2 * Copyright 2000, International Business Machines Corporation and others.
5 * This software has been released under the terms of the IBM Public
6 * License. For details, see the LICENSE file in the top-level source
7 * directory or online at http://www.openafs.org/dl/license10.html
10 /* rx_user.c contains routines specific to the user space UNIX implementation of rx */
12 /* rxi_syscall is currently not prototyped */
14 #include <afsconfig.h>
15 #include <afs/param.h>
22 # include <WINNT/syscfg.h>
26 #if !defined(AFS_AIX_ENV) && !defined(AFS_NT40_ENV)
27 # include <sys/syscall.h>
29 #include <afs/afs_args.h>
30 #include <afs/afsutil.h>
32 #ifndef IPPORT_USERRESERVED
33 /* If in.h doesn't define this, define it anyway. Unfortunately, defining
34 this doesn't put the code into the kernel to restrict kernel assigned
35 port numbers to numbers below IPPORT_USERRESERVED... */
36 #define IPPORT_USERRESERVED 5000
39 #if defined(HAVE_LINUX_ERRQUEUE_H) && defined(ADAPT_PMTU)
40 #include <linux/types.h>
41 #include <linux/errqueue.h>
48 #include "rx_atomic.h"
49 #include "rx_globals.h"
52 #include "rx_packet.h"
54 #ifdef AFS_PTHREAD_ENV
57 * The rx_if_init_mutex mutex protects the following global variables:
61 afs_kmutex_t rx_if_init_mutex;
62 #define LOCK_IF_INIT MUTEX_ENTER(&rx_if_init_mutex)
63 #define UNLOCK_IF_INIT MUTEX_EXIT(&rx_if_init_mutex)
66 * The rx_if_mutex mutex protects the following global variables:
72 afs_kmutex_t rx_if_mutex;
73 #define LOCK_IF MUTEX_ENTER(&rx_if_mutex)
74 #define UNLOCK_IF MUTEX_EXIT(&rx_if_mutex)
77 #define UNLOCK_IF_INIT
80 #endif /* AFS_PTHREAD_ENV */
84 * Make a socket for receiving/sending IP packets. Set it into non-blocking
85 * and large buffering modes. If port isn't specified, the kernel will pick
86 * one. Returns the socket (>= 0) on success. Returns OSI_NULLSOCKET on
87 * failure. Port must be in network byte order.
90 rxi_GetHostUDPSocket(u_int ahost, u_short port)
93 osi_socket socketFd = OSI_NULLSOCKET;
94 struct sockaddr_in taddr;
95 char *name = "rxi_GetUDPSocket: ";
96 #ifdef AFS_LINUX22_ENV
97 #if defined(ADAPT_PMTU)
98 int pmtu=IP_PMTUDISC_WANT;
101 int pmtu=IP_PMTUDISC_DONT;
105 #if !defined(AFS_NT40_ENV)
106 if (ntohs(port) >= IPPORT_RESERVED && ntohs(port) < IPPORT_USERRESERVED) {
107 /* (osi_Msg "%s*WARNING* port number %d is not a reserved port number. Use port numbers above %d\n", name, port, IPPORT_USERRESERVED);
110 if (ntohs(port) > 0 && ntohs(port) < IPPORT_RESERVED && geteuid() != 0) {
112 "%sport number %d is a reserved port number which may only be used by root. Use port numbers above %d\n",
113 name, ntohs(port), IPPORT_USERRESERVED);
117 socketFd = socket(AF_INET, SOCK_DGRAM, IPPROTO_UDP);
119 if (socketFd == OSI_NULLSOCKET) {
121 fprintf(stderr, "socket() failed with error %u\n", WSAGetLastError());
129 rxi_xmit_init(socketFd);
130 #endif /* AFS_NT40_ENV */
132 taddr.sin_addr.s_addr = ahost;
133 taddr.sin_family = AF_INET;
134 taddr.sin_port = (u_short) port;
135 #ifdef STRUCT_SOCKADDR_HAS_SA_LEN
136 taddr.sin_len = sizeof(struct sockaddr_in);
138 #define MAX_RX_BINDS 10
