2 * Copyright (C) 2004, 2005, 2007-2009 Internet Systems Consortium, Inc. ("ISC")
3 * Copyright (C) 1999-2002 Internet Software Consortium.
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.
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.
18 /* $Id: journal.c,v 1.103.48.2 2009/01/18 23:47:37 tbox Exp $ */
28 #include <isc/stdio.h>
29 #include <isc/string.h>
32 #include <dns/compress.h>
34 #include <dns/dbiterator.h>
36 #include <dns/fixedname.h>
37 #include <dns/journal.h>
39 #include <dns/rdataset.h>
40 #include <dns/rdatasetiter.h>
41 #include <dns/result.h>
47 * A journal file consists of
49 * \li A fixed-size header of type journal_rawheader_t.
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.
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.
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.
82 * When true, accept IXFR difference sequences where the
83 * SOA serial number does not change (BIND 8 sends such
86 static isc_boolean_t bind8_compat = ISC_TRUE; /* XXX config */
88 /**************************************************************************/
90 * Miscellaneous utilities.
93 #define JOURNAL_COMMON_LOGARGS \
94 dns_lctx, DNS_LOGCATEGORY_GENERAL, DNS_LOGMODULE_JOURNAL
96 #define JOURNAL_DEBUG_LOGARGS(n) \
97 JOURNAL_COMMON_LOGARGS, ISC_LOG_DEBUG(n)
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".
105 do { result = (code); \
106 if (result != ISC_R_SUCCESS) goto failure; \
110 do { result = (op); \
111 if (result != ISC_R_SUCCESS) goto failure; \
114 static isc_result_t index_to_disk(dns_journal_t *);
116 static inline isc_uint32_t
117 decode_uint32(unsigned char *p) {
118 return ((p[0] << 24) +
125 encode_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);
133 dns_db_createsoatuple(dns_db_t *db, dns_dbversion_t *ver, isc_mem_t *mctx,
134 dns_diffop_t op, dns_difftuple_t **tp)
138 dns_rdataset_t rdataset;
139 dns_rdata_t rdata = DNS_RDATA_INIT;
140 dns_name_t *zonename;
142 zonename = dns_db_origin(db);
145 result = dns_db_findnode(db, zonename, ISC_FALSE, &node);
146 if (result != ISC_R_SUCCESS)
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)
155 result = dns_rdataset_first(&rdataset);
156 if (result != ISC_R_SUCCESS)
159 dns_rdataset_current(&rdataset, &rdata);
161 result = dns_difftuple_create(mctx, op, zonename, rdataset.ttl,
164 dns_rdataset_disassociate(&rdataset);
165 dns_db_detachnode(db, &node);
166 return (ISC_R_SUCCESS);
169 dns_db_detachnode(db, &node);
171 UNEXPECTED_ERROR(__FILE__, __LINE__, "missing SOA");
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.
184 unsigned char serial[4]; /*%< SOA serial before update. */
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
192 unsigned char offset[4]; /*%< Offset from beginning of file. */
197 * The header is of a fixed size, with some spare room for future
200 #define JOURNAL_HEADER_SIZE 64 /* Bytes. */
203 * The on-disk representation of the journal header.
204 * All numbers are stored in big-endian order.
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];
217 /* Pad the header to a fixed size. */
218 unsigned char pad[JOURNAL_HEADER_SIZE];
219 } journal_rawheader_t;
222 * The on-disk representation of the transaction header.
223 * There is one of these at the beginning of each transaction.
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. */
232 * The on-disk representation of the RR header.
233 * There is one of these at the beginning of each RR.
236 unsigned char size[4]; /*%< In bytes, excluding header. */
237 } journal_rawrrhdr_t;
240 * The in-core representation of the journal header.
247 #define POS_VALID(pos) ((pos).offset != 0)
248 #define POS_INVALIDATE(pos) ((pos).offset = 0, (pos).serial = 0)
251 unsigned char format[16];
254 isc_uint32_t index_size;
258 * The in-core representation of the transaction header.
263 isc_uint32_t serial0;
264 isc_uint32_t serial1;
268 * The in-core representation of the RR header.
276 * Initial contents to store in the header of a newly created
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).
