journal.c revision 193149
1/* 2 * Copyright (C) 2004, 2005, 2007-2009 Internet Systems Consortium, Inc. ("ISC") 3 * Copyright (C) 1999-2002 Internet Software Consortium. 4 * 5 * Permission to use, copy, modify, and/or distribute this software for any 6 * purpose with or without fee is hereby granted, provided that the above 7 * copyright notice and this permission notice appear in all copies. 8 * 9 * THE SOFTWARE IS PROVIDED "AS IS" AND ISC DISCLAIMS ALL WARRANTIES WITH 10 * REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY 11 * AND FITNESS. IN NO EVENT SHALL ISC BE LIABLE FOR ANY SPECIAL, DIRECT, 12 * INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM 13 * LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE 14 * OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR 15 * PERFORMANCE OF THIS SOFTWARE. 16 */ 17 18/* $Id: journal.c,v 1.103.48.2 2009/01/18 23:47:37 tbox Exp $ */ 19 20#include <config.h> 21 22#include <stdlib.h> 23#include <unistd.h> 24#include <errno.h> 25 26#include <isc/file.h> 27#include <isc/mem.h> 28#include <isc/stdio.h> 29#include <isc/string.h> 30#include <isc/util.h> 31 32#include <dns/compress.h> 33#include <dns/db.h> 34#include <dns/dbiterator.h> 35#include <dns/diff.h> 36#include <dns/fixedname.h> 37#include <dns/journal.h> 38#include <dns/log.h> 39#include <dns/rdataset.h> 40#include <dns/rdatasetiter.h> 41#include <dns/result.h> 42#include <dns/soa.h> 43 44/*! \file 45 * \brief Journaling. 46 * 47 * A journal file consists of 48 * 49 * \li A fixed-size header of type journal_rawheader_t. 50 * 51 * \li The index. This is an unordered array of index entries 52 * of type journal_rawpos_t giving the locations 53 * of some arbitrary subset of the journal's addressable 54 * transactions. The index entries are used as hints to 55 * speed up the process of locating a transaction with a given 56 * serial number. Unused index entries have an "offset" 57 * field of zero. The size of the index can vary between 58 * journal files, but does not change during the lifetime 59 * of a file. The size can be zero. 60 * 61 * \li The journal data. This consists of one or more transactions. 62 * Each transaction begins with a transaction header of type 63 * journal_rawxhdr_t. The transaction header is followed by a 64 * sequence of RRs, similar in structure to an IXFR difference 65 * sequence (RFC1995). That is, the pre-transaction SOA, 66 * zero or more other deleted RRs, the post-transaction SOA, 67 * and zero or more other added RRs. Unlike in IXFR, each RR 68 * is prefixed with a 32-bit length. 69 * 70 * The journal data part grows as new transactions are 71 * appended to the file. Only those transactions 72 * whose serial number is current-(2^31-1) to current 73 * are considered "addressable" and may be pointed 74 * to from the header or index. They may be preceded 75 * by old transactions that are no longer addressable, 76 * and they may be followed by transactions that were 77 * appended to the journal but never committed by updating 78 * the "end" position in the header. The latter will 79 * be overwritten when new transactions are added. 80 */ 81/*% 82 * When true, accept IXFR difference sequences where the 83 * SOA serial number does not change (BIND 8 sends such 84 * sequences). 85 */ 86static isc_boolean_t bind8_compat = ISC_TRUE; /* XXX config */ 87 88/**************************************************************************/ 89/* 90 * Miscellaneous utilities. 91 */ 92 93#define JOURNAL_COMMON_LOGARGS \ 94 dns_lctx, DNS_LOGCATEGORY_GENERAL, DNS_LOGMODULE_JOURNAL 95 96#define JOURNAL_DEBUG_LOGARGS(n) \ 97 JOURNAL_COMMON_LOGARGS, ISC_LOG_DEBUG(n) 98 99/*% 100 * It would be non-sensical (or at least obtuse) to use FAIL() with an 101 * ISC_R_SUCCESS code, but the test is there to keep the Solaris compiler 102 * from complaining about "end-of-loop code not reached". 103 */ 104#define FAIL(code) \ 105 do { result = (code); \ 106 if (result != ISC_R_SUCCESS) goto failure; \ 107 } while (0) 108 109#define CHECK(op) \ 110 do { result = (op); \ 111 if (result != ISC_R_SUCCESS) goto failure; \ 112 } while (0) 113 114static isc_result_t index_to_disk(dns_journal_t *); 115 116static inline isc_uint32_t 117decode_uint32(unsigned char *p) { 118 return ((p[0] << 24) + 119 (p[1] << 16) + 120 (p[2] << 8) + 121 (p[3] << 0)); 122} 123 124static inline void 125encode_uint32(isc_uint32_t val, unsigned char *p) { 126 p[0] = (isc_uint8_t)(val >> 24); 127 p[1] = (isc_uint8_t)(val >> 16); 128 p[2] = (isc_uint8_t)(val >> 8); 129 p[3] = (isc_uint8_t)(val >> 0); 130} 131 132isc_result_t 133dns_db_createsoatuple(dns_db_t *db, dns_dbversion_t *ver, isc_mem_t *mctx, 134 dns_diffop_t op, dns_difftuple_t **tp) 135{ 136 isc_result_t result; 137 dns_dbnode_t *node; 138 dns_rdataset_t rdataset; 139 dns_rdata_t rdata = DNS_RDATA_INIT; 140 dns_name_t *zonename; 141 142 zonename = dns_db_origin(db); 143 144 node = NULL; 145 result = dns_db_findnode(db, zonename, ISC_FALSE, &node); 146 if (result != ISC_R_SUCCESS) 147 goto nonode; 148 149 dns_rdataset_init(&rdataset); 150 result = dns_db_findrdataset(db, node, ver, dns_rdatatype_soa, 0, 151 (isc_stdtime_t)0, &rdataset, NULL); 152 if (result != ISC_R_SUCCESS) 153 goto freenode; 154 155 result = dns_rdataset_first(&rdataset); 156 if (result != ISC_R_SUCCESS) 157 goto freenode; 158 159 dns_rdataset_current(&rdataset, &rdata); 160 161 result = dns_difftuple_create(mctx, op, zonename, rdataset.ttl, 162 &rdata, tp); 163 164 dns_rdataset_disassociate(&rdataset); 165 dns_db_detachnode(db, &node); 166 return (ISC_R_SUCCESS); 167 168 freenode: 169 dns_db_detachnode(db, &node); 170 nonode: 171 UNEXPECTED_ERROR(__FILE__, __LINE__, "missing SOA"); 172 return (result); 173} 174 175/* Journaling */ 176 177/*% 178 * On-disk representation of a "pointer" to a journal entry. 179 * These are used in the journal header to locate the beginning 180 * and end of the journal, and in the journal index to locate 181 * other transactions. 182 */ 183typedef struct { 184 unsigned char serial[4]; /*%< SOA serial before update. */ 185 /* 186 * XXXRTH Should offset be 8 bytes? 187 * XXXDCL ... probably, since isc_offset_t is 8 bytes on many OSs. 188 * XXXAG ... but we will not be able to seek >2G anyway on many 189 * platforms as long as we are using fseek() rather 190 * than lseek(). 191 */ 192 unsigned char offset[4]; /*%< Offset from beginning of file. */ 193} journal_rawpos_t; 194 195 196/*% 197 * The header is of a fixed size, with some spare room for future 198 * extensions. 199 */ 200#define JOURNAL_HEADER_SIZE 64 /* Bytes. */ 201 202/*% 203 * The on-disk representation of the journal header. 204 * All numbers are stored in big-endian order. 205 */ 206typedef union { 207 struct { 208 /*% File format version ID. */ 209 unsigned char format[16]; 210 /*% Position of the first addressable transaction */ 211 journal_rawpos_t begin; 212 /*% Position of the next (yet nonexistent) transaction. */ 213 journal_rawpos_t end; 214 /*% Number of index entries following the header. */ 215 unsigned char index_size[4]; 216 } h; 217 /* Pad the header to a fixed size. */ 218 unsigned char pad[JOURNAL_HEADER_SIZE]; 219} journal_rawheader_t; 220 221/*% 222 * The on-disk representation of the transaction header. 223 * There is one of these at the beginning of each transaction. 224 */ 225typedef struct { 226 unsigned char size[4]; /*%< In bytes, excluding header. */ 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 */ 235typedef 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 */ 242typedef struct { 243 isc_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 250typedef struct { 251 unsigned char format[16]; 252 journal_pos_t begin; 253 journal_pos_t end; 254 isc_uint32_t index_size; 255} journal_header_t; 256 257/*% 258 * The in-core representation of the transaction header. 259 */ 260 261typedef struct { 262 isc_uint32_t size; 263 isc_uint32_t serial0; 264 isc_uint32_t serial1; 265} journal_xhdr_t; 266 267/*% 268 * The in-core representation of the RR header. 