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"
20 #include "afs/sysincludes.h" /* Standard vendor system headers */
23 #if !defined(AFS_LINUX20_ENV) && !defined(AFS_FBSD_ENV)
25 #include <netinet/in.h>
29 #include "h/hashing.h"
31 #if !defined(AFS_HPUX110_ENV) && !defined(AFS_LINUX20_ENV) && !defined(AFS_FBSD_ENV) && !defined(AFS_DARWIN60_ENV)
32 #include <netinet/in_var.h>
36 #include "afsincludes.h" /* Afs-based standard headers */
37 #include "afs/afs_stats.h" /* afs statistics */
38 #include "afs/afs_util.h"
39 #include "afs/unified_afs.h"
41 #if defined(AFS_SUN56_ENV)
43 #include <inet/common.h>
44 #if defined(AFS_SUN58_ENV)
45 #include <netinet/ip6.h>
51 /* shouldn't do it this way, but for now will do */
52 #ifndef ERROR_TABLE_BASE_U
53 #define ERROR_TABLE_BASE_U (5376L)
54 #endif /* ubik error base define */
56 /* shouldn't do it this way, but for now will do */
57 #ifndef ERROR_TABLE_BASE_uae
58 #define ERROR_TABLE_BASE_uae (49733376L)
59 #endif /* unified afs error base define */
61 /* same hack for vlserver error base as for ubik error base */
62 #ifndef ERROR_TABLE_BASE_VL
63 #define ERROR_TABLE_BASE_VL (363520L)
64 #define VL_NOENT (363524L)
65 #endif /* vlserver error base define */
68 int afs_BusyWaitPeriod = 15; /* poll every 15 seconds */
70 afs_int32 hm_retry_RO = 0; /* don't wait */
71 afs_int32 hm_retry_RW = 0; /* don't wait */
72 afs_int32 hm_retry_int = 0; /* don't wait */
74 #define VSleep(at) afs_osi_Wait((at)*1000, 0, 0)
79 * 0 if the vldb record for a specific volume is different from what
80 * we have cached -- perhaps the volume has moved.
81 * 1 if the vldb record is the same
82 * 2 if we can't tell if it's the same or not.
84 * If 0, the caller will probably start over at the beginning of our
85 * list of servers for this volume and try to find one that is up. If
86 * not 0, we will probably just keep plugging with what we have
87 * cached. If we fail to contact the VL server, we should just keep
88 * trying with the information we have, rather than failing. */
93 VLDB_Same(struct VenusFid *afid, struct vrequest *areq)
96 struct afs_conn *tconn;
100 struct nvldbentry ntve;
101 struct uvldbentry utve;
105 char *bp, tbuf[CVBS]; /* biggest volume id is 2^32, ~ 4*10^9 */
106 unsigned int changed;
107 struct server *(oldhosts[NMAXNSERVERS]);
109 AFS_STATCNT(CheckVLDB);
110 afs_FinalizeReq(areq);
112 if ((i = afs_InitReq(&treq, afs_osi_credp)))
114 v = afs_osi_Alloc(sizeof(*v));
115 tcell = afs_GetCell(afid->Cell, READ_LOCK);
116 bp = afs_cv2string(&tbuf[CVBS], afid->Fid.Volume);
118 VSleep(2); /* Better safe than sorry. */
120 afs_ConnByMHosts(tcell->cellHosts, tcell->vlport, tcell->cellNum,
123 if (tconn->srvr->server->flags & SNO_LHOSTS) {
126 i = VL_GetEntryByNameO(tconn->id, bp, &v->tve);
128 } else if (tconn->srvr->server->flags & SYES_LHOSTS) {
131 i = VL_GetEntryByNameN(tconn->id, bp, &v->ntve);
136 i = VL_GetEntryByNameU(tconn->id, bp, &v->utve);
138 if (!(tconn->srvr->server->flags & SVLSRV_UUID)) {
139 if (i == RXGEN_OPCODE) {
142 i = VL_GetEntryByNameN(tconn->id, bp, &v->ntve);
144 if (i == RXGEN_OPCODE) {
146 tconn->srvr->server->flags |= SNO_LHOSTS;
