Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * visibilitymap.c
4 : * bitmap for tracking visibility of heap tuples
5 : *
6 : * Portions Copyright (c) 1996-2017, 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_truncate - truncate the visibility map
21 : *
22 : * NOTES
23 : *
24 : * The visibility map is a bitmap with two bits (all-visible and all-frozen)
25 : * per heap page. A set all-visible bit means that all tuples on the page are
26 : * known visible to all transactions, and therefore the page doesn't need to
27 : * be vacuumed. A set all-frozen bit means that all tuples on the page are
28 : * completely frozen, and therefore the page doesn't need to be vacuumed even
29 : * if whole table scanning vacuum is required (e.g. anti-wraparound vacuum).
30 : * The all-frozen bit must be set only when the page is already all-visible.
31 : *
32 : * The map is conservative in the sense that we make sure that whenever a bit
33 : * is set, we know the condition is true, but if a bit is not set, it might or
34 : * might not be true.
35 : *
36 : * Clearing visibility map bits is not separately WAL-logged. The callers
37 : * must make sure that whenever a bit is cleared, the bit is cleared on WAL
38 : * replay of the updating operation as well.
39 : *
40 : * When we *set* a visibility map during VACUUM, we must write WAL. This may
41 : * seem counterintuitive, since the bit is basically a hint: if it is clear,
42 : * it may still be the case that every tuple on the page is visible to all
43 : * transactions; we just don't know that for certain. The difficulty is that
44 : * there are two bits which are typically set together: the PD_ALL_VISIBLE bit
45 : * on the page itself, and the visibility map bit. If a crash occurs after the
46 : * visibility map page makes it to disk and before the updated heap page makes
47 : * it to disk, redo must set the bit on the heap page. Otherwise, the next
48 : * insert, update, or delete on the heap page will fail to realize that the
49 : * visibility map bit must be cleared, possibly causing index-only scans to
50 : * return wrong answers.
51 : *
52 : * VACUUM will normally skip pages for which the visibility map bit is set;
53 : * such pages can't contain any dead tuples and therefore don't need vacuuming.
54 : *
55 : * LOCKING
56 : *
57 : * In heapam.c, whenever a page is modified so that not all tuples on the
58 : * page are visible to everyone anymore, the corresponding bit in the
59 : * visibility map is cleared. In order to be crash-safe, we need to do this
60 : * while still holding a lock on the heap page and in the same critical
61 : * section that logs the page modification. However, we don't want to hold
62 : * the buffer lock over any I/O that may be required to read in the visibility
63 : * map page. To avoid this, we examine the heap page before locking it;
64 : * if the page-level PD_ALL_VISIBLE bit is set, we pin the visibility map
65 : * bit. Then, we lock the buffer. But this creates a race condition: there
66 : * is a possibility that in the time it takes to lock the buffer, the
67 : * PD_ALL_VISIBLE bit gets set. If that happens, we have to unlock the
68 : * buffer, pin the visibility map page, and relock the buffer. This shouldn't
69 : * happen often, because only VACUUM currently sets visibility map bits,
70 : * and the race will only occur if VACUUM processes a given page at almost
71 : * exactly the same time that someone tries to further modify it.
72 : *
73 : * To set a bit, you need to hold a lock on the heap page. That prevents
74 : * the race condition where VACUUM sees that all tuples on the page are
75 : * visible to everyone, but another backend modifies the page before VACUUM
76 : * sets the bit in the visibility map.
77 : *
78 : * When a bit is set, the LSN of the visibility map page is updated to make
79 : * sure that the visibility map update doesn't get written to disk before the
80 : * WAL record of the changes that made it possible to set the bit is flushed.
81 : * But when a bit is cleared, we don't have to do that because it's always
82 : * safe to clear a bit in the map from correctness point of view.
83 : *
84 : *-------------------------------------------------------------------------
85 : */
86 : #include "postgres.h"
87 :
88 : #include "access/heapam_xlog.h"
89 : #include "access/visibilitymap.h"
90 : #include "access/xlog.h"
91 : #include "miscadmin.h"
92 : #include "storage/bufmgr.h"
93 : #include "storage/lmgr.h"
94 : #include "storage/smgr.h"
95 : #include "utils/inval.h"
96 :
97 :
98 : /*#define TRACE_VISIBILITYMAP */
99 :
100 : /*
101 : * Size of the bitmap on each visibility map page, in bytes. There's no
102 : * extra headers, so the whole page minus the standard page header is
103 : * used for the bitmap.
