| 1 |
/*------------------------------------------------------------------------- |
| 2 |
* |
| 3 |
* hashsearch.c |
| 4 |
* search code for postgres hash tables |
| 5 |
* |
| 6 |
* Portions Copyright (c) 1996-2025, PostgreSQL Global Development Group |
| 7 |
* Portions Copyright (c) 1994, Regents of the University of California |
| 8 |
* |
| 9 |
* |
| 10 |
* IDENTIFICATION |
| 11 |
* src/backend/access/hash/hashsearch.c |
| 12 |
* |
| 13 |
*------------------------------------------------------------------------- |
| 14 |
*/ |
| 15 |
#include "postgres.h" |
| 16 |
|
| 17 |
#include "access/hash.h" |
| 18 |
#include "access/relscan.h" |
| 19 |
#include "miscadmin.h" |
| 20 |
#include "pgstat.h" |
| 21 |
#include "storage/predicate.h" |
| 22 |
#include "utils/rel.h" |
| 23 |
|
| 24 |
static bool _hash_readpage(IndexScanDesc scan, Buffer *bufP, |
| 25 |
ScanDirection dir); |
| 26 |
static int _hash_load_qualified_items(IndexScanDesc scan, Page page, |
| 27 |
OffsetNumber offnum, ScanDirection dir); |
| 28 |
static inline void _hash_saveitem(HashScanOpaque so, int itemIndex, |
| 29 |
OffsetNumber offnum, IndexTuple itup); |
| 30 |
static void _hash_readnext(IndexScanDesc scan, Buffer *bufp, |
| 31 |
Page *pagep, HashPageOpaque *opaquep); |
| 32 |
|
| 33 |
/* |
| 34 |
* _hash_next() -- Get the next item in a scan. |
| 35 |
* |
| 36 |
* On entry, so->currPos describes the current page, which may |
| 37 |
* be pinned but not locked, and so->currPos.itemIndex identifies |
| 38 |
* which item was previously returned. |
| 39 |
* |
| 40 |
* On successful exit, scan->xs_heaptid is set to the TID of the next |
| 41 |
* heap tuple. so->currPos is updated as needed. |
| 42 |
* |
| 43 |
* On failure exit (no more tuples), we return false with pin |
| 44 |
* held on bucket page but no pins or locks held on overflow |
| 45 |
* page. |
| 46 |
*/ |
| 47 |
bool |
| 48 |
_hash_next(IndexScanDesc scan, ScanDirection dir) |
| 49 |
{ |
| 50 |
Relation rel = scan->indexRelation; |
| 51 |
HashScanOpaque so = (HashScanOpaque) scan->opaque; |
| 52 |
HashScanPosItem *currItem; |
| 53 |
BlockNumber blkno; |
| 54 |
Buffer buf; |
| 55 |
bool end_of_scan = false; |
| 56 |
|
| 57 |
/* |
| 58 |
* Advance to the next tuple on the current page; or if done, try to read |
| 59 |
* data from the next or previous page based on the scan direction. Before |
| 60 |
* moving to the next or previous page make sure that we deal with all the |
| 61 |
* killed items. |
| 62 |
*/ |
| 63 |
if (ScanDirectionIsForward(dir)) |
| 64 |
{ |
| 65 |
if (++so->currPos.itemIndex > so->currPos.lastItem) |
| 66 |
{ |
| 67 |
if (so->numKilled > 0) |
| 68 |
_hash_kill_items(scan); |
| 69 |
|
| 70 |
blkno = so->currPos.nextPage; |
| 71 |
if (BlockNumberIsValid(blkno)) |
| 72 |
{ |
| 73 |
buf = _hash_getbuf(rel, blkno, HASH_READ, LH_OVERFLOW_PAGE); |
| 74 |
if (!_hash_readpage(scan, &buf, dir)) |
| 75 |
end_of_scan = true; |
| 76 |
} |
| 77 |
else |
| 78 |
end_of_scan = true; |
| 79 |
} |
| 80 |
} |
| 81 |
else |
| 82 |
{ |
| 83 |
if (--so->currPos.itemIndex < so->currPos.firstItem) |
| 84 |
{ |
| 85 |
if (so->numKilled > 0) |
| 86 |
_hash_kill_items(scan); |
| 87 |
|
| 88 |
blkno = so->currPos.prevPage; |
| 89 |
if (BlockNumberIsValid(blkno)) |
| 90 |
{ |
| 91 |
