First of all, I'm pretty sure this bug is in ne.c, the reason is that the only change in my machine is a died 3c509 to a ne clone (ne.o), from that point my nfs has become flacky. Description: One machine (head) the server, 486dx4, ne clone, 2.4.15-pre9, knfsd The other machine (quartz) client, 2x200mmx, old 3c503 (3c503.o), 2.5.3 & 2.4.17-rc1. Both machines have a crossover clable using a realtek 8139C (head) and a 3c905B (quartz) wich works ok, but the old ne card is in a coax network with other clients, so nfs over ne.o should work ok. As I said the problem is easy, and I have logs on both ends, the only change that caused this problem is a died 3c509 -> EISA ne clone so being this the only change it should be easy, but I can't find it. The problem seems to be that the end packet of a multi-packet UDP transfer never gets send, TCP works realiable and fast, but I didn't try hard to reproduce it, only some http transfers. An excerpt of the log: first the server (head): 17:29:54.910504 < quartz.4216423526 > head.nfs: 116 read fh Unknown/1 8192 bytes @ 102400 (DF) (ttl 64, id 0) 17:29:54.950484 > head.nfs > quartz.4216423526: reply ok 1472 read REG 100744 ids 0/0 sz 645911 nlink 1 rdev ffffffff fsid 302 nodeid 9c8d2 a/m/ctime 1014395394.000000 1011347457.000000 1011347457.000000 (frag 4075:1480@0+) (ttl 64) 17:29:54.950484 > head > quartz: (frag 14075:1480@1480+) (ttl 64) 17:29:54.950484 > head > quartz: (frag 14075:1480@2960+) (ttl 64) 17:29:54.950484 > head > quartz: (frag 14075:1480@4440+) (ttl 64) 17:29:54.950484 > head > quartz: (frag 14075:1480@5920+) (ttl 64) 17:29:54.950484 > head > quartz: (frag 14075:900@7400) (ttl 64) 17:29:55.610156 < quartz.4216423526 > head.nfs: 116 read fh Unknown/1 8192 bytes @ 102400 (DF) (ttl 64, id 0) then what the client (quartz) sees: 17:30:11.268447 > quartz.4216423526 > head.nfs: 116 read fh Unknown/1 8192 bytes @ 102400 (DF) (ttl 64, id 0) 17:30:11.308348 < head.nfs > quartz.4216423526: reply ok 1472 read REG 100744 ids 0/0 sz 645911 nlink 1 rdev ffffffff fsid 302 nodeid 9c8d2 a/m/ctime 1014395394.000000 1011347457.000000 1011347457.000000 (frag 4075:1480@0+) (ttl 64) 17:30:11.310481 < head > quartz: (frag 14075:1480@1480+) (ttl 64) 17:30:11.312078 < head > quartz: (frag 14075:1480@2960+) (ttl 64) 17:30:11.313692 < head > quartz: (frag 14075:1480@4440+) (ttl 64) 17:30:11.962381 > quartz.4216423526 > head.nfs: 116 read fh Unknown/1 8192 bytes @ 102400 (DF) (ttl 64, id 0) The last packet from the server (head) never arrives, 17:29:54.950484 > head > quartz: (frag 14075:900@7400) (ttl 64) ^ This never reaches the other end. The error is fairly easy to reproduce, just trying to read a file contents dumps an input/output error. As I said TCP works reliable and fast, so does doing an ls -la of the /etc directory, I'm open to more attemps, so if one sees the ligth please mail me ;) P.S. I know the ne driver is a old one, but the machine is a 486, and I don't have free pci slots, and as I said the card works ok because tcp is fast and don't ask for retransmits. I mount with rsize=8192,wsize=8192 but without it it also fails. -- Jorge Nerin