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
17 #include <afsconfig.h>
18 #include "afs/param.h"
20 #include "afs/sysincludes.h" /* Standard vendor system headers */
21 #include "afsincludes.h" /* Afs-based standard headers */
22 #include "afs/afs_stats.h" /* statistics */
23 #include "afs/afs_cbqueue.h"
24 #include "afs/nfsclient.h"
25 #include "afs/exporter.h"
26 #include "afs/afs_osidnlc.h"
27 #include "afs/afs_dynroot.h"
29 extern struct vcache *afs_globalVp;
31 afs_int32 afs_bkvolpref = 0;
32 afs_int32 afs_bulkStatsDone;
33 static int bulkStatCounter = 0; /* counter for bulk stat seq. numbers */
34 int afs_fakestat_enable = 0; /* 1: fakestat-all, 2: fakestat-crosscell */
37 /* this would be faster if it did comparison as int32word, but would be
38 * dependant on byte-order and alignment, and I haven't figured out
39 * what "@sys" is in binary... */
40 #define AFS_EQ_ATSYS(name) (((name)[0]=='@')&&((name)[1]=='s')&&((name)[2]=='y')&&((name)[3]=='s')&&(!(name)[4]))
42 /* call under write lock, evaluate mvid field from a mt pt.
43 * avc is the vnode of the mount point object; must be write-locked.
44 * advc is the vnode of the containing directory (optional; if NULL and
45 * EvalMountPoint succeeds, caller must initialize *avolpp->dotdot)
46 * avolpp is where we return a pointer to the volume named by the mount pt, if success
47 * areq is the identity of the caller.
49 * NOTE: this function returns a held volume structure in *volpp if it returns 0!
52 EvalMountData(char type, char *data, afs_uint32 states, afs_uint32 cellnum,
53 struct volume **avolpp, register struct vrequest *areq,
54 afs_uint32 *acellidxp, afs_uint32 *avolnump,
55 afs_uint32 *avnoidp, afs_uint32 *auniqp)
57 struct volume *tvp = 0;
60 char *cpos, *volnamep;
62 afs_int32 prefetch; /* 1=>None 2=>RO 3=>BK */
63 afs_int32 mtptCell, assocCell = 0, hac = 0;
64 afs_int32 samecell, roname, len;
65 afs_uint32 volid = 0, cellidx, vnoid = 0, uniq = 0;
67 /* Start by figuring out and finding the cell */
68 cpos = afs_strchr(data, ':'); /* if cell name present */
72 if ((afs_strtoi_r(data, &endptr, &cellnum) == 0) &&
74 tcell = afs_GetCell(cellnum, READ_LOCK);
76 tcell = afs_GetCellByName(data, READ_LOCK);
82 tcell = afs_GetCell(cellnum, READ_LOCK);
84 /* No cellname or cellnum; return ENODEV */
88 /* no cell found; return ENODEV */
92 cellidx = tcell->cellIndex;
93 mtptCell = tcell->cellNum; /* The cell for the mountpoint */
95 hac = 1; /* has associated cell */
96 assocCell = tcell->lcellp->cellNum; /* The associated cell */
98 afs_PutCell(tcell, READ_LOCK);
100 /* If there's nothing to look up, we can't proceed */
104 /* cell found. figure out volume */
105 cpos = afs_strchr(volnamep, ':');
109 /* Look for an all-numeric volume ID */
110 if ((afs_strtoi_r(volnamep, &endptr, &volid) == 0) &&
111 ((endptr == cpos) || (!*endptr)))
113 /* Ok. Is there a vnode and uniq? */
115 char *vnodep = (char *)(cpos + 1);
117 if ((!*vnodep) /* no vnode after colon */
118 || !(uniqp = afs_strchr(vnodep, ':')) /* no colon for uniq */
119 || (!*(++uniqp)) /* no uniq after colon */
120 || (afs_strtoi_r(vnodep, &endptr, &vnoid) != 0) /* bad vno */
121 || (*endptr != ':') /* bad vnode field */
122 || (afs_strtoi_r(uniqp, &endptr, &uniq) != 0) /* bad uniq */
123 || (*endptr)) /* anything after uniq */
126 /* sorry. vnode and uniq, or nothing */
134 * If the volume ID was all-numeric, and they didn't ask for a
135 * pointer to the volume structure, then just return the number
136 * as-is. This is currently only used for handling name lookups
137 * in the dynamic mount directory.
