journal.c revision 1.6
1 /* $NetBSD: journal.c,v 1.6 2021/02/19 16:42:16 christos Exp $ */ 2 3 /* 4 * Copyright (C) Internet Systems Consortium, Inc. ("ISC") 5 * 6 * This Source Code Form is subject to the terms of the Mozilla Public 7 * License, v. 2.0. If a copy of the MPL was not distributed with this 8 * file, you can obtain one at https://mozilla.org/MPL/2.0/. 9 * 10 * See the COPYRIGHT file distributed with this work for additional 11 * information regarding copyright ownership. 12 */ 13 14 #include <errno.h> 15 #include <inttypes.h> 16 #include <stdbool.h> 17 #include <stdlib.h> 18 #include <unistd.h> 19 20 #include <isc/file.h> 21 #include <isc/mem.h> 22 #include <isc/print.h> 23 #include <isc/stdio.h> 24 #include <isc/string.h> 25 #include <isc/util.h> 26 27 #include <dns/compress.h> 28 #include <dns/db.h> 29 #include <dns/dbiterator.h> 30 #include <dns/diff.h> 31 #include <dns/fixedname.h> 32 #include <dns/journal.h> 33 #include <dns/log.h> 34 #include <dns/rdataset.h> 35 #include <dns/rdatasetiter.h> 36 #include <dns/result.h> 37 #include <dns/soa.h> 38 39 /*! \file 40 * \brief Journaling. 41 * 42 * A journal file consists of 43 * 44 * \li A fixed-size header of type journal_rawheader_t. 45 * 46 * \li The index. This is an unordered array of index entries 47 * of type journal_rawpos_t giving the locations 48 * of some arbitrary subset of the journal's addressable 49 * transactions. The index entries are used as hints to 50 * speed up the process of locating a transaction with a given 51 * serial number. Unused index entries have an "offset" 52 * field of zero. The size of the index can vary between 53 * journal files, but does not change during the lifetime 54 * of a file. The size can be zero. 55 * 56 * \li The journal data. This consists of one or more transactions. 57 * Each transaction begins with a transaction header of type 58 * journal_rawxhdr_t. The transaction header is followed by a 59 * sequence of RRs, similar in structure to an IXFR difference 60 * sequence (RFC1995). That is, the pre-transaction SOA, 61 * zero or more other deleted RRs, the post-transaction SOA, 62 * and zero or more other added RRs. Unlike in IXFR, each RR 63 * is prefixed with a 32-bit length. 64 * 65 * The journal data part grows as new transactions are 66 * appended to the file. Only those transactions 67 * whose serial number is current-(2^31-1) to current 68 * are considered "addressable" and may be pointed 69 * to from the header or index. They may be preceded 70 * by old transactions that are no longer addressable, 71 * and they may be followed by transactions that were 72 * appended to the journal but never committed by updating 73 * the "end" position in the header. The latter will 74 * be overwritten when new transactions are added. 75 */ 76 77 /**************************************************************************/ 78 /* 79 * Miscellaneous utilities. 80 */ 81 82 #define JOURNAL_COMMON_LOGARGS \ 83 dns_lctx, DNS_LOGCATEGORY_GENERAL, DNS_LOGMODULE_JOURNAL 84 85 #define JOURNAL_DEBUG_LOGARGS(n) JOURNAL_COMMON_LOGARGS, ISC_LOG_DEBUG(n) 86 87 /*% 88 * It would be non-sensical (or at least obtuse) to use FAIL() with an 89 * ISC_R_SUCCESS code, but the test is there to keep the Solaris compiler 90 * from complaining about "end-of-loop code not reached". 91 */ 92 #define FAIL(code) \ 93 do { \ 94 result = (code); \ 95 if (result != ISC_R_SUCCESS) \ 96 goto failure; \ 97 } while (/*CONSTCOND*/0) 98 99 #define CHECK(op) \ 100 do { \ 101 result = (op); \ 102 if (result != ISC_R_SUCCESS) \ 103 goto failure; \ 104 } while (/*CONSTCOND*/0) 105 106 #define JOURNAL_SERIALSET 0x01U 107 108 static isc_result_t 109 index_to_disk(dns_journal_t *); 110 111 static inline uint32_t 112 decode_uint32(unsigned char *p) { 113 return ((p[0] << 24) + (p[1] << 16) + (p[2] << 8) + (p[3] << 0)); 114 } 115 116 static inline void 117 encode_uint32(uint32_t val, unsigned char *p) { 118 p[0] = (uint8_t)(val >> 24); 119 p[1] = (uint8_t)(val >> 16); 120 p[2] = (uint8_t)(val >> 8); 121 p[3] = (uint8_t)(val >> 0); 122 } 123 124 isc_result_t 125 dns_db_createsoatuple(dns_db_t *db, dns_dbversion_t *ver, isc_mem_t *mctx, 126 dns_diffop_t op, dns_difftuple_t **tp) { 127 isc_result_t result; 128 dns_dbnode_t *node; 129 dns_rdataset_t rdataset; 130 dns_rdata_t rdata = DNS_RDATA_INIT; 131 dns_fixedname_t fixed; 132 dns_name_t *zonename; 133 134 zonename = dns_fixedname_initname(&fixed); 135 dns_name_copynf(dns_db_origin(db), zonename); 136 137 node = NULL; 138 result = dns_db_findnode(db, zonename, false, &node); 139 if (result != ISC_R_SUCCESS) { 140 goto nonode; 141 } 142 143 dns_rdataset_init(&rdataset); 144 result = dns_db_findrdataset(db, node, ver, dns_rdatatype_soa, 0, 145 (isc_stdtime_t)0, &rdataset, NULL); 146 if (result != ISC_R_SUCCESS) { 147 goto freenode; 148 } 149 150 result = dns_rdataset_first(&rdataset); 151 if (result != ISC_R_SUCCESS) { 152 goto freenode; 153 } 154 155 dns_rdataset_current(&rdataset, &rdata); 156 dns_rdataset_getownercase(&rdataset, zonename); 157 158 result = dns_difftuple_create(mctx, op, zonename, rdataset.ttl, &rdata, 159 tp); 160 161 dns_rdataset_disassociate(&rdataset); 162 dns_db_detachnode(db, &node); 163 return (result); 164 165 freenode: 166 dns_db_detachnode(db, &node); 167 nonode: 168 UNEXPECTED_ERROR(__FILE__, __LINE__, "missing SOA"); 169 return (result); 170 } 171 172 /* Journaling */ 173 174 /*% 175 * On-disk representation of a "pointer" to a journal entry. 176 * These are used in the journal header to locate the beginning 177 * and end of the journal, and in the journal index to locate 178 * other transactions. 179 */ 180 typedef struct { 181 unsigned char serial[4]; /*%< SOA serial before update. */ 182 /* 183 * XXXRTH Should offset be 8 bytes? 184 * XXXDCL ... probably, since isc_offset_t is 8 bytes on many OSs. 185 * XXXAG ... but we will not be able to seek >2G anyway on many 186 * platforms as long as we are using fseek() rather 187 * than lseek(). 188 */ 189 unsigned char offset[4]; /*%< Offset from beginning of file. */ 190 } journal_rawpos_t; 191 192 /*% 193 * The header is of a fixed size, with some spare room for future 194 * extensions. 195 */ 196 #define JOURNAL_HEADER_SIZE 64 /* Bytes. */ 197 198 /*% 199 * The on-disk representation of the journal header. 200 * All numbers are stored in big-endian order. 201 */ 202 typedef union { 203 struct { 204 /*% File format version ID. */ 205 unsigned char format[16]; 206 /*% Position of the first addressable transaction */ 207 journal_rawpos_t begin; 208 /*% Position of the next (yet nonexistent) transaction. */ 209 journal_rawpos_t end; 210 /*% Number of index entries following the header. */ 211 unsigned char index_size[4]; 212 /*% Source serial number. */ 213 unsigned char sourceserial[4]; 214 unsigned char flags; 215 } h; 216 /* Pad the header to a fixed size. */ 217 unsigned char pad[JOURNAL_HEADER_SIZE]; 218 } journal_rawheader_t; 219 220 /*% 221 * The on-disk representation of the transaction header. 222 * There is one of these at the beginning of each transaction. 223 */ 224 typedef struct { 225 unsigned char size[4]; /*%< In bytes, excluding header. */ 226 unsigned char count[4]; /*%< Number of records in transaction */ 227 unsigned char serial0[4]; /*%< SOA serial before update. */ 228 unsigned char serial1[4]; /*%< SOA serial after update. */ 229 } journal_rawxhdr_t; 230 231 /*% 232 * The on-disk representation of the RR header. 233 * There is one of these at the beginning of each RR. 234 */ 235 typedef struct { 236 unsigned char size[4]; /*%< In bytes, excluding header. */ 237 } journal_rawrrhdr_t; 238 239 /*% 240 * The in-core representation of the journal header. 241 */ 242 typedef struct { 243 uint32_t serial; 244 isc_offset_t offset; 245 } journal_pos_t; 246 247 #define POS_VALID(pos) ((pos).offset != 0) 248 #define POS_INVALIDATE(pos) ((pos).offset = 0, (pos).serial = 0) 249 250 typedef struct { 251 unsigned char format[16]; 252 journal_pos_t begin; 253 journal_pos_t end; 254 uint32_t index_size; 255 uint32_t sourceserial; 256 bool serialset; 257 } journal_header_t; 258 259 /*% 260 * The in-core representation of the transaction header. 261 */ 262 typedef struct { 263 uint32_t size; 264 uint32_t count; 265 uint32_t serial0; 266 uint32_t serial1; 267 } journal_xhdr_t; 268 269 /*% 270 * The in-core representation of the RR header. 271 */ 272 typedef struct { 273 uint32_t size; 274 } journal_rrhdr_t; 275 276 /*% 277 * Initial contents to store in the header of a newly created 278 * journal file. 279 * 280 * The header starts with the magic string ";BIND LOG V9\n" 281 * to identify the file as a BIND 9 journal file. An ASCII 282 * identification string is used rather than a binary magic 283 * number to be consistent with BIND 8 (BIND 8 journal files 284 * are ASCII text files). 