Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * xlogreader.c
4 : * Generic XLog reading facility
5 : *
6 : * Portions Copyright (c) 2013-2020, PostgreSQL Global Development Group
7 : *
8 : * IDENTIFICATION
9 : * src/backend/access/transam/xlogreader.c
10 : *
11 : * NOTES
12 : * See xlogreader.h for more notes on this facility.
13 : *
14 : * This file is compiled as both front-end and backend code, so it
15 : * may not use ereport, server-defined static variables, etc.
16 : *-------------------------------------------------------------------------
17 : */
18 : #include "postgres.h"
19 :
20 : #include <unistd.h>
21 :
22 : #include "access/transam.h"
23 : #include "access/xlog_internal.h"
24 : #include "access/xlogreader.h"
25 : #include "access/xlogrecord.h"
26 : #include "catalog/pg_control.h"
27 : #include "common/pg_lzcompress.h"
28 : #include "replication/origin.h"
29 :
30 : #ifndef FRONTEND
31 : #include "miscadmin.h"
32 : #include "pgstat.h"
33 : #include "utils/memutils.h"
34 : #endif
35 :
36 : static void report_invalid_record(XLogReaderState *state, const char *fmt,...)
37 : pg_attribute_printf(2, 3);
38 : static bool allocate_recordbuf(XLogReaderState *state, uint32 reclength);
39 : static int ReadPageInternal(XLogReaderState *state, XLogRecPtr pageptr,
40 : int reqLen);
41 : static void XLogReaderInvalReadState(XLogReaderState *state);
42 : static bool ValidXLogRecordHeader(XLogReaderState *state, XLogRecPtr RecPtr,
43 : XLogRecPtr PrevRecPtr, XLogRecord *record, bool randAccess);
44 : static bool ValidXLogRecord(XLogReaderState *state, XLogRecord *record,
45 : XLogRecPtr recptr);
46 : static void ResetDecoder(XLogReaderState *state);
47 :
48 : /* size of the buffer allocated for error message. */
49 : #define MAX_ERRORMSG_LEN 1000
50 :
51 : /*
52 : * Construct a string in state->errormsg_buf explaining what's wrong with
53 : * the current record being read.
54 : */
55 : static void
56 0 : report_invalid_record(XLogReaderState *state, const char *fmt,...)
57 : {
58 : va_list args;
59 :
60 0 : fmt = _(fmt);
61 :
62 0 : va_start(args, fmt);
63 0 : vsnprintf(state->errormsg_buf, MAX_ERRORMSG_LEN, fmt, args);
64 0 : va_end(args);
65 0 : }
66 :
67 : /*
68 : * Allocate and initialize a new XLogReader.
69 : *
70 : * Returns NULL if the xlogreader couldn't be allocated.
71 : */
72 : XLogReaderState *
73 72 : XLogReaderAllocate(int wal_segment_size, const char *waldir,
74 : XLogReaderRoutine *routine, void *private_data)
75 : {
76 : XLogReaderState *state;
77 :
78 : state = (XLogReaderState *)
79 72 : palloc_extended(sizeof(XLogReaderState),
80 : MCXT_ALLOC_NO_OOM | MCXT_ALLOC_ZERO);
81 72 : if (!state)
82 0 : return NULL;
83 :
84 : /* initialize caller-provided support functions */
85 72 : state->routine = *routine;
86 :
87 72 : state->max_block_id = -1;
88 :
89 : /*
90 : * Permanently allocate readBuf. We do it this way, rather than just
91 : * making a static array, for two reasons: (1) no need to waste the
92 : * storage in most instantiations of the backend; (2) a static char array
93 : * isn't guaranteed to have any particular alignment, whereas
94 : * palloc_extended() will provide MAXALIGN'd storage.
95 : */
96 72 : state->readBuf = (char *) palloc_extended(XLOG_BLCKSZ,
97 : MCXT_ALLOC_NO_OOM);
98 72 : if (!state->readBuf)
99 : {
100 0 : pfree(state);
101 0 : return NULL;
102 : }
103 :
104 : /* Initialize segment info. */
105 72 : WALOpenSegmentInit(&state->seg, &state->segcxt, wal_segment_size,
106 : waldir);
107 :
108 : /* system_identifier initialized to zeroes above */
109 72 : state->private_data = private_data;
110 : /* ReadRecPtr, EndRecPtr and readLen initialized to zeroes above */
111 72 : state->errormsg_buf = palloc_extended(MAX_ERRORMSG_LEN + 1,
112 : MCXT_ALLOC_NO_OOM);
113 72 : if (!state->errormsg_buf)
114 : {
115 0 : pfree(state->readBuf);
116 0 : pfree(state);
117 0 : return NULL;
118 : }
119 72 : state->errormsg_buf[0] = '\0';
120 :
121 : /*
122 : * Allocate an initial readRecordBuf of minimal size, which can later be
123 : * enlarged if necessary.
124 : */
125 72 : if (!allocate_recordbuf(state, 0))
126 : {
127 0 : pfree(state->errormsg_buf);
128 0 : pfree(state->readBuf);
129 0 : pfree(state);
130 0 : return NULL;
131 : }
132 :
133 72 : return state;
134 : }
135 :
136 : void
137 72 : XLogReaderFree(XLogReaderState *state)
138 : {
139 : int block_id;
140 :
141 72 : if (state->seg.ws_file != -1)
142 72 : state->routine.segment_close(state);
143 :
144 2448 : for (block_id = 0; block_id <= XLR_MAX_BLOCK_ID; block_id++)
145 : {
146 2376 : if (state->blocks[block_id].data)
147 0 : pfree(state->blocks[block_id].data);
148 : }
149 72 : if (state->main_data)
150 72 : pfree(state->main_data);
151 :
152 72 : pfree(state->errormsg_buf);
153 72 : if (state->readRecordBuf)
154 72 : pfree(state->readRecordBuf);
155 72 : pfree(state->readBuf);
156 72 : pfree(state);
157 72 : }
158 :
159 : /*
160 : * Allocate readRecordBuf to fit a record of at least the given length.
161 : * Returns true if successful, false if out of memory.
162 : *
163 : * readRecordBufSize is set to the new buffer size.
164 : *
165 : * To avoid useless small increases, round its size to a multiple of
166 : * XLOG_BLCKSZ, and make sure it's at least 5*Max(BLCKSZ, XLOG_BLCKSZ) to start
167 : * with. (That is enough for all "normal" records, but very large commit or
168 : * abort records might need more space.)
169 : */
170 : static bool
171 72 : allocate_recordbuf(XLogReaderState *state, uint32 reclength)
172 : {
173 72 : uint32 newSize = reclength;
174 :
175 72 : newSize += XLOG_BLCKSZ - (newSize % XLOG_BLCKSZ);
176 72 : newSize = Max(newSize, 5 * Max(BLCKSZ, XLOG_BLCKSZ));
177 :
178 : #ifndef FRONTEND
179 :
180 : /*
181 : * Note that in much unlucky circumstances, the random data read from a
182 : * recycled segment can cause this routine to be called with a size
183 : * causing a hard failure at allocation. For a standby, this would cause
184 : * the instance to stop suddenly with a hard failure, preventing it to
185 : * retry fetching WAL from one of its sources which could allow it to move
186 : * on with replay without a manual restart. If the data comes from a past
187 : * recycled segment and is still valid, then the allocation may succeed
188 : * but record checks are going to fail so this would be short-lived. If
189 : * the allocation fails because of a memory shortage, then this is not a
190 : * hard failure either per the guarantee given by MCXT_ALLOC_NO_OOM.
191 : */
192 : if (!AllocSizeIsValid(newSize))
193 : return false;
194 :
195 : #endif
196 :
197 72 : if (state->readRecordBuf)
198 0 : pfree(state->readRecordBuf);
199 72 : state->readRecordBuf =
200 72 : (char *) palloc_extended(newSize, MCXT_ALLOC_NO_OOM);
201 72 : if (state->readRecordBuf == NULL)
202 : {
203 0 : state->readRecordBufSize = 0;
204 0 : return false;
205 : }
206 72 : state->readRecordBufSize = newSize;
207 72 : return true;
208 : }
209 :
210 : /*
211 : * Initialize the passed segment structs.
