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 "../afs/param.h" /* Should be always first */
23 #include "../afs/sysincludes.h" /* Standard vendor system headers */
24 #include "../afs/afsincludes.h" /* Afs-based standard headers */
25 #include "../afs/afs_stats.h" /* statistics */
26 #include "../afs/afs_cbqueue.h"
27 #include "../afs/nfsclient.h"
28 #include "../afs/exporter.h"
29 #include "../afs/afs_osidnlc.h"
33 * A few definitions. This is until we have a proper header file
34 * which has prototypes for all functions
37 extern struct DirEntry * afs_dir_GetBlob();
39 extern afs_rwlock_t afs_xvcache;
40 extern afs_rwlock_t afs_xcbhash;
41 extern struct afs_exporter *afs_nfsexporter;
42 extern char *afs_sysname;
43 extern struct afs_q VLRU; /*vcache LRU*/
44 #ifdef AFS_LINUX22_ENV
45 extern struct inode_operations afs_symlink_iops, afs_dir_iops;
49 afs_int32 afs_bulkStatsDone;
50 static int bulkStatCounter = 0; /* counter for bulk stat seq. numbers */
53 /* this would be faster if it did comparison as int32word, but would be
54 * dependant on byte-order and alignment, and I haven't figured out
55 * what "@sys" is in binary... */
56 #define AFS_EQ_ATSYS(name) (((name)[0]=='@')&&((name)[1]=='s')&&((name)[2]=='y')&&((name)[3]=='s')&&(!(name)[4]))
60 register char *s1, *s2;
76 register char *a, c; {
78 AFS_STATCNT(afs_index);
80 if (tc == c) return a;
86 /* call under write lock, evaluate mvid field from a mt pt.
87 * avc is the vnode of the mount point object.
88 * advc is the vnode of the containing directory
89 * avolpp is where we return a pointer to the volume named by the mount pt, if success
90 * areq is the identity of the caller.
92 * NOTE: this function returns a held volume structure in *volpp if it returns 0!
94 EvalMountPoint(avc, advc, avolpp, areq)
95 register struct vcache *avc;
96 struct volume **avolpp;
97 struct vcache *advc; /* the containing dir */
98 register struct vrequest *areq;
101 struct volume *tvp = 0;
102 struct VenusFid tfid;
104 char *cpos, *volnamep;
106 afs_int32 prefetchRO; /* 1=>No 2=>Yes */
107 afs_int32 mtptCell, assocCell, hac=0;
108 afs_int32 samecell, roname, len;
110 AFS_STATCNT(EvalMountPoint);
112 if (avc->mvid && (avc->states & CMValid)) return 0; /* done while racing */
114 *avolpp = (struct volume *)0;
115 code = afs_HandleLink(avc, areq);
116 if (code) return code;
118 /* Determine which cell and volume the mointpoint goes to */
119 type = avc->linkData[0]; /* '#'=>Regular '%'=>RW */
120 cpos = afs_index(&avc->linkData[1], ':'); /* if cell name present */
124 tcell = afs_GetCellByName(&avc->linkData[1], READ_LOCK);
127 volnamep = &avc->linkData[1];
128 tcell = afs_GetCell(avc->fid.Cell, READ_LOCK);
130 if (!tcell) return ENODEV;
132 mtptCell = tcell->cell; /* The cell for the mountpoint */
134 hac = 1; /* has associated cell */
135 assocCell = tcell->lcellp->cell; /* The associated cell */
137 afs_PutCell(tcell, READ_LOCK);
139 /* Is volume name a "<n>.backup" or "<n>.readonly" name */
140 len = strlen(volnamep);
141 roname = ((len > 9) && (strcmp(&volnamep[len - 9],".readonly") == 0)) ||
142 ((len > 7) && (strcmp(&volnamep[len - 7],".backup") == 0));
144 /* When we cross mountpoint, do we stay in the same cell */
145 samecell = (avc->fid.Cell == mtptCell) || (hac && (avc->fid.Cell == assocCell));
147 /* Decide whether to prefetch the RO. Also means we want the RO.
148 * If this is a regular mountpoint with a RW volume name and
149 * we cross a cell boundary -or- start from a RO volume, then we will
150 * want to prefetch the RO volume when we get the RW below.
152 if ( (type == '#') && !roname && (!samecell || (avc->states & CRO)) ) {
153 prefetchRO = 2; /* Yes, prefetch the RO */
155 prefetchRO = 1; /* No prefetch of the RO */
158 /* Get the volume struct. Unless this volume name has ".readonly" or
159 * ".backup" in it, this will get the volume struct for the RW volume.
160 * The RO volume will be prefetched if requested (but not returned).
162 tvp = afs_GetVolumeByName(volnamep, mtptCell, prefetchRO, areq, WRITE_LOCK);
164 /* If no volume was found in this cell, try the associated linked cell */
165 if (!tvp && hac && areq->volumeError) {
166 tvp = afs_GetVolumeByName(volnamep, assocCell, prefetchRO, areq, WRITE_LOCK);
169 /* Still not found. If we are looking for the RO, then perhaps the RW
170 * doesn't exist? Try adding ".readonly" to volname and look for that.
171 * Don't know why we do this. Would have still found it in above call - jpm.
