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
11 * rx_kcommon.c - Common kernel RX code for all system types.
14 #include <afsconfig.h>
15 #include "afs/param.h"
18 #include "rx/rx_kcommon.h"
19 #include "rx_atomic.h"
22 #ifdef AFS_HPUX110_ENV
28 #ifndef RXK_LISTENER_ENV
29 int (*rxk_PacketArrivalProc) (struct rx_packet * ahandle, struct sockaddr_in * afrom, struct socket *arock, afs_int32 asize); /* set to packet allocation procedure */
30 int (*rxk_GetPacketProc) (struct rx_packet **ahandle, int asize);
33 osi_socket *rxk_NewSocketHost(afs_uint32 ahost, short aport);
34 extern struct interfaceAddr afs_cb_interface;
36 rxk_ports_t rxk_ports;
37 rxk_portRocks_t rxk_portRocks;
41 #if !defined(AFS_SUN5_ENV) && !defined(AFS_SGI62_ENV)
42 #define ADDRSPERSITE 16
43 static afs_uint32 myNetAddrs[ADDRSPERSITE];
44 static int myNetMTUs[ADDRSPERSITE];
45 static int numMyNetAddrs = 0;
48 #if defined(AFS_DARWIN80_ENV)
49 #define sobind sock_bind
50 #define soclose sock_close
53 /* add a port to the monitored list, port # is in network order */
55 rxk_AddPort(u_short aport, char *arock)
58 unsigned short *tsp, ts;
61 zslot = -1; /* look for an empty slot simultaneously */
62 for (i = 0, tsp = rxk_ports; i < MAXRXPORTS; i++, tsp++) {
63 if (((ts = *tsp) == 0) && (zslot == -1))
69 /* otherwise allocate a new port slot */
71 return E2BIG; /* all full */
72 rxk_ports[zslot] = aport;
73 rxk_portRocks[zslot] = arock;
77 /* remove as port from the monitored list, port # is in network order */
79 rxk_DelPort(u_short aport)
84 for (i = 0, tsp = rxk_ports; i < MAXRXPORTS; i++, tsp++) {
86 /* found it, adjust ref count and free the port reference if all gone */
91 /* otherwise port not found */
96 rxk_shutdownPorts(void)
99 for (i = 0; i < MAXRXPORTS; i++) {
102 #if ! defined(AFS_SUN5_ENV) && ! defined(UKERNEL) && ! defined(RXK_LISTENER_ENV)
103 soclose((struct socket *)rxk_portRocks[i]);
105 rxk_portRocks[i] = NULL;
111 rxi_GetHostUDPSocket(u_int host, u_short port)
114 sockp = (osi_socket *)rxk_NewSocketHost(host, port);
115 if (sockp == (osi_socket *)0)
116 return OSI_NULLSOCKET;
117 rxk_AddPort(port, (char *)sockp);
118 return (osi_socket) sockp;
122 rxi_GetUDPSocket(u_short port)
124 return rxi_GetHostUDPSocket(htonl(INADDR_ANY), port);
128 * osi_utoa() - write the NUL-terminated ASCII decimal form of the given
129 * unsigned long value into the given buffer. Returns 0 on success,
130 * and a value less than 0 on failure. The contents of the buffer is
131 * defined only on success.
135 osi_utoa(char *buf, size_t len, unsigned long val)
137 long k; /* index of first byte of string value */
139 /* we definitely need room for at least one digit and NUL */
145 /* compute the string form from the high end of the buffer */
148 for (k = len - 2; k >= 0; k--) {
149 buf[k] = val % 10 + '0';
156 /* did we finish converting val to string form? */
162 /* this should never happen */
168 /* this should never happen */
174 /* if necessary, relocate string to beginning of buf[] */
179 * We need to achieve the effect of calling
181 * memmove(buf, &buf[k], len - k);
183 * However, since memmove() is not available in all
184 * kernels, we explicitly do an appropriate copy.
190 while ((*dst++ = *src++) != '\0')
197 #ifndef AFS_LINUX26_ENV
199 * osi_AssertFailK() -- used by the osi_Assert() macro.
201 * It essentially does
203 * osi_Panic("assertion failed: %s, file: %s, line: %d", expr, file, line);
205 * Since the kernel version of osi_Panic() only passes its first
206 * argument to the native panic(), we construct a single string and hand
207 * that to osi_Panic().
210 osi_AssertFailK(const char *expr, const char *file, int line)
212 static const char msg0[] = "assertion failed: ";
213 static const char msg1[] = ", file: ";
214 static const char msg2[] = ", line: ";
215 static const char msg3[] = "\n";
218 * These buffers add up to 1K, which is a pleasantly nice round
219 * value, but probably not vital.
