| 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_MAPBLOCK_LIMIT(x) \ |
| 119 |
(((x) + HEAPBLOCKS_PER_PAGE - 1) / HEAPBLOCKS_PER_PAGE) |
| 120 |
#define HEAPBLK_TO_MAPBYTE(x) (((x) % HEAPBLOCKS_PER_PAGE) / HEAPBLOCKS_PER_BYTE) |
| 121 |
#define HEAPBLK_TO_OFFSET(x) (((x) % HEAPBLOCKS_PER_BYTE) * BITS_PER_HEAPBLOCK) |
| 122 |
|
| 123 |
/* Masks for counting subsets of bits in the visibility map. */ |
| 124 |
#define VISIBLE_MASK8 (0x55) /* The lower bit of each bit pair */ |
| 125 |
#define FROZEN_MASK8 (0xaa) /* The upper bit of each bit pair */ |
| 126 |
|
| 127 |
/* prototypes for internal routines */ |
| 128 |
static Buffer vm_readbuf(Relation rel, BlockNumber blkno, bool extend); |
| 129 |
static Buffer vm_extend(Relation rel, BlockNumber vm_nblocks); |
| 130 |
|
| 131 |
|
| 132 |
/* |
| 133 |
* visibilitymap_clear - clear specified bits for one page in visibility map |
| 134 |
* |
| 135 |
* You must pass a buffer containing the correct map page to this function. |
| 136 |
* Call visibilitymap_pin first to pin the right one. This function doesn't do |
| 137 |
* any I/O. Returns true if any bits have been cleared and false otherwise. |
| 138 |
* |
| 139 |
* Most callers should use visibilitymap_clear_locked rather than this |
| 140 |
* function. It is usually necessary to register the VM buffer in the WAL |
| 141 |
* record, and this necessitates holding the lock for longer than it is held |
| 142 |
* here. However, we retain this function for behavioral compatibility on the |
| 143 |
* back branches (out-of-tree callers may expect it to take the lock). And, |
| 144 |
* since we have it, we use it for in-tree callers that don't have to manage |
| 145 |
* the VM buffer lock themselves. |
| 146 |
*/ |
| 147 |
bool |
| 148 |
visibilitymap_clear(Relation rel, BlockNumber heapBlk, Buffer vmbuf, uint8 flags) |
| 149 |
{ |
| 150 |
bool cleared = false; |
| 151 |
|
| 152 |
LockBuffer(vmbuf, BUFFER_LOCK_EXCLUSIVE); |
| 153 |
|
| 154 |
cleared = visibilitymap_clear_locked(rel, heapBlk, vmbuf, flags); |
| 155 |
|
| 156 |
LockBuffer(vmbuf, BUFFER_LOCK_UNLOCK); |
| 157 |
|
| 158 |
return cleared; |
| 159 |
} |
| 160 |
|
| 161 |
/* |
| 162 |
* Clear specified bits, caller holds VM lock |
| 163 |
* |
| 164 |
* Like visibilitymap_clear(), except the caller must already hold the VM |
| 165 |
* buffer exclusive lock and is responsible for unlocking it. |
| 166 |
* Returns true if any bits were actually cleared, false otherwise. |
| 167 |
*/ |
| 168 |
bool |
| 169 |
visibilitymap_clear_locked(Relation rel, BlockNumber heapBlk, Buffer vmbuf, uint8 flags) |
| 170 |
{ |
| 171 |
BlockNumber mapBlock = HEAPBLK_TO_MAPBLOCK(heapBlk); |
| 172 |
int mapByte = HEAPBLK_TO_MAPBYTE(heapBlk); |
| 173 |
int mapOffset = HEAPBLK_TO_OFFSET(heapBlk); |
| 174 |
uint8 mask = flags << mapOffset; |
| 175 |
char *map; |
| 176 |
bool cleared = false; |
| 177 |
|
| 178 |
