| 1 | /*------------------------------------------------------------------------- |
| 2 | * |
| 3 | * pg_backup_directory.c |
| 4 | * |
| 5 | * A directory format dump is a directory, which contains a "toc.dat" file |
| 6 | * for the TOC, and a separate file for each data entry, named "<oid>.dat". |
| 7 | * Large objects (BLOBs) are stored in separate files named "blob_<uid>.dat", |
| 8 | * and there's a plain-text TOC file for them called "blobs.toc". If |
| 9 | * compression is used, each data file is individually compressed and the |
| 10 | * ".gz" suffix is added to the filenames. The TOC files are never |
| 11 | * compressed by pg_dump, however they are accepted with the .gz suffix too, |
| 12 | * in case the user has manually compressed them with 'gzip'. |
| 13 | * |
| 14 | * NOTE: This format is identical to the files written in the tar file in |
| 15 | * the 'tar' format, except that we don't write the restore.sql file (TODO), |
| 16 | * and the tar format doesn't support compression. Please keep the formats in |
| 17 | * sync. |
| 18 | * |
| 19 | * |
| 20 | * Portions Copyright (c) 1996-2019, PostgreSQL Global Development Group |
| 21 | * Portions Copyright (c) 1994, Regents of the University of California |
| 22 | * Portions Copyright (c) 2000, Philip Warner |
| 23 | * |
| 24 | * Rights are granted to use this software in any way so long |
| 25 | * as this notice is not removed. |
| 26 | * |
| 27 | * The author is not responsible for loss or damages that may |
| 28 | * result from its use. |
| 29 | * |
| 30 | * IDENTIFICATION |
| 31 | * src/bin/pg_dump/pg_backup_directory.c |
| 32 | * |
| 33 | *------------------------------------------------------------------------- |
| 34 | */ |
| 35 | #include "postgres_fe.h" |
| 36 | |
| 37 | #include "compress_io.h" |
| 38 | #include "parallel.h" |
| 39 | #include "pg_backup_utils.h" |
| 40 | #include "common/file_utils.h" |
| 41 | |
| 42 | #include <dirent.h> |
| 43 | #include <sys/stat.h> |
| 44 | |
| 45 | typedef struct |
| 46 | { |
| 47 | /* |
| 48 | * Our archive location. This is basically what the user specified as his |
| 49 | * backup file but of course here it is a directory. |
| 50 | */ |
| 51 | char *directory; |
| 52 | |
| 53 | cfp *dataFH; /* currently open data file */ |
| 54 | |
| 55 | cfp *blobsTocFH; /* file handle for blobs.toc */ |
| 56 | ParallelState *pstate; /* for parallel backup / restore */ |
| 57 | } lclContext; |
| 58 | |
| 59 | typedef struct |
| 60 | { |
| 61 | char *filename; /* filename excluding the directory (basename) */ |
| 62 | } lclTocEntry; |
| 63 | |
| 64 | /* prototypes for private functions */ |
| 65 | static void _ArchiveEntry(ArchiveHandle *AH, TocEntry *te); |
| 66 | static void _StartData(ArchiveHandle *AH, TocEntry *te); |
| 67 | static void _EndData(ArchiveHandle *AH, TocEntry *te); |
| 68 | static void _WriteData(ArchiveHandle *AH, const void *data, size_t dLen); |
| 69 | static int _WriteByte(ArchiveHandle *AH, const int i); |
| 70 | static int _ReadByte(ArchiveHandle *); |
| 71 | static void _WriteBuf(ArchiveHandle *AH, const void *buf, size_t len); |
| 72 | static void _ReadBuf(ArchiveHandle *AH, void *buf, size_t len); |
| 73 | static void _CloseArchive(ArchiveHandle *AH); |
| 74 | static void _ReopenArchive(ArchiveHandle *AH); |
| 75 | static void _PrintTocData(ArchiveHandle *AH, TocEntry *te); |
| 76 | |
| 77 | static void _WriteExtraToc(ArchiveHandle *AH, TocEntry *te); |
| 78 | static void _ReadExtraToc(ArchiveHandle *AH, TocEntry *te); |
| 79 | static void _PrintExtraToc(ArchiveHandle *AH, TocEntry *te); |
| 80 | |
| 81 | static void _StartBlobs(ArchiveHandle *AH, TocEntry *te); |
