This series mends some problems in headroom management in ip6_gre
module. The current code base has the following three closely-related

- ip6gretap tunnels neglect to ensure there's enough writable headroom
  before pushing GRE headers.

- ip6erspan does this, but assumes that dev->needed_headroom is primed.
  But that doesn't happen until ip6_tnl_xmit() is called later. Thus for
  the first packet, ip6erspan actually behaves like ip6gretap above.

- ip6erspan shares some of the code with ip6gretap, including
  calculations of needed header length. While there is custom
  ERSPAN-specific code for calculating the headroom, the computed
  values are overwritten by the ip6gretap code.

The first two issues lead to a kernel panic in situations where a packet
is mirrored from a veth device to the device in question. They are
fixed, respectively, in patches #1 and #2, which include the full panic
trace and a reproducer.

The rest of the patchset deals with the last issue. In patches #3 to #6,
several functions are split up into reusable parts. Finally in patch #7
these blocks are used to compose ERSPAN-specific callbacks where
necessary to fix the hlen calculation.

Petr Machata (7):
  net: ip6_gre: Request headroom in __gre6_xmit()
  net: ip6_gre: Fix headroom request in ip6erspan_tunnel_xmit()
  net: ip6_gre: Split up ip6gre_tnl_link_config()
  net: ip6_gre: Split up ip6gre_tnl_change()
  net: ip6_gre: Split up ip6gre_newlink()
  net: ip6_gre: Split up ip6gre_changelink()
  net: ip6_gre: Fix ip6erspan hlen calculation

 net/ipv6/ip6_gre.c | 184 +++++++++++++++++++++++++++++++++++++++++------------
 1 file changed, 145 insertions(+), 39 deletions(-)


Reply via email to