/* Must never clear all_visible bit while leaving all_frozen bit set */ |
| 179 |
Assert(flags & VISIBILITYMAP_VALID_BITS); |
| 180 |
Assert(flags != VISIBILITYMAP_ALL_VISIBLE); |
| 181 |
|
| 182 |
#ifdef TRACE_VISIBILITYMAP |
| 183 |
elog(DEBUG1, "vm_clear %s %d", RelationGetRelationName(rel), heapBlk); |
| 184 |
#endif |
| 185 |
|
| 186 |
if (!BufferIsValid(vmbuf) || BufferGetBlockNumber(vmbuf) != mapBlock) |
| 187 |
elog(ERROR, "wrong buffer passed to visibilitymap_clear"); |
| 188 |
|
| 189 |
Assert(BufferIsExclusiveLocked(vmbuf)); |
| 190 |
|
| 191 |
map = PageGetContents(BufferGetPage(vmbuf)); |
| 192 |
|
| 193 |
if (map[mapByte] & mask) |
| 194 |
{ |
| 195 |
map[mapByte] &= ~mask; |
| 196 |
|
| 197 |
MarkBufferDirty(vmbuf); |
| 198 |
cleared = true; |
| 199 |
} |
| 200 |
|
| 201 |
return cleared; |
| 202 |
} |
| 203 |
|
| 204 |
/* |
| 205 |
* visibilitymap_pin - pin a map page for setting a bit |
| 206 |
* |
| 207 |
* Setting a bit in the visibility map is a two-phase operation. First, call |
| 208 |
* visibilitymap_pin, to pin the visibility map page containing the bit for |
| 209 |
* the heap page. Because that can require I/O to read the map page, you |
| 210 |
* shouldn't hold a lock on the heap page while doing that. Then, call |
| 211 |
* visibilitymap_set to actually set the bit. |
| 212 |
* |
| 213 |
* On entry, *vmbuf should be InvalidBuffer or a valid buffer returned by |
| 214 |
* an earlier call to visibilitymap_pin or visibilitymap_get_status on the same |
| 215 |
* relation. On return, *vmbuf is a valid buffer with the map page containing |
| 216 |
* the bit for heapBlk. |
| 217 |
* |
| 218 |
* If the page doesn't exist in the map file yet, it is extended. |
| 219 |
*/ |
| 220 |
void |
| 221 |
visibilitymap_pin(Relation rel, BlockNumber heapBlk, Buffer *vmbuf) |
| 222 |
{ |
| 223 |
BlockNumber mapBlock = HEAPBLK_TO_MAPBLOCK(heapBlk); |
| 224 |
|
| 225 |
/* Reuse the old pinned buffer if possible */ |
| 226 |
if (BufferIsValid(*vmbuf)) |
| 227 |
{ |
| 228 |
if (BufferGetBlockNumber(*vmbuf) == mapBlock) |
| 229 |
return; |
| 230 |
|
| 231 |
ReleaseBuffer(*vmbuf); |
| 232 |
} |
| 233 |
*vmbuf = vm_readbuf(rel, mapBlock, true); |
| 234 |
} |
| 235 |
|
| 236 |
/* |
| 237 |
* visibilitymap_pin_ok - do we already have the correct page pinned? |
| 238 |
* |
| 239 |
* On entry, vmbuf should be InvalidBuffer or a valid buffer returned by |
| 240 |
* an earlier call to visibilitymap_pin or visibilitymap_get_status on the same |
| 241 |
* relation. The return value indicates whether the buffer covers the |
| 242 |
* given heapBlk. |
| 243 |
*/ |
| 244 |
bool |
| 245 |
visibilitymap_pin_ok(BlockNumber heapBlk, Buffer vmbuf) |
| 246 |
{ |
| 247 |
BlockNumber mapBlock = HEAPBLK_TO_MAPBLOCK(heapBlk); |
| 248 |
|
| 249 |
return BufferIsValid(vmbuf) && BufferGetBlockNumber(vmbuf) == mapBlock; |
| 250 |
} |
| 251 |
|
| 252 |
/* |
| 253 |