buf = _hash_getbuf(rel, blkno, HASH_READ, |
| 92 |
LH_BUCKET_PAGE | LH_OVERFLOW_PAGE); |
| 93 |
|
| 94 |
/* |
| 95 |
* We always maintain the pin on bucket page for whole scan |
| 96 |
* operation, so releasing the additional pin we have acquired |
| 97 |
* here. |
| 98 |
*/ |
| 99 |
if (buf == so->hashso_bucket_buf || |
| 100 |
buf == so->hashso_split_bucket_buf) |
| 101 |
_hash_dropbuf(rel, buf); |
| 102 |
|
| 103 |
if (!_hash_readpage(scan, &buf, dir)) |
| 104 |
end_of_scan = true; |
| 105 |
} |
| 106 |
else |
| 107 |
end_of_scan = true; |
| 108 |
} |
| 109 |
} |
| 110 |
|
| 111 |
if (end_of_scan) |
| 112 |
{ |
| 113 |
_hash_dropscanbuf(rel, so); |
| 114 |
HashScanPosInvalidate(so->currPos); |
| 115 |
return false; |
| 116 |
} |
| 117 |
|
| 118 |
/* OK, itemIndex says what to return */ |
| 119 |
currItem = &so->currPos.items[so->currPos.itemIndex]; |
| 120 |
scan->xs_heaptid = currItem->heapTid; |
| 121 |
|
| 122 |
return true; |
| 123 |
} |
| 124 |
|
| 125 |
/* |
| 126 |
* Advance to next page in a bucket, if any. If we are scanning the bucket |
| 127 |
* being populated during split operation then this function advances to the |
| 128 |
* bucket being split after the last bucket page of bucket being populated. |
| 129 |
*/ |
| 130 |
static void |
| 131 |
_hash_readnext(IndexScanDesc scan, |
| 132 |
Buffer *bufp, Page *pagep, HashPageOpaque *opaquep) |
| 133 |
{ |
| 134 |
BlockNumber blkno; |
| 135 |
Relation rel = scan->indexRelation; |
| 136 |
HashScanOpaque so = (HashScanOpaque) scan->opaque; |
| 137 |
bool block_found = false; |
| 138 |
|
| 139 |
blkno = (*opaquep)->hasho_nextblkno; |
| 140 |
|
| 141 |
/* |
| 142 |
* Retain the pin on primary bucket page till the end of scan. Refer the |
| 143 |
* comments in _hash_first to know the reason of retaining pin. |
| 144 |
*/ |
| 145 |
if (*bufp == so->hashso_bucket_buf || *bufp == so->hashso_split_bucket_buf) |
| 146 |
LockBuffer(*bufp, BUFFER_LOCK_UNLOCK); |
| 147 |
else |
| 148 |
_hash_relbuf(rel, *bufp); |
| 149 |
|
| 150 |
*bufp = InvalidBuffer; |
| 151 |
/* check for interrupts while we're not holding any buffer lock */ |
| 152 |
CHECK_FOR_INTERRUPTS(); |
| 153 |
if (BlockNumberIsValid(blkno)) |
| 154 |
{ |
| 155 |
*bufp = _hash_getbuf(rel, blkno, HASH_READ, LH_OVERFLOW_PAGE); |
| 156 |
block_found = true; |
| 157 |
} |
| 158 |
else if (so->hashso_buc_populated && !so->hashso_buc_split) |
| 159 |
{ |
| 160 |
/* |
| 161 |
* end of bucket, scan bucket being split if there was a split in |
| 162 |
* progress at the start of scan. |
| 163 |
*/ |
| 164 |
*bufp = so->hashso_split_bucket_buf; |
| 165 |
|
| 166 |
/* |
| 167 |
* buffer for bucket being split must be valid as we acquire the pin |
| 168 |
* on it before the start of scan and retain it till end of scan. |
| 169 |
*/ |
| 170 |
Assert(BufferIsValid(*bufp)); |
| 171 |
|
| 172 |
LockBuffer(*bufp, BUFFER_LOCK_SHARE); |
| 173 |
PredicateLockPage(rel, BufferGetBlockNumber(*bufp), scan->xs_snapshot); |
| 174 |
|
| 175 |
/* |
| 176 |
* setting hashso_buc_split to true indicates that we are scanning |
| 177 |
* bucket being split. |
| 178 |
*/ |
| 179 |
so->hashso_buc_split = true; |
| 180 |
|
| 181 |
block_found = true; |
| 182 |
} |
| 183 |
|
| 184 |
if (block_found) |
| 185 |
{ |
| 186 |
*pagep = BufferGetPage(*bufp); |
