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) && !defined(AFS_DJGPP_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);
96 #if !defined(AFS_NT40_ENV) && !defined(AFS_DJGPP_ENV)
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);
128 #if !defined(AFS_NT40_ENV) && !defined(AFS_DJGPP_ENV)
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;
143 #ifndef AFS_DJGPP_ENV
145 (setsockopt(socketFd, SOL_SOCKET, SO_RCVBUF, (char *)&len2,
148 len2=32766; /* fall back to old size... uh-oh! */
152 (setsockopt(socketFd, SOL_SOCKET, SO_SNDBUF, (char *)&len1,
153 sizeof(len1)) >= 0) &&
154 (setsockopt(socketFd, SOL_SOCKET, SO_RCVBUF, (char *)&len2,
156 #endif /* AFS_DJGPP_ENV */
159 #ifndef AFS_DJGPP_ENV
161 (osi_Msg "%s*WARNING* Unable to increase buffering on socket\n", name);
162 #endif /* AFS_DJGPP_ENV */
163 if (rxi_Listen(socketFd) < 0) {
171 if (socketFd >= 0) closesocket(socketFd);
173 if (socketFd >= 0) close(socketFd);
176 MUTEX_ENTER(&rx_stats_mutex);
177 rx_stats.socketGreedy = greedy;
178 MUTEX_EXIT(&rx_stats_mutex);
179 return OSI_NULLSOCKET;
182 void osi_Panic(msg, a1, a2, a3)
185 (osi_Msg "Fatal Rx error: ");
186 (osi_Msg msg, a1, a2, a3);
193 * osi_AssertFailU() -- used by the osi_Assert() macro.
196 void osi_AssertFailU(const char *expr, const char *file, int line)
198 osi_Panic("assertion failed: %s, file: %s, line: %d\n", expr, file, line);
203 static const char memZero;
208 * 0-length allocs may return NULL ptr from osi_kalloc, so we special-case
209 * things so that NULL returned iff an error occurred
211 if (x == 0) return &memZero;
212 return ((char *) malloc(x));
219 if (x == &memZero) return;
223 #endif /* AFS_AIX32_ENV */
225 #define ADDRSPERSITE 16
228 afs_uint32 rxi_NetAddrs[ADDRSPERSITE]; /* host order */
229 static int myNetMTUs[ADDRSPERSITE];
230 static int myNetMasks[ADDRSPERSITE];
231 static int myNetFlags[ADDRSPERSITE];
232 u_int rxi_numNetAddrs;
233 static int Inited = 0;
235 #if defined(AFS_NT40_ENV) || defined(AFS_DJGPP_ENV)
236 int rxi_getaddr(void)
238 if (rxi_numNetAddrs > 0)
239 return htonl(rxi_NetAddrs[0]);
245 ** return number of addresses
246 ** and the addresses themselves in the buffer
247 ** maxSize - max number of interfaces to return.
