Age Owner Branch data TLA Line data Source code
1 : : /*-------------------------------------------------------------------------
2 : : *
3 : : * visibilitymap.c
4 : : * bitmap for tracking visibility of heap tuples
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/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 a bit 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 : : * Clearing visibility map bits is not separately WAL-logged. The callers
38 : : * must make sure that whenever a bit is cleared, the bit is cleared on WAL
39 : : * replay of the updating operation as well.
40 : : *
41 : : * When we *set* a visibility map during VACUUM, we must write WAL. This may
42 : : * seem counterintuitive, since the bit is basically a hint: if it is clear,
43 : : * it may still be the case that every tuple on the page is visible to all
44 : : * transactions; we just don't know that for certain. The difficulty is that
45 : : * there are two bits which are typically set together: the PD_ALL_VISIBLE bit
46 : : * on the page itself, and the visibility map bit. If a crash occurs after the
47 : : * visibility map page makes it to disk and before the updated heap page makes
48 : : * it to disk, redo must set the bit on the heap page. Otherwise, the next
49 : : * insert, update, or delete on the heap page will fail to realize that the
50 : : * visibility map bit must be cleared, possibly causing index-only scans to
51 : : * return wrong answers.
52 : : *
53 : : * VACUUM will normally skip pages for which the visibility map bit is set;
54 : : * such pages can't contain any dead tuples and therefore don't need vacuuming.
55 : : *
56 : : * LOCKING
57 : : *
58 : : * In heapam.c, whenever a page is modified so that not all tuples on the
59 : : * page are visible to everyone anymore, the corresponding bit in the
60 : : * visibility map is cleared. In order to be crash-safe, we need to do this
61 : : * while still holding a lock on the heap page and in the same critical
62 : : * section that logs the page modification. However, we don't want to hold
63 : : * the buffer lock over any I/O that may be required to read in the visibility
64 : : * map page. To avoid this, we examine the heap page before locking it;
65 : : * if the page-level PD_ALL_VISIBLE bit is set, we pin the visibility map
66 : : * bit. Then, we lock the buffer. But this creates a race condition: there
67 : : * is a possibility that in the time it takes to lock the buffer, the
68 : : * PD_ALL_VISIBLE bit gets set. If that happens, we have to unlock the
69 : : * buffer, pin the visibility map page, and relock the buffer. This shouldn't
70 : : * happen often, because only VACUUM currently sets visibility map bits,
71 : : * and the race will only occur if VACUUM processes a given page at almost
72 : : * exactly the same time that someone tries to further modify it.
73 : : *
74 : : * To set a bit, you need to hold a lock on the heap page. That prevents
75 : : * the race condition where VACUUM sees that all tuples on the page are
76 : : * visible to everyone, but another backend modifies the page before VACUUM
77 : : * sets the bit in the visibility map.
78 : : *
79 : : * When a bit is set, the LSN of the visibility map page is updated to make
80 : : * sure that the visibility map update doesn't get written to disk before the
81 : : * WAL record of the changes that made it possible to set the bit is flushed.
82 : : * But when a bit is cleared, we don't have to do that because it's always
83 : : * safe to clear a bit in the map from correctness point of view.
84 : : *
85 : : *-------------------------------------------------------------------------
86 : : */
87 : : #include "postgres.h"
88 : :
89 : : #include "access/heapam_xlog.h"
90 : : #include "access/visibilitymap.h"
91 : : #include "access/xloginsert.h"
92 : : #include "access/xlogutils.h"
93 : : #include "miscadmin.h"
94 : : #include "port/pg_bitutils.h"
95 : : #include "storage/bufmgr.h"
96 : : #include "storage/smgr.h"
97 : : #include "utils/inval.h"
98 : : #include "utils/rel.h"
99 : :
100 : :
101 : : /*#define TRACE_VISIBILITYMAP */
102 : :
103 : : /*
104 : : * Size of the bitmap on each visibility map page, in bytes. There's no
105 : : * extra headers, so the whole page minus the standard page header is
106 : : * used for the bitmap.