269 */ 270typedef struct { 271 isc_uint32_t size; 272} journal_rrhdr_t; 273 274 275/*% 276 * Initial contents to store in the header of a newly created 277 * journal file. 278 * 279 * The header starts with the magic string ";BIND LOG V9\n" 280 * to identify the file as a BIND 9 journal file. An ASCII 281 * identification string is used rather than a binary magic 282 * number to be consistent with BIND 8 (BIND 8 journal files 283 * are ASCII text files). 284 */ 285 286static journal_header_t 287initial_journal_header = { ";BIND LOG V9\n", { 0, 0 }, { 0, 0 }, 0 }; 288 289#define JOURNAL_EMPTY(h) ((h)->begin.offset == (h)->end.offset) 290 291typedef enum { 292 JOURNAL_STATE_INVALID, 293 JOURNAL_STATE_READ, 294 JOURNAL_STATE_WRITE, 295 JOURNAL_STATE_TRANSACTION 296} journal_state_t; 297 298struct dns_journal { 299 unsigned int magic; /*%< JOUR */ 300 isc_mem_t *mctx; /*%< Memory context */ 301 journal_state_t state; 302 const char *filename; /*%< Journal file name */ 303 FILE * fp; /*%< File handle */ 304 isc_offset_t offset; /*%< Current file offset */ 305 journal_header_t header; /*%< In-core journal header */ 306 unsigned char *rawindex; /*%< In-core buffer for journal index in on-disk format */ 307 journal_pos_t *index; /*%< In-core journal index */ 308 309 /*% Current transaction state (when writing). */ 310 struct { 311 unsigned int n_soa; /*%< Number of SOAs seen */ 312 journal_pos_t pos[2]; /*%< Begin/end position */ 313 } x; 314 315 /*% Iteration state (when reading). */ 316 struct { 317 /* These define the part of the journal we iterate over. */ 318 journal_pos_t bpos; /*%< Position before first, */ 319 journal_pos_t epos; /*%< and after last transaction */ 320 /* The rest is iterator state. */ 321 isc_uint32_t current_serial; /*%< Current SOA serial */ 322 isc_buffer_t source; /*%< Data from disk */ 323 isc_buffer_t target; /*%< Data from _fromwire check */ 324 dns_decompress_t dctx; /*%< Dummy decompression ctx */ 325 dns_name_t name; /*%< Current domain name */ 326 dns_rdata_t rdata; /*%< Current rdata */ 327 isc_uint32_t ttl; /*%< Current TTL */ 328 unsigned int xsize; /*%< Size of transaction data */ 329 unsigned int xpos; /*%< Current position in it */ 330 isc_result_t result; /*%< Result of last call */ 331 } it; 332}; 333 334#define DNS_JOURNAL_MAGIC ISC_MAGIC('J', 'O', 'U', 'R') 335#define DNS_JOURNAL_VALID(t) ISC_MAGIC_VALID(t, DNS_JOURNAL_MAGIC) 336 337static void 338journal_pos_decode(journal_rawpos_t *raw, journal_pos_t *cooked) { 339 cooked->serial = decode_uint32(raw->serial); 340 cooked->offset = decode_uint32(raw->offset); 341} 342 343static void 344journal_pos_encode(journal_rawpos_t *raw, journal_pos_t *cooked) { 345 encode_uint32(cooked->serial, raw->serial); 346 encode_uint32(cooked->offset, raw->offset); 347} 348 349static void 350journal_header_decode(journal_rawheader_t *raw, journal_header_t *cooked) { 351 INSIST(sizeof(cooked->format) == sizeof(raw->h.format)); 352 memcpy(cooked->format, raw->h.format, sizeof(cooked->format)); 353 journal_pos_decode(&raw->h.begin, &cooked->begin); 354 journal_pos_decode(&raw->h.end, &cooked->end); 355 cooked->index_size = decode_uint32(raw->h.index_size); 356} 357 358static void 359journal_header_encode(journal_header_t *cooked, journal_rawheader_t *raw) { 360 INSIST(sizeof(cooked->format) == sizeof(raw->h.format)); 361 memset(raw->pad, 0, sizeof(raw->pad)); 362 memcpy(raw->h.format, cooked->format, sizeof(raw->h.format)); 363 journal_pos_encode(&raw->h.begin, &cooked->begin); 364 journal_pos_encode(&raw->h.end, &cooked->end); 365 encode_uint32(cooked->index_size, raw->h.index_size); 366} 367 368/* 369 * Journal file I/O subroutines, with error checking and reporting. 370 */ 371static isc_result_t 372journal_seek(dns_journal_t *j, isc_uint32_t offset) { 373 isc_result_t result; 374 result = isc_stdio_seek(j->fp, (long)offset, SEEK_SET); 375 if (result != ISC_R_SUCCESS) { 376 isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR, 377 "%s: seek: %s", j->filename, 378 isc_result_totext(result)); 379 return (ISC_R_UNEXPECTED); 380 } 381 j->offset = offset; 382 return (ISC_R_SUCCESS); 383} 384 385static isc_result_t 386journal_read(dns_journal_t *j, void *mem, size_t nbytes) { 387 isc_result_t result; 388 389 result = isc_stdio_read(mem, 1, nbytes, j->fp, NULL); 390 if (result != ISC_R_SUCCESS) { 391 if (result == ISC_R_EOF) 392 return (ISC_R_NOMORE); 393 isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR, 394 "%s: read: %s", 395 j->filename, isc_result_totext(result)); 396 return (ISC_R_UNEXPECTED); 397 } 398 j->offset += nbytes; 399 return (ISC_R_SUCCESS); 400} 401 402static isc_result_t 403journal_write(dns_journal_t *j, void *mem, size_t nbytes) { 404 isc_result_t result; 405 406 result = isc_stdio_write(mem, 1, nbytes, j->fp, NULL); 407 if (result != ISC_R_SUCCESS) { 408 isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR, 409 "%s: write: %s", 410 j->filename, isc_result_totext(result)); 411 return (ISC_R_UNEXPECTED); 412 } 413 j->offset += nbytes; 414 return (ISC_R_SUCCESS); 415} 416 417static isc_result_t 418journal_fsync(dns_journal_t *j) { 419 isc_result_t result; 420 result = isc_stdio_flush(j->fp); 421 if (result != ISC_R_SUCCESS) { 422 isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR, 423 "%s: flush: %s", 424 j->filename, isc_result_totext(result)); 425 return (ISC_R_UNEXPECTED); 426 } 427 result = isc_stdio_sync(j->fp); 428 if (result != ISC_R_SUCCESS) { 429 isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR, 430 "%s: fsync: %s", 431 j->filename, isc_result_totext(result)); 432 return (ISC_R_UNEXPECTED); 433 } 434 return (ISC_R_SUCCESS); 435} 436 437/* 438 * Read/write a transaction header at the current file position. 439 */ 440 441static isc_result_t 442journal_read_xhdr(dns_journal_t *j, journal_xhdr_t *xhdr) { 443 journal_rawxhdr_t raw; 444 isc_result_t result; 445 result = journal_read(j, &raw, sizeof(raw)); 446 if (result != ISC_R_SUCCESS) 447 return (result); 448 xhdr->size = decode_uint32(raw.size); 449 xhdr->serial0 = decode_uint32(raw.serial0); 450 xhdr->serial1 = decode_uint32(raw.serial1); 451 return (ISC_R_SUCCESS); 452} 453 454static isc_result_t 455journal_write_xhdr(dns_journal_t *j, isc_uint32_t size, 456 isc_uint32_t serial0, isc_uint32_t serial1) 457{ 458 journal_rawxhdr_t raw; 459 encode_uint32(size, raw.size); 460 encode_uint32(serial0, raw.serial0); 461 encode_uint32(serial1, raw.serial1); 462 return (journal_write(j, &raw, sizeof(raw))); 463} 464 465 466/* 467 * Read an RR header at the current file position. 468 */ 469 470static isc_result_t 471journal_read_rrhdr(dns_journal_t *j, journal_rrhdr_t *rrhdr) { 472 journal_rawrrhdr_t raw; 473 isc_result_t result; 474 result = journal_read(j, &raw, sizeof(raw)); 475 if (result != ISC_R_SUCCESS) 476 return (result); 477 rrhdr->size = decode_uint32(raw.size); 478 return (ISC_R_SUCCESS); 479} 480 481static isc_result_t 482journal_file_create(isc_mem_t *mctx, const char *filename) { 483 FILE *fp = NULL; 484 isc_result_t result; 485 journal_header_t header; 486 journal_rawheader_t rawheader; 487 int index_size = 56; /* XXX configurable */ 488 int size; 489 void *mem; /* Memory for temporary index image. */ 490 491 INSIST(sizeof(journal_rawheader_t) == JOURNAL_HEADER_SIZE); 492 493 result = isc_stdio_open(filename, "wb", &fp); 494 if (result != ISC_R_SUCCESS) { 495 isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR, 496 "%s: create: %s", 497 filename, isc_result_totext(result)); 498 return (ISC_R_UNEXPECTED); 499 } 500 501 header = initial_journal_header; 502 header.index_size = index_size; 503 journal_header_encode(&header, &rawheader); 504 505 size = sizeof(journal_rawheader_t) + 506 index_size * sizeof(journal_rawpos_t); 507 508 mem = isc_mem_get(mctx, size); 509 if (mem == NULL) { 510 (void)isc_stdio_close(fp); 511 (void)isc_file_remove(filename); 512 return (ISC_R_NOMEMORY); 513 } 514 memset(mem, 0, size); 515 memcpy(mem, &rawheader, sizeof(rawheader)); 516 517 result = isc_stdio_write(mem, 1, (size_t) size, fp, NULL); 518 if (result != ISC_R_SUCCESS) { 