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                 (1U << KEY_TYPE_alloc_v4),
 154         [BKEY_TYPE_quotas] =
 155                 (1U << KEY_TYPE_deleted)|
 156                 (1U << KEY_TYPE_quota),
 157         [BKEY_TYPE_stripes] =
 158                 (1U << KEY_TYPE_deleted)|
 159                 (1U << KEY_TYPE_stripe),
 160         [BKEY_TYPE_reflink] =
 161                 (1U << KEY_TYPE_deleted)|
 162                 (1U << KEY_TYPE_reflink_v)|
 163                 (1U << KEY_TYPE_indirect_inline_data),
 164         [BKEY_TYPE_subvolumes] =
 165                 (1U << KEY_TYPE_deleted)|
 166                 (1U << KEY_TYPE_subvolume),
 167         [BKEY_TYPE_snapshots] =
 168                 (1U << KEY_TYPE_deleted)|
 169                 (1U << KEY_TYPE_snapshot),
 170         [BKEY_TYPE_lru] =
 171                 (1U << KEY_TYPE_deleted)|
 172                 (1U << KEY_TYPE_lru),
 173         [BKEY_TYPE_freespace] =
 174                 (1U << KEY_TYPE_deleted)|
 175                 (1U << KEY_TYPE_set),
 176         [BKEY_TYPE_need_discard] =
 177                 (1U << KEY_TYPE_deleted)|
 178                 (1U << KEY_TYPE_set),
 179         [BKEY_TYPE_btree] =
 180                 (1U << KEY_TYPE_deleted)|
 181                 (1U << KEY_TYPE_btree_ptr)|
 182                 (1U << KEY_TYPE_btree_ptr_v2),
 183 };
 184
 185 const char *__bch2_bkey_invalid(struct bch_fs *c, struct bkey_s_c k,
 186                                 enum btree_node_type type)
 187 {
 188         if (k.k->u64s < BKEY_U64s)
 189                 return "u64s too small";
 190
 191         if (!(bch2_key_types_allowed[type] & (1U << k.k->type)))
 192                 return "invalid key type for this btree";
 193
 194         if (type == BKEY_TYPE_btree &&
 195             bkey_val_u64s(k.k) > BKEY_BTREE_PTR_VAL_U64s_MAX)
 196                 return "value too big";
 197
 198         if (btree_node_type_is_extents(type) && !bkey_whiteout(k.k)) {
 199                 if (k.k->size == 0)
 200                         return "bad size field";
 201
 202                 if (k.k->size > k.k->p.offset)
 203                         return "size greater than offset";
 204         } else {
 205                 if (k.k->size)
 206                         return "nonzero size field";
 207         }
 208
 209         if (type != BKEY_TYPE_btree &&
 210             !btree_type_has_snapshots(type) &&
 211             k.k->p.snapshot)
 212                 return "nonzero snapshot";
 213
 214         if (type != BKEY_TYPE_btree &&
 215             btree_type_has_snapshots(type) &&
 216             !k.k->p.snapshot)
 217                 return "invalid snapshot field";
 218
 219         if (type != BKEY_TYPE_btree &&
 220             !bkey_cmp(k.k->p, POS_MAX))
 221                 return "POS_MAX key";
 222
 223         return NULL;
 224 }
 225
 226 const char *bch2_bkey_invalid(struct bch_fs *c, struct bkey_s_c k,
 227                               enum btree_node_type type)
 228 {
 229         return __bch2_bkey_invalid(c, k, type) ?:
 230                 bch2_bkey_val_invalid(c, k);
 231 }
 232
 233 const char *bch2_bkey_in_btree_node(struct btree *b, struct bkey_s_c k)
 234 {
 235         if (bpos_cmp(k.k->p, b->data->min_key) < 0)
 236                 return "key before start of btree node";
 237
 238         if (bpos_cmp(k.k->p, b->data->max_key) > 0)
 239                 return "key past end of btree node";
 240
 241         return NULL;
 242 }
 243
 244 void bch2_bpos_to_text(struct printbuf *out, struct bpos pos)
 245 {
 246         if (!bpos_cmp(pos, POS_MIN))
 247                 pr_buf(out, "POS_MIN");
 248         else if (!bpos_cmp(pos, POS_MAX))
 249                 pr_buf(out, "POS_MAX");
 250         else if (!bpos_cmp(pos, SPOS_MAX))
 251                 pr_buf(out, "SPOS_MAX");
 252         else {
 253                 if (pos.inode == U64_MAX)
 254                         pr_buf(out, "U64_MAX");
 255                 else
 256                         pr_buf(out, "%llu", pos.inode);
 257                 pr_buf(out, ":");
 258                 if (pos.offset == U64_MAX)
 259                         pr_buf(out, "U64_MAX");