212 : */
213 : void
214 72 : WALOpenSegmentInit(WALOpenSegment *seg, WALSegmentContext *segcxt,
215 : int segsize, const char *waldir)
216 : {
217 72 : seg->ws_file = -1;
218 72 : seg->ws_segno = 0;
219 72 : seg->ws_tli = 0;
220 :
221 72 : segcxt->ws_segsize = segsize;
222 72 : if (waldir)
223 72 : snprintf(segcxt->ws_dir, MAXPGPATH, "%s", waldir);
224 72 : }
225 :
226 : /*
227 : * Begin reading WAL at 'RecPtr'.
228 : *
229 : * 'RecPtr' should point to the beginnning of a valid WAL record. Pointing at
230 : * the beginning of a page is also OK, if there is a new record right after
231 : * the page header, i.e. not a continuation.
232 : *
233 : * This does not make any attempt to read the WAL yet, and hence cannot fail.
234 : * If the starting address is not correct, the first call to XLogReadRecord()
235 : * will error out.
236 : */
237 : void
238 144 : XLogBeginRead(XLogReaderState *state, XLogRecPtr RecPtr)
239 : {
240 : Assert(!XLogRecPtrIsInvalid(RecPtr));
241 :
242 144 : ResetDecoder(state);
243 :
244 : /* Begin at the passed-in record pointer. */
245 144 : state->EndRecPtr = RecPtr;
246 144 : state->ReadRecPtr = InvalidXLogRecPtr;
247 144 : }
248 :
249 : /*
250 : * Attempt to read an XLOG record.
251 : *
252 : * XLogBeginRead() or XLogFindNextRecord() must be called before the first call
253 : * to XLogReadRecord().
254 : *
255 : * If the page_read callback fails to read the requested data, NULL is
256 : * returned. The callback is expected to have reported the error; errormsg
257 : * is set to NULL.
258 : *
259 : * If the reading fails for some other reason, NULL is also returned, and
260 : * *errormsg is set to a string with details of the failure.
261 : *
262 : * The returned pointer (or *errormsg) points to an internal buffer that's
263 : * valid until the next call to XLogReadRecord.
264 : */
265 : XLogRecord *
266 360 : XLogReadRecord(XLogReaderState *state, char **errormsg)
267 : {
268 : XLogRecPtr RecPtr;
269 : XLogRecord *record;
270 : XLogRecPtr targetPagePtr;
271 : bool randAccess;
272 : uint32 len,
273 : total_len;
274 : uint32 targetRecOff;
275 : uint32 pageHeaderSize;
276 : bool gotheader;
277 : int readOff;
278 :
279 : /*
280 : * randAccess indicates whether to verify the previous-record pointer of
281 : * the record we're reading. We only do this if we're reading
282 : * sequentially, which is what we initially assume.
283 : */
284 360 : randAccess = false;
285 :
286 : /* reset error state */
287 360 : *errormsg = NULL;
288 360 : state->errormsg_buf[0] = '\0';
289 :
290 360 : ResetDecoder(state);
291 :
292 360 : RecPtr = state->EndRecPtr;
293 :
294 360 : if (state->ReadRecPtr != InvalidXLogRecPtr)
295 : {
296 : /* read the record after the one we just read */
297 :
298 : /*
299 : * EndRecPtr is pointing to end+1 of the previous WAL record. If
300 : * we're at a page boundary, no more records can fit on the current
301 : * page. We must skip over the page header, but we can't do that until
302 : * we've read in the page, since the header size is variable.
303 : */
304 : }
305 : else
306 : {
307 : /*
308 : * Caller supplied a position to start at.
309 : *
310 : * In this case, EndRecPtr should already be pointing to a valid
311 : * record starting position.
312 : */
313 : Assert(XRecOffIsValid(RecPtr));
314 144 : randAccess = true;
315 : }
316 :
317 360 : state->currRecPtr = RecPtr;
318 :
319 360 : targetPagePtr = RecPtr - (RecPtr % XLOG_BLCKSZ);
320 360 : targetRecOff = RecPtr % XLOG_BLCKSZ;
321 :
322 : /*
323 : * Read the page containing the record into state->readBuf. Request enough
324 : * byte to cover the whole record header, or at least the part of it that
325 : * fits on the same page.
326 : */
327 360 : readOff = ReadPageInternal(state, targetPagePtr,
328 360 : Min(targetRecOff + SizeOfXLogRecord, XLOG_BLCKSZ));
329 360 : if (readOff < 0)
330 72 : goto err;
331 :
332 : /*
333 : * ReadPageInternal always returns at least the page header, so we can
334 : * examine it now.
335 : */
336 288 : pageHeaderSize = XLogPageHeaderSize((XLogPageHeader) state->readBuf);
337 288 : if (targetRecOff == 0)
338 : {
339 : /*
340 : * At page start, so skip over page header.
341 : */
342 0 : RecPtr += pageHeaderSize;
343 0 : targetRecOff = pageHeaderSize;
344 : }
345 288 : else if (targetRecOff < pageHeaderSize)
346 : {
347 0 : report_invalid_record(state, "invalid record offset at %X/%X",
348 0 : (uint32) (RecPtr >> 32), (uint32) RecPtr);
349 0 : goto err;
350 : }
351 :
352 288 : if ((((XLogPageHeader) state->readBuf)->xlp_info & XLP_FIRST_IS_CONTRECORD) &&
353 : targetRecOff == pageHeaderSize)
354 : {
355 0 : report_invalid_record(state, "contrecord is requested by %X/%X",
356 0 : (uint32) (RecPtr >> 32), (uint32) RecPtr);
357 0 : goto err;
358 : }
359 :
360 : /* ReadPageInternal has verified the page header */
361 : Assert(pageHeaderSize <= readOff);
362 :
363 : /*
364 : * Read the record length.
365 : *
366 : * NB: Even though we use an XLogRecord pointer here, the whole record
367 : * header might not fit on this page. xl_tot_len is the first field of the
368 : * struct, so it must be on this page (the records are MAXALIGNed), but we
369 : * cannot access any other fields until we've verified that we got the
370 : * whole header.
371 : */
372 288 : record = (XLogRecord *) (state->readBuf + RecPtr % XLOG_BLCKSZ);
373 288 : total_len = record->xl_tot_len;
374 :
375 : /*
376 : * If the whole record header is on this page, validate it immediately.
377 : * Otherwise do just a basic sanity check on xl_tot_len, and validate the
378 : * rest of the header after reading it from the next page. The xl_tot_len
379 : * check is necessary here to ensure that we enter the "Need to reassemble
380 : * record" code path below; otherwise we might fail to apply
381 : * ValidXLogRecordHeader at all.
382 : */
383 288 : if (targetRecOff <= XLOG_BLCKSZ - SizeOfXLogRecord)
384 : {
385 288 : if (!ValidXLogRecordHeader(state, RecPtr, state->ReadRecPtr, record,
386 : randAccess))
387 0 : goto err;
388 288 : gotheader = true;
389 : }
390 : else
391 : {
392 : /* XXX: more validation should be done here */
393 0 : if (total_len < SizeOfXLogRecord)
394 : {
395 0 : report_invalid_record(state,
396 : "invalid record length at %X/%X: wanted %u, got %u",
397 0 : (uint32) (RecPtr >> 32), (uint32) RecPtr,
398 : (uint32) SizeOfXLogRecord, total_len);
399 0 : goto err;
400 : }
401 0 : gotheader = false;
402 : }
403 :
404 288 : len = XLOG_BLCKSZ - RecPtr % XLOG_BLCKSZ;
405 288 : if (total_len > len)
406 : {
407 : /* Need to reassemble record */
408 : char *contdata;
409 : XLogPageHeader pageHeader;
410 : char *buffer;
411 : uint32 gotlen;
412 :
413 : /*
414 : * Enlarge readRecordBuf as needed.
