[comp.unix.bsd] NetBSD, FreeBSD, and OpenBSD FAQ (Part 7 of 10) Posted-By: auto-faq 3.1.1.2 Archive-name: 386bsd-faq/part7 Section 6. (Interaction with MS-DOS) 6.0 Working with DOS and BNR/2 related software. This section is designed to cover some of the more common problems that DOS will have when interacting with BNR/2. There are other sections of the FAQ that deal with indirectly with this . Try looking in sections 0, 1, and 2 to see if something in there (particularly when talking about DOS and *BSD coexisting on a single drive). 6.1 Formatting a floppy Formatting a floppy under NetBSD is possible through the the patch file stored at ftp://cynjut.neonramp.com/pub/fdformat.patch This patch appears to be for NetBSD 1.2 (release). Obviously, use it at your own risk. FreeBSD (and I think OpenBSD) both have floppy formatting built into their systems in fairly recent versions. 6.2 Sharing the Disk with MS-DOS There are a myriad of questions about how to share a disk between *BSD and MS-DOS. They all boils down to one of the following questions: 1) How can I partition my drive for both MS-DOS and *BSD? 2) I can install using the whole disk, but I can't install when I try to share the drive between *BSD and MS-DOS. Why? 3) I can use either MS-DOS or *BSD on my hard drive, but shutdown -todos doesn't seem to work. 6.2.1 How can I partition my drive to support both MS-DOS and *bsd? NOTE: Before attempting to install *bsd on a computer with an active DOS partition, ALWAYS back up your hard drive. No one on the net, no matter how talented, can help you recover a hosed MS-DOS file system. If you lose all of your data, it is YOUR fault. During the install phase, you need to have un-allocated space left on your disk drive. This allows the install program to correctly install the *bsd partition in the partition table and DOS to peacefully co-exist with *bsd. If you do not have any space available on your hard drive, you will not be able to install both. Re-fdisk your hard drive and make sure you have left un-allocated space in the partition table. This WILL wipe out your DOS partition - Permanently. Even though the partition table procedure above may have worked, there are still no guarantees that your system will boot after the install. This problem most often manifests itself as one of the endless reboot problems. You would normally be able to boot DOS from the hard disk, but not *bsd (once that partition is marked as active). Once the partition table has been correctly defined with both DOS and *bsd, there can still be problem. One of the most common is that the disk drive works in some sort of translation mode. This is particularly common with drives that physically have more than 1024 cylinders. DOS cannot access a drive with more than 1024 cylinders. Translation mode will have to be turned off, usually by redefining your hard drive in SETUP as one of the user definable types. This change will normally trash your hard drive, or at least render your DOS partition unreadable. The solution to this problem is to install *bsd at the end of the hard drive. While DOS cannot use cylinders above 1024, *bsd has no such limitations, once it has booted. During the boot-up phase, some of the newer boot blocks will refer to the BIOS for some services. Specifically, the disk is checked for a bad sector map on the last track. Since the BIOS cannot deal with cylinders higher than 1024, your bad sector map will be incorrectly identified as 1023 if the number of cylinders is larger than that. This problem is being worked on, and I hope to change this section with better news later. NOTE: The only people that this problem will effect are those MFM and ESDI users that have drives with more than 1023 tracks. While drives of this type are not the overwhelming majority, neither are they an anomaly. People are working on it. As an example, if your hard disk physically has 8 heads, 16 sectors per track, and 2000 cylinders (128M); you MUST use some sort of disk translation in order to use the entire drive. An obvious geometry for this drive (for DOS) would be 16 heads, 16 sectors, and 1000 cylinders. Unfortunately, *bsd operates using the disk drives native geometry as reported during the probe phase of boot up. This will probably be 8/16/2000, and will NOT agree with your translated disk geometry. This causes an endless reboot cycle. If you change the geometry so that the drive agrees with the disklabel, your DOS partition is toast. The best way to operate in this case would be to (for example) split the disk in half. That leaves 64M for DOS, using a geometry of 8 heads, 16 sectors per track, and the first 1000 cylinders for DOS. The second 1000 cylinders could then safely be used for *bsd. The DOS partition table may even be capable of showing this partition as it actually