139 for (binds = 0; binds < MAX_RX_BINDS; binds++) {
142 code = bind(socketFd, (struct sockaddr *)&taddr, sizeof(taddr));
146 (osi_Msg "%sbind failed\n", name);
149 #if !defined(AFS_NT40_ENV)
151 * Set close-on-exec on rx socket
153 fcntl(socketFd, F_SETFD, 1);
156 /* Use one of three different ways of getting a socket buffer expanded to
164 len2 = rx_UdpBufSize;
166 /* find the size closest to rx_UdpBufSize that will be accepted */
167 while (!greedy && len2 > len1) {
170 (socketFd, SOL_SOCKET, SO_RCVBUF, (char *)&len2,
176 /* but do not let it get smaller than 32K */
186 (socketFd, SOL_SOCKET, SO_SNDBUF, (char *)&len1,
190 (socketFd, SOL_SOCKET, SO_RCVBUF, (char *)&len2,
193 (osi_Msg "%s*WARNING* Unable to increase buffering on socket\n",
196 rx_atomic_set(&rx_stats.socketGreedy, greedy);
199 #ifdef AFS_LINUX22_ENV
200 setsockopt(socketFd, SOL_IP, IP_MTU_DISCOVER, &pmtu, sizeof(pmtu));
201 #if defined(ADAPT_PMTU)
202 setsockopt(socketFd, SOL_IP, IP_RECVERR, &recverr, sizeof(recverr));
205 if (rxi_Listen(socketFd) < 0) {
213 if (socketFd != OSI_NULLSOCKET)
214 closesocket(socketFd);
216 if (socketFd != OSI_NULLSOCKET)
220 return OSI_NULLSOCKET;
224 rxi_GetUDPSocket(u_short port)
226 return rxi_GetHostUDPSocket(htonl(INADDR_ANY), port);
230 osi_Panic(char *msg, ...)
234 (osi_Msg "Fatal Rx error: ");
243 * osi_AssertFailU() -- used by the osi_Assert() macro.
247 osi_AssertFailU(const char *expr, const char *file, int line)
249 osi_Panic("assertion failed: %s, file: %s, line: %d\n", expr,
253 #if defined(AFS_AIX32_ENV) && !defined(KERNEL)
255 static const char memZero;
257 osi_Alloc(afs_int32 x)
260 * 0-length allocs may return NULL ptr from malloc, so we special-case
261 * things so that NULL returned iff an error occurred
264 return (void *)&memZero;
269 osi_Free(void *x, afs_int32 size)
276 #endif /* defined(AFS_AIX32_ENV) && !defined(KERNEL) */
278 #define ADDRSPERSITE 16
281 static afs_uint32 rxi_NetAddrs[ADDRSPERSITE]; /* host order */
282 static int myNetMTUs[ADDRSPERSITE];
283 static int myNetMasks[ADDRSPERSITE];
284 static int myNetFlags[ADDRSPERSITE];
285 static u_int rxi_numNetAddrs;
286 static int Inited = 0;
288 #if defined(AFS_NT40_ENV)
292 /* The IP address list can change so we must query for it */
295 /* we don't want to use the loopback adapter which is first */
296 /* this is a bad bad hack */
297 if (rxi_numNetAddrs > 1)
298 return htonl(rxi_NetAddrs[1]);
299 else if (rxi_numNetAddrs > 0)
300 return htonl(rxi_NetAddrs[0]);
306 ** return number of addresses
307 ** and the addresses themselves in the buffer
308 ** maxSize - max number of interfaces to return.
311 rx_getAllAddr(afs_uint32 * buffer, int maxSize)
313 int count = 0, offset = 0;
315 /* The IP address list can change so we must query for it */
318 for (count = 0; offset < rxi_numNetAddrs && maxSize > 0;
319 count++, offset++, maxSize--)
320 buffer[count] = htonl(rxi_NetAddrs[offset]);
325 /* this function returns the total number of interface addresses
326 * the buffer has to be passed in by the caller. It also returns
327 * the matching interface mask and mtu. All values are returned
328 * in network byte order.
