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
9 * Portions Copyright (c) 2006 Sine Nomine Associates
12 #include <afsconfig.h>
13 #include <afs/param.h>
19 #ifdef HAVE_SYS_FILE_H
26 #include <afs/afsint.h>
27 #define FSINT_COMMON_XG
28 #include <afs/afscbint.h>
29 #include <afs/rxgen_consts.h>
31 #include <afs/errors.h>
32 #include <afs/ihandle.h>
33 #include <afs/vnode.h>
34 #include <afs/volume.h>
36 #include <afs/ptclient.h>
37 #include <afs/ptuser.h>
38 #include <afs/prs_fs.h>
40 #include <afs/afsutil.h>
41 #include <afs/com_err.h>
43 #include <afs/cellconfig.h>
44 #include "viced_prototypes.h"
48 #ifdef AFS_DEMAND_ATTACH_FS
49 #include "../util/afsutil_prototypes.h"
50 #include "serialize_state.h"
51 #endif /* AFS_DEMAND_ATTACH_FS */
53 pthread_mutex_t host_glock_mutex;
56 extern int CurrentConnections;
58 extern int AnonymousID;
59 extern prlist AnonCPS;
61 extern struct afsconf_dir *confDir; /* config dir object */
62 extern int lwps; /* the max number of server threads */
63 extern afsUUID FS_HostUUID;
64 extern char *FS_configPath;
67 int CEs = 0; /* active clients */
68 int CEBlocks = 0; /* number of blocks of CEs */
69 struct client *CEFree = 0; /* first free client */
70 struct host *hostList = 0; /* linked list of all hosts */
71 int hostCount = 0; /* number of hosts in hostList */
75 static struct rx_securityClass *sc = NULL;
77 /* arguments for PerHost_EnumerateClient enumeration */
78 struct enumclient_args {
80 int (*proc)(struct client *client, void *rock);
84 static void h_SetupCallbackConn_r(struct host * host);
85 static int h_threadquota(int);
86 static int initInterfaceAddr_r(struct host *, struct interfaceAddr *);
88 #define CESPERBLOCK 73
89 struct CEBlock { /* block of CESPERBLOCK file entries */
90 struct client entry[CESPERBLOCK];
93 void h_TossStuff_r(struct host *host);
96 * Make sure the subnet macros have been defined.
99 #define IN_SUBNETA(i) ((((afs_int32)(i))&0x80800000)==0x00800000)
102 #ifndef IN_CLASSA_SUBNET
103 #define IN_CLASSA_SUBNET 0xffff0000
107 #define IN_SUBNETB(i) ((((afs_int32)(i))&0xc0008000)==0x80008000)
110 #ifndef IN_CLASSB_SUBNET
111 #define IN_CLASSB_SUBNET 0xffffff00
115 /* get a new block of CEs and chain it on CEFree */
119 struct CEBlock *block;
122 block = malloc(sizeof(struct CEBlock));
124 ViceLog(0, ("Failed malloc in GetCEBlock\n"));
125 ShutDownAndCore(PANIC);
128 for (i = 0; i < (CESPERBLOCK - 1); i++) {
129 Lock_Init(&block->entry[i].lock);
130 block->entry[i].next = &(block->entry[i + 1]);
132 block->entry[CESPERBLOCK - 1].next = 0;
133 Lock_Init(&block->entry[CESPERBLOCK - 1].lock);
134 CEFree = (struct client *)block;
140 /* get the next available CE */
141 static struct client *
144 struct client *entry;
149 ViceLog(0, ("CEFree NULL in GetCE\n"));
150 ShutDownAndCore(PANIC);
154 CEFree = entry->next;
156 memset(entry, 0, CLIENT_TO_ZERO(entry));
162 /* return an entry to the free list */
164 FreeCE(struct client *entry)
166 entry->VenusEpoch = 0;
168 entry->next = CEFree;
175 * The HTs and HTBlocks variables were formerly static, but they are
176 * now referenced elsewhere in the FileServer.
178 int HTs = 0; /* active file entries */
179 int HTBlocks = 0; /* number of blocks of HTs */
180 static struct host *HTFree = 0; /* first free file entry */
183 * Hash tables of host pointers. We need two tables, one
184 * to map IP addresses onto host pointers, and another
185 * to map host UUIDs onto host pointers.
187 static struct h_AddrHashChain *hostAddrHashTable[h_HASHENTRIES];
188 static struct h_UuidHashChain *hostUuidHashTable[h_HASHENTRIES];
189 #define h_HashIndex(hostip) (ntohl(hostip) & (h_HASHENTRIES-1))
190 #define h_UuidHashIndex(uuidp) (((int)(afs_uuid_hash(uuidp))) & (h_HASHENTRIES-1))
192 struct HTBlock { /* block of HTSPERBLOCK file entries */
193 struct host entry[h_HTSPERBLOCK];
197 /* get a new block of HTs and chain it on HTFree */
201 struct HTBlock *block;
203 static int index = 0;
205 if (HTBlocks == h_MAXHOSTTABLES) {
206 ViceLog(0, ("h_MAXHOSTTABLES reached\n"));
210 block = malloc(sizeof(struct HTBlock));
212 ViceLog(0, ("Failed malloc in GetHTBlock\n"));
213 ShutDownAndCore(PANIC);
215 for (i = 0; i < (h_HTSPERBLOCK); i++)
216 CV_INIT(&block->entry[i].cond, "block entry", CV_DEFAULT, 0);
217 for (i = 0; i < (h_HTSPERBLOCK); i++)
218 Lock_Init(&block->entry[i].lock);
219 for (i = 0; i < (h_HTSPERBLOCK - 1); i++)
220 block->entry[i].next = &(block->entry[i + 1]);
221 for (i = 0; i < (h_HTSPERBLOCK); i++)
222 block->entry[i].index = index++;
223 block->entry[h_HTSPERBLOCK - 1].next = 0;
224 HTFree = (struct host *)block;
225 hosttableptrs[HTBlocks++] = block->entry;
230 /* get the next available HT */
241 HTFree = entry->next;
243 memset(entry, 0, HOST_TO_ZERO(entry));
249 /* return an entry to the free list */
251 FreeHT(struct host *entry)
253 entry->next = HTFree;
260 hpr_Initialize(struct ubik_client **uclient)
263 struct rx_connection *serverconns[MAXSERVERS];
264 struct rx_securityClass *sc;
265 struct afsconf_dir *tdir;
267 struct afsconf_cell info;
271 tdir = afsconf_Open(FS_configPath);
274 ("hpr_Initialize: Could not open configuration directory: %s",
279 code = afsconf_GetLocalCell(tdir, cellstr, sizeof(cellstr));
281 ViceLog(0, ("hpr_Initialize: Could not get local cell. [%d]", code));
286 code = afsconf_GetCellInfo(tdir, cellstr, "afsprot", &info);
288 ViceLog(0, ("hpr_Initialize: Could not locate cell %s in %s/%s",
289 cellstr, tdir->name, AFSDIR_CELLSERVDB_FILE));
296 ViceLog(0, ("hpr_Initialize: Could not initialize rx."));
301 /* Most callers use secLevel==1, however, the fileserver uses secLevel==2
302 * to force use of the KeyFile. secLevel == 0 implies -noauth was
304 code = afsconf_ClientAuthSecure(tdir, &sc, &scIndex);
306 ViceLog(0, ("hpr_Initialize: clientauthsecure returns %d %s "
307 "(so trying noauth)", code, afs_error_message(code)));
308 scIndex = RX_SECIDX_NULL;
309 sc = rxnull_NewClientSecurityObject();
312 if (scIndex == RX_SECIDX_NULL)
313 ViceLog(0, ("hpr_Initialize: Could not get afs tokens, "
314 "running unauthenticated. [%d]", code));
316 memset(serverconns, 0, sizeof(serverconns)); /* terminate list!!! */
317 for (i = 0; i < info.numServers; i++) {
319 rx_NewConnection(info.hostAddr[i].sin_addr.s_addr,
320 info.hostAddr[i].sin_port, PRSRV,
324 code = ubik_ClientInit(serverconns, uclient);
326 ViceLog(0, ("hpr_Initialize: ubik client init failed. [%d]", code));
329 code = rxs_Release(sc);
334 hpr_End(struct ubik_client *uclient)
339 code = ubik_ClientDestroy(uclient);
345 getThreadClient(struct ubik_client **client)
349 *client = pthread_getspecific(viced_uclient_key);
353 code = hpr_Initialize(client);
357 opr_Verify(pthread_setspecific(viced_uclient_key, *client) == 0);
363 hpr_GetHostCPS(afs_int32 host, prlist *CPS)
367 struct ubik_client *uclient;
369 code = getThreadClient(&uclient);
374 code = ubik_PR_GetHostCPS(uclient, 0, host, CPS, &over);
375 if (code != PRSUCCESS)
378 /* do something about this, probably make a new call */
379 /* don't forget there's a hard limit in the interface */
381 "membership list for host id %d exceeds display limit\n",
388 hpr_NameToId(namelist *names, idlist *ids)
392 struct ubik_client *uclient;
394 code = getThreadClient(&uclient);
398 for (i = 0; i < names->namelist_len; i++)
399 stolower(names->namelist_val[i]);
400 code = ubik_PR_NameToID(uclient, 0, names, ids);
405 hpr_IdToName(idlist *ids, namelist *names)
408 struct ubik_client *uclient;
410 code = getThreadClient(&uclient);
414 code = ubik_PR_IDToName(uclient, 0, ids, names);
419 hpr_GetCPS(afs_int32 id, prlist *CPS)
423 struct ubik_client *uclient;
425 code = getThreadClient(&uclient);
430 code = ubik_PR_GetCPS(uclient, 0, id, CPS, &over);
431 if (code != PRSUCCESS)
434 /* do something about this, probably make a new call */
435 /* don't forget there's a hard limit in the interface */
436 fprintf(stderr, "membership list for id %d exceeds display limit\n",
442 static short consolePort = 0;
445 h_Lock_r(struct host *host)
455 * returns 1 if already locked
456 * else returns locks and returns 0
460 h_NBLock_r(struct host *host)
462 struct Lock *hostLock = &host->lock;
467 if (!(hostLock->excl_locked) && !(hostLock->readers_reading))
468 hostLock->excl_locked = WRITE_LOCK;
472 LOCK_UNLOCK(hostLock);
481 /*------------------------------------------------------------------------
482 * PRIVATE h_AddrInSameNetwork
485 * Given a target IP address and a candidate IP address (both
486 * in host byte order), return a non-zero value (1) if the
487 * candidate address is in a different network from the target
491 * a_targetAddr : Target address.
492 * a_candAddr : Candidate address.
495 * 1 if the candidate address is in the same net as the target,
499 * The target and candidate addresses are both in host byte
500 * order, NOT network byte order, when passed in. We return
501 * our value as a character, since that's the type of field in
502 * the host structure, where this info will be stored.
506 *------------------------------------------------------------------------*/
509 h_AddrInSameNetwork(afs_uint32 a_targetAddr, afs_uint32 a_candAddr)
510 { /*h_AddrInSameNetwork */
512 afs_uint32 targetNet;
516 * Pull out the network and subnetwork numbers from the target
517 * and candidate addresses. We can short-circuit this whole
518 * affair if the target and candidate addresses are not of the
521 if (IN_CLASSA(a_targetAddr)) {
522 if (!(IN_CLASSA(a_candAddr))) {
525 targetNet = a_targetAddr & IN_CLASSA_NET;
526 candNet = a_candAddr & IN_CLASSA_NET;
527 } else if (IN_CLASSB(a_targetAddr)) {
528 if (!(IN_CLASSB(a_candAddr))) {
531 targetNet = a_targetAddr & IN_CLASSB_NET;
532 candNet = a_candAddr & IN_CLASSB_NET;
533 } /*Class B target */
534 else if (IN_CLASSC(a_targetAddr)) {
535 if (!(IN_CLASSC(a_candAddr))) {
538 targetNet = a_targetAddr & IN_CLASSC_NET;
539 candNet = a_candAddr & IN_CLASSC_NET;
540 } /*Class C target */
542 targetNet = a_targetAddr;
543 candNet = a_candAddr;
544 } /*Class D address */
547 * Now, simply compare the extracted net values for the two addresses
548 * (which at this point are known to be of the same class)
550 if (targetNet == candNet)
555 } /*h_AddrInSameNetwork */
558 /* Assumptions: called with held host */
560 h_gethostcps_r(struct host *host, afs_int32 now)
565 /* wait if somebody else is already doing the getCPS call */
566 while (host->hostFlags & HCPS_INPROGRESS) {
567 slept = 1; /* I did sleep */
568 host->hostFlags |= HCPS_WAITING; /* I am sleeping now */
569 CV_WAIT(&host->cond, &host_glock_mutex);
573 host->hostFlags |= HCPS_INPROGRESS; /* mark as CPSCall in progress */
574 if (host->hcps.prlist_val)
575 free(host->hcps.prlist_val); /* this is for hostaclRefresh */
576 host->hcps.prlist_val = NULL;
577 host->hcps.prlist_len = 0;
578 host->cpsCall = slept ? time(NULL) : (now);
581 code = hpr_GetHostCPS(ntohl(host->host), &host->hcps);
586 * Although ubik_Call (called by pr_GetHostCPS) traverses thru all protection servers
587 * and reevaluates things if no sync server or quorum is found we could still end up
588 * with one of these errors. In such case we would like to reevaluate the rpc call to
589 * find if there's cps for this guy. We treat other errors (except network failures
590 * ones - i.e. code < 0) as an indication that there is no CPS for this host. Ideally
591 * we could like to deal this problem the other way around (i.e. if code == NOCPS
592 * ignore else retry next time) but the problem is that there're other errors (i.e.
593 * EPERM) for which we don't want to retry and we don't know the whole code list!
595 if (code < 0 || code == UNOQUORUM || code == UNOTSYNC) {
597 * We would have preferred to use a while loop and try again since ops in protected
598 * acls for this host will fail now but they'll be reevaluated on any subsequent
599 * call. The attempt to wait for a quorum/sync site or network error won't work
600 * since this problems really should only occurs during a complete fileserver
601 * restart. Since the fileserver will start before the ptservers (and thus before
602 * quorums are complete) clients will be utilizing all the fileserver's lwps!!
