[netctl] Move away from using wpa_actiond

Message ID 20190207144945.GA19874@Mindship-03
State New
Headers show
Series
  • [netctl] Move away from using wpa_actiond
Related show

Commit Message

Eli Schwartz via arch-projects Feb. 7, 2019, 2:49 p.m. UTC
The same functionality is provided by wpa_supplicant, so we do not need
an extra and Arch Linux specific dependency.
---
 README              |  4 ++--
 contrib/PKGBUILD.in |  1 -
 src/lib/auto.action | 31 ++++++++++++++-----------------
 src/netctl-auto     | 30 +++++++++++++-----------------
 4 files changed, 29 insertions(+), 37 deletions(-)

Comments

Eli Schwartz via arch-projects Feb. 8, 2019, 2:32 p.m. UTC | #1
On Thu, 7 Feb 2019 at 14:50, Jouke Witteveen via arch-projects
<arch-projects@archlinux.org> wrote:
>
> The same functionality is provided by wpa_supplicant, so we do not need
> an extra and Arch Linux specific dependency.

The introduction of wpa_actiond [1] hints there are issues with wpa_cli. Namely:
- partial or missing logging capabilities
- race conditions

Sadly I don't know much more about this. It'll be great if can use
wpa_cli, while not introducing new issues.

-Emil

[1] https://git.archlinux.org/wpa_actiond.git/commit/?id=c5c587771403d31ab4538e1c756e0b88a4641a92
Eli Schwartz via arch-projects Feb. 8, 2019, 7:17 p.m. UTC | #2
On Fri, Feb 8, 2019 at 3:36 PM Emil Velikov <emil.l.velikov@gmail.com> wrote:
>
> On Thu, 7 Feb 2019 at 14:50, Jouke Witteveen via arch-projects
> <arch-projects@archlinux.org> wrote:
> >
> > The same functionality is provided by wpa_supplicant, so we do not need
> > an extra and Arch Linux specific dependency.
>
> The introduction of wpa_actiond [1] hints there are issues with wpa_cli. Namely:
> - partial or missing logging capabilities
> - race conditions
>
> Sadly I don't know much more about this. It'll be great if can use
> wpa_cli, while not introducing new issues.
>
> -Emil
>
> [1] https://git.archlinux.org/wpa_actiond.git/commit/?id=c5c587771403d31ab4538e1c756e0b88a4641a92

Thanks! This detective work is highly appreciated! For some reason I
had assumed wpa_cli would have gained this functionality only after
wpa_actiond came into existence. This couldn't really explain why both
use the same parameter names though...
I wonder if the issues with wpa_cli have been taken upstream. In my
experience, the maintainer of wpa_supplicant is very pleasant to work
with.

autowifi and, later, wpa_actiond are important parts of the history of
netctl, but I tend to give over 10 years of development in
wpa_supplicant the benefit of the doubt. This means I would like to
try to move away from wpa_actiond anyway. Currently, except from those
using wpa_actiond directly, I think netctl is the only user of
wpa_actiond.

@Thomas: do you have an oppinion in these matters?

Thanks,
- Jouke
Eli Schwartz via arch-projects Feb. 15, 2019, 12:02 p.m. UTC | #3
On Sat, 9 Feb 2019 at 09:19, Thomas Bächler <thomas.baechler@gmx.de> wrote:
>
> Am 8. Februar 2019 20:17:52 MEZ schrieb Jouke Witteveen <j.witteveen@gmail.com>:
> >On Fri, Feb 8, 2019 at 3:36 PM Emil Velikov <emil.l.velikov@gmail.com>
> >wrote:
> >>
> >> On Thu, 7 Feb 2019 at 14:50, Jouke Witteveen via arch-projects
> >> <arch-projects@archlinux.org> wrote:
> >> >
> >> > The same functionality is provided by wpa_supplicant, so we do not
> >need
> >> > an extra and Arch Linux specific dependency.
> >>
> >> The introduction of wpa_actiond [1] hints there are issues with
> >wpa_cli. Namely:
> >> - partial or missing logging capabilities
> >> - race conditions
> >>
> >> Sadly I don't know much more about this. It'll be great if can use
> >> wpa_cli, while not introducing new issues.
> >>
> >> -Emil
> >>
> >> [1]
> >https://git.archlinux.org/wpa_actiond.git/commit/?id=c5c587771403d31ab4538e1c756e0b88a4641a92
> >
> >Thanks! This detective work is highly appreciated! For some reason I
> >had assumed wpa_cli would have gained this functionality only after
> >wpa_actiond came into existence. This couldn't really explain why both
> >use the same parameter names though...
> >I wonder if the issues with wpa_cli have been taken upstream. In my
> >experience, the maintainer of wpa_supplicant is very pleasant to work
> >with.
> >
> >autowifi and, later, wpa_actiond are important parts of the history of
> >netctl, but I tend to give over 10 years of development in
> >wpa_supplicant the benefit of the doubt. This means I would like to
> >try to move away from wpa_actiond anyway. Currently, except from those
> >using wpa_actiond directly, I think netctl is the only user of
> >wpa_actiond.
> >
> >@Thomas: do you have an oppinion in these matters?
> >
> >Thanks,
> >- Jouke
>
> I honestly don't remember why wpa_cli was insufficient at the time and why I wrote wpa_actiond. That must have been way over 10 years ago. I just remember that my first attempt used wpa_cli, and something obvious was missing.

