| 1 |
/*------------------------------------------------------------------------- |
| 2 |
* |
| 3 |
* bufpage.c |
| 4 |
* POSTGRES standard buffer page code. |
| 5 |
* |
| 6 |
* Portions Copyright (c) 1996-2025, PostgreSQL Global Development Group |
| 7 |
* Portions Copyright (c) 1994, Regents of the University of California |
| 8 |
* |
| 9 |
* |
| 10 |
* IDENTIFICATION |
| 11 |
* src/backend/storage/page/bufpage.c |
| 12 |
* |
| 13 |
*------------------------------------------------------------------------- |
| 14 |
*/ |
| 15 |
#include "postgres.h" |
| 16 |
|
| 17 |
#include "access/htup_details.h" |
| 18 |
#include "access/itup.h" |
| 19 |
#include "access/xlog.h" |
| 20 |
#include "pgstat.h" |
| 21 |
#include "storage/checksum.h" |
| 22 |
#include "utils/memdebug.h" |
| 23 |
#include "utils/memutils.h" |
| 24 |
|
| 25 |
|
| 26 |
/* GUC variable */ |
| 27 |
bool ignore_checksum_failure = false; |
| 28 |
|
| 29 |
|
| 30 |
/* ---------------------------------------------------------------- |
| 31 |
* Page support functions |
| 32 |
* ---------------------------------------------------------------- |
| 33 |
*/ |
| 34 |
|
| 35 |
/* |
| 36 |
* PageInit |
| 37 |
* Initializes the contents of a page. |
| 38 |
* Note that we don't calculate an initial checksum here; that's not done |
| 39 |
* until it's time to write. |
| 40 |
*/ |
| 41 |
void |
| 42 |
PageInit(Page page, Size pageSize, Size specialSize) |
| 43 |
{ |
| 44 |
PageHeader p = (PageHeader) page; |
| 45 |
|
| 46 |
specialSize = MAXALIGN(specialSize); |
| 47 |
|
| 48 |
Assert(pageSize == BLCKSZ); |
| 49 |
Assert(pageSize > specialSize + SizeOfPageHeaderData); |
| 50 |
|
| 51 |
/* Make sure all fields of page are zero, as well as unused space */ |
| 52 |
MemSet(p, 0, pageSize); |
| 53 |
|
| 54 |
p->pd_flags = 0; |
| 55 |
p->pd_lower = SizeOfPageHeaderData; |
| 56 |
p->pd_upper = pageSize - specialSize; |
| 57 |
p->pd_special = pageSize - specialSize; |
| 58 |
PageSetPageSizeAndVersion(page, pageSize, PG_PAGE_LAYOUT_VERSION); |
| 59 |
/* p->pd_prune_xid = InvalidTransactionId; done by above MemSet */ |
| 60 |
} |
| 61 |
|
| 62 |
|
| 63 |
/* |
| 64 |
* PageIsVerified |
| 65 |
* Check that the page header and checksum (if any) appear valid. |
| 66 |
* |
| 67 |
* This is called when a page has just been read in from disk. The idea is |
| 68 |
* to cheaply detect trashed pages before we go nuts following bogus line |
| 69 |
* pointers, testing invalid transaction identifiers, etc. |
| 70 |
* |
| 71 |
* It turns out to be necessary to allow zeroed pages here too. Even though |
| 72 |
* this routine is *not* called when deliberately adding a page to a relation, |
| 73 |
* there are scenarios in which a zeroed page might be found in a table. |
| 74 |
* (Example: a backend extends a relation, then crashes before it can write |
| 75 |
* any WAL entry about the new page. The kernel will already have the |
| 76 |
* zeroed page in the file, and it will stay that way after restart.) So we |
| 77 |
* allow zeroed pages here, and are careful that the page access macros |
| 78 |
* treat such a page as empty and without free space. Eventually, VACUUM |
| 79 |
* will clean up such a page and make it usable. |
| 80 |
* |
| 81 |
* If flag PIV_LOG_WARNING/PIV_LOG_LOG is set, a WARNING/LOG message is logged |
| 82 |
* in the event of a checksum failure. |
| 83 |
* |
| 84 |
* If flag PIV_IGNORE_CHECKSUM_FAILURE is set, checksum failures will cause a |
| 85 |
* message about the failure to be emitted, but will not cause |
| 86 |
* PageIsVerified() to return false. |
| 87 |
* |
| 88 |
* To allow the caller to report statistics about checksum failures, |
| 89 |
* *checksum_failure_p can be passed in. Note that there may be checksum |
| 90 |
* failures even if this function returns true, due to |
| 91 |
* PIV_IGNORE_CHECKSUM_FAILURE. |
| 92 |
*/ |
| 93 |
bool |
| 94 |
PageIsVerified(PageData *page, BlockNumber blkno, int flags, bool *checksum_failure_p) |
| 95 |
{ |
| 96 |
const PageHeaderData *p = (const PageHeaderData *) page; |
| 97 |
size_t *pagebytes; |
| 98 |
bool checksum_failure = false; |
| 99 |
bool header_sane = false; |
| 100 |
uint16 checksum = 0; |
| 101 |
|
| 102 |
if (checksum_failure_p) |
| 103 |
*checksum_failure_p = false; |
| 104 |
|
| 105 |
/* |
| 106 |
* Don't verify page data unless the page passes basic non-zero test |
| 107 |
*/ |
| 108 |
if (!PageIsNew(page)) |
| 109 |
{ |
| 110 |
if (DataChecksumsEnabled()) |
| 111 |
{ |
| 112 |
checksum = pg_checksum_page(page, blkno); |
| 113 |
|
| 114 |
if (checksum != p->pd_checksum) |
| 115 |
{ |
| 116 |
checksum_failure = true; |
| 117 |
if (checksum_failure_p) |
| 118 |
*checksum_failure_p = true; |
| 119 |
} |
| 120 |
} |
| 121 |
|
| 122 |
/* |
| 123 |
* The following checks don't prove the header is correct, only that |
| 124 |
* it looks sane enough to allow into the buffer pool. Later usage of |
| 125 |
* the block can still reveal problems, which is why we offer the |
| 126 |
* checksum option. |
| 127 |
*/ |
| 128 |
if ((p->pd_flags & ~PD_VALID_FLAG_BITS) == 0 && |
| 129 |
p->pd_lower <= p->pd_upper && |
| 130 |
p->pd_upper <= p->pd_special && |
| 131 |
p->pd_special <= BLCKSZ && |
| 132 |
p->pd_special == MAXALIGN(p->pd_special)) |
| 133 |
header_sane = true; |
| 134 |
|
| 135 |
if (header_sane && !checksum_failure) |
| 136 |
return true; |
| 137 |
} |
| 138 |
|
| 139 |
/* Check all-zeroes case */ |
| 140 |
pagebytes = (size_t *) page; |
| 141 |
|
| 142 |
if (pg_memory_is_all_zeros(pagebytes, BLCKSZ)) |
| 143 |
return true; |
| 144 |
|
| 145 |
/* |
| 146 |
* Throw a WARNING/LOG, as instructed by PIV_LOG_*, if the checksum fails, |
| 147 |
* but only after we've checked for the all-zeroes case. |
| 148 |
*/ |
| 149 |
if (checksum_failure) |
| 150 |
{ |
| 151 |
if ((flags & (PIV_LOG_WARNING | PIV_LOG_LOG)) != 0) |
| 152 |
ereport(flags & PIV_LOG_WARNING ? WARNING : LOG, |
| 153 |
(errcode(ERRCODE_DATA_CORRUPTED), |
| 154 |
errmsg("page verification failed, calculated checksum %u but expected %u", |
| 155 |
checksum, p->pd_checksum))); |
| 156 |
|
| 157 |
if (header_sane && (flags & PIV_IGNORE_CHECKSUM_FAILURE)) |
| 158 |
return true; |
| 159 |
} |
| 160 |
|
| 161 |
return false; |
| 162 |
} |
| 163 |
|
| 164 |
|
| 165 |
/* |
| 166 |
* PageAddItemExtended |
| 167 |
* |
| 168 |
* Add an item to a page. Return value is the offset at which it was |
| 169 |
* inserted, or InvalidOffsetNumber if the item is not inserted for any |
| 170 |
* reason. A WARNING is issued indicating the reason for the refusal. |
| 171 |
* |
| 172 |
* offsetNumber must be either InvalidOffsetNumber to specify finding a |
| 173 |
* free line pointer, or a value between FirstOffsetNumber and one past |
| 174 |
* the last existing item, to specify using that particular line pointer. |
| 175 |
* |
| 176 |
* If offsetNumber is valid and flag PAI_OVERWRITE is set, we just store |
| 177 |
* the item at the specified offsetNumber, which must be either a |
| 178 |
* currently-unused line pointer, or one past the last existing item. |
| 179 |
* |
| 180 |
* If offsetNumber is valid and flag PAI_OVERWRITE is not set, insert |
| 181 |
* the item at the specified offsetNumber, moving existing items later |
| 182 |
* in the array to make room. |
| 183 |
* |
| 184 |