415 : */
416 0 : if (total_len > state->readRecordBufSize &&
417 0 : !allocate_recordbuf(state, total_len))
418 : {
419 : /* We treat this as a "bogus data" condition */
420 0 : report_invalid_record(state, "record length %u at %X/%X too long",
421 : total_len,
422 0 : (uint32) (RecPtr >> 32), (uint32) RecPtr);
423 0 : goto err;
424 : }
425 :
426 : /* Copy the first fragment of the record from the first page. */
427 0 : memcpy(state->readRecordBuf,
428 0 : state->readBuf + RecPtr % XLOG_BLCKSZ, len);
429 0 : buffer = state->readRecordBuf + len;
430 0 : gotlen = len;
431 :
432 : do
433 : {
434 : /* Calculate pointer to beginning of next page */
435 0 : targetPagePtr += XLOG_BLCKSZ;
436 :
437 : /* Wait for the next page to become available */
438 0 : readOff = ReadPageInternal(state, targetPagePtr,
439 0 : Min(total_len - gotlen + SizeOfXLogShortPHD,
440 : XLOG_BLCKSZ));
441 :
442 0 : if (readOff < 0)
443 0 : goto err;
444 :
445 : Assert(SizeOfXLogShortPHD <= readOff);
446 :
447 : /* Check that the continuation on next page looks valid */
448 0 : pageHeader = (XLogPageHeader) state->readBuf;
449 0 : if (!(pageHeader->xlp_info & XLP_FIRST_IS_CONTRECORD))
450 : {
451 0 : report_invalid_record(state,
452 : "there is no contrecord flag at %X/%X",
453 0 : (uint32) (RecPtr >> 32), (uint32) RecPtr);
454 0 : goto err;
455 : }
456 :
457 : /*
458 : * Cross-check that xlp_rem_len agrees with how much of the record
459 : * we expect there to be left.
460 : */
461 0 : if (pageHeader->xlp_rem_len == 0 ||
462 0 : total_len != (pageHeader->xlp_rem_len + gotlen))
463 : {
464 0 : report_invalid_record(state,
465 : "invalid contrecord length %u at %X/%X",
466 : pageHeader->xlp_rem_len,
467 0 : (uint32) (RecPtr >> 32), (uint32) RecPtr);
468 0 : goto err;
469 : }
470 :
471 : /* Append the continuation from this page to the buffer */
472 0 : pageHeaderSize = XLogPageHeaderSize(pageHeader);
473 :
474 0 : if (readOff < pageHeaderSize)
475 0 : readOff = ReadPageInternal(state, targetPagePtr,
476 : pageHeaderSize);
477 :
478 : Assert(pageHeaderSize <= readOff);
479 :
480 0 : contdata = (char *) state->readBuf + pageHeaderSize;
481 0 : len = XLOG_BLCKSZ - pageHeaderSize;
482 0 : if (pageHeader->xlp_rem_len < len)
483 0 : len = pageHeader->xlp_rem_len;
484 :
485 0 : if (readOff < pageHeaderSize + len)
486 0 : readOff = ReadPageInternal(state, targetPagePtr,
487 0 : pageHeaderSize + len);
488 :
489 0 : memcpy(buffer, (char *) contdata, len);
490 0 : buffer += len;
491 0 : gotlen += len;
492 :
493 : /* If we just reassembled the record header, validate it. */
494 0 : if (!gotheader)
495 : {
496 0 : record = (XLogRecord *) state->readRecordBuf;
497 0 : if (!ValidXLogRecordHeader(state, RecPtr, state->ReadRecPtr,
498 : record, randAccess))
499 0 : goto err;
500 0 : gotheader = true;
501 : }
502 0 : } while (gotlen < total_len);
503 :
504 : Assert(gotheader);
505 :
506 0 : record = (XLogRecord *) state->readRecordBuf;
507 0 : if (!ValidXLogRecord(state, record, RecPtr))
508 0 : goto err;
509 :
510 0 : pageHeaderSize = XLogPageHeaderSize((XLogPageHeader) state->readBuf);
511 0 : state->ReadRecPtr = RecPtr;
512 0 : state->EndRecPtr = targetPagePtr + pageHeaderSize
513 0 : + MAXALIGN(pageHeader->xlp_rem_len);
514 : }
515 : else
516 : {
517 : /* Wait for the record data to become available */
518 288 : readOff = ReadPageInternal(state, targetPagePtr,
519 288 : Min(targetRecOff + total_len, XLOG_BLCKSZ));
520 288 : if (readOff < 0)
521 0 : goto err;
522 :
523 : /* Record does not cross a page boundary */
524 288 : if (!ValidXLogRecord(state, record, RecPtr))
525 0 : goto err;
526 :
527 288 : state->EndRecPtr = RecPtr + MAXALIGN(total_len);
528 :
529 288 : state->ReadRecPtr = RecPtr;
530 : }
531 :
532 : /*
533 : * Special processing if it's an XLOG SWITCH record
534 : */
535 288 : if (record->xl_rmid == RM_XLOG_ID &&
536 144 : (record->xl_info & ~XLR_INFO_MASK) == XLOG_SWITCH)
537 : {
538 : /* Pretend it extends to end of segment */
539 0 : state->EndRecPtr += state->segcxt.ws_segsize - 1;
540 0 : state->EndRecPtr -= XLogSegmentOffset(state->EndRecPtr, state->segcxt.ws_segsize);
541 : }
542 :
543 288 : if (DecodeXLogRecord(state, record, errormsg))
544 288 : return record;
545 : else
546 0 : return NULL;
547 :
548 72 : err:
549 :
550 : /*
551 : * Invalidate the read state. We might read from a different source after
552 : * failure.
553 : */
554 72 : XLogReaderInvalReadState(state);
555 :
556 72 : if (state->errormsg_buf[0] != '\0')
557 0 : *errormsg = state->errormsg_buf;
558 :
559 72 : return NULL;
560 : }
561 :
562 : /*
563 : * Read a single xlog page including at least [pageptr, reqLen] of valid data
564 : * via the page_read() callback.
565 : *
566 : * Returns -1 if the required page cannot be read for some reason; errormsg_buf
567 : * is set in that case (unless the error occurs in the page_read callback).
568 : *
569 : * We fetch the page from a reader-local cache if we know we have the required
570 : * data and if there hasn't been any error since caching the data.
571 : */
572 : static int
573 792 : ReadPageInternal(XLogReaderState *state, XLogRecPtr pageptr, int reqLen)
574 : {
575 : int readLen;
576 : uint32 targetPageOff;
577 : XLogSegNo targetSegNo;
578 : XLogPageHeader hdr;
579 :
580 : Assert((pageptr % XLOG_BLCKSZ) == 0);
581 :
582 792 : XLByteToSeg(pageptr, targetSegNo, state->segcxt.ws_segsize);
583 792 : targetPageOff = XLogSegmentOffset(pageptr, state->segcxt.ws_segsize);
584 :
585 : /* check whether we have all the requested data already */
586 792 : if (targetSegNo == state->seg.ws_segno &&
587 720 : targetPageOff == state->segoff && reqLen <= state->readLen)
588 648 : return state->readLen;
589 :
590 : /*
591 : * Data is not in our buffer.
592 : *
593 : * Every time we actually read the segment, even if we looked at parts of
594 : * it before, we need to do verification as the page_read callback might
595 : * now be rereading data from a different source.
596 : *
597 : * Whenever switching to a new WAL segment, we read the first page of the
598 : * file and validate its header, even if that's not where the target
599 : * record is. This is so that we can check the additional identification
600 : * info that is present in the first page's "long" header.
601 : */
602 144 : if (targetSegNo != state->seg.ws_segno && targetPageOff != 0)
603 : {
604 0 : XLogRecPtr targetSegmentPtr = pageptr - targetPageOff;
605 :
606 0 : readLen = state->routine.page_read(state, targetSegmentPtr, XLOG_BLCKSZ,
607 : state->currRecPtr,
608 : state->readBuf);
609 0 : if (readLen < 0)
610 0 : goto err;
611 :
612 : /* we can be sure to have enough WAL available, we scrolled back */
613 : Assert(readLen == XLOG_BLCKSZ);
614 :
615 0 : if (!XLogReaderValidatePageHeader(state, targetSegmentPtr,
616 : state->readBuf))
617 0 : goto err;
618 : }
619 :
620 : /*
621 : * First, read the requested data length, but at least a short page header
622 : * so that we can validate it.
