Newer
Older
/* Copyright 1998 by the Massachusetts Institute of Technology.
*
* Permission to use, copy, modify, and distribute this
* software and its documentation for any purpose and without
* fee is hereby granted, provided that the above copyright
* notice appear in all copies and that both that copyright
* notice and this permission notice appear in supporting
* documentation, and that the name of M.I.T. not be used in
* advertising or publicity pertaining to distribution of the
* software without specific, written prior permission.
* M.I.T. makes no representations about the suitability of
* this software for any purpose. It is provided "as is"
* without express or implied warranty.
*/
#include "setup.h"
#if defined(WIN32) && !defined(WATT32)
#else
#include <sys/socket.h>
#ifdef HAVE_SYS_UIO_H
#include <netinet/in.h>
#include <netdb.h>
#include <arpa/nameser.h>
Daniel Stenberg
committed
#ifdef HAVE_ARPA_NAMESER_COMPAT_H
#include <arpa/nameser_compat.h>
#endif
#endif /* WIN32 && !WATT32 */
#ifdef HAVE_UNISTD_H
#include <unistd.h>
#endif
#ifdef HAVE_SYS_IOCTL_H
#include <sys/ioctl.h>
#endif
#ifdef NETWARE
#include <sys/filio.h>
#endif
#include <string.h>
#include <stdlib.h>
#include <fcntl.h>
#include <time.h>
#include <errno.h>
#include "ares.h"
#include "ares_dns.h"
#include "ares_private.h"
William Ahern
committed
static int try_again(int errnum);
static void write_tcp_data(ares_channel channel, fd_set *write_fds,
ares_socket_t write_fd, time_t now);
static void read_tcp_data(ares_channel channel, fd_set *read_fds,
ares_socket_t read_fd, time_t now);
static void read_udp_packets(ares_channel channel, fd_set *read_fds,
ares_socket_t read_fd, time_t now);
static void process_timeouts(ares_channel channel, time_t now);
static void process_answer(ares_channel channel, unsigned char *abuf,
int alen, int whichserver, int tcp, time_t now);
static void handle_error(ares_channel channel, int whichserver, time_t now);
static struct query *next_server(ares_channel channel, struct query *query, time_t now);
static int open_tcp_socket(ares_channel channel, struct server_state *server);
static int open_udp_socket(ares_channel channel, struct server_state *server);
static int same_questions(const unsigned char *qbuf, int qlen,
const unsigned char *abuf, int alen);
static struct query *end_query(ares_channel channel, struct query *query, int status,
unsigned char *abuf, int alen);
/* Something interesting happened on the wire, or there was a timeout.
* See what's up and respond accordingly.
*/
void ares_process(ares_channel channel, fd_set *read_fds, fd_set *write_fds)
{
time_t now;
time(&now);
write_tcp_data(channel, write_fds, ARES_SOCKET_BAD, now);
read_tcp_data(channel, read_fds, ARES_SOCKET_BAD, now);
read_udp_packets(channel, read_fds, ARES_SOCKET_BAD, now);
process_timeouts(channel, now);
}
/* Something interesting happened on the wire, or there was a timeout.
* See what's up and respond accordingly.
*/
void ares_process_fd(ares_channel channel,
ares_socket_t read_fd, /* use ARES_SOCKET_BAD or valid
file descriptors */
ares_socket_t write_fd)
{
time_t now;
time(&now);
write_tcp_data(channel, NULL, write_fd, now);
read_tcp_data(channel, NULL, read_fd, now);
read_udp_packets(channel, NULL, read_fd, now);
process_timeouts(channel, now);
}
/* Return 1 if the specified error number describes a readiness error, or 0
William Ahern
committed
* otherwise. This is mostly for HP-UX, which could return EAGAIN or
* EWOULDBLOCK. See this man page
*
Daniel Stenberg
committed
* http://devrsrc1.external.hp.com/STKS/cgi-bin/man2html?manpage=/usr/share/man/man2.Z/send.2
William Ahern
committed
*/
static int try_again(int errnum)
{
#if !defined EWOULDBLOCK && !defined EAGAIN
#error "Neither EWOULDBLOCK nor EAGAIN defined"
#endif
switch (errnum)
{
#ifdef EWOULDBLOCK
case EWOULDBLOCK:
return 1;
#endif
#if defined EAGAIN && EAGAIN != EWOULDBLOCK
case EAGAIN:
return 1;
#endif
}
return 0;
}
/* If any TCP sockets select true for writing, write out queued data
* we have for them.
