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 "xattr.h"
  15
  16 const char * const bch2_bkey_types[] = {
  17 #define x(name, nr) #name,
  18         BCH_BKEY_TYPES()
  19 #undef x
  20         NULL
  21 };
  22
  23 static const char *deleted_key_invalid(const struct bch_fs *c,
  24                                         struct bkey_s_c k)
  25 {
  26         return NULL;
  27 }
  28
  29 #define bch2_bkey_ops_deleted (struct bkey_ops) {       \
  30         .key_invalid = deleted_key_invalid,             \
  31 }
  32
  33 #define bch2_bkey_ops_discard (struct bkey_ops) {       \
  34         .key_invalid = deleted_key_invalid,             \
  35 }
  36
  37 static const char *empty_val_key_invalid(const struct bch_fs *c, struct bkey_s_c k)
  38 {
  39         if (bkey_val_bytes(k.k))
  40                 return "value size should be zero";
  41
  42         return NULL;
  43 }
  44
  45 #define bch2_bkey_ops_error (struct bkey_ops) {         \
  46         .key_invalid = empty_val_key_invalid,           \
  47 }
  48
  49 static const char *key_type_cookie_invalid(const struct bch_fs *c,
  50                                            struct bkey_s_c k)
  51 {
  52         if (bkey_val_bytes(k.k) != sizeof(struct bch_cookie))
  53                 return "incorrect value size";
  54
  55         return NULL;
  56 }
  57
  58 #define bch2_bkey_ops_cookie (struct bkey_ops) {        \
  59         .key_invalid = key_type_cookie_invalid,         \
  60 }
  61
  62 #define bch2_bkey_ops_hash_whiteout (struct bkey_ops) { \
  63         .key_invalid = empty_val_key_invalid,           \
  64 }
  65
  66 static const char *key_type_inline_data_invalid(const struct bch_fs *c,
  67                                            struct bkey_s_c k)
  68 {
  69         return NULL;
  70 }
  71
  72 static void key_type_inline_data_to_text(struct printbuf *out, struct bch_fs *c,
  73                                          struct bkey_s_c k)
  74 {
  75         struct bkey_s_c_inline_data d = bkey_s_c_to_inline_data(k);
  76         unsigned datalen = bkey_inline_data_bytes(k.k);
  77
  78         pr_buf(out, "datalen %u: %*phN",
  79                datalen, min(datalen, 32U), d.v->data);
  80 }
  81
  82 #define bch2_bkey_ops_inline_data (struct bkey_ops) {   \
  83         .key_invalid    = key_type_inline_data_invalid, \
  84         .val_to_text    = key_type_inline_data_to_text, \
  85 }
  86
  87 const struct bkey_ops bch2_bkey_ops[] = {
  88 #define x(name, nr) [KEY_TYPE_##name]   = bch2_bkey_ops_##name,
  89         BCH_BKEY_TYPES()
  90 #undef x
  91 };
  92
  93 const char *bch2_bkey_val_invalid(struct bch_fs *c, struct bkey_s_c k)
  94 {
  95         if (k.k->type >= KEY_TYPE_MAX)
  96                 return "invalid type";
  97
  98         return bch2_bkey_ops[k.k->type].key_invalid(c, k);
  99 }
 100
 101 static unsigned bch2_key_types_allowed[] = {
 102         [BKEY_TYPE_extents] =
 103                 (1U << KEY_TYPE_error)|
 104                 (1U << KEY_TYPE_cookie)|
 105                 (1U << KEY_TYPE_extent)|
 106                 (1U << KEY_TYPE_reservation)|
 107                 (1U << KEY_TYPE_reflink_p)|
 108                 (1U << KEY_TYPE_inline_data),
 109         [BKEY_TYPE_inodes] =
 110                 (1U << KEY_TYPE_inode)|
 111                 (1U << KEY_TYPE_inode_generation),
 112         [BKEY_TYPE_dirents] =
 113                 (1U << KEY_TYPE_hash_whiteout)|
 114                 (1U << KEY_TYPE_dirent),
 115         [BKEY_TYPE_xattrs] =
 116                 (1U << KEY_TYPE_cookie)|
 117                 (1U << KEY_TYPE_hash_whiteout)|
 118                 (1U << KEY_TYPE_xattr),
 119         [BKEY_TYPE_alloc] =
 120                 (1U << KEY_TYPE_alloc)|
 121                 (1U << KEY_TYPE_alloc_v2),
 122         [BKEY_TYPE_quotas] =
 123                 (1U << KEY_TYPE_quota),
 124         [BKEY_TYPE_stripes] =
 125                 (1U << KEY_TYPE_stripe),
 126         [BKEY_TYPE_reflink] =
