/*
* 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
/* rx_user.c contains routines specific to the user space UNIX implementation of rx */
+/* rxi_syscall is currently not prototyped */
+
#include <afsconfig.h>
#include <afs/param.h>
-RCSID
- ("$Header$");
+#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>
#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
-#ifndef AFS_NT40_ENV
-# include <sys/time.h>
+#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
-# 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:
* Inited
*/
-pthread_mutex_t rx_if_init_mutex;
-#define LOCK_IF_INIT assert(pthread_mutex_lock(&rx_if_init_mutex)==0)
-#define UNLOCK_IF_INIT assert(pthread_mutex_unlock(&rx_if_init_mutex)==0)
+afs_kmutex_t rx_if_init_mutex;
+#define LOCK_IF_INIT MUTEX_ENTER(&rx_if_init_mutex)
+#define UNLOCK_IF_INIT MUTEX_EXIT(&rx_if_init_mutex)
/*
* The rx_if_mutex mutex protects the following global variables:
* myNetMasks
*/
-pthread_mutex_t rx_if_mutex;
-#define LOCK_IF assert(pthread_mutex_lock(&rx_if_mutex)==0)
-#define UNLOCK_IF assert(pthread_mutex_unlock(&rx_if_mutex)==0)
+afs_kmutex_t rx_if_mutex;
+#define LOCK_IF MUTEX_ENTER(&rx_if_mutex)
+#define UNLOCK_IF MUTEX_EXIT(&rx_if_mutex)
#else
#define LOCK_IF_INIT
#define UNLOCK_IF_INIT
* 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
- int pmtu=IP_PMTUDISC_DONT;
+# 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;
}
+#ifdef AFS_NT40_ENV
+ rxi_xmit_init(socketFd);
+#endif /* AFS_NT40_ENV */
+
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
len1 = 32766;
len2 = rx_UdpBufSize;
- greedy =
- (setsockopt
- (socketFd, SOL_SOCKET, SO_RCVBUF, (char *)&len2,
- sizeof(len2)) >= 0);
- if (!greedy) {
- len2 = 32766; /* fall back to old size... uh-oh! */
- }
- greedy =
+ /* find the size closest to rx_UdpBufSize that will be accepted */
+ while (!greedy && len2 > len1) {
+ greedy =
+ (setsockopt
+ (socketFd, SOL_SOCKET, SO_RCVBUF, (char *)&len2,
+ sizeof(len2)) >= 0);
+ if (!greedy)
+ len2 /= 2;
+ }
+
+ /* but do not let it get smaller than 32K */
+ if (len2 < len1)
+ len2 = len1;
+
+ if (len1 < len2)
+ len1 = len2;
+
+
+ greedy =
(setsockopt
(socketFd, SOL_SOCKET, SO_SNDBUF, (char *)&len1,
sizeof(len1)) >= 0)
if (!greedy)
(osi_Msg "%s*WARNING* Unable to increase buffering on socket\n",
name);
- 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));
#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
}
void
-osi_Panic(msg, a1, a2, a3)
- char *msg;
+osi_Panic(char *msg, ...)
{
+ va_list ap;
+ va_start(ap, msg);
(osi_Msg "Fatal Rx error: ");
- (osi_Msg msg, a1, a2, a3);
+ (osi_VMsg msg, ap);
+ 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.
*/
int
-rx_getAllAddr(afs_int32 * buffer, int maxSize)
+rx_getAllAddr(afs_uint32 * buffer, int maxSize)
{
int count = 0, offset = 0;
/* The IP address list can change so we must query for it */
rx_GetIFInfo();
-#ifndef AFS_NT40_ENV
- /* we don't want to use the loopback adapter which is first */
- /* this is a bad bad hack.
- * and doesn't hold true on Windows.
- */
- if ( rxi_numNetAddrs > 1 )
- offset = 1;
-#endif /* AFS_NT40_ENV */
-
for (count = 0; offset < rxi_numNetAddrs && maxSize > 0;
count++, offset++, maxSize--)
buffer[count] = htonl(rxi_NetAddrs[offset]);
return count;
}
+
+/* this function returns the total number of interface addresses
+ * the buffer has to be passed in by the caller. It also returns
+ * the matching interface mask and mtu. All values are returned
+ * in network byte order.
+ */
+int
+rx_getAllAddrMaskMtu(afs_uint32 addrBuffer[], afs_uint32 maskBuffer[],
+ afs_uint32 mtuBuffer[], int maxSize)
+{
+ int count = 0, offset = 0;
+
+ /* The IP address list can change so we must query for it */
+ rx_GetIFInfo();
+
+ for (count = 0;
+ offset < rxi_numNetAddrs && maxSize > 0;
+ count++, offset++, maxSize--) {
+ addrBuffer[count] = htonl(rxi_NetAddrs[offset]);
+ maskBuffer[count] = htonl(myNetMasks[offset]);
+ mtuBuffer[count] = htonl(myNetMTUs[offset]);
+ }
+ return count;
+}
#endif
#ifdef AFS_NT40_ENV
+extern rx_atomic_t rxinit_status;
+void
+rxi_InitMorePackets(void) {
+ int npackets, ncbufs;
+
+ ncbufs = (rx_maxJumboRecvSize - RX_FIRSTBUFFERSIZE);
+ if (ncbufs > 0) {
+ ncbufs = ncbufs / RX_CBUFFERSIZE;
+ npackets = rx_initSendWindow - 1;
+ rxi_MorePackets(npackets * (ncbufs + 1));
+ }
+}
void
rx_GetIFInfo(void)
{
u_int maxsize;
u_int rxsize;
- int npackets, ncbufs;
afs_uint32 i;
LOCK_IF_INIT;
+ if (Inited) {
+ if (Inited < 2 && !rx_atomic_test_bit(&rxinit_status, 0)) {
+ /* We couldn't initialize more packets earlier.
