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 #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++) {
128 if (binds) rxi_Delay (10);
129 code = bind(socketFd, (struct sockaddr *) &taddr, sizeof(taddr));
134 (osi_Msg "%sbind failed\n", name);
138 #if !defined(AFS_NT40_ENV) && !defined(AFS_DJGPP_ENV)
140 * Set close-on-exec on rx socket
142 fcntl(socketFd, F_SETFD, 1);
145 /* Use one of three different ways of getting a socket buffer expanded to
152 len2 = rx_UdpBufSize;
153 #ifndef AFS_DJGPP_ENV
155 (setsockopt(socketFd, SOL_SOCKET, SO_RCVBUF, (char *)&len2,
158 len2=32766; /* fall back to old size... uh-oh! */
162 (setsockopt(socketFd, SOL_SOCKET, SO_SNDBUF, (char *)&len1,
163 sizeof(len1)) >= 0) &&
164 (setsockopt(socketFd, SOL_SOCKET, SO_RCVBUF, (char *)&len2,
166 #endif /* AFS_DJGPP_ENV */
169 #ifndef AFS_DJGPP_ENV
171 (osi_Msg "%s*WARNING* Unable to increase buffering on socket\n", name);
172 #endif /* AFS_DJGPP_ENV */
173 if (rxi_Listen(socketFd) < 0) {
181 if (socketFd >= 0) closesocket(socketFd);
183 if (socketFd >= 0) close(socketFd);
186 MUTEX_ENTER(&rx_stats_mutex);
187 rx_stats.socketGreedy = greedy;
188 MUTEX_EXIT(&rx_stats_mutex);
189 return OSI_NULLSOCKET;
192 void osi_Panic(msg, a1, a2, a3)
195 (osi_Msg "Fatal Rx error: ");
196 (osi_Msg msg, a1, a2, a3);
203 * osi_AssertFailU() -- used by the osi_Assert() macro.
206 void osi_AssertFailU(const char *expr, const char *file, int line)
208 osi_Panic("assertion failed: %s, file: %s, line: %d\n", expr, file, line);
213 static const char memZero;
218 * 0-length allocs may return NULL ptr from osi_kalloc, so we special-case
219 * things so that NULL returned iff an error occurred
221 if (x == 0) return &memZero;
222 return ((char *) malloc(x));
229 if (x == &memZero) return;
233 #endif /* AFS_AIX32_ENV */
235 #define ADDRSPERSITE 16
238 afs_uint32 rxi_NetAddrs[ADDRSPERSITE]; /* host order */
239 static int myNetMTUs[ADDRSPERSITE];
240 static int myNetMasks[ADDRSPERSITE];
241 static int myNetFlags[ADDRSPERSITE];
242 u_int rxi_numNetAddrs;
243 static int Inited = 0;
245 #if defined(AFS_NT40_ENV) || defined(AFS_DJGPP_ENV)
246 int rxi_getaddr(void)
248 if (rxi_numNetAddrs > 0)
249 return htonl(rxi_NetAddrs[0]);
255 ** return number of addresses
256 ** and the addresses themselves in the buffer
257 ** maxSize - max number of interfaces to return.
