Branch data Line data Source code
1 : : /*-------------------------------------------------------------------------
2 : : *
3 : : * visibilitymap.c
4 : : * bitmap for tracking visibility of heap tuples
5 : : *
6 : : * Portions Copyright (c) 1996-2026, PostgreSQL Global Development Group
7 : : * Portions Copyright (c) 1994, Regents of the University of California
8 : : *
9 : : *
10 : : * IDENTIFICATION
11 : : * src/backend/access/heap/visibilitymap.c
12 : : *
13 : : * INTERFACE ROUTINES
14 : : * visibilitymap_clear - clear bits for one page in the visibility map
15 : : * visibilitymap_pin - pin a map page for setting a bit
16 : : * visibilitymap_pin_ok - check whether correct map page is already pinned
17 : : * visibilitymap_set - set bit(s) in a previously pinned page
18 : : * visibilitymap_get_status - get status of bits
19 : : * visibilitymap_count - count number of bits set in visibility map
20 : : * visibilitymap_prepare_truncate -
21 : : * prepare for truncation of the visibility map
22 : : *
23 : : * NOTES
24 : : *
25 : : * The visibility map is a bitmap with two bits (all-visible and all-frozen)
26 : : * per heap page. A set all-visible bit means that all tuples on the page are
27 : : * known visible to all transactions, and therefore the page doesn't need to
28 : : * be vacuumed. A set all-frozen bit means that all tuples on the page are
29 : : * completely frozen, and therefore the page doesn't need to be vacuumed even
30 : : * if whole table scanning vacuum is required (e.g. anti-wraparound vacuum).
31 : : * The all-frozen bit must be set only when the page is already all-visible.
32 : : *
33 : : * The map is conservative in the sense that we make sure that whenever a bit
34 : : * is set, we know the condition is true, but if a bit is not set, it might or
35 : : * might not be true.
36 : : *
37 : : * Changes to the visibility map bits are not separately WAL-logged. Callers
38 : : * must make sure that whenever a visibility map bit is cleared, the bit is
39 : : * cleared on WAL replay of the updating operation. And whenever a visibility
40 : : * map bit is set, the bit is set on WAL replay of the operation that rendered
41 : : * the page all-visible/all-frozen.
42 : : *
43 : : * The visibility map bits operate as a hint in one direction: if they are
44 : : * clear, it may still be the case that every tuple on the page is visible to
45 : : * all transactions (we just don't know that for certain). However, if they
46 : : * are set, we may skip vacuuming pages and advance relfrozenxid or skip
47 : : * reading heap pages for an index-only scan. If they are incorrectly set,
48 : : * this can lead to data corruption and wrong results.
49 : : *
50 : : * Additionally, it is critical that the heap-page level PD_ALL_VISIBLE bit be
51 : : * correctly set and cleared along with the VM bits.
52 : : *
53 : : * When clearing the VM, if a crash occurs after the heap page makes it to
54 : : * disk but before the VM page makes it to disk, replay must clear the VM or
55 : : * the next index-only scan can return wrong results or vacuum may incorrectly
56 : : * advance relfrozenxid.
57 : : *
58 : : * When setting the VM, if a crash occurs after the visibility map page makes
59 : : * it to disk and before the updated heap page makes it to disk, redo must set
60 : : * the bit on the heap page. Otherwise, the next insert, update, or delete on
61 : : * the heap page will fail to realize that the visibility map bit must be
62 : : * cleared, possibly causing index-only scans to return wrong answers.
63 : : *
64 : : * VACUUM will normally skip pages for which the visibility map bit is set;
65 : : * such pages can't contain any dead tuples and therefore don't need vacuuming.