285 */ 286 287 static journal_header_t initial_journal_header = { 288 ";BIND LOG V9\n", { 0, 0 }, { 0, 0 }, 0, 0, 0 289 }; 290 291 #define JOURNAL_EMPTY(h) ((h)->begin.offset == (h)->end.offset) 292 293 typedef enum { 294 JOURNAL_STATE_INVALID, 295 JOURNAL_STATE_READ, 296 JOURNAL_STATE_WRITE, 297 JOURNAL_STATE_TRANSACTION, 298 JOURNAL_STATE_INLINE 299 } journal_state_t; 300 301 struct dns_journal { 302 unsigned int magic; /*%< JOUR */ 303 isc_mem_t *mctx; /*%< Memory context */ 304 journal_state_t state; 305 char *filename; /*%< Journal file name */ 306 FILE *fp; /*%< File handle */ 307 isc_offset_t offset; /*%< Current file offset */ 308 journal_header_t header; /*%< In-core journal header */ 309 unsigned char *rawindex; /*%< In-core buffer for journal index 310 * in on-disk format */ 311 journal_pos_t *index; /*%< In-core journal index */ 312 313 /*% Current transaction state (when writing). */ 314 struct { 315 unsigned int n_soa; /*%< Number of SOAs seen */ 316 unsigned int n_rr; /*%< Number of RRs to write */ 317 journal_pos_t pos[2]; /*%< Begin/end position */ 318 } x; 319 320 /*% Iteration state (when reading). */ 321 struct { 322 /* These define the part of the journal we iterate over. */ 323 journal_pos_t bpos; /*%< Position before first, */ 324 journal_pos_t epos; /*%< and after last transaction */ 325 /* The rest is iterator state. */ 326 uint32_t current_serial; /*%< Current SOA serial 327 * */ 328 isc_buffer_t source; /*%< Data from disk */ 329 isc_buffer_t target; /*%< Data from _fromwire check */ 330 dns_decompress_t dctx; /*%< Dummy decompression ctx */ 331 dns_name_t name; /*%< Current domain name */ 332 dns_rdata_t rdata; /*%< Current rdata */ 333 uint32_t ttl; /*%< Current TTL */ 334 unsigned int xsize; /*%< Size of transaction data */ 335 unsigned int xpos; /*%< Current position in it */ 336 isc_result_t result; /*%< Result of last call */ 337 } it; 338 }; 339 340 #define DNS_JOURNAL_MAGIC ISC_MAGIC('J', 'O', 'U', 'R') 341 #define DNS_JOURNAL_VALID(t) ISC_MAGIC_VALID(t, DNS_JOURNAL_MAGIC) 342 343 static void 344 journal_pos_decode(journal_rawpos_t *raw, journal_pos_t *cooked) { 345 cooked->serial = decode_uint32(raw->serial); 346 cooked->offset = decode_uint32(raw->offset); 347 } 348 349 static void 350 journal_pos_encode(journal_rawpos_t *raw, journal_pos_t *cooked) { 351 encode_uint32(cooked->serial, raw->serial); 352 encode_uint32(cooked->offset, raw->offset); 353 } 354 355 static void 356 journal_header_decode(journal_rawheader_t *raw, journal_header_t *cooked) { 357 INSIST(sizeof(cooked->format) == sizeof(raw->h.format)); 358 memmove(cooked->format, raw->h.format, sizeof(cooked->format)); 359 journal_pos_decode(&raw->h.begin, &cooked->begin); 360 journal_pos_decode(&raw->h.end, &cooked->end); 361 cooked->index_size = decode_uint32(raw->h.index_size); 362 cooked->sourceserial = decode_uint32(raw->h.sourceserial); 363 cooked->serialset = ((raw->h.flags & JOURNAL_SERIALSET) != 0); 364 } 365 366 static void 367 journal_header_encode(journal_header_t *cooked, journal_rawheader_t *raw) { 368 unsigned char flags = 0; 369 370 INSIST(sizeof(cooked->format) == sizeof(raw->h.format)); 371 memset(raw->pad, 0, sizeof(raw->pad)); 372 memmove(raw->h.format, cooked->format, sizeof(raw->h.format)); 373 journal_pos_encode(&raw->h.begin, &cooked->begin); 374 journal_pos_encode(&raw->h.end, &cooked->end); 375 encode_uint32(cooked->index_size, raw->h.index_size); 376 encode_uint32(cooked->sourceserial, raw->h.sourceserial); 377 if (cooked->serialset) { 378 flags |= JOURNAL_SERIALSET; 379 } 380 raw->h.flags = flags; 381 } 382 383 /* 384 * Journal file I/O subroutines, with error checking and reporting. 385 */ 386 static isc_result_t 387 journal_seek(dns_journal_t *j, uint32_t offset) { 388 isc_result_t result; 389 390 result = isc_stdio_seek(j->fp, (off_t)offset, SEEK_SET); 391 if (result != ISC_R_SUCCESS) { 392 isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR, 393 "%s: seek: %s", j->filename, 394 isc_result_totext(result)); 395 return (ISC_R_UNEXPECTED); 396 } 397 j->offset = offset; 398 return (ISC_R_SUCCESS); 399 } 400 401 static isc_result_t 402 journal_read(dns_journal_t *j, void *mem, size_t nbytes) { 403 isc_result_t result; 404 405 result = isc_stdio_read(mem, 1, nbytes, j->fp, NULL); 406 if (result != ISC_R_SUCCESS) { 407 if (result == ISC_R_EOF) { 408 return (ISC_R_NOMORE); 409 } 410 isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR, 411 "%s: read: %s", j->filename, 412 isc_result_totext(result)); 413 return (ISC_R_UNEXPECTED); 414 } 415 j->offset += (isc_offset_t)nbytes; 416 return (ISC_R_SUCCESS); 417 } 418 419 static isc_result_t 420 journal_write(dns_journal_t *j, void *mem, size_t nbytes) { 421 isc_result_t result; 422 423 result = isc_stdio_write(mem, 1, nbytes, j->fp, NULL); 424 if (result != ISC_R_SUCCESS) { 425 isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR, 426 "%s: write: %s", j->filename, 427 isc_result_totext(result)); 428 return (ISC_R_UNEXPECTED); 429 } 430 j->offset += (isc_offset_t)nbytes; 431 return (ISC_R_SUCCESS); 432 } 433 434 static isc_result_t 435 journal_fsync(dns_journal_t *j) { 436 isc_result_t result; 437 result = isc_stdio_flush(j->fp); 438 if (result != ISC_R_SUCCESS) { 439 isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR, 440 "%s: flush: %s", j->filename, 441 isc_result_totext(result)); 442 return (ISC_R_UNEXPECTED); 443 } 444 result = isc_stdio_sync(j->fp); 445 if (result != ISC_R_SUCCESS) { 446 isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR, 447 "%s: fsync: %s", j->filename, 448 isc_result_totext(result)); 449 return (ISC_R_UNEXPECTED); 450 } 451 return (ISC_R_SUCCESS); 452 } 453 454 /* 455 * Read/write a transaction header at the current file position. 456 */ 457 458 static isc_result_t 459 journal_read_xhdr(dns_journal_t *j, journal_xhdr_t *xhdr) { 460 journal_rawxhdr_t raw; 461 isc_result_t result; 462 result = journal_read(j, &raw, sizeof(raw)); 463 if (result != ISC_R_SUCCESS) { 464 return (result); 465 } 466 xhdr->size = decode_uint32(raw.size); 467 xhdr->count = decode_uint32(raw.count); 468 xhdr->serial0 = decode_uint32(raw.serial0); 469 xhdr->serial1 = decode_uint32(raw.serial1); 470 return (ISC_R_SUCCESS); 471 } 472 473 static isc_result_t 474 journal_write_xhdr(dns_journal_t *j, uint32_t size, uint32_t count, 475 uint32_t serial0, uint32_t serial1) { 476 journal_rawxhdr_t raw; 477 encode_uint32(size, raw.size); 478 encode_uint32(count, raw.count); 479 encode_uint32(serial0, raw.serial0); 480 encode_uint32(serial1, raw.serial1); 481 return (journal_write(j, &raw, sizeof(raw))); 482 } 483 484 /* 485 * Read an RR header at the current file position. 486 */ 487 488 static isc_result_t 489 journal_read_rrhdr(dns_journal_t *j, journal_rrhdr_t *rrhdr) { 490 journal_rawrrhdr_t raw; 491 isc_result_t result; 492 result = journal_read(j, &raw, sizeof(raw)); 493 if (result != ISC_R_SUCCESS) { 494 return (result); 495 } 496 rrhdr->size = decode_uint32(raw.size); 497 return (ISC_R_SUCCESS); 498 } 499 500 static isc_result_t 501 journal_file_create(isc_mem_t *mctx, const char *filename) { 502 FILE *fp = NULL; 503 isc_result_t result; 504 journal_header_t header; 505 journal_rawheader_t rawheader; 506 int index_size = 56; /* XXX configurable */ 507 int size; 508 void *mem; /* Memory for temporary index image. */ 509 510 INSIST(sizeof(journal_rawheader_t) == JOURNAL_HEADER_SIZE); 511 512 result = isc_stdio_open(filename, "wb", &fp); 513 if (result != ISC_R_SUCCESS) { 514 isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR, 515 "%s: create: %s", filename, 516 isc_result_totext(result)); 517 return (ISC_R_UNEXPECTED); 518 } 519 520 header = initial_journal_header; 521 header.index_size = index_size; 522 journal_header_encode(&header, &rawheader); 523 524 size = sizeof(journal_rawheader_t) + 525 index_size * sizeof(journal_rawpos_t); 526 527 mem = isc_mem_get(mctx, size); 528 memset(mem, 0, size); 529 memmove(mem, &rawheader, sizeof(rawheader)); 530 531 result = isc_stdio_write(mem, 1, (size_t)size, fp, NULL); 532 if (result != ISC_R_SUCCESS) { 533 isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR, 534 "%s: write: %s", filename, 535 isc_result_totext(result)); 536 (void)isc_stdio_close(fp); 537 (void)isc_file_remove(filename); 538 isc_mem_put(mctx, mem, size); 539 return (ISC_R_UNEXPECTED); 540 } 541 isc_mem_put(mctx, mem, size); 542 543 result = isc_stdio_close(fp); 544 if (result != ISC_R_SUCCESS) { 545 isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR, 546 "%s: close: %s", filename, 547 isc_result_totext(result)); 548 (void)isc_file_remove(filename); 549 return (ISC_R_UNEXPECTED); 550 } 551 552 return (ISC_R_SUCCESS); 553 } 554 555 static isc_result_t 556 journal_open(isc_mem_t *mctx, const char *filename, bool writable, bool create, 557 dns_journal_t **journalp) { 558 FILE *fp = NULL; 559 isc_result_t result; 560 journal_rawheader_t rawheader; 561 dns_journal_t *j; 562 563 INSIST(journalp != NULL && *journalp == NULL); 564 j = isc_mem_get(mctx, sizeof(*j)); 565 566 j->mctx = NULL; 567 isc_mem_attach(mctx, &j->mctx); 568 j->state = JOURNAL_STATE_INVALID; 569 j->fp = NULL; 570 j->filename = isc_mem_strdup(mctx, filename); 571 j->index = NULL; 572 j->rawindex = NULL; 573 574 if (j->filename == NULL) { 575 FAIL(ISC_R_NOMEMORY); 576 } 577 578 result = isc_stdio_open(j->filename, writable ? "rb+" : "rb", &fp); 579 580 if (result == ISC_R_FILENOTFOUND) { 581 if (create) { 582 isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_DEBUG(1), 583 "journal file %s does not exist, " 584 "creating it", 585 j->filename); 586 CHECK(journal_file_create(mctx, filename)); 587 /* 588 * Retry. 