1 /*------------------------------------------------------------------------- 2 * 3 * nbtree.h 4 * header file for postgres btree access method implementation. 5 * 6 * 7 * Portions Copyright (c) 1996-2019, PostgreSQL Global Development Group 8 * Portions Copyright (c) 1994, Regents of the University of California 9 * 10 * src/include/access/nbtree.h 11 * 12 *------------------------------------------------------------------------- 13 */ 14 #ifndef NBTREE_H 15 #define NBTREE_H 16 17 #include "access/amapi.h" 18 #include "access/itup.h" 19 #include "access/sdir.h" 20 #include "access/xlogreader.h" 21 #include "catalog/pg_index.h" 22 #include "lib/stringinfo.h" 23 #include "storage/bufmgr.h" 24 #include "storage/shm_toc.h" 25 26 /* There's room for a 16-bit vacuum cycle ID in BTPageOpaqueData */ 27 typedef uint16 BTCycleId; 28 29 /* 30 * BTPageOpaqueData -- At the end of every page, we store a pointer 31 * to both siblings in the tree. This is used to do forward/backward 32 * index scans. The next-page link is also critical for recovery when 33 * a search has navigated to the wrong page due to concurrent page splits 34 * or deletions; see src/backend/access/nbtree/README for more info. 35 * 36 * In addition, we store the page's btree level (counting upwards from 37 * zero at a leaf page) as well as some flag bits indicating the page type 38 * and status. If the page is deleted, we replace the level with the 39 * next-transaction-ID value indicating when it is safe to reclaim the page. 40 * 41 * We also store a "vacuum cycle ID". When a page is split while VACUUM is 42 * processing the index, a nonzero value associated with the VACUUM run is 43 * stored into both halves of the split page. (If VACUUM is not running, 44 * both pages receive zero cycleids.) This allows VACUUM to detect whether 45 * a page was split since it started, with a small probability of false match 46 * if the page was last split some exact multiple of MAX_BT_CYCLE_ID VACUUMs 47 * ago. Also, during a split, the BTP_SPLIT_END flag is cleared in the left 48 * (original) page, and set in the right page, but only if the next page 49 * to its right has a different cycleid. 50 * 51 * NOTE: the BTP_LEAF flag bit is redundant since level==0 could be tested 52 * instead. 53 */ 54 55 typedef struct BTPageOpaqueData 56 { 57 BlockNumber btpo_prev; /* left sibling, or P_NONE if leftmost */ 58 BlockNumber btpo_next; /* right sibling, or P_NONE if rightmost */ 59 union 60 { 61 uint32 level; /* tree level --- zero for leaf pages */ 62 TransactionId xact; /* next transaction ID, if deleted */ 63 } btpo; 64 uint16 btpo_flags; /* flag bits, see below */ 65 BTCycleId btpo_cycleid; /* vacuum cycle ID of latest split */ 66 } BTPageOpaqueData; 67 68 typedef BTPageOpaqueData *BTPageOpaque; 69 70 /* Bits defined in btpo_flags */ 71 #define BTP_LEAF (1 << 0) /* leaf page, i.e. not internal page */ 72 #define BTP_ROOT (1 << 1) /* root page (has no parent) */ 73 #define BTP_DELETED (1 << 2) /* page has been deleted from tree */ 74 #define BTP_META (1 << 3) /* meta-page */ 75 #define BTP_HALF_DEAD (1 << 4) /* empty, but still in tree */ 76 #define BTP_SPLIT_END (1 << 5) /* rightmost page of split group */ 77 #define BTP_HAS_GARBAGE (1 << 6) /* page has LP_DEAD tuples */ 78 #define BTP_INCOMPLETE_SPLIT (1 << 7) /* right sibling's downlink is missing */ 79 80 /* 81 * The max allowed value of a cycle ID is a bit less than 64K. This is 82 * for convenience of pg_filedump and similar utilities: we want to use 83 * the last 2 bytes of special space as an index type indicator, and 84 * restricting cycle ID lets btree use that space for vacuum cycle IDs 85 * while still allowing index type to be identified. 86 */ 87 #define MAX_BT_CYCLE_ID 0xFF7F 88 89 90 /* 91 * The Meta page is always the first page in the btree index. 92 * Its primary purpose is to point to the location of the btree root page. 93 * We also point to the "fast" root, which is the current effective root; 94 * see README for discussion. 95 */ 96 97 typedef struct BTMetaPageData 98 { 99 uint32 btm_magic; /* should contain BTREE_MAGIC */ 100 uint32 btm_version; /* nbtree version (always <= BTREE_VERSION) */ 101 BlockNumber btm_root; /* current root location */ 102 uint32 btm_level; /* tree level of the root page */ 103 BlockNumber btm_fastroot; /* current "fast" root location */ 104 uint32 btm_fastlevel; /* tree level of the "fast" root page */ 105 /* remaining fields only valid when btm_version >= BTREE_NOVAC_VERSION */ 106 TransactionId btm_oldest_btpo_xact; /* oldest btpo_xact among all deleted 107 * pages */ 108 float8 btm_last_cleanup_num_heap_tuples; /* number of heap tuples 109 * during last cleanup */ 110 } BTMetaPageData; 111 112 #define BTPageGetMeta(p) \ 113 ((BTMetaPageData *) PageGetContents(p)) 114 115 /* 116 * The current Btree version is 4. That's what you'll get when you create 117 * a new index. 