139 if (volid && !avolpp) {
146 * If the volume ID was all-numeric, and the type was '%', then
147 * assume whoever made the mount point knew what they were doing,
148 * and don't second-guess them by forcing use of a RW volume when
149 * they gave the ID of something else.
151 if (volid && type == '%') {
152 tfid.Fid.Volume = volid; /* remember BK volume */
153 tfid.Cell = mtptCell;
154 tvp = afs_GetVolume(&tfid, areq, WRITE_LOCK); /* get the new one */
155 if (cpos) /* one way or another we're done */
158 return ENODEV; /* afs_GetVolume failed; return ENODEV */
162 /* Is volume name a "<n>.backup" or "<n>.readonly" name */
163 len = strlen(volnamep);
164 roname = ((len > 9) && (strcmp(&volnamep[len - 9], ".readonly") == 0))
165 || ((len > 7) && (strcmp(&volnamep[len - 7], ".backup") == 0));
167 /* When we cross mountpoint, do we stay in the same cell */
168 samecell = (cellnum == mtptCell) || (hac && (cellnum == assocCell));
170 /* Decide whether to prefetch the BK, or RO. Also means we want the BK or
172 * If this is a regular mountpoint with a RW volume name
173 * - If BK preference is enabled AND we remain within the same cell AND
174 * start from a BK volume, then we will want to prefetch the BK volume.
175 * - If we cross a cell boundary OR start from a RO volume, then we will
176 * want to prefetch the RO volume.
178 if ((type == '#') && !roname) {
179 if (afs_bkvolpref && samecell && (states & CBackup))
180 prefetch = 3; /* Prefetch the BK */
181 else if (!samecell || (states & CRO))
182 prefetch = 2; /* Prefetch the RO */
184 prefetch = 1; /* Do not prefetch */
186 prefetch = 1; /* Do not prefetch */
189 /* Get the volume struct. Unless this volume name has ".readonly" or
190 * ".backup" in it, this will get the volume struct for the RW volume.
191 * The RO volume will be prefetched if requested (but not returned).
192 * Set up to use volname first.
194 tvp = afs_GetVolumeByName(volnamep, mtptCell, prefetch, areq, WRITE_LOCK);
196 /* If no volume was found in this cell, try the associated linked cell */
197 if (!tvp && hac && areq->volumeError) {
199 afs_GetVolumeByName(volnamep, assocCell, prefetch, areq,
203 /* Still not found. If we are looking for the RO, then perhaps the RW
204 * doesn't exist? Try adding ".readonly" to volname and look for that.
205 * Don't know why we do this. Would have still found it in above call - jpm.
207 if (!tvp && (prefetch == 2) && len < AFS_SMALLOCSIZ - 10) {
208 buf = (char *)osi_AllocSmallSpace(len + 10);
210 strcpy(buf, volnamep);
211 afs_strcat(buf, ".readonly");
213 tvp = afs_GetVolumeByName(buf, mtptCell, 1, areq, WRITE_LOCK);
215 /* Try the associated linked cell if failed */
216 if (!tvp && hac && areq->volumeError) {
217 tvp = afs_GetVolumeByName(buf, assocCell, 1, areq, WRITE_LOCK);
219 osi_FreeSmallSpace(buf);
221 /* done with volname */
225 return ENODEV; /* Couldn't find the volume */
229 /* Don't cross mountpoint from a BK to a BK volume */
230 if ((states & CBackup) && (tvp->states & VBackup)) {
231 afs_PutVolume(tvp, WRITE_LOCK);
235 /* If we want (prefetched) the BK and it exists, then drop the RW volume
237 * Otherwise, if we want (prefetched0 the RO and it exists, then drop the
238 * RW volume and get the RO.