286 static journal_header_t
287 initial_journal_header = { ";BIND LOG V9\n", { 0, 0 }, { 0, 0 }, 0 };
289 #define JOURNAL_EMPTY(h) ((h)->begin.offset == (h)->end.offset)
292 JOURNAL_STATE_INVALID,
295 JOURNAL_STATE_TRANSACTION
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 */
309 /*% Current transaction state (when writing). */
311 unsigned int n_soa; /*%< Number of SOAs seen */
312 journal_pos_t pos[2]; /*%< Begin/end position */
315 /*% Iteration state (when reading). */
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 */
334 #define DNS_JOURNAL_MAGIC ISC_MAGIC('J', 'O', 'U', 'R')
335 #define DNS_JOURNAL_VALID(t) ISC_MAGIC_VALID(t, DNS_JOURNAL_MAGIC)
338 journal_pos_decode(journal_rawpos_t *raw, journal_pos_t *cooked) {
339 cooked->serial = decode_uint32(raw->serial);
340 cooked->offset = decode_uint32(raw->offset);
344 journal_pos_encode(journal_rawpos_t *raw, journal_pos_t *cooked) {
345 encode_uint32(cooked->serial, raw->serial);
346 encode_uint32(cooked->offset, raw->offset);
350 journal_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);
359 journal_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);
369 * Journal file I/O subroutines, with error checking and reporting.
372 journal_seek(dns_journal_t *j, isc_uint32_t offset) {
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);
382 return (ISC_R_SUCCESS);
386 journal_read(dns_journal_t *j, void *mem, size_t nbytes) {
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,
395 j->filename, isc_result_totext(result));
396 return (ISC_R_UNEXPECTED);
399 return (ISC_R_SUCCESS);
403 journal_write(dns_journal_t *j, void *mem, size_t nbytes) {
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,
410 j->filename, isc_result_totext(result));
411 return (ISC_R_UNEXPECTED);
414 return (ISC_R_SUCCESS);
418 journal_fsync(dns_journal_t *j) {
420 result = isc_stdio_flush(j->fp);
421 if (result != ISC_R_SUCCESS) {
422 isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR,
424 j->filename, isc_result_totext(result));
425 return (ISC_R_UNEXPECTED);
427 result = isc_stdio_sync(j->fp);
428 if (result != ISC_R_SUCCESS) {
429 isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR,
431 j->filename, isc_result_totext(result));
432 return (ISC_R_UNEXPECTED);
434 return (ISC_R_SUCCESS);
438 * Read/write a transaction header at the current file position.
442 journal_read_xhdr(dns_journal_t *j, journal_xhdr_t *xhdr) {
443 journal_rawxhdr_t raw;
445 result = journal_read(j, &raw, sizeof(raw));
446 if (result != ISC_R_SUCCESS)
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);
455 journal_write_xhdr(dns_journal_t *j, isc_uint32_t size,
456 isc_uint32_t serial0, isc_uint32_t serial1)
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)));
467 * Read an RR header at the current file position.
471 journal_read_rrhdr(dns_journal_t *j, journal_rrhdr_t *rrhdr) {
472 journal_rawrrhdr_t raw;
474 result = journal_read(j, &raw, sizeof(raw));
475 if (result != ISC_R_SUCCESS)
477 rrhdr->size = decode_uint32(raw.size);
478 return (ISC_R_SUCCESS);
482 journal_file_create(isc_mem_t *mctx, const char *filename) {
485 journal_header_t header;
486 journal_rawheader_t rawheader;
487 int index_size = 56; /* XXX configurable */
489 void *mem; /* Memory for temporary index image. */
491 INSIST(sizeof(journal_rawheader_t) == JOURNAL_HEADER_SIZE);
493 result = isc_stdio_open(filename, "wb", &fp);
494 if (result != ISC_R_SUCCESS) {
495 isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR,
497 filename, isc_result_totext(result));
498 return (ISC_R_UNEXPECTED);
501 header = initial_journal_header;
502 header.index_size = index_size;
503 journal_header_encode(&header, &rawheader);
505 size = sizeof(journal_rawheader_t) +
506 index_size * sizeof(journal_rawpos_t);
508 mem = isc_mem_get(mctx, size);
510 (void)isc_stdio_close(fp);
511 (void)isc_file_remove(filename);
512 return (ISC_R_NOMEMORY);
514 memset(mem, 0, size);
515 memcpy(mem, &rawheader, sizeof(rawheader));
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,
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);
527 isc_mem_put(mctx, mem, size);
529 result = isc_stdio_close(fp);
530 if (result != ISC_R_SUCCESS) {
531 isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR,
533 filename, isc_result_totext(result));
534 (void)isc_file_remove(filename);
535 return (ISC_R_UNEXPECTED);
538 return (ISC_R_SUCCESS);
542 journal_open(isc_mem_t *mctx, const char *filename, isc_boolean_t write,
543 isc_boolean_t create, dns_journal_t **journalp) {
546 journal_rawheader_t rawheader;
549 INSIST(journalp != NULL && *journalp == NULL);
550 j = isc_mem_get(mctx, sizeof(*j));
552 return (ISC_R_NOMEMORY);
555 j->state = JOURNAL_STATE_INVALID;
557 j->filename = filename;
561 result = isc_stdio_open(j->filename, write ? "rb+" : "rb", &fp);
563 if (result == ISC_R_FILENOTFOUND) {
565 isc_log_write(JOURNAL_COMMON_LOGARGS,
567 "journal file %s does not exist, "
570 CHECK(journal_file_create(mctx, filename));
574 result = isc_stdio_open(j->filename, "rb+", &fp);
576 FAIL(ISC_R_NOTFOUND);
579 if (result != ISC_R_SUCCESS) {
580 isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR,
582 j->filename, isc_result_totext(result));
583 FAIL(ISC_R_UNEXPECTED);
589 * Set magic early so that seek/read can succeed.
