Re: More puzzing nmbclusters

[mlelstv%serpens.de@localhost (Michael van Elst)]

nd%syndicat.com@localhost (Niels Dettenbach) writes:

>i can prove that what Konrad wrotes. After regularly problems (out of 
>ressources) with nmbclusters to low on a NetBSD 5 we raised it up to 32xxx and 
>our machines was running over years. May be that there is a theoretical (or 
>even practical) "insecurity" but for many of us it broughts a situation which 
>seems not acceptable.

NetBSD-5 for i386:
#ifndef NMBCLUSTERS
[...]
#ifdef GATEWAY
#define NMBCLUSTERS     2048            /* map size, max cluster allocation */
#else
#define NMBCLUSTERS     1024            /* map size, max cluster allocation */
#endif
...

NetBSD-6 for i386 (the default is no longer a constant):
#ifndef NMBCLUSTERS_MAX
#define NMBCLUSTERS_MAX (0x2000000 / MCLBYTES)  /* Limit to 64MB for clusters */

There is an extra i386-specific limit is introduced in NetBSD-6, probably
to avoid KVA problems. You should be able to override this limit by
compiling a kernel with, e.g. option NMBCLUSTERS_MAX 32768.

I also see that the comment (64MB) doesn't agree with the limit, which
is only 32MB (16384 * 2048 bytes).