104 : */
105 : #define MAPSIZE (BLCKSZ - MAXALIGN(SizeOfPageHeaderData))
106 :
107 : /* Number of heap blocks we can represent in one byte */
108 : #define HEAPBLOCKS_PER_BYTE (BITS_PER_BYTE / BITS_PER_HEAPBLOCK)
109 :
110 : /* Number of heap blocks we can represent in one visibility map page. */
111 : #define HEAPBLOCKS_PER_PAGE (MAPSIZE * HEAPBLOCKS_PER_BYTE)
112 :
113 : /* Mapping from heap block number to the right bit in the visibility map */
114 : #define HEAPBLK_TO_MAPBLOCK(x) ((x) / HEAPBLOCKS_PER_PAGE)
115 : #define HEAPBLK_TO_MAPBYTE(x) (((x) % HEAPBLOCKS_PER_PAGE) / HEAPBLOCKS_PER_BYTE)
116 : #define HEAPBLK_TO_OFFSET(x) (((x) % HEAPBLOCKS_PER_BYTE) * BITS_PER_HEAPBLOCK)
117 :
118 : /* tables for fast counting of set bits for visible and frozen */
119 : static const uint8 number_of_ones_for_visible[256] = {
120 : 0, 1, 0, 1, 1, 2, 1, 2, 0, 1, 0, 1, 1, 2, 1, 2,
121 : 1, 2, 1, 2, 2, 3, 2, 3, 1, 2, 1, 2, 2, 3, 2, 3,
122 : 0, 1, 0, 1, 1, 2, 1, 2, 0, 1, 0, 1, 1, 2, 1, 2,
123 : 1, 2, 1, 2, 2, 3, 2, 3, 1, 2, 1, 2, 2, 3, 2, 3,
124 : 1, 2, 1, 2, 2, 3, 2, 3, 1, 2, 1, 2, 2, 3, 2, 3,
125 : 2, 3, 2, 3, 3, 4, 3, 4, 2, 3, 2, 3, 3, 4, 3, 4,
126 : 1, 2, 1, 2, 2, 3, 2, 3, 1, 2, 1, 2, 2, 3, 2, 3,
127 : 2, 3, 2, 3, 3, 4, 3, 4, 2, 3, 2, 3, 3, 4, 3, 4,
128 : 0, 1, 0, 1, 1, 2, 1, 2, 0, 1, 0, 1, 1, 2, 1, 2,
129 : 1, 2, 1, 2, 2, 3, 2, 3, 1, 2, 1, 2, 2, 3, 2, 3,
130 : 0, 1, 0, 1, 1, 2, 1, 2, 0, 1, 0, 1, 1, 2, 1, 2,
131 : 1, 2, 1, 2, 2, 3, 2, 3, 1, 2, 1, 2, 2, 3, 2, 3,
132 : 1, 2, 1, 2, 2, 3, 2, 3, 1, 2, 1, 2, 2, 3, 2, 3,
133 : 2, 3, 2, 3, 3, 4, 3, 4, 2, 3, 2, 3, 3, 4, 3, 4,
134 : 1, 2, 1, 2, 2, 3, 2, 3, 1, 2, 1, 2, 2, 3, 2, 3,
135 : 2, 3, 2, 3, 3, 4, 3, 4, 2, 3, 2, 3, 3, 4, 3, 4
136 : };
137 : static const uint8 number_of_ones_for_frozen[256] = {
138 : 0, 0, 1, 1, 0, 0, 1, 1, 1, 1, 2, 2, 1, 1, 2, 2,
139 : 0, 0, 1, 1, 0, 0, 1, 1, 1, 1, 2, 2, 1, 1, 2, 2,
140 : 1, 1, 2, 2, 1, 1, 2, 2, 2, 2, 3, 3, 2, 2, 3, 3,
141 : 1, 1, 2, 2, 1, 1, 2, 2, 2, 2, 3, 3, 2, 2, 3, 3,
142 : 0, 0, 1, 1, 0, 0, 1, 1, 1, 1, 2, 2, 1, 1, 2, 2,
143 : 0, 0, 1, 1, 0, 0, 1, 1, 1, 1, 2, 2, 1, 1, 2, 2,
144 : 1, 1, 2, 2, 1, 1, 2, 2, 2, 2, 3, 3, 2, 2, 3, 3,
145 : 1, 1, 2, 2, 1, 1, 2, 2, 2, 2, 3, 3, 2, 2, 3, 3,
146 : 1, 1, 2, 2, 1, 1, 2, 2, 2, 2, 3, 3, 2, 2, 3, 3,
147 : 1, 1, 2, 2, 1, 1, 2, 2, 2, 2, 3, 3, 2, 2, 3, 3,
148 : 2, 2, 3, 3, 2, 2, 3, 3, 3, 3, 4, 4, 3, 3, 4, 4,
