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(u_int ahost, u_short port)
100 osi_socket socketFd = OSI_NULLSOCKET;
101 struct sockaddr_in taddr;
102 char *name = "rxi_GetUDPSocket: ";
105 #if !defined(AFS_NT40_ENV) && !defined(AFS_DJGPP_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, 0);
124 taddr.sin_addr.s_addr = ahost;
125 taddr.sin_family = AF_INET;
126 taddr.sin_port = (u_short) port;
127 #ifdef STRUCT_SOCKADDR_HAS_SA_LEN
128 taddr.sin_len = sizeof(struct sockaddr_in);
130 #define MAX_RX_BINDS 10
131 for (binds = 0; binds < MAX_RX_BINDS; binds++) {
134 code = bind(socketFd, (struct sockaddr *)&taddr, sizeof(taddr));
140 (osi_Msg "%sbind failed\n", name);
143 #if !defined(AFS_NT40_ENV) && !defined(AFS_DJGPP_ENV)
145 * Set close-on-exec on rx socket
147 fcntl(socketFd, F_SETFD, 1);
150 /* Use one of three different ways of getting a socket buffer expanded to
157 len2 = rx_UdpBufSize;
158 #ifndef AFS_DJGPP_ENV
161 (socketFd, SOL_SOCKET, SO_RCVBUF, (char *)&len2,
164 len2 = 32766; /* fall back to old size... uh-oh! */
169 (socketFd, SOL_SOCKET, SO_SNDBUF, (char *)&len1,
173 (socketFd, SOL_SOCKET, SO_RCVBUF, (char *)&len2,
175 #endif /* AFS_DJGPP_ENV */
178 #ifndef AFS_DJGPP_ENV
180 (osi_Msg "%s*WARNING* Unable to increase buffering on socket\n",
182 #endif /* AFS_DJGPP_ENV */
183 if (rxi_Listen(socketFd) < 0) {
192 closesocket(socketFd);
198 MUTEX_ENTER(&rx_stats_mutex);
199 rx_stats.socketGreedy = greedy;
200 MUTEX_EXIT(&rx_stats_mutex);
201 return OSI_NULLSOCKET;
205 rxi_GetUDPSocket(u_short port)
207 return rxi_GetHostUDPSocket(htonl(INADDR_ANY), port);
211 osi_Panic(char *msg, int a1, int a2, int a3)
213 (osi_Msg "Fatal Rx error: ");
214 (osi_Msg msg, a1, a2, a3);
221 * osi_AssertFailU() -- used by the osi_Assert() macro.
225 osi_AssertFailU(const char *expr, const char *file, int line)
227 osi_Panic("assertion failed: %s, file: %s, line: %d\n", (int)expr,
233 static const char memZero;
235 osi_Alloc(afs_int32 x)
238 * 0-length allocs may return NULL ptr from osi_kalloc, so we special-case
239 * things so that NULL returned iff an error occurred
243 return ((char *)malloc(x));
247 osi_Free(char *x, afs_int32 size)
254 #endif /* AFS_AIX32_ENV */
256 #define ADDRSPERSITE 16
259 afs_uint32 rxi_NetAddrs[ADDRSPERSITE]; /* host order */
260 static int myNetMTUs[ADDRSPERSITE];
261 static int myNetMasks[ADDRSPERSITE];
262 static int myNetFlags[ADDRSPERSITE];
263 u_int rxi_numNetAddrs;
264 static int Inited = 0;
266 #if defined(AFS_NT40_ENV) || defined(AFS_DJGPP_ENV)
270 if (rxi_numNetAddrs > 0)
271 return htonl(rxi_NetAddrs[0]);
277 ** return number of addresses
278 ** and the addresses themselves in the buffer
279 ** maxSize - max number of interfaces to return.
