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 ENOENT;
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 void 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 */
402 /* first compute some basic parameters. We dont want to prefetch more
403 * than a fraction of the cache in any given call, and we want to preserve
404 * a portion of the LRU queue in any event, so as to avoid thrashing
405 * the entire stat cache (we will at least leave some of it alone).
406 * presently dont stat more than 1/8 the cache in any one call. */
407 nentries = afs_cacheStats / 8;
409 /* dont bother prefetching more than one calls worth of info */
410 if (nentries > AFSCBMAX) nentries = AFSCBMAX;
412 /* heuristic to make sure that things fit in 4K. This means that
413 * we shouldnt make it any bigger than 47 entries. I am typically
414 * going to keep it a little lower, since we don't want to load
415 * too much of the stat cache.
417 if (nentries > 30) nentries = 30;
419 /* now, to reduce the stack size, well allocate two 4K blocks,
420 * one for fids and callbacks, and one for stat info. Well set
421 * up our pointers to the memory from there, too.
423 statMemp = osi_AllocLargeSpace(nentries * sizeof(AFSFetchStatus));
424 statsp = (struct AFSFetchStatus *) statMemp;
425 cbfMemp = osi_AllocLargeSpace(nentries *
426 (sizeof(AFSCallBack) + sizeof(AFSFid)));
427 fidsp = (AFSFid *) cbfMemp;
428 cbsp = (AFSCallBack *) (cbfMemp + nentries * sizeof(AFSFid));
430 /* next, we must iterate over the directory, starting from the specified
431 * cookie offset (dirCookie), and counting out nentries file entries.
432 * We skip files that already have stat cache entries, since we
433 * dont want to bulk stat files that are already in the cache.
436 code = afs_VerifyVCache(adp, areqp);
439 dcp = afs_GetDCache(adp, 0, areqp, &temp, &temp, 1);
445 /* lock the directory cache entry */
446 ObtainReadLock(&adp->lock);
449 * Make sure that the data in the cache is current. There are two
450 * cases we need to worry about:
451 * 1. The cache data is being fetched by another process.
452 * 2. The cache data is no longer valid
454 while ((adp->states & CStatd)
455 && (dcp->flags & DFFetching)
456 && hsame(adp->m.DataVersion, dcp->f.versionNo)) {
457 dcp->flags |= DFWaiting;
458 ReleaseReadLock(&adp->lock);
459 afs_osi_Sleep(&dcp->validPos);
460 ObtainReadLock(&adp->lock);
462 if (!(adp->states & CStatd)
463 || !hsame(adp->m.DataVersion, dcp->f.versionNo)) {
464 ReleaseReadLock(&adp->lock);
469 /* Generate a sequence number so we can tell whether we should
470 * store the attributes when processing the response. This number is
471 * stored in the file size when we set the CBulkFetching bit. If the
472 * CBulkFetching is still set and this value hasn't changed, then
473 * we know we were the last to set CBulkFetching bit for this file,
474 * and it is safe to set the status information for this file.
476 statSeqNo = bulkStatCounter++;
478 /* now we have dir data in the cache, so scan the dir page */
481 while (1) { /* Should probably have some constant bound */
482 /* look for first safe entry to examine in the directory. BlobScan
483 * looks for a the 1st allocated dir after the dirCookie slot.
485 newIndex = BlobScan(&dcp->f.inode, (dirCookie>>5));
486 if (newIndex == 0) break;
488 /* remember the updated directory cookie */
489 dirCookie = newIndex << 5;
491 /* get a ptr to the dir entry */
492 dirEntryp =(struct DirEntry *)afs_dir_GetBlob(&dcp->f.inode, newIndex);
493 if (!dirEntryp) break;
495 /* dont copy more than we have room for */
496 if (fidIndex >= nentries) {
497 DRelease((char *) dirEntryp, 0);
501 /* now, if the dir entry looks good, copy it out to our list. Vnode
502 * 0 means deleted, although it should also be free were it deleted.
504 if (dirEntryp->fid.vnode != 0) {
505 /* dont copy entries we have in our cache. This check will
506 * also make us skip "." and probably "..", unless it has
507 * disappeared from the cache since we did our namei call.
