/*
* Copyright 2000, International Business Machines Corporation and others.
* All Rights Reserved.
- *
+ *
* This software has been released under the terms of the IBM Public
* License. For details, see the LICENSE file in the top-level source
* directory or online at http://www.openafs.org/dl/license10.html
#include <afsconfig.h>
#include <afs/param.h>
+#include <roken.h>
+
+#include <afs/opr.h>
-# include <sys/types.h>
-# include <errno.h>
-# include <signal.h>
-# include <string.h>
#ifdef AFS_NT40_ENV
# include <WINNT/syscfg.h>
#else
-# include <sys/socket.h>
-# include <sys/file.h>
-# include <netdb.h>
-# include <sys/stat.h>
-# include <netinet/in.h>
-# include <sys/time.h>
# include <net/if.h>
-# include <sys/ioctl.h>
-# include <unistd.h>
#endif
-# include <fcntl.h>
-#if !defined(AFS_AIX_ENV) && !defined(AFS_NT40_ENV)
+#if !defined(AFS_AIX_ENV) && !defined(AFS_NT40_ENV)
# include <sys/syscall.h>
#endif
#include <afs/afs_args.h>
#define IPPORT_USERRESERVED 5000
# endif
-#if defined(HAVE_LINUX_ERRQUEUE_H) && defined(ADAPT_PMTU)
-#include <linux/types.h>
-#include <linux/errqueue.h>
-#ifndef IP_MTU
-#define IP_MTU 14
-#endif
+#if defined(AFS_LINUX22_ENV) && defined(AFS_RXERRQ_ENV)
+# include <linux/types.h>
+# include <linux/errqueue.h>
+# if defined(AFS_ADAPT_PMTU) && !defined(IP_MTU)
+# define IP_MTU 14
+# endif
#endif
-#ifndef AFS_NT40_ENV
-# include <sys/time.h>
-#endif
-# include "rx.h"
-# include "rx_globals.h"
+#include "rx.h"
+#include "rx_atomic.h"
+#include "rx_globals.h"
+#include "rx_stats.h"
+#include "rx_peer.h"
+#include "rx_packet.h"
+#include "rx_internal.h"
#ifdef AFS_PTHREAD_ENV
-#include <assert.h>
/*
* The rx_if_init_mutex mutex protects the following global variables:
* Make a socket for receiving/sending IP packets. Set it into non-blocking
* and large buffering modes. If port isn't specified, the kernel will pick
* one. Returns the socket (>= 0) on success. Returns OSI_NULLSOCKET on
- * failure. Port must be in network byte order.
+ * failure. Port must be in network byte order.
*/
osi_socket
rxi_GetHostUDPSocket(u_int ahost, u_short port)
struct sockaddr_in taddr;
char *name = "rxi_GetUDPSocket: ";
#ifdef AFS_LINUX22_ENV
-#if defined(ADAPT_PMTU)
- int pmtu=IP_PMTUDISC_WANT;
- int recverr=1;
-#else
- int pmtu=IP_PMTUDISC_DONT;
-#endif
+# if defined(AFS_ADAPT_PMTU)
+ int pmtu = IP_PMTUDISC_WANT;
+# else
+ int pmtu = IP_PMTUDISC_DONT;
+# endif
#endif
-#if !defined(AFS_NT40_ENV)
+#if !defined(AFS_NT40_ENV)
if (ntohs(port) >= IPPORT_RESERVED && ntohs(port) < IPPORT_USERRESERVED) {
/* (osi_Msg "%s*WARNING* port number %d is not a reserved port number. Use port numbers above %d\n", name, port, IPPORT_USERRESERVED);
*/ ;
goto error;
}
#endif
- socketFd = socket(AF_INET, SOCK_DGRAM, 0);
+ socketFd = socket(AF_INET, SOCK_DGRAM, IPPROTO_UDP);
- if (socketFd < 0) {
+ if (socketFd == OSI_NULLSOCKET) {
+#ifdef AFS_NT40_ENV
+ fprintf(stderr, "socket() failed with error %u\n", WSAGetLastError());
+#else
perror("socket");
+#endif
