Skip to content

bind should use IP_RECVERR / IPV6_RECVERR to learn about ICMP destination network/host unreachable

Description

bind uses a connect()ed UDP socket when sending recursive queries. The connect()ed socket provides feedback on a variety of ICMP errors (eg port unreachable) which bind can then use to decide what to do with errors (report them to the client, try again with a different nameserver etc).

However, Linux's implementation does not report what it considers "transient" conditions, which is defined as Destination host Unreachable, Destination network unreachable, Source Route Failed and Message Too Big.

stracing and tcpdumping named shows:

; Setting up the socket: 
[pid  7339] 23:06:03.324158 socket(AF_INET, SOCK_DGRAM, IPPROTO_UDP) = 6 <0.000027>
[pid  7339] 23:06:03.324240 fcntl(6, F_DUPFD, 512) = 523 <0.000032>
[pid  7339] 23:06:03.324333 close(6) = 0 <0.000024>
[pid  7339] 23:06:03.324405 fcntl(523, F_GETFL) = 0x2 (flags O_RDWR) <0.000018>
[pid  7339] 23:06:03.324464 fcntl(523, F_SETFL, O_RDWR|O_NONBLOCK) = 0 <0.000018>
[pid  7339] 23:06:03.324520 setsockopt(523, SOL_SOCKET, SO_TIMESTAMP, [1], 4) = 0 <0.000020>
[pid  7339] 23:06:03.324582 setsockopt(523, SOL_IP, IP_MTU_DISCOVER, [5], 4) = 0 <0.000020>
[pid  7339] 23:06:03.324641 getsockopt(523, SOL_SOCKET, SO_RCVBUF, [212992], [4]) = 0 <0.000019>
[pid  7339] 23:06:03.324702 setsockopt(523, SOL_IP, IP_RECVTOS, [1], 4) = 0 <0.000021>
[pid  7339] 23:06:03.324756 bind(523, {sa_family=AF_INET, sin_port=htons(54746), sin_addr=inet_addr("0.0.0.0")}, 16) = 0 <0.000013>
[pid  7339] 23:06:03.324804 recvmsg(523, {msg_namelen=128}, 0) = -1 EAGAIN (Resource temporarily unavailable) <0.000010>
[pid  7339] 23:06:03.324950 connect(523, {sa_family=AF_INET, sin_port=htons(53), sin_addr=inet_addr("192.112.36.4")}, 16) = 0 <0.000023>
[pid  7341] 23:06:03.324972 epoll_ctl(8, EPOLL_CTL_ADD, 523, {EPOLLIN, {u32=523, u64=523}}) = 0 <0.000018>

; Now, send the packet:

[pid  7339] 23:06:03.325017 sendmsg(523, {msg_name=NULL, msg_namelen=0, msg_iov=[{iov_base="\27\201\0\0\0\1\0\0\0\0\0\0\3www\7example\3com\0\0\1\0"..., iov_len=33}], msg_iovlen=1, msg_controllen=0, msg_flags=0}, 0) = 33 <0.000066>

; And here's the packets, as seen from tcpdump:

23:06:03.325049 IP 192.168.0.1.54746 > 192.112.36.4.53: 6017 A? www.example.com. (33)
23:06:03.325081 IP 192.168.0.254 > 192.168.0.1: ICMP net 192.112.36.4 unreachable, length 69

; See where the packet was sent from...

[pid  7339] 23:06:03.325155 getsockname(523, {sa_family=AF_INET, sin_port=htons(36615), sin_addr=inet_addr("192.168.0.1")}, [128->16]) = 0 <0.000028>

; Now we wait ...

; And then we time out and give up

[pid  7341] 23:06:03.854391 epoll_ctl(8, EPOLL_CTL_DEL, 523, 0x7ff0a8bcbdd8) = 0 <0.000024>
[pid  7341] 23:06:03.854460 epoll_ctl(8, EPOLL_CTL_DEL, 523, 0x7ff0a8bcbdd8) = -1 ENOENT (No such file or directory) <0.000022>
[pid  7341] 23:06:03.854534 close(523) = 0 <0.000023>

Request

bind should set setsockopt(fd, SOL_IP, IP_RECVERR, &one, sizeof(one)); and receive a EPOLL_ERR, and recvmsg return value so that it notices that the send has failed immediately rather than waiting for a timeout.

This would let bind react much much more quickly to network events.

Links / references

To upload designs, you'll need to enable LFS and have an admin enable hashed storage. More information