509 tfid.Cell = adp->fid.Cell;
510 tfid.Fid.Volume = adp->fid.Fid.Volume;
511 tfid.Fid.Vnode = ntohl(dirEntryp->fid.vnode);
512 tfid.Fid.Unique = ntohl(dirEntryp->fid.vunique);
515 ObtainWriteLock(&afs_xvcache, 130);
516 tvcp = afs_FindVCache(&tfid, 0, 0, &retry, 0 /* no stats | LRU */);
518 ReleaseWriteLock(&afs_xvcache);
521 } while (tvcp && retry);
522 if (!tvcp) { /* otherwise, create manually */
523 tvcp = afs_NewVCache(&tfid, hostp, 0, 0);
524 ObtainWriteLock(&tvcp->lock, 505);
525 ReleaseWriteLock(&afs_xvcache);
526 afs_RemoveVCB(&tfid);
527 ReleaseWriteLock(&tvcp->lock);
529 ReleaseWriteLock(&afs_xvcache);
532 goto done; /* can't happen at present, more's the pity */
534 /* WARNING: afs_DoBulkStat uses the Length field to store a
535 * sequence number for each bulk status request. Under no
536 * circumstances should afs_DoBulkStat store a sequence number
537 * if the new length will be ignored when afs_ProcessFS is
538 * called with new stats. */
540 if (!(tvcp->states & (CStatd|CBulkFetching))
541 && (tvcp->execsOrWriters <= 0)
542 && !afs_DirtyPages(tvcp)
543 && !AFS_VN_MAPPED((vnode_t*)tvcp))
545 if (!(tvcp->states & (CStatd|CBulkFetching))
546 && (tvcp->execsOrWriters <= 0)
547 && !afs_DirtyPages(tvcp))
551 /* this entry doesnt exist in the cache, and is not
552 * already being fetched by someone else, so add it to the
553 * list of file IDs to obtain.
555 * We detect a callback breaking race condition by checking the
556 * CBulkFetching state bit and the value in the file size.
557 * It is safe to set the status only if the CBulkFetching
558 * flag is still set and the value in the file size does
561 * Don't fetch status for dirty files. We need to
562 * preserve the value of the file size. We could
563 * flush the pages, but it wouldn't be worthwhile.
565 memcpy((char *)(fidsp+fidIndex), (char *) &tfid.Fid, sizeof(*fidsp));
566 tvcp->states |= CBulkFetching;
567 tvcp->m.Length = statSeqNo;
571 } /* if dir vnode has non-zero entry */
573 /* move to the next dir entry by adding in the # of entries
574 * used by this dir entry.
576 temp = afs_dir_NameBlobs(dirEntryp->name) << 5;
577 DRelease((char *) dirEntryp, 0);
578 if (temp <= 0) break;
580 } /* while loop over all dir entries */
582 /* now release the dir lock and prepare to make the bulk RPC */
583 ReleaseReadLock(&adp->lock);
585 /* release the chunk */
588 /* dont make a null call */
589 if (fidIndex == 0) goto done;
592 /* setup the RPC parm structures */
593 fidParm.AFSCBFids_len = fidIndex;
594 fidParm.AFSCBFids_val = fidsp;
595 statParm.AFSBulkStats_len = fidIndex;
596 statParm.AFSBulkStats_val = statsp;
597 cbParm.AFSCBs_len = fidIndex;
598 cbParm.AFSCBs_val = cbsp;
600 /* start the timer; callback expirations are relative to this */
601 startTime = osi_Time();
603 tcp = afs_Conn(&adp->fid, areqp, SHARED_LOCK);
605 hostp = tcp->srvr->server;
606 XSTATS_START_TIME(AFS_STATS_FS_RPCIDX_BULKSTATUS);
607 #ifdef RX_ENABLE_LOCKS
609 #endif /* RX_ENABLE_LOCKS */
