Hi,
I've noticed that ZFS uses:
sys/zfs_znode.h:
uint_t z_mapcnt; /* number of memory maps to file */
while UFS (and the others) do:
sys/fs/ufs_inode.h:
long i_mapcnt; /* mappings to file pages */
Now for UFS (and others), the count of mapped pages on a file can
be 32bit since the filesize there is always small enough, so the
use of "long" is overkill there. You can't mmap a > 1TB file on
UFS, and that's only 128M 8kB pages.
But:
Why is it appropriate to use a 32bit counter on ZFS ? 4G pages
are no more than 32TB on SPARC (16TB on AMD64), far below the
maximum filesize on ZFS, and still significantly below the VA
space on either sparcv9/amd64.
Does this mean ZFS has a bug there ?
Thx,
FrankH.
_______________________________________________
ufs-discuss mailing list
[email protected]
