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
18 * AFS_EQ_ATSYS (macro)
22 #include <afsconfig.h>
23 #include "../afs/param.h"
27 #include "../afs/sysincludes.h" /* Standard vendor system headers */
28 #include "../afs/afsincludes.h" /* Afs-based standard headers */
29 #include "../afs/afs_stats.h" /* statistics */
30 #include "../afs/afs_cbqueue.h"
31 #include "../afs/nfsclient.h"
32 #include "../afs/exporter.h"
33 #include "../afs/afs_osidnlc.h"
37 * A few definitions. This is until we have a proper header file
38 * which has prototypes for all functions
41 extern struct DirEntry * afs_dir_GetBlob();
43 extern afs_rwlock_t afs_xvcache;
44 extern afs_rwlock_t afs_xcbhash;
45 extern struct afs_exporter *afs_nfsexporter;
46 extern char *afs_sysname;
47 extern char *afs_sysnamelist[];
48 extern int afs_sysnamecount;
49 extern struct afs_q VLRU; /*vcache LRU*/
50 #ifdef AFS_LINUX22_ENV
51 extern struct inode_operations afs_symlink_iops, afs_dir_iops;
55 afs_int32 afs_bulkStatsDone;
56 static int bulkStatCounter = 0; /* counter for bulk stat seq. numbers */
59 /* this would be faster if it did comparison as int32word, but would be
60 * dependant on byte-order and alignment, and I haven't figured out
61 * what "@sys" is in binary... */
62 #define AFS_EQ_ATSYS(name) (((name)[0]=='@')&&((name)[1]=='s')&&((name)[2]=='y')&&((name)[3]=='s')&&(!(name)[4]))
66 register char *s1, *s2;
82 register char *a, c; {
84 AFS_STATCNT(afs_index);
86 if (tc == c) return a;
92 /* call under write lock, evaluate mvid field from a mt pt.
93 * avc is the vnode of the mount point object.
94 * advc is the vnode of the containing directory
95 * avolpp is where we return a pointer to the volume named by the mount pt, if success
96 * areq is the identity of the caller.
98 * NOTE: this function returns a held volume structure in *volpp if it returns 0!
100 EvalMountPoint(avc, advc, avolpp, areq)
101 register struct vcache *avc;
102 struct volume **avolpp;
103 struct vcache *advc; /* the containing dir */
104 register struct vrequest *areq;
107 struct volume *tvp = 0;
108 struct VenusFid tfid;
110 char *cpos, *volnamep;
112 afs_int32 prefetchRO; /* 1=>No 2=>Yes */
113 afs_int32 mtptCell, assocCell, hac=0;
114 afs_int32 samecell, roname, len;
116 AFS_STATCNT(EvalMountPoint);
118 if (avc->mvid && (avc->states & CMValid)) return 0; /* done while racing */
120 *avolpp = (struct volume *)0;
121 code = afs_HandleLink(avc, areq);
122 if (code) return code;
124 /* Determine which cell and volume the mointpoint goes to */
125 type = avc->linkData[0]; /* '#'=>Regular '%'=>RW */
126 cpos = afs_index(&avc->linkData[1], ':'); /* if cell name present */
130 tcell = afs_GetCellByName(&avc->linkData[1], READ_LOCK);
133 volnamep = &avc->linkData[1];
134 tcell = afs_GetCell(avc->fid.Cell, READ_LOCK);
136 if (!tcell) return ENODEV;
138 mtptCell = tcell->cell; /* The cell for the mountpoint */
140 hac = 1; /* has associated cell */
141 assocCell = tcell->lcellp->cell; /* The associated cell */
143 afs_PutCell(tcell, READ_LOCK);
145 /* Is volume name a "<n>.backup" or "<n>.readonly" name */
146 len = strlen(volnamep);
147 roname = ((len > 9) && (strcmp(&volnamep[len - 9],".readonly") == 0)) ||
148 ((len > 7) && (strcmp(&volnamep[len - 7],".backup") == 0));
150 /* When we cross mountpoint, do we stay in the same cell */
151 samecell = (avc->fid.Cell == mtptCell) || (hac && (avc->fid.Cell == assocCell));
153 /* Decide whether to prefetch the RO. Also means we want the RO.
154 * If this is a regular mountpoint with a RW volume name and
155 * we cross a cell boundary -or- start from a RO volume, then we will
156 * want to prefetch the RO volume when we get the RW below.
158 if ( (type == '#') && !roname && (!samecell || (avc->states & CRO)) ) {
159 prefetchRO = 2; /* Yes, prefetch the RO */
161 prefetchRO = 1; /* No prefetch of the RO */
164 /* Get the volume struct. Unless this volume name has ".readonly" or
165 * ".backup" in it, this will get the volume struct for the RW volume.
166 * The RO volume will be prefetched if requested (but not returned).
168 tvp = afs_GetVolumeByName(volnamep, mtptCell, prefetchRO, areq, WRITE_LOCK);
170 /* If no volume was found in this cell, try the associated linked cell */
171 if (!tvp && hac && areq->volumeError) {
172 tvp = afs_GetVolumeByName(volnamep, assocCell, prefetchRO, areq, WRITE_LOCK);
175 /* Still not found. If we are looking for the RO, then perhaps the RW
176 * doesn't exist? Try adding ".readonly" to volname and look for that.
177 * Don't know why we do this. Would have still found it in above call - jpm.
179 if (!tvp && (prefetchRO == 2)) {
180 strcpy(buf, volnamep);
181 afs_strcat(buf, ".readonly");
183 tvp = afs_GetVolumeByName(buf, mtptCell, 1, areq, WRITE_LOCK);
185 /* Try the associated linked cell if failed */
186 if (!tvp && hac && areq->volumeError) {
187 tvp = afs_GetVolumeByName(buf, assocCell, 1, areq, WRITE_LOCK);
191 if (!tvp) return ENODEV; /* Couldn't find the volume */
193 /* Don't cross mountpoint from a BK to a BK volume */
194 if ((avc->states & CBackup) && (tvp->states & VBackup)) {
195 afs_PutVolume(tvp, WRITE_LOCK);
199 /* If we want (prefetched) the RO and it exists, then drop the
200 * RW volume and get the RO. Othewise, go with the RW.