331 rx_getAllAddrMaskMtu(afs_uint32 addrBuffer[], afs_uint32 maskBuffer[],
332 afs_uint32 mtuBuffer[], int maxSize)
334 int count = 0, offset = 0;
336 /* The IP address list can change so we must query for it */
340 offset < rxi_numNetAddrs && maxSize > 0;
341 count++, offset++, maxSize--) {
342 addrBuffer[count] = htonl(rxi_NetAddrs[offset]);
343 maskBuffer[count] = htonl(myNetMasks[offset]);
344 mtuBuffer[count] = htonl(myNetMTUs[offset]);
351 extern int rxinit_status;
353 rxi_InitMorePackets(void) {
354 int npackets, ncbufs;
356 ncbufs = (rx_maxJumboRecvSize - RX_FIRSTBUFFERSIZE);
358 ncbufs = ncbufs / RX_CBUFFERSIZE;
359 npackets = rx_initSendWindow - 1;
360 rxi_MorePackets(npackets * (ncbufs + 1));
372 if (Inited < 2 && rxinit_status == 0) {
373 /* We couldn't initialize more packets earlier.
375 rxi_InitMorePackets();
385 rxi_numNetAddrs = ADDRSPERSITE;
386 (void)syscfg_GetIFInfo(&rxi_numNetAddrs, rxi_NetAddrs,
387 myNetMasks, myNetMTUs, myNetFlags);
389 for (i = 0; i < rxi_numNetAddrs; i++) {
390 rxsize = rxi_AdjustIfMTU(myNetMTUs[i] - RX_IPUDP_SIZE);
392 rxi_nRecvFrags * rxsize + (rxi_nRecvFrags - 1) * UDP_HDR_SIZE;
393 maxsize = rxi_AdjustMaxMTU(rxsize, maxsize);
394 if (rx_maxReceiveSize > maxsize) {
395 rx_maxReceiveSize = MIN(RX_MAX_PACKET_SIZE, maxsize);
397 MIN(rx_maxReceiveSize, rx_maxReceiveSizeUser);
399 if (rx_MyMaxSendSize > maxsize) {
400 rx_MyMaxSendSize = MIN(RX_MAX_PACKET_SIZE, maxsize);
406 * If rxinit_status is still set, rx_InitHost() has yet to be called
407 * and we therefore do not have any mutex locks initialized. As a
408 * result we cannot call rxi_MorePackets() without crashing.
413 rxi_InitMorePackets();
418 fudge_netmask(afs_uint32 addr)
424 else if (IN_CLASSB(addr))
426 else if (IN_CLASSC(addr))
436 #if !defined(AFS_AIX_ENV) && !defined(AFS_NT40_ENV) && !defined(AFS_LINUX20_ENV)
438 rxi_syscall(afs_uint32 a3, afs_uint32 a4, void *a5)
443 old = signal(SIGSYS, SIG_IGN);
445 #if defined(AFS_SGI_ENV)
446 rcode = afs_syscall(AFS_SYSCALL, 28, a3, a4, a5);
448 rcode = syscall(AFS_SYSCALL, 28 /* AFSCALL_CALL */ , a3, a4, a5);
449 #endif /* AFS_SGI_ENV */
455 #endif /* AFS_AIX_ENV */
464 struct ifreq ifs[ADDRSPERSITE];
467 char buf[BUFSIZ], *cp, *cplim;
469 struct sockaddr_in *a;
480 memset(rxi_NetAddrs, 0, sizeof(rxi_NetAddrs));
481 memset(myNetFlags, 0, sizeof(myNetFlags));
482 memset(myNetMTUs, 0, sizeof(myNetMTUs));
483 memset(myNetMasks, 0, sizeof(myNetMasks));
485 s = socket(AF_INET, SOCK_DGRAM, IPPROTO_UDP);
486 if (s == OSI_NULLSOCKET)
489 ifc.ifc_len = sizeof(buf);
493 ifc.ifc_len = sizeof(ifs);
494 ifc.ifc_buf = (caddr_t) & ifs[0];