249 int rx_getAllAddr (afs_int32 *buffer, int maxSize)
252 for (count = 0; count < rxi_numNetAddrs && maxSize > 0; count++, maxSize--)
253 buffer[count] = htonl(rxi_NetAddrs[count]);
262 void rx_GetIFInfo(void)
272 int npackets, ncbufs;
277 rxi_numNetAddrs = ADDRSPERSITE;
280 (void) syscfg_GetIFInfo(&rxi_numNetAddrs, rxi_NetAddrs,
281 myNetMasks, myNetMTUs, myNetFlags);
283 for (i=0; i<rxi_numNetAddrs; i++) {
284 rxsize = rxi_AdjustIfMTU(myNetMTUs[i] - RX_IPUDP_SIZE);
285 maxsize = rxi_nRecvFrags*rxsize + (rxi_nRecvFrags-1)*UDP_HDR_SIZE;
286 maxsize = rxi_AdjustMaxMTU(rxsize, maxsize);
287 if (rx_maxReceiveSize < maxsize) {
288 rx_maxReceiveSize = MIN( RX_MAX_PACKET_SIZE, maxsize);
289 rx_maxReceiveSize = MIN( rx_maxReceiveSize,
290 rx_maxReceiveSizeUser);
295 ncbufs = (rx_maxJumboRecvSize - RX_FIRSTBUFFERSIZE);
297 ncbufs = ncbufs / RX_CBUFFERSIZE;
298 npackets = rx_initSendWindow -1;
299 rxi_MorePackets(npackets*(ncbufs+1));
306 static afs_uint32 fudge_netmask(addr)
311 if (IN_CLASSA(addr)) msk = IN_CLASSA_NET;
312 else if (IN_CLASSB(addr)) msk = IN_CLASSB_NET;
313 else if (IN_CLASSC(addr)) msk = IN_CLASSC_NET;
321 #if !defined(AFS_AIX_ENV) && !defined(AFS_NT40_ENV) && !defined(AFS_LINUX20_ENV) && !defined(AFS_DJGPP_ENV)
322 int rxi_syscall(a3, a4, a5)
329 old = (void (*)())signal(SIGSYS, SIG_IGN);
331 #if defined(AFS_SGI_ENV)
332 rcode = afs_syscall(a3, a4, a5);
334 rcode = syscall (AFS_SYSCALL, 28 /* AFSCALL_CALL */, a3, a4, a5);
335 #endif /* AFS_SGI_ENV */
341 #endif /* AFS_AIX_ENV */
348 #ifndef AFS_DJGPP_ENV
350 struct ifreq ifs[ADDRSPERSITE];
351 struct ifreq ifreq, *ifr;
353 char buf[BUFSIZ], *cp, *cplim;
355 struct sockaddr_in *a;
356 #endif /* AFS_DJGPP_ENV */
368 bzero(rxi_NetAddrs, sizeof(rxi_NetAddrs));
369 bzero(myNetFlags, sizeof(myNetFlags));
370 bzero(myNetMTUs, sizeof(myNetMTUs));
371 bzero(myNetMasks, sizeof(myNetMasks));
374 s = socket(AF_INET, SOCK_DGRAM, 0);
377 #ifndef AFS_DJGPP_ENV
379 ifc.ifc_len = sizeof (buf);
383 ifc.ifc_len = sizeof(ifs);
384 ifc.ifc_buf = (caddr_t) &ifs[0];
385 bzero(&ifs[0], sizeof(ifs));
387 res = ioctl(s, SIOCGIFCONF, &ifc);
389 /* fputs(stderr, "ioctl error IFCONF\n"); */
397 #define size(p) MAX((p).sa_len, sizeof(p))
398 cplim = buf + ifc.ifc_len; /*skip over if's with big ifr_addr's */
399 for (cp = buf; cp < cplim ;
400 cp += sizeof(ifr->ifr_name) + MAX(a->sin_len, sizeof(*a))) {
401 if (rxi_numNetAddrs >= ADDRSPERSITE)
404 ifr = (struct ifreq *)cp;
406 len = ifc.ifc_len / sizeof(struct ifreq);
407 if (len > ADDRSPERSITE)
410 for (i = 0; i < len; ++i) {
412 res = ioctl(s, SIOCGIFADDR, ifr);
415 /* fputs(stderr, "ioctl error IFADDR\n");
416 perror(ifr->ifr_name); */
419 a = (struct sockaddr_in *) &ifr->ifr_addr;
420 if (a->sin_family != AF_INET) continue;
421 rxi_NetAddrs[rxi_numNetAddrs] = ntohl(a->sin_addr.s_addr);
422 if (rxi_NetAddrs[rxi_numNetAddrs] == 0x7f000001) {
423 /* we don't really care about "localhost" */
426 for (j=0; j < rxi_numNetAddrs; j++) {
427 if (rxi_NetAddrs[j] == rxi_NetAddrs[rxi_numNetAddrs])
430 if (j < rxi_numNetAddrs) continue;
432 /* fprintf(stderr, "if %s addr=%x\n", ifr->ifr_name,
433 rxi_NetAddrs[rxi_numNetAddrs]); */
436 res = ioctl(s, SIOCGIFFLAGS, ifr);
438 myNetFlags[rxi_numNetAddrs] = ifr->ifr_flags;
440 /* Handle aliased loopbacks as well. */