202 if ((prefetchRO == 2) && tvp->roVol) {
203 tfid.Fid.Volume = tvp->roVol; /* remember RO volume */
204 tfid.Cell = tvp->cell;
205 afs_PutVolume(tvp, WRITE_LOCK); /* release old volume */
206 tvp = afs_GetVolume(&tfid, areq, WRITE_LOCK); /* get the new one */
207 if (!tvp) return ENODEV; /* oops, can't do it */
211 avc->mvid = (struct VenusFid *) osi_AllocSmallSpace(sizeof(struct VenusFid));
212 avc->mvid->Cell = tvp->cell;
213 avc->mvid->Fid.Volume = tvp->volume;
214 avc->mvid->Fid.Vnode = 1;
215 avc->mvid->Fid.Unique = 1;
216 avc->states |= CMValid;
218 /* Used to: if the mount point is stored within a backup volume,
219 * then we should only update the parent pointer information if
220 * there's none already set, so as to avoid updating a volume's ..
221 * info with something in an OldFiles directory.
223 * Next two lines used to be under this if:
225 * if (!(avc->states & CBackup) || tvp->dotdot.Fid.Volume == 0)
227 * Now: update mount point back pointer on every call, so that we handle
228 * multiple mount points better. This way, when du tries to go back
229 * via chddir(".."), it will end up exactly where it started, yet
230 * cd'ing via a new path to a volume will reset the ".." pointer
233 tvp->mtpoint = avc->fid; /* setup back pointer to mtpoint */
234 tvp->dotdot = advc->fid;
241 register char *aname; {
245 AFS_STATCNT(ENameOK);
246 tlen = strlen(aname);
247 if (tlen >= 4 && strcmp(aname+tlen-4, "@sys") == 0) return 0;
251 afs_getsysname(areq, adp, bufp)
252 register struct vrequest *areq;
253 register struct vcache *adp;
256 static char sysname[MAXSYSNAME];
257 register struct unixuser *au;
258 register afs_int32 error;
260 if (!afs_nfsexporter) {
261 strcpy(bufp, afs_sysname);
264 AFS_STATCNT(getsysname);
265 au = afs_GetUser(areq->uid, adp->fid.Cell, 0);
268 error = EXP_SYSNAME(au->exporter, (char *)0, bufp);
270 strcpy(bufp, "@sys");
273 strcpy(bufp, afs_sysname);
278 Check_AtSys(avc, aname, state, areq)
279 register struct vcache *avc;
281 struct sysname_info *state;
282 struct vrequest *areq;
284 if (AFS_EQ_ATSYS(aname)) {
286 state->name = (char *) osi_AllocLargeSpace(AFS_SMALLOCSIZ);
288 state->index = afs_getsysname(areq, avc, state->name);
297 Next_AtSys(avc, areq, state)
298 register struct vcache *avc;
299 struct vrequest *areq;
300 struct sysname_info *state;
302 if (state->index == -1)
303 return 0; /* No list */
305 /* Check for the initial state of aname != "@sys" in Check_AtSys*/
306 if (state->offset == -1 && state->allocked == 0) {
307 register char *tname;
308 /* Check for .*@sys */
309 for (tname=state->name; *tname; tname++)
310 /*Move to the end of the string*/;
311 if ((tname > state->name + 4) && (AFS_EQ_ATSYS(tname-4))) {
312 state->offset = (tname - 4) - state->name;
313 tname = (char *) osi_AllocLargeSpace(AFS_LRALLOCSIZ);
314 strncpy(tname, state->name, state->offset);
317 state->index = afs_getsysname(areq, avc, state->name+state->offset);
320 return 0; /* .*@sys doesn't match either */
321 } else if (++(state->index) >= afs_sysnamecount
322 || !afs_sysnamelist[state->index])
323 return 0; /* end of list */
324 strcpy(state->name+state->offset, afs_sysnamelist[state->index]);
328 #if (defined(AFS_SGI62_ENV) || defined(AFS_SUN57_64BIT_ENV))
329 extern int BlobScan(ino64_t *afile, afs_int32 ablob);
331 #if defined AFS_LINUX_64BIT_KERNEL
332 extern int BlobScan(long *afile, afs_int32 ablob);
334 extern int BlobScan(afs_int32 *afile, afs_int32 ablob);
339 /* called with an unlocked directory and directory cookie. Areqp
340 * describes who is making the call.
341 * Scans the next N (about 30, typically) directory entries, and does
342 * a bulk stat call to stat them all.
344 * Must be very careful when merging in RPC responses, since we dont
345 * want to overwrite newer info that was added by a file system mutating
346 * call that ran concurrently with our bulk stat call.
348 * We do that, as described below, by not merging in our info (always
349 * safe to skip the merge) if the status info is valid in the vcache entry.