I'd imagine Jouke has been running the patch for a bit and has not
seen serious issues.
Guess we could merge this and consider any issues as/if they arise?

If it were me I would have kept the cosmetics separate, but it's
nothing major so.
 - s/wpa_config/WPAConfigFile/
 - RFKill && -> if RFKill; then

Thanks for the work guys.
-Emil
Eli Schwartz via arch-projects May 16, 2019, 9:52 a.m. UTC | #4
Hi team,

On Sat, 9 Feb 2019 at 09:19, Thomas Bächler <thomas.baechler@gmx.de> wrote:
>
> Am 8. Februar 2019 20:17:52 MEZ schrieb Jouke Witteveen <j.witteveen@gmail.com>:
> >On Fri, Feb 8, 2019 at 3:36 PM Emil Velikov <emil.l.velikov@gmail.com>
> >wrote:
> >>
> >> On Thu, 7 Feb 2019 at 14:50, Jouke Witteveen via arch-projects
> >> <arch-projects@archlinux.org> wrote:
> >> >
> >> > The same functionality is provided by wpa_supplicant, so we do not
> >need
> >> > an extra and Arch Linux specific dependency.
> >>
> >> The introduction of wpa_actiond [1] hints there are issues with
> >wpa_cli. Namely:
> >> - partial or missing logging capabilities
> >> - race conditions
> >>
> >> Sadly I don't know much more about this. It'll be great if can use
> >> wpa_cli, while not introducing new issues.
> >>
> >> -Emil
> >>
> >> [1]
> >https://git.archlinux.org/wpa_actiond.git/commit/?id=c5c587771403d31ab4538e1c756e0b88a4641a92
> >
> >Thanks! This detective work is highly appreciated! For some reason I
> >had assumed wpa_cli would have gained this functionality only after
> >wpa_actiond came into existence. This couldn't really explain why both
> >use the same parameter names though...
> >I wonder if the issues with wpa_cli have been taken upstream. In my
> >experience, the maintainer of wpa_supplicant is very pleasant to work
> >with.
> >
> >autowifi and, later, wpa_actiond are important parts of the history of
> >netctl, but I tend to give over 10 years of development in
> >wpa_supplicant the benefit of the doubt. This means I would like to
> >try to move away from wpa_actiond anyway. Currently, except from those
> >using wpa_actiond directly, I think netctl is the only user of
> >wpa_actiond.
> >
> >@Thomas: do you have an oppinion in these matters?
> >
> >Thanks,
> >- Jouke
>
> I honestly don't remember why wpa_cli was insufficient at the time and why I wrote wpa_actiond. That must have been way over 10 years ago. I just remember that my first attempt used wpa_cli, and something obvious was missing.

After a bit of debugging I've noticed the move to wpa_cli broke my setup.
Namely: wpa_cli does not reliably detect when the laptop resumes into
another wireless network.
Issue happens fairly intermittently - from once a week to multiple
times a day - yet I have not seen it even once in the years of using
wpa_actiond.

Should I file a archlinux bug report, any suggestions how to proceed?