* If offsetNumber is not valid, then assign a slot by finding the first |
| 185 |
* one that is both unused and deallocated. |
| 186 |
* |
| 187 |
* If flag PAI_IS_HEAP is set, we enforce that there can't be more than |
| 188 |
* MaxHeapTuplesPerPage line pointers on the page. |
| 189 |
* |
| 190 |
* !!! EREPORT(ERROR) IS DISALLOWED HERE !!! |
| 191 |
*/ |
| 192 |
OffsetNumber |
| 193 |
PageAddItemExtended(Page page, |
| 194 |
Item item, |
| 195 |
Size size, |
| 196 |
OffsetNumber offsetNumber, |
| 197 |
int flags) |
| 198 |
{ |
| 199 |
PageHeader phdr = (PageHeader) page; |
| 200 |
Size alignedSize; |
| 201 |
int lower; |
| 202 |
int upper; |
| 203 |
ItemId itemId; |
| 204 |
OffsetNumber limit; |
| 205 |
bool needshuffle = false; |
| 206 |
|
| 207 |
/* |
| 208 |
* Be wary about corrupted page pointers |
| 209 |
*/ |
| 210 |
if (phdr->pd_lower < SizeOfPageHeaderData || |
| 211 |
phdr->pd_lower > phdr->pd_upper || |
| 212 |
phdr->pd_upper > phdr->pd_special || |
| 213 |
phdr->pd_special > BLCKSZ) |
| 214 |
ereport(PANIC, |
| 215 |
(errcode(ERRCODE_DATA_CORRUPTED), |
| 216 |
errmsg("corrupted page pointers: lower = %u, upper = %u, special = %u", |
| 217 |
phdr->pd_lower, phdr->pd_upper, phdr->pd_special))); |
| 218 |
|
| 219 |
/* |
| 220 |
* Select offsetNumber to place the new item at |
| 221 |
*/ |
| 222 |
limit = OffsetNumberNext(PageGetMaxOffsetNumber(page)); |
| 223 |
|
| 224 |
/* was offsetNumber passed in? */ |
| 225 |
if (OffsetNumberIsValid(offsetNumber)) |
| 226 |
{ |
| 227 |
/* yes, check it */ |
| 228 |
if ((flags & PAI_OVERWRITE) != 0) |
| 229 |
{ |
| 230 |
if (offsetNumber < limit) |
| 231 |
{ |
| 232 |
itemId = PageGetItemId(page, offsetNumber); |
| 233 |
if (ItemIdIsUsed(itemId) || ItemIdHasStorage(itemId)) |
| 234 |
{ |
| 235 |
elog(WARNING, "will not overwrite a used ItemId"); |
| 236 |
return InvalidOffsetNumber; |
| 237 |
} |
| 238 |
} |
| 239 |
} |
| 240 |
else |
| 241 |
{ |
| 242 |
if (offsetNumber < limit) |
| 243 |
needshuffle = true; /* need to move existing linp's */ |
| 244 |
} |
| 245 |
} |
| 246 |
else |
| 247 |
{ |
| 248 |
/* offsetNumber was not passed in, so find a free slot */ |
| 249 |
/* if no free slot, we'll put it at limit (1st open slot) */ |
| 250 |
if (PageHasFreeLinePointers(page)) |
| 251 |
{ |
| 252 |
/* |
| 253 |
* Scan line pointer array to locate a "recyclable" (unused) |
| 254 |
* ItemId. |
| 255 |
* |
| 256 |
* Always use earlier items first. PageTruncateLinePointerArray |
| 257 |
* can only truncate unused items when they appear as a contiguous |
| 258 |
* group at the end of the line pointer array. |
| 259 |
*/ |
| 260 |
for (offsetNumber = FirstOffsetNumber; |
| 261 |
offsetNumber < limit; /* limit is maxoff+1 */ |
| 262 |
offsetNumber++) |
| 263 |
{ |
| 264 |
itemId = PageGetItemId(page, offsetNumber); |
| 265 |
|
| 266 |
/* |
| 267 |
* We check for no storage as well, just to be paranoid; |
| 268 |
* unused items should never have storage. Assert() that the |
| 269 |
* invariant is respected too. |
| 270 |
*/ |
| 271 |
Assert(ItemIdIsUsed(itemId) || !ItemIdHasStorage(itemId)); |
| 272 |
|
| 273 |
if (!ItemIdIsUsed(itemId) && !ItemIdHasStorage(itemId)) |
| 274 |
break; |
| 275 |
} |
| 276 |
if (offsetNumber >= limit) |
| 277 |
{ |
| 278 |
/* the hint is wrong, so reset it */ |
| 279 |
PageClearHasFreeLinePointers(page); |
| 280 |
} |
| 281 |
} |
| 282 |
else |
| 283 |
{ |
| 284 |
/* don't bother searching if hint says there's no free slot */ |
| 285 |
offsetNumber = limit; |
| 286 |
} |
| 287 |
} |
| 288 |
|
| 289 |
/* Reject placing items beyond the first unused line pointer */ |
| 290 |
if (offsetNumber > limit) |
| 291 |
{ |
| 292 |
elog(WARNING, "specified item offset is too large"); |
| 293 |
return InvalidOffsetNumber; |
| 294 |
} |
| 295 |
|
| 296 |
/* Reject placing items beyond heap boundary, if heap */ |
| 297 |
if ((flags & PAI_IS_HEAP) != 0 && offsetNumber > MaxHeapTuplesPerPage) |
| 298 |
{ |
| 299 |
elog(WARNING, "can't put more than MaxHeapTuplesPerPage items in a heap page"); |
| 300 |
return InvalidOffsetNumber; |
| 301 |
} |
| 302 |
|
| 303 |
/* |
| 304 |
* Compute new lower and upper pointers for page, see if it'll fit. |
| 305 |
* |
| 306 |
* Note: do arithmetic as signed ints, to avoid mistakes if, say, |
| 307 |
* alignedSize > pd_upper. |
| 308 |
*/ |
| 309 |
if (offsetNumber == limit || needshuffle) |
| 310 |
lower = phdr->pd_lower + sizeof(ItemIdData); |
| 311 |
else |
| 312 |
lower = phdr->pd_lower; |
| 313 |
|
| 314 |
alignedSize = MAXALIGN(size); |
| 315 |
|
| 316 |
upper = (int) phdr->pd_upper - (int) alignedSize; |
| 317 |
|
| 318 |
if (lower > upper) |
| 319 |
return InvalidOffsetNumber; |
| 320 |
|
| 321 |
/* |
| 322 |
* OK to insert the item. First, shuffle the existing pointers if needed. |
| 323 |
*/ |
| 324 |
itemId = PageGetItemId(page, offsetNumber); |
| 325 |
|
| 326 |
if (needshuffle) |
| 327 |
memmove(itemId + 1, itemId, |
| 328 |
(limit - offsetNumber) * sizeof(ItemIdData)); |
| 329 |
|
| 330 |
/* set the line pointer */ |
| 331 |
ItemIdSetNormal(itemId, upper, size); |
| 332 |
|
| 333 |
/* |
| 334 |
* Items normally contain no uninitialized bytes. Core bufpage consumers |
| 335 |
* conform, but this is not a necessary coding rule; a new index AM could |
| 336 |
* opt to depart from it. However, data type input functions and other |
| 337 |
* C-language functions that synthesize datums should initialize all |
| 338 |
* bytes; datumIsEqual() relies on this. Testing here, along with the |
| 339 |
* similar check in printtup(), helps to catch such mistakes. |
| 340 |
* |
| 341 |
* Values of the "name" type retrieved via index-only scans may contain |
| 342 |
* uninitialized bytes; see comment in btrescan(). Valgrind will report |
| 343 |
* this as an error, but it is safe to ignore. |
| 344 |
*/ |
| 345 |
VALGRIND_CHECK_MEM_IS_DEFINED(item, size); |
| 346 |
|
| 347 |
/* copy the item's data onto the page */ |
| 348 |
memcpy((char *) page + upper, item, size); |
| 349 |
|
| 350 |
/* adjust page header */ |
| 351 |
phdr->pd_lower = (LocationIndex) lower; |
| 352 |
phdr->pd_upper = (LocationIndex) upper; |
| 353 |
|
| 354 |
return offsetNumber; |
| 355 |
} |
| 356 |
|
| 357 |
|
| 358 |
/* |
| 359 |
* PageGetTempPage |
| 360 |
* Get a temporary page in local memory for special processing. |
| 361 |
* The returned page is not initialized at all; caller must do that. |
| 362 |
*/ |
| 363 |
Page |
| 364 |
PageGetTempPage(const PageData *page) |
| 365 |
{ |
| 366 |
Size pageSize; |
| 367 |
Page temp; |
| 368 |
|
| 369 |
pageSize = PageGetPageSize(page); |
| 370 |
temp = (Page) palloc(pageSize); |
| 371 |
|
| 372 |
return temp; |
| 373 |
} |
| 374 |
|
| 375 |
/* |
| 376 |
* PageGetTempPageCopy |
| 377 |
* Get a temporary page in local memory for special processing. |
| 378 |
* The page is initialized by copying the contents of the given page. |
| 379 |
*/ |
| 380 |
Page |
| 381 |
PageGetTempPageCopy(const PageData *page) |
| 382 |
{ |
| 383 |
Size pageSize; |
| 384 |
Page temp; |
| 385 |
|
| 386 |
pageSize = PageGetPageSize(page); |
| 387 |
temp = (Page) palloc(pageSize); |
| 388 |
|
| 389 |
memcpy(temp, page, pageSize); |
| 390 |
|
| 391 |
return temp; |
| 392 |
} |
| 393 |
|
| 394 |
/* |
| 395 |
* PageGetTempPageCopySpecial |
| 396 |