589 */ 590 result = isc_stdio_open(j->filename, "rb+", &fp); 591 } else { 592 FAIL(ISC_R_NOTFOUND); 593 } 594 } 595 if (result != ISC_R_SUCCESS) { 596 isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR, 597 "%s: open: %s", j->filename, 598 isc_result_totext(result)); 599 FAIL(ISC_R_UNEXPECTED); 600 } 601 602 j->fp = fp; 603 604 /* 605 * Set magic early so that seek/read can succeed. 606 */ 607 j->magic = DNS_JOURNAL_MAGIC; 608 609 CHECK(journal_seek(j, 0)); 610 CHECK(journal_read(j, &rawheader, sizeof(rawheader))); 611 612 if (memcmp(rawheader.h.format, initial_journal_header.format, 613 sizeof(initial_journal_header.format)) != 0) 614 { 615 isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR, 616 "%s: journal format not recognized", j->filename); 617 FAIL(ISC_R_UNEXPECTED); 618 } 619 journal_header_decode(&rawheader, &j->header); 620 621 /* 622 * If there is an index, read the raw index into a dynamically 623 * allocated buffer and then convert it into a cooked index. 624 */ 625 if (j->header.index_size != 0) { 626 unsigned int i; 627 unsigned int rawbytes; 628 unsigned char *p; 629 630 rawbytes = j->header.index_size * sizeof(journal_rawpos_t); 631 j->rawindex = isc_mem_get(mctx, rawbytes); 632 633 CHECK(journal_read(j, j->rawindex, rawbytes)); 634 635 j->index = isc_mem_get(mctx, j->header.index_size * 636 sizeof(journal_pos_t)); 637 638 p = j->rawindex; 639 for (i = 0; i < j->header.index_size; i++) { 640 j->index[i].serial = decode_uint32(p); 641 p += 4; 642 j->index[i].offset = decode_uint32(p); 643 p += 4; 644 } 645 INSIST(p == j->rawindex + rawbytes); 646 } 647 j->offset = -1; /* Invalid, must seek explicitly. */ 648 649 /* 650 * Initialize the iterator. 651 */ 652 dns_name_init(&j->it.name, NULL); 653 dns_rdata_init(&j->it.rdata); 654 655 /* 656 * Set up empty initial buffers for unchecked and checked 657 * wire format RR data. They will be reallocated 658 * later. 659 */ 660 isc_buffer_init(&j->it.source, NULL, 0); 661 isc_buffer_init(&j->it.target, NULL, 0); 662 dns_decompress_init(&j->it.dctx, -1, DNS_DECOMPRESS_NONE); 663 664 j->state = writable ? JOURNAL_STATE_WRITE : JOURNAL_STATE_READ; 665 666 *journalp = j; 667 return (ISC_R_SUCCESS); 668 669 failure: 670 j->magic = 0; 671 if (j->rawindex != NULL) { 672 isc_mem_put(j->mctx, j->rawindex, 673 j->header.index_size * sizeof(journal_rawpos_t)); 674 } 675 if (j->index != NULL) { 676 isc_mem_put(j->mctx, j->index, 677 j->header.index_size * sizeof(journal_pos_t)); 678 } 679 if (j->filename != NULL) { 680 isc_mem_free(j->mctx, j->filename); 681 } 682 if (j->fp != NULL) { 683 (void)isc_stdio_close(j->fp); 684 } 685 isc_mem_putanddetach(&j->mctx, j, sizeof(*j)); 686 return (result); 687 } 688 689 isc_result_t 690 dns_journal_open(isc_mem_t *mctx, const char *filename, unsigned int mode, 691 dns_journal_t **journalp) { 692 isc_result_t result; 693 size_t namelen; 694 char backup[1024]; 695 bool writable, create; 696 697 create = ((mode & DNS_JOURNAL_CREATE) != 0); 698 writable = ((mode & (DNS_JOURNAL_WRITE | DNS_JOURNAL_CREATE)) != 0); 699 700 result = journal_open(mctx, filename, writable, create, journalp); 701 if (result == ISC_R_NOTFOUND) { 702 namelen = strlen(filename); 703 if (namelen > 4U && strcmp(filename + namelen - 4, ".jnl") == 0) 704 { 705 namelen -= 4; 706 } 707 708 result = snprintf(backup, sizeof(backup), "%.*s.jbk", 709 (int)namelen, filename); 710 if (result >= sizeof(backup)) { 711 return (ISC_R_NOSPACE); 712 } 713 result = journal_open(mctx, backup, writable, writable, 714 journalp); 715 } 716 return (result); 717 } 718 719 /* 720 * A comparison function defining the sorting order for 721 * entries in the IXFR-style journal file. 722 * 723 * The IXFR format requires that deletions are sorted before 724 * additions, and within either one, SOA records are sorted 725 * before others. 726 * 727 * Also sort the non-SOA records by type as a courtesy to the 728 * server receiving the IXFR - it may help reduce the amount of 729 * rdataset merging it has to do. 730 */ 731 static int 732 ixfr_order(const void *av, const void *bv) { 733 dns_difftuple_t const *const *ap = av; 734 dns_difftuple_t const *const *bp = bv; 735 dns_difftuple_t const *a = *ap; 736 dns_difftuple_t const *b = *bp; 737 int r; 738 int bop = 0, aop = 0; 739 740 switch (a->op) { 741 case DNS_DIFFOP_DEL: 742 case DNS_DIFFOP_DELRESIGN: 743 aop = 1; 744 break; 745 case DNS_DIFFOP_ADD: 746 case DNS_DIFFOP_ADDRESIGN: 747 aop = 0; 748 break; 749 default: 750 INSIST(0); 751 ISC_UNREACHABLE(); 752 } 753 754 switch (b->op) { 755 case DNS_DIFFOP_DEL: 756 case DNS_DIFFOP_DELRESIGN: 757 bop = 1; 758 break; 759 case DNS_DIFFOP_ADD: 760 case DNS_DIFFOP_ADDRESIGN: 761 bop = 0; 762 break; 763 default: 764 INSIST(0); 765 ISC_UNREACHABLE(); 766 } 767 768 r = bop - aop; 769 if (r != 0) { 770 return (r); 771 } 772 773 r = (b->rdata.type == dns_rdatatype_soa) - 774 (a->rdata.type == dns_rdatatype_soa); 775 if (r != 0) { 776 return (r); 777 } 778 779 r = (a->rdata.type - b->rdata.type); 780 return (r); 781 } 782 783 /* 784 * Advance '*pos' to the next journal transaction. 785 * 786 * Requires: 787 * *pos refers to a valid journal transaction. 788 * 789 * Ensures: 790 * When ISC_R_SUCCESS is returned, 791 * *pos refers to the next journal transaction. 792 * 793 * Returns one of: 794 * 795 * ISC_R_SUCCESS 796 * ISC_R_NOMORE *pos pointed at the last transaction 797 * Other results due to file errors are possible. 798 */ 799 static isc_result_t 800 journal_next(dns_journal_t *j, journal_pos_t *pos) { 801 isc_result_t result; 802 journal_xhdr_t xhdr; 803 REQUIRE(DNS_JOURNAL_VALID(j)); 804 805 result = journal_seek(j, pos->offset); 806 if (result != ISC_R_SUCCESS) { 807 return (result); 808 } 809 810 if (pos->serial == j->header.end.serial) { 811 return (ISC_R_NOMORE); 812 } 813 /* 814 * Read the header of the current transaction. 815 * This will return ISC_R_NOMORE if we are at EOF. 816 */ 817 result = journal_read_xhdr(j, &xhdr); 818 if (result != ISC_R_SUCCESS) { 819 return (result); 820 } 821 822 /* 823 * Check serial number consistency. 824 */ 825 if (xhdr.serial0 != pos->serial) { 826 isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR, 827 "%s: journal file corrupt: " 828 "expected serial %u, got %u", 829 j->filename, pos->serial, xhdr.serial0); 830 return (ISC_R_UNEXPECTED); 831 } 832 833 /* 834 * Check for offset wraparound. 835 */ 836 if ((isc_offset_t)(pos->offset + sizeof(journal_rawxhdr_t) + 837 xhdr.size) < pos->offset) 838 { 839 isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR, 840 "%s: offset too large", j->filename); 841 return (ISC_R_UNEXPECTED); 842 } 843 844 pos->offset += sizeof(journal_rawxhdr_t) + xhdr.size; 845 pos->serial = xhdr.serial1; 846 return (ISC_R_SUCCESS); 847 } 848 849 /* 850 * If the index of the journal 'j' contains an entry "better" 851 * than '*best_guess', replace '*best_guess' with it. 852 * 853 * "Better" means having a serial number closer to 'serial' 854 * but not greater than 'serial'. 855 */ 856 static void 857 index_find(dns_journal_t *j, uint32_t serial, journal_pos_t *best_guess) { 858 unsigned int i; 859 if (j->index == NULL) { 860 return; 861 } 862 for (i = 0; i < j->header.index_size; i++) { 863 if (POS_VALID(j->index[i]) && 864 DNS_SERIAL_GE(serial, j->index[i].serial) && 865 DNS_SERIAL_GT(j->index[i].serial, best_guess->serial)) 866 { 867 *best_guess = j->index[i]; 868 } 869 } 870 } 871 872 /* 873 * Add a new index entry. If there is no room, make room by removing 874 * the odd-numbered entries and compacting the others into the first 875 * half of the index. This decimates old index entries exponentially 876 * over time, so that the index always contains a much larger fraction 877 * of recent serial numbers than of old ones. This is deliberate - 878 * most index searches are for outgoing IXFR, and IXFR tends to request 879 * recent versions more often than old ones. 880 */ 881 static void 882 index_add(dns_journal_t *j, journal_pos_t *pos) { 883 unsigned int i; 884 if (j->index == NULL) { 885 return; 886 } 887 /* 888 * Search for a vacant position. 