 260                 else
 261                         pr_buf(out, "%llu", pos.offset);
 262                 pr_buf(out, ":");
 263                 if (pos.snapshot == U32_MAX)
 264                         pr_buf(out, "U32_MAX");
 265                 else
 266                         pr_buf(out, "%u", pos.snapshot);
 267         }
 268 }
 269
 270 void bch2_bkey_to_text(struct printbuf *out, const struct bkey *k)
 271 {
 272         if (k) {
 273                 pr_buf(out, "u64s %u type ", k->u64s);
 274
 275                 if (k->type < KEY_TYPE_MAX)
 276                         pr_buf(out, "%s ", bch2_bkey_types[k->type]);
 277                 else
 278                         pr_buf(out, "%u ", k->type);
 279
 280                 bch2_bpos_to_text(out, k->p);
 281
 282                 pr_buf(out, " len %u ver %llu", k->size, k->version.lo);
 283         } else {
 284                 pr_buf(out, "(null)");
 285         }
 286 }
 287
 288 void bch2_val_to_text(struct printbuf *out, struct bch_fs *c,
 289                       struct bkey_s_c k)
 290 {
 291         if (k.k->type < KEY_TYPE_MAX) {
 292                 const struct bkey_ops *ops = &bch2_bkey_ops[k.k->type];
 293
 294                 if (likely(ops->val_to_text))
 295                         ops->val_to_text(out, c, k);
 296         } else {
 297                 pr_buf(out, "(invalid type %u)", k.k->type);
 298         }
 299 }
 300
 301 void bch2_bkey_val_to_text(struct printbuf *out, struct bch_fs *c,
 302                            struct bkey_s_c k)
 303 {
 304         bch2_bkey_to_text(out, k.k);
 305
 306         if (bkey_val_bytes(k.k)) {
 307                 pr_buf(out, ": ");
 308                 bch2_val_to_text(out, c, k);
 309         }
 310 }
 311
 312 void bch2_bkey_swab_val(struct bkey_s k)
 313 {
 314         const struct bkey_ops *ops = &bch2_bkey_ops[k.k->type];
 315
 316         if (ops->swab)
 317                 ops->swab(k);
 318 }
 319
 320 bool bch2_bkey_normalize(struct bch_fs *c, struct bkey_s k)
 321 {
 322         const struct bkey_ops *ops = &bch2_bkey_ops[k.k->type];
 323
 324         return ops->key_normalize
 325                 ? ops->key_normalize(c, k)
 326                 : false;
 327 }
 328
 329 bool bch2_bkey_merge(struct bch_fs *c, struct bkey_s l, struct bkey_s_c r)
 330 {
 331         const struct bkey_ops *ops = &bch2_bkey_ops[l.k->type];
 332
 333         return bch2_bkey_maybe_mergable(l.k, r.k) && ops->key_merge(c, l, r);
 334 }
 335
 336 static const struct old_bkey_type {
 337         u8              btree_node_type;
 338         u8              old;
 339         u8              new;
 340 } bkey_renumber_table[] = {
 341         {BKEY_TYPE_btree,       128, KEY_TYPE_btree_ptr         },
 342         {BKEY_TYPE_extents,     128, KEY_TYPE_extent            },
 343         {BKEY_TYPE_extents,     129, KEY_TYPE_extent            },
 344         {BKEY_TYPE_extents,     130, KEY_TYPE_reservation       },
 345         {BKEY_TYPE_inodes,      128, KEY_TYPE_inode             },
 346         {BKEY_TYPE_inodes,      130, KEY_TYPE_inode_generation  },
 347         {BKEY_TYPE_dirents,     128, KEY_TYPE_dirent            },
 348         {BKEY_TYPE_dirents,     129, KEY_TYPE_hash_whiteout     },
 349         {BKEY_TYPE_xattrs,      128, KEY_TYPE_xattr             },
 350         {BKEY_TYPE_xattrs,      129, KEY_TYPE_hash_whiteout     },
 351         {BKEY_TYPE_alloc,       128, KEY_TYPE_alloc             },
 352         {BKEY_TYPE_quotas,      128, KEY_TYPE_quota             },
 353 };
 354
 355 void bch2_bkey_renumber(enum btree_node_type btree_node_type,
 356                         struct bkey_packed *k,
 357                         int write)
 358 {
 359         const struct old_bkey_type *i;
 360
 361         for (i = bkey_renumber_table;
 362              i < bkey_renumber_table + ARRAY_SIZE(bkey_renumber_table);
 363              i++)
 364                 if (btree_node_type == i->btree_node_type &&
 365                     k->type == (write ? i->new : i->old)) {
 366                         k->type = write ? i->old : i->new;
 367                         break;
 368                 }
 369 }
 370
