Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * xloginsert.c
4 : * Functions for constructing WAL records
5 : *
6 : * Constructing a WAL record begins with a call to XLogBeginInsert,
7 : * followed by a number of XLogRegister* calls. The registered data is
8 : * collected in private working memory, and finally assembled into a chain
9 : * of XLogRecData structs by a call to XLogRecordAssemble(). See
10 : * access/transam/README for details.
11 : *
12 : * Portions Copyright (c) 1996-2017, PostgreSQL Global Development Group
13 : * Portions Copyright (c) 1994, Regents of the University of California
14 : *
15 : * src/backend/access/transam/xloginsert.c
16 : *
17 : *-------------------------------------------------------------------------
18 : */
19 :
20 : #include "postgres.h"
21 :
22 : #include "access/xact.h"
23 : #include "access/xlog.h"
24 : #include "access/xlog_internal.h"
25 : #include "access/xloginsert.h"
26 : #include "catalog/pg_control.h"
27 : #include "common/pg_lzcompress.h"
28 : #include "miscadmin.h"
29 : #include "replication/origin.h"
30 : #include "storage/bufmgr.h"
31 : #include "storage/proc.h"
32 : #include "utils/memutils.h"
33 : #include "pg_trace.h"
34 :
35 : /* Buffer size required to store a compressed version of backup block image */
36 : #define PGLZ_MAX_BLCKSZ PGLZ_MAX_OUTPUT(BLCKSZ)
37 :
38 : /*
39 : * For each block reference registered with XLogRegisterBuffer, we fill in
40 : * a registered_buffer struct.
41 : */
42 : typedef struct
43 : {
44 : bool in_use; /* is this slot in use? */
45 : uint8 flags; /* REGBUF_* flags */
46 : RelFileNode rnode; /* identifies the relation and block */
47 : ForkNumber forkno;
48 : BlockNumber block;
49 : Page page; /* page content */
50 : uint32 rdata_len; /* total length of data in rdata chain */
51 : XLogRecData *rdata_head; /* head of the chain of data registered with
52 : * this block */
53 : XLogRecData *rdata_tail; /* last entry in the chain, or &rdata_head if
54 : * empty */
55 :
56 : XLogRecData bkp_rdatas[2]; /* temporary rdatas used to hold references to
57 : * backup block data in XLogRecordAssemble() */
58 :
59 : /* buffer to store a compressed version of backup block image */
60 : char compressed_page[PGLZ_MAX_BLCKSZ];
61 : } registered_buffer;
62 :
63 : static registered_buffer *registered_buffers;
64 : static int max_registered_buffers; /* allocated size */
65 : static int max_registered_block_id = 0; /* highest block_id + 1 currently
66 : * registered */
67 :
68 : /*
69 : * A chain of XLogRecDatas to hold the "main data" of a WAL record, registered
70 : * with XLogRegisterData(...).
71 : */
72 : static XLogRecData *mainrdata_head;
73 : static XLogRecData *mainrdata_last = (XLogRecData *) &mainrdata_head;
74 : static uint32 mainrdata_len; /* total # of bytes in chain */
75 :
76 : /* flags for the in-progress insertion */
77 : static uint8 curinsert_flags = 0;
78 :
79 : /*
80 : * These are used to hold the record header while constructing a record.
81 : * 'hdr_scratch' is not a plain variable, but is palloc'd at initialization,
82 : * because we want it to be MAXALIGNed and padding bytes zeroed.
83 : *
84 : * For simplicity, it's allocated large enough to hold the headers for any
85 : * WAL record.
86 : */
87 : static XLogRecData hdr_rdt;
88 : static char *hdr_scratch = NULL;
89 :
90 : #define SizeOfXlogOrigin (sizeof(RepOriginId) + sizeof(char))
91 :
92 : #define HEADER_SCRATCH_SIZE \
93 : (SizeOfXLogRecord + \
94 : MaxSizeOfXLogRecordBlockHeader * (XLR_MAX_BLOCK_ID + 1) + \
95 : SizeOfXLogRecordDataHeaderLong + SizeOfXlogOrigin)
96 :
97 : /*
98 : * An array of XLogRecData structs, to hold registered data.
99 : */
100 : static XLogRecData *rdatas;
101 : static int num_rdatas; /* entries currently used */
102 : static int max_rdatas; /* allocated size */
103 :
104 : static bool begininsert_called = false;
105 :
106 : /* Memory context to hold the registered buffer and data references. */
107 : static MemoryContext xloginsert_cxt;
108 :
109 : static XLogRecData *XLogRecordAssemble(RmgrId rmid, uint8 info,
110 : XLogRecPtr RedoRecPtr, bool doPageWrites,
111 : XLogRecPtr *fpw_lsn);
112 : static bool XLogCompressBackupBlock(char *page, uint16 hole_offset,
113 : uint16 hole_length, char *dest, uint16 *dlen);
114 :
115 : /*
116 : * Begin constructing a WAL record. This must be called before the
117 : * XLogRegister* functions and XLogInsert().
118 : */
119 : void
120 1397899 : XLogBeginInsert(void)
121 : {
122 1397899 : Assert(max_registered_block_id == 0);
123 1397899 : Assert(mainrdata_last == (XLogRecData *) &mainrdata_head);
124 1397899 : Assert(mainrdata_len == 0);
125 :
126 : /* cross-check on whether we should be here or not */
127 1397899 : if (!XLogInsertAllowed())
128 0 : elog(ERROR, "cannot make new WAL entries during recovery");
129 :
130 1397899 : if (begininsert_called)
131 0 : elog(ERROR, "XLogBeginInsert was already called");
132 :
133 1397899 : begininsert_called = true;
134 1397899 : }
135 :
136 : /*
137 : * Ensure that there are enough buffer and data slots in the working area,
138 : * for subsequent XLogRegisterBuffer, XLogRegisterData and XLogRegisterBufData
139 : * calls.