*/
static void write_tcp_data(ares_channel channel,
fd_set *write_fds,
ares_socket_t write_fd,
time_t now)
{
struct server_state *server;
struct send_request *sendreq;
struct iovec *vec;
int i;
size_t n;
Daniel Stenberg
committed
if(!write_fds && (write_fd == ARES_SOCKET_BAD))
/* no possible action */
return;
for (i = 0; i < channel->nservers; i++)
{
/* Make sure server has data to send and is selected in write_fds or
write_fd. */
server = &channel->servers[i];
if (!server->qhead || server->tcp_socket == ARES_SOCKET_BAD)
if(write_fds) {
if(!FD_ISSET(server->tcp_socket, write_fds))
continue;
}
else {
if(server->tcp_socket != write_fd)
continue;
}
Daniel Stenberg
committed
if(write_fds)
/* If there's an error and we close this socket, then open
* another with the same fd to talk to another server, then we
* don't want to think that it was the new socket that was
* ready. This is not disastrous, but is likely to result in
* extra system calls and confusion. */
FD_CLR(server->tcp_socket, write_fds);
/* Count the number of send queue items. */
n = 0;
for (sendreq = server->qhead; sendreq; sendreq = sendreq->next)
/* Allocate iovecs so we can send all our data at once. */
vec = malloc(n * sizeof(struct iovec));
if (vec)
{
/* Fill in the iovecs and send. */
n = 0;
for (sendreq = server->qhead; sendreq; sendreq = sendreq->next)
{
vec[n].iov_base = (char *) sendreq->data;
vec[n].iov_len = sendreq->len;
n++;
}
wcount = (ssize_t)writev(server->tcp_socket, vec, (int)n);
free(vec);
if (wcount < 0)
{
if (!try_again(SOCKERRNO))
William Ahern
committed
handle_error(channel, i, now);
continue;
}
/* Advance the send queue by as many bytes as we sent. */
while (wcount)
{
sendreq = server->qhead;
if ((size_t)wcount >= sendreq->len)
{
wcount -= sendreq->len;
server->qhead = sendreq->next;
if (server->qhead == NULL)
Daniel Stenberg
committed
{
SOCK_STATE_CALLBACK(channel, server->tcp_socket, 1, 0);
server->qtail = NULL;
}
free(sendreq);
}
else
{
sendreq->data += wcount;
sendreq->len -= wcount;
break;
}
}
}
{
/* Can't allocate iovecs; just send the first request. */
sendreq = server->qhead;
scount = swrite(server->tcp_socket, sendreq->data, sendreq->len);
if (scount < 0)
{
if (!try_again(SOCKERRNO))
William Ahern
committed
handle_error(channel, i, now);
continue;
}
/* Advance the send queue by as many bytes as we sent. */
if ((size_t)scount == sendreq->len)
{
server->qhead = sendreq->next;
if (server->qhead == NULL)
Daniel Stenberg
committed
{
SOCK_STATE_CALLBACK(channel, server->tcp_socket, 1, 0);
server->qtail = NULL;
}
free(sendreq);
}
else
{
sendreq->data += scount;
sendreq->len -= scount;
}
}
}
}
/* If any TCP socket selects true for reading, read some data,
* allocate a buffer if we finish reading the length word, and process
* a packet if we finish reading one.