Thanks
Emil
Eli Schwartz via arch-projects May 16, 2019, 12:19 p.m. UTC | #5
On Thu, May 16, 2019 at 11:53 AM Emil Velikov <emil.l.velikov@gmail.com> wrote:
> After a bit of debugging I've noticed the move to wpa_cli broke my setup.
> Namely: wpa_cli does not reliably detect when the laptop resumes into
> another wireless network.
> Issue happens fairly intermittently - from once a week to multiple
> times a day - yet I have not seen it even once in the years of using
> wpa_actiond.

Is this related to your wpa_actiond patch [1] to wait for three
missing PONGs before failing?

> Should I file a archlinux bug report, any suggestions how to proceed?

It would be nice if you could gather some more details about the issue
you are experiencing. Most likely, it indicates a bug in wpa_cli. In
my experience, you are likely to get a helpful response over at the
hostap mailing list.

Regards,
- Jouke

[1] https://patchwork.archlinux.org/patch/230/
Eli Schwartz via arch-projects May 16, 2019, 12:32 p.m. UTC | #6
On Thu, 16 May 2019 at 13:19, Jouke Witteveen <j.witteveen@gmail.com> wrote:
>
> On Thu, May 16, 2019 at 11:53 AM Emil Velikov <emil.l.velikov@gmail.com> wrote:
> > After a bit of debugging I've noticed the move to wpa_cli broke my setup.
> > Namely: wpa_cli does not reliably detect when the laptop resumes into
> > another wireless network.
> > Issue happens fairly intermittently - from once a week to multiple
> > times a day - yet I have not seen it even once in the years of using
> > wpa_actiond.
>
> Is this related to your wpa_actiond patch [1] to wait for three
> missing PONGs before failing?
>
Haven't confirmed (did not hack on wpa_cli yet) but I suspect they're different.
The PONGs are required solely due to resource starvation - high
memory/CPU/fd usage. While here we have a wifi network change across
suspend/resume.

> > Should I file a archlinux bug report, any suggestions how to proceed?
>
> It would be nice if you could gather some more details about the issue
> you are experiencing. Most likely, it indicates a bug in wpa_cli. In
> my experience, you are likely to get a helpful response over at the
> hostap mailing list.
>
Dully noted, will do - thanks.

-Emil

Patch

diff --git a/README b/README
index 95df087..f3ca634 100644
--- a/README
+++ b/README
@@ -9,8 +9,8 @@  Optional:
 - dhcpcd or dhclient: for DHCP support
 - wpa_supplicant: for WPA support
 - dialog: for the interactive assistant
-- ifplugd: for automatic connection
-- wpa_actiond: for automatic connection
+- ifplugd: for automatic wired connections
+- ppp: for PPP support
 
 For documentation generation:
 - asciidoc
diff --git a/contrib/PKGBUILD.in b/contrib/PKGBUILD.in
index b5ce44c..f3359c6 100644
--- a/contrib/PKGBUILD.in
+++ b/contrib/PKGBUILD.in
@@ -15,7 +15,6 @@  optdepends=('dialog: for the menu based wifi assistant'
             'dhcpcd: for DHCP support (or dhclient)'
             'wpa_supplicant: for wireless networking support'
             'ifplugd: for automatic wired connections through netctl-ifplugd'
-            'wpa_actiond: for automatic wireless connections through netctl-auto'
             'ppp: for PPP connections'
             'openvswitch: for Open vSwitch connections'
            )
diff --git a/src/lib/auto.action b/src/lib/auto.action
index f81d4e3..999f66f 100755
--- a/src/lib/auto.action
+++ b/src/lib/auto.action
@@ -5,16 +5,17 @@ 
 . "$SUBR_DIR/ip"
 
 export INTERFACE="$1"
-export SSID="$2"
-export ACTION="$4"
+export ACTION="$2"
+PROFILE_STATE="$STATE_DIR/netctl-auto-$INTERFACE.profile"
 
-load_profile "$3"
 load_interface_config "$INTERFACE"
 
