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"
20 #include "rx_packet.h"
21 #include "rx_internal.h"
25 #ifdef AFS_HPUX110_ENV
31 #ifndef RXK_LISTENER_ENV
32 int (*rxk_PacketArrivalProc) (struct rx_packet * ahandle, struct sockaddr_in * afrom, struct socket *arock, afs_int32 asize); /* set to packet allocation procedure */
33 int (*rxk_GetPacketProc) (struct rx_packet **ahandle, int asize);
36 extern struct interfaceAddr afs_cb_interface;
38 rxk_ports_t rxk_ports;
39 rxk_portRocks_t rxk_portRocks;
43 #if !defined(AFS_SUN5_ENV) && !defined(AFS_SGI_ENV)
44 # define ADDRSPERSITE 16
45 static afs_uint32 myNetAddrs[ADDRSPERSITE];
46 static int myNetMTUs[ADDRSPERSITE];
47 static int numMyNetAddrs = 0;
50 #if defined(AFS_DARWIN80_ENV)
51 # define sobind sock_bind
52 # define soclose sock_close
55 /* add a port to the monitored list, port # is in network order */
57 rxk_AddPort(u_short aport, char *arock)
60 unsigned short *tsp, ts;
63 zslot = -1; /* look for an empty slot simultaneously */
64 for (i = 0, tsp = rxk_ports; i < MAXRXPORTS; i++, tsp++) {
65 if (((ts = *tsp) == 0) && (zslot == -1))
71 /* otherwise allocate a new port slot */
73 return E2BIG; /* all full */
74 rxk_ports[zslot] = aport;
75 rxk_portRocks[zslot] = arock;
79 /* remove as port from the monitored list, port # is in network order */
81 rxk_DelPort(u_short aport)
86 for (i = 0, tsp = rxk_ports; i < MAXRXPORTS; i++, tsp++) {
88 /* found it, adjust ref count and free the port reference if all gone */
93 /* otherwise port not found */
98 rxk_shutdownPorts(void)
101 for (i = 0; i < MAXRXPORTS; i++) {
104 #if ! defined(AFS_SUN5_ENV) && ! defined(UKERNEL) && ! defined(RXK_LISTENER_ENV) && ! defined(AFS_SOCKPROXY_ENV)
105 soclose((struct socket *)rxk_portRocks[i]);
107 rxk_portRocks[i] = NULL;
113 rxi_GetHostUDPSocket(u_int host, u_short port)
116 sockp = (osi_socket *)rxk_NewSocketHost(host, port);
117 if (sockp == (osi_socket *)0)
118 return OSI_NULLSOCKET;
119 rxk_AddPort(port, (char *)sockp);
120 return (osi_socket) sockp;
124 rxi_GetUDPSocket(u_short port)
126 return rxi_GetHostUDPSocket(htonl(INADDR_ANY), port);
130 * osi_utoa() - write the NUL-terminated ASCII decimal form of the given
131 * unsigned long value into the given buffer. Returns 0 on success,
132 * and a value less than 0 on failure. The contents of the buffer is
133 * defined only on success.
137 osi_utoa(char *buf, size_t len, unsigned long val)
139 long k; /* index of first byte of string value */
141 /* we definitely need room for at least one digit and NUL */
147 /* compute the string form from the high end of the buffer */
150 for (k = len - 2; k >= 0; k--) {
151 buf[k] = val % 10 + '0';
158 /* did we finish converting val to string form? */
164 /* this should never happen */
170 /* this should never happen */
176 /* if necessary, relocate string to beginning of buf[] */
181 * We need to achieve the effect of calling
183 * memmove(buf, &buf[k], len - k);
185 * However, since memmove() is not available in all
186 * kernels, we explicitly do an appropriate copy.
192 while ((*dst++ = *src++) != '\0')
199 #ifndef AFS_LINUX_ENV
201 * osi_AssertFailK() -- used by the osi_Assert() macro.
203 * It essentially does
205 * osi_Panic("assertion failed: %s, file: %s, line: %d", expr, file, line);
207 * Since the kernel version of osi_Panic() only passes its first
208 * argument to the native panic(), we construct a single string and hand
209 * that to osi_Panic().
212 osi_AssertFailK(const char *expr, const char *file, int line)
214 static const char msg0[] = "assertion failed: ";
215 static const char msg1[] = ", file: ";
216 static const char msg2[] = ", line: ";
217 static const char msg3[] = "\n";
220 * These buffers add up to 1K, which is a pleasantly nice round
221 * value, but probably not vital.
226 /* check line number conversion */
228 if (osi_utoa(linebuf, sizeof linebuf, line) < 0) {
229 osi_Panic("osi_AssertFailK: error in osi_utoa()\n");
234 # define ADDBUF(BUF, STR) \
235 if (strlen(BUF) + strlen((char *)(STR)) + 1 <= sizeof BUF) { \
236 strlcat(BUF, (char *)(STR), sizeof(BUF)); \
245 ADDBUF(buf, linebuf);
250 osi_Panic("%s", buf);
252 #endif /* !AFS_LINUX_ENV */
255 /* This is the server process request loop. Kernel server
256 * processes never become listener threads */
258 rx_ServerProc(void *unused)
262 rxi_MorePackets(rx_maxReceiveWindow + 2); /* alloc more packets */
263 MUTEX_ENTER(&rx_quota_mutex);
264 rxi_dataQuota += rx_initSendWindow; /* Reserve some pkts for hard times */
265 /* threadID is used for making decisions in GetCall. Get it by bumping
266 * number of threads handling incoming calls */
267 threadID = rxi_availProcs++;
268 MUTEX_EXIT(&rx_quota_mutex);
270 rxi_ServerProc(threadID, NULL, NULL);
274 #endif /* !UKERNEL */
276 #ifndef RXK_LISTENER_ENV
277 /* asize includes the Rx header */
279 MyPacketProc(struct rx_packet **ahandle, int asize)
281 struct rx_packet *tp;
283 /* If this is larger than we expected, increase rx_maxReceiveDataSize */
284 /* If we can't scrounge enough cbufs, then we have to drop the packet,
285 * but we should set a flag so we magic up some more at our leisure.