351 * If adapt ever implements the bulk stat RPC, then this code will need to
352 * ensure that vcaches created for failed RPC's to older servers have the
355 struct vcache * BStvc = (struct vcache *) 0;
356 int afs_DoBulkStat(adp, dirCookie, areqp)
359 struct vrequest *areqp;
361 int nentries; /* # of entries to prefetch */
362 int nskip; /* # of slots in the LRU queue to skip */
363 struct vcache *lruvcp; /* vcache ptr of our goal pos in LRU queue */
364 struct dcache *dcp; /* chunk containing the dir block */
365 char *statMemp; /* status memory block */
366 char *cbfMemp; /* callback and fid memory block */
367 long temp; /* temp for holding chunk length, &c. */
368 struct AFSFid *fidsp; /* file IDs were collecting */
369 struct AFSCallBack *cbsp; /* call back pointers */
370 struct AFSCallBack *tcbp; /* temp callback ptr */
371 struct AFSFetchStatus *statsp; /* file status info */
372 struct AFSVolSync volSync; /* vol sync return info */
373 struct vcache *tvcp; /* temp vcp */
374 struct afs_q *tq; /* temp queue variable */
375 AFSCBFids fidParm; /* file ID parm for bulk stat */
376 AFSBulkStats statParm; /* stat info parm for bulk stat */
377 int fidIndex; /* which file were stating */
378 struct conn *tcp; /* conn for call */
379 AFSCBs cbParm; /* callback parm for bulk stat */
380 struct server *hostp = 0; /* host we got callback from */
381 long origEvenCBs; /* original # of callbacks for even-fid files */
382 long origOddCBs; /* original # of callbacks for odd-fid files */
383 long origEvenZaps; /* original # of recycles for even-fid files */
384 long origOddZaps; /* original # of recycles for odd-fid files */
385 long startTime; /* time we started the call,
386 * for callback expiration base
388 int statSeqNo; /* Valued of file size to detect races */
389 int code; /* error code */
390 long newIndex; /* new index in the dir */
391 struct DirEntry *dirEntryp; /* dir entry we are examining */
393 struct VenusFid afid; /* file ID we are using now */
394 struct VenusFid tfid; /* another temp. file ID */
395 afs_int32 retry; /* handle low-level SGI MP race conditions */
396 long volStates; /* flags from vol structure */
397 struct volume *volp=0; /* volume ptr */
398 struct VenusFid dotdot;
399 int flagIndex; /* First file with bulk fetch flag set */
400 int inlinebulk=0; /* Did we use InlineBulk RPC or not? */
403 /* first compute some basic parameters. We dont want to prefetch more
404 * than a fraction of the cache in any given call, and we want to preserve
405 * a portion of the LRU queue in any event, so as to avoid thrashing
406 * the entire stat cache (we will at least leave some of it alone).
407 * presently dont stat more than 1/8 the cache in any one call. */
408 nentries = afs_cacheStats / 8;
410 /* dont bother prefetching more than one calls worth of info */
411 if (nentries > AFSCBMAX) nentries = AFSCBMAX;
413 /* heuristic to make sure that things fit in 4K. This means that
414 * we shouldnt make it any bigger than 47 entries. I am typically
415 * going to keep it a little lower, since we don't want to load
416 * too much of the stat cache.
418 if (nentries > 30) nentries = 30;
420 /* now, to reduce the stack size, well allocate two 4K blocks,
421 * one for fids and callbacks, and one for stat info. Well set
422 * up our pointers to the memory from there, too.
424 statMemp = osi_AllocLargeSpace(nentries * sizeof(AFSFetchStatus));
425 statsp = (struct AFSFetchStatus *) statMemp;
426 cbfMemp = osi_AllocLargeSpace(nentries *
427 (sizeof(AFSCallBack) + sizeof(AFSFid)));
428 fidsp = (AFSFid *) cbfMemp;
429 cbsp = (AFSCallBack *) (cbfMemp + nentries * sizeof(AFSFid));
431 /* next, we must iterate over the directory, starting from the specified
432 * cookie offset (dirCookie), and counting out nentries file entries.
433 * We skip files that already have stat cache entries, since we
434 * dont want to bulk stat files that are already in the cache.
437 code = afs_VerifyVCache(adp, areqp);
440 dcp = afs_GetDCache(adp, 0, areqp, &temp, &temp, 1);
446 /* lock the directory cache entry */
447 ObtainReadLock(&adp->lock);
450 * Make sure that the data in the cache is current. There are two
451 * cases we need to worry about:
452 * 1. The cache data is being fetched by another process.
453 * 2. The cache data is no longer valid
455 while ((adp->states & CStatd)
456 && (dcp->flags & DFFetching)
457 && hsame(adp->m.DataVersion, dcp->f.versionNo)) {
458 dcp->flags |= DFWaiting;
459 ReleaseReadLock(&adp->lock);
460 afs_osi_Sleep(&dcp->validPos);
461 ObtainReadLock(&adp->lock);
463 if (!(adp->states & CStatd)
464 || !hsame(adp->m.DataVersion, dcp->f.versionNo)) {
465 ReleaseReadLock(&adp->lock);
470 /* Generate a sequence number so we can tell whether we should
471 * store the attributes when processing the response. This number is
472 * stored in the file size when we set the CBulkFetching bit. If the
473 * CBulkFetching is still set and this value hasn't changed, then
474 * we know we were the last to set CBulkFetching bit for this file,
475 * and it is safe to set the status information for this file.
477 statSeqNo = bulkStatCounter++;
479 /* now we have dir data in the cache, so scan the dir page */
482 while (1) { /* Should probably have some constant bound */
483 /* look for first safe entry to examine in the directory. BlobScan
484 * looks for a the 1st allocated dir after the dirCookie slot.
486 newIndex = BlobScan(&dcp->f.inode, (dirCookie>>5));
487 if (newIndex == 0) break;
489 /* remember the updated directory cookie */
490 dirCookie = newIndex << 5;
492 /* get a ptr to the dir entry */
493 dirEntryp =(struct DirEntry *)afs_dir_GetBlob(&dcp->f.inode, newIndex);
494 if (!dirEntryp) break;
496 /* dont copy more than we have room for */
497 if (fidIndex >= nentries) {
498 DRelease((char *) dirEntryp, 0);
502 /* now, if the dir entry looks good, copy it out to our list. Vnode
503 * 0 means deleted, although it should also be free were it deleted.
505 if (dirEntryp->fid.vnode != 0) {
506 /* dont copy entries we have in our cache. This check will
507 * also make us skip "." and probably "..", unless it has
508 * disappeared from the cache since we did our namei call.
