MDEV-37138: Innochecksum fails to handle doublewrite buffer and

		multiple file tablespace

Problem:
=======
- innochecksum was incorrectly interpreting doublewrite buffer
pages as index pages, causing confusion about stale tables
in the system tablespace.

- innochecksum fails to parse the multi-file system tablespace

Solution:
========
1. Rewrite checksum of doublewrite buffer pages
are skipped.

2. Introduced the option --tablespace-flags which can be used
to initialize page size. This option can handle the ibdata2,
ibdata3 etc without parsing ibdata1.

This is a cherry-pick of commit 9f8716a

Author: Thirunarayanan

extra/innochecksum.cc

@@ -48,9 +48,8 @@ The parts not included are excluded by #ifndef UNIV_INNOCHECKSUM. */
 #include "buf0buf.h"             /* buf_page_is_corrupted */
 #include "page0zip.h"            /* page_zip_*() */
 #include "trx0undo.h"            /* TRX_* */
-#include "ut0crc32.h"            /* ut_crc32_init() */
+#include "ut0crc32.h"		 /* ut_crc32_init() */
 #include "fil0crypt.h"           /* fil_space_verify_crypt_checksum */
-
 #include <string.h>
 
 #ifdef UNIV_NONINL
@@ -79,6 +78,8 @@ static ulint extent_size;
 static ulint xdes_size;
 ulong srv_page_size;
 ulong srv_page_size_shift;
+static uint32_t dblwr_1;
+static uint32_t dblwr_2;
 /* Current page number (0 based). */
 uint32_t		cur_page_num;
 /* Current space. */
@@ -102,8 +103,10 @@ FILE*				log_file = NULL;
 /* Enabled for log write option. */
 static bool			is_log_enabled = false;
 static bool			skip_freed_pages;
+static uint32_t 		tablespace_flags= 0;
 static byte field_ref_zero_buf[UNIV_PAGE_SIZE_MAX];
 const byte *field_ref_zero = field_ref_zero_buf;
+constexpr uint32_t USE_FSP_FLAGS{UINT32_MAX};
 
 #ifndef _WIN32
 /* advisory lock for non-window system. */
@@ -258,12 +261,9 @@ void print_leaf_stats(
 }
 
 /** Init the page size for the tablespace.
-@param[in]	buf	buffer used to read the page */
-static void init_page_size(const byte* buf)
+@param[in]	flags	InnoDB tablespace flags */
+static void init_page_size_from_flags(const uint32_t flags)
 {
-	const unsigned	flags = mach_read_from_4(buf + FIL_PAGE_DATA
-						 + FSP_SPACE_FLAGS);
-
 	if (fil_space_t::full_crc32(flags)) {
 		const ulong ssize = FSP_FLAGS_FCRC32_GET_PAGE_SSIZE(flags);
 		srv_page_size_shift = UNIV_ZIP_SIZE_SHIFT_MIN - 1 + ssize;
@@ -550,24 +550,15 @@ is_page_corrupted(
 	return(is_corrupted);
 }
 
-/********************************************//*
- Check if page is doublewrite buffer or not.
- @param [in] page	buffer page
-
- @retval true  if page is doublewrite buffer otherwise false.
-*/
-static
-bool
-is_page_doublewritebuffer(
-	const byte*	page)
+/** Check if page is doublewrite buffer or not.
+@retval true  if page is doublewrite buffer otherwise false. */
+static bool is_page_doublewritebuffer()
 {
-	if ((cur_page_num >= extent_size)
-		&& (cur_page_num < extent_size * 3)) {
-		/* page is doublewrite buffer. */
-		return (true);
-	}
-
-	return (false);
+  if (cur_space != 0) return false;
+  const uint32_t extent{static_cast<uint32_t>(
+    cur_page_num & ~(extent_size - 1))};
+  return cur_page_num > FSP_DICT_HDR_PAGE_NO &&
+	 extent && (extent == dblwr_1 || extent == dblwr_2);
 }
 