exists. ACCESSING MS-DOS PARTITIONS FROM NetBSD-i386 First off, it's important to understand BSD disklabels. The disklabel is a description of the Unix partition layout and other disk parameters stored on-disk, usually somewhere in the first couple of sectors. There is a maximum of 8 partitions, labelled "a" thru "h". Typically partition "a" is assigned to the root partition, partition "b" is configured as a swap area, and partition "c" is defined as the whole disk. You can change these, but it's a good idea to stick with this scheme, as many programs assume that's the way things are going to be. If you're whole disk is dedicated to Unix, then that's all you need to know. But if you're sharing your disk with DOS, then there are a few magical things happening. DOS has it's own partitioning scheme. The way NetBSD co-exists with this is to fit all of the Unix partitions into one DOS partition. So partitions a-h all fit inside one DOS partition, which has a partition type of 165 (each MS-DOS partition has a "partition type" associated with it. The BSD partition type is 165). In this setup, partition "c" refers to the entire BSD partition. But in this scheme, partition "d" refers to the ENTIRE disk, MS-DOS partitions and all. So, if you want to access your MS-DOS partition from NetBSD, first you'll have to create a partition that points to the MS-DOS partition. You'll want to run the command: disklabel -e -r /dev/r??0d (fill in with your disk type). You'll get popped into an editor with all the disklabel stuff in it. Go down to the bottom. You should see something like: 6 partitions: # size offset fstype [fsize bsize cpg] a: 30720 409600 4.2BSD 1024 8192 16 # (Cyl... b: 129024 440320 swap # (Cyl... c: 1617920 409600 unused 0 0 # (Cyl... d: 2029568 0 unused 0 0 # (Cyl... e: 61440 569344 4.2BSD 1024 8192 16 # (Cyl... f: 1396736 630784 4.2BSD 1024 8192 16 # (Cyl... (or whatever it appropriate for your disk). Note that partition "a" starts on cylinder 200. That's where my BSD partition starts on my disk. Also note that partition "c" starts at 200 as well and goes to the end of the disk. You'll also note that partition "d" goes from sector 0 all the way to the end of the disk. Add a new line that looks something like: g: 409568 32 MS-DOS # (Cyl. 0*- 199*) (The comment on the end isn't necessary. Only the partition letter, size, offset, and partition type are needed. You can put unused in instead of MS-DOS if you want). *NOTE* Be sure to change the line that says "6 partitions" to the new number of partitions that you have!!! Otherwise you'll get an obscure error. In my case I'd change that line to be "7 partitions". If you aren't sure what your MS-DOS partition size and offsets are, you can use the NetBSD fdisk to find them out. Don't forget that there's a maximum of 8 partitions. Once you do that and you have MSDOSFS configured into your kernel, you can just do something like "mount -t msdos /dev/sd0g /msdos". Or you can put a line like this in your fstab: /dev/sd0g /msdos msdos rw 0 0 If you want to access a DOS-only HD from NetBSD, here are some instructions posted by Charles Hannum a while back. I haven't tried them myself, but they seem like they would work. Assuming you don't have something (like OS-BS) which uses the extra sectors in the boot track, you can do the following: 1) Use the NetBSD `fdisk' or DOS `pfdisk' to create a NetBSD partition in the MBR which spans the entire disk. 2) Save a copy of the MBR: dd if=/dev/rsd0d of=my-mbr bs=1b count=1 3) Use `disklabel' to create a NetBSD label with the DOS partition and whatnot. Answer `y' when it asks you if you want to `overwrite [a] disk with [a] DOS partition table'. 4) Put back the saved copy of the MBR: dd if=my-mbr of=/dev/rsd0d bs=1b count=1 This works for me. Your mileage may vary. Luke Mewburn has provided the following tutorial on using the pfdisk program and making your *bsd/NetBSD partitions peacefully coexist with DOS. While this is kind of a 'cookbook' approach, please keep in mind that this is probably easily transferable to all BNR derived Unices. 6.2.2 I can install using the whole disk, but I can't install when I try to share the drive between *BSD and MS-DOS. Why? This is an extension of the question above. The most common reason for this is, once again, disk translation problems. If the disklabel does not agree with the disk geometry, the install will fail. Other incarnations of this problem are that you can install DOS, then *BSD, and DOS will be hosed, or vice versa. There are more than a couple of people who will blithely suggest that this is a good thing, and you should install *BSD exclusively, job not withstanding. 