118 * 119 * Btree version 3 was used in PostgreSQL v11. It is mostly the same as 120 * version 4, but heap TIDs were not part of the keyspace. Index tuples 121 * with duplicate keys could be stored in any order. We continue to 122 * support reading and writing Btree versions 2 and 3, so that they don't 123 * need to be immediately re-indexed at pg_upgrade. In order to get the 124 * new heapkeyspace semantics, however, a REINDEX is needed. 125 * 126 * Btree version 2 is mostly the same as version 3. There are two new 127 * fields in the metapage that were introduced in version 3. A version 2 128 * metapage will be automatically upgraded to version 3 on the first 129 * insert to it. INCLUDE indexes cannot use version 2. 130 */ 131 #define BTREE_METAPAGE 0 /* first page is meta */ 132 #define BTREE_MAGIC 0x053162 /* magic number in metapage */ 133 #define BTREE_VERSION 4 /* current version number */ 134 #define BTREE_MIN_VERSION 2 /* minimal supported version number */ 135 #define BTREE_NOVAC_VERSION 3 /* minimal version with all meta fields */ 136 137 /* 138 * Maximum size of a btree index entry, including its tuple header. 139 * 140 * We actually need to be able to fit three items on every page, 141 * so restrict any one item to 1/3 the per-page available space. 142 * 143 * There are rare cases where _bt_truncate() will need to enlarge 144 * a heap index tuple to make space for a tiebreaker heap TID 145 * attribute, which we account for here. 146 */ 147 #define BTMaxItemSize(page) \ 148 MAXALIGN_DOWN((PageGetPageSize(page) - \ 149 MAXALIGN(SizeOfPageHeaderData + \ 150 3*sizeof(ItemIdData) + \ 151 3*sizeof(ItemPointerData)) - \ 152 MAXALIGN(sizeof(BTPageOpaqueData))) / 3) 153 #define BTMaxItemSizeNoHeapTid(page) \ 154 MAXALIGN_DOWN((PageGetPageSize(page) - \ 155 MAXALIGN(SizeOfPageHeaderData + 3*sizeof(ItemIdData)) - \ 156 MAXALIGN(sizeof(BTPageOpaqueData))) / 3) 157 158 /* 159 * The leaf-page fillfactor defaults to 90% but is user-adjustable. 160 * For pages above the leaf level, we use a fixed 70% fillfactor. 161 * The fillfactor is applied during index build and when splitting 162 * a rightmost page; when splitting non-rightmost pages we try to 163 * divide the data equally. When splitting a page that's entirely 164 * filled with a single value (duplicates), the effective leaf-page 165 * fillfactor is 96%, regardless of whether the page is a rightmost 166 * page. 167 */ 168 #define BTREE_MIN_FILLFACTOR 10 169 #define BTREE_DEFAULT_FILLFACTOR 90 170 #define BTREE_NONLEAF_FILLFACTOR 70 171 #define BTREE_SINGLEVAL_FILLFACTOR 96 172 173 /* 174 * In general, the btree code tries to localize its knowledge about 175 * page layout to a couple of routines. However, we need a special 176 * value to indicate "no page number" in those places where we expect 177 * page numbers. We can use zero for this because we never need to 178 * make a pointer to the metadata page. 179 */ 180 181 #define P_NONE 0 182 183 /* 184 * Macros to test whether a page is leftmost or rightmost on its tree level, 185 * as well as other state info kept in the opaque data. 186 */ 187 #define P_LEFTMOST(opaque) ((opaque)->btpo_prev == P_NONE) 188 #define P_RIGHTMOST(opaque) ((opaque)->btpo_next == P_NONE) 189 #define P_ISLEAF(opaque) (((opaque)->btpo_flags & BTP_LEAF) != 0) 190 #define P_ISROOT(opaque) (((opaque)->btpo_flags & BTP_ROOT) != 0) 191 #define P_ISDELETED(opaque) (((opaque)->btpo_flags & BTP_DELETED) != 0) 192 #define P_ISMETA(opaque) (((opaque)->btpo_flags & BTP_META) != 0) 193 #define P_ISHALFDEAD(opaque) (((opaque)->btpo_flags & BTP_HALF_DEAD) != 0) 194 #define P_IGNORE(opaque) (((opaque)->btpo_flags & (BTP_DELETED|BTP_HALF_DEAD)) != 0) 195 #define P_HAS_GARBAGE(opaque) (((opaque)->btpo_flags & BTP_HAS_GARBAGE) != 0) 196 #define P_INCOMPLETE_SPLIT(opaque) (((opaque)->btpo_flags & BTP_INCOMPLETE_SPLIT) != 0) 197 198 /* 199 * Lehman and Yao's algorithm requires a ``high key'' on every non-rightmost 200 * page. The high key is not a tuple that is used to visit the heap. It is 201 * a pivot tuple (see "Notes on B-Tree tuple format" below for definition). 202 * The high key on a page is required to be greater than or equal to any 203 * other key that appears on the page. If we find ourselves trying to 204 * insert a key that is strictly > high key, we know we need to move right 205 * (this should only happen if the page was split since we examined the 206 * parent page). 