66 : : *
67 : : * LOCKING
68 : : *
69 : : * In heapam.c, whenever a page is modified so that not all tuples on the
70 : : * page are visible to everyone anymore, the corresponding bit in the
71 : : * visibility map is cleared. In order to be crash-safe, we need to do this
72 : : * while still holding a lock on the heap page and in the same critical
73 : : * section that logs the page modification. However, we don't want to hold
74 : : * the buffer lock over any I/O that may be required to read in the visibility
75 : : * map page. To avoid this, we examine the heap page before locking it;
76 : : * if the page-level PD_ALL_VISIBLE bit is set, we pin the visibility map
77 : : * bit. Then, we lock the buffer. But this creates a race condition: there
78 : : * is a possibility that in the time it takes to lock the buffer, the
79 : : * PD_ALL_VISIBLE bit gets set. If that happens, we have to unlock the
80 : : * buffer, pin the visibility map page, and relock the buffer. This shouldn't
81 : : * happen often, because only VACUUM currently sets visibility map bits,
82 : : * and the race will only occur if VACUUM processes a given page at almost
83 : : * exactly the same time that someone tries to further modify it.
84 : : *
85 : : * To set a bit, you need to hold a lock on the heap page. That prevents
86 : : * the race condition where VACUUM sees that all tuples on the page are
87 : : * visible to everyone, but another backend modifies the page before VACUUM
88 : : * sets the bit in the visibility map.
89 : : *
90 : : * When a bit is set, the LSN of the visibility map page is updated to make
91 : : * sure that the visibility map update doesn't get written to disk before the
92 : : * WAL record of the changes that made it possible to set the bit is flushed.
93 : : * But when a bit is cleared, we don't have to do that because it's always
94 : : * safe to clear a bit in the map from correctness point of view.
95 : : *
96 : : *-------------------------------------------------------------------------
97 : : */
98 : : #include "postgres.h"
99 : :
100 : : #include "access/heapam_xlog.h"
101 : : #include "access/visibilitymap.h"
102 : : #include "access/xloginsert.h"
103 : : #include "access/xlogutils.h"
104 : : #include "miscadmin.h"
105 : : #include "port/pg_bitutils.h"
106 : : #include "storage/bufmgr.h"
107 : : #include "storage/smgr.h"
108 : : #include "utils/inval.h"
109 : : #include "utils/rel.h"
110 : :
111 : :
112 : : /*#define TRACE_VISIBILITYMAP */
113 : :
114 : : /*
115 : : * Size of the bitmap on each visibility map page, in bytes. There's no
116 : : * extra headers, so the whole page minus the standard page header is
117 : : * used for the bitmap.
118 : : */
119 : : #define MAPSIZE (BLCKSZ - MAXALIGN(SizeOfPageHeaderData))
120 : :
121 : : /* Number of heap blocks we can represent in one byte */
122 : : #define HEAPBLOCKS_PER_BYTE (BITS_PER_BYTE / BITS_PER_HEAPBLOCK)
123 : :
124 : : /* Number of heap blocks we can represent in one visibility map page. */
125 : : #define HEAPBLOCKS_PER_PAGE (MAPSIZE * HEAPBLOCKS_PER_BYTE)
126 : :
127 : : /* Mapping from heap block number to the right bit in the visibility map */
128 : : #define HEAPBLK_TO_MAPBLOCK(x) ((x) / HEAPBLOCKS_PER_PAGE)
129 : : #define HEAPBLK_TO_MAPBLOCK_LIMIT(x) \
130 : : (((x) + HEAPBLOCKS_PER_PAGE - 1) / HEAPBLOCKS_PER_PAGE)
131 : : #define HEAPBLK_TO_MAPBYTE(x) (((x) % HEAPBLOCKS_PER_PAGE) / HEAPBLOCKS_PER_BYTE)
132 : : #define HEAPBLK_TO_OFFSET(x) (((x) % HEAPBLOCKS_PER_BYTE) * BITS_PER_HEAPBLOCK)
133 : :
134 : : /* Masks for counting subsets of bits in the visibility map. */
135 : : #define VISIBLE_MASK8 (0x55) /* The lower bit of each bit pair */
136 : : #define FROZEN_MASK8 (0xaa) /* The upper bit of each bit pair */
137 : :
138 : : /* prototypes for internal routines */
139 : : static Buffer vm_readbuf(Relation rel, BlockNumber blkno, bool extend);
140 : : static Buffer vm_extend(Relation rel, BlockNumber vm_nblocks);
141 : :
142 : : /*
143 : : * visibilitymap_clear - clear specified bits for one page in visibility map
144 : : *
145 : : * You must pass a buffer containing the correct map page to this function,
146 : : * which already needs to be pinned and locked exclusively.