| 82 | static void _StartBlob(ArchiveHandle *AH, TocEntry *te, Oid oid); |
| 83 | static void _EndBlob(ArchiveHandle *AH, TocEntry *te, Oid oid); |
| 84 | static void _EndBlobs(ArchiveHandle *AH, TocEntry *te); |
| 85 | static void _LoadBlobs(ArchiveHandle *AH); |
| 86 | |
| 87 | static void _PrepParallelRestore(ArchiveHandle *AH); |
| 88 | static void _Clone(ArchiveHandle *AH); |
| 89 | static void _DeClone(ArchiveHandle *AH); |
| 90 | |
| 91 | static int _WorkerJobRestoreDirectory(ArchiveHandle *AH, TocEntry *te); |
| 92 | static int _WorkerJobDumpDirectory(ArchiveHandle *AH, TocEntry *te); |
| 93 | |
| 94 | static void setFilePath(ArchiveHandle *AH, char *buf, |
| 95 | const char *relativeFilename); |
| 96 | |
| 97 | /* |
| 98 | * Init routine required by ALL formats. This is a global routine |
| 99 | * and should be declared in pg_backup_archiver.h |
| 100 | * |
| 101 | * Its task is to create any extra archive context (using AH->formatData), |
| 102 | * and to initialize the supported function pointers. |
| 103 | * |
| 104 | * It should also prepare whatever its input source is for reading/writing, |
| 105 | * and in the case of a read mode connection, it should load the Header & TOC. |
| 106 | */ |
| 107 | void |
| 108 | InitArchiveFmt_Directory(ArchiveHandle *AH) |
| 109 | { |
| 110 | lclContext *ctx; |
| 111 | |
| 112 | /* Assuming static functions, this can be copied for each format. */ |
| 113 | AH->ArchiveEntryPtr = _ArchiveEntry; |
| 114 | AH->StartDataPtr = _StartData; |
| 115 | AH->WriteDataPtr = _WriteData; |
| 116 | AH->EndDataPtr = _EndData; |
| 117 | AH->WriteBytePtr = _WriteByte; |
| 118 | AH->ReadBytePtr = _ReadByte; |
| 119 | AH->WriteBufPtr = _WriteBuf; |
| 120 | AH->ReadBufPtr = _ReadBuf; |
| 121 | AH->ClosePtr = _CloseArchive; |
| 122 | AH->ReopenPtr = _ReopenArchive; |
| 123 | AH->PrintTocDataPtr = _PrintTocData; |
| 124 | AH->ReadExtraTocPtr = _ReadExtraToc; |
| 125 | AH->WriteExtraTocPtr = _WriteExtraToc; |
| 126 | AH->PrintExtraTocPtr = _PrintExtraToc; |
| 127 | |
| 128 | AH->StartBlobsPtr = _StartBlobs; |
| 129 | AH->StartBlobPtr = _StartBlob; |
| 130 | AH->EndBlobPtr = _EndBlob; |
| 131 | AH->EndBlobsPtr = _EndBlobs; |
| 132 | |
| 133 | AH->PrepParallelRestorePtr = _PrepParallelRestore; |
| 134 | AH->ClonePtr = _Clone; |
| 135 | AH->DeClonePtr = _DeClone; |
| 136 | |
| 137 | AH->WorkerJobRestorePtr = _WorkerJobRestoreDirectory; |
| 138 | AH->WorkerJobDumpPtr = _WorkerJobDumpDirectory; |
| 139 | |
| 140 | /* Set up our private context */ |
| 141 | ctx = (lclContext *) pg_malloc0(sizeof(lclContext)); |
| 142 | AH->formatData = (void *) ctx; |
| 143 | |
| 144 | ctx->dataFH = NULL; |
| 145 | ctx->blobsTocFH = NULL; |
| 146 | |
| 147 | /* Initialize LO buffering */ |
| 148 | AH->lo_buf_size = LOBBUFSIZE; |
| 149 | AH->lo_buf = (void *) pg_malloc(LOBBUFSIZE); |
| 150 | |
| 151 | /* |
| 152 | * Now open the TOC file |
| 153 | */ |
| 154 | |
| 155 | if (!AH->fSpec || strcmp(AH->fSpec, "" ) == 0) |
| 156 | fatal("no output directory specified" ); |
| 157 | |
| 158 | ctx->directory = AH->fSpec; |
| 159 | |
| 160 | if (AH->mode == archModeWrite) |
| 161 | { |
| 162 | struct stat st; |
| 163 | bool is_empty = false; |
| 164 | |
| 165 | /* we accept an empty existing directory */ |
| 166 | if (stat(ctx->directory, &st) == 0 && S_ISDIR(st.st_mode)) |
| 167 | { |
| 168 | DIR *dir = opendir(ctx->directory); |
| 169 | |
| 170 | if (dir) |
| 171 | { |
| 172 | struct dirent *d; |
| 173 | |
| 174 | is_empty = true; |
| 175 | while (errno = 0, (d = readdir(dir))) |
| 176 | { |
| 177 | if (strcmp(d->d_name, "." ) != 0 && strcmp(d->d_name, ".." ) != 0) |
