Hi Mark, I’ve made the changes in GET-HOST-BY-NAME and GET-RANDOM-HOST-BY-NAME on usocket trunk (master) [1]. Now they only returns IPv4 addresses if the underlying hostname resolves to both IPv4 and IPv6 addresses. I think this change is safe for all users.
I also exported these functions as I saw many users directly use them as exported symbols. This is reasonable too, as GET-RANDOM-HOST-BY-NAME actually gives the ability to do load balancing. Any way, could you please confirm if the current code base works for you? —Chun [1] https://github.com/usocket/usocket <https://github.com/usocket/usocket>, a1f32274686d3749ce491dd768c9d0da1a4479c4 > Il giorno 05 lug 2018, alle ore 20:13, Mark H. David <m...@yv.org> ha scritto: > > Hi, that sounds very good to me. Thanks a lot. -Mark > > ----- Original message ----- > From: "Chun Tian (binghe)" <binghe.l...@gmail.com> > To: "Mark H. David" <m...@yv.org> > Cc: usocket development <usocket-devel@common-lisp.net> > Subject: Re: bug in host-to-vector-quad on lispworks (also > host-to-vector-quad) > Date: Thu, 5 Jul 2018 14:18:54 +0200 > > Hi Mark, > > Now I see your points. So IPv4 addresses should have a higher priority over > IPv6 when returned by USOCKET host resolving functions. I’m a little busy in > this week, *next* week I’ll revise all related USOCKET functions and make the > changes (also need a better way to detect IPv6 addresses in various forms). > Hopefully at the end of July I can make a new release with cumulative updates > in the past... more than one year. > > —Chun > >> Il giorno 04 lug 2018, alle ore 23:44, Mark H. David <m...@yv.org> ha >> scritto: >> >> 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.l...@gmail.com> >> To: "Mark H. David" <m...@yv.org> >> Cc: "usocket-devel" <usocket-devel@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 <m...@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) > > Email had 1 attachment: > + signature.asc > 1k (application/pgp-signature)
signature.asc
Description: Message signed with OpenPGP using GPGMail