148 i = VL_GetEntryByNameO(tconn->id, bp, &v->tve);
151 tconn->srvr->server->flags |= SYES_LHOSTS;
153 tconn->srvr->server->flags |= SVLSRV_UUID;
159 } while (afs_Analyze(tconn, i, NULL, &treq, -1, /* no op code for this */
160 SHARED_LOCK, tcell));
162 afs_PutCell(tcell, READ_LOCK);
163 afs_Trace2(afs_iclSetp, CM_TRACE_CHECKVLDB, ICL_TYPE_FID, &afid,
167 afs_osi_Free(v, sizeof(*v));
170 /* have info, copy into serverHost array */
172 tvp = afs_FindVolume(afid, WRITE_LOCK);
174 ObtainWriteLock(&tvp->lock, 107);
175 for (i = 0; i < NMAXNSERVERS && tvp->serverHost[i]; i++) {
176 oldhosts[i] = tvp->serverHost[i];
180 InstallUVolumeEntry(tvp, &v->utve, afid->Cell, tcell, &treq);
181 } else if (type == 1) {
182 InstallNVolumeEntry(tvp, &v->ntve, afid->Cell);
184 InstallVolumeEntry(tvp, &v->tve, afid->Cell);
187 if (i < NMAXNSERVERS && tvp->serverHost[i]) {
190 for (--i; !changed && i >= 0; i--) {
191 if (tvp->serverHost[i] != oldhosts[i]) {
192 changed = 1; /* also happens if prefs change. big deal. */
196 ReleaseWriteLock(&tvp->lock);
197 afs_PutVolume(tvp, WRITE_LOCK);
198 } else { /* can't find volume */
199 tvp = afs_GetVolume(afid, &treq, WRITE_LOCK);
201 afs_PutVolume(tvp, WRITE_LOCK);
202 afs_osi_Free(v, sizeof(*v));
205 afs_osi_Free(v, sizeof(*v));
210 afs_osi_Free(v, sizeof(*v));
211 return (changed ? DIFFERENT : SAME);
214 /*------------------------------------------------------------------------
218 * Mark a server as invalid for further attempts of this request only.
221 * areq : The request record associated with this operation.
222 * afid : The FID of the file involved in the action. This argument
223 * may be null if none was involved.
224 * tsp : pointer to a server struct for the server we wish to
228 * Non-zero value if further servers are available to try,
232 * This routine is typically called in situations where we believe
233 * one server out of a pool may have an error condition.
239 * The afs_Conn* routines use the list of invalidated servers to
240 * avoid reusing a server marked as invalid for this request.
241 *------------------------------------------------------------------------*/
243 afs_BlackListOnce(struct vrequest *areq, struct VenusFid *afid,
248 afs_int32 serversleft = 0;
251 tvp = afs_FindVolume(afid, READ_LOCK);
253 for (i = 0; i < MAXHOSTS; i++) {
254 if (tvp->serverHost[i] == tsp) {
255 areq->skipserver[i] = 1;
257 if (tvp->serverHost[i] &&
258 (tvp->serverHost[i]->addr->sa_flags &
260 areq->skipserver[i] = 1;
263 afs_PutVolume(tvp, READ_LOCK);
264 for (i = 0; i < MAXHOSTS; i++) {
265 if (tvp->serverHost[i] && areq->skipserver[i] == 0) {
277 /*------------------------------------------------------------------------
278 * EXPORTED afs_Analyze
281 * Analyze the outcome of an RPC operation, taking whatever support
282 * actions are necessary.
285 * aconn : Ptr to the relevant connection on which the call was made.
286 * acode : The return code experienced by the RPC.
287 * afid : The FID of the file involved in the action. This argument
288 * may be null if none was involved.
289 * areq : The request record associated with this operation.
290 * op : which RPC we are analyzing.