207 * 208 * Our insertion algorithm guarantees that we can use the initial least key 209 * on our right sibling as the high key. Once a page is created, its high 210 * key changes only if the page is split. 211 * 212 * On a non-rightmost page, the high key lives in item 1 and data items 213 * start in item 2. Rightmost pages have no high key, so we store data 214 * items beginning in item 1. 215 */ 216 217 #define P_HIKEY ((OffsetNumber) 1) 218 #define P_FIRSTKEY ((OffsetNumber) 2) 219 #define P_FIRSTDATAKEY(opaque) (P_RIGHTMOST(opaque) ? P_HIKEY : P_FIRSTKEY) 220 221 /* 222 * Notes on B-Tree tuple format, and key and non-key attributes: 223 * 224 * INCLUDE B-Tree indexes have non-key attributes. These are extra 225 * attributes that may be returned by index-only scans, but do not influence 226 * the order of items in the index (formally, non-key attributes are not 227 * considered to be part of the key space). Non-key attributes are only 228 * present in leaf index tuples whose item pointers actually point to heap 229 * tuples (non-pivot tuples). _bt_check_natts() enforces the rules 230 * described here. 231 * 232 * Non-pivot tuple format: 233 * 234 * t_tid | t_info | key values | INCLUDE columns, if any 235 * 236 * t_tid points to the heap TID, which is a tiebreaker key column as of 237 * BTREE_VERSION 4. Currently, the INDEX_ALT_TID_MASK status bit is never 238 * set for non-pivot tuples. 239 * 240 * All other types of index tuples ("pivot" tuples) only have key columns, 241 * since pivot tuples only exist to represent how the key space is 242 * separated. In general, any B-Tree index that has more than one level 243 * (i.e. any index that does not just consist of a metapage and a single 244 * leaf root page) must have some number of pivot tuples, since pivot 245 * tuples are used for traversing the tree. Suffix truncation can omit 246 * trailing key columns when a new pivot is formed, which makes minus 247 * infinity their logical value. Since BTREE_VERSION 4 indexes treat heap 248 * TID as a trailing key column that ensures that all index tuples are 249 * physically unique, it is necessary to represent heap TID as a trailing 250 * key column in pivot tuples, though very often this can be truncated 251 * away, just like any other key column. (Actually, the heap TID is 252 * omitted rather than truncated, since its representation is different to 253 * the non-pivot representation.) 254 * 255 * Pivot tuple format: 256 * 257 * t_tid | t_info | key values | [heap TID] 258 * 259 * We store the number of columns present inside pivot tuples by abusing 260 * their t_tid offset field, since pivot tuples never need to store a real 261 * offset (downlinks only need to store a block number in t_tid). The 262 * offset field only stores the number of columns/attributes when the 263 * INDEX_ALT_TID_MASK bit is set, which doesn't count the trailing heap 264 * TID column sometimes stored in pivot tuples -- that's represented by 265 * the presence of BT_HEAP_TID_ATTR. The INDEX_ALT_TID_MASK bit in t_info 266 * is always set on BTREE_VERSION 4. BT_HEAP_TID_ATTR can only be set on 267 * BTREE_VERSION 4. 268 * 269 * In version 3 indexes, the INDEX_ALT_TID_MASK flag might not be set in 270 * pivot tuples. In that case, the number of key columns is implicitly 271 * the same as the number of key columns in the index. It is not usually 272 * set on version 2 indexes, which predate the introduction of INCLUDE 273 * indexes. (Only explicitly truncated pivot tuples explicitly represent 274 * the number of key columns on versions 2 and 3, whereas all pivot tuples 275 * are formed using truncation on version 4. A version 2 index will have 276 * it set for an internal page negative infinity item iff internal page 277 * split occurred after upgrade to Postgres 11+.) 278 * 279 * The 12 least significant offset bits from t_tid are used to represent 280 * the number of columns in INDEX_ALT_TID_MASK tuples, leaving 4 status 281 * bits (BT_RESERVED_OFFSET_MASK bits), 3 of which that are reserved for 282 * future use. BT_N_KEYS_OFFSET_MASK should be large enough to store any 283 * number of columns/attributes <= INDEX_MAX_KEYS. 284 * 285 * Note well: The macros that deal with the number of attributes in tuples 286 * assume that a tuple with INDEX_ALT_TID_MASK set must be a pivot tuple, 287 * and that a tuple without INDEX_ALT_TID_MASK set must be a non-pivot 288 * tuple (or must have the same number of attributes as the index has 289 * generally in the case of !heapkeyspace