610 code = RXAFS_BulkStatus(tcp->id, &fidParm, &statParm, &cbParm,
612 #ifdef RX_ENABLE_LOCKS
614 #endif /* RX_ENABLE_LOCKS */
618 } while (afs_Analyze(tcp, code, &adp->fid, areqp,
619 AFS_STATS_FS_RPCIDX_BULKSTATUS, SHARED_LOCK, (struct cell *)0));
621 /* now, if we didnt get the info, bail out. */
624 /* we need vol flags to create the entries properly */
625 dotdot.Fid.Volume = 0;
626 volp = afs_GetVolume(&adp->fid, areqp, READ_LOCK);
628 volStates = volp->states;
629 if (volp->dotdot.Fid.Volume != 0)
630 dotdot = volp->dotdot;
634 /* find the place to merge the info into We do this by skipping
635 * nskip entries in the LRU queue. The more we skip, the more
636 * we preserve, since the head of the VLRU queue is the most recently
640 nskip = afs_cacheStats / 2; /* preserved fraction of the cache */
641 ObtainReadLock(&afs_xvcache);
643 /* actually a serious error, probably should panic. Probably will
644 * panic soon, oh well. */
645 ReleaseReadLock(&afs_xvcache);
646 afs_warnuser("afs_DoBulkStat: VLRU empty!");
649 if ((VLRU.next->prev != &VLRU) || (VLRU.prev->next != &VLRU)) {
650 refpanic ("Bulkstat VLRU inconsistent");
652 for(tq = VLRU.next; tq != &VLRU; tq = QNext(tq)) {
653 if (--nskip <= 0) break;
654 else if (QNext(QPrev(tq)) != tq) {
656 refpanic ("BulkStat VLRU inconsistent");
659 if (tq != &VLRU) lruvcp = QTOV(tq);
660 else lruvcp = QTOV(VLRU.next);
662 /* now we have to hold this entry, so that it does not get moved
663 * into the free list while we're running. It could still get
664 * moved within the lru queue, but hopefully that will be rare; it
665 * doesn't hurt nearly as much.
668 osi_vnhold(lruvcp, &retry);
669 ReleaseReadLock(&afs_xvcache); /* could be read lock */
673 /* otherwise, merge in the info. We have to be quite careful here,
674 * since we need to ensure that we don't merge old info over newer
675 * stuff in a stat cache entry. We're very conservative here: we don't
676 * do the merge at all unless we ourselves create the stat cache
677 * entry. That's pretty safe, and should work pretty well, since we
678 * typically expect to do the stat cache creation ourselves.
680 * We also have to take into account racing token revocations.
682 for(i=0; i<fidIndex; i++) {
683 afid.Cell = adp->fid.Cell;
684 afid.Fid.Volume = adp->fid.Fid.Volume;
685 afid.Fid.Vnode = fidsp[i].Vnode;
686 afid.Fid.Unique = fidsp[i].Unique;
689 ObtainReadLock(&afs_xvcache);
690 tvcp = afs_FindVCache(&afid, 1, 0, &retry, 0/* !stats&!lru*/);
691 ReleaseReadLock(&afs_xvcache);
692 } while (tvcp && retry);
694 /* The entry may no longer exist */
699 /* now we have the entry held, but we need to fill it in */
700 ObtainWriteLock(&tvcp->lock,131);
702 /* if CBulkFetching is not set, or if the file size no longer
703 * matches the value we placed there when we set the CBulkFetching
704 * flag, then someone else has done something with this node,
705 * and we may not have the latest status information for this
706 * file. Leave the entry alone.
