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 <afs/param.h>
13 # include <sys/types.h>
17 # include <WINNT/syscfg.h>
19 # include <sys/socket.h>
20 # include <sys/file.h>
22 # include <sys/stat.h>
23 # include <netinet/in.h>
24 # include <sys/time.h>
26 # include <sys/ioctl.h>
29 #if !defined(AFS_AIX_ENV) && !defined(AFS_NT40_ENV)
30 # include <sys/syscall.h>
32 #include <afs/afs_args.h>
33 #include <afs/afsutil.h>
35 #ifndef IPPORT_USERRESERVED
36 /* If in.h doesn't define this, define it anyway. Unfortunately, defining
37 this doesn't put the code into the kernel to restrict kernel assigned
38 port numbers to numbers below IPPORT_USERRESERVED... */
39 #define IPPORT_USERRESERVED 5000
43 # include <sys/time.h>
46 # include "rx_globals.h"
49 extern void rxi_Delay();
51 #ifdef AFS_PTHREAD_ENV
55 * The rx_if_init_mutex mutex protects the following global variables:
59 pthread_mutex_t rx_if_init_mutex;
60 #define LOCK_IF_INIT assert(pthread_mutex_lock(&rx_if_init_mutex)==0);
61 #define UNLOCK_IF_INIT assert(pthread_mutex_unlock(&rx_if_init_mutex)==0);
64 * The rx_if_mutex mutex protects the following global variables:
70 pthread_mutex_t rx_if_mutex;
71 #define LOCK_IF assert(pthread_mutex_lock(&rx_if_mutex)==0);
72 #define UNLOCK_IF assert(pthread_mutex_unlock(&rx_if_mutex)==0);
75 #define UNLOCK_IF_INIT
78 #endif /* AFS_PTHREAD_ENV */
82 * Make a socket for receiving/sending IP packets. Set it into non-blocking
83 * and large buffering modes. If port isn't specified, the kernel will pick
84 * one. Returns the socket (>= 0) on success. Returns OSI_NULLSOCKET on
85 * failure. Port must be in network byte order.
87 osi_socket rxi_GetUDPSocket(u_short port)
90 osi_socket socketFd = OSI_NULLSOCKET;
91 struct sockaddr_in taddr;
92 char *name = "rxi_GetUDPSocket: ";
93 extern int rxi_Listen(osi_socket sock);
97 if (ntohs(port) >= IPPORT_RESERVED && ntohs(port) < IPPORT_USERRESERVED) {
98 /* (osi_Msg "%s*WARNING* port number %d is not a reserved port number. Use port numbers above %d\n", name, port, IPPORT_USERRESERVED);
101 if (ntohs(port) > 0 && ntohs(port) < IPPORT_RESERVED && geteuid() != 0) {
102 (osi_Msg "%sport number %d is a reserved port number which may only be used by root. Use port numbers above %d\n", name, ntohs(port), IPPORT_USERRESERVED);
106 socketFd = socket(AF_INET, SOCK_DGRAM, 0);
113 taddr.sin_addr.s_addr = 0;
114 taddr.sin_family = AF_INET;
115 taddr.sin_port = (u_short)port;
116 #define MAX_RX_BINDS 10
117 for (binds=0; binds<MAX_RX_BINDS; binds++) {
118 if (binds) rxi_Delay (10);
119 code = bind(socketFd, (struct sockaddr *) &taddr, sizeof(taddr));
124 (osi_Msg "%sbind failed\n", name);
130 * Set close-on-exec on rx socket
132 fcntl(socketFd, F_SETFD, 1);
135 /* Use one of three different ways of getting a socket buffer expanded to
142 len2 = rx_UdpBufSize;
144 (setsockopt(socketFd, SOL_SOCKET, SO_RCVBUF, (char *)&len2,
147 len2=32766; /* fall back to old size... uh-oh! */
151 (setsockopt(socketFd, SOL_SOCKET, SO_SNDBUF, (char *)&len1,
152 sizeof(len1)) >= 0) &&
153 (setsockopt(socketFd, SOL_SOCKET, SO_RCVBUF, (char *)&len2,
158 (osi_Msg "%s*WARNING* Unable to increase buffering on socket\n", name);
159 if (rxi_Listen(socketFd) < 0) {
167 if (socketFd >= 0) closesocket(socketFd);
169 if (socketFd >= 0) close(socketFd);
172 MUTEX_ENTER(&rx_stats_mutex);
173 rx_stats.socketGreedy = greedy;
174 MUTEX_EXIT(&rx_stats_mutex);
175 return OSI_NULLSOCKET;
178 void osi_Panic(msg, a1, a2, a3)
181 (osi_Msg "Fatal Rx error: ");
182 (osi_Msg msg, a1, a2, a3);
189 * osi_AssertFailU() -- used by the osi_Assert() macro.