goto error;
}
taddr.sin_addr.s_addr = ahost;
taddr.sin_family = AF_INET;
taddr.sin_port = (u_short) port;
+ memset(&taddr.sin_zero, 0, sizeof(taddr.sin_zero));
#ifdef STRUCT_SOCKADDR_HAS_SA_LEN
taddr.sin_len = sizeof(struct sockaddr_in);
#endif
if (binds)
rxi_Delay(10);
code = bind(socketFd, (struct sockaddr *)&taddr, sizeof(taddr));
- if (!code)
- break;
+ break;
}
if (code) {
- perror("bind");
(osi_Msg "%sbind failed\n", name);
goto error;
}
-#if !defined(AFS_NT40_ENV)
+#if !defined(AFS_NT40_ENV)
/*
- * Set close-on-exec on rx socket
+ * Set close-on-exec on rx socket
*/
fcntl(socketFd, F_SETFD, 1);
#endif
len2 /= 2;
}
- /* but do not let it get smaller than 32K */
+ /* but do not let it get smaller than 32K */
if (len2 < len1)
len2 = len1;
if (!greedy)
(osi_Msg "%s*WARNING* Unable to increase buffering on socket\n",
name);
- if (rx_stats_active) {
- MUTEX_ENTER(&rx_stats_mutex);
- rx_stats.socketGreedy = greedy;
- MUTEX_EXIT(&rx_stats_mutex);
- }
+ if (rx_stats_active)
+ rx_atomic_set(&rx_stats.socketGreedy, greedy);
}
#ifdef AFS_LINUX22_ENV
setsockopt(socketFd, SOL_IP, IP_MTU_DISCOVER, &pmtu, sizeof(pmtu));
-#if defined(ADAPT_PMTU)
- setsockopt(socketFd, SOL_IP, IP_RECVERR, &recverr, sizeof(recverr));
#endif
+#ifdef AFS_RXERRQ_ENV
+ {
+ int recverr = 1;
+ setsockopt(socketFd, SOL_IP, IP_RECVERR, &recverr, sizeof(recverr));
+ }
#endif
if (rxi_Listen(socketFd) < 0) {
goto error;
error:
#ifdef AFS_NT40_ENV
- if (socketFd >= 0)
+ if (socketFd != OSI_NULLSOCKET)
closesocket(socketFd);
#else
- if (socketFd >= 0)
+ if (socketFd != OSI_NULLSOCKET)
close(socketFd);
#endif
va_end(ap);
fflush(stderr);
fflush(stdout);
- afs_abort();
+ opr_abort();
}
/*
void *
osi_Alloc(afs_int32 x)
{
- /*
+ /*
* 0-length allocs may return NULL ptr from malloc, so we special-case
- * things so that NULL returned iff an error occurred
+ * things so that NULL returned iff an error occurred
*/
if (x == 0)
return (void *)&memZero;
static u_int rxi_numNetAddrs;
static int Inited = 0;
-#if defined(AFS_NT40_ENV)
+#if defined(AFS_NT40_ENV)
int
rxi_getaddr(void)
{
/* we don't want to use the loopback adapter which is first */
/* this is a bad bad hack */
if (rxi_numNetAddrs > 1)
- return htonl(rxi_NetAddrs[1]);
+ return htonl(rxi_NetAddrs[1]);
else if (rxi_numNetAddrs > 0)
return htonl(rxi_NetAddrs[0]);
else
return 0;
}
-/*
-** return number of addresses
+/*
+** return number of addresses
** and the addresses themselves in the buffer
** maxSize - max number of interfaces to return.
*/
/* The IP address list can change so we must query for it */
rx_GetIFInfo();
- for (count = 0;
+ for (count = 0;
offset < rxi_numNetAddrs && maxSize > 0;
count++, offset++, maxSize--) {
addrBuffer[count] = htonl(rxi_NetAddrs[offset]);
#endif
#ifdef AFS_NT40_ENV
-extern int rxinit_status;
-void
+void
rxi_InitMorePackets(void) {
int npackets, ncbufs;
LOCK_IF_INIT;
if (Inited) {
- if (Inited < 2 && rxinit_status == 0) {
+ if (Inited < 2 && rxi_IsRunning()) {
/* We couldn't initialize more packets earlier.