| 178 | { |
| 179 | is_empty = false; |
| 180 | break; |
| 181 | } |
| 182 | } |
| 183 | |
| 184 | if (errno) |
| 185 | fatal("could not read directory \"%s\": %m" , |
| 186 | ctx->directory); |
| 187 | |
| 188 | if (closedir(dir)) |
| 189 | fatal("could not close directory \"%s\": %m" , |
| 190 | ctx->directory); |
| 191 | } |
| 192 | } |
| 193 | |
| 194 | if (!is_empty && mkdir(ctx->directory, 0700) < 0) |
| 195 | fatal("could not create directory \"%s\": %m" , |
| 196 | ctx->directory); |
| 197 | } |
| 198 | else |
| 199 | { /* Read Mode */ |
| 200 | char fname[MAXPGPATH]; |
| 201 | cfp *tocFH; |
| 202 | |
| 203 | setFilePath(AH, fname, "toc.dat" ); |
| 204 | |
| 205 | tocFH = cfopen_read(fname, PG_BINARY_R); |
| 206 | if (tocFH == NULL) |
| 207 | fatal("could not open input file \"%s\": %m" , fname); |
| 208 | |
| 209 | ctx->dataFH = tocFH; |
| 210 | |
| 211 | /* |
| 212 | * The TOC of a directory format dump shares the format code of the |
| 213 | * tar format. |
| 214 | */ |
| 215 | AH->format = archTar; |
| 216 | ReadHead(AH); |
| 217 | AH->format = archDirectory; |
| 218 | ReadToc(AH); |
| 219 | |
| 220 | /* Nothing else in the file, so close it again... */ |
| 221 | if (cfclose(tocFH) != 0) |
| 222 | fatal("could not close TOC file: %m" ); |
| 223 | ctx->dataFH = NULL; |
| 224 | } |
| 225 | } |
| 226 | |
| 227 | /* |
| 228 | * Called by the Archiver when the dumper creates a new TOC entry. |
| 229 | * |
| 230 | * We determine the filename for this entry. |
| 231 | */ |
| 232 | static void |
| 233 | _ArchiveEntry(ArchiveHandle *AH, TocEntry *te) |
| 234 | { |
| 235 | lclTocEntry *tctx; |
| 236 | char fn[MAXPGPATH]; |
| 237 | |
| 238 | tctx = (lclTocEntry *) pg_malloc0(sizeof(lclTocEntry)); |
| 239 | if (strcmp(te->desc, "BLOBS" ) == 0) |
| 240 | tctx->filename = pg_strdup("blobs.toc" ); |
| 241 | else if (te->dataDumper) |
| 242 | { |
| 243 | snprintf(fn, MAXPGPATH, "%d.dat" , te->dumpId); |
| 244 | tctx->filename = pg_strdup(fn); |
| 245 | } |
| 246 | else |
| 247 | tctx->filename = NULL; |
| 248 | |
| 249 | te->formatData = (void *) tctx; |
| 250 | } |
| 251 | |
| 252 | /* |
| 253 | * Called by the Archiver to save any extra format-related TOC entry |
| 254 | * data. |
| 255 | * |
| 256 | * Use the Archiver routines to write data - they are non-endian, and |
| 257 | * maintain other important file information. |
| 258 | */ |
| 259 | static void |
| 260 | (ArchiveHandle *AH, TocEntry *te) |
| 261 | { |
| 262 | lclTocEntry *tctx = (lclTocEntry *) te->formatData; |
| 263 | |
| 264 | /* |
| 265 | * A dumpable object has set tctx->filename, any other object has not. |
| 266 | * (see _ArchiveEntry). |
| 267 | */ |
| 268 | if (tctx->filename) |
| 269 | WriteStr(AH, tctx->filename); |
| 270 | else |
| 271 | WriteStr(AH, "" ); |
| 272 | } |
| 273 | |
| 274 | /* |
| 275 | * Called by the Archiver to read any extra format-related TOC data. |
| 276 | * |
| 277 | * Needs to match the order defined in _WriteExtraToc, and should also |
| 278 | * use the Archiver input routines. |
| 279 | */ |
| 280 | static void |
| 281 | (ArchiveHandle *AH, TocEntry *te) |
| 282 | { |
| 283 | lclTocEntry *tctx = (lclTocEntry *) te->formatData; |
| 284 | |
| 285 | if (tctx == NULL) |
| 286 | { |
| 287 | tctx = (lclTocEntry *) pg_malloc0(sizeof(lclTocEntry)); |
| 288 | te->formatData = (void *) tctx; |
| 289 | } |
| 290 | |
| 291 | tctx->filename = ReadStr(AH); |
| 292 | if (strlen(tctx->filename) == 0) |
| 293 | { |
| 294 | free(tctx->filename); |
| 295 | tctx->filename = NULL; |
| 296 | } |
| 297 | } |
| 298 | |
| 299 | /* |
| 300 | * Called by the Archiver when restoring an archive to output a comment |
