journal.c

bind 9.3结合mysql数据库

				 "%s: close: %s",
				 filename, isc_result_totext(result));
		(void)isc_file_remove(filename);
		return (ISC_R_UNEXPECTED);
	}

	return (ISC_R_SUCCESS);
}

static isc_result_t
journal_open(isc_mem_t *mctx, const char *filename, isc_boolean_t write,
	     isc_boolean_t create, dns_journal_t **journalp) {
	FILE *fp = NULL;
	isc_result_t result;
	journal_rawheader_t rawheader;
	dns_journal_t *j;

	INSIST(journalp != NULL && *journalp == NULL);
	j = isc_mem_get(mctx, sizeof(*j));
	if (j == NULL)
		return (ISC_R_NOMEMORY);

	j->mctx = mctx;
	j->state = JOURNAL_STATE_INVALID;
	j->fp = NULL;
	j->filename = filename;
	j->index = NULL;
	j->rawindex = NULL;

	result = isc_stdio_open(j->filename, write ? "rb+" : "rb", &fp);

	if (result == ISC_R_FILENOTFOUND) {
		if (create) {
			isc_log_write(JOURNAL_COMMON_LOGARGS,
				      ISC_LOG_INFO,
				      "journal file %s does not exist, "
				      "creating it",
				      j->filename);
			CHECK(journal_file_create(mctx, filename));
			/*
			 * Retry.
			 */
			result = isc_stdio_open(j->filename, "rb+", &fp);
		} else {
			FAIL(ISC_R_NOTFOUND);
		}
	}
	if (result != ISC_R_SUCCESS) {
		isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR,
			      "%s: open: %s",
			      j->filename, isc_result_totext(result));
		FAIL(ISC_R_UNEXPECTED);
	}

	j->fp = fp;

	/*
	 * Set magic early so that seek/read can succeed.
	 */
	j->magic = DNS_JOURNAL_MAGIC;

	CHECK(journal_seek(j, 0));
	CHECK(journal_read(j, &rawheader, sizeof(rawheader)));

	if (memcmp(rawheader.h.format, initial_journal_header.format,
		   sizeof(initial_journal_header.format)) != 0) {
		isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR,
				 "%s: journal format not recognized",
				 j->filename);
		FAIL(ISC_R_UNEXPECTED);
	}
	journal_header_decode(&rawheader, &j->header);

	/*
	 * If there is an index, read the raw index into a dynamically
	 * allocated buffer and then convert it into a cooked index.
	 */
	if (j->header.index_size != 0) {
		unsigned int i;
		unsigned int rawbytes;
		unsigned char *p;

		rawbytes = j->header.index_size * sizeof(journal_rawpos_t);
		j->rawindex = isc_mem_get(mctx, rawbytes);
		if (j->rawindex == NULL)
			FAIL(ISC_R_NOMEMORY);

		CHECK(journal_read(j, j->rawindex, rawbytes));

		j->index = isc_mem_get(mctx, j->header.index_size *
				       sizeof(journal_pos_t));
		if (j->index == NULL)
			FAIL(ISC_R_NOMEMORY);

		p = j->rawindex;
		for (i = 0; i < j->header.index_size; i++) {
			j->index[i].serial = decode_uint32(p);
			p += 4;
			j->index[i].offset = decode_uint32(p);
			p += 4;
		}
		INSIST(p == j->rawindex + rawbytes);
	}
	j->offset = -1; /* Invalid, must seek explicitly. */

	/*
	 * Initialize the iterator.
	 */
	dns_name_init(&j->it.name, NULL);
	dns_rdata_init(&j->it.rdata);

	/*
	 * Set up empty initial buffers for uncheched and checked
	 * wire format RR data.  They will be reallocated
	 * later.
	 */
	isc_buffer_init(&j->it.source, NULL, 0);
	isc_buffer_init(&j->it.target, NULL, 0);
	dns_decompress_init(&j->it.dctx, -1, DNS_DECOMPRESS_NONE);

	j->state =
		write ? JOURNAL_STATE_WRITE : JOURNAL_STATE_READ;

