(am on lkml but not linux-net - linux-net replies please cc: me)
This broke sometime between 2.6.0-test8-mm1 (which works) and -test9.
Userspace is Red Hat Rawhide/Fedora Core, kernel compiled with RH gcc-3.3.2-1
Basic summary - when /sbin/nameif goes to rename an interface, things
go totally pear-shaped. nameif itself croaks, and apparently leaves
data structures corrupted - on a subsequent 'ifup lo' or 'shutdown -r'
the system locks up solid.
Unable to handle kernel NULL pointer dereference at virtual address 000000d8
printing eip:
c033eb32
*pde = 0f670067
*pte = 00000000
Oops: 0000 [#1]
CPU: 0
EIP: 0060:[<c033eb32>] Not tainted
EFLAGS: 00010246
EIP is at addrconf_sysctl_unregister+0x7/0x3a
eax: 0000009c ebx: 0000009c ecx: 00000000 edx: 00000000
esi: 00000000 edi: cfd4a800 ebp: cedabea8 esp: cedabea4
ds: 007b es: 007b ss: 0068
Process nameif (pid: 280, threadinfo=cedaa000 task=ceaeecc0)
Stack: 0000009c cedabec0 c033cdda 0000009c c042e318 cfd4a800 0000000a cedabee0
c012673d c042e318 0000000a cfd4a800 cfd4a800 00000000 cedabf34 cedabf10
c02ec593 c04c5848 0000000a cfd4a800 cfd4a800 cedabf34 00000010 cedabf24
Call Trace:
[<c033cdda>] addrconf_notify+0xc4/0xfb
[<c012673d>] notifier_call_chain+0x1c/0x37
[<c02ec593>] dev_ifsioc+0x2f3/0x391
[<c02ec867>] dev_ioctl+0x236/0x33b
[<c0322188>] inet_ioctl+0xbf/0xcd
[<c02e4ee7>] sock_ioctl+0x27d/0x2a3
[<c0156922>] sys_ioctl+0x200/0x246
[<c010a96b>] syscall_call+0x7/0xb
Code: 8b 58 3c 85 db 74 27 c7 40 3c 00 00 00 00 ff 33 e8 dc 13 de
Reproducible on both -test9 and -test9-mm1. Sorry I didn't catch it sooner,
I hadn't booted the laptop in the docking station under -test9 till
yesterday, so the call to nameif didn't actually have to rename
anything until then.
Attachment:
pgp00116.pgp
Description: PGP signature
