v5-0002-Implementation-of-backup-validator.patch

From 7c8b2f1ad635bc669e95b7a2000d4836468f74b5 Mon Sep 17 00:00:00 2001
From: Suraj Kharage <suraj.kharage@enterprisedb.com>
Date: Fri, 20 Dec 2019 16:19:44 +0530
Subject: [PATCH v5 2/2] Implementation of backup validator

Patch by Suraj Kharage, inputs from Robert Haas, review from Jeevan Chalke,
and Robert Haas.
---
 doc/src/sgml/ref/pg_basebackup.sgml   |  12 +
 src/bin/pg_basebackup/pg_basebackup.c | 461 ++++++++++++++++++++++++++++++++++
 src/common/encode.c                   |  18 ++
 src/include/common/encode.h           |   1 +
 src/tools/pgindent/typedefs.list      |   2 +
 5 files changed, 494 insertions(+)

diff --git a/doc/src/sgml/ref/pg_basebackup.sgml b/doc/src/sgml/ref/pg_basebackup.sgml
index af7c731..043ee39 100644
--- a/doc/src/sgml/ref/pg_basebackup.sgml
+++ b/doc/src/sgml/ref/pg_basebackup.sgml
@@ -548,6 +548,18 @@ PostgreSQL documentation
       </listitem>
      </varlistentry>
 
+     <varlistentry>
+      <term><option>--verify-backup </option></term>
+      <listitem>
+       <para>
+        Validate the given backup directory and detect the modification if any
+        without restarting the server. For plain backup, provide the backup
+        directory path with <option>--pgdata</option> option. Tar format
+        backups can be verified after untarring.
+       </para>
+      </listitem>
+     </varlistentry>
+
     </variablelist>
    </para>
 
diff --git a/src/bin/pg_basebackup/pg_basebackup.c b/src/bin/pg_basebackup/pg_basebackup.c
index 7d5ed0d..aa82c46 100644
--- a/src/bin/pg_basebackup/pg_basebackup.c
+++ b/src/bin/pg_basebackup/pg_basebackup.c
@@ -27,9 +27,12 @@
 #endif
 
 #include "access/xlog_internal.h"
+#include "common/checksum_utils.h"
+#include "common/encode.h"
 #include "common/file_perm.h"
 #include "common/file_utils.h"
 #include "common/logging.h"
+#include "common/sha2.h"
 #include "common/string.h"
 #include "fe_utils/recovery_gen.h"
 #include "fe_utils/string_utils.h"
@@ -97,6 +100,29 @@ typedef struct WriteManifestState
 typedef void (*WriteDataCallback) (size_t nbytes, char *buf,
 								   void *callback_data);
 
+typedef struct DataDirectoryFileInfo
+{
+	char	   *filename;
+	int			filesize;
+	char	   *checksum;
+	bool		matched;
+	uint32		status;			/* hash status */
+} DataDirectoryFileInfo;
+
+struct manifesthash_hash *hashtab;
+
+#define SH_PREFIX manifesthash
+#define SH_ELEMENT_TYPE DataDirectoryFileInfo
+#define SH_KEY_TYPE char*
+#define SH_KEY filename
+#define SH_HASH_KEY(tb, key) string_hash_sdbm(key)
+#define SH_EQUAL(tb, a, b) (strcmp(a, b) == 0)
+#define SH_SCOPE static inline
+#define SH_RAW_ALLOCATOR pg_malloc
+#define SH_DECLARE
+#define SH_DEFINE
+#include "lib/simplehash.h"
+
 /*
  * pg_xlog has been renamed to pg_wal in version 10.  This version number
  * should be compared with PQserverVersion().
@@ -142,6 +168,7 @@ static bool create_slot = false;
 static bool no_slot = false;
 static bool verify_checksums = true;
 static char *manifest_checksums = NULL;
+static ChecksumAlgorithm checksum_type = MC_NONE;
 
 static bool success = false;
 static bool made_new_pgdata = false;
@@ -201,6 +228,15 @@ static bool reached_end_position(XLogRecPtr segendpos, uint32 timeline,
 static const char *get_tablespace_mapping(const char *dir);
 static void tablespace_list_append(const char *arg);
 