| 301 | * that includes useful information about the TOC entry. |
| 302 | */ |
| 303 | static void |
| 304 | (ArchiveHandle *AH, TocEntry *te) |
| 305 | { |
| 306 | lclTocEntry *tctx = (lclTocEntry *) te->formatData; |
| 307 | |
| 308 | if (AH->public.verbose && tctx->filename) |
| 309 | ahprintf(AH, "-- File: %s\n" , tctx->filename); |
| 310 | } |
| 311 | |
| 312 | /* |
| 313 | * Called by the archiver when saving TABLE DATA (not schema). This routine |
| 314 | * should save whatever format-specific information is needed to read |
| 315 | * the archive back. |
| 316 | * |
| 317 | * It is called just prior to the dumper's 'DataDumper' routine being called. |
| 318 | * |
| 319 | * We create the data file for writing. |
| 320 | */ |
| 321 | static void |
| 322 | _StartData(ArchiveHandle *AH, TocEntry *te) |
| 323 | { |
| 324 | lclTocEntry *tctx = (lclTocEntry *) te->formatData; |
| 325 | lclContext *ctx = (lclContext *) AH->formatData; |
| 326 | char fname[MAXPGPATH]; |
| 327 | |
| 328 | setFilePath(AH, fname, tctx->filename); |
| 329 | |
| 330 | ctx->dataFH = cfopen_write(fname, PG_BINARY_W, AH->compression); |
| 331 | if (ctx->dataFH == NULL) |
| 332 | fatal("could not open output file \"%s\": %m" , fname); |
| 333 | } |
| 334 | |
| 335 | /* |
| 336 | * Called by archiver when dumper calls WriteData. This routine is |
| 337 | * called for both BLOB and TABLE data; it is the responsibility of |
| 338 | * the format to manage each kind of data using StartBlob/StartData. |
| 339 | * |
| 340 | * It should only be called from within a DataDumper routine. |
| 341 | * |
| 342 | * We write the data to the open data file. |
| 343 | */ |
| 344 | static void |
| 345 | _WriteData(ArchiveHandle *AH, const void *data, size_t dLen) |
| 346 | { |
| 347 | lclContext *ctx = (lclContext *) AH->formatData; |
| 348 | |
| 349 | if (dLen > 0 && cfwrite(data, dLen, ctx->dataFH) != dLen) |
| 350 | fatal("could not write to output file: %s" , |
| 351 | get_cfp_error(ctx->dataFH)); |
| 352 | |
| 353 | |
| 354 | return; |
| 355 | } |
| 356 | |
| 357 | /* |
| 358 | * Called by the archiver when a dumper's 'DataDumper' routine has |
| 359 | * finished. |
| 360 | * |
| 361 | * We close the data file. |
| 362 | */ |
| 363 | static void |
| 364 | _EndData(ArchiveHandle *AH, TocEntry *te) |
| 365 | { |
| 366 | lclContext *ctx = (lclContext *) AH->formatData; |
| 367 | |
| 368 | /* Close the file */ |
| 369 | cfclose(ctx->dataFH); |
| 370 | |
| 371 | ctx->dataFH = NULL; |
| 372 | } |
| 373 | |
| 374 | /* |
| 375 | * Print data for a given file (can be a BLOB as well) |
| 376 | */ |
| 377 | static void |
| 378 | _PrintFileData(ArchiveHandle *AH, char *filename) |
| 379 | { |
| 380 | size_t cnt; |
| 381 | char *buf; |
| 382 | size_t buflen; |
| 383 | cfp *cfp; |
| 384 | |
| 385 | if (!filename) |
| 386 | return; |
| 387 | |
| 388 | cfp = cfopen_read(filename, PG_BINARY_R); |
| 389 | |
| 390 | if (!cfp) |
| 391 | fatal("could not open input file \"%s\": %m" , filename); |
| 392 | |
| 393 | buf = pg_malloc(ZLIB_OUT_SIZE); |
| 394 | buflen = ZLIB_OUT_SIZE; |
| 395 | |
| 396 | while ((cnt = cfread(buf, buflen, cfp))) |
| 397 | { |
| 398 | ahwrite(buf, 1, cnt, AH); |
| 399 | } |
| 400 | |
| 401 | free(buf); |
| 402 | if (cfclose(cfp) !=0) |
| 403 | fatal("could not close data file: %m" ); |
| 404 | } |
| 405 | |
| 406 | /* |
| 407 | * Print data for a given TOC entry |
| 408 | */ |
| 409 | static void |
| 410 | _PrintTocData(ArchiveHandle *AH, TocEntry *te) |
| 411 | { |
| 412 | lclTocEntry *tctx = (lclTocEntry *) te->formatData; |
| 413 | |
| 414 | if (!tctx->filename) |