224 /* check line number conversion */
226 if (osi_utoa(linebuf, sizeof linebuf, line) < 0) {
227 osi_Panic("osi_AssertFailK: error in osi_utoa()\n");
232 #define ADDBUF(BUF, STR) \
233 if (strlen(BUF) + strlen((char *)(STR)) + 1 <= sizeof BUF) { \
234 strcat(BUF, (char *)(STR)); \
243 ADDBUF(buf, linebuf);
248 osi_Panic("%s", buf);
252 osi_AssertFailK(const char *expr, const char *file, int line)
254 printk(KERN_CRIT "assertion failed: %s, file: %s, line: %d\n", expr, file, line);
259 /* This is the server process request loop. Kernel server
260 * processes never become listener threads */
262 rx_ServerProc(void *unused)
266 rxi_MorePackets(rx_maxReceiveWindow + 2); /* alloc more packets */
267 MUTEX_ENTER(&rx_quota_mutex);
268 rxi_dataQuota += rx_initSendWindow; /* Reserve some pkts for hard times */
269 /* threadID is used for making decisions in GetCall. Get it by bumping
270 * number of threads handling incoming calls */
271 threadID = rxi_availProcs++;
272 MUTEX_EXIT(&rx_quota_mutex);
274 #ifdef RX_ENABLE_LOCKS
276 #endif /* RX_ENABLE_LOCKS */
277 rxi_ServerProc(threadID, NULL, NULL);
278 #ifdef RX_ENABLE_LOCKS
280 #endif /* RX_ENABLE_LOCKS */
284 #endif /* !UKERNEL */
286 #ifndef RXK_LISTENER_ENV
287 /* asize includes the Rx header */
289 MyPacketProc(struct rx_packet **ahandle, int asize)
291 struct rx_packet *tp;
293 /* If this is larger than we expected, increase rx_maxReceiveDataSize */
294 /* If we can't scrounge enough cbufs, then we have to drop the packet,
295 * but we should set a flag so we magic up some more at our leisure.
298 if ((asize >= 0) && (asize <= RX_MAX_PACKET_SIZE)) {
299 tp = rxi_AllocPacket(RX_PACKET_CLASS_RECEIVE);
300 if (tp && (tp->length + RX_HEADER_SIZE) < asize) {
302 rxi_AllocDataBuf(tp, asize - (tp->length + RX_HEADER_SIZE),
303 RX_PACKET_CLASS_RECV_CBUF)) {
306 if (rx_stats_active) {
307 rx_atomic_inc(&rx_stats.noPacketBuffersOnRead);
313 * XXX if packet is too long for our buffer,
314 * should do this at a higher layer and let other
315 * end know we're losing.
317 if (rx_stats_active) {
318 rx_atomic_inc(&rx_stats.bogusPacketOnRead);
320 /* I DON"T LIKE THIS PRINTF -- PRINTFS MAKE THINGS VERY VERY SLOOWWW */
321 dpf(("rx: packet dropped: bad ulen=%d\n", asize));
327 /* otherwise we have a packet, set appropriate values */
333 MyArrivalProc(struct rx_packet *ahandle,
334 struct sockaddr_in *afrom,
335 struct socket *arock,
338 /* handle basic rx packet */
339 ahandle->length = asize - RX_HEADER_SIZE;
340 rxi_DecodePacketHeader(ahandle);
342 rxi_ReceivePacket(ahandle, arock,
343 afrom->sin_addr.s_addr, afrom->sin_port, NULL,
346 /* free the packet if it has been returned */
348 rxi_FreePacket(ahandle);
351 #endif /* !RXK_LISTENER_ENV */
354 rxi_StartListener(void)
356 /* if kernel, give name of appropriate procedures */
357 #ifndef RXK_LISTENER_ENV
358 rxk_GetPacketProc = MyPacketProc;
359 rxk_PacketArrivalProc = MyArrivalProc;
364 /* Called from rxi_FindPeer, when initializing a clear rx_peer structure,
365 to get interesting information. */
367 rxi_InitPeerParams(struct rx_peer *pp)
372 # ifndef AFS_SUN5_ENV
373 # ifdef AFS_USERSPACE_IP_ADDR
377 i = rxi_Findcbi(pp->host);
380 /* pp->timeout.usec = 0; */
381 pp->ifMTU = MIN(RX_REMOTE_PACKET_SIZE, rx_MyMaxSendSize);
384 /* pp->timeout.usec = 0; */
385 pp->ifMTU = MIN(RX_MAX_PACKET_SIZE, rx_MyMaxSendSize);
386 mtu = ntohl(afs_cb_interface.mtu[i]);
387 /* Diminish the packet size to one based on the MTU given by
389 if (mtu > (RX_IPUDP_SIZE + RX_HEADER_SIZE)) {
390 rxmtu = mtu - RX_IPUDP_SIZE;
391 if (rxmtu < pp->ifMTU)
395 # else /* AFS_USERSPACE_IP_ADDR */
398 # if !defined(AFS_SGI62_ENV)
399 if (numMyNetAddrs == 0)
400 (void)rxi_GetIFInfo();
403 ifn = rxi_FindIfnet(pp->host, NULL);
406 /* pp->timeout.usec = 0; */
407 pp->ifMTU = MIN(RX_MAX_PACKET_SIZE, rx_MyMaxSendSize);
408 # ifdef IFF_POINTOPOINT
409 if (rx_ifnet_flags(ifn) & IFF_POINTOPOINT) {