259 int rx_getAllAddr (afs_int32 *buffer, int maxSize)
262 for (count = 0; count < rxi_numNetAddrs && maxSize > 0; count++, maxSize--)
263 buffer[count] = htonl(rxi_NetAddrs[count]);
272 void rx_GetIFInfo(void)
282 int npackets, ncbufs;
287 rxi_numNetAddrs = ADDRSPERSITE;
290 (void) syscfg_GetIFInfo(&rxi_numNetAddrs, rxi_NetAddrs,
291 myNetMasks, myNetMTUs, myNetFlags);
293 for (i=0; i<rxi_numNetAddrs; i++) {
294 rxsize = rxi_AdjustIfMTU(myNetMTUs[i] - RX_IPUDP_SIZE);
295 maxsize = rxi_nRecvFrags*rxsize + (rxi_nRecvFrags-1)*UDP_HDR_SIZE;
296 maxsize = rxi_AdjustMaxMTU(rxsize, maxsize);
297 if (rx_maxReceiveSize < maxsize) {
298 rx_maxReceiveSize = MIN( RX_MAX_PACKET_SIZE, maxsize);
299 rx_maxReceiveSize = MIN( rx_maxReceiveSize,
300 rx_maxReceiveSizeUser);
305 ncbufs = (rx_maxJumboRecvSize - RX_FIRSTBUFFERSIZE);
307 ncbufs = ncbufs / RX_CBUFFERSIZE;
308 npackets = rx_initSendWindow -1;
309 rxi_MorePackets(npackets*(ncbufs+1));
316 static afs_uint32 fudge_netmask(addr)
321 if (IN_CLASSA(addr)) msk = IN_CLASSA_NET;
322 else if (IN_CLASSB(addr)) msk = IN_CLASSB_NET;
323 else if (IN_CLASSC(addr)) msk = IN_CLASSC_NET;
331 #if !defined(AFS_AIX_ENV) && !defined(AFS_NT40_ENV) && !defined(AFS_LINUX20_ENV) && !defined(AFS_DJGPP_ENV)
332 int rxi_syscall(a3, a4, a5)
339 old = (void (*)())signal(SIGSYS, SIG_IGN);
341 #if defined(AFS_SGI_ENV)
342 rcode = afs_syscall(a3, a4, a5);
344 rcode = syscall (AFS_SYSCALL, 28 /* AFSCALL_CALL */, a3, a4, a5);
345 #endif /* AFS_SGI_ENV */
351 #endif /* AFS_AIX_ENV */
358 #ifndef AFS_DJGPP_ENV
360 struct ifreq ifs[ADDRSPERSITE];
363 char buf[BUFSIZ], *cp, *cplim;
365 struct sockaddr_in *a;
366 #endif /* AFS_DJGPP_ENV */
378 memset(rxi_NetAddrs, 0, sizeof(rxi_NetAddrs));
379 memset(myNetFlags, 0, sizeof(myNetFlags));
380 memset(myNetMTUs, 0, sizeof(myNetMTUs));
381 memset(myNetMasks, 0, sizeof(myNetMasks));
384 s = socket(AF_INET, SOCK_DGRAM, 0);
387 #ifndef AFS_DJGPP_ENV
389 ifc.ifc_len = sizeof (buf);
393 ifc.ifc_len = sizeof(ifs);
394 ifc.ifc_buf = (caddr_t) &ifs[0];
395 memset(&ifs[0], 0, sizeof(ifs));
397 res = ioctl(s, SIOCGIFCONF, &ifc);
399 /* fputs(stderr, "ioctl error IFCONF\n"); */
407 #define size(p) MAX((p).sa_len, sizeof(p))
408 cplim = buf + ifc.ifc_len; /*skip over if's with big ifr_addr's */
409 for (cp = buf; cp < cplim ;
410 cp += sizeof(ifr->ifr_name) + MAX(a->sin_len, sizeof(*a))) {
411 if (rxi_numNetAddrs >= ADDRSPERSITE)
414 ifr = (struct ifreq *)cp;
416 len = ifc.ifc_len / sizeof(struct ifreq);
417 if (len > ADDRSPERSITE)
420 for (i = 0; i < len; ++i) {
422 res = ioctl(s, SIOCGIFADDR, ifr);
425 /* fputs(stderr, "ioctl error IFADDR\n");
426 perror(ifr->ifr_name); */
429 a = (struct sockaddr_in *) &ifr->ifr_addr;
430 if (a->sin_family != AF_INET) continue;
431 rxi_NetAddrs[rxi_numNetAddrs] = ntohl(a->sin_addr.s_addr);
432 if (rxi_NetAddrs[rxi_numNetAddrs] == 0x7f000001) {
433 /* we don't really care about "localhost" */
436 for (j=0; j < rxi_numNetAddrs; j++) {
437 if (rxi_NetAddrs[j] == rxi_NetAddrs[rxi_numNetAddrs])
440 if (j < rxi_numNetAddrs) continue;
442 /* fprintf(stderr, "if %s addr=%x\n", ifr->ifr_name,
443 rxi_NetAddrs[rxi_numNetAddrs]); */
446 res = ioctl(s, SIOCGIFFLAGS, ifr);
448 myNetFlags[rxi_numNetAddrs] = ifr->ifr_flags;
450 /* Handle aliased loopbacks as well. */