708 if (!(tvcp->states & CBulkFetching) || (tvcp->m.Length != statSeqNo)) {
710 ReleaseWriteLock(&tvcp->lock);
715 /* now copy ".." entry back out of volume structure, if necessary */
716 if (tvcp->mvstat == 2 && (dotdot.Fid.Volume != 0)) {
718 tvcp->mvid = (struct VenusFid *) osi_AllocSmallSpace(sizeof(struct VenusFid));
719 *tvcp->mvid = dotdot;
722 ObtainWriteLock(&afs_xvcache,132);
723 if ((VLRU.next->prev != &VLRU) || (VLRU.prev->next != &VLRU)) {
724 refpanic ("Bulkstat VLRU inconsistent2");
726 if ((QNext(QPrev(&tvcp->vlruq)) != &tvcp->vlruq)
727 || (QPrev(QNext(&tvcp->vlruq)) != &tvcp->vlruq))
728 refpanic ("Bulkstat VLRU inconsistent4");
729 if ((QNext(QPrev(&lruvcp->vlruq)) != &lruvcp->vlruq)
730 || (QPrev(QNext(&lruvcp->vlruq)) != &lruvcp->vlruq))
731 refpanic ("Bulkstat VLRU inconsistent5");
733 if (tvcp != lruvcp) { /* if they are == don't move it, don't corrupt vlru */
734 QRemove(&tvcp->vlruq);
735 QAdd(&lruvcp->vlruq, &tvcp->vlruq);
738 if ((VLRU.next->prev != &VLRU) || (VLRU.prev->next != &VLRU)) {
739 refpanic ("Bulkstat VLRU inconsistent3");
741 if ((QNext(QPrev(&tvcp->vlruq)) != &tvcp->vlruq)
742 || (QPrev(QNext(&tvcp->vlruq)) != &tvcp->vlruq))
743 refpanic ("Bulkstat VLRU inconsistent5");
744 if ((QNext(QPrev(&lruvcp->vlruq)) != &lruvcp->vlruq)
745 || (QPrev(QNext(&lruvcp->vlruq)) != &lruvcp->vlruq))
746 refpanic ("Bulkstat VLRU inconsistent6");
747 ReleaseWriteLock(&afs_xvcache);
749 ObtainWriteLock(&afs_xcbhash, 494);
751 /* We need to check the flags again. We may have missed
752 * something while we were waiting for a lock.
754 if (!(tvcp->states & CBulkFetching) || (tvcp->m.Length != statSeqNo)) {
756 ReleaseWriteLock(&tvcp->lock);
757 ReleaseWriteLock(&afs_xcbhash);
762 /* now merge in the resulting status back into the vnode.
763 * We only do this if the entry looks clear.
765 afs_ProcessFS(tvcp, &statsp[i], areqp);
766 #ifdef AFS_LINUX22_ENV
767 /* overwrite the ops if it's a directory or symlink. */
768 if (vType(tvcp) == VDIR)
769 tvcp->v.v_op = &afs_dir_iops;
770 else if (vType(tvcp) == VLNK)
771 tvcp->v.v_op = &afs_symlink_iops;
774 /* do some accounting for bulk stats: mark this entry as
775 * loaded, so we can tell if we use it before it gets
778 tvcp->states |= CBulkStat;
779 tvcp->states &= ~CBulkFetching;
783 /* merge in vol info */
784 if (volStates & VRO) tvcp->states |= CRO;
785 if (volStates & VBackup) tvcp->states |= CBackup;
786 if (volStates & VForeign) tvcp->states |= CForeign;
788 /* merge in the callback info */
789 tvcp->states |= CTruth;
791 /* get ptr to the callback we are interested in */
794 if (tcbp->ExpirationTime != 0) {
795 tvcp->cbExpires = tcbp->ExpirationTime+startTime;
796 tvcp->callback = hostp;
797 tvcp->states |= CStatd;
798 afs_QueueCallback(tvcp, CBHash(tcbp->ExpirationTime), volp);
800 else if (tvcp->states & CRO) {
801 /* ordinary callback on a read-only volume -- AFS 3.2 style */
802 tvcp->cbExpires = 3600+startTime;
803 tvcp->callback = hostp;
804 tvcp->states |= CStatd;
805 afs_QueueCallback(tvcp, CBHash(3600), volp);
809 tvcp->states &= ~(CStatd|CUnique);
810 afs_DequeueCallback(tvcp);
811 if ((tvcp->states & CForeign) || (vType(tvcp) == VDIR))
812 osi_dnlc_purgedp (tvcp); /* if it (could be) a directory */
814 ReleaseWriteLock(&afs_xcbhash);
816 ReleaseWriteLock(&tvcp->lock);
817 /* finally, we're done with the entry */
819 } /* for all files we got back */
821 /* finally return the pointer into the LRU queue */
822 afs_PutVCache(lruvcp);
825 /* Be sure to turn off the CBulkFetching flags */