 /*******************************************************//*
@@ -774,7 +765,7 @@ Parse the page and collect/dump the information about page type
 @param [in] file	file for diagnosis.
 @param [in] is_encrypted  tablespace is encrypted
 */
-void
+static void
 parse_page(
 	const byte*	page,
 	byte*		xdes,
@@ -794,6 +785,12 @@ parse_page(
 	str = skip_page ? "Double_write_buffer" : "-";
 	page_no = mach_read_from_4(page + FIL_PAGE_OFFSET);
 	if (skip_freed_pages) {
+
+		/** Skip doublewrite pages when -r is enabled */
+		if (is_page_doublewritebuffer()) {
+			return;
+		}
+
 		const byte *des= xdes + XDES_ARR_OFFSET +
 			xdes_size * ((page_no & (physical_page_size - 1))
 				     / extent_size);
@@ -988,6 +985,18 @@ parse_page(
 			fprintf(file, "#::" UINT32PF "\t\t|\t\tTransaction system "
 				"page\t\t|\t%s\n", cur_page_num, str);
 		}
+
+		if (cur_space == 0 &&
+		    (mach_read_from_4(page + TRX_SYS_DOUBLEWRITE +
+			             TRX_SYS_DOUBLEWRITE_MAGIC) ==
+		     TRX_SYS_DOUBLEWRITE_MAGIC_N))  {
+			dblwr_1 = mach_read_from_4(
+					page + TRX_SYS_DOUBLEWRITE +
+					TRX_SYS_DOUBLEWRITE_BLOCK1);
+			dblwr_2 = mach_read_from_4(
+					page + TRX_SYS_DOUBLEWRITE +
+					TRX_SYS_DOUBLEWRITE_BLOCK2);
+		}
 		break;
 
 	case FIL_PAGE_TYPE_FSP_HDR:
@@ -1230,6 +1239,9 @@ static struct my_option innochecksum_options[] = {
   {"skip-freed-pages", 'r', "skip freed pages for the tablespace",
    &skip_freed_pages, &skip_freed_pages, 0, GET_BOOL, NO_ARG,
    0, 0, 0, 0, 0, 0},
+  {"tablespace-flags", 0, "InnoDB tablespace flags (default: 4294967295 "
+   "= read from page 0)", &tablespace_flags, &tablespace_flags, 0,
+   GET_UINT, REQUIRED_ARG, USE_FSP_FLAGS, 0, USE_FSP_FLAGS, 0, 0, 0},
 
   {0, 0, 0, 0, 0, 0, GET_NO_ARG, NO_ARG, 0, 0, 0, 0, 0, 0}
 };
@@ -1304,6 +1316,14 @@ innochecksum_get_one_option(
 		my_end(0);
 		exit(EXIT_SUCCESS);
 		break;
+	default:
+	  if (tablespace_flags != USE_FSP_FLAGS &&
+	      !fil_space_t::is_valid_flags(tablespace_flags, false) &&
+	      !fil_space_t::is_valid_flags(tablespace_flags, true)) {
+		fprintf(stderr, "Error: Provided --tablespace-flags "
+			"is not valid.");
+		return true;
+	  }
 	}
 
 	return(false);
@@ -1434,6 +1454,87 @@ rewrite_checksum(
 		&& !write_file(filename, fil_in, buf, flags, pos);
 }
 
