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>
23 # include <WINNT/syscfg.h>
25 # include <sys/socket.h>
26 # include <sys/file.h>
28 # include <sys/stat.h>
29 # include <netinet/in.h>
30 # include <sys/time.h>
32 # include <sys/ioctl.h>
35 #if !defined(AFS_AIX_ENV) && !defined(AFS_NT40_ENV)
36 # include <sys/syscall.h>
38 #include <afs/afs_args.h>
39 #include <afs/afsutil.h>
41 #ifndef IPPORT_USERRESERVED
42 /* If in.h doesn't define this, define it anyway. Unfortunately, defining
43 this doesn't put the code into the kernel to restrict kernel assigned
44 port numbers to numbers below IPPORT_USERRESERVED... */
45 #define IPPORT_USERRESERVED 5000
49 # include <sys/time.h>
52 # include "rx_globals.h"
54 #ifdef AFS_PTHREAD_ENV
58 * The rx_if_init_mutex mutex protects the following global variables:
62 pthread_mutex_t rx_if_init_mutex;
63 #define LOCK_IF_INIT assert(pthread_mutex_lock(&rx_if_init_mutex)==0)
64 #define UNLOCK_IF_INIT assert(pthread_mutex_unlock(&rx_if_init_mutex)==0)
67 * The rx_if_mutex mutex protects the following global variables:
73 pthread_mutex_t rx_if_mutex;
74 #define LOCK_IF assert(pthread_mutex_lock(&rx_if_mutex)==0)
75 #define UNLOCK_IF assert(pthread_mutex_unlock(&rx_if_mutex)==0)
78 #define UNLOCK_IF_INIT
81 #endif /* AFS_PTHREAD_ENV */
85 * Make a socket for receiving/sending IP packets. Set it into non-blocking
86 * and large buffering modes. If port isn't specified, the kernel will pick
87 * one. Returns the socket (>= 0) on success. Returns OSI_NULLSOCKET on
88 * failure. Port must be in network byte order.
91 rxi_GetHostUDPSocket(u_int ahost, u_short port)
94 osi_socket socketFd = OSI_NULLSOCKET;
95 struct sockaddr_in taddr;
96 char *name = "rxi_GetUDPSocket: ";
97 #ifdef AFS_LINUX22_ENV
98 int pmtu=IP_PMTUDISC_DONT;
101 #if !defined(AFS_NT40_ENV)
102 if (ntohs(port) >= IPPORT_RESERVED && ntohs(port) < IPPORT_USERRESERVED) {
103 /* (osi_Msg "%s*WARNING* port number %d is not a reserved port number. Use port numbers above %d\n", name, port, IPPORT_USERRESERVED);
106 if (ntohs(port) > 0 && ntohs(port) < IPPORT_RESERVED && geteuid() != 0) {
108 "%sport number %d is a reserved port number which may only be used by root. Use port numbers above %d\n",
109 name, ntohs(port), IPPORT_USERRESERVED);
113 socketFd = socket(AF_INET, SOCK_DGRAM, 0);
120 taddr.sin_addr.s_addr = ahost;
121 taddr.sin_family = AF_INET;
122 taddr.sin_port = (u_short) port;
123 #ifdef STRUCT_SOCKADDR_HAS_SA_LEN
124 taddr.sin_len = sizeof(struct sockaddr_in);
126 #define MAX_RX_BINDS 10
127 for (binds = 0; binds < MAX_RX_BINDS; binds++) {
130 code = bind(socketFd, (struct sockaddr *)&taddr, sizeof(taddr));
136 (osi_Msg "%sbind failed\n", name);
139 #if !defined(AFS_NT40_ENV)
141 * Set close-on-exec on rx socket
143 fcntl(socketFd, F_SETFD, 1);
146 /* Use one of three different ways of getting a socket buffer expanded to
154 len2 = rx_UdpBufSize;
157 (socketFd, SOL_SOCKET, SO_RCVBUF, (char *)&len2,
160 len2 = 32766; /* fall back to old size... uh-oh! */
165 (socketFd, SOL_SOCKET, SO_SNDBUF, (char *)&len1,
169 (socketFd, SOL_SOCKET, SO_RCVBUF, (char *)&len2,
172 (osi_Msg "%s*WARNING* Unable to increase buffering on socket\n",
174 MUTEX_ENTER(&rx_stats_mutex);
175 rx_stats.socketGreedy = greedy;
176 MUTEX_EXIT(&rx_stats_mutex);
179 #ifdef AFS_LINUX22_ENV
180 setsockopt(socketFd, SOL_IP, IP_MTU_DISCOVER, &pmtu, sizeof(pmtu));
183 if (rxi_Listen(socketFd) < 0) {
192 closesocket(socketFd);
198 return OSI_NULLSOCKET;
202 rxi_GetUDPSocket(u_short port)
204 return rxi_GetHostUDPSocket(htonl(INADDR_ANY), port);
208 osi_Panic(msg, a1, a2, a3)
211 (osi_Msg "Fatal Rx error: ");
212 (osi_Msg msg, a1, a2, a3);
219 * osi_AssertFailU() -- used by the osi_Assert() macro.