* Get a temporary page in local memory for special processing. |
| 397 |
* The page is PageInit'd with the same special-space size as the |
| 398 |
* given page, and the special space is copied from the given page. |
| 399 |
*/ |
| 400 |
Page |
| 401 |
PageGetTempPageCopySpecial(const PageData *page) |
| 402 |
{ |
| 403 |
Size pageSize; |
| 404 |
Page temp; |
| 405 |
|
| 406 |
pageSize = PageGetPageSize(page); |
| 407 |
temp = (Page) palloc(pageSize); |
| 408 |
|
| 409 |
PageInit(temp, pageSize, PageGetSpecialSize(page)); |
| 410 |
memcpy(PageGetSpecialPointer(temp), |
| 411 |
PageGetSpecialPointer(page), |
| 412 |
PageGetSpecialSize(page)); |
| 413 |
|
| 414 |
return temp; |
| 415 |
} |
| 416 |
|
| 417 |
/* |
| 418 |
* PageRestoreTempPage |
| 419 |
* Copy temporary page back to permanent page after special processing |
| 420 |
* and release the temporary page. |
| 421 |
*/ |
| 422 |
void |
| 423 |
PageRestoreTempPage(Page tempPage, Page oldPage) |
| 424 |
{ |
| 425 |
Size pageSize; |
| 426 |
|
| 427 |
pageSize = PageGetPageSize(tempPage); |
| 428 |
memcpy(oldPage, tempPage, pageSize); |
| 429 |
|
| 430 |
pfree(tempPage); |
| 431 |
} |
| 432 |
|
| 433 |
/* |
| 434 |
* Tuple defrag support for PageRepairFragmentation and PageIndexMultiDelete |
| 435 |
*/ |
| 436 |
typedef struct itemIdCompactData |
| 437 |
{ |
| 438 |
uint16 offsetindex; /* linp array index */ |
| 439 |
int16 itemoff; /* page offset of item data */ |
| 440 |
uint16 alignedlen; /* MAXALIGN(item data len) */ |
| 441 |
} itemIdCompactData; |
| 442 |
typedef itemIdCompactData *itemIdCompact; |
| 443 |
|
| 444 |
/* |
| 445 |
* After removing or marking some line pointers unused, move the tuples to |
| 446 |
* remove the gaps caused by the removed items and reorder them back into |
| 447 |
* reverse line pointer order in the page. |
| 448 |
* |
| 449 |
* This function can often be fairly hot, so it pays to take some measures to |
| 450 |
* make it as optimal as possible. |
| 451 |
* |
| 452 |
* Callers may pass 'presorted' as true if the 'itemidbase' array is sorted in |
| 453 |
* descending order of itemoff. When this is true we can just memmove() |
| 454 |
* tuples towards the end of the page. This is quite a common case as it's |
| 455 |
* the order that tuples are initially inserted into pages. When we call this |
| 456 |
* function to defragment the tuples in the page then any new line pointers |
| 457 |
* added to the page will keep that presorted order, so hitting this case is |
| 458 |
* still very common for tables that are commonly updated. |
| 459 |
* |
| 460 |
* When the 'itemidbase' array is not presorted then we're unable to just |
| 461 |
* memmove() tuples around freely. Doing so could cause us to overwrite the |
| 462 |
* memory belonging to a tuple we've not moved yet. In this case, we copy all |
| 463 |
* the tuples that need to be moved into a temporary buffer. We can then |
| 464 |
* simply memcpy() out of that temp buffer back into the page at the correct |
| 465 |
* location. Tuples are copied back into the page in the same order as the |
| 466 |
* 'itemidbase' array, so we end up reordering the tuples back into reverse |
| 467 |
* line pointer order. This will increase the chances of hitting the |
| 468 |
* presorted case the next time around. |
| 469 |
* |
| 470 |
* Callers must ensure that nitems is > 0 |
| 471 |
*/ |
| 472 |
static void |
| 473 |
compactify_tuples(itemIdCompact itemidbase, int nitems, Page page, bool presorted) |
| 474 |
{ |
| 475 |
PageHeader phdr = (PageHeader) page; |
| 476 |
Offset upper; |
| 477 |
Offset copy_tail; |
| 478 |
Offset copy_head; |
| 479 |
itemIdCompact itemidptr; |
| 480 |
int i; |
| 481 |
|
| 482 |
/* Code within will not work correctly if nitems == 0 */ |
| 483 |
Assert(nitems > 0); |
| 484 |
|
| 485 |
if (presorted) |
| 486 |
{ |
| 487 |
|
| 488 |
#ifdef USE_ASSERT_CHECKING |
| 489 |
{ |
| 490 |
/* |
| 491 |
* Verify we've not gotten any new callers that are incorrectly |
| 492 |
* passing a true presorted value. |
| 493 |
*/ |
| 494 |
Offset lastoff = phdr->pd_special; |
| 495 |
|
| 496 |
for (i = 0; i < nitems; i++) |
| 497 |
{ |
| 498 |
itemidptr = &itemidbase[i]; |
| 499 |
|
| 500 |
Assert(lastoff > itemidptr->itemoff); |
| 501 |
|
| 502 |
lastoff = itemidptr->itemoff; |
| 503 |
} |
| 504 |
} |
| 505 |
#endif /* USE_ASSERT_CHECKING */ |
| 506 |
|
| 507 |
/* |
| 508 |
* 'itemidbase' is already in the optimal order, i.e, lower item |
| 509 |
* pointers have a higher offset. This allows us to memmove() the |
| 510 |
* tuples up to the end of the page without having to worry about |
| 511 |
* overwriting other tuples that have not been moved yet. |
| 512 |
* |
| 513 |
* There's a good chance that there are tuples already right at the |
| 514 |
* end of the page that we can simply skip over because they're |
| 515 |
* already in the correct location within the page. We'll do that |
| 516 |
* first... |
| 517 |
*/ |
| 518 |
upper = phdr->pd_special; |
| 519 |
i = 0; |
| 520 |
do |
| 521 |
{ |
| 522 |
itemidptr = &itemidbase[i]; |
| 523 |
if (upper != itemidptr->itemoff + itemidptr->alignedlen) |
| 524 |
break; |
| 525 |
upper -= itemidptr->alignedlen; |
| 526 |
|
| 527 |
i++; |
| 528 |
} while (i < nitems); |
| 529 |
|
| 530 |
/* |
| 531 |
* Now that we've found the first tuple that needs to be moved, we can |
| 532 |
* do the tuple compactification. We try and make the least number of |
| 533 |
* memmove() calls and only call memmove() when there's a gap. When |
| 534 |
* we see a gap we just move all tuples after the gap up until the |
| 535 |
* point of the last move operation. |
| 536 |
*/ |
| 537 |
copy_tail = copy_head = itemidptr->itemoff + itemidptr->alignedlen; |
| 538 |
for (; i < nitems; i++) |
| 539 |
{ |
| 540 |
ItemId lp; |
| 541 |
|
| 542 |
itemidptr = &itemidbase[i]; |
| 543 |
lp = PageGetItemId(page, itemidptr->offsetindex + 1); |
| 544 |
|
| 545 |
if (copy_head != itemidptr->itemoff + itemidptr->alignedlen) |
| 546 |
{ |
| 547 |
memmove((char *) page + upper, |
| 548 |
page + copy_head, |
| 549 |
copy_tail - copy_head); |
| 550 |
|
| 551 |
/* |
| 552 |
* We've now moved all tuples already seen, but not the |
| 553 |
* current tuple, so we set the copy_tail to the end of this |
| 554 |
* tuple so it can be moved in another iteration of the loop. |
| 555 |
*/ |
| 556 |
copy_tail = itemidptr->itemoff + itemidptr->alignedlen; |
| 557 |
} |
| 558 |
/* shift the target offset down by the length of this tuple */ |
| 559 |
upper -= itemidptr->alignedlen; |
| 560 |
/* point the copy_head to the start of this tuple */ |
| 561 |
copy_head = itemidptr->itemoff; |
| 562 |
|
| 563 |
/* update the line pointer to reference the new offset */ |
| 564 |
lp->lp_off = upper; |
| 565 |
} |
| 566 |
|
| 567 |
/* move the remaining tuples. */ |
| 568 |
memmove((char *) page + upper, |
| 569 |
page + copy_head, |
| 570 |
copy_tail - copy_head); |
| 571 |
} |
| 572 |
else |
| 573 |
{ |
| 574 |
PGAlignedBlock scratch; |
| 575 |
char *scratchptr = scratch.data; |
| 576 |
|
| 577 |
/* |
| 578 |
* Non-presorted case: The tuples in the itemidbase array may be in |
| 579 |
* any order. So, in order to move these to the end of the page we |
| 580 |
* must make a temp copy of each tuple that needs to be moved before |
| 581 |
