cgit with busybox httpd

cgit with busybox httpd

[Andreas Mahling]

hello,

I'm in the process to configure a cgit instance for my private network at home.
Because my internet router will be the host for git and cgit, I'm
looking for a ligthweight setup.
I would like to use busybox httpd as webserver, which has a very low
footprint and supports cgi, but no other fancy bells and whistles,
especially no url rewrite.

It works in principle, but there is a problem with the URLs generated
by cgit: they seem always containing a trailing slash. This leads to a
404 error thrown by httpd. If I manually remove the slash, everyting
is OK.

Example given:
http://git/cgi-bin/cgit.cgi/linuxadmin.git/ -> 404
http://git/cgi-bin/cgit.cgi/linuxadmin.git -> Page ist shown

Is there a way to tell cgit to give up the trailing slash? I'm running
cgit-1.2.3-r3 on Alpine 3.17.2

Any help greatly appreciated!

BTW: I've also tried lighttp, which had no problem with trailing
slashes, but is more heavyweight than busybox.

Best Regards
Andreas

Re: cgit with busybox httpd

[John Keeping]

On Mon, Mar 06, 2023 at 11:13:10AM +0100, Andreas Mahling wrote:
> I'm in the process to configure a cgit instance for my private network at home.
> Because my internet router will be the host for git and cgit, I'm
> looking for a ligthweight setup.
> I would like to use busybox httpd as webserver, which has a very low
> footprint and supports cgi, but no other fancy bells and whistles,
> especially no url rewrite.
> 
> It works in principle, but there is a problem with the URLs generated
> by cgit: they seem always containing a trailing slash. This leads to a
> 404 error thrown by httpd. If I manually remove the slash, everyting
> is OK.
> 
> Example given:
> http://git/cgi-bin/cgit.cgi/linuxadmin.git/ -> 404
> http://git/cgi-bin/cgit.cgi/linuxadmin.git -> Page ist shown
> 
> Is there a way to tell cgit to give up the trailing slash? I'm running
> cgit-1.2.3-r3 on Alpine 3.17.2

I don't think it's possible to change that behaviour without patching
CGit.  Given how URLs are constructed, it looks a bit complicated to fix
in all cases, but you could remove the setting of "virtual_root" in
cmd_main() to force using query parameters instead of virtual paths in
URLs.

Can you tell where the 404 is generated?  Is this a case of httpd
returning an error when it shouldn't, or is it a behaviour difference
that CGit isn't handling correctly - maybe other CGI hosts strip
trailing slashes before passing a path to CGit but httpd doesn't?

Re: cgit with busybox httpd

[John Keeping]

[Cc: add the mailing list back in]

On Mon, Mar 06, 2023 at 01:01:50PM +0100, Andreas Mahling wrote:
> It seems the 404 is generated by httpd, I think because the url ends with a
> slash httpd treats the part after cgit.cgi not as PATH_INFO (as it should)
> but as a directory. So this seems more a fault of httpd to me.
> 
> Sorry, but I don't understand how to put into QUERY_STRING what now goes
> into PATH_INFO. Do you mean the virtual-root option in cgitrc?
> It is already commented out in my setup, but cgit builds the url with
> PATH_INFO part anyway.

To disable that feature you would have the patch the CGit source and
build your own cgit binary.

There is code in cmd_main() which automatically calculates the virtual
root from other environment variables if they are provided (and it seems
that httpd does provide this detail to CGI scripts).

> Am Mo., 6. März 2023 um 12:41 Uhr schrieb John Keeping <john@keeping.me.uk>:
> 
> > On Mon, Mar 06, 2023 at 11:13:10AM +0100, Andreas Mahling wrote:
> > > I'm in the process to configure a cgit instance for my private network
> > at home.
> > > Because my internet router will be the host for git and cgit, I'm
> > > looking for a ligthweight setup.
> > > I would like to use busybox httpd as webserver, which has a very low
> > > footprint and supports cgi, but no other fancy bells and whistles,
> > > especially no url rewrite.
> > >
> > > It works in principle, but there is a problem with the URLs generated
> > > by cgit: they seem always containing a trailing slash. This leads to a
> > > 404 error thrown by httpd. If I manually remove the slash, everyting
> > > is OK.
> > >
> > > Example given:
> > > http://git/cgi-bin/cgit.cgi/linuxadmin.git/ -> 404
> > > http://git/cgi-bin/cgit.cgi/linuxadmin.git -> Page ist shown
> > >
> > > Is there a way to tell cgit to give up the trailing slash? I'm running
> > > cgit-1.2.3-r3 on Alpine 3.17.2
> >
> > I don't think it's possible to change that behaviour without patching
> > CGit.  Given how URLs are constructed, it looks a bit complicated to fix
> > in all cases, but you could remove the setting of "virtual_root" in
> > cmd_main() to force using query parameters instead of virtual paths in
> > URLs.
> >
> > Can you tell where the 404 is generated?  Is this a case of httpd
> > returning an error when it shouldn't, or is it a behaviour difference
> > that CGit isn't handling correctly - maybe other CGI hosts strip
> > trailing slashes before passing a path to CGit but httpd doesn't?
> >

