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
13 #include <afsconfig.h>
14 #include "afs/param.h"
18 #include "afs/sysincludes.h" /* Standard vendor system headers */
21 #if !defined(AFS_LINUX20_ENV) && !defined(AFS_FBSD_ENV)
23 #include <netinet/in.h>
27 #include "h/hashing.h"
29 #if !defined(AFS_HPUX110_ENV) && !defined(AFS_LINUX20_ENV) && !defined(AFS_FBSD_ENV) && !defined(AFS_DARWIN_ENV)
30 #include <netinet/in_var.h>
34 #include "afsincludes.h" /* Afs-based standard headers */
35 #include "afs/afs_stats.h" /* afs statistics */
36 #include "afs/afs_util.h"
37 #include "afs/unified_afs.h"
39 #if defined(AFS_SUN5_ENV)
41 #include <inet/common.h>
42 #include <netinet/ip6.h>
46 /* shouldn't do it this way, but for now will do */
47 #ifndef ERROR_TABLE_BASE_U
48 #define ERROR_TABLE_BASE_U (5376L)
49 #endif /* ubik error base define */
51 /* shouldn't do it this way, but for now will do */
52 #ifndef ERROR_TABLE_BASE_uae
53 #define ERROR_TABLE_BASE_uae (49733376L)
54 #endif /* unified afs error base define */
56 /* same hack for vlserver error base as for ubik error base */
57 #ifndef ERROR_TABLE_BASE_VL
58 #define ERROR_TABLE_BASE_VL (363520L)
59 #define VL_NOENT (363524L)
60 #endif /* vlserver error base define */
63 int afs_BusyWaitPeriod = 15; /* poll every 15 seconds */
65 afs_int32 hm_retry_RO = 0; /* don't wait */
66 afs_int32 hm_retry_RW = 0; /* don't wait */
67 afs_int32 hm_retry_int = 0; /* don't wait */
69 #define VSleep(at) afs_osi_Wait((at)*1000, 0, 0)
74 * 0 if the vldb record for a specific volume is different from what
75 * we have cached -- perhaps the volume has moved.
76 * 1 if the vldb record is the same
77 * 2 if we can't tell if it's the same or not.
79 * If 0, the caller will probably start over at the beginning of our
80 * list of servers for this volume and try to find one that is up. If
81 * not 0, we will probably just keep plugging with what we have
82 * cached. If we fail to contact the VL server, we should just keep
83 * trying with the information we have, rather than failing. */
88 VLDB_Same(struct VenusFid *afid, struct vrequest *areq)
91 struct afs_conn *tconn;
95 struct nvldbentry ntve;
96 struct uvldbentry utve;
100 char *bp, tbuf[CVBS]; /* biggest volume id is 2^32, ~ 4*10^9 */
101 unsigned int changed;
102 struct server *(oldhosts[NMAXNSERVERS]);
103 struct rx_connection *rxconn;
105 AFS_STATCNT(CheckVLDB);
106 afs_FinalizeReq(areq);
108 if ((i = afs_InitReq(&treq, afs_osi_credp)))
110 v = afs_osi_Alloc(sizeof(*v));
111 osi_Assert(v != NULL);
112 tcell = afs_GetCell(afid->Cell, READ_LOCK);
113 bp = afs_cv2string(&tbuf[CVBS], afid->Fid.Volume);
115 VSleep(2); /* Better safe than sorry. */
117 afs_ConnByMHosts(tcell->cellHosts, tcell->vlport, tcell->cellNum,
118 &treq, SHARED_LOCK, 0, &rxconn);
120 if ( tconn->parent->srvr->server->flags & SNO_LHOSTS) {
123 i = VL_GetEntryByNameO(rxconn, bp, &v->tve);
125 } else if (tconn->parent->srvr->server->flags & SYES_LHOSTS) {
128 i = VL_GetEntryByNameN(rxconn, bp, &v->ntve);
133 i = VL_GetEntryByNameU(rxconn, bp, &v->utve);
135 if (!(tconn->parent->srvr->server->flags & SVLSRV_UUID)) {
136 if (i == RXGEN_OPCODE) {
139 i = VL_GetEntryByNameN(rxconn, bp, &v->ntve);
