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 "quota.h"
  13 #include "reflink.h"
  14 #include "subvolume.h"
  15 #include "xattr.h"
  16
  17 const char * const bch2_bkey_types[] = {
  18 #define x(name, nr) #name,
  19         BCH_BKEY_TYPES()
  20 #undef x
  21         NULL
  22 };
  23
  24 static const char *deleted_key_invalid(const struct bch_fs *c,
  25                                         struct bkey_s_c k)
  26 {
  27         return NULL;
  28 }
  29
  30 #define bch2_bkey_ops_deleted (struct bkey_ops) {       \
  31         .key_invalid = deleted_key_invalid,             \
  32 }
  33
  34 #define bch2_bkey_ops_whiteout (struct bkey_ops) {      \
  35         .key_invalid = deleted_key_invalid,             \
  36 }
  37
  38 static const char *empty_val_key_invalid(const struct bch_fs *c, struct bkey_s_c k)
  39 {
  40         if (bkey_val_bytes(k.k))
  41                 return "value size should be zero";
  42
  43         return NULL;
  44 }
  45
  46 #define bch2_bkey_ops_error (struct bkey_ops) {         \
  47         .key_invalid = empty_val_key_invalid,           \
  48 }
  49
  50 static const char *key_type_cookie_invalid(const struct bch_fs *c,
  51                                            struct bkey_s_c k)
  52 {
  53         if (bkey_val_bytes(k.k) != sizeof(struct bch_cookie))
  54                 return "incorrect value size";
  55
  56         return NULL;
  57 }
  58
  59 #define bch2_bkey_ops_cookie (struct bkey_ops) {        \
  60         .key_invalid = key_type_cookie_invalid,         \
  61 }
  62
  63 #define bch2_bkey_ops_hash_whiteout (struct bkey_ops) { \
  64         .key_invalid = empty_val_key_invalid,           \
  65 }
  66
  67 static const char *key_type_inline_data_invalid(const struct bch_fs *c,
  68                                            struct bkey_s_c k)
  69 {
  70         return NULL;
  71 }
  72
  73 static void key_type_inline_data_to_text(struct printbuf *out, struct bch_fs *c,
  74                                          struct bkey_s_c k)
  75 {
  76         struct bkey_s_c_inline_data d = bkey_s_c_to_inline_data(k);
  77         unsigned datalen = bkey_inline_data_bytes(k.k);
  78
  79         pr_buf(out, "datalen %u: %*phN",
  80                datalen, min(datalen, 32U), d.v->data);
  81 }
  82
  83 #define bch2_bkey_ops_inline_data (struct bkey_ops) {   \
  84         .key_invalid    = key_type_inline_data_invalid, \
  85         .val_to_text    = key_type_inline_data_to_text, \
  86 }
  87
  88 const struct bkey_ops bch2_bkey_ops[] = {
  89 #define x(name, nr) [KEY_TYPE_##name]   = bch2_bkey_ops_##name,
  90         BCH_BKEY_TYPES()
  91 #undef x
  92 };
  93
  94 const char *bch2_bkey_val_invalid(struct bch_fs *c, struct bkey_s_c k)
  95 {
  96         if (k.k->type >= KEY_TYPE_MAX)
  97                 return "invalid type";
  98
  99         return bch2_bkey_ops[k.k->type].key_invalid(c, k);
 100 }
 101
 102 static unsigned bch2_key_types_allowed[] = {
 103         [BKEY_TYPE_extents] =
 104                 (1U << KEY_TYPE_deleted)|
 105                 (1U << KEY_TYPE_whiteout)|
 106                 (1U << KEY_TYPE_error)|
 107                 (1U << KEY_TYPE_cookie)|
 108                 (1U << KEY_TYPE_extent)|
 109                 (1U << KEY_TYPE_reservation)|
 110                 (1U << KEY_TYPE_reflink_p)|
 111                 (1U << KEY_TYPE_inline_data),
 112         [BKEY_TYPE_inodes] =
 113                 (1U << KEY_TYPE_deleted)|
 114                 (1U << KEY_TYPE_whiteout)|
 115                 (1U << KEY_TYPE_inode)|
 116                 (1U << KEY_TYPE_inode_generation),
 117         [BKEY_TYPE_dirents] =
 118                 (1U << KEY_TYPE_deleted)|
 119                 (1U << KEY_TYPE_whiteout)|
 120                 (1U << KEY_TYPE_hash_whiteout)|
 121                 (1U << KEY_TYPE_dirent),
 122         [BKEY_TYPE_xattrs] =
 123                 (1U << KEY_TYPE_deleted)|
 124                 (1U << KEY_TYPE_whiteout)|
 125                 (1U << KEY_TYPE_cookie)|
 126                 (1U << KEY_TYPE_hash_whiteout)|
 127                 (1U << KEY_TYPE_xattr),
 128         [BKEY_TYPE_alloc] =
 129                 (1U << KEY_TYPE_deleted)|
 130                 (1U << KEY_TYPE_alloc)|
