这个数据报套接字有什么问题？

Question

I have some code here which queries the Steam master servers to get a list of IPs of game servers:

#define _BSD_SOURCE
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <stdint.h>
#include <errno.h>
#include <unistd.h>
#include <getopt.h>
#include <sys/types.h>
#include <sys/select.h>
#include <sys/socket.h>
#include <arpa/inet.h>
#include <netinet/in.h>
#include <netdb.h>

struct timeval timeout = {1, 0};
char master[256];
char reply[1500];
uint16_t port;
uint8_t query[] = {0x31, 0x03, 0x30, 0x2e, 0x30, 0x2e, 0x30, 0x2e, 0x30, 0x2e,
                         0x30, 0x3a, 0x30, 0x00, 0x00};
uint8_t replyHeader[] = {0xff, 0xff, 0xff, 0xff, 0x66, 0x0a};
int gotResponse = 0;
int bytesRead = 0;
int verbosity = 0;

int main(int argc, char** argv)
{
    strcpy(master, "hl2master.steampowered.com");
    port = 27011;

    int opt;

    while ((opt = getopt(argc, argv, "s:p:v")) != -1)
    {
        switch (opt)
        {
            case 's':
                strcpy(master, optarg);
                break;
            case 'p':
                port = atoi(optarg);
                break;
            case 'v':
                verbosity++;
                break;
        }
    }

    int sockFD;
    struct sockaddr_in server;
    struct hostent* hostInfo;

    sockFD = socket(PF_INET, SOCK_DGRAM, IPPROTO_UDP);

    if (sockFD == -1)
    {
        perror("socket");
        exit(EXIT_FAILURE);
    }

    if ((setsockopt(sockFD, SOL_SOCKET, SO_RCVTIMEO, &timeout, sizeof(timeout)))
         != 0)
    {
        perror("setsockopt");
        exit(EXIT_FAILURE);
    }

    if ((setsockopt(sockFD, SOL_SOCKET, SO_SNDTIMEO, &timeout, sizeof(timeout)))
         != 0)
    {
        perror("setsockopt");
        exit(EXIT_FAILURE);
    }

    server.sin_family = AF_INET;
    server.sin_port = htons(port);
    hostInfo = gethostbyname(master);

    if (hostInfo == NULL)
    {
        fprintf(stderr, "Unknown host %s\n", master);
        exit(EXIT_FAILURE);
    }

    server.sin_addr = *(struct in_addr*) hostInfo->h_addr_list[0];

    while (gotResponse == 0)
    {
        if ((sendto(sockFD, query, 15, 0, (struct sockaddr*) &server,
                        sizeof(server))) == -1)
        {
            perror("sendto");
            exit(EXIT_FAILURE);
        }

        socklen_t serverSize = sizeof(server);

        if ((bytesRead = recvfrom(sockFD, reply, 1500, 0,
                                          (struct sockaddr*) &server, &serverSize)) == -1)
        {
            if ((errno == EAGAIN) || (errno == EWOULDBLOCK))
            {
                fprintf(stderr, "TIMEOUT\n");
            }
            else
            {
                perror("recvfrom");
                exit(EXIT_FAILURE);
            }
        }
        else
            gotResponse = 1;
    }

    if ((close(sockFD)) == -1)
    {
        perror("close");
        exit(EXIT_FAILURE);
    }

    if ((strncmp(reply, replyHeader, 6)) != 0)
    {
        fprintf(stderr, "Bad reply from master server\n");
        exit(EXIT_FAILURE);
    }

    uint32_t i = 6;

    while (i < bytesRead)
    {
        if (verbosity > 0)
            fprintf(stderr, "%u <= %d\n", i, bytesRead);

        uint8_t ip[4] = {reply[i], reply[i + 1], reply[i + 2], reply[i + 3]};

        printf("%hu.%hu.%hu.%hu:", ip[0], ip[1], ip[2], ip[3]);

        uint16_t thisPort = reply[i + 4] + (reply[i + 5] << 8);

        printf("%hu\n", ntohs(thisPort));

        i += 6;
    }

    return EXIT_SUCCESS;
}

(Note the one second timeout .)

It is fine other than some odd behaviour with the communication. It seems to either work first time or continually timeout again, and again, and never succeed.

The way to fix is to simply run it again and it may work, but I don't understand the reason for it to arbitrarily not work.

Any input would be appreciated!

Answer 1

Just a wild guess: maybe recvfrom is mangling the server argument, and the following sends have an incorrect address? Try to pass a separate struct sockaddr.

strace output could help.

Answer 2

Since you posted the entire runnable code, I tried running it with strace as cdleonard suggested. That immediately show that the gethostbyname call for for hl2master.steampowered.com returns one of two different addresses each time the program is run -- either 63.234.149.83 or 72.165.61.153 -- and when the first address is returned it works fine, whereas the second address fails with the timeouts.

So the issue seems to be that there's two address for the server on DNS but one of them doesn't really work.

I would suggest checking through the h_addr_list returned by gethostbyaddr and going through each address in turn in your loop, rather than just always sending to the first address.