149 : 2, 2, 3, 3, 2, 2, 3, 3, 3, 3, 4, 4, 3, 3, 4, 4,
150 : 1, 1, 2, 2, 1, 1, 2, 2, 2, 2, 3, 3, 2, 2, 3, 3,
151 : 1, 1, 2, 2, 1, 1, 2, 2, 2, 2, 3, 3, 2, 2, 3, 3,
152 : 2, 2, 3, 3, 2, 2, 3, 3, 3, 3, 4, 4, 3, 3, 4, 4,
153 : 2, 2, 3, 3, 2, 2, 3, 3, 3, 3, 4, 4, 3, 3, 4, 4
154 : };
155 :
156 : /* prototypes for internal routines */
157 : static Buffer vm_readbuf(Relation rel, BlockNumber blkno, bool extend);
158 : static void vm_extend(Relation rel, BlockNumber nvmblocks);
159 :
160 :
161 : /*
162 : * visibilitymap_clear - clear specified bits for one page in visibility map
163 : *
164 : * You must pass a buffer containing the correct map page to this function.
165 : * Call visibilitymap_pin first to pin the right one. This function doesn't do
166 : * any I/O. Returns true if any bits have been cleared and false otherwise.
167 : */
168 : bool
169 962 : visibilitymap_clear(Relation rel, BlockNumber heapBlk, Buffer buf, uint8 flags)
170 : {
171 962 : BlockNumber mapBlock = HEAPBLK_TO_MAPBLOCK(heapBlk);
172 962 : int mapByte = HEAPBLK_TO_MAPBYTE(heapBlk);
173 962 : int mapOffset = HEAPBLK_TO_OFFSET(heapBlk);
174 962 : uint8 mask = flags << mapOffset;
175 : char *map;
176 962 : bool cleared = false;
177 :
178 962 : Assert(flags & VISIBILITYMAP_VALID_BITS);
179 :
180 : #ifdef TRACE_VISIBILITYMAP
181 : elog(DEBUG1, "vm_clear %s %d", RelationGetRelationName(rel), heapBlk);
182 : #endif
183 :
184 962 : if (!BufferIsValid(buf) || BufferGetBlockNumber(buf) != mapBlock)
185 0 : elog(ERROR, "wrong buffer passed to visibilitymap_clear");
186 :
187 962 : LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
188 962 : map = PageGetContents(BufferGetPage(buf));
189 :
190 962 : if (map[mapByte] & mask)
191 : {
192 917 : map[mapByte] &= ~mask;
193 :
194 917 : MarkBufferDirty(buf);
195 917 : cleared = true;
196 : }
197 :
198 962 : LockBuffer(buf, BUFFER_LOCK_UNLOCK);
199 :
200 962 : return cleared;
201 : }
202 :
203 : /*
204 : * visibilitymap_pin - pin a map page for setting a bit
205 : *
206 : * Setting a bit in the visibility map is a two-phase operation. First, call
207 : * visibilitymap_pin, to pin the visibility map page containing the bit for
208 : * the heap page. Because that can require I/O to read the map page, you
209 : * shouldn't hold a lock on the heap page while doing that. Then, call
210 : * visibilitymap_set to actually set the bit.