495 memset(&ifs[0], 0, sizeof(ifs));
497 res = ioctl(s, SIOCGIFCONF, &ifc);
499 /* fputs(stderr, "ioctl error IFCONF\n"); */
506 #define size(p) MAX((p).sa_len, sizeof(p))
507 cplim = buf + ifc.ifc_len; /*skip over if's with big ifr_addr's */
508 for (cp = buf; cp < cplim;
509 cp += sizeof(ifr->ifr_name) + MAX(a->sin_len, sizeof(*a))) {
510 if (rxi_numNetAddrs >= ADDRSPERSITE)
513 ifr = (struct ifreq *)cp;
515 len = ifc.ifc_len / sizeof(struct ifreq);
516 if (len > ADDRSPERSITE)
519 for (i = 0; i < len; ++i) {
521 res = ioctl(s, SIOCGIFADDR, ifr);
524 /* fputs(stderr, "ioctl error IFADDR\n");
525 * perror(ifr->ifr_name); */
528 a = (struct sockaddr_in *)&ifr->ifr_addr;
529 if (a->sin_family != AF_INET)
531 rxi_NetAddrs[rxi_numNetAddrs] = ntohl(a->sin_addr.s_addr);
532 if (rx_IsLoopbackAddr(rxi_NetAddrs[rxi_numNetAddrs])) {
533 /* we don't really care about "localhost" */
536 for (j = 0; j < rxi_numNetAddrs; j++) {
537 if (rxi_NetAddrs[j] == rxi_NetAddrs[rxi_numNetAddrs])
540 if (j < rxi_numNetAddrs)
543 /* fprintf(stderr, "if %s addr=%x\n", ifr->ifr_name,
544 * rxi_NetAddrs[rxi_numNetAddrs]); */
547 res = ioctl(s, SIOCGIFFLAGS, ifr);
549 myNetFlags[rxi_numNetAddrs] = ifr->ifr_flags;
551 /* Handle aliased loopbacks as well. */
552 if (ifr->ifr_flags & IFF_LOOPBACK)
555 /* fprintf(stderr, "if %s flags=%x\n",
556 * ifr->ifr_name, ifr->ifr_flags); */
558 * fputs(stderr, "ioctl error IFFLAGS\n");
559 * perror(ifr->ifr_name); */
561 #endif /* SIOCGIFFLAGS */
563 #if !defined(AFS_AIX_ENV) && !defined(AFS_LINUX20_ENV)
564 /* this won't run on an AIX system w/o a cache manager */
565 rxi_syscallp = rxi_syscall;
568 /* If I refer to kernel extensions that aren't loaded on AIX, the
569 * program refuses to load and run, so I simply can't include the
570 * following code. Fortunately, AIX is the one operating system in
571 * which the subsequent ioctl works reliably. */
573 if ((*rxi_syscallp) (20 /*AFSOP_GETMTU */ ,
574 htonl(rxi_NetAddrs[rxi_numNetAddrs]),
575 &(myNetMTUs[rxi_numNetAddrs]))) {
576 /* fputs(stderr, "syscall error GETMTU\n");
577 * perror(ifr->ifr_name); */
578 myNetMTUs[rxi_numNetAddrs] = 0;
580 if ((*rxi_syscallp) (42 /*AFSOP_GETMASK */ ,
581 htonl(rxi_NetAddrs[rxi_numNetAddrs]),
582 &(myNetMasks[rxi_numNetAddrs]))) {
583 /* fputs(stderr, "syscall error GETMASK\n");
584 * perror(ifr->ifr_name); */
585 myNetMasks[rxi_numNetAddrs] = 0;
587 myNetMasks[rxi_numNetAddrs] =
588 ntohl(myNetMasks[rxi_numNetAddrs]);
589 /* fprintf(stderr, "if %s mask=0x%x\n",
590 * ifr->ifr_name, myNetMasks[rxi_numNetAddrs]); */
593 if (myNetMTUs[rxi_numNetAddrs] == 0) {
