Package: wpasupplicant Version: 2:2.10-24 Severity: normal X-Debbugs-Cc: [email protected]
Hi Andrej,
tl;dr: There's an (unlikeley) race in wpa_supplicant<>ifupdown integration.
Currently we start wpa_supplicnat in the "pre-up" PHASE when the interface
is still administratively DOWN (~IFF_UP).
Since wpa_supplicant takes control of UP'ing (IFF_UP) the interface
(cf. i802_set_iface_flags(bss, 1) in wpa_driver_nl80211_finish_drv_init())
delays in other pre-up scripts can race with ifup setting various ipv6
sysctls that need to be initialised *before* the interface actually comes
UP (IFF_RUNNING).
See attached test-ifupdown-race.sh for details. I needed to run it with a
RACE_WINDOW of 6 seconds in my setup to get the race/bug to trigger.
Looking at debian/ifupdown/wpasupplicant/ifupdown.sh, init_wpa_supplicant
seems to be what kicks things off:
case "$MODE" in
start)
do_start
case "$PHASE" in
pre-up)
kill_wpa_supplicant
init_wpa_supplicant || exit 1
conf_wpa_supplicant || { kill_wpa_supplicant; exit 1; }
;;
I believe we should change this to be run at `up` instead.
--Daniel
test-ifupdown-race.sh
Description: Bourne shell script
signature.asc
Description: PGP signature
