Path: utzoo!utgpu!jarvis.csri.toronto.edu!rutgers!mcnc!rti!trt
From: trt@rti.UUCP (Thomas Truscott)
Newsgroups: comp.sys.pyramid
Subject: alignment restrictions when writing to raw magtape
Message-ID: <3106@rti.UUCP>
Date: 4 Aug 89 22:36:22 GMT
Distribution: comp
Organization: Research Triangle Institute, RTP, NC
Lines: 33

The following program, run on on OSx4.4 on a Pyramid
(I forget the exact info but it has 2 10MIP cpus,
and a fast tape unit),
shows that if the write buffer does not begin on
a 4-byte boundary, the write syscall fails with EFAULT.

Why is this? Note that one can write
any number of bytes to the drive, thus the buffer can *end*
on an arbitrary boundary.

main()
{
	char b[9000];
	int i, fd;

	if ((fd = open("/dev/rmt0", 2)) < 0)
		perror("open");
	for (i = 0; i < 8; i++)
		if (write(fd, b+i, 43) != 43)
			perror("write");
		else
			printf("%d\n", i);
	exit(0);
}

I would hope that this will be considered a (admittedly minor)
bug and hope that a kernel fix is not too difficult.
It it is not considered a bug, could someone point to
a specific (e.g. "/dev/odk0, optical paper tape on SolOS 3.1")
instance in which a similar limitation exists?
(Am am aware that some disks must be given write lengths
that are multiples of their sector size, that is not at issue here).
	Tom Truscott