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                                unsigned flags, 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                                  unsigned flags, 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 -BCH_ERR_invalid_bkey;
  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                                    unsigned flags, struct printbuf *err)
  58 {
  59         return 0;
  60 }
  61
  62 #define bch2_bkey_ops_cookie ((struct bkey_ops) {       \
  63         .key_invalid    = key_type_cookie_invalid,      \
  64         .min_val_size   = 8,                            \
  65 })
  66
  67 #define bch2_bkey_ops_hash_whiteout ((struct bkey_ops) {\
  68         .key_invalid = empty_val_key_invalid,           \
  69 })
  70
  71 static int key_type_inline_data_invalid(const struct bch_fs *c, struct bkey_s_c k,
  72                                         unsigned flags, struct printbuf *err)
  73 {
  74         return 0;
  75 }
  76
  77 static void key_type_inline_data_to_text(struct printbuf *out, struct bch_fs *c,
  78                                          struct bkey_s_c k)
  79 {
  80         struct bkey_s_c_inline_data d = bkey_s_c_to_inline_data(k);
  81         unsigned datalen = bkey_inline_data_bytes(k.k);
  82
  83         prt_printf(out, "datalen %u: %*phN",
  84                datalen, min(datalen, 32U), d.v->data);
  85 }
  86
  87 #define bch2_bkey_ops_inline_data ((struct bkey_ops) {  \
  88         .key_invalid    = key_type_inline_data_invalid, \
  89         .val_to_text    = key_type_inline_data_to_text, \
  90 })
  91
  92 static int key_type_set_invalid(const struct bch_fs *c, struct bkey_s_c k,
  93                                 unsigned flags, struct printbuf *err)
  94 {
  95         if (bkey_val_bytes(k.k)) {
  96                 prt_printf(err, "incorrect value size (%zu != %zu)",
  97                        bkey_val_bytes(k.k), sizeof(struct bch_cookie));
  98                 return -BCH_ERR_invalid_bkey;
  99         }
 100
 101         return 0;
 102 }
 103
 104 static bool key_type_set_merge(struct bch_fs *c, struct bkey_s l, struct bkey_s_c r)
 105 {
 106         bch2_key_resize(l.k, l.k->size + r.k->size);
 107         return true;
 108 }
 109
 110 #define bch2_bkey_ops_set ((struct bkey_ops) {          \
 111         .key_invalid    = key_type_set_invalid,         \
 112         .key_merge      = key_type_set_merge,           \
 113 })
 114
 115 const struct bkey_ops bch2_bkey_ops[] = {
 116 #define x(name, nr) [KEY_TYPE_##name]   = bch2_bkey_ops_##name,
 117         BCH_BKEY_TYPES()
 118 #undef x
 119 };
 120
 121 const struct bkey_ops bch2_bkey_null_ops = {
 122         .min_val_size = U8_MAX,
 123 };
 124
 125 int bch2_bkey_val_invalid(struct bch_fs *c, struct bkey_s_c k,
 126                           enum bkey_invalid_flags flags,
 127                           struct printbuf *err)
 128 {
 129         const struct bkey_ops *ops = bch2_bkey_type_ops(k.k->type);
 130
 131         if (bkey_val_bytes(k.k) < ops->min_val_size) {
 132                 prt_printf(err, "bad val size (%zu < %u)",
 133                            bkey_val_bytes(k.k), ops->min_val_size);
 134                 return -BCH_ERR_invalid_bkey;
 135         }
 136
 137         if (!ops->key_invalid)
 138                 return 0;
 139
 140         return ops->key_invalid(c, k, flags, err);
 141 }
 142
 143 static u64 bch2_key_types_allowed[] = {
 144 #define x(name, nr, flags, keys)        [BKEY_TYPE_##name] = BIT_ULL(KEY_TYPE_deleted)|keys,
 145         BCH_BTREE_IDS()
 146 #undef x
 147         [BKEY_TYPE_btree] =
 148                 BIT_ULL(KEY_TYPE_deleted)|
 149                 BIT_ULL(KEY_TYPE_btree_ptr)|
 150                 BIT_ULL(KEY_TYPE_btree_ptr_v2),
 151 };
 152
 153 int __bch2_bkey_invalid(struct bch_fs *c, struct bkey_s_c k,
 154                         enum btree_node_type type,
 155                         enum bkey_invalid_flags flags,
 156                         struct printbuf *err)
 157 {
 158         if (k.k->u64s < BKEY_U64s) {