889 */ 890 for (i = 0; i < j->header.index_size; i++) { 891 if (!POS_VALID(j->index[i])) { 892 break; 893 } 894 } 895 if (i == j->header.index_size) { 896 unsigned int k = 0; 897 /* 898 * Found no vacant position. Make some room. 899 */ 900 for (i = 0; i < j->header.index_size; i += 2) { 901 j->index[k++] = j->index[i]; 902 } 903 i = k; /* 'i' identifies the first vacant position. */ 904 while (k < j->header.index_size) { 905 POS_INVALIDATE(j->index[k]); 906 k++; 907 } 908 } 909 INSIST(i < j->header.index_size); 910 INSIST(!POS_VALID(j->index[i])); 911 912 /* 913 * Store the new index entry. 914 */ 915 j->index[i] = *pos; 916 } 917 918 /* 919 * Invalidate any existing index entries that could become 920 * ambiguous when a new transaction with number 'serial' is added. 921 */ 922 static void 923 index_invalidate(dns_journal_t *j, uint32_t serial) { 924 unsigned int i; 925 if (j->index == NULL) { 926 return; 927 } 928 for (i = 0; i < j->header.index_size; i++) { 929 if (!DNS_SERIAL_GT(serial, j->index[i].serial)) { 930 POS_INVALIDATE(j->index[i]); 931 } 932 } 933 } 934 935 /* 936 * Try to find a transaction with initial serial number 'serial' 937 * in the journal 'j'. 938 * 939 * If found, store its position at '*pos' and return ISC_R_SUCCESS. 940 * 941 * If 'serial' is current (= the ending serial number of the 942 * last transaction in the journal), set '*pos' to 943 * the position immediately following the last transaction and 944 * return ISC_R_SUCCESS. 945 * 946 * If 'serial' is within the range of addressable serial numbers 947 * covered by the journal but that particular serial number is missing 948 * (from the journal, not just from the index), return ISC_R_NOTFOUND. 949 * 950 * If 'serial' is outside the range of addressable serial numbers 951 * covered by the journal, return ISC_R_RANGE. 952 * 953 */ 954 static isc_result_t 955 journal_find(dns_journal_t *j, uint32_t serial, journal_pos_t *pos) { 956 isc_result_t result; 957 journal_pos_t current_pos; 958 REQUIRE(DNS_JOURNAL_VALID(j)); 959 960 if (DNS_SERIAL_GT(j->header.begin.serial, serial)) { 961 return (ISC_R_RANGE); 962 } 963 if (DNS_SERIAL_GT(serial, j->header.end.serial)) { 964 return (ISC_R_RANGE); 965 } 966 if (serial == j->header.end.serial) { 967 *pos = j->header.end; 968 return (ISC_R_SUCCESS); 969 } 970 971 current_pos = j->header.begin; 972 index_find(j, serial, ¤t_pos); 973 974 while (current_pos.serial != serial) { 975 if (DNS_SERIAL_GT(current_pos.serial, serial)) { 976 return (ISC_R_NOTFOUND); 977 } 978 result = journal_next(j, ¤t_pos); 979 if (result != ISC_R_SUCCESS) { 980 return (result); 981 } 982 } 983 *pos = current_pos; 984 return (ISC_R_SUCCESS); 985 } 986 987 isc_result_t 988 dns_journal_begin_transaction(dns_journal_t *j) { 989 uint32_t offset; 990 isc_result_t result; 991 journal_rawxhdr_t hdr; 992 993 REQUIRE(DNS_JOURNAL_VALID(j)); 994 REQUIRE(j->state == JOURNAL_STATE_WRITE || 995 j->state == JOURNAL_STATE_INLINE); 996 997 /* 998 * Find the file offset where the new transaction should 999 * be written, and seek there. 1000 */ 1001 if (JOURNAL_EMPTY(&j->header)) { 1002 offset = sizeof(journal_rawheader_t) + 1003 j->header.index_size * sizeof(journal_rawpos_t); 1004 } else { 1005 offset = j->header.end.offset; 1006 } 1007 j->x.pos[0].offset = offset; 1008 j->x.pos[1].offset = offset; /* Initial value, will be incremented. */ 1009 j->x.n_soa = 0; 1010 1011 CHECK(journal_seek(j, offset)); 1012 1013 /* 1014 * Write a dummy transaction header of all zeroes to reserve 1015 * space. It will be filled in when the transaction is 1016 * finished. 1017 */ 1018 memset(&hdr, 0, sizeof(hdr)); 1019 CHECK(journal_write(j, &hdr, sizeof(hdr))); 1020 j->x.pos[1].offset = j->offset; 1021 1022 j->state = JOURNAL_STATE_TRANSACTION; 1023 result = ISC_R_SUCCESS; 1024 failure: 1025 return (result); 1026 } 1027 1028 isc_result_t 1029 dns_journal_writediff(dns_journal_t *j, dns_diff_t *diff) { 1030 dns_difftuple_t *t; 1031 isc_buffer_t buffer; 1032 void *mem = NULL; 1033 uint64_t size = 0; 1034 uint32_t rrcount = 0; 1035 isc_result_t result; 1036 isc_region_t used; 1037 1038 REQUIRE(DNS_DIFF_VALID(diff)); 1039 REQUIRE(j->state == JOURNAL_STATE_TRANSACTION); 1040 1041 isc_log_write(JOURNAL_DEBUG_LOGARGS(3), "writing to journal"); 1042 (void)dns_diff_print(diff, NULL); 1043 1044 /* 1045 * Pass 1: determine the buffer size needed, and 1046 * keep track of SOA serial numbers. 1047 */ 1048 for (t = ISC_LIST_HEAD(diff->tuples); t != NULL; 1049 t = ISC_LIST_NEXT(t, link)) { 1050 if (t->rdata.type == dns_rdatatype_soa) { 1051 if (j->x.n_soa < 2) { 1052 j->x.pos[j->x.n_soa].serial = 1053 dns_soa_getserial(&t->rdata); 1054 } 1055 j->x.n_soa++; 1056 } 1057 size += sizeof(journal_rawrrhdr_t); 1058 size += t->name.length; /* XXX should have access macro? */ 1059 size += 10; 1060 size += t->rdata.length; 1061 } 1062 1063 if (size >= DNS_JOURNAL_SIZE_MAX) { 1064 isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR, 1065 "dns_journal_writediff: %s: journal entry " 1066 "too big to be stored: %" PRIu64 " bytes", 1067 j->filename, size); 1068 return (ISC_R_NOSPACE); 1069 } 1070 1071 mem = isc_mem_get(j->mctx, size); 1072 1073 isc_buffer_init(&buffer, mem, size); 1074 1075 /* 1076 * Pass 2. Write RRs to buffer. 1077 */ 1078 for (t = ISC_LIST_HEAD(diff->tuples); t != NULL; 1079 t = ISC_LIST_NEXT(t, link)) { 1080 /* 1081 * Write the RR header. 1082 */ 1083 isc_buffer_putuint32(&buffer, 1084 t->name.length + 10 + t->rdata.length); 1085 /* 1086 * Write the owner name, RR header, and RR data. 1087 */ 1088 isc_buffer_putmem(&buffer, t->name.ndata, t->name.length); 1089 isc_buffer_putuint16(&buffer, t->rdata.type); 1090 isc_buffer_putuint16(&buffer, t->rdata.rdclass); 1091 isc_buffer_putuint32(&buffer, t->ttl); 1092 INSIST(t->rdata.length < 65536); 1093 isc_buffer_putuint16(&buffer, (uint16_t)t->rdata.length); 1094 INSIST(isc_buffer_availablelength(&buffer) >= t->rdata.length); 1095 isc_buffer_putmem(&buffer, t->rdata.data, t->rdata.length); 1096 1097 rrcount++; 1098 } 1099 1100 isc_buffer_usedregion(&buffer, &used); 1101 INSIST(used.length == size); 1102 1103 j->x.pos[1].offset += used.length; 1104 j->x.n_rr = rrcount; 1105 1106 /* 1107 * Write the buffer contents to the journal file. 1108 */ 1109 CHECK(journal_write(j, used.base, used.length)); 1110 1111 result = ISC_R_SUCCESS; 1112 1113 failure: 1114 if (mem != NULL) { 1115 isc_mem_put(j->mctx, mem, size); 1116 } 1117 return (result); 1118 } 1119 1120 isc_result_t 1121 dns_journal_commit(dns_journal_t *j) { 1122 isc_result_t result; 1123 journal_rawheader_t rawheader; 1124 uint64_t total; 1125 1126 REQUIRE(DNS_JOURNAL_VALID(j)); 1127 REQUIRE(j->state == JOURNAL_STATE_TRANSACTION || 1128 j->state == JOURNAL_STATE_INLINE); 1129 1130 /* 1131 * Just write out a updated header. 1132 */ 1133 if (j->state == JOURNAL_STATE_INLINE) { 1134 CHECK(journal_fsync(j)); 1135 journal_header_encode(&j->header, &rawheader); 1136 CHECK(journal_seek(j, 0)); 1137 CHECK(journal_write(j, &rawheader, sizeof(rawheader))); 1138 CHECK(journal_fsync(j)); 1139 j->state = JOURNAL_STATE_WRITE; 1140 return (ISC_R_SUCCESS); 1141 } 1142 1143 /* 1144 * Perform some basic consistency checks. 1145 */ 1146 if (j->x.n_soa != 2) { 1147 isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR, 1148 "%s: malformed transaction: %d SOAs", j->filename, 1149 j->x.n_soa); 1150 return (ISC_R_UNEXPECTED); 1151 } 1152 if (!DNS_SERIAL_GT(j->x.pos[1].serial, j->x.pos[0].serial)) { 1153 isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR, 1154 "%s: malformed transaction: serial number " 1155 "did not increase", 1156 j->filename); 1157 return (ISC_R_UNEXPECTED); 1158 } 1159 if (!JOURNAL_EMPTY(&j->header)) { 1160 if (j->x.pos[0].serial != j->header.end.serial) { 1161 isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR, 1162 "malformed transaction: " 1163 "%s last serial %u != " 1164 "transaction first serial %u", 1165 j->filename, j->header.end.serial, 1166 j->x.pos[0].serial); 1167 return (ISC_R_UNEXPECTED); 1168 } 1169 } 1170 1171 /* 1172 * We currently don't support huge journal entries. 1173 */ 1174 total = j->x.pos[1].offset - j->x.pos[0].offset; 1175 if (total >= DNS_JOURNAL_SIZE_MAX) { 1176 isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR, 1177 "transaction too big to be stored in journal: " 1178 "%" PRIu64 "b (max is %" PRIu64 "b)", 1179 total, (uint64_t)DNS_JOURNAL_SIZE_MAX); 1180 return (ISC_R_UNEXPECTED); 1181 } 1182 1183 /* 1184 * Some old journal entries may become non-addressable 1185 * when we increment the current serial number. Purge them 1186 * by stepping header.begin forward to the first addressable 1187 * transaction. Also purge them from the index. 1188 */ 1189 if (!JOURNAL_EMPTY(&j->header)) { 1190 while (!DNS_SERIAL_GT(j->x.pos[1].serial, 1191 j->header.begin.serial)) { 1192 CHECK(journal_next(j, &j->header.begin)); 1193 } 1194 index_invalidate(j, j->x.pos[1].serial); 1195 } 1196 #ifdef notyet 1197 if (DNS_SERIAL_GT(last_dumped_serial, j->x.pos[1].serial)) { 1198 force_dump(...); 1199 } 1200 #endif /* ifdef notyet */ 1201 1202 /* 1203 * Commit the transaction data to stable storage. 