510 tfid.Cell = adp->fid.Cell;
511 tfid.Fid.Volume = adp->fid.Fid.Volume;
512 tfid.Fid.Vnode = ntohl(dirEntryp->fid.vnode);
513 tfid.Fid.Unique = ntohl(dirEntryp->fid.vunique);
516 ObtainWriteLock(&afs_xvcache, 130);
517 tvcp = afs_FindVCache(&tfid, 0, 0, &retry, 0 /* no stats | LRU */);
519 ReleaseWriteLock(&afs_xvcache);
522 } while (tvcp && retry);
523 if (!tvcp) { /* otherwise, create manually */
524 tvcp = afs_NewVCache(&tfid, hostp, 0, 0);
525 ObtainWriteLock(&tvcp->lock, 505);
526 ReleaseWriteLock(&afs_xvcache);
527 afs_RemoveVCB(&tfid);
528 ReleaseWriteLock(&tvcp->lock);
530 ReleaseWriteLock(&afs_xvcache);
533 goto done; /* can't happen at present, more's the pity */
535 /* WARNING: afs_DoBulkStat uses the Length field to store a
536 * sequence number for each bulk status request. Under no
537 * circumstances should afs_DoBulkStat store a sequence number
538 * if the new length will be ignored when afs_ProcessFS is
539 * called with new stats. */
541 if (!(tvcp->states & (CStatd|CBulkFetching))
542 && (tvcp->execsOrWriters <= 0)
543 && !afs_DirtyPages(tvcp)
544 && !AFS_VN_MAPPED((vnode_t*)tvcp))
546 if (!(tvcp->states & (CStatd|CBulkFetching))
547 && (tvcp->execsOrWriters <= 0)
548 && !afs_DirtyPages(tvcp))
552 /* this entry doesnt exist in the cache, and is not
553 * already being fetched by someone else, so add it to the
554 * list of file IDs to obtain.
556 * We detect a callback breaking race condition by checking the
557 * CBulkFetching state bit and the value in the file size.
558 * It is safe to set the status only if the CBulkFetching
559 * flag is still set and the value in the file size does
562 * Don't fetch status for dirty files. We need to
563 * preserve the value of the file size. We could
564 * flush the pages, but it wouldn't be worthwhile.
566 memcpy((char *)(fidsp+fidIndex), (char *) &tfid.Fid, sizeof(*fidsp));
567 tvcp->states |= CBulkFetching;
568 tvcp->m.Length = statSeqNo;
572 } /* if dir vnode has non-zero entry */
574 /* move to the next dir entry by adding in the # of entries
575 * used by this dir entry.
577 temp = afs_dir_NameBlobs(dirEntryp->name) << 5;
578 DRelease((char *) dirEntryp, 0);
579 if (temp <= 0) break;
581 } /* while loop over all dir entries */
583 /* now release the dir lock and prepare to make the bulk RPC */
584 ReleaseReadLock(&adp->lock);
586 /* release the chunk */
589 /* dont make a null call */
590 if (fidIndex == 0) goto done;
593 /* setup the RPC parm structures */
594 fidParm.AFSCBFids_len = fidIndex;
595 fidParm.AFSCBFids_val = fidsp;
596 statParm.AFSBulkStats_len = fidIndex;
597 statParm.AFSBulkStats_val = statsp;
598 cbParm.AFSCBs_len = fidIndex;
599 cbParm.AFSCBs_val = cbsp;
601 /* start the timer; callback expirations are relative to this */
602 startTime = osi_Time();
604 tcp = afs_Conn(&adp->fid, areqp, SHARED_LOCK);
606 hostp = tcp->srvr->server;
607 XSTATS_START_TIME(AFS_STATS_FS_RPCIDX_BULKSTATUS);
608 #ifdef RX_ENABLE_LOCKS
610 #endif /* RX_ENABLE_LOCKS */
611 code = RXAFS_InlineBulkStatus(tcp->id, &fidParm, &statParm,
613 if (code == RXGEN_OPCODE) {
614 code = RXAFS_BulkStatus(tcp->id, &fidParm, &statParm, &cbParm,
621 #ifdef RX_ENABLE_LOCKS
623 #endif /* RX_ENABLE_LOCKS */
627 } while (afs_Analyze(tcp, code, &adp->fid, areqp,
628 AFS_STATS_FS_RPCIDX_BULKSTATUS, SHARED_LOCK, (struct cell *)0));
630 /* now, if we didnt get the info, bail out. */
633 /* we need vol flags to create the entries properly */
634 dotdot.Fid.Volume = 0;
635 volp = afs_GetVolume(&adp->fid, areqp, READ_LOCK);
637 volStates = volp->states;
638 if (volp->dotdot.Fid.Volume != 0)
639 dotdot = volp->dotdot;
643 /* find the place to merge the info into We do this by skipping
644 * nskip entries in the LRU queue. The more we skip, the more
645 * we preserve, since the head of the VLRU queue is the most recently
649 nskip = afs_cacheStats / 2; /* preserved fraction of the cache */
650 ObtainReadLock(&afs_xvcache);
652 /* actually a serious error, probably should panic. Probably will
653 * panic soon, oh well. */
654 ReleaseReadLock(&afs_xvcache);
655 afs_warnuser("afs_DoBulkStat: VLRU empty!");
658 if ((VLRU.next->prev != &VLRU) || (VLRU.prev->next != &VLRU)) {
659 refpanic ("Bulkstat VLRU inconsistent");
661 for(tq = VLRU.next; tq != &VLRU; tq = QNext(tq)) {
662 if (--nskip <= 0) break;
663 else if (QNext(QPrev(tq)) != tq) {
665 refpanic ("BulkStat VLRU inconsistent");
668 if (tq != &VLRU) lruvcp = QTOV(tq);
669 else lruvcp = QTOV(VLRU.next);
671 /* now we have to hold this entry, so that it does not get moved
672 * into the free list while we're running. It could still get
673 * moved within the lru queue, but hopefully that will be rare; it
674 * doesn't hurt nearly as much.