* Do it now. */
rxi_InitMorePackets();
UNLOCK_IF;
/*
- * If rxinit_status is still set, rx_InitHost() has yet to be called
+ * If rxi_IsRunning is false, rx_InitHost() has yet to be called
* and we therefore do not have any mutex locks initialized. As a
* result we cannot call rxi_MorePackets() without crashing.
*/
- if (rxinit_status)
+ if (!rxi_IsRunning())
return;
rxi_InitMorePackets();
-#if !defined(AFS_AIX_ENV) && !defined(AFS_NT40_ENV) && !defined(AFS_LINUX20_ENV)
+#if !defined(AFS_AIX_ENV) && !defined(AFS_NT40_ENV) && !defined(AFS_LINUX20_ENV) && !defined(AFS_DARWIN160_ENV)
int
rxi_syscall(afs_uint32 a3, afs_uint32 a4, void *a5)
{
#if defined(AFS_SGI_ENV)
rcode = afs_syscall(AFS_SYSCALL, 28, a3, a4, a5);
-#else
+#elif defined(AFS_SYSCALL)
rcode = syscall(AFS_SYSCALL, 28 /* AFSCALL_CALL */ , a3, a4, a5);
+#else
+ rcode = -1;
#endif /* AFS_SGI_ENV */
signal(SIGSYS, old);
memset(myNetMTUs, 0, sizeof(myNetMTUs));
memset(myNetMasks, 0, sizeof(myNetMasks));
UNLOCK_IF;
- s = socket(AF_INET, SOCK_DGRAM, 0);
- if (s < 0)
+ s = socket(AF_INET, SOCK_DGRAM, IPPROTO_UDP);
+ if (s == OSI_NULLSOCKET)
return;
-
#ifdef AFS_AIX41_ENV
ifc.ifc_len = sizeof(buf);
ifc.ifc_buf = buf;
if (ifr->ifr_flags & IFF_LOOPBACK)
continue;
#endif
- /* fprintf(stderr, "if %s flags=%x\n",
+ /* fprintf(stderr, "if %s flags=%x\n",
* ifr->ifr_name, ifr->ifr_flags); */
} else { /*
* fputs(stderr, "ioctl error IFFLAGS\n");
}
#endif /* SIOCGIFFLAGS */
-#if !defined(AFS_AIX_ENV) && !defined(AFS_LINUX20_ENV)
+#if !defined(AFS_AIX_ENV) && !defined(AFS_LINUX20_ENV) && !defined(AFS_DARWIN160_ENV)
/* this won't run on an AIX system w/o a cache manager */
rxi_syscallp = rxi_syscall;
#endif
- /* If I refer to kernel extensions that aren't loaded on AIX, the
- * program refuses to load and run, so I simply can't include the
+ /* If I refer to kernel extensions that aren't loaded on AIX, the
+ * program refuses to load and run, so I simply can't include the
* following code. Fortunately, AIX is the one operating system in
* which the subsequent ioctl works reliably. */
if (rxi_syscallp) {
} else
myNetMasks[rxi_numNetAddrs] =
ntohl(myNetMasks[rxi_numNetAddrs]);
- /* fprintf(stderr, "if %s mask=0x%x\n",
+ /* fprintf(stderr, "if %s mask=0x%x\n",
* ifr->ifr_name, myNetMasks[rxi_numNetAddrs]); */
}
res = ioctl(s, SIOCGIFMTU, ifr);
if ((res == 0) && (ifr->ifr_metric > 128)) { /* sanity check */
myNetMTUs[rxi_numNetAddrs] = ifr->ifr_metric;
