log0recv.c

这是linux下运行的mysql软件包,可用于linux 下安装 php + mysql + apach 的网络配置

	        /* The address hash table is externally chained */
		recv_addr = hash_get_nth_cell(recv_sys->addr_hash, i)->node;

		while (recv_addr != NULL) {

			if (!fil_tablespace_exists_in_mem(recv_addr->space)) {
/*
				fprintf(stderr,
"InnoDB: Warning: cannot apply log record to tablespace %lu page %lu,\n"
"InnoDB: because tablespace with that id does not exist.\n",
				      recv_addr->space, recv_addr->page_no);
*/
				recv_addr->state = RECV_PROCESSED;

				ut_a(recv_sys->n_addrs);
				recv_sys->n_addrs--;

				goto skip_this_recv_addr;
			}

			/* We simulate a page read made by the buffer pool, to
			make sure the recovery apparatus works ok, for
			example, the buf_frame_align() function. We must init
			the block corresponding to buf_pool->frame_zero
			(== page). */

			buf_page_init_for_backup_restore(recv_addr->space,
						recv_addr->page_no,
						buf_block_align(page));

			/* Extend the tablespace's last file if the page_no
			does not fall inside its bounds; we assume the last
			file is auto-extending, and ibbackup copied the file
			when it still was smaller */

			success = fil_extend_space_to_desired_size(
						&actual_size,
						recv_addr->space,
						recv_addr->page_no + 1);
			if (!success) {
			  fprintf(stderr,
"InnoDB: Fatal error: cannot extend tablespace %lu to hold %lu pages\n",
				     recv_addr->space, recv_addr->page_no);
				     
				exit(1);
			}

			/* Read the page from the tablespace file using the
			fil0fil.c routines */

			error = fil_io(OS_FILE_READ, TRUE, recv_addr->space,
					recv_addr->page_no, 0, UNIV_PAGE_SIZE,
					page, NULL);
			if (error != DB_SUCCESS) {
			  fprintf(stderr,
"InnoDB: Fatal error: cannot read from tablespace %lu page number %lu\n",
				     (ulong) recv_addr->space, (ulong) recv_addr->page_no);
				     
				exit(1);
			}

			/* Apply the log records to this page */
			recv_recover_page(TRUE, FALSE, page, recv_addr->space,
						       recv_addr->page_no);

			/* Write the page back to the tablespace file using the
			fil0fil.c routines */

			buf_flush_init_for_writing(page,
				mach_read_from_8(page + FIL_PAGE_LSN),
				recv_addr->space, recv_addr->page_no);

			error = fil_io(OS_FILE_WRITE, TRUE, recv_addr->space,
					recv_addr->page_no, 0, UNIV_PAGE_SIZE,
					page, NULL);
skip_this_recv_addr:
			recv_addr = HASH_GET_NEXT(addr_hash, recv_addr);
		}

		if ((100 * i) / n_hash_cells
				!= (100 * (i + 1)) / n_hash_cells) {
			fprintf(stderr, "%lu ",
                                (ulong) ((100 * i) / n_hash_cells));
			fflush(stderr);
		}
	}

	recv_sys_empty_hash();
}

#ifdef notdefined
/***********************************************************************
In the debug version, updates the replica of a file page, based on a log
record. */
static
void
recv_update_replicate(
/*==================*/
	byte	type,	/* in: log record type */
	ulint	space,	/* in: space id */
	ulint	page_no,/* in: page number */
	byte*	body,	/* in: log record body */
	byte*	end_ptr)/* in: log record end */
{
	page_t*	replica;
	mtr_t	mtr;
	byte*	ptr;

	mtr_start(&mtr);

	mtr_set_log_mode(&mtr, MTR_LOG_NONE);

	replica = buf_page_get(space + RECV_REPLICA_SPACE_ADD, page_no,
							RW_X_LATCH, &mtr);
#ifdef UNIV_SYNC_DEBUG
	buf_page_dbg_add_level(replica, SYNC_NO_ORDER_CHECK);
#endif /* UNIV_SYNC_DEBUG */
							
	ptr = recv_parse_or_apply_log_rec_body(type, body, end_ptr, replica,
									&mtr);
	ut_a(ptr == end_ptr);

