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 #include <afsconfig.h>
13 #include <afs/param.h>
18 # include <sys/types.h>
22 # include <WINNT/syscfg.h>
24 # include <sys/socket.h>
25 # include <sys/file.h>
27 # include <sys/stat.h>
28 # include <netinet/in.h>
29 # include <sys/time.h>
31 # include <sys/ioctl.h>
34 #if !defined(AFS_AIX_ENV) && !defined(AFS_NT40_ENV) && !defined(AFS_DJGPP_ENV)
35 # include <sys/syscall.h>
37 #include <afs/afs_args.h>
38 #include <afs/afsutil.h>
47 #ifndef IPPORT_USERRESERVED
48 /* If in.h doesn't define this, define it anyway. Unfortunately, defining
49 this doesn't put the code into the kernel to restrict kernel assigned
50 port numbers to numbers below IPPORT_USERRESERVED... */
51 #define IPPORT_USERRESERVED 5000
55 # include <sys/time.h>
58 # include "rx_globals.h"
60 #ifdef AFS_PTHREAD_ENV
64 * The rx_if_init_mutex mutex protects the following global variables:
68 pthread_mutex_t rx_if_init_mutex;
69 #define LOCK_IF_INIT assert(pthread_mutex_lock(&rx_if_init_mutex)==0)
70 #define UNLOCK_IF_INIT assert(pthread_mutex_unlock(&rx_if_init_mutex)==0)
73 * The rx_if_mutex mutex protects the following global variables:
79 pthread_mutex_t rx_if_mutex;
80 #define LOCK_IF assert(pthread_mutex_lock(&rx_if_mutex)==0)
81 #define UNLOCK_IF assert(pthread_mutex_unlock(&rx_if_mutex)==0)
84 #define UNLOCK_IF_INIT
87 #endif /* AFS_PTHREAD_ENV */
91 * Make a socket for receiving/sending IP packets. Set it into non-blocking
92 * and large buffering modes. If port isn't specified, the kernel will pick
93 * one. Returns the socket (>= 0) on success. Returns OSI_NULLSOCKET on
94 * failure. Port must be in network byte order.
97 rxi_GetHostUDPSocket(struct sockaddr_storage *saddr, int salen)
100 osi_socket socketFd = OSI_NULLSOCKET;
101 struct sockaddr_in taddr;
102 char *name = "rxi_GetUDPSocket: ";
103 #ifdef AFS_LINUX22_ENV
104 int pmtu=IP_PMTUDISC_DONT;
108 #if !defined(AFS_NT40_ENV) && !defined(AFS_DJGPP_ENV)
109 if (ntohs(port) >= IPPORT_RESERVED && ntohs(port) < IPPORT_USERRESERVED) {
110 /* (osi_Msg "%s*WARNING* port number %d is not a reserved port number. Use port numbers above %d\n", name, port, IPPORT_USERRESERVED);
113 if (ntohs(port) > 0 && ntohs(port) < IPPORT_RESERVED && geteuid() != 0) {
115 "%sport number %d is a reserved port number which may only be used by root. Use port numbers above %d\n",
116 name, ntohs(port), IPPORT_USERRESERVED);
121 socketFd = socket(rx_ssfamily(saddr), SOCK_DGRAM, 0);
128 #define MAX_RX_BINDS 10
129 for (binds = 0; binds < MAX_RX_BINDS; binds++) {
132 code = bind(socketFd, (struct sockaddr *) saddr, salen);
138 (osi_Msg "%sbind failed\n", name);
141 #if !defined(AFS_NT40_ENV) && !defined(AFS_DJGPP_ENV)
143 * Set close-on-exec on rx socket
145 fcntl(socketFd, F_SETFD, 1);
148 #ifndef AFS_DJGPP_ENV
149 /* Use one of three different ways of getting a socket buffer expanded to
157 len2 = rx_UdpBufSize;
160 (socketFd, SOL_SOCKET, SO_RCVBUF, (char *)&len2,
163 len2 = 32766; /* fall back to old size... uh-oh! */