 131                 (1U << KEY_TYPE_alloc_v2),
 132         [BKEY_TYPE_quotas] =
 133                 (1U << KEY_TYPE_deleted)|
 134                 (1U << KEY_TYPE_quota),
 135         [BKEY_TYPE_stripes] =
 136                 (1U << KEY_TYPE_deleted)|
 137                 (1U << KEY_TYPE_stripe),
 138         [BKEY_TYPE_reflink] =
 139                 (1U << KEY_TYPE_deleted)|
 140                 (1U << KEY_TYPE_reflink_v)|
 141                 (1U << KEY_TYPE_indirect_inline_data),
 142         [BKEY_TYPE_subvolumes] =
 143                 (1U << KEY_TYPE_deleted)|
 144                 (1U << KEY_TYPE_subvolume),
 145         [BKEY_TYPE_snapshots] =
 146                 (1U << KEY_TYPE_deleted)|
 147                 (1U << KEY_TYPE_snapshot),
 148         [BKEY_TYPE_btree] =
 149                 (1U << KEY_TYPE_deleted)|
 150                 (1U << KEY_TYPE_btree_ptr)|
 151                 (1U << KEY_TYPE_btree_ptr_v2),
 152 };
 153
 154 const char *__bch2_bkey_invalid(struct bch_fs *c, struct bkey_s_c k,
 155                                 enum btree_node_type type)
 156 {
 157         if (k.k->u64s < BKEY_U64s)
 158                 return "u64s too small";
 159
 160         if (!(bch2_key_types_allowed[type] & (1U << k.k->type)))
 161                 return "invalid key type for this btree";
 162
 163         if (type == BKEY_TYPE_btree &&
 164             bkey_val_u64s(k.k) > BKEY_BTREE_PTR_VAL_U64s_MAX)
 165                 return "value too big";
 166
 167         if (btree_node_type_is_extents(type) && !bkey_whiteout(k.k)) {
 168                 if (k.k->size == 0)
 169                         return "bad size field";
 170
 171                 if (k.k->size > k.k->p.offset)
 172                         return "size greater than offset";
 173         } else {
 174                 if (k.k->size)
 175                         return "nonzero size field";
 176         }
 177
 178         if (type != BKEY_TYPE_btree &&
 179             !btree_type_has_snapshots(type) &&
 180             k.k->p.snapshot)
 181                 return "nonzero snapshot";
 182
 183         if (type != BKEY_TYPE_btree &&
 184             btree_type_has_snapshots(type) &&
 185             !k.k->p.snapshot)
 186                 return "invalid snapshot field";
 187
 188         if (type != BKEY_TYPE_btree &&
 189             !bkey_cmp(k.k->p, POS_MAX))
 190                 return "POS_MAX key";
 191
 192         return NULL;
 193 }
 194
 195 const char *bch2_bkey_invalid(struct bch_fs *c, struct bkey_s_c k,
 196                               enum btree_node_type type)
 197 {
 198         return __bch2_bkey_invalid(c, k, type) ?:
 199                 bch2_bkey_val_invalid(c, k);
 200 }
 201
 202 const char *bch2_bkey_in_btree_node(struct btree *b, struct bkey_s_c k)
 203 {
 204         if (bpos_cmp(k.k->p, b->data->min_key) < 0)
 205                 return "key before start of btree node";
 206
 207         if (bpos_cmp(k.k->p, b->data->max_key) > 0)
 208                 return "key past end of btree node";
 209
 210         return NULL;
 211 }
 212
 213 void bch2_bkey_debugcheck(struct bch_fs *c, struct btree *b, struct bkey_s_c k)
 214 {
 215         const char *invalid;
 216
 217         BUG_ON(!k.k->u64s);
 218
 219         invalid = bch2_bkey_invalid(c, k, btree_node_type(b)) ?:
 220                 bch2_bkey_in_btree_node(b, k);
 221         if (invalid) {
 222                 char buf[160];
 223
 224                 bch2_bkey_val_to_text(&PBUF(buf), c, k);
 225                 bch2_fs_inconsistent(c, "invalid bkey %s: %s", buf, invalid);
 226         }
 227 }
 228
 229 void bch2_bpos_to_text(struct printbuf *out, struct bpos pos)
 230 {
 231         if (!bpos_cmp(pos, POS_MIN))
 232                 pr_buf(out, "POS_MIN");
 233         else if (!bpos_cmp(pos, POS_MAX))
 234                 pr_buf(out, "POS_MAX");
 235         else if (!bpos_cmp(pos, SPOS_MAX))
 236                 pr_buf(out, "SPOS_MAX");
 237         else {
 238                 if (pos.inode == U64_MAX)
 239                         pr_buf(out, "U64_MAX");
 240                 else
 241                         pr_buf(out, "%llu", pos.inode);
 242                 pr_buf(out, ":");