* visibilitymap_set - set bit(s) on a previously pinned page |
| 254 |
* |
| 255 |
* recptr is the LSN of the XLOG record we're replaying, if we're in recovery, |
| 256 |
* or InvalidXLogRecPtr in normal running. The VM page LSN is advanced to the |
| 257 |
* one provided; in normal running, we generate a new XLOG record and set the |
| 258 |
* page LSN to that value (though the heap page's LSN may *not* be updated; |
| 259 |
* see below). cutoff_xid is the largest xmin on the page being marked |
| 260 |
* all-visible; it is needed for Hot Standby, and can be InvalidTransactionId |
| 261 |
* if the page contains no tuples. It can also be set to InvalidTransactionId |
| 262 |
* when a page that is already all-visible is being marked all-frozen. |
| 263 |
* |
| 264 |
* Caller is expected to set the heap page's PD_ALL_VISIBLE bit before calling |
| 265 |
* this function. Except in recovery, caller should also pass the heap |
| 266 |
* buffer. When checksums are enabled and we're not in recovery, we must add |
| 267 |
* the heap buffer to the WAL chain to protect it from being torn. |
| 268 |
* |
| 269 |
* You must pass a buffer containing the correct map page to this function. |
| 270 |
* Call visibilitymap_pin first to pin the right one. This function doesn't do |
| 271 |
* any I/O. |
| 272 |
* |
| 273 |
* Returns the state of the page's VM bits before setting flags. |
| 274 |
*/ |
| 275 |
uint8 |
| 276 |
visibilitymap_set(Relation rel, BlockNumber heapBlk, Buffer heapBuf, |
| 277 |
XLogRecPtr recptr, Buffer vmBuf, TransactionId cutoff_xid, |
| 278 |
uint8 flags) |
| 279 |
{ |
| 280 |
BlockNumber mapBlock = HEAPBLK_TO_MAPBLOCK(heapBlk); |
| 281 |
uint32 mapByte = HEAPBLK_TO_MAPBYTE(heapBlk); |
| 282 |
uint8 mapOffset = HEAPBLK_TO_OFFSET(heapBlk); |
| 283 |
Page page; |
| 284 |
uint8 *map; |
| 285 |
uint8 status; |
| 286 |
|
| 287 |
#ifdef TRACE_VISIBILITYMAP |
| 288 |
elog(DEBUG1, "vm_set %s %d", RelationGetRelationName(rel), heapBlk); |
| 289 |
#endif |
| 290 |
|
| 291 |
Assert(InRecovery || XLogRecPtrIsInvalid(recptr)); |
| 292 |
Assert(InRecovery || PageIsAllVisible((Page) BufferGetPage(heapBuf))); |
| 293 |
Assert((flags & VISIBILITYMAP_VALID_BITS) == flags); |
| 294 |
|
| 295 |
/* Must never set all_frozen bit without also setting all_visible bit */ |
| 296 |
Assert(flags != VISIBILITYMAP_ALL_FROZEN); |
| 297 |
|
| 298 |
/* Check that we have the right heap page pinned, if present */ |
| 299 |
if (BufferIsValid(heapBuf) && BufferGetBlockNumber(heapBuf) != heapBlk) |
| 300 |
elog(ERROR, "wrong heap buffer passed to visibilitymap_set"); |
| 301 |
|
| 302 |
/* Check that we have the right VM page pinned */ |
| 303 |
if (!BufferIsValid(vmBuf) || BufferGetBlockNumber(vmBuf) != mapBlock) |
| 304 |
elog(ERROR, "wrong VM buffer passed to visibilitymap_set"); |
| 305 |
|
| 306 |
page = BufferGetPage(vmBuf); |
| 307 |
map = (uint8 *) PageGetContents(page); |
