This... is not any of the things I said I would be working on. I can
only say that a herd of very hairy yaks led me astray. Looking at bug
66 I spotted some problems with our handling of MTUs / maximum frame
sizes. Looking at that I found some weirdness and some real, if
minor, bugs in the sizing and handling of the packet pools.
Changes in v2:
* Stefano convinced me that packet_check_range() is still worthwhile.
* So don't remove it... but in looking at it I spotted various
flaws in the checks, so address those in a number of new patches.
David Gibson (12):
test focus
hack: stop on fail, but not perf fail
make passt dumpable
packet: Use flexible array member in struct pool
packet: Don't pass start and offset separately too
packet_check_range()
packet: Don't hard code maximum packet size to UINT16_MAX
packet: Remove unhelpful packet_get_try() macro
util: Add abort_with_msg() and ASSERT_WITH_MSG() helpers
packet: Distinguish severities of different packet_{add,git}_do()
errors
packet: Move packet length checks into packet_check_range()
tap: Don't size pool_tap[46] for the maximum number of packets
packet: More cautious checks to avoid pointer arithmetic UB
dhcpv6.c | 2 +-
ip.c | 2 +-
isolation.c | 2 +-
packet.c | 106 ++++++++++++++++++++++----------------------------
packet.h | 19 ++++++---
passt.h | 2 -
tap.c | 18 +++++++--
tap.h | 3 +-
test/lib/term | 1 +
test/lib/test | 4 +-
test/run | 38 +++++++++---------
util.c | 19 +++++++++
util.h | 25 +++++-------
vu_common.c | 34 ++++++++++------
14 files changed, 153 insertions(+), 122 deletions(-)
--
2.47.1