* we copy them back into the page at the new offset. |
| 582 |
* |
| 583 |
* If a large percentage of tuples have been pruned (>75%) then we'll |
| 584 |
* copy these into the temp buffer tuple-by-tuple, otherwise, we'll |
| 585 |
* just do a single memcpy() for all tuples that need to be moved. |
| 586 |
* When so many tuples have been removed there's likely to be a lot of |
| 587 |
* gaps and it's unlikely that many non-movable tuples remain at the |
| 588 |
* end of the page. |
| 589 |
*/ |
| 590 |
if (nitems < PageGetMaxOffsetNumber(page) / 4) |
| 591 |
{ |
| 592 |
i = 0; |
| 593 |
do |
| 594 |
{ |
| 595 |
itemidptr = &itemidbase[i]; |
| 596 |
memcpy(scratchptr + itemidptr->itemoff, page + itemidptr->itemoff, |
| 597 |
itemidptr->alignedlen); |
| 598 |
i++; |
| 599 |
} while (i < nitems); |
| 600 |
|
| 601 |
/* Set things up for the compactification code below */ |
| 602 |
i = 0; |
| 603 |
itemidptr = &itemidbase[0]; |
| 604 |
upper = phdr->pd_special; |
| 605 |
} |
| 606 |
else |
| 607 |
{ |
| 608 |
upper = phdr->pd_special; |
| 609 |
|
| 610 |
/* |
| 611 |
* Many tuples are likely to already be in the correct location. |
| 612 |
* There's no need to copy these into the temp buffer. Instead |
| 613 |
* we'll just skip forward in the itemidbase array to the position |
| 614 |
* that we do need to move tuples from so that the code below just |
| 615 |
* leaves these ones alone. |
| 616 |
*/ |
| 617 |
i = 0; |
| 618 |
do |
| 619 |
{ |
| 620 |
itemidptr = &itemidbase[i]; |
| 621 |
if (upper != itemidptr->itemoff + itemidptr->alignedlen) |
| 622 |
break; |
| 623 |
upper -= itemidptr->alignedlen; |
| 624 |
|
| 625 |
i++; |
| 626 |
} while (i < nitems); |
| 627 |
|
| 628 |
/* Copy all tuples that need to be moved into the temp buffer */ |
| 629 |
memcpy(scratchptr + phdr->pd_upper, |
| 630 |
page + phdr->pd_upper, |
| 631 |
upper - phdr->pd_upper); |
| 632 |
} |
| 633 |
|
| 634 |
/* |
| 635 |
* Do the tuple compactification. itemidptr is already pointing to |
| 636 |
* the first tuple that we're going to move. Here we collapse the |
| 637 |
* memcpy calls for adjacent tuples into a single call. This is done |
| 638 |
* by delaying the memcpy call until we find a gap that needs to be |
| 639 |
* closed. |
| 640 |
*/ |
| 641 |
copy_tail = copy_head = itemidptr->itemoff + itemidptr->alignedlen; |
| 642 |
for (; i < nitems; i++) |
| 643 |
{ |
| 644 |
ItemId lp; |
| 645 |
|
| 646 |
itemidptr = &itemidbase[i]; |
| 647 |
lp = PageGetItemId(page, itemidptr->offsetindex + 1); |
| 648 |
|
| 649 |
/* copy pending tuples when we detect a gap */ |
| 650 |
if (copy_head != itemidptr->itemoff + itemidptr->alignedlen) |
| 651 |
{ |
| 652 |
memcpy((char *) page + upper, |
| 653 |
scratchptr + copy_head, |
| 654 |
copy_tail - copy_head); |
| 655 |
|
| 656 |
/* |
| 657 |
* We've now copied all tuples already seen, but not the |
| 658 |
* current tuple, so we set the copy_tail to the end of this |
| 659 |
* tuple. |
| 660 |
*/ |
| 661 |
copy_tail = itemidptr->itemoff + itemidptr->alignedlen; |
| 662 |
} |
| 663 |
/* shift the target offset down by the length of this tuple */ |
| 664 |
upper -= itemidptr->alignedlen; |
| 665 |
/* point the copy_head to the start of this tuple */ |
| 666 |
copy_head = itemidptr->itemoff; |
| 667 |
|
| 668 |
/* update the line pointer to reference the new offset */ |
| 669 |
lp->lp_off = upper; |
| 670 |
} |
| 671 |
|
| 672 |
/* Copy the remaining chunk */ |
| 673 |
memcpy((char *) page + upper, |
| 674 |
scratchptr + copy_head, |
| 675 |
copy_tail - copy_head); |
| 676 |
} |
| 677 |
|
| 678 |
phdr->pd_upper = upper; |
| 679 |
} |
| 680 |
|
| 681 |
/* |
| 682 |
* PageRepairFragmentation |
| 683 |
* |
| 684 |
* Frees fragmented space on a heap page following pruning. |
| 685 |
* |
| 686 |
* This routine is usable for heap pages only, but see PageIndexMultiDelete. |
| 687 |
* |
| 688 |
* This routine removes unused line pointers from the end of the line pointer |
| 689 |
* array. This is possible when dead heap-only tuples get removed by pruning, |
| 690 |
* especially when there were HOT chains with several tuples each beforehand. |
| 691 |
* |
| 692 |
* Caller had better have a full cleanup lock on page's buffer. As a side |
| 693 |
* effect the page's PD_HAS_FREE_LINES hint bit will be set or unset as |
| 694 |
* needed. Caller might also need to account for a reduction in the length of |
| 695 |
* the line pointer array following array truncation. |
| 696 |
*/ |
| 697 |
void |
| 698 |
PageRepairFragmentation(Page page) |
| 699 |
{ |
| 700 |
Offset pd_lower = ((PageHeader) page)->pd_lower; |
| 701 |
Offset pd_upper = ((PageHeader) page)->pd_upper; |
| 702 |
Offset pd_special = ((PageHeader) page)->pd_special; |
| 703 |
Offset last_offset; |
| 704 |
itemIdCompactData itemidbase[MaxHeapTuplesPerPage]; |
| 705 |
itemIdCompact itemidptr; |
| 706 |
ItemId lp; |
| 707 |
int nline, |
| 708 |
nstorage, |
| 709 |
nunused; |
| 710 |
OffsetNumber finalusedlp = InvalidOffsetNumber; |
| 711 |
int i; |
| 712 |
Size totallen; |
| 713 |
bool presorted = true; /* For now */ |
| 714 |
|
| 715 |
/* |
| 716 |
* It's worth the trouble to be more paranoid here than in most places, |
| 717 |
* because we are about to reshuffle data in (what is usually) a shared |
| 718 |
* disk buffer. If we aren't careful then corrupted pointers, lengths, |
| 719 |
* etc could cause us to clobber adjacent disk buffers, spreading the data |
| 720 |
* loss further. So, check everything. |
| 721 |
*/ |
| 722 |
if (pd_lower < SizeOfPageHeaderData || |
| 723 |
pd_lower > pd_upper || |
| 724 |
pd_upper > pd_special || |
| 725 |
pd_special > BLCKSZ || |
| 726 |
pd_special != MAXALIGN(pd_special)) |
| 727 |
ereport(ERROR, |
| 728 |
(errcode(ERRCODE_DATA_CORRUPTED), |
| 729 |
errmsg("corrupted page pointers: lower = %u, upper = %u, special = %u", |
| 730 |
pd_lower, pd_upper, pd_special))); |
| 731 |
|
| 732 |
/* |
| 733 |
* Run through the line pointer array and collect data about live items. |
| 734 |
*/ |
| 735 |
nline = PageGetMaxOffsetNumber(page); |
| 736 |
itemidptr = itemidbase; |
| 737 |
nunused = totallen = 0; |
| 738 |
last_offset = pd_special; |
| 739 |
for (i = FirstOffsetNumber; i <= nline; i++) |
| 740 |
{ |
| 741 |
lp = PageGetItemId(page, i); |
| 742 |
if (ItemIdIsUsed(lp)) |
| 743 |
{ |
| 744 |
if (ItemIdHasStorage(lp)) |
| 745 |
{ |
| 746 |
itemidptr->offsetindex = i - 1; |
| 747 |
itemidptr->itemoff = ItemIdGetOffset(lp); |
| 748 |
|
| 749 |
if (last_offset > itemidptr->itemoff) |
| 750 |
last_offset = itemidptr->itemoff; |
| 751 |
else |
| 752 |
presorted = false; |
| 753 |
|
| 754 |
if (unlikely(itemidptr->itemoff < (int) pd_upper || |
| 755 |
itemidptr->itemoff >= (int) pd_special)) |
| 756 |
ereport(ERROR, |
| 757 |
(errcode(ERRCODE_DATA_CORRUPTED), |
| 758 |
errmsg("corrupted line pointer: %u", |
| 759 |
itemidptr->itemoff))); |
| 760 |
itemidptr->alignedlen = MAXALIGN(ItemIdGetLength(lp)); |
| 761 |
totallen += itemidptr->alignedlen; |
| 762 |
itemidptr++; |
| 763 |
} |
| 764 |
|
| 765 |
finalusedlp = i; /* Could be the final non-LP_UNUSED item */ |
| 766 |
} |
| 767 |
else |
| 768 |
{ |
| 769 |
/* Unused entries should have lp_len = 0, but make sure */ |
| 770 |
Assert(!ItemIdHasStorage(lp)); |
| 771 |
ItemIdSetUnused(lp); |
| 772 |
nunused++; |