| 308 |
LockBuffer(vmBuf, BUFFER_LOCK_EXCLUSIVE); |
| 309 |
|
| 310 |
status = (map[mapByte] >> mapOffset) & VISIBILITYMAP_VALID_BITS; |
| 311 |
if (flags != status) |
| 312 |
{ |
| 313 |
START_CRIT_SECTION(); |
| 314 |
|
| 315 |
map[mapByte] |= (flags << mapOffset); |
| 316 |
MarkBufferDirty(vmBuf); |
| 317 |
|
| 318 |
if (RelationNeedsWAL(rel)) |
| 319 |
{ |
| 320 |
if (XLogRecPtrIsInvalid(recptr)) |
| 321 |
{ |
| 322 |
Assert(!InRecovery); |
| 323 |
recptr = log_heap_visible(rel, heapBuf, vmBuf, cutoff_xid, flags); |
| 324 |
|
| 325 |
/* |
| 326 |
* If data checksums are enabled (or wal_log_hints=on), we |
| 327 |
* need to protect the heap page from being torn. |
| 328 |
* |
| 329 |
* If not, then we must *not* update the heap page's LSN. In |
| 330 |
* this case, the FPI for the heap page was omitted from the |
| 331 |
* WAL record inserted above, so it would be incorrect to |
| 332 |
* update the heap page's LSN. |
| 333 |
*/ |
| 334 |
if (XLogHintBitIsNeeded()) |
| 335 |
{ |
| 336 |
Page heapPage = BufferGetPage(heapBuf); |
| 337 |
|
| 338 |
PageSetLSN(heapPage, recptr); |
| 339 |
} |
| 340 |
} |
| 341 |
PageSetLSN(page, recptr); |
| 342 |
} |
| 343 |
|
| 344 |
END_CRIT_SECTION(); |
| 345 |
} |
| 346 |
|
| 347 |
LockBuffer(vmBuf, BUFFER_LOCK_UNLOCK); |
| 348 |
return status; |
| 349 |
} |
| 350 |
|
| 351 |
/* |
| 352 |
* visibilitymap_get_status - get status of bits |
| 353 |
* |
| 354 |
* Are all tuples on heapBlk visible to all or are marked frozen, according |
| 355 |
* to the visibility map? |
| 356 |
* |
| 357 |
* On entry, *vmbuf should be InvalidBuffer or a valid buffer returned by an |
| 358 |
* earlier call to visibilitymap_pin or visibilitymap_get_status on the same |
| 359 |
* relation. On return, *vmbuf is a valid buffer with the map page containing |
| 360 |
* the bit for heapBlk, or InvalidBuffer. The caller is responsible for |
| 361 |
* releasing *vmbuf after it's done testing and setting bits. |
| 362 |
* |
| 363 |
* NOTE: This function is typically called without a lock on the heap page, |
| 364 |
* so somebody else could change the bit just after we look at it. In fact, |
| 365 |
* since we don't lock the visibility map page either, it's even possible that |
| 366 |
* someone else could have changed the bit just before we look at it, but yet |
| 367 |
* we might see the old value. It is the caller's responsibility to deal with |
| 368 |
* all concurrency issues! |
| 369 |
*/ |
| 370 |
uint8 |
| 371 |
visibilitymap_get_status(Relation rel, BlockNumber heapBlk, Buffer *vmbuf) |
| 372 |
{ |
| 373 |
BlockNumber mapBlock = HEAPBLK_TO_MAPBLOCK(heapBlk); |
| 374 |
uint32 mapByte = HEAPBLK_TO_MAPBYTE(heapBlk); |
| 375 |
uint8 mapOffset = HEAPBLK_TO_OFFSET(heapBlk); |
| 376 |
char *map; |
| 377 |
uint8 result; |
| 378 |
|
| 379 |
#ifdef TRACE_VISIBILITYMAP |
| 380 |