594 myNetMTUs[rxi_numNetAddrs] = OLD_MAX_PACKET_SIZE + RX_IPUDP_SIZE;
596 res = ioctl(s, SIOCGIFMTU, ifr);
597 if ((res == 0) && (ifr->ifr_metric > 128)) { /* sanity check */
598 myNetMTUs[rxi_numNetAddrs] = ifr->ifr_metric;
599 /* fprintf(stderr, "if %s mtu=%d\n",
600 * ifr->ifr_name, ifr->ifr_metric); */
602 /* fputs(stderr, "ioctl error IFMTU\n");
603 * perror(ifr->ifr_name); */
608 if (myNetMasks[rxi_numNetAddrs] == 0) {
609 myNetMasks[rxi_numNetAddrs] =
610 fudge_netmask(rxi_NetAddrs[rxi_numNetAddrs]);
611 #ifdef SIOCGIFNETMASK
612 res = ioctl(s, SIOCGIFNETMASK, ifr);
614 a = (struct sockaddr_in *)&ifr->ifr_addr;
615 myNetMasks[rxi_numNetAddrs] = ntohl(a->sin_addr.s_addr);
616 /* fprintf(stderr, "if %s subnetmask=0x%x\n",
617 * ifr->ifr_name, myNetMasks[rxi_numNetAddrs]); */
619 /* fputs(stderr, "ioctl error IFMASK\n");
620 * perror(ifr->ifr_name); */
625 if (!rx_IsLoopbackAddr(rxi_NetAddrs[rxi_numNetAddrs])) { /* ignore lo0 */
628 rxi_nRecvFrags * (myNetMTUs[rxi_numNetAddrs] - RX_IP_SIZE);
629 maxsize -= UDP_HDR_SIZE; /* only the first frag has a UDP hdr */
630 if (rx_maxReceiveSize < maxsize)
631 rx_maxReceiveSize = MIN(RX_MAX_PACKET_SIZE, maxsize);
638 /* have to allocate at least enough to allow a single packet to reach its
639 * maximum size, so ReadPacket will work. Allocate enough for a couple
640 * of packets to do so, for good measure */
642 int npackets, ncbufs;
644 rx_maxJumboRecvSize =
645 RX_HEADER_SIZE + rxi_nDgramPackets * RX_JUMBOBUFFERSIZE +
646 (rxi_nDgramPackets - 1) * RX_JUMBOHEADERSIZE;
647 rx_maxJumboRecvSize = MAX(rx_maxJumboRecvSize, rx_maxReceiveSize);
648 ncbufs = (rx_maxJumboRecvSize - RX_FIRSTBUFFERSIZE);
650 ncbufs = ncbufs / RX_CBUFFERSIZE;
651 npackets = rx_initSendWindow - 1;
652 rxi_MorePackets(npackets * (ncbufs + 1));
656 #endif /* AFS_NT40_ENV */
658 /* Called from rxi_FindPeer, when initializing a clear rx_peer structure,
659 * to get interesting information.
660 * Curiously enough, the rx_peerHashTable_lock currently protects the
661 * Inited variable (and hence rx_GetIFInfo). When the fs suite uses
662 * pthreads, this issue will need to be revisited.
666 rxi_InitPeerParams(struct rx_peer *pp)
671 #if defined(ADAPT_PMTU) && defined(IP_MTU)
673 struct sockaddr_in addr;
680 * there's a race here since more than one thread could call
681 * rx_GetIFInfo. The race stops in rx_GetIFInfo.
689 /* try to second-guess IP, and identify which link is most likely to
690 * be used for traffic to/from this host. */
691 ppaddr = ntohl(pp->host);
694 rx_rto_setPeerTimeoutSecs(pp, 2);
695 /* I don't initialize these, because I presume they are bzero'd...