+ * Do it now. */
+ rxi_InitMorePackets();
+ Inited = 2;
+ }
+ UNLOCK_IF_INIT;
+ return;
+ }
Inited = 1;
UNLOCK_IF_INIT;
maxsize =
rxi_nRecvFrags * rxsize + (rxi_nRecvFrags - 1) * UDP_HDR_SIZE;
maxsize = rxi_AdjustMaxMTU(rxsize, maxsize);
- if (rx_maxReceiveSize < maxsize) {
+ if (rx_maxReceiveSize > maxsize) {
rx_maxReceiveSize = MIN(RX_MAX_PACKET_SIZE, maxsize);
rx_maxReceiveSize =
MIN(rx_maxReceiveSize, rx_maxReceiveSizeUser);
}
-
+ if (rx_MyMaxSendSize > maxsize) {
+ rx_MyMaxSendSize = MIN(RX_MAX_PACKET_SIZE, maxsize);
+ }
}
UNLOCK_IF;
- ncbufs = (rx_maxJumboRecvSize - RX_FIRSTBUFFERSIZE);
- if (ncbufs > 0) {
- ncbufs = ncbufs / RX_CBUFFERSIZE;
- npackets = rx_initSendWindow - 1;
- rxi_MorePackets(npackets * (ncbufs + 1));
- }
+
+ /*
+ * If rxinit_status is still set, 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 (rx_atomic_test_bit(&rxinit_status, 0))
+ return;
+
+ rxi_InitMorePackets();
}
#endif
-#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)
int
-rxi_syscall(a3, a4, a5)
- afs_uint32 a3, a4;
- void *a5;
+rxi_syscall(afs_uint32 a3, afs_uint32 a4, void *a5)
{
afs_uint32 rcode;
- void (*old) ();
+ void (*old) (int);
- old = (void (*)())signal(SIGSYS, SIG_IGN);
+ old = signal(SIGSYS, SIG_IGN);
#if defined(AFS_SGI_ENV)
- rcode = afs_syscall(a3, a4, a5);
+ rcode = afs_syscall(AFS_SYSCALL, 28, a3, a4, a5);
#else
rcode = syscall(AFS_SYSCALL, 28 /* AFSCALL_CALL */ , a3, a4, a5);
#endif /* AFS_SGI_ENV */
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 (a->sin_family != AF_INET)
continue;
rxi_NetAddrs[rxi_numNetAddrs] = ntohl(a->sin_addr.s_addr);
- if (rxi_NetAddrs[rxi_numNetAddrs] == 0x7f000001) {
+ if (rx_IsLoopbackAddr(rxi_NetAddrs[rxi_numNetAddrs])) {
/* we don't really care about "localhost" */
continue;
}
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");
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");
#endif
}
- if (rxi_NetAddrs[rxi_numNetAddrs] != 0x7f000001) { /* ignore lo0 */
+ if (!rx_IsLoopbackAddr(rxi_NetAddrs[rxi_numNetAddrs])) { /* ignore lo0 */
int maxsize;
maxsize =
rxi_nRecvFrags * (myNetMTUs[rxi_numNetAddrs] - RX_IP_SIZE);
afs_uint32 ppaddr;
u_short rxmtu;
int ix;
+#ifdef AFS_ADAPT_PMTU
+ int sock;
+ struct sockaddr_in addr;
+#endif
LOCK_IF_INIT;
if (!Inited) {
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;
+ if (!pp->ifMTU) { /* not local */
+ 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 */
+#ifdef AFS_ADAPT_PMTU
+ sock=socket(PF_INET, SOCK_DGRAM, IPPROTO_UDP);
+ 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);
pp->maxMTU = OLD_MAX_PACKET_SIZE; /* for compatibility with old guys */
pp->natMTU = MIN((int)pp->ifMTU, OLD_MAX_PACKET_SIZE);
pp->maxDgramPackets =
MIN(rxi_nDgramPackets,
- rxi_AdjustDgramPackets(RX_MAX_FRAGS, pp->ifMTU));
+ rxi_AdjustDgramPackets(rxi_nSendFrags, pp->ifMTU));
pp->ifDgramPackets =
MIN(rxi_nDgramPackets,
- rxi_AdjustDgramPackets(RX_MAX_FRAGS, pp->ifMTU));
+ rxi_AdjustDgramPackets(rxi_nSendFrags, pp->ifMTU));
pp->maxDgramPackets = 1;
/* Initialize slow start parameters */
pp->MTU = MIN(pp->natMTU, pp->maxMTU);
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;
}
+
+#ifdef AFS_RXERRQ_ENV
+int
+rxi_HandleSocketError(int socket)
+{
+ struct msghdr msg;
+ struct cmsghdr *cmsg;
+ struct sock_extended_err *err;
+ struct sockaddr_in addr;
+ char controlmsgbuf[256];
+ int code;
+
+ msg.msg_name = &addr;
+ msg.msg_namelen = sizeof(addr);
+ msg.msg_iov = NULL;
+ msg.msg_iovlen = 0;
+ msg.msg_control = controlmsgbuf;
+ msg.msg_controllen = 256;
+ msg.msg_flags = 0;
+ code = recvmsg(socket, &msg, MSG_ERRQUEUE|MSG_DONTWAIT|MSG_TRUNC);
+
+ if (code < 0 || !(msg.msg_flags & MSG_ERRQUEUE))
+ return 0;
+
+ for (cmsg = CMSG_FIRSTHDR(&msg); cmsg; cmsg = CMSG_NXTHDR(&msg, cmsg)) {
+ 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);
+ }
+ }
+
+ return 1;
+}
+#endif
git://git.openafs.org / openafs.git / blobdiff