| 187 |
*opaquep = HashPageGetOpaque(*pagep); |
| 188 |
} |
| 189 |
} |
| 190 |
|
| 191 |
/* |
| 192 |
* Advance to previous page in a bucket, if any. If the current scan has |
| 193 |
* started during split operation then this function advances to bucket |
| 194 |
* being populated after the first bucket page of bucket being split. |
| 195 |
*/ |
| 196 |
static void |
| 197 |
_hash_readprev(IndexScanDesc scan, |
| 198 |
Buffer *bufp, Page *pagep, HashPageOpaque *opaquep) |
| 199 |
{ |
| 200 |
BlockNumber blkno; |
| 201 |
Relation rel = scan->indexRelation; |
| 202 |
HashScanOpaque so = (HashScanOpaque) scan->opaque; |
| 203 |
bool haveprevblk; |
| 204 |
|
| 205 |
blkno = (*opaquep)->hasho_prevblkno; |
| 206 |
|
| 207 |
/* |
| 208 |
* Retain the pin on primary bucket page till the end of scan. Refer the |
| 209 |
* comments in _hash_first to know the reason of retaining pin. |
| 210 |
*/ |
| 211 |
if (*bufp == so->hashso_bucket_buf || *bufp == so->hashso_split_bucket_buf) |
| 212 |
{ |
| 213 |
LockBuffer(*bufp, BUFFER_LOCK_UNLOCK); |
| 214 |
haveprevblk = false; |
| 215 |
} |
| 216 |
else |
| 217 |
{ |
| 218 |
_hash_relbuf(rel, *bufp); |
| 219 |
haveprevblk = true; |
| 220 |
} |
| 221 |
|
| 222 |
*bufp = InvalidBuffer; |
| 223 |
/* check for interrupts while we're not holding any buffer lock */ |
| 224 |
CHECK_FOR_INTERRUPTS(); |
| 225 |
|
| 226 |
if (haveprevblk) |
| 227 |
{ |
| 228 |
Assert(BlockNumberIsValid(blkno)); |
| 229 |
*bufp = _hash_getbuf(rel, blkno, HASH_READ, |
| 230 |
LH_BUCKET_PAGE | LH_OVERFLOW_PAGE); |
| 231 |
*pagep = BufferGetPage(*bufp); |
| 232 |
*opaquep = HashPageGetOpaque(*pagep); |
| 233 |
|
| 234 |
/* |
| 235 |
* We always maintain the pin on bucket page for whole scan operation, |
| 236 |
* so releasing the additional pin we have acquired here. |
| 237 |
*/ |
| 238 |
if (*bufp == so->hashso_bucket_buf || *bufp == so->hashso_split_bucket_buf) |
| 239 |
_hash_dropbuf(rel, *bufp); |
| 240 |
} |
| 241 |
else if (so->hashso_buc_populated && so->hashso_buc_split) |
| 242 |
{ |
| 243 |
/* |
| 244 |
* end of bucket, scan bucket being populated if there was a split in |
| 245 |
* progress at the start of scan. |
| 246 |
*/ |
| 247 |
*bufp = so->hashso_bucket_buf; |
| 248 |
|
| 249 |
/* |
| 250 |
* buffer for bucket being populated must be valid as we acquire the |
| 251 |
* pin on it before the start of scan and retain it till end of scan. |
| 252 |
*/ |
| 253 |
Assert(BufferIsValid(*bufp)); |
| 254 |
|
| 255 |
LockBuffer(*bufp, BUFFER_LOCK_SHARE); |
| 256 |
*pagep = BufferGetPage(*bufp); |
| 257 |
*opaquep = HashPageGetOpaque(*pagep); |
| 258 |
|
| 259 |
/* move to the end of bucket chain */ |
| 260 |
while (BlockNumberIsValid((*opaquep)->hasho_nextblkno)) |
| 261 |
_hash_readnext(scan, bufp, pagep, opaquep); |
| 262 |
|
| 263 |
/* |
| 264 |
* setting hashso_buc_split to false indicates that we are scanning |
| 265 |
* bucket being populated. |
| 266 |
*/ |
| 267 |
so->hashso_buc_split = false; |
| 268 |
} |
| 269 |
} |
| 270 |
|
| 271 |
/* |
| 272 |
* _hash_first() -- Find the first item in a scan. |
| 273 |
* |
| 274 |
* We find the first item (or, if backward scan, the last item) in the |
| 275 |
* index that satisfies the qualification associated with the scan |
| 276 |
* descriptor. |
| 277 |