107 : : */
108 : : #define MAPSIZE (BLCKSZ - MAXALIGN(SizeOfPageHeaderData))
109 : :
110 : : /* Number of heap blocks we can represent in one byte */
111 : : #define HEAPBLOCKS_PER_BYTE (BITS_PER_BYTE / BITS_PER_HEAPBLOCK)
112 : :
113 : : /* Number of heap blocks we can represent in one visibility map page. */
114 : : #define HEAPBLOCKS_PER_PAGE (MAPSIZE * HEAPBLOCKS_PER_BYTE)
115 : :
116 : : /* Mapping from heap block number to the right bit in the visibility map */
117 : : #define HEAPBLK_TO_MAPBLOCK(x) ((x) / HEAPBLOCKS_PER_PAGE)
118 : : #define HEAPBLK_TO_MAPBYTE(x) (((x) % HEAPBLOCKS_PER_PAGE) / HEAPBLOCKS_PER_BYTE)
119 : : #define HEAPBLK_TO_OFFSET(x) (((x) % HEAPBLOCKS_PER_BYTE) * BITS_PER_HEAPBLOCK)
120 : :
121 : : /* Masks for counting subsets of bits in the visibility map. */
122 : : #define VISIBLE_MASK8 (0x55) /* The lower bit of each bit pair */
123 : : #define FROZEN_MASK8 (0xaa) /* The upper bit of each bit pair */
124 : :
125 : : /* prototypes for internal routines */
126 : : static Buffer vm_readbuf(Relation rel, BlockNumber blkno, bool extend);
127 : : static Buffer vm_extend(Relation rel, BlockNumber vm_nblocks);
128 : :
129 : :
130 : : /*
131 : : * visibilitymap_clear - clear specified bits for one page in visibility map
132 : : *
133 : : * You must pass a buffer containing the correct map page to this function.
134 : : * Call visibilitymap_pin first to pin the right one. This function doesn't do
135 : : * any I/O. Returns true if any bits have been cleared and false otherwise.
136 : : */
137 : : bool
1083 pg@bowt.ie 138 :CBC 19227 : visibilitymap_clear(Relation rel, BlockNumber heapBlk, Buffer vmbuf, uint8 flags)
139 : : {
6121 heikki.linnakangas@i 140 : 19227 : BlockNumber mapBlock = HEAPBLK_TO_MAPBLOCK(heapBlk);
141 : 19227 : int mapByte = HEAPBLK_TO_MAPBYTE(heapBlk);
3382 rhaas@postgresql.org 142 : 19227 : int mapOffset = HEAPBLK_TO_OFFSET(heapBlk);
3337 andres@anarazel.de 143 : 19227 : uint8 mask = flags << mapOffset;
144 : : char *map;
145 : 19227 : bool cleared = false;
146 : :
147 : : /* Must never clear all_visible bit while leaving all_frozen bit set */
148 [ - + ]: 19227 : Assert(flags & VISIBILITYMAP_VALID_BITS);
964 pg@bowt.ie 149 [ - + ]: 19227 : Assert(flags != VISIBILITYMAP_ALL_VISIBLE);
150 : :
151 : : #ifdef TRACE_VISIBILITYMAP
152 : : elog(DEBUG1, "vm_clear %s %d", RelationGetRelationName(rel), heapBlk);
153 : : #endif
154 : :
1083 155 [ + - - + ]: 19227 : if (!BufferIsValid(vmbuf) || BufferGetBlockNumber(vmbuf) != mapBlock)
5191 rhaas@postgresql.org 156 [ # # ]:UBC 0 : elog(ERROR, "wrong buffer passed to visibilitymap_clear");
157 : :
1083 pg@bowt.ie 158 :CBC 19227 : LockBuffer(vmbuf, BUFFER_LOCK_EXCLUSIVE);
159 : 19227 : map = PageGetContents(BufferGetPage(vmbuf));
160 : :
6121 heikki.linnakangas@i 161 [ + + ]: 19227 : if (map[mapByte] & mask)
162 : : {
163 : 17207 : map[mapByte] &= ~mask;
164 : :
1083 pg@bowt.ie 165 : 17207 : MarkBufferDirty(vmbuf);
3337 andres@anarazel.de 166 : 17207 : cleared = true;
167 : : }
168 : :
1083 pg@bowt.ie 169 : 19227 : LockBuffer(vmbuf, BUFFER_LOCK_UNLOCK);
170 : :
3337 andres@anarazel.de 171 : 19227 : return cleared;
172 : : }
173 : :
174 : : /*
175 : : * visibilitymap_pin - pin a map page for setting a bit
176 : : *
177 : : * Setting a bit in the visibility map is a two-phase operation. First, call
178 : : * visibilitymap_pin, to pin the visibility map page containing the bit for
179 : : * the heap page. Because that can require I/O to read the map page, you
180 : : * shouldn't hold a lock on the heap page while doing that. Then, call
181 : : * visibilitymap_set to actually set the bit.