elog(DEBUG1, "vm_get_status %s %d", RelationGetRelationName(rel), heapBlk); |
| 381 |
#endif |
| 382 |
|
| 383 |
/* Reuse the old pinned buffer if possible */ |
| 384 |
if (BufferIsValid(*vmbuf)) |
| 385 |
{ |
| 386 |
if (BufferGetBlockNumber(*vmbuf) != mapBlock) |
| 387 |
{ |
| 388 |
ReleaseBuffer(*vmbuf); |
| 389 |
*vmbuf = InvalidBuffer; |
| 390 |
} |
| 391 |
} |
| 392 |
|
| 393 |
if (!BufferIsValid(*vmbuf)) |
| 394 |
{ |
| 395 |
*vmbuf = vm_readbuf(rel, mapBlock, false); |
| 396 |
if (!BufferIsValid(*vmbuf)) |
| 397 |
return false; |
| 398 |
} |
| 399 |
|
| 400 |
map = PageGetContents(BufferGetPage(*vmbuf)); |
| 401 |
|
| 402 |
/* |
| 403 |
* A single byte read is atomic. There could be memory-ordering effects |
| 404 |
* here, but for performance reasons we make it the caller's job to worry |
| 405 |
* about that. |
| 406 |
*/ |
| 407 |
result = ((map[mapByte] >> mapOffset) & VISIBILITYMAP_VALID_BITS); |
| 408 |
return result; |
| 409 |
} |
| 410 |
|
| 411 |
/* |
| 412 |
* visibilitymap_count - count number of bits set in visibility map |
| 413 |
* |
| 414 |
* Note: we ignore the possibility of race conditions when the table is being |
| 415 |
* extended concurrently with the call. New pages added to the table aren't |
| 416 |
* going to be marked all-visible or all-frozen, so they won't affect the result. |
| 417 |
*/ |
| 418 |
void |
| 419 |
visibilitymap_count(Relation rel, BlockNumber *all_visible, BlockNumber *all_frozen) |
| 420 |
{ |
| 421 |
BlockNumber mapBlock; |
| 422 |
BlockNumber nvisible = 0; |
| 423 |
BlockNumber nfrozen = 0; |
| 424 |
|
| 425 |
/* all_visible must be specified */ |
| 426 |
Assert(all_visible); |
| 427 |
|
| 428 |
for (mapBlock = 0;; mapBlock++) |
| 429 |
{ |
| 430 |
Buffer mapBuffer; |
| 431 |
uint64 *map; |
| 432 |
|
| 433 |
/* |
| 434 |
* Read till we fall off the end of the map. We assume that any extra |
| 435 |
* bytes in the last page are zeroed, so we don't bother excluding |
| 436 |
* them from the count. |
| 437 |
*/ |
| 438 |
mapBuffer = vm_readbuf(rel, mapBlock, false); |
| 439 |
if (!BufferIsValid(mapBuffer)) |
| 440 |
break; |
| 441 |
|
| 442 |
/* |
| 443 |
* We choose not to lock the page, since the result is going to be |
| 444 |
* immediately stale anyway if anyone is concurrently setting or |
| 445 |
* clearing bits, and we only really need an approximate value. |
| 446 |
*/ |
| 447 |
map = (uint64 *) PageGetContents(BufferGetPage(mapBuffer)); |
| 448 |
|
| 449 |
nvisible += pg_popcount_masked((const char *) map, MAPSIZE, VISIBLE_MASK8); |
| 450 |
if (all_frozen) |
| 451 |
nfrozen += pg_popcount_masked((const char *) map, MAPSIZE, FROZEN_MASK8); |
| 452 |
|
| 453 |
ReleaseBuffer(mapBuffer); |
| 454 |
} |
| 455 |
|
| 456 |
*all_visible = nvisible; |
| 457 |
if (all_frozen) |
| 458 |
*all_frozen = nfrozen; |
| 459 |
} |
| 460 |
|
| 461 |
/* |
| 462 |
* visibilitymap_prepare_truncate - |
| 463 |
* prepare for truncation of the visibility map |