* |
| 278 |
* On successful exit, if the page containing current index tuple is an |
| 279 |
* overflow page, both pin and lock are released whereas if it is a bucket |
| 280 |
* page then it is pinned but not locked and data about the matching |
| 281 |
* tuple(s) on the page has been loaded into so->currPos, |
| 282 |
* scan->xs_heaptid is set to the heap TID of the current tuple. |
| 283 |
* |
| 284 |
* On failure exit (no more tuples), we return false, with pin held on |
| 285 |
* bucket page but no pins or locks held on overflow page. |
| 286 |
*/ |
| 287 |
bool |
| 288 |
_hash_first(IndexScanDesc scan, ScanDirection dir) |
| 289 |
{ |
| 290 |
Relation rel = scan->indexRelation; |
| 291 |
HashScanOpaque so = (HashScanOpaque) scan->opaque; |
| 292 |
ScanKey cur; |
| 293 |
uint32 hashkey; |
| 294 |
Bucket bucket; |
| 295 |
Buffer buf; |
| 296 |
Page page; |
| 297 |
HashPageOpaque opaque; |
| 298 |
HashScanPosItem *currItem; |
| 299 |
|
| 300 |
pgstat_count_index_scan(rel); |
| 301 |
if (scan->instrument) |
| 302 |
scan->instrument->nsearches++; |
| 303 |
|
| 304 |
/* |
| 305 |
* We do not support hash scans with no index qualification, because we |
| 306 |
* would have to read the whole index rather than just one bucket. That |
| 307 |
* creates a whole raft of problems, since we haven't got a practical way |
| 308 |
* to lock all the buckets against splits or compactions. |
| 309 |
*/ |
| 310 |
if (scan->numberOfKeys < 1) |
| 311 |
ereport(ERROR, |
| 312 |
(errcode(ERRCODE_FEATURE_NOT_SUPPORTED), |
| 313 |
errmsg("hash indexes do not support whole-index scans"))); |
| 314 |
|
| 315 |
/* There may be more than one index qual, but we hash only the first */ |
| 316 |
cur = &scan->keyData[0]; |
| 317 |
|
| 318 |
/* We support only single-column hash indexes */ |
| 319 |
Assert(cur->sk_attno == 1); |
| 320 |
/* And there's only one operator strategy, too */ |
| 321 |
Assert(cur->sk_strategy == HTEqualStrategyNumber); |
| 322 |
|
| 323 |
/* |
| 324 |
* If the constant in the index qual is NULL, assume it cannot match any |
| 325 |
* items in the index. |
| 326 |
*/ |
| 327 |
if (cur->sk_flags & SK_ISNULL) |
| 328 |
return false; |
| 329 |
|
| 330 |
/* |
| 331 |
* Okay to compute the hash key. We want to do this before acquiring any |
| 332 |
* locks, in case a user-defined hash function happens to be slow. |
| 333 |
* |
| 334 |
* If scankey operator is not a cross-type comparison, we can use the |
| 335 |
* cached hash function; otherwise gotta look it up in the catalogs. |
| 336 |
* |
| 337 |
* We support the convention that sk_subtype == InvalidOid means the |
| 338 |
* opclass input type; this is a hack to simplify life for ScanKeyInit(). |
| 339 |
*/ |
| 340 |
if (cur->sk_subtype == rel->rd_opcintype[0] || |
| 341 |
cur->sk_subtype == InvalidOid) |
| 342 |
hashkey = _hash_datum2hashkey(rel, cur->sk_argument); |
| 343 |
else |
| 344 |
hashkey = _hash_datum2hashkey_type(rel, cur->sk_argument, |
| 345 |
cur->sk_subtype); |
| 346 |
|
| 347 |
so->hashso_sk_hash = hashkey; |
| 348 |
|
| 349 |
buf = _hash_getbucketbuf_from_hashkey(rel, hashkey, HASH_READ, NULL); |
| 350 |
PredicateLockPage(rel, BufferGetBlockNumber(buf), scan->xs_snapshot); |
| 351 |
page = BufferGetPage(buf); |
| 352 |
opaque = HashPageGetOpaque(page); |