288 if ((asize >= 0) && (asize <= RX_MAX_PACKET_SIZE)) {
289 tp = rxi_AllocPacket(RX_PACKET_CLASS_RECEIVE);
290 if (tp && (tp->length + RX_HEADER_SIZE) < asize) {
292 rxi_AllocDataBuf(tp, asize - (tp->length + RX_HEADER_SIZE),
293 RX_PACKET_CLASS_RECV_CBUF)) {
296 if (rx_stats_active) {
297 rx_atomic_inc(&rx_stats.noPacketBuffersOnRead);
303 * XXX if packet is too long for our buffer,
304 * should do this at a higher layer and let other
305 * end know we're losing.
307 if (rx_stats_active) {
308 rx_atomic_inc(&rx_stats.bogusPacketOnRead);
310 /* I DON"T LIKE THIS PRINTF -- PRINTFS MAKE THINGS VERY VERY SLOOWWW */
311 dpf(("rx: packet dropped: bad ulen=%d\n", asize));
317 /* otherwise we have a packet, set appropriate values */
323 MyArrivalProc(struct rx_packet *ahandle,
324 struct sockaddr_in *afrom,
325 struct socket *arock,
328 /* handle basic rx packet */
329 ahandle->length = asize - RX_HEADER_SIZE;
330 rxi_DecodePacketHeader(ahandle);
332 rxi_ReceivePacket(ahandle, arock,
333 afrom->sin_addr.s_addr, afrom->sin_port, NULL,
336 /* free the packet if it has been returned */
338 rxi_FreePacket(ahandle);
341 #endif /* !RXK_LISTENER_ENV */
344 rxi_StartListener(void)
346 #if !defined(RXK_LISTENER_ENV) && !defined(RXK_UPCALL_ENV)
347 /* if kernel, give name of appropriate procedures */
348 rxk_GetPacketProc = MyPacketProc;
349 rxk_PacketArrivalProc = MyArrivalProc;
354 /* Called from rxi_FindPeer, when initializing a clear rx_peer structure,
355 to get interesting information. */
357 rxi_InitPeerParams(struct rx_peer *pp)
362 # ifdef AFS_USERSPACE_IP_ADDR
366 i = rxi_Findcbi(pp->host);
368 rx_rto_setPeerTimeoutSecs(pp, 3);
369 pp->ifMTU = MIN(RX_REMOTE_PACKET_SIZE, rx_MyMaxSendSize);
371 rx_rto_setPeerTimeoutSecs(pp, 2);
372 pp->ifMTU = MIN(RX_MAX_PACKET_SIZE, rx_MyMaxSendSize);
373 mtu = ntohl(afs_cb_interface.mtu[i]);
374 /* Diminish the packet size to one based on the MTU given by
376 if (mtu > (RX_IPUDP_SIZE + RX_HEADER_SIZE)) {
377 rxmtu = mtu - RX_IPUDP_SIZE;
378 if (rxmtu < pp->ifMTU)
382 # else /* AFS_USERSPACE_IP_ADDR */
385 RX_NET_EPOCH_ENTER();
387 # if !defined(AFS_SGI_ENV)
388 if (numMyNetAddrs == 0)
389 (void)rxi_GetIFInfo();
392 ifn = rxi_FindIfnet(pp->host, NULL);
394 rx_rto_setPeerTimeoutSecs(pp, 2);
395 pp->ifMTU = MIN(RX_MAX_PACKET_SIZE, rx_MyMaxSendSize);
396 # ifdef IFF_POINTOPOINT
397 if (rx_ifnet_flags(ifn) & IFF_POINTOPOINT) {
398 /* wish we knew the bit rate and the chunk size, sigh. */
399 rx_rto_setPeerTimeoutSecs(pp, 4);
400 pp->ifMTU = RX_PP_PACKET_SIZE;
402 # endif /* IFF_POINTOPOINT */
403 /* Diminish the packet size to one based on the MTU given by
405 if (rx_ifnet_mtu(ifn) > (RX_IPUDP_SIZE + RX_HEADER_SIZE)) {
406 rxmtu = rx_ifnet_mtu(ifn) - RX_IPUDP_SIZE;
407 if (rxmtu < pp->ifMTU)
410 } else { /* couldn't find the interface, so assume the worst */
411 rx_rto_setPeerTimeoutSecs(pp, 3);
412 pp->ifMTU = MIN(RX_REMOTE_PACKET_SIZE, rx_MyMaxSendSize);
417 # endif /* else AFS_USERSPACE_IP_ADDR */
418 #else /* AFS_SUN5_ENV */
421 mtu = rxi_FindIfMTU(pp->host);
424 rx_rto_setPeerTimeoutSecs(pp, 3);
425 pp->ifMTU = MIN(RX_REMOTE_PACKET_SIZE, rx_MyMaxSendSize);
427 rx_rto_setPeerTimeoutSecs(pp, 2);
428 pp->ifMTU = MIN(RX_MAX_PACKET_SIZE, rx_MyMaxSendSize);
430 /* Diminish the packet size to one based on the MTU given by
432 if (mtu > (RX_IPUDP_SIZE + RX_HEADER_SIZE)) {
433 rxmtu = mtu - RX_IPUDP_SIZE;
434 if (rxmtu < pp->ifMTU)
438 #endif /* AFS_SUN5_ENV */
439 pp->ifMTU = rxi_AdjustIfMTU(pp->ifMTU);
440 pp->maxMTU = OLD_MAX_PACKET_SIZE; /* for compatibility with old guys */
441 pp->natMTU = MIN(pp->ifMTU, OLD_MAX_PACKET_SIZE);
443 MIN(rxi_nDgramPackets,
444 rxi_AdjustDgramPackets(rxi_nSendFrags, pp->ifMTU));
445 pp->maxDgramPackets = 1;
447 /* Initialize slow start parameters */
448 pp->MTU = MIN(pp->natMTU, pp->maxMTU);
450 pp->nDgramPackets = 1;
455 /* The following code is common to several system types, but not all. The
456 * separate ones are found in the system specific subdirectories.
