To make the code a bit more flexible, let's get rid of the forking-code
in serve() and do it in main(). This way, we are more liberal in the
future to possibly handle it in a different way.
And many other things, too many to list here. For example, it now
properly logs uds instead of erroring out.
Separating concerns in many places definitely improves the readability.
- add missing header netinet/in.h for socket declarations (POSIX).
- rename sendfile to responsefile, sendfile(2) is a syscall on FreeBSD.
- remove _XOPEN_SOURCE: this will give a warning about strptime on Linux
glibc, but unbreaks the build on NetBSD and FreeBSD.
thanks also to josuah and quinq for testing!
Thanks Michael Forney for reporting this! We cannot use identifiers
beginning with an underscore, says the C99-standard, section 7.1.3:
"All identifiers that begin with an underscore are always reserved for
use as identifiers with file scope in both the ordinary and tag name
spaces."
We go around this by putting the underscore at the end.
The (c)-symbol has become more of a remnant after the Berne convention
has been signed. Given the ISC exploits some simplifications introduced
with the Berne convention, it just makes sense to drop this relict as
well and just state our Copyright without much ado about nothing.
https://opensource.org/licenses/ISC
Check for its presence and bail out if found.
If the socket file is present, either a server is already bound to it,
or the last one errored out and we'd want to inspect this.
Also it could be an unrelated file given by error.
As given in the config, we match a regex of hosts to a canonical host
which points to an internal directory.
Regexes are compiled on initialization, so we can error out early.
The rest is just modifications to use relative directories rather than
absolute ones, as we chdir() into the vhost directories dynamically.
Given we normalize the targets beforehand, there is no danger of
malformed requests escaping the vhost-context.
