Fix tar files emitted by pg_basebackup to be POSIX conformant.

tglsfdc · tglsfdc · commit dfa6eda5e44f · 2012-09-28T15:35:51.000-04:00
Back-patch portions of commit 05b555d. There doesn't seem to be any reason not to fix pg_basebackup fully, but we can't change pg_dump's "magic" string without breaking older versions of pg_restore. Instead, just patch pg_restore to accept either version of the magic string, in hopes of avoiding compatibility problems when 9.3 comes out. I also fixed pg_dump to write the correct 2-block EOF marker, since that won't create a compatibility problem with pg_restore and it could help with some versions of tar. Brian Weaver and Tom Lane
diff --git a/doc/src/sgml/protocol.sgml b/doc/src/sgml/protocol.sgml
@@ -1726,9 +1726,11 @@ The commands accepted in walsender mode are:
       After the second regular result set, one or more CopyResponse results
       will be sent, one for PGDATA and one for each additional tablespace other
       than <literal>pg_default</> and <literal>pg_global</>. The data in
-      the CopyResponse results will be a tar format (using ustar00
-      extensions) dump of the tablespace contents. After the tar data is
-      complete, a final ordinary result set will be sent.
+      the CopyResponse results will be a tar format (following the
+      <quote>ustar interchange format</> specified in the POSIX 1003.1-2008
+      standard) dump of the tablespace contents, except that the two trailing
+      blocks of zeroes specified in the standard are omitted.
+      After the tar data is complete, a final ordinary result set will be sent.
      </para>
 
      <para>
diff --git a/src/backend/replication/basebackup.c b/src/backend/replication/basebackup.c
@@ -554,7 +554,7 @@ sendFileWithContent(const char *filename, const char *content)
 
 /*
  * Include all files from the given directory in the output tar stream. If
- * 'sizeonly' is true, we just calculate a total length and return ig, without
+ * 'sizeonly' is true, we just calculate a total length and return it, without
  * actually sending anything.
  */
 static int64
@@ -744,11 +744,16 @@ _tarChecksum(char *header)
 	int			i,
 				sum;
 
-	sum = 0;
+	/*
+	 * Per POSIX, the checksum is the simple sum of all bytes in the header,
+	 * treating the bytes as unsigned, and treating the checksum field (at
+	 * offset 148) as though it contained 8 spaces.
+	 */
+	sum = 8 * ' ';				/* presumed value for checksum field */
 	for (i = 0; i < 512; i++)
 		if (i < 148 || i >= 156)
 			sum += 0xFF & header[i];
-	return sum + 256;			/* Assume 8 blanks in checksum field */
+	return sum;
 }
 
 /* Given the member, write the TAR header & send the file */
@@ -827,9 +832,13 @@ _tarWriteHeader(const char *filename, const char *linktarget,
 				struct stat * statbuf)
 {
 	char		h[512];
-	int			lastSum = 0;
-	int			sum;
 
+	/*
+	 * Note: most of the fields in a tar header are not supposed to be
+	 * null-terminated.  We use sprintf, which will write a null after the
+	 * required bytes; that null goes into the first byte of the next field.
+	 * This is okay as long as we fill the fields in order.
+	 */
 	memset(h, 0, sizeof(h));
 
 	/* Name 100 */
@@ -841,8 +850,11 @@ _tarWriteHeader(const char *filename, const char *linktarget,
 		 * indicated in the tar format by adding a slash at the end of the
 		 * name, the same as for regular directories.
 		 */
-		h[strlen(filename)] = '/';
-		h[strlen(filename) + 1] = '\0';
+		int			flen = strlen(filename);
+
+		flen = Min(flen, 99);
+		h[flen] = '/';
+		h[flen + 1] = '\0';
 	}
 
 	/* Mode 8 */
@@ -852,9 +864,9 @@ _tarWriteHeader(const char *filename, const char *linktarget,
 	sprintf(&h[108], "%07o ", statbuf->st_uid);
 
 	/* Group 8 */
-	sprintf(&h[117], "%07o ", statbuf->st_gid);
+	sprintf(&h[116], "%07o ", statbuf->st_gid);
 
-	/* File size 12 - 11 digits, 1 space, no NUL */
+	/* File size 12 - 11 digits, 1 space; use print_val for 64 bit support */
 	if (linktarget != NULL || S_ISDIR(statbuf->st_mode))
 		/* Symbolic link or directory has size zero */
 		print_val(&h[124], 0, 8, 11);
@@ -865,13 +877,13 @@ _tarWriteHeader(const char *filename, const char *linktarget,
 	/* Mod Time 12 */
 	sprintf(&h[136], "%011o ", (int) statbuf->st_mtime);
 