*/
static void read_tcp_data(ares_channel channel, fd_set *read_fds,
ares_socket_t read_fd, time_t now)
{
struct server_state *server;
Daniel Stenberg
committed
if(!read_fds && (read_fd == ARES_SOCKET_BAD))
/* no possible action */
return;
for (i = 0; i < channel->nservers; i++)
{
/* Make sure the server has a socket and is selected in read_fds. */
server = &channel->servers[i];
if (server->tcp_socket == ARES_SOCKET_BAD)
if(read_fds) {
if(!FD_ISSET(server->tcp_socket, read_fds))
continue;
}
else {
if(server->tcp_socket != read_fd)
continue;
}
Daniel Stenberg
committed
if(read_fds)
/* If there's an error and we close this socket, then open
* another with the same fd to talk to another server, then we
* don't want to think that it was the new socket that was
* ready. This is not disastrous, but is likely to result in
* extra system calls and confusion. */
FD_CLR(server->tcp_socket, read_fds);
if (server->tcp_lenbuf_pos != 2)
{
/* We haven't yet read a length word, so read that (or
* what's left to read of it).
*/
count = sread(server->tcp_socket,
server->tcp_lenbuf + server->tcp_lenbuf_pos,
2 - server->tcp_lenbuf_pos);
if (count <= 0)
{
if (!(count == -1 && try_again(SOCKERRNO)))
William Ahern
committed
handle_error(channel, i, now);
if (server->tcp_lenbuf_pos == 2)
{
/* We finished reading the length word. Decode the
* length and allocate a buffer for the data.
*/
server->tcp_length = server->tcp_lenbuf[0] << 8
| server->tcp_lenbuf[1];
server->tcp_buffer = malloc(server->tcp_length);
if (!server->tcp_buffer)
handle_error(channel, i, now);
server->tcp_buffer_pos = 0;
}
}
{
/* Read data into the allocated buffer. */
count = sread(server->tcp_socket,
server->tcp_buffer + server->tcp_buffer_pos,
server->tcp_length - server->tcp_buffer_pos);
if (count <= 0)
{
if (!(count == -1 && try_again(SOCKERRNO)))
William Ahern
committed
handle_error(channel, i, now);
if (server->tcp_buffer_pos == server->tcp_length)
{
/* We finished reading this answer; process it and
* prepare to read another length word.
*/
process_answer(channel, server->tcp_buffer, server->tcp_length,
i, 1, now);
if (server->tcp_buffer)
free(server->tcp_buffer);
server->tcp_buffer = NULL;
server->tcp_lenbuf_pos = 0;
}
}
/* If any UDP sockets select true for reading, process them. */
static void read_udp_packets(ares_channel channel, fd_set *read_fds,
ares_socket_t read_fd, time_t now)
{
struct server_state *server;
unsigned char buf[PACKETSZ + 1];
Daniel Stenberg
committed
if(!read_fds && (read_fd == ARES_SOCKET_BAD))
/* no possible action */
return;
for (i = 0; i < channel->nservers; i++)
{
/* Make sure the server has a socket and is selected in read_fds. */
server = &channel->servers[i];
if (server->udp_socket == ARES_SOCKET_BAD)
if(read_fds) {
if(!FD_ISSET(server->udp_socket, read_fds))
continue;
}
else {
if(server->udp_socket != read_fd)
continue;
}
Daniel Stenberg
committed
if(read_fds)
/* If there's an error and we close this socket, then open
* another with the same fd to talk to another server, then we
* don't want to think that it was the new socket that was
* ready. This is not disastrous, but is likely to result in
* extra system calls and confusion. */
FD_CLR(server->udp_socket, read_fds);
count = sread(server->udp_socket, buf, sizeof(buf));
if (count == -1 && try_again(SOCKERRNO))
William Ahern
committed
continue;
else if (count <= 0)
handle_error(channel, i, now);
}
}
/* If any queries have timed out, note the timeout and move them on. */
