Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * hio.c
4 : * POSTGRES heap access method input/output code.
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/heap/hio.c
12 : *
13 : *-------------------------------------------------------------------------
14 : */
15 :
16 : #include "postgres.h"
17 :
18 : #include "access/heapam.h"
19 : #include "access/hio.h"
20 : #include "access/htup_details.h"
21 : #include "access/visibilitymap.h"
22 : #include "storage/bufmgr.h"
23 : #include "storage/freespace.h"
24 : #include "storage/lmgr.h"
25 :
26 :
27 : /*
28 : * RelationPutHeapTuple - place tuple at specified page
29 : *
30 : * !!! EREPORT(ERROR) IS DISALLOWED HERE !!! Must PANIC on failure!!!
31 : *
32 : * Note - caller must hold BUFFER_LOCK_EXCLUSIVE on the buffer.
33 : */
34 : void
35 20246498 : RelationPutHeapTuple(Relation relation,
36 : Buffer buffer,
37 : HeapTuple tuple,
38 : bool token)
39 : {
40 : Page pageHeader;
41 : OffsetNumber offnum;
42 :
43 : /*
44 : * A tuple that's being inserted speculatively should already have its
45 : * token set.
46 : */
47 : Assert(!token || HeapTupleHeaderIsSpeculative(tuple->t_data));
48 :
49 : /*
50 : * Do not allow tuples with invalid combinations of hint bits to be placed
51 : * on a page. This combination is detected as corruption by the
52 : * contrib/amcheck logic, so if you disable this assertion, make
53 : * corresponding changes there.
54 : */
55 : Assert(!((tuple->t_data->t_infomask & HEAP_XMAX_COMMITTED) &&
56 : (tuple->t_data->t_infomask & HEAP_XMAX_IS_MULTI)));
57 :
58 : /* Add the tuple to the page */
59 20246498 : pageHeader = BufferGetPage(buffer);
60 :
61 20246498 : offnum = PageAddItem(pageHeader, tuple->t_data, tuple->t_len, InvalidOffsetNumber, false, true);
62 20246498 : if (offnum == InvalidOffsetNumber)
63 0 : elog(PANIC, "failed to add tuple to page");
64 :
65 : /* Update tuple->t_self to the actual position where it was stored */
66 20246498 : ItemPointerSet(&(tuple->t_self), BufferGetBlockNumber(buffer), offnum);
67 :
68 : /*
69 : * Insert the correct position into CTID of the stored tuple, too (unless
70 : * this is a speculative insertion, in which case the token is held in
71 : * CTID field instead)
72 : */
73 20246498 : if (!token)
74 : {
75 20242340 : ItemId itemId = PageGetItemId(pageHeader, offnum);
76 20242340 : HeapTupleHeader item = (HeapTupleHeader) PageGetItem(pageHeader, itemId);
77 :
78 20242340 : item->t_ctid = tuple->t_self;
79 : }
80 20246498 : }
81 :
82 : /*
83 : * Read in a buffer in mode, using bulk-insert strategy if bistate isn't NULL.
84 : */
85 : static Buffer
86 17455910 : ReadBufferBI(Relation relation, BlockNumber targetBlock,
87 : ReadBufferMode mode, BulkInsertState bistate)
88 : {
89 : Buffer buffer;
90 :
91 : /* If not bulk-insert, exactly like ReadBuffer */
92 17455910 : if (!bistate)
93 14991048 : return ReadBufferExtended(relation, MAIN_FORKNUM, targetBlock,
94 : mode, NULL);
95 :
96 : /* If we have the desired block already pinned, re-pin and return it */
97 2464862 : if (bistate->current_buf != InvalidBuffer)
98 : {
99 2404488 : if (BufferGetBlockNumber(bistate->current_buf) == targetBlock)
100 : {
101 : /*
102 : * Currently the LOCK variants are only used for extending
103 : * relation, which should never reach this branch.
104 : */
105 : Assert(mode != RBM_ZERO_AND_LOCK &&
106 : mode != RBM_ZERO_AND_CLEANUP_LOCK);
107 :
108 2379598 : IncrBufferRefCount(bistate->current_buf);
109 2379598 : return bistate->current_buf;
110 : }
111 : /* ... else drop the old buffer */
112 24890 : ReleaseBuffer(bistate->current_buf);
113 24890 : bistate->current_buf = InvalidBuffer;
114 : }
115 :
116 : /* Perform a read using the buffer strategy */
117 85264 : buffer = ReadBufferExtended(relation, MAIN_FORKNUM, targetBlock,
118 : mode, bistate->strategy);
119 :
120 : /* Save the selected block as target for future inserts */
121 85264 : IncrBufferRefCount(buffer);
122 85264 : bistate->current_buf = buffer;
123 :
124 85264 : return buffer;
125 : }
126 :
127 : /*
128 : * For each heap page which is all-visible, acquire a pin on the appropriate
129 : * visibility map page, if we haven't already got one.
