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>
17 # include <sys/types.h>
21 # include <WINNT/syscfg.h>
23 # include <sys/socket.h>
24 # include <sys/file.h>
26 # include <sys/stat.h>
27 # include <netinet/in.h>
28 # include <sys/time.h>
30 # include <sys/ioctl.h>
33 #if !defined(AFS_AIX_ENV) && !defined(AFS_NT40_ENV) && !defined(AFS_DJGPP_ENV)
34 # include <sys/syscall.h>
36 #include <afs/afs_args.h>
37 #include <afs/afsutil.h>
46 #ifndef IPPORT_USERRESERVED
47 /* If in.h doesn't define this, define it anyway. Unfortunately, defining
48 this doesn't put the code into the kernel to restrict kernel assigned
49 port numbers to numbers below IPPORT_USERRESERVED... */
50 #define IPPORT_USERRESERVED 5000
54 # include <sys/time.h>
57 # include "rx_globals.h"
59 #ifdef AFS_PTHREAD_ENV
63 * The rx_if_init_mutex mutex protects the following global variables:
67 pthread_mutex_t rx_if_init_mutex;
68 #define LOCK_IF_INIT assert(pthread_mutex_lock(&rx_if_init_mutex)==0);
69 #define UNLOCK_IF_INIT assert(pthread_mutex_unlock(&rx_if_init_mutex)==0);
72 * The rx_if_mutex mutex protects the following global variables:
78 pthread_mutex_t rx_if_mutex;
79 #define LOCK_IF assert(pthread_mutex_lock(&rx_if_mutex)==0);
80 #define UNLOCK_IF assert(pthread_mutex_unlock(&rx_if_mutex)==0);
83 #define UNLOCK_IF_INIT
86 #endif /* AFS_PTHREAD_ENV */
90 * Make a socket for receiving/sending IP packets. Set it into non-blocking
91 * and large buffering modes. If port isn't specified, the kernel will pick
92 * one. Returns the socket (>= 0) on success. Returns OSI_NULLSOCKET on
93 * failure. Port must be in network byte order.
95 osi_socket rxi_GetUDPSocket(u_short port)
98 osi_socket socketFd = OSI_NULLSOCKET;
99 struct sockaddr_in taddr;
100 char *name = "rxi_GetUDPSocket: ";
103 #if !defined(AFS_NT40_ENV) && !defined(AFS_DJGPP_ENV)
104 if (ntohs(port) >= IPPORT_RESERVED && ntohs(port) < IPPORT_USERRESERVED) {
105 /* (osi_Msg "%s*WARNING* port number %d is not a reserved port number. Use port numbers above %d\n", name, port, IPPORT_USERRESERVED);
108 if (ntohs(port) > 0 && ntohs(port) < IPPORT_RESERVED && geteuid() != 0) {
109 (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);
113 socketFd = socket(AF_INET, SOCK_DGRAM, 0);
120 taddr.sin_addr.s_addr = 0;
121 taddr.sin_family = AF_INET;
122 taddr.sin_port = (u_short)port;
123 #define MAX_RX_BINDS 10
124 for (binds=0; binds<MAX_RX_BINDS; binds++) {
125 if (binds) rxi_Delay (10);
126 code = bind(socketFd, (struct sockaddr *) &taddr, sizeof(taddr));
131 (osi_Msg "%sbind failed\n", name);
135 #if !defined(AFS_NT40_ENV) && !defined(AFS_DJGPP_ENV)
137 * Set close-on-exec on rx socket
139 fcntl(socketFd, F_SETFD, 1);
142 /* Use one of three different ways of getting a socket buffer expanded to
149 len2 = rx_UdpBufSize;
150 #ifndef AFS_DJGPP_ENV
152 (setsockopt(socketFd, SOL_SOCKET, SO_RCVBUF, (char *)&len2,
155 len2=32766; /* fall back to old size... uh-oh! */
159 (setsockopt(socketFd, SOL_SOCKET, SO_SNDBUF, (char *)&len1,
160 sizeof(len1)) >= 0) &&
161 (setsockopt(socketFd, SOL_SOCKET, SO_RCVBUF, (char *)&len2,
163 #endif /* AFS_DJGPP_ENV */
166 #ifndef AFS_DJGPP_ENV
168 (osi_Msg "%s*WARNING* Unable to increase buffering on socket\n", name);
169 #endif /* AFS_DJGPP_ENV */
170 if (rxi_Listen(socketFd) < 0) {
178 if (socketFd >= 0) closesocket(socketFd);
180 if (socketFd >= 0) close(socketFd);
183 MUTEX_ENTER(&rx_stats_mutex);
184 rx_stats.socketGreedy = greedy;
185 MUTEX_EXIT(&rx_stats_mutex);
186 return OSI_NULLSOCKET;
189 void osi_Panic(msg, a1, a2, a3)
192 (osi_Msg "Fatal Rx error: ");
193 (osi_Msg msg, a1, a2, a3);
200 * osi_AssertFailU() -- used by the osi_Assert() macro.