141 if (i == RXGEN_OPCODE) {
143 tconn->parent->srvr->server->flags |= SNO_LHOSTS;
145 i = VL_GetEntryByNameO(rxconn, bp, &v->tve);
148 tconn->parent->srvr->server->flags |= SYES_LHOSTS;
150 tconn->parent->srvr->server->flags |= SVLSRV_UUID;
156 } while (afs_Analyze(tconn, rxconn, i, NULL, &treq, -1, /* no op code for this */
157 SHARED_LOCK, tcell));
159 afs_PutCell(tcell, READ_LOCK);
160 afs_Trace2(afs_iclSetp, CM_TRACE_CHECKVLDB, ICL_TYPE_FID, &afid,
164 afs_osi_Free(v, sizeof(*v));
167 /* have info, copy into serverHost array */
169 tvp = afs_FindVolume(afid, WRITE_LOCK);
171 ObtainWriteLock(&tvp->lock, 107);
172 for (i = 0; i < NMAXNSERVERS && tvp->serverHost[i]; i++) {
173 oldhosts[i] = tvp->serverHost[i];
175 ReleaseWriteLock(&tvp->lock);
178 LockAndInstallUVolumeEntry(tvp, &v->utve, afid->Cell, tcell, &treq);
179 } else if (type == 1) {
180 LockAndInstallNVolumeEntry(tvp, &v->ntve, afid->Cell);
182 LockAndInstallVolumeEntry(tvp, &v->tve, afid->Cell);
185 if (i < NMAXNSERVERS && tvp->serverHost[i]) {
188 for (--i; !changed && i >= 0; i--) {
189 if (tvp->serverHost[i] != oldhosts[i]) {
190 changed = 1; /* also happens if prefs change. big deal. */
194 ReleaseWriteLock(&tvp->lock);
195 afs_PutVolume(tvp, WRITE_LOCK);
196 } else { /* can't find volume */
197 tvp = afs_GetVolume(afid, &treq, WRITE_LOCK);
199 afs_PutVolume(tvp, WRITE_LOCK);
200 afs_osi_Free(v, sizeof(*v));
203 afs_osi_Free(v, sizeof(*v));
208 afs_osi_Free(v, sizeof(*v));
209 return (changed ? DIFFERENT : SAME);
212 /*------------------------------------------------------------------------
216 * Mark a server as invalid for further attempts of this request only.
219 * areq : The request record associated with this operation.
220 * afid : The FID of the file involved in the action. This argument
221 * may be null if none was involved.
222 * tsp : pointer to a server struct for the server we wish to
226 * Non-zero value if further servers are available to try,
230 * This routine is typically called in situations where we believe
231 * one server out of a pool may have an error condition.
237 * The afs_Conn* routines use the list of invalidated servers to
238 * avoid reusing a server marked as invalid for this request.
239 *------------------------------------------------------------------------*/
241 afs_BlackListOnce(struct vrequest *areq, struct VenusFid *afid,
246 afs_int32 serversleft = 0;
249 tvp = afs_FindVolume(afid, READ_LOCK);
251 for (i = 0; i < AFS_MAXHOSTS; i++) {
252 if (tvp->serverHost[i] == tsp) {
253 areq->skipserver[i] = 1;
255 if (tvp->serverHost[i] &&
256 (tvp->serverHost[i]->addr->sa_flags &
258 areq->skipserver[i] = 1;
261 for (i = 0; i < AFS_MAXHOSTS; i++) {
262 if (tvp->serverHost[i] && areq->skipserver[i] == 0) {
267 afs_PutVolume(tvp, READ_LOCK);
274 /*------------------------------------------------------------------------
278 * Analyze the outcome of an RPC operation, taking whatever support
279 * actions are necessary.
282 * afid : The FID of the file involved in the action. This argument
283 * may be null if none was involved.
284 * op : which RPC we are analyzing.
285 * avp : A pointer to the struct volume, if we already have one.