130 : *
131 : * To avoid complexity in the callers, either buffer1 or buffer2 may be
132 : * InvalidBuffer if only one buffer is involved. For the same reason, block2
133 : * may be smaller than block1.
134 : *
135 : * Returns whether buffer locks were temporarily released.
136 : */
137 : static bool
138 17765704 : GetVisibilityMapPins(Relation relation, Buffer buffer1, Buffer buffer2,
139 : BlockNumber block1, BlockNumber block2,
140 : Buffer *vmbuffer1, Buffer *vmbuffer2)
141 : {
142 : bool need_to_pin_buffer1;
143 : bool need_to_pin_buffer2;
144 17765704 : bool released_locks = false;
145 :
146 : /*
147 : * Swap buffers around to handle case of a single block/buffer, and to
148 : * handle if lock ordering rules require to lock block2 first.
149 : */
150 35530072 : if (!BufferIsValid(buffer1) ||
151 18072826 : (BufferIsValid(buffer2) && block1 > block2))
152 : {
153 290784 : Buffer tmpbuf = buffer1;
154 290784 : Buffer *tmpvmbuf = vmbuffer1;
155 290784 : BlockNumber tmpblock = block1;
156 :
157 290784 : buffer1 = buffer2;
158 290784 : vmbuffer1 = vmbuffer2;
159 290784 : block1 = block2;
160 :
161 290784 : buffer2 = tmpbuf;
162 290784 : vmbuffer2 = tmpvmbuf;
163 290784 : block2 = tmpblock;
164 : }
165 :
166 : Assert(BufferIsValid(buffer1));
167 : Assert(buffer2 == InvalidBuffer || block1 <= block2);
168 :
169 : while (1)
170 : {
171 : /* Figure out which pins we need but don't have. */
172 17765704 : need_to_pin_buffer1 = PageIsAllVisible(BufferGetPage(buffer1))
173 17765704 : && !visibilitymap_pin_ok(block1, *vmbuffer1);
174 17765704 : need_to_pin_buffer2 = buffer2 != InvalidBuffer
175 308458 : && PageIsAllVisible(BufferGetPage(buffer2))
176 18074162 : && !visibilitymap_pin_ok(block2, *vmbuffer2);
177 17765704 : if (!need_to_pin_buffer1 && !need_to_pin_buffer2)
178 17765704 : break;
179 :
180 : /* We must unlock both buffers before doing any I/O. */
181 0 : released_locks = true;
182 0 : LockBuffer(buffer1, BUFFER_LOCK_UNLOCK);
183 0 : if (buffer2 != InvalidBuffer && buffer2 != buffer1)
184 0 : LockBuffer(buffer2, BUFFER_LOCK_UNLOCK);
185 :
186 : /* Get pins. */
187 0 : if (need_to_pin_buffer1)
188 0 : visibilitymap_pin(relation, block1, vmbuffer1);
189 0 : if (need_to_pin_buffer2)
190 0 : visibilitymap_pin(relation, block2, vmbuffer2);
191 :
192 : /* Relock buffers. */
193 0 : LockBuffer(buffer1, BUFFER_LOCK_EXCLUSIVE);
194 0 : if (buffer2 != InvalidBuffer && buffer2 != buffer1)
195 0 : LockBuffer(buffer2, BUFFER_LOCK_EXCLUSIVE);
196 :
197 : /*
198 : * If there are two buffers involved and we pinned just one of them,
199 : * it's possible that the second one became all-visible while we were
200 : * busy pinning the first one. If it looks like that's a possible
201 : * scenario, we'll need to make a second pass through this loop.
202 : */
203 0 : if (buffer2 == InvalidBuffer || buffer1 == buffer2
204 0 : || (need_to_pin_buffer1 && need_to_pin_buffer2))
205 : break;
206 : }
207 :
208 17765704 : return released_locks;
209 : }
210 :
211 : /*
212 : * Extend the relation. By multiple pages, if beneficial.
213 : *
214 : * If the caller needs multiple pages (num_pages > 1), we always try to extend
215 : * by at least that much.
216 : *
217 : * If there is contention on the extension lock, we don't just extend "for
218 : * ourselves", but we try to help others. We can do so by adding empty pages
219 : * into the FSM. Typically there is no contention when we can't use the FSM.
220 : *
221 : * We do have to limit the number of pages to extend by to some value, as the
222 : * buffers for all the extended pages need to, temporarily, be pinned. For now
223 : * we define MAX_BUFFERS_TO_EXTEND_BY to be 64 buffers, it's hard to see
224 : * benefits with higher numbers. This partially is because copyfrom.c's
225 : * MAX_BUFFERED_TUPLES / MAX_BUFFERED_BYTES prevents larger multi_inserts.