- /* fprintf(stderr, "if %s mtu=%d\n",
+ /* fprintf(stderr, "if %s mtu=%d\n",
* ifr->ifr_name, ifr->ifr_metric); */
} else {
/* fputs(stderr, "ioctl error IFMTU\n");
fudge_netmask(rxi_NetAddrs[rxi_numNetAddrs]);
#ifdef SIOCGIFNETMASK
res = ioctl(s, SIOCGIFNETMASK, ifr);
- if ((res == 0)) {
+ if (res == 0) {
a = (struct sockaddr_in *)&ifr->ifr_addr;
myNetMasks[rxi_numNetAddrs] = ntohl(a->sin_addr.s_addr);
- /* fprintf(stderr, "if %s subnetmask=0x%x\n",
+ /* fprintf(stderr, "if %s subnetmask=0x%x\n",
* ifr->ifr_name, myNetMasks[rxi_numNetAddrs]); */
} else {
/* fputs(stderr, "ioctl error IFMASK\n");
afs_uint32 ppaddr;
u_short rxmtu;
int ix;
-#if defined(ADAPT_PMTU) && defined(IP_MTU)
+#ifdef AFS_ADAPT_PMTU
int sock;
struct sockaddr_in addr;
#endif
-
-
LOCK_IF_INIT;
if (!Inited) {
UNLOCK_IF_INIT;
UNLOCK_IF_INIT;
}
-#ifdef ADAPT_MTU
/* try to second-guess IP, and identify which link is most likely to
* be used for traffic to/from this host. */
ppaddr = ntohl(pp->host);
pp->ifMTU = 0;
- pp->timeout.sec = 2;
- pp->rateFlag = 2; /* start timing after two full packets */
- /* I don't initialize these, because I presume they are bzero'd...
+ rx_rto_setPeerTimeoutSecs(pp, 2);
+ /* I don't initialize these, because I presume they are bzero'd...
* pp->burstSize pp->burst pp->burstWait.sec pp->burstWait.usec
- * pp->timeout.usec */
+ */
LOCK_IF;
for (ix = 0; ix < rxi_numNetAddrs; ++ix) {
if ((rxi_NetAddrs[ix] & myNetMasks[ix]) == (ppaddr & myNetMasks[ix])) {
#ifdef IFF_POINTOPOINT
if (myNetFlags[ix] & IFF_POINTOPOINT)
- pp->timeout.sec = 4;
+ rx_rto_setPeerTimeoutSecs(pp, 4);
#endif /* IFF_POINTOPOINT */
+
rxmtu = myNetMTUs[ix] - RX_IPUDP_SIZE;
if (rxmtu < RX_MIN_PACKET_SIZE)
rxmtu = RX_MIN_PACKET_SIZE;
}
UNLOCK_IF;
if (!pp->ifMTU) { /* not local */
- pp->timeout.sec = 3;
+ rx_rto_setPeerTimeoutSecs(pp, 3);
pp->ifMTU = MIN(rx_MyMaxSendSize, RX_REMOTE_PACKET_SIZE);
}
-#else /* ADAPT_MTU */
- pp->rateFlag = 2; /* start timing after two full packets */
- pp->timeout.sec = 2;
- pp->ifMTU = MIN(rx_MyMaxSendSize, OLD_MAX_PACKET_SIZE);
-#endif /* ADAPT_MTU */
-#if defined(ADAPT_PMTU) && defined(IP_MTU)
+#ifdef AFS_ADAPT_PMTU
sock=socket(PF_INET, SOCK_DGRAM, IPPROTO_UDP);
- if (sock >= 0) {
- addr.sin_family = AF_INET;
- addr.sin_addr.s_addr = pp->host;
- addr.sin_port = pp->port;
- if (connect(sock, (struct sockaddr *)&addr, sizeof(addr)) == 0) {
- int mtu=0;
- socklen_t s = sizeof(mtu);