211 : *
212 : * On entry, *buf should be InvalidBuffer or a valid buffer returned by
213 : * an earlier call to visibilitymap_pin or visibilitymap_get_status on the same
214 : * relation. On return, *buf is a valid buffer with the map page containing
215 : * the bit for heapBlk.
216 : *
217 : * If the page doesn't exist in the map file yet, it is extended.
218 : */
219 : void
220 5642 : visibilitymap_pin(Relation rel, BlockNumber heapBlk, Buffer *buf)
221 : {
222 5642 : BlockNumber mapBlock = HEAPBLK_TO_MAPBLOCK(heapBlk);
223 :
224 : /* Reuse the old pinned buffer if possible */
225 5642 : if (BufferIsValid(*buf))
226 : {
227 4562 : if (BufferGetBlockNumber(*buf) == mapBlock)
228 10204 : return;
229 :
230 0 : ReleaseBuffer(*buf);
231 : }
232 1080 : *buf = vm_readbuf(rel, mapBlock, true);
233 : }
234 :
235 : /*
236 : * visibilitymap_pin_ok - do we already have the correct page pinned?
237 : *
238 : * On entry, buf should be InvalidBuffer or a valid buffer returned by
239 : * an earlier call to visibilitymap_pin or visibilitymap_get_status on the same
240 : * relation. The return value indicates whether the buffer covers the
241 : * given heapBlk.
242 : */
243 : bool
244 582 : visibilitymap_pin_ok(BlockNumber heapBlk, Buffer buf)
245 : {
246 582 : BlockNumber mapBlock = HEAPBLK_TO_MAPBLOCK(heapBlk);
247 :
248 582 : return BufferIsValid(buf) && BufferGetBlockNumber(buf) == mapBlock;
249 : }
250 :
251 : /*
252 : * visibilitymap_set - set bit(s) on a previously pinned page
253 : *
254 : * recptr is the LSN of the XLOG record we're replaying, if we're in recovery,
255 : * or InvalidXLogRecPtr in normal running. The page LSN is advanced to the
256 : * one provided; in normal running, we generate a new XLOG record and set the
257 : * page LSN to that value. cutoff_xid is the largest xmin on the page being
258 : * marked all-visible; it is needed for Hot Standby, and can be
259 : * InvalidTransactionId if the page contains no tuples. It can also be set
260 : * to InvalidTransactionId when a page that is already all-visible is being
261 : * marked all-frozen.
262 : *
263 : * Caller is expected to set the heap page's PD_ALL_VISIBLE bit before calling
264 : * this function. Except in recovery, caller should also pass the heap
265 : * buffer. When checksums are enabled and we're not in recovery, we must add
266 : * the heap buffer to the WAL chain to protect it from being torn.
267 : *
268 : * You must pass a buffer containing the correct map page to this function.
269 : * Call visibilitymap_pin first to pin the right one. This function doesn't do
270 : * any I/O.
271 : */
272 : void
273 4025 : visibilitymap_set(Relation rel, BlockNumber heapBlk, Buffer heapBuf,
274 : XLogRecPtr recptr, Buffer vmBuf, TransactionId cutoff_xid,
275 : uint8 flags)
276 : {
277 4025 : BlockNumber mapBlock = HEAPBLK_TO_MAPBLOCK(heapBlk);
278 4025 : uint32 mapByte = HEAPBLK_TO_MAPBYTE(heapBlk);
279 4025 : uint8 mapOffset = HEAPBLK_TO_OFFSET(heapBlk);
280 : Page page;
281 : uint8 *map;
282 :
283 : #ifdef TRACE_VISIBILITYMAP
284 : elog(DEBUG1, "vm_set %s %d", RelationGetRelationName(rel), heapBlk);
285 : #endif
286 :
287 4025 : Assert(InRecovery || XLogRecPtrIsInvalid(recptr));
288 4025 : Assert(InRecovery || BufferIsValid(heapBuf));
289 4025 : Assert(flags & VISIBILITYMAP_VALID_BITS);
290 :
291 : /* Check that we have the right heap page pinned, if present */
292 4025 : if (BufferIsValid(heapBuf) && BufferGetBlockNumber(heapBuf) != heapBlk)
293 0 : elog(ERROR, "wrong heap buffer passed to visibilitymap_set");
294 :