460 #if ! defined(AFS_AIX_ENV) && ! defined(AFS_SUN5_ENV) && ! defined(UKERNEL) && ! defined(AFS_LINUX_ENV) && !defined (AFS_DARWIN_ENV) && !defined (AFS_XBSD_ENV)
461 /* Routine called during the afsd "-shutdown" process to put things back to
464 static struct protosw parent_proto; /* udp proto switch */
467 shutdown_rxkernel(void)
469 struct protosw *tpro, *last;
470 last = inetdomain.dom_protoswNPROTOSW;
471 for (tpro = inetdomain.dom_protosw; tpro < last; tpro++)
472 if (tpro->pr_protocol == IPPROTO_UDP) {
473 /* restore original udp protocol switch */
474 memcpy((void *)tpro, (void *)&parent_proto, sizeof(parent_proto));
475 memset((void *)&parent_proto, 0, sizeof(parent_proto));
480 dpf(("shutdown_rxkernel: no udp proto\n"));
482 #endif /* !AIX && !SUN && !NCR && !UKERNEL */
484 #if !defined(AFS_SUN5_ENV) && !defined(AFS_SGI_ENV)
485 /* Determine what the network interfaces are for this machine. */
487 # ifdef AFS_USERSPACE_IP_ADDR
491 int i, j, different = 0, num = ADDRSPERSITE;
494 afs_uint32 addrs[ADDRSPERSITE];
495 int mtus[ADDRSPERSITE];
497 memset((void *)addrs, 0, sizeof(addrs));
498 memset((void *)mtus, 0, sizeof(mtus));
500 if (afs_cb_interface.numberOfInterfaces < num)
501 num = afs_cb_interface.numberOfInterfaces;
502 for (i = 0; i < num; i++) {
503 if (!afs_cb_interface.mtu[i])
504 afs_cb_interface.mtu[i] = htonl(1500);
505 rxmtu = (ntohl(afs_cb_interface.mtu[i]) - RX_IPUDP_SIZE);
506 ifinaddr = ntohl(afs_cb_interface.addr_in[i]);
507 if (myNetAddrs[i] != ifinaddr)
511 rxmtu = rxi_AdjustIfMTU(rxmtu);
513 rxmtu * rxi_nRecvFrags + ((rxi_nRecvFrags - 1) * UDP_HDR_SIZE);
514 maxmtu = rxi_AdjustMaxMTU(rxmtu, maxmtu);
515 addrs[i++] = ifinaddr;
516 if (!rx_IsLoopbackAddr(ifinaddr) && (maxmtu > rx_maxReceiveSize)) {
517 rx_maxReceiveSize = MIN(RX_MAX_PACKET_SIZE, maxmtu);
518 rx_maxReceiveSize = MIN(rx_maxReceiveSize, rx_maxReceiveSizeUser);
522 rx_maxJumboRecvSize =
523 RX_HEADER_SIZE + (rxi_nDgramPackets * RX_JUMBOBUFFERSIZE) +
524 ((rxi_nDgramPackets - 1) * RX_JUMBOHEADERSIZE);
525 rx_maxJumboRecvSize = MAX(rx_maxJumboRecvSize, rx_maxReceiveSize);
528 for (j = 0; j < i; j++) {
529 myNetMTUs[j] = mtus[j];
530 myNetAddrs[j] = addrs[j];
537 /* Returns the afs_cb_interface inxex which best matches address.
538 * If none is found, we return -1.