| 415 | return; |
| 416 | |
| 417 | if (strcmp(te->desc, "BLOBS" ) == 0) |
| 418 | _LoadBlobs(AH); |
| 419 | else |
| 420 | { |
| 421 | char fname[MAXPGPATH]; |
| 422 | |
| 423 | setFilePath(AH, fname, tctx->filename); |
| 424 | _PrintFileData(AH, fname); |
| 425 | } |
| 426 | } |
| 427 | |
| 428 | static void |
| 429 | _LoadBlobs(ArchiveHandle *AH) |
| 430 | { |
| 431 | Oid oid; |
| 432 | lclContext *ctx = (lclContext *) AH->formatData; |
| 433 | char fname[MAXPGPATH]; |
| 434 | char line[MAXPGPATH]; |
| 435 | |
| 436 | StartRestoreBlobs(AH); |
| 437 | |
| 438 | setFilePath(AH, fname, "blobs.toc" ); |
| 439 | |
| 440 | ctx->blobsTocFH = cfopen_read(fname, PG_BINARY_R); |
| 441 | |
| 442 | if (ctx->blobsTocFH == NULL) |
| 443 | fatal("could not open large object TOC file \"%s\" for input: %m" , |
| 444 | fname); |
| 445 | |
| 446 | /* Read the blobs TOC file line-by-line, and process each blob */ |
| 447 | while ((cfgets(ctx->blobsTocFH, line, MAXPGPATH)) != NULL) |
| 448 | { |
| 449 | char fname[MAXPGPATH]; |
| 450 | char path[MAXPGPATH]; |
| 451 | |
| 452 | /* Can't overflow because line and fname are the same length. */ |
| 453 | if (sscanf(line, "%u %s\n" , &oid, fname) != 2) |
| 454 | fatal("invalid line in large object TOC file \"%s\": \"%s\"" , |
| 455 | fname, line); |
| 456 | |
| 457 | StartRestoreBlob(AH, oid, AH->public.ropt->dropSchema); |
| 458 | snprintf(path, MAXPGPATH, "%s/%s" , ctx->directory, fname); |
| 459 | _PrintFileData(AH, path); |
| 460 | EndRestoreBlob(AH, oid); |
| 461 | } |
| 462 | if (!cfeof(ctx->blobsTocFH)) |
| 463 | fatal("error reading large object TOC file \"%s\"" , |
| 464 | fname); |
| 465 | |
| 466 | if (cfclose(ctx->blobsTocFH) != 0) |
| 467 | fatal("could not close large object TOC file \"%s\": %m" , |
| 468 | fname); |
| 469 | |
| 470 | ctx->blobsTocFH = NULL; |
| 471 | |
| 472 | EndRestoreBlobs(AH); |
| 473 | } |
| 474 | |
| 475 | |
| 476 | /* |
| 477 | * Write a byte of data to the archive. |
| 478 | * Called by the archiver to do integer & byte output to the archive. |
| 479 | * These routines are only used to read & write the headers & TOC. |
| 480 | */ |
| 481 | static int |
| 482 | _WriteByte(ArchiveHandle *AH, const int i) |
| 483 | { |
| 484 | unsigned char c = (unsigned char) i; |
| 485 | lclContext *ctx = (lclContext *) AH->formatData; |
| 486 | |
| 487 | if (cfwrite(&c, 1, ctx->dataFH) != 1) |
| 488 | fatal("could not write to output file: %s" , |
| 489 | get_cfp_error(ctx->dataFH)); |
| 490 | |
| 491 | return 1; |
| 492 | } |
| 493 | |
| 494 | /* |
| 495 | * Read a byte of data from the archive. |
| 496 | * Called by the archiver to read bytes & integers from the archive. |
| 497 | * These routines are only used to read & write headers & TOC. |
| 498 | * EOF should be treated as a fatal error. |
| 499 | */ |
| 500 | static int |
| 501 | _ReadByte(ArchiveHandle *AH) |
| 502 | { |
| 503 | lclContext *ctx = (lclContext *) AH->formatData; |
| 504 | |
| 505 | return cfgetc(ctx->dataFH); |
| 506 | } |
| 507 | |
| 508 | /* |
| 509 | * Write a buffer of data to the archive. |
| 510 | * Called by the archiver to write a block of bytes to the TOC or a data file. |
| 511 | */ |
| 512 | static void |
| 513 | _WriteBuf(ArchiveHandle *AH, const void *buf, size_t len) |
| 514 | { |
| 515 | lclContext *ctx = (lclContext *) AH->formatData; |
| 516 | |
| 517 | if (cfwrite(buf, len, ctx->dataFH) != len) |
| 518 | fatal("could not write to output file: %s" , |
| 519 | get_cfp_error(ctx->dataFH)); |
| 520 | |
| 521 | return; |
| 522 | } |
| 523 | |
| 524 | /* |
| 525 | * Read a block of bytes from the archive. |
| 526 | * |
