Age Owner Branch data TLA Line data Source code
1 : : /*-------------------------------------------------------------------------
2 : : *
3 : : * reconstruct.c
4 : : * Reconstruct full file from incremental file and backup chain.
5 : : *
6 : : * Copyright (c) 2017-2025, PostgreSQL Global Development Group
7 : : *
8 : : * IDENTIFICATION
9 : : * src/bin/pg_combinebackup/reconstruct.c
10 : : *
11 : : *-------------------------------------------------------------------------
12 : : */
13 : : #include "postgres_fe.h"
14 : :
15 : : #include <unistd.h>
16 : :
17 : : #include "backup/basebackup_incremental.h"
18 : : #include "common/file_perm.h"
19 : : #include "common/logging.h"
20 : : #include "copy_file.h"
21 : : #include "lib/stringinfo.h"
22 : : #include "reconstruct.h"
23 : : #include "storage/block.h"
24 : :
25 : : /*
26 : : * An rfile stores the data that we need in order to be able to use some file
27 : : * on disk for reconstruction. For any given output file, we create one rfile
28 : : * per backup that we need to consult when we constructing that output file.
29 : : *
30 : : * If we find a full version of the file in the backup chain, then only
31 : : * filename and fd are initialized; the remaining fields are 0 or NULL.
32 : : * For an incremental file, header_length, num_blocks, relative_block_numbers,
33 : : * and truncation_block_length are also set.
34 : : *
35 : : * num_blocks_read and highest_offset_read always start out as 0.
36 : : */
37 : : typedef struct rfile
38 : : {
39 : : char *filename;
40 : : int fd;
41 : : size_t header_length;
42 : : unsigned num_blocks;
43 : : BlockNumber *relative_block_numbers;
44 : : unsigned truncation_block_length;
45 : : unsigned num_blocks_read;
46 : : off_t highest_offset_read;
47 : : } rfile;
48 : :
49 : : static void debug_reconstruction(int n_source,
50 : : rfile **sources,
51 : : bool dry_run);
52 : : static unsigned find_reconstructed_block_length(rfile *s);
53 : : static rfile *make_incremental_rfile(char *filename);
54 : : static rfile *make_rfile(char *filename, bool missing_ok);
55 : : static void write_reconstructed_file(char *input_filename,
56 : : char *output_filename,
57 : : unsigned block_length,
58 : : rfile **sourcemap,
59 : : off_t *offsetmap,
60 : : pg_checksum_context *checksum_ctx,
61 : : CopyMethod copy_method,
62 : : bool debug,
63 : : bool dry_run);
64 : : static void read_bytes(rfile *rf, void *buffer, unsigned length);
65 : : static void write_block(int fd, char *output_filename,
66 : : uint8 *buffer,
67 : : pg_checksum_context *checksum_ctx);
68 : : static void read_block(rfile *s, off_t off, uint8 *buffer);
69 : :
70 : : /*
71 : : * Reconstruct a full file from an incremental file and a chain of prior
72 : : * backups.
73 : : *
74 : : * input_filename should be the path to the incremental file, and
75 : : * output_filename should be the path where the reconstructed file is to be
76 : : * written.
77 : : *
78 : : * relative_path should be the path to the directory containing this file,
79 : : * relative to the root of the backup (NOT relative to the root of the
80 : : * tablespace). It must always end with a trailing slash. bare_file_name
81 : : * should be the name of the file within that directory, without
82 : : * "INCREMENTAL.".
83 : : *
84 : : * n_prior_backups is the number of prior backups, and prior_backup_dirs is
85 : : * an array of pathnames where those backups can be found.
86 : : */
87 : : void
677 rhaas@postgresql.org 88 :CBC 6206 : reconstruct_from_incremental_file(char *input_filename,
89 : : char *output_filename,
90 : : char *relative_path,
91 : : char *bare_file_name,
92 : : int n_prior_backups,
93 : : char **prior_backup_dirs,
94 : : manifest_data **manifests,
95 : : char *manifest_path,
96 : : pg_checksum_type checksum_type,
97 : : int *checksum_length,
98 : : uint8 **checksum_payload,
99 : : CopyMethod copy_method,
100 : : bool debug,
101 : : bool dry_run)
102 : : {
103 : : rfile **source;
104 : 6206 : rfile *latest_source = NULL;
105 : : rfile **sourcemap;
106 : : off_t *offsetmap;
107 : : unsigned block_length;
108 : : unsigned i;
109 : 6206 : unsigned sidx = n_prior_backups;
110 : 6206 : bool full_copy_possible = true;
111 : 6206 : int copy_source_index = -1;
112 : 6206 : rfile *copy_source = NULL;
113 : : pg_checksum_context checksum_ctx;
114 : :
115 : : /* Sanity check the relative_path. */
357 116 [ - + ]: 6206 : Assert(relative_path[0] != '\0');
117 [ - + ]: 6206 : Assert(relative_path[strlen(relative_path) - 1] == '/');
118 : :
119 : : /*
120 : : * Every block must come either from the latest version of the file or
121 : : * from one of the prior backups.