indexes). They will need to be 290 * updated if non-pivot tuples ever get taught to use INDEX_ALT_TID_MASK 291 * for something else. 292 */ 293 #define INDEX_ALT_TID_MASK INDEX_AM_RESERVED_BIT 294 295 /* Item pointer offset bits */ 296 #define BT_RESERVED_OFFSET_MASK 0xF000 297 #define BT_N_KEYS_OFFSET_MASK 0x0FFF 298 #define BT_HEAP_TID_ATTR 0x1000 299 300 /* Get/set downlink block number */ 301 #define BTreeInnerTupleGetDownLink(itup) \ 302 ItemPointerGetBlockNumberNoCheck(&((itup)->t_tid)) 303 #define BTreeInnerTupleSetDownLink(itup, blkno) \ 304 ItemPointerSetBlockNumber(&((itup)->t_tid), (blkno)) 305 306 /* 307 * Get/set leaf page highkey's link. During the second phase of deletion, the 308 * target leaf page's high key may point to an ancestor page (at all other 309 * times, the leaf level high key's link is not used). See the nbtree README 310 * for full details. 311 */ 312 #define BTreeTupleGetTopParent(itup) \ 313 ItemPointerGetBlockNumberNoCheck(&((itup)->t_tid)) 314 #define BTreeTupleSetTopParent(itup, blkno) \ 315 do { \ 316 ItemPointerSetBlockNumber(&((itup)->t_tid), (blkno)); \ 317 BTreeTupleSetNAtts((itup), 0); \ 318 } while(0) 319 320 /* 321 * Get/set number of attributes within B-tree index tuple. 322 * 323 * Note that this does not include an implicit tiebreaker heap TID 324 * attribute, if any. Note also that the number of key attributes must be 325 * explicitly represented in all heapkeyspace pivot tuples. 326 */ 327 #define BTreeTupleGetNAtts(itup, rel) \ 328 ( \ 329 (itup)->t_info & INDEX_ALT_TID_MASK ? \ 330 ( \ 331 ItemPointerGetOffsetNumberNoCheck(&(itup)->t_tid) & BT_N_KEYS_OFFSET_MASK \ 332 ) \ 333 : \ 334 IndexRelationGetNumberOfAttributes(rel) \ 335 ) 336 #define BTreeTupleSetNAtts(itup, n) \ 337 do { \ 338 (itup)->t_info |= INDEX_ALT_TID_MASK; \ 339 ItemPointerSetOffsetNumber(&(itup)->t_tid, (n) & BT_N_KEYS_OFFSET_MASK); \ 340 } while(0) 341 342 /* 343 * Get tiebreaker heap TID attribute, if any. Macro works with both pivot 344 * and non-pivot tuples, despite differences in how heap TID is represented. 345 */ 346 #define BTreeTupleGetHeapTID(itup) \ 347 ( \ 348 (itup)->t_info & INDEX_ALT_TID_MASK && \ 349 (ItemPointerGetOffsetNumberNoCheck(&(itup)->t_tid) & BT_HEAP_TID_ATTR) != 0 ? \ 350 ( \ 351 (ItemPointer) (((char *) (itup) + IndexTupleSize(itup)) - \ 352 sizeof(ItemPointerData)) \ 353 ) \ 354 : (itup)->t_info & INDEX_ALT_TID_MASK ? NULL : (ItemPointer) &((itup)->t_tid) \ 355 ) 356 /* 357 * Set the heap TID attribute for a tuple that uses the INDEX_ALT_TID_MASK 358 * representation (currently limited to pivot tuples) 359 */ 360 #define BTreeTupleSetAltHeapTID(itup) \ 361 do { \ 362 Assert((itup)->t_info & INDEX_ALT_TID_MASK); \ 363 ItemPointerSetOffsetNumber(&(itup)->t_tid, \ 364 ItemPointerGetOffsetNumberNoCheck(&(itup)->t_tid) | BT_HEAP_TID_ATTR); \ 365 } while(0) 366 367 /* 368 * Operator strategy numbers for B-tree have been moved to access/stratnum.h, 369 * because many places need to use them in ScanKeyInit() calls. 370 * 371 * The strategy numbers are chosen so that we can commute them by 372 * subtraction, thus: 373 */ 374 #define BTCommuteStrategyNumber(strat) (BTMaxStrategyNumber + 1 - (strat)) 375 376 /* 377 * When a new operator class is declared, we require that the user 378 * supply us with an amproc procedure (BTORDER_PROC) for determining 379 * whether, for two keys a and b, a < b, a = b, or a > b. This routine 380 * must return < 0, 0, > 0, respectively, in these three cases. 381 * 382 * To facilitate accelerated sorting, an operator class may choose to 383 * offer a second procedure (BTSORTSUPPORT_PROC). For full details, see 384 * src/include/utils/sortsupport.h. 385 * 386 * To support window frames defined by "RANGE offset PRECEDING/FOLLOWING", 387 * an operator class may choose to offer a third amproc procedure 388 * (BTINRANGE_PROC), independently of whether it offers sortsupport. 389 * For full details, see doc/src/sgml/btree.sgml. 390 */ 391 392 #define BTORDER_PROC 1 393 #define BTSORTSUPPORT_PROC 2 394 #define BTINRANGE_PROC 3 395 #define BTNProcs 3 396 397 /* 398 * We need to be able to tell the difference between read and write 399 * requests for pages, in order to do locking correctly. 400 */ 401 402 #define BT_READ BUFFER_LOCK_SHARE 403 #define BT_WRITE BUFFER_LOCK_EXCLUSIVE 404 405 /* 406 * BTStackData -- As we descend a tree, we push the (location, downlink) 407 * pairs from internal pages onto a private stack. If we split a 408 * leaf, we use this stack to walk back up the tree and insert data 409 * into parent pages (and possibly to split them, too). Lehman and 410 * Yao's update algorithm guarantees that under no circumstances can 411 * our private stack give us an irredeemably bad picture up the tree. 412 * Again, see the paper for details. 