441 if (ifr->ifr_flags & IFF_LOOPBACK)
444 /* fprintf(stderr, "if %s flags=%x\n",
445 ifr->ifr_name, ifr->ifr_flags); */
448 fputs(stderr, "ioctl error IFFLAGS\n");
449 perror(ifr->ifr_name); */
451 #endif /* SIOCGIFFLAGS */
453 #if !defined(AFS_AIX_ENV) && !defined(AFS_LINUX20_ENV)
454 /* this won't run on an AIX system w/o a cache manager */
455 rxi_syscallp = rxi_syscall;
458 /* If I refer to kernel extensions that aren't loaded on AIX, the
459 * program refuses to load and run, so I simply can't include the
460 * following code. Fortunately, AIX is the one operating system in
461 * which the subsequent ioctl works reliably. */
463 if ((*rxi_syscallp)( 20 /*AFSOP_GETMTU*/,
464 htonl(rxi_NetAddrs[rxi_numNetAddrs]),
465 &(myNetMTUs[rxi_numNetAddrs]))) {
466 /* fputs(stderr, "syscall error GETMTU\n");
467 perror(ifr->ifr_name); */
468 myNetMTUs[rxi_numNetAddrs] = 0;
470 if ((*rxi_syscallp)(42 /*AFSOP_GETMASK*/,
471 htonl(rxi_NetAddrs[rxi_numNetAddrs]),
472 &(myNetMasks[rxi_numNetAddrs]))) {
473 /* fputs(stderr, "syscall error GETMASK\n");
474 perror(ifr->ifr_name); */
475 myNetMasks[rxi_numNetAddrs] = 0;
477 else myNetMasks[rxi_numNetAddrs] = ntohl( myNetMasks[rxi_numNetAddrs]);
478 /* fprintf(stderr, "if %s mask=0x%x\n",
479 ifr->ifr_name, myNetMasks[rxi_numNetAddrs]);*/
482 if (myNetMTUs[rxi_numNetAddrs] == 0) {
483 myNetMTUs[rxi_numNetAddrs] = OLD_MAX_PACKET_SIZE + RX_IPUDP_SIZE;
485 res = ioctl(s, SIOCGIFMTU, ifr);
486 if ((res == 0) && (ifr->ifr_metric > 128)) { /* sanity check */
487 myNetMTUs[rxi_numNetAddrs] = ifr->ifr_metric;
488 /* fprintf(stderr, "if %s mtu=%d\n",
489 ifr->ifr_name, ifr->ifr_metric); */
492 /* fputs(stderr, "ioctl error IFMTU\n");
493 perror(ifr->ifr_name); */
498 if (myNetMasks[rxi_numNetAddrs] == 0) {
499 myNetMasks[rxi_numNetAddrs] = fudge_netmask(rxi_NetAddrs[rxi_numNetAddrs]);
500 #ifdef SIOCGIFNETMASK
501 res = ioctl(s, SIOCGIFNETMASK, ifr);
503 a = (struct sockaddr_in *) &ifr->ifr_addr;
504 myNetMasks[rxi_numNetAddrs] = ntohl(a->sin_addr.s_addr);
505 /* fprintf(stderr, "if %s subnetmask=0x%x\n",
506 ifr->ifr_name, myNetMasks[rxi_numNetAddrs]); */
509 /* fputs(stderr, "ioctl error IFMASK\n");
510 perror(ifr->ifr_name); */
515 if (rxi_NetAddrs[rxi_numNetAddrs] != 0x7f000001) { /* ignore lo0 */
517 maxsize = rxi_nRecvFrags*(myNetMTUs[rxi_numNetAddrs] - RX_IP_SIZE);
518 maxsize -= UDP_HDR_SIZE; /* only the first frag has a UDP hdr */
519 if (rx_maxReceiveSize < maxsize)
520 rx_maxReceiveSize = MIN( RX_MAX_PACKET_SIZE, maxsize);
527 /* have to allocate at least enough to allow a single packet to reach its
528 * maximum size, so ReadPacket will work. Allocate enough for a couple
529 * of packets to do so, for good measure */
531 int npackets, ncbufs;
533 rx_maxJumboRecvSize = RX_HEADER_SIZE
534 + rxi_nDgramPackets * RX_JUMBOBUFFERSIZE
535 + (rxi_nDgramPackets-1) * RX_JUMBOHEADERSIZE;
536 rx_maxJumboRecvSize = MAX(rx_maxJumboRecvSize, rx_maxReceiveSize);
537 ncbufs = (rx_maxJumboRecvSize - RX_FIRSTBUFFERSIZE);
539 ncbufs = ncbufs / RX_CBUFFERSIZE;
540 npackets = rx_initSendWindow -1;
541 rxi_MorePackets(npackets*(ncbufs+1));
544 #else /* AFS_DJGPP_ENV */
547 #endif /* AFS_DJGPP_ENV */
549 #endif /* AFS_NT40_ENV */
551 /* Called from rxi_FindPeer, when initializing a clear rx_peer structure,
552 * to get interesting information.