410 /* wish we knew the bit rate and the chunk size, sigh. */
412 pp->ifMTU = RX_PP_PACKET_SIZE;
414 # endif /* IFF_POINTOPOINT */
415 /* Diminish the packet size to one based on the MTU given by
417 if (rx_ifnet_mtu(ifn) > (RX_IPUDP_SIZE + RX_HEADER_SIZE)) {
418 rxmtu = rx_ifnet_mtu(ifn) - RX_IPUDP_SIZE;
419 if (rxmtu < pp->ifMTU)
422 } else { /* couldn't find the interface, so assume the worst */
424 /* pp->timeout.usec = 0; */
425 pp->ifMTU = MIN(RX_REMOTE_PACKET_SIZE, rx_MyMaxSendSize);
427 # endif /* else AFS_USERSPACE_IP_ADDR */
428 # else /* AFS_SUN5_ENV */
431 mtu = rxi_FindIfMTU(pp->host);
435 /* pp->timeout.usec = 0; */
436 pp->ifMTU = MIN(RX_REMOTE_PACKET_SIZE, rx_MyMaxSendSize);
439 /* pp->timeout.usec = 0; */
440 pp->ifMTU = MIN(RX_MAX_PACKET_SIZE, rx_MyMaxSendSize);
442 /* Diminish the packet size to one based on the MTU given by
444 if (mtu > (RX_IPUDP_SIZE + RX_HEADER_SIZE)) {
445 rxmtu = mtu - RX_IPUDP_SIZE;
446 if (rxmtu < pp->ifMTU)
450 # endif /* AFS_SUN5_ENV */
451 #else /* ADAPT_MTU */
452 pp->rateFlag = 2; /* start timing after two full packets */
454 pp->ifMTU = OLD_MAX_PACKET_SIZE;
455 #endif /* else ADAPT_MTU */
456 pp->ifMTU = rxi_AdjustIfMTU(pp->ifMTU);
457 pp->maxMTU = OLD_MAX_PACKET_SIZE; /* for compatibility with old guys */
458 pp->natMTU = MIN(pp->ifMTU, OLD_MAX_PACKET_SIZE);
460 MIN(rxi_nDgramPackets,
461 rxi_AdjustDgramPackets(rxi_nSendFrags, pp->ifMTU));
462 pp->maxDgramPackets = 1;
464 /* Initialize slow start parameters */
465 pp->MTU = MIN(pp->natMTU, pp->maxMTU);
467 pp->nDgramPackets = 1;
472 /* The following code is common to several system types, but not all. The
473 * separate ones are found in the system specific subdirectories.
477 #if ! defined(AFS_AIX_ENV) && ! defined(AFS_SUN5_ENV) && ! defined(UKERNEL) && ! defined(AFS_LINUX20_ENV) && !defined (AFS_DARWIN_ENV) && !defined (AFS_XBSD_ENV)
478 /* Routine called during the afsd "-shutdown" process to put things back to
481 static struct protosw parent_proto; /* udp proto switch */
484 shutdown_rxkernel(void)
486 struct protosw *tpro, *last;
487 last = inetdomain.dom_protoswNPROTOSW;
488 for (tpro = inetdomain.dom_protosw; tpro < last; tpro++)
489 if (tpro->pr_protocol == IPPROTO_UDP) {
490 /* restore original udp protocol switch */
491 memcpy((void *)tpro, (void *)&parent_proto, sizeof(parent_proto));
492 memset((void *)&parent_proto, 0, sizeof(parent_proto));
497 dpf(("shutdown_rxkernel: no udp proto\n"));
499 #endif /* !AIX && !SUN && !NCR && !UKERNEL */
501 #if !defined(AFS_SUN5_ENV) && !defined(AFS_SGI62_ENV)
502 /* Determine what the network interfaces are for this machine. */
504 #ifdef AFS_USERSPACE_IP_ADDR
508 int i, j, different = 0, num = ADDRSPERSITE;
511 afs_uint32 addrs[ADDRSPERSITE];
512 int mtus[ADDRSPERSITE];
514 memset((void *)addrs, 0, sizeof(addrs));
515 memset((void *)mtus, 0, sizeof(mtus));
517 if (afs_cb_interface.numberOfInterfaces < num)
518 num = afs_cb_interface.numberOfInterfaces;
519 for (i = 0; i < num; i++) {
520 if (!afs_cb_interface.mtu[i])
521 afs_cb_interface.mtu[i] = htonl(1500);
522 rxmtu = (ntohl(afs_cb_interface.mtu[i]) - RX_IPUDP_SIZE);
523 ifinaddr = ntohl(afs_cb_interface.addr_in[i]);
524 if (myNetAddrs[i] != ifinaddr)
528 rxmtu = rxi_AdjustIfMTU(rxmtu);
530 rxmtu * rxi_nRecvFrags + ((rxi_nRecvFrags - 1) * UDP_HDR_SIZE);
531 maxmtu = rxi_AdjustMaxMTU(rxmtu, maxmtu);
532 addrs[i++] = ifinaddr;
533 if (!rx_IsLoopbackAddr(ifinaddr) && (maxmtu > rx_maxReceiveSize)) {
534 rx_maxReceiveSize = MIN(RX_MAX_PACKET_SIZE, maxmtu);
535 rx_maxReceiveSize = MIN(rx_maxReceiveSize, rx_maxReceiveSizeUser);
539 rx_maxJumboRecvSize =
540 RX_HEADER_SIZE + (rxi_nDgramPackets * RX_JUMBOBUFFERSIZE) +
541 ((rxi_nDgramPackets - 1) * RX_JUMBOHEADERSIZE);
542 rx_maxJumboRecvSize = MAX(rx_maxJumboRecvSize, rx_maxReceiveSize);
545 for (j = 0; j < i; j++) {
546 myNetMTUs[j] = mtus[j];
547 myNetAddrs[j] = addrs[j];
554 /* Returns the afs_cb_interface inxex which best matches address.