519 isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR, 520 "%s: write: %s", 521 filename, isc_result_totext(result)); 522 (void)isc_stdio_close(fp); 523 (void)isc_file_remove(filename); 524 isc_mem_put(mctx, mem, size); 525 return (ISC_R_UNEXPECTED); 526 } 527 isc_mem_put(mctx, mem, size); 528 529 result = isc_stdio_close(fp); 530 if (result != ISC_R_SUCCESS) { 531 isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR, 532 "%s: close: %s", 533 filename, isc_result_totext(result)); 534 (void)isc_file_remove(filename); 535 return (ISC_R_UNEXPECTED); 536 } 537 538 return (ISC_R_SUCCESS); 539} 540 541static isc_result_t 542journal_open(isc_mem_t *mctx, const char *filename, isc_boolean_t write, 543 isc_boolean_t create, dns_journal_t **journalp) { 544 FILE *fp = NULL; 545 isc_result_t result; 546 journal_rawheader_t rawheader; 547 dns_journal_t *j; 548 549 INSIST(journalp != NULL && *journalp == NULL); 550 j = isc_mem_get(mctx, sizeof(*j)); 551 if (j == NULL) 552 return (ISC_R_NOMEMORY); 553 554 j->mctx = mctx; 555 j->state = JOURNAL_STATE_INVALID; 556 j->fp = NULL; 557 j->filename = filename; 558 j->index = NULL; 559 j->rawindex = NULL; 560 561 result = isc_stdio_open(j->filename, write ? "rb+" : "rb", &fp); 562 563 if (result == ISC_R_FILENOTFOUND) { 564 if (create) { 565 isc_log_write(JOURNAL_COMMON_LOGARGS, 566 ISC_LOG_INFO, 567 "journal file %s does not exist, " 568 "creating it", 569 j->filename); 570 CHECK(journal_file_create(mctx, filename)); 571 /* 572 * Retry. 573 */ 574 result = isc_stdio_open(j->filename, "rb+", &fp); 575 } else { 576 FAIL(ISC_R_NOTFOUND); 577 } 578 } 579 if (result != ISC_R_SUCCESS) { 580 isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR, 581 "%s: open: %s", 582 j->filename, isc_result_totext(result)); 583 FAIL(ISC_R_UNEXPECTED); 584 } 585 586 j->fp = fp; 587 588 /* 589 * Set magic early so that seek/read can succeed. 590 */ 591 j->magic = DNS_JOURNAL_MAGIC; 592 593 CHECK(journal_seek(j, 0)); 594 CHECK(journal_read(j, &rawheader, sizeof(rawheader))); 595 596 if (memcmp(rawheader.h.format, initial_journal_header.format, 597 sizeof(initial_journal_header.format)) != 0) { 598 isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR, 599 "%s: journal format not recognized", 600 j->filename); 601 FAIL(ISC_R_UNEXPECTED); 602 } 603 journal_header_decode(&rawheader, &j->header); 604 605 /* 606 * If there is an index, read the raw index into a dynamically 607 * allocated buffer and then convert it into a cooked index. 608 */ 609 if (j->header.index_size != 0) { 610 unsigned int i; 611 unsigned int rawbytes; 612 unsigned char *p; 613 614 rawbytes = j->header.index_size * sizeof(journal_rawpos_t); 615 j->rawindex = isc_mem_get(mctx, rawbytes); 616 if (j->rawindex == NULL) 617 FAIL(ISC_R_NOMEMORY); 618 619 CHECK(journal_read(j, j->rawindex, rawbytes)); 620 621 j->index = isc_mem_get(mctx, j->header.index_size * 622 sizeof(journal_pos_t)); 623 if (j->index == NULL) 624 FAIL(ISC_R_NOMEMORY); 625 626 p = j->rawindex; 627 for (i = 0; i < j->header.index_size; i++) { 628 j->index[i].serial = decode_uint32(p); 629 p += 4; 630 j->index[i].offset = decode_uint32(p); 631 p += 4; 632 } 633 INSIST(p == j->rawindex + rawbytes); 634 } 635 j->offset = -1; /* Invalid, must seek explicitly. */ 636 637 /* 638 * Initialize the iterator. 639 */ 640 dns_name_init(&j->it.name, NULL); 641 dns_rdata_init(&j->it.rdata); 642 643 /* 644 * Set up empty initial buffers for unchecked and checked 645 * wire format RR data. They will be reallocated 646 * later. 647 */ 648 isc_buffer_init(&j->it.source, NULL, 0); 649 isc_buffer_init(&j->it.target, NULL, 0); 650 dns_decompress_init(&j->it.dctx, -1, DNS_DECOMPRESS_NONE); 651 652 j->state = 653 write ? JOURNAL_STATE_WRITE : JOURNAL_STATE_READ; 654 655 *journalp = j; 656 return (ISC_R_SUCCESS); 657 658 failure: 659 j->magic = 0; 660 if (j->index != NULL) { 661 isc_mem_put(j->mctx, j->index, j->header.index_size * 662 sizeof(journal_rawpos_t)); 663 j->index = NULL; 664 } 665 if (j->fp != NULL) 666 (void)isc_stdio_close(j->fp); 667 isc_mem_put(j->mctx, j, sizeof(*j)); 668 return (result); 669} 670 671isc_result_t 672dns_journal_open(isc_mem_t *mctx, const char *filename, isc_boolean_t write, 673 dns_journal_t **journalp) { 674 isc_result_t result; 675 int namelen; 676 char backup[1024]; 677 678 result = journal_open(mctx, filename, write, write, journalp); 679 if (result == ISC_R_NOTFOUND) { 680 namelen = strlen(filename); 681 if (namelen > 4 && strcmp(filename + namelen - 4, ".jnl") == 0) 682 namelen -= 4; 683 684 result = isc_string_printf(backup, sizeof(backup), "%.*s.jbk", 685 namelen, filename); 686 if (result != ISC_R_SUCCESS) 687 return (result); 688 result = journal_open(mctx, backup, write, write, journalp); 689 } 690 return (result); 691} 692 693/* 694 * A comparison function defining the sorting order for 695 * entries in the IXFR-style journal file. 696 * 697 * The IXFR format requires that deletions are sorted before 698 * additions, and within either one, SOA records are sorted 699 * before others. 700 * 701 * Also sort the non-SOA records by type as a courtesy to the 702 * server receiving the IXFR - it may help reduce the amount of 703 * rdataset merging it has to do. 704 */ 705static int 706ixfr_order(const void *av, const void *bv) { 707 dns_difftuple_t const * const *ap = av; 708 dns_difftuple_t const * const *bp = bv; 709 dns_difftuple_t const *a = *ap; 710 dns_difftuple_t const *b = *bp; 711 int r; 712 int bop = 0, aop = 0; 713 714 switch (a->op) { 715 case DNS_DIFFOP_DEL: 716 case DNS_DIFFOP_DELRESIGN: 717 aop = 1; 718 break; 719 case DNS_DIFFOP_ADD: 720 case DNS_DIFFOP_ADDRESIGN: 721 aop = 0; 722 break; 723 default: 724 INSIST(0); 725 } 726 727 switch (b->op) { 728 case DNS_DIFFOP_DEL: 729 case DNS_DIFFOP_DELRESIGN: 730 bop = 1; 731 break; 732 case DNS_DIFFOP_ADD: 733 case DNS_DIFFOP_ADDRESIGN: 734 bop = 0; 735 break; 736 default: 737 INSIST(0); 738 } 739 740 r = bop - aop; 741 if (r != 0) 742 return (r); 743 744 r = (b->rdata.type == dns_rdatatype_soa) - 745 (a->rdata.type == dns_rdatatype_soa); 746 if (r != 0) 747 return (r); 748 749 r = (a->rdata.type - b->rdata.type); 750 return (r); 751} 752 753/* 754 * Advance '*pos' to the next journal transaction. 755 * 756 * Requires: 757 * *pos refers to a valid journal transaction. 758 * 759 * Ensures: 760 * When ISC_R_SUCCESS is returned, 761 * *pos refers to the next journal transaction. 762 * 763 * Returns one of: 764 * 765 * ISC_R_SUCCESS 766 * ISC_R_NOMORE *pos pointed at the last transaction 767 * Other results due to file errors are possible. 768 */ 769static isc_result_t 770journal_next(dns_journal_t *j, journal_pos_t *pos) { 771 isc_result_t result; 772 journal_xhdr_t xhdr; 773 REQUIRE(DNS_JOURNAL_VALID(j)); 774 775 result = journal_seek(j, pos->offset); 776 if (result != ISC_R_SUCCESS) 777 return (result); 778 779 if (pos->serial == j->header.end.serial) 780 return (ISC_R_NOMORE); 781 /* 782 * Read the header of the current transaction. 783 * This will return ISC_R_NOMORE if we are at EOF. 784 */ 785 result = journal_read_xhdr(j, &xhdr); 786 if (result != ISC_R_SUCCESS) 787 return (result); 788 789 /* 790 * Check serial number consistency. 791 */ 792 if (xhdr.serial0 != pos->serial) { 793 isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR, 794 "%s: journal file corrupt: " 795 "expected serial %u, got %u", 796 j->filename, pos->serial, xhdr.serial0); 797 return (ISC_R_UNEXPECTED); 798 } 799 800 /* 801 * Check for offset wraparound. 802 */ 803 if ((isc_offset_t)(pos->offset + sizeof(journal_rawxhdr_t) + xhdr.size) 804 < pos->offset) { 805 isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR, 806 "%s: offset too large", j->filename); 807 return (ISC_R_UNEXPECTED); 808 } 809 810 pos->offset += sizeof(journal_rawxhdr_t) + xhdr.size; 811 pos->serial = xhdr.serial1; 812 return (ISC_R_SUCCESS); 813} 814 815/* 816 * If the index of the journal 'j' contains an entry "better" 817 * than '*best_guess', replace '*best_guess' with it. 818 * 819 * "Better" means having a serial number closer to 'serial' 820 * but not greater than 'serial'. 