282 rx_getAllAddr(afs_int32 * buffer, int maxSize)
285 for (count = 0; count < rxi_numNetAddrs && maxSize > 0;
287 buffer[count] = htonl(rxi_NetAddrs[count]);
299 LOCK_IF_INIT if (Inited) {
300 UNLOCK_IF_INIT return;
304 int npackets, ncbufs;
308 UNLOCK_IF_INIT rxi_numNetAddrs = ADDRSPERSITE;
310 LOCK_IF(void) syscfg_GetIFInfo(&rxi_numNetAddrs, rxi_NetAddrs,
311 myNetMasks, myNetMTUs, myNetFlags);
313 for (i = 0; i < rxi_numNetAddrs; i++) {
314 rxsize = rxi_AdjustIfMTU(myNetMTUs[i] - RX_IPUDP_SIZE);
316 rxi_nRecvFrags * rxsize + (rxi_nRecvFrags - 1) * UDP_HDR_SIZE;
317 maxsize = rxi_AdjustMaxMTU(rxsize, maxsize);
318 if (rx_maxReceiveSize < maxsize) {
319 rx_maxReceiveSize = MIN(RX_MAX_PACKET_SIZE, maxsize);
321 MIN(rx_maxReceiveSize, rx_maxReceiveSizeUser);
325 UNLOCK_IF ncbufs = (rx_maxJumboRecvSize - RX_FIRSTBUFFERSIZE);
327 ncbufs = ncbufs / RX_CBUFFERSIZE;
328 npackets = rx_initSendWindow - 1;
329 rxi_MorePackets(npackets * (ncbufs + 1));
337 fudge_netmask(afs_uint32 addr)
343 else if (IN_CLASSB(addr))
345 else if (IN_CLASSC(addr))
355 #if !defined(AFS_AIX_ENV) && !defined(AFS_NT40_ENV) && !defined(AFS_LINUX20_ENV) && !defined(AFS_DJGPP_ENV)
357 rxi_syscall(a3, a4, a5)
364 old = (void (*)())signal(SIGSYS, SIG_IGN);
366 #if defined(AFS_SGI_ENV)
367 rcode = afs_syscall(a3, a4, a5);
369 rcode = syscall(AFS_SYSCALL, 28 /* AFSCALL_CALL */ , a3, a4, a5);
370 #endif /* AFS_SGI_ENV */
376 #endif /* AFS_AIX_ENV */
384 #ifndef AFS_DJGPP_ENV
386 struct ifreq ifs[ADDRSPERSITE];
389 char buf[BUFSIZ], *cp, *cplim;
391 struct sockaddr_in *a;
392 #endif /* AFS_DJGPP_ENV */
394 LOCK_IF_INIT if (Inited) {
395 UNLOCK_IF_INIT return;
398 UNLOCK_IF_INIT LOCK_IF rxi_numNetAddrs = 0;
399 memset(rxi_NetAddrs, 0, sizeof(rxi_NetAddrs));
400 memset(myNetFlags, 0, sizeof(myNetFlags));
401 memset(myNetMTUs, 0, sizeof(myNetMTUs));
402 memset(myNetMasks, 0, sizeof(myNetMasks));
403 UNLOCK_IF s = socket(AF_INET, SOCK_DGRAM, 0);
407 #ifndef AFS_DJGPP_ENV
409 ifc.ifc_len = sizeof(buf);
413 ifc.ifc_len = sizeof(ifs);
414 ifc.ifc_buf = (caddr_t) & ifs[0];
415 memset(&ifs[0], 0, sizeof(ifs));
417 res = ioctl(s, SIOCGIFCONF, &ifc);
419 /* fputs(stderr, "ioctl error IFCONF\n"); */
426 #define size(p) MAX((p).sa_len, sizeof(p))
427 cplim = buf + ifc.ifc_len; /*skip over if's with big ifr_addr's */
428 for (cp = buf; cp < cplim;
429 cp += sizeof(ifr->ifr_name) + MAX(a->sin_len, sizeof(*a))) {
430 if (rxi_numNetAddrs >= ADDRSPERSITE)
433 ifr = (struct ifreq *)cp;
435 len = ifc.ifc_len / sizeof(struct ifreq);
436 if (len > ADDRSPERSITE)
439 for (i = 0; i < len; ++i) {
441 res = ioctl(s, SIOCGIFADDR, ifr);
444 /* fputs(stderr, "ioctl error IFADDR\n");
445 * perror(ifr->ifr_name); */
448 a = (struct sockaddr_in *)&ifr->ifr_addr;
449 if (a->sin_family != AF_INET)
451 rxi_NetAddrs[rxi_numNetAddrs] = ntohl(a->sin_addr.s_addr);
452 if (rxi_NetAddrs[rxi_numNetAddrs] == 0x7f000001) {
453 /* we don't really care about "localhost" */
456 for (j = 0; j < rxi_numNetAddrs; j++) {
457 if (rxi_NetAddrs[j] == rxi_NetAddrs[rxi_numNetAddrs])
460 if (j < rxi_numNetAddrs)
463 /* fprintf(stderr, "if %s addr=%x\n", ifr->ifr_name,
464 * rxi_NetAddrs[rxi_numNetAddrs]); */
467 res = ioctl(s, SIOCGIFFLAGS, ifr);
469 myNetFlags[rxi_numNetAddrs] = ifr->ifr_flags;
471 /* Handle aliased loopbacks as well. */