226 : *
227 : * Returns a buffer for a newly extended block. If possible, the buffer is
228 : * returned exclusively locked. *did_unlock is set to true if the lock had to
229 : * be released, false otherwise.
230 : *
231 : *
232 : * XXX: It would likely be beneficial for some workloads to extend more
233 : * aggressively, e.g. using a heuristic based on the relation size.
234 : */
235 : static Buffer
236 227960 : RelationAddBlocks(Relation relation, BulkInsertState bistate,
237 : int num_pages, bool use_fsm, bool *did_unlock)
238 : {
239 : #define MAX_BUFFERS_TO_EXTEND_BY 64
240 : Buffer victim_buffers[MAX_BUFFERS_TO_EXTEND_BY];
241 227960 : BlockNumber first_block = InvalidBlockNumber;
242 227960 : BlockNumber last_block = InvalidBlockNumber;
243 : uint32 extend_by_pages;
244 : uint32 not_in_fsm_pages;
245 : Buffer buffer;
246 : Page page;
247 :
248 : /*
249 : * Determine by how many pages to try to extend by.
250 : */
251 227960 : if (bistate == NULL && !use_fsm)
252 : {
253 : /*
254 : * If we have neither bistate, nor can use the FSM, we can't bulk
255 : * extend - there'd be no way to find the additional pages.
256 : */
257 344 : extend_by_pages = 1;
258 : }
259 : else
260 : {
261 : uint32 waitcount;
262 :
263 : /*
264 : * Try to extend at least by the number of pages the caller needs. We
265 : * can remember the additional pages (either via FSM or bistate).
266 : */
267 227616 : extend_by_pages = num_pages;
268 :
269 227616 : if (!RELATION_IS_LOCAL(relation))
270 135074 : waitcount = RelationExtensionLockWaiterCount(relation);
271 : else
272 92542 : waitcount = 0;
273 :
274 : /*
275 : * Multiply the number of pages to extend by the number of waiters. Do
276 : * this even if we're not using the FSM, as it still relieves
277 : * contention, by deferring the next time this backend needs to
278 : * extend. In that case the extended pages will be found via
279 : * bistate->next_free.
280 : */
281 227616 : extend_by_pages += extend_by_pages * waitcount;
282 :
283 : /* ---
284 : * If we previously extended using the same bistate, it's very likely
285 : * we'll extend some more. Try to extend by as many pages as
286 : * before. This can be important for performance for several reasons,
287 : * including:
288 : *
289 : * - It prevents mdzeroextend() switching between extending the
290 : * relation in different ways, which is inefficient for some
291 : * filesystems.
292 : *
293 : * - Contention is often intermittent. Even if we currently don't see
294 : * other waiters (see above), extending by larger amounts can
295 : * prevent future contention.
296 : * ---
297 : */
298 227616 : if (bistate)
299 11832 : extend_by_pages = Max(extend_by_pages, bistate->already_extended_by);
300 :
301 : /*
302 : * Can't extend by more than MAX_BUFFERS_TO_EXTEND_BY, we need to pin
303 : * them all concurrently.
304 : */
305 227616 : extend_by_pages = Min(extend_by_pages, MAX_BUFFERS_TO_EXTEND_BY);
306 : }
307 :
308 : /*
309 : * How many of the extended pages should be entered into the FSM?
310 : *
311 : * If we have a bistate, only enter pages that we don't need ourselves
312 : * into the FSM. Otherwise every other backend will immediately try to
313 : * use the pages this backend needs for itself, causing unnecessary
314 : * contention. If we don't have a bistate, we can't avoid the FSM.
315 : *
316 : * Never enter the page returned into the FSM, we'll immediately use it.
317 : */
318 227960 : if (num_pages > 1 && bistate == NULL)
319 508 : not_in_fsm_pages = 1;
320 : else
321 227452 : not_in_fsm_pages = num_pages;
322 :
323 : /* prepare to put another buffer into the bistate */
324 227960 : if (bistate && bistate->current_buf != InvalidBuffer)
325 : {
326 8622 : ReleaseBuffer(bistate->current_buf);
327 8622 : bistate->current_buf = InvalidBuffer;
328 : }
329 :
330 : /*
331 : * Extend the relation. We ask for the first returned page to be locked,
332 : * so that we are sure that nobody has inserted into the page
333 : * concurrently.
334 : *
335 : * With the current MAX_BUFFERS_TO_EXTEND_BY there's no danger of
336 : * [auto]vacuum trying to truncate later pages as REL_TRUNCATE_MINIMUM is
337 : * way larger.