295 : /* Check that we have the right VM page pinned */
296 4025 : if (!BufferIsValid(vmBuf) || BufferGetBlockNumber(vmBuf) != mapBlock)
297 0 : elog(ERROR, "wrong VM buffer passed to visibilitymap_set");
298 :
299 4025 : page = BufferGetPage(vmBuf);
300 4025 : map = (uint8 *) PageGetContents(page);
301 4025 : LockBuffer(vmBuf, BUFFER_LOCK_EXCLUSIVE);
302 :
303 4025 : if (flags != (map[mapByte] >> mapOffset & VISIBILITYMAP_VALID_BITS))
304 : {
305 4025 : START_CRIT_SECTION();
306 :
307 4025 : map[mapByte] |= (flags << mapOffset);
308 4025 : MarkBufferDirty(vmBuf);
309 :
310 4025 : if (RelationNeedsWAL(rel))
311 : {
312 4025 : if (XLogRecPtrIsInvalid(recptr))
313 : {
314 4025 : Assert(!InRecovery);
315 4025 : recptr = log_heap_visible(rel->rd_node, heapBuf, vmBuf,
316 : cutoff_xid, flags);
317 :
318 : /*
319 : * If data checksums are enabled (or wal_log_hints=on), we
320 : * need to protect the heap page from being torn.
321 : */
322 4025 : if (XLogHintBitIsNeeded())
323 : {
324 0 : Page heapPage = BufferGetPage(heapBuf);
325 :
326 : /* caller is expected to set PD_ALL_VISIBLE first */
327 0 : Assert(PageIsAllVisible(heapPage));
328 0 : PageSetLSN(heapPage, recptr);
329 : }
330 : }
331 4025 : PageSetLSN(page, recptr);
332 : }
333 :
334 4025 : END_CRIT_SECTION();
335 : }
336 :
337 4025 : LockBuffer(vmBuf, BUFFER_LOCK_UNLOCK);
338 4025 : }
339 :
340 : /*
341 : * visibilitymap_get_status - get status of bits
342 : *
343 : * Are all tuples on heapBlk visible to all or are marked frozen, according
344 : * to the visibility map?
345 : *
346 : * On entry, *buf should be InvalidBuffer or a valid buffer returned by an
347 : * earlier call to visibilitymap_pin or visibilitymap_get_status on the same
348 : * relation. On return, *buf is a valid buffer with the map page containing
349 : * the bit for heapBlk, or InvalidBuffer. The caller is responsible for
350 : * releasing *buf after it's done testing and setting bits.
351 : *
352 : * NOTE: This function is typically called without a lock on the heap page,
353 : * so somebody else could change the bit just after we look at it. In fact,
354 : * since we don't lock the visibility map page either, it's even possible that
355 : * someone else could have changed the bit just before we look at it, but yet
356 : * we might see the old value. It is the caller's responsibility to deal with
357 : * all concurrency issues!
358 : */
359 : uint8
360 596075 : visibilitymap_get_status(Relation rel, BlockNumber heapBlk, Buffer *buf)
361 : {
362 596075 : BlockNumber mapBlock = HEAPBLK_TO_MAPBLOCK(heapBlk);
363 596075 : uint32 mapByte = HEAPBLK_TO_MAPBYTE(heapBlk);
364 596075 : uint8 mapOffset = HEAPBLK_TO_OFFSET(heapBlk);
365 : char *map;
366 : uint8 result;
367 :
368 : #ifdef TRACE_VISIBILITYMAP
369 : elog(DEBUG1, "vm_get_status %s %d", RelationGetRelationName(rel), heapBlk);
370 : #endif
371 :
372 : /* Reuse the old pinned buffer if possible */
373 596075 : if (BufferIsValid(*buf))
374 : {
375 450986 : if (BufferGetBlockNumber(*buf) != mapBlock)
376 : {
377 0 : ReleaseBuffer(*buf);
378 0 : *buf = InvalidBuffer;
379 : }
380 : }
381 :
382 596075 : if (!BufferIsValid(*buf))
383 : {
384 145089 : *buf = vm_readbuf(rel, mapBlock, false);
385 145089 : if (!BufferIsValid(*buf))
386 144724 : return false;
387 : }
388 :
389 451351 : map = PageGetContents(BufferGetPage(*buf));
390 :
391 : /*
392 : * A single byte read is atomic. There could be memory-ordering effects
393 : * here, but for performance reasons we make it the caller's job to worry
394 : * about that.