623 : */
624 144 : readLen = state->routine.page_read(state, pageptr, Max(reqLen, SizeOfXLogShortPHD),
625 : state->currRecPtr,
626 : state->readBuf);
627 144 : if (readLen < 0)
628 72 : goto err;
629 :
630 : Assert(readLen <= XLOG_BLCKSZ);
631 :
632 : /* Do we have enough data to check the header length? */
633 72 : if (readLen <= SizeOfXLogShortPHD)
634 0 : goto err;
635 :
636 : Assert(readLen >= reqLen);
637 :
638 72 : hdr = (XLogPageHeader) state->readBuf;
639 :
640 : /* still not enough */
641 72 : if (readLen < XLogPageHeaderSize(hdr))
642 : {
643 0 : readLen = state->routine.page_read(state, pageptr, XLogPageHeaderSize(hdr),
644 : state->currRecPtr,
645 : state->readBuf);
646 0 : if (readLen < 0)
647 0 : goto err;
648 : }
649 :
650 : /*
651 : * Now that we know we have the full header, validate it.
652 : */
653 72 : if (!XLogReaderValidatePageHeader(state, pageptr, (char *) hdr))
654 0 : goto err;
655 :
656 : /* update read state information */
657 72 : state->seg.ws_segno = targetSegNo;
658 72 : state->segoff = targetPageOff;
659 72 : state->readLen = readLen;
660 :
661 72 : return readLen;
662 :
663 72 : err:
664 72 : XLogReaderInvalReadState(state);
665 72 : return -1;
666 : }
667 :
668 : /*
669 : * Invalidate the xlogreader's read state to force a re-read.
670 : */
671 : static void
672 144 : XLogReaderInvalReadState(XLogReaderState *state)
673 : {
674 144 : state->seg.ws_segno = 0;
675 144 : state->segoff = 0;
676 144 : state->readLen = 0;
677 144 : }
678 :
679 : /*
680 : * Validate an XLOG record header.
681 : *
682 : * This is just a convenience subroutine to avoid duplicated code in
683 : * XLogReadRecord. It's not intended for use from anywhere else.
684 : */
685 : static bool
686 288 : ValidXLogRecordHeader(XLogReaderState *state, XLogRecPtr RecPtr,
687 : XLogRecPtr PrevRecPtr, XLogRecord *record,
688 : bool randAccess)
689 : {
690 288 : if (record->xl_tot_len < SizeOfXLogRecord)
691 : {
692 0 : report_invalid_record(state,
693 : "invalid record length at %X/%X: wanted %u, got %u",
694 0 : (uint32) (RecPtr >> 32), (uint32) RecPtr,
695 : (uint32) SizeOfXLogRecord, record->xl_tot_len);
696 0 : return false;
697 : }
698 288 : if (record->xl_rmid > RM_MAX_ID)
699 : {
700 0 : report_invalid_record(state,
701 : "invalid resource manager ID %u at %X/%X",
702 0 : record->xl_rmid, (uint32) (RecPtr >> 32),
703 : (uint32) RecPtr);
704 0 : return false;
705 : }
706 288 : if (randAccess)
707 : {
708 : /*
709 : * We can't exactly verify the prev-link, but surely it should be less
710 : * than the record's own address.
711 : */
712 144 : if (!(record->xl_prev < RecPtr))
713 : {
714 0 : report_invalid_record(state,
715 : "record with incorrect prev-link %X/%X at %X/%X",
716 0 : (uint32) (record->xl_prev >> 32),
717 0 : (uint32) record->xl_prev,
718 0 : (uint32) (RecPtr >> 32), (uint32) RecPtr);
719 0 : return false;
720 : }
721 : }
722 : else
723 : {
724 : /*
725 : * Record's prev-link should exactly match our previous location. This
726 : * check guards against torn WAL pages where a stale but valid-looking
727 : * WAL record starts on a sector boundary.
728 : */
729 144 : if (record->xl_prev != PrevRecPtr)
730 : {
731 0 : report_invalid_record(state,
732 : "record with incorrect prev-link %X/%X at %X/%X",
733 0 : (uint32) (record->xl_prev >> 32),
734 0 : (uint32) record->xl_prev,
735 0 : (uint32) (RecPtr >> 32), (uint32) RecPtr);
736 0 : return false;
737 : }
738 : }
739 :
740 288 : return true;
741 : }
742 :
743 :
744 : /*
745 : * CRC-check an XLOG record. We do not believe the contents of an XLOG
746 : * record (other than to the minimal extent of computing the amount of
747 : * data to read in) until we've checked the CRCs.
748 : *
749 : * We assume all of the record (that is, xl_tot_len bytes) has been read
750 : * into memory at *record. Also, ValidXLogRecordHeader() has accepted the
751 : * record's header, which means in particular that xl_tot_len is at least
752 : * SizeOfXLogRecord.
753 : */
754 : static bool
755 288 : ValidXLogRecord(XLogReaderState *state, XLogRecord *record, XLogRecPtr recptr)
756 : {
757 : pg_crc32c crc;
758 :
759 : /* Calculate the CRC */
760 288 : INIT_CRC32C(crc);
761 288 : COMP_CRC32C(crc, ((char *) record) + SizeOfXLogRecord, record->xl_tot_len - SizeOfXLogRecord);
762 : /* include the record header last */
763 288 : COMP_CRC32C(crc, (char *) record, offsetof(XLogRecord, xl_crc));
764 288 : FIN_CRC32C(crc);
765 :
766 288 : if (!EQ_CRC32C(record->xl_crc, crc))
767 : {
768 0 : report_invalid_record(state,
769 : "incorrect resource manager data checksum in record at %X/%X",
770 0 : (uint32) (recptr >> 32), (uint32) recptr);
771 0 : return false;
772 : }
773 :
774 288 : return true;
775 : }
776 :
777 : /*
778 : * Validate a page header.
779 : *
780 : * Check if 'phdr' is valid as the header of the XLog page at position
781 : * 'recptr'.
782 : */
783 : bool
784 72 : XLogReaderValidatePageHeader(XLogReaderState *state, XLogRecPtr recptr,
785 : char *phdr)
786 : {
787 : XLogRecPtr recaddr;
788 : XLogSegNo segno;
789 : int32 offset;
790 72 : XLogPageHeader hdr = (XLogPageHeader) phdr;
791 :
792 : Assert((recptr % XLOG_BLCKSZ) == 0);
793 :
794 72 : XLByteToSeg(recptr, segno, state->segcxt.ws_segsize);
795 72 : offset = XLogSegmentOffset(recptr, state->segcxt.ws_segsize);
796 :
797 72 : XLogSegNoOffsetToRecPtr(segno, offset, state->segcxt.ws_segsize, recaddr);
798 :
799 72 : if (hdr->xlp_magic != XLOG_PAGE_MAGIC)
800 : {
801 : char fname[MAXFNAMELEN];
802 :
803 0 : XLogFileName(fname, state->seg.ws_tli, segno, state->segcxt.ws_segsize);
804 :
805 0 : report_invalid_record(state,
806 : "invalid magic number %04X in log segment %s, offset %u",
807 0 : hdr->xlp_magic,
808 : fname,
809 : offset);
810 0 : return false;
811 : }
812 :
813 72 : if ((hdr->xlp_info & ~XLP_ALL_FLAGS) != 0)
814 : {
815 : char fname[MAXFNAMELEN];
816 :
817 0 : XLogFileName(fname, state->seg.ws_tli, segno, state->segcxt.ws_segsize);
818 :
819 0 : report_invalid_record(state,
820 : "invalid info bits %04X in log segment %s, offset %u",
821 0 : hdr->xlp_info,
822 : fname,
823 : offset);
824 0 : return false;
825 : }
826 :
827 72 : if (hdr->xlp_info & XLP_LONG_HEADER)
828 : {
829 72 : XLogLongPageHeader longhdr = (XLogLongPageHeader) hdr;
830 :
831 72 : if (state->system_identifier &&
832 0 : longhdr->xlp_sysid != state->system_identifier)
833 : {
834 0 : report_invalid_record(state,
835 : "WAL file is from different database system: WAL file database system identifier is %llu, pg_control database system identifier is %llu",
836 0 : (unsigned long long) longhdr->xlp_sysid,
837 0 : (unsigned long long) state->system_identifier);
838 0 : return false;
839 : }
840 72 : else if (longhdr->xlp_seg_size != state->segcxt.ws_segsize)
841 : {
842 0 : report_invalid_record(state,
843 : "WAL file is from different database system: incorrect segment size in page header");
844 0 : return false;
845 : }
846 72 : else if (longhdr->xlp_xlog_blcksz != XLOG_BLCKSZ)
847 : {
848 0 : report_invalid_record(state,
849 : "WAL file is from different database system: incorrect XLOG_BLCKSZ in page header");
850 0 : return false;
851 : }
852 : }
853 0 : else if (offset == 0)
854 : {
855 : char fname[MAXFNAMELEN];
856 :
857 0 : XLogFileName(fname, state->seg.ws_tli, segno, state->segcxt.ws_segsize);
858 :
859 : /* hmm, first page of file doesn't have a long header? */
860 0 : report_invalid_record(state,
861 : "invalid info bits %04X in log segment %s, offset %u",
862 0 : hdr->xlp_info,
863 : fname,
864 : offset);
865 0 : return false;
866 : }
867 :
868 : /*
869 : * Check that the address on the page agrees with what we expected. This
870 : * check typically fails when an old WAL segment is recycled, and hasn't
871 : * yet been overwritten with new data yet.