 case $ACTION in
-  CONNECT)
+  CONNECTED)
+    load_profile "$WPA_ID_STR"
     DhcpcdOptions+=" -K -L"
     ip_set || exit 1
+    echo "$Profile" > "$PROFILE_STATE"
     # Sandbox the eval
     if ! ( do_debug eval "$ExecUpPost" ); then
         # Failing ExecUpPost will take the connection down
@@ -22,20 +23,16 @@  case $ACTION in
         exit 1
     fi
   ;;
-  DISCONNECT)
-    # Sandbox the eval
-    if ! ( do_debug eval "$ExecDownPre" ); then
-        exit 1
+  DISCONNECTED)
+    if [[ -s "$PROFILE_STATE" ]]; then
+        load_profile "$(< "$PROFILE_STATE")"
+        rm -f "$PROFILE_STATE"
+        # Sandbox the eval
+        if ! ( do_debug eval "$ExecDownPre" ); then
+            exit 1
+        fi
+        ip_unset
     fi
-    ip_unset
-  ;;
-  LOST|REESTABLISHED)
-    # Not handled.
-    exit 0
-  ;;
-  *)
-    # ???
-    exit 1
   ;;
 esac
 
diff --git a/src/netctl-auto b/src/netctl-auto
index 18d2e10..7e4062e 100755
--- a/src/netctl-auto
+++ b/src/netctl-auto
@@ -6,7 +6,6 @@ 
 . "$SUBR_DIR/rfkill"
 . "$SUBR_DIR/wpa"
 
-: ${ACTIOND:=wpa_actiond -p /run/wpa_supplicant}
 : ${ACTION_SCRIPT:=$SUBR_DIR/auto.action}
 
 
@@ -193,11 +192,10 @@  list() {
     done
 }
 
-## Start and generate config file for the WPA supplicant, start wpa_actiond
+## Start and generate config file for the WPA supplicant, monitor for changes
 # $1: interface
 start() {
     local interface="$1"
-    local pidfile="$STATE_DIR/wpa_actiond-$interface.pid"
 
     if interface_is_up "$interface"; then
         exit_error "The interface '$interface' is already up"
@@ -206,12 +204,11 @@  start() {
         rf_enable "$interface" "$RFKill" || return 1
     fi
 
-    local wpa_conf
-    if ! wpa_conf=$(wpa_make_config_file "$interface"); then
+    if ! WPAConfigFile=$(wpa_make_config_file "$interface"); then
         exit_error "Could not create the configuration file for interface '$interface'"
     fi
-    # Disable p2p to prevent wpa_supplicant from creating another control interface.
-    echo "p2p_disabled=1" >> "$wpa_conf"
+    # Disable p2p to prevent wpa_supplicant from creating another control interface
+    echo "p2p_disabled=1" >> "$WPAConfigFile"
 
     local profile
     list_profiles | while IFS= read -r profile; do
@@ -222,16 +219,16 @@  start() {
           is_yes "${ExcludeAuto:-no}" && exit
           # Set default and exclude wpa-config as it does not fit this scheme
           [[ ${Security:=none} != "wpa-config" ]] || exit
-          printf '%s\n' "network={" "$(wpa_make_config_block)" "id_str=\"$profile\"" "}" >> "$wpa_conf"
+          printf '%s\n' "network={" "$(wpa_make_config_block)" "id_str=\"$profile\"" "}" >> "$WPAConfigFile"
           report_notice "Included profile '$profile'"
         )
     done
 
-    # Start the WPA supplicant and wpa_actiond
+    # Start the WPA supplicant and wpa_cli
     : ${WPADriver:=nl80211,wext}
     WPAOptions+=" -W"
-    if wpa_start "$interface" "$WPADriver" "$wpa_conf"; then
-        if $ACTIOND -i "$interface" -P "$pidfile" -a "$ACTION_SCRIPT"; then
+    if wpa_start "$interface" "$WPADriver" "$WPAConfigFile"; then
+        if wpa_call "$interface" -B -a "$ACTION_SCRIPT"; then
             return 0
         fi
         wpa_stop "$interface"
@@ -241,17 +238,16 @@  start() {
     return 1
 }
 
-## Stop the WPA supplicant and wpa_actiond
+## Stop the WPA supplicant, which automatically stops wpa_cli
 # $1: interface
 stop() {
     local interface="$1"
-    local pidfile="$STATE_DIR/wpa_actiond-$interface.pid"
 
-    [[ -e "$pidfile" ]] && kill "$(< "$pidfile")"
-    timeout_wait 1 '! wpa_is_active "$interface"' || wpa_stop "$interface"
+    wpa_stop "$interface"
     bring_interface_down "$interface"
-    [[ $RFKill ]] && rf_disable "$interface" "$RFKill"
-    return 0
+    if [[ $RFKill ]]; then
+        rf_disable "$interface" "$RFKill"
+    fi
 }
 
 ## Remove WPA supplicant configuration files