 127                 (1U << KEY_TYPE_reflink_v)|
 128                 (1U << KEY_TYPE_indirect_inline_data),
 129         [BKEY_TYPE_btree] =
 130                 (1U << KEY_TYPE_btree_ptr)|
 131                 (1U << KEY_TYPE_btree_ptr_v2),
 132 };
 133
 134 const char *__bch2_bkey_invalid(struct bch_fs *c, struct bkey_s_c k,
 135                                 enum btree_node_type type)
 136 {
 137         unsigned key_types_allowed = (1U << KEY_TYPE_deleted)|
 138                 bch2_key_types_allowed[type] ;
 139
 140         if (k.k->u64s < BKEY_U64s)
 141                 return "u64s too small";
 142
 143         if (!(key_types_allowed & (1U << k.k->type)))
 144                 return "invalid key type for this btree";
 145
 146         if (type == BKEY_TYPE_btree &&
 147             bkey_val_u64s(k.k) > BKEY_BTREE_PTR_VAL_U64s_MAX)
 148                 return "value too big";
 149
 150         if (btree_node_type_is_extents(type)) {
 151                 if ((k.k->size == 0) != bkey_deleted(k.k))
 152                         return "bad size field";
 153
 154                 if (k.k->size > k.k->p.offset)
 155                         return "size greater than offset";
 156         } else {
 157                 if (k.k->size)
 158                         return "nonzero size field";
 159         }
 160
 161         if (type != BKEY_TYPE_btree &&
 162             !btree_type_has_snapshots(type) &&
 163             k.k->p.snapshot)
 164                 return "nonzero snapshot";
 165
 166         if (type != BKEY_TYPE_btree &&
 167             btree_type_has_snapshots(type) &&
 168             k.k->p.snapshot != U32_MAX)
 169                 return "invalid snapshot field";
 170
 171         if (type != BKEY_TYPE_btree &&
 172             !bkey_cmp(k.k->p, POS_MAX))
 173                 return "POS_MAX key";
 174
 175         return NULL;
 176 }
 177
 178 const char *bch2_bkey_invalid(struct bch_fs *c, struct bkey_s_c k,
 179                               enum btree_node_type type)
 180 {
 181         return __bch2_bkey_invalid(c, k, type) ?:
 182                 bch2_bkey_val_invalid(c, k);
 183 }
 184
 185 const char *bch2_bkey_in_btree_node(struct btree *b, struct bkey_s_c k)
 186 {
 187         if (bpos_cmp(k.k->p, b->data->min_key) < 0)
 188                 return "key before start of btree node";
 189
 190         if (bpos_cmp(k.k->p, b->data->max_key) > 0)
 191                 return "key past end of btree node";
 192
 193         return NULL;
 194 }
 195
 196 void bch2_bkey_debugcheck(struct bch_fs *c, struct btree *b, struct bkey_s_c k)
 197 {
 198         const char *invalid;
 199
 200         BUG_ON(!k.k->u64s);
 201
 202         invalid = bch2_bkey_invalid(c, k, btree_node_type(b)) ?:
 203                 bch2_bkey_in_btree_node(b, k);
 204         if (invalid) {
 205                 char buf[160];
 206
 207                 bch2_bkey_val_to_text(&PBUF(buf), c, k);
 208                 bch2_fs_inconsistent(c, "invalid bkey %s: %s", buf, invalid);
 209         }
 210 }
 211
 212 void bch2_bpos_to_text(struct printbuf *out, struct bpos pos)
 213 {
 214         if (!bpos_cmp(pos, POS_MIN))
 215                 pr_buf(out, "POS_MIN");
 216         else if (!bpos_cmp(pos, POS_MAX))
 217                 pr_buf(out, "POS_MAX");
 218         else {
 219                 if (pos.inode == U64_MAX)
 220                         pr_buf(out, "U64_MAX");
 221                 else
 222                         pr_buf(out, "%llu", pos.inode);
 223                 pr_buf(out, ":");
 224                 if (pos.offset == U64_MAX)
 225                         pr_buf(out, "U64_MAX");
 226                 else
 227                         pr_buf(out, "%llu", pos.offset);
 228                 pr_buf(out, ":");
 229                 if (pos.snapshot == U32_MAX)
 230                         pr_buf(out, "U32_MAX");
 231                 else
 232                         pr_buf(out, "%u", pos.snapshot);
 233         }
 234 }
 235