872 : */
873 72 : if (hdr->xlp_pageaddr != recaddr)
874 : {
875 : char fname[MAXFNAMELEN];
876 :
877 0 : XLogFileName(fname, state->seg.ws_tli, segno, state->segcxt.ws_segsize);
878 :
879 0 : report_invalid_record(state,
880 : "unexpected pageaddr %X/%X in log segment %s, offset %u",
881 0 : (uint32) (hdr->xlp_pageaddr >> 32), (uint32) hdr->xlp_pageaddr,
882 : fname,
883 : offset);
884 0 : return false;
885 : }
886 :
887 : /*
888 : * Since child timelines are always assigned a TLI greater than their
889 : * immediate parent's TLI, we should never see TLI go backwards across
890 : * successive pages of a consistent WAL sequence.
891 : *
892 : * Sometimes we re-read a segment that's already been (partially) read. So
893 : * we only verify TLIs for pages that are later than the last remembered
894 : * LSN.
895 : */
896 72 : if (recptr > state->latestPagePtr)
897 : {
898 72 : if (hdr->xlp_tli < state->latestPageTLI)
899 : {
900 : char fname[MAXFNAMELEN];
901 :
902 0 : XLogFileName(fname, state->seg.ws_tli, segno, state->segcxt.ws_segsize);
903 :
904 0 : report_invalid_record(state,
905 : "out-of-sequence timeline ID %u (after %u) in log segment %s, offset %u",
906 : hdr->xlp_tli,
907 : state->latestPageTLI,
908 : fname,
909 : offset);
910 0 : return false;
911 : }
912 : }
913 72 : state->latestPagePtr = recptr;
914 72 : state->latestPageTLI = hdr->xlp_tli;
915 :
916 72 : return true;
917 : }
918 :
919 : #ifdef FRONTEND
920 : /*
921 : * Functions that are currently not needed in the backend, but are better
922 : * implemented inside xlogreader.c because of the internal facilities available
923 : * here.
924 : */
925 :
926 : /*
927 : * Find the first record with an lsn >= RecPtr.
928 : *
929 : * This is different from XLogBeginRead() in that RecPtr doesn't need to point
930 : * to a valid record boundary. Useful for checking whether RecPtr is a valid
931 : * xlog address for reading, and to find the first valid address after some
932 : * address when dumping records for debugging purposes.
933 : *
934 : * This positions the reader, like XLogBeginRead(), so that the next call to
935 : * XLogReadRecord() will read the next valid record.
936 : */
937 : XLogRecPtr
938 72 : XLogFindNextRecord(XLogReaderState *state, XLogRecPtr RecPtr)
939 : {
940 : XLogRecPtr tmpRecPtr;
941 72 : XLogRecPtr found = InvalidXLogRecPtr;
942 : XLogPageHeader header;
943 : char *errormsg;
944 :
945 : Assert(!XLogRecPtrIsInvalid(RecPtr));
946 :
947 : /*
948 : * skip over potential continuation data, keeping in mind that it may span
949 : * multiple pages
950 : */
951 72 : tmpRecPtr = RecPtr;
952 : while (true)
953 0 : {
954 : XLogRecPtr targetPagePtr;
955 : int targetRecOff;
956 : uint32 pageHeaderSize;
957 : int readLen;
958 :
959 : /*
960 : * Compute targetRecOff. It should typically be equal or greater than
961 : * short page-header since a valid record can't start anywhere before
962 : * that, except when caller has explicitly specified the offset that
963 : * falls somewhere there or when we are skipping multi-page
964 : * continuation record. It doesn't matter though because
965 : * ReadPageInternal() is prepared to handle that and will read at
966 : * least short page-header worth of data
967 : */
968 72 : targetRecOff = tmpRecPtr % XLOG_BLCKSZ;
969 :
970 : /* scroll back to page boundary */
971 72 : targetPagePtr = tmpRecPtr - targetRecOff;
972 :
973 : /* Read the page containing the record */
974 72 : readLen = ReadPageInternal(state, targetPagePtr, targetRecOff);
975 72 : if (readLen < 0)
976 0 : goto err;
977 :
978 72 : header = (XLogPageHeader) state->readBuf;
979 :
980 72 : pageHeaderSize = XLogPageHeaderSize(header);
981 :
982 : /* make sure we have enough data for the page header */
983 72 : readLen = ReadPageInternal(state, targetPagePtr, pageHeaderSize);
984 72 : if (readLen < 0)
985 0 : goto err;
986 :
987 : /* skip over potential continuation data */
988 72 : if (header->xlp_info & XLP_FIRST_IS_CONTRECORD)
989 : {
990 : /*
991 : * If the length of the remaining continuation data is more than
992 : * what can fit in this page, the continuation record crosses over
993 : * this page. Read the next page and try again. xlp_rem_len in the
994 : * next page header will contain the remaining length of the
995 : * continuation data
996 : *
997 : * Note that record headers are MAXALIGN'ed
998 : */
999 0 : if (MAXALIGN(header->xlp_rem_len) >= (XLOG_BLCKSZ - pageHeaderSize))
1000 0 : tmpRecPtr = targetPagePtr + XLOG_BLCKSZ;
1001 : else
1002 : {
1003 : /*
1004 : * The previous continuation record ends in this page. Set
1005 : * tmpRecPtr to point to the first valid record
1006 : */
1007 0 : tmpRecPtr = targetPagePtr + pageHeaderSize
1008 0 : + MAXALIGN(header->xlp_rem_len);
1009 0 : break;
1010 : }
1011 : }
1012 : else
1013 : {
1014 72 : tmpRecPtr = targetPagePtr + pageHeaderSize;
1015 72 : break;
1016 : }
1017 : }
1018 :
1019 : /*
1020 : * we know now that tmpRecPtr is an address pointing to a valid XLogRecord
1021 : * because either we're at the first record after the beginning of a page
1022 : * or we just jumped over the remaining data of a continuation.
1023 : */
1024 72 : XLogBeginRead(state, tmpRecPtr);
1025 72 : while (XLogReadRecord(state, &errormsg) != NULL)
1026 : {
1027 : /* past the record we've found, break out */
1028 72 : if (RecPtr <= state->ReadRecPtr)
1029 : {
1030 : /* Rewind the reader to the beginning of the last record. */
1031 72 : found = state->ReadRecPtr;
1032 72 : XLogBeginRead(state, found);
1033 72 : return found;
1034 : }
1035 : }
1036 :
1037 0 : err:
1038 0 : XLogReaderInvalReadState(state);
1039 :
1040 0 : return InvalidXLogRecPtr;
1041 : }
1042 :
1043 : #endif /* FRONTEND */
1044 :
1045 : /*
1046 : * Helper function to ease writing of XLogRoutine->page_read callbacks.