338 : */
339 227960 : first_block = ExtendBufferedRelBy(BMR_REL(relation), MAIN_FORKNUM,
340 : bistate ? bistate->strategy : NULL,
341 : EB_LOCK_FIRST,
342 : extend_by_pages,
343 : victim_buffers,
344 : &extend_by_pages);
345 227960 : buffer = victim_buffers[0]; /* the buffer the function will return */
346 227960 : last_block = first_block + (extend_by_pages - 1);
347 : Assert(first_block == BufferGetBlockNumber(buffer));
348 :
349 : /*
350 : * Relation is now extended. Initialize the page. We do this here, before
351 : * potentially releasing the lock on the page, because it allows us to
352 : * double check that the page contents are empty (this should never
353 : * happen, but if it does we don't want to risk wiping out valid data).
354 : */
355 227960 : page = BufferGetPage(buffer);
356 227960 : if (!PageIsNew(page))
357 0 : elog(ERROR, "page %u of relation \"%s\" should be empty but is not",
358 : first_block,
359 : RelationGetRelationName(relation));
360 :
361 227960 : PageInit(page, BufferGetPageSize(buffer), 0);
362 227960 : MarkBufferDirty(buffer);
363 :
364 : /*
365 : * If we decided to put pages into the FSM, release the buffer lock (but
366 : * not pin), we don't want to do IO while holding a buffer lock. This will
367 : * necessitate a bit more extensive checking in our caller.
368 : */
369 227960 : if (use_fsm && not_in_fsm_pages < extend_by_pages)
370 : {
371 784 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
372 784 : *did_unlock = true;
373 : }
374 : else
375 227176 : *did_unlock = false;
376 :
377 : /*
378 : * Relation is now extended. Release pins on all buffers, except for the
379 : * first (which we'll return). If we decided to put pages into the FSM,
380 : * we can do that as part of the same loop.
381 : */
382 262324 : for (uint32 i = 1; i < extend_by_pages; i++)
383 : {
384 34364 : BlockNumber curBlock = first_block + i;
385 :
386 : Assert(curBlock == BufferGetBlockNumber(victim_buffers[i]));
387 : Assert(BlockNumberIsValid(curBlock));
388 :
389 34364 : ReleaseBuffer(victim_buffers[i]);
390 :
391 34364 : if (use_fsm && i >= not_in_fsm_pages)
392 : {
393 10606 : Size freespace = BufferGetPageSize(victim_buffers[i]) -
394 : SizeOfPageHeaderData;
395 :
396 10606 : RecordPageWithFreeSpace(relation, curBlock, freespace);
397 : }
398 : }
399 :
400 227960 : if (use_fsm && not_in_fsm_pages < extend_by_pages)
401 : {
402 784 : BlockNumber first_fsm_block = first_block + not_in_fsm_pages;
403 :
404 784 : FreeSpaceMapVacuumRange(relation, first_fsm_block, last_block);
405 : }
406 :
407 227960 : if (bistate)
408 : {
409 : /*
410 : * Remember the additional pages we extended by, so we later can use
411 : * them without looking into the FSM.
412 : */
413 11832 : if (extend_by_pages > 1)
414 : {
415 1814 : bistate->next_free = first_block + 1;
416 1814 : bistate->last_free = last_block;
417 : }
418 : else
419 : {
420 10018 : bistate->next_free = InvalidBlockNumber;
421 10018 : bistate->last_free = InvalidBlockNumber;
422 : }
423 :
424 : /* maintain bistate->current_buf */
425 11832 : IncrBufferRefCount(buffer);
426 11832 : bistate->current_buf = buffer;
427 11832 : bistate->already_extended_by += extend_by_pages;
428 : }
429 :
430 227960 : return buffer;
431 : #undef MAX_BUFFERS_TO_EXTEND_BY
432 : }
433 :
434 : /*
435 : * RelationGetBufferForTuple
436 : *
437 : * Returns pinned and exclusive-locked buffer of a page in given relation
438 : * with free space >= given len.
439 : *
440 : * If num_pages is > 1, we will try to extend the relation by at least that
441 : * many pages when we decide to extend the relation. This is more efficient
442 : * for callers that know they will need multiple pages
443 : * (e.g. heap_multi_insert()).
444 : *
445 : * If otherBuffer is not InvalidBuffer, then it references a previously
446 : * pinned buffer of another page in the same relation; on return, this
447 : * buffer will also be exclusive-locked. (This case is used by heap_update;
448 : * the otherBuffer contains the tuple being updated.)
449 : *
450 : * The reason for passing otherBuffer is that if two backends are doing
451 : * concurrent heap_update operations, a deadlock could occur if they try
452 : * to lock the same two buffers in opposite orders. To ensure that this
453 : * can't happen, we impose the rule that buffers of a relation must be
454 : * locked in increasing page number order. This is most conveniently done
455 : * by having RelationGetBufferForTuple lock them both, with suitable care
456 : * for ordering.