	/* Notify the buffer manager that the page has been updated */

	buf_flush_recv_note_modification(buf_block_align(replica),
					log_sys->old_lsn, log_sys->old_lsn);

	/* Make sure that committing mtr does not call log routines, as
	we currently own the log mutex */
	
	mtr.modifications = FALSE;

	mtr_commit(&mtr);
}

/***********************************************************************
Checks that two strings are identical. */
static
void
recv_check_identical(
/*=================*/
	byte*	str1,	/* in: first string */
	byte*	str2,	/* in: second string */
	ulint	len)	/* in: length of strings */
{
	ulint	i;

	for (i = 0; i < len; i++) {

		if (str1[i] != str2[i]) {
			fprintf(stderr,
				"Strings do not match at offset %lu\n", i);
			ut_print_buf(str1 + i, 16);
			fprintf(stderr, "\n");
			ut_print_buf(str2 + i, 16);

			ut_error;
		}
	}
}	
			
/***********************************************************************
In the debug version, checks that the replica of a file page is identical
to the original page. */
static
void
recv_compare_replicate(
/*===================*/
	ulint	space,	/* in: space id */
	ulint	page_no)/* in: page number */
{
	page_t*	replica;
	page_t*	page;
	mtr_t	mtr;

	mtr_start(&mtr);

	mutex_enter(&(buf_pool->mutex));

	page = buf_page_hash_get(space, page_no)->frame;

	mutex_exit(&(buf_pool->mutex));

	replica = buf_page_get(space + RECV_REPLICA_SPACE_ADD, page_no,
							RW_X_LATCH, &mtr);
#ifdef UNIV_SYNC_DEBUG
	buf_page_dbg_add_level(replica, SYNC_NO_ORDER_CHECK);
#endif /* UNIV_SYNC_DEBUG */

	recv_check_identical(page + FIL_PAGE_DATA,
			replica + FIL_PAGE_DATA,
			PAGE_HEADER + PAGE_MAX_TRX_ID - FIL_PAGE_DATA);

	recv_check_identical(page + PAGE_HEADER + PAGE_MAX_TRX_ID + 8,
			replica + PAGE_HEADER + PAGE_MAX_TRX_ID + 8,
			UNIV_PAGE_SIZE - FIL_PAGE_DATA_END
			- PAGE_HEADER - PAGE_MAX_TRX_ID - 8);
	mtr_commit(&mtr);
}

/***********************************************************************
Checks that a replica of a space is identical to the original space. */

void
recv_compare_spaces(
/*================*/
	ulint	space1,	/* in: space id */
	ulint	space2,	/* in: space id */
	ulint	n_pages)/* in: number of pages */
{
	page_t*	replica;
	page_t*	page;
	mtr_t	mtr;
	page_t*	frame;
	ulint	page_no;

	replica = buf_frame_alloc();
	page = buf_frame_alloc();

	for (page_no = 0; page_no < n_pages; page_no++) {
	
		mtr_start(&mtr);

		frame = buf_page_get_gen(space1, page_no, RW_S_LATCH, NULL,
						BUF_GET_IF_IN_POOL,
						__FILE__, __LINE__,
						&mtr);
		if (frame) {
#ifdef UNIV_SYNC_DEBUG
			buf_page_dbg_add_level(frame, SYNC_NO_ORDER_CHECK);
#endif /* UNIV_SYNC_DEBUG */
			ut_memcpy(page, frame, UNIV_PAGE_SIZE);
		} else {
			/* Read it from file */
			fil_io(OS_FILE_READ, TRUE, space1, page_no, 0,
						UNIV_PAGE_SIZE, page, NULL);
		}

		frame = buf_page_get_gen(space2, page_no, RW_S_LATCH, NULL,
						BUF_GET_IF_IN_POOL,
						__FILE__, __LINE__,
						&mtr);
		if (frame) {
#ifdef UNIV_SYNC_DEBUG
			buf_page_dbg_add_level(frame, SYNC_NO_ORDER_CHECK);
#endif /* UNIV_SYNC_DEBUG */
			ut_memcpy(replica, frame, UNIV_PAGE_SIZE);
		} else {
			/* Read it from file */
			fil_io(OS_FILE_READ, TRUE, space2, page_no, 0,
				UNIV_PAGE_SIZE, replica, NULL);
		}
		