140 : *
141 : * There is always space for a small number of buffers and data chunks, enough
142 : * for most record types. This function is for the exceptional cases that need
143 : * more.
144 : */
145 : void
146 1321 : XLogEnsureRecordSpace(int max_block_id, int ndatas)
147 : {
148 : int nbuffers;
149 :
150 : /*
151 : * This must be called before entering a critical section, because
152 : * allocating memory inside a critical section can fail. repalloc() will
153 : * check the same, but better to check it here too so that we fail
154 : * consistently even if the arrays happen to be large enough already.
155 : */
156 1321 : Assert(CritSectionCount == 0);
157 :
158 : /* the minimum values can't be decreased */
159 1321 : if (max_block_id < XLR_NORMAL_MAX_BLOCK_ID)
160 1282 : max_block_id = XLR_NORMAL_MAX_BLOCK_ID;
161 1321 : if (ndatas < XLR_NORMAL_RDATAS)
162 1309 : ndatas = XLR_NORMAL_RDATAS;
163 :
164 1321 : if (max_block_id > XLR_MAX_BLOCK_ID)
165 0 : elog(ERROR, "maximum number of WAL record block references exceeded");
166 1321 : nbuffers = max_block_id + 1;
167 :
168 1321 : if (nbuffers > max_registered_buffers)
169 : {
170 2 : registered_buffers = (registered_buffer *)
171 2 : repalloc(registered_buffers, sizeof(registered_buffer) * nbuffers);
172 :
173 : /*
174 : * At least the padding bytes in the structs must be zeroed, because
175 : * they are included in WAL data, but initialize it all for tidiness.
176 : */
177 2 : MemSet(®istered_buffers[max_registered_buffers], 0,
178 : (nbuffers - max_registered_buffers) * sizeof(registered_buffer));
179 2 : max_registered_buffers = nbuffers;
180 : }
181 :
182 1321 : if (ndatas > max_rdatas)
183 : {
184 3 : rdatas = (XLogRecData *) repalloc(rdatas, sizeof(XLogRecData) * ndatas);
185 3 : max_rdatas = ndatas;
186 : }
187 1321 : }
188 :
189 : /*
190 : * Reset WAL record construction buffers.
191 : */
192 : void
193 1401528 : XLogResetInsertion(void)
194 : {
195 : int i;
196 :
197 2895300 : for (i = 0; i < max_registered_block_id; i++)
198 1493772 : registered_buffers[i].in_use = false;
199 :
200 1401528 : num_rdatas = 0;
201 1401528 : max_registered_block_id = 0;
202 1401528 : mainrdata_len = 0;
203 1401528 : mainrdata_last = (XLogRecData *) &mainrdata_head;
204 1401528 : curinsert_flags = 0;
205 1401528 : begininsert_called = false;
206 1401528 : }
207 :
208 : /*
209 : * Register a reference to a buffer with the WAL record being constructed.
210 : * This must be called for every page that the WAL-logged operation modifies.
211 : */
212 : void
213 1488047 : XLogRegisterBuffer(uint8 block_id, Buffer buffer, uint8 flags)
214 : {
215 : registered_buffer *regbuf;
216 :
217 : /* NO_IMAGE doesn't make sense with FORCE_IMAGE */
218 1488047 : Assert(!((flags & REGBUF_FORCE_IMAGE) && (flags & (REGBUF_NO_IMAGE))));
219 1488047 : Assert(begininsert_called);
220 :
221 1488047 : if (block_id >= max_registered_block_id)
222 : {
223 1384697 : if (block_id >= max_registered_buffers)
224 0 : elog(ERROR, "too many registered buffers");
225 1384697 : max_registered_block_id = block_id + 1;
226 : }
227 :
228 1488047 : regbuf = ®istered_buffers[block_id];
229 :
230 1488047 : BufferGetTag(buffer, ®buf->rnode, ®buf->forkno, ®buf->block);
231 1488047 : regbuf->page = BufferGetPage(buffer);
232 1488047 : regbuf->flags = flags;
233 1488047 : regbuf->rdata_tail = (XLogRecData *) ®buf->rdata_head;
234 1488047 : regbuf->rdata_len = 0;
235 :
236 : /*
237 : * Check that this page hasn't already been registered with some other
238 : * block_id.
239 : */
240 : #ifdef USE_ASSERT_CHECKING
241 : {
242 : int i;
243 :
244 3202696 : for (i = 0; i < max_registered_block_id; i++)
245 : {
246 1714649 : registered_buffer *regbuf_old = ®istered_buffers[i];
247 :
248 1714649 : if (i == block_id || !regbuf_old->in_use)
249 1594748 : continue;
250 :
251 119901 : Assert(!RelFileNodeEquals(regbuf_old->rnode, regbuf->rnode) ||
252 : regbuf_old->forkno != regbuf->forkno ||
253 : regbuf_old->block != regbuf->block);
254 : }
255 : }
256 : #endif
257 :
258 1488047 : regbuf->in_use = true;
259 1488047 : }
260 :
261 : /*
262 : * Like XLogRegisterBuffer, but for registering a block that's not in the
263 : * shared buffer pool (i.e. when you don't have a Buffer for it).