Answer 3

The host hl2master.steampowered.com resolves to three IP addresses:

syzdek@blackenhawk$ dig +short hl2master.steampowered.com
63.234.149.83
63.234.149.90
72.165.61.153
syzdek@blackenhawk$

Two of the three IP address are responding to queries, the third is not:

syzdek@blackenhawk$ ./a.out -s 63.234.149.83 |head -2
66.189.187.173:27012
216.6.229.173:27015
syzdek@blackenhawk$ ./a.out -s 63.234.149.90 |head -2
66.189.187.173:27012
216.6.229.173:27015
syzdek@blackenhawk$ ./a.out -s 72.165.61.153
recvfrom: TIMEOUT: Resource temporarily unavailable
recvfrom: TIMEOUT: Resource temporarily unavailable
^C
syzdek@blackenhawk$

Small note, I changed fprintf(stderr, "TIMEOUT\\n"); to perror("recvfrom: TIMEOUT"); in the course of trying your code.

Maybe try a using a different server after the timeout:

int retryCount = 0;
while (gotResponse == 0)
{
    // verify that next address exists
    if (hostInfo->h_addr_list[retryCount/2] == NULL)
    {
        fprintf(stderr, "All servers are not responding.");
        exit(EXIT_FAILURE);
    };

    // Attempt each address twice before moving to next IP address
    server.sin_addr = *(struct in_addr*) hostInfo->h_addr_list[retryCount/2];
    retryCount++;

    if ((sendto(sockFD, query, 15, 0, (struct sockaddr*) &server,
                            sizeof(server))) == -1)
    {
        perror("sendto");
        exit(EXIT_FAILURE);
    }

    / * rest of code */

The above edits will try each address returned by gethostbyame() twice before moving to the next returned IP address.

Answer 4

In case you are getting EAGAIN from your call to recvfrom() that is 'Resouce temporarily unavailable' you may try to re-call gethostbyname() , just in case the DNS might give you a different IP for that hostname you are passing (in a round-robin way).

If this is the case and at least one of the addresses returned by the DNS is not reachable you will have exactly the behavior you are facing.

Just as a note: gethostbyname() may return static data, so it's a good idea to copy the result and not just reference it.

Answer 5

Instead of gethostbyname() , which returns only one IP address, you could try with getaddrinfo() , which gives you all of them, and even with all supported address families.

Here it is how it works:

    int sockFD;
    struct hostent* hostInfo;

    struct addrinfo hints = {
        .ai_socktype = SOCK_DGRAM,
        .ai_protocol = IPPROTO_UDP /* MHO redundant*/
    };
    struct addrinfo * ai_chain, *ai;

    int gai_ret = getaddrinfo(master, NULL, &hints, &ai_chain);
    if (gai_ret != 0) {
        fprintf(stderr, "getaddrinfo: %s", gai_strerror(gai_ret));
        exit(EXIT_FAILURE);
    }

    for (ai = ai_chain; ai; ai = ai->ai_next) {
        printf("try %s\n", ai->ai_canonname ? ai->ai_canonname : "");
        sockFD = socket(ai->ai_family, ai->ai_socktype, ai->ai_protocol);
        if (sockFD == -1)
        {
            perror("socket");
            continue;
        }

        if ((setsockopt(sockFD, SOL_SOCKET, SO_RCVTIMEO, &timeout, sizeof(timeout))) != 0)
        {
            perror("setsockopt 1");
            continue;
        }

        if ((setsockopt(sockFD, SOL_SOCKET, SO_SNDTIMEO, &timeout,
    sizeof(timeout)))
            != 0)
        {
            perror("setsockopt 2");
            continue;
        }

        if ((sendto(sockFD, query, 15, 0, ai->ai_addr, ai->ai_addrlen)) == -1)
        {
            perror("sendto");
            continue;
        }

        struct sockaddr_in6 server;
        socklen_t serverSize = sizeof(server);

        if ((bytesRead = recvfrom(sockFD, reply, 1500, 0,
                                        (struct sockaddr*) &server,
&serverSize)) == -1)
        {
            if ((errno == EAGAIN) || (errno == EWOULDBLOCK))
            {
                fprintf(stderr, "TIMEOUT\n");
                continue;
            }
            else
            {
                perror("recvfrom");
                continue;
            }
        }
        else
            break;
    }

    if ((close(sockFD)) == -1)
    {
        perror("close");
        exit(EXIT_FAILURE);
    }

    if ((strncmp(reply, replyHeader, 6)) != 0)
    {
        fprintf(stderr, "Bad reply from master server\n");
        exit(EXIT_FAILURE);
    }

    uint32_t i = 6;

    while (i < bytesRead)
    {
        if (verbosity > 0)
            fprintf(stderr, "%u <= %d\n", i, bytesRead);

        uint8_t ip[4] = {reply[i], reply[i + 1], reply[i + 2], reply[i + 3]};

        printf("%hu.%hu.%hu.%hu:", ip[0], ip[1], ip[2], ip[3]);

        uint16_t thisPort = reply[i + 4] + (reply[i + 5] << 8);

        printf("%hu\n", ntohs(thisPort));

        i += 6;
    }

    return EXIT_SUCCESS;
}

While the code still has the one or other flaw (till now, it never closes the sockets; buffer overflow vulnerability in master due to strcpy() instead of strncpy() ), it shows you how things work.