288 * Non-zero value if the related RPC operation can be retried,
292 * This routine is called when we got a network error,
293 * and discards state if the operation was a data-mutating
295 *------------------------------------------------------------------------*/
297 afs_ClearStatus(struct VenusFid *afid, int op, struct volume *avp)
299 struct volume *tvp = NULL;
301 /* if it's not a write op, we have nothing to veto and shouldn't clear. */
302 if (!AFS_STATS_FS_RPCIDXES_ISWRITE(op)) {
309 tvp = afs_FindVolume(afid, READ_LOCK);
311 /* don't assume just discarding will fix if no cached volume */
314 ObtainReadLock(&afs_xvcache);
315 if ((tvc = afs_FindVCache(afid, 0, 0))) {
316 ReleaseReadLock(&afs_xvcache);
317 tvc->f.states &= ~(CStatd | CUnique);
320 ReleaseReadLock(&afs_xvcache);
323 afs_PutVolume(tvp, READ_LOCK);
326 if (AFS_STATS_FS_RPCIDXES_WRITE_RETRIABLE(op))
329 /* not retriable: we may have raced ourselves */
333 /*------------------------------------------------------------------------
334 * EXPORTED afs_Analyze
337 * Analyze the outcome of an RPC operation, taking whatever support
338 * actions are necessary.
341 * aconn : Ptr to the relevant connection on which the call was made.
342 * acode : The return code experienced by the RPC.
343 * afid : The FID of the file involved in the action. This argument
344 * may be null if none was involved.
345 * areq : The request record associated with this operation.
346 * op : which RPC we are analyzing.
347 * cellp : pointer to a cell struct. Must provide either fid or cell.
350 * Non-zero value if the related RPC operation should be retried,
354 * This routine is typically called in a do-while loop, causing the
355 * embedded RPC operation to be called repeatedly if appropriate
356 * until whatever error condition (if any) is intolerable.
362 * The retry return value is used by afs_StoreAllSegments to determine
363 * if this is a temporary or permanent error.
364 *------------------------------------------------------------------------*/
366 afs_Analyze(struct afs_conn *aconn, struct rx_connection *rxconn,
367 afs_int32 acode, struct VenusFid *afid, struct vrequest *areq,
368 int op, afs_int32 locktype, struct cell *cellp)
373 struct volume *tvp = NULL;
374 afs_int32 shouldRetry = 0;
375 afs_int32 serversleft = 1;
376 struct afs_stats_RPCErrors *aerrP;
379 if (AFS_IS_DISCONNECTED && !AFS_IN_SYNC) {
380 /* On reconnection, act as connected. XXX: for now.... */
381 /* SXW - This may get very tired after a while. We should try and
382 * intercept all RPCs before they get here ... */
383 /*printf("afs_Analyze: disconnected\n");*/
384 afs_FinalizeReq(areq);
386 /* SXW - I suspect that this will _never_ happen - we shouldn't
387 * get a connection because we're disconnected !!!*/
388 afs_PutConn(aconn, rxconn, locktype);
393 AFS_STATCNT(afs_Analyze);
394 afs_Trace4(afs_iclSetp, CM_TRACE_ANALYZE, ICL_TYPE_INT32, op,
395 ICL_TYPE_POINTER, aconn, ICL_TYPE_INT32, acode, ICL_TYPE_LONG,
398 aerrP = (struct afs_stats_RPCErrors *)0;
400 if ((op >= 0) && (op < AFS_STATS_NUM_FS_RPC_OPS))
401 aerrP = &(afs_stats_cmfullperf.rpc.fsRPCErrors[op]);
403 afs_FinalizeReq(areq);
404 if (!aconn && areq->busyCount) { /* one RPC or more got VBUSY/VRESTARTING */
406 tvp = afs_FindVolume(afid, READ_LOCK);
408 afs_warnuser("afs: Waiting for busy volume %u (%s) in cell %s\n",
409 (afid ? afid->Fid.Volume : 0),
410 (tvp->name ? tvp->name : ""),
412 && tvp->serverHost[0]->cell) ? tvp->serverHost[0]->
413 cell->cellName : ""));
415 for (i = 0; i < AFS_MAXHOSTS; i++) {