677 osi_vnhold(lruvcp, &retry);
678 ReleaseReadLock(&afs_xvcache); /* could be read lock */
682 /* otherwise, merge in the info. We have to be quite careful here,
683 * since we need to ensure that we don't merge old info over newer
684 * stuff in a stat cache entry. We're very conservative here: we don't
685 * do the merge at all unless we ourselves create the stat cache
686 * entry. That's pretty safe, and should work pretty well, since we
687 * typically expect to do the stat cache creation ourselves.
689 * We also have to take into account racing token revocations.
691 for(i=0; i<fidIndex; i++) {
692 if ((&statsp[i])->errorCode)
694 afid.Cell = adp->fid.Cell;
695 afid.Fid.Volume = adp->fid.Fid.Volume;
696 afid.Fid.Vnode = fidsp[i].Vnode;
697 afid.Fid.Unique = fidsp[i].Unique;
700 ObtainReadLock(&afs_xvcache);
701 tvcp = afs_FindVCache(&afid, 1, 0, &retry, 0/* !stats&!lru*/);
702 ReleaseReadLock(&afs_xvcache);
703 } while (tvcp && retry);
705 /* The entry may no longer exist */
710 /* now we have the entry held, but we need to fill it in */
711 ObtainWriteLock(&tvcp->lock,131);
713 /* if CBulkFetching is not set, or if the file size no longer
714 * matches the value we placed there when we set the CBulkFetching
715 * flag, then someone else has done something with this node,
716 * and we may not have the latest status information for this
717 * file. Leave the entry alone.
719 if (!(tvcp->states & CBulkFetching) || (tvcp->m.Length != statSeqNo)) {
721 ReleaseWriteLock(&tvcp->lock);
726 /* now copy ".." entry back out of volume structure, if necessary */
727 if (tvcp->mvstat == 2 && (dotdot.Fid.Volume != 0)) {
729 tvcp->mvid = (struct VenusFid *) osi_AllocSmallSpace(sizeof(struct VenusFid));
730 *tvcp->mvid = dotdot;
733 ObtainWriteLock(&afs_xvcache,132);
734 if ((VLRU.next->prev != &VLRU) || (VLRU.prev->next != &VLRU)) {
735 refpanic ("Bulkstat VLRU inconsistent2");
737 if ((QNext(QPrev(&tvcp->vlruq)) != &tvcp->vlruq)
738 || (QPrev(QNext(&tvcp->vlruq)) != &tvcp->vlruq))
739 refpanic ("Bulkstat VLRU inconsistent4");
740 if ((QNext(QPrev(&lruvcp->vlruq)) != &lruvcp->vlruq)
741 || (QPrev(QNext(&lruvcp->vlruq)) != &lruvcp->vlruq))
742 refpanic ("Bulkstat VLRU inconsistent5");
744 if (tvcp != lruvcp) { /* if they are == don't move it, don't corrupt vlru */
745 QRemove(&tvcp->vlruq);
746 QAdd(&lruvcp->vlruq, &tvcp->vlruq);
749 if ((VLRU.next->prev != &VLRU) || (VLRU.prev->next != &VLRU)) {
750 refpanic ("Bulkstat VLRU inconsistent3");
752 if ((QNext(QPrev(&tvcp->vlruq)) != &tvcp->vlruq)
753 || (QPrev(QNext(&tvcp->vlruq)) != &tvcp->vlruq))
754 refpanic ("Bulkstat VLRU inconsistent5");
755 if ((QNext(QPrev(&lruvcp->vlruq)) != &lruvcp->vlruq)
756 || (QPrev(QNext(&lruvcp->vlruq)) != &lruvcp->vlruq))
757 refpanic ("Bulkstat VLRU inconsistent6");
758 ReleaseWriteLock(&afs_xvcache);
760 ObtainWriteLock(&afs_xcbhash, 494);
762 /* We need to check the flags again. We may have missed
763 * something while we were waiting for a lock.
765 if (!(tvcp->states & CBulkFetching) || (tvcp->m.Length != statSeqNo)) {
767 ReleaseWriteLock(&tvcp->lock);
768 ReleaseWriteLock(&afs_xcbhash);
773 /* now merge in the resulting status back into the vnode.
774 * We only do this if the entry looks clear.