+/** Read and validate page 0, then initialize tablespace flags
+and page size.
+@param	fil_in        File pointer
+@param	buf           Buffer to read page into
+@return whether the page was read successfully */
+static bool read_and_validate_page0(FILE *fil_in, byte *buf)
+{
+  /* Read the minimum page size first */
+  size_t initial_page_size= UNIV_ZIP_SIZE_MIN;
+  if (tablespace_flags != USE_FSP_FLAGS)
+  {
+    init_page_size_from_flags(tablespace_flags);
+    initial_page_size= physical_page_size;
+  }
+
+  /* Read just enough to get the tablespace flags */
+  size_t bytes= fread(buf, 1, initial_page_size, fil_in);
+
+  if (bytes != initial_page_size)
+  {
+    fprintf(stderr, "Error: Was not able to read the "
+            "minimum page size of %zu bytes. Bytes read "
+            "was %zu\n", initial_page_size, bytes);
+    return false;
+  }
+
+  /* Read space_id and page offset */
+  cur_space= mach_read_from_4(buf + FIL_PAGE_SPACE_ID);
+  cur_page_num= mach_read_from_4(buf + FIL_PAGE_OFFSET);
+
+  /* Get tablespace flags from the FSP header */
+  uint32_t flags= mach_read_from_4(buf + FSP_HEADER_OFFSET +
+                                   FSP_SPACE_FLAGS);
+
+  if (tablespace_flags != USE_FSP_FLAGS)
+  {
+    if (cur_page_num == 0 && flags != tablespace_flags)
+      fprintf(stderr, "Error: Mismatch between provided tablespace "
+              "flags (0x%x) and file flags (0x%x)\n",
+              tablespace_flags, flags);
+  }
+  else
+  {
+    if (cur_page_num)
+    {
+      fprintf(stderr, "Error: First page of the tablespace file "
+              "should be 0, but encountered page number %" PRIu32 ". "
+              "If you are checking multi file system "
+              "tablespace files, please specify the correct "
+              "tablespace flags using --tablespace-flags option.\n",
+              cur_page_num);
+      return false;
+    }
+    /* Initialize page size parameters based on flags */
+    init_page_size_from_flags(flags);
+    /* Read the rest of the page if it's larger than the minimum size */
+    if (physical_page_size > UNIV_ZIP_SIZE_MIN)
+    {
+      /* Read rest of the page 0 to determine crypt_data */
+      ulint bytes= read_file(buf, true, physical_page_size, fil_in);
+      if (bytes != physical_page_size)
+      {
+        fprintf(stderr, "Error: Was not able to read the rest of the "
+		"page of " ULINTPF " bytes. Bytes read was " ULINTPF "\n",
+		physical_page_size - UNIV_ZIP_SIZE_MIN, bytes);
+        return false;
+      }
+    }
+    tablespace_flags= flags;
+  }
+
+  if (physical_page_size < UNIV_ZIP_SIZE_MIN ||
+      physical_page_size > UNIV_PAGE_SIZE_MAX)
+  {
+    fprintf(stderr, "Error: Invalid page size " ULINTPF
+            " encountered\n", physical_page_size);
+    return false;
+  }
+  return true;
+}
+
 int main(
 	int	argc,
 	char	**argv)
@@ -1569,51 +1670,13 @@ int main(
 			}
 		}
 
-		/* Read the minimum page size. */
-		bytes = fread(buf, 1, UNIV_ZIP_SIZE_MIN, fil_in);
-		partial_page_read = true;
-
-		if (bytes != UNIV_ZIP_SIZE_MIN) {
-			fprintf(stderr, "Error: Was not able to read the "
-				"minimum page size ");
-			fprintf(stderr, "of %d bytes.  Bytes read was " ULINTPF "\n",
-				UNIV_ZIP_SIZE_MIN, bytes);
-
+		/* Read and validate page 0 */
+		if (!read_and_validate_page0(fil_in, buf)) {
 			exit_status = 1;
 			goto my_exit;
 		}
 
-		/* enable variable is_system_tablespace when space_id of given
-		file is zero. Use to skip the checksum verification and rewrite
-		for doublewrite pages. */
-		cur_space = mach_read_from_4(buf + FIL_PAGE_SPACE_ID);
-		cur_page_num = mach_read_from_4(buf + FIL_PAGE_OFFSET);
-
-		/* Determine page size, zip_size and page compression
-		from fsp_flags and encryption metadata from page 0 */
-		init_page_size(buf);
-
-		ulint flags = mach_read_from_4(FSP_HEADER_OFFSET + FSP_SPACE_FLAGS + buf);
-
-		if (physical_page_size == UNIV_ZIP_SIZE_MIN) {
-			partial_page_read = false;
-		} else {
-			/* Read rest of the page 0 to determine crypt_data */
-			bytes = read_file(buf, partial_page_read, physical_page_size, fil_in);
-			if (bytes != physical_page_size) {
-				fprintf(stderr, "Error: Was not able to read the "
-					"rest of the page ");
-				fprintf(stderr, "of " ULINTPF " bytes.  Bytes read was " ULINTPF "\n",
-					physical_page_size - UNIV_ZIP_SIZE_MIN, bytes);
-
-				exit_status = 1;
-				goto my_exit;
-			}
-			partial_page_read = false;
-		}
-
-
-		/* Now that we have full page 0 in buffer, check encryption */
+		/* Check if tablespace is encrypted */
 		bool is_encrypted = check_encryption(filename, buf);
 