291 * cellp : pointer to a cell struct. Must provide either fid or cell.
294 * Non-zero value if the related RPC operation should be retried,
298 * This routine is typically called in a do-while loop, causing the
299 * embedded RPC operation to be called repeatedly if appropriate
300 * until whatever error condition (if any) is intolerable.
306 * The retry return value is used by afs_StoreAllSegments to determine
307 * if this is a temporary or permanent error.
308 *------------------------------------------------------------------------*/
310 afs_Analyze(register struct afs_conn *aconn, afs_int32 acode,
311 struct VenusFid *afid, register struct vrequest *areq, int op,
312 afs_int32 locktype, struct cell *cellp)
318 afs_int32 shouldRetry = 0;
319 afs_int32 serversleft = 1;
320 struct afs_stats_RPCErrors *aerrP;
321 afs_int32 markeddown;
325 if (AFS_IS_DISCONNECTED && !AFS_IN_SYNC) {
326 /* On reconnection, act as connected. XXX: for now.... */
327 /* SXW - This may get very tired after a while. We should try and
328 * intercept all RPCs before they get here ... */
329 /*printf("afs_Analyze: disconnected\n");*/
330 afs_FinalizeReq(areq);
332 /* SXW - I suspect that this will _never_ happen - we shouldn't
333 * get a connection because we're disconnected !!!*/
334 afs_PutConn(aconn, locktype);
339 AFS_STATCNT(afs_Analyze);
340 afs_Trace4(afs_iclSetp, CM_TRACE_ANALYZE, ICL_TYPE_INT32, op,
341 ICL_TYPE_POINTER, aconn, ICL_TYPE_INT32, acode, ICL_TYPE_LONG,
344 aerrP = (struct afs_stats_RPCErrors *)0;
346 if ((op >= 0) && (op < AFS_STATS_NUM_FS_RPC_OPS))
347 aerrP = &(afs_stats_cmfullperf.rpc.fsRPCErrors[op]);
349 afs_FinalizeReq(areq);
350 if (!aconn && areq->busyCount) { /* one RPC or more got VBUSY/VRESTARTING */
352 tvp = afs_FindVolume(afid, READ_LOCK);
354 afs_warnuser("afs: Waiting for busy volume %u (%s) in cell %s\n",
355 (afid ? afid->Fid.Volume : 0),
356 (tvp->name ? tvp->name : ""),
358 && tvp->serverHost[0]->cell) ? tvp->serverHost[0]->
359 cell->cellName : ""));
361 for (i = 0; i < MAXHOSTS; i++) {
362 if (tvp->status[i] != not_busy && tvp->status[i] != offline) {
363 tvp->status[i] = not_busy;
365 if (tvp->status[i] == not_busy)
368 afs_PutVolume(tvp, READ_LOCK);
370 afs_warnuser("afs: Waiting for busy volume %u\n",
371 (afid ? afid->Fid.Volume : 0));
374 if (areq->busyCount > 100) {
376 (aerrP->err_Volume)++;
377 areq->volumeError = VOLBUSY;
380 VSleep(afs_BusyWaitPeriod); /* poll periodically */
382 if (shouldRetry != 0)
385 return shouldRetry; /* should retry */
388 if (!aconn || !aconn->srvr) {
389 if (!areq->volumeError) {
391 (aerrP->err_Network)++;
392 if (hm_retry_int && !(areq->flags & O_NONBLOCK) && /* "hard" mount */
393 ((afid && afs_IsPrimaryCellNum(afid->Cell))
394 || (cellp && afs_IsPrimaryCell(cellp)))) {