203 void osi_AssertFailU(const char *expr, const char *file, int line)
205 osi_Panic("assertion failed: %s, file: %s, line: %d\n", expr, file, line);
210 static const char memZero;
215 * 0-length allocs may return NULL ptr from osi_kalloc, so we special-case
216 * things so that NULL returned iff an error occurred
218 if (x == 0) return &memZero;
219 return ((char *) malloc(x));
226 if (x == &memZero) return;
230 #endif /* AFS_AIX32_ENV */
232 #define ADDRSPERSITE 16
235 afs_uint32 rxi_NetAddrs[ADDRSPERSITE]; /* host order */
236 static int myNetMTUs[ADDRSPERSITE];
237 static int myNetMasks[ADDRSPERSITE];
238 static int myNetFlags[ADDRSPERSITE];
239 u_int rxi_numNetAddrs;
240 static int Inited = 0;
242 #if defined(AFS_NT40_ENV) || defined(AFS_DJGPP_ENV)
243 int rxi_getaddr(void)
245 if (rxi_numNetAddrs > 0)
246 return htonl(rxi_NetAddrs[0]);
252 ** return number of addresses
253 ** and the addresses themselves in the buffer
254 ** maxSize - max number of interfaces to return.
256 int rx_getAllAddr (afs_int32 *buffer, int maxSize)
259 for (count = 0; count < rxi_numNetAddrs && maxSize > 0; count++, maxSize--)
260 buffer[count] = htonl(rxi_NetAddrs[count]);
269 void rx_GetIFInfo(void)
279 int npackets, ncbufs;
284 rxi_numNetAddrs = ADDRSPERSITE;
287 (void) syscfg_GetIFInfo(&rxi_numNetAddrs, rxi_NetAddrs,
288 myNetMasks, myNetMTUs, myNetFlags);
290 for (i=0; i<rxi_numNetAddrs; i++) {
291 rxsize = rxi_AdjustIfMTU(myNetMTUs[i] - RX_IPUDP_SIZE);
292 maxsize = rxi_nRecvFrags*rxsize + (rxi_nRecvFrags-1)*UDP_HDR_SIZE;
293 maxsize = rxi_AdjustMaxMTU(rxsize, maxsize);
294 if (rx_maxReceiveSize < maxsize) {
295 rx_maxReceiveSize = MIN( RX_MAX_PACKET_SIZE, maxsize);
296 rx_maxReceiveSize = MIN( rx_maxReceiveSize,
297 rx_maxReceiveSizeUser);
302 ncbufs = (rx_maxJumboRecvSize - RX_FIRSTBUFFERSIZE);
304 ncbufs = ncbufs / RX_CBUFFERSIZE;
305 npackets = rx_initSendWindow -1;
306 rxi_MorePackets(npackets*(ncbufs+1));
313 static afs_uint32 fudge_netmask(addr)
318 if (IN_CLASSA(addr)) msk = IN_CLASSA_NET;
319 else if (IN_CLASSB(addr)) msk = IN_CLASSB_NET;
320 else if (IN_CLASSC(addr)) msk = IN_CLASSC_NET;
328 #if !defined(AFS_AIX_ENV) && !defined(AFS_NT40_ENV) && !defined(AFS_LINUX20_ENV) && !defined(AFS_DJGPP_ENV)
329 int rxi_syscall(a3, a4, a5)
336 old = (void (*)())signal(SIGSYS, SIG_IGN);
338 #if defined(AFS_SGI_ENV)
339 rcode = afs_syscall(a3, a4, a5);
341 rcode = syscall (AFS_SYSCALL, 28 /* AFSCALL_CALL */, a3, a4, a5);
342 #endif /* AFS_SGI_ENV */
348 #endif /* AFS_AIX_ENV */
355 #ifndef AFS_DJGPP_ENV