182 : : *
183 : : * On entry, *vmbuf should be InvalidBuffer or a valid buffer returned by
184 : : * an earlier call to visibilitymap_pin or visibilitymap_get_status on the same
185 : : * relation. On return, *vmbuf is a valid buffer with the map page containing
186 : : * the bit for heapBlk.
187 : : *
188 : : * If the page doesn't exist in the map file yet, it is extended.
189 : : */
190 : : void
1083 pg@bowt.ie 191 : 114645 : visibilitymap_pin(Relation rel, BlockNumber heapBlk, Buffer *vmbuf)
192 : : {
6121 heikki.linnakangas@i 193 : 114645 : BlockNumber mapBlock = HEAPBLK_TO_MAPBLOCK(heapBlk);
194 : :
195 : : /* Reuse the old pinned buffer if possible */
1083 pg@bowt.ie 196 [ + + ]: 114645 : if (BufferIsValid(*vmbuf))
197 : : {
198 [ + - ]: 89714 : if (BufferGetBlockNumber(*vmbuf) == mapBlock)
6121 heikki.linnakangas@i 199 : 89714 : return;
200 : :
1083 pg@bowt.ie 201 :UBC 0 : ReleaseBuffer(*vmbuf);
202 : : }
1083 pg@bowt.ie 203 :CBC 24931 : *vmbuf = vm_readbuf(rel, mapBlock, true);
204 : : }
205 : :
206 : : /*
207 : : * visibilitymap_pin_ok - do we already have the correct page pinned?
208 : : *
209 : : * On entry, vmbuf should be InvalidBuffer or a valid buffer returned by
210 : : * an earlier call to visibilitymap_pin or visibilitymap_get_status on the same
211 : : * relation. The return value indicates whether the buffer covers the
212 : : * given heapBlk.
213 : : */
214 : : bool
215 : 16967 : visibilitymap_pin_ok(BlockNumber heapBlk, Buffer vmbuf)
216 : : {
5191 rhaas@postgresql.org 217 : 16967 : BlockNumber mapBlock = HEAPBLK_TO_MAPBLOCK(heapBlk);
218 : :
1083 pg@bowt.ie 219 [ + + + - ]: 16967 : return BufferIsValid(vmbuf) && BufferGetBlockNumber(vmbuf) == mapBlock;
220 : : }
221 : :
222 : : /*
223 : : * visibilitymap_set - set bit(s) on a previously pinned page
224 : : *
225 : : * recptr is the LSN of the XLOG record we're replaying, if we're in recovery,
226 : : * or InvalidXLogRecPtr in normal running. The VM page LSN is advanced to the
227 : : * one provided; in normal running, we generate a new XLOG record and set the
228 : : * page LSN to that value (though the heap page's LSN may *not* be updated;
229 : : * see below). cutoff_xid is the largest xmin on the page being marked
230 : : * all-visible; it is needed for Hot Standby, and can be InvalidTransactionId
231 : : * if the page contains no tuples. It can also be set to InvalidTransactionId
232 : : * when a page that is already all-visible is being marked all-frozen.
233 : : *
234 : : * Caller is expected to set the heap page's PD_ALL_VISIBLE bit before calling
235 : : * this function. Except in recovery, caller should also pass the heap
236 : : * buffer. When checksums are enabled and we're not in recovery, we must add
237 : : * the heap buffer to the WAL chain to protect it from being torn.
238 : : *
239 : : * You must pass a buffer containing the correct map page to this function.
240 : : * Call visibilitymap_pin first to pin the right one. This function doesn't do
241 : : * any I/O.
242 : : *
243 : : * Returns the state of the page's VM bits before setting flags.