591 j->magic = DNS_JOURNAL_MAGIC;
593 CHECK(journal_seek(j, 0));
594 CHECK(journal_read(j, &rawheader, sizeof(rawheader)));
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",
601 FAIL(ISC_R_UNEXPECTED);
603 journal_header_decode(&rawheader, &j->header);
606 * If there is an index, read the raw index into a dynamically
607 * allocated buffer and then convert it into a cooked index.
609 if (j->header.index_size != 0) {
611 unsigned int rawbytes;
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);
619 CHECK(journal_read(j, j->rawindex, rawbytes));
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);
627 for (i = 0; i < j->header.index_size; i++) {
628 j->index[i].serial = decode_uint32(p);
630 j->index[i].offset = decode_uint32(p);
633 INSIST(p == j->rawindex + rawbytes);
635 j->offset = -1; /* Invalid, must seek explicitly. */
638 * Initialize the iterator.
640 dns_name_init(&j->it.name, NULL);
641 dns_rdata_init(&j->it.rdata);
644 * Set up empty initial buffers for unchecked and checked
645 * wire format RR data. They will be reallocated
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);
653 write ? JOURNAL_STATE_WRITE : JOURNAL_STATE_READ;
656 return (ISC_R_SUCCESS);
660 if (j->index != NULL) {
661 isc_mem_put(j->mctx, j->index, j->header.index_size *
662 sizeof(journal_rawpos_t));
666 (void)isc_stdio_close(j->fp);
667 isc_mem_put(j->mctx, j, sizeof(*j));
672 dns_journal_open(isc_mem_t *mctx, const char *filename, isc_boolean_t write,
673 dns_journal_t **journalp) {
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)
684 result = isc_string_printf(backup, sizeof(backup), "%.*s.jbk",
686 if (result != ISC_R_SUCCESS)
688 result = journal_open(mctx, backup, write, write, journalp);
694 * A comparison function defining the sorting order for
695 * entries in the IXFR-style journal file.
697 * The IXFR format requires that deletions are sorted before
698 * additions, and within either one, SOA records are sorted
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.
706 ixfr_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;
712 int bop = 0, aop = 0;
716 case DNS_DIFFOP_DELRESIGN:
720 case DNS_DIFFOP_ADDRESIGN:
729 case DNS_DIFFOP_DELRESIGN:
733 case DNS_DIFFOP_ADDRESIGN:
744 r = (b->rdata.type == dns_rdatatype_soa) -
745 (a->rdata.type == dns_rdatatype_soa);
749 r = (a->rdata.type - b->rdata.type);
754 * Advance '*pos' to the next journal transaction.
757 * *pos refers to a valid journal transaction.
760 * When ISC_R_SUCCESS is returned,
761 * *pos refers to the next journal transaction.
766 * ISC_R_NOMORE *pos pointed at the last transaction
767 * Other results due to file errors are possible.
770 journal_next(dns_journal_t *j, journal_pos_t *pos) {
773 REQUIRE(DNS_JOURNAL_VALID(j));
775 result = journal_seek(j, pos->offset);
776 if (result != ISC_R_SUCCESS)
779 if (pos->serial == j->header.end.serial)
780 return (ISC_R_NOMORE);
782 * Read the header of the current transaction.
783 * This will return ISC_R_NOMORE if we are at EOF.
785 result = journal_read_xhdr(j, &xhdr);
786 if (result != ISC_R_SUCCESS)
790 * Check serial number consistency.
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);
801 * Check for offset wraparound.
803 if ((isc_offset_t)(pos->offset + sizeof(journal_rawxhdr_t) + xhdr.size)
805 isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR,
806 "%s: offset too large", j->filename);
807 return (ISC_R_UNEXPECTED);
810 pos->offset += sizeof(journal_rawxhdr_t) + xhdr.size;
811 pos->serial = xhdr.serial1;
812 return (ISC_R_SUCCESS);
816 * If the index of the journal 'j' contains an entry "better"
817 * than '*best_guess', replace '*best_guess' with it.
819 * "Better" means having a serial number closer to 'serial'
820 * but not greater than 'serial'.
823 index_find(dns_journal_t *j, isc_uint32_t serial, journal_pos_t *best_guess) {
825 if (j->index == NULL)
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];
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.
