bug in host-to-vector-quad on lispworks (also host-to-vector-quad)

[Mark H. David]

Hi,
Kind of asking for a friend. I thought the thing to do would be to loop through the list and choose the first non-ipv6 result whereas now it simply chooses the first result.  For the random host case, it should similarly filter out ipv6 entries in the result.
Thank you very much!
-Mark

----- Original message -----
From: "Chun Tian (binghe)" <binghe.lisp at gmail.com>
To: "Mark H. David" <mhd at yv.org>
Cc: "usocket-devel" <usocket-devel at common-lisp.net>
Subject: Re: bug in host-to-vector-quad on lispworks (also host-to-vector-quad)
Date: Wed, 4 Jul 2018 22:47:04 +0200

Hi Mark,

thanks for reporting this issue.

It’s obvious that #(0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1) means IPv6 address “::1”, which is indeed a valid address mapping to “localhost”. The function HOST-TO-HBO is not an exported function, it’s used internally by some backends and deals only with integers representing IPv4 addresses. But on LispWorks (>=6.1) the function lw-hbo-to-vector-quad may pass longer vectors back to HOST-TO-HBO and then caused this issue.

On my Mac OS X laptop, I also got the same: (the first returned address is “::1”)

CL-USER 5 > (comm:get-host-entry "localhost" :fields '(:addresses))
(#<COMM:IPV6-ADDRESS ::1 4020187E73> 2130706433)

So what’s your desired behavior here?  In the quick fixes, I could do something to make get-random-host-by-name return a random IPv4 address in presence of mixed IPv4/v6 address lists. But generally speaking get-hosts-by-name should support IPv6 addresses, especially for those IPv6-only hostnames.

On the other side, HOST-TO-HBO is not an exported function, and it’s also not called by LispWorks backends, I don’t understand how your call chain finally touched it and let LispWorks throw an exception.  It’s hard to support IPv6 in HOST-TO-HBO.

Let me know what you think here, and with your help/inputs we can make usocket better.

Regards,

Chun

> Il giorno 04 lug 2018, alle ore 21:04, Mark H. David <mhd at yv.org> ha scritto:
> 
> Running usocket-0.7.0.1 in LispWorks 7.1.0. In my system it happens currently that
> 
> (get-hosts-by-name "localhost")
> 
> returns the following list:
> 
> (#(0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1) #(127 0 0 1))
> 
> Then get-host-by-name simply takes the first of that list and returns it.
> 
> Host-to-hbo expects the result of get-host-by-name to always be a vector quad, so the result of calling it with "localhost" is an error on the recursive call with the result of (get-host-by-name "localhost") => #(0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1)
> 
> The error message is due to falling out of the ecase:
> 
>  #(0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1) fell through ETYPECASE expression inside USOCKET::HOST-TO-HBO.
> 
> I observed that host-to-vector-quad gives the wrong result, a 16-byte instead of a 4-byte vector, due to the same root cause, but it will happen "randomly". The reason is it uses get-random-host-by-name, which chooses a random element of the same result list returned by (get-hosts-by-name "localhost").  For example, I just ran this in the REPL -- log:
> 
>  USOCKET> (loop repeat 10 for r = (host-to-vector-quad "localhost") when (not (= (length r) 4)) collect r)
>  (#(0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1) #(0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1) #(0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1))
> 
> This seems to be LispWorks only. I tried this on SBCL and CCL, but I could not recreate similar problems. Also, not all MacOS installs seem to get this weird ipv6 localhost the way I do. I'm not sure why I'm getting this. I do not remember doing anything special or any kind of stuff with ipv6, but apparently something I've done has "infected" me with it.  However, I can say I do not notice other networking issues on my system.
> 
> Thanks,
> Mark
> 

Email had 1 attachment:
+ signature.asc
  1k (application/pgp-signature)