147 : : *
148 : : * This function doesn't do any I/O. Returns true if any bits have been
149 : : * cleared and false otherwise.
150 : : */
151 : : bool
152 : 26180 : visibilitymap_clear(RelFileLocator rlocator, BlockNumber heapBlk,
153 : : Buffer vmbuf, uint8 flags)
154 : : {
155 : 26180 : int mapByte = HEAPBLK_TO_MAPBYTE(heapBlk);
156 : 26180 : int mapOffset = HEAPBLK_TO_OFFSET(heapBlk);
157 : 26180 : BlockNumber mapBlock = HEAPBLK_TO_MAPBLOCK(heapBlk);
158 : 26180 : uint8 mask = flags << mapOffset;
159 : : Page page;
160 : : char *map;
161 : 26180 : bool cleared = false;
162 : :
163 : : /* Must never clear all_visible bit while leaving all_frozen bit set */
164 : : Assert(flags & VISIBILITYMAP_VALID_BITS);
165 : : Assert(flags != VISIBILITYMAP_ALL_VISIBLE);
166 : :
167 : : #ifdef TRACE_VISIBILITYMAP
168 : : elog(DEBUG1, "vm_clear %s %d",
169 : : relpathbackend(rlocator, MyProcNumber, MAIN_FORKNUM).str,
170 : : heapBlk);
171 : : #endif
172 : :
173 [ + - - + ]: 26180 : if (!BufferIsValid(vmbuf) || BufferGetBlockNumber(vmbuf) != mapBlock)
174 [ # # ]: 0 : elog(ERROR, "wrong buffer passed to visibilitymap_clear");
175 : :
176 : : Assert(BufferIsLockedByMeInMode(vmbuf, BUFFER_LOCK_EXCLUSIVE));
177 : :
178 : 26180 : page = BufferGetPage(vmbuf);
179 : 26180 : map = PageGetContents(page);
180 : :
181 [ + + ]: 26180 : if (map[mapByte] & mask)
182 : : {
183 : 21981 : map[mapByte] &= ~mask;
184 : :
185 : 21981 : MarkBufferDirty(vmbuf);
186 : 21981 : cleared = true;
187 : : }
188 : :
189 : 26180 : return cleared;
190 : : }
191 : :
192 : : /*
193 : : * visibilitymap_pin - pin a map page for setting a bit
194 : : *
195 : : * Setting a bit in the visibility map is a two-phase operation. First, call
196 : : * visibilitymap_pin, to pin the visibility map page containing the bit for
197 : : * the heap page. Because that can require I/O to read the map page, you
198 : : * shouldn't hold a lock on the heap page while doing that. Then, call
199 : : * visibilitymap_set to actually set the bit.
200 : : *
201 : : * On entry, *vmbuf should be InvalidBuffer or a valid buffer returned by
202 : : * an earlier call to visibilitymap_pin or visibilitymap_get_status on the same
203 : : * relation. On return, *vmbuf is a valid buffer with the map page containing
204 : : * the bit for heapBlk.
205 : : *
206 : : * If the page doesn't exist in the map file yet, it is extended.
207 : : */
208 : : void
209 : 1077084 : visibilitymap_pin(Relation rel, BlockNumber heapBlk, Buffer *vmbuf)
210 : : {
211 : 1077084 : BlockNumber mapBlock = HEAPBLK_TO_MAPBLOCK(heapBlk);
212 : :
213 : : /* Reuse the old pinned buffer if possible */
214 [ + + ]: 1077084 : if (BufferIsValid(*vmbuf))
215 : : {
216 [ + - ]: 532483 : if (BufferGetBlockNumber(*vmbuf) == mapBlock)
217 : 532483 : return;
218 : :
219 : 0 : ReleaseBuffer(*vmbuf);
220 : : }
221 : 544601 : *vmbuf = vm_readbuf(rel, mapBlock, true);
222 : : }
223 : :
224 : : /*
225 : : * visibilitymap_pin_ok - do we already have the correct page pinned?
226 : : *
227 : : * On entry, vmbuf should be InvalidBuffer or a valid buffer returned by
228 : : * an earlier call to visibilitymap_pin or visibilitymap_get_status on the same
229 : : * relation. The return value indicates whether the buffer covers the
230 : : * given heapBlk.