845 index_add(dns_journal_t *j, journal_pos_t *pos) {
847 if (j->index == NULL)
850 * Search for a vacant position.
852 for (i = 0; i < j->header.index_size; i++) {
853 if (! POS_VALID(j->index[i]))
856 if (i == j->header.index_size) {
859 * Found no vacant position. Make some room.
861 for (i = 0; i < j->header.index_size; i += 2) {
862 j->index[k++] = j->index[i];
864 i = k; /* 'i' identifies the first vacant position. */
865 while (k < j->header.index_size) {
866 POS_INVALIDATE(j->index[k]);
870 INSIST(i < j->header.index_size);
871 INSIST(! POS_VALID(j->index[i]));
874 * Store the new index entry.
880 * Invalidate any existing index entries that could become
881 * ambiguous when a new transaction with number 'serial' is added.
884 index_invalidate(dns_journal_t *j, isc_uint32_t serial) {
886 if (j->index == NULL)
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]);
895 * Try to find a transaction with initial serial number 'serial'
896 * in the journal 'j'.
898 * If found, store its position at '*pos' and return ISC_R_SUCCESS.
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.
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.
909 * If 'serial' is outside the range of addressable serial numbers
910 * covered by the journal, return ISC_R_RANGE.
914 journal_find(dns_journal_t *j, isc_uint32_t serial, journal_pos_t *pos) {
916 journal_pos_t current_pos;
917 REQUIRE(DNS_JOURNAL_VALID(j));
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);
928 current_pos = j->header.begin;
929 index_find(j, serial, ¤t_pos);
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)
939 return (ISC_R_SUCCESS);
943 dns_journal_begin_transaction(dns_journal_t *j) {
946 journal_rawxhdr_t hdr;
948 REQUIRE(DNS_JOURNAL_VALID(j));
949 REQUIRE(j->state == JOURNAL_STATE_WRITE);
952 * Find the file offset where the new transaction should
953 * be written, and seek there.
955 if (JOURNAL_EMPTY(&j->header)) {
956 offset = sizeof(journal_rawheader_t) +
957 j->header.index_size * sizeof(journal_rawpos_t);
959 offset = j->header.end.offset;
961 j->x.pos[0].offset = offset;
962 j->x.pos[1].offset = offset; /* Initial value, will be incremented. */
965 CHECK(journal_seek(j, offset));
968 * Write a dummy transaction header of all zeroes to reserve
969 * space. It will be filled in when the transaction is
972 memset(&hdr, 0, sizeof(hdr));
973 CHECK(journal_write(j, &hdr, sizeof(hdr)));
974 j->x.pos[1].offset = j->offset;
976 j->state = JOURNAL_STATE_TRANSACTION;
977 result = ISC_R_SUCCESS;
983 dns_journal_writediff(dns_journal_t *j, dns_diff_t *diff) {
991 REQUIRE(DNS_DIFF_VALID(diff));
992 REQUIRE(j->state == JOURNAL_STATE_TRANSACTION);
994 isc_log_write(JOURNAL_DEBUG_LOGARGS(3), "writing to journal");
995 (void)dns_diff_print(diff, NULL);
998 * Pass 1: determine the buffer size needed, and
999 * keep track of SOA serial numbers.
1002 for (t = ISC_LIST_HEAD(diff->tuples); t != NULL;
1003 t = ISC_LIST_NEXT(t, link))
1005 if (t->rdata.type == dns_rdatatype_soa) {
1007 j->x.pos[j->x.n_soa].serial =
1008 dns_soa_getserial(&t->rdata);
1011 size += sizeof(journal_rawrrhdr_t);
1012 size += t->name.length; /* XXX should have access macro? */
1014 size += t->rdata.length;
1017 mem = isc_mem_get(j->mctx, size);
1019 return (ISC_R_NOMEMORY);
1021 isc_buffer_init(&buffer, mem, size);
1024 * Pass 2. Write RRs to buffer.
1026 for (t = ISC_LIST_HEAD(diff->tuples); t != NULL;
1027 t = ISC_LIST_NEXT(t, link))
1030 * Write the RR header.
1032 isc_buffer_putuint32(&buffer, t->name.length + 10 +
1035 * Write the owner name, RR header, and RR data.
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);
1047 isc_buffer_usedregion(&buffer, &used);
1048 INSIST(used.length == size);
1050 j->x.pos[1].offset += used.length;
1053 * Write the buffer contents to the journal file.
1055 CHECK(journal_write(j, used.base, used.length));
1057 result = ISC_R_SUCCESS;
1061 isc_mem_put(j->mctx, mem, size);
1067 dns_journal_commit(dns_journal_t *j) {
1068 isc_result_t result;
1069 journal_rawheader_t rawheader;
1071 REQUIRE(DNS_JOURNAL_VALID(j));
1072 REQUIRE(j->state == JOURNAL_STATE_TRANSACTION);
1075 * Perform some basic consistency checks.
