Comment 1 Aaron Sowry 2012-06-07 14:16:14 CEST

Noticed this on castro too with 3.3.0.

Comment 2 Aaron Sowry 2013-07-26 13:43:58 CEST

This is because tlstunnel's child handling is all kinds of broken.

Inside the main loop, we call a non-blocking waitpid() on our children, but then call a blocking accept() call while we wait for new connections. This means that children which exit after waitpid() but during accept() will not be reaped until the next iteration of the loop; i.e. until a new connection is made. After this, new zombie processes will eventually be created.

We can probably do better than this.

Comment 3 Aaron Sowry 2013-07-26 14:58:38 CEST

(In reply to comment #2)
> We can probably do better than this.

We should either write our own SIGCHLD handler, or ignore SIGCHLD completely. Probably the latter.

Comment 4 Aaron Sowry 2013-07-26 15:51:52 CEST

Created attachment 484 [details]
Suggested patch to avoid having to wait on child processes

Comment 5 Aaron Sowry 2013-07-30 11:38:46 CEST

r27743

Comment 6 Henrik Andersson 2013-10-11 10:50:36 CEST

Verified and works as expected on SLED11 Sp2 ThinLinc build 4110.
Tested both webadmin and webaccess.