Bug 1550097

Hi Eric,

(In reply to Eric Garver from comment #0)
> Description of problem:
> 
> See bug 1501418 comment 11.
> 
> Setting the MAC of an interface fails with the following: 
> 
>   # ip link set dev p0 address "f0:00:00:01:01:01"
>   Invalid address length 6 - must be 42401 bytes
> 
> This was introduced with iproute-4.11.0-9.el7.
> 
> Marking urgent as it blocks/breaks a lot of OVS scenarios and tests.

I can't reproduce this on my local RHEL7 VM. Maybe it is a kernel issue? With which kernel version are you able to reproduce the problem?

Does this happen only with veth type interfaces or others as well?

Thanks, Phil

(In reply to Phil Sutter from comment #3)
> Hi Eric,
> 
> (In reply to Eric Garver from comment #0)
> > Description of problem:
> > 
> > See bug 1501418 comment 11.
> > 
> > Setting the MAC of an interface fails with the following: 
> > 
> >   # ip link set dev p0 address "f0:00:00:01:01:01"
> >   Invalid address length 6 - must be 42401 bytes
> > 
> > This was introduced with iproute-4.11.0-9.el7.
> > 
> > Marking urgent as it blocks/breaks a lot of OVS scenarios and tests.
> 
> I can't reproduce this on my local RHEL7 VM. Maybe it is a kernel issue?
> With which kernel version are you able to reproduce the problem?
> 
> Does this happen only with veth type interfaces or others as well?
> 
> Thanks, Phil

I was also unable to reproduce it outside of the OVS testsuite. So I went looking for suspicious code in iproute and found a use after free in nl_get_ll_addr_len(). tb array is filled from the dynamically allocated answer. But tb[IFLA_ADDRESS] is accessed after answer is freed.

This was introduced by

  86bf43c7c2fd ("lib/libnetlink: update rtnl_talk to support malloc buff at run time")

which was backported in iproute-4.11.0-9

(In reply to Eric Garver from comment #4)
> (In reply to Phil Sutter from comment #3)
> > Hi Eric,
> > 
> > (In reply to Eric Garver from comment #0)
> > > Description of problem:
> > > 
> > > See bug 1501418 comment 11.
> > > 
> > > Setting the MAC of an interface fails with the following: 
> > > 
> > >   # ip link set dev p0 address "f0:00:00:01:01:01"
> > >   Invalid address length 6 - must be 42401 bytes
> > > 
> > > This was introduced with iproute-4.11.0-9.el7.
> > > 
> > > Marking urgent as it blocks/breaks a lot of OVS scenarios and tests.
> > 
> > I can't reproduce this on my local RHEL7 VM. Maybe it is a kernel issue?
> > With which kernel version are you able to reproduce the problem?
> > 
> > Does this happen only with veth type interfaces or others as well?
> > 
> > Thanks, Phil
> 
> I was also unable to reproduce it outside of the OVS testsuite. So I went
> looking for suspicious code in iproute and found a use after free in
> nl_get_ll_addr_len(). tb array is filled from the dynamically allocated
> answer. But tb[IFLA_ADDRESS] is accessed after answer is freed.
> 
> This was introduced by
> 
>   86bf43c7c2fd ("lib/libnetlink: update rtnl_talk to support malloc buff at
> run time")
> 
> which was backported in iproute-4.11.0-9

ElectricFence and gdb agree.

$ cat bz1550097.sh 
#!/bin/sh
set -e

ip link delete ovs-p0 || true
ip netns delete at_ns0 || true

ip netns add at_ns0
ip link add p0 type veth peer name ovs-p0
ip link set p0 netns at_ns0
ip link set dev ovs-p0 up

ip netns exec at_ns0 ip addr add 10.1.1.1/24 dev p0
ip netns exec at_ns0 ip link set dev p0 up
ip netns exec at_ns0 ef ip link set dev p0 address f0:00:00:01:01:01

Note the "ef" in the last command.

$ sudo sh bz1550097.sh

  Electric Fence 2.2.2 Copyright (C) 1987-1999 Bruce Perens <bruce>
/bin/ef: line 20: 24650 Segmentation fault      (core dumped) ( export LD_PRELOAD=libefence.so.0.0; exec "$@" )

$ gdb -c core.24650  /usr/sbin/ip
...
Core was generated by `ip link set dev p0 address f0 00 00 01 01 01'.
Program terminated with signal 11, Segmentation fault.
#0  0x000000000041db1c in nl_get_ll_addr_len ()
Missing separate debuginfos, use: debuginfo-install iproute-4.11.0-9.el7.x86_64
(gdb) bt
#0  0x000000000041db1c in nl_get_ll_addr_len ()
#1  0x000000000041e5dc in iplink_parse ()
#2  0x0000000000420146 in iplink_modify ()
#3  0x00000000004207b2 in do_iplink ()
#4  0x0000000000408384 in do_cmd ()
#5  0x0000000000407e97 in main ()

(gdb) layout asm
...
   |0x41db00 <nl_get_ll_addr_len+160>       callq  0x448890 <parse_rtattr_flags>
   |0x41db05 <nl_get_ll_addr_len+165>       cmpq   $0x0,0x18(%rsp)
   |0x41db0b <nl_get_ll_addr_len+171>       mov    0x8(%rsp),%rdi
   |0x41db10 <nl_get_ll_addr_len+176>       je     0x41db40 <nl_get_ll_addr_len+224>
   |0x41db12 <nl_get_ll_addr_len+178>       callq  0x405e60 <free@plt>
   |0x41db17 <nl_get_ll_addr_len+183>       mov    0x18(%rsp),%rax
  >|0x41db1c <nl_get_ll_addr_len+188>       movzwl (%rax),%eax
   |0x41db1f <nl_get_ll_addr_len+191>       sub    $0x4,%eax
   |0x41db22 <nl_get_ll_addr_len+194>       mov    0x598(%rsp),%rdx
   |0x41db2a <nl_get_ll_addr_len+202>       xor    %fs:0x28,%rdx
   |0x41db33 <nl_get_ll_addr_len+211>       jne    0x41db53 <nl_get_ll_addr_len+243>
   |0x41db35 <nl_get_ll_addr_len+213>       add    $0x5a8,%rsp
   |0x41db3c <nl_get_ll_addr_len+220>       retq
...

Hi Eric,

(In reply to Eric Garver from comment #5)
> (In reply to Eric Garver from comment #4)
[...]
> > I was also unable to reproduce it outside of the OVS testsuite. So I went
> > looking for suspicious code in iproute and found a use after free in
> > nl_get_ll_addr_len(). tb array is filled from the dynamically allocated
> > answer. But tb[IFLA_ADDRESS] is accessed after answer is freed.
> > 
> > This was introduced by
> > 
> >   86bf43c7c2fd ("lib/libnetlink: update rtnl_talk to support malloc buff at
> > run time")
> > 
> > which was backported in iproute-4.11.0-9

Oh, I see! That also explains why some builds don't expose the issue - it
simply depends on how the code was compiled.

> ElectricFence and gdb agree.

Thanks for analyzing the issue!

Patch sent upstream: https://marc.info/?l=linux-netdev&m=151989693717192&w=2

Since the problem described in this bug report should be
resolved in a recent advisory, it has been closed with a
resolution of ERRATA.

For information on the advisory, and where to find the updated
files, follow the link below.

If the solution does not work for you, open a new bug report.

https://access.redhat.com/errata/RHEA-2018:0815