static void process_timeouts(ares_channel channel, time_t now)
{
struct query *query, *next;
for (query = channel->queries; query; query = next)
{
next = query->next;
if (query->timeout != 0 && now >= query->timeout)
{
query->error_status = ARES_ETIMEOUT;
next = next_server(channel, query, now);
}
}
}
/* Handle an answer from a server. */
static void process_answer(ares_channel channel, unsigned char *abuf,
int alen, int whichserver, int tcp, time_t now)
int tc, rcode;
unsigned short id;
struct query *query;
/* If there's no room in the answer for a header, we can't do much
* with it. */
if (alen < HFIXEDSZ)
return;
/* Grab the query ID, truncate bit, and response code from the packet. */
id = DNS_HEADER_QID(abuf);
tc = DNS_HEADER_TC(abuf);
rcode = DNS_HEADER_RCODE(abuf);
/* Find the query corresponding to this packet. */
for (query = channel->queries; query; query = query->next)
{
if (query->qid == id)
}
if (!query)
return;
/* If we got a truncated UDP packet and are not ignoring truncation,
* don't accept the packet, and switch the query to TCP if we hadn't
* done so already.
*/
if ((tc || alen > PACKETSZ) && !tcp && !(channel->flags & ARES_FLAG_IGNTC))
{
if (!query->using_tcp)
{
query->using_tcp = 1;
ares__send_query(channel, query, now);
}
return;
}
/* Limit alen to PACKETSZ if we aren't using TCP (only relevant if we
* are ignoring truncation.
*/
if (alen > PACKETSZ && !tcp)
alen = PACKETSZ;
/* If we aren't passing through all error packets, discard packets
* with SERVFAIL, NOTIMP, or REFUSED response codes.
*/
if (!(channel->flags & ARES_FLAG_NOCHECKRESP))
{
if (rcode == SERVFAIL || rcode == NOTIMP || rcode == REFUSED)
{
query->skip_server[whichserver] = 1;
if (query->server == whichserver)
next_server(channel, query, now);
return;
}
if (!same_questions(query->qbuf, query->qlen, abuf, alen))
{
if (query->server == whichserver)
next_server(channel, query, now);
return;
}
}
end_query(channel, query, ARES_SUCCESS, abuf, alen);
}
static void handle_error(ares_channel channel, int whichserver, time_t now)
{
/* Reset communications with this server. */
Daniel Stenberg
committed
ares__close_sockets(channel, &channel->servers[whichserver]);
/* Tell all queries talking to this server to move on and not try
* this server again.
*/
if (query->server == whichserver)
{
query->skip_server[whichserver] = 1;
next = next_server(channel, query, now);
}
static struct query *next_server(ares_channel channel, struct query *query, time_t now)
{
/* Advance to the next server or try. */
query->server++;
for (; query->try < channel->tries; query->try++)
{
for (; query->server < channel->nservers; query->server++)
{
if (!query->skip_server[query->server])
{
ares__send_query(channel, query, now);
return (query->next);
}
}
Daniel Stenberg
committed
/* You might think that with TCP we only need one try. However,
* even when using TCP, servers can time-out our connection just
* as we're sending a request, or close our connection because
* they die, or never send us a reply because they get wedged or
* tickle a bug that drops our request.
*/
return end_query(channel, query, query->error_status, NULL, 0);
}
void ares__send_query(ares_channel channel, struct query *query, time_t now)
{
struct send_request *sendreq;
struct server_state *server;
server = &channel->servers[query->server];
if (query->using_tcp)
{
/* Make sure the TCP socket for this server is set up and queue
* a send request.