776 afs_ProcessFS(tvcp, &statsp[i], areqp);
777 #ifdef AFS_LINUX22_ENV
778 /* overwrite the ops if it's a directory or symlink. */
779 if (vType(tvcp) == VDIR)
780 tvcp->v.v_op = &afs_dir_iops;
781 else if (vType(tvcp) == VLNK)
782 tvcp->v.v_op = &afs_symlink_iops;
785 /* do some accounting for bulk stats: mark this entry as
786 * loaded, so we can tell if we use it before it gets
789 tvcp->states |= CBulkStat;
790 tvcp->states &= ~CBulkFetching;
794 /* merge in vol info */
795 if (volStates & VRO) tvcp->states |= CRO;
796 if (volStates & VBackup) tvcp->states |= CBackup;
797 if (volStates & VForeign) tvcp->states |= CForeign;
799 /* merge in the callback info */
800 tvcp->states |= CTruth;
802 /* get ptr to the callback we are interested in */
805 if (tcbp->ExpirationTime != 0) {
806 tvcp->cbExpires = tcbp->ExpirationTime+startTime;
807 tvcp->callback = hostp;
808 tvcp->states |= CStatd;
809 afs_QueueCallback(tvcp, CBHash(tcbp->ExpirationTime), volp);
811 else if (tvcp->states & CRO) {
812 /* ordinary callback on a read-only volume -- AFS 3.2 style */
813 tvcp->cbExpires = 3600+startTime;
814 tvcp->callback = hostp;
815 tvcp->states |= CStatd;
816 afs_QueueCallback(tvcp, CBHash(3600), volp);
820 tvcp->states &= ~(CStatd|CUnique);
821 afs_DequeueCallback(tvcp);
822 if ((tvcp->states & CForeign) || (vType(tvcp) == VDIR))
823 osi_dnlc_purgedp (tvcp); /* if it (could be) a directory */
825 ReleaseWriteLock(&afs_xcbhash);
827 ReleaseWriteLock(&tvcp->lock);
828 /* finally, we're done with the entry */
830 } /* for all files we got back */
832 /* finally return the pointer into the LRU queue */
833 afs_PutVCache(lruvcp);
836 /* Be sure to turn off the CBulkFetching flags */
837 for(i=flagIndex; i<fidIndex; i++) {
838 afid.Cell = adp->fid.Cell;
839 afid.Fid.Volume = adp->fid.Fid.Volume;
840 afid.Fid.Vnode = fidsp[i].Vnode;
841 afid.Fid.Unique = fidsp[i].Unique;
844 ObtainReadLock(&afs_xvcache);
845 tvcp = afs_FindVCache(&afid, 1, 0, &retry, 0/* !stats&!lru*/);
846 ReleaseReadLock(&afs_xvcache);
847 } while (tvcp && retry);
849 && (tvcp->states & CBulkFetching)
850 && (tvcp->m.Length == statSeqNo)) {
851 tvcp->states &= ~CBulkFetching;
858 afs_PutVolume(volp, READ_LOCK);
860 /* If we did the InlineBulk RPC pull out the return code */
861 if (inlinebulk && (&statsp[0])->errorCode) {
862 afs_Analyze(tcp, (&statsp[0])->errorCode, &adp->fid, areqp,
863 AFS_STATS_FS_RPCIDX_BULKSTATUS, SHARED_LOCK,
865 code = (&statsp[0])->errorCode;
867 osi_FreeLargeSpace(statMemp);
868 osi_FreeLargeSpace(cbfMemp);
872 /* was: (AFS_DEC_ENV) || defined(AFS_OSF30_ENV) || defined(AFS_NCR_ENV) */
878 struct nameidata *ndp; {
879 char aname[MAXNAMLEN+1]; /* XXX */
880 struct vcache **avcp = (struct vcache **)&(ndp->ni_vp);
881 struct ucred *acred = ndp->ni_cred;
882 int wantparent = ndp->ni_nameiop & WANTPARENT;
883 int opflag = ndp->ni_nameiop & OPFLAG;
884 #else /* AFS_OSF_ENV */
885 #if defined(AFS_SUN5_ENV) || defined(AFS_SGI_ENV)
886 afs_lookup(OSI_VC_ARG(adp), aname, avcp, pnp, flags, rdir, acred)
887 struct pathname *pnp;
892 afs_lookup(adp, aname, avcp, acred, flags)
895 afs_lookup(adp, aname, avcp, acred)
897 #endif /* SUN5 || SGI */
899 struct vcache **avcp;
901 struct AFS_UCRED *acred; {
903 struct vrequest treq;
904 char *tname = (char *)0;
905 register struct vcache *tvc=0;
906 register afs_int32 code;
907 register afs_int32 bulkcode = 0;
908 int pass = 0, hit = 0;
910 extern afs_int32 afs_mariner; /*Writing activity to log?*/
912 afs_hyper_t versionNo;
913 int no_read_access = 0;
914 struct sysname_info sysState; /* used only for @sys checking */
916 AFS_STATCNT(afs_lookup);
918 ndp->ni_dvp = (struct vnode *)adp;
919 memcpy(aname, ndp->ni_ptr, ndp->ni_namelen);
920 aname[ndp->ni_namelen] = '\0';
921 #endif /* AFS_OSF_ENV */
923 *avcp = (struct vcache *) 0; /* Since some callers don't initialize it */
925 if (code = afs_InitReq(&treq, acred)) {
929 /* come back to here if we encounter a non-existent object in a read-only
930 volume's directory */
933 *avcp = (struct vcache *) 0; /* Since some callers don't initialize it */
936 if (!(adp->states & CStatd)) {
937 if (code = afs_VerifyVCache2(adp, &treq)) {
943 /* watch for ".." in a volume root */
944 if (adp->mvstat == 2 && aname[0] == '.' && aname[1] == '.' && !aname[2]) {
945 /* looking up ".." in root via special hacks */
946 if (adp->mvid == (struct VenusFid *) 0 || adp->mvid->Fid.Volume == 0) {
948 extern struct vcache *afs_globalVp;
949 if (adp == afs_globalVp) {
950 struct vnode *rvp = (struct vnode *)adp;
952 ndp->ni_vp = rvp->v_vfsp->vfs_vnodecovered;
953 ndp->ni_dvp = ndp->ni_vp;
963 /* otherwise we have the fid here, so we use it */
964 tvc = afs_GetVCache(adp->mvid, &treq, (afs_int32 *)0,
965 (struct vcache*)0, 0);
966 afs_Trace3(afs_iclSetp, CM_TRACE_GETVCDOTDOT,
967 ICL_TYPE_FID, adp->mvid, ICL_TYPE_POINTER, tvc,
968 ICL_TYPE_INT32, code);
970 code = (tvc ? 0 : ENOENT);
972 if (tvc && !tvc->vrefCount) {
976 /*printf("LOOKUP GETVCDOTDOT -> %d\n", code);*/
981 /* now check the access */
982 if (treq.uid != adp->last_looker) {
983 if (!afs_AccessOK(adp, PRSFS_LOOKUP, &treq, CHECK_MODE_BITS)) {
984 *avcp = (struct vcache *)0;
988 else adp->last_looker = treq.uid;
991 /* Check for read access as well. We need read access in order to
992 stat files, but not to stat subdirectories. */
993 if (!afs_AccessOK(adp, PRSFS_READ, &treq, CHECK_MODE_BITS))
996 /* special case lookup of ".". Can we check for it sooner in this code,
997 * for instance, way up before "redo:" ??