 		/* Verify page 0 contents. Note that we can't allow
@@ -1624,7 +1687,8 @@ int main(
 			allow_mismatches = 0;
 
 			exit_status = verify_checksum(buf, is_encrypted,
-						      &mismatch_count, flags);
+						      &mismatch_count,
+						      tablespace_flags);
 
 			if (exit_status) {
 				fprintf(stderr, "Error: Page 0 checksum mismatch, can't continue. \n");
@@ -1635,7 +1699,8 @@ int main(
 
 		if ((exit_status = rewrite_checksum(
 					filename, fil_in, buf,
-					&pos, is_encrypted, flags))) {
+					&pos, is_encrypted,
+					tablespace_flags))) {
 			goto my_exit;
 		}
 
@@ -1831,7 +1896,7 @@ int main(
 first_non_zero:
 			if (is_system_tablespace) {
 				/* enable when page is double write buffer.*/
-				skip_page = is_page_doublewritebuffer(buf);
+				skip_page = is_page_doublewritebuffer();
 			} else {
 				skip_page = false;
 			}
@@ -1852,13 +1917,16 @@ int main(
 			    && !is_page_free(xdes, physical_page_size, cur_page_num)
 			    && (exit_status = verify_checksum(
 						buf, is_encrypted,
-						&mismatch_count, flags))) {
+						&mismatch_count,
+						tablespace_flags))) {
 				goto my_exit;
 			}
 
-			if ((exit_status = rewrite_checksum(
-						filename, fil_in, buf,
-						&pos, is_encrypted, flags))) {
+			if (!is_page_doublewritebuffer() &&
+			    (exit_status = rewrite_checksum(
+				    filename, fil_in, buf,
+				    &pos, is_encrypted,
+				    tablespace_flags))) {
 				goto my_exit;
 			}
 

mysql-test/suite/innodb/r/innochecksum_flags_and_skip.result

@@ -0,0 +1,28 @@
+CREATE TABLE t1(a INT PRIMARY KEY) ENGINE=InnoDB STATS_PERSISTENT=0;
+SET GLOBAL innodb_log_checkpoint_now=ON;
+FLUSH TABLE t1 FOR EXPORT;
+UNLOCK TABLES;
+INSERT INTO t1 SET a=1;
+FLUSH TABLE t1 FOR EXPORT;
+UNLOCK TABLES;
+# Skip InnoDB Doublewrite Buffer
+Should not exist when summary excludes dblwr pages
+Should exist when summary includes dblwr pages
+# restart
+CREATE TABLE ibd_1(f1 INT PRIMARY KEY)ENGINE=InnoDB;
+INSERT INTO ibd_1 VALUES(1), (2), (3), (4);
+# Pass wrong tablespace flag for ibdata2
+FOUND 1 /Error: Page 0 checksum mismatch/ in result.log
+# Pass wrong tablespace flag for ibdata1
+FOUND 1 /Error: Mismatch between provided tablespace flags/ in result.log
+# Pass invalid tablespace flag for ibdata1
+FOUND 1 /Error: Provided --tablespace-flags is not valid/ in result.log
+# innochecksum should be succesfull
+NOT FOUND /Fail/ in result.log
+# Create a tablespace with page number > 2^31
+# Test innochecksum with the modified ibdata3
+FOUND 1 /Error: First page of the tablespace file should be 0, but encountered page number 2147483649/ in result.log
+# Test innochecksum with the modified ibdata3 with tablespace flags
+FOUND 1 /Exceeded the maximum allowed checksum mismatch/ in result.log
+# restart
+DROP TABLE t1, ibd_1;

mysql-test/suite/innodb/t/innochecksum_flags_and_skip.opt

@@ -0,0 +1,2 @@
+--innodb_data_file_path=ibdata1:3M;ibdata2:1M:autoextend
+--innodb_sys_tablespaces