*/
if (server->tcp_socket == ARES_SOCKET_BAD)
{
if (open_tcp_socket(channel, server) == -1)
{
query->skip_server[query->server] = 1;
next_server(channel, query, now);
return;
}
}
end_query(channel, query, ARES_ENOMEM, NULL, 0);
sendreq->data = query->tcpbuf;
sendreq->len = query->tcplen;
sendreq->next = NULL;
if (server->qtail)
server->qtail->next = sendreq;
Daniel Stenberg
committed
{
SOCK_STATE_CALLBACK(channel, server->tcp_socket, 1, 1);
server->qhead = sendreq;
}
server->qtail = sendreq;
query->timeout = 0;
}
else
{
if (server->udp_socket == ARES_SOCKET_BAD)
{
if (open_udp_socket(channel, server) == -1)
{
query->skip_server[query->server] = 1;
next_server(channel, query, now);
return;
}
}
if (swrite(server->udp_socket, query->qbuf, query->qlen) == -1)
William Ahern
committed
/* FIXME: Handle EAGAIN here since it likely can happen. */
query->skip_server[query->server] = 1;
next_server(channel, query, now);
return;
}
+ ((query->try == 0) ? channel->timeout
: channel->timeout << query->try / channel->nservers);
/*
* nonblock() set the given socket to either blocking or non-blocking mode
* based on the 'nonblock' boolean argument. This function is highly portable.
*/
static int nonblock(ares_socket_t sockfd, /* operate on this */
int nonblock /* TRUE or FALSE */)
#undef SETBLOCK
#define SETBLOCK 0
#ifdef HAVE_O_NONBLOCK
/* most recent unix versions */
int flags;
flags = fcntl(sockfd, F_GETFL, 0);
if (FALSE != nonblock)
return fcntl(sockfd, F_SETFL, flags | O_NONBLOCK);
else
return fcntl(sockfd, F_SETFL, flags & (~O_NONBLOCK));
#undef SETBLOCK
#define SETBLOCK 1
#endif
#if defined(HAVE_FIONBIO) && (SETBLOCK == 0)
/* older unix versions */
int flags;
flags = nonblock;
return ioctl(sockfd, FIONBIO, &flags);
#undef SETBLOCK
#define SETBLOCK 2
#endif
#if defined(HAVE_IOCTLSOCKET) && (SETBLOCK == 0)
#ifdef WATT32
char flags;
#else
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
flags = nonblock;
return ioctlsocket(sockfd, FIONBIO, &flags);
#undef SETBLOCK
#define SETBLOCK 3
#endif
#if defined(HAVE_IOCTLSOCKET_CASE) && (SETBLOCK == 0)
/* presumably for Amiga */
return IoctlSocket(sockfd, FIONBIO, (long)nonblock);
#undef SETBLOCK
#define SETBLOCK 4
#endif
#if defined(HAVE_SO_NONBLOCK) && (SETBLOCK == 0)
/* BeOS */
long b = nonblock ? 1 : 0;
return setsockopt(sockfd, SOL_SOCKET, SO_NONBLOCK, &b, sizeof(b));
#undef SETBLOCK
#define SETBLOCK 5
#endif
#ifdef HAVE_DISABLED_NONBLOCKING
return 0; /* returns success */
#undef SETBLOCK
#define SETBLOCK 6
#endif
#if (SETBLOCK == 0)
#error "no non-blocking method was found/used/set"
}
static int open_tcp_socket(ares_channel channel, struct server_state *server)
{
struct sockaddr_in sockin;
/* Acquire a socket. */
s = socket(AF_INET, SOCK_STREAM, 0);
if (s == ARES_SOCKET_BAD)
return -1;
/* Set the socket non-blocking. */
/* Connect to the server. */
memset(&sockin, 0, sizeof(sockin));
sockin.sin_family = AF_INET;
sockin.sin_addr = server->addr;
sockin.sin_port = (unsigned short)(channel->tcp_port & 0xffff);