- if (getsockopt(sock, SOL_IP, IP_MTU, &mtu, &s)== 0) {
- pp->ifMTU = MIN(mtu - RX_IPUDP_SIZE, pp->ifMTU);
- }
- }
- close(sock);
+ if (sock != OSI_NULLSOCKET) {
+ addr.sin_family = AF_INET;
+ addr.sin_addr.s_addr = pp->host;
+ addr.sin_port = pp->port;
+ memset(&addr.sin_zero, 0, sizeof(addr.sin_zero));
+ if (connect(sock, (struct sockaddr *)&addr, sizeof(addr)) == 0) {
+ int mtu=0;
+ socklen_t s = sizeof(mtu);
+ if (getsockopt(sock, SOL_IP, IP_MTU, &mtu, &s)== 0) {
+ pp->ifMTU = MIN(mtu - RX_IPUDP_SIZE, pp->ifMTU);
+ }
+ }
+# ifdef AFS_NT40_ENV
+ closesocket(sock);
+# else
+ close(sock);
+# endif
}
#endif
pp->ifMTU = rxi_AdjustIfMTU(pp->ifMTU);
rxi_nSendFrags = rxi_nRecvFrags = 1;
}
-/* Override max MTU. If rx_SetNoJumbo is called, it must be
+/* Override max MTU. If rx_SetNoJumbo is called, it must be
called before calling rx_SetMaxMTU since SetNoJumbo clobbers rx_maxReceiveSize */
-void
+int
rx_SetMaxMTU(int mtu)
{
+ if (mtu < RX_MIN_PACKET_SIZE || mtu > RX_MAX_PACKET_DATA_SIZE)
+ return EINVAL;
+
rx_MyMaxSendSize = rx_maxReceiveSizeUser = rx_maxReceiveSize = mtu;
+
+ return 0;
}
-#if defined(HAVE_LINUX_ERRQUEUE_H) && defined(ADAPT_PMTU)
+#ifdef AFS_RXERRQ_ENV
int
rxi_HandleSocketError(int socket)
{
struct cmsghdr *cmsg;
struct sock_extended_err *err;
struct sockaddr_in addr;
- struct sockaddr *offender;
char controlmsgbuf[256];
- int ret=0;
int code;
msg.msg_name = &addr;
code = recvmsg(socket, &msg, MSG_ERRQUEUE|MSG_DONTWAIT|MSG_TRUNC);
if (code < 0 || !(msg.msg_flags & MSG_ERRQUEUE))
- goto out;
+ return 0;
for (cmsg = CMSG_FIRSTHDR(&msg); cmsg; cmsg = CMSG_NXTHDR(&msg, cmsg)) {
- if ((char *)cmsg - controlmsgbuf > msg.msg_controllen - CMSG_SPACE(0) ||
- (char *)cmsg - controlmsgbuf > msg.msg_controllen - CMSG_SPACE(cmsg->cmsg_len) ||
- cmsg->cmsg_len == 0) {
- cmsg = 0;
- break;
+ if (cmsg->cmsg_level == SOL_IP && cmsg->cmsg_type == IP_RECVERR) {
+ err = (struct sock_extended_err *)CMSG_DATA(cmsg);
+ rxi_ProcessNetError(err, addr.sin_addr.s_addr, addr.sin_port);
}
- if (cmsg->cmsg_level == SOL_IP && cmsg->cmsg_type == IP_RECVERR)
- break;
}
- if (!cmsg)
- goto out;
- ret=1;
- err =(struct sock_extended_err *) CMSG_DATA(cmsg);
-
- if (err->ee_errno == EMSGSIZE && err->ee_info >= 68) {
- rxi_SetPeerMtu(NULL, addr.sin_addr.s_addr, addr.sin_port,
- err->ee_info - RX_IPUDP_SIZE);
- }
- /* other DEST_UNREACH's and TIME_EXCEEDED should be dealt with too */
-
-out:
- return ret;
+
+ return 1;
}
#endif