 236 void bch2_bkey_to_text(struct printbuf *out, const struct bkey *k)
 237 {
 238         if (k) {
 239                 pr_buf(out, "u64s %u type ", k->u64s);
 240
 241                 if (k->type < KEY_TYPE_MAX)
 242                         pr_buf(out, "%s ", bch2_bkey_types[k->type]);
 243                 else
 244                         pr_buf(out, "%u ", k->type);
 245
 246                 bch2_bpos_to_text(out, k->p);
 247
 248                 pr_buf(out, " len %u ver %llu", k->size, k->version.lo);
 249         } else {
 250                 pr_buf(out, "(null)");
 251         }
 252 }
 253
 254 void bch2_val_to_text(struct printbuf *out, struct bch_fs *c,
 255                       struct bkey_s_c k)
 256 {
 257         if (k.k->type < KEY_TYPE_MAX) {
 258                 const struct bkey_ops *ops = &bch2_bkey_ops[k.k->type];
 259
 260                 if (likely(ops->val_to_text))
 261                         ops->val_to_text(out, c, k);
 262         } else {
 263                 pr_buf(out, "(invalid type %u)", k.k->type);
 264         }
 265 }
 266
 267 void bch2_bkey_val_to_text(struct printbuf *out, struct bch_fs *c,
 268                            struct bkey_s_c k)
 269 {
 270         bch2_bkey_to_text(out, k.k);
 271
 272         if (k.k) {
 273                 pr_buf(out, ": ");
 274                 bch2_val_to_text(out, c, k);
 275         }
 276 }
 277
 278 void bch2_bkey_swab_val(struct bkey_s k)
 279 {
 280         const struct bkey_ops *ops = &bch2_bkey_ops[k.k->type];
 281
 282         if (ops->swab)
 283                 ops->swab(k);
 284 }
 285
 286 bool bch2_bkey_normalize(struct bch_fs *c, struct bkey_s k)
 287 {
 288         const struct bkey_ops *ops = &bch2_bkey_ops[k.k->type];
 289
 290         return ops->key_normalize
 291                 ? ops->key_normalize(c, k)
 292                 : false;
 293 }
 294
 295 bool bch2_bkey_merge(struct bch_fs *c, struct bkey_s l, struct bkey_s_c r)
 296 {
 297         const struct bkey_ops *ops = &bch2_bkey_ops[l.k->type];
 298
 299         return bch2_bkey_maybe_mergable(l.k, r.k) && ops->key_merge(c, l, r);
 300 }
 301
 302 static const struct old_bkey_type {
 303         u8              btree_node_type;
 304         u8              old;
 305         u8              new;
 306 } bkey_renumber_table[] = {
 307         {BKEY_TYPE_btree,       128, KEY_TYPE_btree_ptr         },
 308         {BKEY_TYPE_extents,     128, KEY_TYPE_extent            },
 309         {BKEY_TYPE_extents,     129, KEY_TYPE_extent            },
 310         {BKEY_TYPE_extents,     130, KEY_TYPE_reservation       },
 311         {BKEY_TYPE_inodes,      128, KEY_TYPE_inode             },
 312         {BKEY_TYPE_inodes,      130, KEY_TYPE_inode_generation  },
 313         {BKEY_TYPE_dirents,     128, KEY_TYPE_dirent            },
 314         {BKEY_TYPE_dirents,     129, KEY_TYPE_hash_whiteout     },
 315         {BKEY_TYPE_xattrs,      128, KEY_TYPE_xattr             },
 316         {BKEY_TYPE_xattrs,      129, KEY_TYPE_hash_whiteout     },
 317         {BKEY_TYPE_alloc,       128, KEY_TYPE_alloc             },
 318         {BKEY_TYPE_quotas,      128, KEY_TYPE_quota             },
 319 };
 320
 321 void bch2_bkey_renumber(enum btree_node_type btree_node_type,
 322                         struct bkey_packed *k,
 323                         int write)
 324 {
 325         const struct old_bkey_type *i;
 326
 327         for (i = bkey_renumber_table;
 328              i < bkey_renumber_table + ARRAY_SIZE(bkey_renumber_table);
 329              i++)
 330                 if (btree_node_type == i->btree_node_type &&
 331                     k->type == (write ? i->new : i->old)) {
 332                         k->type = write ? i->old : i->new;
 333                         break;
 334                 }
 335 }
 336
 337 void __bch2_bkey_compat(unsigned level, enum btree_id btree_id,
 338                         unsigned version, unsigned big_endian,
 339                         int write,
 340                         struct bkey_format *f,