231 : : */
232 : : bool
233 : 18729 : visibilitymap_pin_ok(BlockNumber heapBlk, Buffer vmbuf)
234 : : {
235 : 18729 : BlockNumber mapBlock = HEAPBLK_TO_MAPBLOCK(heapBlk);
236 : :
237 [ + + + - ]: 18729 : return BufferIsValid(vmbuf) && BufferGetBlockNumber(vmbuf) == mapBlock;
238 : : }
239 : :
240 : : /*
241 : : * Set VM (visibility map) flags in the VM block in vmBuf.
242 : : *
243 : : * This function is intended for callers that log VM changes together
244 : : * with the heap page modifications that rendered the page all-visible.
245 : : *
246 : : * vmBuf must be pinned and exclusively locked, and it must cover the VM bits
247 : : * corresponding to heapBlk.
248 : : *
249 : : * In normal operation (not recovery), this must be called inside a critical
250 : : * section that also applies the necessary heap page changes and, if
251 : : * applicable, emits WAL.
252 : : *
253 : : * The caller is responsible for ensuring consistency between the heap page
254 : : * and the VM page by holding a pin and exclusive lock on the buffer
255 : : * containing heapBlk.
256 : : *
257 : : * rlocator is used only for debugging messages.
258 : : */
259 : : void
260 : 91750 : visibilitymap_set(BlockNumber heapBlk,
261 : : Buffer vmBuf, uint8 flags,
262 : : RelFileLocator rlocator)
263 : : {
264 : 91750 : BlockNumber mapBlock = HEAPBLK_TO_MAPBLOCK(heapBlk);
265 : 91750 : uint32 mapByte = HEAPBLK_TO_MAPBYTE(heapBlk);
266 : 91750 : uint8 mapOffset = HEAPBLK_TO_OFFSET(heapBlk);
267 : : Page page;
268 : : uint8 *map;
269 : : uint8 status;
270 : :
271 : : #ifdef TRACE_VISIBILITYMAP
272 : : elog(DEBUG1, "vm_set flags 0x%02X for %s %d",
273 : : flags,
274 : : relpathbackend(rlocator, MyProcNumber, MAIN_FORKNUM).str,
275 : : heapBlk);
276 : : #endif
277 : :
278 : : /* Call in same critical section where WAL is emitted. */
279 : : Assert(InRecovery || CritSectionCount > 0);
280 : :
281 : : /* Flags should be valid. Also never clear bits with this function */
282 : : Assert((flags & VISIBILITYMAP_VALID_BITS) == flags);
283 : :
284 : : /* Must never set all_frozen bit without also setting all_visible bit */
285 : : Assert(flags != VISIBILITYMAP_ALL_FROZEN);
286 : :
287 : : /* Check that we have the right VM page pinned */
288 [ + - - + ]: 91750 : if (!BufferIsValid(vmBuf) || BufferGetBlockNumber(vmBuf) != mapBlock)
289 [ # # ]: 0 : elog(ERROR, "wrong VM buffer passed to visibilitymap_set");
290 : :
291 : : Assert(BufferIsLockedByMeInMode(vmBuf, BUFFER_LOCK_EXCLUSIVE));
292 : :
293 : 91750 : page = BufferGetPage(vmBuf);
294 : 91750 : map = (uint8 *) PageGetContents(page);
295 : :
296 : 91750 : status = (map[mapByte] >> mapOffset) & VISIBILITYMAP_VALID_BITS;
297 [ + - ]: 91750 : if (flags != status)
298 : : {
299 : 91750 : map[mapByte] |= (flags << mapOffset);
300 : 91750 : MarkBufferDirty(vmBuf);
301 : : }
302 : 91750 : }
303 : :
304 : : /*
305 : : * visibilitymap_get_status - get status of bits
306 : : *
307 : : * Are all tuples on heapBlk visible to all or are marked frozen, according
308 : : * to the visibility map?
309 : : *
310 : : * On entry, *vmbuf should be InvalidBuffer or a valid buffer returned by an
311 : : * earlier call to visibilitymap_pin or visibilitymap_get_status on the same
312 : : * relation. On return, *vmbuf is a valid buffer with the map page containing
313 : : * the bit for heapBlk, or InvalidBuffer. The caller is responsible for
314 : : * releasing *vmbuf after it's done testing and setting bits.