6.2.4 Is there any hope of ever running MS-DOS applications under any of the free BSD systems? There is currently a project in development to port the Windows program environment to Linux and the *BSD systems. Here is an excerpt from the original message announcing the project: As many of you already know, we are in the process of creating a Windows emulator. This emulator is similar to Sun's Wabi product, but is being developed completely independent of them. Many of you are anxious to hear the latest status of the project. I have created a mailing list for those of you. To join the list, simply send mail to: wine-project-info@amscons.com If your mailing address is not easy to deduce from the mail headers, then place the following line in the body of the message that you send. Reply-To: youraddress@yourmachine where youraddress@yourmachine should be replaced by your actual mailing address. 6.2.5 How do I get Linux executables to run under NetBSD? First, you need to make certain your kernel has LINUX_COMPAT as one of the options for your kernel. Second, you will need the libraries for Linux. You can find the Linux supporting binaries for NetBSD i386 at ftp://ftp.enigma.net/pub/netbsd_i386. There are instructions there to tell you how to get the libraries working correctly. With VERY new versions of NetBSD (NetBSD 1.3A and above) the kernel option for Linux compatibility have changed. All ELF and COFF executables, as well as native Linux executables will be included as part of the same option. 6.3 Accessing the MS-DOS filesystem One of the most common MS-DOS related questions (with the possible exception of 6.2 above) is how to access the DOS disk partitions from *BSD. -------------------------------------------------------------------- How to mount your DOS partition from FreeBSD 1. First, be root. The following won't work as an ordinary user. 2. Second, use 'fdisk' to see where your DOS partition starts. It will be labeled as type DOS. On my system, 'fdisk /dev/sd0d' produces the following: ... (extraneous output, not of interest) ... The data for partition 0 is: sysid 6,(Primary 'big' DOS (> 32MB)) start 32, size 306400 (149 Meg), flag 0 beg: cyl 0/ sector 1/ head 1; end: cyl 149/ sector 32/ head 39 This shows me that my DOS partition starts at sector 32, and is 306400 (512 byte) sectors long. NOTE: If you're trying to mount a DOS `EXTENDED' partition, then you need to add the number of sectors per track to this start address you got from fdisk in subsequent calculations, I.E. in the above example (assuming it was an EXTENDED partition rather than the Primary), you'd use `start 64, size 306400'. [Ed.Note. This example assumes a SCSI disk. For disks with a number of sectors per track which is different than 32, you will probably see the 32s above replaced with your number of sectors per track. IDE, MFM, and ESDI drives are all examples where the number of sectors per track is likely to NOT be 32.] 3. Next, using this information, you craft a new disk entry in your /etc/disktab file that assigns one of your unused "UNIX" partitions to this DOS region. Again, using my system as a default, you see I've created: disk0|DEC 5501:\ :ty=winchester:dt=SCSI:se#512:nt#8:ns#256:nc#1001:rm#3600:\ :pa#956416:oa#307200:ba#8192:fa#1024:ta=4.2BSD:\ :pb#131072:ob#1263616:tb=swap:\ :pc#1087488:oc#307200:tc=UNUSED:\ :pe#306400:oe#32:te=MSDOS: As you can see, partition 'e' now points to the DOS partition as pointed out by fdisk. [Ed.Note again. Remember what I said about the 32 above...] Also, there may be a problem with some versions of disklabel not recognizing the MSDOS (or MS-DOS, depending) in the te: entry above. You may need to run a "disklabel -e" to get the partition type to 'stick'. 4. Now we have to actually stick the label on the disk, which is done with disklabel. Using my example, this would be: disklabel -r -w sd0 disk0 SCSI /usr/mdec/sdboot /usr/mdec/bootsd 5. Reboot your system to see the new disk label. 6. Mount the DOS partition. I do: mount -t pcfs /dev/sd0e /dos_c Where /dos_c is just a convenient directory to mount it. 7. You're set! With the exception that the '-t' option is msdos in NetBSD, these instructions seem to work with the same facility for NetBSD. I also received a note a couple of weeks ago (that I promptly deleted because I new that I would remember what it said) that DOS extended partitions are readable if you skip the first 'n' blocks in your computations (where 'n' is your number of sectors per track). This way, you skip over the 'new' part of the DOS file system. That means that instead of the oe:32 above, you would need an oe:48 instead. Also remember that the compressed file system in DOS 6 will probably be completely Greek to your NetBSD/FreeBSD system. I seriously doubt that you will be able to read the compressed DOS file system anytime in the foreseeable future. 6.4 NFS/PC-NFS support The problems normally associated with PC-NFS are also associated with NFS in general. 