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);
1083 if (! (DNS_SERIAL_GT(j->x.pos[1].serial, j->x.pos[0].serial) ||
1085 j->x.pos[1].serial == j->x.pos[0].serial)))
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);
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",
1099 j->header.end.serial,
1100 j->x.pos[0].serial);
1101 return (ISC_R_UNEXPECTED);
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.
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));
1116 index_invalidate(j, j->x.pos[1].serial);
1119 if (DNS_SERIAL_GT(last_dumped_serial, j->x.pos[1].serial)) {
1125 * Commit the transaction data to stable storage.
1127 CHECK(journal_fsync(j));
1130 * Update the transaction header.
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));
1138 * Update the journal header.
1140 if (JOURNAL_EMPTY(&j->header)) {
1141 j->header.begin = j->x.pos[0];
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)));
1151 index_add(j, &j->x.pos[0]);
1154 * Convert the index into on-disk format and write
1157 CHECK(index_to_disk(j));
1160 * Commit the header to stable storage.
1162 CHECK(journal_fsync(j));
1165 * We no longer have a transaction open.
1167 j->state = JOURNAL_STATE_WRITE;
1169 result = ISC_R_SUCCESS;
1176 dns_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;
1188 dns_journal_destroy(dns_journal_t **journalp) {
1189 dns_journal_t *j = *journalp;
1190 REQUIRE(DNS_JOURNAL_VALID(j));
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);
1207 (void)isc_stdio_close(j->fp);
1209 isc_mem_put(j->mctx, j, sizeof(*j));
1214 * Roll the open journal 'j' into the database 'db'.
1215 * A new database version will be created.
1218 /* XXX Share code with incoming IXFR? */
1221 roll_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;
1230 unsigned int n_soa = 0;
1231 unsigned int n_put = 0;
1234 REQUIRE(DNS_JOURNAL_VALID(j));
1235 REQUIRE(DNS_DB_VALID(db));
1237 dns_diff_init(j->mctx, &diff);
1240 * Set up empty initial buffers for unchecked and checked
1241 * wire format transaction data. They will be reallocated
1244 isc_buffer_init(&source, NULL, 0);
1245 isc_buffer_init(&target, NULL, 0);
1248 * Create the new database version.
1250 CHECK(dns_db_newversion(db, &ver));
1253 * Get the current database SOA serial number.
1255 CHECK(dns_db_getsoaserial(db, ver, &db_serial));
1258 * Locate a journal entry for the current database serial.
1260 CHECK(journal_find(j, db_serial, &pos));
1262 * XXX do more drastic things, like marking zone stale,
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.
1271 end_serial = dns_journal_last_serial(j);
1272 if (db_serial == end_serial)
1273 CHECK(DNS_R_UPTODATE);
1275 CHECK(dns_journal_iter_init(j, db_serial, end_serial));
1277 for (result = dns_journal_first_rr(j);
1278 result == ISC_R_SUCCESS;
1279 result = dns_journal_next_rr(j))
1284 dns_difftuple_t *tuple = NULL;
1288 dns_journal_current_rr(j, &name, &ttl, &rdata);
1290 if (rdata->type == dns_rdatatype_soa) {
1293 db_serial = j->it.current_serial;
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);
1304 if ((options & DNS_JOURNALOPT_RESIGN) != 0)
1305 op = (n_soa == 1) ? DNS_DIFFOP_DELRESIGN :
1306 DNS_DIFFOP_ADDRESIGN;
1308 op = (n_soa == 1) ? DNS_DIFFOP_DEL : DNS_DIFFOP_ADD;
1310 CHECK(dns_difftuple_create(diff.mctx, op, name, ttl, rdata,
1312 dns_diff_append(&diff, &tuple);
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);
1324 if (result == ISC_R_NOMORE)
1325 result = ISC_R_SUCCESS;
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);
1339 dns_db_closeversion(db, &ver, result == ISC_R_SUCCESS ?