998 * I'm not fiddling with the LRUQ here, either, perhaps I should, or else
999 * invent a lightweight version of GetVCache.
1001 if (aname[0] == '.' && !aname[1]) { /* special case */
1002 ObtainReadLock(&afs_xvcache);
1004 ReleaseReadLock(&afs_xvcache);
1008 if (adp && !adp->vrefCount) {
1014 Check_AtSys(adp, aname, &sysState, &treq);
1015 tname = sysState.name;
1017 /* 1st Check_AtSys and lookup by tname is required here, for now,
1018 because the dnlc is *not* told to remove entries for the parent
1019 dir of file/dir op that afs_LocalHero likes, but dnlc is informed
1020 if the cached entry for the parent dir is invalidated for a
1022 Otherwise, we'd be able to do a dnlc lookup on an entry ending
1023 w/@sys and know the dnlc was consistent with reality. */
1024 tvc = osi_dnlc_lookup (adp, tname, WRITE_LOCK);
1025 *avcp = tvc; /* maybe wasn't initialized, but it is now */
1027 if (no_read_access && vType(tvc) != VDIR && vType(tvc) != VLNK) {
1028 /* need read access on dir to stat non-directory / non-link */
1029 afs_PutVCache(tvc, WRITE_LOCK);
1030 *avcp = (struct vcache *)0;
1034 #ifdef AFS_LINUX22_ENV
1035 if (tvc->mvstat == 2) { /* we don't trust the dnlc for root vcaches */
1044 #else /* non - LINUX */
1048 #endif /* linux22 */
1052 register struct dcache *tdc;
1053 afs_int32 dirOffset, dirLen;
1055 struct VenusFid tfid;
1057 /* now we have to lookup the next fid */
1058 tdc = afs_GetDCache(adp, 0, &treq, &dirOffset, &dirLen, 1);
1060 *avcp = (struct vcache *)0; /* redundant, but harmless */
1065 /* now we will just call dir package with appropriate inode.
1066 Dirs are always fetched in their entirety for now */
1067 ObtainReadLock(&adp->lock);
1070 * Make sure that the data in the cache is current. There are two
1071 * cases we need to worry about:
1072 * 1. The cache data is being fetched by another process.
1073 * 2. The cache data is no longer valid
1075 while ((adp->states & CStatd)
1076 && (tdc->flags & DFFetching)
1077 && hsame(adp->m.DataVersion, tdc->f.versionNo)) {
1078 tdc->flags |= DFWaiting;
1079 ReleaseReadLock(&adp->lock);
1080 afs_osi_Sleep(&tdc->validPos);
1081 ObtainReadLock(&adp->lock);
1083 if (!(adp->states & CStatd)
1084 || !hsame(adp->m.DataVersion, tdc->f.versionNo)) {
1085 ReleaseReadLock(&adp->lock);
1090 /* Save the version number for when we call osi_dnlc_enter */
1091 hset(versionNo, tdc->f.versionNo);
1094 * check for, and handle "@sys" if it's there. We should be able
1095 * to avoid the alloc and the strcpy with a little work, but it's
1096 * not pressing. If there aren't any remote users (ie, via the
1097 * NFS translator), we have a slightly easier job.
1098 * the faster way to do this is to check for *aname == '@' and if
1099 * it's there, check for @sys, otherwise, assume there's no @sys
1100 * then, if the lookup fails, check for .*@sys...
1102 /* above now implemented by Check_AtSys and Next_AtSys */
1104 /* lookup the name in the appropriate dir, and return a cache entry
1105 on the resulting fid */
1106 theDir = tdc->f.inode;
1107 code = afs_dir_LookupOffset(&theDir, sysState.name, &tfid.Fid, &dirCookie);
1109 /* If the first lookup doesn't succeed, maybe it's got @sys in the name */
1110 while (code == ENOENT && Next_AtSys(adp, &treq, &sysState)) {
1111 code = afs_dir_LookupOffset(&theDir, sysState.name, &tfid.Fid, &dirCookie);
1113 tname = sysState.name;
1115 ReleaseReadLock(&adp->lock);
1118 /* new fid has same cell and volume */
1119 tfid.Cell = adp->fid.Cell;
1120 tfid.Fid.Volume = adp->fid.Fid.Volume;
1121 afs_Trace4(afs_iclSetp, CM_TRACE_LOOKUP, ICL_TYPE_POINTER, adp,
1122 ICL_TYPE_STRING, tname,
1123 ICL_TYPE_FID, &tfid, ICL_TYPE_INT32, code);
1126 if (code != ENOENT) {
1127 printf("LOOKUP dirLookupOff -> %d\n", code);
1132 /* prefetch some entries, if the dir is currently open. The variable
1133 * dirCookie tells us where to start prefetching from.
1135 if (AFSDOBULK && adp->opens > 0 && !(adp->states & CForeign)) {
1137 /* if the entry is not in the cache, or is in the cache,
1138 * but hasn't been statd, then do a bulk stat operation.
1142 ObtainReadLock(&afs_xvcache);
1143 tvc = afs_FindVCache(&tfid, 1, 0, &retry, 0/* !stats,!lru */);
1144 ReleaseReadLock(&afs_xvcache);
1145 } while (tvc && retry);
1147 if (!tvc || !(tvc->states & CStatd))
1148 bulkcode = afs_DoBulkStat(adp, dirCookie, &treq);
1152 /* if the vcache isn't usable, release it */
1153 if (tvc && !(tvc->states & CStatd)) {
1155 tvc = (struct vcache *) 0;
1158 tvc = (struct vcache *) 0;
1162 /* now get the status info, if we don't already have it */
1163 /* This is kind of weird, but we might wind up accidentally calling
1164 * RXAFS_Lookup because we happened upon a file which legitimately
1165 * has a 0 uniquifier. That is the result of allowing unique to wrap
1166 * to 0. This was fixed in AFS 3.4. For CForeign, Unique == 0 means that
1167 * the file has not yet been looked up.