+static void VerifyBackup(void);
+static manifesthash_hash *create_manifest_hash(char *manifest_path);
+static void scan_data_directory(char *basedir, const char *subdirpath,
+								manifesthash_hash *hashtab);
+static void verify_file(struct dirent *de, char fn[MAXPGPATH],
+						struct stat st, char relative_path[MAXPGPATH],
+						manifesthash_hash *hashtab);
+static char *nextLine(char *buf);
+static char *nextWord(char *line);
 
 static void
 cleanup_directories_atexit(void)
@@ -401,6 +437,7 @@ usage(void)
 			 "                         do not verify checksums\n"));
 	printf(_("      --manifest-checksums=SHA256|CRC32C|NONE\n"
 			 "                         calculate checksums for manifest files using provided algorithm\n"));
+	printf(_("      --verify-backup    validate the backup\n"));
 	printf(_("  -?, --help             show this help, then exit\n"));
 	printf(_("\nConnection options:\n"));
 	printf(_("  -d, --dbname=CONNSTR   connection string\n"));
@@ -2167,11 +2204,13 @@ main(int argc, char **argv)
 		{"no-slot", no_argument, NULL, 2},
 		{"no-verify-checksums", no_argument, NULL, 3},
 		{"manifest-checksums", required_argument, NULL, 'm'},
+		{"verify-backup", no_argument, NULL, 4},
 		{NULL, 0, NULL, 0}
 	};
 	int			c;
 
 	int			option_index;
+	bool		verify_backup = false;
 
 	pg_logging_init(argv[0]);
 	progname = get_progname(argv[0]);
@@ -2338,6 +2377,9 @@ main(int argc, char **argv)
 			case 'm':
 				manifest_checksums = pg_strdup(optarg);
 				break;
+			case 4:
+				verify_backup = true;
+				break;
 			default:
 