| 527 | * Called by the archiver to read a block of bytes from the archive |
| 528 | */ |
| 529 | static void |
| 530 | _ReadBuf(ArchiveHandle *AH, void *buf, size_t len) |
| 531 | { |
| 532 | lclContext *ctx = (lclContext *) AH->formatData; |
| 533 | |
| 534 | /* |
| 535 | * If there was an I/O error, we already exited in cfread(), so here we |
| 536 | * exit on short reads. |
| 537 | */ |
| 538 | if (cfread(buf, len, ctx->dataFH) != len) |
| 539 | fatal("could not read from input file: end of file" ); |
| 540 | |
| 541 | return; |
| 542 | } |
| 543 | |
| 544 | /* |
| 545 | * Close the archive. |
| 546 | * |
| 547 | * When writing the archive, this is the routine that actually starts |
| 548 | * the process of saving it to files. No data should be written prior |
| 549 | * to this point, since the user could sort the TOC after creating it. |
| 550 | * |
| 551 | * If an archive is to be written, this routine must call: |
| 552 | * WriteHead to save the archive header |
| 553 | * WriteToc to save the TOC entries |
| 554 | * WriteDataChunks to save all DATA & BLOBs. |
| 555 | */ |
| 556 | static void |
| 557 | _CloseArchive(ArchiveHandle *AH) |
| 558 | { |
| 559 | lclContext *ctx = (lclContext *) AH->formatData; |
| 560 | |
| 561 | if (AH->mode == archModeWrite) |
| 562 | { |
| 563 | cfp *tocFH; |
| 564 | char fname[MAXPGPATH]; |
| 565 | |
| 566 | setFilePath(AH, fname, "toc.dat" ); |
| 567 | |
| 568 | /* this will actually fork the processes for a parallel backup */ |
| 569 | ctx->pstate = ParallelBackupStart(AH); |
| 570 | |
| 571 | /* The TOC is always created uncompressed */ |
| 572 | tocFH = cfopen_write(fname, PG_BINARY_W, 0); |
| 573 | if (tocFH == NULL) |
| 574 | fatal("could not open output file \"%s\": %m" , fname); |
| 575 | ctx->dataFH = tocFH; |
| 576 | |
| 577 | /* |
| 578 | * Write 'tar' in the format field of the toc.dat file. The directory |
| 579 | * is compatible with 'tar', so there's no point having a different |
| 580 | * format code for it. |
| 581 | */ |
| 582 | AH->format = archTar; |
| 583 | WriteHead(AH); |
| 584 | AH->format = archDirectory; |
| 585 | WriteToc(AH); |
| 586 | if (cfclose(tocFH) != 0) |
| 587 | fatal("could not close TOC file: %m" ); |
| 588 | WriteDataChunks(AH, ctx->pstate); |
| 589 | |
| 590 | ParallelBackupEnd(AH, ctx->pstate); |
| 591 | |
| 592 | /* |
| 593 | * In directory mode, there is no need to sync all the entries |
| 594 | * individually. Just recurse once through all the files generated. |
| 595 | */ |
| 596 | if (AH->dosync) |
| 597 | fsync_dir_recurse(ctx->directory); |
| 598 | } |
| 599 | AH->FH = NULL; |
| 600 | } |
| 601 | |
| 602 | /* |
| 603 | * Reopen the archive's file handle. |
| 604 | */ |
| 605 | static void |
| 606 | _ReopenArchive(ArchiveHandle *AH) |
| 607 | { |
| 608 | /* |
| 609 | * Our TOC is in memory, our data files are opened by each child anyway as |
| 610 | * they are separate. We support reopening the archive by just doing |
| 611 | * nothing. |
| 612 | */ |
| 613 | } |
| 614 | |
| 615 | /* |
| 616 | * BLOB support |
| 617 | */ |
| 618 | |
| 619 | /* |
| 620 | * Called by the archiver when starting to save all BLOB DATA (not schema). |
| 621 | * It is called just prior to the dumper's DataDumper routine. |
| 622 | * |
| 623 | * We open the large object TOC file here, so that we can append a line to |
| 624 | * it for each blob. |
| 625 | */ |
| 626 | static void |
| 627 | _StartBlobs(ArchiveHandle *AH, TocEntry *te) |
| 628 | { |
| 629 | lclContext *ctx = (lclContext *) AH->formatData; |
| 630 | char fname[MAXPGPATH]; |
| 631 | |
| 632 | setFilePath(AH, fname, "blobs.toc" ); |
| 633 | |