821 */ 822static void 823index_find(dns_journal_t *j, isc_uint32_t serial, journal_pos_t *best_guess) { 824 unsigned int i; 825 if (j->index == NULL) 826 return; 827 for (i = 0; i < j->header.index_size; i++) { 828 if (POS_VALID(j->index[i]) && 829 DNS_SERIAL_GE(serial, j->index[i].serial) && 830 DNS_SERIAL_GT(j->index[i].serial, best_guess->serial)) 831 *best_guess = j->index[i]; 832 } 833} 834 835/* 836 * Add a new index entry. If there is no room, make room by removing 837 * the odd-numbered entries and compacting the others into the first 838 * half of the index. This decimates old index entries exponentially 839 * over time, so that the index always contains a much larger fraction 840 * of recent serial numbers than of old ones. This is deliberate - 841 * most index searches are for outgoing IXFR, and IXFR tends to request 842 * recent versions more often than old ones. 843 */ 844static void 845index_add(dns_journal_t *j, journal_pos_t *pos) { 846 unsigned int i; 847 if (j->index == NULL) 848 return; 849 /* 850 * Search for a vacant position. 851 */ 852 for (i = 0; i < j->header.index_size; i++) { 853 if (! POS_VALID(j->index[i])) 854 break; 855 } 856 if (i == j->header.index_size) { 857 unsigned int k = 0; 858 /* 859 * Found no vacant position. Make some room. 860 */ 861 for (i = 0; i < j->header.index_size; i += 2) { 862 j->index[k++] = j->index[i]; 863 } 864 i = k; /* 'i' identifies the first vacant position. */ 865 while (k < j->header.index_size) { 866 POS_INVALIDATE(j->index[k]); 867 k++; 868 } 869 } 870 INSIST(i < j->header.index_size); 871 INSIST(! POS_VALID(j->index[i])); 872 873 /* 874 * Store the new index entry. 875 */ 876 j->index[i] = *pos; 877} 878 879/* 880 * Invalidate any existing index entries that could become 881 * ambiguous when a new transaction with number 'serial' is added. 882 */ 883static void 884index_invalidate(dns_journal_t *j, isc_uint32_t serial) { 885 unsigned int i; 886 if (j->index == NULL) 887 return; 888 for (i = 0; i < j->header.index_size; i++) { 889 if (! DNS_SERIAL_GT(serial, j->index[i].serial)) 890 POS_INVALIDATE(j->index[i]); 891 } 892} 893 894/* 895 * Try to find a transaction with initial serial number 'serial' 896 * in the journal 'j'. 897 * 898 * If found, store its position at '*pos' and return ISC_R_SUCCESS. 899 * 900 * If 'serial' is current (= the ending serial number of the 901 * last transaction in the journal), set '*pos' to 902 * the position immediately following the last transaction and 903 * return ISC_R_SUCCESS. 904 * 905 * If 'serial' is within the range of addressable serial numbers 906 * covered by the journal but that particular serial number is missing 907 * (from the journal, not just from the index), return ISC_R_NOTFOUND. 908 * 909 * If 'serial' is outside the range of addressable serial numbers 910 * covered by the journal, return ISC_R_RANGE. 911 * 912 */ 913static isc_result_t 914journal_find(dns_journal_t *j, isc_uint32_t serial, journal_pos_t *pos) { 915 isc_result_t result; 916 journal_pos_t current_pos; 917 REQUIRE(DNS_JOURNAL_VALID(j)); 918 919 if (DNS_SERIAL_GT(j->header.begin.serial, serial)) 920 return (ISC_R_RANGE); 921 if (DNS_SERIAL_GT(serial, j->header.end.serial)) 922 return (ISC_R_RANGE); 923 if (serial == j->header.end.serial) { 924 *pos = j->header.end; 925 return (ISC_R_SUCCESS); 926 } 927 928 current_pos = j->header.begin; 929 index_find(j, serial, ¤t_pos); 930 931 while (current_pos.serial != serial) { 932 if (DNS_SERIAL_GT(current_pos.serial, serial)) 933 return (ISC_R_NOTFOUND); 934 result = journal_next(j, ¤t_pos); 935 if (result != ISC_R_SUCCESS) 936 return (result); 937 } 938 *pos = current_pos; 939 return (ISC_R_SUCCESS); 940} 941 942isc_result_t 943dns_journal_begin_transaction(dns_journal_t *j) { 944 isc_uint32_t offset; 945 isc_result_t result; 946 journal_rawxhdr_t hdr; 947 948 REQUIRE(DNS_JOURNAL_VALID(j)); 949 REQUIRE(j->state == JOURNAL_STATE_WRITE); 950 951 /* 952 * Find the file offset where the new transaction should 953 * be written, and seek there. 954 */ 955 if (JOURNAL_EMPTY(&j->header)) { 956 offset = sizeof(journal_rawheader_t) + 957 j->header.index_size * sizeof(journal_rawpos_t); 958 } else { 959 offset = j->header.end.offset; 960 } 961 j->x.pos[0].offset = offset; 962 j->x.pos[1].offset = offset; /* Initial value, will be incremented. */ 963 j->x.n_soa = 0; 964 965 CHECK(journal_seek(j, offset)); 966 967 /* 968 * Write a dummy transaction header of all zeroes to reserve 969 * space. It will be filled in when the transaction is 970 * finished. 971 */ 972 memset(&hdr, 0, sizeof(hdr)); 973 CHECK(journal_write(j, &hdr, sizeof(hdr))); 974 j->x.pos[1].offset = j->offset; 975 976 j->state = JOURNAL_STATE_TRANSACTION; 977 result = ISC_R_SUCCESS; 978 failure: 979 return (result); 980} 981 982isc_result_t 983dns_journal_writediff(dns_journal_t *j, dns_diff_t *diff) { 984 dns_difftuple_t *t; 985 isc_buffer_t buffer; 986 void *mem = NULL; 987 unsigned int size; 988 isc_result_t result; 989 isc_region_t used; 990 991 REQUIRE(DNS_DIFF_VALID(diff)); 992 REQUIRE(j->state == JOURNAL_STATE_TRANSACTION); 993 994 isc_log_write(JOURNAL_DEBUG_LOGARGS(3), "writing to journal"); 995 (void)dns_diff_print(diff, NULL); 996 997 /* 998 * Pass 1: determine the buffer size needed, and 999 * keep track of SOA serial numbers. 1000 */ 1001 size = 0; 1002 for (t = ISC_LIST_HEAD(diff->tuples); t != NULL; 1003 t = ISC_LIST_NEXT(t, link)) 1004 { 1005 if (t->rdata.type == dns_rdatatype_soa) { 1006 if (j->x.n_soa < 2) 1007 j->x.pos[j->x.n_soa].serial = 1008 dns_soa_getserial(&t->rdata); 1009 j->x.n_soa++; 1010 } 1011 size += sizeof(journal_rawrrhdr_t); 1012 size += t->name.length; /* XXX should have access macro? */ 1013 size += 10; 1014 size += t->rdata.length; 1015 } 1016 1017 mem = isc_mem_get(j->mctx, size); 1018 if (mem == NULL) 1019 return (ISC_R_NOMEMORY); 1020 1021 isc_buffer_init(&buffer, mem, size); 1022 1023 /* 1024 * Pass 2. Write RRs to buffer. 1025 */ 1026 for (t = ISC_LIST_HEAD(diff->tuples); t != NULL; 1027 t = ISC_LIST_NEXT(t, link)) 1028 { 1029 /* 1030 * Write the RR header. 1031 */ 1032 isc_buffer_putuint32(&buffer, t->name.length + 10 + 1033 t->rdata.length); 1034 /* 1035 * Write the owner name, RR header, and RR data. 1036 */ 1037 isc_buffer_putmem(&buffer, t->name.ndata, t->name.length); 1038 isc_buffer_putuint16(&buffer, t->rdata.type); 1039 isc_buffer_putuint16(&buffer, t->rdata.rdclass); 1040 isc_buffer_putuint32(&buffer, t->ttl); 1041 INSIST(t->rdata.length < 65536); 1042 isc_buffer_putuint16(&buffer, (isc_uint16_t)t->rdata.length); 1043 INSIST(isc_buffer_availablelength(&buffer) >= t->rdata.length); 1044 isc_buffer_putmem(&buffer, t->rdata.data, t->rdata.length); 1045 } 1046 1047 isc_buffer_usedregion(&buffer, &used); 1048 INSIST(used.length == size); 1049 1050 j->x.pos[1].offset += used.length; 1051 1052 /* 1053 * Write the buffer contents to the journal file. 1054 */ 1055 CHECK(journal_write(j, used.base, used.length)); 1056 1057 result = ISC_R_SUCCESS; 1058 1059 failure: 1060 if (mem != NULL) 1061 isc_mem_put(j->mctx, mem, size); 1062 return (result); 1063 1064} 1065 1066isc_result_t 1067dns_journal_commit(dns_journal_t *j) { 1068 isc_result_t result; 1069 journal_rawheader_t rawheader; 1070 1071 REQUIRE(DNS_JOURNAL_VALID(j)); 1072 REQUIRE(j->state == JOURNAL_STATE_TRANSACTION); 1073 1074 /* 1075 * Perform some basic consistency checks. 1076 */ 1077 if (j->x.n_soa != 2) { 1078 isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR, 1079 "%s: malformed transaction: %d SOAs", 1080 j->filename, j->x.n_soa); 1081 return (ISC_R_UNEXPECTED); 1082 } 1083 if (! (DNS_SERIAL_GT(j->x.pos[1].serial, j->x.pos[0].serial) || 1084 (bind8_compat && 1085 j->x.pos[1].serial == j->x.pos[0].serial))) 1086 { 1087 isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR, 1088 "%s: malformed transaction: serial number " 1089 "would decrease", j->filename); 1090 return (ISC_R_UNEXPECTED); 1091 } 1092 if (! JOURNAL_EMPTY(&j->header)) { 1093 if (j->x.pos[0].serial != j->header.end.serial) { 1094 isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR, 1095 "malformed transaction: " 1096 "%s last serial %u != " 1097 "transaction first serial %u", 1098 j->filename, 1099 j->header.end.serial, 1100 j->x.pos[0].serial); 1101 return (ISC_R_UNEXPECTED); 1102 } 1103 } 1104 1105 /* 1106 * Some old journal entries may become non-addressable 1107 * when we increment the current serial number. Purge them 1108 * by stepping header.begin forward to the first addressable 1109 * transaction. Also purge them from the index. 