826 for(i=flagIndex; i<fidIndex; i++) {
827 afid.Cell = adp->fid.Cell;
828 afid.Fid.Volume = adp->fid.Fid.Volume;
829 afid.Fid.Vnode = fidsp[i].Vnode;
830 afid.Fid.Unique = fidsp[i].Unique;
833 ObtainReadLock(&afs_xvcache);
834 tvcp = afs_FindVCache(&afid, 1, 0, &retry, 0/* !stats&!lru*/);
835 ReleaseReadLock(&afs_xvcache);
836 } while (tvcp && retry);
838 && (tvcp->states & CBulkFetching)
839 && (tvcp->m.Length == statSeqNo)) {
840 tvcp->states &= ~CBulkFetching;
847 afs_PutVolume(volp, READ_LOCK);
849 osi_FreeLargeSpace(statMemp);
850 osi_FreeLargeSpace(cbfMemp);
853 /* was: (AFS_DEC_ENV) || defined(AFS_OSF30_ENV) || defined(AFS_NCR_ENV) */
859 struct nameidata *ndp; {
860 char aname[MAXNAMLEN+1]; /* XXX */
861 struct vcache **avcp = (struct vcache **)&(ndp->ni_vp);
862 struct ucred *acred = ndp->ni_cred;
863 int wantparent = ndp->ni_nameiop & WANTPARENT;
864 int opflag = ndp->ni_nameiop & OPFLAG;
865 #else /* AFS_OSF_ENV */
866 #if defined(AFS_SUN5_ENV) || defined(AFS_SGI_ENV)
867 afs_lookup(OSI_VC_ARG(adp), aname, avcp, pnp, flags, rdir, acred)
868 struct pathname *pnp;
873 afs_lookup(adp, aname, avcp, acred, flags)
876 afs_lookup(adp, aname, avcp, acred)
878 #endif /* SUN5 || SGI */
880 struct vcache **avcp;
882 struct AFS_UCRED *acred; {
884 struct vrequest treq;
885 char *tname = (char *)0;
886 register struct vcache *tvc=0;
887 register afs_int32 code;
888 int pass = 0, hit = 0;
890 extern afs_int32 afs_mariner; /*Writing activity to log?*/
892 afs_hyper_t versionNo;
893 int no_read_access = 0;
894 struct sysname_info sysState; /* used only for @sys checking */
896 AFS_STATCNT(afs_lookup);
898 ndp->ni_dvp = (struct vnode *)adp;
899 memcpy(aname, ndp->ni_ptr, ndp->ni_namelen);
900 aname[ndp->ni_namelen] = '\0';
901 #endif /* AFS_OSF_ENV */
903 *avcp = (struct vcache *) 0; /* Since some callers don't initialize it */
905 if (code = afs_InitReq(&treq, acred)) {
909 /* come back to here if we encounter a non-existent object in a read-only
910 volume's directory */
913 *avcp = (struct vcache *) 0; /* Since some callers don't initialize it */
915 if (!(adp->states & CStatd)) {
916 if (code = afs_VerifyVCache2(adp, &treq))
921 /* watch for ".." in a volume root */
922 if (adp->mvstat == 2 && aname[0] == '.' && aname[1] == '.' && !aname[2]) {
923 /* looking up ".." in root via special hacks */
924 if (adp->mvid == (struct VenusFid *) 0 || adp->mvid->Fid.Volume == 0) {
926 extern struct vcache *afs_globalVp;
927 if (adp == afs_globalVp) {
928 struct vnode *rvp = (struct vnode *)adp;
930 ndp->ni_vp = rvp->v_vfsp->vfs_vnodecovered;
931 ndp->ni_dvp = ndp->ni_vp;
941 /* otherwise we have the fid here, so we use it */
942 tvc = afs_GetVCache(adp->mvid, &treq, (afs_int32 *)0,
943 (struct vcache*)0, 0);
944 afs_Trace3(afs_iclSetp, CM_TRACE_GETVCDOTDOT,
945 ICL_TYPE_FID, adp->mvid, ICL_TYPE_POINTER, tvc,
946 ICL_TYPE_INT32, code);
948 code = (tvc ? 0 : ENOENT);
950 if (tvc && !tvc->vrefCount) {
954 /*printf("LOOKUP GETVCDOTDOT -> %d\n", code);*/
959 /* now check the access */
960 if (treq.uid != adp->last_looker) {
961 if (!afs_AccessOK(adp, PRSFS_LOOKUP, &treq, CHECK_MODE_BITS)) {
962 *avcp = (struct vcache *)0;
966 else adp->last_looker = treq.uid;
969 /* Check for read access as well. We need read access in order to
970 stat files, but not to stat subdirectories. */
971 if (!afs_AccessOK(adp, PRSFS_READ, &treq, CHECK_MODE_BITS))
974 /* special case lookup of ".". Can we check for it sooner in this code,
975 * for instance, way up before "redo:" ??