| 353 |
bucket = opaque->hasho_bucket; |
| 354 |
|
| 355 |
so->hashso_bucket_buf = buf; |
| 356 |
|
| 357 |
/* |
| 358 |
* If a bucket split is in progress, then while scanning the bucket being |
| 359 |
* populated, we need to skip tuples that were copied from bucket being |
| 360 |
* split. We also need to maintain a pin on the bucket being split to |
| 361 |
* ensure that split-cleanup work done by vacuum doesn't remove tuples |
| 362 |
* from it till this scan is done. We need to maintain a pin on the |
| 363 |
* bucket being populated to ensure that vacuum doesn't squeeze that |
| 364 |
* bucket till this scan is complete; otherwise, the ordering of tuples |
| 365 |
* can't be maintained during forward and backward scans. Here, we have |
| 366 |
* to be cautious about locking order: first, acquire the lock on bucket |
| 367 |
* being split; then, release the lock on it but not the pin; then, |
| 368 |
* acquire a lock on bucket being populated and again re-verify whether |
| 369 |
* the bucket split is still in progress. Acquiring the lock on bucket |
| 370 |
* being split first ensures that the vacuum waits for this scan to |
| 371 |
* finish. |
| 372 |
*/ |
| 373 |
if (H_BUCKET_BEING_POPULATED(opaque)) |
| 374 |
{ |
| 375 |
BlockNumber old_blkno; |
| 376 |
Buffer old_buf; |
| 377 |
|
| 378 |
old_blkno = _hash_get_oldblock_from_newbucket(rel, bucket); |
| 379 |
|
| 380 |
/* |
| 381 |
* release the lock on new bucket and re-acquire it after acquiring |
| 382 |
* the lock on old bucket. |
| 383 |
*/ |
| 384 |
LockBuffer(buf, BUFFER_LOCK_UNLOCK); |
| 385 |
|
| 386 |
old_buf = _hash_getbuf(rel, old_blkno, HASH_READ, LH_BUCKET_PAGE); |
| 387 |
|
| 388 |
/* |
| 389 |
* remember the split bucket buffer so as to use it later for |
| 390 |
* scanning. |
| 391 |
*/ |
| 392 |
so->hashso_split_bucket_buf = old_buf; |
| 393 |
LockBuffer(old_buf, BUFFER_LOCK_UNLOCK); |
| 394 |
|
| 395 |
LockBuffer(buf, BUFFER_LOCK_SHARE); |
| 396 |
page = BufferGetPage(buf); |
| 397 |
opaque = HashPageGetOpaque(page); |
| 398 |
Assert(opaque->hasho_bucket == bucket); |
| 399 |
|
| 400 |
if (H_BUCKET_BEING_POPULATED(opaque)) |
| 401 |
so->hashso_buc_populated = true; |
| 402 |
else |
| 403 |
{ |
| 404 |
_hash_dropbuf(rel, so->hashso_split_bucket_buf); |
| 405 |
so->hashso_split_bucket_buf = InvalidBuffer; |
| 406 |
} |
| 407 |
} |
| 408 |
|
| 409 |
/* If a backwards scan is requested, move to the end of the chain */ |
| 410 |
if (ScanDirectionIsBackward(dir)) |
| 411 |
{ |
| 412 |
/* |
| 413 |
* Backward scans that start during split needs to start from end of |
| 414 |
* bucket being split. |
| 415 |
*/ |
| 416 |
while (BlockNumberIsValid(opaque->hasho_nextblkno) || |
| 417 |
(so->hashso_buc_populated && !so->hashso_buc_split)) |
| 418 |
_hash_readnext(scan, &buf, &page, &opaque); |
| 419 |
} |
| 420 |
|
| 421 |
/* remember which buffer we have pinned, if any */ |
| 422 |
Assert(BufferIsInvalid(so->currPos.buf)); |
| 423 |
so->currPos.buf = buf; |
| 424 |
|
| 425 |
/* Now find all the tuples satisfying the qualification from a page */ |
| 426 |
if (!_hash_readpage(scan, &buf, dir)) |
| 427 |
return false; |
| 428 |
|
| 429 |
/* OK, itemIndex says what to return */ |
| 430 |
currItem = &so->currPos.items[so->currPos.itemIndex]; |