192 void osi_AssertFailU(const char *expr, const char *file, int line)
194 osi_Panic("assertion failed: %s, file: %s, line: %d\n", expr, file, line);
199 static const char memZero;
204 * 0-length allocs may return NULL ptr from osi_kalloc, so we special-case
205 * things so that NULL returned iff an error occurred
207 if (x == 0) return &memZero;
208 return ((char *) malloc(x));
215 if (x == &memZero) return;
219 #endif /* AFS_AIX32_ENV */
221 #define ADDRSPERSITE 16
224 afs_uint32 rxi_NetAddrs[ADDRSPERSITE]; /* host order */
225 static int myNetMTUs[ADDRSPERSITE];
226 static int myNetMasks[ADDRSPERSITE];
227 static int myNetFlags[ADDRSPERSITE];
228 u_int rxi_numNetAddrs;
229 static int Inited = 0;
231 #if defined(AFS_NT40_ENV)
232 int rxi_getaddr(void)
234 if (rxi_numNetAddrs > 0)
235 return htonl(rxi_NetAddrs[0]);
241 ** return number of addresses
242 ** and the addresses themselves in the buffer
243 ** maxSize - max number of interfaces to return.
245 int rx_getAllAddr (afs_int32 *buffer, int maxSize)
248 for (count = 0; count < rxi_numNetAddrs && maxSize > 0; count++, maxSize--)
249 buffer[count] = htonl(rxi_NetAddrs[count]);
258 void rx_GetIFInfo(void)
268 int npackets, ncbufs;
273 rxi_numNetAddrs = ADDRSPERSITE;
276 (void) syscfg_GetIFInfo(&rxi_numNetAddrs, rxi_NetAddrs,
277 myNetMasks, myNetMTUs, myNetFlags);
279 for (i=0; i<rxi_numNetAddrs; i++) {
280 rxsize = rxi_AdjustIfMTU(myNetMTUs[i] - RX_IPUDP_SIZE);
281 maxsize = rxi_nRecvFrags*rxsize + (rxi_nRecvFrags-1)*UDP_HDR_SIZE;
282 maxsize = rxi_AdjustMaxMTU(rxsize, maxsize);
283 if (rx_maxReceiveSize < maxsize) {
284 rx_maxReceiveSize = MIN( RX_MAX_PACKET_SIZE, maxsize);
285 rx_maxReceiveSize = MIN( rx_maxReceiveSize,
286 rx_maxReceiveSizeUser);
291 ncbufs = (rx_maxJumboRecvSize - RX_FIRSTBUFFERSIZE);
293 ncbufs = ncbufs / RX_CBUFFERSIZE;
294 npackets = rx_initSendWindow -1;
295 rxi_MorePackets(npackets*(ncbufs+1));
302 static afs_uint32 fudge_netmask(addr)
307 if (IN_CLASSA(addr)) msk = IN_CLASSA_NET;
308 else if (IN_CLASSB(addr)) msk = IN_CLASSB_NET;
309 else if (IN_CLASSC(addr)) msk = IN_CLASSC_NET;
317 #if !defined(AFS_AIX_ENV) && !defined(AFS_NT40_ENV) && !defined(AFS_LINUX20_ENV)
318 int rxi_syscall(a3, a4, a5)
325 old = (void (*)())signal(SIGSYS, SIG_IGN);
327 #if defined(AFS_SGI_ENV)
328 rcode = afs_syscall(a3, a4, a5);
330 rcode = syscall (AFS_SYSCALL, 28 /* AFSCALL_CALL */, a3, a4, a5);
331 #endif /* AFS_SGI_ENV */
337 #endif /* AFS_AIX_ENV */
345 struct ifreq ifs[ADDRSPERSITE];
346 struct ifreq ifreq, *ifr;
348 char buf[BUFSIZ], *cp, *cplim;
350 struct sockaddr_in *a;
362 bzero(rxi_NetAddrs, sizeof(rxi_NetAddrs));
363 bzero(myNetFlags, sizeof(myNetFlags));
364 bzero(myNetMTUs, sizeof(myNetMTUs));
365 bzero(myNetMasks, sizeof(myNetMasks));
368 s = socket(AF_INET, SOCK_DGRAM, 0);
372 ifc.ifc_len = sizeof (buf);
376 ifc.ifc_len = sizeof(ifs);
377 ifc.ifc_buf = (caddr_t) &ifs[0];
378 bzero(&ifs[0], sizeof(ifs));
380 res = ioctl(s, SIOCGIFCONF, &ifc);
382 /* fputs(stderr, "ioctl error IFCONF\n"); */
390 #define size(p) MAX((p).sa_len, sizeof(p))
391 cplim = buf + ifc.ifc_len; /*skip over if's with big ifr_addr's */
392 for (cp = buf; cp < cplim ;