604 host->hcpsfailed = 1;
606 ("Warning: GetHostCPS failed (%d) for %p (%s:%d); will retry\n",
607 code, host, afs_inet_ntoa_r(host->host, hoststr), ntohs(host->port)));
609 host->hcpsfailed = 0;
611 ("gethost: GetHostCPS failed (%d) for %p (%s:%d); ignored\n",
612 code, host, afs_inet_ntoa_r(host->host, hoststr), ntohs(host->port)));
614 if (host->hcps.prlist_val)
615 free(host->hcps.prlist_val);
616 host->hcps.prlist_val = NULL;
617 host->hcps.prlist_len = 0; /* Make sure it's zero */
619 host->hcpsfailed = 0;
621 host->hostFlags &= ~HCPS_INPROGRESS;
622 /* signal all who are waiting */
623 if (host->hostFlags & HCPS_WAITING) { /* somebody is waiting */
624 host->hostFlags &= ~HCPS_WAITING;
625 CV_BROADCAST(&host->cond);
629 /* args in net byte order */
631 h_flushhostcps(afs_uint32 hostaddr, afs_uint16 hport)
636 h_Lookup_r(hostaddr, hport, &host);
638 host->hcpsfailed = 1;
647 * Allocate a host. It will be identified by the peer (ip,port) info in the
648 * rx connection provided. The host is returned held and locked
650 #define DEF_ROPCONS 2115
653 h_Alloc_r(struct rx_connection *r_con)
655 struct servent *serverentry;
657 afs_uint32 newHostAddr_HBO; /*New host IP addr, in host byte order */
664 /* acquire the host lock withot dropping H_LOCK. we can do this here
665 * because we know we will not block; we just created this host and
666 * nobody else knows about it. */
667 ObtainWriteLock(&host->lock);
669 host->host = rxr_HostOf(r_con);
670 host->port = rxr_PortOf(r_con);
672 h_AddHostToAddrHashTable_r(host->host, host->port, host);
674 if (consolePort == 0) { /* find the portal number for console */
675 #if defined(AFS_OSF_ENV)
676 serverentry = getservbyname("ropcons", "");
678 serverentry = getservbyname("ropcons", 0);
681 consolePort = serverentry->s_port;
683 consolePort = htons(DEF_ROPCONS); /* Use a default */
685 if (host->port == consolePort)
687 /* Make a callback channel even for the console, on the off chance that it
688 * makes a request that causes a break call back. It shouldn't. */
689 h_SetupCallbackConn_r(host);
690 host->LastCall = host->cpsCall = host->ActiveCall = time(NULL);
692 host->hcps.prlist_val = NULL;
693 host->hcps.prlist_len = 0;
694 host->interface = NULL;
696 host->hcpsfailed = 0; /* save cycles */
697 h_gethostcps(host); /* do this under host hold/lock */
699 host->FirstClient = NULL;
700 h_InsertList_r(host); /* update global host List */
702 * Compare the new host's IP address (in host byte order) with ours
703 * (the File Server's), remembering if they are in the same network.
705 newHostAddr_HBO = (afs_uint32) ntohl(host->host);
706 host->InSameNetwork =
707 h_AddrInSameNetwork(FS_HostAddr_HBO, newHostAddr_HBO);
714 /* Make a callback channel even for the console, on the off chance that it
715 * makes a request that causes a break call back. It shouldn't. */
717 h_SetupCallbackConn_r(struct host * host)
720 sc = rxnull_NewClientSecurityObject();
721 host->callback_rxcon =
722 rx_NewConnection(host->host, host->port, 1, sc, 0);
723 rx_SetConnDeadTime(host->callback_rxcon, 50);
724 rx_SetConnHardDeadTime(host->callback_rxcon, AFS_HARDDEADTIME);
725 rx_SetConnSecondsUntilNatPing(host->callback_rxcon, 20);
729 * Lookup a host given an IP address and UDP port number.
730 * hostaddr and hport are in network order
731 * hostaddr and hport are in network order
732 * On return, refCount is incremented.
735 h_Lookup_r(afs_uint32 haddr, afs_uint16 hport, struct host **hostp)
738 struct host *host = NULL;
739 struct h_AddrHashChain *chain;
740 int index = h_HashIndex(haddr);
741 extern int hostaclRefresh;
744 for (chain = hostAddrHashTable[index]; chain; chain = chain->next) {
745 host = chain->hostPtr;
747 if (!(host->hostFlags & HOSTDELETED) && chain->addr == haddr
748 && chain->port == hport) {
749 if ((host->hostFlags & HWHO_INPROGRESS) &&
750 h_threadquota(host->lock.num_waiting)) {
756 if (host->hostFlags & HOSTDELETED) {
763 now = time(NULL); /* always evaluate "now" */
764 if (host->hcpsfailed || (host->cpsCall + hostaclRefresh < now)) {
766 * Every hostaclRefresh period (def 2 hrs) get the new
767 * membership list for the host. Note this could be the
768 * first time that the host is added to a group. Also
769 * here we also retry on previous legitimate hcps failures.
771 * If we get here refCount is elevated.
773 h_gethostcps_r(host, now);
783 /* Lookup a host given its UUID. */
785 h_LookupUuid_r(afsUUID * uuidp)
787 struct host *host = 0;
788 struct h_UuidHashChain *chain;
789 int index = h_UuidHashIndex(uuidp);
791 for (chain = hostUuidHashTable[index]; chain; chain = chain->next) {
792 host = chain->hostPtr;
794 if (!(host->hostFlags & HOSTDELETED) && host->interface
795 && afs_uuid_equal(&host->interface->uuid, uuidp)) {
803 /* h_TossStuff_r: Toss anything in the host structure (the host or
804 * clients marked for deletion. Called from h_Release_r ONLY.
805 * To be called, there must be no holds, and either host->deleted
806 * or host->clientDeleted must be set.
809 h_TossStuff_r(struct host *host)
811 struct client **cp, *client;
815 if ((host->hostFlags & HOSTDELETED)) {
819 /* make sure host doesn't go away over h_NBLock_r */
822 code = h_NBLock_r(host);
824 /* don't use h_Release_r, since that may call h_TossStuff_r again */
827 /* if somebody still has this host locked */
831 /* someone locked the host while HOSTDELETED was set; that is bad */
832 ViceLog(0, ("Warning: h_TossStuff_r failed; Host %" AFS_PTR_FMT
833 " (%s:%d flags 0x%x) was locked.\n",
834 host, afs_inet_ntoa_r(host->host, hoststr), ntohs(host->port),
835 (unsigned)host->hostFlags));
842 /* if somebody still has this host held */
843 /* we must check this _after_ h_NBLock_r, since h_NBLock_r can drop and
844 * reacquire H_LOCK */
845 if (host->refCount > 0) {
848 /* someone grabbed a ref while HOSTDELETED was set; that is bad */
849 ViceLog(0, ("Warning: h_TossStuff_r failed; Host %" AFS_PTR_FMT
850 " (%s:%d flags 0x%x) was held.\n",
851 host, afs_inet_ntoa_r(host->host, hoststr), ntohs(host->port),
852 (unsigned)host->hostFlags));
857 /* ASSUMPTION: rxi_FreeConnection() does not yield */
858 for (cp = &host->FirstClient; (client = *cp);) {
859 if ((host->hostFlags & HOSTDELETED) || client->deleted) {
861 ObtainWriteLockNoBlock(&client->lock, code);
865 ("Warning: h_TossStuff_r failed: Host %p (%s:%d) "
866 "client %p was locked.\n",
867 host, afs_inet_ntoa_r(host->host, hoststr),
868 ntohs(host->port), client));
872 if (client->refCount) {
875 ("Warning: h_TossStuff_r failed: Host %p (%s:%d) "
876 "client %p refcount %d.\n",
877 host, afs_inet_ntoa_r(host->host, hoststr),
878 ntohs(host->port), client, client->refCount));
879 /* This is the same thing we do if the host is locked */
880 ReleaseWriteLock(&client->lock);
883 client->CPS.prlist_len = 0;
884 if ((client->ViceId != ANONYMOUSID) && client->CPS.prlist_val)
885 free(client->CPS.prlist_val);
886 client->CPS.prlist_val = NULL;
887 CurrentConnections--;
889 ReleaseWriteLock(&client->lock);
895 /* We've just cleaned out all the deleted clients; clear the flag */
896 host->hostFlags &= ~CLIENTDELETED;
898 if (host->hostFlags & HOSTDELETED) {
899 struct rx_connection *rxconn;
900 struct AddrPort hostAddrPort;
903 if (host->Console & 1)
905 if ((rxconn = host->callback_rxcon)) {
906 host->callback_rxcon = (struct rx_connection *)0;
907 rx_DestroyConnection(rxconn);
909 if (host->hcps.prlist_val)
910 free(host->hcps.prlist_val);
911 host->hcps.prlist_val = NULL;
912 host->hcps.prlist_len = 0;
913 DeleteAllCallBacks_r(host, 1);
914 host->hostFlags &= ~RESETDONE; /* just to be safe */
916 /* if alternate addresses do not exist */
917 if (!(host->interface)) {
918 h_DeleteHostFromAddrHashTable_r(host->host, host->port, host);
920 h_DeleteHostFromUuidHashTable_r(host);
921 h_DeleteHostFromAddrHashTable_r(host->host, host->port, host);
922 /* delete the hash entry for each valid alternate addresses */
923 for (i = 0; i < host->interface->numberOfInterfaces; i++) {
924 hostAddrPort = host->interface->interface[i];
926 * if the interface addr/port is the primary, we already
927 * removed it. If the addr/port is not valid, its not
930 if (hostAddrPort.valid &&
931 (host->host != hostAddrPort.addr ||
932 host->port != hostAddrPort.port))
933 h_DeleteHostFromAddrHashTable_r(hostAddrPort.addr, hostAddrPort.port, host);
935 free(host->interface);
936 host->interface = NULL;
937 } /* if alternate address exists */
939 h_DeleteList_r(host); /* remove host from global host List */
946 /* h_Enumerate: Calls (*proc)(host, param) for at least each host in the
947 * system at the start of the enumeration (perhaps more). Hosts may be deleted
948 * (have delete flag set); ditto for clients. refCount is always incremented
949 * before (*proc) is called.
951 * The return value of the proc is a set of flags. The proc should set
952 * H_ENUMERATE_BAIL(foo) if the enumeration of hosts should be stopped early.
955 h_Enumerate(int (*proc) (struct host*, void *), void *param)
957 struct host *host, **list;
962 if (hostCount == 0) {
966 list = malloc(hostCount * sizeof(struct host *));
968 ViceLogThenPanic(0, ("Failed malloc in h_Enumerate (list)\n"));
970 for (totalCount = count = 0, host = hostList;
971 host && totalCount < hostCount;
972 host = host->next, totalCount++) {
974 if (!(host->hostFlags & HOSTDELETED)) {
980 if (totalCount != hostCount) {
981 ViceLog(0, ("h_Enumerate found %d of %d hosts\n", totalCount, hostCount));
982 } else if (host != NULL) {
983 ViceLog(0, ("h_Enumerate found more than %d hosts\n", hostCount));
984 ShutDownAndCore(PANIC);
987 for (i = 0; i < count; i++) {
989 flags = (*proc) (list[i], param);
991 h_Release_r(list[i]);
993 /* bail out of the enumeration early */
994 if (H_ENUMERATE_ISSET_BAIL(flags)) {
997 ViceLog(0, ("h_Enumerate got back invalid return value %d\n", flags));
998 ShutDownAndCore(PANIC);
1002 /* we bailed out of enumerating hosts early; we still have holds on
1003 * some of the hosts in 'list', so release them */
1006 for ( ; i < count; i++) {
1007 h_Release_r(list[i]);
1015 /* h_Enumerate_r (revised):
1016 * Calls (*proc)(host, param) for each host in hostList, starting
1017 * at enumstart. Called only under H_LOCK. Hosts may be deleted (have
1018 * delete flag set); ditto for clients. refCount is always incremented
1019 * before (*proc) is called.
1021 * @note Assumes that hostList is only prepended to, that a host is never
1022 * inserted into the middle. Otherwise this would not be guaranteed to
1025 * The return value of the proc is a set of flags. The proc should set
1026 * H_ENUMERATE_BAIL(foo) if the enumeration of hosts should be stopped early.
1029 h_Enumerate_r(int (*proc) (struct host *, void *),
1030 struct host *enumstart, void *param)
1032 struct host *host, *next;
1036 if (hostCount == 0) {
1043 /* find the first non-deleted host, so we know where to actually start
1045 for (count = 0; host && count < hostCount; count++) {
1046 if (!(host->hostFlags & HOSTDELETED)) {
1053 /* we didn't find a non-deleted host... */
1055 if (host && count >= hostCount) {
1056 /* ...because we found a loop */
1057 ViceLog(0, ("h_Enumerate_r found more than %d hosts\n", hostCount));
1058 ShutDownAndCore(PANIC);
1061 /* ...because the hostList is full of deleted hosts */
1065 h_Hold_r(enumstart);
1067 /* remember hostCount, lest it change over the potential H_LOCK drop in
1069 origHostCount = hostCount;
1071 for (count = 0, host = enumstart; host && count < origHostCount; host = next, count++) {
1074 /* find the next non-deleted host */
1075 while (next && (next->hostFlags & HOSTDELETED)) {
1077 /* inc count for the skipped-over host */
1078 if (++count > origHostCount) {
1079 ViceLog(0, ("h_Enumerate_r found more than %d hosts\n", origHostCount));
1080 ShutDownAndCore(PANIC);
1086 if (!(host->hostFlags & HOSTDELETED)) {
1088 flags = (*proc) (host, param);
1089 if (H_ENUMERATE_ISSET_BAIL(flags)) {
1090 h_Release_r(host); /* this might free up the host */
1096 ViceLog(0, ("h_Enumerate_r got back invalid return value %d\n", flags));
1097 ShutDownAndCore(PANIC);
1100 h_Release_r(host); /* this might free up the host */
1102 if (host != NULL && count >= origHostCount) {
1103 ViceLog(0, ("h_Enumerate_r found more than %d hosts\n", origHostCount));
1104 ShutDownAndCore(PANIC);
1106 } /*h_Enumerate_r */
1109 /* inserts a new HashChain structure corresponding to this UUID */
1111 h_AddHostToUuidHashTable_r(struct afsUUID *uuid, struct host *host)
1114 struct h_UuidHashChain *chain;
1115 char uuid1[128], uuid2[128];
1118 /* hash into proper bucket */
1119 index = h_UuidHashIndex(uuid);
1121 /* don't add the same entry multiple times */
1122 for (chain = hostUuidHashTable[index]; chain; chain = chain->next) {
1123 if (!chain->hostPtr)
1126 if (chain->hostPtr->interface &&
1127 afs_uuid_equal(&chain->hostPtr->interface->uuid, uuid)) {
1128 if (LogLevel >= 125) {
1129 afsUUID_to_string(&chain->hostPtr->interface->uuid, uuid1,
1131 afsUUID_to_string(uuid, uuid2, 127);
1132 ViceLog(125, ("h_AddHostToUuidHashTable_r: host %" AFS_PTR_FMT " (uuid %s) exists as %s:%d (uuid %s)\n",
1134 afs_inet_ntoa_r(chain->hostPtr->host, hoststr),
1135 ntohs(chain->hostPtr->port), uuid2));
1141 /* insert into beginning of list for this bucket */
1142 chain = malloc(sizeof(struct h_UuidHashChain));
1144 ViceLogThenPanic(0, ("Failed malloc in h_AddHostToUuidHashTable_r\n"));
1146 chain->hostPtr = host;
1147 chain->next = hostUuidHashTable[index];
1148 hostUuidHashTable[index] = chain;
1151 afsUUID_to_string(uuid, uuid2, 127);
1153 ("h_AddHostToUuidHashTable_r: host %p (%s:%d) added as uuid %s\n",
1154 host, afs_inet_ntoa_r(chain->hostPtr->host, hoststr),
1155 ntohs(chain->hostPtr->port), uuid2));
1158 /* deletes a HashChain structure corresponding to this host */
1160 h_DeleteHostFromUuidHashTable_r(struct host *host)
1163 struct h_UuidHashChain **uhp, *uth;
1167 if (!host->interface)
1170 /* hash into proper bucket */
1171 index = h_UuidHashIndex(&host->interface->uuid);
1173 if (LogLevel >= 125)
1174 afsUUID_to_string(&host->interface->uuid, uuid1, 127);
1175 for (uhp = &hostUuidHashTable[index]; (uth = *uhp); uhp = &uth->next) {
1176 opr_Assert(uth->hostPtr);
1177 if (uth->hostPtr == host) {
1179 ("h_DeleteHostFromUuidHashTable_r: host %" AFS_PTR_FMT " (uuid %s %s:%d)\n",
1180 host, uuid1, afs_inet_ntoa_r(host->host, hoststr),
1181 ntohs(host->port)));
1188 ("h_DeleteHostFromUuidHashTable_r: host %" AFS_PTR_FMT " (uuid %s %s:%d) not found\n",
1189 host, uuid1, afs_inet_ntoa_r(host->host, hoststr),
1190 ntohs(host->port)));
1195 * This is called with host locked and held.