 				/*
@@ -2460,6 +2502,12 @@ main(int argc, char **argv)
 	}
 #endif
 
+	if (verify_backup)
+	{
+		VerifyBackup();
+		return 0;
+	}
+
 	/* connection in replication mode to server */
 	conn = GetConnection();
 	if (!conn)
@@ -2524,3 +2572,416 @@ main(int argc, char **argv)
 	success = true;
 	return 0;
 }
+
+static void
+VerifyBackup(void)
+{
+	char		manifest_path[MAXPGPATH];
+	manifesthash_hash *hashtab;
+	manifesthash_iterator i;
+	DataDirectoryFileInfo *entry;
+
+	snprintf(manifest_path, sizeof(manifest_path), "%s/%s", basedir,
+			 "backup_manifest");
+
+	/* create hash table */
+	hashtab = create_manifest_hash(manifest_path);
+
+	scan_data_directory(basedir, NULL, hashtab);
+
+	manifesthash_start_iterate(hashtab, &i);
+	while ((entry = manifesthash_iterate(hashtab, &i)) != NULL)
+	{
+		if (!entry->matched)
+			pg_log_info("file \"%s\" is present in manifest but missing from the backup",
+						entry->filename);
+	}
+}
+
+/*
+ * Given a file path, read that file and generate the hash table for same.
+ * Also generate the checksum for the records that are read from file and
+ * compare that with checksum written in backup_manifest file. If both
+ * checksums are identical then proceed, otherwise throw an error and abort.
+ */
+static manifesthash_hash *
+create_manifest_hash(char *manifest_path)
+{
+	manifesthash_hash *hashtab;
+	DataDirectoryFileInfo *entry;
+	char	   *buf;
+	int			fd;
+	struct stat stat;
+
+	fd = open(manifest_path, O_RDONLY, 0);
+
+	if (fstat(fd, &stat))
+	{
+		pg_log_error("could not stat file \"%s\": %m", manifest_path);
+		close(fd);
+		exit(1);
+	}
+
+	buf = pg_malloc(stat.st_size);
+
+	if (buf == NULL)
+	{
+		pg_log_error("out of memory - Could not allocate enough memory to read file \"%s\".",
+					 manifest_path);
+		close(fd);
+		exit(1);
+	}
+
+	hashtab = manifesthash_create(1024, NULL);
+
+	if (read(fd, buf, stat.st_size) != stat.st_size)
+	{
+		pg_log_error("could not read file \"%s\": %m", manifest_path);
+
+		close(fd);
+		exit(1);
+	}
+
+	/* Read the buffer line by line and parse each line into fields */
+	if (*buf != '\0')
+	{
+		char	   *header_line;
+		int			header_length;
+		bool		checksum_initiated = false;
+		ChecksumCtx cCtx;
+		int			checksum_label_length = 0;
+
+		/*
+		 * Read the header from file, here header_line is pointing to start of
+		 * file. Advanced the buffer to next line and then buf - header_line
+		 * will give us the header length.
+		 */
+		header_line = buf;
+
+		buf = nextLine(buf);
+
+		header_length = buf - header_line;
+
+		/*
+		 * Once we read the header, then read the buffer line by line and check
+		 * whether it is a File record or Manifest-Checksum entry and parse
+		 * accordingly.
+		 */
+		while (*buf != '\0')
+		{
+			int			length;
+			char	   *line;
+
+			line = buf;
+			/* read the next line and calculate the length for the line */
+			buf = nextLine(buf);
+			length = buf - line;
+
+			/*
+			 * If it is a Manifest-Checksum entry, then finalize the checksum
+			 * and compare it with the manifest checksum parsed from the file
+			 */
+			if (strncmp(line, "Manifest-Checksum", 17) == 0)
+			{
+				char		checksumbuf[256];
+				int			checksumbuflen;
+				char		encoded_checksum[256];
+
+				checksumbuflen = finalize_checksum(&cCtx, checksum_type,
+												   (char *) checksumbuf);
+				checksumbuflen = hex_encode(checksumbuf, checksumbuflen,
+											encoded_checksum);
+				encoded_checksum[checksumbuflen] = '\0';
+
+				line[length - 1] = '\0';
+
+				line = nextWord(line);
+
+				line = line + checksum_label_length;
+
+				if (strcmp(encoded_checksum, line) != 0)
+				{
+					pg_log_error("backup manifest checksum difference. Aborting");
+					exit(1);
+				}
+			}
+
+			/*
+			 * If it is a File record, then parse it into fields. With this we
+			 * will get the pointers for filename, checksum and size in long
+			 */
+			else if (strncmp(line, "File", 4) == 0)
+			{
+				char	   *filename;
+				char	   *checksum;
+				char	   *size;
+				long		filesize;
+				bool		found;
+				char	   *record; /* pointer for a single record */
+				long		filelength,
+							sizelength;
+
+				record = line;
+