976 * I'm not fiddling with the LRUQ here, either, perhaps I should, or else
977 * invent a lightweight version of GetVCache.
979 if (aname[0] == '.' && !aname[1]) { /* special case */
980 ObtainReadLock(&afs_xvcache);
982 ReleaseReadLock(&afs_xvcache);
986 if (adp && !adp->vrefCount) {
992 Check_AtSys(adp, aname, &sysState, &treq);
993 tname = sysState.name;
995 /* 1st Check_AtSys and lookup by tname is required here, for now,
996 because the dnlc is *not* told to remove entries for the parent
997 dir of file/dir op that afs_LocalHero likes, but dnlc is informed
998 if the cached entry for the parent dir is invalidated for a
1000 Otherwise, we'd be able to do a dnlc lookup on an entry ending
1001 w/@sys and know the dnlc was consistent with reality. */
1002 tvc = osi_dnlc_lookup (adp, tname, WRITE_LOCK);
1003 *avcp = tvc; /* maybe wasn't initialized, but it is now */
1005 if (no_read_access && vType(tvc) != VDIR && vType(tvc) != VLNK) {
1006 /* need read access on dir to stat non-directory / non-link */
1007 afs_PutVCache(tvc, WRITE_LOCK);
1008 *avcp = (struct vcache *)0;
1012 #ifdef AFS_LINUX22_ENV
1013 if (tvc->mvstat == 2) { /* we don't trust the dnlc for root vcaches */
1022 #else /* non - LINUX */
1026 #endif /* linux22 */
1030 register struct dcache *tdc;
1031 afs_int32 dirOffset, dirLen;
1033 struct VenusFid tfid;
1035 /* now we have to lookup the next fid */
1036 tdc = afs_GetDCache(adp, 0, &treq, &dirOffset, &dirLen, 1);
1038 *avcp = (struct vcache *)0; /* redundant, but harmless */
1043 /* now we will just call dir package with appropriate inode.
1044 Dirs are always fetched in their entirety for now */
1045 ObtainReadLock(&adp->lock);
1048 * Make sure that the data in the cache is current. There are two
1049 * cases we need to worry about:
1050 * 1. The cache data is being fetched by another process.
1051 * 2. The cache data is no longer valid
1053 while ((adp->states & CStatd)
1054 && (tdc->flags & DFFetching)
1055 && hsame(adp->m.DataVersion, tdc->f.versionNo)) {
1056 tdc->flags |= DFWaiting;
1057 ReleaseReadLock(&adp->lock);
1058 afs_osi_Sleep(&tdc->validPos);
1059 ObtainReadLock(&adp->lock);
1061 if (!(adp->states & CStatd)
1062 || !hsame(adp->m.DataVersion, tdc->f.versionNo)) {
1063 ReleaseReadLock(&adp->lock);
1068 /* Save the version number for when we call osi_dnlc_enter */
1069 hset(versionNo, tdc->f.versionNo);
1072 * check for, and handle "@sys" if it's there. We should be able
1073 * to avoid the alloc and the strcpy with a little work, but it's
1074 * not pressing. If there aren't any remote users (ie, via the
1075 * NFS translator), we have a slightly easier job.
1076 * the faster way to do this is to check for *aname == '@' and if
1077 * it's there, check for @sys, otherwise, assume there's no @sys
1078 * then, if the lookup fails, check for .*@sys...