1204 */ 1205 CHECK(journal_fsync(j)); 1206 1207 if (j->state == JOURNAL_STATE_TRANSACTION) { 1208 isc_offset_t offset; 1209 offset = (j->x.pos[1].offset - j->x.pos[0].offset) - 1210 sizeof(journal_rawxhdr_t); 1211 /* 1212 * Update the transaction header. 1213 */ 1214 CHECK(journal_seek(j, j->x.pos[0].offset)); 1215 CHECK(journal_write_xhdr(j, offset, j->x.n_rr, 1216 j->x.pos[0].serial, 1217 j->x.pos[1].serial)); 1218 } 1219 1220 /* 1221 * Update the journal header. 1222 */ 1223 if (JOURNAL_EMPTY(&j->header)) { 1224 j->header.begin = j->x.pos[0]; 1225 } 1226 j->header.end = j->x.pos[1]; 1227 journal_header_encode(&j->header, &rawheader); 1228 CHECK(journal_seek(j, 0)); 1229 CHECK(journal_write(j, &rawheader, sizeof(rawheader))); 1230 1231 /* 1232 * Update the index. 1233 */ 1234 index_add(j, &j->x.pos[0]); 1235 1236 /* 1237 * Convert the index into on-disk format and write 1238 * it to disk. 1239 */ 1240 CHECK(index_to_disk(j)); 1241 1242 /* 1243 * Commit the header to stable storage. 1244 */ 1245 CHECK(journal_fsync(j)); 1246 1247 /* 1248 * We no longer have a transaction open. 1249 */ 1250 j->state = JOURNAL_STATE_WRITE; 1251 1252 result = ISC_R_SUCCESS; 1253 1254 failure: 1255 return (result); 1256 } 1257 1258 isc_result_t 1259 dns_journal_write_transaction(dns_journal_t *j, dns_diff_t *diff) { 1260 isc_result_t result; 1261 CHECK(dns_diff_sort(diff, ixfr_order)); 1262 CHECK(dns_journal_begin_transaction(j)); 1263 CHECK(dns_journal_writediff(j, diff)); 1264 CHECK(dns_journal_commit(j)); 1265 result = ISC_R_SUCCESS; 1266 failure: 1267 return (result); 1268 } 1269 1270 void 1271 dns_journal_destroy(dns_journal_t **journalp) { 1272 dns_journal_t *j = *journalp; 1273 *journalp = NULL; 1274 REQUIRE(DNS_JOURNAL_VALID(j)); 1275 1276 j->it.result = ISC_R_FAILURE; 1277 dns_name_invalidate(&j->it.name); 1278 dns_decompress_invalidate(&j->it.dctx); 1279 if (j->rawindex != NULL) { 1280 isc_mem_put(j->mctx, j->rawindex, 1281 j->header.index_size * sizeof(journal_rawpos_t)); 1282 } 1283 if (j->index != NULL) { 1284 isc_mem_put(j->mctx, j->index, 1285 j->header.index_size * sizeof(journal_pos_t)); 1286 } 1287 if (j->it.target.base != NULL) { 1288 isc_mem_put(j->mctx, j->it.target.base, j->it.target.length); 1289 } 1290 if (j->it.source.base != NULL) { 1291 isc_mem_put(j->mctx, j->it.source.base, j->it.source.length); 1292 } 1293 if (j->filename != NULL) { 1294 isc_mem_free(j->mctx, j->filename); 1295 } 1296 if (j->fp != NULL) { 1297 (void)isc_stdio_close(j->fp); 1298 } 1299 j->magic = 0; 1300 isc_mem_putanddetach(&j->mctx, j, sizeof(*j)); 1301 } 1302 1303 /* 1304 * Roll the open journal 'j' into the database 'db'. 1305 * A new database version will be created. 1306 */ 1307 1308 /* XXX Share code with incoming IXFR? */ 1309 1310 static isc_result_t 1311 roll_forward(dns_journal_t *j, dns_db_t *db, unsigned int options) { 1312 isc_buffer_t source; /* Transaction data from disk */ 1313 isc_buffer_t target; /* Ditto after _fromwire check */ 1314 uint32_t db_serial; /* Database SOA serial */ 1315 uint32_t end_serial; /* Last journal SOA serial */ 1316 isc_result_t result; 1317 dns_dbversion_t *ver = NULL; 1318 journal_pos_t pos; 1319 dns_diff_t diff; 1320 unsigned int n_soa = 0; 1321 unsigned int n_put = 0; 1322 dns_diffop_t op; 1323 1324 REQUIRE(DNS_JOURNAL_VALID(j)); 1325 REQUIRE(DNS_DB_VALID(db)); 1326 1327 dns_diff_init(j->mctx, &diff); 1328 1329 /* 1330 * Set up empty initial buffers for unchecked and checked 1331 * wire format transaction data. They will be reallocated 1332 * later. 1333 */ 1334 isc_buffer_init(&source, NULL, 0); 1335 isc_buffer_init(&target, NULL, 0); 1336 1337 /* 1338 * Create the new database version. 1339 */ 1340 CHECK(dns_db_newversion(db, &ver)); 1341 1342 /* 1343 * Get the current database SOA serial number. 1344 */ 1345 CHECK(dns_db_getsoaserial(db, ver, &db_serial)); 1346 1347 /* 1348 * Locate a journal entry for the current database serial. 1349 */ 1350 CHECK(journal_find(j, db_serial, &pos)); 1351 /* 1352 * XXX do more drastic things, like marking zone stale, 1353 * if this fails? 1354 */ 1355 /* 1356 * XXXRTH The zone code should probably mark the zone as bad and 1357 * scream loudly into the log if this is a dynamic update 1358 * log reply that failed. 1359 */ 1360 1361 end_serial = dns_journal_last_serial(j); 1362 if (db_serial == end_serial) { 1363 CHECK(DNS_R_UPTODATE); 1364 } 1365 1366 CHECK(dns_journal_iter_init(j, db_serial, end_serial, NULL)); 1367 1368 for (result = dns_journal_first_rr(j); result == ISC_R_SUCCESS; 1369 result = dns_journal_next_rr(j)) 1370 { 1371 dns_name_t *name; 1372 uint32_t ttl; 1373 dns_rdata_t *rdata; 1374 dns_difftuple_t *tuple = NULL; 1375 1376 name = NULL; 1377 rdata = NULL; 1378 dns_journal_current_rr(j, &name, &ttl, &rdata); 1379 1380 if (rdata->type == dns_rdatatype_soa) { 1381 n_soa++; 1382 if (n_soa == 2) { 1383 db_serial = j->it.current_serial; 1384 } 1385 } 1386 1387 if (n_soa == 3) { 1388 n_soa = 1; 1389 } 1390 if (n_soa == 0) { 1391 isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR, 1392 "%s: journal file corrupt: missing " 1393 "initial SOA", 1394 j->filename); 1395 FAIL(ISC_R_UNEXPECTED); 1396 } 1397 if ((options & DNS_JOURNALOPT_RESIGN) != 0) { 1398 op = (n_soa == 1) ? DNS_DIFFOP_DELRESIGN 1399 : DNS_DIFFOP_ADDRESIGN; 1400 } else { 1401 op = (n_soa == 1) ? DNS_DIFFOP_DEL : DNS_DIFFOP_ADD; 1402 } 1403 1404 CHECK(dns_difftuple_create(diff.mctx, op, name, ttl, rdata, 1405 &tuple)); 1406 dns_diff_append(&diff, &tuple); 1407 1408 if (++n_put > 100) { 1409 isc_log_write(JOURNAL_DEBUG_LOGARGS(3), 1410 "%s: applying diff to database (%u)", 1411 j->filename, db_serial); 1412 (void)dns_diff_print(&diff, NULL); 1413 CHECK(dns_diff_apply(&diff, db, ver)); 1414 dns_diff_clear(&diff); 1415 n_put = 0; 1416 } 1417 } 1418 if (result == ISC_R_NOMORE) { 1419 result = ISC_R_SUCCESS; 1420 } 1421 CHECK(result); 1422 1423 if (n_put != 0) { 1424 isc_log_write(JOURNAL_DEBUG_LOGARGS(3), 1425 "%s: applying final diff to database (%u)", 1426 j->filename, db_serial); 1427 (void)dns_diff_print(&diff, NULL); 1428 CHECK(dns_diff_apply(&diff, db, ver)); 1429 dns_diff_clear(&diff); 1430 } 1431 1432 failure: 1433 if (ver != NULL) { 1434 dns_db_closeversion(db, &ver, 1435 result == ISC_R_SUCCESS ? true : false); 1436 } 1437 1438 if (source.base != NULL) { 1439 isc_mem_put(j->mctx, source.base, source.length); 1440 } 1441 if (target.base != NULL) { 1442 isc_mem_put(j->mctx, target.base, target.length); 1443 } 1444 1445 dns_diff_clear(&diff); 1446 1447 INSIST(ver == NULL); 1448 1449 return (result); 1450 } 1451 1452 isc_result_t 1453 dns_journal_rollforward(isc_mem_t *mctx, dns_db_t *db, unsigned int options, 1454 const char *filename) { 1455 dns_journal_t *j; 1456 isc_result_t result; 1457 1458 REQUIRE(DNS_DB_VALID(db)); 1459 REQUIRE(filename != NULL); 1460 1461 j = NULL; 1462 result = dns_journal_open(mctx, filename, DNS_JOURNAL_READ, &j); 1463 if (result == ISC_R_NOTFOUND) { 1464 isc_log_write(JOURNAL_DEBUG_LOGARGS(3), "no journal file, but " 1465 "that's OK"); 1466 return (DNS_R_NOJOURNAL); 1467 } 1468 if (result != ISC_R_SUCCESS) { 1469 return (result); 1470 } 1471 if (JOURNAL_EMPTY(&j->header)) { 1472 result = DNS_R_UPTODATE; 1473 } else { 1474 result = roll_forward(j, db, options); 1475 } 1476 1477 dns_journal_destroy(&j); 1478 1479 return (result); 1480 } 1481 1482 isc_result_t 1483 dns_journal_print(isc_mem_t *mctx, const char *filename, FILE *file) { 1484 dns_journal_t *j; 1485 isc_buffer_t source; /* Transaction data from disk */ 1486 isc_buffer_t target; /* Ditto after _fromwire check */ 1487 uint32_t start_serial; /* Database SOA serial */ 1488 uint32_t end_serial; /* Last journal SOA serial */ 1489 isc_result_t result; 1490 dns_diff_t diff; 1491 unsigned int n_soa = 0; 1492 unsigned int n_put = 0; 1493 1494 REQUIRE(filename != NULL); 1495 1496 j = NULL; 1497 result = dns_journal_open(mctx, filename, DNS_JOURNAL_READ, &j); 1498 if (result == ISC_R_NOTFOUND) { 1499 isc_log_write(JOURNAL_DEBUG_LOGARGS(3), "no journal file"); 1500 return (DNS_R_NOJOURNAL); 1501 } 1502 1503 if (result != ISC_R_SUCCESS) { 1504 isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR, 1505 "journal open failure: %s: %s", 1506 isc_result_totext(result), filename); 1507 return (result); 1508 } 1509 1510 if (j->header.serialset) { 1511 fprintf(file, "Source serial = %u\n", j->header.sourceserial); 1512 } 1513 dns_diff_init(j->mctx, &diff); 1514 1515 /* 1516 * Set up empty initial buffers for unchecked and checked 1517 * wire format transaction data. They will be reallocated 1518 * later. 