1340 ISC_TRUE : ISC_FALSE);
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);
1347 dns_diff_clear(&diff);
1353 dns_journal_rollforward(isc_mem_t *mctx, dns_db_t *db,
1354 unsigned int options, const char *filename)
1357 isc_result_t result;
1359 REQUIRE(DNS_DB_VALID(db));
1360 REQUIRE(filename != 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);
1369 if (result != ISC_R_SUCCESS)
1371 if (JOURNAL_EMPTY(&j->header))
1372 result = DNS_R_UPTODATE;
1374 result = roll_forward(j, db, options);
1376 dns_journal_destroy(&j);
1382 dns_journal_print(isc_mem_t *mctx, const char *filename, FILE *file) {
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;
1390 unsigned int n_soa = 0;
1391 unsigned int n_put = 0;
1393 REQUIRE(filename != 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);
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);
1409 dns_diff_init(j->mctx, &diff);
1412 * Set up empty initial buffers for unchecked and checked
1413 * wire format transaction data. They will be reallocated
1416 isc_buffer_init(&source, NULL, 0);
1417 isc_buffer_init(&target, NULL, 0);
1419 start_serial = dns_journal_first_serial(j);
1420 end_serial = dns_journal_last_serial(j);
1422 CHECK(dns_journal_iter_init(j, start_serial, end_serial));
1424 for (result = dns_journal_first_rr(j);
1425 result == ISC_R_SUCCESS;
1426 result = dns_journal_next_rr(j))
1431 dns_difftuple_t *tuple = NULL;
1435 dns_journal_current_rr(j, &name, &ttl, &rdata);
1437 if (rdata->type == dns_rdatatype_soa)
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);
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);
1453 if (++n_put > 100) {
1454 result = dns_diff_print(&diff, file);
1455 dns_diff_clear(&diff);
1457 if (result != ISC_R_SUCCESS)
1461 if (result == ISC_R_NOMORE)
1462 result = ISC_R_SUCCESS;
1466 result = dns_diff_print(&diff, file);
1467 dns_diff_clear(&diff);
1472 isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR,
1473 "%s: cannot print: journal file corrupt", j->filename);
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);
1481 dns_diff_clear(&diff);
1482 dns_journal_destroy(&j);
1487 /**************************************************************************/
1489 * Miscellaneous accessors.
1491 isc_uint32_t dns_journal_first_serial(dns_journal_t *j) {
1492 return (j->header.begin.serial);
1495 isc_uint32_t dns_journal_last_serial(dns_journal_t *j) {
1496 return (j->header.end.serial);
1499 /**************************************************************************/
1501 * Iteration support.
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.
1511 static isc_result_t read_one_rr(dns_journal_t *j);
1514 * Make sure the buffer 'b' is has at least 'size' bytes
1515 * allocated, and clear it.
1518 * Either b->base is NULL, or it points to b->length bytes of memory
1519 * previously allocated by isc_mem_get().
1523 size_buffer(isc_mem_t *mctx, isc_buffer_t *b, unsigned size) {
1524 if (b->length < size) {
1525 void *mem = isc_mem_get(mctx, size);
1527 return (ISC_R_NOMEMORY);
1528 if (b->base != NULL)
1529 isc_mem_put(mctx, b->base, b->length);
1533 isc_buffer_clear(b);
1534 return (ISC_R_SUCCESS);
1538 dns_journal_iter_init(dns_journal_t *j,
1539 isc_uint32_t begin_serial, isc_uint32_t end_serial)
1541 isc_result_t result;
1543 CHECK(journal_find(j, begin_serial, &j->it.bpos));
1544 INSIST(j->it.bpos.serial == begin_serial);
1546 CHECK(journal_find(j, end_serial, &j->it.epos));
1547 INSIST(j->it.epos.serial == end_serial);
1549 result = ISC_R_SUCCESS;
1551 j->it.result = result;
1552 return (j->it.result);
1557 dns_journal_first_rr(dns_journal_t *j) {
1558 isc_result_t result;
1561 * Seek to the beginning of the first transaction we are
1564 CHECK(journal_seek(j, j->it.bpos.offset));
1565 j->it.current_serial = j->it.bpos.serial;
1567 j->it.xsize = 0; /* We have no transaction data yet... */
1568 j->it.xpos = 0; /* ...and haven't used any of it. */
1570 return (read_one_rr(j));
1577 read_one_rr(dns_journal_t *j) {
1578 isc_result_t result;
1580 dns_rdatatype_t rdtype;
1581 dns_rdataclass_t rdclass;
1584 journal_xhdr_t xhdr;
1585 journal_rrhdr_t rrhdr;
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) {
1592 * We are at a transaction boundary.
1593 * Read another transaction header.
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",
1600 FAIL(ISC_R_UNEXPECTED);
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",
1607 j->it.current_serial, xhdr.serial0);
1608 FAIL(ISC_R_UNEXPECTED);
1610 j->it.xsize = xhdr.size;
1616 CHECK(journal_read_rrhdr(j, &rrhdr));
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.
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);
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);
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.
1641 CHECK(size_buffer(j->mctx, &j->it.target, rrhdr.size));
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".
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));
1654 * Check that the RR header is there, and parse it.