	*journalp = j;
	return (ISC_R_SUCCESS);

 failure:
	j->magic = 0;
	if (j->index != NULL) {
		isc_mem_put(j->mctx, j->index, j->header.index_size *
			    sizeof(journal_rawpos_t));
		j->index = NULL;
	}
	if (j->fp != NULL)
		(void)isc_stdio_close(j->fp);
	isc_mem_put(j->mctx, j, sizeof(*j));
	return (result);
}

isc_result_t
dns_journal_open(isc_mem_t *mctx, const char *filename, isc_boolean_t write,
		 dns_journal_t **journalp) {
	return (journal_open(mctx, filename, write, write, journalp));
}

/*
 * A comparison function defining the sorting order for
 * entries in the IXFR-style journal file.
 *
 * The IXFR format requires that deletions are sorted before
 * additions, and within either one, SOA records are sorted
 * before others.
 *
 * Also sort the non-SOA records by type as a courtesy to the
 * server receiving the IXFR - it may help reduce the amount of
 * rdataset merging it has to do.
 */
static int
ixfr_order(const void *av, const void *bv) {
	dns_difftuple_t const * const *ap = av;
	dns_difftuple_t const * const *bp = bv;
	dns_difftuple_t const *a = *ap;
	dns_difftuple_t const *b = *bp;
	int r;

	r = (b->op == DNS_DIFFOP_DEL) - (a->op == DNS_DIFFOP_DEL);
	if (r != 0)
		return (r);

	r = (b->rdata.type == dns_rdatatype_soa) -
		(a->rdata.type == dns_rdatatype_soa);
	if (r != 0)
		return (r);

	r = (a->rdata.type - b->rdata.type);
	return (r);
}

/*
 * Advance '*pos' to the next journal transaction.
 *
 * Requires:
 *	*pos refers to a valid journal transaction.
 *
 * Ensures:
 *	When ISC_R_SUCCESS is returned,
 *	*pos refers to the next journal transaction.
 *
 * Returns one of:
 *
 *    ISC_R_SUCCESS
 *    ISC_R_NOMORE 	*pos pointed at the last transaction
 *    Other results due to file errors are possible.
 */
static isc_result_t
journal_next(dns_journal_t *j, journal_pos_t *pos) {
	isc_result_t result;
	journal_xhdr_t xhdr;
	REQUIRE(DNS_JOURNAL_VALID(j));

	result = journal_seek(j, pos->offset);
	if (result != ISC_R_SUCCESS)
		return (result);

	if (pos->serial == j->header.end.serial)
		return (ISC_R_NOMORE);
	/*
	 * Read the header of the current transaction.
	 * This will return ISC_R_NOMORE if we are at EOF.
	 */
	result = journal_read_xhdr(j, &xhdr);
	if (result != ISC_R_SUCCESS)
		return (result);

	/*
	 * Check serial number consistency.
	 */
	if (xhdr.serial0 != pos->serial) {
		isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR,
			      "%s: journal file corrupt: "
			      "expected serial %u, got %u",
			      j->filename, pos->serial, xhdr.serial0);
		return (ISC_R_UNEXPECTED);
	}

	/*
	 * Check for offset wraparound.
	 */
	if ((isc_offset_t)(pos->offset + sizeof(journal_rawxhdr_t) + xhdr.size)
	    < pos->offset) {
		isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR,
			      "%s: offset too large", j->filename);
		return (ISC_R_UNEXPECTED);
	}

	pos->offset += sizeof(journal_rawxhdr_t) + xhdr.size;
	pos->serial = xhdr.serial1;
	return (ISC_R_SUCCESS);
}

/*
 * If the index of the journal 'j' contains an entry "better"
 * than '*best_guess', replace '*best_guess' with it.
 *
 * "Better" means having a serial number closer to 'serial'
 * but not greater than 'serial'.
 */
static void
index_find(dns_journal_t *j, isc_uint32_t serial, journal_pos_t *best_guess) {
	unsigned int i;
	if (j->index == NULL)
		return;
	for (i = 0; i < j->header.index_size; i++) {
		if (POS_VALID(j->index[i]) &&
		    DNS_SERIAL_GE(serial, j->index[i].serial) &&
		    DNS_SERIAL_GT(j->index[i].serial, best_guess->serial))
			*best_guess = j->index[i];
	}
}