1047 : * If this function is used, caller must supply a segment_open callback in
1048 : * 'state', as that is used here.
1049 : *
1050 : * Read 'count' bytes into 'buf', starting at location 'startptr', from WAL
1051 : * fetched from timeline 'tli'.
1052 : *
1053 : * Returns true if succeeded, false if an error occurs, in which case
1054 : * 'errinfo' receives error details.
1055 : *
1056 : * XXX probably this should be improved to suck data directly from the
1057 : * WAL buffers when possible.
1058 : */
1059 : bool
1060 72 : WALRead(XLogReaderState *state,
1061 : char *buf, XLogRecPtr startptr, Size count, TimeLineID tli,
1062 : WALReadError *errinfo)
1063 : {
1064 : char *p;
1065 : XLogRecPtr recptr;
1066 : Size nbytes;
1067 :
1068 72 : p = buf;
1069 72 : recptr = startptr;
1070 72 : nbytes = count;
1071 :
1072 144 : while (nbytes > 0)
1073 : {
1074 : uint32 startoff;
1075 : int segbytes;
1076 : int readbytes;
1077 :
1078 72 : startoff = XLogSegmentOffset(recptr, state->segcxt.ws_segsize);
1079 :
1080 : /*
1081 : * If the data we want is not in a segment we have open, close what we
1082 : * have (if anything) and open the next one, using the caller's
1083 : * provided openSegment callback.
1084 : */
1085 72 : if (state->seg.ws_file < 0 ||
1086 0 : !XLByteInSeg(recptr, state->seg.ws_segno, state->segcxt.ws_segsize) ||
1087 0 : tli != state->seg.ws_tli)
1088 : {
1089 : XLogSegNo nextSegNo;
1090 :
1091 72 : if (state->seg.ws_file >= 0)
1092 0 : state->routine.segment_close(state);
1093 :
1094 72 : XLByteToSeg(recptr, nextSegNo, state->segcxt.ws_segsize);
1095 72 : state->routine.segment_open(state, nextSegNo, &tli);
1096 :
1097 : /* This shouldn't happen -- indicates a bug in segment_open */
1098 : Assert(state->seg.ws_file >= 0);
1099 :
1100 : /* Update the current segment info. */
1101 72 : state->seg.ws_tli = tli;
1102 72 : state->seg.ws_segno = nextSegNo;
1103 : }
1104 :
1105 : /* How many bytes are within this segment? */
1106 72 : if (nbytes > (state->segcxt.ws_segsize - startoff))
1107 0 : segbytes = state->segcxt.ws_segsize - startoff;
1108 : else
1109 72 : segbytes = nbytes;
1110 :
1111 : #ifndef FRONTEND
1112 : pgstat_report_wait_start(WAIT_EVENT_WAL_READ);
1113 : #endif
1114 :
1115 : /* Reset errno first; eases reporting non-errno-affecting errors */
1116 72 : errno = 0;
1117 72 : readbytes = pg_pread(state->seg.ws_file, p, segbytes, (off_t) startoff);
1118 :
1119 : #ifndef FRONTEND
1120 : pgstat_report_wait_end();
1121 : #endif
1122 :
1123 72 : if (readbytes <= 0)
1124 : {
1125 0 : errinfo->wre_errno = errno;
1126 0 : errinfo->wre_req = segbytes;
1127 0 : errinfo->wre_read = readbytes;
1128 0 : errinfo->wre_off = startoff;
1129 0 : errinfo->wre_seg = state->seg;
1130 0 : return false;
1131 : }
1132 :
1133 : /* Update state for read */
1134 72 : recptr += readbytes;
1135 72 : nbytes -= readbytes;
1136 72 : p += readbytes;
1137 : }
1138 :
1139 72 : return true;
1140 : }
1141 :
1142 : /* ----------------------------------------
1143 : * Functions for decoding the data and block references in a record.
1144 : * ----------------------------------------
1145 : */
1146 :
1147 : /* private function to reset the state between records */
1148 : static void
1149 792 : ResetDecoder(XLogReaderState *state)
1150 : {
1151 : int block_id;
1152 :
1153 792 : state->decoded_record = NULL;
1154 :
1155 792 : state->main_data_len = 0;
1156 :
1157 792 : for (block_id = 0; block_id <= state->max_block_id; block_id++)
1158 : {
1159 0 : state->blocks[block_id].in_use = false;
1160 0 : state->blocks[block_id].has_image = false;
1161 0 : state->blocks[block_id].has_data = false;
1162 0 : state->blocks[block_id].apply_image = false;
1163 : }
1164 792 : state->max_block_id = -1;
1165 792 : }
1166 :
1167 : /*
1168 : * Decode the previously read record.
1169 : *
1170 : * On error, a human-readable error message is returned in *errormsg, and
1171 : * the return value is false.
1172 : */
1173 : bool
1174 288 : DecodeXLogRecord(XLogReaderState *state, XLogRecord *record, char **errormsg)
1175 : {
1176 : /*
1177 : * read next _size bytes from record buffer, but check for overrun first.
1178 : */
1179 : #define COPY_HEADER_FIELD(_dst, _size) \
1180 : do { \
1181 : if (remaining < _size) \
1182 : goto shortdata_err; \
1183 : memcpy(_dst, ptr, _size); \
1184 : ptr += _size; \
1185 : remaining -= _size; \
1186 : } while(0)
1187 :
1188 : char *ptr;
1189 : uint32 remaining;
1190 : uint32 datatotal;
1191 288 : RelFileNode *rnode = NULL;
1192 : uint8 block_id;
1193 :
1194 288 : ResetDecoder(state);
1195 :
1196 288 : state->decoded_record = record;
1197 288 : state->record_origin = InvalidRepOriginId;
1198 :
1199 288 : ptr = (char *) record;
1200 288 : ptr += SizeOfXLogRecord;
1201 288 : remaining = record->xl_tot_len - SizeOfXLogRecord;
1202 :
1203 : /* Decode the headers */
1204 288 : datatotal = 0;
1205 288 : while (remaining > datatotal)
1206 : {
1207 288 : COPY_HEADER_FIELD(&block_id, sizeof(uint8));
1208 :
1209 288 : if (block_id == XLR_BLOCK_ID_DATA_SHORT)
1210 : {
1211 : /* XLogRecordDataHeaderShort */
1212 : uint8 main_data_len;
1213 :
1214 288 : COPY_HEADER_FIELD(&main_data_len, sizeof(uint8));
1215 :
1216 288 : state->main_data_len = main_data_len;
1217 288 : datatotal += main_data_len;
1218 288 : break; /* by convention, the main data fragment is
1219 : * always last */
1220 : }
1221 0 : else if (block_id == XLR_BLOCK_ID_DATA_LONG)
1222 : {
1223 : /* XLogRecordDataHeaderLong */
1224 : uint32 main_data_len;
1225 :
1226 0 : COPY_HEADER_FIELD(&main_data_len, sizeof(uint32));
1227 0 : state->main_data_len = main_data_len;
1228 0 : datatotal += main_data_len;
1229 0 : break; /* by convention, the main data fragment is
1230 : * always last */
1231 : }
1232 0 : else if (block_id == XLR_BLOCK_ID_ORIGIN)
1233 : {
1234 0 : COPY_HEADER_FIELD(&state->record_origin, sizeof(RepOriginId));