553 * Curiously enough, the rx_peerHashTable_lock currently protects the
554 * Inited variable (and hence rx_GetIFInfo). When the fs suite uses
555 * pthreads, this issue will need to be revisited.
558 void rxi_InitPeerParams(struct rx_peer *pp)
570 * there's a race here since more than one thread could call
571 * rx_GetIFInfo. The race stops in rx_GetIFInfo.
579 /* try to second-guess IP, and identify which link is most likely to
580 * be used for traffic to/from this host. */
581 ppaddr = ntohl(pp->host);
585 pp->rateFlag = 2; /* start timing after two full packets */
586 /* I don't initialize these, because I presume they are bzero'd...
587 * pp->burstSize pp->burst pp->burstWait.sec pp->burstWait.usec
588 * pp->timeout.usec */
591 for (ix = 0; ix < rxi_numNetAddrs; ++ix) {
592 if ((rxi_NetAddrs[ix] & myNetMasks[ix]) == (ppaddr & myNetMasks[ix])) {
593 #ifdef IFF_POINTOPOINT
594 if (myNetFlags[ix] & IFF_POINTOPOINT)
596 #endif /* IFF_POINTOPOINT */
597 rxmtu = myNetMTUs[ix] - RX_IPUDP_SIZE;
598 if ( rxmtu < RX_MIN_PACKET_SIZE )
599 rxmtu = RX_MIN_PACKET_SIZE ;
600 if (pp->ifMTU < rxmtu)
601 pp->ifMTU = MIN(rx_MyMaxSendSize, rxmtu);
606 if (!pp->ifMTU) { /* not local */
608 pp->ifMTU = RX_REMOTE_PACKET_SIZE;
610 #else /* ADAPT_MTU */
611 pp->rateFlag = 2; /* start timing after two full packets */
613 pp->ifMTU = OLD_MAX_PACKET_SIZE;
614 #endif /* ADAPT_MTU */
615 pp->ifMTU = rxi_AdjustIfMTU(pp->ifMTU);
616 pp->maxMTU = OLD_MAX_PACKET_SIZE; /* for compatibility with old guys */
617 pp->natMTU = MIN((int)pp->ifMTU, OLD_MAX_PACKET_SIZE);
618 pp->maxDgramPackets = MIN(rxi_nDgramPackets,
619 rxi_AdjustDgramPackets(RX_MAX_FRAGS, pp->ifMTU));
620 pp->ifDgramPackets = MIN(rxi_nDgramPackets,
621 rxi_AdjustDgramPackets(RX_MAX_FRAGS, pp->ifMTU));
622 pp->maxDgramPackets = 1;
623 /* Initialize slow start parameters */
624 pp->MTU = MIN(pp->natMTU, pp->maxMTU);
626 pp->nDgramPackets = 1;
630 /* Don't expose jumobgram internals. */
631 void rx_SetNoJumbo(void)
633 rx_maxReceiveSize = OLD_MAX_PACKET_SIZE;
634 rxi_nSendFrags = rxi_nRecvFrags = 1;