 243                 if (pos.offset == U64_MAX)
 244                         pr_buf(out, "U64_MAX");
 245                 else
 246                         pr_buf(out, "%llu", pos.offset);
 247                 pr_buf(out, ":");
 248                 if (pos.snapshot == U32_MAX)
 249                         pr_buf(out, "U32_MAX");
 250                 else
 251                         pr_buf(out, "%u", pos.snapshot);
 252         }
 253 }
 254
 255 void bch2_bkey_to_text(struct printbuf *out, const struct bkey *k)
 256 {
 257         if (k) {
 258                 pr_buf(out, "u64s %u type ", k->u64s);
 259
 260                 if (k->type < KEY_TYPE_MAX)
 261                         pr_buf(out, "%s ", bch2_bkey_types[k->type]);
 262                 else
 263                         pr_buf(out, "%u ", k->type);
 264
 265                 bch2_bpos_to_text(out, k->p);
 266
 267                 pr_buf(out, " len %u ver %llu", k->size, k->version.lo);
 268         } else {
 269                 pr_buf(out, "(null)");
 270         }
 271 }
 272
 273 void bch2_val_to_text(struct printbuf *out, struct bch_fs *c,
 274                       struct bkey_s_c k)
 275 {
 276         if (k.k->type < KEY_TYPE_MAX) {
 277                 const struct bkey_ops *ops = &bch2_bkey_ops[k.k->type];
 278
 279                 if (likely(ops->val_to_text))
 280                         ops->val_to_text(out, c, k);
 281         } else {
 282                 pr_buf(out, "(invalid type %u)", k.k->type);
 283         }
 284 }
 285
 286 void bch2_bkey_val_to_text(struct printbuf *out, struct bch_fs *c,
 287                            struct bkey_s_c k)
 288 {
 289         bch2_bkey_to_text(out, k.k);
 290
 291         if (bkey_val_bytes(k.k)) {
 292                 pr_buf(out, ": ");
 293                 bch2_val_to_text(out, c, k);
 294         }
 295 }
 296
 297 void bch2_bkey_swab_val(struct bkey_s k)
 298 {
 299         const struct bkey_ops *ops = &bch2_bkey_ops[k.k->type];
 300
 301         if (ops->swab)
 302                 ops->swab(k);
 303 }
 304
 305 bool bch2_bkey_normalize(struct bch_fs *c, struct bkey_s k)
 306 {
 307         const struct bkey_ops *ops = &bch2_bkey_ops[k.k->type];
 308
 309         return ops->key_normalize
 310                 ? ops->key_normalize(c, k)
 311                 : false;
 312 }
 313
 314 bool bch2_bkey_merge(struct bch_fs *c, struct bkey_s l, struct bkey_s_c r)
 315 {
 316         const struct bkey_ops *ops = &bch2_bkey_ops[l.k->type];
 317
 318         return bch2_bkey_maybe_mergable(l.k, r.k) && ops->key_merge(c, l, r);
 319 }
 320
 321 static const struct old_bkey_type {
 322         u8              btree_node_type;
 323         u8              old;
 324         u8              new;
 325 } bkey_renumber_table[] = {
 326         {BKEY_TYPE_btree,       128, KEY_TYPE_btree_ptr         },
 327         {BKEY_TYPE_extents,     128, KEY_TYPE_extent            },
 328         {BKEY_TYPE_extents,     129, KEY_TYPE_extent            },
 329         {BKEY_TYPE_extents,     130, KEY_TYPE_reservation       },
 330         {BKEY_TYPE_inodes,      128, KEY_TYPE_inode             },
 331         {BKEY_TYPE_inodes,      130, KEY_TYPE_inode_generation  },
 332         {BKEY_TYPE_dirents,     128, KEY_TYPE_dirent            },
 333         {BKEY_TYPE_dirents,     129, KEY_TYPE_hash_whiteout     },
 334         {BKEY_TYPE_xattrs,      128, KEY_TYPE_xattr             },
 335         {BKEY_TYPE_xattrs,      129, KEY_TYPE_hash_whiteout     },
 336         {BKEY_TYPE_alloc,       128, KEY_TYPE_alloc             },
 337         {BKEY_TYPE_quotas,      128, KEY_TYPE_quota             },
 338 };
 339
 340 void bch2_bkey_renumber(enum btree_node_type btree_node_type,
 341                         struct bkey_packed *k,
 342                         int write)
 343 {
 344         const struct old_bkey_type *i;
 345
 346         for (i = bkey_renumber_table;
 347              i < bkey_renumber_table + ARRAY_SIZE(bkey_renumber_table);
 348              i++)
 349                 if (btree_node_type == i->btree_node_type &&
 350                     k->type == (write ? i->new : i->old)) {
 351                         k->type = write ? i->old : i->new;
 352                         break;