696 * pp->burstSize pp->burst pp->burstWait.sec pp->burstWait.usec
700 for (ix = 0; ix < rxi_numNetAddrs; ++ix) {
701 if ((rxi_NetAddrs[ix] & myNetMasks[ix]) == (ppaddr & myNetMasks[ix])) {
702 #ifdef IFF_POINTOPOINT
703 if (myNetFlags[ix] & IFF_POINTOPOINT)
704 rx_rto_setPeerTimeoutSecs(pp, 4);
705 #endif /* IFF_POINTOPOINT */
707 rxmtu = myNetMTUs[ix] - RX_IPUDP_SIZE;
708 if (rxmtu < RX_MIN_PACKET_SIZE)
709 rxmtu = RX_MIN_PACKET_SIZE;
710 if (pp->ifMTU < rxmtu)
711 pp->ifMTU = MIN(rx_MyMaxSendSize, rxmtu);
715 if (!pp->ifMTU) { /* not local */
716 rx_rto_setPeerTimeoutSecs(pp, 3);
717 pp->ifMTU = MIN(rx_MyMaxSendSize, RX_REMOTE_PACKET_SIZE);
719 #else /* ADAPT_MTU */
720 rx_rto_setPeerTimeoutSecs(pp, 2);
721 pp->ifMTU = MIN(rx_MyMaxSendSize, OLD_MAX_PACKET_SIZE);
722 #endif /* ADAPT_MTU */
723 #if defined(ADAPT_PMTU) && defined(IP_MTU)
724 sock=socket(PF_INET, SOCK_DGRAM, IPPROTO_UDP);
725 if (sock != OSI_NULLSOCKET) {
726 addr.sin_family = AF_INET;
727 addr.sin_addr.s_addr = pp->host;
728 addr.sin_port = pp->port;
729 if (connect(sock, (struct sockaddr *)&addr, sizeof(addr)) == 0) {
731 socklen_t s = sizeof(mtu);
732 if (getsockopt(sock, SOL_IP, IP_MTU, &mtu, &s)== 0) {
733 pp->ifMTU = MIN(mtu - RX_IPUDP_SIZE, pp->ifMTU);
743 pp->ifMTU = rxi_AdjustIfMTU(pp->ifMTU);
744 pp->maxMTU = OLD_MAX_PACKET_SIZE; /* for compatibility with old guys */
745 pp->natMTU = MIN((int)pp->ifMTU, OLD_MAX_PACKET_SIZE);
746 pp->maxDgramPackets =
747 MIN(rxi_nDgramPackets,
748 rxi_AdjustDgramPackets(rxi_nSendFrags, pp->ifMTU));
750 MIN(rxi_nDgramPackets,
751 rxi_AdjustDgramPackets(rxi_nSendFrags, pp->ifMTU));
752 pp->maxDgramPackets = 1;
753 /* Initialize slow start parameters */
754 pp->MTU = MIN(pp->natMTU, pp->maxMTU);
756 pp->nDgramPackets = 1;
760 /* Don't expose jumobgram internals. */
764 rx_maxReceiveSize = OLD_MAX_PACKET_SIZE;
765 rxi_nSendFrags = rxi_nRecvFrags = 1;
768 /* Override max MTU. If rx_SetNoJumbo is called, it must be
769 called before calling rx_SetMaxMTU since SetNoJumbo clobbers rx_maxReceiveSize */
771 rx_SetMaxMTU(int mtu)
773 if (mtu < RX_MIN_PACKET_SIZE || mtu > RX_MAX_PACKET_DATA_SIZE)
776 rx_MyMaxSendSize = rx_maxReceiveSizeUser = rx_maxReceiveSize = mtu;
781 #if defined(ADAPT_PMTU)
783 rxi_HandleSocketError(int socket)
786 #if defined(HAVE_LINUX_ERRQUEUE_H)
788 struct cmsghdr *cmsg;
789 struct sock_extended_err *err;
790 struct sockaddr_in addr;
791 char controlmsgbuf[256];
794 msg.msg_name = &addr;
795 msg.msg_namelen = sizeof(addr);
798 msg.msg_control = controlmsgbuf;
799 msg.msg_controllen = 256;
801 code = recvmsg(socket, &msg, MSG_ERRQUEUE|MSG_DONTWAIT|MSG_TRUNC);
803 if (code < 0 || !(msg.msg_flags & MSG_ERRQUEUE))
806 for (cmsg = CMSG_FIRSTHDR(&msg); cmsg; cmsg = CMSG_NXTHDR(&msg, cmsg)) {
807 if ((char *)cmsg - controlmsgbuf > msg.msg_controllen - CMSG_SPACE(0) ||
808 (char *)cmsg - controlmsgbuf > msg.msg_controllen - CMSG_SPACE(cmsg->cmsg_len) ||
809 cmsg->cmsg_len == 0) {
813 if (cmsg->cmsg_level == SOL_IP && cmsg->cmsg_type == IP_RECVERR)
819 err =(struct sock_extended_err *) CMSG_DATA(cmsg);
821 if (err->ee_errno == EMSGSIZE && err->ee_info >= 68) {
822 rxi_SetPeerMtu(NULL, addr.sin_addr.s_addr, addr.sin_port,
823 err->ee_info - RX_IPUDP_SIZE);
825 /* other DEST_UNREACH's and TIME_EXCEEDED should be dealt with too */