555 * If none is found, we return -1.
558 rxi_Findcbi(afs_uint32 addr)
561 afs_uint32 myAddr, thisAddr, netMask, subnetMask;
562 afs_int32 rvalue = -1;
565 if (numMyNetAddrs == 0)
566 (void)rxi_GetcbiInfo();
568 myAddr = ntohl(addr);
570 if (IN_CLASSA(myAddr))
571 netMask = IN_CLASSA_NET;
572 else if (IN_CLASSB(myAddr))
573 netMask = IN_CLASSB_NET;
574 else if (IN_CLASSC(myAddr))
575 netMask = IN_CLASSC_NET;
579 for (j = 0; j < afs_cb_interface.numberOfInterfaces; j++) {
580 thisAddr = ntohl(afs_cb_interface.addr_in[j]);
581 subnetMask = ntohl(afs_cb_interface.subnetmask[j]);
582 if ((myAddr & netMask) == (thisAddr & netMask)) {
583 if ((myAddr & subnetMask) == (thisAddr & subnetMask)) {
584 if (myAddr == thisAddr) {
589 if (match_value < 3) {
594 if (match_value < 2) {
605 #else /* AFS_USERSPACE_IP_ADDR */
607 #if !defined(AFS_AIX41_ENV) && !defined(AFS_DUX40_ENV) && !defined(AFS_DARWIN_ENV) && !defined(AFS_XBSD_ENV)
608 #define IFADDR2SA(f) (&((f)->ifa_addr))
609 #else /* AFS_AIX41_ENV */
610 #define IFADDR2SA(f) ((f)->ifa_addr)
620 afs_uint32 addrs[ADDRSPERSITE];
621 int mtus[ADDRSPERSITE];
623 #if defined(AFS_DARWIN80_ENV)
629 struct sockaddr sout;
630 struct sockaddr_in *sin;
633 rx_ifaddr_t ifad; /* ifnet points to a if_addrlist of ifaddrs */
637 memset(addrs, 0, sizeof(addrs));
638 memset(mtus, 0, sizeof(mtus));
640 #if defined(AFS_DARWIN80_ENV)
641 if (!ifnet_list_get(AF_INET, &ifns, &count)) {
642 for (m = 0; m < count; m++) {
643 if (!ifnet_get_address_list(ifns[m], &ifads)) {
644 for (j = 0; ifads[j] != NULL && cnt < ADDRSPERSITE; j++) {
645 if ((t = ifaddr_address(ifads[j], &sout, sizeof(struct sockaddr))) == 0) {
646 sin = (struct sockaddr_in *)&sout;
647 rxmtu = rx_ifnet_mtu(rx_ifaddr_ifnet(ifads[j])) - RX_IPUDP_SIZE;
648 ifinaddr = ntohl(sin->sin_addr.s_addr);
649 if (myNetAddrs[i] != ifinaddr) {
653 rxmtu = rxi_AdjustIfMTU(rxmtu);
655 rxmtu * rxi_nRecvFrags +
656 ((rxi_nRecvFrags - 1) * UDP_HDR_SIZE);
657 maxmtu = rxi_AdjustMaxMTU(rxmtu, maxmtu);
658 addrs[i++] = ifinaddr;
659 if (!rx_IsLoopbackAddr(ifinaddr) &&
660 (maxmtu > rx_maxReceiveSize)) {
662 MIN(RX_MAX_PACKET_SIZE, maxmtu);
664 MIN(rx_maxReceiveSize, rx_maxReceiveSizeUser);
669 ifnet_free_address_list(ifads);
672 ifnet_list_free(ifns);
675 #if defined(AFS_DARWIN_ENV) || defined(AFS_FBSD_ENV)
676 TAILQ_FOREACH(ifn, &ifnet, if_link) {
677 if (i >= ADDRSPERSITE)
679 #elif defined(AFS_OBSD_ENV) || defined(AFS_NBSD_ENV)
680 for (ifn = ifnet.tqh_first; i < ADDRSPERSITE && ifn != NULL;
681 ifn = ifn->if_list.tqe_next) {
683 for (ifn = ifnet; ifn != NULL && i < ADDRSPERSITE; ifn = ifn->if_next) {
685 rxmtu = (ifn->if_mtu - RX_IPUDP_SIZE);
686 #if defined(AFS_DARWIN_ENV) || defined(AFS_FBSD_ENV)
687 TAILQ_FOREACH(ifad, &ifn->if_addrhead, ifa_link) {
688 if (i >= ADDRSPERSITE)
690 #elif defined(AFS_OBSD_ENV) || defined(AFS_NBSD_ENV)
691 for (ifad = ifn->if_addrlist.tqh_first;
692 ifad != NULL && i < ADDRSPERSITE;
693 ifad = ifad->ifa_list.tqe_next) {
695 for (ifad = ifn->if_addrlist; ifad != NULL && i < ADDRSPERSITE;
696 ifad = ifad->ifa_next) {
698 if (IFADDR2SA(ifad)->sa_family == AF_INET) {
700 ntohl(((struct sockaddr_in *)IFADDR2SA(ifad))->sin_addr.