122 : : */
677 123 : 6206 : source = pg_malloc0(sizeof(rfile *) * (1 + n_prior_backups));
124 : :
125 : : /*
126 : : * Use the information from the latest incremental file to figure out how
127 : : * long the reconstructed file should be.
128 : : */
129 : 6206 : latest_source = make_incremental_rfile(input_filename);
130 : 6206 : source[n_prior_backups] = latest_source;
131 : 6206 : block_length = find_reconstructed_block_length(latest_source);
132 : :
133 : : /*
134 : : * For each block in the output file, we need to know from which file we
135 : : * need to obtain it and at what offset in that file it's stored.
136 : : * sourcemap gives us the first of these things, and offsetmap the latter.
137 : : */
138 : 6206 : sourcemap = pg_malloc0(sizeof(rfile *) * block_length);
139 : 6206 : offsetmap = pg_malloc0(sizeof(off_t) * block_length);
140 : :
141 : : /*
142 : : * Every block that is present in the newest incremental file should be
143 : : * sourced from that file. If it precedes the truncation_block_length,
144 : : * it's a block that we would otherwise have had to find in an older
145 : : * backup and thus reduces the number of blocks remaining to be found by
146 : : * one; otherwise, it's an extra block that needs to be included in the
147 : : * output but would not have needed to be found in an older backup if it
148 : : * had not been present.
149 : : */
150 [ + + ]: 6248 : for (i = 0; i < latest_source->num_blocks; ++i)
151 : : {
152 : 42 : BlockNumber b = latest_source->relative_block_numbers[i];
153 : :
154 [ - + ]: 42 : Assert(b < block_length);
155 : 42 : sourcemap[b] = latest_source;
156 : 42 : offsetmap[b] = latest_source->header_length + (i * BLCKSZ);
157 : :
158 : : /*
159 : : * A full copy of a file from an earlier backup is only possible if no
160 : : * blocks are needed from any later incremental file.
161 : : */
162 : 42 : full_copy_possible = false;
163 : : }
164 : :
165 : : while (1)
166 : 1332 : {
167 : : char source_filename[MAXPGPATH];
168 : : rfile *s;
169 : :
170 : : /*
171 : : * Move to the next backup in the chain. If there are no more, then
172 : : * we're done.
173 : : */
174 [ + + ]: 7538 : if (sidx == 0)
175 : 1 : break;
176 : 7537 : --sidx;
177 : :
178 : : /*
179 : : * Look for the full file in the previous backup. If not found, then
180 : : * look for an incremental file instead.
181 : : */
357 182 : 7537 : snprintf(source_filename, MAXPGPATH, "%s/%s%s",
677 183 : 7537 : prior_backup_dirs[sidx], relative_path, bare_file_name);
184 [ + + ]: 7537 : if ((s = make_rfile(source_filename, true)) == NULL)
185 : : {
357 186 : 1332 : snprintf(source_filename, MAXPGPATH, "%s/%sINCREMENTAL.%s",
677 187 : 1332 : prior_backup_dirs[sidx], relative_path, bare_file_name);
188 : 1332 : s = make_incremental_rfile(source_filename);
189 : : }
190 : 7537 : source[sidx] = s;
191 : :
192 : : /*
193 : : * If s->header_length == 0, then this is a full file; otherwise, it's
194 : : * an incremental file.
195 : : */
196 [ + + ]: 7537 : if (s->header_length == 0)
197 : : {
198 : : struct stat sb;
199 : : BlockNumber b;
200 : : BlockNumber blocklength;
201 : :
202 : : /* We need to know the length of the file. */
203 [ - + ]: 6205 : if (fstat(s->fd, &sb) < 0)
493 peter@eisentraut.org 204 :UBC 0 : pg_fatal("could not stat file \"%s\": %m", s->filename);
205 : :
206 : : /*
207 : : * Since we found a full file, source all blocks from it that
208 : : * exist in the file.
209 : : *
210 : : * Note that there may be blocks that don't exist either in this
211 : : * file or in any incremental file but that precede
212 : : * truncation_block_length. These are, presumably, zero-filled
213 : : * blocks that result from the server extending the file but
214 : : * taking no action on those blocks that generated any WAL.