168 (socketFd, SOL_SOCKET, SO_SNDBUF, (char *)&len1,
172 (socketFd, SOL_SOCKET, SO_RCVBUF, (char *)&len2,
175 (osi_Msg "%s*WARNING* Unable to increase buffering on socket\n",
177 MUTEX_ENTER(&rx_stats_mutex);
178 rx_stats.socketGreedy = greedy;
179 MUTEX_EXIT(&rx_stats_mutex);
181 #endif /* AFS_DJGPP_ENV */
183 #ifdef AFS_LINUX22_ENV
184 setsockopt(socketFd, SOL_IP, IP_MTU_DISCOVER, &pmtu, sizeof(pmtu));
187 if (rxi_Listen(socketFd) < 0) {
196 closesocket(socketFd);
202 return OSI_NULLSOCKET;
206 rxi_GetUDPSocket(u_short port)
208 struct sockaddr_storage saddr;
209 struct sockaddr_in *sin = (struct sockaddr_in *) &saddr;
211 memset((void *) &saddr, 0, sizeof(saddr));
213 sin->sin_family = AF_INET;
214 sin->sin_addr.s_addr = htonl(INADDR_ANY);
215 sin->sin_port = port;
217 return rxi_GetHostUDPSocket(&saddr, sizeof(struct sockaddr_in));
221 osi_Panic(msg, a1, a2, a3)
224 (osi_Msg "Fatal Rx error: ");
225 (osi_Msg msg, a1, a2, a3);
232 * osi_AssertFailU() -- used by the osi_Assert() macro.
236 osi_AssertFailU(const char *expr, const char *file, int line)
238 osi_Panic("assertion failed: %s, file: %s, line: %d\n", expr,
242 #if defined(AFS_AIX32_ENV) && !defined(KERNEL)
244 static const char memZero;
246 osi_Alloc(afs_int32 x)
249 * 0-length allocs may return NULL ptr from malloc, so we special-case
250 * things so that NULL returned iff an error occurred
253 return (void *)&memZero;
258 osi_Free(void *x, afs_int32 size)
265 #endif /* defined(AFS_AIX32_ENV) && !defined(KERNEL) */
267 #define ADDRSPERSITE 16
270 static afs_uint32 rxi_NetAddrs[ADDRSPERSITE]; /* host order */
271 static int myNetMTUs[ADDRSPERSITE];
272 static int myNetMasks[ADDRSPERSITE];
273 static int myNetFlags[ADDRSPERSITE];
274 static u_int rxi_numNetAddrs;
275 static int Inited = 0;
277 #if defined(AFS_NT40_ENV) || defined(AFS_DJGPP_ENV)
281 /* The IP address list can change so we must query for it */
284 /* we don't want to use the loopback adapter which is first */
285 /* this is a bad bad hack */
286 if (rxi_numNetAddrs > 1)
287 return htonl(rxi_NetAddrs[1]);
288 else if (rxi_numNetAddrs > 0)
289 return htonl(rxi_NetAddrs[0]);
295 ** return number of addresses
296 ** and the addresses themselves in the buffer
297 ** maxSize - max number of interfaces to return.
300 rx_getAllAddr(afs_int32 * buffer, int maxSize)
302 int count = 0, offset = 0;
304 /* The IP address list can change so we must query for it */
307 /* we don't want to use the loopback adapter which is first */
308 /* this is a bad bad hack */
309 if ( rxi_numNetAddrs > 1 )
312 for (count = 0; offset < rxi_numNetAddrs && maxSize > 0;
313 count++, offset++, maxSize--)
314 buffer[count] = htonl(rxi_NetAddrs[offset]);
326 int npackets, ncbufs;
334 rxi_numNetAddrs = ADDRSPERSITE;
335 (void)syscfg_GetIFInfo(&rxi_numNetAddrs, rxi_NetAddrs,
336 myNetMasks, myNetMTUs, myNetFlags);
338 for (i = 0; i < rxi_numNetAddrs; i++) {
339 rxsize = rxi_AdjustIfMTU(myNetMTUs[i] - RX_IPUDP_SIZE);