1197 * All addresses are in network byte order.
1200 invalidateInterfaceAddr_r(struct host *host, afs_uint32 addr, afs_uint16 port)
1204 struct Interface *interface;
1205 char hoststr[16], hoststr2[16];
1208 opr_Assert(host->interface);
1210 ViceLog(125, ("invalidateInterfaceAddr : host %" AFS_PTR_FMT " (%s:%d) addr %s:%d\n",
1211 host, afs_inet_ntoa_r(host->host, hoststr),
1212 ntohs(host->port), afs_inet_ntoa_r(addr, hoststr2),
1216 * Make sure this address is on the list of known addresses
1219 interface = host->interface;
1220 number = host->interface->numberOfInterfaces;
1221 for (i = 0; i < number; i++) {
1222 if (interface->interface[i].addr == addr &&
1223 interface->interface[i].port == port) {
1224 if (interface->interface[i].valid) {
1225 h_DeleteHostFromAddrHashTable_r(addr, port, host);
1226 interface->interface[i].valid = 0;
1237 * This is called with host locked and held. This function differs
1238 * from removeInterfaceAddr_r in that it is called when the address
1239 * is being removed from the host regardless of whether or not there
1240 * is an interface list for the host. This function will delete the
1241 * host if there are no addresses left on it.
1243 * All addresses are in network byte order.
1246 removeAddress_r(struct host *host, afs_uint32 addr, afs_uint16 port)
1249 char hoststr[16], hoststr2[16];
1250 struct rx_connection *rxconn;
1252 if (!host->interface || host->interface->numberOfInterfaces == 1) {
1253 if (host->host == addr && host->port == port) {
1255 ("Removing only address for host %" AFS_PTR_FMT " (%s:%d), deleting host.\n",
1256 host, afs_inet_ntoa_r(host->host, hoststr), ntohs(host->port)));
1257 host->hostFlags |= HOSTDELETED;
1259 * Do not remove the primary addr/port from the hash table.
1260 * It will be ignored due to the HOSTDELETED flag and will
1261 * be removed when h_TossStuff_r() cleans up the HOSTDELETED
1262 * host. Removing it here will only result in a search for
1263 * the host/addr/port in the hash chain which will fail.
1267 ("Removing address that does not belong to host %" AFS_PTR_FMT " (%s:%d).\n",
1268 host, afs_inet_ntoa_r(host->host, hoststr), ntohs(host->port)));
1271 if (host->host == addr && host->port == port) {
1272 removeInterfaceAddr_r(host, addr, port);
1274 for (i=0; i < host->interface->numberOfInterfaces; i++) {
1275 if (host->interface->interface[i].valid) {
1277 ("Removed address for host %" AFS_PTR_FMT " (%s:%d), new primary interface %s:%d.\n",
1278 host, afs_inet_ntoa_r(host->host, hoststr), ntohs(host->port),
1279 afs_inet_ntoa_r(host->interface->interface[i].addr, hoststr2),
1280 ntohs(host->interface->interface[i].port)));
1281 host->host = host->interface->interface[i].addr;
1282 host->port = host->interface->interface[i].port;
1283 h_AddHostToAddrHashTable_r(host->host, host->port, host);
1288 if (i == host->interface->numberOfInterfaces) {
1290 ("Removed only address for host %" AFS_PTR_FMT " (%s:%d), no valid alternate interfaces, deleting host.\n",
1291 host, afs_inet_ntoa_r(host->host, hoststr), ntohs(host->port)));
1292 host->hostFlags |= HOSTDELETED;
1293 /* addr/port was removed from the hash table */
1297 rxconn = host->callback_rxcon;
1298 host->callback_rxcon = NULL;
1301 rx_DestroyConnection(rxconn);
1305 h_SetupCallbackConn_r(host);
1308 /* not the primary addr/port, just invalidate it */
1309 invalidateInterfaceAddr_r(host, addr, port);
1317 createHostAddrHashChain_r(int index, afs_uint32 addr, afs_uint16 port, struct host *host)
1319 struct h_AddrHashChain *chain;
1322 /* insert into beginning of list for this bucket */
1323 chain = malloc(sizeof(struct h_AddrHashChain));
1325 ViceLogThenPanic(0, ("Failed malloc in h_AddHostToAddrHashTable_r\n"));
1327 chain->hostPtr = host;
1328 chain->next = hostAddrHashTable[index];
1331 hostAddrHashTable[index] = chain;
1332 ViceLog(125, ("h_AddHostToAddrHashTable_r: host %" AFS_PTR_FMT " added as %s:%d\n",
1333 host, afs_inet_ntoa_r(addr, hoststr), ntohs(port)));
1337 * Resolve host address conflicts when hashing by address.
1339 * @param[in] addr an ip address of the interface
1340 * @param[in] port the port of the interface
1341 * @param[in] newHost the host being added with this address
1342 * @param[in] oldHost the host previously added with this address
1345 reconcileHosts_r(afs_uint32 addr, afs_uint16 port, struct host *newHost,
1346 struct host *oldHost)
1348 struct rx_connection *cb = NULL;
1350 struct interfaceAddr interf;
1352 afsUUID *newHostUuid = &nulluuid;
1353 afsUUID *oldHostUuid = &nulluuid;
1357 ("reconcileHosts_r: addr %s:%d newHost %" AFS_PTR_FMT " oldHost %"
1358 AFS_PTR_FMT, afs_inet_ntoa_r(addr, hoststr), ntohs(port),
1361 opr_Assert(oldHost != newHost);
1362 caps.Capabilities_val = NULL;
1365 sc = rxnull_NewClientSecurityObject();
1368 cb = rx_NewConnection(addr, port, 1, sc, 0);
1369 rx_SetConnDeadTime(cb, 50);
1370 rx_SetConnHardDeadTime(cb, AFS_HARDDEADTIME);
1375 code = RXAFSCB_TellMeAboutYourself(cb, &interf, &caps);
1376 if (code == RXGEN_OPCODE) {
1377 code = RXAFSCB_WhoAreYou(cb, &interf);
1381 if (code == RXGEN_OPCODE ||
1382 (code == 0 && afs_uuid_equal(&interf.uuid, &nulluuid))) {
1384 ("reconcileHosts_r: WhoAreYou not supported for connection (%s:%d), error %d\n",
1385 afs_inet_ntoa_r(addr, hoststr), ntohs(port), code));
1390 ("reconcileHosts_r: WhoAreYou failed for connection (%s:%d), error %d\n",
1391 afs_inet_ntoa_r(addr, hoststr), ntohs(port), code));
1395 /* Since lock was dropped, the hosts may have been deleted during the rpcs. */
1396 if ((newHost->hostFlags & HOSTDELETED)
1397 && (oldHost->hostFlags & HOSTDELETED)) {
1399 ("reconcileHosts_r: new and old hosts were deleted during probe.\n"));
1403 /* A check can be done if at least one of the hosts has a uuid. It
1404 * is an error if the hosts have the same (not null) uuid. */
1405 if ((!(newHost->hostFlags & HOSTDELETED)) && newHost->interface) {
1406 newHostUuid = &(newHost->interface->uuid);
1408 if ((!(oldHost->hostFlags & HOSTDELETED)) && oldHost->interface) {
1409 oldHostUuid = &(oldHost->interface->uuid);
1411 if (afs_uuid_equal(newHostUuid, &nulluuid) &&
1412 afs_uuid_equal(oldHostUuid, &nulluuid)) {
1414 ("reconcileHosts_r: Cannot reconcile hosts for connection (%s:%d), no uuids\n",
1415 afs_inet_ntoa_r(addr, hoststr), ntohs(port)));
1418 if (afs_uuid_equal(newHostUuid, oldHostUuid)) {
1420 ("reconcileHosts_r: Cannot reconcile hosts for connection (%s:%d), same uuids\n",
1421 afs_inet_ntoa_r(addr, hoststr), ntohs(port)));
1425 /* Determine which host should be hashed */
1426 if ((!(newHost->hostFlags & HOSTDELETED))
1427 && afs_uuid_equal(newHostUuid, &(interf.uuid))) {
1428 /* Install the new host into the hash before removing the stale
1429 * addresses. Walk the hash chain again since the hash table may have
1430 * been changed when the host lock was dropped to get the uuid. */
1431 struct h_AddrHashChain *chain;
1432 int index = h_HashIndex(addr);
1433 for (chain = hostAddrHashTable[index]; chain; chain = chain->next) {
1434 if (chain->addr == addr && chain->port == port) {
1435 chain->hostPtr = newHost;
1436 removeAddress_r(oldHost, addr, port);
1440 createHostAddrHashChain_r(index, addr, port, newHost);
1441 removeAddress_r(oldHost, addr, port);
1444 if ((!(oldHost->hostFlags & HOSTDELETED))
1445 && afs_uuid_equal(oldHostUuid, &(interf.uuid))) {
1446 removeAddress_r(newHost, addr, port);
1451 if (!(newHost->hostFlags & HOSTDELETED)) {
1452 removeAddress_r(newHost, addr, port);
1454 if (!(oldHost->hostFlags & HOSTDELETED)) {
1455 removeAddress_r(oldHost, addr, port);
1459 h_Release_r(newHost);
1460 h_Release_r(oldHost);
1461 rx_DestroyConnection(cb);
1465 /* inserts a new HashChain structure corresponding to this address */
1467 h_AddHostToAddrHashTable_r(afs_uint32 addr, afs_uint16 port, struct host *host)
1470 struct h_AddrHashChain *chain;
1473 /* hash into proper bucket */
1474 index = h_HashIndex(addr);
1476 /* don't add the same address:port pair entry multiple times */
1477 for (chain = hostAddrHashTable[index]; chain; chain = chain->next) {
1478 if (chain->addr == addr && chain->port == port) {
1479 if (chain->hostPtr == host) {
1481 ("h_AddHostToAddrHashTable_r: host %" AFS_PTR_FMT " (%s:%d) already hashed\n",
1482 host, afs_inet_ntoa_r(chain->addr, hoststr),
1483 ntohs(chain->port)));
1486 if (!(chain->hostPtr->hostFlags & HOSTDELETED)) {
1487 /* attempt to resolve host address collision */
1488 reconcileHosts_r(addr, port, host, chain->hostPtr);
1493 createHostAddrHashChain_r(index, addr, port, host);
1497 * This is called with host locked and held.
1498 * It is called to either validate or add an additional interface
1499 * address/port on the specified host.
1501 * All addresses are in network byte order.
1504 addInterfaceAddr_r(struct host *host, afs_uint32 addr, afs_uint16 port)
1508 struct Interface *interface;
1509 char hoststr[16], hoststr2[16];
1512 opr_Assert(host->interface);
1515 * Make sure this address is on the list of known addresses
1518 number = host->interface->numberOfInterfaces;
1519 for (i = 0; i < number; i++) {
1520 if (host->interface->interface[i].addr == addr &&
1521 host->interface->interface[i].port == port) {
1523 ("addInterfaceAddr : found host %" AFS_PTR_FMT " (%s:%d) adding %s:%d%s\n",
1524 host, afs_inet_ntoa_r(host->host, hoststr),
1525 ntohs(host->port), afs_inet_ntoa_r(addr, hoststr2),
1526 ntohs(port), host->interface->interface[i].valid ? "" :
1529 if (host->interface->interface[i].valid == 0) {
1530 host->interface->interface[i].valid = 1;
1531 h_AddHostToAddrHashTable_r(addr, port, host);
1537 ViceLog(125, ("addInterfaceAddr : host %" AFS_PTR_FMT " (%s:%d) adding %s:%d\n",
1538 host, afs_inet_ntoa_r(host->host, hoststr),
1539 ntohs(host->port), afs_inet_ntoa_r(addr, hoststr2),
1542 interface = malloc(sizeof(struct Interface)
1543 + (sizeof(struct AddrPort) * number));
1545 ViceLogThenPanic(0, ("Failed malloc in addInterfaceAddr_r\n"));
1547 interface->numberOfInterfaces = number + 1;
1548 interface->uuid = host->interface->uuid;
1549 for (i = 0; i < number; i++)
1550 interface->interface[i] = host->interface->interface[i];
1552 /* Add the new valid interface */
1553 interface->interface[number].addr = addr;
1554 interface->interface[number].port = port;
1555 interface->interface[number].valid = 1;
1556 h_AddHostToAddrHashTable_r(addr, port, host);
1557 free(host->interface);
1558 host->interface = interface;
1565 * This is called with host locked and held.
1567 * All addresses are in network byte order.
1570 removeInterfaceAddr_r(struct host *host, afs_uint32 addr, afs_uint16 port)
1574 struct Interface *interface;
1575 char hoststr[16], hoststr2[16];
1578 opr_Assert(host->interface);
1580 ViceLog(125, ("removeInterfaceAddr : host %" AFS_PTR_FMT " (%s:%d) addr %s:%d\n",
1581 host, afs_inet_ntoa_r(host->host, hoststr),
1582 ntohs(host->port), afs_inet_ntoa_r(addr, hoststr2),
1586 * Make sure this address is on the list of known addresses
1589 interface = host->interface;
1590 number = host->interface->numberOfInterfaces;
1591 for (i = 0; i < number; i++) {
1592 if (interface->interface[i].addr == addr &&
1593 interface->interface[i].port == port) {
1594 if (interface->interface[i].valid)
1595 h_DeleteHostFromAddrHashTable_r(addr, port, host);
1597 for (; i < number; i++) {
1598 interface->interface[i] = interface->interface[i+1];
1600 interface->numberOfInterfaces = number;
1611 h_threadquota(int waiting)
1616 } else if (lwps > 32) {
1619 } else if (lwps > 16) {
1629 /* If found, host is returned with refCount incremented */
1631 h_GetHost_r(struct rx_connection *tcon)
1634 struct host *oldHost;
1636 struct interfaceAddr interf;
1637 int interfValid = 0;
1638 struct Identity *identP = NULL;
1641 char hoststr[16], hoststr2[16];
1643 struct rx_connection *cb_conn = NULL;
1644 struct rx_connection *cb_in = NULL;
1646 caps.Capabilities_val = NULL;
1648 haddr = rxr_HostOf(tcon);
1649 hport = rxr_PortOf(tcon);
1652 rx_DestroyConnection(cb_in);
1655 if (caps.Capabilities_val)
1656 free(caps.Capabilities_val);
1657 caps.Capabilities_val = NULL;
1658 caps.Capabilities_len = 0;
1661 if (h_Lookup_r(haddr, hport, &host))
1663 identP = (struct Identity *)rx_GetSpecific(tcon, rxcon_ident_key);
1664 if (host && !identP && !(host->Console & 1)) {
1665 /* This is a new connection, and we already have a host
1666 * structure for this address. Verify that the identity
1667 * of the caller matches the identity in the host structure.