357 struct ifreq ifs[ADDRSPERSITE];
360 char buf[BUFSIZ], *cp, *cplim;
362 struct sockaddr_in *a;
363 #endif /* AFS_DJGPP_ENV */
375 memset(rxi_NetAddrs, 0, sizeof(rxi_NetAddrs));
376 memset(myNetFlags, 0, sizeof(myNetFlags));
377 memset(myNetMTUs, 0, sizeof(myNetMTUs));
378 memset(myNetMasks, 0, sizeof(myNetMasks));
381 s = socket(AF_INET, SOCK_DGRAM, 0);
384 #ifndef AFS_DJGPP_ENV
386 ifc.ifc_len = sizeof (buf);
390 ifc.ifc_len = sizeof(ifs);
391 ifc.ifc_buf = (caddr_t) &ifs[0];
392 memset(&ifs[0], 0, sizeof(ifs));
394 res = ioctl(s, SIOCGIFCONF, &ifc);
396 /* fputs(stderr, "ioctl error IFCONF\n"); */
404 #define size(p) MAX((p).sa_len, sizeof(p))
405 cplim = buf + ifc.ifc_len; /*skip over if's with big ifr_addr's */
406 for (cp = buf; cp < cplim ;
407 cp += sizeof(ifr->ifr_name) + MAX(a->sin_len, sizeof(*a))) {
408 if (rxi_numNetAddrs >= ADDRSPERSITE)
411 ifr = (struct ifreq *)cp;
413 len = ifc.ifc_len / sizeof(struct ifreq);
414 if (len > ADDRSPERSITE)
417 for (i = 0; i < len; ++i) {
419 res = ioctl(s, SIOCGIFADDR, ifr);
422 /* fputs(stderr, "ioctl error IFADDR\n");
423 perror(ifr->ifr_name); */
426 a = (struct sockaddr_in *) &ifr->ifr_addr;
427 if (a->sin_family != AF_INET) continue;
428 rxi_NetAddrs[rxi_numNetAddrs] = ntohl(a->sin_addr.s_addr);
429 if (rxi_NetAddrs[rxi_numNetAddrs] == 0x7f000001) {
430 /* we don't really care about "localhost" */
433 for (j=0; j < rxi_numNetAddrs; j++) {
434 if (rxi_NetAddrs[j] == rxi_NetAddrs[rxi_numNetAddrs])
437 if (j < rxi_numNetAddrs) continue;
439 /* fprintf(stderr, "if %s addr=%x\n", ifr->ifr_name,
440 rxi_NetAddrs[rxi_numNetAddrs]); */
443 res = ioctl(s, SIOCGIFFLAGS, ifr);
445 myNetFlags[rxi_numNetAddrs] = ifr->ifr_flags;
447 /* Handle aliased loopbacks as well. */
448 if (ifr->ifr_flags & IFF_LOOPBACK)
451 /* fprintf(stderr, "if %s flags=%x\n",
452 ifr->ifr_name, ifr->ifr_flags); */
455 fputs(stderr, "ioctl error IFFLAGS\n");
456 perror(ifr->ifr_name); */
458 #endif /* SIOCGIFFLAGS */
460 #if !defined(AFS_AIX_ENV) && !defined(AFS_LINUX20_ENV)
461 /* this won't run on an AIX system w/o a cache manager */
462 rxi_syscallp = rxi_syscall;
465 /* If I refer to kernel extensions that aren't loaded on AIX, the
466 * program refuses to load and run, so I simply can't include the
467 * following code. Fortunately, AIX is the one operating system in
468 * which the subsequent ioctl works reliably. */
470 if ((*rxi_syscallp)( 20 /*AFSOP_GETMTU*/,
471 htonl(rxi_NetAddrs[rxi_numNetAddrs]),
472 &(myNetMTUs[rxi_numNetAddrs]))) {