416 if (tvp->status[i] != not_busy && tvp->status[i] != offline) {
417 tvp->status[i] = not_busy;
419 if (tvp->status[i] == not_busy)
422 afs_PutVolume(tvp, READ_LOCK);
424 afs_warnuser("afs: Waiting for busy volume %u\n",
425 (afid ? afid->Fid.Volume : 0));
428 if (areq->busyCount > 100) {
430 (aerrP->err_Volume)++;
431 areq->volumeError = VOLBUSY;
434 VSleep(afs_BusyWaitPeriod); /* poll periodically */
436 if (shouldRetry != 0)
439 return shouldRetry; /* should retry */
442 if (!aconn || !aconn->parent->srvr) {
443 if (!areq->volumeError) {
445 (aerrP->err_Network)++;
446 if (hm_retry_int && !(areq->flags & O_NONBLOCK) && /* "hard" mount */
447 ((afid && afs_IsPrimaryCellNum(afid->Cell))
448 || (cellp && afs_IsPrimaryCell(cellp)))) {
450 static int afs_vl_hm = 0;
453 afs_vl_hm = warn = 1;
457 ("afs: hard-mount waiting for a vlserver to return to service\n");
459 VSleep(hm_retry_int);
460 afs_CheckServers(1, cellp);
467 static int afs_unknown_vhm = 0;
468 int warn = 0, vp_vhm = 0;
470 tvp = afs_FindVolume(afid, READ_LOCK);
471 if (!tvp || (tvp->states & VRO)) {
472 shouldRetry = hm_retry_RO;
474 shouldRetry = hm_retry_RW;
477 /* Set 'warn' if we should afs_warnuser. Only let one
478 * caller call afs_warnuser per hm_retry_int interval per
482 if (!(tvp->states & VHardMount)) {
483 tvp->states |= VHardMount;
487 if (!afs_unknown_vhm) {
495 afs_PutVolume(tvp, READ_LOCK);
500 ("afs: hard-mount waiting for volume %u\n",
504 VSleep(hm_retry_int);
505 afs_CheckServers(1, cellp);
506 /* clear the black listed servers on this request. */
507 memset(areq->skipserver, 0, sizeof(areq->skipserver));
510 tvp = afs_FindVolume(afid, READ_LOCK);
512 tvp->states &= ~VHardMount;
513 afs_PutVolume(tvp, READ_LOCK);
520 } /* if (hm_retry_int ... */
522 if (acode == RX_MSGSIZE)
525 areq->networkError = 1;
526 /* do not promote to shouldRetry if not already */
527 if (afs_ClearStatus(afid, op, NULL) == 0)
535 /* Find server associated with this connection. */
536 sa = aconn->parent->srvr;
538 address = ntohl(sa->sa_ip);
540 /* Before we do anything with acode, make sure we translate it back to
542 if ((acode & ~0xff) == ERROR_TABLE_BASE_uae)
543 acode = et_to_sys_error(acode);
546 /* If we previously took an error, mark this volume not busy */
547 if (areq->volumeError) {
548 tvp = afs_FindVolume(afid, READ_LOCK);
550 for (i = 0; i < AFS_MAXHOSTS; i++) {
551 if (tvp->serverHost[i] == tsp) {
552 tvp->status[i] = not_busy;
555 afs_PutVolume(tvp, READ_LOCK);
559 afs_PutConn(aconn, rxconn, locktype);
563 /* If network troubles, mark server as having bogued out again. */
564 /* VRESTARTING is < 0 because of backward compatibility issues
565 * with 3.4 file servers and older cache managers */
566 #ifdef AFS_64BIT_CLIENT
569 #endif /* AFS_64BIT_CLIENT */
570 if ((acode < 0) && (acode != VRESTARTING)) {
571 if (acode == RX_MSGSIZE) {
575 if (acode == RX_CALL_TIMEOUT || acode == RX_CALL_IDLE || acode == RX_CALL_BUSY) {
576 serversleft = afs_BlackListOnce(areq, afid, tsp);
578 tvp = afs_FindVolume(afid, READ_LOCK);
579 if ((serversleft == 0) && tvp &&
580 ((tvp->states & VRO) || (tvp->states & VBackup))) {
585 if (!afid || !tvp || (tvp->states & VRO))
587 else if (afs_ClearStatus(afid, op, tvp) == 0)
591 afs_PutVolume(tvp, READ_LOCK);
592 /* By doing this, we avoid ever marking a server down
593 * in an idle timeout case. That's because the server is
594 * still responding and may only be letting a single vnode
595 * time out. We otherwise risk having the server continually
596 * be marked down, then up, then down again...