264 : */
265 : void
266 3827 : XLogRegisterBlock(uint8 block_id, RelFileNode *rnode, ForkNumber forknum,
267 : BlockNumber blknum, Page page, uint8 flags)
268 : {
269 : registered_buffer *regbuf;
270 :
271 : /* This is currently only used to WAL-log a full-page image of a page */
272 3827 : Assert(flags & REGBUF_FORCE_IMAGE);
273 3827 : Assert(begininsert_called);
274 :
275 3827 : if (block_id >= max_registered_block_id)
276 3827 : max_registered_block_id = block_id + 1;
277 :
278 3827 : if (block_id >= max_registered_buffers)
279 0 : elog(ERROR, "too many registered buffers");
280 :
281 3827 : regbuf = ®istered_buffers[block_id];
282 :
283 3827 : regbuf->rnode = *rnode;
284 3827 : regbuf->forkno = forknum;
285 3827 : regbuf->block = blknum;
286 3827 : regbuf->page = page;
287 3827 : regbuf->flags = flags;
288 3827 : regbuf->rdata_tail = (XLogRecData *) ®buf->rdata_head;
289 3827 : regbuf->rdata_len = 0;
290 :
291 : /*
292 : * Check that this page hasn't already been registered with some other
293 : * block_id.
294 : */
295 : #ifdef USE_ASSERT_CHECKING
296 : {
297 : int i;
298 :
299 7654 : for (i = 0; i < max_registered_block_id; i++)
300 : {
301 3827 : registered_buffer *regbuf_old = ®istered_buffers[i];
302 :
303 3827 : if (i == block_id || !regbuf_old->in_use)
304 3827 : continue;
305 :
306 0 : Assert(!RelFileNodeEquals(regbuf_old->rnode, regbuf->rnode) ||
307 : regbuf_old->forkno != regbuf->forkno ||
308 : regbuf_old->block != regbuf->block);
309 : }
310 : }
311 : #endif
312 :
313 3827 : regbuf->in_use = true;
314 3827 : }
315 :
316 : /*
317 : * Add data to the WAL record that's being constructed.
318 : *
319 : * The data is appended to the "main chunk", available at replay with
320 : * XLogRecGetData().
321 : */
322 : void
323 1520170 : XLogRegisterData(char *data, int len)
324 : {
325 : XLogRecData *rdata;
326 :
327 1520170 : Assert(begininsert_called);
328 :
329 1520170 : if (num_rdatas >= max_rdatas)
330 0 : elog(ERROR, "too much WAL data");
331 1520170 : rdata = &rdatas[num_rdatas++];
332 :
333 1520170 : rdata->data = data;
334 1520170 : rdata->len = len;
335 :
336 : /*
337 : * we use the mainrdata_last pointer to track the end of the chain, so no
338 : * need to clear 'next' here.
339 : */
340 :
341 1520170 : mainrdata_last->next = rdata;
342 1520170 : mainrdata_last = rdata;
343 :
344 1520170 : mainrdata_len += len;
345 1520170 : }
346 :
347 : /*
348 : * Add buffer-specific data to the WAL record that's being constructed.
349 : *
350 : * Block_id must reference a block previously registered with
351 : * XLogRegisterBuffer(). If this is called more than once for the same
352 : * block_id, the data is appended.
353 : *
354 : * The maximum amount of data that can be registered per block is 65535
355 : * bytes. That should be plenty; if you need more than BLCKSZ bytes to
356 : * reconstruct the changes to the page, you might as well just log a full
357 : * copy of it. (the "main data" that's not associated with a block is not
358 : * limited)
359 : */
360 : void
361 2032147 : XLogRegisterBufData(uint8 block_id, char *data, int len)
362 : {
363 : registered_buffer *regbuf;
364 : XLogRecData *rdata;
365 :
366 2032147 : Assert(begininsert_called);
367 :
368 : /* find the registered buffer struct */
369 2032147 : regbuf = ®istered_buffers[block_id];
370 2032147 : if (!regbuf->in_use)
371 0 : elog(ERROR, "no block with id %d registered with WAL insertion",
372 : block_id);
373 :
374 2032147 : if (num_rdatas >= max_rdatas)
375 0 : elog(ERROR, "too much WAL data");
376 2032147 : rdata = &rdatas[num_rdatas++];
377 :
378 2032147 : rdata->data = data;
379 2032147 : rdata->len = len;
380 :
381 2032147 : regbuf->rdata_tail->next = rdata;
382 2032147 : regbuf->rdata_tail = rdata;
383 2032147 : regbuf->rdata_len += len;
384 2032147 : }
385 :
386 : /*
387 : * Set insert status flags for the upcoming WAL record.
388 : *
389 : * The flags that can be used here are:
390 : * - XLOG_INCLUDE_ORIGIN, to determine if the replication origin should be
391 : * included in the record.
392 : * - XLOG_MARK_UNIMPORTANT, to signal that the record is not important for
393 : * durability, which allows to avoid triggering WAL archiving and other
394 : * background activity.
395 : */
396 : void
397 733889 : XLogSetRecordFlags(uint8 flags)
398 : {
399 733889 : Assert(begininsert_called);
400 733889 : curinsert_flags = flags;
401 733889 : }
402 :
403 : /*
404 : * Insert an XLOG record having the specified RMID and info bytes, with the
405 : * body of the record being the data and buffer references registered earlier
406 : * with XLogRegister* calls.