1656 if (isc_buffer_remaininglength(&j->it.source) < 10)
1657 FAIL(DNS_R_FORMERR);
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);
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,
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);
1682 result = ISC_R_SUCCESS;
1685 j->it.result = result;
1690 dns_journal_next_rr(dns_journal_t *j) {
1691 j->it.result = read_one_rr(j);
1692 return (j->it.result);
1696 dns_journal_current_rr(dns_journal_t *j, dns_name_t **name, isc_uint32_t *ttl,
1697 dns_rdata_t **rdata)
1699 REQUIRE(j->it.result == ISC_R_SUCCESS);
1700 *name = &j->it.name;
1702 *rdata = &j->it.rdata;
1705 /**************************************************************************/
1707 * Generating diffs from databases
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'.
1716 * Requires: 'name' must have buffer large enough to hold the name.
1717 * Typically, a dns_fixedname_t would be used.
1720 get_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,
1724 isc_result_t result;
1725 dns_dbnode_t *node = NULL;
1726 dns_rdatasetiter_t *rdsiter = NULL;
1727 dns_difftuple_t *tuple = NULL;
1729 result = dns_dbiterator_current(dbit, &node, name);
1730 if (result != ISC_R_SUCCESS)
1733 result = dns_db_allrdatasets(db, node, ver, now, &rdsiter);
1734 if (result != ISC_R_SUCCESS)
1737 for (result = dns_rdatasetiter_first(rdsiter);
1738 result == ISC_R_SUCCESS;
1739 result = dns_rdatasetiter_next(rdsiter))
1741 dns_rdataset_t rdataset;
1743 dns_rdataset_init(&rdataset);
1744 dns_rdatasetiter_current(rdsiter, &rdataset);
1746 for (result = dns_rdataset_first(&rdataset);
1747 result == ISC_R_SUCCESS;
1748 result = dns_rdataset_next(&rdataset))
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,
1755 if (result != ISC_R_SUCCESS) {
1756 dns_rdataset_disassociate(&rdataset);
1757 goto cleanup_iterator;
1759 dns_diff_append(diff, &tuple);
1761 dns_rdataset_disassociate(&rdataset);
1762 if (result != ISC_R_NOMORE)
1763 goto cleanup_iterator;
1765 if (result != ISC_R_NOMORE)
1766 goto cleanup_iterator;
1768 result = ISC_R_SUCCESS;
1771 dns_rdatasetiter_destroy(&rdsiter);
1774 dns_db_detachnode(db, &node);
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.
1786 rdata_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;
1792 r = (b->rdata.type - a->rdata.type);
1795 r = dns_rdata_compare(&a->rdata, &b->rdata);
1800 dns_diff_subtract(dns_diff_t diff[2], dns_diff_t *r) {
1801 isc_result_t result;
1802 dns_difftuple_t *p[2];
1804 isc_boolean_t append;
1806 CHECK(dns_diff_sort(&diff[0], rdata_order));
1807 CHECK(dns_diff_sort(&diff[1], rdata_order));
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)
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);
1821 t = rdata_order(&p[0], &p[1]);
1823 ISC_LIST_UNLINK(diff[0].tuples, p[0], link);
1824 ISC_LIST_APPEND(r->tuples, p[0], link);
1828 ISC_LIST_UNLINK(diff[1].tuples, p[1], link);
1829 ISC_LIST_APPEND(r->tuples, p[1], link);
1834 * Identical RRs in both databases; skip them both
1835 * if the ttl differs.
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);
1841 ISC_LIST_APPEND(r->tuples, p[i], link);
1843 dns_difftuple_free(&p[i]);
1848 result = ISC_R_SUCCESS;
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.
