I'm interested to know, how does WSAIoctl() with the SIO_ROUTING_INTERFACE_QUERY control code create the list of IP addresses of my host machine? In particular, what criteria does it use to order the IP addresses?
It only returns one! From MSDN (emphasis mine):
SIO_ROUTING_INTERFACE_QUERY (opcode setting: I, O, T==1)
To obtain the address of the local interface (represented as sockaddr structure) that should be used to send to the remote address specified in the input buffer ...
It's true that multiple routes to the destination address might exist, in which case it will no doubt pick the cheapest (routing table entries each contain a cost, or metric, see here).
Or did you mean SIO_ADDRESS_LIST_QUERY?, in which case Windows knows full well what network interfaces you have installed on your machine and the order in which they are returned is supremely unimportant.
Related
I'd like to store ip2location database in a postgres database.
Guide on ip2location website suggests to store IP addresses that specify the range (ip_from and ip_to) in columns of type DECIMAL(39, 0).
I would have expected this to be of type INET.
Is there any advantage (e.g. in terms of speed, size, ...) to use DECIMAL(39, 0) instead of INET? Other than the ip2location database format contains IPs converted to integers already and one would have to convert that back to IP addresses obviously.
The advantage of using the inet data type are
it takes up only 4 bytes to store an IPv4 address
you can make use of all the useful built-in functions and operators for IP addresses and CIDRs
There is no advantage in using NUMERIC(39), except that it seems to be required by the software you want to use.
Say I've got a file descriptor from socket(2) and I've done a connect(2) on it -- is there any way later to determine (from inside the running program) what network device might be in use for the underlying transport? A call to stat(2) on the fd gives a device number of 0; none of the ioctl(2) or getsockopt(2) options seem applicable.
There's no foolproof way to do so -- certainly not a posix-compliant way.
However, in practice, you can easily determine the interface 99% of the time. After you've done the connect, use getsockname to obtain the IP address, then look through the list of available interfaces on the box for one with a matching IP address.
From the accept call you should be able to get the remote client's ip address (seen here on Beej's). Assuming you do not have any asymmetric routes, you can look up the route to that address in your local routing table. The routing table should tell you what ethernet device is used to send packets to the remote client.
EDIT:
you could use the following command line tool to query your local routing table with the remote client's address:
ip route get <remote-client-ip-addr>
I have a PostgreSQL table with a column (of type Text) 'ip_port'. This column has values like:
167.123.24.25:59874
This is IP:Port
And I'm trying to write a query to list all the subnets from those IP and a count of IPs in each subnet. Is there an elegant way of doing this in Postgres?
Thanks in advance!
You cannot do this - in PostgreSQL or any other tool - if the only information you have is the IP address. You must also know:
The broadcast address for the network;
The network subnet mask or mask prefix length; or
that pre-CIDR class-based address allocation is being used
Consider that a given IP is actually a member of multiple networks of increasing size. For the purpose of local area Ethernet-based IP networking there's a broadcast domain defined by the subnet mask, but there are also a nested set of increasingly broad routing aggregation domains for that address, some of which are visible on the public Internet and some of which are private to an internal network. The line between "public" and "private" can be blurry; for example, within my ISP's network their large network blocks are split into smaller chunks that are not visible from the outside world but are visible to the ISP's customers.
For example, a non-exhaustive list of networks for 167.123.24.25 might be, according to some quick calculations with the sipcalc tool:
167.123.24.25/32 (the host its self)
167.123.24.25/28: (a common ISP allocation): usable 167.123.24.17 - 167.123.24.30, netmask 255.255.255.240
167.123.24.25/24: (a common LAN subnet size): usable 167.123.24.1 - 167.123.24.254, netmask 255.255.255.0
167.123.24.25/20: (a midlevel routing aggregation): usable 167.123.16.1 - 167.123.31.254, netmask 255.255.240.0
167.123.24.25/16: The top-level APNIC allocation of this address block according to whois): usable 167.123.0.1 - 167.123.255.254, netmask 255.255.0.0
If you traceroute that address, consider that any of those routers might (but also might not) be the point at which the address becomes part of a narrower, more specific subnet. See Classless inter-domain routing on Wikipedia.