244 : : */
245 : : uint8
4551 simon@2ndQuadrant.co 246 : 61703 : visibilitymap_set(Relation rel, BlockNumber heapBlk, Buffer heapBuf,
247 : : XLogRecPtr recptr, Buffer vmBuf, TransactionId cutoff_xid,
248 : : uint8 flags)
249 : : {
6121 heikki.linnakangas@i 250 : 61703 : BlockNumber mapBlock = HEAPBLK_TO_MAPBLOCK(heapBlk);
251 : 61703 : uint32 mapByte = HEAPBLK_TO_MAPBYTE(heapBlk);
3382 rhaas@postgresql.org 252 : 61703 : uint8 mapOffset = HEAPBLK_TO_OFFSET(heapBlk);
253 : : Page page;
254 : : uint8 *map;
255 : : uint8 status;
256 : :
257 : : #ifdef TRACE_VISIBILITYMAP
258 : : elog(DEBUG1, "vm_set flags 0x%02X for %s %d",
259 : : flags, RelationGetRelationName(rel), heapBlk);
260 : : #endif
261 : :
5191 262 [ + + - + ]: 61703 : Assert(InRecovery || XLogRecPtrIsInvalid(recptr));
8 peter@eisentraut.org 263 [ + + - + ]:GNC 61703 : Assert(InRecovery || PageIsAllVisible(BufferGetPage(heapBuf)));
889 andres@anarazel.de 264 [ - + ]:CBC 61703 : Assert((flags & VISIBILITYMAP_VALID_BITS) == flags);
265 : :
266 : : /* Must never set all_frozen bit without also setting all_visible bit */
964 pg@bowt.ie 267 [ - + ]: 61703 : Assert(flags != VISIBILITYMAP_ALL_FROZEN);
268 : :
269 : : /* Check that we have the right heap page pinned, if present */
4551 simon@2ndQuadrant.co 270 [ + + - + ]: 61703 : if (BufferIsValid(heapBuf) && BufferGetBlockNumber(heapBuf) != heapBlk)
4551 simon@2ndQuadrant.co 271 [ # # ]:UBC 0 : elog(ERROR, "wrong heap buffer passed to visibilitymap_set");
272 : :
1 melanieplageman@gmai 273 [ + + - + ]:GNC 61703 : Assert(!BufferIsValid(heapBuf) || BufferIsExclusiveLocked(heapBuf));
274 : :
275 : : /* Check that we have the right VM page pinned */
4551 simon@2ndQuadrant.co 276 [ + - - + ]:CBC 61703 : if (!BufferIsValid(vmBuf) || BufferGetBlockNumber(vmBuf) != mapBlock)
4551 simon@2ndQuadrant.co 277 [ # # ]:UBC 0 : elog(ERROR, "wrong VM buffer passed to visibilitymap_set");
278 : :
3426 kgrittn@postgresql.o 279 :CBC 61703 : page = BufferGetPage(vmBuf);
3376 rhaas@postgresql.org 280 : 61703 : map = (uint8 *) PageGetContents(page);
4551 simon@2ndQuadrant.co 281 : 61703 : LockBuffer(vmBuf, BUFFER_LOCK_EXCLUSIVE);
282 : :
263 melanieplageman@gmai 283 : 61703 : status = (map[mapByte] >> mapOffset) & VISIBILITYMAP_VALID_BITS;
284 [ + - ]: 61703 : if (flags != status)
285 : : {
5191 rhaas@postgresql.org 286 : 61703 : START_CRIT_SECTION();
287 : :
3382 288 : 61703 : map[mapByte] |= (flags << mapOffset);
4551 simon@2ndQuadrant.co 289 : 61703 : MarkBufferDirty(vmBuf);
290 : :
5191 rhaas@postgresql.org 291 [ + + + + : 61703 : if (RelationNeedsWAL(rel))
+ - + + ]
292 : : {
293 [ + + ]: 57647 : if (XLogRecPtrIsInvalid(recptr))
294 : : {
4551 simon@2ndQuadrant.co 295 [ - + ]: 49891 : Assert(!InRecovery);
889 andres@anarazel.de 296 : 49891 : recptr = log_heap_visible(rel, heapBuf, vmBuf, cutoff_xid, flags);
297 : :
298 : : /*
299 : : * If data checksums are enabled (or wal_log_hints=on), we
300 : : * need to protect the heap page from being torn.
301 : : *
302 : : * If not, then we must *not* update the heap page's LSN. In
303 : : * this case, the FPI for the heap page was omitted from the
304 : : * WAL record inserted above, so it would be incorrect to
305 : : * update the heap page's LSN.