| 634 | /* The blob TOC file is never compressed */ |
| 635 | ctx->blobsTocFH = cfopen_write(fname, "ab" , 0); |
| 636 | if (ctx->blobsTocFH == NULL) |
| 637 | fatal("could not open output file \"%s\": %m" , fname); |
| 638 | } |
| 639 | |
| 640 | /* |
| 641 | * Called by the archiver when we're about to start dumping a blob. |
| 642 | * |
| 643 | * We create a file to write the blob to. |
| 644 | */ |
| 645 | static void |
| 646 | _StartBlob(ArchiveHandle *AH, TocEntry *te, Oid oid) |
| 647 | { |
| 648 | lclContext *ctx = (lclContext *) AH->formatData; |
| 649 | char fname[MAXPGPATH]; |
| 650 | |
| 651 | snprintf(fname, MAXPGPATH, "%s/blob_%u.dat" , ctx->directory, oid); |
| 652 | |
| 653 | ctx->dataFH = cfopen_write(fname, PG_BINARY_W, AH->compression); |
| 654 | |
| 655 | if (ctx->dataFH == NULL) |
| 656 | fatal("could not open output file \"%s\": %m" , fname); |
| 657 | } |
| 658 | |
| 659 | /* |
| 660 | * Called by the archiver when the dumper is finished writing a blob. |
| 661 | * |
| 662 | * We close the blob file and write an entry to the blob TOC file for it. |
| 663 | */ |
| 664 | static void |
| 665 | _EndBlob(ArchiveHandle *AH, TocEntry *te, Oid oid) |
| 666 | { |
| 667 | lclContext *ctx = (lclContext *) AH->formatData; |
| 668 | char buf[50]; |
| 669 | int len; |
| 670 | |
| 671 | /* Close the BLOB data file itself */ |
| 672 | cfclose(ctx->dataFH); |
| 673 | ctx->dataFH = NULL; |
| 674 | |
| 675 | /* register the blob in blobs.toc */ |
| 676 | len = snprintf(buf, sizeof(buf), "%u blob_%u.dat\n" , oid, oid); |
| 677 | if (cfwrite(buf, len, ctx->blobsTocFH) != len) |
| 678 | fatal("could not write to blobs TOC file" ); |
| 679 | } |
| 680 | |
| 681 | /* |
| 682 | * Called by the archiver when finishing saving all BLOB DATA. |
| 683 | * |
| 684 | * We close the blobs TOC file. |
| 685 | */ |
| 686 | static void |
| 687 | _EndBlobs(ArchiveHandle *AH, TocEntry *te) |
| 688 | { |
| 689 | lclContext *ctx = (lclContext *) AH->formatData; |
| 690 | |
| 691 | cfclose(ctx->blobsTocFH); |
| 692 | ctx->blobsTocFH = NULL; |
| 693 | } |
| 694 | |
| 695 | /* |
| 696 | * Gets a relative file name and prepends the output directory, writing the |
| 697 | * result to buf. The caller needs to make sure that buf is MAXPGPATH bytes |
| 698 | * big. Can't use a static char[MAXPGPATH] inside the function because we run |
| 699 | * multithreaded on Windows. |
| 700 | */ |
| 701 | static void |
| 702 | setFilePath(ArchiveHandle *AH, char *buf, const char *relativeFilename) |
| 703 | { |
| 704 | lclContext *ctx = (lclContext *) AH->formatData; |
| 705 | char *dname; |
| 706 | |
| 707 | dname = ctx->directory; |
| 708 | |
| 709 | if (strlen(dname) + 1 + strlen(relativeFilename) + 1 > MAXPGPATH) |
| 710 | fatal("file name too long: \"%s\"" , dname); |
| 711 | |
| 712 | strcpy(buf, dname); |
| 713 | strcat(buf, "/" ); |
| 714 | strcat(buf, relativeFilename); |
| 715 | } |
| 716 | |
| 717 | /* |
| 718 | * Prepare for parallel restore. |
| 719 | * |
| 720 | * The main thing that needs to happen here is to fill in TABLE DATA and BLOBS |
| 721 | * TOC entries' dataLength fields with appropriate values to guide the |
| 722 | * ordering of restore jobs. The source of said data is format-dependent, |
| 723 | * as is the exact meaning of the values. |
| 724 | * |
| 725 | * A format module might also choose to do other setup here. |
| 726 | */ |
| 727 | static void |
| 728 | _PrepParallelRestore(ArchiveHandle *AH) |
| 729 | { |
| 730 | TocEntry *te; |
| 731 | |
| 732 | for (te = AH->toc->next; te != AH->toc; te = te->next) |
| 733 | { |
| 734 | lclTocEntry *tctx = (lclTocEntry *) te->formatData; |
