Subject: frag packets with 2.0C over nfs
To: None <>
From: MLH <>
List: current-users
Date: 04/16/2004 11:22:55
Since loading -current 2.0C, I've started seeing all my 2.0C hosts
sending fragged packets over NFS resulting in device timeouts *only*
on the nfs server interface.

Both the nfs clients and servers are sending them, i386 and macppc,
with a collection of different nics. No other OS'es or earlier
versions of NetBSD that I have available are generating them.

It appears that the nfs server interface timeouts only occur when
the ex0-based client is accessing the  nfs server.

11:04:04.284775 > udp (frag 12416:1480@1480+)
11:04:04.284776 > udp (frag 12416:1480@2960+)
11:04:04.284778 > udp (frag 12416:1480@4440+)
11:04:04.284779 > udp (frag 12416:1480@5920+)
11:04:04.284781 > udp (frag 12416:272@7400)
11:04:04.287981 > 1472 write [|nfs] (frag 22405:1480@0+)


nfs server:

NetBSD 2.0C NetBSD 2.0C (GENERIC.MP) #0: Mon Apr 5 14:16:43 CDT 2004 i386

fxp0 at pci2 dev 4 function 0: i82550 Ethernet, rev 16
fxp0: interrupting at ioapic0 pin 16 (irq 11)
inphy0 at fxp0 phy 1: i82555 10/100 media interface, rev. 4

        media: Ethernet autoselect (100baseTX full-duplex)
        status: active

client 1:
NetBSD 2.0C NetBSD 2.0 (GENERIC) - i386

ex0 at pci0 dev 12 function 0: 3Com 3c905B-TX 10/100 Ethernet (rev. 0x30)
ex0: interrupting at irq 10
ex0: MAC address 00:50:04:0b:82:75
exphy0 at ex0 phy 24: 3Com internal media interface
exphy0: 10baseT, 10baseT-FDX, 100baseTX, 100baseTX-FDX, auto

        media: Ethernet autoselect (100baseTX full-duplex)

client 2:
NetBSD 2.0C NetBSD 2.0C (GENERIC.MP) - macppc

gem0 at pci3 dev 15 function 0: Apple Computer GMAC Ethernet (rev. 0x01)
gem0: interrupting at irq 41
gem0: Ethernet address 00:30:65:e4:33:52, 10KB RX fifo, 4KB TX fifo
brgphy0 at gem0 phy 0: BCM5400 1000BASE-T media interface, rev. 7

        media: Ethernet autoselect (100baseTX full-duplex)
        status: active


What did I miss or what's changed in 2.0C ?