if (connect(s, (struct sockaddr *) &sockin, sizeof(sockin)) == -1) {
int err = SOCKERRNO;
if (err != EINPROGRESS && err != EWOULDBLOCK) {
closesocket(s);
return -1;
}
}
Daniel Stenberg
committed
SOCK_STATE_CALLBACK(channel, s, 1, 0);
server->tcp_socket = s;
return 0;
}
static int open_udp_socket(ares_channel channel, struct server_state *server)
{
ares_socket_t s;
struct sockaddr_in sockin;
/* Acquire a socket. */
s = socket(AF_INET, SOCK_DGRAM, 0);
if (s == ARES_SOCKET_BAD)
/* Set the socket non-blocking. */
nonblock(s, TRUE);
/* Connect to the server. */
memset(&sockin, 0, sizeof(sockin));
sockin.sin_family = AF_INET;
sockin.sin_addr = server->addr;
sockin.sin_port = (unsigned short)(channel->udp_port & 0xffff);
if (connect(s, (struct sockaddr *) &sockin, sizeof(sockin)) == -1)
{
closesocket(s);
return -1;
}
Daniel Stenberg
committed
SOCK_STATE_CALLBACK(channel, s, 1, 0);
server->udp_socket = s;
return 0;
}
static int same_questions(const unsigned char *qbuf, int qlen,
const unsigned char *abuf, int alen)
{
struct {
const unsigned char *p;
int qdcount;
char *name;
long namelen;
int type;
int dnsclass;
} q, a;
int i, j;
if (qlen < HFIXEDSZ || alen < HFIXEDSZ)
return 0;
/* Extract qdcount from the request and reply buffers and compare them. */
q.qdcount = DNS_HEADER_QDCOUNT(qbuf);
a.qdcount = DNS_HEADER_QDCOUNT(abuf);
if (q.qdcount != a.qdcount)
return 0;
/* For each question in qbuf, find it in abuf. */
q.p = qbuf + HFIXEDSZ;
for (i = 0; i < q.qdcount; i++)
{
/* Decode the question in the query. */
if (ares_expand_name(q.p, qbuf, qlen, &q.name, &q.namelen)
!= ARES_SUCCESS)
return 0;
q.p += q.namelen;
if (q.p + QFIXEDSZ > qbuf + qlen)
{
free(q.name);
return 0;
}
q.type = DNS_QUESTION_TYPE(q.p);
q.dnsclass = DNS_QUESTION_CLASS(q.p);
q.p += QFIXEDSZ;
/* Search for this question in the answer. */
a.p = abuf + HFIXEDSZ;
for (j = 0; j < a.qdcount; j++)
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
{
/* Decode the question in the answer. */
if (ares_expand_name(a.p, abuf, alen, &a.name, &a.namelen)
!= ARES_SUCCESS)
{
free(q.name);
return 0;
}
a.p += a.namelen;
if (a.p + QFIXEDSZ > abuf + alen)
{
free(q.name);
free(a.name);
return 0;
}
a.type = DNS_QUESTION_TYPE(a.p);
a.dnsclass = DNS_QUESTION_CLASS(a.p);
a.p += QFIXEDSZ;
/* Compare the decoded questions. */
if (strcasecmp(q.name, a.name) == 0 && q.type == a.type
&& q.dnsclass == a.dnsclass)
{
free(a.name);
break;
}
free(a.name);
}
free(q.name);
if (j == a.qdcount)
static struct query *end_query (ares_channel channel, struct query *query, int status,
unsigned char *abuf, int alen)
int i;
query->callback(query->arg, status, abuf, alen);
for (q = &channel->queries; *q; q = &(*q)->next)
{
if (*q == query)
free(query->tcpbuf);
free(query->skip_server);
free(query);
/* Simple cleanup policy: if no queries are remaining, close all
* network sockets unless STAYOPEN is set.
*/
if (!channel->queries && !(channel->flags & ARES_FLAG_STAYOPEN))
{
for (i = 0; i < channel->nservers; i++)
Daniel Stenberg
committed
ares__close_sockets(channel, &channel->servers[i]);