541 rxi_Findcbi(afs_uint32 addr)
544 afs_uint32 myAddr, thisAddr, netMask, subnetMask;
545 afs_int32 rvalue = -1;
548 if (numMyNetAddrs == 0)
549 (void)rxi_GetcbiInfo();
551 myAddr = ntohl(addr);
553 if (IN_CLASSA(myAddr))
554 netMask = IN_CLASSA_NET;
555 else if (IN_CLASSB(myAddr))
556 netMask = IN_CLASSB_NET;
557 else if (IN_CLASSC(myAddr))
558 netMask = IN_CLASSC_NET;
562 for (j = 0; j < afs_cb_interface.numberOfInterfaces; j++) {
563 thisAddr = ntohl(afs_cb_interface.addr_in[j]);
564 subnetMask = ntohl(afs_cb_interface.subnetmask[j]);
565 if ((myAddr & netMask) == (thisAddr & netMask)) {
566 if ((myAddr & subnetMask) == (thisAddr & subnetMask)) {
567 if (myAddr == thisAddr) {
572 if (match_value < 3) {
577 if (match_value < 2) {
588 # else /* AFS_USERSPACE_IP_ADDR */
590 # if !defined(AFS_AIX41_ENV) && !defined(AFS_DARWIN_ENV) && !defined(AFS_XBSD_ENV)
591 # define IFADDR2SA(f) (&((f)->ifa_addr))
593 # define IFADDR2SA(f) ((f)->ifa_addr)
603 afs_uint32 addrs[ADDRSPERSITE];
604 int mtus[ADDRSPERSITE];
606 # if defined(AFS_DARWIN80_ENV)
612 struct sockaddr sout;
613 struct sockaddr_in *sin;
616 rx_ifaddr_t ifad; /* ifnet points to a if_addrlist of ifaddrs */
620 memset(addrs, 0, sizeof(addrs));
621 memset(mtus, 0, sizeof(mtus));
623 # if defined(AFS_DARWIN80_ENV)
624 if (!ifnet_list_get(AF_INET, &ifns, &count)) {
625 for (m = 0; m < count; m++) {
626 if (!ifnet_get_address_list(ifns[m], &ifads)) {
627 for (j = 0; ifads[j] != NULL && cnt < ADDRSPERSITE; j++) {
628 if ((t = ifaddr_address(ifads[j], &sout, sizeof(struct sockaddr))) == 0) {
629 sin = (struct sockaddr_in *)&sout;
630 rxmtu = rx_ifnet_mtu(rx_ifaddr_ifnet(ifads[j])) - RX_IPUDP_SIZE;
631 ifinaddr = ntohl(sin->sin_addr.s_addr);
632 if (myNetAddrs[i] != ifinaddr) {
636 rxmtu = rxi_AdjustIfMTU(rxmtu);
638 rxmtu * rxi_nRecvFrags +
639 ((rxi_nRecvFrags - 1) * UDP_HDR_SIZE);
640 maxmtu = rxi_AdjustMaxMTU(rxmtu, maxmtu);
641 addrs[i++] = ifinaddr;
642 if (!rx_IsLoopbackAddr(ifinaddr) &&
643 (maxmtu > rx_maxReceiveSize)) {
645 MIN(RX_MAX_PACKET_SIZE, maxmtu);
647 MIN(rx_maxReceiveSize, rx_maxReceiveSizeUser);
652 ifnet_free_address_list(ifads);
655 ifnet_list_free(ifns);
657 # else /* AFS_DARWIN80_ENV */
658 # if defined(AFS_DARWIN_ENV)
659 TAILQ_FOREACH(ifn, &ifnet, if_link) {
660 if (i >= ADDRSPERSITE)
662 # elif defined(AFS_FBSD_ENV)
663 CURVNET_SET(rx_socket->so_vnet);
665 AFS_FBSD_NET_FOREACH(ifn, &V_ifnet, if_link) {
666 if (i >= ADDRSPERSITE)
668 # elif defined(AFS_OBSD_ENV) || defined(AFS_NBSD_ENV)
669 for (ifn = ifnet.tqh_first; i < ADDRSPERSITE && ifn != NULL;
670 ifn = ifn->if_list.tqe_next) {
672 for (ifn = ifnet; ifn != NULL && i < ADDRSPERSITE; ifn = ifn->if_next) {
674 rxmtu = (ifn->if_mtu - RX_IPUDP_SIZE);
675 # if defined(AFS_DARWIN_ENV)
676 TAILQ_FOREACH(ifad, &ifn->if_addrhead, ifa_link) {
677 if (i >= ADDRSPERSITE)
679 # elif defined(AFS_FBSD_ENV)
681 AFS_FBSD_NET_FOREACH(ifad, &ifn->if_addrhead, ifa_link) {
682 if (i >= ADDRSPERSITE)
684 # elif defined(AFS_OBSD_ENV) || defined(AFS_NBSD_ENV)
685 for (ifad = ifn->if_addrlist.tqh_first;
686 ifad != NULL && i < ADDRSPERSITE;
687 ifad = ifad->ifa_list.tqe_next) {
689 for (ifad = ifn->if_addrlist; ifad != NULL && i < ADDRSPERSITE;
690 ifad = ifad->ifa_next) {
692 if (IFADDR2SA(ifad)->sa_family == AF_INET) {
694 ntohl(((struct sockaddr_in *)IFADDR2SA(ifad))->sin_addr.