1110 */ 1111 if (! JOURNAL_EMPTY(&j->header)) { 1112 while (! DNS_SERIAL_GT(j->x.pos[1].serial, 1113 j->header.begin.serial)) { 1114 CHECK(journal_next(j, &j->header.begin)); 1115 } 1116 index_invalidate(j, j->x.pos[1].serial); 1117 } 1118#ifdef notyet 1119 if (DNS_SERIAL_GT(last_dumped_serial, j->x.pos[1].serial)) { 1120 force_dump(...); 1121 } 1122#endif 1123 1124 /* 1125 * Commit the transaction data to stable storage. 1126 */ 1127 CHECK(journal_fsync(j)); 1128 1129 /* 1130 * Update the transaction header. 1131 */ 1132 CHECK(journal_seek(j, j->x.pos[0].offset)); 1133 CHECK(journal_write_xhdr(j, (j->x.pos[1].offset - j->x.pos[0].offset) - 1134 sizeof(journal_rawxhdr_t), 1135 j->x.pos[0].serial, j->x.pos[1].serial)); 1136 1137 /* 1138 * Update the journal header. 1139 */ 1140 if (JOURNAL_EMPTY(&j->header)) { 1141 j->header.begin = j->x.pos[0]; 1142 } 1143 j->header.end = j->x.pos[1]; 1144 journal_header_encode(&j->header, &rawheader); 1145 CHECK(journal_seek(j, 0)); 1146 CHECK(journal_write(j, &rawheader, sizeof(rawheader))); 1147 1148 /* 1149 * Update the index. 1150 */ 1151 index_add(j, &j->x.pos[0]); 1152 1153 /* 1154 * Convert the index into on-disk format and write 1155 * it to disk. 1156 */ 1157 CHECK(index_to_disk(j)); 1158 1159 /* 1160 * Commit the header to stable storage. 1161 */ 1162 CHECK(journal_fsync(j)); 1163 1164 /* 1165 * We no longer have a transaction open. 1166 */ 1167 j->state = JOURNAL_STATE_WRITE; 1168 1169 result = ISC_R_SUCCESS; 1170 1171 failure: 1172 return (result); 1173} 1174 1175isc_result_t 1176dns_journal_write_transaction(dns_journal_t *j, dns_diff_t *diff) { 1177 isc_result_t result; 1178 CHECK(dns_diff_sort(diff, ixfr_order)); 1179 CHECK(dns_journal_begin_transaction(j)); 1180 CHECK(dns_journal_writediff(j, diff)); 1181 CHECK(dns_journal_commit(j)); 1182 result = ISC_R_SUCCESS; 1183 failure: 1184 return (result); 1185} 1186 1187void 1188dns_journal_destroy(dns_journal_t **journalp) { 1189 dns_journal_t *j = *journalp; 1190 REQUIRE(DNS_JOURNAL_VALID(j)); 1191 1192 j->it.result = ISC_R_FAILURE; 1193 dns_name_invalidate(&j->it.name); 1194 dns_decompress_invalidate(&j->it.dctx); 1195 if (j->rawindex != NULL) 1196 isc_mem_put(j->mctx, j->rawindex, j->header.index_size * 1197 sizeof(journal_rawpos_t)); 1198 if (j->index != NULL) 1199 isc_mem_put(j->mctx, j->index, j->header.index_size * 1200 sizeof(journal_pos_t)); 1201 if (j->it.target.base != NULL) 1202 isc_mem_put(j->mctx, j->it.target.base, j->it.target.length); 1203 if (j->it.source.base != NULL) 1204 isc_mem_put(j->mctx, j->it.source.base, j->it.source.length); 1205 1206 if (j->fp != NULL) 1207 (void)isc_stdio_close(j->fp); 1208 j->magic = 0; 1209 isc_mem_put(j->mctx, j, sizeof(*j)); 1210 *journalp = NULL; 1211} 1212 1213/* 1214 * Roll the open journal 'j' into the database 'db'. 1215 * A new database version will be created. 1216 */ 1217 1218/* XXX Share code with incoming IXFR? */ 1219 1220static isc_result_t 1221roll_forward(dns_journal_t *j, dns_db_t *db, unsigned int options) { 1222 isc_buffer_t source; /* Transaction data from disk */ 1223 isc_buffer_t target; /* Ditto after _fromwire check */ 1224 isc_uint32_t db_serial; /* Database SOA serial */ 1225 isc_uint32_t end_serial; /* Last journal SOA serial */ 1226 isc_result_t result; 1227 dns_dbversion_t *ver = NULL; 1228 journal_pos_t pos; 1229 dns_diff_t diff; 1230 unsigned int n_soa = 0; 1231 unsigned int n_put = 0; 1232 dns_diffop_t op; 1233 1234 REQUIRE(DNS_JOURNAL_VALID(j)); 1235 REQUIRE(DNS_DB_VALID(db)); 1236 1237 dns_diff_init(j->mctx, &diff); 1238 1239 /* 1240 * Set up empty initial buffers for unchecked and checked 1241 * wire format transaction data. They will be reallocated 1242 * later. 1243 */ 1244 isc_buffer_init(&source, NULL, 0); 1245 isc_buffer_init(&target, NULL, 0); 1246 1247 /* 1248 * Create the new database version. 1249 */ 1250 CHECK(dns_db_newversion(db, &ver)); 1251 1252 /* 1253 * Get the current database SOA serial number. 1254 */ 1255 CHECK(dns_db_getsoaserial(db, ver, &db_serial)); 1256 1257 /* 1258 * Locate a journal entry for the current database serial. 1259 */ 1260 CHECK(journal_find(j, db_serial, &pos)); 1261 /* 1262 * XXX do more drastic things, like marking zone stale, 1263 * if this fails? 1264 */ 1265 /* 1266 * XXXRTH The zone code should probably mark the zone as bad and 1267 * scream loudly into the log if this is a dynamic update 1268 * log reply that failed. 1269 */ 1270 1271 end_serial = dns_journal_last_serial(j); 1272 if (db_serial == end_serial) 1273 CHECK(DNS_R_UPTODATE); 1274 1275 CHECK(dns_journal_iter_init(j, db_serial, end_serial)); 1276 1277 for (result = dns_journal_first_rr(j); 1278 result == ISC_R_SUCCESS; 1279 result = dns_journal_next_rr(j)) 1280 { 1281 dns_name_t *name; 1282 isc_uint32_t ttl; 1283 dns_rdata_t *rdata; 1284 dns_difftuple_t *tuple = NULL; 1285 1286 name = NULL; 1287 rdata = NULL; 1288 dns_journal_current_rr(j, &name, &ttl, &rdata); 1289 1290 if (rdata->type == dns_rdatatype_soa) { 1291 n_soa++; 1292 if (n_soa == 2) 1293 db_serial = j->it.current_serial; 1294 } 1295 1296 if (n_soa == 3) 1297 n_soa = 1; 1298 if (n_soa == 0) { 1299 isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR, 1300 "%s: journal file corrupt: missing " 1301 "initial SOA", j->filename); 1302 FAIL(ISC_R_UNEXPECTED); 1303 } 1304 if ((options & DNS_JOURNALOPT_RESIGN) != 0) 1305 op = (n_soa == 1) ? DNS_DIFFOP_DELRESIGN : 1306 DNS_DIFFOP_ADDRESIGN; 1307 else 1308 op = (n_soa == 1) ? DNS_DIFFOP_DEL : DNS_DIFFOP_ADD; 1309 1310 CHECK(dns_difftuple_create(diff.mctx, op, name, ttl, rdata, 1311 &tuple)); 1312 dns_diff_append(&diff, &tuple); 1313 1314 if (++n_put > 100) { 1315 isc_log_write(JOURNAL_DEBUG_LOGARGS(3), 1316 "%s: applying diff to database (%u)", 1317 j->filename, db_serial); 1318 (void)dns_diff_print(&diff, NULL); 1319 CHECK(dns_diff_apply(&diff, db, ver)); 1320 dns_diff_clear(&diff); 1321 n_put = 0; 1322 } 1323 } 1324 if (result == ISC_R_NOMORE) 1325 result = ISC_R_SUCCESS; 1326 CHECK(result); 1327 1328 if (n_put != 0) { 1329 isc_log_write(JOURNAL_DEBUG_LOGARGS(3), 1330 "%s: applying final diff to database (%u)", 1331 j->filename, db_serial); 1332 (void)dns_diff_print(&diff, NULL); 1333 CHECK(dns_diff_apply(&diff, db, ver)); 1334 dns_diff_clear(&diff); 1335 } 1336 1337 failure: 1338 if (ver != NULL) 1339 dns_db_closeversion(db, &ver, result == ISC_R_SUCCESS ? 1340 ISC_TRUE : ISC_FALSE); 1341 1342 if (source.base != NULL) 1343 isc_mem_put(j->mctx, source.base, source.length); 1344 if (target.base != NULL) 1345 isc_mem_put(j->mctx, target.base, target.length); 1346 1347 dns_diff_clear(&diff); 1348 1349 return (result); 1350} 1351 1352isc_result_t 1353dns_journal_rollforward(isc_mem_t *mctx, dns_db_t *db, 1354 unsigned int options, const char *filename) 1355{ 1356 dns_journal_t *j; 1357 isc_result_t result; 1358 1359 REQUIRE(DNS_DB_VALID(db)); 1360 REQUIRE(filename != NULL); 1361 1362 j = NULL; 1363 result = dns_journal_open(mctx, filename, ISC_FALSE, &j); 1364 if (result == ISC_R_NOTFOUND) { 1365 isc_log_write(JOURNAL_DEBUG_LOGARGS(3), 1366 "no journal file, but that's OK"); 1367 return (DNS_R_NOJOURNAL); 1368 } 1369 if (result != ISC_R_SUCCESS) 1370 return (result); 1371 if (JOURNAL_EMPTY(&j->header)) 1372 result = DNS_R_UPTODATE; 1373 else 1374 result = roll_forward(j, db, options); 1375 1376 dns_journal_destroy(&j); 1377 1378 return (result); 1379} 1380 1381isc_result_t 1382dns_journal_print(isc_mem_t *mctx, const char *filename, FILE *file) { 1383 dns_journal_t *j; 1384 isc_buffer_t source; /* Transaction data from disk */ 1385 isc_buffer_t target; /* Ditto after _fromwire check */ 1386 isc_uint32_t start_serial; /* Database SOA serial */ 1387 isc_uint32_t end_serial; /* Last journal SOA serial */ 1388 isc_result_t result; 1389 dns_diff_t diff; 1390 unsigned int n_soa = 0; 1391 unsigned int n_put = 0; 1392 1393 REQUIRE(filename != NULL); 1394 1395 j = NULL; 1396 result = dns_journal_open(mctx, filename, ISC_FALSE, &j); 1397 if (result == ISC_R_NOTFOUND) { 1398 isc_log_write(JOURNAL_DEBUG_LOGARGS(3), "no journal file"); 1399 return (DNS_R_NOJOURNAL); 1400 } 1401 1402 if (result != ISC_R_SUCCESS) { 1403 isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR, 1404 "journal open failure: %s: %s", 1405 isc_result_totext(result), filename); 1406 return (result); 1407 } 1408 1409 dns_diff_init(j->mctx, &diff); 1410 1411 /* 1412 * Set up empty initial buffers for unchecked and checked 1413 * wire format transaction data. They will be reallocated 1414 * later. 