1519 */ 1520 isc_buffer_init(&source, NULL, 0); 1521 isc_buffer_init(&target, NULL, 0); 1522 1523 start_serial = dns_journal_first_serial(j); 1524 end_serial = dns_journal_last_serial(j); 1525 1526 CHECK(dns_journal_iter_init(j, start_serial, end_serial, NULL)); 1527 1528 for (result = dns_journal_first_rr(j); result == ISC_R_SUCCESS; 1529 result = dns_journal_next_rr(j)) 1530 { 1531 dns_name_t *name; 1532 uint32_t ttl; 1533 dns_rdata_t *rdata; 1534 dns_difftuple_t *tuple = NULL; 1535 1536 name = NULL; 1537 rdata = NULL; 1538 dns_journal_current_rr(j, &name, &ttl, &rdata); 1539 1540 if (rdata->type == dns_rdatatype_soa) { 1541 n_soa++; 1542 } 1543 1544 if (n_soa == 3) { 1545 n_soa = 1; 1546 } 1547 if (n_soa == 0) { 1548 isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR, 1549 "%s: journal file corrupt: missing " 1550 "initial SOA", 1551 j->filename); 1552 FAIL(ISC_R_UNEXPECTED); 1553 } 1554 CHECK(dns_difftuple_create( 1555 diff.mctx, n_soa == 1 ? DNS_DIFFOP_DEL : DNS_DIFFOP_ADD, 1556 name, ttl, rdata, &tuple)); 1557 dns_diff_append(&diff, &tuple); 1558 1559 if (++n_put > 100) { 1560 result = dns_diff_print(&diff, file); 1561 dns_diff_clear(&diff); 1562 n_put = 0; 1563 if (result != ISC_R_SUCCESS) { 1564 break; 1565 } 1566 } 1567 } 1568 if (result == ISC_R_NOMORE) { 1569 result = ISC_R_SUCCESS; 1570 } 1571 CHECK(result); 1572 1573 if (n_put != 0) { 1574 result = dns_diff_print(&diff, file); 1575 dns_diff_clear(&diff); 1576 } 1577 goto cleanup; 1578 1579 failure: 1580 isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR, 1581 "%s: cannot print: journal file corrupt", j->filename); 1582 1583 cleanup: 1584 if (source.base != NULL) { 1585 isc_mem_put(j->mctx, source.base, source.length); 1586 } 1587 if (target.base != NULL) { 1588 isc_mem_put(j->mctx, target.base, target.length); 1589 } 1590 1591 dns_diff_clear(&diff); 1592 dns_journal_destroy(&j); 1593 1594 return (result); 1595 } 1596 1597 /**************************************************************************/ 1598 /* 1599 * Miscellaneous accessors. 1600 */ 1601 bool 1602 dns_journal_empty(dns_journal_t *j) { 1603 return (JOURNAL_EMPTY(&j->header)); 1604 } 1605 1606 uint32_t 1607 dns_journal_first_serial(dns_journal_t *j) { 1608 return (j->header.begin.serial); 1609 } 1610 1611 uint32_t 1612 dns_journal_last_serial(dns_journal_t *j) { 1613 return (j->header.end.serial); 1614 } 1615 1616 void 1617 dns_journal_set_sourceserial(dns_journal_t *j, uint32_t sourceserial) { 1618 REQUIRE(j->state == JOURNAL_STATE_WRITE || 1619 j->state == JOURNAL_STATE_INLINE || 1620 j->state == JOURNAL_STATE_TRANSACTION); 1621 1622 j->header.sourceserial = sourceserial; 1623 j->header.serialset = true; 1624 if (j->state == JOURNAL_STATE_WRITE) { 1625 j->state = JOURNAL_STATE_INLINE; 1626 } 1627 } 1628 1629 bool 1630 dns_journal_get_sourceserial(dns_journal_t *j, uint32_t *sourceserial) { 1631 REQUIRE(sourceserial != NULL); 1632 1633 if (!j->header.serialset) { 1634 return (false); 1635 } 1636 *sourceserial = j->header.sourceserial; 1637 return (true); 1638 } 1639 1640 /**************************************************************************/ 1641 /* 1642 * Iteration support. 1643 * 1644 * When serving an outgoing IXFR, we transmit a part the journal starting 1645 * at the serial number in the IXFR request and ending at the serial 1646 * number that is current when the IXFR request arrives. The ending 1647 * serial number is not necessarily at the end of the journal: 1648 * the journal may grow while the IXFR is in progress, but we stop 1649 * when we reach the serial number that was current when the IXFR started. 1650 */ 1651 1652 static isc_result_t 1653 read_one_rr(dns_journal_t *j); 1654 1655 /* 1656 * Make sure the buffer 'b' is has at least 'size' bytes 1657 * allocated, and clear it. 1658 * 1659 * Requires: 1660 * Either b->base is NULL, or it points to b->length bytes of memory 1661 * previously allocated by isc_mem_get(). 1662 */ 1663 1664 static isc_result_t 1665 size_buffer(isc_mem_t *mctx, isc_buffer_t *b, unsigned size) { 1666 if (b->length < size) { 1667 void *mem = isc_mem_get(mctx, size); 1668 if (mem == NULL) { 1669 return (ISC_R_NOMEMORY); 1670 } 1671 if (b->base != NULL) { 1672 isc_mem_put(mctx, b->base, b->length); 1673 } 1674 b->base = mem; 1675 b->length = size; 1676 } 1677 isc_buffer_clear(b); 1678 return (ISC_R_SUCCESS); 1679 } 1680 1681 isc_result_t 1682 dns_journal_iter_init(dns_journal_t *j, uint32_t begin_serial, 1683 uint32_t end_serial, size_t *xfrsizep) { 1684 isc_result_t result; 1685 1686 CHECK(journal_find(j, begin_serial, &j->it.bpos)); 1687 INSIST(j->it.bpos.serial == begin_serial); 1688 1689 CHECK(journal_find(j, end_serial, &j->it.epos)); 1690 INSIST(j->it.epos.serial == end_serial); 1691 1692 if (xfrsizep != NULL) { 1693 journal_pos_t pos = j->it.bpos; 1694 journal_xhdr_t xhdr; 1695 uint64_t size = 0; 1696 uint32_t count = 0; 1697 1698 /* 1699 * We already know the beginning and ending serial 1700 * numbers are in the journal. Scan through them, 1701 * adding up sizes and RR counts so we can calculate 1702 * the IXFR size. 1703 */ 1704 CHECK(journal_seek(j, pos.offset)); 1705 do { 1706 CHECK(journal_read_xhdr(j, &xhdr)); 1707 1708 size += xhdr.size; 1709 count += xhdr.count; 1710 1711 result = journal_next(j, &pos); 1712 if (result == ISC_R_NOMORE) { 1713 result = ISC_R_SUCCESS; 1714 } 1715 CHECK(result); 1716 } while (pos.serial != end_serial); 1717 1718 /* 1719 * For each RR, subtract the length of the RR header, 1720 * as this would not be present in IXFR messages. 1721 * (We don't need to worry about the transaction header 1722 * because that was already excluded from xdr.size.) 1723 */ 1724 *xfrsizep = size - (count * sizeof(journal_rawrrhdr_t)); 1725 } 1726 1727 result = ISC_R_SUCCESS; 1728 failure: 1729 j->it.result = result; 1730 return (j->it.result); 1731 } 1732 1733 isc_result_t 1734 dns_journal_first_rr(dns_journal_t *j) { 1735 isc_result_t result; 1736 1737 /* 1738 * Seek to the beginning of the first transaction we are 1739 * interested in. 1740 */ 1741 CHECK(journal_seek(j, j->it.bpos.offset)); 1742 j->it.current_serial = j->it.bpos.serial; 1743 1744 j->it.xsize = 0; /* We have no transaction data yet... */ 1745 j->it.xpos = 0; /* ...and haven't used any of it. */ 1746 1747 return (read_one_rr(j)); 1748 1749 failure: 1750 return (result); 1751 } 1752 1753 static isc_result_t 1754 read_one_rr(dns_journal_t *j) { 1755 isc_result_t result; 1756 1757 dns_rdatatype_t rdtype; 1758 dns_rdataclass_t rdclass; 1759 unsigned int rdlen; 1760 uint32_t ttl; 1761 journal_xhdr_t xhdr; 1762 journal_rrhdr_t rrhdr; 1763 1764 if (j->offset > j->it.epos.offset) { 1765 isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR, 1766 "%s: journal corrupt: possible integer overflow", 1767 j->filename); 1768 return (ISC_R_UNEXPECTED); 1769 } 1770 if (j->offset == j->it.epos.offset) { 1771 return (ISC_R_NOMORE); 1772 } 1773 if (j->it.xpos == j->it.xsize) { 1774 /* 1775 * We are at a transaction boundary. 1776 * Read another transaction header. 1777 */ 1778 CHECK(journal_read_xhdr(j, &xhdr)); 1779 if (xhdr.size == 0) { 1780 isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR, 1781 "%s: journal corrupt: empty transaction", 1782 j->filename); 1783 FAIL(ISC_R_UNEXPECTED); 1784 } 1785 if (xhdr.serial0 != j->it.current_serial) { 1786 isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR, 1787 "%s: journal file corrupt: " 1788 "expected serial %u, got %u", 1789 j->filename, j->it.current_serial, 1790 xhdr.serial0); 1791 FAIL(ISC_R_UNEXPECTED); 1792 } 1793 j->it.xsize = xhdr.size; 1794 j->it.xpos = 0; 1795 } 1796 /* 1797 * Read an RR. 1798 */ 1799 CHECK(journal_read_rrhdr(j, &rrhdr)); 1800 /* 1801 * Perform a sanity check on the journal RR size. 1802 * The smallest possible RR has a 1-byte owner name 1803 * and a 10-byte header. The largest possible 1804 * RR has 65535 bytes of data, a header, and a maximum- 1805 * size owner name, well below 70 k total. 1806 */ 1807 if (rrhdr.size < 1 + 10 || rrhdr.size > 70000) { 1808 isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR, 1809 "%s: journal corrupt: impossible RR size " 1810 "(%d bytes)", 1811 j->filename, rrhdr.size); 1812 FAIL(ISC_R_UNEXPECTED); 1813 } 1814 1815 CHECK(size_buffer(j->mctx, &j->it.source, rrhdr.size)); 1816 CHECK(journal_read(j, j->it.source.base, rrhdr.size)); 1817 isc_buffer_add(&j->it.source, rrhdr.size); 1818 1819 /* 1820 * The target buffer is made the same size 1821 * as the source buffer, with the assumption that when 1822 * no compression in present, the output of dns_*_fromwire() 1823 * is no larger than the input. 1824 */ 1825 CHECK(size_buffer(j->mctx, &j->it.target, rrhdr.size)); 1826 1827 /* 1828 * Parse the owner name. We don't know where it 1829 * ends yet, so we make the entire "remaining" 1830 * part of the buffer "active". 