696 if (myNetAddrs[i] != ifinaddr) {
700 rxmtu = rxi_AdjustIfMTU(rxmtu);
702 rxmtu * rxi_nRecvFrags +
703 ((rxi_nRecvFrags - 1) * UDP_HDR_SIZE);
704 maxmtu = rxi_AdjustMaxMTU(rxmtu, maxmtu);
705 addrs[i++] = ifinaddr;
706 if (!rx_IsLoopbackAddr(ifinaddr) && (maxmtu > rx_maxReceiveSize)) {
707 rx_maxReceiveSize = MIN(RX_MAX_PACKET_SIZE, maxmtu);
709 MIN(rx_maxReceiveSize, rx_maxReceiveSizeUser);
714 if_addr_runlock(ifn);
717 # endif /* !AFS_DARWIN80_ENV */
719 rx_maxJumboRecvSize =
720 RX_HEADER_SIZE + rxi_nDgramPackets * RX_JUMBOBUFFERSIZE +
721 (rxi_nDgramPackets - 1) * RX_JUMBOHEADERSIZE;
722 rx_maxJumboRecvSize = MAX(rx_maxJumboRecvSize, rx_maxReceiveSize);
726 for (l = 0; l < i; l++) {
727 myNetMTUs[l] = mtus[l];
728 myNetAddrs[l] = addrs[l];
740 # if defined(AFS_DARWIN_ENV) || defined(AFS_XBSD_ENV)
741 /* Returns ifnet which best matches address */
743 rxi_FindIfnet(afs_uint32 addr, afs_uint32 * maskp)
745 struct sockaddr_in s, sr;
750 CURVNET_SET(rx_socket->so_vnet);
753 s.sin_family = AF_INET;
754 s.sin_addr.s_addr = addr;
755 ifad = rx_ifaddr_withnet((struct sockaddr *)&s);
758 rx_ifaddr_netmask(ifad, (struct sockaddr *)&sr, sizeof(sr));
759 *maskp = sr.sin_addr.s_addr;
762 ret = (ifad ? rx_ifaddr_ifnet(ifad) : NULL);
771 # else /* DARWIN || XBSD */
773 /* Returns ifnet which best matches address */
775 rxi_FindIfnet(afs_uint32 addr, afs_uint32 * maskp)
778 extern struct in_ifaddr *in_ifaddr;
779 struct in_ifaddr *ifa, *ifad = NULL;
783 for (ifa = in_ifaddr; ifa; ifa = ifa->ia_next) {
784 if ((addr & ifa->ia_netmask) == ifa->ia_net) {
785 if ((addr & ifa->ia_subnetmask) == ifa->ia_subnet) {
786 if (IA_SIN(ifa)->sin_addr.s_addr == addr) { /* ie, ME!!! */
791 if (match_value < 3) {
796 if (match_value < 2) {
801 } /* if net matches */
802 } /* for all in_ifaddrs */
806 *maskp = ifad->ia_subnetmask;
807 return (ifad ? ifad->ia_ifp : NULL);
809 # endif /* else DARWIN || XBSD */
810 # endif /* else AFS_USERSPACE_IP_ADDR */
811 #endif /* !SUN5 && !SGI */
814 /* rxk_NewSocket, rxk_FreeSocket and osi_NetSend are from the now defunct
815 * afs_osinet.c. One could argue that rxi_NewSocket could go into the
816 * system specific subdirectories for all systems. But for the moment,
817 * most of it is simple to follow common code.
819 #if !defined(UKERNEL)
820 # if !defined(AFS_SUN5_ENV) && !defined(AFS_LINUX_ENV) && !defined(AFS_SOCKPROXY_ENV)
821 /* rxk_NewSocket creates a new socket on the specified port. The port is
822 * in network byte order.
825 rxk_NewSocketHost(afs_uint32 ahost, short aport)
828 # ifdef AFS_DARWIN80_ENV
831 struct socket *newSocket;
833 # if (!defined(AFS_HPUX1122_ENV) && !defined(AFS_FBSD_ENV))
836 struct sockaddr_in myaddr;
837 # ifdef AFS_HPUX110_ENV
838 /* prototype copied from kernel source file streams/str_proto.h */
839 extern MBLKP allocb_wait(int, int);
841 int addrsize = sizeof(struct sockaddr_in);
843 extern struct fileops socketops;
849 AFS_STATCNT(osi_NewSocket);
850 # if (defined(AFS_DARWIN_ENV) || defined(AFS_XBSD_ENV)) && defined(KERNEL_FUNNEL)
851 thread_funnel_switch(KERNEL_FUNNEL, NETWORK_FUNNEL);
853 # if defined(AFS_HPUX102_ENV)
854 # if defined(AFS_HPUX110_ENV)
855 /* we need a file associated with the socket so sosend in NetSend
857 /* blocking socket */
858 code = socreate(AF_INET, &newSocket, SOCK_DGRAM, 0, 0);
862 fp->f_flag = FREAD | FWRITE;
863 fp->f_type = DTYPE_SOCKET;
864 fp->f_ops = &socketops;
866 fp->f_data = (void *)newSocket;
867 newSocket->so_fp = (void *)fp;
869 # else /* AFS_HPUX110_ENV */
870 code = socreate(AF_INET, &newSocket, SOCK_DGRAM, 0, SS_NOWAIT);
871 # endif /* else AFS_HPUX110_ENV */