| 464 |
* |
| 465 |
* nheapblocks is the new size of the heap. |
| 466 |
* |
| 467 |
* Return the number of blocks of new visibility map. |
| 468 |
* If it's InvalidBlockNumber, there is nothing to truncate; |
| 469 |
* otherwise the caller is responsible for calling smgrtruncate() |
| 470 |
* to truncate the visibility map pages. |
| 471 |
*/ |
| 472 |
BlockNumber |
| 473 |
visibilitymap_prepare_truncate(Relation rel, BlockNumber nheapblocks) |
| 474 |
{ |
| 475 |
BlockNumber newnblocks; |
| 476 |
|
| 477 |
/* last remaining block, byte, and bit */ |
| 478 |
BlockNumber truncBlock = HEAPBLK_TO_MAPBLOCK(nheapblocks); |
| 479 |
uint32 truncByte = HEAPBLK_TO_MAPBYTE(nheapblocks); |
| 480 |
uint8 truncOffset = HEAPBLK_TO_OFFSET(nheapblocks); |
| 481 |
|
| 482 |
#ifdef TRACE_VISIBILITYMAP |
| 483 |
elog(DEBUG1, "vm_truncate %s %d", RelationGetRelationName(rel), nheapblocks); |
| 484 |
#endif |
| 485 |
|
| 486 |
/* |
| 487 |
* If no visibility map has been created yet for this relation, there's |
| 488 |
* nothing to truncate. |
| 489 |
*/ |
| 490 |
if (!smgrexists(RelationGetSmgr(rel), VISIBILITYMAP_FORKNUM)) |
| 491 |
return InvalidBlockNumber; |
| 492 |
|
| 493 |
/* |
| 494 |
* Unless the new size is exactly at a visibility map page boundary, the |
| 495 |
* tail bits in the last remaining map page, representing truncated heap |
| 496 |
* blocks, need to be cleared. This is not only tidy, but also necessary |
| 497 |
* because we don't get a chance to clear the bits if the heap is extended |
| 498 |
* again. |
| 499 |
*/ |
| 500 |
if (truncByte != 0 || truncOffset != 0) |
| 501 |
{ |
| 502 |
Buffer mapBuffer; |
| 503 |
Page page; |
| 504 |
char *map; |
| 505 |
|
| 506 |
newnblocks = truncBlock + 1; |
| 507 |
|
| 508 |
mapBuffer = vm_readbuf(rel, truncBlock, false); |
| 509 |
if (!BufferIsValid(mapBuffer)) |
| 510 |
{ |
| 511 |
/* nothing to do, the file was already smaller */ |
| 512 |
return InvalidBlockNumber; |
| 513 |
} |
| 514 |
|
| 515 |
page = BufferGetPage(mapBuffer); |
| 516 |
map = PageGetContents(page); |
| 517 |
|
| 518 |
LockBuffer(mapBuffer, BUFFER_LOCK_EXCLUSIVE); |
| 519 |
|
| 520 |
/* NO EREPORT(ERROR) from here till changes are logged */ |
| 521 |
START_CRIT_SECTION(); |
| 522 |
|
| 523 |
/* Clear out the unwanted bytes. */ |
| 524 |
MemSet(&map[truncByte + 1], 0, MAPSIZE - (truncByte + 1)); |
| 525 |
|
| 526 |
/*---- |
| 527 |
* Mask out the unwanted bits of the last remaining byte. |
| 528 |
* |
| 529 |
* ((1 << 0) - 1) = 00000000 |
| 530 |
* ((1 << 1) - 1) = 00000001 |
| 531 |
* ... |
| 532 |
* ((1 << 6) - 1) = 00111111 |
| 533 |
* ((1 << 7) - 1) = 01111111 |
| 534 |
*---- |
| 535 |
*/ |
| 536 |
map[truncByte] &= (1 << truncOffset) - 1; |
| 537 |
|
| 538 |
/* |
| 539 |
* Truncation of a relation is WAL-logged at a higher-level, and we |
| 540 |
* will be called at WAL replay. But if checksums are enabled, we need |