173 if (!tvp && (prefetchRO == 2)) {
174 strcpy(buf, volnamep);
175 afs_strcat(buf, ".readonly");
177 tvp = afs_GetVolumeByName(buf, mtptCell, 1, areq, WRITE_LOCK);
179 /* Try the associated linked cell if failed */
180 if (!tvp && hac && areq->volumeError) {
181 tvp = afs_GetVolumeByName(buf, assocCell, 1, areq, WRITE_LOCK);
185 if (!tvp) return ENOENT; /* Couldn't find the volume */
187 /* Don't cross mountpoint from a BK to a BK volume */
188 if ((avc->states & CBackup) && (tvp->states & VBackup)) {
189 afs_PutVolume(tvp, WRITE_LOCK);
193 /* If we want (prefetched) the RO and it exists, then drop the
194 * RW volume and get the RO. Othewise, go with the RW.
196 if ((prefetchRO == 2) && tvp->roVol) {
197 tfid.Fid.Volume = tvp->roVol; /* remember RO volume */
198 tfid.Cell = tvp->cell;
199 afs_PutVolume(tvp, WRITE_LOCK); /* release old volume */
200 tvp = afs_GetVolume(&tfid, areq, WRITE_LOCK); /* get the new one */
201 if (!tvp) return ENOENT; /* oops, can't do it */
205 avc->mvid = (struct VenusFid *) osi_AllocSmallSpace(sizeof(struct VenusFid));
206 avc->mvid->Cell = tvp->cell;
207 avc->mvid->Fid.Volume = tvp->volume;
208 avc->mvid->Fid.Vnode = 1;
209 avc->mvid->Fid.Unique = 1;
210 avc->states |= CMValid;
212 /* Used to: if the mount point is stored within a backup volume,
213 * then we should only update the parent pointer information if
214 * there's none already set, so as to avoid updating a volume's ..
215 * info with something in an OldFiles directory.
217 * Next two lines used to be under this if:
219 * if (!(avc->states & CBackup) || tvp->dotdot.Fid.Volume == 0)
221 * Now: update mount point back pointer on every call, so that we handle
222 * multiple mount points better. This way, when du tries to go back
223 * via chddir(".."), it will end up exactly where it started, yet
224 * cd'ing via a new path to a volume will reset the ".." pointer
227 tvp->mtpoint = avc->fid; /* setup back pointer to mtpoint */
228 tvp->dotdot = advc->fid;
235 register char *aname; {
239 AFS_STATCNT(ENameOK);
240 tlen = strlen(aname);
241 if (tlen >= 4 && strcmp(aname+tlen-4, "@sys") == 0) return 0;
245 Check_AtSys(avc, aname, outb, areq)
246 register struct vcache *avc;
248 struct vrequest *areq;
250 register char *tname;
251 register int error = 0, offset = -1;
253 for (tname=aname; *tname; tname++) /*Move to the end of the string*/;
256 * If the current string is 4 chars long or more, check to see if the
257 * tail end is "@sys".
259 if ((tname >= aname + 4) && (AFS_EQ_ATSYS(tname-4)))
260 offset = (tname - 4) - aname;
264 tname = (char *) osi_AllocLargeSpace(AFS_LRALLOCSIZ);
266 strncpy(tname, aname, offset);
267 if (!afs_nfsexporter)
268 strcpy(tname+offset, (afs_sysname ? afs_sysname : SYS_NAME ));
270 register struct unixuser *au;
271 register afs_int32 error;
272 au = afs_GetUser(areq->uid, avc->fid.Cell, 0); afs_PutUser(au, 0);
274 error = EXP_SYSNAME(au->exporter, (char *)0, tname+offset);
276 strcpy(tname+offset, "@sys");
278 strcpy(tname+offset, (afs_sysname ? afs_sysname : SYS_NAME ));
288 char *afs_getsysname(areq, adp)
289 register struct vrequest *areq;
290 register struct vcache *adp; {
291 static char sysname[MAXSYSNAME];
292 register struct unixuser *au;
293 register afs_int32 error;
295 AFS_STATCNT(getsysname);
296 /* this whole interface is wrong, it should take a buffer ptr and copy
299 au = afs_GetUser(areq->uid, adp->fid.Cell, 0);
302 error = EXP_SYSNAME(au->exporter, (char *)0, sysname);
303 if (error) return "@sys";
306 return (afs_sysname == 0? SYS_NAME : afs_sysname);
310 void afs_HandleAtName(aname, aresult, areq, adp)
311 register char *aname;
312 register char *aresult;
313 register struct vrequest *areq;
314 register struct vcache *adp; {
316 AFS_STATCNT(HandleAtName);
317 tlen = strlen(aname);
318 if (tlen >= 4 && strcmp(aname+tlen-4, "@sys")==0) {
319 strncpy(aresult, aname, tlen-4);
320 strcpy(aresult+tlen-4, afs_getsysname(areq, adp));
322 else strcpy(aresult, aname);
325 #if (defined(AFS_SGI62_ENV) || defined(AFS_SUN57_64BIT_ENV))
326 extern int BlobScan(ino64_t *afile, afs_int32 ablob);
328 #if defined AFS_LINUX_64BIT_KERNEL
329 extern int BlobScan(long *afile, afs_int32 ablob);
331 extern int BlobScan(afs_int32 *afile, afs_int32 ablob);
336 /* called with an unlocked directory and directory cookie. Areqp
337 * describes who is making the call.
338 * Scans the next N (about 30, typically) directory entries, and does
339 * a bulk stat call to stat them all.
341 * Must be very careful when merging in RPC responses, since we dont
342 * want to overwrite newer info that was added by a file system mutating
343 * call that ran concurrently with our bulk stat call.
345 * We do that, as described below, by not merging in our info (always
346 * safe to skip the merge) if the status info is valid in the vcache entry.