6.4.1 Can I use 8K packets for NFS? When I try, I have all kinds of problems. Specifically, I get 'ring buffer overflows' or the performance is real bad. In addition to the NE2000 card, this problem can also manifest itself on other ISA networks cards that have a limited amount of memory. Ken Raeburn (raeburn@cambridge.cygnus.com) has identified a common problem with the NE2000 card and provided us with a work around: -------------------------------------------------------------------- I reported previously that I was seeing problems reading files over NFS using the NE2000 driver; timeouts would eventually be reported, no data would be read. Listing files and directories (small ones anyway) were not a problem. In the meantime, mounting NFS file systems with "rsize=1024" does get rid of this problem. Ken -------------------------------------------------------------------- As a matter of policy, specifying "rsize=1024,wsize=1024" works very well also, and makes the transfers seem to run faster. This is probably because there are fewer collisions. The disadvantage of this method comes from the kernel 'sync'ing after all NFS writes. This can slow NFS accesses considerably. As with most generalizations, this one too can do nearly as much harm as good. If you have trouble, reduce your default packet size until the problem goes away. With the newer drivers (especially the ed driver) most of these problems are solved automagically. If you are still using the original 386bsd 0.1 release, you REALLY need to upgrade. See section 6.4.5 on 'ring buffer overflows' and the 3C503 for more discussion on this problem. 6.4.2 How do I get around the NFS "Permission denied" error? The problem is not the configuration of the server (unless there is no real requirement to run it in "secure" mode, and you happen to be running it that way anyway). The problem is the fact that, even though mount request are sent on a privileged port, NFS connections are not. 6.4.3 What does the message "BAD MNT RPC: RPC Authentication error; why = Invalid client credential" mean when I try to mount something from another machine? Hellmuth Michaelis (hm@hcshh.hcs.de) offers the solution to this relatively common problem: You have to make sure that the user "root" is not present in more than 8 entries in the "/etc/group" - file on the *BSD machine. Simply remove some entries and the NFS mounts will succeed. 6.4.4 What does the message "Bad MNT RPC: RPC: Authentication error; why = Client credential too weak" mean when I try to mount something from another machine? This problem is a standard NFS problem; it simply means that your user number is not one of the ones that can mount this NFS. Normally, you will get this message when you are trying to mount a filesystem from a machine that allows 'root' to mount an NFS, but limits other users. Another documented problem with "client credentials being too weak" is the dichotomy of SunOS and 4.4 based systems. SunOS, and other commercial systems, do not allow NFS commands to come in on anything but a reserved port. There are several places that need to be addressed if weak credentials are a problem. The first is the mount command. The mount itself may work, but all references to files in the NFS will fail. This is usually the most common symptom of this problem. The solution for this is to either include the '-o resvport' keyword pair on the mount command, or the -P option. In addition to the resvport command on the mount, it may become important to include an NFS volume in your fstab. If this is the case, you will need to ensure that the resvport keyword is added on the mount line in the fstab. Finally, if you are using the automounter, you will need to make absolutely certain that you have included the resvport option in your automount maps as the default. 6.4.5 I get a lot of 'ring buffer overflow' messages using NFS and the ed0 driver. Is there a problem? David Greenman (davidg@implode.rain.com), the original author of the ed0 driver, provides us with some insight into the inner workings of the ed0 driver. It always surprises me that people don't just ask the original author these questions. :-) Anyway, the reason these are happening is that the access to the 8bit boards shared memory simply isn't fast enough to deal with full wire speeds...but the driver tries hard...so even though packets get dropped, your performance only drops to about what the ethernet board is capable of (should be in the 400-600k range with an 8bit card). NFS is especially bad because the UDP window is quite large (40k last time I looked), so the overflow condition can happen easily. I've explained this for the most part in the release notes for the driver, but these didn't make it into either the FreeBSD or NetBSD releases (we couldn't find an appropriate place to put them). >From