315 : : *
316 : : * NOTE: This function is typically called without a lock on the heap page,
317 : : * so somebody else could change the bit just after we look at it. In fact,
318 : : * since we don't lock the visibility map page either, it's even possible that
319 : : * someone else could have changed the bit just before we look at it, but yet
320 : : * we might see the old value. It is the caller's responsibility to deal with
321 : : * all concurrency issues! In practice it can't be stale enough to matter for
322 : : * the primary use case: index-only scans that check whether a heap fetch can
323 : : * be skipped.
324 : : *
325 : : * The argument for why it can't be stale enough to matter for the primary use
326 : : * case is as follows:
327 : : *
328 : : * Inserts: we need to detect that a VM bit was cleared by an insert right
329 : : * away, because the new tuple is present in the index but not yet visible.
330 : : * Reading the TID from the index page (under a shared lock on the index
331 : : * buffer) is serialized with the insertion of the TID into the index (under
332 : : * an exclusive lock on the same index buffer). Because the VM bit is cleared
333 : : * before the index is updated, and locking/unlocking of the index page acts
334 : : * as a full memory barrier, we are sure to see the cleared bit whenever we
335 : : * see a recently-inserted TID.
336 : : *
337 : : * Deletes: the clearing of the VM bit by a delete is NOT serialized with the
338 : : * index page access, because deletes do not update the index page (only
339 : : * VACUUM removes the index TID). So we may see a significantly stale value.
340 : : * However, we don't need to detect the delete right away, because the tuple
341 : : * remains visible until the deleting transaction commits or the statement
342 : : * ends (if it's our own transaction). In either case, the lock on the VM
343 : : * buffer will have been released (acting as a write barrier) after clearing
344 : : * the bit. And for us to have a snapshot that includes the deleting
345 : : * transaction (making the tuple invisible), we must have acquired
346 : : * ProcArrayLock after that time, acting as a read barrier.
347 : : */
348 : : uint8
349 : 5458320 : visibilitymap_get_status(Relation rel, BlockNumber heapBlk, Buffer *vmbuf)
350 : : {
351 : 5458320 : BlockNumber mapBlock = HEAPBLK_TO_MAPBLOCK(heapBlk);
352 : 5458320 : uint32 mapByte = HEAPBLK_TO_MAPBYTE(heapBlk);
353 : 5458320 : uint8 mapOffset = HEAPBLK_TO_OFFSET(heapBlk);
354 : : char *map;
355 : : uint8 result;
356 : :
357 : : #ifdef TRACE_VISIBILITYMAP
358 : : elog(DEBUG1, "vm_get_status %s %d", RelationGetRelationName(rel), heapBlk);
359 : : #endif
360 : :
361 : : /* Reuse the old pinned buffer if possible */
362 [ + + ]: 5458320 : if (BufferIsValid(*vmbuf))
363 : : {
364 [ - + ]: 5277287 : if (BufferGetBlockNumber(*vmbuf) != mapBlock)
365 : : {
366 : 0 : ReleaseBuffer(*vmbuf);
367 : 0 : *vmbuf = InvalidBuffer;
368 : : }
369 : : }
370 : :
371 [ + + ]: 5458320 : if (!BufferIsValid(*vmbuf))
372 : : {
373 : 181033 : *vmbuf = vm_readbuf(rel, mapBlock, false);
374 [ + + ]: 181033 : if (!BufferIsValid(*vmbuf))
375 : 55500 : return (uint8) 0;
376 : : }
377 : :
378 : 5402820 : map = PageGetContents(BufferGetPage(*vmbuf));
379 : :
380 : : /*
381 : : * A single byte read is atomic. There could be memory-ordering effects
382 : : * here, but for performance reasons we make it the caller's job to worry
383 : : * about that.
384 : : */
385 : 5402820 : result = ((map[mapByte] >> mapOffset) & VISIBILITYMAP_VALID_BITS);
386 : 5402820 : return result;
387 : : }
388 : :
389 : : /*
390 : : * visibilitymap_count - count number of bits set in visibility map
391 : : *
392 : : * Note: we ignore the possibility of race conditions when the table is being
393 : : * extended concurrently with the call. New pages added to the table aren't
394 : : * going to be marked all-visible or all-frozen, so they won't affect the result.