348 * If adapt ever implements the bulk stat RPC, then this code will need to
349 * ensure that vcaches created for failed RPC's to older servers have the
352 struct vcache * BStvc = (struct vcache *) 0;
353 void afs_DoBulkStat(adp, dirCookie, areqp)
356 struct vrequest *areqp;
358 int nentries; /* # of entries to prefetch */
359 int nskip; /* # of slots in the LRU queue to skip */
360 struct vcache *lruvcp; /* vcache ptr of our goal pos in LRU queue */
361 struct dcache *dcp; /* chunk containing the dir block */
362 char *statMemp; /* status memory block */
363 char *cbfMemp; /* callback and fid memory block */
364 long temp; /* temp for holding chunk length, &c. */
365 struct AFSFid *fidsp; /* file IDs were collecting */
366 struct AFSCallBack *cbsp; /* call back pointers */
367 struct AFSCallBack *tcbp; /* temp callback ptr */
368 struct AFSFetchStatus *statsp; /* file status info */
369 struct AFSVolSync volSync; /* vol sync return info */
370 struct vcache *tvcp; /* temp vcp */
371 struct afs_q *tq; /* temp queue variable */
372 AFSCBFids fidParm; /* file ID parm for bulk stat */
373 AFSBulkStats statParm; /* stat info parm for bulk stat */
374 int fidIndex; /* which file were stating */
375 struct conn *tcp; /* conn for call */
376 AFSCBs cbParm; /* callback parm for bulk stat */
377 struct server *hostp = 0; /* host we got callback from */
378 long origEvenCBs; /* original # of callbacks for even-fid files */
379 long origOddCBs; /* original # of callbacks for odd-fid files */
380 long origEvenZaps; /* original # of recycles for even-fid files */
381 long origOddZaps; /* original # of recycles for odd-fid files */
382 long startTime; /* time we started the call,
383 * for callback expiration base
385 int statSeqNo; /* Valued of file size to detect races */
386 int code; /* error code */
387 long newIndex; /* new index in the dir */
388 struct DirEntry *dirEntryp; /* dir entry we are examining */
390 struct VenusFid afid; /* file ID we are using now */
391 struct VenusFid tfid; /* another temp. file ID */
392 afs_int32 retry; /* handle low-level SGI MP race conditions */
393 long volStates; /* flags from vol structure */
394 struct volume *volp=0; /* volume ptr */
395 struct VenusFid dotdot;
396 int flagIndex; /* First file with bulk fetch flag set */
399 /* first compute some basic parameters. We dont want to prefetch more
400 * than a fraction of the cache in any given call, and we want to preserve
401 * a portion of the LRU queue in any event, so as to avoid thrashing
402 * the entire stat cache (we will at least leave some of it alone).
403 * presently dont stat more than 1/8 the cache in any one call. */
404 nentries = afs_cacheStats / 8;
406 /* dont bother prefetching more than one calls worth of info */
407 if (nentries > AFSCBMAX) nentries = AFSCBMAX;
409 /* heuristic to make sure that things fit in 4K. This means that
410 * we shouldnt make it any bigger than 47 entries. I am typically
411 * going to keep it a little lower, since we don't want to load
412 * too much of the stat cache.
414 if (nentries > 30) nentries = 30;
416 /* now, to reduce the stack size, well allocate two 4K blocks,
417 * one for fids and callbacks, and one for stat info. Well set
418 * up our pointers to the memory from there, too.
420 statMemp = osi_AllocLargeSpace(nentries * sizeof(AFSFetchStatus));
421 statsp = (struct AFSFetchStatus *) statMemp;
422 cbfMemp = osi_AllocLargeSpace(nentries *
423 (sizeof(AFSCallBack) + sizeof(AFSFid)));
424 fidsp = (AFSFid *) cbfMemp;
425 cbsp = (AFSCallBack *) (cbfMemp + nentries * sizeof(AFSFid));
427 /* next, we must iterate over the directory, starting from the specified
428 * cookie offset (dirCookie), and counting out nentries file entries.
429 * We skip files that already have stat cache entries, since we
430 * dont want to bulk stat files that are already in the cache.
433 code = afs_VerifyVCache(adp, areqp);
436 dcp = afs_GetDCache(adp, 0, areqp, &temp, &temp, 1);
442 /* lock the directory cache entry */
443 ObtainReadLock(&adp->lock);
446 * Make sure that the data in the cache is current. There are two
447 * cases we need to worry about:
448 * 1. The cache data is being fetched by another process.
449 * 2. The cache data is no longer valid
451 while ((adp->states & CStatd)
452 && (dcp->flags & DFFetching)
453 && hsame(adp->m.DataVersion, dcp->f.versionNo)) {
454 dcp->flags |= DFWaiting;
455 ReleaseReadLock(&adp->lock);
456 afs_osi_Sleep(&dcp->validPos);
457 ObtainReadLock(&adp->lock);
459 if (!(adp->states & CStatd)
460 || !hsame(adp->m.DataVersion, dcp->f.versionNo)) {
461 ReleaseReadLock(&adp->lock);
466 /* Generate a sequence number so we can tell whether we should
467 * store the attributes when processing the response. This number is
468 * stored in the file size when we set the CBulkFetching bit. If the
469 * CBulkFetching is still set and this value hasn't changed, then
470 * we know we were the last to set CBulkFetching bit for this file,
471 * and it is safe to set the status information for this file.
473 statSeqNo = bulkStatCounter++;
475 /* now we have dir data in the cache, so scan the dir page */
478 while (1) { /* Should probably have some constant bound */
479 /* look for first safe entry to examine in the directory. BlobScan
480 * looks for a the 1st allocated dir after the dirCookie slot.