1235 : }
1236 0 : else if (block_id <= XLR_MAX_BLOCK_ID)
1237 : {
1238 : /* XLogRecordBlockHeader */
1239 : DecodedBkpBlock *blk;
1240 : uint8 fork_flags;
1241 :
1242 0 : if (block_id <= state->max_block_id)
1243 : {
1244 0 : report_invalid_record(state,
1245 : "out-of-order block_id %u at %X/%X",
1246 : block_id,
1247 0 : (uint32) (state->ReadRecPtr >> 32),
1248 0 : (uint32) state->ReadRecPtr);
1249 0 : goto err;
1250 : }
1251 0 : state->max_block_id = block_id;
1252 :
1253 0 : blk = &state->blocks[block_id];
1254 0 : blk->in_use = true;
1255 0 : blk->apply_image = false;
1256 :
1257 0 : COPY_HEADER_FIELD(&fork_flags, sizeof(uint8));
1258 0 : blk->forknum = fork_flags & BKPBLOCK_FORK_MASK;
1259 0 : blk->flags = fork_flags;
1260 0 : blk->has_image = ((fork_flags & BKPBLOCK_HAS_IMAGE) != 0);
1261 0 : blk->has_data = ((fork_flags & BKPBLOCK_HAS_DATA) != 0);
1262 :
1263 0 : COPY_HEADER_FIELD(&blk->data_len, sizeof(uint16));
1264 : /* cross-check that the HAS_DATA flag is set iff data_length > 0 */
1265 0 : if (blk->has_data && blk->data_len == 0)
1266 : {
1267 0 : report_invalid_record(state,
1268 : "BKPBLOCK_HAS_DATA set, but no data included at %X/%X",
1269 0 : (uint32) (state->ReadRecPtr >> 32), (uint32) state->ReadRecPtr);
1270 0 : goto err;
1271 : }
1272 0 : if (!blk->has_data && blk->data_len != 0)
1273 : {
1274 0 : report_invalid_record(state,
1275 : "BKPBLOCK_HAS_DATA not set, but data length is %u at %X/%X",
1276 0 : (unsigned int) blk->data_len,
1277 0 : (uint32) (state->ReadRecPtr >> 32), (uint32) state->ReadRecPtr);
1278 0 : goto err;
1279 : }
1280 0 : datatotal += blk->data_len;
1281 :
1282 0 : if (blk->has_image)
1283 : {
1284 0 : COPY_HEADER_FIELD(&blk->bimg_len, sizeof(uint16));
1285 0 : COPY_HEADER_FIELD(&blk->hole_offset, sizeof(uint16));
1286 0 : COPY_HEADER_FIELD(&blk->bimg_info, sizeof(uint8));
1287 :
1288 0 : blk->apply_image = ((blk->bimg_info & BKPIMAGE_APPLY) != 0);
1289 :
1290 0 : if (blk->bimg_info & BKPIMAGE_IS_COMPRESSED)
1291 : {
1292 0 : if (blk->bimg_info & BKPIMAGE_HAS_HOLE)
1293 0 : COPY_HEADER_FIELD(&blk->hole_length, sizeof(uint16));
1294 : else
1295 0 : blk->hole_length = 0;
1296 : }
1297 : else
1298 0 : blk->hole_length = BLCKSZ - blk->bimg_len;
1299 0 : datatotal += blk->bimg_len;
1300 :
1301 : /*
1302 : * cross-check that hole_offset > 0, hole_length > 0 and
1303 : * bimg_len < BLCKSZ if the HAS_HOLE flag is set.
1304 : */
1305 0 : if ((blk->bimg_info & BKPIMAGE_HAS_HOLE) &&
1306 0 : (blk->hole_offset == 0 ||
1307 0 : blk->hole_length == 0 ||
1308 0 : blk->bimg_len == BLCKSZ))
1309 : {
1310 0 : report_invalid_record(state,
1311 : "BKPIMAGE_HAS_HOLE set, but hole offset %u length %u block image length %u at %X/%X",
1312 0 : (unsigned int) blk->hole_offset,
1313 0 : (unsigned int) blk->hole_length,
1314 0 : (unsigned int) blk->bimg_len,
1315 0 : (uint32) (state->ReadRecPtr >> 32), (uint32) state->ReadRecPtr);
1316 0 : goto err;
1317 : }
1318 :
1319 : /*
1320 : * cross-check that hole_offset == 0 and hole_length == 0 if
1321 : * the HAS_HOLE flag is not set.
1322 : */
1323 0 : if (!(blk->bimg_info & BKPIMAGE_HAS_HOLE) &&
1324 0 : (blk->hole_offset != 0 || blk->hole_length != 0))
1325 : {
1326 0 : report_invalid_record(state,
1327 : "BKPIMAGE_HAS_HOLE not set, but hole offset %u length %u at %X/%X",
1328 0 : (unsigned int) blk->hole_offset,
1329 0 : (unsigned int) blk->hole_length,
1330 0 : (uint32) (state->ReadRecPtr >> 32), (uint32) state->ReadRecPtr);
1331 0 : goto err;
1332 : }
1333 :
1334 : /*
1335 : * cross-check that bimg_len < BLCKSZ if the IS_COMPRESSED
1336 : * flag is set.
1337 : */
1338 0 : if ((blk->bimg_info & BKPIMAGE_IS_COMPRESSED) &&
1339 0 : blk->bimg_len == BLCKSZ)
1340 : {
1341 0 : report_invalid_record(state,
1342 : "BKPIMAGE_IS_COMPRESSED set, but block image length %u at %X/%X",
1343 0 : (unsigned int) blk->bimg_len,
1344 0 : (uint32) (state->ReadRecPtr >> 32), (uint32) state->ReadRecPtr);
1345 0 : goto err;
1346 : }
1347 :
1348 : /*
1349 : * cross-check that bimg_len = BLCKSZ if neither HAS_HOLE nor
1350 : * IS_COMPRESSED flag is set.
1351 : */
1352 0 : if (!(blk->bimg_info & BKPIMAGE_HAS_HOLE) &&
1353 0 : !(blk->bimg_info & BKPIMAGE_IS_COMPRESSED) &&
1354 0 : blk->bimg_len != BLCKSZ)
1355 : {
1356 0 : report_invalid_record(state,
1357 : "neither BKPIMAGE_HAS_HOLE nor BKPIMAGE_IS_COMPRESSED set, but block image length is %u at %X/%X",
1358 0 : (unsigned int) blk->data_len,
1359 0 : (uint32) (state->ReadRecPtr >> 32), (uint32) state->ReadRecPtr);
1360 0 : goto err;
1361 : }
1362 : }
1363 0 : if (!(fork_flags & BKPBLOCK_SAME_REL))
1364 : {
1365 0 : COPY_HEADER_FIELD(&blk->rnode, sizeof(RelFileNode));
1366 0 : rnode = &blk->rnode;
1367 : }
1368 : else
1369 : {
1370 0 : if (rnode == NULL)
1371 : {
1372 0 : report_invalid_record(state,
1373 : "BKPBLOCK_SAME_REL set but no previous rel at %X/%X",
1374 0 : (uint32) (state->ReadRecPtr >> 32), (uint32) state->ReadRecPtr);
1375 0 : goto err;
1376 : }
1377 :
1378 0 : blk->rnode = *rnode;
1379 : }
1380 0 : COPY_HEADER_FIELD(&blk->blkno, sizeof(BlockNumber));
1381 : }
1382 : else
1383 : {
1384 0 : report_invalid_record(state,
1385 : "invalid block_id %u at %X/%X",
1386 : block_id,
1387 0 : (uint32) (state->ReadRecPtr >> 32),
1388 0 : (uint32) state->ReadRecPtr);
1389 0 : goto err;
1390 : }
1391 : }
1392 :
1393 288 : if (remaining != datatotal)
1394 0 : goto shortdata_err;
1395 :
1396 : /*
1397 : * Ok, we've parsed the fragment headers, and verified that the total
1398 : * length of the payload in the fragments is equal to the amount of data
1399 : * left. Copy the data of each fragment to a separate buffer.
1400 : *
1401 : * We could just set up pointers into readRecordBuf, but we want to align
1402 : * the data for the convenience of the callers. Backup images are not
1403 : * copied, however; they don't need alignment.