 159                 prt_printf(err, "u64s too small (%u < %zu)", k.k->u64s, BKEY_U64s);
 160                 return -BCH_ERR_invalid_bkey;
 161         }
 162
 163         if (flags & BKEY_INVALID_COMMIT  &&
 164             !(bch2_key_types_allowed[type] & BIT_ULL(k.k->type))) {
 165                 prt_printf(err, "invalid key type for btree %s (%s)",
 166                            bch2_btree_ids[type], bch2_bkey_types[k.k->type]);
 167                 return -BCH_ERR_invalid_bkey;
 168         }
 169
 170         if (btree_node_type_is_extents(type) && !bkey_whiteout(k.k)) {
 171                 if (k.k->size == 0) {
 172                         prt_printf(err, "size == 0");
 173                         return -BCH_ERR_invalid_bkey;
 174                 }
 175
 176                 if (k.k->size > k.k->p.offset) {
 177                         prt_printf(err, "size greater than offset (%u > %llu)",
 178                                k.k->size, k.k->p.offset);
 179                         return -BCH_ERR_invalid_bkey;
 180                 }
 181         } else {
 182                 if (k.k->size) {
 183                         prt_printf(err, "size != 0");
 184                         return -BCH_ERR_invalid_bkey;
 185                 }
 186         }
 187
 188         if (type != BKEY_TYPE_btree) {
 189                 if (!btree_type_has_snapshots((enum btree_id) type) &&
 190                     k.k->p.snapshot) {
 191                         prt_printf(err, "nonzero snapshot");
 192                         return -BCH_ERR_invalid_bkey;
 193                 }
 194
 195                 if (btree_type_has_snapshots((enum btree_id) type) &&
 196                     !k.k->p.snapshot) {
 197                         prt_printf(err, "snapshot == 0");
 198                         return -BCH_ERR_invalid_bkey;
 199                 }
 200
 201                 if (bkey_eq(k.k->p, POS_MAX)) {
 202                         prt_printf(err, "key at POS_MAX");
 203                         return -BCH_ERR_invalid_bkey;
 204                 }
 205         }
 206
 207         return 0;
 208 }
 209
 210 int bch2_bkey_invalid(struct bch_fs *c, struct bkey_s_c k,
 211                       enum btree_node_type type,
 212                       enum bkey_invalid_flags flags,
 213                       struct printbuf *err)
 214 {
 215         return __bch2_bkey_invalid(c, k, type, flags, err) ?:
 216                 bch2_bkey_val_invalid(c, k, flags, err);
 217 }
 218
 219 int bch2_bkey_in_btree_node(struct btree *b, struct bkey_s_c k,
 220                             struct printbuf *err)
 221 {
 222         if (bpos_lt(k.k->p, b->data->min_key)) {
 223                 prt_printf(err, "key before start of btree node");
 224                 return -BCH_ERR_invalid_bkey;
 225         }
 226
 227         if (bpos_gt(k.k->p, b->data->max_key)) {
 228                 prt_printf(err, "key past end of btree node");
 229                 return -BCH_ERR_invalid_bkey;
 230         }
 231
 232         return 0;
 233 }
 234
 235 void bch2_bpos_to_text(struct printbuf *out, struct bpos pos)
 236 {
 237         if (bpos_eq(pos, POS_MIN))
 238                 prt_printf(out, "POS_MIN");
 239         else if (bpos_eq(pos, POS_MAX))
 240                 prt_printf(out, "POS_MAX");
 241         else if (bpos_eq(pos, SPOS_MAX))
 242                 prt_printf(out, "SPOS_MAX");
 243         else {
 244                 if (pos.inode == U64_MAX)
 245                         prt_printf(out, "U64_MAX");
 246                 else
 247                         prt_printf(out, "%llu", pos.inode);
 248                 prt_printf(out, ":");
 249                 if (pos.offset == U64_MAX)
 250                         prt_printf(out, "U64_MAX");
 251                 else
 252                         prt_printf(out, "%llu", pos.offset);
 253                 prt_printf(out, ":");
 254                 if (pos.snapshot == U32_MAX)
 255                         prt_printf(out, "U32_MAX");
 256                 else
 257                         prt_printf(out, "%u", pos.snapshot);