1669 if ((host->hostFlags & HWHO_INPROGRESS) &&
1670 h_threadquota(host->lock.num_waiting)) {
1676 if (!(host->hostFlags & ALTADDR) ||
1677 (host->hostFlags & HOSTDELETED)) {
1678 /* Another thread is doing initialization
1679 * or this host was deleted while we
1680 * waited for the lock. */
1683 ("Host %" AFS_PTR_FMT " (%s:%d) starting h_Lookup again\n",
1684 host, afs_inet_ntoa_r(host->host, hoststr),
1685 ntohs(host->port)));
1689 host->hostFlags |= HWHO_INPROGRESS;
1690 host->hostFlags &= ~ALTADDR;
1692 /* We received a new connection from an IP address/port
1693 * that is associated with 'host' but the address/port of
1694 * the callback connection does not have to match it.
1695 * If there is a match, we can use the existing callback
1696 * connection to verify the UUID. If they do not match
1697 * we need to use a new callback connection to verify the
1698 * UUID of the incoming caller and perhaps use the old
1699 * callback connection to verify that the old address/port
1703 cb_conn = host->callback_rxcon;
1704 rx_GetConnection(cb_conn);
1706 if (haddr == host->host && hport == host->port) {
1707 /* The existing callback connection matches the
1708 * incoming connection so just use it.
1711 RXAFSCB_TellMeAboutYourself(cb_conn, &interf, &caps);
1712 if (code == RXGEN_OPCODE)
1713 code = RXAFSCB_WhoAreYou(cb_conn, &interf);
1715 /* We do not have a match. Create a new connection
1716 * for the new addr/port and use multi_Rx to probe
1717 * both of them simultaneously.
1720 sc = rxnull_NewClientSecurityObject();
1721 cb_in = rx_NewConnection(haddr, hport, 1, sc, 0);
1722 rx_SetConnDeadTime(cb_in, 50);
1723 rx_SetConnHardDeadTime(cb_in, AFS_HARDDEADTIME);
1724 rx_SetConnSecondsUntilNatPing(cb_in, 20);
1727 RXAFSCB_TellMeAboutYourself(cb_in, &interf, &caps);
1728 if (code == RXGEN_OPCODE)
1729 code = RXAFSCB_WhoAreYou(cb_in, &interf);
1731 rx_PutConnection(cb_conn);
1734 if ((code == RXGEN_OPCODE) ||
1735 ((code == 0) && (afs_uuid_equal(&interf.uuid, &nulluuid)))) {
1736 identP = malloc(sizeof(struct Identity));
1738 ViceLogThenPanic(0, ("Failed malloc in h_GetHost_r\n"));
1741 rx_SetSpecific(tcon, rxcon_ident_key, identP);
1742 if (cb_in == NULL) {
1743 /* The host on this connection was unable to respond to
1744 * the WhoAreYou. We will treat this as a new connection
1745 * from the existing host. The worst that can happen is
1746 * that we maintain some extra callback state information */
1747 if (host->interface) {
1749 ("Host %" AFS_PTR_FMT " (%s:%d) used to support WhoAreYou, deleting.\n",
1751 afs_inet_ntoa_r(host->host, hoststr),
1752 ntohs(host->port)));
1753 host->hostFlags |= HOSTDELETED;
1754 host->hostFlags &= ~HWHO_INPROGRESS;
1761 /* The incoming connection does not support WhoAreYou but
1762 * the original one might have. Use removeAddress_r() to
1763 * remove this addr/port from the host that was found.
1764 * If there are no more addresses left for the host it
1765 * will be deleted. Then we retry.
1767 removeAddress_r(host, haddr, hport);
1768 host->hostFlags &= ~HWHO_INPROGRESS;
1769 host->hostFlags |= ALTADDR;
1775 } else if (code == 0) {
1777 identP = malloc(sizeof(struct Identity));
1779 ViceLogThenPanic(0, ("Failed malloc in h_GetHost_r\n"));
1782 identP->uuid = interf.uuid;
1783 rx_SetSpecific(tcon, rxcon_ident_key, identP);
1784 /* Check whether the UUID on this connection matches
1785 * the UUID in the host structure. If they don't match
1786 * then this is not the same host as before. */
1787 if (!host->interface
1788 || !afs_uuid_equal(&interf.uuid, &host->interface->uuid)) {
1791 ("Uuid doesn't match connection (%s:%d).\n",
1792 afs_inet_ntoa_r(haddr, hoststr), ntohs(hport)));
1793 removeAddress_r(host, haddr, hport);
1796 ("Uuid doesn't match host %" AFS_PTR_FMT " (%s:%d).\n",
1797 host, afs_inet_ntoa_r(host->host, hoststr), ntohs(host->port)));
1799 removeAddress_r(host, host->host, host->port);
1801 host->hostFlags &= ~HWHO_INPROGRESS;
1802 host->hostFlags |= ALTADDR;
1808 /* the UUID matched the client at the incoming addr/port
1809 * but this is not the address of the active callback
1810 * connection. Try that connection and see if the client
1811 * is still there and if the reported UUID is the same.
1814 afsUUID uuid = host->interface->uuid;
1815 cb_conn = host->callback_rxcon;
1816 rx_GetConnection(cb_conn);
1817 rx_SetConnDeadTime(cb_conn, 2);
1818 rx_SetConnHardDeadTime(cb_conn, AFS_HARDDEADTIME);
1820 code2 = RXAFSCB_ProbeUuid(cb_conn, &uuid);
1822 rx_SetConnDeadTime(cb_conn, 50);
1823 rx_SetConnHardDeadTime(cb_conn, AFS_HARDDEADTIME);
1824 rx_PutConnection(cb_conn);
1827 /* The primary address is either not responding or
1828 * is not the client we are looking for. Need to
1829 * remove the primary address and add swap in the new
1830 * callback connection, and destroy the old one.
1832 struct rx_connection *rxconn;
1833 ViceLog(0,("CB: ProbeUuid for host %" AFS_PTR_FMT " (%s:%d) failed %d\n",
1835 afs_inet_ntoa_r(host->host, hoststr),
1836 ntohs(host->port),code2));
1839 * make sure we add and then remove. otherwise, we
1840 * might end up with no valid interfaces after the
1841 * remove and the host will have been marked deleted.
1843 addInterfaceAddr_r(host, haddr, hport);
1844 removeInterfaceAddr_r(host, host->host, host->port);
1847 rxconn = host->callback_rxcon;
1848 host->callback_rxcon = cb_in;
1853 * If rx_DestroyConnection calls h_FreeConnection we
1854 * will deadlock on the host_glock_mutex. Work around
1855 * the problem by unhooking the client from the
1856 * connection before destroying the connection.
1858 rx_SetSpecific(rxconn, rxcon_client_key, (void *)0);
1859 rx_DestroyConnection(rxconn);
1865 /* A callback to the incoming connection address is failing.
1866 * Assume that the addr/port is no longer associated with the host
1867 * returned by h_Lookup_r.
1870 ("CB: WhoAreYou failed for connection (%s:%d) , error %d\n",
1871 afs_inet_ntoa_r(haddr, hoststr), ntohs(hport), code));
1872 removeAddress_r(host, haddr, hport);
1873 host->hostFlags &= ~HWHO_INPROGRESS;
1874 host->hostFlags |= ALTADDR;
1878 rx_DestroyConnection(cb_in);
1883 ("CB: WhoAreYou failed for host %" AFS_PTR_FMT " (%s:%d), error %d\n",
1884 host, afs_inet_ntoa_r(host->host, hoststr),
1885 ntohs(host->port), code));
1886 host->hostFlags |= VENUSDOWN;
1889 if (caps.Capabilities_val
1890 && (caps.Capabilities_val[0] & CLIENT_CAPABILITY_ERRORTRANS))
1891 host->hostFlags |= HERRORTRANS;
1893 host->hostFlags &= ~(HERRORTRANS);
1894 host->hostFlags |= ALTADDR;
1895 host->hostFlags &= ~HWHO_INPROGRESS;
1898 if (!(host->hostFlags & ALTADDR)) {
1899 /* another thread is doing the initialisation */
1901 ("Host %" AFS_PTR_FMT " (%s:%d) waiting for host-init to complete\n",
1902 host, afs_inet_ntoa_r(host->host, hoststr),
1903 ntohs(host->port)));
1907 ("Host %" AFS_PTR_FMT " (%s:%d) starting h_Lookup again\n",
1908 host, afs_inet_ntoa_r(host->host, hoststr),
1909 ntohs(host->port)));
1913 /* We need to check whether the identity in the host structure
1914 * matches the identity on the connection. If they don't match
1915 * then treat this a new host. */
1916 if (!(host->Console & 1)
1917 && ((!identP->valid && host->interface)
1918 || (identP->valid && !host->interface)
1920 && !afs_uuid_equal(&identP->uuid,
1921 &host->interface->uuid)))) {
1922 char uuid1[128], uuid2[128];
1924 afsUUID_to_string(&identP->uuid, uuid1, 127);
1925 if (host->interface)
1926 afsUUID_to_string(&host->interface->uuid, uuid2, 127);
1928 ("CB: new identity for host %p (%s:%d), "
1929 "deleting(%x %p %s %s)\n",
1930 host, afs_inet_ntoa_r(host->host, hoststr), ntohs(host->port),
1931 identP->valid, host->interface,
1932 identP->valid ? uuid1 : "no_uuid",
1933 host->interface ? uuid2 : "no_uuid"));
1935 /* The host in the cache is not the host for this connection */
1937 host->hostFlags |= HOSTDELETED;
1943 host = h_Alloc_r(tcon); /* returned held and locked */
1946 h_gethostcps_r(host, time(NULL));
1947 if (!(host->Console & 1)) {
1949 cb_conn = host->callback_rxcon;
1950 rx_GetConnection(cb_conn);
1951 host->hostFlags |= HWHO_INPROGRESS;
1954 RXAFSCB_TellMeAboutYourself(cb_conn, &interf, &caps);
1955 if (code == RXGEN_OPCODE)
1956 code = RXAFSCB_WhoAreYou(cb_conn, &interf);
1957 rx_PutConnection(cb_conn);
1960 if ((code == RXGEN_OPCODE) ||
1961 ((code == 0) && (afs_uuid_equal(&interf.uuid, &nulluuid)))) {
1963 identP = malloc(sizeof(struct Identity));
1968 ViceLogThenPanic(0, ("Failed malloc in h_GetHost_r\n"));
1972 rx_SetSpecific(tcon, rxcon_ident_key, identP);
1974 ("Host %" AFS_PTR_FMT " (%s:%d) does not support WhoAreYou.\n",
1975 host, afs_inet_ntoa_r(host->host, hoststr),
1976 ntohs(host->port)));
1978 } else if (code == 0) {
1980 identP = malloc(sizeof(struct Identity));
1985 ViceLogThenPanic(0, ("Failed malloc in h_GetHost_r\n"));
1989 identP->uuid = interf.uuid;
1991 rx_SetSpecific(tcon, rxcon_ident_key, identP);
1993 ("WhoAreYou success on host %" AFS_PTR_FMT " (%s:%d)\n",
1994 host, afs_inet_ntoa_r(host->host, hoststr),
1995 ntohs(host->port)));
1997 if (code == 0 && !identP->valid) {
1998 cb_conn = host->callback_rxcon;
1999 rx_GetConnection(cb_conn);
2001 code = RXAFSCB_InitCallBackState(cb_conn);
2002 rx_PutConnection(cb_conn);
2005 } else if (code == 0) {
2006 oldHost = h_LookupUuid_r(&identP->uuid);
2011 if (oldHost->hostFlags & HOSTDELETED) {
2012 h_Unlock_r(oldHost);
2013 h_Release_r(oldHost);
2021 /* This is a new address for an existing host. Update
2022 * the list of interfaces for the existing host and
2023 * delete the host structure we just allocated. */
2025 /* mark the duplicate host as deleted before we do
2026 * anything. The probing code below may try to change
2027 * "oldHost" to the same IP address as "host" currently
2028 * has, and we do not want a pseudo-"collision" to be
2030 host->hostFlags |= HOSTDELETED;
2032 oldHost->hostFlags |= HWHO_INPROGRESS;
2034 if (oldHost->interface) {
2036 afsUUID uuid = oldHost->interface->uuid;
2037 cb_conn = oldHost->callback_rxcon;
2038 rx_GetConnection(cb_conn);
2039 rx_SetConnDeadTime(cb_conn, 2);
2040 rx_SetConnHardDeadTime(cb_conn, AFS_HARDDEADTIME);
2042 code2 = RXAFSCB_ProbeUuid(cb_conn, &uuid);
2044 rx_SetConnDeadTime(cb_conn, 50);
2045 rx_SetConnHardDeadTime(cb_conn, AFS_HARDDEADTIME);
2046 rx_PutConnection(cb_conn);
2049 /* The primary address is either not responding or
2050 * is not the client we are looking for.
2051 * MultiProbeAlternateAddress_r() will remove the
2052 * alternate interfaces that do not have the same
2054 ViceLog(0,("CB: ProbeUuid for host %" AFS_PTR_FMT " (%s:%d) failed %d\n",
2056 afs_inet_ntoa_r(oldHost->host, hoststr),
2057 ntohs(oldHost->port),code2));
2059 if (MultiProbeAlternateAddress_r(oldHost)) {
2060 /* If MultiProbeAlternateAddress_r succeeded,
2061 * it updated oldHost->host and oldHost->port
2062 * to an address that responded successfully to
2063 * a ProbeUuid, so it is as if the ProbeUuid
2064 * call above returned success. So, only set
2065 * 'probefail' if MultiProbeAlternateAddress_r
2074 if (oldHost->host != haddr || oldHost->port != hport) {
2075 struct rx_connection *rxconn;
2078 ("CB: Host %" AFS_PTR_FMT " (%s:%d) has new addr %s:%d\n",
2080 afs_inet_ntoa_r(oldHost->host, hoststr2),
2081 ntohs(oldHost->port),
2082 afs_inet_ntoa_r(haddr, hoststr),
2085 * add then remove. otherwise the host may get marked
2086 * deleted if we removed the only valid address.
2088 addInterfaceAddr_r(oldHost, haddr, hport);
2089 if (probefail || oldHost->host == haddr) {
2091 * The probe failed which means that the old
2092 * address is either unreachable or is not the
2093 * same host we were just contacted by. We will
2094 * also remove addresses if only the port has
2095 * changed because that indicates the client
2098 removeInterfaceAddr_r(oldHost, oldHost->host, oldHost->port);
2101 struct Interface *interface = oldHost->interface;
2102 int number = oldHost->interface->numberOfInterfaces;
2103 for (i = 0; i < number; i++) {
2104 if (interface->interface[i].addr == haddr &&
2105 interface->interface[i].port != hport) {
2107 * We have just been contacted by a client
2108 * that has been seen from behind a NAT
2109 * and at least one other address.