All that is before you even consider NAT (blech) and internal WAN routing. 167.123.24.25 might be a router for a whole hidden network of NATed hosts. These hosts are invisible to you from the outside without extremely sophisticated passive mapping that only someone close to them along their routing path can do.
To learn more about the address's network you have to do some serious probing to find out what addresses might be treated as broadcast addresses. Of course, most routers will filter packets destined for broadcast addresses from outside the local routing domain, so you might not be able to tell the difference between "host doesn't exist", "is a broadcast address" and "host exists but filters the packet I'm using for probing". Such probing is slow, and it'll also tend to make the network owners grumpy at you. Since this network is owned by the US Department of Public Works, I'd recommend not making them grumpy.
http://publib.boulder.ibm.com/infocenter/aix/v7r1/index.jsp?topic=%2Fcom.ibm.aix.progcomm%2Fdoc%2Fprogcomc%2Fskt_bind.htm
Internet domain:
Produces an association composed of local and foreign addresses and local and foreign ports.
UNIX domain:
Produces an association composed of local and foreign path names.
NDD domain (Network Device Driver of the operating system):
Provides an association composed of local device name (operating system NDD
name) and foreign addresses, the form of which depends on the protocol
being used.
They talk about association composed of local and foreign addresses.
What exactly does that mean.
I can understand the local address, but what's with the foreigh addresses, how does it find them, and which are they, and how does it create their association?
The local and foreign socket addresses are simply the address of your program (the local one) and the address of the other program you are communicating with (the foreign one).
Different programs use different mechanisms to determine that foreign address. For example:
To view a Web page, you type in a desired URL or click on a link; the URL contains a name, and the address associated with the name is looked up and connected to in order to retrieve the resource.
To look up an address from a name, your computer will contact a name server at a different designated address. The address of the name server is itself often obtained from another protocol, DHCP.
To locate a DHCP server, your computer will send a broadcast message to all machines on the local network, then wait for a server to reply.
Associations can also vary depending upon whether a protocol is connection-based or connectionless.
In a connection-based protocol such as TCP, this association is called a connection; messages are exchanged between the network subsystems of the two machines to negotiate the set-up of the connection, and this connection persists until explicitly closed.
In a connectionless protocol such as UDP, a simple facility to send and receive individual messages to or from any addressing-compatible endpoint is provided; any associations are entirely at the convenience of the software using the protocol.
Below is a sample invoking of getaddrinfo()
status = getaddrinfo("www.example.net","1234", &hints, &server_info);
After that, the server_info will point to a linked list with all kinds of address information.
I have the following questions:
Since I have clearly specified the host name and port number, the only address infos I can think of are IPv4 and IPv6 addresses. So, the length of the linked list should be 2. Is there any other kind of address info besides them?
Thanks.
The name could resolve to more than one IPv4 or IPv6 address, there is nothing to say that only one IPv4 address will be returned, for example (try it with "www.google.com" for example, you'll likely get more than one IPv4 address).
But in any case, I think the basic premise of your question is wrong. Even if there was no possibility to return more than one IPv4 and one IPv6 address today, the function is documented to return an arbitrarily-long linked list of addrinfo objects. Therefore, even if your code worked today in every situation, there is no guarantee that it would continue to work tomorrow. If the function is documented to return an arbitrarily-long linked list, then you need to be able to handle that.
You want to disconnect the physical configuration of machines with names in your mind. DNS merely maps a name to a set of addresses. A lot of hosts will only have one interface. Many hosts will have multiple (called multi-homing). DNS doesn't care about what the configuration of the machine or machines that the addresses it maps a name to is. As simple examples often a server will have interfaces on multiple networks with different addresses that will all map to the same name. Sometimes when collapsing services from different machines onto one different names will map to the same address. So don't assume any sort of 1:1 mapping between names and machines much less interfaces.