215 : : *
216 : : * Sadly, we have no way of validating that this is really what
217 : : * happened, and neither does the server. From its perspective,
218 : : * an unmodified block that contains data looks exactly the same
219 : : * as a zero-filled block that never had any data: either way,
220 : : * it's not mentioned in any WAL summary and the server has no
221 : : * reason to read it. From our perspective, all we know is that
222 : : * nobody had a reason to back up the block. That certainly means
223 : : * that the block didn't exist at the time of the full backup, but
224 : : * the supposition that it was all zeroes at the time of every
225 : : * later backup is one that we can't validate.
226 : : */
677 rhaas@postgresql.org 227 :CBC 6205 : blocklength = sb.st_size / BLCKSZ;
228 [ + + ]: 30513 : for (b = 0; b < latest_source->truncation_block_length; ++b)
229 : : {
230 [ + + + - ]: 24308 : if (sourcemap[b] == NULL && b < blocklength)
231 : : {
232 : 24266 : sourcemap[b] = s;
233 : 24266 : offsetmap[b] = b * BLCKSZ;
234 : : }
235 : : }
236 : :
237 : : /*
238 : : * If a full copy looks possible, check whether the resulting file
239 : : * should be exactly as long as the source file is. If so, a full
240 : : * copy is acceptable, otherwise not.
241 : : */
242 [ + + ]: 6205 : if (full_copy_possible)
243 : : {
244 : : uint64 expected_length;
245 : :
246 : 6180 : expected_length =
247 : 6180 : (uint64) latest_source->truncation_block_length;
248 : 6180 : expected_length *= BLCKSZ;
249 [ + - ]: 6180 : if (expected_length == sb.st_size)
250 : : {
251 : 6180 : copy_source = s;
252 : 6180 : copy_source_index = sidx;
253 : : }
254 : : }
255 : :
256 : : /* We don't need to consider any further sources. */
257 : 6205 : break;
258 : : }
259 : :
260 : : /*
261 : : * Since we found another incremental file, source all blocks from it
262 : : * that we need but don't yet have.
263 : : */
264 [ - + ]: 1332 : for (i = 0; i < s->num_blocks; ++i)
265 : : {
677 rhaas@postgresql.org 266 :UBC 0 : BlockNumber b = s->relative_block_numbers[i];
267 : :
268 [ # # ]: 0 : if (b < latest_source->truncation_block_length &&
269 [ # # ]: 0 : sourcemap[b] == NULL)
270 : : {
271 : 0 : sourcemap[b] = s;
272 : 0 : offsetmap[b] = s->header_length + (i * BLCKSZ);
273 : :
274 : : /*
275 : : * A full copy of a file from an earlier backup is only
276 : : * possible if no blocks are needed from any later incremental
277 : : * file.
278 : : */
279 : 0 : full_copy_possible = false;
280 : : }
281 : : }
282 : : }
283 : :
284 : : /*
285 : : * If a checksum of the required type already exists in the
286 : : * backup_manifest for the relevant input directory, we can save some work
287 : : * by reusing that checksum instead of computing a new one.
288 : : */
677 rhaas@postgresql.org 289 [ + + + - :CBC 6206 : if (copy_source_index >= 0 && manifests[copy_source_index] != NULL &&
+ - ]
290 : : checksum_type != CHECKSUM_TYPE_NONE)
291 : : {
292 : : manifest_file *mfile;
293 : :
294 : 6180 : mfile = manifest_files_lookup(manifests[copy_source_index]->files,
295 : : manifest_path);
296 [ - + ]: 6180 : if (mfile == NULL)
297 : : {
677 rhaas@postgresql.org 298 :UBC 0 : char *path = psprintf("%s/backup_manifest",
299 : 0 : prior_backup_dirs[copy_source_index]);
300 : :
301 : : /*
302 : : * The directory is out of sync with the backup_manifest, so emit
303 : : * a warning.
304 : : */
493 peter@eisentraut.org 305 : 0 : pg_log_warning("manifest file \"%s\" contains no entry for file \"%s\"",
306 : : path,
307 : : manifest_path);
677 rhaas@postgresql.org 308 : 0 : pfree(path);
309 : : }
677 rhaas@postgresql.org 310 [ + - ]:CBC 6180 : else if (mfile->checksum_type == checksum_type)
311 : : {
312 : 6180 : *checksum_length = mfile->checksum_length;
313 : 6180 : *checksum_payload = pg_malloc(*checksum_length);
314 : 6180 : memcpy(*checksum_payload, mfile->checksum_payload,
315 : 6180 : *checksum_length);
316 : 6180 : checksum_type = CHECKSUM_TYPE_NONE;
317 : : }
318 : : }
319 : :
320 : : /* Prepare for checksum calculation, if required. */
321 : 6206 : pg_checksum_init(&checksum_ctx, checksum_type);
322 : :
323 : : /*
324 : : * If the full file can be created by copying a file from an older backup
325 : : * in the chain without needing to overwrite any blocks or truncate the
326 : : * result, then forget about performing reconstruction and just copy that
327 : : * file in its entirety.