395 : */
396 451351 : result = ((map[mapByte] >> mapOffset) & VISIBILITYMAP_VALID_BITS);
397 451351 : return result;
398 : }
399 :
400 : /*
401 : * visibilitymap_count - count number of bits set in visibility map
402 : *
403 : * Note: we ignore the possibility of race conditions when the table is being
404 : * extended concurrently with the call. New pages added to the table aren't
405 : * going to be marked all-visible or all-frozen, so they won't affect the result.
406 : */
407 : void
408 2090 : visibilitymap_count(Relation rel, BlockNumber *all_visible, BlockNumber *all_frozen)
409 : {
410 : BlockNumber mapBlock;
411 :
412 : /* all_visible must be specified */
413 2090 : Assert(all_visible);
414 :
415 2090 : *all_visible = 0;
416 2090 : if (all_frozen)
417 0 : *all_frozen = 0;
418 :
419 2407 : for (mapBlock = 0;; mapBlock++)
420 : {
421 : Buffer mapBuffer;
422 : unsigned char *map;
423 : int i;
424 :
425 : /*
426 : * Read till we fall off the end of the map. We assume that any extra
427 : * bytes in the last page are zeroed, so we don't bother excluding
428 : * them from the count.
429 : */
430 2407 : mapBuffer = vm_readbuf(rel, mapBlock, false);
431 2407 : if (!BufferIsValid(mapBuffer))
432 2090 : break;
433 :
434 : /*
435 : * We choose not to lock the page, since the result is going to be
436 : * immediately stale anyway if anyone is concurrently setting or
437 : * clearing bits, and we only really need an approximate value.
438 : */
439 317 : map = (unsigned char *) PageGetContents(BufferGetPage(mapBuffer));
440 :
441 2589573 : for (i = 0; i < MAPSIZE; i++)
442 : {
443 2589256 : *all_visible += number_of_ones_for_visible[map[i]];
444 2589256 : if (all_frozen)
445 0 : *all_frozen += number_of_ones_for_frozen[map[i]];
446 : }
447 :
448 317 : ReleaseBuffer(mapBuffer);
449 317 : }
450 2090 : }
451 :
452 : /*
453 : * visibilitymap_truncate - truncate the visibility map
454 : *
455 : * The caller must hold AccessExclusiveLock on the relation, to ensure that
456 : * other backends receive the smgr invalidation event that this function sends
457 : * before they access the VM again.
458 : *
459 : * nheapblocks is the new size of the heap.
460 : */
461 : void
462 12 : visibilitymap_truncate(Relation rel, BlockNumber nheapblocks)
463 : {
464 : BlockNumber newnblocks;
465 :
466 : /* last remaining block, byte, and bit */
467 12 : BlockNumber truncBlock = HEAPBLK_TO_MAPBLOCK(nheapblocks);
468 12 : uint32 truncByte = HEAPBLK_TO_MAPBYTE(nheapblocks);
469 12 : uint8 truncOffset = HEAPBLK_TO_OFFSET(nheapblocks);
470 :
471 : #ifdef TRACE_VISIBILITYMAP
472 : elog(DEBUG1, "vm_truncate %s %d", RelationGetRelationName(rel), nheapblocks);
473 : #endif
474 :
475 12 : RelationOpenSmgr(rel);
476 :
477 : /*
478 : * If no visibility map has been created yet for this relation, there's
479 : * nothing to truncate.
480 : */
481 12 : if (!smgrexists(rel->rd_smgr, VISIBILITYMAP_FORKNUM))
482 0 : return;
483 :
484 : /*
485 : * Unless the new size is exactly at a visibility map page boundary, the
486 : * tail bits in the last remaining map page, representing truncated heap
487 : * blocks, need to be cleared. This is not only tidy, but also necessary
488 : * because we don't get a chance to clear the bits if the heap is extended
489 : * again.