Re: cgit with busybox httpd

[Andreas Mahling]

I've preferred to patch busybox httpd.
http://lists.busybox.net/pipermail/busybox/2023-March/090197.html

thanks for your help
Andreas
Am Mo., 6. März 2023 um 15:42 Uhr schrieb John Keeping <john@keeping.me.uk>:
>
> [Cc: add the mailing list back in]
>
> On Mon, Mar 06, 2023 at 01:01:50PM +0100, Andreas Mahling wrote:
> > It seems the 404 is generated by httpd, I think because the url ends with a
> > slash httpd treats the part after cgit.cgi not as PATH_INFO (as it should)
> > but as a directory. So this seems more a fault of httpd to me.
> >
> > Sorry, but I don't understand how to put into QUERY_STRING what now goes
> > into PATH_INFO. Do you mean the virtual-root option in cgitrc?
> > It is already commented out in my setup, but cgit builds the url with
> > PATH_INFO part anyway.
>
> To disable that feature you would have the patch the CGit source and
> build your own cgit binary.
>
> There is code in cmd_main() which automatically calculates the virtual
> root from other environment variables if they are provided (and it seems
> that httpd does provide this detail to CGI scripts).
>

cgit with busybox httpd

[Andreas Mahling]

sorry, I've missed sending to the list address

I wrote a script test.sh which prints the environment.
When called as http://git/cgi-bin/test.sh/xyz PATH_INFO ist set to '/xyz'
When called as  http://git/cgi-bin/test.sh/xyz/ 404 is thrown. So the
problem seems to be caused by httpd

hello John,

It seems the 404 is generated by httpd, I think because the url ends
with a slash httpd treats the part after cgit.cgi not as PATH_INFO (as
it should) but as a directory. So this seems more a fault of httpd to
me.

Sorry, but I don't understand how to put into QUERY_STRING what now
goes into PATH_INFO. Do you mean the virtual-root option in cgitrc?
It is already commented out in my setup, but cgit builds the url with
PATH_INFO part anyway.

best regards Andreas



Am Mo., 6. März 2023 um 12:41 Uhr schrieb John Keeping <john@keeping.me.uk>:
>
> On Mon, Mar 06, 2023 at 11:13:10AM +0100, Andreas Mahling wrote:
> > I'm in the process to configure a cgit instance for my private network at home.
> > Because my internet router will be the host for git and cgit, I'm
> > looking for a ligthweight setup.
> > I would like to use busybox httpd as webserver, which has a very low
> > footprint and supports cgi, but no other fancy bells and whistles,
> > especially no url rewrite.
> >
> > It works in principle, but there is a problem with the URLs generated
> > by cgit: they seem always containing a trailing slash. This leads to a
> > 404 error thrown by httpd. If I manually remove the slash, everyting
> > is OK.
> >
> > Example given:
> > http://git/cgi-bin/cgit.cgi/linuxadmin.git/ -> 404
> > http://git/cgi-bin/cgit.cgi/linuxadmin.git -> Page ist shown
> >
> > Is there a way to tell cgit to give up the trailing slash? I'm running
> > cgit-1.2.3-r3 on Alpine 3.17.2
>
> I don't think it's possible to change that behaviour without patching
> CGit.  Given how URLs are constructed, it looks a bit complicated to fix
> in all cases, but you could remove the setting of "virtual_root" in
> cmd_main() to force using query parameters instead of virtual paths in
> URLs.
>
> Can you tell where the 404 is generated?  Is this a case of httpd
> returning an error when it shouldn't, or is it a behaviour difference
> that CGit isn't handling correctly - maybe other CGI hosts strip
> trailing slashes before passing a path to CGit but httpd doesn't?
>