1415 */ 1416 isc_buffer_init(&source, NULL, 0); 1417 isc_buffer_init(&target, NULL, 0); 1418 1419 start_serial = dns_journal_first_serial(j); 1420 end_serial = dns_journal_last_serial(j); 1421 1422 CHECK(dns_journal_iter_init(j, start_serial, end_serial)); 1423 1424 for (result = dns_journal_first_rr(j); 1425 result == ISC_R_SUCCESS; 1426 result = dns_journal_next_rr(j)) 1427 { 1428 dns_name_t *name; 1429 isc_uint32_t ttl; 1430 dns_rdata_t *rdata; 1431 dns_difftuple_t *tuple = NULL; 1432 1433 name = NULL; 1434 rdata = NULL; 1435 dns_journal_current_rr(j, &name, &ttl, &rdata); 1436 1437 if (rdata->type == dns_rdatatype_soa) 1438 n_soa++; 1439 1440 if (n_soa == 3) 1441 n_soa = 1; 1442 if (n_soa == 0) { 1443 isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR, 1444 "%s: journal file corrupt: missing " 1445 "initial SOA", j->filename); 1446 FAIL(ISC_R_UNEXPECTED); 1447 } 1448 CHECK(dns_difftuple_create(diff.mctx, n_soa == 1 ? 1449 DNS_DIFFOP_DEL : DNS_DIFFOP_ADD, 1450 name, ttl, rdata, &tuple)); 1451 dns_diff_append(&diff, &tuple); 1452 1453 if (++n_put > 100) { 1454 result = dns_diff_print(&diff, file); 1455 dns_diff_clear(&diff); 1456 n_put = 0; 1457 if (result != ISC_R_SUCCESS) 1458 break; 1459 } 1460 } 1461 if (result == ISC_R_NOMORE) 1462 result = ISC_R_SUCCESS; 1463 CHECK(result); 1464 1465 if (n_put != 0) { 1466 result = dns_diff_print(&diff, file); 1467 dns_diff_clear(&diff); 1468 } 1469 goto cleanup; 1470 1471 failure: 1472 isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR, 1473 "%s: cannot print: journal file corrupt", j->filename); 1474 1475 cleanup: 1476 if (source.base != NULL) 1477 isc_mem_put(j->mctx, source.base, source.length); 1478 if (target.base != NULL) 1479 isc_mem_put(j->mctx, target.base, target.length); 1480 1481 dns_diff_clear(&diff); 1482 dns_journal_destroy(&j); 1483 1484 return (result); 1485} 1486 1487/**************************************************************************/ 1488/* 1489 * Miscellaneous accessors. 1490 */ 1491isc_uint32_t dns_journal_first_serial(dns_journal_t *j) { 1492 return (j->header.begin.serial); 1493} 1494 1495isc_uint32_t dns_journal_last_serial(dns_journal_t *j) { 1496 return (j->header.end.serial); 1497} 1498 1499/**************************************************************************/ 1500/* 1501 * Iteration support. 1502 * 1503 * When serving an outgoing IXFR, we transmit a part the journal starting 1504 * at the serial number in the IXFR request and ending at the serial 1505 * number that is current when the IXFR request arrives. The ending 1506 * serial number is not necessarily at the end of the journal: 1507 * the journal may grow while the IXFR is in progress, but we stop 1508 * when we reach the serial number that was current when the IXFR started. 1509 */ 1510 1511static isc_result_t read_one_rr(dns_journal_t *j); 1512 1513/* 1514 * Make sure the buffer 'b' is has at least 'size' bytes 1515 * allocated, and clear it. 1516 * 1517 * Requires: 1518 * Either b->base is NULL, or it points to b->length bytes of memory 1519 * previously allocated by isc_mem_get(). 1520 */ 1521 1522static isc_result_t 1523size_buffer(isc_mem_t *mctx, isc_buffer_t *b, unsigned size) { 1524 if (b->length < size) { 1525 void *mem = isc_mem_get(mctx, size); 1526 if (mem == NULL) 1527 return (ISC_R_NOMEMORY); 1528 if (b->base != NULL) 1529 isc_mem_put(mctx, b->base, b->length); 1530 b->base = mem; 1531 b->length = size; 1532 } 1533 isc_buffer_clear(b); 1534 return (ISC_R_SUCCESS); 1535} 1536 1537isc_result_t 1538dns_journal_iter_init(dns_journal_t *j, 1539 isc_uint32_t begin_serial, isc_uint32_t end_serial) 1540{ 1541 isc_result_t result; 1542 1543 CHECK(journal_find(j, begin_serial, &j->it.bpos)); 1544 INSIST(j->it.bpos.serial == begin_serial); 1545 1546 CHECK(journal_find(j, end_serial, &j->it.epos)); 1547 INSIST(j->it.epos.serial == end_serial); 1548 1549 result = ISC_R_SUCCESS; 1550 failure: 1551 j->it.result = result; 1552 return (j->it.result); 1553} 1554 1555 1556isc_result_t 1557dns_journal_first_rr(dns_journal_t *j) { 1558 isc_result_t result; 1559 1560 /* 1561 * Seek to the beginning of the first transaction we are 1562 * interested in. 1563 */ 1564 CHECK(journal_seek(j, j->it.bpos.offset)); 1565 j->it.current_serial = j->it.bpos.serial; 1566 1567 j->it.xsize = 0; /* We have no transaction data yet... */ 1568 j->it.xpos = 0; /* ...and haven't used any of it. */ 1569 1570 return (read_one_rr(j)); 1571 1572 failure: 1573 return (result); 1574} 1575 1576static isc_result_t 1577read_one_rr(dns_journal_t *j) { 1578 isc_result_t result; 1579 1580 dns_rdatatype_t rdtype; 1581 dns_rdataclass_t rdclass; 1582 unsigned int rdlen; 1583 isc_uint32_t ttl; 1584 journal_xhdr_t xhdr; 1585 journal_rrhdr_t rrhdr; 1586 1587 INSIST(j->offset <= j->it.epos.offset); 1588 if (j->offset == j->it.epos.offset) 1589 return (ISC_R_NOMORE); 1590 if (j->it.xpos == j->it.xsize) { 1591 /* 1592 * We are at a transaction boundary. 1593 * Read another transaction header. 1594 */ 1595 CHECK(journal_read_xhdr(j, &xhdr)); 1596 if (xhdr.size == 0) { 1597 isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR, 1598 "%s: journal corrupt: empty transaction", 1599 j->filename); 1600 FAIL(ISC_R_UNEXPECTED); 1601 } 1602 if (xhdr.serial0 != j->it.current_serial) { 1603 isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR, 1604 "%s: journal file corrupt: " 1605 "expected serial %u, got %u", 1606 j->filename, 1607 j->it.current_serial, xhdr.serial0); 1608 FAIL(ISC_R_UNEXPECTED); 1609 } 1610 j->it.xsize = xhdr.size; 1611 j->it.xpos = 0; 1612 } 1613 /* 1614 * Read an RR. 1615 */ 1616 CHECK(journal_read_rrhdr(j, &rrhdr)); 1617 /* 1618 * Perform a sanity check on the journal RR size. 1619 * The smallest possible RR has a 1-byte owner name 1620 * and a 10-byte header. The largest possible 1621 * RR has 65535 bytes of data, a header, and a maximum- 1622 * size owner name, well below 70 k total. 1623 */ 1624 if (rrhdr.size < 1+10 || rrhdr.size > 70000) { 1625 isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR, 1626 "%s: journal corrupt: impossible RR size " 1627 "(%d bytes)", j->filename, rrhdr.size); 1628 FAIL(ISC_R_UNEXPECTED); 1629 } 1630 1631 CHECK(size_buffer(j->mctx, &j->it.source, rrhdr.size)); 1632 CHECK(journal_read(j, j->it.source.base, rrhdr.size)); 1633 isc_buffer_add(&j->it.source, rrhdr.size); 1634 1635 /* 1636 * The target buffer is made the same size 1637 * as the source buffer, with the assumption that when 1638 * no compression in present, the output of dns_*_fromwire() 1639 * is no larger than the input. 1640 */ 1641 CHECK(size_buffer(j->mctx, &j->it.target, rrhdr.size)); 1642 1643 /* 1644 * Parse the owner name. We don't know where it 1645 * ends yet, so we make the entire "remaining" 1646 * part of the buffer "active". 