341 rxi_nRecvFrags * rxsize + (rxi_nRecvFrags - 1) * UDP_HDR_SIZE;
342 maxsize = rxi_AdjustMaxMTU(rxsize, maxsize);
343 if (rx_maxReceiveSize < maxsize) {
344 rx_maxReceiveSize = MIN(RX_MAX_PACKET_SIZE, maxsize);
346 MIN(rx_maxReceiveSize, rx_maxReceiveSizeUser);
351 ncbufs = (rx_maxJumboRecvSize - RX_FIRSTBUFFERSIZE);
353 ncbufs = ncbufs / RX_CBUFFERSIZE;
354 npackets = rx_initSendWindow - 1;
355 rxi_MorePackets(npackets * (ncbufs + 1));
361 fudge_netmask(afs_uint32 addr)
367 else if (IN_CLASSB(addr))
369 else if (IN_CLASSC(addr))
379 #if !defined(AFS_AIX_ENV) && !defined(AFS_NT40_ENV) && !defined(AFS_LINUX20_ENV) && !defined(AFS_DJGPP_ENV)
381 rxi_syscall(a3, a4, a5)
388 old = (void (*)())signal(SIGSYS, SIG_IGN);
390 #if defined(AFS_SGI_ENV)
391 rcode = afs_syscall(a3, a4, a5);
393 rcode = syscall(AFS_SYSCALL, 28 /* AFSCALL_CALL */ , a3, a4, a5);
394 #endif /* AFS_SGI_ENV */
400 #endif /* AFS_AIX_ENV */
408 #ifndef AFS_DJGPP_ENV
410 struct ifreq ifs[ADDRSPERSITE];
413 char buf[BUFSIZ], *cp, *cplim;
415 struct sockaddr_in *a;
416 #endif /* AFS_DJGPP_ENV */
427 memset(rxi_NetAddrs, 0, sizeof(rxi_NetAddrs));
428 memset(myNetFlags, 0, sizeof(myNetFlags));
429 memset(myNetMTUs, 0, sizeof(myNetMTUs));
430 memset(myNetMasks, 0, sizeof(myNetMasks));
432 s = socket(AF_INET, SOCK_DGRAM, 0);
436 #ifndef AFS_DJGPP_ENV
438 ifc.ifc_len = sizeof(buf);
442 ifc.ifc_len = sizeof(ifs);
443 ifc.ifc_buf = (caddr_t) & ifs[0];
444 memset(&ifs[0], 0, sizeof(ifs));
446 res = ioctl(s, SIOCGIFCONF, &ifc);
448 /* fputs(stderr, "ioctl error IFCONF\n"); */
455 #define size(p) MAX((p).sa_len, sizeof(p))
456 cplim = buf + ifc.ifc_len; /*skip over if's with big ifr_addr's */
457 for (cp = buf; cp < cplim;
458 cp += sizeof(ifr->ifr_name) + MAX(a->sin_len, sizeof(*a))) {
459 if (rxi_numNetAddrs >= ADDRSPERSITE)
462 ifr = (struct ifreq *)cp;
464 len = ifc.ifc_len / sizeof(struct ifreq);
465 if (len > ADDRSPERSITE)
468 for (i = 0; i < len; ++i) {
470 res = ioctl(s, SIOCGIFADDR, ifr);
473 /* fputs(stderr, "ioctl error IFADDR\n");
474 * perror(ifr->ifr_name); */
477 a = (struct sockaddr_in *)&ifr->ifr_addr;
478 if (a->sin_family != AF_INET)
480 rxi_NetAddrs[rxi_numNetAddrs] = ntohl(a->sin_addr.s_addr);
481 if (rxi_NetAddrs[rxi_numNetAddrs] == 0x7f000001) {
482 /* we don't really care about "localhost" */
485 for (j = 0; j < rxi_numNetAddrs; j++) {
486 if (rxi_NetAddrs[j] == rxi_NetAddrs[rxi_numNetAddrs])
489 if (j < rxi_numNetAddrs)
492 /* fprintf(stderr, "if %s addr=%x\n", ifr->ifr_name,
493 * rxi_NetAddrs[rxi_numNetAddrs]); */
496 res = ioctl(s, SIOCGIFFLAGS, ifr);
498 myNetFlags[rxi_numNetAddrs] = ifr->ifr_flags;
500 /* Handle aliased loopbacks as well. */
501 if (ifr->ifr_flags & IFF_LOOPBACK)
504 /* fprintf(stderr, "if %s flags=%x\n",
505 * ifr->ifr_name, ifr->ifr_flags); */
507 * fputs(stderr, "ioctl error IFFLAGS\n");
508 * perror(ifr->ifr_name); */
510 #endif /* SIOCGIFFLAGS */
512 #if !defined(AFS_AIX_ENV) && !defined(AFS_LINUX20_ENV)
513 /* this won't run on an AIX system w/o a cache manager */