1170 afs_int32 cached = 0;
1171 if (!tfid.Fid.Unique && (adp->states & CForeign)) {
1172 tvc = afs_LookupVCache(&tfid, &treq, &cached, WRITE_LOCK,
1175 if (!tvc && !bulkcode) { /* lookup failed or wasn't called */
1176 tvc = afs_GetVCache(&tfid, &treq, &cached, (struct vcache*)0,
1180 } /* sub-block just to reduce stack usage */
1183 if (adp->states & CForeign)
1184 tvc->states |= CForeign;
1185 tvc->parentVnode = adp->fid.Fid.Vnode;
1186 tvc->parentUnique = adp->fid.Fid.Unique;
1187 tvc->states &= ~CBulkStat;
1189 #if defined(UKERNEL) && defined(AFS_WEB_ENHANCEMENTS)
1190 if (!(flags & AFS_LOOKUP_NOEVAL))
1191 /* don't eval mount points */
1192 #endif /* UKERNEL && AFS_WEB_ENHANCEMENTS */
1193 if (tvc->mvstat == 1) {
1194 /* a mt point, possibly unevaluated */
1195 struct volume *tvolp;
1197 ObtainWriteLock(&tvc->lock,133);
1198 code = EvalMountPoint(tvc, adp, &tvolp, &treq);
1199 ReleaseWriteLock(&tvc->lock);
1202 afs_PutVCache(tvc, WRITE_LOCK);
1203 if (tvolp) afs_PutVolume(tvolp, WRITE_LOCK);
1207 /* next, we want to continue using the target of the mt point */
1208 if (tvc->mvid && (tvc->states & CMValid)) {
1210 /* now lookup target, to set .. pointer */
1211 afs_Trace2(afs_iclSetp, CM_TRACE_LOOKUP1,
1212 ICL_TYPE_POINTER, tvc, ICL_TYPE_FID, &tvc->fid);
1213 uvc = tvc; /* remember for later */
1215 if (tvolp && (tvolp->states & VForeign)) {
1216 /* XXXX tvolp has ref cnt on but not locked! XXX */
1217 tvc = afs_GetRootVCache(tvc->mvid, &treq, (afs_int32 *)0, tvolp, WRITE_LOCK);
1219 tvc = afs_GetVCache(tvc->mvid, &treq, (afs_int32 *)0,
1220 (struct vcache*)0, WRITE_LOCK);
1222 afs_PutVCache(uvc, WRITE_LOCK); /* we're done with it */
1227 afs_PutVolume(tvolp, WRITE_LOCK);
1232 /* now, if we came via a new mt pt (say because of a new
1233 * release of a R/O volume), we must reevaluate the ..
1234 * ptr to point back to the appropriate place */
1236 ObtainWriteLock(&tvc->lock,134);
1237 if (tvc->mvid == (struct VenusFid *) 0) {
1238 tvc->mvid = (struct VenusFid *) osi_AllocSmallSpace(sizeof(struct VenusFid));
1240 /* setup backpointer */
1241 *tvc->mvid = tvolp->dotdot;
1242 ReleaseWriteLock(&tvc->lock);
1243 afs_PutVolume(tvolp, WRITE_LOCK);
1247 afs_PutVCache(tvc, WRITE_LOCK);
1249 if (tvolp) afs_PutVolume(tvolp, WRITE_LOCK);
1254 if (tvc && !tvc->vrefCount) {
1260 /* if we get here, we found something in a directory that couldn't
1261 be located (a Multics "connection failure"). If the volume is
1262 read-only, we try flushing this entry from the cache and trying
1266 tv = afs_GetVolume(&adp->fid, &treq, READ_LOCK);
1268 if (tv->states & VRO) {
1269 pass = 1; /* try this *once* */
1270 ObtainWriteLock(&afs_xcbhash, 495);
1271 afs_DequeueCallback(adp);
1272 /* re-stat to get later version */
1273 adp->states &= ~CStatd;
1274 ReleaseWriteLock(&afs_xcbhash);
1275 osi_dnlc_purgedp(adp);
1276 afs_PutVolume(tv, READ_LOCK);
1279 afs_PutVolume(tv, READ_LOCK);
1286 /* put the network buffer back, if need be */
1287 if (tname != aname && tname) osi_FreeLargeSpace(tname);
1290 /* Handle RENAME; only need to check rename "." */
1291 if (opflag == RENAME && wantparent && *ndp->ni_next == 0) {
1292 if (!FidCmp(&(tvc->fid), &(adp->fid))) {
1293 afs_PutVCache(*avcp, WRITE_LOCK);
1295 return afs_CheckCode(EISDIR, &treq, 18);
1298 #endif /* AFS_OSF_ENV */
1301 afs_AddMarinerName(aname, tvc);
1303 #if defined(UKERNEL) && defined(AFS_WEB_ENHANCEMENTS)
1304 if (!(flags & AFS_LOOKUP_NOEVAL))
1305 /* Here we don't enter the name into the DNLC because we want the
1306 evaluated mount dir to be there (the vcache for the mounted volume)
1307 rather than the vc of the mount point itself. we can still find the
1308 mount point's vc in the vcache by its fid. */
1309 #endif /* UKERNEL && AFS_WEB_ENHANCEMENTS */
1311 osi_dnlc_enter (adp, aname, tvc, &versionNo);
1314 #ifdef AFS_LINUX20_ENV
1315 /* So Linux inode cache is up to date. */
1316 code = afs_VerifyVCache(tvc, &treq);
1318 return 0; /* can't have been any errors if hit and !code */
1322 if (bulkcode) code = bulkcode; else
1323 code = afs_CheckCode(code, &treq, 19);
1325 /* If there is an error, make sure *avcp is null.
1326 * Alphas panic otherwise - defect 10719.
1328 *avcp = (struct vcache *)0;