457 : *
458 : * NOTE: it is unlikely, but not quite impossible, for otherBuffer to be the
459 : * same buffer we select for insertion of the new tuple (this could only
460 : * happen if space is freed in that page after heap_update finds there's not
461 : * enough there). In that case, the page will be pinned and locked only once.
462 : *
463 : * We also handle the possibility that the all-visible flag will need to be
464 : * cleared on one or both pages. If so, pin on the associated visibility map
465 : * page must be acquired before acquiring buffer lock(s), to avoid possibly
466 : * doing I/O while holding buffer locks. The pins are passed back to the
467 : * caller using the input-output arguments vmbuffer and vmbuffer_other.
468 : * Note that in some cases the caller might have already acquired such pins,
469 : * which is indicated by these arguments not being InvalidBuffer on entry.
470 : *
471 : * We normally use FSM to help us find free space. However,
472 : * if HEAP_INSERT_SKIP_FSM is specified, we just append a new empty page to
473 : * the end of the relation if the tuple won't fit on the current target page.
474 : * This can save some cycles when we know the relation is new and doesn't
475 : * contain useful amounts of free space.
476 : *
477 : * HEAP_INSERT_SKIP_FSM is also useful for non-WAL-logged additions to a
478 : * relation, if the caller holds exclusive lock and is careful to invalidate
479 : * relation's smgr_targblock before the first insertion --- that ensures that
480 : * all insertions will occur into newly added pages and not be intermixed
481 : * with tuples from other transactions. That way, a crash can't risk losing
482 : * any committed data of other transactions. (See heap_insert's comments
483 : * for additional constraints needed for safe usage of this behavior.)
484 : *
485 : * The caller can also provide a BulkInsertState object to optimize many
486 : * insertions into the same relation. This keeps a pin on the current
487 : * insertion target page (to save pin/unpin cycles) and also passes a
488 : * BULKWRITE buffer selection strategy object to the buffer manager.
489 : * Passing NULL for bistate selects the default behavior.
490 : *
491 : * We don't fill existing pages further than the fillfactor, except for large
492 : * tuples in nearly-empty pages. This is OK since this routine is not
493 : * consulted when updating a tuple and keeping it on the same page, which is
494 : * the scenario fillfactor is meant to reserve space for.
495 : *
496 : * ereport(ERROR) is allowed here, so this routine *must* be called
497 : * before any (unlogged) changes are made in buffer pool.
498 : */
499 : Buffer
500 17733858 : RelationGetBufferForTuple(Relation relation, Size len,
501 : Buffer otherBuffer, int options,
502 : BulkInsertState bistate,
503 : Buffer *vmbuffer, Buffer *vmbuffer_other,
504 : int num_pages)
505 : {
506 17733858 : bool use_fsm = !(options & HEAP_INSERT_SKIP_FSM);
507 17733858 : Buffer buffer = InvalidBuffer;
508 : Page page;
509 : Size nearlyEmptyFreeSpace,
510 17733858 : pageFreeSpace = 0,
511 17733858 : saveFreeSpace = 0,
512 17733858 : targetFreeSpace = 0;
513 : BlockNumber targetBlock,
514 : otherBlock;
515 : bool unlockedTargetBuffer;
516 : bool recheckVmPins;
517 :
518 17733858 : len = MAXALIGN(len); /* be conservative */
519 :
520 : /* if the caller doesn't know by how many pages to extend, extend by 1 */
521 17733858 : if (num_pages <= 0)
522 16987082 : num_pages = 1;
523 :
524 : /* Bulk insert is not supported for updates, only inserts. */
525 : Assert(otherBuffer == InvalidBuffer || !bistate);
526 :
527 : /*
528 : * If we're gonna fail for oversize tuple, do it right away
529 : */
530 17733858 : if (len > MaxHeapTupleSize)
531 0 : ereport(ERROR,
532 : (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
533 : errmsg("row is too big: size %zu, maximum size %zu",
534 : len, MaxHeapTupleSize)));
535 :
536 : /* Compute desired extra freespace due to fillfactor option */
537 17733858 : saveFreeSpace = RelationGetTargetPageFreeSpace(relation,
538 : HEAP_DEFAULT_FILLFACTOR);
539 :
540 : /*
541 : * Since pages without tuples can still have line pointers, we consider
542 : * pages "empty" when the unavailable space is slight. This threshold is
543 : * somewhat arbitrary, but it should prevent most unnecessary relation
544 : * extensions while inserting large tuples into low-fillfactor tables.