| 541 |
* to still write a WAL record to protect against a torn page, if the |
| 542 |
* page is flushed to disk before the truncation WAL record. We cannot |
| 543 |
* use MarkBufferDirtyHint here, because that will not dirty the page |
| 544 |
* during recovery. |
| 545 |
*/ |
| 546 |
MarkBufferDirty(mapBuffer); |
| 547 |
if (!InRecovery && RelationNeedsWAL(rel) && XLogHintBitIsNeeded()) |
| 548 |
log_newpage_buffer(mapBuffer, false); |
| 549 |
|
| 550 |
END_CRIT_SECTION(); |
| 551 |
|
| 552 |
UnlockReleaseBuffer(mapBuffer); |
| 553 |
} |
| 554 |
else |
| 555 |
newnblocks = truncBlock; |
| 556 |
|
| 557 |
if (smgrnblocks(RelationGetSmgr(rel), VISIBILITYMAP_FORKNUM) <= newnblocks) |
| 558 |
{ |
| 559 |
/* nothing to do, the file was already smaller than requested size */ |
| 560 |
return InvalidBlockNumber; |
| 561 |
} |
| 562 |
|
| 563 |
return newnblocks; |
| 564 |
} |
| 565 |
|
| 566 |
/* |
| 567 |
* visibilitymap_truncation_length - |
| 568 |
* compute truncation length for visibility map |
| 569 |
* |
| 570 |
* Given a proposed truncation length for the main fork, compute the |
| 571 |
* correct truncation length for the visibility map. Should return the |
| 572 |
* same answer as visibilitymap_prepare_truncate(), but without modifying |
| 573 |
* anything. |
| 574 |
*/ |
| 575 |
BlockNumber |
| 576 |
visibilitymap_truncation_length(BlockNumber nheapblocks) |
| 577 |
{ |
| 578 |
return HEAPBLK_TO_MAPBLOCK_LIMIT(nheapblocks); |
| 579 |
} |
| 580 |
|
| 581 |
/* |
| 582 |
* Read a visibility map page. |
| 583 |
* |
| 584 |
* If the page doesn't exist, InvalidBuffer is returned, or if 'extend' is |
| 585 |
* true, the visibility map file is extended. |
| 586 |
*/ |
| 587 |
static Buffer |
| 588 |
vm_readbuf(Relation rel, BlockNumber blkno, bool extend) |
| 589 |
{ |
| 590 |
Buffer buf; |
| 591 |
SMgrRelation reln; |
| 592 |
|
| 593 |
/* |
| 594 |
* Caution: re-using this smgr pointer could fail if the relcache entry |
| 595 |
* gets closed. It's safe as long as we only do smgr-level operations |
| 596 |
* between here and the last use of the pointer. |
| 597 |
*/ |
| 598 |
reln = RelationGetSmgr(rel); |
| 599 |
|
| 600 |
/* |
| 601 |
* If we haven't cached the size of the visibility map fork yet, check it |
| 602 |
* first. |
| 603 |
*/ |
| 604 |
if (reln->smgr_cached_nblocks[VISIBILITYMAP_FORKNUM] == InvalidBlockNumber) |
| 605 |
{ |
| 606 |
if (smgrexists(reln, VISIBILITYMAP_FORKNUM)) |
| 607 |
smgrnblocks(reln, VISIBILITYMAP_FORKNUM); |
| 608 |
else |
| 609 |
reln->smgr_cached_nblocks[VISIBILITYMAP_FORKNUM] = 0; |
| 610 |
} |
| 611 |
|
| 612 |
/* |
| 613 |
* For reading we use ZERO_ON_ERROR mode, and initialize the page if |
| 614 |
* necessary. It's always safe to clear bits, so it's better to clear |
| 615 |
* corrupt pages than error out. |
| 616 |
* |
| 617 |
* We use the same path below to initialize pages when extending the |