 353                 }
 354 }
 355
 356 void __bch2_bkey_compat(unsigned level, enum btree_id btree_id,
 357                         unsigned version, unsigned big_endian,
 358                         int write,
 359                         struct bkey_format *f,
 360                         struct bkey_packed *k)
 361 {
 362         const struct bkey_ops *ops;
 363         struct bkey uk;
 364         struct bkey_s u;
 365         unsigned nr_compat = 5;
 366         int i;
 367
 368         /*
 369          * Do these operations in reverse order in the write path:
 370          */
 371
 372         for (i = 0; i < nr_compat; i++)
 373         switch (!write ? i : nr_compat - 1 - i) {
 374         case 0:
 375                 if (big_endian != CPU_BIG_ENDIAN)
 376                         bch2_bkey_swab_key(f, k);
 377                 break;
 378         case 1:
 379                 if (version < bcachefs_metadata_version_bkey_renumber)
 380                         bch2_bkey_renumber(__btree_node_type(level, btree_id), k, write);
 381                 break;
 382         case 2:
 383                 if (version < bcachefs_metadata_version_inode_btree_change &&
 384                     btree_id == BTREE_ID_inodes) {
 385                         if (!bkey_packed(k)) {
 386                                 struct bkey_i *u = packed_to_bkey(k);
 387                                 swap(u->k.p.inode, u->k.p.offset);
 388                         } else if (f->bits_per_field[BKEY_FIELD_INODE] &&
 389                                    f->bits_per_field[BKEY_FIELD_OFFSET]) {
 390                                 struct bkey_format tmp = *f, *in = f, *out = &tmp;
 391
 392                                 swap(tmp.bits_per_field[BKEY_FIELD_INODE],
 393                                      tmp.bits_per_field[BKEY_FIELD_OFFSET]);
 394                                 swap(tmp.field_offset[BKEY_FIELD_INODE],
 395                                      tmp.field_offset[BKEY_FIELD_OFFSET]);
 396
 397                                 if (!write)
 398                                         swap(in, out);
 399
 400                                 uk = __bch2_bkey_unpack_key(in, k);
 401                                 swap(uk.p.inode, uk.p.offset);
 402                                 BUG_ON(!bch2_bkey_pack_key(k, &uk, out));
 403                         }
 404                 }
 405                 break;
 406         case 3:
 407                 if (version < bcachefs_metadata_version_snapshot &&
 408                     (level || btree_type_has_snapshots(btree_id))) {
 409                         struct bkey_i *u = packed_to_bkey(k);
 410
 411                         if (u) {
 412                                 u->k.p.snapshot = write
 413                                         ? 0 : U32_MAX;
 414                         } else {
 415                                 u64 min_packed = f->field_offset[BKEY_FIELD_SNAPSHOT];
 416                                 u64 max_packed = min_packed +
 417                                         ~(~0ULL << f->bits_per_field[BKEY_FIELD_SNAPSHOT]);
 418
 419                                 uk = __bch2_bkey_unpack_key(f, k);
 420                                 uk.p.snapshot = write
 421                                         ? min_packed : min_t(u64, U32_MAX, max_packed);
 422
 423                                 BUG_ON(!bch2_bkey_pack_key(k, &uk, f));
 424                         }
 425                 }
 426
 427                 break;
 428         case 4:
 429                 if (!bkey_packed(k)) {
 430                         u = bkey_i_to_s(packed_to_bkey(k));
 431                 } else {
 432                         uk = __bch2_bkey_unpack_key(f, k);
 433                         u.k = &uk;
 434                         u.v = bkeyp_val(f, k);
 435                 }
 436
 437                 if (big_endian != CPU_BIG_ENDIAN)
 438                         bch2_bkey_swab_val(u);
 439
 440                 ops = &bch2_bkey_ops[k->type];
 441
 442                 if (ops->compat)
 443                         ops->compat(btree_id, version, big_endian, write, u);
 444                 break;
 445         default:
 446                 BUG();
 447         }
 448 }