514 rxi_syscallp = rxi_syscall;
517 /* If I refer to kernel extensions that aren't loaded on AIX, the
518 * program refuses to load and run, so I simply can't include the
519 * following code. Fortunately, AIX is the one operating system in
520 * which the subsequent ioctl works reliably. */
522 if ((*rxi_syscallp) (20 /*AFSOP_GETMTU */ ,
523 htonl(rxi_NetAddrs[rxi_numNetAddrs]),
524 &(myNetMTUs[rxi_numNetAddrs]))) {
525 /* fputs(stderr, "syscall error GETMTU\n");
526 * perror(ifr->ifr_name); */
527 myNetMTUs[rxi_numNetAddrs] = 0;
529 if ((*rxi_syscallp) (42 /*AFSOP_GETMASK */ ,
530 htonl(rxi_NetAddrs[rxi_numNetAddrs]),
531 &(myNetMasks[rxi_numNetAddrs]))) {
532 /* fputs(stderr, "syscall error GETMASK\n");
533 * perror(ifr->ifr_name); */
534 myNetMasks[rxi_numNetAddrs] = 0;
536 myNetMasks[rxi_numNetAddrs] =
537 ntohl(myNetMasks[rxi_numNetAddrs]);
538 /* fprintf(stderr, "if %s mask=0x%x\n",
539 * ifr->ifr_name, myNetMasks[rxi_numNetAddrs]); */
542 if (myNetMTUs[rxi_numNetAddrs] == 0) {
543 myNetMTUs[rxi_numNetAddrs] = OLD_MAX_PACKET_SIZE + RX_IPUDP_SIZE;
545 res = ioctl(s, SIOCGIFMTU, ifr);
546 if ((res == 0) && (ifr->ifr_metric > 128)) { /* sanity check */
547 myNetMTUs[rxi_numNetAddrs] = ifr->ifr_metric;
548 /* fprintf(stderr, "if %s mtu=%d\n",
549 * ifr->ifr_name, ifr->ifr_metric); */
551 /* fputs(stderr, "ioctl error IFMTU\n");
552 * perror(ifr->ifr_name); */
557 if (myNetMasks[rxi_numNetAddrs] == 0) {
558 myNetMasks[rxi_numNetAddrs] =
559 fudge_netmask(rxi_NetAddrs[rxi_numNetAddrs]);
560 #ifdef SIOCGIFNETMASK
561 res = ioctl(s, SIOCGIFNETMASK, ifr);
563 a = (struct sockaddr_in *)&ifr->ifr_addr;
564 myNetMasks[rxi_numNetAddrs] = ntohl(a->sin_addr.s_addr);
565 /* fprintf(stderr, "if %s subnetmask=0x%x\n",
566 * ifr->ifr_name, myNetMasks[rxi_numNetAddrs]); */
568 /* fputs(stderr, "ioctl error IFMASK\n");
569 * perror(ifr->ifr_name); */
574 if (rxi_NetAddrs[rxi_numNetAddrs] != 0x7f000001) { /* ignore lo0 */
577 rxi_nRecvFrags * (myNetMTUs[rxi_numNetAddrs] - RX_IP_SIZE);
578 maxsize -= UDP_HDR_SIZE; /* only the first frag has a UDP hdr */
579 if (rx_maxReceiveSize < maxsize)
580 rx_maxReceiveSize = MIN(RX_MAX_PACKET_SIZE, maxsize);
587 /* have to allocate at least enough to allow a single packet to reach its
588 * maximum size, so ReadPacket will work. Allocate enough for a couple
589 * of packets to do so, for good measure */
591 int npackets, ncbufs;
593 rx_maxJumboRecvSize =
594 RX_HEADER_SIZE + rxi_nDgramPackets * RX_JUMBOBUFFERSIZE +
595 (rxi_nDgramPackets - 1) * RX_JUMBOHEADERSIZE;
596 rx_maxJumboRecvSize = MAX(rx_maxJumboRecvSize, rx_maxReceiveSize);
597 ncbufs = (rx_maxJumboRecvSize - RX_FIRSTBUFFERSIZE);
599 ncbufs = ncbufs / RX_CBUFFERSIZE;
600 npackets = rx_initSendWindow - 1;
601 rxi_MorePackets(npackets * (ncbufs + 1));
604 #else /* AFS_DJGPP_ENV */
607 #endif /* AFS_DJGPP_ENV */
609 #endif /* AFS_NT40_ENV */
611 /* Called from rxi_FindPeer, when initializing a clear rx_peer structure,
612 * to get interesting information.