545 : */
546 17733858 : nearlyEmptyFreeSpace = MaxHeapTupleSize -
547 : (MaxHeapTuplesPerPage / 8 * sizeof(ItemIdData));
548 17733858 : if (len + saveFreeSpace > nearlyEmptyFreeSpace)
549 6092 : targetFreeSpace = Max(len, nearlyEmptyFreeSpace);
550 : else
551 17727766 : targetFreeSpace = len + saveFreeSpace;
552 :
553 17733858 : if (otherBuffer != InvalidBuffer)
554 298342 : otherBlock = BufferGetBlockNumber(otherBuffer);
555 : else
556 17435516 : otherBlock = InvalidBlockNumber; /* just to keep compiler quiet */
557 :
558 : /*
559 : * We first try to put the tuple on the same page we last inserted a tuple
560 : * on, as cached in the BulkInsertState or relcache entry. If that
561 : * doesn't work, we ask the Free Space Map to locate a suitable page.
562 : * Since the FSM's info might be out of date, we have to be prepared to
563 : * loop around and retry multiple times. (To ensure this isn't an infinite
564 : * loop, we must update the FSM with the correct amount of free space on
565 : * each page that proves not to be suitable.) If the FSM has no record of
566 : * a page with enough free space, we give up and extend the relation.
567 : *
568 : * When use_fsm is false, we either put the tuple onto the existing target
569 : * page or extend the relation.
570 : */
571 17733858 : if (bistate && bistate->current_buf != InvalidBuffer)
572 2379598 : targetBlock = BufferGetBlockNumber(bistate->current_buf);
573 : else
574 15354260 : targetBlock = RelationGetTargetBlock(relation);
575 :
576 17733858 : if (targetBlock == InvalidBlockNumber && use_fsm)
577 : {
578 : /*
579 : * We have no cached target page, so ask the FSM for an initial
580 : * target.
581 : */
582 94750 : targetBlock = GetPageWithFreeSpace(relation, targetFreeSpace);
583 : }
584 :
585 : /*
586 : * If the FSM knows nothing of the rel, try the last page before we give
587 : * up and extend. This avoids one-tuple-per-page syndrome during
588 : * bootstrapping or in a recently-started system.
589 : */
590 17733858 : if (targetBlock == InvalidBlockNumber)
591 : {
592 75534 : BlockNumber nblocks = RelationGetNumberOfBlocks(relation);
593 :
594 75534 : if (nblocks > 0)
595 45520 : targetBlock = nblocks - 1;
596 : }
597 :
598 17733858 : loop:
599 17979788 : while (targetBlock != InvalidBlockNumber)
600 : {
601 : /*
602 : * Read and exclusive-lock the target block, as well as the other
603 : * block if one was given, taking suitable care with lock ordering and
604 : * the possibility they are the same block.
605 : *
606 : * If the page-level all-visible flag is set, caller will need to
607 : * clear both that and the corresponding visibility map bit. However,
608 : * by the time we return, we'll have x-locked the buffer, and we don't
609 : * want to do any I/O while in that state. So we check the bit here
610 : * before taking the lock, and pin the page if it appears necessary.
611 : * Checking without the lock creates a risk of getting the wrong
612 : * answer, so we'll have to recheck after acquiring the lock.
613 : */
614 17758074 : if (otherBuffer == InvalidBuffer)
615 : {
616 : /* easy case */
617 17455910 : buffer = ReadBufferBI(relation, targetBlock, RBM_NORMAL, bistate);
618 17455910 : if (PageIsAllVisible(BufferGetPage(buffer)))
619 23994 : visibilitymap_pin(relation, targetBlock, vmbuffer);
620 :
621 : /*
622 : * If the page is empty, pin vmbuffer to set all_frozen bit later.
623 : */
624 17466154 : if ((options & HEAP_INSERT_FROZEN) &&
625 10244 : (PageGetMaxOffsetNumber(BufferGetPage(buffer)) == 0))
626 3226 : visibilitymap_pin(relation, targetBlock, vmbuffer);
627 :
628 17455910 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
629 : }
630 302164 : else if (otherBlock == targetBlock)
631 : {
632 : /* also easy case */
633 2950 : buffer = otherBuffer;
634 2950 : if (PageIsAllVisible(BufferGetPage(buffer)))
635 0 : visibilitymap_pin(relation, targetBlock, vmbuffer);
636 2950 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
637 : }
638 299214 : else if (otherBlock < targetBlock)
639 : {
640 : /* lock other buffer first */
641 289448 : buffer = ReadBuffer(relation, targetBlock);
642 289448 : if (PageIsAllVisible(BufferGetPage(buffer)))
643 1386 : visibilitymap_pin(relation, targetBlock, vmbuffer);
644 289448 : LockBuffer(otherBuffer, BUFFER_LOCK_EXCLUSIVE);
645 289448 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
646 : }
647 : else
648 : {
649 : /* lock target buffer first */
650 9766 : buffer = ReadBuffer(relation, targetBlock);
651 9766 : if (PageIsAllVisible(BufferGetPage(buffer)))
652 204 : visibilitymap_pin(relation, targetBlock, vmbuffer);
653 9766 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
654 9766 : LockBuffer(otherBuffer, BUFFER_LOCK_EXCLUSIVE);
655 : }
656 :
657 : /*
658 : * We now have the target page (and the other buffer, if any) pinned
659 : * and locked. However, since our initial PageIsAllVisible checks
660 : * were performed before acquiring the lock, the results might now be
661 : * out of date, either for the selected victim buffer, or for the
662 : * other buffer passed by the caller. In that case, we'll need to
663 : * give up our locks, go get the pin(s) we failed to get earlier, and
664 : * re-lock. That's pretty painful, but hopefully shouldn't happen
665 : * often.