 371 void __bch2_bkey_compat(unsigned level, enum btree_id btree_id,
 372                         unsigned version, unsigned big_endian,
 373                         int write,
 374                         struct bkey_format *f,
 375                         struct bkey_packed *k)
 376 {
 377         const struct bkey_ops *ops;
 378         struct bkey uk;
 379         struct bkey_s u;
 380         unsigned nr_compat = 5;
 381         int i;
 382
 383         /*
 384          * Do these operations in reverse order in the write path:
 385          */
 386
 387         for (i = 0; i < nr_compat; i++)
 388         switch (!write ? i : nr_compat - 1 - i) {
 389         case 0:
 390                 if (big_endian != CPU_BIG_ENDIAN)
 391                         bch2_bkey_swab_key(f, k);
 392                 break;
 393         case 1:
 394                 if (version < bcachefs_metadata_version_bkey_renumber)
 395                         bch2_bkey_renumber(__btree_node_type(level, btree_id), k, write);
 396                 break;
 397         case 2:
 398                 if (version < bcachefs_metadata_version_inode_btree_change &&
 399                     btree_id == BTREE_ID_inodes) {
 400                         if (!bkey_packed(k)) {
 401                                 struct bkey_i *u = packed_to_bkey(k);
 402                                 swap(u->k.p.inode, u->k.p.offset);
 403                         } else if (f->bits_per_field[BKEY_FIELD_INODE] &&
 404                                    f->bits_per_field[BKEY_FIELD_OFFSET]) {
 405                                 struct bkey_format tmp = *f, *in = f, *out = &tmp;
 406
 407                                 swap(tmp.bits_per_field[BKEY_FIELD_INODE],
 408                                      tmp.bits_per_field[BKEY_FIELD_OFFSET]);
 409                                 swap(tmp.field_offset[BKEY_FIELD_INODE],
 410                                      tmp.field_offset[BKEY_FIELD_OFFSET]);
 411
 412                                 if (!write)
 413                                         swap(in, out);
 414
 415                                 uk = __bch2_bkey_unpack_key(in, k);
 416                                 swap(uk.p.inode, uk.p.offset);
 417                                 BUG_ON(!bch2_bkey_pack_key(k, &uk, out));
 418                         }
 419                 }
 420                 break;
 421         case 3:
 422                 if (version < bcachefs_metadata_version_snapshot &&
 423                     (level || btree_type_has_snapshots(btree_id))) {
 424                         struct bkey_i *u = packed_to_bkey(k);
 425
 426                         if (u) {
 427                                 u->k.p.snapshot = write
 428                                         ? 0 : U32_MAX;
 429                         } else {
 430                                 u64 min_packed = f->field_offset[BKEY_FIELD_SNAPSHOT];
 431                                 u64 max_packed = min_packed +
 432                                         ~(~0ULL << f->bits_per_field[BKEY_FIELD_SNAPSHOT]);
 433
 434                                 uk = __bch2_bkey_unpack_key(f, k);
 435                                 uk.p.snapshot = write
 436                                         ? min_packed : min_t(u64, U32_MAX, max_packed);
 437
 438                                 BUG_ON(!bch2_bkey_pack_key(k, &uk, f));
 439                         }
 440                 }
 441
 442                 break;
 443         case 4:
 444                 if (!bkey_packed(k)) {
 445                         u = bkey_i_to_s(packed_to_bkey(k));
 446                 } else {
 447                         uk = __bch2_bkey_unpack_key(f, k);
 448                         u.k = &uk;
 449                         u.v = bkeyp_val(f, k);
 450                 }
 451
 452                 if (big_endian != CPU_BIG_ENDIAN)
 453                         bch2_bkey_swab_val(u);
 454
 455                 ops = &bch2_bkey_ops[k->type];
 456
 457                 if (ops->compat)
 458                         ops->compat(btree_id, version, big_endian, write, u);
 459                 break;
 460         default:
 461                 BUG();
 462         }
 463 }