482 newIndex = BlobScan(&dcp->f.inode, (dirCookie>>5));
483 if (newIndex == 0) break;
485 /* remember the updated directory cookie */
486 dirCookie = newIndex << 5;
488 /* get a ptr to the dir entry */
489 dirEntryp =(struct DirEntry *)afs_dir_GetBlob(&dcp->f.inode, newIndex);
490 if (!dirEntryp) break;
492 /* dont copy more than we have room for */
493 if (fidIndex >= nentries) {
494 DRelease((char *) dirEntryp, 0);
498 /* now, if the dir entry looks good, copy it out to our list. Vnode
499 * 0 means deleted, although it should also be free were it deleted.
501 if (dirEntryp->fid.vnode != 0) {
502 /* dont copy entries we have in our cache. This check will
503 * also make us skip "." and probably "..", unless it has
504 * disappeared from the cache since we did our namei call.
506 tfid.Cell = adp->fid.Cell;
507 tfid.Fid.Volume = adp->fid.Fid.Volume;
508 tfid.Fid.Vnode = ntohl(dirEntryp->fid.vnode);
509 tfid.Fid.Unique = ntohl(dirEntryp->fid.vunique);
512 ObtainWriteLock(&afs_xvcache, 130);
513 tvcp = afs_FindVCache(&tfid, 0, 0, &retry, 0 /* no stats | LRU */);
515 ReleaseWriteLock(&afs_xvcache);
518 } while (tvcp && retry);
519 if (!tvcp) { /* otherwise, create manually */
520 tvcp = afs_NewVCache(&tfid, hostp, 0, 0);
521 ObtainWriteLock(&tvcp->lock, 505);
522 ReleaseWriteLock(&afs_xvcache);
523 afs_RemoveVCB(&tfid);
524 ReleaseWriteLock(&tvcp->lock);
526 ReleaseWriteLock(&afs_xvcache);
529 goto done; /* can't happen at present, more's the pity */
531 /* WARNING: afs_DoBulkStat uses the Length field to store a
532 * sequence number for each bulk status request. Under no
533 * circumstances should afs_DoBulkStat store a sequence number
534 * if the new length will be ignored when afs_ProcessFS is
535 * called with new stats. */
537 if (!(tvcp->states & (CStatd|CBulkFetching))
538 && (tvcp->execsOrWriters <= 0)
539 && !afs_DirtyPages(tvcp)
540 && !AFS_VN_MAPPED((vnode_t*)tvcp))
542 if (!(tvcp->states & (CStatd|CBulkFetching))
543 && (tvcp->execsOrWriters <= 0)
544 && !afs_DirtyPages(tvcp))
548 /* this entry doesnt exist in the cache, and is not
549 * already being fetched by someone else, so add it to the
550 * list of file IDs to obtain.
552 * We detect a callback breaking race condition by checking the
553 * CBulkFetching state bit and the value in the file size.
554 * It is safe to set the status only if the CBulkFetching
555 * flag is still set and the value in the file size does
558 * Don't fetch status for dirty files. We need to
559 * preserve the value of the file size. We could
560 * flush the pages, but it wouldn't be worthwhile.
562 bcopy((char *) &tfid.Fid, (char *)(fidsp+fidIndex),
564 tvcp->states |= CBulkFetching;
565 tvcp->m.Length = statSeqNo;
569 } /* if dir vnode has non-zero entry */
571 /* move to the next dir entry by adding in the # of entries
572 * used by this dir entry.
574 temp = afs_dir_NameBlobs(dirEntryp->name) << 5;
575 DRelease((char *) dirEntryp, 0);
576 if (temp <= 0) break;
578 } /* while loop over all dir entries */
580 /* now release the dir lock and prepare to make the bulk RPC */
581 ReleaseReadLock(&adp->lock);
583 /* release the chunk */
586 /* dont make a null call */
587 if (fidIndex == 0) goto done;
590 /* setup the RPC parm structures */
591 fidParm.AFSCBFids_len = fidIndex;
592 fidParm.AFSCBFids_val = fidsp;
593 statParm.AFSBulkStats_len = fidIndex;
594 statParm.AFSBulkStats_val = statsp;
595 cbParm.AFSCBs_len = fidIndex;
596 cbParm.AFSCBs_val = cbsp;
598 /* start the timer; callback expirations are relative to this */
599 startTime = osi_Time();
601 tcp = afs_Conn(&adp->fid, areqp, SHARED_LOCK);
603 hostp = tcp->srvr->server;
604 XSTATS_START_TIME(AFS_STATS_FS_RPCIDX_BULKSTATUS);
605 #ifdef RX_ENABLE_LOCKS
607 #endif /* RX_ENABLE_LOCKS */
608 code = RXAFS_BulkStatus(tcp->id, &fidParm, &statParm, &cbParm,
610 #ifdef RX_ENABLE_LOCKS
612 #endif /* RX_ENABLE_LOCKS */
616 } while (afs_Analyze(tcp, code, &adp->fid, areqp,
617 AFS_STATS_FS_RPCIDX_BULKSTATUS, SHARED_LOCK, (struct cell *)0));
619 /* now, if we didnt get the info, bail out. */
622 /* we need vol flags to create the entries properly */
623 dotdot.Fid.Volume = 0;
624 volp = afs_GetVolume(&adp->fid, areqp, READ_LOCK);
626 volStates = volp->states;
627 if (volp->dotdot.Fid.Volume != 0)
628 dotdot = volp->dotdot;
632 /* find the place to merge the info into We do this by skipping
633 * nskip entries in the LRU queue. The more we skip, the more
634 * we preserve, since the head of the VLRU queue is the most recently
638 nskip = afs_cacheStats / 2; /* preserved fraction of the cache */
639 ObtainReadLock(&afs_xvcache);
641 /* actually a serious error, probably should panic. Probably will
642 * panic soon, oh well. */
643 ReleaseReadLock(&afs_xvcache);
644 afs_warnuser("afs_DoBulkStat: VLRU empty!");
647 if ((VLRU.next->prev != &VLRU) || (VLRU.prev->next != &VLRU)) {
648 refpanic ("Bulkstat VLRU inconsistent");
650 for(tq = VLRU.next; tq != &VLRU; tq = QNext(tq)) {
651 if (--nskip <= 0) break;
652 else if (QNext(QPrev(tq)) != tq) {
654 refpanic ("BulkStat VLRU inconsistent");
657 if (tq != &VLRU) lruvcp = QTOV(tq);
658 else lruvcp = QTOV(VLRU.next);
660 /* now we have to hold this entry, so that it does not get moved
661 * into the free list while we're running. It could still get
662 * moved within the lru queue, but hopefully that will be rare; it
663 * doesn't hurt nearly as much.