+				/* skip the "File" field */
+				line = nextWord(line);
+
+				filename = line;
+				line = nextWord(line);
+				filelength = line - filename;
+
+				size = line;
+				line = nextWord(line);
+				sizelength = line - size;
+
+				/* skip mtime field */
+				line = nextWord(line);
+
+				checksum = line;
+
+				if (!checksum_initiated)
+				{
+					if (strncmp(checksum, "SHA256", 6) == 0)
+					{
+						checksum_type = MC_SHA256;
+						checksum_label_length = strlen("SHA256:");
+					}
+					else if (strncmp(checksum, "CRC32C", 6) == 0)
+					{
+						checksum_type = MC_CRC32C;
+						checksum_label_length = strlen("CRC32C:");
+					}
+					else if (strncmp(checksum, "-", 1) == 0)
+						checksum_type = MC_NONE;
+					else
+					{
+						pg_log_error("unknown checksum method");
+						exit(1);
+					}
+
+					/*
+					 * we don't have checksum type in the header, so need to
+					 * read through the first file entry to find the checksum
+					 * type for the manifest file and initialize the checksum
+					 * for the manifest file itself.
+					 */
+					if (checksum_type != MC_NONE)
+					{
+						/* initialize the checksum for the first time */
+						initialize_checksum(&cCtx, checksum_type);
+
+						/* feed the header to the checksum machinery */
+						update_checksum(&cCtx, checksum_type, header_line,
+										header_length);
+					}
+
+					checksum_initiated = true;
+				}
+
+				/* feed line to the checksum machinery. */
+				if (checksum_type != MC_NONE)
+					update_checksum(&cCtx, checksum_type, record, length);
+
+				/* terminate the each field */
+				record[length - 1] = '\0';
+				filename[filelength - 1] = '\0';
+				size[sizelength - 1] = '\0';
+
+				/* convert from string to long */
+				filesize = atol(size);
+
+				/*
+				 * Increase the checksum by its label length so that we can
+				 * get the exact checksum
+				 */
+				checksum = checksum + checksum_label_length;
+
+				/* insert the hash record */
+				entry = manifesthash_insert(hashtab, filename, &found);
+				entry->filesize = filesize;
+				entry->checksum = checksum;
+			}
+			else
+			{
+				pg_log_error("invalid record found in \"%s\"", manifest_path);
+				exit(1);
+			}
+		}
+	}
+	close(fd);
+	return hashtab;
+}
+
+/*
+ * Scan the data directory and check whether each file entry present in hash
+ * table with the correct details, i.e. filesize and checksum.
+ */
+static void
+scan_data_directory(char *basedir, const char *subdirpath,
+					manifesthash_hash *hashtab)
+{
+	char		path[MAXPGPATH];
+	char		relative_path[MAXPGPATH] = "";
+	DIR		   *dir;
+	struct dirent *de;
+
+	if (subdirpath)
+	{
+		snprintf(path, sizeof(path), "%s/%s", basedir,
+				 subdirpath);
+		snprintf(relative_path, sizeof(relative_path), "%s/", subdirpath);
+
+	}
+	else
+		snprintf(path, sizeof(path), "%s", basedir);
+
+	dir = opendir(path);
+	if (!dir)
+	{
+		pg_log_error("could not open directory \"%s\": %m", path);
+		exit(1);
+	}
+
+	while ((de = readdir(dir)) != NULL)
+	{
+		char		fn[MAXPGPATH];
+		struct stat st;
+
+		if (strcmp(de->d_name, ".") == 0 || strcmp(de->d_name, "..") == 0 ||
+			(strcmp(de->d_name, "pg_wal") == 0 && !subdirpath))
+			continue;
+
+		snprintf(fn, sizeof(fn), "%s/%s", path, de->d_name);
+		if (stat(fn, &st) < 0)
+		{
+			pg_log_error("could not stat file \"%s\": %m", fn);
+			exit(1);
+		}
+		if (S_ISREG(st.st_mode))
+		{
+			verify_file(de, fn, st, relative_path, hashtab);
+		}
+		else if (S_ISDIR(st.st_mode))
+		{
+			char		newsubdirpath[MAXPGPATH];
+
+			if (subdirpath)
+				snprintf(newsubdirpath, MAXPGPATH, "%s/%s", subdirpath,
+						 de->d_name);
+			else
+				snprintf(newsubdirpath, MAXPGPATH, "%s", de->d_name);
+
+			scan_data_directory(basedir, newsubdirpath, hashtab);
+		}
+	}
+	closedir(dir);
+}
+
+/*
+ * Given the file and its details, check whether it is present in hash table
+ * and if yes, then compare its details with hash table entry.
+ */
+static void
+verify_file(struct dirent *de, char fn[MAXPGPATH], struct stat st,
+			char relative_path[MAXPGPATH], manifesthash_hash *hashtab)
+{
+	char		filename[MAXPGPATH];
+	DataDirectoryFileInfo *record;
+
+	/* Skip backup manifest file. */