2111 removeInterfaceAddr_r(oldHost, haddr,
2112 interface->interface[i].port);
2117 oldHost->host = haddr;
2118 oldHost->port = hport;
2119 rxconn = oldHost->callback_rxcon;
2120 oldHost->callback_rxcon = host->callback_rxcon;
2121 host->callback_rxcon = rxconn;
2123 /* don't destroy rxconn here; let h_TossStuff_r
2124 * take care of that via h_Release_r below */
2126 host->hostFlags &= ~HWHO_INPROGRESS;
2128 /* release host because it was allocated by h_Alloc_r */
2131 /* the new host is held and locked */
2133 /* This really is a new host */
2134 h_AddHostToUuidHashTable_r(&identP->uuid, host);
2135 cb_conn = host->callback_rxcon;
2136 rx_GetConnection(cb_conn);
2139 RXAFSCB_InitCallBackState3(cb_conn,
2141 rx_PutConnection(cb_conn);
2146 ("InitCallBackState3 success on host %" AFS_PTR_FMT " (%s:%d)\n",
2147 host, afs_inet_ntoa_r(host->host, hoststr),
2148 ntohs(host->port)));
2149 opr_Assert(interfValid == 1);
2150 initInterfaceAddr_r(host, &interf);
2156 ("CB: RCallBackConnectBack failed for %" AFS_PTR_FMT " (%s:%d)\n",
2157 host, afs_inet_ntoa_r(host->host, hoststr), ntohs(host->port)));
2158 host->hostFlags |= VENUSDOWN;
2161 ("CB: RCallBackConnectBack succeeded for %" AFS_PTR_FMT " (%s:%d)\n",
2162 host, afs_inet_ntoa_r(host->host, hoststr), ntohs(host->port)));
2163 host->hostFlags |= RESETDONE;
2166 if (caps.Capabilities_val
2167 && (caps.Capabilities_val[0] & CLIENT_CAPABILITY_ERRORTRANS))
2168 host->hostFlags |= HERRORTRANS;
2170 host->hostFlags &= ~(HERRORTRANS);
2171 host->hostFlags |= ALTADDR; /* host structure initialization complete */
2172 host->hostFlags &= ~HWHO_INPROGRESS;
2177 if (caps.Capabilities_val)
2178 free(caps.Capabilities_val);
2179 caps.Capabilities_val = NULL;
2180 caps.Capabilities_len = 0;
2182 rx_DestroyConnection(cb_in);
2192 h_InitHostPackage(void)
2194 memset(&nulluuid, 0, sizeof(afsUUID));
2195 rxcon_ident_key = rx_KeyCreate((rx_destructor_t) free);
2196 rxcon_client_key = rx_KeyCreate((rx_destructor_t) 0);
2197 MUTEX_INIT(&host_glock_mutex, "host glock", MUTEX_DEFAULT, 0);
2201 MapName_r(char *uname, afs_int32 * aval)
2207 lnames.namelist_len = 1;
2208 lnames.namelist_val = (prname *) uname;
2209 lids.idlist_len = 0;
2210 lids.idlist_val = NULL;
2213 code = hpr_NameToId(&lnames, &lids);
2216 if (lids.idlist_val) {
2217 *aval = lids.idlist_val[0];
2218 if (*aval == AnonymousID) {
2220 ("MapName: NameToId on %s returns anonymousID\n",
2221 lnames.namelist_val[0]));
2223 free(lids.idlist_val); /* return parms are not malloced in stub if server proc aborts */
2226 ("MapName: NameToId on '%s' is unknown\n",
2227 lnames.namelist_val[0]));
2238 PerHost_EnumerateClient(struct host *host, void *arock)
2240 struct enumclient_args *args = arock;
2241 struct client *client;
2244 for (client = host->FirstClient; client; client = client->next) {
2245 if (!client->deleted && client->ViceId == args->vid) {
2250 code = (*args->proc)(client, args->rock);
2253 h_ReleaseClient_r(client);
2256 return H_ENUMERATE_BAIL(0);
2265 h_EnumerateClients(afs_int32 vid,
2266 int (*proc)(struct client *client, void *rock),
2269 struct enumclient_args args;
2275 h_Enumerate_r(PerHost_EnumerateClient, hostList, &args);
2280 format_vname(char *vname, int usize, const char *tname, const char *tinst,
2281 const char *tcell, afs_int32 islocal)
2285 len = strlcpy(vname, tname, usize);
2289 len = strlcat(vname, ".", usize);
2292 len = strlcat(vname, tinst, usize);
2296 if (tcell[0] && !islocal) {
2297 len = strlcat(vname, "@", usize);
2300 len = strlcat(vname, tcell, usize);
2308 getPeerDetails(struct rx_connection *conn,
2309 afs_int32 *viceid, afs_int32 *expTime, int authClass)
2312 #if (64-MAXKTCNAMELEN)
2313 ticket name length != 64
2317 char tcell[MAXKTCREALMLEN];
2318 char uname[PR_MAXNAMELEN];
2320 *viceid = AnonymousID;
2321 *expTime = 0x7fffffff;
2324 ("FindClient: authenticating connection: authClass=%d\n",
2326 if (authClass == 1) {
2327 /* A bcrypt tickets, no longer supported */
2328 ViceLog(1, ("FindClient: bcrypt ticket, using AnonymousID\n"));
2332 if (authClass == 2) {
2333 /* an rxkad ticket */
2337 /* kerberos ticket */
2338 code = rxkad_GetServerInfo(conn, /*level */ 0, (afs_uint32 *)expTime,
2339 tname, tinst, tcell, &kvno);
2341 ViceLog(1, ("Failed to get rxkad ticket info\n"));
2346 ("FindClient: rxkad conn: name=%s,inst=%s,cell=%s,exp=%d,kvno=%d\n",
2347 tname, tinst, tcell, *expTime, kvno));
2348 code = afsconf_IsLocalRealmMatch(confDir, &islocal, tname, tinst, tcell);
2351 ViceLog(0, ("FindClient: local realm check failed; code=%d", code));
2355 code = format_vname(uname, sizeof(uname), tname, tinst, tcell, islocal);
2357 ViceLog(0, ("FindClient: uname truncated."));
2361 /* translate the name to a vice id */
2362 code = MapName_r(uname, viceid);
2364 ViceLog(1, ("failed to map name=%s -> code=%d\n", uname,
2366 return code; /* Actually flag this is a failure */
2376 * Called by the server main loop. Returns a h_Held client, which must be
2377 * released later the main loop. Allocates a client if the matching one
2378 * isn't around. The client is returned with its reference count incremented
2379 * by one. The caller must call h_ReleaseClient_r when finished with
2382 * The refCount on client->host is returned incremented. h_ReleaseClient_r
2383 * does not decrement the refCount on client->host.
2386 h_FindClient_r(struct rx_connection *tcon)
2388 struct client *client;
2389 struct host *host = NULL;
2390 struct client *oldClient;
2391 afs_int32 viceid = 0;
2399 client = (struct client *)rx_GetSpecific(tcon, rxcon_client_key);
2400 if (client && client->sid == rx_GetConnectionId(tcon)
2401 && client->VenusEpoch == rx_GetConnectionEpoch(tcon)
2402 && !(client->host->hostFlags & HOSTDELETED)
2403 && !client->deleted) {
2406 h_Hold_r(client->host);
2407 if (client->prfail != 2) {
2408 /* Could add shared lock on client here */
2409 /* note that we don't have to lock entry in this path to
2410 * ensure CPS is initialized, since we don't call rx_SetSpecific
2411 * until initialization is done, and we only get here if
2412 * rx_GetSpecific located the client structure.
2417 ObtainWriteLock(&client->lock); /* released at end */
2423 authClass = rx_SecurityClassOf(tcon);
2425 code = getPeerDetails(tcon, &viceid, &expTime, authClass);
2429 if (!client) { /* loop */
2430 host = h_GetHost_r(tcon); /* Returns with incremented refCount */
2436 /* First try to find the client structure */
2437 for (client = host->FirstClient; client; client = client->next) {
2438 if (!client->deleted && (client->sid == rx_GetConnectionId(tcon))
2439 && (client->VenusEpoch == rx_GetConnectionEpoch(tcon))) {
2442 ObtainWriteLock(&client->lock);
2448 /* Still no client structure - get one */
2451 if (host->hostFlags & HOSTDELETED) {
2456 /* Retry to find the client structure */
2457 for (client = host->FirstClient; client; client = client->next) {
2458 if (!client->deleted && (client->sid == rx_GetConnectionId(tcon))
2459 && (client->VenusEpoch == rx_GetConnectionEpoch(tcon))) {
2461 goto retryfirstclient;
2466 ObtainWriteLock(&client->lock);
2467 client->refCount = 1;
2468 client->host = host;
2469 client->InSameNetwork = host->InSameNetwork;
2470 client->ViceId = viceid;
2471 client->expTime = expTime; /* rx only */
2472 client->authClass = authClass; /* rx only */
2473 client->sid = rx_GetConnectionId(tcon);
2474 client->VenusEpoch = rx_GetConnectionEpoch(tcon);
2475 client->CPS.prlist_val = NULL;
2476 client->CPS.prlist_len = 0;
2480 client->prfail = fail;
2482 if (!(client->CPS.prlist_val) || (viceid != client->ViceId)) {
2483 client->CPS.prlist_len = 0;
2484 if (client->CPS.prlist_val && (client->ViceId != ANONYMOUSID))
2485 free(client->CPS.prlist_val);
2486 client->CPS.prlist_val = NULL;
2487 client->ViceId = viceid;
2488 client->expTime = expTime;
2490 if (viceid == ANONYMOUSID) {
2491 client->CPS.prlist_len = AnonCPS.prlist_len;
2492 client->CPS.prlist_val = AnonCPS.prlist_val;
2495 code = hpr_GetCPS(viceid, &client->CPS);
2500 ("pr_GetCPS failed(%d) for user %d, host %" AFS_PTR_FMT " (%s:%d)\n",
2501 code, viceid, client->host,
2502 afs_inet_ntoa_r(client->host->host,hoststr),
2503 ntohs(client->host->port)));
2505 /* Although ubik_Call (called by pr_GetCPS) traverses thru
2506 * all protection servers and reevaluates things if no
2507 * sync server or quorum is found we could still end up
2508 * with one of these errors. In such case we would like to
2509 * reevaluate the rpc call to find if there's cps for this
2510 * guy. We treat other errors (except network failures
2511 * ones - i.e. code < 0) as an indication that there is no
2512 * CPS for this host. Ideally we could like to deal this
2513 * problem the other way around (i.e. if code == NOCPS
2514 * ignore else retry next time) but the problem is that
2515 * there're other errors (i.e. EPERM) for which we don't
2516 * want to retry and we don't know the whole code list!
2518 if (code < 0 || code == UNOQUORUM || code == UNOTSYNC)
2522 /* the disabling of system:administrators is so iffy and has so many
2523 * possible failure modes that we will disable it again */
2524 /* Turn off System:Administrator for safety
2525 * if (AL_IsAMember(SystemId, client->CPS) == 0)
2526 * osi_Assert(AL_DisableGroup(SystemId, client->CPS) == 0); */
2529 /* Now, tcon may already be set to a rock, since we blocked with no host
2530 * or client locks set above in pr_GetCPS (XXXX some locking is probably
2531 * required). So, before setting the RPC's rock, we should disconnect
2532 * the RPC from the other client structure's rock.
2534 oldClient = (struct client *)rx_GetSpecific(tcon, rxcon_client_key);
2535 if (oldClient && oldClient != client
2536 && oldClient->sid == rx_GetConnectionId(tcon)
2537 && oldClient->VenusEpoch == rx_GetConnectionEpoch(tcon)
2538 && !(oldClient->host->hostFlags & HOSTDELETED)) {
2540 if (!oldClient->deleted) {
2541 /* if we didn't create it, it's not ours to put back */
2543 ViceLog(0, ("FindClient: stillborn client %p(%x); "
2544 "conn %p (host %s:%d) had client %p(%x)\n",
2545 client, client->sid, tcon,
2546 afs_inet_ntoa_r(rxr_HostOf(tcon), hoststr),
2547 ntohs(rxr_PortOf(tcon)),
2548 oldClient, oldClient->sid));
2549 if ((client->ViceId != ANONYMOUSID) && client->CPS.prlist_val)
2550 free(client->CPS.prlist_val);
2551 client->CPS.prlist_val = NULL;
2552 client->CPS.prlist_len = 0;
2554 /* We should perhaps check for 0 here */
2556 ReleaseWriteLock(&client->lock);
2561 oldClient->refCount++;
2563 h_Hold_r(oldClient->host);
2564 h_Release_r(client->host);
2567 ObtainWriteLock(&oldClient->lock);
2570 host = oldClient->host;
2572 ViceLog(0, ("FindClient: deleted client %p(%x ref %d host %p href "
2573 "%d) already had conn %p (host %s:%d, cid %x), stolen "
2574 "by client %p(%x, ref %d host %p href %d)\n",
2575 oldClient, oldClient->sid, oldClient->refCount,
2576 oldClient->host, oldClient->host->refCount, tcon,
2577 afs_inet_ntoa_r(rxr_HostOf(tcon), hoststr),
2578 ntohs(rxr_PortOf(tcon)), rx_GetConnectionId(tcon),
2579 client, client->sid, client->refCount,
2580 client->host, client->host->refCount));
2581 /* rx_SetSpecific will be done immediately below */
2584 /* Avoid chaining in more than once. */
2588 if (host->hostFlags & HOSTDELETED) {
2593 client->host = NULL;
2595 if ((client->ViceId != ANONYMOUSID) && client->CPS.prlist_val)
2596 free(client->CPS.prlist_val);
2597 client->CPS.prlist_val = NULL;
2598 client->CPS.prlist_len = 0;
2601 ReleaseWriteLock(&client->lock);
2606 client->next = host->FirstClient;
2607 host->FirstClient = client;
2609 CurrentConnections++; /* increment number of connections */
2611 rx_SetSpecific(tcon, rxcon_client_key, client);
2612 ReleaseWriteLock(&client->lock);
2616 } /*h_FindClient_r */
2619 h_ReleaseClient_r(struct client *client)
2621 opr_Assert(client->refCount > 0);
2628 * Sigh: this one is used to get the client AGAIN within the individual
2629 * server routines. This does not bother h_Holding the host, since
2630 * this is assumed already have been done by the server main loop.