490 : */
491 12 : if (truncByte != 0 || truncOffset != 0)
492 8 : {
493 : Buffer mapBuffer;
494 : Page page;
495 : char *map;
496 :
497 8 : newnblocks = truncBlock + 1;
498 :
499 8 : mapBuffer = vm_readbuf(rel, truncBlock, false);
500 8 : if (!BufferIsValid(mapBuffer))
501 : {
502 : /* nothing to do, the file was already smaller */
503 0 : return;
504 : }
505 :
506 8 : page = BufferGetPage(mapBuffer);
507 8 : map = PageGetContents(page);
508 :
509 8 : LockBuffer(mapBuffer, BUFFER_LOCK_EXCLUSIVE);
510 :
511 : /* NO EREPORT(ERROR) from here till changes are logged */
512 8 : START_CRIT_SECTION();
513 :
514 : /* Clear out the unwanted bytes. */
515 8 : MemSet(&map[truncByte + 1], 0, MAPSIZE - (truncByte + 1));
516 :
517 : /*----
518 : * Mask out the unwanted bits of the last remaining byte.
519 : *
520 : * ((1 << 0) - 1) = 00000000
521 : * ((1 << 1) - 1) = 00000001
522 : * ...
523 : * ((1 << 6) - 1) = 00111111
524 : * ((1 << 7) - 1) = 01111111
525 : *----
526 : */
527 8 : map[truncByte] &= (1 << truncOffset) - 1;
528 :
529 : /*
530 : * Truncation of a relation is WAL-logged at a higher-level, and we
531 : * will be called at WAL replay. But if checksums are enabled, we need
532 : * to still write a WAL record to protect against a torn page, if the
533 : * page is flushed to disk before the truncation WAL record. We cannot
534 : * use MarkBufferDirtyHint here, because that will not dirty the page
535 : * during recovery.
536 : */
537 8 : MarkBufferDirty(mapBuffer);
538 8 : if (!InRecovery && RelationNeedsWAL(rel) && XLogHintBitIsNeeded())
539 0 : log_newpage_buffer(mapBuffer, false);
540 :
541 8 : END_CRIT_SECTION();
542 :
543 8 : UnlockReleaseBuffer(mapBuffer);
544 : }
545 : else
546 4 : newnblocks = truncBlock;
547 :
548 12 : if (smgrnblocks(rel->rd_smgr, VISIBILITYMAP_FORKNUM) <= newnblocks)
549 : {
550 : /* nothing to do, the file was already smaller than requested size */
551 8 : return;
552 : }
553 :
554 : /* Truncate the unused VM pages, and send smgr inval message */
555 4 : smgrtruncate(rel->rd_smgr, VISIBILITYMAP_FORKNUM, newnblocks);
556 :
557 : /*
558 : * We might as well update the local smgr_vm_nblocks setting. smgrtruncate
559 : * sent an smgr cache inval message, which will cause other backends to
560 : * invalidate their copy of smgr_vm_nblocks, and this one too at the next
561 : * command boundary. But this ensures it isn't outright wrong until then.
562 : */
563 4 : if (rel->rd_smgr)
564 4 : rel->rd_smgr->smgr_vm_nblocks = newnblocks;
565 : }
566 :
567 : /*
568 : * Read a visibility map page.
569 : *
570 : * If the page doesn't exist, InvalidBuffer is returned, or if 'extend' is
571 : * true, the visibility map file is extended.
572 : */
573 : static Buffer
574 148584 : vm_readbuf(Relation rel, BlockNumber blkno, bool extend)
575 : {
576 : Buffer buf;
577 :
578 : /*
579 : * We might not have opened the relation at the smgr level yet, or we
580 : * might have been forced to close it by a sinval message. The code below
581 : * won't necessarily notice relation extension immediately when extend =
582 : * false, so we rely on sinval messages to ensure that our ideas about the
583 : * size of the map aren't too far out of date.
584 : */
585 148584 : RelationOpenSmgr(rel);
586 :
587 : /*
588 : * If we haven't cached the size of the visibility map fork yet, check it
589 : * first.