		recv_check_identical(page + FIL_PAGE_DATA,
			replica + FIL_PAGE_DATA,
			PAGE_HEADER + PAGE_MAX_TRX_ID - FIL_PAGE_DATA);

		recv_check_identical(page + PAGE_HEADER + PAGE_MAX_TRX_ID + 8,
			replica + PAGE_HEADER + PAGE_MAX_TRX_ID + 8,
			UNIV_PAGE_SIZE - FIL_PAGE_DATA_END
			- PAGE_HEADER - PAGE_MAX_TRX_ID - 8);

		mtr_commit(&mtr);
	}

	buf_frame_free(replica);
	buf_frame_free(page);
}

/***********************************************************************
Checks that a replica of a space is identical to the original space. Disables
ibuf operations and flushes and invalidates the buffer pool pages after the
test. This function can be used to check the recovery before dict or trx
systems are initialized. */

void
recv_compare_spaces_low(
/*====================*/
	ulint	space1,	/* in: space id */
	ulint	space2,	/* in: space id */
	ulint	n_pages)/* in: number of pages */
{
	mutex_enter(&(log_sys->mutex));

	recv_apply_hashed_log_recs(FALSE);
	
	mutex_exit(&(log_sys->mutex));

	recv_compare_spaces(space1, space2, n_pages);
}
#endif /* UNIV_LOG_REPLICATE */

/***********************************************************************
Tries to parse a single log record and returns its length. */
static
ulint
recv_parse_log_rec(
/*===============*/
			/* out: length of the record, or 0 if the record was
			not complete */
	byte*	ptr,	/* in: pointer to a buffer */
	byte*	end_ptr,/* in: pointer to the buffer end */
	byte*	type,	/* out: type */
	ulint*	space,	/* out: space id */
	ulint*	page_no,/* out: page number */
	byte**	body)	/* out: log record body start */
{
	byte*	new_ptr;

	*body = NULL;

	if (ptr == end_ptr) {

		return(0);
	}

	if (*ptr == MLOG_MULTI_REC_END) {
	
		*type = *ptr;

		return(1);
	}

	if (*ptr == MLOG_DUMMY_RECORD) {
		*type = *ptr;

		*space = ULINT_UNDEFINED - 1; /* For debugging */

		return(1);
	}

	new_ptr = mlog_parse_initial_log_record(ptr, end_ptr, type, space,
								page_no);
	*body = new_ptr;

	if (UNIV_UNLIKELY(!new_ptr)) {

	        return(0);
	}

	/* Check that page_no is sensible */

	if (UNIV_UNLIKELY(*page_no > 0x8FFFFFFFUL)) {

		recv_sys->found_corrupt_log = TRUE;

		return(0);
	}

	new_ptr = recv_parse_or_apply_log_rec_body(*type, new_ptr, end_ptr,
								NULL, NULL);
	if (UNIV_UNLIKELY(new_ptr == NULL)) {

		return(0);
	}

	if (*page_no > recv_max_parsed_page_no) {
		recv_max_parsed_page_no = *page_no;
	}
	
	return(new_ptr - ptr);
}

/***********************************************************
Calculates the new value for lsn when more data is added to the log. */
static
dulint
recv_calc_lsn_on_data_add(
/*======================*/
	dulint	lsn,	/* in: old lsn */
	ulint	len)	/* in: this many bytes of data is added, log block
			headers not included */
{
	ulint	frag_len;
	ulint	lsn_len;
	
	frag_len = (ut_dulint_get_low(lsn) % OS_FILE_LOG_BLOCK_SIZE)
		   					- LOG_BLOCK_HDR_SIZE;
	ut_ad(frag_len < OS_FILE_LOG_BLOCK_SIZE - LOG_BLOCK_HDR_SIZE
		      					- LOG_BLOCK_TRL_SIZE);
	lsn_len = len + ((len + frag_len)
		    	 / (OS_FILE_LOG_BLOCK_SIZE - LOG_BLOCK_HDR_SIZE
		      					- LOG_BLOCK_TRL_SIZE))
		     	 * (LOG_BLOCK_HDR_SIZE + LOG_BLOCK_TRL_SIZE);