1080 /* above now implemented by Check_AtSys and Next_AtSys */
1082 /* lookup the name in the appropriate dir, and return a cache entry
1083 on the resulting fid */
1084 theDir = tdc->f.inode;
1085 code = afs_dir_LookupOffset(&theDir, sysState.name, &tfid.Fid, &dirCookie);
1087 /* If the first lookup doesn't succeed, maybe it's got @sys in the name */
1088 while (code == ENOENT && Next_AtSys(adp, &treq, &sysState)) {
1089 code = afs_dir_LookupOffset(&theDir, sysState.name, &tfid.Fid, &dirCookie);
1091 tname = sysState.name;
1093 ReleaseReadLock(&adp->lock);
1096 /* new fid has same cell and volume */
1097 tfid.Cell = adp->fid.Cell;
1098 tfid.Fid.Volume = adp->fid.Fid.Volume;
1099 afs_Trace4(afs_iclSetp, CM_TRACE_LOOKUP, ICL_TYPE_POINTER, adp,
1100 ICL_TYPE_STRING, tname,
1101 ICL_TYPE_FID, &tfid, ICL_TYPE_INT32, code);
1104 if (code != ENOENT) {
1105 printf("LOOKUP dirLookupOff -> %d\n", code);
1110 /* prefetch some entries, if the dir is currently open. The variable
1111 * dirCookie tells us where to start prefetching from.
1113 if (AFSDOBULK && adp->opens > 0 && !(adp->states & CForeign)) {
1115 /* if the entry is not in the cache, or is in the cache,
1116 * but hasn't been statd, then do a bulk stat operation.
1120 ObtainReadLock(&afs_xvcache);
1121 tvc = afs_FindVCache(&tfid, 1, 0, &retry, 0/* !stats,!lru */);
1122 ReleaseReadLock(&afs_xvcache);
1123 } while (tvc && retry);
1125 if (!tvc || !(tvc->states & CStatd)) {
1126 afs_DoBulkStat(adp, dirCookie, &treq);
1129 /* if the vcache isn't usable, release it */
1130 if (tvc && !(tvc->states & CStatd)) {
1132 tvc = (struct vcache *) 0;
1135 else tvc = (struct vcache *) 0;
1137 /* now get the status info, if we don't already have it */
1138 /* This is kind of weird, but we might wind up accidentally calling
1139 * RXAFS_Lookup because we happened upon a file which legitimately
1140 * has a 0 uniquifier. That is the result of allowing unique to wrap
1141 * to 0. This was fixed in AFS 3.4. For CForeigh, Unique == 0 means that
1142 * the file has not yet been looked up.
1145 afs_int32 cached = 0;
1146 if (!tfid.Fid.Unique && (adp->states & CForeign)) {
1147 tvc = afs_LookupVCache(&tfid, &treq, &cached, WRITE_LOCK,
1150 if (!tvc) { /* lookup failed or wasn't called */
1151 tvc = afs_GetVCache(&tfid, &treq, &cached, (struct vcache*)0,
1155 } /* sub-block just to reduce stack usage */
1158 if (adp->states & CForeign)
1159 tvc->states |= CForeign;
1160 tvc->parentVnode = adp->fid.Fid.Vnode;
1161 tvc->parentUnique = adp->fid.Fid.Unique;
1162 tvc->states &= ~CBulkStat;
1164 #if defined(UKERNEL) && defined(AFS_WEB_ENHANCEMENTS)
1165 if (!(flags & AFS_LOOKUP_NOEVAL))
1166 /* don't eval mount points */
1167 #endif /* UKERNEL && AFS_WEB_ENHANCEMENTS */
1168 if (tvc->mvstat == 1) {
1169 /* a mt point, possibly unevaluated */
1170 struct volume *tvolp;
1172 ObtainWriteLock(&tvc->lock,133);
1173 code = EvalMountPoint(tvc, adp, &tvolp, &treq);
1174 ReleaseWriteLock(&tvc->lock);
1177 if (tvolp) afs_PutVolume(tvolp, WRITE_LOCK);
1181 /* next, we want to continue using the target of the mt point */
1182 if (tvc->mvid && (tvc->states & CMValid)) {
1184 /* now lookup target, to set .. pointer */
1185 afs_Trace2(afs_iclSetp, CM_TRACE_LOOKUP1,
1186 ICL_TYPE_POINTER, tvc, ICL_TYPE_FID, &tvc->fid);
1187 uvc = tvc; /* remember for later */
1189 if (tvolp && (tvolp->states & VForeign)) {
1190 /* XXXX tvolp has ref cnt on but not locked! XXX */
1191 tvc = afs_GetRootVCache(tvc->mvid, &treq, (afs_int32 *)0, tvolp, WRITE_LOCK);
1193 tvc = afs_GetVCache(tvc->mvid, &treq, (afs_int32 *)0,
1194 (struct vcache*)0, WRITE_LOCK);
1196 afs_PutVCache(uvc, WRITE_LOCK); /* we're done with it */
1201 afs_PutVolume(tvolp, WRITE_LOCK);
1206 /* now, if we came via a new mt pt (say because of a new
1207 * release of a R/O volume), we must reevaluate the ..
