Using Linux and DOS Together

I found it effective to use loadlin, a DOS-based loader; this has the distinct advantage of always booting a working system before running Linux. My experience with LILO has been if you don't do it right, your system is only useful as a paper weight. In my MS-DOS config.sys, I take advantage of the menus, and the result is shown in Listing 1.

menuitem=dos
menuitem=simple.dos
menuitem=linux.1.2.8
menuitem=scandisk
menudefault=linux.1.2.8,5

[linux.1.2.8]
SHELL=C:\loadlin\loadlin.exe \loadlin\zimage.128 root=/dev/hdc2 -v ro

[simple.dos]

[dos]
DEVICE=C:\DOS\HIMEM.SYS
DEVICEHIGH=C:\MAGICS20\CDIFINIT.SYS /T:X
DEVICEHIGH=C:\MTM\MTMCDAI.SYS /D:MTMIDE01
DEVICEHIGH /L:2,12048 =C:\DOS\SETVER.EXE
DOS=HIGH
STACKS=9,256

[scandisk]
SHELL=c:\dos\scandisk.exe /all /checkonly

I boot Linux with a delay of 5 seconds, the advantage being that the system can always boot DOS and will work in some capacity. I find this preferable to using LILO and modifying the master boot record on your hard disk (if you do anything wrong, you need to boot from floppy to recover).

One can easily select several kernels and/or configurations from the command line. Using loadlin, you have to make a compressed kernel (make zImage), and then put it on a DOS partition. I find this strategy effective even when installing Linux the first time (instead of dealing with a boot and root floppy, the system can boot the kernel with only a root floppy needed). You can easily add to the menu to have several different kernels to boot from. Remember, you can use the rdev utility to build defaults (like the root device) into the kernel.

In your autoexec.bat you can use the strategy:

goto %config%

:simple.dos
PATH=C:\marty\bin;C:\gnu\bin;C:\dos;C:\
goto end

:dos
SET SOUND16=C:\MAGICS20
C:\MAGICS20\SNDINIT /b
SET BLASTER=A220 I7 D1 T4
C:\DOS\SMARTDRV.EXE 512 512 /C
C:\DOS\IMOUSE.COM
PROMPT $p$g
SET PATH=c:\gnu\bin;C:\MARTY\BIN;C:\WINDOWS;C:\DOS;

SET TEMP=C:\DOS
JOIN d: \marty
JOIN f: \gnu
goto end

:end

The simple.dos setting is conceptually the same as booting Linux in single user mode. I find it very useful for debugging a DOS system. If you want, you can add config.sys menu entries to boot different kernels, boot Linux in single user mode, boot Linux from floppies, etc.

In the standard Linux kernel configuration, the UMSDOS file system isn't enabled. UMSDOS has a number of major advantages if you need file systems to be shared between Linux and DOS. It retains full Unix semantics, so you don't have to always be handicapped by DOS problems such as:

Using UMSDOS you can take advantage of a file system shared between DOS and Linux, with the appearance of being a Linux file system when you run Linux. If you want files to be portable between MS-DOS and Linux, restrict yourself to DOS filenames (8+3 characters). Don't use links if you want the files to appear under DOS. With a Linux file system, it's easy do things like create “dot files”, do gzip-r on trees, and create links and backup files. Any file is readable in MS-DOS; however, if you don't conform to the MS-DOS file naming conventions, files are “munged” (that is, their names are squeezed to fit within the 8+3 namespace). This munging is similar to what happens in mfs; those who use PC-NFS are probably familiar with this.

When you start running the UMSDOS file systems, remember to run the application called umssync, which creates consistency between the --linux-.--- files and the directory contents. You can have problems if you add or delete files under DOS without Linux knowing about it. Call umssync from /etc/rc.d/rc.local or /etc/rc.d/rc.M after the mount takes place, and this shouldn't be a problem.

I've noticed a problem in UMSDOS files systems—the mount points are owned by root, only writable by root, and the date is the beginning of the epoch. A simple workaround is after mounting, do chown/chmod to the mount points as appropriate (in your /etc/rc.d/rc.local file. Also, I find it useful to occasionally run scandisk from DOS (notice the scandisk target in config.sys).

There is a performance penalty for DOS and UMSDOS file systems compared to normal ext2. The penalty becomes severe if you have several hundred files in a single directory (when you do an ls, get a cup of coffee). What I've noticed is sequential I/O (with a tester called Bonnie) is marginally faster on ext2 than UMSDOS.

But UMSDOS is ideal if you're doing work with DOSEMU. You put DOS files on UMSDOS partitions, and you can easily access them from DOS, DOSEMU or Linux. If they keep within the DOS file system bounds of 8+3 characters, they look the same on both DOS and Linux. UMSDOS partitions provide a big advantage when sharing files with DOS (much more so then the MSDOS file system, since it treats Linux files as Linux files), but performance has to be watched.