| 773 |
} |
| 774 |
} |
| 775 |
|
| 776 |
nstorage = itemidptr - itemidbase; |
| 777 |
if (nstorage == 0) |
| 778 |
{ |
| 779 |
/* Page is completely empty, so just reset it quickly */ |
| 780 |
((PageHeader) page)->pd_upper = pd_special; |
| 781 |
} |
| 782 |
else |
| 783 |
{ |
| 784 |
/* Need to compact the page the hard way */ |
| 785 |
if (totallen > (Size) (pd_special - pd_lower)) |
| 786 |
ereport(ERROR, |
| 787 |
(errcode(ERRCODE_DATA_CORRUPTED), |
| 788 |
errmsg("corrupted item lengths: total %u, available space %u", |
| 789 |
(unsigned int) totallen, pd_special - pd_lower))); |
| 790 |
|
| 791 |
compactify_tuples(itemidbase, nstorage, page, presorted); |
| 792 |
} |
| 793 |
|
| 794 |
if (finalusedlp != nline) |
| 795 |
{ |
| 796 |
/* The last line pointer is not the last used line pointer */ |
| 797 |
int nunusedend = nline - finalusedlp; |
| 798 |
|
| 799 |
Assert(nunused >= nunusedend && nunusedend > 0); |
| 800 |
|
| 801 |
/* remove trailing unused line pointers from the count */ |
| 802 |
nunused -= nunusedend; |
| 803 |
/* truncate the line pointer array */ |
| 804 |
((PageHeader) page)->pd_lower -= (sizeof(ItemIdData) * nunusedend); |
| 805 |
} |
| 806 |
|
| 807 |
/* Set hint bit for PageAddItemExtended */ |
| 808 |
if (nunused > 0) |
| 809 |
PageSetHasFreeLinePointers(page); |
| 810 |
else |
| 811 |
PageClearHasFreeLinePointers(page); |
| 812 |
} |
| 813 |
|
| 814 |
/* |
| 815 |
* PageTruncateLinePointerArray |
| 816 |
* |
| 817 |
* Removes unused line pointers at the end of the line pointer array. |
| 818 |
* |
| 819 |
* This routine is usable for heap pages only. It is called by VACUUM during |
| 820 |
* its second pass over the heap. We expect at least one LP_UNUSED line |
| 821 |
* pointer on the page (if VACUUM didn't have an LP_DEAD item on the page that |
| 822 |
* it just set to LP_UNUSED then it should not call here). |
| 823 |
* |
| 824 |
* We avoid truncating the line pointer array to 0 items, if necessary by |
| 825 |
* leaving behind a single remaining LP_UNUSED item. This is a little |
| 826 |
* arbitrary, but it seems like a good idea to avoid leaving a PageIsEmpty() |
| 827 |
* page behind. |
| 828 |
* |
| 829 |
* Caller can have either an exclusive lock or a full cleanup lock on page's |
| 830 |
* buffer. The page's PD_HAS_FREE_LINES hint bit will be set or unset based |
| 831 |
* on whether or not we leave behind any remaining LP_UNUSED items. |
| 832 |
*/ |
| 833 |
void |
| 834 |
PageTruncateLinePointerArray(Page page) |
| 835 |
{ |
| 836 |
PageHeader phdr = (PageHeader) page; |
| 837 |
bool countdone = false, |
| 838 |
sethint = false; |
| 839 |
int nunusedend = 0; |
| 840 |
|
| 841 |
/* Scan line pointer array back-to-front */ |
| 842 |
for (int i = PageGetMaxOffsetNumber(page); i >= FirstOffsetNumber; i--) |
| 843 |
{ |
| 844 |
ItemId lp = PageGetItemId(page, i); |
| 845 |
|
| 846 |
if (!countdone && i > FirstOffsetNumber) |
| 847 |
{ |
| 848 |
/* |
| 849 |
* Still determining which line pointers from the end of the array |
| 850 |
* will be truncated away. Either count another line pointer as |
| 851 |
* safe to truncate, or notice that it's not safe to truncate |
| 852 |
* additional line pointers (stop counting line pointers). |
| 853 |
*/ |
| 854 |
if (!ItemIdIsUsed(lp)) |
| 855 |
nunusedend++; |
| 856 |
else |
| 857 |
countdone = true; |
| 858 |
} |
| 859 |
else |
| 860 |
{ |
| 861 |
/* |
| 862 |
* Once we've stopped counting we still need to figure out if |
| 863 |
* there are any remaining LP_UNUSED line pointers somewhere more |
| 864 |
* towards the front of the array. |
| 865 |
*/ |
| 866 |
if (!ItemIdIsUsed(lp)) |
| 867 |
{ |
| 868 |
/* |
| 869 |
* This is an unused line pointer that we won't be truncating |
| 870 |
* away -- so there is at least one. Set hint on page. |
| 871 |
*/ |
| 872 |
sethint = true; |
| 873 |
break; |
| 874 |
} |
| 875 |
} |
| 876 |
} |
| 877 |
|
| 878 |
if (nunusedend > 0) |
| 879 |
{ |
| 880 |
phdr->pd_lower -= sizeof(ItemIdData) * nunusedend; |
| 881 |
|
| 882 |
#ifdef CLOBBER_FREED_MEMORY |
| 883 |
memset((char *) page + phdr->pd_lower, 0x7F, |
| 884 |
sizeof(ItemIdData) * nunusedend); |
| 885 |
#endif |
| 886 |
} |
| 887 |
else |
| 888 |
Assert(sethint); |
| 889 |
|
| 890 |
/* Set hint bit for PageAddItemExtended */ |
| 891 |
if (sethint) |
| 892 |
PageSetHasFreeLinePointers(page); |
| 893 |
else |
| 894 |
PageClearHasFreeLinePointers(page); |
| 895 |
} |
| 896 |
|
| 897 |
/* |
| 898 |
* PageGetFreeSpace |
| 899 |
* Returns the size of the free (allocatable) space on a page, |
| 900 |
* reduced by the space needed for a new line pointer. |
| 901 |
* |
| 902 |
* Note: this should usually only be used on index pages. Use |
| 903 |
* PageGetHeapFreeSpace on heap pages. |
| 904 |
*/ |
| 905 |
Size |
| 906 |
PageGetFreeSpace(const PageData *page) |
| 907 |
{ |
| 908 |
const PageHeaderData *phdr = (const PageHeaderData *) page; |
| 909 |
int space; |
| 910 |
|
| 911 |
/* |
| 912 |
* Use signed arithmetic here so that we behave sensibly if pd_lower > |
| 913 |
* pd_upper. |
| 914 |
*/ |
| 915 |
space = (int) phdr->pd_upper - (int) phdr->pd_lower; |
| 916 |
|
| 917 |
if (space < (int) sizeof(ItemIdData)) |
| 918 |
return 0; |
| 919 |
space -= sizeof(ItemIdData); |
| 920 |
|
| 921 |
return (Size) space; |
| 922 |
} |
| 923 |
|
| 924 |
/* |
| 925 |
* PageGetFreeSpaceForMultipleTuples |
| 926 |
* Returns the size of the free (allocatable) space on a page, |
| 927 |
* reduced by the space needed for multiple new line pointers. |
| 928 |
* |
| 929 |
* Note: this should usually only be used on index pages. Use |
| 930 |
* PageGetHeapFreeSpace on heap pages. |
| 931 |
*/ |
| 932 |
Size |
| 933 |
PageGetFreeSpaceForMultipleTuples(const PageData *page, int ntups) |
| 934 |
{ |
| 935 |
const PageHeaderData *phdr = (const PageHeaderData *) page; |
| 936 |
int space; |
| 937 |
|
| 938 |
/* |
| 939 |
* Use signed arithmetic here so that we behave sensibly if pd_lower > |
| 940 |
* pd_upper. |
| 941 |
*/ |
| 942 |
space = (int) phdr->pd_upper - (int) phdr->pd_lower; |
| 943 |
|
| 944 |
if (space < (int) (ntups * sizeof(ItemIdData))) |
| 945 |
return 0; |
| 946 |
space -= ntups * sizeof(ItemIdData); |
| 947 |
|
| 948 |
return (Size) space; |
| 949 |
} |
| 950 |
|
| 951 |
/* |
| 952 |
* PageGetExactFreeSpace |
| 953 |
* Returns the size of the free (allocatable) space on a page, |
| 954 |
* without any consideration for adding/removing line pointers. |
| 955 |
*/ |
| 956 |
Size |
| 957 |
PageGetExactFreeSpace(const PageData *page) |
| 958 |
{ |
| 959 |
const PageHeaderData *phdr = (const PageHeaderData *) page; |
| 960 |
int space; |
| 961 |
|
| 962 |
/* |
| 963 |
* Use signed arithmetic here so that we behave sensibly if pd_lower > |
| 964 |
* pd_upper. |
| 965 |
*/ |
| 966 |
space = (int) phdr->pd_upper - (int) phdr->pd_lower; |
| 967 |
|
| 968 |
if (space < 0) |
| 969 |
return 0; |
| 970 |
|
| 971 |
return (Size) space; |
| 972 |
} |
| 973 |
|
| 974 |
|
| 975 |
/* |
| 976 |
* PageGetHeapFreeSpace |
| 977 |
* Returns the size of the free (allocatable) space on a page, |
| 978 |
* reduced by the space needed for a new line pointer. |
| 979 |
* |
| 980 |
* The difference between this and PageGetFreeSpace is that this will return |
| 981 |