702 if (myNetAddrs[i] != ifinaddr) {
706 rxmtu = rxi_AdjustIfMTU(rxmtu);
708 rxmtu * rxi_nRecvFrags +
709 ((rxi_nRecvFrags - 1) * UDP_HDR_SIZE);
710 maxmtu = rxi_AdjustMaxMTU(rxmtu, maxmtu);
711 addrs[i++] = ifinaddr;
712 if (!rx_IsLoopbackAddr(ifinaddr) && (maxmtu > rx_maxReceiveSize)) {
713 rx_maxReceiveSize = MIN(RX_MAX_PACKET_SIZE, maxmtu);
715 MIN(rx_maxReceiveSize, rx_maxReceiveSizeUser);
722 rx_maxJumboRecvSize =
723 RX_HEADER_SIZE + rxi_nDgramPackets * RX_JUMBOBUFFERSIZE +
724 (rxi_nDgramPackets - 1) * RX_JUMBOHEADERSIZE;
725 rx_maxJumboRecvSize = MAX(rx_maxJumboRecvSize, rx_maxReceiveSize);
729 for (l = 0; l < i; l++) {
730 myNetMTUs[l] = mtus[l];
731 myNetAddrs[l] = addrs[l];
737 #if defined(AFS_DARWIN_ENV) || defined(AFS_XBSD_ENV)
738 /* Returns ifnet which best matches address */
740 rxi_FindIfnet(afs_uint32 addr, afs_uint32 * maskp)
742 struct sockaddr_in s, sr;
745 s.sin_family = AF_INET;
746 s.sin_addr.s_addr = addr;
747 ifad = rx_ifaddr_withnet((struct sockaddr *)&s);
750 rx_ifaddr_netmask(ifad, (struct sockaddr *)&sr, sizeof(sr));
751 *maskp = sr.sin_addr.s_addr;
753 return (ifad ? rx_ifaddr_ifnet(ifad) : NULL);
756 #else /* DARWIN || XBSD */
758 /* Returns ifnet which best matches address */
760 rxi_FindIfnet(afs_uint32 addr, afs_uint32 * maskp)
763 extern struct in_ifaddr *in_ifaddr;
764 struct in_ifaddr *ifa, *ifad = NULL;
768 #if defined(AFS_DARWIN_ENV)
769 for (ifa = TAILQ_FIRST(&in_ifaddrhead); ifa;
770 ifa = TAILQ_NEXT(ifa, ia_link)) {
772 for (ifa = in_ifaddr; ifa; ifa = ifa->ia_next) {
774 if ((addr & ifa->ia_netmask) == ifa->ia_net) {
775 if ((addr & ifa->ia_subnetmask) == ifa->ia_subnet) {
776 if (IA_SIN(ifa)->sin_addr.s_addr == addr) { /* ie, ME!!! */
781 if (match_value < 3) {
786 if (match_value < 2) {
791 } /* if net matches */
792 } /* for all in_ifaddrs */
796 *maskp = ifad->ia_subnetmask;
797 return (ifad ? ifad->ia_ifp : NULL);
799 #endif /* else DARWIN || XBSD */
800 #endif /* else AFS_USERSPACE_IP_ADDR */
801 #endif /* !SUN5 && !SGI62 */
804 /* rxk_NewSocket, rxk_FreeSocket and osi_NetSend are from the now defunct
805 * afs_osinet.c. One could argue that rxi_NewSocket could go into the
806 * system specific subdirectories for all systems. But for the moment,
807 * most of it is simple to follow common code.
809 #if !defined(UKERNEL)
810 #if !defined(AFS_SUN5_ENV) && !defined(AFS_LINUX20_ENV)
811 /* rxk_NewSocket creates a new socket on the specified port. The port is
812 * in network byte order.
815 rxk_NewSocketHost(afs_uint32 ahost, short aport)
818 #ifdef AFS_DARWIN80_ENV
821 struct socket *newSocket;
823 #if (!defined(AFS_HPUX1122_ENV) && !defined(AFS_FBSD_ENV))
826 struct sockaddr_in myaddr;
827 #ifdef AFS_HPUX110_ENV
828 /* prototype copied from kernel source file streams/str_proto.h */
829 extern MBLKP allocb_wait(int, int);
831 int addrsize = sizeof(struct sockaddr_in);
833 extern struct fileops socketops;
839 AFS_STATCNT(osi_NewSocket);
840 #if (defined(AFS_DARWIN_ENV) || defined(AFS_XBSD_ENV)) && defined(KERNEL_FUNNEL)
841 thread_funnel_switch(KERNEL_FUNNEL, NETWORK_FUNNEL);
845 #if defined(AFS_HPUX102_ENV)
846 #if defined(AFS_HPUX110_ENV)
847 /* we need a file associated with the socket so sosend in NetSend
849 /* blocking socket */
850 code = socreate(AF_INET, &newSocket, SOCK_DGRAM, 0, 0);
854 fp->f_flag = FREAD | FWRITE;
855 fp->f_type = DTYPE_SOCKET;
856 fp->f_ops = &socketops;
858 fp->f_data = (void *)newSocket;
859 newSocket->so_fp = (void *)fp;
861 #else /* AFS_HPUX110_ENV */
862 code = socreate(AF_INET, &newSocket, SOCK_DGRAM, 0, SS_NOWAIT);