666 : *
667 : * Note that there's a small possibility that we didn't pin the page
668 : * above but still have the correct page pinned anyway, either because
669 : * we've already made a previous pass through this loop, or because
670 : * caller passed us the right page anyway.
671 : *
672 : * Note also that it's possible that by the time we get the pin and
673 : * retake the buffer locks, the visibility map bit will have been
674 : * cleared by some other backend anyway. In that case, we'll have
675 : * done a bit of extra work for no gain, but there's no real harm
676 : * done.
677 : */
678 17758074 : GetVisibilityMapPins(relation, buffer, otherBuffer,
679 : targetBlock, otherBlock, vmbuffer,
680 : vmbuffer_other);
681 :
682 : /*
683 : * Now we can check to see if there's enough free space here. If so,
684 : * we're done.
685 : */
686 17758074 : page = BufferGetPage(buffer);
687 :
688 : /*
689 : * If necessary initialize page, it'll be used soon. We could avoid
690 : * dirtying the buffer here, and rely on the caller to do so whenever
691 : * it puts a tuple onto the page, but there seems not much benefit in
692 : * doing so.
693 : */
694 17758074 : if (PageIsNew(page))
695 : {
696 27836 : PageInit(page, BufferGetPageSize(buffer), 0);
697 27836 : MarkBufferDirty(buffer);
698 : }
699 :
700 17758074 : pageFreeSpace = PageGetHeapFreeSpace(page);
701 17758074 : if (targetFreeSpace <= pageFreeSpace)
702 : {
703 : /* use this page as future insert target, too */
704 17505898 : RelationSetTargetBlock(relation, targetBlock);
705 17505898 : return buffer;
706 : }
707 :
708 : /*
709 : * Not enough space, so we must give up our page locks and pin (if
710 : * any) and prepare to look elsewhere. We don't care which order we
711 : * unlock the two buffers in, so this can be slightly simpler than the
712 : * code above.
713 : */
714 252176 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
715 252176 : if (otherBuffer == InvalidBuffer)
716 242060 : ReleaseBuffer(buffer);
717 10116 : else if (otherBlock != targetBlock)
718 : {
719 7166 : LockBuffer(otherBuffer, BUFFER_LOCK_UNLOCK);
720 7166 : ReleaseBuffer(buffer);
721 : }
722 :
723 : /* Is there an ongoing bulk extension? */
724 252176 : if (bistate && bistate->next_free != InvalidBlockNumber)
725 : {
726 : Assert(bistate->next_free <= bistate->last_free);
727 :
728 : /*
729 : * We bulk extended the relation before, and there are still some
730 : * unused pages from that extension, so we don't need to look in
731 : * the FSM for a new page. But do record the free space from the
732 : * last page, somebody might insert narrower tuples later.
733 : */
734 24610 : if (use_fsm)
735 8720 : RecordPageWithFreeSpace(relation, targetBlock, pageFreeSpace);
736 :
737 24610 : targetBlock = bistate->next_free;
738 24610 : if (bistate->next_free >= bistate->last_free)
739 : {
740 1456 : bistate->next_free = InvalidBlockNumber;
741 1456 : bistate->last_free = InvalidBlockNumber;
742 : }
743 : else
744 23154 : bistate->next_free++;
745 : }
746 227566 : else if (!use_fsm)
747 : {
748 : /* Without FSM, always fall out of the loop and extend */
749 6246 : break;
750 : }
751 : else
752 : {
753 : /*
754 : * Update FSM as to condition of this page, and ask for another
755 : * page to try.
756 : */
757 221320 : targetBlock = RecordAndGetPageWithFreeSpace(relation,
758 : targetBlock,
759 : pageFreeSpace,
760 : targetFreeSpace);
761 : }
762 : }
763 :
764 : /* Have to extend the relation */
765 227960 : buffer = RelationAddBlocks(relation, bistate, num_pages, use_fsm,
766 : &unlockedTargetBuffer);
767 :
768 227960 : targetBlock = BufferGetBlockNumber(buffer);
769 227960 : page = BufferGetPage(buffer);
770 :
771 : /*
772 : * The page is empty, pin vmbuffer to set all_frozen bit. We don't want to
773 : * do IO while the buffer is locked, so we unlock the page first if IO is
774 : * needed (necessitating checks below).