* zero if there are already MaxHeapTuplesPerPage line pointers in the page |
| 982 |
* and none are free. We use this to enforce that no more than |
| 983 |
* MaxHeapTuplesPerPage line pointers are created on a heap page. (Although |
| 984 |
* no more tuples than that could fit anyway, in the presence of redirected |
| 985 |
* or dead line pointers it'd be possible to have too many line pointers. |
| 986 |
* To avoid breaking code that assumes MaxHeapTuplesPerPage is a hard limit |
| 987 |
* on the number of line pointers, we make this extra check.) |
| 988 |
*/ |
| 989 |
Size |
| 990 |
PageGetHeapFreeSpace(const PageData *page) |
| 991 |
{ |
| 992 |
Size space; |
| 993 |
|
| 994 |
space = PageGetFreeSpace(page); |
| 995 |
if (space > 0) |
| 996 |
{ |
| 997 |
OffsetNumber offnum, |
| 998 |
nline; |
| 999 |
|
| 1000 |
/* |
| 1001 |
* Are there already MaxHeapTuplesPerPage line pointers in the page? |
| 1002 |
*/ |
| 1003 |
nline = PageGetMaxOffsetNumber(page); |
| 1004 |
if (nline >= MaxHeapTuplesPerPage) |
| 1005 |
{ |
| 1006 |
if (PageHasFreeLinePointers(page)) |
| 1007 |
{ |
| 1008 |
/* |
| 1009 |
* Since this is just a hint, we must confirm that there is |
| 1010 |
* indeed a free line pointer |
| 1011 |
*/ |
| 1012 |
for (offnum = FirstOffsetNumber; offnum <= nline; offnum = OffsetNumberNext(offnum)) |
| 1013 |
{ |
| 1014 |
ItemId lp = PageGetItemId(unconstify(PageData *, page), offnum); |
| 1015 |
|
| 1016 |
if (!ItemIdIsUsed(lp)) |
| 1017 |
break; |
| 1018 |
} |
| 1019 |
|
| 1020 |
if (offnum > nline) |
| 1021 |
{ |
| 1022 |
/* |
| 1023 |
* The hint is wrong, but we can't clear it here since we |
| 1024 |
* don't have the ability to mark the page dirty. |
| 1025 |
*/ |
| 1026 |
space = 0; |
| 1027 |
} |
| 1028 |
} |
| 1029 |
else |
| 1030 |
{ |
| 1031 |
/* |
| 1032 |
* Although the hint might be wrong, PageAddItem will believe |
| 1033 |
* it anyway, so we must believe it too. |
| 1034 |
*/ |
| 1035 |
space = 0; |
| 1036 |
} |
| 1037 |
} |
| 1038 |
} |
| 1039 |
return space; |
| 1040 |
} |
| 1041 |
|
| 1042 |
|
| 1043 |
/* |
| 1044 |
* PageIndexTupleDelete |
| 1045 |
* |
| 1046 |
* This routine does the work of removing a tuple from an index page. |
| 1047 |
* |
| 1048 |
* Unlike heap pages, we compact out the line pointer for the removed tuple. |
| 1049 |
*/ |
| 1050 |
void |
| 1051 |
PageIndexTupleDelete(Page page, OffsetNumber offnum) |
| 1052 |
{ |
| 1053 |
PageHeader phdr = (PageHeader) page; |
| 1054 |
char *addr; |
| 1055 |
ItemId tup; |
| 1056 |
Size size; |
| 1057 |
unsigned offset; |
| 1058 |
int nbytes; |
| 1059 |
int offidx; |
| 1060 |
int nline; |
| 1061 |
|
| 1062 |
/* |
| 1063 |
* As with PageRepairFragmentation, paranoia seems justified. |
| 1064 |
*/ |
| 1065 |
if (phdr->pd_lower < SizeOfPageHeaderData || |
| 1066 |
phdr->pd_lower > phdr->pd_upper || |
| 1067 |
phdr->pd_upper > phdr->pd_special || |
| 1068 |
phdr->pd_special > BLCKSZ || |
| 1069 |
phdr->pd_special != MAXALIGN(phdr->pd_special)) |
| 1070 |
ereport(ERROR, |
| 1071 |
(errcode(ERRCODE_DATA_CORRUPTED), |
| 1072 |
errmsg("corrupted page pointers: lower = %u, upper = %u, special = %u", |
| 1073 |
phdr->pd_lower, phdr->pd_upper, phdr->pd_special))); |
| 1074 |
|
| 1075 |
nline = PageGetMaxOffsetNumber(page); |
| 1076 |
if ((int) offnum <= 0 || (int) offnum > nline) |
| 1077 |
elog(ERROR, "invalid index offnum: %u", offnum); |
| 1078 |
|
| 1079 |
/* change offset number to offset index */ |
| 1080 |
offidx = offnum - 1; |
| 1081 |
|
| 1082 |
tup = PageGetItemId(page, offnum); |
| 1083 |
Assert(ItemIdHasStorage(tup)); |
| 1084 |
size = ItemIdGetLength(tup); |
| 1085 |
offset = ItemIdGetOffset(tup); |
| 1086 |
|
| 1087 |
if (offset < phdr->pd_upper || (offset + size) > phdr->pd_special || |
| 1088 |
offset != MAXALIGN(offset)) |
| 1089 |
ereport(ERROR, |
| 1090 |
(errcode(ERRCODE_DATA_CORRUPTED), |
| 1091 |
errmsg("corrupted line pointer: offset = %u, size = %u", |
| 1092 |
offset, (unsigned int) size))); |
| 1093 |
|
| 1094 |
/* Amount of space to actually be deleted */ |
| 1095 |
size = MAXALIGN(size); |
| 1096 |
|
| 1097 |
/* |
| 1098 |
* First, we want to get rid of the pd_linp entry for the index tuple. We |
| 1099 |
* copy all subsequent linp's back one slot in the array. We don't use |
| 1100 |
* PageGetItemId, because we are manipulating the _array_, not individual |
| 1101 |
* linp's. |
| 1102 |
*/ |
| 1103 |
nbytes = phdr->pd_lower - |
| 1104 |
((char *) &phdr->pd_linp[offidx + 1] - (char *) phdr); |
| 1105 |
|
| 1106 |
if (nbytes > 0) |
| 1107 |
memmove(&(phdr->pd_linp[offidx]), |
| 1108 |
&(phdr->pd_linp[offidx + 1]), |
| 1109 |
nbytes); |
| 1110 |
|
| 1111 |
/* |
| 1112 |
* Now move everything between the old upper bound (beginning of tuple |
| 1113 |
* space) and the beginning of the deleted tuple forward, so that space in |
| 1114 |
* the middle of the page is left free. If we've just deleted the tuple |
| 1115 |
* at the beginning of tuple space, then there's no need to do the copy. |
| 1116 |
*/ |
| 1117 |
|
| 1118 |
/* beginning of tuple space */ |
| 1119 |
addr = (char *) page + phdr->pd_upper; |
| 1120 |
|
| 1121 |
if (offset > phdr->pd_upper) |
| 1122 |
memmove(addr + size, addr, offset - phdr->pd_upper); |
| 1123 |
|
| 1124 |
/* adjust free space boundary pointers */ |
| 1125 |
phdr->pd_upper += size; |
| 1126 |
phdr->pd_lower -= sizeof(ItemIdData); |
| 1127 |
|
| 1128 |
/* |
| 1129 |
* Finally, we need to adjust the linp entries that remain. |
| 1130 |
* |
| 1131 |
* Anything that used to be before the deleted tuple's data was moved |
| 1132 |
* forward by the size of the deleted tuple. |
| 1133 |
*/ |
| 1134 |
if (!PageIsEmpty(page)) |
| 1135 |
{ |
| 1136 |
int i; |
| 1137 |
|
| 1138 |
nline--; /* there's one less than when we started */ |
| 1139 |
for (i = 1; i <= nline; i++) |
| 1140 |
{ |
| 1141 |
ItemId ii = PageGetItemId(page, i); |
| 1142 |
|
| 1143 |
Assert(ItemIdHasStorage(ii)); |
| 1144 |
if (ItemIdGetOffset(ii) <= offset) |
| 1145 |
ii->lp_off += size; |
| 1146 |
} |
| 1147 |
} |
| 1148 |
} |
| 1149 |
|
| 1150 |
|
| 1151 |
/* |
| 1152 |
* PageIndexMultiDelete |
| 1153 |
* |
| 1154 |
* This routine handles the case of deleting multiple tuples from an |
| 1155 |
* index page at once. It is considerably faster than a loop around |
| 1156 |
* PageIndexTupleDelete ... however, the caller *must* supply the array |
| 1157 |
* of item numbers to be deleted in item number order! |
| 1158 |
*/ |
| 1159 |
void |
| 1160 |
PageIndexMultiDelete(Page page, OffsetNumber *itemnos, int nitems) |
| 1161 |
{ |
| 1162 |
PageHeader phdr = (PageHeader) page; |
| 1163 |
Offset pd_lower = phdr->pd_lower; |
| 1164 |
Offset pd_upper = phdr->pd_upper; |
| 1165 |
Offset pd_special = phdr->pd_special; |
| 1166 |
Offset last_offset; |
| 1167 |
itemIdCompactData itemidbase[MaxIndexTuplesPerPage]; |
| 1168 |
ItemIdData newitemids[MaxIndexTuplesPerPage]; |
| 1169 |
itemIdCompact itemidptr; |
| 1170 |
ItemId lp; |
| 1171 |
int nline, |
| 1172 |
nused; |
| 1173 |
Size totallen; |
| 1174 |
Size size; |
| 1175 |
unsigned offset; |
| 1176 |
int nextitm; |
| 1177 |
OffsetNumber offnum; |
| 1178 |
bool presorted = true; /* For now */ |
| 1179 |
|
| 1180 |
Assert(nitems <= MaxIndexTuplesPerPage); |
| 1181 |
|
| 1182 |
/* |
| 1183 |
* If there aren't very many items to delete, then retail |