223 osi_AssertFailU(const char *expr, const char *file, int line)
225 osi_Panic("assertion failed: %s, file: %s, line: %d\n", expr,
229 #if defined(AFS_AIX32_ENV) && !defined(KERNEL)
231 static const char memZero;
233 osi_Alloc(afs_int32 x)
236 * 0-length allocs may return NULL ptr from malloc, so we special-case
237 * things so that NULL returned iff an error occurred
240 return (void *)&memZero;
245 osi_Free(void *x, afs_int32 size)
252 #endif /* defined(AFS_AIX32_ENV) && !defined(KERNEL) */
254 #define ADDRSPERSITE 16
257 static afs_uint32 rxi_NetAddrs[ADDRSPERSITE]; /* host order */
258 static int myNetMTUs[ADDRSPERSITE];
259 static int myNetMasks[ADDRSPERSITE];
260 static int myNetFlags[ADDRSPERSITE];
261 static u_int rxi_numNetAddrs;
262 static int Inited = 0;
264 #if defined(AFS_NT40_ENV)
268 /* The IP address list can change so we must query for it */
271 /* we don't want to use the loopback adapter which is first */
272 /* this is a bad bad hack */
273 if (rxi_numNetAddrs > 1)
274 return htonl(rxi_NetAddrs[1]);
275 else if (rxi_numNetAddrs > 0)
276 return htonl(rxi_NetAddrs[0]);
282 ** return number of addresses
283 ** and the addresses themselves in the buffer
284 ** maxSize - max number of interfaces to return.
287 rx_getAllAddr(afs_int32 * buffer, int maxSize)
289 int count = 0, offset = 0;
291 /* The IP address list can change so we must query for it */
295 /* we don't want to use the loopback adapter which is first */
296 /* this is a bad bad hack.
297 * and doesn't hold true on Windows.