-	/* Checksum 8 */
-	sprintf(&h[148], "%06o ", lastSum);
+	/* Checksum 8 cannot be calculated until we've filled all other fields */
 
 	if (linktarget != NULL)
 	{
 		/* Type - Symbolic link */
 		sprintf(&h[156], "2");
+		/* Link Name 100 */
 		sprintf(&h[157], "%.99s", linktarget);
 	}
 	else if (S_ISDIR(statbuf->st_mode))
@@ -881,10 +893,11 @@ _tarWriteHeader(const char *filename, const char *linktarget,
 		/* Type - regular file */
 		sprintf(&h[156], "0");
 
-	/* Link tag 100 (NULL) */
+	/* Magic 6 */
+	sprintf(&h[257], "ustar");
 
-	/* Magic 6 + Version 2 */
-	sprintf(&h[257], "ustar00");
+	/* Version 2 */
+	sprintf(&h[263], "00");
 
 	/* User 32 */
 	/* XXX: Do we need to care about setting correct username? */
@@ -894,17 +907,21 @@ _tarWriteHeader(const char *filename, const char *linktarget,
 	/* XXX: Do we need to care about setting correct group name? */
 	sprintf(&h[297], "%.31s", "postgres");
 
-	/* Maj Dev 8 */
-	sprintf(&h[329], "%6o ", 0);
+	/* Major Dev 8 */
+	sprintf(&h[329], "%07o ", 0);
 
-	/* Min Dev 8 */
-	sprintf(&h[337], "%6o ", 0);
+	/* Minor Dev 8 */
+	sprintf(&h[337], "%07o ", 0);
 
-	while ((sum = _tarChecksum(h)) != lastSum)
-	{
-		sprintf(&h[148], "%06o ", sum);
-		lastSum = sum;
-	}
+	/* Prefix 155 - not used, leave as nulls */
+
+	/*
+	 * We mustn't overwrite the next field while inserting the checksum.
+	 * Fortunately, the checksum can't exceed 6 octal digits, so we just write
+	 * 6 digits, a space, and a null, which is legal per POSIX.
+	 */
+	sprintf(&h[148], "%06o ", _tarChecksum(h));
 
+	/* Now send the completed header. */
 	pq_putmessage('d', h, 512);
 }
diff --git a/src/bin/pg_dump/pg_backup_tar.c b/src/bin/pg_dump/pg_backup_tar.c
@@ -876,8 +876,10 @@ _CloseArchive(ArchiveHandle *AH)
 
 		tarClose(AH, th);
 
-		/* Add a block of NULLs since it's de-rigeur. */
-		for (i = 0; i < 512; i++)
+		/*
+		 * EOF marker for tar files is two blocks of NULLs.
+		 */
+		for (i = 0; i < 512 * 2; i++)
 		{
 			if (fputc(0, ctx->tarFH) == EOF)
 				die_horribly(AH, modulename,
@@ -1028,11 +1030,16 @@ _tarChecksum(char *header)
 	int			i,
 				sum;
 
-	sum = 0;
+	/*
+	 * Per POSIX, the checksum is the simple sum of all bytes in the header,
+	 * treating the bytes as unsigned, and treating the checksum field (at
+	 * offset 148) as though it contained 8 spaces.
+	 */
+	sum = 8 * ' ';				/* presumed value for checksum field */
 	for (i = 0; i < 512; i++)
 		if (i < 148 || i >= 156)
 			sum += 0xFF & header[i];
-	return sum + 256;			/* Assume 8 blanks in checksum field */
+	return sum;
 }
 
 bool
@@ -1046,11 +1053,15 @@ isValidTarHeader(char *header)
 	if (sum != chk)
 		return false;
 
-	/* POSIX format */
-	if (strncmp(&header[257], "ustar00", 7) == 0)
+	/* POSIX tar format */
+	if (memcmp(&header[257], "ustar\0", 6) == 0 &&
+		memcmp(&header[263], "00", 2) == 0)
+		return true;
+	/* GNU tar format */
+	if (memcmp(&header[257], "ustar  \0", 8) == 0)
 		return true;
-	/* older format */
-	if (strncmp(&header[257], "ustar  ", 7) == 0)
+	/* not-quite-POSIX format written by pre-9.3 pg_dump */
+	if (memcmp(&header[257], "ustar00\0", 8) == 0)
 		return true;
 
 	return false;