2631 * It does check tokens, since only the server routines can return the
2632 * VICETOKENDEAD error code
2635 GetClient(struct rx_connection *tcon, struct client **cp)
2637 struct client *client;
2642 client = (struct client *)rx_GetSpecific(tcon, rxcon_client_key);
2643 if (client == NULL) {
2645 ("GetClient: no client in conn %p (host %s:%d), VBUSYING\n",
2646 tcon, afs_inet_ntoa_r(rxr_HostOf(tcon), hoststr),
2647 ntohs(rxr_PortOf(tcon))));
2651 if (rx_GetConnectionId(tcon) != client->sid
2652 || rx_GetConnectionEpoch(tcon) != client->VenusEpoch) {
2654 ("GetClient: tcon %p tcon sid %d client sid %d\n",
2655 tcon, rx_GetConnectionId(tcon), client->sid));
2659 if (client && client->LastCall > client->expTime && client->expTime) {
2661 ("Token for %s at %s:%d expired %d\n", h_UserName(client),
2662 afs_inet_ntoa_r(client->host->host, hoststr),
2663 ntohs(client->host->port), client->expTime));
2665 return VICETOKENDEAD;
2667 if (client->deleted) {
2668 ViceLog(0, ("GetClient: got deleted client, connection will appear "
2669 "anonymous; tcon %p cid %x client %p ref %d host %p "
2670 "(%s:%d) href %d ViceId %d\n",
2671 tcon, rx_GetConnectionId(tcon), client, client->refCount,
2673 afs_inet_ntoa_r(client->host->host, hoststr),
2674 (int)ntohs(client->host->port), client->host->refCount,
2675 (int)client->ViceId));
2685 PutClient(struct client **cp)
2691 h_ReleaseClient_r(*cp);
2698 /* Client user name for short term use. Note that this is NOT inexpensive */
2700 h_UserName(struct client *client)
2702 static char User[PR_MAXNAMELEN + 1];
2706 lids.idlist_len = 1;
2707 lids.idlist_val = malloc(1 * sizeof(afs_int32));
2708 if (!lids.idlist_val) {
2709 ViceLogThenPanic(0, ("Failed malloc in h_UserName\n"));
2711 lnames.namelist_len = 0;
2712 lnames.namelist_val = (prname *) 0;
2713 lids.idlist_val[0] = client->ViceId;
2714 if (hpr_IdToName(&lids, &lnames)) {
2715 /* We need to free id we alloced above! */
2716 free(lids.idlist_val);
2717 return "*UNKNOWN USER NAME*";
2719 strncpy(User, lnames.namelist_val[0], PR_MAXNAMELEN);
2720 free(lids.idlist_val);
2721 free(lnames.namelist_val);
2730 ("Total Client entries = %d, blocks = %d; Host entries = %d, blocks = %d\n",
2731 CEs, CEBlocks, HTs, HTBlocks));
2737 h_PrintClient(struct host *host, void *rock)
2739 StreamHandle_t *file = (StreamHandle_t *)rock;
2740 struct client *client;
2745 time_t LastCall, expTime;
2749 LastCall = host->LastCall;
2750 if (host->hostFlags & HOSTDELETED) {
2754 strftime(tbuffer, sizeof(tbuffer), "%a %b %d %H:%M:%S %Y",
2755 localtime_r(&LastCall, &tm));
2756 snprintf(tmpStr, sizeof tmpStr, "Host %s:%d down = %d, LastCall %s\n",
2757 afs_inet_ntoa_r(host->host, hoststr),
2758 ntohs(host->port), (host->hostFlags & VENUSDOWN),
2760 (void)STREAM_WRITE(tmpStr, strlen(tmpStr), 1, file);
2761 for (client = host->FirstClient; client; client = client->next) {
2762 if (!client->deleted) {
2763 expTime = client->expTime;
2764 strftime(tbuffer, sizeof(tbuffer), "%a %b %d %H:%M:%S %Y",
2765 localtime_r(&expTime, &tm));
2766 snprintf(tmpStr, sizeof tmpStr,
2767 " user id=%d, name=%s, sl=%s till %s\n",
2768 client->ViceId, h_UserName(client),
2769 client->authClass ? "Authenticated"
2770 : "Not authenticated",
2771 client->authClass ? tbuffer : "No Limit");
2772 (void)STREAM_WRITE(tmpStr, strlen(tmpStr), 1, file);
2773 snprintf(tmpStr, sizeof tmpStr, " CPS-%d is [",
2774 client->CPS.prlist_len);
2775 (void)STREAM_WRITE(tmpStr, strlen(tmpStr), 1, file);
2776 if (client->CPS.prlist_val) {
2777 for (i = 0; i < client->CPS.prlist_len; i++) {
2778 snprintf(tmpStr, sizeof tmpStr, " %d",
2779 client->CPS.prlist_val[i]);
2780 (void)STREAM_WRITE(tmpStr, strlen(tmpStr), 1, file);
2783 sprintf(tmpStr, "]\n");
2784 (void)STREAM_WRITE(tmpStr, strlen(tmpStr), 1, file);
2790 } /*h_PrintClient */
2795 * Print a list of clients, with last security level and token value seen,
2799 h_PrintClients(void)
2806 StreamHandle_t *file = STREAM_OPEN(AFSDIR_SERVER_CLNTDUMP_FILEPATH, "w");
2810 ("Couldn't create client dump file %s\n",
2811 AFSDIR_SERVER_CLNTDUMP_FILEPATH));
2815 strftime(tbuffer, sizeof(tbuffer), "%a %b %d %H:%M:%S %Y",
2816 localtime_r(&now, &tm));
2817 snprintf(tmpStr, sizeof tmpStr, "List of active users at %s\n\n",
2819 (void)STREAM_WRITE(tmpStr, strlen(tmpStr), 1, file);
2820 h_Enumerate(h_PrintClient, (char *)file);
2821 STREAM_REALLYCLOSE(file);
2822 ViceLog(0, ("Created client dump %s\n", AFSDIR_SERVER_CLNTDUMP_FILEPATH));
2829 h_DumpHost(struct host *host, void *rock)
2831 StreamHandle_t *file = (StreamHandle_t *)rock;
2838 snprintf(tmpStr, sizeof tmpStr,
2839 "ip:%s port:%d hidx:%d cbid:%d lock:%x last:%u active:%u "
2840 "down:%d del:%d cons:%d cldel:%d\n\t hpfailed:%d hcpsCall:%u "
2842 afs_inet_ntoa_r(host->host, hoststr), ntohs(host->port),
2843 host->index, host->cblist, CheckLock(&host->lock),
2844 host->LastCall, host->ActiveCall, (host->hostFlags & VENUSDOWN),
2845 host->hostFlags & HOSTDELETED, host->Console,
2846 host->hostFlags & CLIENTDELETED, host->hcpsfailed,
2848 (void)STREAM_WRITE(tmpStr, strlen(tmpStr), 1, file);
2849 if (host->hcps.prlist_val)
2850 for (i = 0; i < host->hcps.prlist_len; i++) {
2851 snprintf(tmpStr, sizeof tmpStr, " %d", host->hcps.prlist_val[i]);
2852 (void)STREAM_WRITE(tmpStr, strlen(tmpStr), 1, file);
2854 sprintf(tmpStr, "] [");
2855 (void)STREAM_WRITE(tmpStr, strlen(tmpStr), 1, file);
2856 if (host->interface)
2857 for (i = 0; i < host->interface->numberOfInterfaces; i++) {
2859 sprintf(tmpStr, " %s:%d",
2860 afs_inet_ntoa_r(host->interface->interface[i].addr, hoststr),
2861 ntohs(host->interface->interface[i].port));
2862 (void)STREAM_WRITE(tmpStr, strlen(tmpStr), 1, file);
2864 sprintf(tmpStr, "] refCount:%d hostFlags:%hu\n", host->refCount, host->hostFlags);
2865 (void)STREAM_WRITE(tmpStr, strlen(tmpStr), 1, file);
2877 StreamHandle_t *file = STREAM_OPEN(AFSDIR_SERVER_HOSTDUMP_FILEPATH, "w");
2884 ("Couldn't create host dump file %s\n",
2885 AFSDIR_SERVER_HOSTDUMP_FILEPATH));
2889 strftime(tbuffer, sizeof(tbuffer), "%a %b %d %H:%M:%S %Y",
2890 localtime_r(&now, &tm));
2891 snprintf(tmpStr, sizeof tmpStr, "List of active hosts at %s\n\n", tbuffer);
2892 (void)STREAM_WRITE(tmpStr, strlen(tmpStr), 1, file);
2893 h_Enumerate(h_DumpHost, (char *)file);
2894 STREAM_REALLYCLOSE(file);
2895 ViceLog(0, ("Created host dump %s\n", AFSDIR_SERVER_HOSTDUMP_FILEPATH));
2899 #ifdef AFS_DEMAND_ATTACH_FS
2902 * host state serialization
2904 static int h_stateFillHeader(struct host_state_header * hdr);
2905 static int h_stateCheckHeader(struct host_state_header * hdr);
2906 static int h_stateAllocMap(struct fs_dump_state * state);
2907 static int h_stateSaveHost(struct host * host, void *rock);
2908 static int h_stateRestoreHost(struct fs_dump_state * state);
2909 static int h_stateRestoreIndex(struct host * h, void *rock);
2910 static int h_stateVerifyHost(struct host * h, void *rock);
2911 static int h_stateVerifyAddrHash(struct fs_dump_state * state, struct host * h,
2912 afs_uint32 addr, afs_uint16 port, int valid);
2913 static int h_stateVerifyUuidHash(struct fs_dump_state * state, struct host * h);
2914 static void h_hostToDiskEntry_r(struct host * in, struct hostDiskEntry * out);
2915 static void h_diskEntryToHost_r(struct hostDiskEntry * in, struct host * out);
2918 * Is this host busy?
2920 * This is just a hint and should not be trusted; this should probably only be
2921 * used by the host state serialization code when trying to detect if a host
2922 * can be sanely serialized to disk or not. If this function returns 1, the
2923 * host may be in an invalid state and thus should not be saved to disk.
2926 h_isBusy_r(struct host *host)
2928 struct Lock *hostLock = &host->lock;
2931 LOCK_LOCK(hostLock);
2932 if (hostLock->excl_locked || hostLock->readers_reading) {
2935 LOCK_UNLOCK(hostLock);
2941 if ((host->hostFlags & HWHO_INPROGRESS) || !(host->hostFlags & ALTADDR)) {
2942 /* We shouldn't hit this if the host wasn't locked, but just in case... */
2949 /* this procedure saves all host state to disk for fast startup */
2951 h_stateSave(struct fs_dump_state * state)
2953 AssignInt64(state->eof_offset, &state->hdr->h_offset);
2956 ViceLog(0, ("h_stateSave: hostCount=%d\n", hostCount));
2958 /* invalidate host state header */
2959 memset(state->h_hdr, 0, sizeof(struct host_state_header));
2961 if (fs_stateWriteHeader(state, &state->hdr->h_offset, state->h_hdr,
2962 sizeof(struct host_state_header))) {
2967 fs_stateIncEOF(state, sizeof(struct host_state_header));
2969 h_Enumerate_r(h_stateSaveHost, hostList, (char *)state);
2974 h_stateFillHeader(state->h_hdr);
2976 /* write the real header to disk */
2977 state->bail = fs_stateWriteHeader(state, &state->hdr->h_offset, state->h_hdr,
2978 sizeof(struct host_state_header));
2985 * host state serialization
2987 * this procedure restores all host state from a disk for fast startup
2990 h_stateRestore(struct fs_dump_state * state)
2994 /* seek to the right position and read in the host state header */
2995 if (fs_stateReadHeader(state, &state->hdr->h_offset, state->h_hdr,
2996 sizeof(struct host_state_header))) {
3001 /* check the validity of the header */
3002 if (h_stateCheckHeader(state->h_hdr)) {
3007 records = state->h_hdr->records;
3009 if (h_stateAllocMap(state)) {
3014 /* iterate over records restoring host state */
3015 for (i=0; i < records; i++) {
3016 if (h_stateRestoreHost(state) != 0) {
3027 h_stateRestoreIndices(struct fs_dump_state * state)
3029 h_Enumerate_r(h_stateRestoreIndex, hostList, (char *)state);
3034 h_stateRestoreIndex(struct host * h, void *rock)
3036 struct fs_dump_state *state = (struct fs_dump_state *)rock;
3037 if (cb_OldToNew(state, h->cblist, &h->cblist)) {
3038 return H_ENUMERATE_BAIL(0);
3044 h_stateVerify(struct fs_dump_state * state)
3046 h_Enumerate_r(h_stateVerifyHost, hostList, (char *)state);
3051 h_stateVerifyHost(struct host * h, void* rock)
3053 struct fs_dump_state *state = (struct fs_dump_state *)rock;
3057 ViceLog(0, ("h_stateVerifyHost: error: NULL host pointer in linked list\n"));
3058 return H_ENUMERATE_BAIL(0);
3062 for (i = h->interface->numberOfInterfaces-1; i >= 0; i--) {
3063 if (h_stateVerifyAddrHash(state, h, h->interface->interface[i].addr,
3064 h->interface->interface[i].port,
3065 h->interface->interface[i].valid)) {
3069 if (h_stateVerifyUuidHash(state, h)) {
3072 } else if (h_stateVerifyAddrHash(state, h, h->host, h->port, 1)) {
3076 if (cb_stateVerifyHCBList(state, h)) {
3084 * verify a host is either in, or absent from, the addr hash table.