328 : : *
329 : : * If we have only incremental files, and there's no full file at any
330 : : * point in the backup chain, something has gone wrong. Emit an error.
331 : : *
332 : : * Otherwise, reconstruct.
333 : : */
334 [ + + ]: 6206 : if (copy_source != NULL)
335 : 6180 : copy_file(copy_source->filename, output_filename,
336 : : &checksum_ctx, copy_method, dry_run);
357 337 [ + - + + ]: 26 : else if (sidx == 0 && source[0]->header_length != 0)
338 : : {
339 : 1 : pg_fatal("full backup contains unexpected incremental file \"%s\"",
340 : : source[0]->filename);
341 : : }
342 : : else
343 : : {
677 344 : 25 : write_reconstructed_file(input_filename, output_filename,
345 : : block_length, sourcemap, offsetmap,
346 : : &checksum_ctx, copy_method,
347 : : debug, dry_run);
348 : 25 : debug_reconstruction(n_prior_backups + 1, source, dry_run);
349 : : }
350 : :
351 : : /* Save results of checksum calculation. */
352 [ + + ]: 6205 : if (checksum_type != CHECKSUM_TYPE_NONE)
353 : : {
354 : 25 : *checksum_payload = pg_malloc(PG_CHECKSUM_MAX_LENGTH);
355 : 25 : *checksum_length = pg_checksum_final(&checksum_ctx,
356 : : *checksum_payload);
357 : : }
358 : :
359 : : /*
360 : : * Close files and release memory.
361 : : */
362 [ + + ]: 19946 : for (i = 0; i <= n_prior_backups; ++i)
363 : : {
364 : 13741 : rfile *s = source[i];
365 : :
366 [ - + ]: 13741 : if (s == NULL)
677 rhaas@postgresql.org 367 :UBC 0 : continue;
677 rhaas@postgresql.org 368 [ - + ]:CBC 13741 : if (close(s->fd) != 0)
493 peter@eisentraut.org 369 :UBC 0 : pg_fatal("could not close file \"%s\": %m", s->filename);
677 rhaas@postgresql.org 370 [ + + ]:CBC 13741 : if (s->relative_block_numbers != NULL)
371 : 25 : pfree(s->relative_block_numbers);
372 : 13741 : pg_free(s->filename);
5 tgl@sss.pgh.pa.us 373 : 13741 : pg_free(s);
374 : : }
677 rhaas@postgresql.org 375 : 6205 : pfree(sourcemap);
376 : 6205 : pfree(offsetmap);
377 : 6205 : pfree(source);
378 : 6205 : }
379 : :
380 : : /*
381 : : * Perform post-reconstruction logging and sanity checks.
382 : : */
383 : : static void
384 : 25 : debug_reconstruction(int n_source, rfile **sources, bool dry_run)
385 : : {
386 : : unsigned i;
387 : :
388 [ + + ]: 78 : for (i = 0; i < n_source; ++i)
389 : : {
390 : 53 : rfile *s = sources[i];
391 : :
392 : : /* Ignore source if not used. */
393 [ - + ]: 53 : if (s == NULL)
677 rhaas@postgresql.org 394 :UBC 0 : continue;
395 : :
396 : : /* If no data is needed from this file, we can ignore it. */
677 rhaas@postgresql.org 397 [ + + ]:CBC 53 : if (s->num_blocks_read == 0)
398 : 3 : continue;
399 : :
400 : : /* Debug logging. */
401 [ - + ]: 50 : if (dry_run)
677 rhaas@postgresql.org 402 [ # # ]:UBC 0 : pg_log_debug("would have read %u blocks from \"%s\"",
403 : : s->num_blocks_read, s->filename);
404 : : else
677 rhaas@postgresql.org 405 [ + - ]:CBC 50 : pg_log_debug("read %u blocks from \"%s\"",
406 : : s->num_blocks_read, s->filename);
407 : :
408 : : /*
409 : : * In dry-run mode, we don't actually try to read data from the file,
410 : : * but we do try to verify that the file is long enough that we could
411 : : * have read the data if we'd tried.
412 : : *
413 : : * If this fails, then it means that a non-dry-run attempt would fail,
414 : : * complaining of not being able to read the required bytes from the
415 : : * file.
