4 ms·

Getaddrinfo sucks. everything else is much worse

34 points by todsacerdoti 2y ago

koakuma-chan 2y ago
Why not a user-space DNS resolver, such as Hickory DNS?
vetrom 2y ago
Doesn't systemd's resolver api hit most of the wishlist items talked about in the slides there? Granted, then you're pulling in dbus, a dbus api, and all that 'fun'.
- kazinator 2y ago
  Before you know it, your docker build environment is pulling in X11, Wayland, QT, TeX, man pages, Postgresql, Apache, Python3, ...
- theamk 2y ago
  > Linux - res_nquery. This function is well documented, thread-safe, and works as expected. The only platform which has systemd resolver api already has a nice working function.
krackers 2y ago
macOS definitely has an async version for getaddrinfo, it's just not public API. CoreFoundation uses it, but what they publicly give you is a connect-by-hostname API that does happy eyeballs, see https://news.ycombinator.com/item?id=41572770 https://news.ycombinator.com/item?id=41572770
kazinator 2y ago
Looks fine from my armchair: This is the TXR Lisp interactive listener of TXR 298. Quit with :quit or Ctrl-D on an empty line. Ctrl-X ? for cheatsheet. 1> (getaddrinfo "news.ycombinator.com") (#S(sockaddr-in canonname nil addr 3520653007 port 0 prefix 32) #S(sockaddr-in canonname nil addr 3520653007 port 0 prefix 32) #S(sockaddr-in6 canonname nil addr 50544109533264435600708011529282256934 port 0 flow-info 0 scope-id 0 prefix 128) #S(sockaddr-in6 canonname nil addr 50544109533264435600708011529282256934 port 0 flow-info 0 scope-id 0 prefix 128)) It would be tidy to deduplicate that, though it is overhead.
- titanomachy 2y ago
  Why is your lisp rendering ipv4 and ipv6 addresses as integers instead of octets...?
  - kazinator 2y ago
    Because that's what they are; that's the representation. Look at the classic struct sockaddr_in structure it has a member sin_addr of type struct in_addr. What is that? In my header files here I have it as: typedef uint32_t in_addr_t; struct in_addr { in_addr_t s_addr; }; The C structure has an integer, so the Lisp analog does also. C socket code does things like addr.sin_addr.s_addr = htonl(INADDR_LOOPBACK); INADDR_LOOPBACK is 127.0.0.1 encoded as an integer. It's easy to to do bitwise arithmetic on it. E.g. to apply a netmask like 255.0.0.0, you literally just form the netmask as 0xFF000000, and use your & operator: addr & 0xFF00000000. In the case of IPv6, there is a union: a view of the address as 16 bytes, 8 words, or 4 double words. Since I have bignum integers, I just made it a simple integer, so it looks like the IPv4 case. You can do bitwise arithmetic on it in a straightforward. We can get the addresses in text format like this: 1> (getaddrinfo "news.ycombinator.com") (#S(sockaddr-in canonname nil addr 3520653007 port 0 prefix 32) #S(sockaddr-in canonname nil addr 3520653007 port 0 prefix 32) #S(sockaddr-in6 canonname nil addr 50544109533264435600708011529282256934 port 0 flow-info 0 scope-id 0 prefix 128) #S(sockaddr-in6 canonname nil addr 50544109533264435600708011529282256934 port 0 flow-info 0 scope-id 0 prefix 128)) 2> (first *1) #S(sockaddr-in canonname nil addr 3520653007 port 0 prefix 32) 3> (third *1) #S(sockaddr-in6 canonname nil addr 50544109533264435600708011529282256934 port 0 flow-info 0 scope-id 0 prefix 128) 4> (str-inaddr *2.addr) "209.216.230.207" 5> (str-in6addr *3.addr) "2606:7100:1:67::26" str-inaddr and str-in6addr* take an integer and return a string. Their rough opposites return a structure, because the general text format can encode prefix and port number info. 6> (inaddr-str *4) #S(sockaddr-in canonname nil addr 3520653007 port 0 prefix 32) 7> (in6addr-str *5) #S(sockaddr-in6 canonname nil addr 50544109533264435600708011529282256934 port 0 flow-info 0 scope-id 0 prefix 128) The structures also support a generic method str-addr: 8> (mapcar .(str-addr) (getaddrinfo "news.ycombinator.com")) ("209.216.230.207" "209.216.230.207" "2606:7100:1:67::26" "2606:7100:1:67::26") This will take into account prefix and port number if they are present and nonzero. We could define a custom print method in the sockaddr base: 9> (defmeth sockaddr print (self stream pretty-p) (put-string `#<@(typeof self) @{self.(str-addr)}>` stream)) (meth sockaddr print) 10> (getaddrinfo "news.ycombinator.com") (#<sockaddr-in 209.216.230.207> #<sockaddr-in 209.216.230.207> #<sockaddr-in6 2606:7100:1:67::26> #<sockaddr-in6 2606:7100:1:67::26>)
shakna 2y ago
I got the macOS hang a few years back (a decade?). On attempting to report it, I just got it closed with "can't reproduce". And attempting to reopen or discuss got permission denied errors. So I kinda doubt we'll ever see that one fixed.
greatgib 2y ago
In my opinion, Firefox is trying to do things that it should not do. That explains issues that we get with weird behaviors sometimes. Like it is not the role of FF to mess with DNS TTL, that should only be the system resolver to decide if the request has to really be done or the cache used for DNS. In the same way, it looks ridiculous to me that in the http3, Firefox will silently target a different port than the one that is requested and displayed...