299 if ( rxi_numNetAddrs > 1 )
301 #endif /* AFS_NT40_ENV */
303 for (count = 0; offset < rxi_numNetAddrs && maxSize > 0;
304 count++, offset++, maxSize--)
305 buffer[count] = htonl(rxi_NetAddrs[offset]);
317 int npackets, ncbufs;
325 rxi_numNetAddrs = ADDRSPERSITE;
326 (void)syscfg_GetIFInfo(&rxi_numNetAddrs, rxi_NetAddrs,
327 myNetMasks, myNetMTUs, myNetFlags);
329 for (i = 0; i < rxi_numNetAddrs; i++) {
330 rxsize = rxi_AdjustIfMTU(myNetMTUs[i] - RX_IPUDP_SIZE);
332 rxi_nRecvFrags * rxsize + (rxi_nRecvFrags - 1) * UDP_HDR_SIZE;
333 maxsize = rxi_AdjustMaxMTU(rxsize, maxsize);
334 if (rx_maxReceiveSize < maxsize) {
335 rx_maxReceiveSize = MIN(RX_MAX_PACKET_SIZE, maxsize);
337 MIN(rx_maxReceiveSize, rx_maxReceiveSizeUser);
342 ncbufs = (rx_maxJumboRecvSize - RX_FIRSTBUFFERSIZE);
344 ncbufs = ncbufs / RX_CBUFFERSIZE;
345 npackets = rx_initSendWindow - 1;
346 rxi_MorePackets(npackets * (ncbufs + 1));
352 fudge_netmask(afs_uint32 addr)
358 else if (IN_CLASSB(addr))
360 else if (IN_CLASSC(addr))
370 #if !defined(AFS_AIX_ENV) && !defined(AFS_NT40_ENV) && !defined(AFS_LINUX20_ENV)
372 rxi_syscall(a3, a4, a5)
379 old = (void (*)())signal(SIGSYS, SIG_IGN);
381 #if defined(AFS_SGI_ENV)
382 rcode = afs_syscall(a3, a4, a5);
384 rcode = syscall(AFS_SYSCALL, 28 /* AFSCALL_CALL */ , a3, a4, a5);
385 #endif /* AFS_SGI_ENV */
391 #endif /* AFS_AIX_ENV */
400 struct ifreq ifs[ADDRSPERSITE];
403 char buf[BUFSIZ], *cp, *cplim;
405 struct sockaddr_in *a;
416 memset(rxi_NetAddrs, 0, sizeof(rxi_NetAddrs));
417 memset(myNetFlags, 0, sizeof(myNetFlags));
418 memset(myNetMTUs, 0, sizeof(myNetMTUs));
419 memset(myNetMasks, 0, sizeof(myNetMasks));
421 s = socket(AF_INET, SOCK_DGRAM, 0);
426 ifc.ifc_len = sizeof(buf);
430 ifc.ifc_len = sizeof(ifs);
431 ifc.ifc_buf = (caddr_t) & ifs[0];
432 memset(&ifs[0], 0, sizeof(ifs));
434 res = ioctl(s, SIOCGIFCONF, &ifc);
436 /* fputs(stderr, "ioctl error IFCONF\n"); */
443 #define size(p) MAX((p).sa_len, sizeof(p))
444 cplim = buf + ifc.ifc_len; /*skip over if's with big ifr_addr's */
445 for (cp = buf; cp < cplim;
446 cp += sizeof(ifr->ifr_name) + MAX(a->sin_len, sizeof(*a))) {
447 if (rxi_numNetAddrs >= ADDRSPERSITE)
450 ifr = (struct ifreq *)cp;
452 len = ifc.ifc_len / sizeof(struct ifreq);
453 if (len > ADDRSPERSITE)
456 for (i = 0; i < len; ++i) {
458 res = ioctl(s, SIOCGIFADDR, ifr);
461 /* fputs(stderr, "ioctl error IFADDR\n");
462 * perror(ifr->ifr_name); */
465 a = (struct sockaddr_in *)&ifr->ifr_addr;
466 if (a->sin_family != AF_INET)
468 rxi_NetAddrs[rxi_numNetAddrs] = ntohl(a->sin_addr.s_addr);
469 if (rxi_NetAddrs[rxi_numNetAddrs] == 0x7f000001) {
470 /* we don't really care about "localhost" */
473 for (j = 0; j < rxi_numNetAddrs; j++) {
474 if (rxi_NetAddrs[j] == rxi_NetAddrs[rxi_numNetAddrs])
477 if (j < rxi_numNetAddrs)
480 /* fprintf(stderr, "if %s addr=%x\n", ifr->ifr_name,
481 * rxi_NetAddrs[rxi_numNetAddrs]); */
484 res = ioctl(s, SIOCGIFFLAGS, ifr);
486 myNetFlags[rxi_numNetAddrs] = ifr->ifr_flags;
488 /* Handle aliased loopbacks as well. */
489 if (ifr->ifr_flags & IFF_LOOPBACK)
492 /* fprintf(stderr, "if %s flags=%x\n",
493 * ifr->ifr_name, ifr->ifr_flags); */