| 1184 |
* PageIndexTupleDelete is the best way. Delete the items in reverse |
| 1185 |
* order so we don't have to think about adjusting item numbers for |
| 1186 |
* previous deletions. |
| 1187 |
* |
| 1188 |
* TODO: tune the magic number here |
| 1189 |
*/ |
| 1190 |
if (nitems <= 2) |
| 1191 |
{ |
| 1192 |
while (--nitems >= 0) |
| 1193 |
PageIndexTupleDelete(page, itemnos[nitems]); |
| 1194 |
return; |
| 1195 |
} |
| 1196 |
|
| 1197 |
/* |
| 1198 |
* As with PageRepairFragmentation, paranoia seems justified. |
| 1199 |
*/ |
| 1200 |
if (pd_lower < SizeOfPageHeaderData || |
| 1201 |
pd_lower > pd_upper || |
| 1202 |
pd_upper > pd_special || |
| 1203 |
pd_special > BLCKSZ || |
| 1204 |
pd_special != MAXALIGN(pd_special)) |
| 1205 |
ereport(ERROR, |
| 1206 |
(errcode(ERRCODE_DATA_CORRUPTED), |
| 1207 |
errmsg("corrupted page pointers: lower = %u, upper = %u, special = %u", |
| 1208 |
pd_lower, pd_upper, pd_special))); |
| 1209 |
|
| 1210 |
/* |
| 1211 |
* Scan the line pointer array and build a list of just the ones we are |
| 1212 |
* going to keep. Notice we do not modify the page yet, since we are |
| 1213 |
* still validity-checking. |
| 1214 |
*/ |
| 1215 |
nline = PageGetMaxOffsetNumber(page); |
| 1216 |
itemidptr = itemidbase; |
| 1217 |
totallen = 0; |
| 1218 |
nused = 0; |
| 1219 |
nextitm = 0; |
| 1220 |
last_offset = pd_special; |
| 1221 |
for (offnum = FirstOffsetNumber; offnum <= nline; offnum = OffsetNumberNext(offnum)) |
| 1222 |
{ |
| 1223 |
lp = PageGetItemId(page, offnum); |
| 1224 |
Assert(ItemIdHasStorage(lp)); |
| 1225 |
size = ItemIdGetLength(lp); |
| 1226 |
offset = ItemIdGetOffset(lp); |
| 1227 |
if (offset < pd_upper || |
| 1228 |
(offset + size) > pd_special || |
| 1229 |
offset != MAXALIGN(offset)) |
| 1230 |
ereport(ERROR, |
| 1231 |
(errcode(ERRCODE_DATA_CORRUPTED), |
| 1232 |
errmsg("corrupted line pointer: offset = %u, size = %u", |
| 1233 |
offset, (unsigned int) size))); |
| 1234 |
|
| 1235 |
if (nextitm < nitems && offnum == itemnos[nextitm]) |
| 1236 |
{ |
| 1237 |
/* skip item to be deleted */ |
| 1238 |
nextitm++; |
| 1239 |
} |
| 1240 |
else |
| 1241 |
{ |
| 1242 |
itemidptr->offsetindex = nused; /* where it will go */ |
| 1243 |
itemidptr->itemoff = offset; |
| 1244 |
|
| 1245 |
if (last_offset > itemidptr->itemoff) |
| 1246 |
last_offset = itemidptr->itemoff; |
| 1247 |
else |
| 1248 |
presorted = false; |
| 1249 |
|
| 1250 |
itemidptr->alignedlen = MAXALIGN(size); |
| 1251 |
totallen += itemidptr->alignedlen; |
| 1252 |
newitemids[nused] = *lp; |
| 1253 |
itemidptr++; |
| 1254 |
nused++; |
| 1255 |
} |
| 1256 |
} |
| 1257 |
|
| 1258 |
/* this will catch invalid or out-of-order itemnos[] */ |
| 1259 |
if (nextitm != nitems) |
| 1260 |
elog(ERROR, "incorrect index offsets supplied"); |
| 1261 |
|
| 1262 |
if (totallen > (Size) (pd_special - pd_lower)) |
| 1263 |
ereport(ERROR, |
| 1264 |
(errcode(ERRCODE_DATA_CORRUPTED), |
| 1265 |
errmsg("corrupted item lengths: total %u, available space %u", |
| 1266 |
(unsigned int) totallen, pd_special - pd_lower))); |
| 1267 |
|
| 1268 |
/* |
| 1269 |
* Looks good. Overwrite the line pointers with the copy, from which we've |
| 1270 |
* removed all the unused items. |
| 1271 |
*/ |
| 1272 |
memcpy(phdr->pd_linp, newitemids, nused * sizeof(ItemIdData)); |
| 1273 |
phdr->pd_lower = SizeOfPageHeaderData + nused * sizeof(ItemIdData); |
| 1274 |
|
| 1275 |
/* and compactify the tuple data */ |
| 1276 |
if (nused > 0) |
| 1277 |
compactify_tuples(itemidbase, nused, page, presorted); |
| 1278 |
else |
| 1279 |
phdr->pd_upper = pd_special; |
| 1280 |
} |
| 1281 |
|
| 1282 |
|
| 1283 |
/* |
| 1284 |
* PageIndexTupleDeleteNoCompact |
| 1285 |
* |
| 1286 |
* Remove the specified tuple from an index page, but set its line pointer |
| 1287 |
* to "unused" instead of compacting it out, except that it can be removed |
| 1288 |
* if it's the last line pointer on the page. |
| 1289 |
* |
| 1290 |
* This is used for index AMs that require that existing TIDs of live tuples |
| 1291 |
* remain unchanged, and are willing to allow unused line pointers instead. |
| 1292 |
*/ |
| 1293 |
void |
| 1294 |
PageIndexTupleDeleteNoCompact(Page page, OffsetNumber offnum) |
| 1295 |
{ |
| 1296 |
PageHeader phdr = (PageHeader) page; |
| 1297 |
char *addr; |
| 1298 |
ItemId tup; |
| 1299 |
Size size; |
| 1300 |
unsigned offset; |
| 1301 |
int nline; |
| 1302 |
|
| 1303 |
/* |
| 1304 |
* As with PageRepairFragmentation, paranoia seems justified. |
| 1305 |
*/ |
| 1306 |
if (phdr->pd_lower < SizeOfPageHeaderData || |
| 1307 |
phdr->pd_lower > phdr->pd_upper || |
| 1308 |
phdr->pd_upper > phdr->pd_special || |
| 1309 |
phdr->pd_special > BLCKSZ || |
| 1310 |
phdr->pd_special != MAXALIGN(phdr->pd_special)) |
| 1311 |
ereport(ERROR, |
| 1312 |
(errcode(ERRCODE_DATA_CORRUPTED), |
| 1313 |
errmsg("corrupted page pointers: lower = %u, upper = %u, special = %u", |
| 1314 |
phdr->pd_lower, phdr->pd_upper, phdr->pd_special))); |
| 1315 |
|
| 1316 |
nline = PageGetMaxOffsetNumber(page); |
| 1317 |
if ((int) offnum <= 0 || (int) offnum > nline) |
| 1318 |
elog(ERROR, "invalid index offnum: %u", offnum); |
| 1319 |
|
| 1320 |
tup = PageGetItemId(page, offnum); |
| 1321 |
Assert(ItemIdHasStorage(tup)); |
| 1322 |
size = ItemIdGetLength(tup); |
| 1323 |
offset = ItemIdGetOffset(tup); |
| 1324 |
|
| 1325 |
if (offset < phdr->pd_upper || (offset + size) > phdr->pd_special || |
| 1326 |
offset != MAXALIGN(offset)) |
| 1327 |
ereport(ERROR, |
| 1328 |
(errcode(ERRCODE_DATA_CORRUPTED), |
| 1329 |
errmsg("corrupted line pointer: offset = %u, size = %u", |
| 1330 |
offset, (unsigned int) size))); |
| 1331 |
|
| 1332 |
/* Amount of space to actually be deleted */ |
| 1333 |
size = MAXALIGN(size); |
| 1334 |
|
| 1335 |
/* |
| 1336 |
* Either set the line pointer to "unused", or zap it if it's the last |
| 1337 |
* one. (Note: it's possible that the next-to-last one(s) are already |
| 1338 |
* unused, but we do not trouble to try to compact them out if so.) |
| 1339 |
*/ |
| 1340 |
if ((int) offnum < nline) |
| 1341 |
ItemIdSetUnused(tup); |
| 1342 |
else |
| 1343 |
{ |
| 1344 |
phdr->pd_lower -= sizeof(ItemIdData); |
| 1345 |
nline--; /* there's one less than when we started */ |
| 1346 |
} |
| 1347 |
|
| 1348 |
/* |
| 1349 |
* Now move everything between the old upper bound (beginning of tuple |
| 1350 |
* space) and the beginning of the deleted tuple forward, so that space in |
| 1351 |
* the middle of the page is left free. If we've just deleted the tuple |
| 1352 |
* at the beginning of tuple space, then there's no need to do the copy. |
| 1353 |
*/ |
| 1354 |
|
| 1355 |
/* beginning of tuple space */ |
| 1356 |
addr = (char *) page + phdr->pd_upper; |
| 1357 |
|
| 1358 |
if (offset > phdr->pd_upper) |
| 1359 |
memmove(addr + size, addr, offset - phdr->pd_upper); |
| 1360 |
|
| 1361 |
/* adjust free space boundary pointer */ |
| 1362 |
phdr->pd_upper += size; |
| 1363 |
|
| 1364 |
/* |
| 1365 |
* Finally, we need to adjust the linp entries that remain. |
| 1366 |
* |
| 1367 |
* Anything that used to be before the deleted tuple's data was moved |
| 1368 |
* forward by the size of the deleted tuple. |
| 1369 |
*/ |
| 1370 |
if (!PageIsEmpty(page)) |
| 1371 |
{ |
| 1372 |
int i; |
| 1373 |
|
| 1374 |
