There seems to be a problem with nfsd in nfs-server-2.2beta37 (the one included in RedHat 5.2) here are the exports from Linux: /tmp/try-export *.cs.huji.ac.il(rw,insecure) *** note the 'insecure' you then mount it from BSDI (3.1 or 4.0) and try to write to a file there (under the mount point). the file gets created with size 0 no matter what you do. the following message is logged on the linux box: Mar 11 01:06:42 fury nfsd[280]: strange write req from amos-04.cs.huji.ac.il: co unt 1717856768 len 5 (fury is the linux box, amos-04 is the BSDI) additional facts: 1. when writing from BSDI to any other OS we have here everything is fine. 2. this problem is *not occuring* when you try to write from any other OS (Sun0S 4, Solaris, IRIX) to linux 3. the problem doesn't exist neither in the older nfs-server-2.2beta29 (shipped in RedHat 5.1) nor in the new nfsd from linux-nfs-0.4.22 based on knfsd (kernel 2.2.x level nfsd support) There seems to be a new condition in nfsd.c (in 2.2beta37): 492: if (count > NFS_MAXDATA || count > argp->data.data_len) { 493: Dprintf(L_NOTICE, 494: "strange write req from %s: count %lu len %lu", 495: nfsclient->clnt_name, count, argp->data.data_len); 496: return NFSERR_IO; 497: } which BSDI nfs client obviously doesn't satisfy. (i'm posting this also to the relevant mailing-list, please post any suggestions on comp.os.linux.development.system) kreso ---- CS System Group Hebrew University of Jerusalem
Does this bug persist with the knfsd running on a 2.2 kernel? Please reopen this bug if so.