1861 dns_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)
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;
1874 dns_journal_t *journal = NULL;
1876 db[0] = dba, db[1] = dbb;
1877 ver[0] = dbvera, ver[1] = dbverb;
1879 dns_diff_init(mctx, &diff[0]);
1880 dns_diff_init(mctx, &diff[1]);
1881 dns_diff_init(mctx, &resultdiff);
1883 dns_fixedname_init(&fixname[0]);
1884 dns_fixedname_init(&fixname[1]);
1886 result = dns_journal_open(mctx, journal_filename, ISC_TRUE, &journal);
1887 if (result != ISC_R_SUCCESS)
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;
1897 itresult[0] = dns_dbiterator_first(dbit[0]);
1898 itresult[1] = dns_dbiterator_first(dbit[1]);
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]),
1909 itresult[i] = dns_dbiterator_next(dbit[i]);
1914 if (! have[0] && ! have[1]) {
1915 INSIST(ISC_LIST_EMPTY(diff[0].tuples));
1916 INSIST(ISC_LIST_EMPTY(diff[1].tuples));
1920 for (i = 0; i < 2; 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;
1930 t = dns_name_compare(dns_fixedname_name(&fixname[0]),
1931 dns_fixedname_name(&fixname[1]));
1933 ISC_LIST_APPENDLIST(resultdiff.tuples,
1934 diff[0].tuples, link);
1935 INSIST(ISC_LIST_EMPTY(diff[0].tuples));
1936 have[0] = ISC_FALSE;
1940 ISC_LIST_APPENDLIST(resultdiff.tuples,
1941 diff[1].tuples, link);
1942 INSIST(ISC_LIST_EMPTY(diff[1].tuples));
1943 have[1] = ISC_FALSE;
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;
1953 if (itresult[0] != ISC_R_NOMORE)
1955 if (itresult[1] != ISC_R_NOMORE)
1958 if (ISC_LIST_EMPTY(resultdiff.tuples)) {
1959 isc_log_write(JOURNAL_DEBUG_LOGARGS(3), "no changes");
1961 CHECK(dns_journal_write_transaction(journal, &resultdiff));
1963 INSIST(ISC_LIST_EMPTY(diff[0].tuples));
1964 INSIST(ISC_LIST_EMPTY(diff[1].tuples));
1967 dns_diff_clear(&resultdiff);
1968 dns_dbiterator_destroy(&dbit[1]);
1970 dns_dbiterator_destroy(&dbit[0]);
1972 dns_journal_destroy(&journal);
1977 dns_journal_compact(isc_mem_t *mctx, char *filename, isc_uint32_t serial,
1978 isc_uint32_t target_size)
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;
1989 unsigned int size = 0;
1990 isc_result_t result;
1991 unsigned int indexend;
1994 isc_boolean_t is_backup = ISC_FALSE;
1996 namelen = strlen(filename);
1997 if (namelen > 4 && strcmp(filename + namelen - 4, ".jnl") == 0)
2000 result = isc_string_printf(newname, sizeof(newname), "%.*s.jnw",
2002 if (result != ISC_R_SUCCESS)
2005 result = isc_string_printf(backup, sizeof(backup), "%.*s.jbk",
2007 if (result != ISC_R_SUCCESS)
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);
2015 if (result != ISC_R_SUCCESS)
2018 if (JOURNAL_EMPTY(&j->header)) {
2019 dns_journal_destroy(&j);
2020 return (ISC_R_SUCCESS);
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);
2030 * Cope with very small target sizes.
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;
2038 * See if there is any work to do.
2040 if ((isc_uint32_t) j->header.end.offset < target_size) {
2041 dns_journal_destroy(&j);
2042 return (ISC_R_SUCCESS);
2045 CHECK(journal_open(mctx, newname, ISC_TRUE, ISC_TRUE, &new));
2048 * Remove overhead so space test below can succeed.
2050 if (target_size >= indexend)
2051 target_size -= indexend;
2054 * Find if we can create enough free space.
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];
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)
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;
2081 INSIST(best_guess.serial != j->header.end.serial);
2082 if (best_guess.serial != serial)
2083 CHECK(journal_next(j, &best_guess));
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.
2090 copy_length = j->header.end.offset - best_guess.offset;
2092 if (copy_length != 0) {
2094 * Copy best_guess to end into space just freed.
2097 if (copy_length < size)
2099 buf = isc_mem_get(mctx, size);
2101 result = ISC_R_NOMEMORY;
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 :
2110 CHECK(journal_read(j, buf, len));
2111 CHECK(journal_write(new, buf, len));
2114 CHECK(journal_fsync(new));
2117 * Compute new header.
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;
2125 * Update the journal header.
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));
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));
2144 CHECK(index_to_disk(new));
2145 CHECK(journal_fsync(new));
2147 indexend = new->header.end.offset;
2149 dns_journal_destroy(&new);
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.
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.
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)
2166 if (rename(filename, backup) == -1)
2168 if (rename(newname, filename) == -1)
2170 (void)isc_file_remove(backup);
2173 result = ISC_R_FAILURE;
2178 dns_journal_destroy(&j);
2179 result = ISC_R_SUCCESS;
2182 (void)isc_file_remove(newname);
2184 isc_mem_put(mctx, buf, size);
2186 dns_journal_destroy(&j);
2188 dns_journal_destroy(&new);
2193 index_to_disk(dns_journal_t *j) {
2194 isc_result_t result = ISC_R_SUCCESS;
2196 if (j->header.index_size != 0) {
2199 unsigned int rawbytes;
2201 rawbytes = j->header.index_size * sizeof(journal_rawpos_t);
2204 for (i = 0; i < j->header.index_size; i++) {
2205 encode_uint32(j->index[i].serial, p);
2207 encode_uint32(j->index[i].offset, p);
2210 INSIST(p == j->rawindex + rawbytes);
2212 CHECK(journal_seek(j, sizeof(journal_rawheader_t)));
2213 CHECK(journal_write(j, j->rawindex, rawbytes));