 258         }
 259 }
 260
 261 void bch2_bkey_to_text(struct printbuf *out, const struct bkey *k)
 262 {
 263         if (k) {
 264                 prt_printf(out, "u64s %u type ", k->u64s);
 265
 266                 if (k->type < KEY_TYPE_MAX)
 267                         prt_printf(out, "%s ", bch2_bkey_types[k->type]);
 268                 else
 269                         prt_printf(out, "%u ", k->type);
 270
 271                 bch2_bpos_to_text(out, k->p);
 272
 273                 prt_printf(out, " len %u ver %llu", k->size, k->version.lo);
 274         } else {
 275                 prt_printf(out, "(null)");
 276         }
 277 }
 278
 279 void bch2_val_to_text(struct printbuf *out, struct bch_fs *c,
 280                       struct bkey_s_c k)
 281 {
 282         const struct bkey_ops *ops = bch2_bkey_type_ops(k.k->type);
 283
 284         if (likely(ops->val_to_text))
 285                 ops->val_to_text(out, c, k);
 286 }
 287
 288 void bch2_bkey_val_to_text(struct printbuf *out, struct bch_fs *c,
 289                            struct bkey_s_c k)
 290 {
 291         bch2_bkey_to_text(out, k.k);
 292
 293         if (bkey_val_bytes(k.k)) {
 294                 prt_printf(out, ": ");
 295                 bch2_val_to_text(out, c, k);
 296         }
 297 }
 298
 299 void bch2_bkey_swab_val(struct bkey_s k)
 300 {
 301         const struct bkey_ops *ops = bch2_bkey_type_ops(k.k->type);
 302
 303         if (ops->swab)
 304                 ops->swab(k);
 305 }
 306
 307 bool bch2_bkey_normalize(struct bch_fs *c, struct bkey_s k)
 308 {
 309         const struct bkey_ops *ops = bch2_bkey_type_ops(k.k->type);
 310
 311         return ops->key_normalize
 312                 ? ops->key_normalize(c, k)
 313                 : false;
 314 }
 315
 316 bool bch2_bkey_merge(struct bch_fs *c, struct bkey_s l, struct bkey_s_c r)
 317 {
 318         const struct bkey_ops *ops = bch2_bkey_type_ops(l.k->type);
 319
 320         return ops->key_merge &&
 321                 bch2_bkey_maybe_mergable(l.k, r.k) &&
 322                 (u64) l.k->size + r.k->size <= KEY_SIZE_MAX &&
 323                 !bch2_key_merging_disabled &&
 324                 ops->key_merge(c, l, r);
 325 }
 326
 327 static const struct old_bkey_type {
 328         u8              btree_node_type;
 329         u8              old;
 330         u8              new;
 331 } bkey_renumber_table[] = {
 332         {BKEY_TYPE_btree,       128, KEY_TYPE_btree_ptr         },
 333         {BKEY_TYPE_extents,     128, KEY_TYPE_extent            },
 334         {BKEY_TYPE_extents,     129, KEY_TYPE_extent            },
 335         {BKEY_TYPE_extents,     130, KEY_TYPE_reservation       },
 336         {BKEY_TYPE_inodes,      128, KEY_TYPE_inode             },
 337         {BKEY_TYPE_inodes,      130, KEY_TYPE_inode_generation  },
 338         {BKEY_TYPE_dirents,     128, KEY_TYPE_dirent            },
 339         {BKEY_TYPE_dirents,     129, KEY_TYPE_hash_whiteout     },
 340         {BKEY_TYPE_xattrs,      128, KEY_TYPE_xattr             },
 341         {BKEY_TYPE_xattrs,      129, KEY_TYPE_hash_whiteout     },
 342         {BKEY_TYPE_alloc,       128, KEY_TYPE_alloc             },
 343         {BKEY_TYPE_quotas,      128, KEY_TYPE_quota             },
 344 };
 345
 346 void bch2_bkey_renumber(enum btree_node_type btree_node_type,
 347                         struct bkey_packed *k,
 348                         int write)
 349 {
 350         const struct old_bkey_type *i;
 351
 352         for (i = bkey_renumber_table;
 353              i < bkey_renumber_table + ARRAY_SIZE(bkey_renumber_table);
 354              i++)
 355                 if (btree_node_type == i->btree_node_type &&
 356                     k->type == (write ? i->new : i->old)) {
 357                         k->type = write ? i->old : i->new;
 358                         break;
 359                 }
 360 }
 361
 362 void __bch2_bkey_compat(unsigned level, enum btree_id btree_id,
 363                         unsigned version, unsigned big_endian,