/*
 * Add a new index entry.  If there is no room, make room by removing
 * the odd-numbered entries and compacting the others into the first
 * half of the index.  This decimates old index entries exponentially
 * over time, so that the index always contains a much larger fraction
 * of recent serial numbers than of old ones.  This is deliberate -
 * most index searches are for outgoing IXFR, and IXFR tends to request
 * recent versions more often than old ones.
 */
static void
index_add(dns_journal_t *j, journal_pos_t *pos) {
	unsigned int i;
	if (j->index == NULL)
		return;
	/*
	 * Search for a vacant position.
	 */
	for (i = 0; i < j->header.index_size; i++) {
		if (! POS_VALID(j->index[i]))
			break;
	}
	if (i == j->header.index_size) {
		unsigned int k = 0;
		/*
		 * Found no vacant position.  Make some room.
		 */
		for (i = 0; i < j->header.index_size; i += 2) {
			j->index[k++] = j->index[i];
		}
		i = k; /* 'i' identifies the first vacant position. */
		while (k < j->header.index_size) {
			POS_INVALIDATE(j->index[k]);
			k++;
		}
	}
	INSIST(i < j->header.index_size);
	INSIST(! POS_VALID(j->index[i]));

	/*
	 * Store the new index entry.
	 */
	j->index[i] = *pos;
}

/*
 * Invalidate any existing index entries that could become
 * ambiguous when a new transaction with number 'serial' is added.
 */
static void
index_invalidate(dns_journal_t *j, isc_uint32_t serial) {
	unsigned int i;
	if (j->index == NULL)
		return;
	for (i = 0; i < j->header.index_size; i++) {
		if (! DNS_SERIAL_GT(serial, j->index[i].serial))
			POS_INVALIDATE(j->index[i]);
	}
}

/*
 * Try to find a transaction with initial serial number 'serial'
 * in the journal 'j'.
 *
 * If found, store its position at '*pos' and return ISC_R_SUCCESS.
 *
 * If 'serial' is current (= the ending serial number of the
 * last transaction in the journal), set '*pos' to
 * the position immediately following the last transaction and
 * return ISC_R_SUCCESS.
 *
 * If 'serial' is within the range of addressable serial numbers
 * covered by the journal but that particular serial number is missing
 * (from the journal, not just from the index), return ISC_R_NOTFOUND.
 *
 * If 'serial' is outside the range of addressable serial numbers
 * covered by the journal, return ISC_R_RANGE.
 *
 */
static isc_result_t
journal_find(dns_journal_t *j, isc_uint32_t serial, journal_pos_t *pos) {
	isc_result_t result;
	journal_pos_t current_pos;
	REQUIRE(DNS_JOURNAL_VALID(j));

	if (DNS_SERIAL_GT(j->header.begin.serial, serial))
		return (ISC_R_RANGE);
	if (DNS_SERIAL_GT(serial, j->header.end.serial))
		return (ISC_R_RANGE);
	if (serial == j->header.end.serial) {
		*pos = j->header.end;
		return (ISC_R_SUCCESS);
	}

	current_pos = j->header.begin;
	index_find(j, serial, &current_pos);

	while (current_pos.serial != serial) {
		if (DNS_SERIAL_GT(current_pos.serial, serial))
			return (ISC_R_NOTFOUND);
		result = journal_next(j, &current_pos);
		if (result != ISC_R_SUCCESS)
			return (result);
	}
	*pos = current_pos;
	return (ISC_R_SUCCESS);
}

isc_result_t
dns_journal_begin_transaction(dns_journal_t *j) {
	isc_uint32_t offset;
	isc_result_t result;
	journal_rawxhdr_t hdr;

	REQUIRE(DNS_JOURNAL_VALID(j));
	REQUIRE(j->state == JOURNAL_STATE_WRITE);

	/*
	 * Find the file offset where the new transaction should
	 * be written, and seek there.
	 */
	if (JOURNAL_EMPTY(&j->header)) {
		offset = sizeof(journal_rawheader_t) +
			j->header.index_size * sizeof(journal_rawpos_t);
	} else {
		offset = j->header.end.offset;
	}
	j->x.pos[0].offset = offset;
	j->x.pos[1].offset = offset; /* Initial value, will be incremented. */
	j->x.n_soa = 0;

	CHECK(journal_seek(j, offset));

	/*
	 * Write a dummy transaction header of all zeroes to reserve
	 * space.  It will be filled in when the transaction is
	 * finished.
	 */
	memset(&hdr, 0, sizeof(hdr));
	CHECK(journal_write(j, &hdr, sizeof(hdr)));
	j->x.pos[1].offset = j->offset;

	j->state = JOURNAL_STATE_TRANSACTION;
	result = ISC_R_SUCCESS;
 failure:
	return (result);
}

isc_result_t
dns_journal_writediff(dns_journal_t *j, dns_diff_t *diff) {
	dns_difftuple_t *t;
	isc_buffer_t buffer;
	void *mem = NULL;
	unsigned int size;
	isc_result_t result;
	isc_region_t used;