395 : : */
396 : : void
397 : 149290 : visibilitymap_count(Relation rel, BlockNumber *all_visible, BlockNumber *all_frozen)
398 : : {
399 : : BlockNumber mapBlock;
400 : 149290 : BlockNumber nvisible = 0;
401 : 149290 : BlockNumber nfrozen = 0;
402 : :
403 : : /* all_visible must be specified */
404 : : Assert(all_visible);
405 : :
406 : 149290 : for (mapBlock = 0;; mapBlock++)
407 : 53711 : {
408 : : Buffer mapBuffer;
409 : : uint64 *map;
410 : :
411 : : /*
412 : : * Read till we fall off the end of the map. We assume that any extra
413 : : * bytes in the last page are zeroed, so we don't bother excluding
414 : : * them from the count.
415 : : */
416 : 203001 : mapBuffer = vm_readbuf(rel, mapBlock, false);
417 [ + + ]: 203001 : if (!BufferIsValid(mapBuffer))
418 : 149290 : break;
419 : :
420 : : /*
421 : : * We choose not to lock the page, since the result is going to be
422 : : * immediately stale anyway if anyone is concurrently setting or
423 : : * clearing bits, and we only really need an approximate value.
424 : : */
425 : 53711 : map = (uint64 *) PageGetContents(BufferGetPage(mapBuffer));
426 : :
427 : 53711 : nvisible += pg_popcount_masked((const char *) map, MAPSIZE, VISIBLE_MASK8);
428 [ + - ]: 53711 : if (all_frozen)
429 : 53711 : nfrozen += pg_popcount_masked((const char *) map, MAPSIZE, FROZEN_MASK8);
430 : :
431 : 53711 : ReleaseBuffer(mapBuffer);
432 : : }
433 : :
434 : 149290 : *all_visible = nvisible;
435 [ + - ]: 149290 : if (all_frozen)
436 : 149290 : *all_frozen = nfrozen;
437 : 149290 : }
438 : :
439 : : /*
440 : : * visibilitymap_prepare_truncate -
441 : : * prepare for truncation of the visibility map
442 : : *
443 : : * nheapblocks is the new size of the heap.
444 : : *
445 : : * Return the number of blocks of new visibility map.
446 : : * If it's InvalidBlockNumber, there is nothing to truncate;
447 : : * otherwise the caller is responsible for calling smgrtruncate()
448 : : * to truncate the visibility map pages.
449 : : */
450 : : BlockNumber
451 : 250 : visibilitymap_prepare_truncate(Relation rel, BlockNumber nheapblocks)
452 : : {
453 : : BlockNumber newnblocks;
454 : :
455 : : /* last remaining block, byte, and bit */
456 : 250 : BlockNumber truncBlock = HEAPBLK_TO_MAPBLOCK(nheapblocks);
457 : 250 : uint32 truncByte = HEAPBLK_TO_MAPBYTE(nheapblocks);
458 : 250 : uint8 truncOffset = HEAPBLK_TO_OFFSET(nheapblocks);
459 : :
460 : : #ifdef TRACE_VISIBILITYMAP
461 : : elog(DEBUG1, "vm_truncate %s %d", RelationGetRelationName(rel), nheapblocks);
462 : : #endif
463 : :
464 : : /*
465 : : * If no visibility map has been created yet for this relation, there's
466 : : * nothing to truncate.
467 : : */
468 [ - + ]: 250 : if (!smgrexists(RelationGetSmgr(rel), VISIBILITYMAP_FORKNUM))
469 : 0 : return InvalidBlockNumber;
470 : :
471 : : /*
472 : : * Unless the new size is exactly at a visibility map page boundary, the
473 : : * tail bits in the last remaining map page, representing truncated heap
474 : : * blocks, need to be cleared. This is not only tidy, but also necessary
475 : : * because we don't get a chance to clear the bits if the heap is extended
476 : : * again.