34 points by todsacerdoti 2y ago

Why not a user-space DNS resolver, such as Hickory DNS?

Doesn't systemd's resolver api hit most of the wishlist items talked about in the slides there? Granted, then you're pulling in dbus, a dbus api, and all that 'fun'.

kazinator 2y ago
Before you know it, your docker build environment is pulling in X11, Wayland, QT, TeX, man pages, Postgresql, Apache, Python3, ...
theamk 2y ago
> Linux - res_nquery. This function is well documented, thread-safe, and works as expected. The only platform which has systemd resolver api already has a nice working function.

krackers 2y ago

macOS definitely has an async version for getaddrinfo, it's just not public API. CoreFoundation uses it, but what they publicly give you is a connect-by-hostname API that does happy eyeballs, see https://news.ycombinator.com/item?id=41572770 https://news.ycombinator.com/item?id=41572770

kazinator 2y ago

Looks fine from my armchair: This is the TXR Lisp interactive listener of TXR 298. Quit with :quit or Ctrl-D on an empty line. Ctrl-X ? for cheatsheet. 1> (getaddrinfo "news.ycombinator.com") (#S(sockaddr-in canonname nil addr 3520653007 port 0 prefix 32) #S(sockaddr-in canonname nil addr 3520653007 port 0 prefix 32) #S(sockaddr-in6 canonname nil addr 50544109533264435600708011529282256934 port 0 flow-info 0 scope-id 0 prefix 128) #S(sockaddr-in6 canonname nil addr 50544109533264435600708011529282256934 port 0 flow-info 0 scope-id 0 prefix 128)) It would be tidy to deduplicate that, though it is overhead.

titanomachy 2y ago
Why is your lisp rendering ipv4 and ipv6 addresses as integers instead of octets...?
- kazinator 2y ago
  Because that's what they are; that's the representation. Look at the classic struct sockaddr_in structure it has a member sin_addr of type struct in_addr. What is that? In my header files here I have it as: typedef uint32_t in_addr_t; struct in_addr { in_addr_t s_addr; }; The C structure has an integer, so the Lisp analog does also. C socket code does things like addr.sin_addr.s_addr = htonl(INADDR_LOOPBACK); INADDR_LOOPBACK is 127.0.0.1 encoded as an integer. It's easy to to do bitwise arithmetic on it. E.g. to apply a netmask like 255.0.0.0, you literally just form the netmask as 0xFF000000, and use your & operator: addr & 0xFF00000000. In the case of IPv6, there is a union: a view of the address as 16 bytes, 8 words, or 4 double words. Since I have bignum integers, I just made it a simple integer, so it looks like the IPv4 case. You can do bitwise arithmetic on it in a straightforward. We can get the addresses in text format like this: 1> (getaddrinfo "news.ycombinator.com") (#S(sockaddr-in canonname nil addr 3520653007 port 0 prefix 32) #S(sockaddr-in canonname nil addr 3520653007 port 0 prefix 32) #S(sockaddr-in6 canonname nil addr 50544109533264435600708011529282256934 port 0 flow-info 0 scope-id 0 prefix 128) #S(sockaddr-in6 canonname nil addr 50544109533264435600708011529282256934 port 0 flow-info 0 scope-id 0 prefix 128)) 2> (first *1) #S(sockaddr-in canonname nil addr 3520653007 port 0 prefix 32) 3> (third *1) #S(sockaddr-in6 canonname nil addr 50544109533264435600708011529282256934 port 0 flow-info 0 scope-id 0 prefix 128) 4> (str-inaddr *2.addr) "209.216.230.207" 5> (str-in6addr *3.addr) "2606:7100:1:67::26" str-inaddr and str-in6addr* take an integer and return a string. Their rough opposites return a structure, because the general text format can encode prefix and port number info. 6> (inaddr-str *4) #S(sockaddr-in canonname nil addr 3520653007 port 0 prefix 32) 7> (in6addr-str *5) #S(sockaddr-in6 canonname nil addr 50544109533264435600708011529282256934 port 0 flow-info 0 scope-id 0 prefix 128) The structures also support a generic method str-addr: 8> (mapcar .(str-addr) (getaddrinfo "news.ycombinator.com")) ("209.216.230.207" "209.216.230.207" "2606:7100:1:67::26" "2606:7100:1:67::26") This will take into account prefix and port number if they are present and nonzero. We could define a custom print method in the sockaddr base: 9> (defmeth sockaddr print (self stream pretty-p) (put-string `#<@(typeof self) @{self.(str-addr)}>` stream)) (meth sockaddr print) 10> (getaddrinfo "news.ycombinator.com") (#<sockaddr-in 209.216.230.207> #<sockaddr-in 209.216.230.207> #<sockaddr-in6 2606:7100:1:67::26> #<sockaddr-in6 2606:7100:1:67::26>)

shakna 2y ago

I got the macOS hang a few years back (a decade?). On attempting to report it, I just got it closed with "can't reproduce". And attempting to reopen or discuss got permission denied errors. So I kinda doubt we'll ever see that one fixed.

greatgib 2y ago

In my opinion, Firefox is trying to do things that it should not do. That explains issues that we get with weird behaviors sometimes. Like it is not the role of FF to mess with DNS TTL, that should only be the system resolver to decide if the request has to really be done or the cache used for DNS. In the same way, it looks ridiculous to me that in the http3, Firefox will silently target a different port than the one that is requested and displayed...