407 : *
408 : * Returns XLOG pointer to end of record (beginning of next record).
409 : * This can be used as LSN for data pages affected by the logged action.
410 : * (LSN is the XLOG point up to which the XLOG must be flushed to disk
411 : * before the data page can be written out. This implements the basic
412 : * WAL rule "write the log before the data".)
413 : */
414 : XLogRecPtr
415 1397899 : XLogInsert(RmgrId rmid, uint8 info)
416 : {
417 : XLogRecPtr EndPos;
418 :
419 : /* XLogBeginInsert() must have been called. */
420 1397899 : if (!begininsert_called)
421 0 : elog(ERROR, "XLogBeginInsert was not called");
422 :
423 : /*
424 : * The caller can set rmgr bits, XLR_SPECIAL_REL_UPDATE and
425 : * XLR_CHECK_CONSISTENCY; the rest are reserved for use by me.
426 : */
427 1397899 : if ((info & ~(XLR_RMGR_INFO_MASK |
428 : XLR_SPECIAL_REL_UPDATE |
429 : XLR_CHECK_CONSISTENCY)) != 0)
430 0 : elog(PANIC, "invalid xlog info mask %02X", info);
431 :
432 : TRACE_POSTGRESQL_WAL_INSERT(rmid, info);
433 :
434 : /*
435 : * In bootstrap mode, we don't actually log anything but XLOG resources;
436 : * return a phony record pointer.
437 : */
438 1397899 : if (IsBootstrapProcessingMode() && rmid != RM_XLOG_ID)
439 : {
440 7793 : XLogResetInsertion();
441 7793 : EndPos = SizeOfXLogLongPHD; /* start of 1st chkpt record */
442 7793 : return EndPos;
443 : }
444 :
445 : do
446 : {
447 : XLogRecPtr RedoRecPtr;
448 : bool doPageWrites;
449 : XLogRecPtr fpw_lsn;
450 : XLogRecData *rdt;
451 :
452 : /*
453 : * Get values needed to decide whether to do full-page writes. Since
454 : * we don't yet have an insertion lock, these could change under us,
455 : * but XLogInsertRecord will recheck them once it has a lock.
456 : */
457 1390106 : GetFullPageWriteInfo(&RedoRecPtr, &doPageWrites);
458 :
459 1390106 : rdt = XLogRecordAssemble(rmid, info, RedoRecPtr, doPageWrites,
460 : &fpw_lsn);
461 :
462 1390106 : EndPos = XLogInsertRecord(rdt, fpw_lsn, curinsert_flags);
463 1390106 : } while (EndPos == InvalidXLogRecPtr);
464 :
465 1390106 : XLogResetInsertion();
466 :
467 1390106 : return EndPos;
468 : }
469 :
470 : /*
471 : * Assemble a WAL record from the registered data and buffers into an
472 : * XLogRecData chain, ready for insertion with XLogInsertRecord().
473 : *
474 : * The record header fields are filled in, except for the xl_prev field. The
475 : * calculated CRC does not include the record header yet.
476 : *
477 : * If there are any registered buffers, and a full-page image was not taken
478 : * of all of them, *fpw_lsn is set to the lowest LSN among such pages. This
479 : * signals that the assembled record is only good for insertion on the
480 : * assumption that the RedoRecPtr and doPageWrites values were up-to-date.
481 : */
482 : static XLogRecData *
483 1390106 : XLogRecordAssemble(RmgrId rmid, uint8 info,
484 : XLogRecPtr RedoRecPtr, bool doPageWrites,
485 : XLogRecPtr *fpw_lsn)
486 : {
487 : XLogRecData *rdt;
488 1390106 : uint32 total_len = 0;
489 : int block_id;
490 : pg_crc32c rdata_crc;
491 1390106 : registered_buffer *prev_regbuf = NULL;
492 : XLogRecData *rdt_datas_last;
493 : XLogRecord *rechdr;
494 1390106 : char *scratch = hdr_scratch;
495 :
496 : /*
497 : * Note: this function can be called multiple times for the same record.
498 : * All the modifications we do to the rdata chains below must handle that.
499 : */
500 :
501 : /* The record begins with the fixed-size header */
502 1390106 : rechdr = (XLogRecord *) scratch;
503 1390106 : scratch += SizeOfXLogRecord;
504 :
505 1390106 : hdr_rdt.next = NULL;
506 1390106 : rdt_datas_last = &hdr_rdt;
507 1390106 : hdr_rdt.data = hdr_scratch;
508 :
509 : /*
510 : * Enforce consistency checks for this record if user is looking for it.
511 : * Do this before at the beginning of this routine to give the possibility
512 : * for callers of XLogInsert() to pass XLR_CHECK_CONSISTENCY directly for
513 : * a record.
514 : */
515 1390106 : if (wal_consistency_checking[rmid])
516 0 : info |= XLR_CHECK_CONSISTENCY;
517 :
518 : /*
519 : * Make an rdata chain containing all the data portions of all block
520 : * references. This includes the data for full-page images. Also append
521 : * the headers for the block references in the scratch buffer.