393 cp += sizeof(ifr->ifr_name) + MAX(a->sin_len, sizeof(*a))) {
394 if (rxi_numNetAddrs >= ADDRSPERSITE)
397 ifr = (struct ifreq *)cp;
399 len = ifc.ifc_len / sizeof(struct ifreq);
400 if (len > ADDRSPERSITE)
403 for (i = 0; i < len; ++i) {
405 res = ioctl(s, SIOCGIFADDR, ifr);
408 /* fputs(stderr, "ioctl error IFADDR\n");
409 perror(ifr->ifr_name); */
412 a = (struct sockaddr_in *) &ifr->ifr_addr;
413 if (a->sin_family != AF_INET) continue;
414 rxi_NetAddrs[rxi_numNetAddrs] = ntohl(a->sin_addr.s_addr);
415 if (rxi_NetAddrs[rxi_numNetAddrs] == 0x7f000001) {
416 /* we don't really care about "localhost" */
419 for (j=0; j < rxi_numNetAddrs; j++) {
420 if (rxi_NetAddrs[j] == rxi_NetAddrs[rxi_numNetAddrs])
423 if (j < rxi_numNetAddrs) continue;
425 /* fprintf(stderr, "if %s addr=%x\n", ifr->ifr_name,
426 rxi_NetAddrs[rxi_numNetAddrs]); */
429 res = ioctl(s, SIOCGIFFLAGS, ifr);
431 myNetFlags[rxi_numNetAddrs] = ifr->ifr_flags;
433 /* Handle aliased loopbacks as well. */
434 if (ifr->ifr_flags & IFF_LOOPBACK)
437 /* fprintf(stderr, "if %s flags=%x\n",
438 ifr->ifr_name, ifr->ifr_flags); */
441 fputs(stderr, "ioctl error IFFLAGS\n");
442 perror(ifr->ifr_name); */
444 #endif /* SIOCGIFFLAGS */
446 #if !defined(AFS_AIX_ENV) && !defined(AFS_LINUX20_ENV)
447 /* this won't run on an AIX system w/o a cache manager */
448 rxi_syscallp = rxi_syscall;
451 /* If I refer to kernel extensions that aren't loaded on AIX, the
452 * program refuses to load and run, so I simply can't include the
453 * following code. Fortunately, AIX is the one operating system in
454 * which the subsequent ioctl works reliably. */
456 if ((*rxi_syscallp)( 20 /*AFSOP_GETMTU*/,
457 htonl(rxi_NetAddrs[rxi_numNetAddrs]),
458 &(myNetMTUs[rxi_numNetAddrs]))) {
459 /* fputs(stderr, "syscall error GETMTU\n");
460 perror(ifr->ifr_name); */
461 myNetMTUs[rxi_numNetAddrs] = 0;
463 if ((*rxi_syscallp)(42 /*AFSOP_GETMASK*/,
464 htonl(rxi_NetAddrs[rxi_numNetAddrs]),
465 &(myNetMasks[rxi_numNetAddrs]))) {
466 /* fputs(stderr, "syscall error GETMASK\n");
467 perror(ifr->ifr_name); */
468 myNetMasks[rxi_numNetAddrs] = 0;
470 else myNetMasks[rxi_numNetAddrs] = ntohl( myNetMasks[rxi_numNetAddrs]);
471 /* fprintf(stderr, "if %s mask=0x%x\n",
472 ifr->ifr_name, myNetMasks[rxi_numNetAddrs]);*/
475 if (myNetMTUs[rxi_numNetAddrs] == 0) {
476 myNetMTUs[rxi_numNetAddrs] = OLD_MAX_PACKET_SIZE + RX_IPUDP_SIZE;
478 res = ioctl(s, SIOCGIFMTU, ifr);
479 if ((res == 0) && (ifr->ifr_metric > 128)) { /* sanity check */
480 myNetMTUs[rxi_numNetAddrs] = ifr->ifr_metric;
481 /* fprintf(stderr, "if %s mtu=%d\n",
482 ifr->ifr_name, ifr->ifr_metric); */
485 /* fputs(stderr, "ioctl error IFMTU\n");
486 perror(ifr->ifr_name); */
491 if (myNetMasks[rxi_numNetAddrs] == 0) {
492 myNetMasks[rxi_numNetAddrs] = fudge_netmask(rxi_NetAddrs[rxi_numNetAddrs]);
493 #ifdef SIOCGIFNETMASK
494 res = ioctl(s, SIOCGIFNETMASK, ifr);
496 a = (struct sockaddr_in *) &ifr->ifr_addr;
497 myNetMasks[rxi_numNetAddrs] = ntohl(a->sin_addr.s_addr);
498 /* fprintf(stderr, "if %s subnetmask=0x%x\n",
499 ifr->ifr_name, myNetMasks[rxi_numNetAddrs]); */
502 /* fputs(stderr, "ioctl error IFMASK\n");
