key_normalize(c, k)
 325                 : false;
 326 }
 327
 328 bool bch2_bkey_merge(struct bch_fs *c, struct bkey_s l, struct bkey_s_c r)
 329 {
 330         const struct bkey_ops *ops = &bch2_bkey_ops[l.k->type];
 331
 332         return bch2_bkey_maybe_mergable(l.k, r.k) && ops->key_merge(c, l, r);
 333 }
 334
 335 static const struct old_bkey_type {
 336         u8              btree_node_type;
 337         u8              old;
 338         u8              new;
 339 } bkey_renumber_table[] = {
 340         {BKEY_TYPE_btree,       128, KEY_TYPE_btree_ptr         },
 341         {BKEY_TYPE_extents,     128, KEY_TYPE_extent            },
 342         {BKEY_TYPE_extents,     129, KEY_TYPE_extent            },
 343         {BKEY_TYPE_extents,     130, KEY_TYPE_reservation       },
 344         {BKEY_TYPE_inodes,      128, KEY_TYPE_inode             },
 345         {BKEY_TYPE_inodes,      130, KEY_TYPE_inode_generation  },
 346         {BKEY_TYPE_dirents,     128, KEY_TYPE_dirent            },
 347         {BKEY_TYPE_dirents,     129, KEY_TYPE_hash_whiteout     },
 348         {BKEY_TYPE_xattrs,      128, KEY_TYPE_xattr             },
 349         {BKEY_TYPE_xattrs,      129, KEY_TYPE_hash_whiteout     },
 350         {BKEY_TYPE_alloc,       128, KEY_TYPE_alloc             },
 351         {BKEY_TYPE_quotas,      128, KEY_TYPE_quota             },
 352 };
 353
 354 void bch2_bkey_renumber(enum btree_node_type btree_node_type,
 355                         struct bkey_packed *k,
 356                         int write)
 357 {
 358         const struct old_bkey_type *i;
 359
 360         for (i = bkey_renumber_table;
 361              i < bkey_renumber_table + ARRAY_SIZE(bkey_renumber_table);
 362              i++)
 363                 if (btree_node_type == i->btree_node_type &&
 364                     k->type == (write ? i->new : i->old)) {
 365                         k->type = write ? i->old : i->new;
 366                         break;
 367                 }
 368 }
 369
 370 void __bch2_bkey_compat(unsigned level, enum btree_id btree_id,
 371                         unsigned version, unsigned big_endian,
 372                         int write,
 373                         struct bkey_format *f,
 374                         struct bkey_packed *k)
 375 {
 376         const struct bkey_ops *ops;
 377         struct bkey uk;
 378         struct bkey_s u;
 379         unsigned nr_compat = 5;
 380         int i;
 381
 382         /*
 383          * Do these operations in reverse order in the write path:
 384          */
 385
 386         for (i = 0; i < nr_compat; i++)
 387         switch (!write ? i : nr_compat - 1 - i) {
 388         case 0:
 389                 if (big_endian != CPU_BIG_ENDIAN)
 390                         bch2_bkey_swab_key(f, k);
 391                 break;
 392         case 1:
 393                 if (version < bcachefs_metadata_version_bkey_renumber)
 394                         bch2_bkey_renumber(__btree_node_type(level, btree_id), k, write);
 395                 break;
 396         case 2:
 397                 if (version < bcachefs_metadata_version_inode_btree_change &&
 398                     btree_id == BTREE_ID_inodes) {
 399                         if (!bkey_packed(k)) {
 400                                 struct bkey_i *u = packed_to_bkey(k);
 401                                 swap(u->k.p.inode, u->k.p.offset);
 402                         } else if (f->bits_per_field[BKEY_FIELD_INODE] &&
 403                                    f->bits_per_field[BKEY_FIELD_OFFSET]) {
 404                                 struct bkey_format tmp = *f, *in = f, *out = &tmp;
 405
 406                                 swap(tmp.bits_per_field[BKEY_FIELD_INODE],
 407                                      tmp.bits_per_field[BKEY_FIELD_OFFSET]);
 408                                 swap(tmp.field_offset[BKEY_FIELD_INODE],
 409                                      tmp.field_offset[BKEY_FIELD_OFFSET]);
 410
 411                                 if (!write)
 412                                         swap(in, out);
 413
 414                                 uk = __bch2_bkey_unpack_key(in, k);
 415                                 swap(uk.p.inode, uk.p.offset);
 416                                 BUG_ON(!bch2_bkey_pack_key(k, &uk, out));
 417                         }
 418                 }
 419                 break;
 420         case 3:
 421                 if (version < bcachefs_metadata_version_snapshot &&
 422                     (level || btree_type_has_snapshots(btree_id))) {
 423                         struct bkey_i *u = packed_to_bkey(k);
 424
 425                         if (u) {
 426                                 u->k.p.snapshot = write
 427                                         ? 0 : U32_MAX;
 428                         } else {
 429                                 u64 min_packed = f->field_offset[BKEY_FIELD_SNAPSHOT];
 430                                 u64 max_packed = min_packed +
 431                                         ~(~0ULL << f->bits_per_field[BKEY_FIELD_SNAPSHOT]);
 432
 433                                 uk = __bch2_bkey_unpack_key(f, k);
 434                                 uk.p.snapshot = write
 435                                         ? min_packed : min_t(u64, U32_MAX, max_packed);
 436
 437                                 BUG_ON(!bch2_bkey_pack_key(k, &uk, f));
 438                         }
 439                 }
 440
 441                 break;
 442         case 4:
 443                 if (!bkey_packed(k)) {
 444                         u = bkey_i_to_s(packed_to_bkey(k));
 445                 } else {
 446                         uk = __bch2_bkey_unpack_key(f, k);
 447                         u.k = &uk;
 448                         u.v = bkeyp_val(f, k);
 449                 }
 450
 451                 if (big_endian != CPU_BIG_ENDIAN)
 452                         bch2_bkey_swab_val(u);
 453
 454                 ops = &bch2_bkey_ops[k->type];
 455
 456                 if (ops->compat)
 457                         ops->compat(btree_id, version, big_endian, write, u);
 458                 break;
 459         default:
 460                 BUG();
 461         }
 462 }