522 : */
523 1390106 : *fpw_lsn = InvalidXLogRecPtr;
524 2876426 : for (block_id = 0; block_id < max_registered_block_id; block_id++)
525 : {
526 1486320 : registered_buffer *regbuf = ®istered_buffers[block_id];
527 : bool needs_backup;
528 : bool needs_data;
529 : XLogRecordBlockHeader bkpb;
530 : XLogRecordBlockImageHeader bimg;
531 1486320 : XLogRecordBlockCompressHeader cbimg = {0};
532 : bool samerel;
533 1486320 : bool is_compressed = false;
534 : bool include_image;
535 :
536 1486320 : if (!regbuf->in_use)
537 1898 : continue;
538 :
539 : /* Determine if this block needs to be backed up */
540 1484422 : if (regbuf->flags & REGBUF_FORCE_IMAGE)
541 4470 : needs_backup = true;
542 1479952 : else if (regbuf->flags & REGBUF_NO_IMAGE)
543 40316 : needs_backup = false;
544 1439636 : else if (!doPageWrites)
545 0 : needs_backup = false;
546 : else
547 : {
548 : /*
549 : * We assume page LSN is first data on *every* page that can be
550 : * passed to XLogInsert, whether it has the standard page layout
551 : * or not.
552 : */
553 1439636 : XLogRecPtr page_lsn = PageGetLSN(regbuf->page);
554 :
555 1439636 : needs_backup = (page_lsn <= RedoRecPtr);
556 1439636 : if (!needs_backup)
557 : {
558 1438994 : if (*fpw_lsn == InvalidXLogRecPtr || page_lsn < *fpw_lsn)
559 1352842 : *fpw_lsn = page_lsn;
560 : }
561 : }
562 :
563 : /* Determine if the buffer data needs to included */
564 1484422 : if (regbuf->rdata_len == 0)
565 321428 : needs_data = false;
566 1162994 : else if ((regbuf->flags & REGBUF_KEEP_DATA) != 0)
567 0 : needs_data = true;
568 : else
569 1162994 : needs_data = !needs_backup;
570 :
571 1484422 : bkpb.id = block_id;
572 1484422 : bkpb.fork_flags = regbuf->forkno;
573 1484422 : bkpb.data_length = 0;
574 :
575 1484422 : if ((regbuf->flags & REGBUF_WILL_INIT) == REGBUF_WILL_INIT)
576 36256 : bkpb.fork_flags |= BKPBLOCK_WILL_INIT;
577 :
578 : /*
579 : * If needs_backup is true or WAL checking is enabled for current
580 : * resource manager, log a full-page write for the current block.
581 : */
582 1484422 : include_image = needs_backup || (info & XLR_CHECK_CONSISTENCY) != 0;
583 :
584 1484422 : if (include_image)
585 : {
586 5112 : Page page = regbuf->page;
587 : uint16 compressed_len;
588 :
589 : /*
590 : * The page needs to be backed up, so calculate its hole length
591 : * and offset.
592 : */
593 5112 : if (regbuf->flags & REGBUF_STANDARD)
594 : {
595 : /* Assume we can omit data between pd_lower and pd_upper */
596 4885 : uint16 lower = ((PageHeader) page)->pd_lower;
597 4885 : uint16 upper = ((PageHeader) page)->pd_upper;
598 :
599 4885 : if (lower >= SizeOfPageHeaderData &&
600 4883 : upper > lower &&
601 : upper <= BLCKSZ)
602 : {
603 4883 : bimg.hole_offset = lower;
604 4883 : cbimg.hole_length = upper - lower;
605 : }
606 : else
607 : {
608 : /* No "hole" to compress out */
609 2 : bimg.hole_offset = 0;
610 2 : cbimg.hole_length = 0;
611 : }
612 : }
613 : else
614 : {
615 : /* Not a standard page header, don't try to eliminate "hole" */
616 227 : bimg.hole_offset = 0;
617 227 : cbimg.hole_length = 0;
618 : }
619 :
620 : /*
621 : * Try to compress a block image if wal_compression is enabled
622 : */
623 5112 : if (wal_compression)
624 : {
625 0 : is_compressed =
626 0 : XLogCompressBackupBlock(page, bimg.hole_offset,
627 0 : cbimg.hole_length,
628 0 : regbuf->compressed_page,
629 : &compressed_len);
630 : }
631 :
632 : /*
633 : * Fill in the remaining fields in the XLogRecordBlockHeader
634 : * struct
635 : */
636 5112 : bkpb.fork_flags |= BKPBLOCK_HAS_IMAGE;
637 :
638 : /*
639 : * Construct XLogRecData entries for the page content.
640 : */
641 5112 : rdt_datas_last->next = ®buf->bkp_rdatas[0];
642 5112 : rdt_datas_last = rdt_datas_last->next;
643 :
644 5112 : bimg.bimg_info = (cbimg.hole_length == 0) ? 0 : BKPIMAGE_HAS_HOLE;
645 :
646 : /*
647 : * If WAL consistency checking is enabled for the resource manager
648 : * of this WAL record, a full-page image is included in the record
649 : * for the block modified. During redo, the full-page is replayed
650 : * only if BKPIMAGE_APPLY is set.