613 * Curiously enough, the rx_peerHashTable_lock currently protects the
614 * Inited variable (and hence rx_GetIFInfo). When the fs suite uses
615 * pthreads, this issue will need to be revisited.
619 rxi_InitPeerParams(struct rx_peer *pp)
631 * there's a race here since more than one thread could call
632 * rx_GetIFInfo. The race stops in rx_GetIFInfo.
640 /* try to second-guess IP, and identify which link is most likely to
641 * be used for traffic to/from this host. */
642 switch (rx_ssfamily(&pp->saddr)) {
644 ppaddr = ntohl(((struct sockaddr_in * ) &pp->saddr)->sin_addr.s_addr);
648 pp->rateFlag = 2; /* start timing after two full packets */
649 /* I don't initialize these, because I presume they are bzero'd...
650 * pp->burstSize pp->burst pp->burstWait.sec pp->burstWait.usec
651 * pp->timeout.usec */
654 for (ix = 0; ix < rxi_numNetAddrs; ++ix) {
655 if ((rxi_NetAddrs[ix] & myNetMasks[ix]) ==
656 (ppaddr & myNetMasks[ix])) {
657 #ifdef IFF_POINTOPOINT
658 if (myNetFlags[ix] & IFF_POINTOPOINT)
660 #endif /* IFF_POINTOPOINT */
661 rxmtu = myNetMTUs[ix] - RX_IPUDP_SIZE;
662 if (rxmtu < RX_MIN_PACKET_SIZE)
663 rxmtu = RX_MIN_PACKET_SIZE;
664 if (pp->ifMTU < rxmtu)
665 pp->ifMTU = MIN(rx_MyMaxSendSize, rxmtu);
669 if (!pp->ifMTU) { /* not local */
671 pp->ifMTU = MIN(rx_MyMaxSendSize, RX_REMOTE_PACKET_SIZE);
678 pp->rateFlag = 2; /* start timing after two full packets */
680 pp->ifMTU = MIN(rx_MyMaxSendSize, OLD_MAX_PACKET_SIZE);
683 #else /* ADAPT_MTU */
684 pp->rateFlag = 2; /* start timing after two full packets */
686 pp->ifMTU = MIN(rx_MyMaxSendSize, OLD_MAX_PACKET_SIZE);
687 #endif /* ADAPT_MTU */
688 pp->ifMTU = rxi_AdjustIfMTU(pp->ifMTU);
689 pp->maxMTU = OLD_MAX_PACKET_SIZE; /* for compatibility with old guys */
690 pp->natMTU = MIN((int)pp->ifMTU, OLD_MAX_PACKET_SIZE);
691 pp->maxDgramPackets =
692 MIN(rxi_nDgramPackets,
693 rxi_AdjustDgramPackets(RX_MAX_FRAGS, pp->ifMTU));
695 MIN(rxi_nDgramPackets,
696 rxi_AdjustDgramPackets(RX_MAX_FRAGS, pp->ifMTU));
697 pp->maxDgramPackets = 1;
698 /* Initialize slow start parameters */
699 pp->MTU = MIN(pp->natMTU, pp->maxMTU);
701 pp->nDgramPackets = 1;
705 /* Don't expose jumobgram internals. */
709 rx_maxReceiveSize = OLD_MAX_PACKET_SIZE;
710 rxi_nSendFrags = rxi_nRecvFrags = 1;
713 /* Override max MTU. If rx_SetNoJumbo is called, it must be
714 called before calling rx_SetMaxMTU since SetNoJumbo clobbers rx_maxReceiveSize */
716 rx_SetMaxMTU(int mtu)
718 rx_MyMaxSendSize = rx_maxReceiveSizeUser = rx_maxReceiveSize = mtu;