600 afs_ServerDown(sa, acode);
601 ForceNewConnections(sa); /**multi homed clients lock:afs_xsrvAddr? */
603 (aerrP->err_Server)++;
606 if (acode == VBUSY || acode == VRESTARTING) {
607 if (acode == VBUSY) {
610 (aerrP->err_VolumeBusies)++;
614 tvp = afs_FindVolume(afid, READ_LOCK);
616 for (i = 0; i < AFS_MAXHOSTS; i++) {
617 if (tvp->serverHost[i] == tsp) {
618 tvp->status[i] = rdwr_busy; /* can't tell which yet */
619 /* to tell which, have to look at the op code. */
622 afs_PutVolume(tvp, READ_LOCK);
624 afs_warnuser("afs: Waiting for busy volume %u in cell %s (server %d.%d.%d.%d)\n",
625 (afid ? afid->Fid.Volume : 0), tsp->cell->cellName,
626 (address >> 24), (address >> 16) & 0xff,
627 (address >> 8) & 0xff, (address) & 0xff);
628 VSleep(afs_BusyWaitPeriod); /* poll periodically */
632 } else if (acode == VICETOKENDEAD
633 || (acode & ~0xff) == ERROR_TABLE_BASE_RXK) {
634 /* any rxkad error is treated as token expiration */
637 * I'm calling these errors protection errors, since they involve
638 * faulty authentication.
641 (aerrP->err_Protection)++;
643 tu = afs_FindUser(areq->uid, tsp->cell->cellNum, READ_LOCK);
645 if (acode == VICETOKENDEAD) {
646 aconn->forceConnectFS = 1;
647 } else if (acode == RXKADEXPIRED) {
648 aconn->forceConnectFS = 0; /* don't check until new tokens set */
649 aconn->parent->user->states |= UTokensBad;
650 afs_NotifyUser(tu, UTokensDropped);
652 ("afs: Tokens for user of AFS id %d for cell %s have expired (server %d.%d.%d.%d)\n",
653 tu->viceId, aconn->parent->srvr->server->cell->cellName,
654 (address >> 24), (address >> 16) & 0xff,
655 (address >> 8) & 0xff, (address) & 0xff);
657 serversleft = afs_BlackListOnce(areq, afid, tsp);
662 ("afs: Tokens for user of AFS id %d for cell %s: rxkad error=%d (server %d.%d.%d.%d)\n",
663 tu->viceId, aconn->parent->srvr->server->cell->cellName, acode,
664 (address >> 24), (address >> 16) & 0xff,
665 (address >> 8) & 0xff, (address) & 0xff);
668 areq->tokenError = 0;
669 aconn->forceConnectFS = 0; /* don't check until new tokens set */
670 aconn->parent->user->states |= UTokensBad;
671 afs_NotifyUser(tu, UTokensDropped);
673 ("afs: Tokens for user of AFS id %d for cell %s are discarded (rxkad error=%d, server %d.%d.%d.%d)\n",
674 tu->viceId, aconn->parent->srvr->server->cell->cellName, acode,
675 (address >> 24), (address >> 16) & 0xff,
676 (address >> 8) & 0xff, (address) & 0xff);
679 afs_PutUser(tu, READ_LOCK);
681 /* The else case shouldn't be possible and should probably be replaced by a panic? */
682 if (acode == VICETOKENDEAD) {
683 aconn->forceConnectFS = 1;
684 } else if (acode == RXKADEXPIRED) {
685 aconn->forceConnectFS = 0; /* don't check until new tokens set */
686 aconn->parent->user->states |= UTokensBad;
687 afs_NotifyUser(tu, UTokensDropped);
689 ("afs: Tokens for user %d for cell %s have expired (server %d.%d.%d.%d)\n",
690 areq->uid, aconn->parent->srvr->server->cell->cellName,
691 (address >> 24), (address >> 16) & 0xff,