651 : */
652 5112 : if (needs_backup)
653 5112 : bimg.bimg_info |= BKPIMAGE_APPLY;
654 :
655 5112 : if (is_compressed)
656 : {
657 0 : bimg.length = compressed_len;
658 0 : bimg.bimg_info |= BKPIMAGE_IS_COMPRESSED;
659 :
660 0 : rdt_datas_last->data = regbuf->compressed_page;
661 0 : rdt_datas_last->len = compressed_len;
662 : }
663 : else
664 : {
665 5112 : bimg.length = BLCKSZ - cbimg.hole_length;
666 :
667 5112 : if (cbimg.hole_length == 0)
668 : {
669 229 : rdt_datas_last->data = page;
670 229 : rdt_datas_last->len = BLCKSZ;
671 : }
672 : else
673 : {
674 : /* must skip the hole */
675 4883 : rdt_datas_last->data = page;
676 4883 : rdt_datas_last->len = bimg.hole_offset;
677 :
678 4883 : rdt_datas_last->next = ®buf->bkp_rdatas[1];
679 4883 : rdt_datas_last = rdt_datas_last->next;
680 :
681 4883 : rdt_datas_last->data =
682 4883 : page + (bimg.hole_offset + cbimg.hole_length);
683 4883 : rdt_datas_last->len =
684 4883 : BLCKSZ - (bimg.hole_offset + cbimg.hole_length);
685 : }
686 : }
687 :
688 5112 : total_len += bimg.length;
689 : }
690 :
691 1484422 : if (needs_data)
692 : {
693 : /*
694 : * Link the caller-supplied rdata chain for this buffer to the
695 : * overall list.
696 : */
697 1162679 : bkpb.fork_flags |= BKPBLOCK_HAS_DATA;
698 1162679 : bkpb.data_length = regbuf->rdata_len;
699 1162679 : total_len += regbuf->rdata_len;
700 :
701 1162679 : rdt_datas_last->next = regbuf->rdata_head;
702 1162679 : rdt_datas_last = regbuf->rdata_tail;
703 : }
704 :
705 1484422 : if (prev_regbuf && RelFileNodeEquals(regbuf->rnode, prev_regbuf->rnode))
706 : {
707 117371 : samerel = true;
708 117371 : bkpb.fork_flags |= BKPBLOCK_SAME_REL;
709 : }
710 : else
711 1367051 : samerel = false;
712 1484422 : prev_regbuf = regbuf;
713 :
714 : /* Ok, copy the header to the scratch buffer */
715 1484422 : memcpy(scratch, &bkpb, SizeOfXLogRecordBlockHeader);
716 1484422 : scratch += SizeOfXLogRecordBlockHeader;
717 1484422 : if (include_image)
718 : {
719 5112 : memcpy(scratch, &bimg, SizeOfXLogRecordBlockImageHeader);
720 5112 : scratch += SizeOfXLogRecordBlockImageHeader;
721 5112 : if (cbimg.hole_length != 0 && is_compressed)
722 : {
723 0 : memcpy(scratch, &cbimg,
724 : SizeOfXLogRecordBlockCompressHeader);
725 0 : scratch += SizeOfXLogRecordBlockCompressHeader;
726 : }
727 : }
728 1484422 : if (!samerel)
729 : {
730 1367051 : memcpy(scratch, ®buf->rnode, sizeof(RelFileNode));
731 1367051 : scratch += sizeof(RelFileNode);
732 : }
733 1484422 : memcpy(scratch, ®buf->block, sizeof(BlockNumber));
734 1484422 : scratch += sizeof(BlockNumber);
735 : }
736 :
737 : /* followed by the record's origin, if any */
738 2107911 : if ((curinsert_flags & XLOG_INCLUDE_ORIGIN) &&
739 717805 : replorigin_session_origin != InvalidRepOriginId)
740 : {
741 0 : *(scratch++) = (char) XLR_BLOCK_ID_ORIGIN;
742 0 : memcpy(scratch, &replorigin_session_origin, sizeof(replorigin_session_origin));
743 0 : scratch += sizeof(replorigin_session_origin);
744 : }
745 :
746 : /* followed by main data, if any */
747 1390106 : if (mainrdata_len > 0)
748 : {
749 1385953 : if (mainrdata_len > 255)
750 : {
751 3761 : *(scratch++) = (char) XLR_BLOCK_ID_DATA_LONG;
752 3761 : memcpy(scratch, &mainrdata_len, sizeof(uint32));
753 3761 : scratch += sizeof(uint32);
754 : }
755 : else
756 : {
757 1382192 : *(scratch++) = (char) XLR_BLOCK_ID_DATA_SHORT;
758 1382192 : *(scratch++) = (uint8) mainrdata_len;
759 : }
760 1385953 : rdt_datas_last->next = mainrdata_head;
761 1385953 : rdt_datas_last = mainrdata_last;
762 1385953 : total_len += mainrdata_len;
763 : }
764 1390106 : rdt_datas_last->next = NULL;
765 :
766 1390106 : hdr_rdt.len = (scratch - hdr_scratch);
767 1390106 : total_len += hdr_rdt.len;
768 :
769 : /*
770 : * Calculate CRC of the data
771 : *
772 : * Note that the record header isn't added into the CRC initially since we
773 : * don't know the prev-link yet. Thus, the CRC will represent the CRC of
774 : * the whole record in the order: rdata, then backup blocks, then record
775 : * header.
776 : */
777 1390106 : INIT_CRC32C(rdata_crc);
778 1390106 : COMP_CRC32C(rdata_crc, hdr_scratch + SizeOfXLogRecord, hdr_rdt.len - SizeOfXLogRecord);
779 4929382 : for (rdt = hdr_rdt.next; rdt != NULL; rdt = rdt->next)
780 3539276 : COMP_CRC32C(rdata_crc, rdt->data, rdt->len);
781 :
782 : /*
783 : * Fill in the fields in the record header. Prev-link is filled in later,
784 : * once we know where in the WAL the record will be inserted. The CRC does
785 : * not include the record header yet.