1647 */ 1648 isc_buffer_setactive(&j->it.source, 1649 j->it.source.used - j->it.source.current); 1650 CHECK(dns_name_fromwire(&j->it.name, &j->it.source, 1651 &j->it.dctx, 0, &j->it.target)); 1652 1653 /* 1654 * Check that the RR header is there, and parse it. 1655 */ 1656 if (isc_buffer_remaininglength(&j->it.source) < 10) 1657 FAIL(DNS_R_FORMERR); 1658 1659 rdtype = isc_buffer_getuint16(&j->it.source); 1660 rdclass = isc_buffer_getuint16(&j->it.source); 1661 ttl = isc_buffer_getuint32(&j->it.source); 1662 rdlen = isc_buffer_getuint16(&j->it.source); 1663 1664 /* 1665 * Parse the rdata. 1666 */ 1667 if (isc_buffer_remaininglength(&j->it.source) != rdlen) 1668 FAIL(DNS_R_FORMERR); 1669 isc_buffer_setactive(&j->it.source, rdlen); 1670 dns_rdata_reset(&j->it.rdata); 1671 CHECK(dns_rdata_fromwire(&j->it.rdata, rdclass, 1672 rdtype, &j->it.source, &j->it.dctx, 1673 0, &j->it.target)); 1674 j->it.ttl = ttl; 1675 1676 j->it.xpos += sizeof(journal_rawrrhdr_t) + rrhdr.size; 1677 if (rdtype == dns_rdatatype_soa) { 1678 /* XXX could do additional consistency checks here */ 1679 j->it.current_serial = dns_soa_getserial(&j->it.rdata); 1680 } 1681 1682 result = ISC_R_SUCCESS; 1683 1684 failure: 1685 j->it.result = result; 1686 return (result); 1687} 1688 1689isc_result_t 1690dns_journal_next_rr(dns_journal_t *j) { 1691 j->it.result = read_one_rr(j); 1692 return (j->it.result); 1693} 1694 1695void 1696dns_journal_current_rr(dns_journal_t *j, dns_name_t **name, isc_uint32_t *ttl, 1697 dns_rdata_t **rdata) 1698{ 1699 REQUIRE(j->it.result == ISC_R_SUCCESS); 1700 *name = &j->it.name; 1701 *ttl = j->it.ttl; 1702 *rdata = &j->it.rdata; 1703} 1704 1705/**************************************************************************/ 1706/* 1707 * Generating diffs from databases 1708 */ 1709 1710/* 1711 * Construct a diff containing all the RRs at the current name of the 1712 * database iterator 'dbit' in database 'db', version 'ver'. 1713 * Set '*name' to the current name, and append the diff to 'diff'. 1714 * All new tuples will have the operation 'op'. 1715 * 1716 * Requires: 'name' must have buffer large enough to hold the name. 1717 * Typically, a dns_fixedname_t would be used. 1718 */ 1719static isc_result_t 1720get_name_diff(dns_db_t *db, dns_dbversion_t *ver, isc_stdtime_t now, 1721 dns_dbiterator_t *dbit, dns_name_t *name, dns_diffop_t op, 1722 dns_diff_t *diff) 1723{ 1724 isc_result_t result; 1725 dns_dbnode_t *node = NULL; 1726 dns_rdatasetiter_t *rdsiter = NULL; 1727 dns_difftuple_t *tuple = NULL; 1728 1729 result = dns_dbiterator_current(dbit, &node, name); 1730 if (result != ISC_R_SUCCESS) 1731 return (result); 1732 1733 result = dns_db_allrdatasets(db, node, ver, now, &rdsiter); 1734 if (result != ISC_R_SUCCESS) 1735 goto cleanup_node; 1736 1737 for (result = dns_rdatasetiter_first(rdsiter); 1738 result == ISC_R_SUCCESS; 1739 result = dns_rdatasetiter_next(rdsiter)) 1740 { 1741 dns_rdataset_t rdataset; 1742 1743 dns_rdataset_init(&rdataset); 1744 dns_rdatasetiter_current(rdsiter, &rdataset); 1745 1746 for (result = dns_rdataset_first(&rdataset); 1747 result == ISC_R_SUCCESS; 1748 result = dns_rdataset_next(&rdataset)) 1749 { 1750 dns_rdata_t rdata = DNS_RDATA_INIT; 1751 dns_rdataset_current(&rdataset, &rdata); 1752 result = dns_difftuple_create(diff->mctx, op, name, 1753 rdataset.ttl, &rdata, 1754 &tuple); 1755 if (result != ISC_R_SUCCESS) { 1756 dns_rdataset_disassociate(&rdataset); 1757 goto cleanup_iterator; 1758 } 1759 dns_diff_append(diff, &tuple); 1760 } 1761 dns_rdataset_disassociate(&rdataset); 1762 if (result != ISC_R_NOMORE) 1763 goto cleanup_iterator; 1764 } 1765 if (result != ISC_R_NOMORE) 1766 goto cleanup_iterator; 1767 1768 result = ISC_R_SUCCESS; 1769 1770 cleanup_iterator: 1771 dns_rdatasetiter_destroy(&rdsiter); 1772 1773 cleanup_node: 1774 dns_db_detachnode(db, &node); 1775 1776 return (result); 1777} 1778 1779/* 1780 * Comparison function for use by dns_diff_subtract when sorting 1781 * the diffs to be subtracted. The sort keys are the rdata type 1782 * and the rdata itself. The owner name is ignored, because 1783 * it is known to be the same for all tuples. 1784 */ 1785static int 1786rdata_order(const void *av, const void *bv) { 1787 dns_difftuple_t const * const *ap = av; 1788 dns_difftuple_t const * const *bp = bv; 1789 dns_difftuple_t const *a = *ap; 1790 dns_difftuple_t const *b = *bp; 1791 int r; 1792 r = (b->rdata.type - a->rdata.type); 1793 if (r != 0) 1794 return (r); 1795 r = dns_rdata_compare(&a->rdata, &b->rdata); 1796 return (r); 1797} 1798 1799static isc_result_t 1800dns_diff_subtract(dns_diff_t diff[2], dns_diff_t *r) { 1801 isc_result_t result; 1802 dns_difftuple_t *p[2]; 1803 int i, t; 1804 isc_boolean_t append; 1805 1806 CHECK(dns_diff_sort(&diff[0], rdata_order)); 1807 CHECK(dns_diff_sort(&diff[1], rdata_order)); 1808 1809 for (;;) { 1810 p[0] = ISC_LIST_HEAD(diff[0].tuples); 1811 p[1] = ISC_LIST_HEAD(diff[1].tuples); 1812 if (p[0] == NULL && p[1] == NULL) 1813 break; 1814 1815 for (i = 0; i < 2; i++) 1816 if (p[!i] == NULL) { 1817 ISC_LIST_UNLINK(diff[i].tuples, p[i], link); 1818 ISC_LIST_APPEND(r->tuples, p[i], link); 1819 goto next; 1820 } 1821 t = rdata_order(&p[0], &p[1]); 1822 if (t < 0) { 1823 ISC_LIST_UNLINK(diff[0].tuples, p[0], link); 1824 ISC_LIST_APPEND(r->tuples, p[0], link); 1825 goto next; 1826 } 1827 if (t > 0) { 1828 ISC_LIST_UNLINK(diff[1].tuples, p[1], link); 1829 ISC_LIST_APPEND(r->tuples, p[1], link); 1830 goto next; 1831 } 1832 INSIST(t == 0); 1833 /* 1834 * Identical RRs in both databases; skip them both 1835 * if the ttl differs. 1836 */ 1837 append = ISC_TF(p[0]->ttl != p[1]->ttl); 1838 for (i = 0; i < 2; i++) { 1839 ISC_LIST_UNLINK(diff[i].tuples, p[i], link); 1840 if (append) { 1841 ISC_LIST_APPEND(r->tuples, p[i], link); 1842 } else { 1843 dns_difftuple_free(&p[i]); 1844 } 1845 } 1846 next: ; 1847 } 1848 result = ISC_R_SUCCESS; 1849 failure: 1850 return (result); 1851} 1852 1853/* 1854 * Compare the databases 'dba' and 'dbb' and generate a journal 1855 * entry containing the changes to make 'dba' from 'dbb' (note 1856 * the order). This journal entry will consist of a single, 1857 * possibly very large transaction. 1858 */ 1859 1860isc_result_t 1861dns_db_diff(isc_mem_t *mctx, 1862 dns_db_t *dba, dns_dbversion_t *dbvera, 1863 dns_db_t *dbb, dns_dbversion_t *dbverb, 1864 const char *journal_filename) 1865{ 1866 dns_db_t *db[2]; 1867 dns_dbversion_t *ver[2]; 1868 dns_dbiterator_t *dbit[2] = { NULL, NULL }; 1869 isc_boolean_t have[2] = { ISC_FALSE, ISC_FALSE }; 1870 dns_fixedname_t fixname[2]; 1871 isc_result_t result, itresult[2]; 1872 dns_diff_t diff[2], resultdiff; 1873 int i, t; 1874 dns_journal_t *journal = NULL; 1875 1876 db[0] = dba, db[1] = dbb; 1877 ver[0] = dbvera, ver[1] = dbverb; 1878 1879 dns_diff_init(mctx, &diff[0]); 1880 dns_diff_init(mctx, &diff[1]); 1881 dns_diff_init(mctx, &resultdiff); 1882 1883 dns_fixedname_init(&fixname[0]); 1884 dns_fixedname_init(&fixname[1]); 1885 1886 result = dns_journal_open(mctx, journal_filename, ISC_TRUE, &journal); 1887 if (result != ISC_R_SUCCESS) 1888 return (result); 1889 1890 result = dns_db_createiterator(db[0], 0, &dbit[0]); 1891 if (result != ISC_R_SUCCESS) 1892 goto cleanup_journal; 1893 result = dns_db_createiterator(db[1], 0, &dbit[1]); 1894 if (result != ISC_R_SUCCESS) 1895 goto cleanup_interator0; 1896 1897 itresult[0] = dns_dbiterator_first(dbit[0]); 1898 itresult[1] = dns_dbiterator_first(dbit[1]); 1899 1900 for (;;) { 1901 for (i = 0; i < 2; i++) { 1902 if (! have[i] && itresult[i] == ISC_R_SUCCESS) { 1903 CHECK(get_name_diff(db[i], ver[i], 0, dbit[i], 1904 dns_fixedname_name(&fixname[i]), 1905 i == 0 ? 