239 * Otherwise, go with the RW.
241 if ((prefetch == 3) && tvp->backVol) {
242 tfid.Fid.Volume = tvp->backVol; /* remember BK volume */
243 tfid.Cell = tvp->cell;
244 afs_PutVolume(tvp, WRITE_LOCK); /* release old volume */
245 tvp = afs_GetVolume(&tfid, areq, WRITE_LOCK); /* get the new one */
247 return ENODEV; /* oops, can't do it */
248 } else if ((prefetch >= 2) && tvp->roVol) {
249 tfid.Fid.Volume = tvp->roVol; /* remember RO volume */
250 tfid.Cell = tvp->cell;
251 afs_PutVolume(tvp, WRITE_LOCK); /* release old volume */
252 tvp = afs_GetVolume(&tfid, areq, WRITE_LOCK); /* get the new one */
254 return ENODEV; /* oops, can't do it */
259 *acellidxp = cellidx;
269 afs_PutVolume(tvp, WRITE_LOCK);
274 EvalMountPoint(register struct vcache *avc, struct vcache *advc,
275 struct volume **avolpp, register struct vrequest *areq)
278 afs_uint32 avnoid, auniq;
280 AFS_STATCNT(EvalMountPoint);
282 if (avc->mvid && (avc->f.states & CMValid))
283 return 0; /* done while racing */
286 code = afs_HandleLink(avc, areq);
290 /* Determine which cell and volume the mointpoint goes to */
291 code = EvalMountData(avc->linkData[0], avc->linkData + 1,
292 avc->f.states, avc->f.fid.Cell, avolpp, areq, 0, 0,
294 if (code) return code;
304 (struct VenusFid *)osi_AllocSmallSpace(sizeof(struct VenusFid));
305 avc->mvid->Cell = (*avolpp)->cell;
306 avc->mvid->Fid.Volume = (*avolpp)->volume;
307 avc->mvid->Fid.Vnode = avnoid;
308 avc->mvid->Fid.Unique = auniq;
309 avc->f.states |= CMValid;
311 /* Used to: if the mount point is stored within a backup volume,
312 * then we should only update the parent pointer information if
313 * there's none already set, so as to avoid updating a volume's ..
314 * info with something in an OldFiles directory.
316 * Next two lines used to be under this if:
318 * if (!(avc->f.states & CBackup) || tvp->dotdot.Fid.Volume == 0)
320 * Now: update mount point back pointer on every call, so that we handle
321 * multiple mount points better. This way, when du tries to go back
322 * via chddir(".."), it will end up exactly where it started, yet
323 * cd'ing via a new path to a volume will reset the ".." pointer
326 (*avolpp)->mtpoint = avc->f.fid; /* setup back pointer to mtpoint */
329 (*avolpp)->dotdot = advc->f.fid;
337 * Must be called on an afs_fakestat_state object before calling
338 * afs_EvalFakeStat or afs_PutFakeStat. Calling afs_PutFakeStat
339 * without calling afs_EvalFakeStat is legal, as long as this
340 * function is called.
343 afs_InitFakeStat(struct afs_fakestat_state *state)
345 if (!afs_fakestat_enable)
350 state->need_release = 0;
354 * afs_EvalFakeStat_int
356 * The actual implementation of afs_EvalFakeStat and afs_TryEvalFakeStat,
357 * which is called by those wrapper functions.
359 * Only issues RPCs if canblock is non-zero.