416 : : */
417 [ - + ]: 50 : if (dry_run)
418 : : {
419 : : struct stat sb;
420 : :
677 rhaas@postgresql.org 421 [ # # ]:UBC 0 : if (fstat(s->fd, &sb) < 0)
493 peter@eisentraut.org 422 : 0 : pg_fatal("could not stat file \"%s\": %m", s->filename);
677 rhaas@postgresql.org 423 [ # # ]: 0 : if (sb.st_size < s->highest_offset_read)
424 : 0 : pg_fatal("file \"%s\" is too short: expected %llu, found %llu",
425 : : s->filename,
426 : : (unsigned long long) s->highest_offset_read,
427 : : (unsigned long long) sb.st_size);
428 : : }
429 : : }
677 rhaas@postgresql.org 430 :CBC 25 : }
431 : :
432 : : /*
433 : : * When we perform reconstruction using an incremental file, the output file
434 : : * should be at least as long as the truncation_block_length. Any blocks
435 : : * present in the incremental file increase the output length as far as is
436 : : * necessary to include those blocks.
437 : : */
438 : : static unsigned
439 : 6206 : find_reconstructed_block_length(rfile *s)
440 : : {
441 : 6206 : unsigned block_length = s->truncation_block_length;
442 : : unsigned i;
443 : :
444 [ + + ]: 6248 : for (i = 0; i < s->num_blocks; ++i)
445 [ - + ]: 42 : if (s->relative_block_numbers[i] >= block_length)
677 rhaas@postgresql.org 446 :UBC 0 : block_length = s->relative_block_numbers[i] + 1;
447 : :
677 rhaas@postgresql.org 448 :CBC 6206 : return block_length;
449 : : }
450 : :
451 : : /*
452 : : * Initialize an incremental rfile, reading the header so that we know which
453 : : * blocks it contains.
454 : : */
455 : : static rfile *
456 : 7538 : make_incremental_rfile(char *filename)
457 : : {
458 : : rfile *rf;
459 : : unsigned magic;
460 : :
461 : 7538 : rf = make_rfile(filename, false);
462 : :
463 : : /* Read and validate magic number. */
464 : 7538 : read_bytes(rf, &magic, sizeof(magic));
465 [ - + ]: 7538 : if (magic != INCREMENTAL_MAGIC)
426 peter@eisentraut.org 466 :UBC 0 : pg_fatal("file \"%s\" has bad incremental magic number (0x%x, expected 0x%x)",
467 : : filename, magic, INCREMENTAL_MAGIC);
468 : :
469 : : /* Read block count. */
677 rhaas@postgresql.org 470 :CBC 7538 : read_bytes(rf, &rf->num_blocks, sizeof(rf->num_blocks));
471 [ - + ]: 7538 : if (rf->num_blocks > RELSEG_SIZE)
677 rhaas@postgresql.org 472 :UBC 0 : pg_fatal("file \"%s\" has block count %u in excess of segment size %u",
473 : : filename, rf->num_blocks, RELSEG_SIZE);
474 : :
475 : : /* Read truncation block length. */
677 rhaas@postgresql.org 476 :CBC 7538 : read_bytes(rf, &rf->truncation_block_length,
477 : : sizeof(rf->truncation_block_length));
478 [ - + ]: 7538 : if (rf->truncation_block_length > RELSEG_SIZE)
677 rhaas@postgresql.org 479 :UBC 0 : pg_fatal("file \"%s\" has truncation block length %u in excess of segment size %u",
480 : : filename, rf->truncation_block_length, RELSEG_SIZE);
481 : :
482 : : /* Read block numbers if there are any. */
677 rhaas@postgresql.org 483 [ + + ]:CBC 7538 : if (rf->num_blocks > 0)
484 : : {
485 : 25 : rf->relative_block_numbers =
486 : 25 : pg_malloc0(sizeof(BlockNumber) * rf->num_blocks);
487 : 25 : read_bytes(rf, rf->relative_block_numbers,
488 : 25 : sizeof(BlockNumber) * rf->num_blocks);
489 : : }
490 : :
491 : : /* Remember length of header. */
492 : 7538 : rf->header_length = sizeof(magic) + sizeof(rf->num_blocks) +
493 : 7538 : sizeof(rf->truncation_block_length) +
494 : 7538 : sizeof(BlockNumber) * rf->num_blocks;
495 : :
496 : : /*
497 : : * Round header length to a multiple of BLCKSZ, so that blocks contents
498 : : * are properly aligned. Only do this when the file actually has data for
499 : : * some blocks.