872 # elif defined(AFS_SGI_ENV) || defined(AFS_OBSD_ENV)
873 code = socreate(AF_INET, &newSocket, SOCK_DGRAM, IPPROTO_UDP);
874 # elif defined(AFS_FBSD_ENV)
875 code = socreate(AF_INET, &newSocket, SOCK_DGRAM, IPPROTO_UDP,
876 afs_osi_credp, curthread);
877 # elif defined(AFS_DARWIN80_ENV)
878 # ifdef RXK_LISTENER_ENV
879 code = sock_socket(AF_INET, SOCK_DGRAM, IPPROTO_UDP, NULL, NULL, &newSocket);
881 code = sock_socket(AF_INET, SOCK_DGRAM, IPPROTO_UDP, rx_upcall, NULL, &newSocket);
883 # elif defined(AFS_NBSD50_ENV)
884 code = socreate(AF_INET, &newSocket, SOCK_DGRAM, 0, osi_curproc(), NULL);
885 # elif defined(AFS_NBSD40_ENV)
886 code = socreate(AF_INET, &newSocket, SOCK_DGRAM, 0, osi_curproc());
888 code = socreate(AF_INET, &newSocket, SOCK_DGRAM, 0);
889 # endif /* AFS_HPUX102_ENV */
893 memset(&myaddr, 0, sizeof myaddr);
894 myaddr.sin_family = AF_INET;
895 myaddr.sin_port = aport;
896 myaddr.sin_addr.s_addr = ahost;
897 # ifdef STRUCT_SOCKADDR_HAS_SA_LEN
898 myaddr.sin_len = sizeof(myaddr);
901 # ifdef AFS_HPUX110_ENV
902 bindnam = allocb_wait((addrsize + SO_MSGOFFSET + 1), BPRI_MED);
907 memcpy((caddr_t) bindnam->b_rptr + SO_MSGOFFSET, (caddr_t) & myaddr,
909 bindnam->b_wptr = bindnam->b_rptr + (addrsize + SO_MSGOFFSET + 1);
910 code = sobind(newSocket, bindnam, addrsize);
913 # if !defined(AFS_HPUX1122_ENV)
920 # else /* AFS_HPUX110_ENV */
921 # if defined(AFS_DARWIN80_ENV)
926 code = sock_setsockopt(newSocket, SOL_SOCKET, SO_SNDBUF,
927 &buflen, sizeof(buflen));
928 code2 = sock_setsockopt(newSocket, SOL_SOCKET, SO_RCVBUF,
929 &buflen, sizeof(buflen));
933 osi_Panic("osi_NewSocket: last attempt to reserve 32K failed!\n");
938 # if defined(AFS_NBSD_ENV)
941 code = soreserve(newSocket, 50000, 50000);
943 code = soreserve(newSocket, 32766, 32766);
945 osi_Panic("osi_NewSocket: last attempt to reserve 32K failed!\n");
947 # if defined(AFS_NBSD_ENV)
951 # if defined(AFS_DARWIN_ENV) || defined(AFS_FBSD_ENV)
952 # if defined(AFS_FBSD_ENV)
953 code = sobind(newSocket, (struct sockaddr *)&myaddr, curthread);
955 code = sobind(newSocket, (struct sockaddr *)&myaddr);
958 dpf(("sobind fails (%d)\n", (int)code));
962 # else /* defined(AFS_DARWIN_ENV) || defined(AFS_FBSD_ENV) */
963 nam = m_get(M_WAIT, MT_SONAME);
965 # if defined(KERNEL_HAVE_UERROR)
970 nam->m_len = sizeof(myaddr);
971 memcpy(mtod(nam, caddr_t), &myaddr, sizeof(myaddr));
972 # if defined(AFS_SGI_ENV)
973 BHV_PDATA(&bhv) = (void *)newSocket;
974 code = sobind(&bhv, nam);
976 # elif defined(AFS_OBSD44_ENV) || defined(AFS_NBSD40_ENV)
977 code = sobind(newSocket, nam, osi_curproc());
979 code = sobind(newSocket, nam);
982 dpf(("sobind fails (%d)\n", (int)code));
989 # endif /* else defined(AFS_DARWIN_ENV) || defined(AFS_FBSD_ENV) */
990 # endif /* else AFS_HPUX110_ENV */
992 # if defined(AFS_DARWIN_ENV) && defined(KERNEL_FUNNEL)
993 thread_funnel_switch(NETWORK_FUNNEL, KERNEL_FUNNEL);
995 return (osi_socket *)newSocket;
998 # if defined(AFS_DARWIN_ENV) && defined(KERNEL_FUNNEL)
999 thread_funnel_switch(NETWORK_FUNNEL, KERNEL_FUNNEL);
1001 return (osi_socket *)0;
1005 rxk_NewSocket(short aport)
1007 return rxk_NewSocketHost(0, aport);
1010 /* free socket allocated by rxk_NewSocket */
1012 rxk_FreeSocket(struct socket *asocket)
1014 AFS_STATCNT(osi_FreeSocket);
1015 # if defined(AFS_DARWIN_ENV) && defined(KERNEL_FUNNEL)
1016 thread_funnel_switch(KERNEL_FUNNEL, NETWORK_FUNNEL);
1018 # ifdef AFS_HPUX110_ENV
1019 if (asocket->so_fp) {
1020 struct file *fp = asocket->so_fp;
1021 # if !defined(AFS_HPUX1123_ENV)
1022 /* 11.23 still has falloc, but not FPENTRYFREE !
