cvs commit: src/sys/gnu/ext2fs ext2_extern.h ext2_vfsops.c ext2_vnops.c

Julian Elischer julian at
Mon Sep 27 14:35:58 PDT 2004

Alfred Perlstein wrote:

>* Poul-Henning Kamp <phk at> [040927 13:38] wrote:
>>phk         2004-09-27 20:38:46 UTC
>>  FreeBSD src repository
>>  Modified files:
>>    sys/gnu/ext2fs       ext2_extern.h ext2_vfsops.c ext2_vnops.c 
>>  Log:
>>  Desupport device nodes on EXT2 filesystems.
>Are you removing mknod(2) for device numbers from our filesystems?
>I'm sorry if I missed out on where this was discussed, do you have
>a pointer explaining what's going on here?

Since devfs is the way that one gets to devices,  and major number 
effectively become random numbers that
mean nothing, then having nodes in filesystems jumping into random 
device drivers is not a good thing.
You still need to have the nodes exist so you can export them via NFS to 
other systems that need them,
but you do not want the freeBSD machine to consider them as devices, 
just as notes that you might
need to pass through to someone.

Consider a solaris machine exporting a root filesystem (with /dev) to 

the device nodes need to have linux major number s and not Solaris major 
That's ok as long as you don't try actually USE them on the solaris machine.
It's the same here except there is no "Native" device node in a filesystem.

So don't allow interpretting of such nodes a devices because there is a 
100% chance that it will be wrong.

More information about the cvs-src mailing list