362 afs_EvalFakeStat_int(struct vcache **avcp, struct afs_fakestat_state *state,
363 struct vrequest *areq, int canblock)
365 struct vcache *tvc, *root_vp;
366 struct volume *tvolp = NULL;
369 if (!afs_fakestat_enable)
372 osi_Assert(state->valid == 1);
373 osi_Assert(state->did_eval == 0);
377 if (tvc->mvstat != 1)
381 /* Is the call to VerifyVCache really necessary? */
382 code = afs_VerifyVCache(tvc, areq);
386 ObtainWriteLock(&tvc->lock, 599);
387 code = EvalMountPoint(tvc, NULL, &tvolp, areq);
388 ReleaseWriteLock(&tvc->lock);
392 tvolp->dotdot = tvc->f.fid;
393 tvolp->dotdot.Fid.Vnode = tvc->f.parent.vnode;
394 tvolp->dotdot.Fid.Unique = tvc->f.parent.unique;
397 if (tvc->mvid && (tvc->f.states & CMValid)) {
403 ObtainWriteLock(&afs_xvcache, 597);
404 root_vp = afs_FindVCache(tvc->mvid, &retry, IS_WLOCK);
405 if (root_vp && retry) {
406 ReleaseWriteLock(&afs_xvcache);
407 afs_PutVCache(root_vp);
409 } while (root_vp && retry);
410 ReleaseWriteLock(&afs_xvcache);
412 root_vp = afs_GetVCache(tvc->mvid, areq, NULL, NULL);
415 code = canblock ? ENOENT : 0;
418 #ifdef AFS_DARWIN80_ENV
419 root_vp->f.m.Type = VDIR;
421 code = afs_darwin_finalizevnode(root_vp, NULL, NULL, 0, 0);
424 vnode_ref(AFSTOV(root_vp));
426 if (tvolp && !afs_InReadDir(root_vp)) {
427 /* Is this always kosher? Perhaps we should instead use
428 * NBObtainWriteLock to avoid potential deadlock.
430 ObtainWriteLock(&root_vp->lock, 598);
432 root_vp->mvid = osi_AllocSmallSpace(sizeof(struct VenusFid));
433 *root_vp->mvid = tvolp->dotdot;
434 ReleaseWriteLock(&root_vp->lock);
436 state->need_release = 1;
437 state->root_vp = root_vp;
441 code = canblock ? ENOENT : 0;
446 afs_PutVolume(tvolp, WRITE_LOCK);
453 * Automatically does the equivalent of EvalMountPoint for vcache entries
454 * which are mount points. Remembers enough state to properly release
455 * the volume root vcache when afs_PutFakeStat() is called.
457 * State variable must be initialized by afs_InitFakeState() beforehand.
459 * Returns 0 when everything succeeds and *avcp points to the vcache entry
460 * that should be used for the real vnode operation. Returns non-zero if
461 * something goes wrong and the error code should be returned to the user.
464 afs_EvalFakeStat(struct vcache **avcp, struct afs_fakestat_state *state,
465 struct vrequest *areq)
467 return afs_EvalFakeStat_int(avcp, state, areq, 1);
471 * afs_TryEvalFakeStat
473 * Same as afs_EvalFakeStat, but tries not to talk to remote servers
474 * and only evaluate the mount point if all the data is already in
477 * Returns 0 if everything succeeds and *avcp points to a valid
478 * vcache entry (possibly evaluated).
481 afs_TryEvalFakeStat(struct vcache **avcp, struct afs_fakestat_state *state,
482 struct vrequest *areq)
484 return afs_EvalFakeStat_int(avcp, state, areq, 0);
490 * Perform any necessary cleanup at the end of a vnode op, given that
491 * afs_InitFakeStat was previously called with this state.