477 : : */
478 [ + + + + ]: 250 : if (truncByte != 0 || truncOffset != 0)
479 : 155 : {
480 : : Buffer mapBuffer;
481 : : Page page;
482 : : char *map;
483 : :
484 : 155 : newnblocks = truncBlock + 1;
485 : :
486 : 155 : mapBuffer = vm_readbuf(rel, truncBlock, false);
487 [ - + ]: 155 : if (!BufferIsValid(mapBuffer))
488 : : {
489 : : /* nothing to do, the file was already smaller */
490 : 0 : return InvalidBlockNumber;
491 : : }
492 : :
493 : 155 : page = BufferGetPage(mapBuffer);
494 : 155 : map = PageGetContents(page);
495 : :
496 : 155 : LockBuffer(mapBuffer, BUFFER_LOCK_EXCLUSIVE);
497 : :
498 : : /* NO EREPORT(ERROR) from here till changes are logged */
499 : 155 : START_CRIT_SECTION();
500 : :
501 : : /* Clear out the unwanted bytes. */
502 [ + + + - : 155 : MemSet(&map[truncByte + 1], 0, MAPSIZE - (truncByte + 1));
+ - - + -
- ]
503 : :
504 : : /*----
505 : : * Mask out the unwanted bits of the last remaining byte.
506 : : *
507 : : * ((1 << 0) - 1) = 00000000
508 : : * ((1 << 1) - 1) = 00000001
509 : : * ...
510 : : * ((1 << 6) - 1) = 00111111
511 : : * ((1 << 7) - 1) = 01111111
512 : : *----
513 : : */
514 : 155 : map[truncByte] &= (1 << truncOffset) - 1;
515 : :
516 : : /*
517 : : * Truncation of a relation is WAL-logged at a higher-level, and we
518 : : * will be called at WAL replay. But if checksums are enabled, we need
519 : : * to still write a WAL record to protect against a torn page, if the
520 : : * page is flushed to disk before the truncation WAL record. We cannot
521 : : * use MarkBufferDirtyHint here, because that will not dirty the page
522 : : * during recovery.
523 : : */
524 : 155 : MarkBufferDirty(mapBuffer);
525 [ + + + + : 155 : if (!InRecovery && RelationNeedsWAL(rel) && XLogHintBitIsNeeded())
+ + + - +
- + + +
- ]
526 : 136 : log_newpage_buffer(mapBuffer, false);
527 : :
528 : 155 : END_CRIT_SECTION();
529 : :
530 : 155 : UnlockReleaseBuffer(mapBuffer);
531 : : }
532 : : else
533 : 95 : newnblocks = truncBlock;
534 : :
535 [ + + ]: 250 : if (smgrnblocks(RelationGetSmgr(rel), VISIBILITYMAP_FORKNUM) <= newnblocks)
536 : : {
537 : : /* nothing to do, the file was already smaller than requested size */
538 : 155 : return InvalidBlockNumber;
539 : : }
540 : :
541 : 95 : return newnblocks;
542 : : }
543 : :
544 : : /*
545 : : * visibilitymap_truncation_length -
546 : : * compute truncation length for visibility map
547 : : *
548 : : * Given a proposed truncation length for the main fork, compute the
549 : : * correct truncation length for the visibility map. Should return the
550 : : * same answer as visibilitymap_prepare_truncate(), but without modifying
551 : : * anything.
552 : : */
553 : : BlockNumber
554 : 1 : visibilitymap_truncation_length(BlockNumber nheapblocks)
555 : : {
556 : 1 : return HEAPBLK_TO_MAPBLOCK_LIMIT(nheapblocks);
557 : : }
558 : :
559 : : /*
560 : : * Read a visibility map page.
561 : : *
562 : : * If the page doesn't exist, InvalidBuffer is returned, or if 'extend' is
563 : : * true, the visibility map file is extended.
564 : : */
565 : : static Buffer
566 : 928790 : vm_readbuf(Relation rel, BlockNumber blkno, bool extend)
567 : : {
568 : : Buffer buf;
569 : : SMgrRelation reln;
570 : :
571 : : /*
572 : : * Caution: re-using this smgr pointer could fail if the relcache entry
573 : : * gets closed. It's safe as long as we only do smgr-level operations
574 : : * between here and the last use of the pointer.