472 if (ifr->ifr_flags & IFF_LOOPBACK)
475 /* fprintf(stderr, "if %s flags=%x\n",
476 * ifr->ifr_name, ifr->ifr_flags); */
478 * fputs(stderr, "ioctl error IFFLAGS\n");
479 * perror(ifr->ifr_name); */
481 #endif /* SIOCGIFFLAGS */
483 #if !defined(AFS_AIX_ENV) && !defined(AFS_LINUX20_ENV)
484 /* this won't run on an AIX system w/o a cache manager */
485 rxi_syscallp = rxi_syscall;
488 /* If I refer to kernel extensions that aren't loaded on AIX, the
489 * program refuses to load and run, so I simply can't include the
490 * following code. Fortunately, AIX is the one operating system in
491 * which the subsequent ioctl works reliably. */
493 if ((*rxi_syscallp) (20 /*AFSOP_GETMTU */ ,
494 htonl(rxi_NetAddrs[rxi_numNetAddrs]),
495 &(myNetMTUs[rxi_numNetAddrs]))) {
496 /* fputs(stderr, "syscall error GETMTU\n");
497 * perror(ifr->ifr_name); */
498 myNetMTUs[rxi_numNetAddrs] = 0;
500 if ((*rxi_syscallp) (42 /*AFSOP_GETMASK */ ,
501 htonl(rxi_NetAddrs[rxi_numNetAddrs]),
502 &(myNetMasks[rxi_numNetAddrs]))) {
503 /* fputs(stderr, "syscall error GETMASK\n");
504 * perror(ifr->ifr_name); */
505 myNetMasks[rxi_numNetAddrs] = 0;
507 myNetMasks[rxi_numNetAddrs] =
508 ntohl(myNetMasks[rxi_numNetAddrs]);
509 /* fprintf(stderr, "if %s mask=0x%x\n",
510 * ifr->ifr_name, myNetMasks[rxi_numNetAddrs]); */
513 if (myNetMTUs[rxi_numNetAddrs] == 0) {
514 myNetMTUs[rxi_numNetAddrs] = OLD_MAX_PACKET_SIZE + RX_IPUDP_SIZE;
516 res = ioctl(s, SIOCGIFMTU, ifr);
517 if ((res == 0) && (ifr->ifr_metric > 128)) { /* sanity check */
518 myNetMTUs[rxi_numNetAddrs] = ifr->ifr_metric;
519 /* fprintf(stderr, "if %s mtu=%d\n",
520 * ifr->ifr_name, ifr->ifr_metric); */
522 /* fputs(stderr, "ioctl error IFMTU\n");
523 * perror(ifr->ifr_name); */
528 if (myNetMasks[rxi_numNetAddrs] == 0) {
529 myNetMasks[rxi_numNetAddrs] =
530 fudge_netmask(rxi_NetAddrs[rxi_numNetAddrs]);
531 #ifdef SIOCGIFNETMASK
532 res = ioctl(s, SIOCGIFNETMASK, ifr);
534 a = (struct sockaddr_in *)&ifr->ifr_addr;
535 myNetMasks[rxi_numNetAddrs] = ntohl(a->sin_addr.s_addr);
536 /* fprintf(stderr, "if %s subnetmask=0x%x\n",
537 * ifr->ifr_name, myNetMasks[rxi_numNetAddrs]); */
539 /* fputs(stderr, "ioctl error IFMASK\n");
540 * perror(ifr->ifr_name); */
545 if (rxi_NetAddrs[rxi_numNetAddrs] != 0x7f000001) { /* ignore lo0 */
548 rxi_nRecvFrags * (myNetMTUs[rxi_numNetAddrs] - RX_IP_SIZE);
549 maxsize -= UDP_HDR_SIZE; /* only the first frag has a UDP hdr */
550 if (rx_maxReceiveSize < maxsize)
551 rx_maxReceiveSize = MIN(RX_MAX_PACKET_SIZE, maxsize);
557 /* have to allocate at least enough to allow a single packet to reach its
558 * maximum size, so ReadPacket will work. Allocate enough for a couple
559 * of packets to do so, for good measure */
561 int npackets, ncbufs;
563 rx_maxJumboRecvSize =
564 RX_HEADER_SIZE + rxi_nDgramPackets * RX_JUMBOBUFFERSIZE +
565 (rxi_nDgramPackets - 1) * RX_JUMBOHEADERSIZE;
566 rx_maxJumboRecvSize = MAX(rx_maxJumboRecvSize, rx_maxReceiveSize);
567 ncbufs = (rx_maxJumboRecvSize - RX_FIRSTBUFFERSIZE);
569 ncbufs = ncbufs / RX_CBUFFERSIZE;
570 npackets = rx_initSendWindow - 1;
571 rxi_MorePackets(npackets * (ncbufs + 1));
574 #else /* AFS_DJGPP_ENV */
577 #endif /* AFS_DJGPP_ENV */
579 #endif /* AFS_NT40_ENV */
581 /* Called from rxi_FindPeer, when initializing a clear rx_peer structure,
582 * to get interesting information.