666 osi_vnhold(lruvcp, &retry);
667 ReleaseReadLock(&afs_xvcache); /* could be read lock */
671 /* otherwise, merge in the info. We have to be quite careful here,
672 * since we need to ensure that we don't merge old info over newer
673 * stuff in a stat cache entry. We're very conservative here: we don't
674 * do the merge at all unless we ourselves create the stat cache
675 * entry. That's pretty safe, and should work pretty well, since we
676 * typically expect to do the stat cache creation ourselves.
678 * We also have to take into account racing token revocations.
680 for(i=0; i<fidIndex; i++) {
681 afid.Cell = adp->fid.Cell;
682 afid.Fid.Volume = adp->fid.Fid.Volume;
683 afid.Fid.Vnode = fidsp[i].Vnode;
684 afid.Fid.Unique = fidsp[i].Unique;
687 ObtainReadLock(&afs_xvcache);
688 tvcp = afs_FindVCache(&afid, 1, 0, &retry, 0/* !stats&!lru*/);
689 ReleaseReadLock(&afs_xvcache);
690 } while (tvcp && retry);
692 /* The entry may no longer exist */
697 /* now we have the entry held, but we need to fill it in */
698 ObtainWriteLock(&tvcp->lock,131);
700 /* if CBulkFetching is not set, or if the file size no longer
701 * matches the value we placed there when we set the CBulkFetching
702 * flag, then someone else has done something with this node,
703 * and we may not have the latest status information for this
704 * file. Leave the entry alone.
706 if (!(tvcp->states & CBulkFetching) || (tvcp->m.Length != statSeqNo)) {
708 ReleaseWriteLock(&tvcp->lock);
713 /* now copy ".." entry back out of volume structure, if necessary */
714 if (tvcp->mvstat == 2 && (dotdot.Fid.Volume != 0)) {
716 tvcp->mvid = (struct VenusFid *) osi_AllocSmallSpace(sizeof(struct VenusFid));
717 *tvcp->mvid = dotdot;
720 ObtainWriteLock(&afs_xvcache,132);
721 if ((VLRU.next->prev != &VLRU) || (VLRU.prev->next != &VLRU)) {
722 refpanic ("Bulkstat VLRU inconsistent2");
724 if ((QNext(QPrev(&tvcp->vlruq)) != &tvcp->vlruq)
725 || (QPrev(QNext(&tvcp->vlruq)) != &tvcp->vlruq))
726 refpanic ("Bulkstat VLRU inconsistent4");
727 if ((QNext(QPrev(&lruvcp->vlruq)) != &lruvcp->vlruq)
728 || (QPrev(QNext(&lruvcp->vlruq)) != &lruvcp->vlruq))
729 refpanic ("Bulkstat VLRU inconsistent5");
731 if (tvcp != lruvcp) { /* if they are == don't move it, don't corrupt vlru */
732 QRemove(&tvcp->vlruq);
733 QAdd(&lruvcp->vlruq, &tvcp->vlruq);
736 if ((VLRU.next->prev != &VLRU) || (VLRU.prev->next != &VLRU)) {
737 refpanic ("Bulkstat VLRU inconsistent3");
739 if ((QNext(QPrev(&tvcp->vlruq)) != &tvcp->vlruq)
740 || (QPrev(QNext(&tvcp->vlruq)) != &tvcp->vlruq))
741 refpanic ("Bulkstat VLRU inconsistent5");
742 if ((QNext(QPrev(&lruvcp->vlruq)) != &lruvcp->vlruq)
743 || (QPrev(QNext(&lruvcp->vlruq)) != &lruvcp->vlruq))
744 refpanic ("Bulkstat VLRU inconsistent6");
745 ReleaseWriteLock(&afs_xvcache);
747 ObtainWriteLock(&afs_xcbhash, 494);
749 /* We need to check the flags again. We may have missed
750 * something while we were waiting for a lock.
752 if (!(tvcp->states & CBulkFetching) || (tvcp->m.Length != statSeqNo)) {
754 ReleaseWriteLock(&tvcp->lock);
755 ReleaseWriteLock(&afs_xcbhash);
760 /* now merge in the resulting status back into the vnode.
761 * We only do this if the entry looks clear.