1404 : */
1405 :
1406 : /* block data first */
1407 288 : for (block_id = 0; block_id <= state->max_block_id; block_id++)
1408 : {
1409 0 : DecodedBkpBlock *blk = &state->blocks[block_id];
1410 :
1411 0 : if (!blk->in_use)
1412 0 : continue;
1413 :
1414 : Assert(blk->has_image || !blk->apply_image);
1415 :
1416 0 : if (blk->has_image)
1417 : {
1418 0 : blk->bkp_image = ptr;
1419 0 : ptr += blk->bimg_len;
1420 : }
1421 0 : if (blk->has_data)
1422 : {
1423 0 : if (!blk->data || blk->data_len > blk->data_bufsz)
1424 : {
1425 0 : if (blk->data)
1426 0 : pfree(blk->data);
1427 :
1428 : /*
1429 : * Force the initial request to be BLCKSZ so that we don't
1430 : * waste time with lots of trips through this stanza as a
1431 : * result of WAL compression.
1432 : */
1433 0 : blk->data_bufsz = MAXALIGN(Max(blk->data_len, BLCKSZ));
1434 0 : blk->data = palloc(blk->data_bufsz);
1435 : }
1436 0 : memcpy(blk->data, ptr, blk->data_len);
1437 0 : ptr += blk->data_len;
1438 : }
1439 : }
1440 :
1441 : /* and finally, the main data */
1442 288 : if (state->main_data_len > 0)
1443 : {
1444 288 : if (!state->main_data || state->main_data_len > state->main_data_bufsz)
1445 : {
1446 72 : if (state->main_data)
1447 0 : pfree(state->main_data);
1448 :
1449 : /*
1450 : * main_data_bufsz must be MAXALIGN'ed. In many xlog record
1451 : * types, we omit trailing struct padding on-disk to save a few
1452 : * bytes; but compilers may generate accesses to the xlog struct
1453 : * that assume that padding bytes are present. If the palloc
1454 : * request is not large enough to include such padding bytes then
1455 : * we'll get valgrind complaints due to otherwise-harmless fetches
1456 : * of the padding bytes.
1457 : *
1458 : * In addition, force the initial request to be reasonably large
1459 : * so that we don't waste time with lots of trips through this
1460 : * stanza. BLCKSZ / 2 seems like a good compromise choice.
1461 : */
1462 72 : state->main_data_bufsz = MAXALIGN(Max(state->main_data_len,
1463 : BLCKSZ / 2));
1464 72 : state->main_data = palloc(state->main_data_bufsz);
1465 : }
1466 288 : memcpy(state->main_data, ptr, state->main_data_len);
1467 288 : ptr += state->main_data_len;
1468 : }
1469 :
1470 288 : return true;
1471 :
1472 0 : shortdata_err:
1473 0 : report_invalid_record(state,
1474 : "record with invalid length at %X/%X",
1475 0 : (uint32) (state->ReadRecPtr >> 32), (uint32) state->ReadRecPtr);
1476 0 : err:
1477 0 : *errormsg = state->errormsg_buf;
1478 :
1479 0 : return false;
1480 : }
1481 :
1482 : /*
1483 : * Returns information about the block that a block reference refers to.
1484 : *
1485 : * If the WAL record contains a block reference with the given ID, *rnode,
1486 : * *forknum, and *blknum are filled in (if not NULL), and returns true.
1487 : * Otherwise returns false.
1488 : */
1489 : bool
1490 0 : XLogRecGetBlockTag(XLogReaderState *record, uint8 block_id,
1491 : RelFileNode *rnode, ForkNumber *forknum, BlockNumber *blknum)
1492 : {
1493 : DecodedBkpBlock *bkpb;
1494 :
1495 0 : if (!record->blocks[block_id].in_use)
1496 0 : return false;
1497 :
1498 0 : bkpb = &record->blocks[block_id];
1499 0 : if (rnode)
1500 0 : *rnode = bkpb->rnode;
1501 0 : if (forknum)
1502 0 : *forknum = bkpb->forknum;
1503 0 : if (blknum)
1504 0 : *blknum = bkpb->blkno;
1505 0 : return true;
1506 : }
1507 :
1508 : /*
1509 : * Returns the data associated with a block reference, or NULL if there is
1510 : * no data (e.g. because a full-page image was taken instead). The returned
1511 : * pointer points to a MAXALIGNed buffer.
1512 : */
1513 : char *
1514 0 : XLogRecGetBlockData(XLogReaderState *record, uint8 block_id, Size *len)
1515 : {
1516 : DecodedBkpBlock *bkpb;
1517 :
1518 0 : if (!record->blocks[block_id].in_use)
1519 0 : return NULL;
1520 :
1521 0 : bkpb = &record->blocks[block_id];
1522 :
1523 0 : if (!bkpb->has_data)
1524 : {
1525 0 : if (len)
1526 0 : *len = 0;
1527 0 : return NULL;
1528 : }
1529 : else
1530 : {
1531 0 : if (len)
1532 0 : *len = bkpb->data_len;
1533 0 : return bkpb->data;
1534 : }
1535 : }
1536 :
1537 : /*
1538 : * Restore a full-page image from a backup block attached to an XLOG record.
1539 : *
1540 : * Returns the buffer number containing the page.
1541 : */
1542 : bool
1543 0 : RestoreBlockImage(XLogReaderState *record, uint8 block_id, char *page)
1544 : {
1545 : DecodedBkpBlock *bkpb;
1546 : char *ptr;
1547 : PGAlignedBlock tmp;
1548 :
1549 0 : if (!record->blocks[block_id].in_use)
1550 0 : return false;
1551 0 : if (!record->blocks[block_id].has_image)
1552 0 : return false;
1553 :
1554 0 : bkpb = &record->blocks[block_id];
1555 0 : ptr = bkpb->bkp_image;
1556 :
1557 0 : if (bkpb->bimg_info & BKPIMAGE_IS_COMPRESSED)
1558 : {
1559 : /* If a backup block image is compressed, decompress it */
1560 0 : if (pglz_decompress(ptr, bkpb->bimg_len, tmp.data,
1561 0 : BLCKSZ - bkpb->hole_length, true) < 0)
1562 : {
1563 0 : report_invalid_record(record, "invalid compressed image at %X/%X, block %d",
1564 0 : (uint32) (record->ReadRecPtr >> 32),
1565 0 : (uint32) record->ReadRecPtr,
1566 : block_id);
1567 0 : return false;
1568 : }
1569 0 : ptr = tmp.data;
1570 : }
1571 :
1572 : /* generate page, taking into account hole if necessary */
1573 0 : if (bkpb->hole_length == 0)
1574 : {
1575 0 : memcpy(page, ptr, BLCKSZ);
1576 : }
1577 : else
1578 : {
1579 0 : memcpy(page, ptr, bkpb->hole_offset);
1580 : /* must zero-fill the hole */
1581 0 : MemSet(page + bkpb->hole_offset, 0, bkpb->hole_length);
1582 0 : memcpy(page + (bkpb->hole_offset + bkpb->hole_length),
1583 0 : ptr + bkpb->hole_offset,
1584 0 : BLCKSZ - (bkpb->hole_offset + bkpb->hole_length));
1585 : }
1586 :
1587 0 : return true;
1588 : }
1589 :
1590 : #ifndef FRONTEND
1591 :
1592 : /*
1593 : * Extract the FullTransactionId from a WAL record.
1594 : */
1595 : FullTransactionId
1596 : XLogRecGetFullXid(XLogReaderState *record)
1597 : {
1598 : TransactionId xid,
1599 : next_xid;
1600 : uint32 epoch;
1601 :
1602 : /*
1603 : * This function is only safe during replay, because it depends on the
1604 : * replay state. See AdvanceNextFullTransactionIdPastXid() for more.
1605 : */
1606 : Assert(AmStartupProcess() || !IsUnderPostmaster);
1607 :
1608 : xid = XLogRecGetXid(record);
1609 : next_xid = XidFromFullTransactionId(ShmemVariableCache->nextFullXid);
1610 : epoch = EpochFromFullTransactionId(ShmemVariableCache->nextFullXid);
1611 :
1612 : /*
1613 : * If xid is numerically greater than next_xid, it has to be from the last
1614 : * epoch.
1615 : */
1616 : if (unlikely(xid > next_xid))
1617 : --epoch;
1618 :
1619 : return FullTransactionIdFromEpochAndXid(epoch, xid);
1620 : }
1621 :
1622 : #endif
|