1208 * ptr to point back to the appropriate place */
1210 ObtainWriteLock(&tvc->lock,134);
1211 if (tvc->mvid == (struct VenusFid *) 0) {
1212 tvc->mvid = (struct VenusFid *) osi_AllocSmallSpace(sizeof(struct VenusFid));
1214 /* setup backpointer */
1215 *tvc->mvid = tvolp->dotdot;
1216 ReleaseWriteLock(&tvc->lock);
1217 afs_PutVolume(tvolp, WRITE_LOCK);
1221 afs_PutVCache(tvc, WRITE_LOCK);
1223 if (tvolp) afs_PutVolume(tvolp, WRITE_LOCK);
1228 if (tvc && !tvc->vrefCount) {
1234 /* if we get here, we found something in a directory that couldn't
1235 be located (a Multics "connection failure"). If the volume is
1236 read-only, we try flushing this entry from the cache and trying
1240 tv = afs_GetVolume(&adp->fid, &treq, READ_LOCK);
1242 if (tv->states & VRO) {
1243 pass = 1; /* try this *once* */
1244 ObtainWriteLock(&afs_xcbhash, 495);
1245 afs_DequeueCallback(adp);
1246 /* re-stat to get later version */
1247 adp->states &= ~CStatd;
1248 ReleaseWriteLock(&afs_xcbhash);
1249 osi_dnlc_purgedp(adp);
1250 afs_PutVolume(tv, READ_LOCK);
1253 afs_PutVolume(tv, READ_LOCK);
1260 /* put the network buffer back, if need be */
1261 if (tname != aname && tname) osi_FreeLargeSpace(tname);
1264 /* Handle RENAME; only need to check rename "." */
1265 if (opflag == RENAME && wantparent && *ndp->ni_next == 0) {
1266 if (!FidCmp(&(tvc->fid), &(adp->fid))) {
1267 afs_PutVCache(*avcp, WRITE_LOCK);
1269 return afs_CheckCode(EISDIR, &treq, 18);
1272 #endif /* AFS_OSF_ENV */
1275 afs_AddMarinerName(aname, tvc);
1277 #if defined(UKERNEL) && defined(AFS_WEB_ENHANCEMENTS)
1278 if (!(flags & AFS_LOOKUP_NOEVAL))
1279 /* Here we don't enter the name into the DNLC because we want the
1280 evaluated mount dir to be there (the vcache for the mounted volume)
1281 rather than the vc of the mount point itself. we can still find the
1282 mount point's vc in the vcache by its fid. */
1283 #endif /* UKERNEL && AFS_WEB_ENHANCEMENTS */
1285 osi_dnlc_enter (adp, aname, tvc, &versionNo);
1288 #ifdef AFS_LINUX20_ENV
1289 /* So Linux inode cache is up to date. */
1290 code = afs_VerifyVCache(tvc, &treq);
1292 return 0; /* can't have been any errors if hit and !code */
1296 code = afs_CheckCode(code, &treq, 19);
1298 /* If there is an error, make sure *avcp is null.
1299 * Alphas panic otherwise - defect 10719.
1301 *avcp = (struct vcache *)0;