3086 * @param[in] state fs dump state
3087 * @param[in] h host we're dealing with
3088 * @param[in] addr addr to look for (NBO)
3089 * @param[in] port port to look for (NBO)
3090 * @param[in] valid 1 if we're verifying that the specified addr and port
3091 * in the hash table point to the specified host. 0 if we're
3092 * verifying that the specified addr and port do NOT point
3093 * to the specified host
3095 * @return operation status
3096 * @retval 1 failed to verify, bail out
3097 * @retval 0 verified successfully, all is well
3100 h_stateVerifyAddrHash(struct fs_dump_state * state, struct host * h,
3101 afs_uint32 addr, afs_uint16 port, int valid)
3103 int ret = 0, found = 0;
3104 struct host *host = NULL;
3105 struct h_AddrHashChain *chain;
3106 int index = h_HashIndex(addr);
3110 for (chain = hostAddrHashTable[index]; chain; chain = chain->next) {
3111 host = chain->hostPtr;
3113 afs_inet_ntoa_r(addr, tmp);
3114 ViceLog(0, ("h_stateVerifyAddrHash: error: addr hash chain has NULL host ptr (lookup addr %s)\n", tmp));
3118 if ((chain->addr == addr) && (chain->port == port)) {
3121 ViceLog(0, ("h_stateVerifyAddrHash: warning: addr hash entry "
3122 "points to different host struct (%d, %d)\n",
3123 h->index, host->index));
3124 state->flags.warnings_generated = 1;
3128 ViceLog(0, ("h_stateVerifyAddrHash: error: addr %s:%u is "
3129 "marked invalid, but points to the containing "
3130 "host\n", afs_inet_ntoa_r(addr, tmp),
3131 (unsigned)htons(port)));
3139 if (chain_len > FS_STATE_H_MAX_ADDR_HASH_CHAIN_LEN) {
3140 ViceLog(0, ("h_stateVerifyAddrHash: error: hash chain length exceeds %d; assuming there's a loop\n",
3141 FS_STATE_H_MAX_ADDR_HASH_CHAIN_LEN));
3148 if (!found && valid) {
3149 afs_inet_ntoa_r(addr, tmp);
3150 if (state->mode == FS_STATE_LOAD_MODE) {
3151 ViceLog(0, ("h_stateVerifyAddrHash: error: addr %s:%u not found in hash\n",
3152 tmp, (unsigned)htons(port)));
3156 ViceLog(0, ("h_stateVerifyAddrHash: warning: addr %s:%u not found in hash\n",
3157 tmp, (unsigned)htons(port)));
3158 state->flags.warnings_generated = 1;
3167 h_stateVerifyUuidHash(struct fs_dump_state * state, struct host * h)
3169 int ret = 0, found = 0;
3170 struct host *host = NULL;
3171 struct h_UuidHashChain *chain;
3172 afsUUID * uuidp = &h->interface->uuid;
3173 int index = h_UuidHashIndex(uuidp);
3177 for (chain = hostUuidHashTable[index]; chain; chain = chain->next) {
3178 host = chain->hostPtr;
3180 afsUUID_to_string(uuidp, tmp, sizeof(tmp));
3181 ViceLog(0, ("h_stateVerifyUuidHash: error: uuid hash chain has NULL host ptr (lookup uuid %s)\n", tmp));
3185 if (host->interface &&
3186 afs_uuid_equal(&host->interface->uuid, uuidp)) {
3188 ViceLog(0, ("h_stateVerifyUuidHash: warning: uuid hash entry points to different host struct (%d, %d)\n",
3189 h->index, host->index));
3190 state->flags.warnings_generated = 1;
3195 if (chain_len > FS_STATE_H_MAX_UUID_HASH_CHAIN_LEN) {
3196 ViceLog(0, ("h_stateVerifyUuidHash: error: hash chain length exceeds %d; assuming there's a loop\n",
3197 FS_STATE_H_MAX_UUID_HASH_CHAIN_LEN));
3205 afsUUID_to_string(uuidp, tmp, sizeof(tmp));
3206 if (state->mode == FS_STATE_LOAD_MODE) {
3207 ViceLog(0, ("h_stateVerifyUuidHash: error: uuid %s not found in hash\n", tmp));
3211 ViceLog(0, ("h_stateVerifyUuidHash: warning: uuid %s not found in hash\n", tmp));
3212 state->flags.warnings_generated = 1;
3220 /* create the host state header structure */
3222 h_stateFillHeader(struct host_state_header * hdr)
3224 hdr->stamp.magic = HOST_STATE_MAGIC;
3225 hdr->stamp.version = HOST_STATE_VERSION;
3229 /* check the contents of the host state header structure */
3231 h_stateCheckHeader(struct host_state_header * hdr)
3235 if (hdr->stamp.magic != HOST_STATE_MAGIC) {
3236 ViceLog(0, ("check_host_state_header: invalid state header\n"));
3239 else if (hdr->stamp.version != HOST_STATE_VERSION) {
3240 ViceLog(0, ("check_host_state_header: unknown version number\n"));
3246 /* allocate the host id mapping table */
3248 h_stateAllocMap(struct fs_dump_state * state)
3250 state->h_map.len = state->h_hdr->index_max + 1;
3251 state->h_map.entries = (struct idx_map_entry_t *)
3252 calloc(state->h_map.len, sizeof(struct idx_map_entry_t));
3253 return (state->h_map.entries != NULL) ? 0 : 1;
3256 /* function called by h_Enumerate to save a host to disk */
3258 h_stateSaveHost(struct host * host, void* rock)
3260 struct fs_dump_state *state = (struct fs_dump_state *) rock;
3261 int if_len=0, hcps_len=0;
3262 struct hostDiskEntry hdsk;
3263 struct host_state_entry_header hdr;
3264 struct Interface * ifp = NULL;
3265 afs_int32 * hcps = NULL;
3266 struct iovec iov[4];
3269 if (h_isBusy_r(host)) {
3271 ViceLog(1, ("Not saving host %s:%d to disk; host appears busy\n",
3272 afs_inet_ntoa_r(host->host, hoststr), (int)ntohs(host->port)));
3273 /* Make sure we don't try to save callbacks to disk for this host, or
3274 * we'll get confused on restore */
3275 DeleteAllCallBacks_r(host, 1);
3279 memset(&hdr, 0, sizeof(hdr));
3281 if (state->h_hdr->index_max < host->index) {
3282 state->h_hdr->index_max = host->index;
3285 h_hostToDiskEntry_r(host, &hdsk);
3286 if (host->interface) {
3287 if_len = sizeof(struct Interface) +
3288 ((host->interface->numberOfInterfaces-1) * sizeof(struct AddrPort));
3289 ifp = malloc(if_len);
3290 opr_Assert(ifp != NULL);
3291 memcpy(ifp, host->interface, if_len);
3292 hdr.interfaces = host->interface->numberOfInterfaces;
3293 iov[iovcnt].iov_base = (char *) ifp;
3294 iov[iovcnt].iov_len = if_len;
3297 if (host->hcps.prlist_val) {
3298 hdr.hcps = host->hcps.prlist_len;
3299 hcps_len = hdr.hcps * sizeof(afs_int32);
3300 hcps = malloc(hcps_len);
3301 opr_Assert(hcps != NULL);
3302 memcpy(hcps, host->hcps.prlist_val, hcps_len);
3303 iov[iovcnt].iov_base = (char *) hcps;
3304 iov[iovcnt].iov_len = hcps_len;
3308 if (hdsk.index > state->h_hdr->index_max)
3309 state->h_hdr->index_max = hdsk.index;
3311 hdr.len = sizeof(struct host_state_entry_header) +
3312 sizeof(struct hostDiskEntry) + if_len + hcps_len;
3313 hdr.magic = HOST_STATE_ENTRY_MAGIC;
3315 iov[0].iov_base = (char *) &hdr;
3316 iov[0].iov_len = sizeof(hdr);
3317 iov[1].iov_base = (char *) &hdsk;
3318 iov[1].iov_len = sizeof(struct hostDiskEntry);
3320 if (fs_stateWriteV(state, iov, iovcnt)) {
3321 ViceLog(0, ("h_stateSaveHost: failed to save host %d", host->index));
3325 fs_stateIncEOF(state, hdr.len);
3327 state->h_hdr->records++;
3334 return H_ENUMERATE_BAIL(0);
3339 /* restores a host from disk */
3341 h_stateRestoreHost(struct fs_dump_state * state)
3343 int ifp_len=0, hcps_len=0, bail=0;
3344 struct host_state_entry_header hdr;
3345 struct hostDiskEntry hdsk;
3346 struct host *host = NULL;
3347 struct Interface *ifp = NULL;
3348 afs_int32 * hcps = NULL;
3349 struct iovec iov[3];
3352 if (fs_stateRead(state, &hdr, sizeof(hdr))) {
3353 ViceLog(0, ("h_stateRestoreHost: failed to read host entry header from dump file '%s'\n",
3359 if (hdr.magic != HOST_STATE_ENTRY_MAGIC) {
3360 ViceLog(0, ("h_stateRestoreHost: fileserver state dump file '%s' is corrupt.\n",
3366 iov[0].iov_base = (char *) &hdsk;
3367 iov[0].iov_len = sizeof(struct hostDiskEntry);
3369 if (hdr.interfaces) {
3370 ifp_len = sizeof(struct Interface) +
3371 ((hdr.interfaces-1) * sizeof(struct AddrPort));
3372 ifp = malloc(ifp_len);
3373 opr_Assert(ifp != NULL);
3374 iov[iovcnt].iov_base = (char *) ifp;
3375 iov[iovcnt].iov_len = ifp_len;
3379 hcps_len = hdr.hcps * sizeof(afs_int32);
3380 hcps = malloc(hcps_len);
3381 opr_Assert(hcps != NULL);
3382 iov[iovcnt].iov_base = (char *) hcps;
3383 iov[iovcnt].iov_len = hcps_len;
3387 if ((ifp_len + hcps_len + sizeof(hdsk) + sizeof(hdr)) != hdr.len) {
3388 ViceLog(0, ("h_stateRestoreHost: host entry header length fields are inconsistent\n"));
3393 if (fs_stateReadV(state, iov, iovcnt)) {
3394 ViceLog(0, ("h_stateRestoreHost: failed to read host entry\n"));
3399 if (!hdr.hcps && hdsk.hcps_valid) {
3400 /* valid, zero-length host cps ; does this ever happen? */
3401 hcps = malloc(sizeof(afs_int32));
3402 opr_Assert(hcps != NULL);
3405 if ((hdsk.hostFlags & HWHO_INPROGRESS) || !(hdsk.hostFlags & ALTADDR)) {
3407 ViceLog(0, ("h_stateRestoreHost: skipping host %s:%d due to invalid flags 0x%x\n",
3408 afs_inet_ntoa_r(hdsk.host, hoststr), (int)ntohs(hdsk.port),
3409 (unsigned)hdsk.hostFlags));
3411 state->h_map.entries[hdsk.index].valid = FS_STATE_IDX_SKIPPED;
3415 /* for restoring state, we better be able to get a host! */
3417 opr_Assert(host != NULL);
3420 host->interface = ifp;
3423 host->hcps.prlist_val = hcps;
3424 host->hcps.prlist_len = hdr.hcps;
3427 h_diskEntryToHost_r(&hdsk, host);
3428 h_SetupCallbackConn_r(host);
3430 h_AddHostToAddrHashTable_r(host->host, host->port, host);
3433 for (i = ifp->numberOfInterfaces-1; i >= 0; i--) {
3434 if (ifp->interface[i].valid &&
3435 !(ifp->interface[i].addr == host->host &&
3436 ifp->interface[i].port == host->port)) {
3437 h_AddHostToAddrHashTable_r(ifp->interface[i].addr,
3438 ifp->interface[i].port,
3442 h_AddHostToUuidHashTable_r(&ifp->uuid, host);
3444 h_InsertList_r(host);
3446 /* setup host id map entry */
3447 state->h_map.entries[hdsk.index].valid = FS_STATE_IDX_VALID;
3448 state->h_map.entries[hdsk.index].old_idx = hdsk.index;
3449 state->h_map.entries[hdsk.index].new_idx = host->index;
3461 /* serialize a host structure to disk */
3463 h_hostToDiskEntry_r(struct host * in, struct hostDiskEntry * out)
3465 out->host = in->host;
3466 out->port = in->port;
3467 out->hostFlags = in->hostFlags;
3468 out->Console = in->Console;
3469 out->hcpsfailed = in->hcpsfailed;
3470 out->LastCall = in->LastCall;
3471 out->ActiveCall = in->ActiveCall;
3472 out->cpsCall = in->cpsCall;
3473 out->cblist = in->cblist;
3474 out->InSameNetwork = in->InSameNetwork;
3476 /* special fields we save, but are not memcpy'd back on restore */
3477 out->index = in->index;
3478 out->hcps_len = in->hcps.prlist_len;
3479 out->hcps_valid = (in->hcps.prlist_val == NULL) ? 0 : 1;
3482 /* restore a host structure from disk */
3484 h_diskEntryToHost_r(struct hostDiskEntry * in, struct host * out)
3486 out->host = in->host;
3487 out->port = in->port;
3488 out->hostFlags = in->hostFlags;
3489 out->Console = in->Console;
3490 out->hcpsfailed = in->hcpsfailed;
3491 out->LastCall = in->LastCall;
3492 out->ActiveCall = in->ActiveCall;
3493 out->cpsCall = in->cpsCall;
3494 out->cblist = in->cblist;
3495 out->InSameNetwork = in->InSameNetwork;
3498 /* index translation routines */
3500 h_OldToNew(struct fs_dump_state * state, afs_uint32 old, afs_uint32 * new)
3504 /* hosts use a zero-based index, so old==0 is valid */
3506 if (old >= state->h_map.len) {
3507 ViceLog(0, ("h_OldToNew: index %d is out of range\n", old));
3509 } else if (state->h_map.entries[old].valid != FS_STATE_IDX_VALID ||
3510 state->h_map.entries[old].old_idx != old) { /* sanity check */
3511 ViceLog(0, ("h_OldToNew: index %d points to an invalid host record\n", old));
3514 *new = state->h_map.entries[old].new_idx;
3519 #endif /* AFS_DEMAND_ATTACH_FS */
3523 * This counts the number of workstations, the number of active workstations,
3524 * and the number of workstations declared "down" (i.e. not heard from
3525 * recently). An active workstation has received a call since the cutoff
3526 * time argument passed.
3529 h_GetWorkStats(int *nump, int *activep, int *delp, afs_int32 cutofftime)
3532 int num = 0, active = 0, del = 0;
3536 for (count = 0, host = hostList; host && count < hostCount; host = host->next, count++) {
3537 if (!(host->hostFlags & HOSTDELETED)) {
3539 if (host->ActiveCall > cutofftime)
3541 if (host->hostFlags & VENUSDOWN)
3545 if (count != hostCount) {
3546 ViceLog(0, ("h_GetWorkStats found %d of %d hosts\n", count, hostCount));
3547 } else if (host != NULL) {
3548 ViceLog(0, ("h_GetWorkStats found more than %d hosts\n", hostCount));
3549 ShutDownAndCore(PANIC);
3559 } /*h_GetWorkStats */
3562 h_GetWorkStats64(afs_uint64 *nump, afs_uint64 *activep, afs_uint64 *delp,
3563 afs_int32 cutofftime)
3565 int num, active, del;
3566 h_GetWorkStats(&num, &active, &del, cutofftime);
3575 /*------------------------------------------------------------------------
3576 * PRIVATE h_ClassifyAddress
3579 * Given a target IP address and a candidate IP address (both
3580 * in host byte order), classify the candidate into one of three
3581 * buckets in relation to the target by bumping the counters passed
3585 * a_targetAddr : Target address.
3586 * a_candAddr : Candidate address.
3587 * a_sameNetOrSubnetP : Ptr to counter to bump when the two
3588 * addresses are either in the same network
3589 * or the same subnet.
3590 * a_diffSubnetP : ...when the candidate is in a different
3592 * a_diffNetworkP : ...when the candidate is in a different
3599 * The target and candidate addresses are both in host byte
3600 * order, NOT network byte order, when passed in.
3604 *------------------------------------------------------------------------*/
3607 h_ClassifyAddress(afs_uint32 a_targetAddr, afs_uint32 a_candAddr,
3608 afs_int32 * a_sameNetOrSubnetP, afs_int32 * a_diffSubnetP,
3609 afs_int32 * a_diffNetworkP)
3610 { /*h_ClassifyAddress */
3612 afs_uint32 targetNet;
3613 afs_uint32 targetSubnet;
3615 afs_uint32 candSubnet;
3618 * Put bad values into the subnet info to start with.
3620 targetSubnet = (afs_uint32) 0;
3621 candSubnet = (afs_uint32) 0;
3624 * Pull out the network and subnetwork numbers from the target
3625 * and candidate addresses. We can short-circuit this whole
3626 * affair if the target and candidate addresses are not of the
3629 if (IN_CLASSA(a_targetAddr)) {
3630 if (!(IN_CLASSA(a_candAddr))) {
3631 (*a_diffNetworkP)++;
3634 targetNet = a_targetAddr & IN_CLASSA_NET;
3635 candNet = a_candAddr & IN_CLASSA_NET;
3636 if (IN_SUBNETA(a_targetAddr))
3637 targetSubnet = a_targetAddr & IN_CLASSA_SUBNET;
3638 if (IN_SUBNETA(a_candAddr))
3639 candSubnet = a_candAddr & IN_CLASSA_SUBNET;
3640 } else if (IN_CLASSB(a_targetAddr)) {
3641 if (!(IN_CLASSB(a_candAddr))) {
3642 (*a_diffNetworkP)++;
3645 targetNet = a_targetAddr & IN_CLASSB_NET;
3646 candNet = a_candAddr & IN_CLASSB_NET;
3647 if (IN_SUBNETB(a_targetAddr))
3648 targetSubnet = a_targetAddr & IN_CLASSB_SUBNET;
3649 if (IN_SUBNETB(a_candAddr))
3650 candSubnet = a_candAddr & IN_CLASSB_SUBNET;
3651 } /*Class B target */
3652 else if (IN_CLASSC(a_targetAddr)) {
3653 if (!(IN_CLASSC(a_candAddr))) {
3654 (*a_diffNetworkP)++;
3657 targetNet = a_targetAddr & IN_CLASSC_NET;
3658 candNet = a_candAddr & IN_CLASSC_NET;
3661 * Note that class C addresses can't have subnets,
3662 * so we leave the defaults untouched.