| 431 |
scan->xs_heaptid = currItem->heapTid; |
| 432 |
|
| 433 |
/* if we're here, _hash_readpage found a valid tuples */ |
| 434 |
return true; |
| 435 |
} |
| 436 |
|
| 437 |
/* |
| 438 |
* _hash_readpage() -- Load data from current index page into so->currPos |
| 439 |
* |
| 440 |
* We scan all the items in the current index page and save them into |
| 441 |
* so->currPos if it satisfies the qualification. If no matching items |
| 442 |
* are found in the current page, we move to the next or previous page |
| 443 |
* in a bucket chain as indicated by the direction. |
| 444 |
* |
| 445 |
* Return true if any matching items are found else return false. |
| 446 |
*/ |
| 447 |
static bool |
| 448 |
_hash_readpage(IndexScanDesc scan, Buffer *bufP, ScanDirection dir) |
| 449 |
{ |
| 450 |
Relation rel = scan->indexRelation; |
| 451 |
HashScanOpaque so = (HashScanOpaque) scan->opaque; |
| 452 |
Buffer buf; |
| 453 |
Page page; |
| 454 |
HashPageOpaque opaque; |
| 455 |
OffsetNumber offnum; |
| 456 |
uint16 itemIndex; |
| 457 |
|
| 458 |
buf = *bufP; |
| 459 |
Assert(BufferIsValid(buf)); |
| 460 |
_hash_checkpage(rel, buf, LH_BUCKET_PAGE | LH_OVERFLOW_PAGE); |
| 461 |
page = BufferGetPage(buf); |
| 462 |
opaque = HashPageGetOpaque(page); |
| 463 |
|
| 464 |
so->currPos.buf = buf; |
| 465 |
so->currPos.currPage = BufferGetBlockNumber(buf); |
| 466 |
|
| 467 |
if (ScanDirectionIsForward(dir)) |
| 468 |
{ |
| 469 |
BlockNumber prev_blkno = InvalidBlockNumber; |
| 470 |
|
| 471 |
for (;;) |
| 472 |
{ |
| 473 |
/* new page, locate starting position by binary search */ |
| 474 |
offnum = _hash_binsearch(page, so->hashso_sk_hash); |
| 475 |
|
| 476 |
itemIndex = _hash_load_qualified_items(scan, page, offnum, dir); |
| 477 |
|
| 478 |
if (itemIndex != 0) |
| 479 |
break; |
| 480 |
|
| 481 |
/* |
| 482 |
* Could not find any matching tuples in the current page, move to |
| 483 |
* the next page. Before leaving the current page, deal with any |
| 484 |
* killed items. |
| 485 |
*/ |
| 486 |
if (so->numKilled > 0) |
| 487 |
_hash_kill_items(scan); |
| 488 |
|
| 489 |
/* |
| 490 |
* If this is a primary bucket page, hasho_prevblkno is not a real |
| 491 |
* block number. |
| 492 |
*/ |
| 493 |
if (so->currPos.buf == so->hashso_bucket_buf || |
| 494 |
so->currPos.buf == so->hashso_split_bucket_buf) |
| 495 |
prev_blkno = InvalidBlockNumber; |
| 496 |
else |
| 497 |
prev_blkno = opaque->hasho_prevblkno; |
| 498 |
|
| 499 |
_hash_readnext(scan, &buf, &page, &opaque); |
| 500 |
if (BufferIsValid(buf)) |
| 501 |
{ |
| 502 |
so->currPos.buf = buf; |
| 503 |
so->currPos.currPage = BufferGetBlockNumber(buf); |
| 504 |
} |
| 505 |
else |
| 506 |
{ |
| 507 |
/* |
| 508 |
* Remember next and previous block numbers for scrollable |
| 509 |
* cursors to know the start position and return false |
| 510 |
* indicating that no more matching tuples were found. Also, |
| 511 |
* don't reset currPage or lsn, because we expect |
| 512 |
* _hash_kill_items to be called for the old page after this |
| 513 |
* function returns. |
| 514 |
*/ |
| 515 |
so->currPos.prevPage = prev_blkno; |
| 516 |
so->currPos.nextPage = InvalidBlockNumber; |
| 517 |
so->currPos.buf = buf; |
| 518 |
return false; |
| 519 |
} |
| 520 |
} |
| 521 |
|
| 522 |
so->currPos.firstItem = 0; |