495 * fputs(stderr, "ioctl error IFFLAGS\n");
496 * perror(ifr->ifr_name); */
498 #endif /* SIOCGIFFLAGS */
500 #if !defined(AFS_AIX_ENV) && !defined(AFS_LINUX20_ENV)
501 /* this won't run on an AIX system w/o a cache manager */
502 rxi_syscallp = rxi_syscall;
505 /* If I refer to kernel extensions that aren't loaded on AIX, the
506 * program refuses to load and run, so I simply can't include the
507 * following code. Fortunately, AIX is the one operating system in
508 * which the subsequent ioctl works reliably. */
510 if ((*rxi_syscallp) (20 /*AFSOP_GETMTU */ ,
511 htonl(rxi_NetAddrs[rxi_numNetAddrs]),
512 &(myNetMTUs[rxi_numNetAddrs]))) {
513 /* fputs(stderr, "syscall error GETMTU\n");
514 * perror(ifr->ifr_name); */
515 myNetMTUs[rxi_numNetAddrs] = 0;
517 if ((*rxi_syscallp) (42 /*AFSOP_GETMASK */ ,
518 htonl(rxi_NetAddrs[rxi_numNetAddrs]),
519 &(myNetMasks[rxi_numNetAddrs]))) {
520 /* fputs(stderr, "syscall error GETMASK\n");
521 * perror(ifr->ifr_name); */
522 myNetMasks[rxi_numNetAddrs] = 0;
524 myNetMasks[rxi_numNetAddrs] =
525 ntohl(myNetMasks[rxi_numNetAddrs]);
526 /* fprintf(stderr, "if %s mask=0x%x\n",
527 * ifr->ifr_name, myNetMasks[rxi_numNetAddrs]); */
530 if (myNetMTUs[rxi_numNetAddrs] == 0) {
531 myNetMTUs[rxi_numNetAddrs] = OLD_MAX_PACKET_SIZE + RX_IPUDP_SIZE;
533 res = ioctl(s, SIOCGIFMTU, ifr);
534 if ((res == 0) && (ifr->ifr_metric > 128)) { /* sanity check */
535 myNetMTUs[rxi_numNetAddrs] = ifr->ifr_metric;
536 /* fprintf(stderr, "if %s mtu=%d\n",
537 * ifr->ifr_name, ifr->ifr_metric); */
539 /* fputs(stderr, "ioctl error IFMTU\n");
540 * perror(ifr->ifr_name); */
545 if (myNetMasks[rxi_numNetAddrs] == 0) {
546 myNetMasks[rxi_numNetAddrs] =
547 fudge_netmask(rxi_NetAddrs[rxi_numNetAddrs]);
548 #ifdef SIOCGIFNETMASK
549 res = ioctl(s, SIOCGIFNETMASK, ifr);
551 a = (struct sockaddr_in *)&ifr->ifr_addr;
552 myNetMasks[rxi_numNetAddrs] = ntohl(a->sin_addr.s_addr);
553 /* fprintf(stderr, "if %s subnetmask=0x%x\n",
554 * ifr->ifr_name, myNetMasks[rxi_numNetAddrs]); */
556 /* fputs(stderr, "ioctl error IFMASK\n");
557 * perror(ifr->ifr_name); */
562 if (rxi_NetAddrs[rxi_numNetAddrs] != 0x7f000001) { /* ignore lo0 */
565 rxi_nRecvFrags * (myNetMTUs[rxi_numNetAddrs] - RX_IP_SIZE);
566 maxsize -= UDP_HDR_SIZE; /* only the first frag has a UDP hdr */
567 if (rx_maxReceiveSize < maxsize)
568 rx_maxReceiveSize = MIN(RX_MAX_PACKET_SIZE, maxsize);
575 /* have to allocate at least enough to allow a single packet to reach its
576 * maximum size, so ReadPacket will work. Allocate enough for a couple
577 * of packets to do so, for good measure */
579 int npackets, ncbufs;
581 rx_maxJumboRecvSize =
582 RX_HEADER_SIZE + rxi_nDgramPackets * RX_JUMBOBUFFERSIZE +
583 (rxi_nDgramPackets - 1) * RX_JUMBOHEADERSIZE;
584 rx_maxJumboRecvSize = MAX(rx_maxJumboRecvSize, rx_maxReceiveSize);
585 ncbufs = (rx_maxJumboRecvSize - RX_FIRSTBUFFERSIZE);
587 ncbufs = ncbufs / RX_CBUFFERSIZE;
588 npackets = rx_initSendWindow - 1;
589 rxi_MorePackets(npackets * (ncbufs + 1));
593 #endif /* AFS_NT40_ENV */
595 /* Called from rxi_FindPeer, when initializing a clear rx_peer structure,
596 * to get interesting information.