1831 */ 1832 isc_buffer_setactive(&j->it.source, 1833 j->it.source.used - j->it.source.current); 1834 CHECK(dns_name_fromwire(&j->it.name, &j->it.source, &j->it.dctx, 0, 1835 &j->it.target)); 1836 1837 /* 1838 * Check that the RR header is there, and parse it. 1839 */ 1840 if (isc_buffer_remaininglength(&j->it.source) < 10) { 1841 FAIL(DNS_R_FORMERR); 1842 } 1843 1844 rdtype = isc_buffer_getuint16(&j->it.source); 1845 rdclass = isc_buffer_getuint16(&j->it.source); 1846 ttl = isc_buffer_getuint32(&j->it.source); 1847 rdlen = isc_buffer_getuint16(&j->it.source); 1848 1849 /* 1850 * Parse the rdata. 1851 */ 1852 if (isc_buffer_remaininglength(&j->it.source) != rdlen) { 1853 FAIL(DNS_R_FORMERR); 1854 } 1855 isc_buffer_setactive(&j->it.source, rdlen); 1856 dns_rdata_reset(&j->it.rdata); 1857 CHECK(dns_rdata_fromwire(&j->it.rdata, rdclass, rdtype, &j->it.source, 1858 &j->it.dctx, 0, &j->it.target)); 1859 j->it.ttl = ttl; 1860 1861 j->it.xpos += sizeof(journal_rawrrhdr_t) + rrhdr.size; 1862 if (rdtype == dns_rdatatype_soa) { 1863 /* XXX could do additional consistency checks here */ 1864 j->it.current_serial = dns_soa_getserial(&j->it.rdata); 1865 } 1866 1867 result = ISC_R_SUCCESS; 1868 1869 failure: 1870 j->it.result = result; 1871 return (result); 1872 } 1873 1874 isc_result_t 1875 dns_journal_next_rr(dns_journal_t *j) { 1876 j->it.result = read_one_rr(j); 1877 return (j->it.result); 1878 } 1879 1880 void 1881 dns_journal_current_rr(dns_journal_t *j, dns_name_t **name, uint32_t *ttl, 1882 dns_rdata_t **rdata) { 1883 REQUIRE(j->it.result == ISC_R_SUCCESS); 1884 *name = &j->it.name; 1885 *ttl = j->it.ttl; 1886 *rdata = &j->it.rdata; 1887 } 1888 1889 /**************************************************************************/ 1890 /* 1891 * Generating diffs from databases 1892 */ 1893 1894 /* 1895 * Construct a diff containing all the RRs at the current name of the 1896 * database iterator 'dbit' in database 'db', version 'ver'. 1897 * Set '*name' to the current name, and append the diff to 'diff'. 1898 * All new tuples will have the operation 'op'. 1899 * 1900 * Requires: 'name' must have buffer large enough to hold the name. 1901 * Typically, a dns_fixedname_t would be used. 1902 */ 1903 static isc_result_t 1904 get_name_diff(dns_db_t *db, dns_dbversion_t *ver, isc_stdtime_t now, 1905 dns_dbiterator_t *dbit, dns_name_t *name, dns_diffop_t op, 1906 dns_diff_t *diff) { 1907 isc_result_t result; 1908 dns_dbnode_t *node = NULL; 1909 dns_rdatasetiter_t *rdsiter = NULL; 1910 dns_difftuple_t *tuple = NULL; 1911 1912 result = dns_dbiterator_current(dbit, &node, name); 1913 if (result != ISC_R_SUCCESS) { 1914 return (result); 1915 } 1916 1917 result = dns_db_allrdatasets(db, node, ver, now, &rdsiter); 1918 if (result != ISC_R_SUCCESS) { 1919 goto cleanup_node; 1920 } 1921 1922 for (result = dns_rdatasetiter_first(rdsiter); result == ISC_R_SUCCESS; 1923 result = dns_rdatasetiter_next(rdsiter)) 1924 { 1925 dns_rdataset_t rdataset; 1926 1927 dns_rdataset_init(&rdataset); 1928 dns_rdatasetiter_current(rdsiter, &rdataset); 1929 1930 for (result = dns_rdataset_first(&rdataset); 1931 result == ISC_R_SUCCESS; 1932 result = dns_rdataset_next(&rdataset)) 1933 { 1934 dns_rdata_t rdata = DNS_RDATA_INIT; 1935 dns_rdataset_current(&rdataset, &rdata); 1936 result = dns_difftuple_create(diff->mctx, op, name, 1937 rdataset.ttl, &rdata, 1938 &tuple); 1939 if (result != ISC_R_SUCCESS) { 1940 dns_rdataset_disassociate(&rdataset); 1941 goto cleanup_iterator; 1942 } 1943 dns_diff_append(diff, &tuple); 1944 } 1945 dns_rdataset_disassociate(&rdataset); 1946 if (result != ISC_R_NOMORE) { 1947 goto cleanup_iterator; 1948 } 1949 } 1950 if (result != ISC_R_NOMORE) { 1951 goto cleanup_iterator; 1952 } 1953 1954 result = ISC_R_SUCCESS; 1955 1956 cleanup_iterator: 1957 dns_rdatasetiter_destroy(&rdsiter); 1958 1959 cleanup_node: 1960 dns_db_detachnode(db, &node); 1961 1962 return (result); 1963 } 1964 1965 /* 1966 * Comparison function for use by dns_diff_subtract when sorting 1967 * the diffs to be subtracted. The sort keys are the rdata type 1968 * and the rdata itself. The owner name is ignored, because 1969 * it is known to be the same for all tuples. 1970 */ 1971 static int 1972 rdata_order(const void *av, const void *bv) { 1973 dns_difftuple_t const *const *ap = av; 1974 dns_difftuple_t const *const *bp = bv; 1975 dns_difftuple_t const *a = *ap; 1976 dns_difftuple_t const *b = *bp; 1977 int r; 1978 r = (b->rdata.type - a->rdata.type); 1979 if (r != 0) { 1980 return (r); 1981 } 1982 r = dns_rdata_compare(&a->rdata, &b->rdata); 1983 return (r); 1984 } 1985 1986 static isc_result_t 1987 dns_diff_subtract(dns_diff_t diff[2], dns_diff_t *r) { 1988 isc_result_t result; 1989 dns_difftuple_t *p[2]; 1990 int i, t; 1991 bool append; 1992 1993 CHECK(dns_diff_sort(&diff[0], rdata_order)); 1994 CHECK(dns_diff_sort(&diff[1], rdata_order)); 1995 1996 for (;;) { 1997 p[0] = ISC_LIST_HEAD(diff[0].tuples); 1998 p[1] = ISC_LIST_HEAD(diff[1].tuples); 1999 if (p[0] == NULL && p[1] == NULL) { 2000 break; 2001 } 2002 2003 for (i = 0; i < 2; i++) { 2004 if (p[!i] == NULL) { 2005 { 2006 ISC_LIST_UNLINK(diff[i].tuples, p[i], 2007 link); 2008 ISC_LIST_APPEND(r->tuples, p[i], link); 2009 goto next; 2010 } 2011 } 2012 } 2013 t = rdata_order(&p[0], &p[1]); 2014 if (t < 0) { 2015 ISC_LIST_UNLINK(diff[0].tuples, p[0], link); 2016 ISC_LIST_APPEND(r->tuples, p[0], link); 2017 goto next; 2018 } 2019 if (t > 0) { 2020 ISC_LIST_UNLINK(diff[1].tuples, p[1], link); 2021 ISC_LIST_APPEND(r->tuples, p[1], link); 2022 goto next; 2023 } 2024 INSIST(t == 0); 2025 /* 2026 * Identical RRs in both databases; skip them both 2027 * if the ttl differs. 2028 */ 2029 append = (p[0]->ttl != p[1]->ttl); 2030 for (i = 0; i < 2; i++) { 2031 ISC_LIST_UNLINK(diff[i].tuples, p[i], link); 2032 if (append) { 2033 ISC_LIST_APPEND(r->tuples, p[i], link); 2034 } else { 2035 dns_difftuple_free(&p[i]); 2036 } 2037 } 2038 next:; 2039 } 2040 result = ISC_R_SUCCESS; 2041 failure: 2042 return (result); 2043 } 2044 2045 static isc_result_t 2046 diff_namespace(dns_db_t *dba, dns_dbversion_t *dbvera, dns_db_t *dbb, 2047 dns_dbversion_t *dbverb, unsigned int options, 2048 dns_diff_t *resultdiff) { 2049 dns_db_t *db[2]; 2050 dns_dbversion_t *ver[2]; 2051 dns_dbiterator_t *dbit[2] = { NULL, NULL }; 2052 bool have[2] = { false, false }; 2053 dns_fixedname_t fixname[2]; 2054 isc_result_t result, itresult[2]; 2055 dns_diff_t diff[2]; 2056 int i, t; 2057 2058 db[0] = dba, db[1] = dbb; 2059 ver[0] = dbvera, ver[1] = dbverb; 2060 2061 dns_diff_init(resultdiff->mctx, &diff[0]); 2062 dns_diff_init(resultdiff->mctx, &diff[1]); 2063 2064 dns_fixedname_init(&fixname[0]); 2065 dns_fixedname_init(&fixname[1]); 2066 2067 result = dns_db_createiterator(db[0], options, &dbit[0]); 2068 if (result != ISC_R_SUCCESS) { 2069 return (result); 2070 } 2071 result = dns_db_createiterator(db[1], options, &dbit[1]); 2072 if (result != ISC_R_SUCCESS) { 2073 goto cleanup_iterator; 2074 } 2075 2076 itresult[0] = dns_dbiterator_first(dbit[0]); 2077 itresult[1] = dns_dbiterator_first(dbit[1]); 2078 2079 for (;;) { 2080 for (i = 0; i < 2; i++) { 2081 if (!have[i] && itresult[i] == ISC_R_SUCCESS) { 2082 CHECK(get_name_diff( 2083 db[i], ver[i], 0, dbit[i], 2084 dns_fixedname_name(&fixname[i]), 2085 i == 0 ? DNS_DIFFOP_ADD 2086 : DNS_DIFFOP_DEL, 2087 &diff[i])); 2088 itresult[i] = dns_dbiterator_next(dbit[i]); 2089 have[i] = true; 2090 } 2091 } 2092 2093 if (!have[0] && !have[1]) { 2094 INSIST(ISC_LIST_EMPTY(diff[0].tuples)); 2095 INSIST(ISC_LIST_EMPTY(diff[1].tuples)); 2096 break; 2097 } 2098 2099 for (i = 0; i < 2; i++) { 2100 if (!have[!i]) { 2101 ISC_LIST_APPENDLIST(resultdiff->tuples, 2102 diff[i].tuples, link); 2103 INSIST(ISC_LIST_EMPTY(diff[i].tuples)); 2104 have[i] = false; 2105 goto next; 2106 } 2107 } 2108 2109 t = dns_name_compare(dns_fixedname_name(&fixname[0]), 2110 dns_fixedname_name(&fixname[1])); 2111 if (t < 0) { 2112 ISC_LIST_APPENDLIST(resultdiff->tuples, diff[0].tuples, 2113 link); 2114 INSIST(ISC_LIST_EMPTY(diff[0].tuples)); 2115 have[0] = false; 2116 continue; 2117 } 2118 if (t > 0) { 2119 ISC_LIST_APPENDLIST(resultdiff->tuples, diff[1].tuples, 2120 link); 2121 INSIST(ISC_LIST_EMPTY(diff[1].tuples)); 2122 have[1] = false; 2123 continue; 2124 } 2125 INSIST(t == 0); 2126 CHECK(dns_diff_subtract(diff, resultdiff)); 2127 INSIST(ISC_LIST_EMPTY(diff[0].tuples)); 2128 INSIST(ISC_LIST_EMPTY(diff[1].tuples)); 2129 have[0] = have[1] = false; 2130 next:; 2131 } 2132 if (itresult[0] != ISC_R_NOMORE) { 2133 FAIL(itresult[0]); 2134 } 2135 if (itresult[1] != ISC_R_NOMORE) { 2136 FAIL(itresult[1]); 2137 } 2138 2139 INSIST(ISC_LIST_EMPTY(diff[0].tuples)); 2140 INSIST(ISC_LIST_EMPTY(diff[1].tuples)); 2141 2142 failure: 2143 dns_dbiterator_destroy(&dbit[1]); 2144 2145 cleanup_iterator: 2146 dns_dbiterator_destroy(&dbit[0]); 2147 dns_diff_clear(&diff[0]); 2148 dns_diff_clear(&diff[1]); 2149 return (result); 2150 } 2151 2152 /* 2153 * Compare the databases 'dba' and 'dbb' and generate a journal 2154 * entry containing the changes to make 'dba' from 'dbb' (note 2155 * the order). This journal entry will consist of a single, 2156 * possibly very large transaction. 