3664 } /*Class C target */
3666 targetNet = a_targetAddr;
3667 candNet = a_candAddr;
3668 } /*Class D address */
3671 * Now, simply compare the extracted net and subnet values for
3672 * the two addresses (which at this point are known to be of the
3675 if (targetNet == candNet) {
3676 if (targetSubnet == candSubnet)
3677 (*a_sameNetOrSubnetP)++;
3681 (*a_diffNetworkP)++;
3683 } /*h_ClassifyAddress */
3686 /*------------------------------------------------------------------------
3687 * EXPORTED h_GetHostNetStats
3690 * Iterate through the host table, and classify each (non-deleted)
3691 * host entry into ``proximity'' categories (same net or subnet,
3692 * different subnet, different network).
3695 * a_numHostsP : Set to total number of (non-deleted) hosts.
3696 * a_sameNetOrSubnetP : Set to # hosts on same net/subnet as server.
3697 * a_diffSubnetP : Set to # hosts on diff subnet as server.
3698 * a_diffNetworkP : Set to # hosts on diff network as server.
3704 * We only count non-deleted hosts. The storage pointed to by our
3705 * parameters is zeroed upon entry.
3709 *------------------------------------------------------------------------*/
3712 h_GetHostNetStats(afs_int32 * a_numHostsP, afs_int32 * a_sameNetOrSubnetP,
3713 afs_int32 * a_diffSubnetP, afs_int32 * a_diffNetworkP)
3714 { /*h_GetHostNetStats */
3716 struct host *hostP; /*Ptr to current host entry */
3717 afs_uint32 currAddr_HBO; /*Curr host addr, host byte order */
3721 * Clear out the storage pointed to by our parameters.
3723 *a_numHostsP = (afs_int32) 0;
3724 *a_sameNetOrSubnetP = (afs_int32) 0;
3725 *a_diffSubnetP = (afs_int32) 0;
3726 *a_diffNetworkP = (afs_int32) 0;
3729 for (count = 0, hostP = hostList; hostP && count < hostCount; hostP = hostP->next, count++) {
3730 if (!(hostP->hostFlags & HOSTDELETED)) {
3732 * Bump the number of undeleted host entries found.
3733 * In classifying the current entry's address, make
3734 * sure to first convert to host byte order.
3737 currAddr_HBO = (afs_uint32) ntohl(hostP->host);
3738 h_ClassifyAddress(FS_HostAddr_HBO, currAddr_HBO,
3739 a_sameNetOrSubnetP, a_diffSubnetP,
3741 } /*Only look at non-deleted hosts */
3742 } /*For each host record hashed to this index */
3743 if (count != hostCount) {
3744 ViceLog(0, ("h_GetHostNetStats found %d of %d hosts\n", count, hostCount));
3745 } else if (hostP != NULL) {
3746 ViceLog(0, ("h_GetHostNetStats found more than %d hosts\n", hostCount));
3747 ShutDownAndCore(PANIC);
3750 } /*h_GetHostNetStats */
3752 static afs_uint32 checktime;
3753 static afs_uint32 clientdeletetime;
3754 static struct AFSFid zerofid;
3758 * XXXX: This routine could use Multi-Rx to avoid serializing the timeouts.
3759 * Since it can serialize them, and pile up, it should be a separate LWP
3760 * from other events.
3764 CheckHost(struct host *host, int flags, void *rock)
3766 struct client *client;
3767 struct rx_connection *cb_conn = NULL;
3770 #ifdef AFS_DEMAND_ATTACH_FS
3771 /* kill the checkhost lwp ASAP during shutdown */
3773 if (fs_state.mode == FS_MODE_SHUTDOWN) {
3775 return H_ENUMERATE_BAIL(flags);
3780 /* Host is held by h_Enumerate */
3782 for (client = host->FirstClient; client; client = client->next) {
3783 if (client->refCount == 0 && client->LastCall < clientdeletetime) {
3784 client->deleted = 1;
3785 host->hostFlags |= CLIENTDELETED;
3788 if (host->LastCall < checktime) {
3790 if (!(host->hostFlags & HOSTDELETED)) {
3791 host->hostFlags |= HWHO_INPROGRESS;
3792 cb_conn = host->callback_rxcon;
3793 rx_GetConnection(cb_conn);
3794 if (host->LastCall < clientdeletetime) {
3795 host->hostFlags |= HOSTDELETED;
3796 if (!(host->hostFlags & VENUSDOWN)) {
3797 host->hostFlags &= ~ALTADDR; /* alternate address invalid */
3798 if (host->interface) {
3801 RXAFSCB_InitCallBackState3(cb_conn,
3807 RXAFSCB_InitCallBackState(cb_conn);
3810 host->hostFlags |= ALTADDR; /* alternate addresses valid */
3813 (void)afs_inet_ntoa_r(host->host, hoststr);
3815 ("CB: RCallBackConnectBack (host.c) failed for host %s:%d\n",
3816 hoststr, ntohs(host->port)));
3817 host->hostFlags |= VENUSDOWN;
3819 /* Note: it's safe to delete hosts even if they have call
3820 * back state, because break delayed callbacks (called when a
3821 * message is received from the workstation) will always send a
3822 * break all call backs to the workstation if there is no
3827 if (!(host->hostFlags & VENUSDOWN) && host->cblist) {
3829 (void)afs_inet_ntoa_r(host->host, hoststr);
3830 if (host->interface) {
3831 afsUUID uuid = host->interface->uuid;
3833 code = RXAFSCB_ProbeUuid(cb_conn, &uuid);
3836 if (MultiProbeAlternateAddress_r(host)) {
3837 ViceLog(0,("CheckHost: Probing all interfaces of host %s:%d failed, code %d\n",
3838 hoststr, ntohs(host->port), code));
3839 host->hostFlags |= VENUSDOWN;
3844 code = RXAFSCB_Probe(cb_conn);
3848 ("CheckHost: Probe failed for host %s:%d, code %d\n",
3849 hoststr, ntohs(host->port), code));
3850 host->hostFlags |= VENUSDOWN;
3856 rx_PutConnection(cb_conn);
3859 host->hostFlags &= ~HWHO_INPROGRESS;
3870 CheckHost_r(struct host *host, void *dummy)
3872 struct client *client;
3873 struct rx_connection *cb_conn = NULL;
3876 #ifdef AFS_DEMAND_ATTACH_FS
3877 /* kill the checkhost lwp ASAP during shutdown */
3879 if (fs_state.mode == FS_MODE_SHUTDOWN) {
3881 return H_ENUMERATE_BAIL(0);
3886 /* Host is held by h_Enumerate_r */
3887 for (client = host->FirstClient; client; client = client->next) {
3888 if (client->refCount == 0 && client->LastCall < clientdeletetime) {
3889 client->deleted = 1;
3890 host->hostFlags |= CLIENTDELETED;
3893 if (host->LastCall < checktime) {
3895 if (!(host->hostFlags & HOSTDELETED)) {
3896 host->hostFlags |= HWHO_INPROGRESS;
3897 cb_conn = host->callback_rxcon;
3898 rx_GetConnection(cb_conn);
3899 if (host->LastCall < clientdeletetime) {
3900 host->hostFlags |= HOSTDELETED;
3901 if (!(host->hostFlags & VENUSDOWN)) {
3902 host->hostFlags &= ~ALTADDR; /* alternate address invalid */
3903 if (host->interface) {
3906 RXAFSCB_InitCallBackState3(cb_conn,
3912 RXAFSCB_InitCallBackState(cb_conn);
3915 host->hostFlags |= ALTADDR; /* alternate addresses valid */
3918 (void)afs_inet_ntoa_r(host->host, hoststr);
3920 ("CB: RCallBackConnectBack (host.c) failed for host %s:%d\n",
3921 hoststr, ntohs(host->port)));
3922 host->hostFlags |= VENUSDOWN;
3924 /* Note: it's safe to delete hosts even if they have call
3925 * back state, because break delayed callbacks (called when a
3926 * message is received from the workstation) will always send a
3927 * break all call backs to the workstation if there is no
3932 if (!(host->hostFlags & VENUSDOWN) && host->cblist) {
3934 (void)afs_inet_ntoa_r(host->host, hoststr);
3935 if (host->interface) {
3936 afsUUID uuid = host->interface->uuid;
3938 code = RXAFSCB_ProbeUuid(cb_conn, &uuid);
3941 if (MultiProbeAlternateAddress_r(host)) {
3942 ViceLog(0,("CheckHost_r: Probing all interfaces of host %s:%d failed, code %d\n",
3943 hoststr, ntohs(host->port), code));
3944 host->hostFlags |= VENUSDOWN;
3949 code = RXAFSCB_Probe(cb_conn);
3953 ("CheckHost_r: Probe failed for host %s:%d, code %d\n",
3954 hoststr, ntohs(host->port), code));
3955 host->hostFlags |= VENUSDOWN;
3961 rx_PutConnection(cb_conn);
3964 host->hostFlags &= ~HWHO_INPROGRESS;
3974 * Set VenusDown for any hosts that have not had a call in 15 minutes and
3975 * don't respond to a probe. Note that VenusDown can only be cleared if
3976 * a message is received from the host (see ServerLWP in file.c).
3977 * Delete hosts that have not had any calls in 1 hour, clients that
3978 * have not had any calls in 15 minutes.
3980 * This routine is called roughly every 5 minutes.
3985 afs_uint32 now = time(NULL);
3987 memset(&zerofid, 0, sizeof(zerofid));
3989 * Send a probe to the workstation if it hasn't been heard from in
3992 checktime = now - 15 * 60;
3993 clientdeletetime = now - 120 * 60; /* 2 hours ago */
3996 h_Enumerate_r(CheckHost_r, hostList, NULL);
4001 * This is called with host locked and held. At this point, the
4002 * hostAddrHashTable has an entry for the primary addr/port inserted
4003 * by h_Alloc_r(). No other interfaces should be considered valid.
4005 * The addresses in the interfaceAddr list are in host byte order.
4008 initInterfaceAddr_r(struct host *host, struct interfaceAddr *interf)
4015 struct Interface *interface;
4018 afs_uint16 port7001 = htons(7001);
4023 number = interf->numberOfInterfaces;
4024 myAddr = host->host; /* current interface address */
4025 myPort = host->port; /* current port */
4028 ("initInterfaceAddr : host %s:%d numAddr %d\n",
4029 afs_inet_ntoa_r(myAddr, hoststr), ntohs(myPort), number));
4031 /* validation checks */
4032 if (number < 0 || number > AFS_MAX_INTERFACE_ADDR) {
4034 ("Invalid number of alternate addresses is %d\n", number));
4039 * The client's notion of its own IP addresses is not reliable.
4041 * 1. The client list might contain private address ranges which
4042 * are likely to be re-used by many clients allocated addresses
4045 * 2. The client list will not include any public addresses that
4046 * are hidden by a NAT.
4048 * 3. Private address ranges that are exposed to the server will
4049 * be obtained from the rx connections that use them.
4051 * 4. Lists provided by the client are not necessarily truthful.
4052 * Many existing clients (UNIX) do not refresh the IP address
4053 * list as the actual assigned addresses change. The end result
4054 * is that they report the initial address list for the lifetime
4055 * of the process. In other words, a client can report addresses
4056 * that they are in fact not using. Adding these addresses to
4057 * the host interface list without verification is not only
4058 * pointless, it is downright dangerous.
4060 * We therefore do not add alternate addresses to the addr hash table.
4061 * We only use them for multi-rx callback breaks.
4065 * Convert IP addresses to network byte order, and remove
4066 * duplicate and loopback IP addresses from the interface list, and
4067 * determine whether or not the incoming addr/port is
4068 * listed. Note that if the address matches it is not
4069 * truly a match because the port number for the entries
4070 * in the interface list are port 7001 and the port number
4071 * for this connection might not be 7001.
4073 for (i = 0, count = 0, found = 0; i < number; i++) {
4074 if (rx_IsLoopbackAddr(interf->addr_in[i])) {
4077 interf->addr_in[i] = htonl(interf->addr_in[i]);
4078 for (j = 0; j < count; j++) {
4079 if (interf->addr_in[j] == interf->addr_in[i])
4083 interf->addr_in[count] = interf->addr_in[i];
4084 if (interf->addr_in[count] == myAddr &&
4092 * Allocate and initialize an interface structure for this host.
4095 interface = malloc(sizeof(struct Interface) +
4096 (sizeof(struct AddrPort) * (count - 1)));
4098 ViceLogThenPanic(0, ("Failed malloc in initInterfaceAddr_r 1\n"));
4100 interface->numberOfInterfaces = count;
4102 interface = malloc(sizeof(struct Interface) +
4103 (sizeof(struct AddrPort) * count));
4105 ViceLogThenPanic(0, ("Failed malloc in initInterfaceAddr_r 2\n"));
4107 interface->numberOfInterfaces = count + 1;
4108 interface->interface[count].addr = myAddr;
4109 interface->interface[count].port = myPort;
4110 interface->interface[count].valid = 1;
4113 for (i = 0; i < count; i++) {
4115 interface->interface[i].addr = interf->addr_in[i];
4116 /* We store the port as 7001 because the addresses reported by
4117 * TellMeAboutYourself and WhoAreYou RPCs are only valid if they
4118 * are coming from fully connected hosts (no NAT/PATs)
4120 interface->interface[i].port = port7001;
4121 interface->interface[i].valid =
4122 (interf->addr_in[i] == myAddr && port7001 == myPort) ? 1 : 0;
4125 interface->uuid = interf->uuid;
4127 opr_Assert(!host->interface);
4128 host->interface = interface;
4130 if (LogLevel >= 125) {
4131 afsUUID_to_string(&interface->uuid, uuidstr, 127);
4133 ViceLog(125, ("--- uuid %s\n", uuidstr));
4134 for (i = 0; i < host->interface->numberOfInterfaces; i++) {
4135 ViceLog(125, ("--- alt address %s:%d\n",
4136 afs_inet_ntoa_r(host->interface->interface[i].addr, hoststr),
4137 ntohs(host->interface->interface[i].port)));
4144 /* deleted a HashChain structure for this address and host */
4145 /* returns 1 on success */
4147 h_DeleteHostFromAddrHashTable_r(afs_uint32 addr, afs_uint16 port,
4151 struct h_AddrHashChain **hp, *th;
4153 if (addr == 0 && port == 0)
4156 for (hp = &hostAddrHashTable[h_HashIndex(addr)]; (th = *hp);
4158 opr_Assert(th->hostPtr);
4159 if (th->hostPtr == host && th->addr == addr && th->port == port) {
4160 ViceLog(125, ("h_DeleteHostFromAddrHashTable_r: host %" AFS_PTR_FMT " (%s:%d)\n",
4161 host, afs_inet_ntoa_r(host->host, hoststr),
4162 ntohs(host->port)));
4169 ("h_DeleteHostFromAddrHashTable_r: host %" AFS_PTR_FMT " (%s:%d) not found\n",
4170 host, afs_inet_ntoa_r(host->host, hoststr),
4171 ntohs(host->port)));
4177 ** prints out all alternate interface address for the host. The 'level'
4178 ** parameter indicates what level of debugging sets this output
4181 printInterfaceAddr(struct host *host, int level)
4186 if (host->interface) {
4187 /* check alternate addresses */
4188 number = host->interface->numberOfInterfaces;
4190 ViceLog(level, ("no-addresses "));
4192 for (i = 0; i < number; i++)
4193 ViceLog(level, ("%s:%d ", afs_inet_ntoa_r(host->interface->interface[i].addr, hoststr),
4194 ntohs(host->interface->interface[i].port)));
4197 ViceLog(level, ("\n"));