306 : : */
4285 heikki.linnakangas@i 307 [ + + - + ]: 49891 : if (XLogHintBitIsNeeded())
308 : : {
3426 kgrittn@postgresql.o 309 : 46425 : Page heapPage = BufferGetPage(heapBuf);
310 : :
4551 simon@2ndQuadrant.co 311 : 46425 : PageSetLSN(heapPage, recptr);
312 : : }
313 : : }
6121 heikki.linnakangas@i 314 : 57647 : PageSetLSN(page, recptr);
315 : : }
316 : :
5191 rhaas@postgresql.org 317 [ - + ]: 61703 : END_CRIT_SECTION();
318 : : }
319 : :
4551 simon@2ndQuadrant.co 320 : 61703 : LockBuffer(vmBuf, BUFFER_LOCK_UNLOCK);
263 melanieplageman@gmai 321 : 61703 : return status;
322 : : }
323 : :
324 : : /*
325 : : * visibilitymap_get_status - get status of bits
326 : : *
327 : : * Are all tuples on heapBlk visible to all or are marked frozen, according
328 : : * to the visibility map?
329 : : *
330 : : * On entry, *vmbuf should be InvalidBuffer or a valid buffer returned by an
331 : : * earlier call to visibilitymap_pin or visibilitymap_get_status on the same
332 : : * relation. On return, *vmbuf is a valid buffer with the map page containing
333 : : * the bit for heapBlk, or InvalidBuffer. The caller is responsible for
334 : : * releasing *vmbuf after it's done testing and setting bits.
335 : : *
336 : : * NOTE: This function is typically called without a lock on the heap page,
337 : : * so somebody else could change the bit just after we look at it. In fact,
338 : : * since we don't lock the visibility map page either, it's even possible that
339 : : * someone else could have changed the bit just before we look at it, but yet
340 : : * we might see the old value. It is the caller's responsibility to deal with
341 : : * all concurrency issues!
342 : : */
343 : : uint8
1083 pg@bowt.ie 344 : 3011548 : visibilitymap_get_status(Relation rel, BlockNumber heapBlk, Buffer *vmbuf)
345 : : {
6121 heikki.linnakangas@i 346 : 3011548 : BlockNumber mapBlock = HEAPBLK_TO_MAPBLOCK(heapBlk);
347 : 3011548 : uint32 mapByte = HEAPBLK_TO_MAPBYTE(heapBlk);
3382 rhaas@postgresql.org 348 : 3011548 : uint8 mapOffset = HEAPBLK_TO_OFFSET(heapBlk);
349 : : char *map;
350 : : uint8 result;
351 : :
352 : : #ifdef TRACE_VISIBILITYMAP
353 : : elog(DEBUG1, "vm_get_status %s %d", RelationGetRelationName(rel), heapBlk);
354 : : #endif
355 : :
356 : : /* Reuse the old pinned buffer if possible */
1083 pg@bowt.ie 357 [ + + ]: 3011548 : if (BufferIsValid(*vmbuf))
358 : : {
359 [ - + ]: 2181725 : if (BufferGetBlockNumber(*vmbuf) != mapBlock)
360 : : {
1083 pg@bowt.ie 361 :UBC 0 : ReleaseBuffer(*vmbuf);
362 : 0 : *vmbuf = InvalidBuffer;
363 : : }
364 : : }
365 : :
1083 pg@bowt.ie 366 [ + + ]:CBC 3011548 : if (!BufferIsValid(*vmbuf))
367 : : {
368 : 829823 : *vmbuf = vm_readbuf(rel, mapBlock, false);
369 [ + + ]: 829823 : if (!BufferIsValid(*vmbuf))
44 nathan@postgresql.or 370 :GNC 767042 : return (uint8) 0;
371 : : }
372 : :
1083 pg@bowt.ie 373 :CBC 2244506 : map = PageGetContents(BufferGetPage(*vmbuf));
374 : :
375 : : /*
376 : : * A single byte read is atomic. There could be memory-ordering effects
377 : : * here, but for performance reasons we make it the caller's job to worry
378 : : * about that.
379 : : */
3382 rhaas@postgresql.org 380 : 2244506 : result = ((map[mapByte] >> mapOffset) & VISIBILITYMAP_VALID_BITS);
381 : 2244506 : return result;
382 : : }
383 : :
384 : : /*
385 : : * visibilitymap_count - count number of bits set in visibility map
386 : : *
387 : : * Note: we ignore the possibility of race conditions when the table is being
388 : : * extended concurrently with the call. New pages added to the table aren't
389 : : * going to be marked all-visible or all-frozen, so they won't affect the result.