786 : */
787 1390106 : rechdr->xl_xid = GetCurrentTransactionIdIfAny();
788 1390106 : rechdr->xl_tot_len = total_len;
789 1390106 : rechdr->xl_info = info;
790 1390106 : rechdr->xl_rmid = rmid;
791 1390106 : rechdr->xl_prev = InvalidXLogRecPtr;
792 1390106 : rechdr->xl_crc = rdata_crc;
793 :
794 1390106 : return &hdr_rdt;
795 : }
796 :
797 : /*
798 : * Create a compressed version of a backup block image.
799 : *
800 : * Returns FALSE if compression fails (i.e., compressed result is actually
801 : * bigger than original). Otherwise, returns TRUE and sets 'dlen' to
802 : * the length of compressed block image.
803 : */
804 : static bool
805 0 : XLogCompressBackupBlock(char *page, uint16 hole_offset, uint16 hole_length,
806 : char *dest, uint16 *dlen)
807 : {
808 0 : int32 orig_len = BLCKSZ - hole_length;
809 : int32 len;
810 0 : int32 extra_bytes = 0;
811 : char *source;
812 : char tmp[BLCKSZ];
813 :
814 0 : if (hole_length != 0)
815 : {
816 : /* must skip the hole */
817 0 : source = tmp;
818 0 : memcpy(source, page, hole_offset);
819 0 : memcpy(source + hole_offset,
820 0 : page + (hole_offset + hole_length),
821 0 : BLCKSZ - (hole_length + hole_offset));
822 :
823 : /*
824 : * Extra data needs to be stored in WAL record for the compressed
825 : * version of block image if the hole exists.
826 : */
827 0 : extra_bytes = SizeOfXLogRecordBlockCompressHeader;
828 : }
829 : else
830 0 : source = page;
831 :
832 : /*
833 : * We recheck the actual size even if pglz_compress() reports success and
834 : * see if the number of bytes saved by compression is larger than the
835 : * length of extra data needed for the compressed version of block image.
836 : */
837 0 : len = pglz_compress(source, orig_len, dest, PGLZ_strategy_default);
838 0 : if (len >= 0 &&
839 0 : len + extra_bytes < orig_len)
840 : {
841 0 : *dlen = (uint16) len; /* successful compression */
842 0 : return true;
843 : }
844 0 : return false;
845 : }
846 :
847 : /*
848 : * Determine whether the buffer referenced has to be backed up.
849 : *
850 : * Since we don't yet have the insert lock, fullPageWrites and forcePageWrites
851 : * could change later, so the result should be used for optimization purposes
852 : * only.
853 : */
854 : bool
855 6222 : XLogCheckBufferNeedsBackup(Buffer buffer)
856 : {
857 : XLogRecPtr RedoRecPtr;
858 : bool doPageWrites;
859 : Page page;
860 :
861 6222 : GetFullPageWriteInfo(&RedoRecPtr, &doPageWrites);
862 :
863 6222 : page = BufferGetPage(buffer);
864 :
865 6222 : if (doPageWrites && PageGetLSN(page) <= RedoRecPtr)
866 5 : return true; /* buffer requires backup */
867 :
868 6217 : return false; /* buffer does not need to be backed up */
869 : }
870 :
871 : /*
872 : * Write a backup block if needed when we are setting a hint. Note that
873 : * this may be called for a variety of page types, not just heaps.
874 : *
875 : * Callable while holding just share lock on the buffer content.
876 : *
877 : * We can't use the plain backup block mechanism since that relies on the
878 : * Buffer being exclusively locked. Since some modifications (setting LSN, hint
879 : * bits) are allowed in a sharelocked buffer that can lead to wal checksum
880 : * failures. So instead we copy the page and insert the copied data as normal
881 : * record data.
882 : *
883 : * We only need to do something if page has not yet been full page written in
884 : * this checkpoint round. The LSN of the inserted wal record is returned if we
885 : * had to write, InvalidXLogRecPtr otherwise.
886 : *
887 : * It is possible that multiple concurrent backends could attempt to write WAL
888 : * records. In that case, multiple copies of the same block would be recorded
889 : * in separate WAL records by different backends, though that is still OK from
890 : * a correctness perspective.
891 : */
892 : XLogRecPtr
893 0 : XLogSaveBufferForHint(Buffer buffer, bool buffer_std)
894 : {
895 0 : XLogRecPtr recptr = InvalidXLogRecPtr;
896 : XLogRecPtr lsn;
897 : XLogRecPtr RedoRecPtr;
898 :
899 : /*
900 : * Ensure no checkpoint can change our view of RedoRecPtr.
901 : */
902 0 : Assert(MyPgXact->delayChkpt);
903 :
904 : /*
905 : * Update RedoRecPtr so that we can make the right decision
906 : */
907 0 : RedoRecPtr = GetRedoRecPtr();
908 :
909 : /*
910 : * We assume page LSN is first data on *every* page that can be passed to
911 : * XLogInsert, whether it has the standard page layout or not. Since we're
912 : * only holding a share-lock on the page, we must take the buffer header
913 : * lock when we look at the LSN.
914 : */
915 0 : lsn = BufferGetLSNAtomic(buffer);
916 :
917 0 : if (lsn <= RedoRecPtr)
918 : {
919 : int flags;
920 : char copied_buffer[BLCKSZ];
921 0 : char *origdata = (char *) BufferGetBlock(buffer);
922 : RelFileNode rnode;
923 : ForkNumber forkno;
924 : BlockNumber blkno;
925 :
926 : /*
927 : * Copy buffer so we don't have to worry about concurrent hint bit or
928 : * lsn updates. We assume pd_lower/upper cannot be changed without an
929 : * exclusive lock, so the contents bkp are not racy.
