Change blocksize to 512

1024 bytes was originally chosen as fitting nicely inside an ethernet
frame. That didn't really make much sense, but the issue was masked by
the kernel actually ignoring block sizes due to a bug.

Now that the kernel bug is fixed, the result is that partitions are
parsed incorrectly, resulting in confusing behaviour.

Fix by defaulting to 512. For the rare case where 1024 bytes still makes
sense, the -b parameter still exists.

nbd-client.c

@@ -910,7 +910,7 @@ void usage(char* errmsg, ...) {
 #if HAVE_GNUTLS && !defined(NOTLS)
 	fprintf(stderr, "All commands that connect to a host also take:\n\t[-F|-certfile certfile] [-K|-keyfile keyfile]\n\t[-A|-cacertfile cacertfile] [-H|-tlshostname hostname] [-x|-enable-tls]\n");
 #endif
-	fprintf(stderr, "Default value for blocksize is 1024 (recommended for ethernet)\n");
+	fprintf(stderr, "Default value for blocksize is 512\n");
 	fprintf(stderr, "Allowed values for blocksize are 512,1024,2048,4096\n"); /* will be checked in kernel :) */
 	fprintf(stderr, "Note, that kernel 2.4.2 and older ones do not work correctly with\n");
 	fprintf(stderr, "blocksizes other than 1024 without patches\n");
@@ -941,7 +941,7 @@ static const char *short_opts = "-A:b:c:C:d:gH:hK:lnN:pSst:uVx";
 int main(int argc, char *argv[]) {
 	char* port=NBD_DEFAULT_PORT;
 	int sock, nbd;
-	int blocksize=1024;
+	int blocksize=512;
 	char *hostname=NULL;
 	char *nbddev=NULL;
 	int swap=0;