575 : : */
576 : 928790 : reln = RelationGetSmgr(rel);
577 : :
578 : : /*
579 : : * If we haven't cached the size of the visibility map fork yet, check it
580 : : * first.
581 : : */
582 [ + + ]: 928790 : if (reln->smgr_cached_nblocks[VISIBILITYMAP_FORKNUM] == InvalidBlockNumber)
583 : : {
584 [ + + ]: 170187 : if (smgrexists(reln, VISIBILITYMAP_FORKNUM))
585 : 71681 : smgrnblocks(reln, VISIBILITYMAP_FORKNUM);
586 : : else
587 : 98506 : reln->smgr_cached_nblocks[VISIBILITYMAP_FORKNUM] = 0;
588 : : }
589 : :
590 : : /*
591 : : * For reading we use ZERO_ON_ERROR mode, and initialize the page if
592 : : * necessary. It's always safe to clear bits, so it's better to clear
593 : : * corrupt pages than error out.
594 : : *
595 : : * We use the same path below to initialize pages when extending the
596 : : * relation, as a concurrent extension can end up with vm_extend()
597 : : * returning an already-initialized page.
598 : : */
599 [ + + ]: 928790 : if (blkno >= reln->smgr_cached_nblocks[VISIBILITYMAP_FORKNUM])
600 : : {
601 [ + + ]: 209367 : if (extend)
602 : 4577 : buf = vm_extend(rel, blkno + 1);
603 : : else
604 : 204790 : return InvalidBuffer;
605 : : }
606 : : else
607 : 719423 : buf = ReadBufferExtended(rel, VISIBILITYMAP_FORKNUM, blkno,
608 : : RBM_ZERO_ON_ERROR, NULL);
609 : :
610 : : /*
611 : : * Initializing the page when needed is trickier than it looks, because of
612 : : * the possibility of multiple backends doing this concurrently, and our
613 : : * desire to not uselessly take the buffer lock in the normal path where
614 : : * the page is OK. We must take the lock to initialize the page, so
615 : : * recheck page newness after we have the lock, in case someone else
616 : : * already did it. Also, because we initially check PageIsNew with no
617 : : * lock, it's possible to fall through and return the buffer while someone
618 : : * else is still initializing the page (i.e., we might see pd_upper as set
619 : : * but other page header fields are still zeroes). This is harmless for
620 : : * callers that will take a buffer lock themselves, but some callers
621 : : * inspect the page without any lock at all. The latter is OK only so
622 : : * long as it doesn't depend on the page header having correct contents.
623 : : * Current usage is safe because PageGetContents() does not require that.
624 : : */
625 [ + + ]: 724000 : if (PageIsNew(BufferGetPage(buf)))
626 : : {
627 : 4957 : LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
628 [ + - ]: 4957 : if (PageIsNew(BufferGetPage(buf)))
629 : 4957 : PageInit(BufferGetPage(buf), BLCKSZ, 0);
630 : 4957 : LockBuffer(buf, BUFFER_LOCK_UNLOCK);
631 : : }
632 : 724000 : return buf;
633 : : }
634 : :
635 : : /*
636 : : * Ensure that the visibility map fork is at least vm_nblocks long, extending
637 : : * it if necessary with zeroed pages.
638 : : */
639 : : static Buffer
640 : 4577 : vm_extend(Relation rel, BlockNumber vm_nblocks)
641 : : {
642 : : Buffer buf;
643 : :
644 : 4577 : buf = ExtendBufferedRelTo(BMR_REL(rel), VISIBILITYMAP_FORKNUM, NULL,
645 : : EB_CREATE_FORK_IF_NEEDED |
646 : : EB_CLEAR_SIZE_CACHE,
647 : : vm_nblocks,
648 : : RBM_ZERO_ON_ERROR);
649 : :
650 : : /*
651 : : * Send a shared-inval message to force other backends to close any smgr
652 : : * references they may have for this rel, which we are about to change.
653 : : * This is a useful optimization because it means that backends don't have
654 : : * to keep checking for creation or extension of the file, which happens
655 : : * infrequently.
656 : : */
657 : 4577 : CacheInvalidateSmgr(RelationGetSmgr(rel)->smgr_rlocator);
658 : :
659 : 4577 : return buf;
660 : : }
|