930 : */
931 0 : if (buffer_std)
932 : {
933 : /* Assume we can omit data between pd_lower and pd_upper */
934 0 : Page page = BufferGetPage(buffer);
935 0 : uint16 lower = ((PageHeader) page)->pd_lower;
936 0 : uint16 upper = ((PageHeader) page)->pd_upper;
937 :
938 0 : memcpy(copied_buffer, origdata, lower);
939 0 : memcpy(copied_buffer + upper, origdata + upper, BLCKSZ - upper);
940 : }
941 : else
942 0 : memcpy(copied_buffer, origdata, BLCKSZ);
943 :
944 0 : XLogBeginInsert();
945 :
946 0 : flags = REGBUF_FORCE_IMAGE;
947 0 : if (buffer_std)
948 0 : flags |= REGBUF_STANDARD;
949 :
950 0 : BufferGetTag(buffer, &rnode, &forkno, &blkno);
951 0 : XLogRegisterBlock(0, &rnode, forkno, blkno, copied_buffer, flags);
952 :
953 0 : recptr = XLogInsert(RM_XLOG_ID, XLOG_FPI_FOR_HINT);
954 : }
955 :
956 0 : return recptr;
957 : }
958 :
959 : /*
960 : * Write a WAL record containing a full image of a page. Caller is responsible
961 : * for writing the page to disk after calling this routine.
962 : *
963 : * Note: If you're using this function, you should be building pages in private
964 : * memory and writing them directly to smgr. If you're using buffers, call
965 : * log_newpage_buffer instead.
966 : *
967 : * If the page follows the standard page layout, with a PageHeader and unused
968 : * space between pd_lower and pd_upper, set 'page_std' to TRUE. That allows
969 : * the unused space to be left out from the WAL record, making it smaller.
970 : */
971 : XLogRecPtr
972 3827 : log_newpage(RelFileNode *rnode, ForkNumber forkNum, BlockNumber blkno,
973 : Page page, bool page_std)
974 : {
975 : int flags;
976 : XLogRecPtr recptr;
977 :
978 3827 : flags = REGBUF_FORCE_IMAGE;
979 3827 : if (page_std)
980 3804 : flags |= REGBUF_STANDARD;
981 :
982 3827 : XLogBeginInsert();
983 3827 : XLogRegisterBlock(0, rnode, forkNum, blkno, page, flags);
984 3827 : recptr = XLogInsert(RM_XLOG_ID, XLOG_FPI);
985 :
986 : /*
987 : * The page may be uninitialized. If so, we can't set the LSN because that
988 : * would corrupt the page.
989 : */
990 3827 : if (!PageIsNew(page))
991 : {
992 3827 : PageSetLSN(page, recptr);
993 : }
994 :
995 3827 : return recptr;
996 : }
997 :
998 : /*
999 : * Write a WAL record containing a full image of a page.
1000 : *
1001 : * Caller should initialize the buffer and mark it dirty before calling this
1002 : * function. This function will set the page LSN.
1003 : *
1004 : * If the page follows the standard page layout, with a PageHeader and unused
1005 : * space between pd_lower and pd_upper, set 'page_std' to TRUE. That allows
1006 : * the unused space to be left out from the WAL record, making it smaller.
1007 : */
1008 : XLogRecPtr
1009 0 : log_newpage_buffer(Buffer buffer, bool page_std)
1010 : {
1011 0 : Page page = BufferGetPage(buffer);
1012 : RelFileNode rnode;
1013 : ForkNumber forkNum;
1014 : BlockNumber blkno;
1015 :
1016 : /* Shared buffers should be modified in a critical section. */
1017 0 : Assert(CritSectionCount > 0);
1018 :
1019 0 : BufferGetTag(buffer, &rnode, &forkNum, &blkno);
1020 :
1021 0 : return log_newpage(&rnode, forkNum, blkno, page, page_std);
1022 : }
1023 :
1024 : /*
1025 : * Allocate working buffers needed for WAL record construction.
1026 : */
1027 : void
1028 359 : InitXLogInsert(void)
1029 : {
1030 : /* Initialize the working areas */
1031 359 : if (xloginsert_cxt == NULL)
1032 : {
1033 342 : xloginsert_cxt = AllocSetContextCreate(TopMemoryContext,
1034 : "WAL record construction",
1035 : ALLOCSET_DEFAULT_SIZES);
1036 : }
1037 :
1038 359 : if (registered_buffers == NULL)
1039 : {
1040 342 : registered_buffers = (registered_buffer *)
1041 342 : MemoryContextAllocZero(xloginsert_cxt,
1042 : sizeof(registered_buffer) * (XLR_NORMAL_MAX_BLOCK_ID + 1));
1043 342 : max_registered_buffers = XLR_NORMAL_MAX_BLOCK_ID + 1;
1044 : }
1045 359 : if (rdatas == NULL)
1046 : {
1047 342 : rdatas = MemoryContextAlloc(xloginsert_cxt,
1048 : sizeof(XLogRecData) * XLR_NORMAL_RDATAS);
1049 342 : max_rdatas = XLR_NORMAL_RDATAS;
1050 : }
1051 :
1052 : /*
1053 : * Allocate a buffer to hold the header information for a WAL record.
1054 : */
1055 359 : if (hdr_scratch == NULL)
1056 342 : hdr_scratch = MemoryContextAllocZero(xloginsert_cxt,
1057 : HEADER_SCRATCH_SIZE);
1058 359 : }
|