763 afs_ProcessFS(tvcp, &statsp[i], areqp);
764 #ifdef AFS_LINUX22_ENV
765 /* overwrite the ops if it's a directory or symlink. */
766 if (vType(tvcp) == VDIR)
767 tvcp->v.v_op = &afs_dir_iops;
768 else if (vType(tvcp) == VLNK)
769 tvcp->v.v_op = &afs_symlink_iops;
772 /* do some accounting for bulk stats: mark this entry as
773 * loaded, so we can tell if we use it before it gets
776 tvcp->states |= CBulkStat;
777 tvcp->states &= ~CBulkFetching;
781 /* merge in vol info */
782 if (volStates & VRO) tvcp->states |= CRO;
783 if (volStates & VBackup) tvcp->states |= CBackup;
784 if (volStates & VForeign) tvcp->states |= CForeign;
786 /* merge in the callback info */
787 tvcp->states |= CTruth;
789 /* get ptr to the callback we are interested in */
792 if (tcbp->ExpirationTime != 0) {
793 tvcp->cbExpires = tcbp->ExpirationTime+startTime;
794 tvcp->callback = hostp;
795 tvcp->states |= CStatd;
796 afs_QueueCallback(tvcp, CBHash(tcbp->ExpirationTime), volp);
798 else if (tvcp->states & CRO) {
799 /* ordinary callback on a read-only volume -- AFS 3.2 style */
800 tvcp->cbExpires = 3600+startTime;
801 tvcp->callback = hostp;
802 tvcp->states |= CStatd;
803 afs_QueueCallback(tvcp, CBHash(3600), volp);
807 tvcp->states &= ~(CStatd|CUnique);
808 afs_DequeueCallback(tvcp);
809 if ((tvcp->states & CForeign) || (vType(tvcp) == VDIR))
810 osi_dnlc_purgedp (tvcp); /* if it (could be) a directory */
812 ReleaseWriteLock(&afs_xcbhash);
814 ReleaseWriteLock(&tvcp->lock);
815 /* finally, we're done with the entry */
817 } /* for all files we got back */
819 /* finally return the pointer into the LRU queue */
820 afs_PutVCache(lruvcp);
823 /* Be sure to turn off the CBulkFetching flags */
824 for(i=flagIndex; i<fidIndex; i++) {
825 afid.Cell = adp->fid.Cell;
826 afid.Fid.Volume = adp->fid.Fid.Volume;
827 afid.Fid.Vnode = fidsp[i].Vnode;
828 afid.Fid.Unique = fidsp[i].Unique;
831 ObtainReadLock(&afs_xvcache);
832 tvcp = afs_FindVCache(&afid, 1, 0, &retry, 0/* !stats&!lru*/);
833 ReleaseReadLock(&afs_xvcache);
834 } while (tvcp && retry);
836 && (tvcp->states & CBulkFetching)
837 && (tvcp->m.Length == statSeqNo)) {
838 tvcp->states &= ~CBulkFetching;
845 afs_PutVolume(volp, READ_LOCK);
847 osi_FreeLargeSpace(statMemp);
848 osi_FreeLargeSpace(cbfMemp);
851 /* was: (AFS_DEC_ENV) || defined(AFS_OSF30_ENV) || defined(AFS_NCR_ENV) */
857 struct nameidata *ndp; {
858 char aname[MAXNAMLEN+1]; /* XXX */
859 struct vcache **avcp = (struct vcache **)&(ndp->ni_vp);
860 struct ucred *acred = ndp->ni_cred;
861 int wantparent = ndp->ni_nameiop & WANTPARENT;
862 int opflag = ndp->ni_nameiop & OPFLAG;
863 #else /* AFS_OSF_ENV */
864 #if defined(AFS_SUN5_ENV) || defined(AFS_SGI_ENV)
865 afs_lookup(OSI_VC_ARG(adp), aname, avcp, pnp, flags, rdir, acred)
866 struct pathname *pnp;
870 afs_lookup(adp, aname, avcp, acred)
873 struct vcache **avcp;
875 struct AFS_UCRED *acred; {
877 struct vrequest treq;
878 char *tname = (char *)0;
879 register struct vcache *tvc=0;
880 register afs_int32 code;
881 int pass = 0, hit = 0;
883 extern afs_int32 afs_mariner; /*Writing activity to log?*/
885 afs_hyper_t versionNo;
887 AFS_STATCNT(afs_lookup);
889 ndp->ni_dvp = (struct vnode *)adp;
890 bcopy(ndp->ni_ptr, aname, ndp->ni_namelen);
891 aname[ndp->ni_namelen] = '\0';
892 #endif /* AFS_OSF_ENV */
894 *avcp = (struct vcache *) 0; /* Since some callers don't initialize it */
896 if (code = afs_InitReq(&treq, acred)) {
900 /* lookup the name aname in the appropriate dir, and return a cache entry
901 on the resulting fid */
904 * check for, and handle "@sys" if it's there. We should be able
905 * to avoid the alloc and the strcpy with a little work, but it's
906 * not pressing. If there aren't any remote users (ie, via the
907 * NFS translator), we have a slightly easier job.
908 * the faster way to do this is to check for *aname == '@' and if
909 * it's there, check for @sys, otherwise, assume there's no @sys
910 * then, if the lookup fails, check for .*@sys...