413 */ 414 415 typedef struct BTStackData 416 { 417 BlockNumber bts_blkno; 418 OffsetNumber bts_offset; 419 BlockNumber bts_btentry; 420 struct BTStackData *bts_parent; 421 } BTStackData; 422 423 typedef BTStackData *BTStack; 424 425 /* 426 * BTScanInsertData is the btree-private state needed to find an initial 427 * position for an indexscan, or to insert new tuples -- an "insertion 428 * scankey" (not to be confused with a search scankey). It's used to descend 429 * a B-Tree using _bt_search. 430 * 431 * heapkeyspace indicates if we expect all keys in the index to be physically 432 * unique because heap TID is used as a tiebreaker attribute, and if index may 433 * have truncated key attributes in pivot tuples. This is actually a property 434 * of the index relation itself (not an indexscan). heapkeyspace indexes are 435 * indexes whose version is >= version 4. It's convenient to keep this close 436 * by, rather than accessing the metapage repeatedly. 437 * 438 * anynullkeys indicates if any of the keys had NULL value when scankey was 439 * built from index tuple (note that already-truncated tuple key attributes 440 * set NULL as a placeholder key value, which also affects value of 441 * anynullkeys). This is a convenience for unique index non-pivot tuple 442 * insertion, which usually temporarily unsets scantid, but shouldn't iff 443 * anynullkeys is true. Value generally matches non-pivot tuple's HasNulls 444 * bit, but may not when inserting into an INCLUDE index (tuple header value 445 * is affected by the NULL-ness of both key and non-key attributes). 446 * 447 * When nextkey is false (the usual case), _bt_search and _bt_binsrch will 448 * locate the first item >= scankey. When nextkey is true, they will locate 449 * the first item > scan key. 450 * 451 * pivotsearch is set to true by callers that want to re-find a leaf page 452 * using a scankey built from a leaf page's high key. Most callers set this 453 * to false. 454 * 455 * scantid is the heap TID that is used as a final tiebreaker attribute. It 456 * is set to NULL when index scan doesn't need to find a position for a 457 * specific physical tuple. Must be set when inserting new tuples into 458 * heapkeyspace indexes, since every tuple in the tree unambiguously belongs 459 * in one exact position (it's never set with !heapkeyspace indexes, though). 460 * Despite the representational difference, nbtree search code considers 461 * scantid to be just another insertion scankey attribute. 462 * 463 * scankeys is an array of scan key entries for attributes that are compared 464 * before scantid (user-visible attributes). keysz is the size of the array. 465 * During insertion, there must be a scan key for every attribute, but when 466 * starting a regular index scan some can be omitted. The array is used as a 467 * flexible array member, though it's sized in a way that makes it possible to 468 * use stack allocations. See nbtree/README for full details. 469 */ 470 typedef struct BTScanInsertData 471 { 472 bool heapkeyspace; 473 bool anynullkeys; 474 bool nextkey; 475 bool pivotsearch; 476 ItemPointer scantid; /* tiebreaker for scankeys */ 477 int keysz; /* Size of scankeys array */ 478 ScanKeyData scankeys[INDEX_MAX_KEYS]; /* Must appear last */ 479 } BTScanInsertData; 480 481 typedef BTScanInsertData *BTScanInsert; 482 483 /* 484 * BTInsertStateData is a working area used during insertion. 485 * 486 * This is filled in after descending the tree to the first leaf page the new 487 * tuple might belong on. Tracks the current position while performing 488 * uniqueness check, before we have determined which exact page to insert 489 * to. 490 * 491 * (This should be private to nbtinsert.c, but it's also used by 492 * _bt_binsrch_insert) 493 */ 494 typedef struct BTInsertStateData 495 { 496 IndexTuple itup; /* Item we're inserting */ 497 Size itemsz; /* Size of itup -- should be MAXALIGN()'d */ 498 BTScanInsert itup_key; /* Insertion scankey */ 499 500 /* Buffer containing leaf page we're likely to insert itup on */ 501 Buffer buf; 502 503 /* 504 * Cache of bounds within the current buffer. Only used for insertions 505 * where _bt_check_unique is called. See _bt_binsrch_insert and 506 * _bt_findinsertloc for details. 507 */ 508 bool bounds_valid; 509 OffsetNumber low; 510 OffsetNumber stricthigh; 511 } BTInsertStateData; 512 513 typedef BTInsertStateData *BTInsertState; 514 515 /* 516 * BTScanOpaqueData is the btree-private state needed for an indexscan. 