1023 * so for now if we shutdown, we will waist a file
1026 asocket->so_fp = NULL;
1029 # endif /* AFS_HPUX110_ENV */
1031 # if defined(AFS_DARWIN_ENV) && defined(KERNEL_FUNNEL)
1032 thread_funnel_switch(NETWORK_FUNNEL, KERNEL_FUNNEL);
1036 # endif /* !AFS_SUN5_ENV && !AFS_LINUX_ENV && !AFS_SOCKPROXY_ENV */
1038 # if defined(RXK_LISTENER_ENV) || defined(AFS_SUN5_ENV) || defined(RXK_UPCALL_ENV)
1039 # ifdef RXK_TIMEDSLEEP_ENV
1040 /* Shutting down should wake us up, as should an earlier event. */
1042 rxi_ReScheduleEvents(void)
1044 /* needed to allow startup */
1045 int glock = ISAFS_GLOCK();
1048 osi_rxWakeup(&afs_termState);
1054 * Run RX event daemon every second (5 times faster than rest of systems)
1057 afs_rxevent_daemon(void)
1063 # ifdef RX_ENABLE_LOCKS
1065 # endif /* RX_ENABLE_LOCKS */
1067 rxevent_RaiseEvents(&temp);
1069 # ifdef RX_ENABLE_LOCKS
1071 # endif /* RX_ENABLE_LOCKS */
1072 # ifdef RX_KERNEL_TRACE
1073 afs_Trace1(afs_iclSetp, CM_TRACE_TIMESTAMP, ICL_TYPE_STRING,
1074 "before afs_osi_Wait()");
1076 # ifdef RXK_TIMEDSLEEP_ENV
1077 afs_osi_TimedSleep(&afs_termState, MAX(500, ((temp.sec * 1000) +
1078 (temp.usec / 1000))), 0);
1080 afs_osi_Wait(500, NULL, 0);
1082 # ifdef RX_KERNEL_TRACE
1083 afs_Trace1(afs_iclSetp, CM_TRACE_TIMESTAMP, ICL_TYPE_STRING,
1084 "after afs_osi_Wait()");
1086 if (afs_termState == AFSOP_STOP_RXEVENT) {
1087 # ifdef RXK_LISTENER_ENV
1088 afs_termState = AFSOP_STOP_RXK_LISTENER;
1089 # elif defined(AFS_SOCKPROXY_ENV)
1090 afs_termState = AFSOP_STOP_SOCKPROXY;
1091 # elif defined(AFS_SUN510_ENV) || defined(RXK_UPCALL_ENV)
1092 afs_termState = AFSOP_STOP_NETIF;
1094 afs_termState = AFSOP_STOP_COMPLETE;
1096 osi_rxWakeup(&afs_termState);
1101 # endif /* RXK_LISTENER_ENV || AFS_SUN5_ENV || RXK_UPCALL_ENV */
1103 # ifdef RXK_LISTENER_ENV
1105 /* rxk_ReadPacket returns 1 if valid packet, 0 on error. */
1107 rxk_ReadPacket(osi_socket so, struct rx_packet *p, int *host, int *port)
1110 struct sockaddr_in from;
1114 afs_int32 savelen; /* was using rlen but had aliasing problems */
1115 rx_computelen(p, tlen);
1116 rx_SetDataSize(p, tlen); /* this is the size of the user data area */
1118 tlen += RX_HEADER_SIZE; /* now this is the size of the entire packet */
1119 rlen = rx_maxJumboRecvSize; /* this is what I am advertising. Only check
1120 * it once in order to avoid races. */
1123 tlen = rxi_AllocDataBuf(p, tlen, RX_PACKET_CLASS_RECV_CBUF);
1131 /* add some padding to the last iovec, it's just to make sure that the
1132 * read doesn't return more data than we expect, and is done to get around
1133 * our problems caused by the lack of a length field in the rx header. */
1134 savelen = p->wirevec[p->niovecs - 1].iov_len;
1135 p->wirevec[p->niovecs - 1].iov_len = savelen + RX_EXTRABUFFERSIZE;
1137 nbytes = tlen + sizeof(afs_int32);
1138 # ifdef RX_KERNEL_TRACE
1139 if (ICL_SETACTIVE(afs_iclSetp)) {
1141 afs_Trace1(afs_iclSetp, CM_TRACE_TIMESTAMP, ICL_TYPE_STRING,
1142 "before osi_NetRecive()");
1146 code = osi_NetReceive(rx_socket, &from, p->wirevec, p->niovecs, &nbytes);
1148 # ifdef RX_KERNEL_TRACE
1149 if (ICL_SETACTIVE(afs_iclSetp)) {
1151 afs_Trace1(afs_iclSetp, CM_TRACE_TIMESTAMP, ICL_TYPE_STRING,
1152 "after osi_NetRecive()");
1156 /* restore the vec to its correct state */
1157 p->wirevec[p->niovecs - 1].iov_len = savelen;
1160 p->length = nbytes - RX_HEADER_SIZE;;
1161 if ((nbytes > tlen) || (p->length & 0x8000)) { /* Bogus packet */
1163 if (rx_stats_active) {
1164 MUTEX_ENTER(&rx_stats_mutex);
1165 rx_atomic_inc(&rx_stats.bogusPacketOnRead);
1166 rx_stats.bogusHost = from.sin_addr.s_addr;
1167 MUTEX_EXIT(&rx_stats_mutex);
1169 dpf(("B: bogus packet from [%x,%d] nb=%d\n",
1170 from.sin_addr.s_addr, from.sin_port, nbytes));
1174 /* Extract packet header. */
1175 rxi_DecodePacketHeader(p);
1177 *host = from.sin_addr.s_addr;
1178 *port = from.sin_port;
1179 if (p->header.type > 0 && p->header.type <= RX_N_PACKET_TYPES) {
1180 if (rx_stats_active) {
1181 rx_atomic_inc(&rx_stats.packetsRead[p->header.type - 1]);
1185 # ifdef RX_TRIMDATABUFS
1186 /* Free any empty packet buffers at the end of this packet */
1187 rxi_TrimDataBufs(p, 1);
1197 * Listen for packets on socket. This thread is typically started after
1198 * rx_Init has called rxi_StartListener(), but nevertheless, ensures that
1199 * the start state is set before proceeding.