912 if (!AFS_EQ_ATSYS(aname)) {
916 tname = (char *) osi_AllocLargeSpace(AFS_SMALLOCSIZ);
917 if (!afs_nfsexporter)
918 strcpy(tname, (afs_sysname ? afs_sysname : SYS_NAME ));
920 register struct unixuser *au;
921 register afs_int32 error;
922 au = afs_GetUser(treq.uid, adp->fid.Cell, 0); afs_PutUser(au, 0);
924 error = EXP_SYSNAME(au->exporter, (char *)0, tname);
926 strcpy(tname, "@sys");
928 strcpy(tname, (afs_sysname ? afs_sysname : SYS_NAME ));
933 /* come back to here if we encounter a non-existent object in a read-only
934 volume's directory */
937 *avcp = (struct vcache *) 0; /* Since some callers don't initialize it */
939 if (!(adp->states & CStatd)) {
940 if (code = afs_VerifyVCache2(adp, &treq))
945 /* watch for ".." in a volume root */
946 if (adp->mvstat == 2 && tname[0] == '.' && tname[1] == '.' && !tname[2]) {
947 /* looking up ".." in root via special hacks */
948 if (adp->mvid == (struct VenusFid *) 0 || adp->mvid->Fid.Volume == 0) {
950 extern struct vcache *afs_globalVp;
951 if (adp == afs_globalVp) {
952 struct vnode *rvp = (struct vnode *)adp;
954 ndp->ni_vp = rvp->v_vfsp->vfs_vnodecovered;
955 ndp->ni_dvp = ndp->ni_vp;
965 /* otherwise we have the fid here, so we use it */
966 tvc = afs_GetVCache(adp->mvid, &treq, (afs_int32 *)0,
967 (struct vcache*)0, 0);
968 afs_Trace3(afs_iclSetp, CM_TRACE_GETVCDOTDOT,
969 ICL_TYPE_FID, adp->mvid, ICL_TYPE_POINTER, tvc,
970 ICL_TYPE_INT32, code);
972 code = (tvc ? 0 : ENOENT);
974 if (tvc && !tvc->vrefCount) {
978 /*printf("LOOKUP GETVCDOTDOT -> %d\n", code);*/
983 /* now check the access */
984 if (treq.uid != adp->last_looker) {
985 if (!afs_AccessOK(adp, PRSFS_LOOKUP, &treq, CHECK_MODE_BITS)) {
986 *avcp = (struct vcache *)0;
990 else adp->last_looker = treq.uid;
994 /* special case lookup of ".". Can we check for it sooner in this code,
995 * for instance, way up before "redo:" ??
996 * I'm not fiddling with the LRUQ here, either, perhaps I should, or else
997 * invent a lightweight version of GetVCache.
999 if (tname[0] == '.' && !tname[1]) { /* special case */
1000 ObtainReadLock(&afs_xvcache);
1002 ReleaseReadLock(&afs_xvcache);
1006 if (adp && !adp->vrefCount) {
1012 tvc = osi_dnlc_lookup (adp, tname, WRITE_LOCK);
1013 *avcp = tvc; /* maybe wasn't initialized, but it is now */
1014 #ifdef AFS_LINUX22_ENV
1016 if (tvc->mvstat == 2) { /* we don't trust the dnlc for root vcaches */
1026 #else /* non - LINUX */
1032 #endif /* linux22 */
1035 register struct dcache *tdc;
1036 afs_int32 dirOffset, dirLen;
1038 struct VenusFid tfid;
1040 /* now we have to lookup the next fid */
1041 tdc = afs_GetDCache(adp, 0, &treq, &dirOffset, &dirLen, 1);
1043 *avcp = (struct vcache *)0; /* redundant, but harmless */
1048 /* now we will just call dir package with appropriate inode.
1049 Dirs are always fetched in their entirety for now */
1050 /* If the first lookup doesn't succeed, maybe it's got @sys in the name */
1051 ObtainReadLock(&adp->lock);
1054 * Make sure that the data in the cache is current. There are two
1055 * cases we need to worry about:
1056 * 1. The cache data is being fetched by another process.
1057 * 2. The cache data is no longer valid
1059 while ((adp->states & CStatd)
1060 && (tdc->flags & DFFetching)
1061 && hsame(adp->m.DataVersion, tdc->f.versionNo)) {
1062 tdc->flags |= DFWaiting;
1063 ReleaseReadLock(&adp->lock);
1064 afs_osi_Sleep(&tdc->validPos);
1065 ObtainReadLock(&adp->lock);
1067 if (!(adp->states & CStatd)
1068 || !hsame(adp->m.DataVersion, tdc->f.versionNo)) {
1069 ReleaseReadLock(&adp->lock);
1074 /* Save the version number for when we call osi_dnlc_enter */
1075 hset(versionNo, tdc->f.versionNo);
1077 theDir = tdc->f.inode;
1078 code = afs_dir_LookupOffset(&theDir, tname, &tfid.Fid, &dirCookie);
1079 if (code == ENOENT && tname == aname) {
1081 len = strlen(aname);
1082 if (len >= 4 && AFS_EQ_ATSYS(aname+len-4)) {
1083 tname = (char *) osi_AllocLargeSpace(AFS_LRALLOCSIZ);
1084 afs_HandleAtName(aname, tname, &treq, adp);
1085 code = afs_dir_LookupOffset(&theDir, tname, &tfid.Fid, &dirCookie);
1088 ReleaseReadLock(&adp->lock);
1091 /* new fid has same cell and volume */
1092 tfid.Cell = adp->fid.Cell;
1093 tfid.Fid.Volume = adp->fid.Fid.Volume;
1094 afs_Trace4(afs_iclSetp, CM_TRACE_LOOKUP, ICL_TYPE_POINTER, adp,
1095 ICL_TYPE_STRING, tname,
1096 ICL_TYPE_FID, &tfid, ICL_TYPE_INT32, code);
1099 if (code != ENOENT) {
1100 printf("LOOKUP dirLookupOff -> %d\n", code);
1105 /* prefetch some entries, if the dir is currently open. The variable
1106 * dirCookie tells us where to start prefetching from.
1108 if (AFSDOBULK && adp->opens > 0 && !(adp->states & CForeign)) {
1110 /* if the entry is not in the cache, or is in the cache,
1111 * but hasn't been statd, then do a bulk stat operation.