2157 */ 2158 isc_result_t 2159 dns_db_diff(isc_mem_t *mctx, dns_db_t *dba, dns_dbversion_t *dbvera, 2160 dns_db_t *dbb, dns_dbversion_t *dbverb, const char *filename) { 2161 isc_result_t result; 2162 dns_diff_t diff; 2163 2164 dns_diff_init(mctx, &diff); 2165 2166 result = dns_db_diffx(&diff, dba, dbvera, dbb, dbverb, filename); 2167 2168 dns_diff_clear(&diff); 2169 2170 return (result); 2171 } 2172 2173 isc_result_t 2174 dns_db_diffx(dns_diff_t *diff, dns_db_t *dba, dns_dbversion_t *dbvera, 2175 dns_db_t *dbb, dns_dbversion_t *dbverb, const char *filename) { 2176 isc_result_t result; 2177 dns_journal_t *journal = NULL; 2178 2179 if (filename != NULL) { 2180 result = dns_journal_open(diff->mctx, filename, 2181 DNS_JOURNAL_CREATE, &journal); 2182 if (result != ISC_R_SUCCESS) { 2183 return (result); 2184 } 2185 } 2186 2187 CHECK(diff_namespace(dba, dbvera, dbb, dbverb, DNS_DB_NONSEC3, diff)); 2188 CHECK(diff_namespace(dba, dbvera, dbb, dbverb, DNS_DB_NSEC3ONLY, diff)); 2189 2190 if (journal != NULL) { 2191 if (ISC_LIST_EMPTY(diff->tuples)) { 2192 isc_log_write(JOURNAL_DEBUG_LOGARGS(3), "no changes"); 2193 } else { 2194 CHECK(dns_journal_write_transaction(journal, diff)); 2195 } 2196 } 2197 2198 failure: 2199 if (journal != NULL) { 2200 dns_journal_destroy(&journal); 2201 } 2202 return (result); 2203 } 2204 2205 isc_result_t 2206 dns_journal_compact(isc_mem_t *mctx, char *filename, uint32_t serial, 2207 uint32_t target_size) { 2208 unsigned int i; 2209 journal_pos_t best_guess; 2210 journal_pos_t current_pos; 2211 dns_journal_t *j1 = NULL; 2212 dns_journal_t *j2 = NULL; 2213 journal_rawheader_t rawheader; 2214 unsigned int copy_length; 2215 size_t namelen; 2216 char *buf = NULL; 2217 unsigned int size = 0; 2218 isc_result_t result; 2219 unsigned int indexend; 2220 char newname[PATH_MAX]; 2221 char backup[PATH_MAX]; 2222 bool is_backup = false; 2223 2224 REQUIRE(filename != NULL); 2225 2226 namelen = strlen(filename); 2227 if (namelen > 4U && strcmp(filename + namelen - 4, ".jnl") == 0) { 2228 namelen -= 4; 2229 } 2230 2231 result = snprintf(newname, sizeof(newname), "%.*s.jnw", (int)namelen, 2232 filename); 2233 RUNTIME_CHECK(result < sizeof(newname)); 2234 2235 result = snprintf(backup, sizeof(backup), "%.*s.jbk", (int)namelen, 2236 filename); 2237 RUNTIME_CHECK(result < sizeof(backup)); 2238 2239 result = journal_open(mctx, filename, false, false, &j1); 2240 if (result == ISC_R_NOTFOUND) { 2241 is_backup = true; 2242 result = journal_open(mctx, backup, false, false, &j1); 2243 } 2244 if (result != ISC_R_SUCCESS) { 2245 return (result); 2246 } 2247 2248 if (JOURNAL_EMPTY(&j1->header)) { 2249 dns_journal_destroy(&j1); 2250 return (ISC_R_SUCCESS); 2251 } 2252 2253 if (DNS_SERIAL_GT(j1->header.begin.serial, serial) || 2254 DNS_SERIAL_GT(serial, j1->header.end.serial)) 2255 { 2256 dns_journal_destroy(&j1); 2257 return (ISC_R_RANGE); 2258 } 2259 2260 /* 2261 * Cope with very small target sizes. 2262 */ 2263 indexend = sizeof(journal_rawheader_t) + 2264 j1->header.index_size * sizeof(journal_rawpos_t); 2265 if (target_size < DNS_JOURNAL_SIZE_MIN) { 2266 target_size = DNS_JOURNAL_SIZE_MIN; 2267 } 2268 if (target_size < indexend * 2) { 2269 target_size = target_size / 2 + indexend; 2270 } 2271 2272 /* 2273 * See if there is any work to do. 2274 */ 2275 if ((uint32_t)j1->header.end.offset < target_size) { 2276 dns_journal_destroy(&j1); 2277 return (ISC_R_SUCCESS); 2278 } 2279 2280 CHECK(journal_open(mctx, newname, true, true, &j2)); 2281 2282 /* 2283 * Remove overhead so space test below can succeed. 2284 */ 2285 if (target_size >= indexend) { 2286 target_size -= indexend; 2287 } 2288 2289 /* 2290 * Find if we can create enough free space. 2291 */ 2292 best_guess = j1->header.begin; 2293 for (i = 0; i < j1->header.index_size; i++) { 2294 if (POS_VALID(j1->index[i]) && 2295 DNS_SERIAL_GE(serial, j1->index[i].serial) && 2296 ((uint32_t)(j1->header.end.offset - j1->index[i].offset) >= 2297 target_size / 2) && 2298 j1->index[i].offset > best_guess.offset) 2299 { 2300 best_guess = j1->index[i]; 2301 } 2302 } 2303 2304 current_pos = best_guess; 2305 while (current_pos.serial != serial) { 2306 CHECK(journal_next(j1, ¤t_pos)); 2307 if (current_pos.serial == j1->header.end.serial) { 2308 break; 2309 } 2310 2311 if (DNS_SERIAL_GE(serial, current_pos.serial) && 2312 ((uint32_t)(j1->header.end.offset - current_pos.offset) >= 2313 (target_size / 2)) && 2314 current_pos.offset > best_guess.offset) 2315 { 2316 best_guess = current_pos; 2317 } else { 2318 break; 2319 } 2320 } 2321 2322 INSIST(best_guess.serial != j1->header.end.serial); 2323 if (best_guess.serial != serial) { 2324 CHECK(journal_next(j1, &best_guess)); 2325 } 2326 2327 /* 2328 * We should now be roughly half target_size provided 2329 * we did not reach 'serial'. If not we will just copy 2330 * all uncommitted deltas regardless of the size. 2331 */ 2332 copy_length = j1->header.end.offset - best_guess.offset; 2333 2334 if (copy_length != 0) { 2335 /* 2336 * Copy best_guess to end into space just freed. 2337 */ 2338 size = 64 * 1024; 2339 if (copy_length < size) { 2340 size = copy_length; 2341 } 2342 buf = isc_mem_get(mctx, size); 2343 2344 CHECK(journal_seek(j1, best_guess.offset)); 2345 CHECK(journal_seek(j2, indexend)); 2346 for (i = 0; i < copy_length; i += size) { 2347 unsigned int len = (copy_length - i) > size 2348 ? size 2349 : (copy_length - i); 2350 CHECK(journal_read(j1, buf, len)); 2351 CHECK(journal_write(j2, buf, len)); 2352 } 2353 2354 CHECK(journal_fsync(j2)); 2355 2356 /* 2357 * Compute new header. 2358 */ 2359 j2->header.begin.serial = best_guess.serial; 2360 j2->header.begin.offset = indexend; 2361 j2->header.end.serial = j1->header.end.serial; 2362 j2->header.end.offset = indexend + copy_length; 2363 j2->header.sourceserial = j1->header.sourceserial; 2364 j2->header.serialset = j1->header.serialset; 2365 2366 /* 2367 * Update the journal header. 2368 */ 2369 journal_header_encode(&j2->header, &rawheader); 2370 CHECK(journal_seek(j2, 0)); 2371 CHECK(journal_write(j2, &rawheader, sizeof(rawheader))); 2372 CHECK(journal_fsync(j2)); 2373 2374 /* 2375 * Build new index. 2376 */ 2377 current_pos = j2->header.begin; 2378 while (current_pos.serial != j2->header.end.serial) { 2379 index_add(j2, ¤t_pos); 2380 CHECK(journal_next(j2, ¤t_pos)); 2381 } 2382 2383 /* 2384 * Write index. 2385 */ 2386 CHECK(index_to_disk(j2)); 2387 CHECK(journal_fsync(j2)); 2388 2389 indexend = j2->header.end.offset; 2390 POST(indexend); 2391 } 2392 2393 /* 2394 * Close both journals before trying to rename files (this is 2395 * necessary on WIN32). 2396 */ 2397 dns_journal_destroy(&j1); 2398 dns_journal_destroy(&j2); 2399 2400 /* 2401 * With a UFS file system this should just succeed and be atomic. 2402 * Any IXFR outs will just continue and the old journal will be 2403 * removed on final close. 2404 * 2405 * With MSDOS / NTFS we need to do a two stage rename, triggered 2406 * by EEXIST. (If any IXFR's are running in other threads, however, 2407 * this will fail, and the journal will not be compacted. But 2408 * if so, hopefully they'll be finished by the next time we 2409 * compact.) 2410 */ 2411 if (rename(newname, filename) == -1) { 2412 if (errno == EEXIST && !is_backup) { 2413 result = isc_file_remove(backup); 2414 if (result != ISC_R_SUCCESS && 2415 result != ISC_R_FILENOTFOUND) { 2416 goto failure; 2417 } 2418 if (rename(filename, backup) == -1) { 2419 goto maperrno; 2420 } 2421 if (rename(newname, filename) == -1) { 2422 goto maperrno; 2423 } 2424 (void)isc_file_remove(backup); 2425 } else { 2426 maperrno: 2427 result = ISC_R_FAILURE; 2428 goto failure; 2429 } 2430 } 2431 2432 result = ISC_R_SUCCESS; 2433 2434 failure: 2435 (void)isc_file_remove(newname); 2436 if (buf != NULL) { 2437 isc_mem_put(mctx, buf, size); 2438 } 2439 if (j1 != NULL) { 2440 dns_journal_destroy(&j1); 2441 } 2442 if (j2 != NULL) { 2443 dns_journal_destroy(&j2); 2444 } 2445 return (result); 2446 } 2447 2448 static isc_result_t 2449 index_to_disk(dns_journal_t *j) { 2450 isc_result_t result = ISC_R_SUCCESS; 2451 2452 if (j->header.index_size != 0) { 2453 unsigned int i; 2454 unsigned char *p; 2455 unsigned int rawbytes; 2456 2457 rawbytes = j->header.index_size * sizeof(journal_rawpos_t); 2458 2459 p = j->rawindex; 2460 for (i = 0; i < j->header.index_size; i++) { 2461 encode_uint32(j->index[i].serial, p); 2462 p += 4; 2463 encode_uint32(j->index[i].offset, p); 2464 p += 4; 2465 } 2466 INSIST(p == j->rawindex + rawbytes); 2467 2468 CHECK(journal_seek(j, sizeof(journal_rawheader_t))); 2469 CHECK(journal_write(j, j->rawindex, rawbytes)); 2470 } 2471 failure: 2472 return (result); 2473 } 2474
Indexes created Fri Mar 20 00:23:26 UTC 2026