 364                         int write,
 365                         struct bkey_format *f,
 366                         struct bkey_packed *k)
 367 {
 368         const struct bkey_ops *ops;
 369         struct bkey uk;
 370         struct bkey_s u;
 371         unsigned nr_compat = 5;
 372         int i;
 373
 374         /*
 375          * Do these operations in reverse order in the write path:
 376          */
 377
 378         for (i = 0; i < nr_compat; i++)
 379         switch (!write ? i : nr_compat - 1 - i) {
 380         case 0:
 381                 if (big_endian != CPU_BIG_ENDIAN)
 382                         bch2_bkey_swab_key(f, k);
 383                 break;
 384         case 1:
 385                 if (version < bcachefs_metadata_version_bkey_renumber)
 386                         bch2_bkey_renumber(__btree_node_type(level, btree_id), k, write);
 387                 break;
 388         case 2:
 389                 if (version < bcachefs_metadata_version_inode_btree_change &&
 390                     btree_id == BTREE_ID_inodes) {
 391                         if (!bkey_packed(k)) {
 392                                 struct bkey_i *u = packed_to_bkey(k);
 393
 394                                 swap(u->k.p.inode, u->k.p.offset);
 395                         } else if (f->bits_per_field[BKEY_FIELD_INODE] &&
 396                                    f->bits_per_field[BKEY_FIELD_OFFSET]) {
 397                                 struct bkey_format tmp = *f, *in = f, *out = &tmp;
 398
 399                                 swap(tmp.bits_per_field[BKEY_FIELD_INODE],
 400                                      tmp.bits_per_field[BKEY_FIELD_OFFSET]);
 401                                 swap(tmp.field_offset[BKEY_FIELD_INODE],
 402                                      tmp.field_offset[BKEY_FIELD_OFFSET]);
 403
 404                                 if (!write)
 405                                         swap(in, out);
 406
 407                                 uk = __bch2_bkey_unpack_key(in, k);
 408                                 swap(uk.p.inode, uk.p.offset);
 409                                 BUG_ON(!bch2_bkey_pack_key(k, &uk, out));
 410                         }
 411                 }
 412                 break;
 413         case 3:
 414                 if (version < bcachefs_metadata_version_snapshot &&
 415                     (level || btree_type_has_snapshots(btree_id))) {
 416                         struct bkey_i *u = packed_to_bkey(k);
 417
 418                         if (u) {
 419                                 u->k.p.snapshot = write
 420                                         ? 0 : U32_MAX;
 421                         } else {
 422                                 u64 min_packed = le64_to_cpu(f->field_offset[BKEY_FIELD_SNAPSHOT]);
 423                                 u64 max_packed = min_packed +
 424                                         ~(~0ULL << f->bits_per_field[BKEY_FIELD_SNAPSHOT]);
 425
 426                                 uk = __bch2_bkey_unpack_key(f, k);
 427                                 uk.p.snapshot = write
 428                                         ? min_packed : min_t(u64, U32_MAX, max_packed);
 429
 430                                 BUG_ON(!bch2_bkey_pack_key(k, &uk, f));
 431                         }
 432                 }
 433
 434                 break;
 435         case 4:
 436                 if (!bkey_packed(k)) {
 437                         u = bkey_i_to_s(packed_to_bkey(k));
 438                 } else {
 439                         uk = __bch2_bkey_unpack_key(f, k);
 440                         u.k = &uk;
 441                         u.v = bkeyp_val(f, k);
 442                 }
 443
 444                 if (big_endian != CPU_BIG_ENDIAN)
 445                         bch2_bkey_swab_val(u);
 446
 447                 ops = bch2_bkey_type_ops(k->type);
 448
 449                 if (ops->compat)
 450                         ops->compat(btree_id, version, big_endian, write, u);
 451                 break;
 452         default:
 453                 BUG();
 454         }
 455 }