863 #endif /* else AFS_HPUX110_ENV */
864 #elif defined(AFS_SGI65_ENV) || defined(AFS_OBSD_ENV)
865 code = socreate(AF_INET, &newSocket, SOCK_DGRAM, IPPROTO_UDP);
866 #elif defined(AFS_FBSD_ENV)
867 code = socreate(AF_INET, &newSocket, SOCK_DGRAM, IPPROTO_UDP,
868 afs_osi_credp, curthread);
869 #elif defined(AFS_DARWIN80_ENV)
870 code = sock_socket(AF_INET, SOCK_DGRAM, IPPROTO_UDP, NULL, NULL, &newSocket);
871 #elif defined(AFS_NBSD50_ENV)
872 code = socreate(AF_INET, &newSocket, SOCK_DGRAM, 0, osi_curproc(), NULL);
873 #elif defined(AFS_NBSD40_ENV)
874 code = socreate(AF_INET, &newSocket, SOCK_DGRAM, 0, osi_curproc());
876 code = socreate(AF_INET, &newSocket, SOCK_DGRAM, 0);
877 #endif /* AFS_HPUX102_ENV */
881 memset(&myaddr, 0, sizeof myaddr);
882 myaddr.sin_family = AF_INET;
883 myaddr.sin_port = aport;
884 myaddr.sin_addr.s_addr = ahost;
885 #ifdef STRUCT_SOCKADDR_HAS_SA_LEN
886 myaddr.sin_len = sizeof(myaddr);
889 #ifdef AFS_HPUX110_ENV
890 bindnam = allocb_wait((addrsize + SO_MSGOFFSET + 1), BPRI_MED);
895 memcpy((caddr_t) bindnam->b_rptr + SO_MSGOFFSET, (caddr_t) & myaddr,
897 bindnam->b_wptr = bindnam->b_rptr + (addrsize + SO_MSGOFFSET + 1);
898 #if defined(AFS_NBSD40_ENV)
899 code = sobind(newSocket, bindnam, addrsize, osi_curproc());
901 code = sobind(newSocket, bindnam, addrsize);
905 #if !defined(AFS_HPUX1122_ENV)
912 #else /* AFS_HPUX110_ENV */
913 #if defined(AFS_DARWIN80_ENV)
918 code = sock_setsockopt(newSocket, SOL_SOCKET, SO_SNDBUF,
919 &buflen, sizeof(buflen));
920 code2 = sock_setsockopt(newSocket, SOL_SOCKET, SO_RCVBUF,
921 &buflen, sizeof(buflen));
925 osi_Panic("osi_NewSocket: last attempt to reserve 32K failed!\n");
930 code = soreserve(newSocket, 50000, 50000);
932 code = soreserve(newSocket, 32766, 32766);
934 osi_Panic("osi_NewSocket: last attempt to reserve 32K failed!\n");
937 #if defined(AFS_DARWIN_ENV) || defined(AFS_FBSD_ENV)
938 #if defined(AFS_FBSD_ENV)
939 code = sobind(newSocket, (struct sockaddr *)&myaddr, curthread);
941 code = sobind(newSocket, (struct sockaddr *)&myaddr);
944 dpf(("sobind fails (%d)\n", (int)code));
948 #else /* defined(AFS_DARWIN_ENV) || defined(AFS_FBSD_ENV) */
950 nam = m_getclr(M_WAIT, MT_SONAME);
951 #else /* AFS_OSF_ENV */
952 nam = m_get(M_WAIT, MT_SONAME);
955 #if defined(KERNEL_HAVE_UERROR)
960 nam->m_len = sizeof(myaddr);
961 memcpy(mtod(nam, caddr_t), &myaddr, sizeof(myaddr));
962 #if defined(AFS_SGI65_ENV)
963 BHV_PDATA(&bhv) = (void *)newSocket;
964 code = sobind(&bhv, nam);
966 #elif defined(AFS_OBSD44_ENV) || defined(AFS_NBSD40_ENV)
967 code = sobind(newSocket, nam, osi_curproc());
969 code = sobind(newSocket, nam);
972 dpf(("sobind fails (%d)\n", (int)code));
974 #ifndef AFS_SGI65_ENV
979 #endif /* else AFS_DARWIN_ENV */
980 #endif /* else AFS_HPUX110_ENV */
983 #if defined(AFS_DARWIN_ENV) && defined(KERNEL_FUNNEL)
984 thread_funnel_switch(NETWORK_FUNNEL, KERNEL_FUNNEL);
986 return (osi_socket *)newSocket;
990 #if defined(AFS_DARWIN_ENV) && defined(KERNEL_FUNNEL)
991 thread_funnel_switch(NETWORK_FUNNEL, KERNEL_FUNNEL);
993 return (osi_socket *)0;
997 rxk_NewSocket(short aport)
999 return rxk_NewSocketHost(0, aport);
1002 /* free socket allocated by rxk_NewSocket */
1004 rxk_FreeSocket(struct socket *asocket)
1006 AFS_STATCNT(osi_FreeSocket);
1007 #if defined(AFS_DARWIN_ENV) && defined(KERNEL_FUNNEL)
1008 thread_funnel_switch(KERNEL_FUNNEL, NETWORK_FUNNEL);
1010 #ifdef AFS_HPUX110_ENV
1011 if (asocket->so_fp) {
1012 struct file *fp = asocket->so_fp;
1013 #if !defined(AFS_HPUX1123_ENV)
1014 /* 11.23 still has falloc, but not FPENTRYFREE !