| 618 |
* relation, as a concurrent extension can end up with vm_extend() |
| 619 |
* returning an already-initialized page. |
| 620 |
*/ |
| 621 |
if (blkno >= reln->smgr_cached_nblocks[VISIBILITYMAP_FORKNUM]) |
| 622 |
{ |
| 623 |
if (extend) |
| 624 |
buf = vm_extend(rel, blkno + 1); |
| 625 |
else |
| 626 |
return InvalidBuffer; |
| 627 |
} |
| 628 |
else |
| 629 |
buf = ReadBufferExtended(rel, VISIBILITYMAP_FORKNUM, blkno, |
| 630 |
RBM_ZERO_ON_ERROR, NULL); |
| 631 |
|
| 632 |
/* |
| 633 |
* Initializing the page when needed is trickier than it looks, because of |
| 634 |
* the possibility of multiple backends doing this concurrently, and our |
| 635 |
* desire to not uselessly take the buffer lock in the normal path where |
| 636 |
* the page is OK. We must take the lock to initialize the page, so |
| 637 |
* recheck page newness after we have the lock, in case someone else |
| 638 |
* already did it. Also, because we initially check PageIsNew with no |
| 639 |
* lock, it's possible to fall through and return the buffer while someone |
| 640 |
* else is still initializing the page (i.e., we might see pd_upper as set |
| 641 |
* but other page header fields are still zeroes). This is harmless for |
| 642 |
* callers that will take a buffer lock themselves, but some callers |
| 643 |
* inspect the page without any lock at all. The latter is OK only so |
| 644 |
* long as it doesn't depend on the page header having correct contents. |
| 645 |
* Current usage is safe because PageGetContents() does not require that. |
| 646 |
*/ |
| 647 |
if (PageIsNew(BufferGetPage(buf))) |
| 648 |
{ |
| 649 |
LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE); |
| 650 |
if (PageIsNew(BufferGetPage(buf))) |
| 651 |
PageInit(BufferGetPage(buf), BLCKSZ, 0); |
| 652 |
LockBuffer(buf, BUFFER_LOCK_UNLOCK); |
| 653 |
} |
| 654 |
return buf; |
| 655 |
} |
| 656 |
|
| 657 |
/* |
| 658 |
* Ensure that the visibility map fork is at least vm_nblocks long, extending |
| 659 |
* it if necessary with zeroed pages. |
| 660 |
*/ |
| 661 |
static Buffer |
| 662 |
vm_extend(Relation rel, BlockNumber vm_nblocks) |
| 663 |
{ |
| 664 |
Buffer buf; |
| 665 |
|
| 666 |
buf = ExtendBufferedRelTo(BMR_REL(rel), VISIBILITYMAP_FORKNUM, NULL, |
| 667 |
EB_CREATE_FORK_IF_NEEDED | |
| 668 |
EB_CLEAR_SIZE_CACHE, |
| 669 |
vm_nblocks, |
| 670 |
RBM_ZERO_ON_ERROR); |
| 671 |
|
| 672 |
/* |
| 673 |
* Send a shared-inval message to force other backends to close any smgr |
| 674 |
* references they may have for this rel, which we are about to change. |
| 675 |
* This is a useful optimization because it means that backends don't have |
| 676 |
* to keep checking for creation or extension of the file, which happens |
| 677 |
* infrequently. |
| 678 |
*/ |
| 679 |
CacheInvalidateSmgr(RelationGetSmgr(rel)->smgr_rlocator); |
| 680 |
|
| 681 |
return buf; |
| 682 |
} |