+	if (strcmp(de->d_name, "backup_manifest") == 0)
+		return;
+
+	snprintf(filename, MAXPGPATH, "%s%s", relative_path, de->d_name);
+
+	/*
+	 * Compare the hash and if record found then we match the file size and
+	 * checksum (if enabled). Modified time cannot be compared with the file
+	 * in the backup directory and its entry in the manifest as manifest entry
+	 * gives mtime from server file whereas the same file in the backup will
+	 * have different mtime.
+	 */
+	record = manifesthash_lookup(hashtab, filename);;
+	if (record)
+	{
+		record->matched = true;
+		if (record->filesize != st.st_size)
+			pg_log_info("file \"%s\" has size %d in manifest but size %lu in backup",
+						filename, record->filesize, st.st_size);
+
+		/*
+		 * Read the file and generate the checksum based on checksum method
+		 * and compare that with the checksum present in hash entry.
+		 */
+		if (checksum_type != MC_NONE)
+		{
+			FILE	   *fp;
+			char		buf[1048576];  /* 1MB chunk */
+			pgoff_t		len = 0;
+			off_t		cnt;
+			char 		checksumbuf[256];
+			char 		encode_checksumbuf[256];
+			int	   		checksumbuflen;
+			ChecksumCtx	cCtx;
+
+			initialize_checksum(&cCtx, checksum_type);
+
+			fp = fopen(fn, "r");
+			if (!fp)
+			{
+				pg_log_error("could not open file \"%s\": %m", de->d_name);
+				exit(1);
+			}
+
+			/* Read file in chunks [1 MB each chunk] */
+			while ((cnt = fread(buf, 1, Min(sizeof(buf), st.st_size - len), fp)) > 0)
+			{
+				update_checksum(&cCtx, checksum_type, buf, cnt);
+				len += cnt;
+			}
+
+			checksumbuflen = finalize_checksum(&cCtx, checksum_type,
+											   checksumbuf);
+
+			/* Convert checksum to hexadecimal. */
+			checksumbuflen = hex_encode(checksumbuf, checksumbuflen,
+										encode_checksumbuf);
+			encode_checksumbuf[checksumbuflen] = '\0';
+
+			fclose(fp);
+
+			if (strcmp(record->checksum, encode_checksumbuf) != 0)
+				pg_log_info("file \"%s\" has checksum %s in manifest but checksum %s in backup",
+							filename, record->checksum, encode_checksumbuf);
+		}
+	}
+	else
+		pg_log_info("file \"%s\" is present in backup but not in manifest",
+					filename);
+}
+
+/*
+ * Find out the next new line character from the provided string and return
+ * char pointer pointing to next character after that.
+ */
+static char *
+nextLine(char *buf)
+{
+	while (*buf != '\0' && *buf != '\n')
+		buf = buf + 1;
+
+	return buf + 1;
+}
+
+/*
+ * Find out the next tab or new line character from the provided string and
+ * return char pointer pointing to next character after that.
+ */
+static char *
+nextWord(char *line)
+{
+	while (*line != '\0' && *line != '\t' && *line != '\n')
+		line = line + 1;
+
+	return line + 1;
+}
diff --git a/src/common/encode.c b/src/common/encode.c
index a450c53..14f2ec2 100644
--- a/src/common/encode.c
+++ b/src/common/encode.c
@@ -36,3 +36,21 @@ hex_encode(const char *src, unsigned len, char *dst)
 	}
 	return len * 2;
 }
+
+/*
+ * Simple string hash function from http://www.cse.yorku.ca/~oz/hash.html
+ *
+ * The backend uses a more sophisticated function for hashing strings,
+ * but we don't really need that complexity here.
+ */
+uint32
+string_hash_sdbm(const char *key)
+{
+	uint32		hash = 0;
+	int			c;
+
+	while ((c = *key++))
+		hash = c + (hash << 6) + (hash << 16) - hash;
+
+	return hash;
+}
diff --git a/src/include/common/encode.h b/src/include/common/encode.h
index 63328bc..44a062f 100644
--- a/src/include/common/encode.h
+++ b/src/include/common/encode.h
@@ -16,5 +16,6 @@
 static const char hextbl[] = "0123456789abcdef";
 
 extern unsigned hex_encode(const char *src, unsigned len, char *dst);
+extern uint32 string_hash_sdbm(const char *key);
 
 #endif							/* COMMON_ENCODE_H */
diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list
index 87556f6..14e475e 100644
--- a/src/tools/pgindent/typedefs.list
+++ b/src/tools/pgindent/typedefs.list
@@ -487,6 +487,7 @@ DR_sqlfunction
 DR_transientrel
 DSA
 DWORD
+DataDirectoryFileInfo
 DataDumperPtr
 DataPageDeleteStack
 DateADT
@@ -1353,6 +1354,7 @@ MultiXactOffset
 MultiXactStateData
 MultiXactStatus
 MyData
+manifesthash_hash
 manifestinfo
 NDBOX
 NODE
-- 
1.8.3.1