473 /* fputs(stderr, "syscall error GETMTU\n");
474 perror(ifr->ifr_name); */
475 myNetMTUs[rxi_numNetAddrs] = 0;
477 if ((*rxi_syscallp)(42 /*AFSOP_GETMASK*/,
478 htonl(rxi_NetAddrs[rxi_numNetAddrs]),
479 &(myNetMasks[rxi_numNetAddrs]))) {
480 /* fputs(stderr, "syscall error GETMASK\n");
481 perror(ifr->ifr_name); */
482 myNetMasks[rxi_numNetAddrs] = 0;
484 else myNetMasks[rxi_numNetAddrs] = ntohl( myNetMasks[rxi_numNetAddrs]);
485 /* fprintf(stderr, "if %s mask=0x%x\n",
486 ifr->ifr_name, myNetMasks[rxi_numNetAddrs]);*/
489 if (myNetMTUs[rxi_numNetAddrs] == 0) {
490 myNetMTUs[rxi_numNetAddrs] = OLD_MAX_PACKET_SIZE + RX_IPUDP_SIZE;
492 res = ioctl(s, SIOCGIFMTU, ifr);
493 if ((res == 0) && (ifr->ifr_metric > 128)) { /* sanity check */
494 myNetMTUs[rxi_numNetAddrs] = ifr->ifr_metric;
495 /* fprintf(stderr, "if %s mtu=%d\n",
496 ifr->ifr_name, ifr->ifr_metric); */
499 /* fputs(stderr, "ioctl error IFMTU\n");
500 perror(ifr->ifr_name); */
505 if (myNetMasks[rxi_numNetAddrs] == 0) {
506 myNetMasks[rxi_numNetAddrs] = fudge_netmask(rxi_NetAddrs[rxi_numNetAddrs]);
507 #ifdef SIOCGIFNETMASK
508 res = ioctl(s, SIOCGIFNETMASK, ifr);
510 a = (struct sockaddr_in *) &ifr->ifr_addr;
511 myNetMasks[rxi_numNetAddrs] = ntohl(a->sin_addr.s_addr);
512 /* fprintf(stderr, "if %s subnetmask=0x%x\n",
513 ifr->ifr_name, myNetMasks[rxi_numNetAddrs]); */
516 /* fputs(stderr, "ioctl error IFMASK\n");
517 perror(ifr->ifr_name); */
522 if (rxi_NetAddrs[rxi_numNetAddrs] != 0x7f000001) { /* ignore lo0 */
524 maxsize = rxi_nRecvFrags*(myNetMTUs[rxi_numNetAddrs] - RX_IP_SIZE);
525 maxsize -= UDP_HDR_SIZE; /* only the first frag has a UDP hdr */
526 if (rx_maxReceiveSize < maxsize)
527 rx_maxReceiveSize = MIN( RX_MAX_PACKET_SIZE, maxsize);
534 /* have to allocate at least enough to allow a single packet to reach its
535 * maximum size, so ReadPacket will work. Allocate enough for a couple
536 * of packets to do so, for good measure */
538 int npackets, ncbufs;
540 rx_maxJumboRecvSize = RX_HEADER_SIZE
541 + rxi_nDgramPackets * RX_JUMBOBUFFERSIZE
542 + (rxi_nDgramPackets-1) * RX_JUMBOHEADERSIZE;
543 rx_maxJumboRecvSize = MAX(rx_maxJumboRecvSize, rx_maxReceiveSize);
544 ncbufs = (rx_maxJumboRecvSize - RX_FIRSTBUFFERSIZE);
546 ncbufs = ncbufs / RX_CBUFFERSIZE;
547 npackets = rx_initSendWindow -1;
548 rxi_MorePackets(npackets*(ncbufs+1));
551 #else /* AFS_DJGPP_ENV */
554 #endif /* AFS_DJGPP_ENV */
556 #endif /* AFS_NT40_ENV */
558 /* Called from rxi_FindPeer, when initializing a clear rx_peer structure,
559 * to get interesting information.