517 * This consists of preprocessed scan keys (see _bt_preprocess_keys() for 518 * details of the preprocessing), information about the current location 519 * of the scan, and information about the marked location, if any. (We use 520 * BTScanPosData to represent the data needed for each of current and marked 521 * locations.) In addition we can remember some known-killed index entries 522 * that must be marked before we can move off the current page. 523 * 524 * Index scans work a page at a time: we pin and read-lock the page, identify 525 * all the matching items on the page and save them in BTScanPosData, then 526 * release the read-lock while returning the items to the caller for 527 * processing. This approach minimizes lock/unlock traffic. Note that we 528 * keep the pin on the index page until the caller is done with all the items 529 * (this is needed for VACUUM synchronization, see nbtree/README). When we 530 * are ready to step to the next page, if the caller has told us any of the 531 * items were killed, we re-lock the page to mark them killed, then unlock. 532 * Finally we drop the pin and step to the next page in the appropriate 533 * direction. 534 * 535 * If we are doing an index-only scan, we save the entire IndexTuple for each 536 * matched item, otherwise only its heap TID and offset. The IndexTuples go 537 * into a separate workspace array; each BTScanPosItem stores its tuple's 538 * offset within that array. 539 */ 540 541 typedef struct BTScanPosItem /* what we remember about each match */ 542 { 543 ItemPointerData heapTid; /* TID of referenced heap item */ 544 OffsetNumber indexOffset; /* index item's location within page */ 545 LocationIndex tupleOffset; /* IndexTuple's offset in workspace, if any */ 546 } BTScanPosItem; 547 548 typedef struct BTScanPosData 549 { 550 Buffer buf; /* if valid, the buffer is pinned */ 551 552 XLogRecPtr lsn; /* pos in the WAL stream when page was read */ 553 BlockNumber currPage; /* page referenced by items array */ 554 BlockNumber nextPage; /* page's right link when we scanned it */ 555 556 /* 557 * moreLeft and moreRight track whether we think there may be matching 558 * index entries to the left and right of the current page, respectively. 559 * We can clear the appropriate one of these flags when _bt_checkkeys() 560 * returns continuescan = false. 561 */ 562 bool moreLeft; 563 bool moreRight; 564 565 /* 566 * If we are doing an index-only scan, nextTupleOffset is the first free 567 * location in the associated tuple storage workspace. 568 */ 569 int nextTupleOffset; 570 571 /* 572 * The items array is always ordered in index order (ie, increasing 573 * indexoffset). When scanning backwards it is convenient to fill the 574 * array back-to-front, so we start at the last slot and fill downwards. 575 * Hence we need both a first-valid-entry and a last-valid-entry counter. 576 * itemIndex is a cursor showing which entry was last returned to caller. 577 */ 578 int firstItem; /* first valid index in items[] */ 579 int lastItem; /* last valid index in items[] */ 580 int itemIndex; /* current index in items[] */ 581 582 BTScanPosItem items[MaxIndexTuplesPerPage]; /* MUST BE LAST */ 583 } BTScanPosData; 584 585 typedef BTScanPosData *BTScanPos; 586 587 #define BTScanPosIsPinned(scanpos) \ 588 ( \ 589 AssertMacro(BlockNumberIsValid((scanpos).currPage) || \ 590 !BufferIsValid((scanpos).buf)), \ 591 BufferIsValid((scanpos).buf) \ 592 ) 593 #define BTScanPosUnpin(scanpos) \ 594 do { \ 595 ReleaseBuffer((scanpos).buf); \ 596 (scanpos).buf = InvalidBuffer; \ 597 } while (0) 598 #define BTScanPosUnpinIfPinned(scanpos) \ 599 do { \ 600 if (BTScanPosIsPinned(scanpos)) \ 601 BTScanPosUnpin(scanpos); \ 602 } while (0) 603 604 #define BTScanPosIsValid(scanpos) \ 605 ( \ 606 AssertMacro(BlockNumberIsValid((scanpos).currPage) || \ 607 !BufferIsValid((scanpos).buf)), \ 608 BlockNumberIsValid((scanpos).currPage) \ 609 ) 610 #define BTScanPosInvalidate(scanpos) \ 611 do { \ 612 (scanpos).currPage = InvalidBlockNumber; \ 613 (scanpos).nextPage = InvalidBlockNumber; \ 614 (scanpos).buf = InvalidBuffer; \ 615 (scanpos).lsn = InvalidXLogRecPtr; \ 616 (scanpos).nextTupleOffset = 0; \ 617 } while (0) 618 619 /* We need one of these for each equality-type SK_SEARCHARRAY scan key */ 620 typedef struct BTArrayKeyInfo 621 { 622 int scan_key; /* index of associated key in arrayKeyData */ 623 int cur_elem; /* index of current element in elem_values */ 624 int mark_elem; /* index of marked element in elem_values */ 625 int num_elems; /* number of elems in current array