| 523 |
so->currPos.lastItem = itemIndex - 1; |
| 524 |
so->currPos.itemIndex = 0; |
| 525 |
} |
| 526 |
else |
| 527 |
{ |
| 528 |
BlockNumber next_blkno = InvalidBlockNumber; |
| 529 |
|
| 530 |
for (;;) |
| 531 |
{ |
| 532 |
/* new page, locate starting position by binary search */ |
| 533 |
offnum = _hash_binsearch_last(page, so->hashso_sk_hash); |
| 534 |
|
| 535 |
itemIndex = _hash_load_qualified_items(scan, page, offnum, dir); |
| 536 |
|
| 537 |
if (itemIndex != MaxIndexTuplesPerPage) |
| 538 |
break; |
| 539 |
|
| 540 |
/* |
| 541 |
* Could not find any matching tuples in the current page, move to |
| 542 |
* the previous page. Before leaving the current page, deal with |
| 543 |
* any killed items. |
| 544 |
*/ |
| 545 |
if (so->numKilled > 0) |
| 546 |
_hash_kill_items(scan); |
| 547 |
|
| 548 |
if (so->currPos.buf == so->hashso_bucket_buf || |
| 549 |
so->currPos.buf == so->hashso_split_bucket_buf) |
| 550 |
next_blkno = opaque->hasho_nextblkno; |
| 551 |
|
| 552 |
_hash_readprev(scan, &buf, &page, &opaque); |
| 553 |
if (BufferIsValid(buf)) |
| 554 |
{ |
| 555 |
so->currPos.buf = buf; |
| 556 |
so->currPos.currPage = BufferGetBlockNumber(buf); |
| 557 |
} |
| 558 |
else |
| 559 |
{ |
| 560 |
/* |
| 561 |
* Remember next and previous block numbers for scrollable |
| 562 |
* cursors to know the start position and return false |
| 563 |
* indicating that no more matching tuples were found. Also, |
| 564 |
* don't reset currPage or lsn, because we expect |
| 565 |
* _hash_kill_items to be called for the old page after this |
| 566 |
* function returns. |
| 567 |
*/ |
| 568 |
so->currPos.prevPage = InvalidBlockNumber; |
| 569 |
so->currPos.nextPage = next_blkno; |
| 570 |
so->currPos.buf = buf; |
| 571 |
return false; |
| 572 |
} |
| 573 |
} |
| 574 |
|
| 575 |
so->currPos.firstItem = itemIndex; |
| 576 |
so->currPos.lastItem = MaxIndexTuplesPerPage - 1; |
| 577 |
so->currPos.itemIndex = MaxIndexTuplesPerPage - 1; |
| 578 |
} |
| 579 |
|
| 580 |
if (so->currPos.buf == so->hashso_bucket_buf || |
| 581 |
so->currPos.buf == so->hashso_split_bucket_buf) |
| 582 |
{ |
| 583 |
so->currPos.prevPage = InvalidBlockNumber; |
| 584 |
so->currPos.nextPage = opaque->hasho_nextblkno; |
| 585 |
LockBuffer(so->currPos.buf, BUFFER_LOCK_UNLOCK); |
| 586 |
} |
| 587 |
else |
| 588 |
{ |
| 589 |
so->currPos.prevPage = opaque->hasho_prevblkno; |
| 590 |
so->currPos.nextPage = opaque->hasho_nextblkno; |
| 591 |
_hash_relbuf(rel, so->currPos.buf); |
| 592 |
so->currPos.buf = InvalidBuffer; |
| 593 |
} |
| 594 |
|
| 595 |
Assert(so->currPos.firstItem <= so->currPos.lastItem); |
| 596 |
return true; |
| 597 |
} |
| 598 |
|
| 599 |
/* |
| 600 |
* Load all the qualified items from a current index page |
| 601 |
* into so->currPos. Helper function for _hash_readpage. |
| 602 |
*/ |
| 603 |
static int |
| 604 |
_hash_load_qualified_items(IndexScanDesc scan, Page page, |
| 605 |
OffsetNumber offnum, ScanDirection dir) |
| 606 |
{ |
| 607 |
HashScanOpaque so = (HashScanOpaque) scan->opaque; |
| 608 |
IndexTuple itup; |
| 609 |
int itemIndex; |
| 610 |
OffsetNumber maxoff; |
| 611 |
|
| 612 |
maxoff = PageGetMaxOffsetNumber(page); |
| 613 |
|
| 614 |
if (ScanDirectionIsForward(dir)) |
| 615 |
{ |
| 616 |