560 * Curiously enough, the rx_peerHashTable_lock currently protects the
561 * Inited variable (and hence rx_GetIFInfo). When the fs suite uses
562 * pthreads, this issue will need to be revisited.
565 void rxi_InitPeerParams(struct rx_peer *pp)
577 * there's a race here since more than one thread could call
578 * rx_GetIFInfo. The race stops in rx_GetIFInfo.
586 /* try to second-guess IP, and identify which link is most likely to
587 * be used for traffic to/from this host. */
588 ppaddr = ntohl(pp->host);
592 pp->rateFlag = 2; /* start timing after two full packets */
593 /* I don't initialize these, because I presume they are bzero'd...
594 * pp->burstSize pp->burst pp->burstWait.sec pp->burstWait.usec
595 * pp->timeout.usec */
598 for (ix = 0; ix < rxi_numNetAddrs; ++ix) {
599 if ((rxi_NetAddrs[ix] & myNetMasks[ix]) == (ppaddr & myNetMasks[ix])) {
600 #ifdef IFF_POINTOPOINT
601 if (myNetFlags[ix] & IFF_POINTOPOINT)
603 #endif /* IFF_POINTOPOINT */
604 rxmtu = myNetMTUs[ix] - RX_IPUDP_SIZE;
605 if ( rxmtu < RX_MIN_PACKET_SIZE )
606 rxmtu = RX_MIN_PACKET_SIZE ;
607 if (pp->ifMTU < rxmtu)
608 pp->ifMTU = MIN(rx_MyMaxSendSize, rxmtu);
613 if (!pp->ifMTU) { /* not local */
615 pp->ifMTU = RX_REMOTE_PACKET_SIZE;
617 #else /* ADAPT_MTU */
618 pp->rateFlag = 2; /* start timing after two full packets */
620 pp->ifMTU = OLD_MAX_PACKET_SIZE;
621 #endif /* ADAPT_MTU */
622 pp->ifMTU = rxi_AdjustIfMTU(pp->ifMTU);
623 pp->maxMTU = OLD_MAX_PACKET_SIZE; /* for compatibility with old guys */
624 pp->natMTU = MIN((int)pp->ifMTU, OLD_MAX_PACKET_SIZE);
625 pp->maxDgramPackets = MIN(rxi_nDgramPackets,
626 rxi_AdjustDgramPackets(RX_MAX_FRAGS, pp->ifMTU));
627 pp->ifDgramPackets = MIN(rxi_nDgramPackets,
628 rxi_AdjustDgramPackets(RX_MAX_FRAGS, pp->ifMTU));
629 pp->maxDgramPackets = 1;
630 /* Initialize slow start parameters */
631 pp->MTU = MIN(pp->natMTU, pp->maxMTU);
633 pp->nDgramPackets = 1;
637 /* Don't expose jumobgram internals. */
638 void rx_SetNoJumbo(void)
640 rx_maxReceiveSize = OLD_MAX_PACKET_SIZE;
641 rxi_nSendFrags = rxi_nRecvFrags = 1;