775 : */
776 227960 : if (options & HEAP_INSERT_FROZEN)
777 : {
778 : Assert(PageGetMaxOffsetNumber(page) == 0);
779 :
780 606 : if (!visibilitymap_pin_ok(targetBlock, *vmbuffer))
781 : {
782 552 : if (!unlockedTargetBuffer)
783 552 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
784 552 : unlockedTargetBuffer = true;
785 552 : visibilitymap_pin(relation, targetBlock, vmbuffer);
786 : }
787 : }
788 :
789 : /*
790 : * Reacquire locks if necessary.
791 : *
792 : * If the target buffer was unlocked above, or is unlocked while
793 : * reacquiring the lock on otherBuffer below, it's unlikely, but possible,
794 : * that another backend used space on this page. We check for that below,
795 : * and retry if necessary.
796 : */
797 227960 : recheckVmPins = false;
798 227960 : if (unlockedTargetBuffer)
799 : {
800 : /* released lock on target buffer above */
801 1336 : if (otherBuffer != InvalidBuffer)
802 0 : LockBuffer(otherBuffer, BUFFER_LOCK_EXCLUSIVE);
803 1336 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
804 1336 : recheckVmPins = true;
805 : }
806 226624 : else if (otherBuffer != InvalidBuffer)
807 : {
808 : /*
809 : * We did not release the target buffer, and otherBuffer is valid,
810 : * need to lock the other buffer. It's guaranteed to be of a lower
811 : * page number than the new page. To conform with the deadlock
812 : * prevent rules, we ought to lock otherBuffer first, but that would
813 : * give other backends a chance to put tuples on our page. To reduce
814 : * the likelihood of that, attempt to lock the other buffer
815 : * conditionally, that's very likely to work.
816 : *
817 : * Alternatively, we could acquire the lock on otherBuffer before
818 : * extending the relation, but that'd require holding the lock while
819 : * performing IO, which seems worse than an unlikely retry.
820 : */
821 : Assert(otherBuffer != buffer);
822 : Assert(targetBlock > otherBlock);
823 :
824 6294 : if (unlikely(!ConditionalLockBuffer(otherBuffer)))
825 : {
826 2 : unlockedTargetBuffer = true;
827 2 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
828 2 : LockBuffer(otherBuffer, BUFFER_LOCK_EXCLUSIVE);
829 2 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
830 : }
831 6294 : recheckVmPins = true;
832 : }
833 :
834 : /*
835 : * If one of the buffers was unlocked (always the case if otherBuffer is
836 : * valid), it's possible, although unlikely, that an all-visible flag
837 : * became set. We can use GetVisibilityMapPins to deal with that. It's
838 : * possible that GetVisibilityMapPins() might need to temporarily release
839 : * buffer locks, in which case we'll need to check if there's still enough
840 : * space on the page below.
841 : */
842 227960 : if (recheckVmPins)
843 : {
844 7630 : if (GetVisibilityMapPins(relation, otherBuffer, buffer,
845 : otherBlock, targetBlock, vmbuffer_other,
846 : vmbuffer))
847 0 : unlockedTargetBuffer = true;
848 : }
849 :
850 : /*
851 : * If the target buffer was temporarily unlocked since the relation
852 : * extension, it's possible, although unlikely, that all the space on the
853 : * page was already used. If so, we just retry from the start. If we
854 : * didn't unlock, something has gone wrong if there's not enough space -
855 : * the test at the top should have prevented reaching this case.
856 : */
857 227960 : pageFreeSpace = PageGetHeapFreeSpace(page);
858 227960 : if (len > pageFreeSpace)
859 : {
860 0 : if (unlockedTargetBuffer)
861 : {
862 0 : if (otherBuffer != InvalidBuffer)
863 0 : LockBuffer(otherBuffer, BUFFER_LOCK_UNLOCK);
864 0 : UnlockReleaseBuffer(buffer);
865 :
866 0 : goto loop;
867 : }
868 0 : elog(PANIC, "tuple is too big: size %zu", len);
869 : }
870 :
871 : /*
872 : * Remember the new page as our target for future insertions.
873 : *
874 : * XXX should we enter the new page into the free space map immediately,
875 : * or just keep it for this backend's exclusive use in the short run
876 : * (until VACUUM sees it)? Seems to depend on whether you expect the
877 : * current backend to make more insertions or not, which is probably a
878 : * good bet most of the time. So for now, don't add it to FSM yet.
879 : */
880 227960 : RelationSetTargetBlock(relation, targetBlock);
881 :
882 227960 : return buffer;
883 : }
|