	REQUIRE(DNS_DIFF_VALID(diff));
	REQUIRE(j->state == JOURNAL_STATE_TRANSACTION);

	isc_log_write(JOURNAL_DEBUG_LOGARGS(3), "writing to journal");
	(void)dns_diff_print(diff, NULL);

	/*
	 * Pass 1: determine the buffer size needed, and
	 * keep track of SOA serial numbers.
	 */
	size = 0;
	for (t = ISC_LIST_HEAD(diff->tuples); t != NULL;
	     t = ISC_LIST_NEXT(t, link))
	{
		if (t->rdata.type == dns_rdatatype_soa) {
			if (j->x.n_soa < 2)
				j->x.pos[j->x.n_soa].serial =
					dns_soa_getserial(&t->rdata);
			j->x.n_soa++;
		}
		size += sizeof(journal_rawrrhdr_t);
		size += t->name.length; /* XXX should have access macro? */
		size += 10;
		size += t->rdata.length;
	}

	mem = isc_mem_get(j->mctx, size);
	if (mem == NULL)
		return (ISC_R_NOMEMORY);

	isc_buffer_init(&buffer, mem, size);

	/*
	 * Pass 2.  Write RRs to buffer.
	 */
	for (t = ISC_LIST_HEAD(diff->tuples); t != NULL;
	     t = ISC_LIST_NEXT(t, link))
	{
		/*
		 * Write the RR header.
		 */
		isc_buffer_putuint32(&buffer, t->name.length + 10 +
				     t->rdata.length);
		/*
		 * Write the owner name, RR header, and RR data.
		 */
		isc_buffer_putmem(&buffer, t->name.ndata, t->name.length);
		isc_buffer_putuint16(&buffer, t->rdata.type);
		isc_buffer_putuint16(&buffer, t->rdata.rdclass);
		isc_buffer_putuint32(&buffer, t->ttl);
		INSIST(t->rdata.length < 65536);
		isc_buffer_putuint16(&buffer, (isc_uint16_t)t->rdata.length);
		INSIST(isc_buffer_availablelength(&buffer) >= t->rdata.length);
		isc_buffer_putmem(&buffer, t->rdata.data, t->rdata.length);
	}

	isc_buffer_usedregion(&buffer, &used);
	INSIST(used.length == size);

	j->x.pos[1].offset += used.length;

	/*
	 * Write the buffer contents to the journal file.
	 */
	CHECK(journal_write(j, used.base, used.length));

	result = ISC_R_SUCCESS;

 failure:
	if (mem != NULL)
		isc_mem_put(j->mctx, mem, size);
	return (result);

}

isc_result_t
dns_journal_commit(dns_journal_t *j) {
	isc_result_t result;
	journal_rawheader_t rawheader;

	REQUIRE(DNS_JOURNAL_VALID(j));
	REQUIRE(j->state == JOURNAL_STATE_TRANSACTION);

	/*
	 * Perform some basic consistency checks.
	 */
	if (j->x.n_soa != 2) {
		isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR,
			      "malformed transaction: %d SOAs",
			      j->x.n_soa);
		return (ISC_R_UNEXPECTED);
	}
	if (! (DNS_SERIAL_GT(j->x.pos[1].serial, j->x.pos[0].serial) ||
	       (bind8_compat &&
		j->x.pos[1].serial == j->x.pos[0].serial)))
	{
		isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR,
			      "malformed transaction: serial number "
			      "would decrease");
		return (ISC_R_UNEXPECTED);
	}
	if (! JOURNAL_EMPTY(&j->header)) {
		if (j->x.pos[0].serial != j->header.end.serial) {
			isc_log_write(JOURNAL_COMMON_LOGARGS, ISC_LOG_ERROR,
					 "malformed transaction: "
					 "%s last serial %u != "
					 "transaction first serial %u",
					 j->filename,
					 j->header.end.serial,
					 j->x.pos[0].serial);
			return (ISC_R_UNEXPECTED);
		}
	}

	/*
	 * Some old journal entries may become non-addressable
	 * when we increment the current serial number.  Purge them