451 if (ifr->ifr_flags & IFF_LOOPBACK)
454 /* fprintf(stderr, "if %s flags=%x\n",
455 ifr->ifr_name, ifr->ifr_flags); */
458 fputs(stderr, "ioctl error IFFLAGS\n");
459 perror(ifr->ifr_name); */
461 #endif /* SIOCGIFFLAGS */
463 #if !defined(AFS_AIX_ENV) && !defined(AFS_LINUX20_ENV)
464 /* this won't run on an AIX system w/o a cache manager */
465 rxi_syscallp = rxi_syscall;
468 /* If I refer to kernel extensions that aren't loaded on AIX, the
469 * program refuses to load and run, so I simply can't include the
470 * following code. Fortunately, AIX is the one operating system in
471 * which the subsequent ioctl works reliably. */
473 if ((*rxi_syscallp)( 20 /*AFSOP_GETMTU*/,
474 htonl(rxi_NetAddrs[rxi_numNetAddrs]),
475 &(myNetMTUs[rxi_numNetAddrs]))) {
476 /* fputs(stderr, "syscall error GETMTU\n");
477 perror(ifr->ifr_name); */
478 myNetMTUs[rxi_numNetAddrs] = 0;
480 if ((*rxi_syscallp)(42 /*AFSOP_GETMASK*/,
481 htonl(rxi_NetAddrs[rxi_numNetAddrs]),
482 &(myNetMasks[rxi_numNetAddrs]))) {
483 /* fputs(stderr, "syscall error GETMASK\n");
484 perror(ifr->ifr_name); */
485 myNetMasks[rxi_numNetAddrs] = 0;
487 else myNetMasks[rxi_numNetAddrs] = ntohl( myNetMasks[rxi_numNetAddrs]);
488 /* fprintf(stderr, "if %s mask=0x%x\n",
489 ifr->ifr_name, myNetMasks[rxi_numNetAddrs]);*/
492 if (myNetMTUs[rxi_numNetAddrs] == 0) {
493 myNetMTUs[rxi_numNetAddrs] = OLD_MAX_PACKET_SIZE + RX_IPUDP_SIZE;
495 res = ioctl(s, SIOCGIFMTU, ifr);
496 if ((res == 0) && (ifr->ifr_metric > 128)) { /* sanity check */
497 myNetMTUs[rxi_numNetAddrs] = ifr->ifr_metric;
498 /* fprintf(stderr, "if %s mtu=%d\n",
499 ifr->ifr_name, ifr->ifr_metric); */
502 /* fputs(stderr, "ioctl error IFMTU\n");
503 perror(ifr->ifr_name); */
508 if (myNetMasks[rxi_numNetAddrs] == 0) {
509 myNetMasks[rxi_numNetAddrs] = fudge_netmask(rxi_NetAddrs[rxi_numNetAddrs]);
510 #ifdef SIOCGIFNETMASK
511 res = ioctl(s, SIOCGIFNETMASK, ifr);
513 a = (struct sockaddr_in *) &ifr->ifr_addr;
514 myNetMasks[rxi_numNetAddrs] = ntohl(a->sin_addr.s_addr);
515 /* fprintf(stderr, "if %s subnetmask=0x%x\n",
516 ifr->ifr_name, myNetMasks[rxi_numNetAddrs]); */
519 /* fputs(stderr, "ioctl error IFMASK\n");
520 perror(ifr->ifr_name); */
525 if (rxi_NetAddrs[rxi_numNetAddrs] != 0x7f000001) { /* ignore lo0 */
527 maxsize = rxi_nRecvFrags*(myNetMTUs[rxi_numNetAddrs] - RX_IP_SIZE);
528 maxsize -= UDP_HDR_SIZE; /* only the first frag has a UDP hdr */
529 if (rx_maxReceiveSize < maxsize)
530 rx_maxReceiveSize = MIN( RX_MAX_PACKET_SIZE, maxsize);
537 /* have to allocate at least enough to allow a single packet to reach its
538 * maximum size, so ReadPacket will work. Allocate enough for a couple
539 * of packets to do so, for good measure */
541 int npackets, ncbufs;
543 rx_maxJumboRecvSize = RX_HEADER_SIZE
544 + rxi_nDgramPackets * RX_JUMBOBUFFERSIZE
545 + (rxi_nDgramPackets-1) * RX_JUMBOHEADERSIZE;
546 rx_maxJumboRecvSize = MAX(rx_maxJumboRecvSize, rx_maxReceiveSize);
547 ncbufs = (rx_maxJumboRecvSize - RX_FIRSTBUFFERSIZE);
549 ncbufs = ncbufs / RX_CBUFFERSIZE;
550 npackets = rx_initSendWindow -1;
551 rxi_MorePackets(npackets*(ncbufs+1));
554 #else /* AFS_DJGPP_ENV */
557 #endif /* AFS_DJGPP_ENV */
559 #endif /* AFS_NT40_ENV */
561 /* Called from rxi_FindPeer, when initializing a clear rx_peer structure,
562 * to get interesting information.