597 * Curiously enough, the rx_peerHashTable_lock currently protects the
598 * Inited variable (and hence rx_GetIFInfo). When the fs suite uses
599 * pthreads, this issue will need to be revisited.
603 rxi_InitPeerParams(struct rx_peer *pp)
613 * there's a race here since more than one thread could call
614 * rx_GetIFInfo. The race stops in rx_GetIFInfo.
622 /* try to second-guess IP, and identify which link is most likely to
623 * be used for traffic to/from this host. */
624 ppaddr = ntohl(pp->host);
628 pp->rateFlag = 2; /* start timing after two full packets */
629 /* I don't initialize these, because I presume they are bzero'd...
630 * pp->burstSize pp->burst pp->burstWait.sec pp->burstWait.usec
631 * pp->timeout.usec */
634 for (ix = 0; ix < rxi_numNetAddrs; ++ix) {
635 if ((rxi_NetAddrs[ix] & myNetMasks[ix]) == (ppaddr & myNetMasks[ix])) {
636 #ifdef IFF_POINTOPOINT
637 if (myNetFlags[ix] & IFF_POINTOPOINT)
639 #endif /* IFF_POINTOPOINT */
640 rxmtu = myNetMTUs[ix] - RX_IPUDP_SIZE;
641 if (rxmtu < RX_MIN_PACKET_SIZE)
642 rxmtu = RX_MIN_PACKET_SIZE;
643 if (pp->ifMTU < rxmtu)
644 pp->ifMTU = MIN(rx_MyMaxSendSize, rxmtu);
648 if (!pp->ifMTU) { /* not local */
650 pp->ifMTU = MIN(rx_MyMaxSendSize, RX_REMOTE_PACKET_SIZE);
652 #else /* ADAPT_MTU */
653 pp->rateFlag = 2; /* start timing after two full packets */
655 pp->ifMTU = MIN(rx_MyMaxSendSize, OLD_MAX_PACKET_SIZE);
656 #endif /* ADAPT_MTU */
657 pp->ifMTU = rxi_AdjustIfMTU(pp->ifMTU);
658 pp->maxMTU = OLD_MAX_PACKET_SIZE; /* for compatibility with old guys */
659 pp->natMTU = MIN((int)pp->ifMTU, OLD_MAX_PACKET_SIZE);
660 pp->maxDgramPackets =
661 MIN(rxi_nDgramPackets,
662 rxi_AdjustDgramPackets(RX_MAX_FRAGS, pp->ifMTU));
664 MIN(rxi_nDgramPackets,
665 rxi_AdjustDgramPackets(RX_MAX_FRAGS, pp->ifMTU));
666 pp->maxDgramPackets = 1;
667 /* Initialize slow start parameters */
668 pp->MTU = MIN(pp->natMTU, pp->maxMTU);
670 pp->nDgramPackets = 1;
674 /* Don't expose jumobgram internals. */
678 rx_maxReceiveSize = OLD_MAX_PACKET_SIZE;
679 rxi_nSendFrags = rxi_nRecvFrags = 1;
682 /* Override max MTU. If rx_SetNoJumbo is called, it must be
683 called before calling rx_SetMaxMTU since SetNoJumbo clobbers rx_maxReceiveSize */
685 rx_SetMaxMTU(int mtu)
687 rx_MyMaxSendSize = rx_maxReceiveSizeUser = rx_maxReceiveSize = mtu;