390 : : */
391 : : void
3476 392 : 26899 : visibilitymap_count(Relation rel, BlockNumber *all_visible, BlockNumber *all_frozen)
393 : : {
394 : : BlockNumber mapBlock;
2395 tgl@sss.pgh.pa.us 395 : 26899 : BlockNumber nvisible = 0;
396 : 26899 : BlockNumber nfrozen = 0;
397 : :
398 : : /* all_visible must be specified */
3476 rhaas@postgresql.org 399 [ - + ]: 26899 : Assert(all_visible);
400 : :
4836 bruce@momjian.us 401 : 26899 : for (mapBlock = 0;; mapBlock++)
5076 tgl@sss.pgh.pa.us 402 : 8088 : {
403 : : Buffer mapBuffer;
404 : : uint64 *map;
405 : :
406 : : /*
407 : : * Read till we fall off the end of the map. We assume that any extra
408 : : * bytes in the last page are zeroed, so we don't bother excluding
409 : : * them from the count.
410 : : */
411 : 34987 : mapBuffer = vm_readbuf(rel, mapBlock, false);
412 [ + + ]: 34987 : if (!BufferIsValid(mapBuffer))
413 : 26899 : break;
414 : :
415 : : /*
416 : : * We choose not to lock the page, since the result is going to be
417 : : * immediately stale anyway if anyone is concurrently setting or
418 : : * clearing bits, and we only really need an approximate value.
419 : : */
2395 420 : 8088 : map = (uint64 *) PageGetContents(BufferGetPage(mapBuffer));
421 : :
518 nathan@postgresql.or 422 : 8088 : nvisible += pg_popcount_masked((const char *) map, MAPSIZE, VISIBLE_MASK8);
423 [ + - ]: 8088 : if (all_frozen)
424 : 8088 : nfrozen += pg_popcount_masked((const char *) map, MAPSIZE, FROZEN_MASK8);
425 : :
5076 tgl@sss.pgh.pa.us 426 : 8088 : ReleaseBuffer(mapBuffer);
427 : : }
428 : :
2395 429 : 26899 : *all_visible = nvisible;
430 [ + - ]: 26899 : if (all_frozen)
431 : 26899 : *all_frozen = nfrozen;
5076 432 : 26899 : }
433 : :
434 : : /*
435 : : * visibilitymap_prepare_truncate -
436 : : * prepare for truncation of the visibility map
437 : : *
438 : : * nheapblocks is the new size of the heap.
439 : : *
440 : : * Return the number of blocks of new visibility map.
441 : : * If it's InvalidBlockNumber, there is nothing to truncate;
442 : : * otherwise the caller is responsible for calling smgrtruncate()
443 : : * to truncate the visibility map pages.
444 : : */
445 : : BlockNumber
2174 fujii@postgresql.org 446 : 203 : visibilitymap_prepare_truncate(Relation rel, BlockNumber nheapblocks)
447 : : {
448 : : BlockNumber newnblocks;
449 : :
450 : : /* last remaining block, byte, and bit */
6121 heikki.linnakangas@i 451 : 203 : BlockNumber truncBlock = HEAPBLK_TO_MAPBLOCK(nheapblocks);
5931 bruce@momjian.us 452 : 203 : uint32 truncByte = HEAPBLK_TO_MAPBYTE(nheapblocks);
3382 rhaas@postgresql.org 453 : 203 : uint8 truncOffset = HEAPBLK_TO_OFFSET(nheapblocks);
454 : :
455 : : #ifdef TRACE_VISIBILITYMAP
456 : : elog(DEBUG1, "vm_truncate %s %d", RelationGetRelationName(rel), nheapblocks);
457 : : #endif
458 : :
459 : : /*
460 : : * If no visibility map has been created yet for this relation, there's
461 : : * nothing to truncate.
462 : : */
1517 tgl@sss.pgh.pa.us 463 [ - + ]: 203 : if (!smgrexists(RelationGetSmgr(rel), VISIBILITYMAP_FORKNUM))
2174 fujii@postgresql.org 464 :UBC 0 : return InvalidBlockNumber;
465 : :
466 : : /*
467 : : * Unless the new size is exactly at a visibility map page boundary, the
468 : : * tail bits in the last remaining map page, representing truncated heap
469 : : * blocks, need to be cleared. This is not only tidy, but also necessary
470 : : * because we don't get a chance to clear the bits if the heap is extended
471 : : * again.