value */ 626 Datum *elem_values; /* array of num_elems Datums */ 627 } BTArrayKeyInfo; 628 629 typedef struct BTScanOpaqueData 630 { 631 /* these fields are set by _bt_preprocess_keys(): */ 632 bool qual_ok; /* false if qual can never be satisfied */ 633 int numberOfKeys; /* number of preprocessed scan keys */ 634 ScanKey keyData; /* array of preprocessed scan keys */ 635 636 /* workspace for SK_SEARCHARRAY support */ 637 ScanKey arrayKeyData; /* modified copy of scan->keyData */ 638 int numArrayKeys; /* number of equality-type array keys (-1 if 639 * there are any unsatisfiable array keys) */ 640 int arrayKeyCount; /* count indicating number of array scan keys 641 * processed */ 642 BTArrayKeyInfo *arrayKeys; /* info about each equality-type array key */ 643 MemoryContext arrayContext; /* scan-lifespan context for array data */ 644 645 /* info about killed items if any (killedItems is NULL if never used) */ 646 int *killedItems; /* currPos.items indexes of killed items */ 647 int numKilled; /* number of currently stored items */ 648 649 /* 650 * If we are doing an index-only scan, these are the tuple storage 651 * workspaces for the currPos and markPos respectively. Each is of size 652 * BLCKSZ, so it can hold as much as a full page's worth of tuples. 653 */ 654 char *currTuples; /* tuple storage for currPos */ 655 char *markTuples; /* tuple storage for markPos */ 656 657 /* 658 * If the marked position is on the same page as current position, we 659 * don't use markPos, but just keep the marked itemIndex in markItemIndex 660 * (all the rest of currPos is valid for the mark position). Hence, to 661 * determine if there is a mark, first look at markItemIndex, then at 662 * markPos. 663 */ 664 int markItemIndex; /* itemIndex, or -1 if not valid */ 665 666 /* keep these last in struct for efficiency */ 667 BTScanPosData currPos; /* current position data */ 668 BTScanPosData markPos; /* marked position, if any */ 669 } BTScanOpaqueData; 670 671 typedef BTScanOpaqueData *BTScanOpaque; 672 673 /* 674 * We use some private sk_flags bits in preprocessed scan keys. We're allowed 675 * to use bits 16-31 (see skey.h). The uppermost bits are copied from the 676 * index's indoption[] array entry for the index attribute. 677 */ 678 #define SK_BT_REQFWD 0x00010000 /* required to continue forward scan */ 679 #define SK_BT_REQBKWD 0x00020000 /* required to continue backward scan */ 680 #define SK_BT_INDOPTION_SHIFT 24 /* must clear the above bits */ 681 #define SK_BT_DESC (INDOPTION_DESC << SK_BT_INDOPTION_SHIFT) 682 #define SK_BT_NULLS_FIRST (INDOPTION_NULLS_FIRST << SK_BT_INDOPTION_SHIFT) 683 684 /* 685 * Constant definition for progress reporting. Phase numbers must match 686 * btbuildphasename. 687 */ 688 /* PROGRESS_CREATEIDX_SUBPHASE_INITIALIZE is 1 (see progress.h) */ 689 #define PROGRESS_BTREE_PHASE_INDEXBUILD_TABLESCAN 2 690 #define PROGRESS_BTREE_PHASE_PERFORMSORT_1 3 691 #define PROGRESS_BTREE_PHASE_PERFORMSORT_2 4 692 #define PROGRESS_BTREE_PHASE_LEAF_LOAD 5 693 694 /* 695 * external entry points for btree, in nbtree.c 696 */ 697 extern void btbuildempty(Relation index); 698 extern bool btinsert(Relation rel, Datum *values, bool *isnull, 699 ItemPointer ht_ctid, Relation heapRel, 700 IndexUniqueCheck checkUnique, 701 struct IndexInfo *indexInfo); 702 extern IndexScanDesc btbeginscan(Relation rel, int nkeys, int norderbys); 703 extern Size btestimateparallelscan(void); 704 extern void btinitparallelscan(void *target); 705 extern bool btgettuple(IndexScanDesc scan, ScanDirection dir); 706 extern int64 btgetbitmap(IndexScanDesc scan, TIDBitmap *tbm); 707 extern void btrescan(IndexScanDesc scan, ScanKey scankey, int nscankeys, 708 ScanKey orderbys, int norderbys); 709 extern void btparallelrescan(IndexScanDesc scan); 710 extern void btendscan(IndexScanDesc scan); 711 extern void btmarkpos(IndexScanDesc scan); 712 extern void btrestrpos(IndexScanDesc scan); 713 extern IndexBulkDeleteResult *btbulkdelete(IndexVacuumInfo *info, 714 IndexBulkDeleteResult *stats, 715 IndexBulkDeleteCallback callback, 716 void *callback_state); 717 extern IndexBulkDeleteResult *btvacuumcleanup(IndexVacuumInfo *info, 718 IndexBulkDeleteResult *stats); 719 extern bool btcanreturn(Relation index, int attno); 720 721 /* 722 * prototypes for internal functions in nbtree.c 723 */ 724 extern bool _bt_parallel_seize(IndexScanDesc scan, BlockNumber *pageno); 725 extern void _bt_parallel_release(IndexScanDesc scan, BlockNumber scan_page); 726 extern void _bt_parallel_done(IndexScanDesc scan); 727 extern void _bt_parallel_advance_array_keys(IndexScanDesc scan); 728 729 /* 