500 : : */
570 tomas.vondra@postgre 501 [ + + + - ]: 7538 : if ((rf->num_blocks > 0) && ((rf->header_length % BLCKSZ) != 0))
502 : 25 : rf->header_length += (BLCKSZ - (rf->header_length % BLCKSZ));
503 : :
677 rhaas@postgresql.org 504 : 7538 : return rf;
505 : : }
506 : :
507 : : /*
508 : : * Allocate and perform basic initialization of an rfile.
509 : : */
510 : : static rfile *
511 : 15075 : make_rfile(char *filename, bool missing_ok)
512 : : {
513 : : rfile *rf;
514 : :
515 : 15075 : rf = pg_malloc0(sizeof(rfile));
516 : 15075 : rf->filename = pstrdup(filename);
517 [ + + ]: 15075 : if ((rf->fd = open(filename, O_RDONLY | PG_BINARY, 0)) < 0)
518 : : {
519 [ + - + - ]: 1332 : if (missing_ok && errno == ENOENT)
520 : : {
5 tgl@sss.pgh.pa.us 521 : 1332 : pg_free(rf->filename);
677 rhaas@postgresql.org 522 : 1332 : pg_free(rf);
523 : 1332 : return NULL;
524 : : }
677 rhaas@postgresql.org 525 :UBC 0 : pg_fatal("could not open file \"%s\": %m", filename);
526 : : }
527 : :
677 rhaas@postgresql.org 528 :CBC 13743 : return rf;
529 : : }
530 : :
531 : : /*
532 : : * Read the indicated number of bytes from an rfile into the buffer.
533 : : */
534 : : static void
535 : 22639 : read_bytes(rfile *rf, void *buffer, unsigned length)
536 : : {
670 tgl@sss.pgh.pa.us 537 : 22639 : int rb = read(rf->fd, buffer, length);
538 : :
677 rhaas@postgresql.org 539 [ - + ]: 22639 : if (rb != length)
540 : : {
677 rhaas@postgresql.org 541 [ # # ]:UBC 0 : if (rb < 0)
542 : 0 : pg_fatal("could not read file \"%s\": %m", rf->filename);
543 : : else
493 peter@eisentraut.org 544 : 0 : pg_fatal("could not read file \"%s\": read %d of %u",
545 : : rf->filename, rb, length);
546 : : }
677 rhaas@postgresql.org 547 :CBC 22639 : }
548 : :
549 : : /*
550 : : * Write out a reconstructed file.
551 : : */
552 : : static void
553 : 25 : write_reconstructed_file(char *input_filename,
554 : : char *output_filename,
555 : : unsigned block_length,
556 : : rfile **sourcemap,
557 : : off_t *offsetmap,
558 : : pg_checksum_context *checksum_ctx,
559 : : CopyMethod copy_method,
560 : : bool debug,
561 : : bool dry_run)
562 : : {
563 : 25 : int wfd = -1;
564 : : unsigned i;
565 : 25 : unsigned zero_blocks = 0;
566 : :
567 : : /* Debugging output. */
568 [ + - ]: 25 : if (debug)
569 : : {
570 : : StringInfoData debug_buf;
571 : 25 : unsigned start_of_range = 0;
572 : 25 : unsigned current_block = 0;
573 : :
574 : : /* Basic information about the output file to be produced. */
575 [ - + ]: 25 : if (dry_run)
677 rhaas@postgresql.org 576 [ # # ]:UBC 0 : pg_log_debug("would reconstruct \"%s\" (%u blocks, checksum %s)",
577 : : output_filename, block_length,
578 : : pg_checksum_type_name(checksum_ctx->type));
579 : : else
677 rhaas@postgresql.org 580 [ + - ]:CBC 25 : pg_log_debug("reconstructing \"%s\" (%u blocks, checksum %s)",
581 : : output_filename, block_length,
582 : : pg_checksum_type_name(checksum_ctx->type));
583 : :
584 : : /* Print out the plan for reconstructing this file. */
585 : 25 : initStringInfo(&debug_buf);
586 [ + + ]: 369 : while (current_block < block_length)
587 : : {
588 : 344 : rfile *s = sourcemap[current_block];
589 : :
590 : : /* Extend range, if possible. */
591 [ + + ]: 344 : if (current_block + 1 < block_length &&
592 [ + + ]: 319 : s == sourcemap[current_block + 1])
593 : : {
594 : 277 : ++current_block;
595 : 277 : continue;
596 : : }
597 : :
598 : : /* Add details about this range. */
599 [ - + ]: 67 : if (s == NULL)
600 : : {
677 rhaas@postgresql.org 601 [ # # ]:UBC 0 : if (current_block == start_of_range)