1201 * Note that this thread is outside the AFS global lock for much of
1204 * In many OS's, the socket receive code sleeps interruptibly. That's not what
1205 * we want here. So we need to either block all signals (including SIGKILL
1206 * and SIGSTOP) or reset the thread's signal state to unsignalled when the
1207 * OS's socket receive routine returns as a result of a signal.
1209 int rxk_ListenerPid; /* Used to signal process to wakeup at shutdown */
1210 # ifdef AFS_LINUX_ENV
1211 struct task_struct *rxk_ListenerTask;
1217 struct rx_packet *rxp = NULL;
1221 # ifdef AFS_LINUX_ENV
1222 rxk_ListenerPid = current->pid;
1223 rxk_ListenerTask = current;
1224 allow_signal(SIGKILL); /* Allowed, but blocked until shutdown */
1226 # ifdef AFS_SUN5_ENV
1227 rxk_ListenerPid = 1; /* No PID, just a flag that we're alive */
1228 # endif /* AFS_SUN5_ENV */
1229 # ifdef AFS_XBSD_ENV
1230 rxk_ListenerPid = curproc->p_pid;
1231 # endif /* AFS_FBSD_ENV */
1232 # ifdef AFS_DARWIN80_ENV
1233 rxk_ListenerPid = proc_selfpid();
1234 # elif defined(AFS_DARWIN_ENV)
1235 rxk_ListenerPid = current_proc()->p_pid;
1237 # ifdef RX_ENABLE_LOCKS
1239 # endif /* RX_ENABLE_LOCKS */
1240 while (afs_termState != AFSOP_STOP_RXK_LISTENER) {
1241 /* See if a check for additional packets was issued */
1245 rxi_RestoreDataBufs(rxp);
1247 rxp = rxi_AllocPacket(RX_PACKET_CLASS_RECEIVE);
1249 osi_Panic("rxk_Listener: No more Rx buffers!\n");
1251 if (!(code = rxk_ReadPacket(rx_socket, rxp, &host, &port))) {
1252 rxp = rxi_ReceivePacket(rxp, rx_socket, host, port, 0, 0);
1256 # ifdef RX_ENABLE_LOCKS
1258 # endif /* RX_ENABLE_LOCKS */
1259 if (afs_termState == AFSOP_STOP_RXK_LISTENER) {
1260 # ifdef AFS_SUN510_ENV
1261 afs_termState = AFSOP_STOP_NETIF;
1263 afs_termState = AFSOP_STOP_COMPLETE;
1265 osi_rxWakeup(&afs_termState);
1267 rxk_ListenerPid = 0;
1268 # ifdef AFS_LINUX_ENV
1269 rxk_ListenerTask = 0;
1270 osi_rxWakeup(&rxk_ListenerTask);
1272 # if defined(AFS_SUN5_ENV) || defined(AFS_FBSD_ENV)
1273 osi_rxWakeup(&rxk_ListenerPid);
1277 # if !defined(AFS_LINUX_ENV) && !defined(AFS_SUN5_ENV) && !defined(AFS_DARWIN_ENV) && !defined(AFS_XBSD_ENV)
1278 /* The manner of stopping the rx listener thread may vary. Most unix's should
1279 * be able to call soclose.
1282 osi_StopListener(void)
1287 # endif /* RXK_LISTENER_ENV */
1288 #endif /* !UKERNEL */
1290 #if !defined(AFS_AIX_ENV) || (defined(AFS_AIX_ENV) && (!defined(KERNEL) || defined(UKERNEL)))
1292 osi_Msg(const char *fmt, ...)
1296 # if defined(AFS_LINUX_ENV)
1305 #if !defined(AFS_LINUX_ENV)
1307 # if defined(AFS_AIX_ENV)
1308 osi_Panic(char *msg, void *a1, void *a2, void *a3)
1310 osi_Panic(char *msg, ...)
1315 msg = "Unknown AFS panic";
1317 * we should probably use the errsave facility here. it is not
1321 printf(msg, a1, a2, a3);
1323 # elif defined(AFS_SGI_ENV)
1326 /* Solaris has vcmn_err, Sol10 01/06 may have issues. Beware. */
1328 cmn_err(CE_PANIC, "Unknown AFS panic");
1331 icmn_err(CE_PANIC, msg, ap);
1334 # elif defined(AFS_DARWIN80_ENV) || defined(AFS_LINUX_ENV) || defined(AFS_FBSD_ENV) || defined(UKERNEL)
1338 msg = "Unknown AFS panic";
1341 vsnprintf(buf, sizeof(buf), msg, ap);
1345 # else /* DARWIN80 || LINUX || FBSD || UKERNEL */
1348 msg = "Unknown AFS panic";
1353 # ifdef AFS_LINUX_ENV
1358 # endif /* else DARWIN80 || LINUX || FBSD || UKERNEL */
1361 #endif /* !AFS_LINUX_ENV */