Heya, It seems I'm experiencing some data corruption on nfs when -w or -r aren't powers of 2.
I have a local file with these settings: % md5 sunclock.diff MD5 (sunclock.diff) = 9f002849da08cd6ab76032a8cf2726e1 now, if I export the filesystem (nfsd -tu -n 4) it's on I get data corruption when I try to use a readsize or writesize that's not a power of 2: % mount_nfs -3 -T spectre:/home /mnt % md5 /mnt/inphobia/sunclock.diff MD5 (/mnt/inphobia/sunclock.diff) = 9f002849da08cd6ab76032a8cf2726e1 % umount /mnt % mount_nfs -3 -T -r 32768 -w 32768 spectre:/home /mnt % md5 /mnt/inphobia/sunclock.diff MD5 (/mnt/inphobia/sunclock.diff) = 9f002849da08cd6ab76032a8cf2726e1 % umount /mnt % mount_nfs -3 -T -r 32000 -w 32000 spectre:/home /mnt % md5 /mnt/inphobia/sunclock.diff MD5 (/mnt/inphobia/sunclock.diff) = d9bfc86665d9619e19c2a317f12b0c09 % umount /mnt % mount_nfs -3 -U -r 32000 -w 32000 spectre:/home /mnt % md5 /mnt/inphobia/sunclock.diff MD5 (/mnt/inphobia/sunclock.diff) = d9bfc86665d9619e19c2a317f12b0c09 % umount /mnt The nfs client is i386 -current, the server is i386 4.0 stable. Yes, the manual page says I should use a power of 2 greater than or equal to 1024, but perhaps it could warn that if I don't my data will get corrupted? Or perhaps have mount_nfs refuse mounts with an incorrect read/writesize - sane defaults? And the background of this: I was lazy and just wanted to do a mount -r 32k -w 32k, but since that is refused I was lazy again and just mounted with -r 32000 -w 32000, which makes the actual mount show up as: spectre:/home on /mnt type nfs (v3, udp, wsize=31744, rsize=31744, rdirsize=31744, timeo=100) Ohw, and it's not just md5 that fails. While the contents of text files look sane, I actually stumbled upon this while trying to recode a flac file to an mp3. Mounting with wrong readsizes caused flac to spew out errors, after remounting with correct sizes it worked fine. // nick