602 : 0 : appendStringInfo(&debug_buf, " %u:zero", current_block);
603 : : else
604 : 0 : appendStringInfo(&debug_buf, " %u-%u:zero",
605 : : start_of_range, current_block);
606 : : }
607 : : else
608 : : {
677 rhaas@postgresql.org 609 [ + + ]:CBC 67 : if (current_block == start_of_range)
610 : 39 : appendStringInfo(&debug_buf, " %u:%s@" UINT64_FORMAT,
611 : : current_block, s->filename,
612 : 39 : (uint64) offsetmap[current_block]);
613 : : else
614 : 28 : appendStringInfo(&debug_buf, " %u-%u:%s@" UINT64_FORMAT,
615 : : start_of_range, current_block,
616 : : s->filename,
617 : 28 : (uint64) offsetmap[current_block]);
618 : : }
619 : :
620 : : /* Begin new range. */
621 : 67 : start_of_range = ++current_block;
622 : :
623 : : /* If the output is very long or we are done, dump it now. */
624 [ + + - + ]: 67 : if (current_block == block_length || debug_buf.len > 1024)
625 : : {
626 [ + - ]: 25 : pg_log_debug("reconstruction plan:%s", debug_buf.data);
627 : 25 : resetStringInfo(&debug_buf);
628 : : }
629 : : }
630 : :
631 : : /* Free memory. */
632 : 25 : pfree(debug_buf.data);
633 : : }
634 : :
635 : : /* Open the output file, except in dry_run mode. */
636 [ + - - + ]: 50 : if (!dry_run &&
637 : 25 : (wfd = open(output_filename,
638 : : O_RDWR | PG_BINARY | O_CREAT | O_EXCL,
639 : : pg_file_create_mode)) < 0)
677 rhaas@postgresql.org 640 :UBC 0 : pg_fatal("could not open file \"%s\": %m", output_filename);
641 : :
642 : : /* Read and write the blocks as required. */
677 rhaas@postgresql.org 643 [ + + ]:CBC 369 : for (i = 0; i < block_length; ++i)
644 : : {
645 : : uint8 buffer[BLCKSZ];
646 : 344 : rfile *s = sourcemap[i];
647 : :
648 : : /* Update accounting information. */
649 [ - + ]: 344 : if (s == NULL)
677 rhaas@postgresql.org 650 :UBC 0 : ++zero_blocks;
651 : : else
652 : : {
677 rhaas@postgresql.org 653 :CBC 344 : s->num_blocks_read++;
654 : 344 : s->highest_offset_read = Max(s->highest_offset_read,
655 : : offsetmap[i] + BLCKSZ);
656 : : }
657 : :
658 : : /* Skip the rest of this in dry-run mode. */
659 [ - + ]: 344 : if (dry_run)
677 rhaas@postgresql.org 660 :UBC 0 : continue;
661 : :
662 : : /* Read or zero-fill the block as appropriate. */
677 rhaas@postgresql.org 663 [ - + ]:CBC 344 : if (s == NULL)
664 : : {
665 : : /*
666 : : * New block not mentioned in the WAL summary. Should have been an
667 : : * uninitialized block, so just zero-fill it.
668 : : */
677 rhaas@postgresql.org 669 :UBC 0 : memset(buffer, 0, BLCKSZ);
670 : :
671 : : /* Write out the block, update the checksum if needed. */
570 tomas.vondra@postgre 672 : 0 : write_block(wfd, output_filename, buffer, checksum_ctx);
673 : :
674 : : /* Nothing else to do for zero-filled blocks. */
675 : 0 : continue;
676 : : }
677 : :
678 : : /* Copy the block using the appropriate copy method. */
570 tomas.vondra@postgre 679 [ + - ]:CBC 344 : if (copy_method != COPY_METHOD_COPY_FILE_RANGE)
680 : : {
681 : : /*
682 : : * Read the block from the correct source file, and then write it
683 : : * out, possibly with a checksum update.
684 : : */
685 : 344 : read_block(s, offsetmap[i], buffer);
686 : 344 : write_block(wfd, output_filename, buffer, checksum_ctx);
687 : : }
688 : : else /* use copy_file_range */
689 : : {
690 : : #if defined(HAVE_COPY_FILE_RANGE)
691 : : /* copy_file_range modifies the offset, so use a local copy */
570 tomas.vondra@postgre 692 :UBC 0 : off_t off = offsetmap[i];
693 : 0 : size_t nwritten = 0;
694 : :
695 : : /*
696 : : * Retry until we've written all the bytes (the offset is updated
697 : : * by copy_file_range, and so is the wfd file offset).