472 : : */
3382 rhaas@postgresql.org 473 [ + + + + ]:CBC 203 : if (truncByte != 0 || truncOffset != 0)
6121 heikki.linnakangas@i 474 : 129 : {
475 : : Buffer mapBuffer;
476 : : Page page;
477 : : char *map;
478 : :
479 : 129 : newnblocks = truncBlock + 1;
480 : :
481 : 129 : mapBuffer = vm_readbuf(rel, truncBlock, false);
482 [ - + ]: 129 : if (!BufferIsValid(mapBuffer))
483 : : {
484 : : /* nothing to do, the file was already smaller */
2174 fujii@postgresql.org 485 :UBC 0 : return InvalidBlockNumber;
486 : : }
487 : :
3426 kgrittn@postgresql.o 488 :CBC 129 : page = BufferGetPage(mapBuffer);
6121 heikki.linnakangas@i 489 : 129 : map = PageGetContents(page);
490 : :
491 : 129 : LockBuffer(mapBuffer, BUFFER_LOCK_EXCLUSIVE);
492 : :
493 : : /* NO EREPORT(ERROR) from here till changes are logged */
3244 494 : 129 : START_CRIT_SECTION();
495 : :
496 : : /* Clear out the unwanted bytes. */
6121 497 [ + + + - : 129 : MemSet(&map[truncByte + 1], 0, MAPSIZE - (truncByte + 1));
+ - - + -
- ]
498 : :
499 : : /*----
500 : : * Mask out the unwanted bits of the last remaining byte.
501 : : *
502 : : * ((1 << 0) - 1) = 00000000
503 : : * ((1 << 1) - 1) = 00000001
504 : : * ...
505 : : * ((1 << 6) - 1) = 00111111
506 : : * ((1 << 7) - 1) = 01111111
507 : : *----
508 : : */
3382 rhaas@postgresql.org 509 : 129 : map[truncByte] &= (1 << truncOffset) - 1;
510 : :
511 : : /*
512 : : * Truncation of a relation is WAL-logged at a higher-level, and we
513 : : * will be called at WAL replay. But if checksums are enabled, we need
514 : : * to still write a WAL record to protect against a torn page, if the
515 : : * page is flushed to disk before the truncation WAL record. We cannot
516 : : * use MarkBufferDirtyHint here, because that will not dirty the page
517 : : * during recovery.
518 : : */
6121 heikki.linnakangas@i 519 : 129 : MarkBufferDirty(mapBuffer);
3244 520 [ + + + + : 129 : if (!InRecovery && RelationNeedsWAL(rel) && XLogHintBitIsNeeded())
+ + + - +
- - + -
- ]
521 : 108 : log_newpage_buffer(mapBuffer, false);
522 : :
523 [ - + ]: 129 : END_CRIT_SECTION();
524 : :
6121 525 : 129 : UnlockReleaseBuffer(mapBuffer);
526 : : }
527 : : else
528 : 74 : newnblocks = truncBlock;
529 : :
1517 tgl@sss.pgh.pa.us 530 [ + + ]: 203 : if (smgrnblocks(RelationGetSmgr(rel), VISIBILITYMAP_FORKNUM) <= newnblocks)
531 : : {
532 : : /* nothing to do, the file was already smaller than requested size */
2174 fujii@postgresql.org 533 : 129 : return InvalidBlockNumber;
534 : : }
535 : :
536 : 74 : return newnblocks;
537 : : }
538 : :
539 : : /*
540 : : * Read a visibility map page.
541 : : *
542 : : * If the page doesn't exist, InvalidBuffer is returned, or if 'extend' is
543 : : * true, the visibility map file is extended.
544 : : */
545 : : static Buffer
6121 heikki.linnakangas@i 546 : 889870 : vm_readbuf(Relation rel, BlockNumber blkno, bool extend)
547 : : {
548 : : Buffer buf;
549 : : SMgrRelation reln;
550 : :
551 : : /*
552 : : * Caution: re-using this smgr pointer could fail if the relcache entry
553 : : * gets closed. It's safe as long as we only do smgr-level operations
554 : : * between here and the last use of the pointer.
555 : : */
1517 tgl@sss.pgh.pa.us 556 : 889870 : reln = RelationGetSmgr(rel);
557 : :
558 : : /*
559 : : * If we haven't cached the size of the visibility map fork yet, check it
560 : : * first.