590 : */
591 148584 : if (rel->rd_smgr->smgr_vm_nblocks == InvalidBlockNumber)
592 : {
593 2512 : if (smgrexists(rel->rd_smgr, VISIBILITYMAP_FORKNUM))
594 829 : rel->rd_smgr->smgr_vm_nblocks = smgrnblocks(rel->rd_smgr,
595 : VISIBILITYMAP_FORKNUM);
596 : else
597 1683 : rel->rd_smgr->smgr_vm_nblocks = 0;
598 : }
599 :
600 : /* Handle requests beyond EOF */
601 148584 : if (blkno >= rel->rd_smgr->smgr_vm_nblocks)
602 : {
603 146991 : if (extend)
604 177 : vm_extend(rel, blkno + 1);
605 : else
606 146814 : return InvalidBuffer;
607 : }
608 :
609 : /*
610 : * Use ZERO_ON_ERROR mode, and initialize the page if necessary. It's
611 : * always safe to clear bits, so it's better to clear corrupt pages than
612 : * error out.
613 : */
614 1770 : buf = ReadBufferExtended(rel, VISIBILITYMAP_FORKNUM, blkno,
615 : RBM_ZERO_ON_ERROR, NULL);
616 1770 : if (PageIsNew(BufferGetPage(buf)))
617 0 : PageInit(BufferGetPage(buf), BLCKSZ, 0);
618 1770 : return buf;
619 : }
620 :
621 : /*
622 : * Ensure that the visibility map fork is at least vm_nblocks long, extending
623 : * it if necessary with zeroed pages.
624 : */
625 : static void
626 177 : vm_extend(Relation rel, BlockNumber vm_nblocks)
627 : {
628 : BlockNumber vm_nblocks_now;
629 : Page pg;
630 :
631 177 : pg = (Page) palloc(BLCKSZ);
632 177 : PageInit(pg, BLCKSZ, 0);
633 :
634 : /*
635 : * We use the relation extension lock to lock out other backends trying to
636 : * extend the visibility map at the same time. It also locks out extension
637 : * of the main fork, unnecessarily, but extending the visibility map
638 : * happens seldom enough that it doesn't seem worthwhile to have a
639 : * separate lock tag type for it.
640 : *
641 : * Note that another backend might have extended or created the relation
642 : * by the time we get the lock.
643 : */
644 177 : LockRelationForExtension(rel, ExclusiveLock);
645 :
646 : /* Might have to re-open if a cache flush happened */
647 177 : RelationOpenSmgr(rel);
648 :
649 : /*
650 : * Create the file first if it doesn't exist. If smgr_vm_nblocks is
651 : * positive then it must exist, no need for an smgrexists call.
652 : */
653 177 : if ((rel->rd_smgr->smgr_vm_nblocks == 0 ||
654 177 : rel->rd_smgr->smgr_vm_nblocks == InvalidBlockNumber) &&
655 177 : !smgrexists(rel->rd_smgr, VISIBILITYMAP_FORKNUM))
656 126 : smgrcreate(rel->rd_smgr, VISIBILITYMAP_FORKNUM, false);
657 :
658 177 : vm_nblocks_now = smgrnblocks(rel->rd_smgr, VISIBILITYMAP_FORKNUM);
659 :
660 : /* Now extend the file */
661 531 : while (vm_nblocks_now < vm_nblocks)
662 : {
663 177 : PageSetChecksumInplace(pg, vm_nblocks_now);
664 :
665 177 : smgrextend(rel->rd_smgr, VISIBILITYMAP_FORKNUM, vm_nblocks_now,
666 : (char *) pg, false);
667 177 : vm_nblocks_now++;
668 : }
669 :
670 : /*
671 : * Send a shared-inval message to force other backends to close any smgr
672 : * references they may have for this rel, which we are about to change.
673 : * This is a useful optimization because it means that backends don't have
674 : * to keep checking for creation or extension of the file, which happens
675 : * infrequently.
676 : */
677 177 : CacheInvalidateSmgr(rel->rd_smgr->smgr_rnode);
678 :
679 : /* Update local cache with the up-to-date size */
680 177 : rel->rd_smgr->smgr_vm_nblocks = vm_nblocks_now;
681 :
682 177 : UnlockRelationForExtension(rel, ExclusiveLock);
683 :
684 177 : pfree(pg);
685 177 : }
|