583 * Curiously enough, the rx_peerHashTable_lock currently protects the
584 * Inited variable (and hence rx_GetIFInfo). When the fs suite uses
585 * pthreads, this issue will need to be revisited.
589 rxi_InitPeerParams(struct rx_peer *pp)
597 LOCK_IF_INIT if (!Inited) {
600 * there's a race here since more than one thread could call
601 * rx_GetIFInfo. The race stops in rx_GetIFInfo.
608 /* try to second-guess IP, and identify which link is most likely to
609 * be used for traffic to/from this host. */
610 ppaddr = ntohl(pp->host);
614 pp->rateFlag = 2; /* start timing after two full packets */
615 /* I don't initialize these, because I presume they are bzero'd...
616 * pp->burstSize pp->burst pp->burstWait.sec pp->burstWait.usec
617 * pp->timeout.usec */
619 LOCK_IF for (ix = 0; ix < rxi_numNetAddrs; ++ix) {
620 if ((rxi_NetAddrs[ix] & myNetMasks[ix]) == (ppaddr & myNetMasks[ix])) {
621 #ifdef IFF_POINTOPOINT
622 if (myNetFlags[ix] & IFF_POINTOPOINT)
624 #endif /* IFF_POINTOPOINT */
625 rxmtu = myNetMTUs[ix] - RX_IPUDP_SIZE;
626 if (rxmtu < RX_MIN_PACKET_SIZE)
627 rxmtu = RX_MIN_PACKET_SIZE;
628 if (pp->ifMTU < rxmtu)
629 pp->ifMTU = MIN(rx_MyMaxSendSize, rxmtu);
632 UNLOCK_IF if (!pp->ifMTU) { /* not local */
634 pp->ifMTU = MIN(rx_MyMaxSendSize, RX_REMOTE_PACKET_SIZE);
636 #else /* ADAPT_MTU */
637 pp->rateFlag = 2; /* start timing after two full packets */
639 pp->ifMTU = MIN(rx_MyMaxSendSize, OLD_MAX_PACKET_SIZE);
640 #endif /* ADAPT_MTU */
641 pp->ifMTU = rxi_AdjustIfMTU(pp->ifMTU);
642 pp->maxMTU = OLD_MAX_PACKET_SIZE; /* for compatibility with old guys */
643 pp->natMTU = MIN((int)pp->ifMTU, OLD_MAX_PACKET_SIZE);
644 pp->maxDgramPackets =
645 MIN(rxi_nDgramPackets,
646 rxi_AdjustDgramPackets(RX_MAX_FRAGS, pp->ifMTU));
648 MIN(rxi_nDgramPackets,
649 rxi_AdjustDgramPackets(RX_MAX_FRAGS, pp->ifMTU));
650 pp->maxDgramPackets = 1;
651 /* Initialize slow start parameters */
652 pp->MTU = MIN(pp->natMTU, pp->maxMTU);
654 pp->nDgramPackets = 1;
658 /* Don't expose jumobgram internals. */
662 rx_maxReceiveSize = OLD_MAX_PACKET_SIZE;
663 rxi_nSendFrags = rxi_nRecvFrags = 1;
666 /* Override max MTU. If rx_SetNoJumbo is called, it must be
667 called before calling rx_SetMaxMTU since SetNoJumbo clobbers rx_maxReceiveSize */
668 void rx_SetMaxMTU(int mtu)
670 rx_MyMaxSendSize = rx_maxReceiveSizeUser = rx_maxReceiveSize = mtu;