698 : : */
699 : : do
700 : : {
701 : : int wb;
702 : :
703 : 0 : wb = copy_file_range(s->fd, &off, wfd, NULL, BLCKSZ - nwritten, 0);
704 : :
705 [ # # ]: 0 : if (wb < 0)
706 : 0 : pg_fatal("error while copying file range from \"%s\" to \"%s\": %m",
707 : : input_filename, output_filename);
708 : :
709 : 0 : nwritten += wb;
710 : :
711 [ # # ]: 0 : } while (BLCKSZ > nwritten);
712 : :
713 : : /*
714 : : * When checksum calculation not needed, we're done, otherwise
715 : : * read the block and pass it to the checksum calculation.
716 : : */
717 [ # # ]: 0 : if (checksum_ctx->type == CHECKSUM_TYPE_NONE)
718 : 0 : continue;
719 : :
720 : 0 : read_block(s, offsetmap[i], buffer);
721 : :
722 [ # # ]: 0 : if (pg_checksum_update(checksum_ctx, buffer, BLCKSZ) < 0)
723 : 0 : pg_fatal("could not update checksum of file \"%s\"",
724 : : output_filename);
725 : : #else
726 : : pg_fatal("copy_file_range not supported on this platform");
727 : : #endif
728 : : }
729 : : }
730 : :
731 : : /* Debugging output. */
677 rhaas@postgresql.org 732 [ - + ]:CBC 25 : if (zero_blocks > 0)
733 : : {
677 rhaas@postgresql.org 734 [ # # ]:UBC 0 : if (dry_run)
735 [ # # ]: 0 : pg_log_debug("would have zero-filled %u blocks", zero_blocks);
736 : : else
737 [ # # ]: 0 : pg_log_debug("zero-filled %u blocks", zero_blocks);
738 : : }
739 : :
740 : : /* Close the output file. */
677 rhaas@postgresql.org 741 [ + - - + ]:CBC 25 : if (wfd >= 0 && close(wfd) != 0)
493 peter@eisentraut.org 742 :UBC 0 : pg_fatal("could not close file \"%s\": %m", output_filename);
677 rhaas@postgresql.org 743 :CBC 25 : }
744 : :
745 : : /*
746 : : * Write the block into the file (using the file descriptor), and
747 : : * if needed update the checksum calculation.
748 : : *
749 : : * The buffer is expected to contain BLCKSZ bytes. The filename is
750 : : * provided only for the error message.
751 : : */
752 : : static void
570 tomas.vondra@postgre 753 : 344 : write_block(int fd, char *output_filename,
754 : : uint8 *buffer, pg_checksum_context *checksum_ctx)
755 : : {
756 : : int wb;
757 : :
758 [ - + ]: 344 : if ((wb = write(fd, buffer, BLCKSZ)) != BLCKSZ)
759 : : {
570 tomas.vondra@postgre 760 [ # # ]:UBC 0 : if (wb < 0)
761 : 0 : pg_fatal("could not write file \"%s\": %m", output_filename);
762 : : else
493 peter@eisentraut.org 763 : 0 : pg_fatal("could not write file \"%s\": wrote %d of %d",
764 : : output_filename, wb, BLCKSZ);
765 : : }
766 : :
767 : : /* Update the checksum computation. */
570 tomas.vondra@postgre 768 [ - + ]:CBC 344 : if (pg_checksum_update(checksum_ctx, buffer, BLCKSZ) < 0)
570 tomas.vondra@postgre 769 :UBC 0 : pg_fatal("could not update checksum of file \"%s\"",
770 : : output_filename);
570 tomas.vondra@postgre 771 :CBC 344 : }
772 : :
773 : : /*
774 : : * Read a block of data (BLCKSZ bytes) into the buffer.
775 : : */
776 : : static void
777 : 344 : read_block(rfile *s, off_t off, uint8 *buffer)
778 : : {
779 : : int rb;
780 : :
781 : : /* Read the block from the correct source, except if dry-run. */
782 : 344 : rb = pg_pread(s->fd, buffer, BLCKSZ, off);
783 [ - + ]: 344 : if (rb != BLCKSZ)
784 : : {
570 tomas.vondra@postgre 785 [ # # ]:UBC 0 : if (rb < 0)
493 peter@eisentraut.org 786 : 0 : pg_fatal("could not read from file \"%s\": %m", s->filename);
787 : : else
788 : 0 : pg_fatal("could not read from file \"%s\", offset %llu: read %d of %d",
789 : : s->filename, (unsigned long long) off, rb, BLCKSZ);
790 : : }
570 tomas.vondra@postgre 791 :CBC 344 : }
|