692 (address >> 8) & 0xff, (address) & 0xff);
694 aconn->forceConnectFS = 0; /* don't check until new tokens set */
695 aconn->parent->user->states |= UTokensBad;
696 afs_NotifyUser(tu, UTokensDropped);
698 ("afs: Tokens for user %d for cell %s are discarded (rxkad error = %d, server %d.%d.%d.%d)\n",
699 areq->uid, aconn->parent->srvr->server->cell->cellName,
701 (address >> 24), (address >> 16) & 0xff,
702 (address >> 8) & 0xff, (address) & 0xff);
706 shouldRetry = 1; /* Try again (as root). */
708 /* Check for access violation. */
709 else if (acode == EACCES) {
710 /* should mark access error in non-existent per-user global structure */
712 (aerrP->err_Protection)++;
713 areq->accessError = 1;
714 if (op == AFS_STATS_FS_RPCIDX_STOREDATA)
715 areq->permWriteError = 1;
718 /* check for ubik errors; treat them like crashed servers */
719 else if (acode >= ERROR_TABLE_BASE_U && acode < ERROR_TABLE_BASE_U + 255) {
720 afs_ServerDown(sa, acode);
722 (aerrP->err_Server)++;
723 shouldRetry = 1; /* retryable (maybe one is working) */
724 VSleep(1); /* just in case */
726 /* Check for bad volume data base / missing volume. */
727 else if (acode == VSALVAGE || acode == VOFFLINE || acode == VNOVOL
728 || acode == VNOSERVICE || acode == VMOVED) {
733 areq->volumeError = VOLMISSING;
735 (aerrP->err_Volume)++;
736 if (afid && (tcell = afs_GetCell(afid->Cell, 0))) {
737 same = VLDB_Same(afid, areq);
738 tvp = afs_FindVolume(afid, READ_LOCK);
740 for (i = 0; i < AFS_MAXHOSTS && tvp->serverHost[i]; i++) {
741 if (tvp->serverHost[i] == tsp) {
742 if (tvp->status[i] == end_not_busy)
743 tvp->status[i] = offline;
747 tvp->status[i] = not_busy; /* reset the others */
750 afs_PutVolume(tvp, READ_LOCK);
753 } else if (acode >= ERROR_TABLE_BASE_VL && acode <= ERROR_TABLE_BASE_VL + 255) { /* vlserver errors */
755 areq->volumeError = VOLMISSING;
756 } else if (acode >= 0) {
758 (aerrP->err_Other)++;
759 if (op == AFS_STATS_FS_RPCIDX_STOREDATA)
760 areq->permWriteError = 1;
761 shouldRetry = 0; /* Other random Vice error. */
762 } else if (acode == RX_MSGSIZE) { /* same meaning as EMSGSIZE... */
764 ("afs: Path MTU may have been exceeded, retrying (server %d.%d.%d.%d)\n",
765 (address >> 24), (address >> 16) & 0xff,
766 (address >> 8) & 0xff, (address) & 0xff);
768 VSleep(1); /* Just a hack for desperate times. */
770 (aerrP->err_Other)++;
771 shouldRetry = 1; /* packet was too big, please retry call */
774 if (acode < 0 && acode != RX_MSGSIZE && acode != VRESTARTING) {
775 /* If we get here, code < 0 and we have network/Server troubles.
776 * areq->networkError is not set here, since we always
777 * retry in case there is another server. However, if we find
778 * no connection (aconn == 0) we set the networkError flag.
780 afs_MarkServerUpOrDown(sa, SRVR_ISDOWN);
782 (aerrP->err_Server)++;
783 VSleep(1); /* Just a hack for desperate times. */
787 /* now unlock the connection and return */
788 afs_PutConn(aconn, rxconn, locktype);
789 return (shouldRetry);