 341                         struct bkey_packed *k)
 342 {
 343         const struct bkey_ops *ops;
 344         struct bkey uk;
 345         struct bkey_s u;
 346         unsigned nr_compat = 5;
 347         int i;
 348
 349         /*
 350          * Do these operations in reverse order in the write path:
 351          */
 352
 353         for (i = 0; i < nr_compat; i++)
 354         switch (!write ? i : nr_compat - 1 - i) {
 355         case 0:
 356                 if (big_endian != CPU_BIG_ENDIAN)
 357                         bch2_bkey_swab_key(f, k);
 358                 break;
 359         case 1:
 360                 if (version < bcachefs_metadata_version_bkey_renumber)
 361                         bch2_bkey_renumber(__btree_node_type(level, btree_id), k, write);
 362                 break;
 363         case 2:
 364                 if (version < bcachefs_metadata_version_inode_btree_change &&
 365                     btree_id == BTREE_ID_inodes) {
 366                         if (!bkey_packed(k)) {
 367                                 struct bkey_i *u = packed_to_bkey(k);
 368                                 swap(u->k.p.inode, u->k.p.offset);
 369                         } else if (f->bits_per_field[BKEY_FIELD_INODE] &&
 370                                    f->bits_per_field[BKEY_FIELD_OFFSET]) {
 371                                 struct bkey_format tmp = *f, *in = f, *out = &tmp;
 372
 373                                 swap(tmp.bits_per_field[BKEY_FIELD_INODE],
 374                                      tmp.bits_per_field[BKEY_FIELD_OFFSET]);
 375                                 swap(tmp.field_offset[BKEY_FIELD_INODE],
 376                                      tmp.field_offset[BKEY_FIELD_OFFSET]);
 377
 378                                 if (!write)
 379                                         swap(in, out);
 380
 381                                 uk = __bch2_bkey_unpack_key(in, k);
 382                                 swap(uk.p.inode, uk.p.offset);
 383                                 BUG_ON(!bch2_bkey_pack_key(k, &uk, out));
 384                         }
 385                 }
 386                 break;
 387         case 3:
 388                 if (version < bcachefs_metadata_version_snapshot &&
 389                     (level || btree_type_has_snapshots(btree_id))) {
 390                         struct bkey_i *u = packed_to_bkey(k);
 391
 392                         if (u) {
 393                                 u->k.p.snapshot = write
 394                                         ? 0 : U32_MAX;
 395                         } else {
 396                                 u64 min_packed = f->field_offset[BKEY_FIELD_SNAPSHOT];
 397                                 u64 max_packed = min_packed +
 398                                         ~(~0ULL << f->bits_per_field[BKEY_FIELD_SNAPSHOT]);
 399
 400                                 uk = __bch2_bkey_unpack_key(f, k);
 401                                 uk.p.snapshot = write
 402                                         ? min_packed : min_t(u64, U32_MAX, max_packed);
 403
 404                                 BUG_ON(!bch2_bkey_pack_key(k, &uk, f));
 405                         }
 406                 }
 407
 408                 break;
 409         case 4:
 410                 if (!bkey_packed(k)) {
 411                         u = bkey_i_to_s(packed_to_bkey(k));
 412                 } else {
 413                         uk = __bch2_bkey_unpack_key(f, k);
 414                         u.k = &uk;
 415                         u.v = bkeyp_val(f, k);
 416                 }
 417
 418                 if (big_endian != CPU_BIG_ENDIAN)
 419                         bch2_bkey_swab_val(u);
 420
 421                 ops = &bch2_bkey_ops[k->type];
 422
 423                 if (ops->compat)
 424                         ops->compat(btree_id, version, big_endian, write, u);
 425                 break;
 426         default:
 427                 BUG();
 428         }
 429 }