563 * Curiously enough, the rx_peerHashTable_lock currently protects the
564 * Inited variable (and hence rx_GetIFInfo). When the fs suite uses
565 * pthreads, this issue will need to be revisited.
568 void rxi_InitPeerParams(struct rx_peer *pp)
580 * there's a race here since more than one thread could call
581 * rx_GetIFInfo. The race stops in rx_GetIFInfo.
589 /* try to second-guess IP, and identify which link is most likely to
590 * be used for traffic to/from this host. */
591 ppaddr = ntohl(pp->host);
595 pp->rateFlag = 2; /* start timing after two full packets */
596 /* I don't initialize these, because I presume they are bzero'd...
597 * pp->burstSize pp->burst pp->burstWait.sec pp->burstWait.usec
598 * pp->timeout.usec */
601 for (ix = 0; ix < rxi_numNetAddrs; ++ix) {
602 if ((rxi_NetAddrs[ix] & myNetMasks[ix]) == (ppaddr & myNetMasks[ix])) {
603 #ifdef IFF_POINTOPOINT
604 if (myNetFlags[ix] & IFF_POINTOPOINT)
606 #endif /* IFF_POINTOPOINT */
607 rxmtu = myNetMTUs[ix] - RX_IPUDP_SIZE;
608 if ( rxmtu < RX_MIN_PACKET_SIZE )
609 rxmtu = RX_MIN_PACKET_SIZE ;
610 if (pp->ifMTU < rxmtu)
611 pp->ifMTU = MIN(rx_MyMaxSendSize, rxmtu);
616 if (!pp->ifMTU) { /* not local */
618 pp->ifMTU = RX_REMOTE_PACKET_SIZE;
620 #else /* ADAPT_MTU */
621 pp->rateFlag = 2; /* start timing after two full packets */
623 pp->ifMTU = OLD_MAX_PACKET_SIZE;
624 #endif /* ADAPT_MTU */
625 pp->ifMTU = rxi_AdjustIfMTU(pp->ifMTU);
626 pp->maxMTU = OLD_MAX_PACKET_SIZE; /* for compatibility with old guys */
627 pp->natMTU = MIN((int)pp->ifMTU, OLD_MAX_PACKET_SIZE);
628 pp->maxDgramPackets = MIN(rxi_nDgramPackets,
629 rxi_AdjustDgramPackets(RX_MAX_FRAGS, pp->ifMTU));
630 pp->ifDgramPackets = MIN(rxi_nDgramPackets,
631 rxi_AdjustDgramPackets(RX_MAX_FRAGS, pp->ifMTU));
632 pp->maxDgramPackets = 1;
633 /* Initialize slow start parameters */
634 pp->MTU = MIN(pp->natMTU, pp->maxMTU);
636 pp->nDgramPackets = 1;
640 /* Don't expose jumobgram internals. */
641 void rx_SetNoJumbo(void)
643 rx_maxReceiveSize = OLD_MAX_PACKET_SIZE;
644 rxi_nSendFrags = rxi_nRecvFrags = 1;