for (i = 1; i <= nline; i++) |
| 1375 |
{ |
| 1376 |
ItemId ii = PageGetItemId(page, i); |
| 1377 |
|
| 1378 |
if (ItemIdHasStorage(ii) && ItemIdGetOffset(ii) <= offset) |
| 1379 |
ii->lp_off += size; |
| 1380 |
} |
| 1381 |
} |
| 1382 |
} |
| 1383 |
|
| 1384 |
|
| 1385 |
/* |
| 1386 |
* PageIndexTupleOverwrite |
| 1387 |
* |
| 1388 |
* Replace a specified tuple on an index page. |
| 1389 |
* |
| 1390 |
* The new tuple is placed exactly where the old one had been, shifting |
| 1391 |
* other tuples' data up or down as needed to keep the page compacted. |
| 1392 |
* This is better than deleting and reinserting the tuple, because it |
| 1393 |
* avoids any data shifting when the tuple size doesn't change; and |
| 1394 |
* even when it does, we avoid moving the line pointers around. |
| 1395 |
* This could be used by an index AM that doesn't want to unset the |
| 1396 |
* LP_DEAD bit when it happens to be set. It could conceivably also be |
| 1397 |
* used by an index AM that cares about the physical order of tuples as |
| 1398 |
* well as their logical/ItemId order. |
| 1399 |
* |
| 1400 |
* If there's insufficient space for the new tuple, return false. Other |
| 1401 |
* errors represent data-corruption problems, so we just elog. |
| 1402 |
*/ |
| 1403 |
bool |
| 1404 |
PageIndexTupleOverwrite(Page page, OffsetNumber offnum, |
| 1405 |
Item newtup, Size newsize) |
| 1406 |
{ |
| 1407 |
PageHeader phdr = (PageHeader) page; |
| 1408 |
ItemId tupid; |
| 1409 |
int oldsize; |
| 1410 |
unsigned offset; |
| 1411 |
Size alignednewsize; |
| 1412 |
int size_diff; |
| 1413 |
int itemcount; |
| 1414 |
|
| 1415 |
/* |
| 1416 |
* As with PageRepairFragmentation, paranoia seems justified. |
| 1417 |
*/ |
| 1418 |
if (phdr->pd_lower < SizeOfPageHeaderData || |
| 1419 |
phdr->pd_lower > phdr->pd_upper || |
| 1420 |
phdr->pd_upper > phdr->pd_special || |
| 1421 |
phdr->pd_special > BLCKSZ || |
| 1422 |
phdr->pd_special != MAXALIGN(phdr->pd_special)) |
| 1423 |
ereport(ERROR, |
| 1424 |
(errcode(ERRCODE_DATA_CORRUPTED), |
| 1425 |
errmsg("corrupted page pointers: lower = %u, upper = %u, special = %u", |
| 1426 |
phdr->pd_lower, phdr->pd_upper, phdr->pd_special))); |
| 1427 |
|
| 1428 |
itemcount = PageGetMaxOffsetNumber(page); |
| 1429 |
if ((int) offnum <= 0 || (int) offnum > itemcount) |
| 1430 |
elog(ERROR, "invalid index offnum: %u", offnum); |
| 1431 |
|
| 1432 |
tupid = PageGetItemId(page, offnum); |
| 1433 |
Assert(ItemIdHasStorage(tupid)); |
| 1434 |
oldsize = ItemIdGetLength(tupid); |
| 1435 |
offset = ItemIdGetOffset(tupid); |
| 1436 |
|
| 1437 |
if (offset < phdr->pd_upper || (offset + oldsize) > phdr->pd_special || |
| 1438 |
offset != MAXALIGN(offset)) |
| 1439 |
ereport(ERROR, |
| 1440 |
(errcode(ERRCODE_DATA_CORRUPTED), |
| 1441 |
errmsg("corrupted line pointer: offset = %u, size = %u", |
| 1442 |
offset, (unsigned int) oldsize))); |
| 1443 |
|
| 1444 |
/* |
| 1445 |
* Determine actual change in space requirement, check for page overflow. |
| 1446 |
*/ |
| 1447 |
oldsize = MAXALIGN(oldsize); |
| 1448 |
alignednewsize = MAXALIGN(newsize); |
| 1449 |
if (alignednewsize > oldsize + (phdr->pd_upper - phdr->pd_lower)) |
| 1450 |
return false; |
| 1451 |
|
| 1452 |
/* |
| 1453 |
* Relocate existing data and update line pointers, unless the new tuple |
| 1454 |
* is the same size as the old (after alignment), in which case there's |
| 1455 |
* nothing to do. Notice that what we have to relocate is data before the |
| 1456 |
* target tuple, not data after, so it's convenient to express size_diff |
| 1457 |
* as the amount by which the tuple's size is decreasing, making it the |
| 1458 |
* delta to add to pd_upper and affected line pointers. |
| 1459 |
*/ |
| 1460 |
size_diff = oldsize - (int) alignednewsize; |
| 1461 |
if (size_diff != 0) |
| 1462 |
{ |
| 1463 |
char *addr = (char *) page + phdr->pd_upper; |
| 1464 |
int i; |
| 1465 |
|
| 1466 |
/* relocate all tuple data before the target tuple */ |
| 1467 |
memmove(addr + size_diff, addr, offset - phdr->pd_upper); |
| 1468 |
|
| 1469 |
/* adjust free space boundary pointer */ |
| 1470 |
phdr->pd_upper += size_diff; |
| 1471 |
|
| 1472 |
/* adjust affected line pointers too */ |
| 1473 |
for (i = FirstOffsetNumber; i <= itemcount; i++) |
| 1474 |
{ |
| 1475 |
ItemId ii = PageGetItemId(page, i); |
| 1476 |
|
| 1477 |
/* Allow items without storage; currently only BRIN needs that */ |
| 1478 |
if (ItemIdHasStorage(ii) && ItemIdGetOffset(ii) <= offset) |
| 1479 |
ii->lp_off += size_diff; |
| 1480 |
} |
| 1481 |
} |
| 1482 |
|
| 1483 |
/* Update the item's tuple length without changing its lp_flags field */ |
| 1484 |
tupid->lp_off = offset + size_diff; |
| 1485 |
tupid->lp_len = newsize; |
| 1486 |
|
| 1487 |
/* Copy new tuple data onto page */ |
| 1488 |
memcpy(PageGetItem(page, tupid), newtup, newsize); |
| 1489 |
|
| 1490 |
return true; |
| 1491 |
} |
| 1492 |
|
| 1493 |
|
| 1494 |
/* |
| 1495 |
* Set checksum for a page in shared buffers. |
| 1496 |
* |
| 1497 |
* If checksums are disabled, or if the page is not initialized, just return |
| 1498 |
* the input. Otherwise, we must make a copy of the page before calculating |
| 1499 |
* the checksum, to prevent concurrent modifications (e.g. setting hint bits) |
| 1500 |
* from making the final checksum invalid. It doesn't matter if we include or |
| 1501 |
* exclude hints during the copy, as long as we write a valid page and |
| 1502 |
* associated checksum. |
| 1503 |
* |
| 1504 |
* Returns a pointer to the block-sized data that needs to be written. Uses |
| 1505 |
* statically-allocated memory, so the caller must immediately write the |
| 1506 |
* returned page and not refer to it again. |
| 1507 |
*/ |
| 1508 |
char * |
| 1509 |
PageSetChecksumCopy(Page page, BlockNumber blkno) |
| 1510 |
{ |
| 1511 |
static char *pageCopy = NULL; |
| 1512 |
|
| 1513 |
/* If we don't need a checksum, just return the passed-in data */ |
| 1514 |
if (PageIsNew(page) || !DataChecksumsEnabled()) |
| 1515 |
return page; |
| 1516 |
|
| 1517 |
/* |
| 1518 |
* We allocate the copy space once and use it over on each subsequent |
| 1519 |
* call. The point of palloc'ing here, rather than having a static char |
| 1520 |
* array, is first to ensure adequate alignment for the checksumming code |
| 1521 |
* and second to avoid wasting space in processes that never call this. |
| 1522 |
*/ |
| 1523 |
if (pageCopy == NULL) |
| 1524 |
pageCopy = MemoryContextAllocAligned(TopMemoryContext, |
| 1525 |
BLCKSZ, |
| 1526 |
PG_IO_ALIGN_SIZE, |
| 1527 |
0); |
| 1528 |
|
| 1529 |
memcpy(pageCopy, page, BLCKSZ); |
| 1530 |
((PageHeader) pageCopy)->pd_checksum = pg_checksum_page(pageCopy, blkno); |
| 1531 |
return pageCopy; |
| 1532 |
} |
| 1533 |
|
| 1534 |
/* |
| 1535 |
* Set checksum for a page in private memory. |
| 1536 |
* |
| 1537 |
* This must only be used when we know that no other process can be modifying |
| 1538 |
* the page buffer. |
| 1539 |
*/ |
| 1540 |
void |
| 1541 |
PageSetChecksumInplace(Page page, BlockNumber blkno) |
| 1542 |
{ |
| 1543 |
/* If we don't need a checksum, just return */ |
| 1544 |
if (PageIsNew(page) || !DataChecksumsEnabled()) |
| 1545 |
return; |
| 1546 |
|
| 1547 |
((PageHeader) page)->pd_checksum = pg_checksum_page(page, blkno); |
| 1548 |
} |