the release notes: receive ------- The 8390 implements a shared memory ring-buffer to store incoming packets. The 8bit boards (3c503, and 8003) usually have only 8k bytes of shared memory. This is only enough room for about 4 full size (1500 byte) packets. This can sometimes be a problem, especially on the original WD8003E and 3c503. This is because these boards' shared memory access speed is also quite slow compared to newer boards - typically only about 1MB/second. The additional overhead of this slow memory access, and the fact that there is only room for 4 full-sized packets means that the ring-buffer will occasionally overflow. When this happens, the board must be reset to avoid a lockup problem in early revision 8390's. Resetting the board will cause all of the data in the ring-buffer to be lost - requiring it to be re-transmitted/received...slowing things even further. Because of these problems, maximum throughput on boards of this type is only about 400-600k per second. The 16bit boards (8013 series), however, have 16k of memory as well as much faster memory access speed. Typical memory access speed on these boards is about 4MB/second. These boards generally have no problems keeping up with full ethernet speed. The only problem I've seen with these boards is related to the (slow) performance of the system's malloc code when additional mbufs must be added to the pool. This can sometimes increase the total time to remove a packet enough for a ring-buffer overflow to occur. With NFS, the problem is really bad, though. The 3c503 does not have enough memory on the card to support the default 8k packets that NFS and other protocols use as their default. The solution for folks that are having a problem with ring buffer overflows in NFS is for them to either use the -r and -w flags to limit the packet size or use the define "NFS_BOOT_RWSIZE=8192". If NFS doesn't work with this defined, the code will automatically step down to the next smaller increment. If you KNOW that you will always be running a 3c503, you can set this define to 4096 instead, just to make sure. This should eliminate the bulk of the ring buffer overflows in NFS. 6.4.6 I am getting really poor performance out of my network, especially when talking to older networks or when performing short file transfers. What's the problem? Try turning off rfc1323 support: sysctl w net.inet.tcp.rfc1323=0 only in newer builds. In older versions you have to edit a kernel config file. RFC1323 is not yet supported by all networks, and can cause TCP performance degradation when tried. This, for example, is a known problem on older Sun and VAX hosted networks, and on many networks which are using Linux as an intermediate host. 6.4.7 Is there any PC software that will allow me to use my enormous PC with all of the unsupported hardware as a PC-NFS server? Yes. It is called SOSS, and is available from MANY FTP sources. You will need the aforementioned Clarkson Packet Drivers for it to work, but that shouldn't cause too many problems for most people. 6.5 How can I use mtools with the 'new' floppy naming convention? With the adoption of BSD 4.4, there is a new way of accessing the floppy disk drive types. The method uses the minor device number to specify different media sizes and densities. These densities are established by a table from the file /usr/src/sys/arch/i386/isa/fd.c in NetBSD (your mileage may vary). The table in FreeBSD's fd.c is likely to be slightly different. The order of the entries defines the order of the minor numbers, so the table below has the following characteristics: /dev/fd0a 0 /* default disk type */ /dev/fd0b 1 /* 1.44MB diskette */ /dev/fd0c 2 /* 1.2 MB AT-diskettes */ /dev/fd0d 3 /* 360KB in 1.2MB drive */ /dev/fd0e 4 /* 360KB PC diskettes */ /dev/fd0f 5 /* 3.5" 720KB diskette */ /dev/fd0g 6 /* 720KB in 1.2MB drive */ /dev/fd0h 7 /* 360KB in 720KB drive */ struct fd_type fd_types[] = { { 18,2,0xff,0xcf,0x1b,0x6c,80,2880,1,FDC_500KBPS,2,"1.44MB" }, { 15,2,0xff,0xdf,0x1b,0x54,80,2400,1,FDC_500KBPS,2,"1.2MB" }, { 9,2,0xff,0xdf,0x23,0x50,40, 720,2,FDC_300KBPS,2,"360KB/AT"}, { 9,2,0xff,0xdf,0x2a,0x50,40, 720,1,FDC_250KBPS,2,"360KB/PC"}, { 9,2,0xff,0xdf,0x2a,0x50,80,1440,1,FDC_250KBPS,2,"720KB" }, { 9,2,0xff,0xdf,0x23,0x50,80,1440,1,FDC_300KBPS,2,"720KB/x" }, { 9,2,0xff,0xdf,0x2a,0x50,40, 720,2,FDC_250KBPS,2,"360KB/x" }, }; In order to add a new device (such as a 2.44 Meg floppy) new tables entries are theoretically all that would be needed. As new entries are created, the minor device numbers would increase and the associated device names would be added. -- Dave Burgess Network Engineer - Nebraska On-Ramp, Inc. *bsd FAQ Maintainer / SysAdmin for the NetBSD system in my spare bedroom "Just because something is stupid doesn't mean there isn't someone that doesn't want to do it...."