503 perror(ifr->ifr_name); */
508 if (rxi_NetAddrs[rxi_numNetAddrs] != 0x7f000001) { /* ignore lo0 */
510 maxsize = rxi_nRecvFrags*(myNetMTUs[rxi_numNetAddrs] - RX_IP_SIZE);
511 maxsize -= UDP_HDR_SIZE; /* only the first frag has a UDP hdr */
512 if (rx_maxReceiveSize < maxsize)
513 rx_maxReceiveSize = MIN( RX_MAX_PACKET_SIZE, maxsize);
520 /* have to allocate at least enough to allow a single packet to reach its
521 * maximum size, so ReadPacket will work. Allocate enough for a couple
522 * of packets to do so, for good measure */
524 int npackets, ncbufs;
526 rx_maxJumboRecvSize = RX_HEADER_SIZE
527 + rxi_nDgramPackets * RX_JUMBOBUFFERSIZE
528 + (rxi_nDgramPackets-1) * RX_JUMBOHEADERSIZE;
529 rx_maxJumboRecvSize = MAX(rx_maxJumboRecvSize, rx_maxReceiveSize);
530 ncbufs = (rx_maxJumboRecvSize - RX_FIRSTBUFFERSIZE);
532 ncbufs = ncbufs / RX_CBUFFERSIZE;
533 npackets = rx_initSendWindow -1;
534 rxi_MorePackets(npackets*(ncbufs+1));
538 #endif /* AFS_NT40_ENV */
540 /* Called from rxi_FindPeer, when initializing a clear rx_peer structure,
541 * to get interesting information.
542 * Curiously enough, the rx_peerHashTable_lock currently protects the
543 * Inited variable (and hence rx_GetIFInfo). When the fs suite uses
544 * pthreads, this issue will need to be revisited.
547 void rxi_InitPeerParams(struct rx_peer *pp)
559 * there's a race here since more than one thread could call
560 * rx_GetIFInfo. The race stops in rx_GetIFInfo.
568 /* try to second-guess IP, and identify which link is most likely to
569 * be used for traffic to/from this host. */
570 ppaddr = ntohl(pp->host);
574 pp->rateFlag = 2; /* start timing after two full packets */
575 /* I don't initialize these, because I presume they are bzero'd...
576 * pp->burstSize pp->burst pp->burstWait.sec pp->burstWait.usec
577 * pp->timeout.usec */
580 for (ix = 0; ix < rxi_numNetAddrs; ++ix) {
581 if ((rxi_NetAddrs[ix] & myNetMasks[ix]) == (ppaddr & myNetMasks[ix])) {
582 #ifdef IFF_POINTOPOINT
583 if (myNetFlags[ix] & IFF_POINTOPOINT)
585 #endif /* IFF_POINTOPOINT */
586 rxmtu = myNetMTUs[ix] - RX_IPUDP_SIZE;
587 if ( rxmtu < RX_MIN_PACKET_SIZE )
588 rxmtu = RX_MIN_PACKET_SIZE ;
589 if (pp->ifMTU < rxmtu)
590 pp->ifMTU = MIN(rx_MyMaxSendSize, rxmtu);
595 if (!pp->ifMTU) { /* not local */
597 pp->ifMTU = RX_REMOTE_PACKET_SIZE;
599 #else /* ADAPT_MTU */
600 pp->rateFlag = 2; /* start timing after two full packets */
602 pp->ifMTU = OLD_MAX_PACKET_SIZE;
603 #endif /* ADAPT_MTU */
604 pp->ifMTU = rxi_AdjustIfMTU(pp->ifMTU);
605 pp->maxMTU = OLD_MAX_PACKET_SIZE; /* for compatibility with old guys */
606 pp->natMTU = MIN((int)pp->ifMTU, OLD_MAX_PACKET_SIZE);
607 pp->maxDgramPackets = MIN(rxi_nDgramPackets,
608 rxi_AdjustDgramPackets(RX_MAX_FRAGS, pp->ifMTU));
609 pp->ifDgramPackets = MIN(rxi_nDgramPackets,
610 rxi_AdjustDgramPackets(RX_MAX_FRAGS, pp->ifMTU));
611 pp->maxDgramPackets = 1;
612 /* Initialize slow start parameters */
613 pp->MTU = MIN(pp->natMTU, pp->maxMTU);
615 pp->nDgramPackets = 1;
619 /* Don't expose jumobgram internals. */
620 void rx_SetNoJumbo(void)
622 rx_maxReceiveSize = OLD_MAX_PACKET_SIZE;
623 rxi_nSendFrags = rxi_nRecvFrags = 1;