730 * prototypes for functions in nbtinsert.c 731 */ 732 extern bool _bt_doinsert(Relation rel, IndexTuple itup, 733 IndexUniqueCheck checkUnique, Relation heapRel); 734 extern Buffer _bt_getstackbuf(Relation rel, BTStack stack); 735 extern void _bt_finish_split(Relation rel, Buffer bbuf, BTStack stack); 736 737 /* 738 * prototypes for functions in nbtsplitloc.c 739 */ 740 extern OffsetNumber _bt_findsplitloc(Relation rel, Page page, 741 OffsetNumber newitemoff, Size newitemsz, IndexTuple newitem, 742 bool *newitemonleft); 743 744 /* 745 * prototypes for functions in nbtpage.c 746 */ 747 extern void _bt_initmetapage(Page page, BlockNumber rootbknum, uint32 level); 748 extern void _bt_update_meta_cleanup_info(Relation rel, 749 TransactionId oldestBtpoXact, float8 numHeapTuples); 750 extern void _bt_upgrademetapage(Page page); 751 extern Buffer _bt_getroot(Relation rel, int access); 752 extern Buffer _bt_gettrueroot(Relation rel); 753 extern int _bt_getrootheight(Relation rel); 754 extern bool _bt_heapkeyspace(Relation rel); 755 extern void _bt_checkpage(Relation rel, Buffer buf); 756 extern Buffer _bt_getbuf(Relation rel, BlockNumber blkno, int access); 757 extern Buffer _bt_relandgetbuf(Relation rel, Buffer obuf, 758 BlockNumber blkno, int access); 759 extern void _bt_relbuf(Relation rel, Buffer buf); 760 extern void _bt_pageinit(Page page, Size size); 761 extern bool _bt_page_recyclable(Page page); 762 extern void _bt_delitems_delete(Relation rel, Buffer buf, 763 OffsetNumber *itemnos, int nitems, Relation heapRel); 764 extern void _bt_delitems_vacuum(Relation rel, Buffer buf, 765 OffsetNumber *itemnos, int nitems, 766 BlockNumber lastBlockVacuumed); 767 extern uint32 _bt_pagedel(Relation rel, Buffer leafbuf, 768 TransactionId *oldestBtpoXact); 769 770 /* 771 * prototypes for functions in nbtsearch.c 772 */ 773 extern BTStack _bt_search(Relation rel, BTScanInsert key, Buffer *bufP, 774 int access, Snapshot snapshot); 775 extern Buffer _bt_moveright(Relation rel, BTScanInsert key, Buffer buf, 776 bool forupdate, BTStack stack, int access, Snapshot snapshot); 777 extern OffsetNumber _bt_binsrch_insert(Relation rel, BTInsertState insertstate); 778 extern int32 _bt_compare(Relation rel, BTScanInsert key, Page page, OffsetNumber offnum); 779 extern bool _bt_first(IndexScanDesc scan, ScanDirection dir); 780 extern bool _bt_next(IndexScanDesc scan, ScanDirection dir); 781 extern Buffer _bt_get_endpoint(Relation rel, uint32 level, bool rightmost, 782 Snapshot snapshot); 783 784 /* 785 * prototypes for functions in nbtutils.c 786 */ 787 extern BTScanInsert _bt_mkscankey(Relation rel, IndexTuple itup); 788 extern void _bt_freestack(BTStack stack); 789 extern void _bt_preprocess_array_keys(IndexScanDesc scan); 790 extern void _bt_start_array_keys(IndexScanDesc scan, ScanDirection dir); 791 extern bool _bt_advance_array_keys(IndexScanDesc scan, ScanDirection dir); 792 extern void _bt_mark_array_keys(IndexScanDesc scan); 793 extern void _bt_restore_array_keys(IndexScanDesc scan); 794 extern void _bt_preprocess_keys(IndexScanDesc scan); 795 extern bool _bt_checkkeys(IndexScanDesc scan, IndexTuple tuple, 796 int tupnatts, ScanDirection dir, bool *continuescan); 797 extern void _bt_killitems(IndexScanDesc scan); 798 extern BTCycleId _bt_vacuum_cycleid(Relation rel); 799 extern BTCycleId _bt_start_vacuum(Relation rel); 800 extern void _bt_end_vacuum(Relation rel); 801 extern void _bt_end_vacuum_callback(int code, Datum arg); 802 extern Size BTreeShmemSize(void); 803 extern void BTreeShmemInit(void); 804 extern bytea *btoptions(Datum reloptions, bool validate); 805 extern bool btproperty(Oid index_oid, int attno, 806 IndexAMProperty prop, const char *propname, 807 bool *res, bool *isnull); 808 extern char *btbuildphasename(int64 phasenum); 809 extern IndexTuple _bt_truncate(Relation rel, IndexTuple lastleft, 810 IndexTuple firstright, BTScanInsert itup_key); 811 extern int _bt_keep_natts_fast(Relation rel, IndexTuple lastleft, 812 IndexTuple firstright); 813 extern bool _bt_check_natts(Relation rel, bool heapkeyspace, Page page, 814 OffsetNumber offnum); 815 extern void _bt_check_third_page(Relation rel, Relation heap, 816 bool needheaptidspace, Page page, IndexTuple newtup); 817 818 /* 819 * prototypes for functions in nbtvalidate.c 820 */ 821 extern bool btvalidate(Oid opclassoid); 822 823 /* 824 * prototypes for functions in nbtsort.c 825 */ 826 extern IndexBuildResult *btbuild(Relation heap, Relation index, 827 struct IndexInfo *indexInfo); 828 extern void _bt_parallel_build_main(dsm_segment *seg, shm_toc *toc); 829 830 #endif /* NBTREE_H */ 831