actkbd randomly stops working, until restarted.

[fff7d1bc]

Hello,

I am hitting this bug for many years, on many computers. Randomly, sometimes 3 times a day, sometimes once a month, actkbd does not 'react' to events. Oddly enough, the process does work, when I attach to ith with strace I do see that it does work, lines like:

read(3, "\227\311AX\0\0\0\0\1n\3\0\0\0\0\0\4\0\4\0\240\0\0\0\227\311AX\0\0\0\0"..., 4096) = 72

Pops up, but nothing more happens, And after I restart the actkbd process I can see that right after read, it runs sendto, along with many, many other calls that results in executing what was configured.

read(3, "\227\311AX\0\0\0\0\1n\3\0\0\0\0\0\4\0\4\0\240\0\0\0\227\311AX\0\0\0\0"..., 4096) = 72
sendto(4, "<29>Dec  2 20:20:55 actkbd[3227]"..., 88, MSG_NOSIGNAL, NULL, 0) = 88
rt_sigaction(SIGINT, {SIG_IGN, [], SA_RESTORER, 0x3f51233250}, {SIG_DFL, [], SA_RESTORER, 0x3f51233250}, 8) = 0
rt_sigaction(SIGQUIT, {SIG_IGN, [], SA_RESTORER, 0x3f51233250}, {SIG_DFL, [], SA_RESTORER, 0x3f51233250}, 8) = 0
rt_sigprocmask(SIG_BLOCK, [CHLD], [], 8) = 0
clone(strace: Process 3900 attached

actkbd is started as:

/usr/bin/actkbd -c /etc/actkbd.conf -D -q -x -l -p /var/run/actkbd.pid -d /dev/input/event4

Can you please comment on this?

[thkala]

I am hitting this bug for many years, on many computers. Randomly, sometimes 3 times a day, sometimes once a month, actkbd does not 'react' to events. Oddly enough, the process does work, when I attach to ith with strace I do see that it does work, lines like: read(3, "\227\311AX\0\0\0\0\1n\3\0\0\0\0\0\4\0\4\0\240\0\0\0\227\311AX\0\0\0\0"..., 4096) = 72

If it's reading, then it's not completely blocked...

Pops up, but nothing more happens, And after I restart the actkbd process I can see that right after read, it runs sendto, along with many, many other calls that results in executing what was configured. read(3, "\227\311AX\0\0\0\0\1n\3\0\0\0\0\0\4\0\4\0\240\0\0\0\227\311AX\0\0\0\0"..., 4096) = 72 sendto(4, "<29>Dec 2 20:20:55 actkbd[3227]"..., 88, MSG_NOSIGNAL, NULL, 0) = 88 rt_sigaction(SIGINT, {SIG_IGN, [], SA_RESTORER, 0x3f51233250}, {SIG_DFL, [], SA_RESTORER, 0x3f51233250}, 8) = 0 rt_sigaction(SIGQUIT, {SIG_IGN, [], SA_RESTORER, 0x3f51233250}, {SIG_DFL, [], SA_RESTORER, 0x3f51233250}, 8) = 0 rt_sigprocmask(SIG_BLOCK, [CHLD], [], 8) = 0 clone(strace: Process 3900 attached actkbd is started as: /usr/bin/actkbd -c /etc/actkbd.conf -D -q -x -l -p /var/run/actkbd.pid -d /dev/input/event4 Can you please comment on this?

...but here's the catch: actkbd (a) is single-threaded and (b) uses system() in the same thread to execute commands. Yes, I know, not nice; the whole program is basically a hack - it just worked well enough for me and a few friends back when I wrote it. Bottom-line is, it can conceivably miss key events, esp. if you are launching commands without the '&' shell character at the end that pushes them to the background. Missing a key _release_ means that it then keeps thinking that some key is still pressed, which messes-up the key mask matching. If you can figure out which key is wedged, then pressing and releasing it should fix the issue. You can probably confirm whether this is what is happening by increasing the verbosity level e.g. to 9 and checking your syslog messages. Keep in mind that doing that will essentially dump all key presses - including passwords and credit card numbers - to your system log. Beware. I have not really worked on actkbd for a very very long time. I have some vague plans about bringing it into this decade by replacing libcfg+ with something that still exists, adding proper threading and multi-device support, maybe even dropping all the macros for some clean C++. But all of that basically amounts to a full-blown rewrite, and I find myself rather pressed for time, so I'd not hold my breath if I were you... Of course, I am open to pull requests, should any find their way to my inbox :-)

[rehno-lindeque]

I think I've got the same issue, and weirdly it seems to resolve itself as soon as I look at the logs with journalctl. I.e. I'll just run journalctl -u actkbd@-dev-input-event2.service -f and then suddenly everything works again, almost as if it's been buffering or something. Very strange indeed.

esp. if you are launching commands without the '&' shell character at the end that
pushes them to the background.

Just to note: for my case, I do add & to everything that is executed through actkbd. (Doesn't solve the issue unfortunately).