1906 DNS_DIFFOP_ADD : 1907 DNS_DIFFOP_DEL, 1908 &diff[i])); 1909 itresult[i] = dns_dbiterator_next(dbit[i]); 1910 have[i] = ISC_TRUE; 1911 } 1912 } 1913 1914 if (! have[0] && ! have[1]) { 1915 INSIST(ISC_LIST_EMPTY(diff[0].tuples)); 1916 INSIST(ISC_LIST_EMPTY(diff[1].tuples)); 1917 break; 1918 } 1919 1920 for (i = 0; i < 2; i++) { 1921 if (! have[!i]) { 1922 ISC_LIST_APPENDLIST(resultdiff.tuples, 1923 diff[i].tuples, link); 1924 INSIST(ISC_LIST_EMPTY(diff[i].tuples)); 1925 have[i] = ISC_FALSE; 1926 goto next; 1927 } 1928 } 1929 1930 t = dns_name_compare(dns_fixedname_name(&fixname[0]), 1931 dns_fixedname_name(&fixname[1])); 1932 if (t < 0) { 1933 ISC_LIST_APPENDLIST(resultdiff.tuples, 1934 diff[0].tuples, link); 1935 INSIST(ISC_LIST_EMPTY(diff[0].tuples)); 1936 have[0] = ISC_FALSE; 1937 continue; 1938 } 1939 if (t > 0) { 1940 ISC_LIST_APPENDLIST(resultdiff.tuples, 1941 diff[1].tuples, link); 1942 INSIST(ISC_LIST_EMPTY(diff[1].tuples)); 1943 have[1] = ISC_FALSE; 1944 continue; 1945 } 1946 INSIST(t == 0); 1947 CHECK(dns_diff_subtract(diff, &resultdiff)); 1948 INSIST(ISC_LIST_EMPTY(diff[0].tuples)); 1949 INSIST(ISC_LIST_EMPTY(diff[1].tuples)); 1950 have[0] = have[1] = ISC_FALSE; 1951 next: ; 1952 } 1953 if (itresult[0] != ISC_R_NOMORE) 1954 FAIL(itresult[0]); 1955 if (itresult[1] != ISC_R_NOMORE) 1956 FAIL(itresult[1]); 1957 1958 if (ISC_LIST_EMPTY(resultdiff.tuples)) { 1959 isc_log_write(JOURNAL_DEBUG_LOGARGS(3), "no changes"); 1960 } else { 1961 CHECK(dns_journal_write_transaction(journal, &resultdiff)); 1962 } 1963 INSIST(ISC_LIST_EMPTY(diff[0].tuples)); 1964 INSIST(ISC_LIST_EMPTY(diff[1].tuples)); 1965 1966 failure: 1967 dns_diff_clear(&resultdiff); 1968 dns_dbiterator_destroy(&dbit[1]); 1969 cleanup_interator0: 1970 dns_dbiterator_destroy(&dbit[0]); 1971 cleanup_journal: 1972 dns_journal_destroy(&journal); 1973 return (result); 1974} 1975 1976isc_result_t 1977dns_journal_compact(isc_mem_t *mctx, char *filename, isc_uint32_t serial, 1978 isc_uint32_t target_size) 1979{ 1980 unsigned int i; 1981 journal_pos_t best_guess; 1982 journal_pos_t current_pos; 1983 dns_journal_t *j = NULL; 1984 dns_journal_t *new = NULL; 1985 journal_rawheader_t rawheader; 1986 unsigned int copy_length; 1987 int namelen; 1988 char *buf = NULL; 1989 unsigned int size = 0; 1990 isc_result_t result; 1991 unsigned int indexend; 1992 char newname[1024]; 1993 char backup[1024]; 1994 isc_boolean_t is_backup = ISC_FALSE; 1995 1996 namelen = strlen(filename); 1997 if (namelen > 4 && strcmp(filename + namelen - 4, ".jnl") == 0) 1998 namelen -= 4; 1999 2000 result = isc_string_printf(newname, sizeof(newname), "%.*s.jnw", 2001 namelen, filename); 2002 if (result != ISC_R_SUCCESS) 2003 return (result); 2004 2005 result = isc_string_printf(backup, sizeof(backup), "%.*s.jbk", 2006 namelen, filename); 2007 if (result != ISC_R_SUCCESS) 2008 return (result); 2009 2010 result = journal_open(mctx, filename, ISC_FALSE, ISC_FALSE, &j); 2011 if (result == ISC_R_NOTFOUND) { 2012 is_backup = ISC_TRUE; 2013 result = journal_open(mctx, backup, ISC_FALSE, ISC_FALSE, &j); 2014 } 2015 if (result != ISC_R_SUCCESS) 2016 return (result); 2017 2018 if (JOURNAL_EMPTY(&j->header)) { 2019 dns_journal_destroy(&j); 2020 return (ISC_R_SUCCESS); 2021 } 2022 2023 if (DNS_SERIAL_GT(j->header.begin.serial, serial) || 2024 DNS_SERIAL_GT(serial, j->header.end.serial)) { 2025 dns_journal_destroy(&j); 2026 return (ISC_R_RANGE); 2027 } 2028 2029 /* 2030 * Cope with very small target sizes. 2031 */ 2032 indexend = sizeof(journal_rawheader_t) + 2033 j->header.index_size * sizeof(journal_rawpos_t); 2034 if (target_size < indexend * 2) 2035 target_size = target_size/2 + indexend; 2036 2037 /* 2038 * See if there is any work to do. 2039 */ 2040 if ((isc_uint32_t) j->header.end.offset < target_size) { 2041 dns_journal_destroy(&j); 2042 return (ISC_R_SUCCESS); 2043 } 2044 2045 CHECK(journal_open(mctx, newname, ISC_TRUE, ISC_TRUE, &new)); 2046 2047 /* 2048 * Remove overhead so space test below can succeed. 2049 */ 2050 if (target_size >= indexend) 2051 target_size -= indexend; 2052 2053 /* 2054 * Find if we can create enough free space. 2055 */ 2056 best_guess = j->header.begin; 2057 for (i = 0; i < j->header.index_size; i++) { 2058 if (POS_VALID(j->index[i]) && 2059 DNS_SERIAL_GE(serial, j->index[i].serial) && 2060 ((isc_uint32_t)(j->header.end.offset - j->index[i].offset) 2061 >= target_size / 2) && 2062 j->index[i].offset > best_guess.offset) 2063 best_guess = j->index[i]; 2064 } 2065 2066 current_pos = best_guess; 2067 while (current_pos.serial != serial) { 2068 CHECK(journal_next(j, ¤t_pos)); 2069 if (current_pos.serial == j->header.end.serial) 2070 break; 2071 2072 if (DNS_SERIAL_GE(serial, current_pos.serial) && 2073 ((isc_uint32_t)(j->header.end.offset - current_pos.offset) 2074 >= (target_size / 2)) && 2075 current_pos.offset > best_guess.offset) 2076 best_guess = current_pos; 2077 else 2078 break; 2079 } 2080 2081 INSIST(best_guess.serial != j->header.end.serial); 2082 if (best_guess.serial != serial) 2083 CHECK(journal_next(j, &best_guess)); 2084 2085 /* 2086 * We should now be roughly half target_size provided 2087 * we did not reach 'serial'. If not we will just copy 2088 * all uncommitted deltas regardless of the size. 2089 */ 2090 copy_length = j->header.end.offset - best_guess.offset; 2091 2092 if (copy_length != 0) { 2093 /* 2094 * Copy best_guess to end into space just freed. 2095 */ 2096 size = 64*1024; 2097 if (copy_length < size) 2098 size = copy_length; 2099 buf = isc_mem_get(mctx, size); 2100 if (buf == NULL) { 2101 result = ISC_R_NOMEMORY; 2102 goto failure; 2103 } 2104 2105 CHECK(journal_seek(j, best_guess.offset)); 2106 CHECK(journal_seek(new, indexend)); 2107 for (i = 0; i < copy_length; i += size) { 2108 unsigned int len = (copy_length - i) > size ? size : 2109 (copy_length - i); 2110 CHECK(journal_read(j, buf, len)); 2111 CHECK(journal_write(new, buf, len)); 2112 } 2113 2114 CHECK(journal_fsync(new)); 2115 2116 /* 2117 * Compute new header. 2118 */ 2119 new->header.begin.serial = best_guess.serial; 2120 new->header.begin.offset = indexend; 2121 new->header.end.serial = j->header.end.serial; 2122 new->header.end.offset = indexend + copy_length; 2123 2124 /* 2125 * Update the journal header. 2126 */ 2127 journal_header_encode(&new->header, &rawheader); 2128 CHECK(journal_seek(new, 0)); 2129 CHECK(journal_write(new, &rawheader, sizeof(rawheader))); 2130 CHECK(journal_fsync(new)); 2131 2132 /* 2133 * Build new index. 2134 */ 2135 current_pos = new->header.begin; 2136 while (current_pos.serial != new->header.end.serial) { 2137 index_add(new, ¤t_pos); 2138 CHECK(journal_next(new, ¤t_pos)); 2139 } 2140 2141 /* 2142 * Write index. 2143 */ 2144 CHECK(index_to_disk(new)); 2145 CHECK(journal_fsync(new)); 2146 2147 indexend = new->header.end.offset; 2148 } 2149 dns_journal_destroy(&new); 2150 2151 /* 2152 * With a UFS file system this should just succeed and be atomic. 2153 * Any IXFR outs will just continue and the old journal will be 2154 * removed on final close. 2155 * 2156 * With MSDOS / NTFS we need to do a two stage rename triggered 2157 * bu EEXISTS. Hopefully all IXFR's that were active at the last 2158 * rename are now complete. 2159 */ 2160 if (rename(newname, filename) == -1) { 2161 if (errno == EACCES && !is_backup) { 2162 result = isc_file_remove(backup); 2163 if (result != ISC_R_SUCCESS && 2164 result != ISC_R_FILENOTFOUND) 2165 goto failure; 2166 if (rename(filename, backup) == -1) 2167 goto maperrno; 2168 if (rename(newname, filename) == -1) 2169 goto maperrno; 2170 (void)isc_file_remove(backup); 2171 } else { 2172 maperrno: 2173 result = ISC_R_FAILURE; 2174 goto failure; 2175 } 2176 } 2177 2178 dns_journal_destroy(&j); 2179 result = ISC_R_SUCCESS; 2180 2181 failure: 2182 (void)isc_file_remove(newname); 2183 if (buf != NULL) 2184 isc_mem_put(mctx, buf, size); 2185 if (j != NULL) 2186 dns_journal_destroy(&j); 2187 if (new != NULL) 2188 dns_journal_destroy(&new); 2189 return (result); 2190} 2191 2192static isc_result_t 2193index_to_disk(dns_journal_t *j) { 2194 isc_result_t result = ISC_R_SUCCESS; 2195 2196 if (j->header.index_size != 0) { 2197 unsigned int i; 2198 unsigned char *p; 2199 unsigned int rawbytes; 2200 2201 rawbytes = j->header.index_size * sizeof(journal_rawpos_t); 2202 2203 p = j->rawindex; 2204 for (i = 0; i < j->header.index_size; i++) { 2205 encode_uint32(j->index[i].serial, p); 2206 p += 4; 2207 encode_uint32(j->index[i].offset, p); 2208 p += 4; 2209 } 2210 INSIST(p == j->rawindex + rawbytes); 2211 2212 CHECK(journal_seek(j, sizeof(journal_rawheader_t))); 2213 CHECK(journal_write(j, j->rawindex, rawbytes)); 2214 } 2215failure: 2216 return (result); 2217} 2218