/* load items[] in ascending order */ |
| 617 |
itemIndex = 0; |
| 618 |
|
| 619 |
while (offnum <= maxoff) |
| 620 |
{ |
| 621 |
Assert(offnum >= FirstOffsetNumber); |
| 622 |
itup = (IndexTuple) PageGetItem(page, PageGetItemId(page, offnum)); |
| 623 |
|
| 624 |
/* |
| 625 |
* skip the tuples that are moved by split operation for the scan |
| 626 |
* that has started when split was in progress. Also, skip the |
| 627 |
* tuples that are marked as dead. |
| 628 |
*/ |
| 629 |
if ((so->hashso_buc_populated && !so->hashso_buc_split && |
| 630 |
(itup->t_info & INDEX_MOVED_BY_SPLIT_MASK)) || |
| 631 |
(scan->ignore_killed_tuples && |
| 632 |
(ItemIdIsDead(PageGetItemId(page, offnum))))) |
| 633 |
{ |
| 634 |
offnum = OffsetNumberNext(offnum); /* move forward */ |
| 635 |
continue; |
| 636 |
} |
| 637 |
|
| 638 |
if (so->hashso_sk_hash == _hash_get_indextuple_hashkey(itup) && |
| 639 |
_hash_checkqual(scan, itup)) |
| 640 |
{ |
| 641 |
/* tuple is qualified, so remember it */ |
| 642 |
_hash_saveitem(so, itemIndex, offnum, itup); |
| 643 |
itemIndex++; |
| 644 |
} |
| 645 |
else |
| 646 |
{ |
| 647 |
/* |
| 648 |
* No more matching tuples exist in this page. so, exit while |
| 649 |
* loop. |
| 650 |
*/ |
| 651 |
break; |
| 652 |
} |
| 653 |
|
| 654 |
offnum = OffsetNumberNext(offnum); |
| 655 |
} |
| 656 |
|
| 657 |
Assert(itemIndex <= MaxIndexTuplesPerPage); |
| 658 |
return itemIndex; |
| 659 |
} |
| 660 |
else |
| 661 |
{ |
| 662 |
/* load items[] in descending order */ |
| 663 |
itemIndex = MaxIndexTuplesPerPage; |
| 664 |
|
| 665 |
while (offnum >= FirstOffsetNumber) |
| 666 |
{ |
| 667 |
Assert(offnum <= maxoff); |
| 668 |
itup = (IndexTuple) PageGetItem(page, PageGetItemId(page, offnum)); |
| 669 |
|
| 670 |
/* |
| 671 |
* skip the tuples that are moved by split operation for the scan |
| 672 |
* that has started when split was in progress. Also, skip the |
| 673 |
* tuples that are marked as dead. |
| 674 |
*/ |
| 675 |
if ((so->hashso_buc_populated && !so->hashso_buc_split && |
| 676 |
(itup->t_info & INDEX_MOVED_BY_SPLIT_MASK)) || |
| 677 |
(scan->ignore_killed_tuples && |
| 678 |
(ItemIdIsDead(PageGetItemId(page, offnum))))) |
| 679 |
{ |
| 680 |
offnum = OffsetNumberPrev(offnum); /* move back */ |
| 681 |
continue; |
| 682 |
} |
| 683 |
|
| 684 |
if (so->hashso_sk_hash == _hash_get_indextuple_hashkey(itup) && |
| 685 |
_hash_checkqual(scan, itup)) |
| 686 |
{ |
| 687 |
itemIndex--; |
| 688 |
/* tuple is qualified, so remember it */ |
| 689 |
_hash_saveitem(so, itemIndex, offnum, itup); |
| 690 |
} |
| 691 |
else |
| 692 |
{ |
| 693 |
/* |
| 694 |
* No more matching tuples exist in this page. so, exit while |
| 695 |
* loop. |
| 696 |
*/ |
| 697 |
break; |
| 698 |
} |
| 699 |
|
| 700 |
offnum = OffsetNumberPrev(offnum); |
| 701 |
} |
| 702 |
|
| 703 |
Assert(itemIndex >= 0); |
| 704 |
return itemIndex; |
| 705 |
} |
| 706 |
} |
| 707 |
|
| 708 |
/* Save an index item into so->currPos.items[itemIndex] */ |
| 709 |
static inline void |
| 710 |
_hash_saveitem(HashScanOpaque so, int itemIndex, |
| 711 |
OffsetNumber offnum, IndexTuple itup) |
| 712 |
{ |
| 713 |
HashScanPosItem *currItem = &so->currPos.items[itemIndex]; |
| 714 |
|
| 715 |
currItem->heapTid = itup->t_tid; |
| 716 |
currItem->indexOffset = offnum; |
| 717 |
} |