1015 * so for now if we shutdown, we will waist a file
1018 asocket->so_fp = NULL;
1021 #endif /* AFS_HPUX110_ENV */
1023 #if defined(AFS_DARWIN_ENV) && defined(KERNEL_FUNNEL)
1024 thread_funnel_switch(NETWORK_FUNNEL, KERNEL_FUNNEL);
1028 #endif /* !SUN5 && !LINUX20 */
1030 #if defined(RXK_LISTENER_ENV) || defined(AFS_SUN5_ENV)
1031 #ifdef AFS_DARWIN80_ENV
1032 /* Shutting down should wake us up, as should an earlier event. */
1034 rxi_ReScheduleEvents(void)
1036 /* needed to allow startup */
1037 int glock = ISAFS_GLOCK();
1040 osi_rxWakeup(&afs_termState);
1046 * Run RX event daemon every second (5 times faster than rest of systems)
1049 afs_rxevent_daemon(void)
1055 #ifdef RX_ENABLE_LOCKS
1057 #endif /* RX_ENABLE_LOCKS */
1059 rxevent_RaiseEvents(&temp);
1061 #ifdef RX_ENABLE_LOCKS
1063 #endif /* RX_ENABLE_LOCKS */
1064 #ifdef RX_KERNEL_TRACE
1065 afs_Trace1(afs_iclSetp, CM_TRACE_TIMESTAMP, ICL_TYPE_STRING,
1066 "before afs_osi_Wait()");
1068 #ifdef AFS_DARWIN80_ENV
1069 afs_osi_TimedSleep(&afs_termState, MAX(500, ((temp.sec * 1000) +
1070 (temp.usec / 1000))), 0);
1072 afs_osi_Wait(500, NULL, 0);
1074 #ifdef RX_KERNEL_TRACE
1075 afs_Trace1(afs_iclSetp, CM_TRACE_TIMESTAMP, ICL_TYPE_STRING,
1076 "after afs_osi_Wait()");
1078 if (afs_termState == AFSOP_STOP_RXEVENT) {
1079 #ifdef RXK_LISTENER_ENV
1080 afs_termState = AFSOP_STOP_RXK_LISTENER;
1082 #ifdef AFS_SUN510_ENV
1083 afs_termState = AFSOP_STOP_NETIF;
1085 afs_termState = AFSOP_STOP_COMPLETE;
1088 osi_rxWakeup(&afs_termState);
1095 #ifdef RXK_LISTENER_ENV
1097 /* rxk_ReadPacket returns 1 if valid packet, 0 on error. */
1099 rxk_ReadPacket(osi_socket so, struct rx_packet *p, int *host, int *port)
1102 struct sockaddr_in from;
1106 afs_int32 savelen; /* was using rlen but had aliasing problems */
1107 rx_computelen(p, tlen);
1108 rx_SetDataSize(p, tlen); /* this is the size of the user data area */
1110 tlen += RX_HEADER_SIZE; /* now this is the size of the entire packet */
1111 rlen = rx_maxJumboRecvSize; /* this is what I am advertising. Only check
1112 * it once in order to avoid races. */
1115 tlen = rxi_AllocDataBuf(p, tlen, RX_PACKET_CLASS_RECV_CBUF);
1123 /* add some padding to the last iovec, it's just to make sure that the
1124 * read doesn't return more data than we expect, and is done to get around
1125 * our problems caused by the lack of a length field in the rx header. */
1126 savelen = p->wirevec[p->niovecs - 1].iov_len;
1127 p->wirevec[p->niovecs - 1].iov_len = savelen + RX_EXTRABUFFERSIZE;
1129 nbytes = tlen + sizeof(afs_int32);
1130 #ifdef RX_KERNEL_TRACE
1131 if (ICL_SETACTIVE(afs_iclSetp)) {
1133 afs_Trace1(afs_iclSetp, CM_TRACE_TIMESTAMP, ICL_TYPE_STRING,
1134 "before osi_NetRecive()");
1138 code = osi_NetReceive(rx_socket, &from, p->wirevec, p->niovecs, &nbytes);
1140 #ifdef RX_KERNEL_TRACE
1141 if (ICL_SETACTIVE(afs_iclSetp)) {
1143 afs_Trace1(afs_iclSetp, CM_TRACE_TIMESTAMP, ICL_TYPE_STRING,
1144 "after osi_NetRecive()");
1148 /* restore the vec to its correct state */
1149 p->wirevec[p->niovecs - 1].iov_len = savelen;
1152 p->length = nbytes - RX_HEADER_SIZE;;
1153 if ((nbytes > tlen) || (p->length & 0x8000)) { /* Bogus packet */
1155 if (rx_stats_active) {
1156 MUTEX_ENTER(&rx_stats_mutex);
1157 rx_atomic_inc(&rx_stats.bogusPacketOnRead);
1158 rx_stats.bogusHost = from.sin_addr.s_addr;
1159 MUTEX_EXIT(&rx_stats_mutex);
1161 dpf(("B: bogus packet from [%x,%d] nb=%d\n",
1162 from.sin_addr.s_addr, from.sin_port, nbytes));
1166 /* Extract packet header. */
1167 rxi_DecodePacketHeader(p);
1169 *host = from.sin_addr.s_addr;
1170 *port = from.sin_port;
1171 if (p->header.type > 0 && p->header.type < RX_N_PACKET_TYPES) {
1172 if (rx_stats_active) {
1173 rx_atomic_inc(&rx_stats.packetsRead[p->header.type - 1]);
1177 #ifdef RX_TRIMDATABUFS
1178 /* Free any empty packet buffers at the end of this packet */
1179 rxi_TrimDataBufs(p, 1);
1189 * Listen for packets on socket. This thread is typically started after
1190 * rx_Init has called rxi_StartListener(), but nevertheless, ensures that
1191 * the start state is set before proceeding.