| 735 | char fname[MAXPGPATH]; |
| 736 | struct stat st; |
| 737 | |
| 738 | /* |
| 739 | * A dumpable object has set tctx->filename, any other object has not. |
| 740 | * (see _ArchiveEntry). |
| 741 | */ |
| 742 | if (tctx->filename == NULL) |
| 743 | continue; |
| 744 | |
| 745 | /* We may ignore items not due to be restored */ |
| 746 | if ((te->reqs & REQ_DATA) == 0) |
| 747 | continue; |
| 748 | |
| 749 | /* |
| 750 | * Stat the file and, if successful, put its size in dataLength. When |
| 751 | * using compression, the physical file size might not be a very good |
| 752 | * guide to the amount of work involved in restoring the file, but we |
| 753 | * only need an approximate indicator of that. |
| 754 | */ |
| 755 | setFilePath(AH, fname, tctx->filename); |
| 756 | |
| 757 | if (stat(fname, &st) == 0) |
| 758 | te->dataLength = st.st_size; |
| 759 | else |
| 760 | { |
| 761 | /* It might be compressed */ |
| 762 | strlcat(fname, ".gz" , sizeof(fname)); |
| 763 | if (stat(fname, &st) == 0) |
| 764 | te->dataLength = st.st_size; |
| 765 | } |
| 766 | |
| 767 | /* |
| 768 | * If this is the BLOBS entry, what we stat'd was blobs.toc, which |
| 769 | * most likely is a lot smaller than the actual blob data. We don't |
| 770 | * have a cheap way to estimate how much smaller, but fortunately it |
| 771 | * doesn't matter too much as long as we get the blobs processed |
| 772 | * reasonably early. Arbitrarily scale up by a factor of 1K. |
| 773 | */ |
| 774 | if (strcmp(te->desc, "BLOBS" ) == 0) |
| 775 | te->dataLength *= 1024; |
| 776 | } |
| 777 | } |
| 778 | |
| 779 | /* |
| 780 | * Clone format-specific fields during parallel restoration. |
| 781 | */ |
| 782 | static void |
| 783 | _Clone(ArchiveHandle *AH) |
| 784 | { |
| 785 | lclContext *ctx = (lclContext *) AH->formatData; |
| 786 | |
| 787 | AH->formatData = (lclContext *) pg_malloc(sizeof(lclContext)); |
| 788 | memcpy(AH->formatData, ctx, sizeof(lclContext)); |
| 789 | ctx = (lclContext *) AH->formatData; |
| 790 | |
| 791 | /* |
| 792 | * Note: we do not make a local lo_buf because we expect at most one BLOBS |
| 793 | * entry per archive, so no parallelism is possible. Likewise, |
| 794 | * TOC-entry-local state isn't an issue because any one TOC entry is |
| 795 | * touched by just one worker child. |
| 796 | */ |
| 797 | |
| 798 | /* |
| 799 | * We also don't copy the ParallelState pointer (pstate), only the master |
| 800 | * process ever writes to it. |
| 801 | */ |
| 802 | } |
| 803 | |
| 804 | static void |
| 805 | _DeClone(ArchiveHandle *AH) |
| 806 | { |
| 807 | lclContext *ctx = (lclContext *) AH->formatData; |
| 808 | |
| 809 | free(ctx); |
| 810 | } |
| 811 | |
| 812 | /* |
| 813 | * This function is executed in the child of a parallel backup for a |
| 814 | * directory-format archive and dumps the actual data for one TOC entry. |
| 815 | */ |
| 816 | static int |
| 817 | _WorkerJobDumpDirectory(ArchiveHandle *AH, TocEntry *te) |
| 818 | { |
| 819 | /* |
| 820 | * This function returns void. We either fail and die horribly or |
| 821 | * succeed... A failure will be detected by the parent when the child dies |
| 822 | * unexpectedly. |
| 823 | */ |
| 824 | WriteDataChunksForTocEntry(AH, te); |
| 825 | |
| 826 | return 0; |
| 827 | } |
| 828 | |
| 829 | /* |
| 830 | * This function is executed in the child of a parallel restore from a |
| 831 | * directory-format archive and restores the actual data for one TOC entry. |
| 832 | */ |
| 833 | static int |
| 834 | _WorkerJobRestoreDirectory(ArchiveHandle *AH, TocEntry *te) |
| 835 | { |
| 836 | return parallel_restore(AH, te); |
| 837 | } |
| 838 | |