397 ("afs: hard-mount waiting for a vlserver to return to service\n");
398 VSleep(hm_retry_int);
399 afs_CheckServers(1, cellp);
402 tvp = afs_FindVolume(afid, READ_LOCK);
403 if (!tvp || (tvp->states & VRO)) {
404 shouldRetry = hm_retry_RO;
406 shouldRetry = hm_retry_RW;
409 afs_PutVolume(tvp, READ_LOCK);
412 ("afs: hard-mount waiting for volume %u\n",
414 VSleep(hm_retry_int);
415 afs_CheckServers(1, cellp);
418 } /* if (hm_retry_int ... */
420 areq->networkError = 1;
426 /* Find server associated with this connection. */
430 /* Before we do anything with acode, make sure we translate it back to
432 if ((acode & ~0xff) == ERROR_TABLE_BASE_uae)
433 acode = et_to_sys_error(acode);
436 /* If we previously took an error, mark this volume not busy */
437 if (areq->volumeError) {
438 tvp = afs_FindVolume(afid, READ_LOCK);
440 for (i = 0; i < MAXHOSTS; i++) {
441 if (tvp->serverHost[i] == tsp) {
442 tvp->status[i] = not_busy;
445 afs_PutVolume(tvp, READ_LOCK);
449 afs_PutConn(aconn, locktype);
453 /* If network troubles, mark server as having bogued out again. */
454 /* VRESTARTING is < 0 because of backward compatibility issues
455 * with 3.4 file servers and older cache managers */
456 #ifdef AFS_64BIT_CLIENT
459 #endif /* AFS_64BIT_CLIENT */
460 if ((acode < 0) && (acode != VRESTARTING)) {
461 if (acode == RX_CALL_TIMEOUT) {
462 serversleft = afs_BlackListOnce(areq, afid, tsp);
469 /* By doing this, we avoid ever marking a server down
470 * in an idle timeout case. That's because the server is
471 * still responding and may only be letting a single vnode
472 * time out. We otherwise risk having the server continually
473 * be marked down, then up, then down again...
477 markeddown = afs_ServerDown(sa);
478 ForceNewConnections(sa); /**multi homed clients lock:afs_xsrvAddr? */
480 (aerrP->err_Server)++;
482 /* retry *once* when the server is timed out in case of NAT */
483 if (markeddown && acode == RX_CALL_DEAD) {
484 aconn->forceConnectFS = 1;
490 if (acode == VBUSY || acode == VRESTARTING) {
491 if (acode == VBUSY) {
494 (aerrP->err_VolumeBusies)++;
498 tvp = afs_FindVolume(afid, READ_LOCK);
500 for (i = 0; i < MAXHOSTS; i++) {
501 if (tvp->serverHost[i] == tsp) {
502 tvp->status[i] = rdwr_busy; /* can't tell which yet */
503 /* to tell which, have to look at the op code. */
506 afs_PutVolume(tvp, READ_LOCK);
508 afs_warnuser("afs: Waiting for busy volume %u in cell %s\n",
509 (afid ? afid->Fid.Volume : 0), tsp->cell->cellName);
510 VSleep(afs_BusyWaitPeriod); /* poll periodically */
514 } else if (acode == VICETOKENDEAD
515 || (acode & ~0xff) == ERROR_TABLE_BASE_RXK) {
516 /* any rxkad error is treated as token expiration */
519 * I'm calling these errors protection errors, since they involve
520 * faulty authentication.