1193 * Note that this thread is outside the AFS global lock for much of
1196 * In many OS's, the socket receive code sleeps interruptibly. That's not what
1197 * we want here. So we need to either block all signals (including SIGKILL
1198 * and SIGSTOP) or reset the thread's signal state to unsignalled when the
1199 * OS's socket receive routine returns as a result of a signal.
1201 int rxk_ListenerPid; /* Used to signal process to wakeup at shutdown */
1202 #ifdef AFS_LINUX20_ENV
1203 struct task_struct *rxk_ListenerTask;
1208 * Run the listener as a kernel thread.
1214 void rxk_ListenerProc(void);
1216 (NULL, DEFAULTSTKSZ, rxk_ListenerProc, 0, 0, &p0, TS_RUN,
1217 minclsyspri) == NULL)
1218 osi_Panic("rxk_Listener: failed to start listener thread!\n");
1222 rxk_ListenerProc(void)
1223 #else /* AFS_SUN5_ENV */
1226 #endif /* AFS_SUN5_ENV */
1228 struct rx_packet *rxp = NULL;
1232 #ifdef AFS_LINUX20_ENV
1233 rxk_ListenerPid = current->pid;
1234 rxk_ListenerTask = current;
1237 rxk_ListenerPid = 1; /* No PID, just a flag that we're alive */
1238 #endif /* AFS_SUN5_ENV */
1240 rxk_ListenerPid = curproc->p_pid;
1241 #endif /* AFS_FBSD_ENV */
1242 #ifdef AFS_DARWIN80_ENV
1243 rxk_ListenerPid = proc_selfpid();
1244 #elif defined(AFS_DARWIN_ENV)
1245 rxk_ListenerPid = current_proc()->p_pid;
1247 #if defined(RX_ENABLE_LOCKS) && !defined(AFS_SUN5_ENV)
1249 #endif /* RX_ENABLE_LOCKS && !AFS_SUN5_ENV */
1250 while (afs_termState != AFSOP_STOP_RXK_LISTENER) {
1251 /* See if a check for additional packets was issued */
1255 rxi_RestoreDataBufs(rxp);
1257 rxp = rxi_AllocPacket(RX_PACKET_CLASS_RECEIVE);
1259 osi_Panic("rxk_Listener: No more Rx buffers!\n");
1261 if (!(code = rxk_ReadPacket(rx_socket, rxp, &host, &port))) {
1262 rxp = rxi_ReceivePacket(rxp, rx_socket, host, port, 0, 0);
1266 #ifdef RX_ENABLE_LOCKS
1268 #endif /* RX_ENABLE_LOCKS */
1269 if (afs_termState == AFSOP_STOP_RXK_LISTENER) {
1270 #ifdef AFS_SUN510_ENV
1271 afs_termState = AFSOP_STOP_NETIF;
1273 afs_termState = AFSOP_STOP_COMPLETE;
1275 osi_rxWakeup(&afs_termState);
1277 rxk_ListenerPid = 0;
1278 #ifdef AFS_LINUX20_ENV
1279 rxk_ListenerTask = 0;
1280 osi_rxWakeup(&rxk_ListenerTask);
1282 #if defined(AFS_SUN5_ENV)
1283 osi_rxWakeup(&rxk_ListenerPid);
1287 #endif /* AFS_SUN5_ENV */
1290 #if !defined(AFS_LINUX20_ENV) && !defined(AFS_SUN5_ENV) && !defined(AFS_DARWIN_ENV) && !defined(AFS_XBSD_ENV)
1291 /* The manner of stopping the rx listener thread may vary. Most unix's should
1292 * be able to call soclose.
1295 osi_StopListener(void)
1300 #endif /* RXK_LISTENER_ENV */
1301 #endif /* !NCR && !UKERNEL */
1303 #if !defined(AFS_LINUX26_ENV)
1305 #if defined(AFS_AIX_ENV)
1306 osi_Panic(char *msg, void *a1, void *a2, void *a3)
1308 osi_Panic(char *msg, ...)
1313 msg = "Unknown AFS panic";
1315 * we should probably use the errsave facility here. it is not
1319 printf(msg, a1, a2, a3);
1321 #elif defined(AFS_SGI_ENV)
1324 /* Solaris has vcmn_err, Sol10 01/06 may have issues. Beware. */
1326 cmn_err(CE_PANIC, "Unknown AFS panic");
1329 icmn_err(CE_PANIC, msg, ap);
1332 #elif defined(AFS_DARWIN80_ENV) || (defined(AFS_LINUX22_ENV) && !defined(AFS_LINUX_26_ENV))
1336 msg = "Unknown AFS panic";
1339 vsnprintf(buf, sizeof(buf), msg, ap);
1346 msg = "Unknown AFS panic";
1351 # ifdef AFS_LINUX20_ENV