	return(ut_dulint_add(lsn, lsn_len));
}

/***********************************************************
Checks that the parser recognizes incomplete initial segments of a log
record as incomplete. */

void
recv_check_incomplete_log_recs(
/*===========================*/
	byte*	ptr,	/* in: pointer to a complete log record */
	ulint	len)	/* in: length of the log record */
{
	ulint	i;
	byte	type;
	ulint	space;
	ulint	page_no;
	byte*	body;
	
	for (i = 0; i < len; i++) {
		ut_a(0 == recv_parse_log_rec(ptr, ptr + i, &type, &space,
							&page_no, &body));
	}
}		

/***********************************************************
Prints diagnostic info of corrupt log. */
static
void
recv_report_corrupt_log(
/*====================*/
	byte*	ptr,	/* in: pointer to corrupt log record */
	byte	type,	/* in: type of the record */
	ulint	space,	/* in: space id, this may also be garbage */
	ulint	page_no)/* in: page number, this may also be garbage */
{
	fprintf(stderr,
"InnoDB: ############### CORRUPT LOG RECORD FOUND\n"
"InnoDB: Log record type %lu, space id %lu, page number %lu\n"
"InnoDB: Log parsing proceeded successfully up to %lu %lu\n"
"InnoDB: Previous log record type %lu, is multi %lu\n"
"InnoDB: Recv offset %lu, prev %lu\n",
	(ulong) type, (ulong) space, (ulong) page_no,
	(ulong) ut_dulint_get_high(recv_sys->recovered_lsn),
        (ulong) ut_dulint_get_low(recv_sys->recovered_lsn),
	(ulong) recv_previous_parsed_rec_type,
	(ulong) recv_previous_parsed_rec_is_multi,
	(ulong) (ptr - recv_sys->buf),
	(ulong) recv_previous_parsed_rec_offset);

	if ((ulint)(ptr - recv_sys->buf + 100)
					> recv_previous_parsed_rec_offset
	    && (ulint)(ptr - recv_sys->buf + 100
					- recv_previous_parsed_rec_offset)
	       < 200000) {
		fputs(
"InnoDB: Hex dump of corrupt log starting 100 bytes before the start\n"
"InnoDB: of the previous log rec,\n"
"InnoDB: and ending 100 bytes after the start of the corrupt rec:\n",
			stderr);
 
		ut_print_buf(stderr,
		     recv_sys->buf + recv_previous_parsed_rec_offset - 100,
		     ptr - recv_sys->buf + 200 -
					recv_previous_parsed_rec_offset);
		putc('\n', stderr);
	}

	fputs(
	"InnoDB: WARNING: the log file may have been corrupt and it\n"
	"InnoDB: is possible that the log scan did not proceed\n"
	"InnoDB: far enough in recovery! Please run CHECK TABLE\n"
	"InnoDB: on your InnoDB tables to check that they are ok!\n"
	"InnoDB: If mysqld crashes after this recovery, look at\n"
	"InnoDB: http://dev.mysql.com/doc/mysql/en/Forcing_recovery.html\n"
	"InnoDB: about forcing recovery.\n", stderr);

	fflush(stderr);
}

/***********************************************************
Parses log records from a buffer and stores them to a hash table to wait
merging to file pages. */
static
ibool
recv_parse_log_recs(
/*================*/
				/* out: currently always returns FALSE */
	ibool	store_to_hash)	/* in: TRUE if the records should be stored
				to the hash table; this is set to FALSE if just
				debug checking is needed */
{
	byte*	ptr;
	byte*	end_ptr;
	ulint	single_rec;
	ulint	len;
	ulint	total_len;
	dulint	new_recovered_lsn;
	dulint	old_lsn;
	byte	type;
	ulint	space;
	ulint	page_no;
	byte*	body;
	ulint	n_recs;
	
#ifdef UNIV_SYNC_DEBUG
	ut_ad(mutex_own(&(log_sys->mutex)));
#endif /* UNIV_SYNC_DEBUG */
	ut_ad(!ut_dulint_is_zero(recv_sys->parse_start_lsn));
loop:
	ptr = recv_sys->buf + recv_sys->recovered_offset;

	end_ptr = recv_sys->buf + recv_sys->len;