561 : : */
562 [ + + ]: 889870 : if (reln->smgr_cached_nblocks[VISIBILITYMAP_FORKNUM] == InvalidBlockNumber)
563 : : {
564 [ + + ]: 38088 : if (smgrexists(reln, VISIBILITYMAP_FORKNUM))
565 : 17428 : smgrnblocks(reln, VISIBILITYMAP_FORKNUM);
566 : : else
567 : 20660 : reln->smgr_cached_nblocks[VISIBILITYMAP_FORKNUM] = 0;
568 : : }
569 : :
570 : : /*
571 : : * For reading we use ZERO_ON_ERROR mode, and initialize the page if
572 : : * necessary. It's always safe to clear bits, so it's better to clear
573 : : * corrupt pages than error out.
574 : : *
575 : : * We use the same path below to initialize pages when extending the
576 : : * relation, as a concurrent extension can end up with vm_extend()
577 : : * returning an already-initialized page.
578 : : */
579 [ + + ]: 889870 : if (blkno >= reln->smgr_cached_nblocks[VISIBILITYMAP_FORKNUM])
580 : : {
6121 heikki.linnakangas@i 581 [ + + ]: 796944 : if (extend)
885 andres@anarazel.de 582 : 3003 : buf = vm_extend(rel, blkno + 1);
583 : : else
6121 heikki.linnakangas@i 584 : 793941 : return InvalidBuffer;
585 : : }
586 : : else
885 andres@anarazel.de 587 : 92926 : buf = ReadBufferExtended(rel, VISIBILITYMAP_FORKNUM, blkno,
588 : : RBM_ZERO_ON_ERROR, NULL);
589 : :
590 : : /*
591 : : * Initializing the page when needed is trickier than it looks, because of
592 : : * the possibility of multiple backends doing this concurrently, and our
593 : : * desire to not uselessly take the buffer lock in the normal path where
594 : : * the page is OK. We must take the lock to initialize the page, so
595 : : * recheck page newness after we have the lock, in case someone else
596 : : * already did it. Also, because we initially check PageIsNew with no
597 : : * lock, it's possible to fall through and return the buffer while someone
598 : : * else is still initializing the page (i.e., we might see pd_upper as set
599 : : * but other page header fields are still zeroes). This is harmless for
600 : : * callers that will take a buffer lock themselves, but some callers
601 : : * inspect the page without any lock at all. The latter is OK only so
602 : : * long as it doesn't depend on the page header having correct contents.
603 : : * Current usage is safe because PageGetContents() does not require that.
604 : : */
3426 kgrittn@postgresql.o 605 [ + + ]: 95929 : if (PageIsNew(BufferGetPage(buf)))
606 : : {
2612 tgl@sss.pgh.pa.us 607 : 3061 : LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
608 [ + - ]: 3061 : if (PageIsNew(BufferGetPage(buf)))
609 : 3061 : PageInit(BufferGetPage(buf), BLCKSZ, 0);
610 : 3061 : LockBuffer(buf, BUFFER_LOCK_UNLOCK);
611 : : }
6121 heikki.linnakangas@i 612 : 95929 : return buf;
613 : : }
614 : :
615 : : /*
616 : : * Ensure that the visibility map fork is at least vm_nblocks long, extending
617 : : * it if necessary with zeroed pages.
618 : : */
619 : : static Buffer
620 : 3003 : vm_extend(Relation rel, BlockNumber vm_nblocks)
621 : : {
622 : : Buffer buf;
623 : :
745 tmunro@postgresql.or 624 : 3003 : buf = ExtendBufferedRelTo(BMR_REL(rel), VISIBILITYMAP_FORKNUM, NULL,
625 : : EB_CREATE_FORK_IF_NEEDED |
626 : : EB_CLEAR_SIZE_CACHE,
627 : : vm_nblocks,
628 : : RBM_ZERO_ON_ERROR);
629 : :
630 : : /*
631 : : * Send a shared-inval message to force other backends to close any smgr
632 : : * references they may have for this rel, which we are about to change.
633 : : * This is a useful optimization because it means that backends don't have
634 : : * to keep checking for creation or extension of the file, which happens
635 : : * infrequently.
636 : : */
885 andres@anarazel.de 637 : 3003 : CacheInvalidateSmgr(RelationGetSmgr(rel)->smgr_rlocator);
638 : :
639 : 3003 : return buf;
640 : : }
|