523 (aerrP->err_Protection)++;
525 tu = afs_FindUser(areq->uid, tsp->cell->cellNum, READ_LOCK);
527 if (acode == VICETOKENDEAD) {
528 aconn->forceConnectFS = 1;
529 } else if (acode == RXKADEXPIRED) {
530 aconn->forceConnectFS = 0; /* don't check until new tokens set */
531 aconn->user->states |= UTokensBad;
533 ("afs: Tokens for user of AFS id %d for cell %s have expired\n",
534 tu->vid, aconn->srvr->server->cell->cellName);
536 serversleft = afs_BlackListOnce(areq, afid, tsp);
541 ("afs: Tokens for user of AFS id %d for cell %s: rxkad error=%d\n",
542 tu->vid, aconn->srvr->server->cell->cellName, acode);
545 areq->tokenError = 0;
546 aconn->forceConnectFS = 0; /* don't check until new tokens set */
547 aconn->user->states |= UTokensBad;
549 ("afs: Tokens for user of AFS id %d for cell %s are discarded (rxkad error=%d)\n",
550 tu->vid, aconn->srvr->server->cell->cellName, acode);
553 afs_PutUser(tu, READ_LOCK);
555 /* The else case shouldn't be possible and should probably be replaced by a panic? */
556 if (acode == VICETOKENDEAD) {
557 aconn->forceConnectFS = 1;
558 } else if (acode == RXKADEXPIRED) {
559 aconn->forceConnectFS = 0; /* don't check until new tokens set */
560 aconn->user->states |= UTokensBad;
562 ("afs: Tokens for user %d for cell %s have expired\n",
563 areq->uid, aconn->srvr->server->cell->cellName);
565 aconn->forceConnectFS = 0; /* don't check until new tokens set */
566 aconn->user->states |= UTokensBad;
568 ("afs: Tokens for user %d for cell %s are discarded (rxkad error = %d)\n",
569 areq->uid, aconn->srvr->server->cell->cellName, acode);
572 shouldRetry = 1; /* Try again (as root). */
574 /* Check for access violation. */
575 else if (acode == EACCES) {
576 /* should mark access error in non-existent per-user global structure */
578 (aerrP->err_Protection)++;
579 areq->accessError = 1;
580 if (op == AFS_STATS_FS_RPCIDX_STOREDATA)
581 areq->permWriteError = 1;
584 /* check for ubik errors; treat them like crashed servers */
585 else if (acode >= ERROR_TABLE_BASE_U && acode < ERROR_TABLE_BASE_U + 255) {
588 (aerrP->err_Server)++;
589 shouldRetry = 1; /* retryable (maybe one is working) */
590 VSleep(1); /* just in case */
592 /* Check for bad volume data base / missing volume. */
593 else if (acode == VSALVAGE || acode == VOFFLINE || acode == VNOVOL
594 || acode == VNOSERVICE || acode == VMOVED) {
599 areq->volumeError = VOLMISSING;
601 (aerrP->err_Volume)++;
602 if (afid && (tcell = afs_GetCell(afid->Cell, 0))) {
603 same = VLDB_Same(afid, areq);
604 tvp = afs_FindVolume(afid, READ_LOCK);
606 for (i = 0; i < MAXHOSTS && tvp->serverHost[i]; i++) {
607 if (tvp->serverHost[i] == tsp) {
608 if (tvp->status[i] == end_not_busy)
609 tvp->status[i] = offline;
613 tvp->status[i] = not_busy; /* reset the others */
616 afs_PutVolume(tvp, READ_LOCK);
619 } else if (acode >= ERROR_TABLE_BASE_VL && acode <= ERROR_TABLE_BASE_VL + 255) { /* vlserver errors */
621 areq->volumeError = VOLMISSING;
622 } else if (acode >= 0) {
624 (aerrP->err_Other)++;
625 if (op == AFS_STATS_FS_RPCIDX_STOREDATA)
626 areq->permWriteError = 1;
627 shouldRetry = 0; /* Other random Vice error. */
628 } else if (acode == RX_MSGSIZE) { /* same meaning as EMSGSIZE... */
629 VSleep(1); /* Just a hack for desperate times. */
631 (aerrP->err_Other)++;
632 shouldRetry = 1; /* packet was too big, please retry call */
635 if (acode < 0 && acode != RX_MSGSIZE && acode != VRESTARTING) {
636 /* If we get here, code < 0 and we have network/Server troubles.
637 * areq->networkError is not set here, since we always
638 * retry in case there is another server. However, if we find
639 * no connection (aconn == 0) we set the networkError flag.
641 afs_MarkServerUpOrDown(sa, SRVR_ISDOWN);
643 (aerrP->err_Server)++;
644 VSleep(1); /* Just a hack for desperate times. */
648 /* now unlock the connection and return */
649 afs_PutConn(aconn, locktype);
650 return (shouldRetry);