1115 ObtainReadLock(&afs_xvcache);
1116 tvc = afs_FindVCache(&tfid, 1, 0, &retry, 0/* !stats,!lru */);
1117 ReleaseReadLock(&afs_xvcache);
1118 } while (tvc && retry);
1120 if (!tvc || !(tvc->states & CStatd)) {
1121 afs_DoBulkStat(adp, dirCookie, &treq);
1124 /* if the vcache isn't usable, release it */
1125 if (tvc && !(tvc->states & CStatd)) {
1127 tvc = (struct vcache *) 0;
1130 else tvc = (struct vcache *) 0;
1132 /* now get the status info, if we don't already have it */
1133 /* This is kind of weird, but we might wind up accidentally calling
1134 * RXAFS_Lookup because we happened upon a file which legitimately
1135 * has a 0 uniquifier. That is the result of allowing unique to wrap
1136 * to 0. This was fixed in AFS 3.4. For CForeigh, Unique == 0 means that
1137 * the file has not yet been looked up.
1140 afs_int32 cached = 0;
1141 if (!tfid.Fid.Unique && (adp->states & CForeign)) {
1142 tvc = afs_LookupVCache(&tfid, &treq, &cached, WRITE_LOCK,
1145 if (!tvc) { /* lookup failed or wasn't called */
1146 tvc = afs_GetVCache(&tfid, &treq, &cached, (struct vcache*)0,
1150 } /* sub-block just to reduce stack usage */
1153 if (adp->states & CForeign)
1154 tvc->states |= CForeign;
1155 tvc->parentVnode = adp->fid.Fid.Vnode;
1156 tvc->parentUnique = adp->fid.Fid.Unique;
1157 tvc->states &= ~CBulkStat;
1158 if (tvc->mvstat == 1) {
1159 /* a mt point, possibly unevaluated */
1160 struct volume *tvolp;
1162 ObtainWriteLock(&tvc->lock,133);
1163 code = EvalMountPoint(tvc, adp, &tvolp, &treq);
1164 ReleaseWriteLock(&tvc->lock);
1165 /* next, we want to continue using the target of the mt point */
1166 if (tvc->mvid && (tvc->states & CMValid)) {
1168 /* now lookup target, to set .. pointer */
1169 afs_Trace2(afs_iclSetp, CM_TRACE_LOOKUP1,
1170 ICL_TYPE_POINTER, tvc, ICL_TYPE_FID, &tvc->fid);
1171 uvc = tvc; /* remember for later */
1173 if (tvolp && (tvolp->states & VForeign)) {
1174 /* XXXX tvolp has ref cnt on but not locked! XXX */
1175 tvc = afs_GetRootVCache(tvc->mvid, &treq, (afs_int32 *)0, tvolp, WRITE_LOCK);
1177 tvc = afs_GetVCache(tvc->mvid, &treq, (afs_int32 *)0,
1178 (struct vcache*)0, WRITE_LOCK);
1180 afs_PutVCache(uvc, WRITE_LOCK); /* we're done with it */
1185 afs_PutVolume(tvolp, WRITE_LOCK);
1190 /* now, if we came via a new mt pt (say because of a new
1191 * release of a R/O volume), we must reevaluate the ..
1192 * ptr to point back to the appropriate place */
1194 ObtainWriteLock(&tvc->lock,134);
1195 if (tvc->mvid == (struct VenusFid *) 0) {
1196 tvc->mvid = (struct VenusFid *) osi_AllocSmallSpace(sizeof(struct VenusFid));
1198 /* setup backpointer */
1199 *tvc->mvid = tvolp->dotdot;
1200 ReleaseWriteLock(&tvc->lock);
1201 afs_PutVolume(tvolp, WRITE_LOCK);
1205 afs_PutVCache(tvc, WRITE_LOCK);
1207 if (tvolp) afs_PutVolume(tvolp, WRITE_LOCK);
1212 if (tvc && !tvc->vrefCount) {
1218 /* if we get here, we found something in a directory that couldn't
1219 be located (a Multics "connection failure"). If the volume is
1220 read-only, we try flushing this entry from the cache and trying
1224 tv = afs_GetVolume(&adp->fid, &treq, READ_LOCK);
1226 if (tv->states & VRO) {
1227 pass = 1; /* try this *once* */
1228 ObtainWriteLock(&afs_xcbhash, 495);
1229 afs_DequeueCallback(adp);
1230 /* re-stat to get later version */
1231 adp->states &= ~CStatd;
1232 ReleaseWriteLock(&afs_xcbhash);
1233 osi_dnlc_purgedp(adp);
1234 afs_PutVolume(tv, READ_LOCK);
1237 afs_PutVolume(tv, READ_LOCK);
1244 /* put the network buffer back, if need be */
1245 if (tname != aname && tname) osi_FreeLargeSpace(tname);
1248 /* Handle RENAME; only need to check rename "." */
1249 if (opflag == RENAME && wantparent && *ndp->ni_next == 0) {
1250 if (!FidCmp(&(tvc->fid), &(adp->fid))) {
1251 afs_PutVCache(*avcp, WRITE_LOCK);
1253 return afs_CheckCode(EISDIR, &treq, 18);
1256 #endif /* AFS_OSF_ENV */
1259 afs_AddMarinerName(aname, tvc);
1261 osi_dnlc_enter (adp, aname, tvc, &versionNo);
1264 #ifdef AFS_LINUX20_ENV
1265 /* So Linux inode cache is up to date. */
1266 code = afs_VerifyVCache(tvc, &treq);
1268 return 0; /* can't have been any errors if hit and !code */
1272 code = afs_CheckCode(code, &treq, 19);
1274 /* If there is an error, make sure *avcp is null.
1275 * Alphas panic otherwise - defect 10719.
1277 *avcp = (struct vcache *)0;