494 afs_PutFakeStat(struct afs_fakestat_state *state)
496 if (!afs_fakestat_enable)
499 osi_Assert(state->valid == 1);
500 if (state->need_release)
501 afs_PutVCache(state->root_vp);
506 afs_ENameOK(register char *aname)
510 AFS_STATCNT(ENameOK);
511 tlen = strlen(aname);
512 if (tlen >= 4 && strcmp(aname + tlen - 4, "@sys") == 0)
518 afs_getsysname(register struct vrequest *areq, register struct vcache *adp,
519 register char *bufp, int *num, char **sysnamelist[])
521 register struct unixuser *au;
522 register afs_int32 error;
524 AFS_STATCNT(getsysname);
526 *sysnamelist = afs_sysnamelist;
528 if (!afs_nfsexporter)
529 strcpy(bufp, (*sysnamelist)[0]);
531 au = afs_GetUser(areq->uid, adp->f.fid.Cell, 0);
533 error = EXP_SYSNAME(au->exporter, (char *)0, sysnamelist, num, 0);
535 strcpy(bufp, "@sys");
539 strcpy(bufp, (*sysnamelist)[0]);
542 strcpy(bufp, afs_sysname);
549 Check_AtSys(register struct vcache *avc, const char *aname,
550 struct sysname_info *state, struct vrequest *areq)
553 char **sysnamelist[MAXNUMSYSNAMES];
555 if (AFS_EQ_ATSYS(aname)) {
557 state->name = (char *)osi_AllocLargeSpace(MAXSYSNAME);
560 afs_getsysname(areq, avc, state->name, &num, sysnamelist);
565 state->name = (char *)aname;
570 Next_AtSys(register struct vcache *avc, struct vrequest *areq,
571 struct sysname_info *state)
573 int num = afs_sysnamecount;
574 char **sysnamelist[MAXNUMSYSNAMES];
576 if (state->index == -1)
577 return 0; /* No list */
579 /* Check for the initial state of aname != "@sys" in Check_AtSys */
580 if (state->offset == -1 && state->allocked == 0) {
581 register char *tname;
583 /* Check for .*@sys */
584 for (tname = state->name; *tname; tname++)
585 /*Move to the end of the string */ ;
587 if ((tname > state->name + 4) && (AFS_EQ_ATSYS(tname - 4))) {
588 state->offset = (tname - 4) - state->name;
589 tname = (char *)osi_AllocLargeSpace(AFS_LRALLOCSIZ);
590 strncpy(tname, state->name, state->offset);
595 afs_getsysname(areq, avc, state->name + state->offset, &num,
599 return 0; /* .*@sys doesn't match either */
601 register struct unixuser *au;
602 register afs_int32 error;
604 *sysnamelist = afs_sysnamelist;
606 if (afs_nfsexporter) {
607 au = afs_GetUser(areq->uid, avc->f.fid.Cell, 0);
610 EXP_SYSNAME(au->exporter, (char *)0, sysnamelist, &num, 0);
618 if (++(state->index) >= num || !(*sysnamelist)[(unsigned int)state->index])
619 return 0; /* end of list */
621 strcpy(state->name + state->offset, (*sysnamelist)[(unsigned int)state->index]);
625 extern int BlobScan(struct dcache * afile, afs_int32 ablob);
627 /* called with an unlocked directory and directory cookie. Areqp
628 * describes who is making the call.
629 * Scans the next N (about 30, typically) directory entries, and does
630 * a bulk stat call to stat them all.
632 * Must be very careful when merging in RPC responses, since we dont
633 * want to overwrite newer info that was added by a file system mutating
634 * call that ran concurrently with our bulk stat call.
636 * We do that, as described below, by not merging in our info (always
637 * safe to skip the merge) if the status info is valid in the vcache entry.
639 * If adapt ever implements the bulk stat RPC, then this code will need to
640 * ensure that vcaches created for failed RPC's to older servers have the
643 static struct vcache *BStvc = NULL;
646 afs_DoBulkStat(struct vcache *adp, long dirCookie, struct vrequest *areqp)
648 int nentries; /* # of entries to prefetch */
649 int nskip; /* # of slots in the LRU queue to skip */
650 #ifdef AFS_DARWIN80_ENV
653 struct vcache *lruvcp; /* vcache ptr of our goal pos in LRU queue */
654 struct dcache *dcp; /* chunk containing the dir block */
655 afs_size_t temp; /* temp for holding chunk length, &c. */
656 struct AFSFid *fidsp; /* file IDs were collecting */
657 struct AFSCallBack *cbsp; /* call back pointers */
658 struct AFSCallBack *tcbp; /* temp callback ptr */
659 struct AFSFetchStatus *statsp; /* file status info */
660 struct AFSVolSync volSync; /* vol sync return info */
661 struct vcache *tvcp; /* temp vcp */
662 struct afs_q *tq; /* temp queue variable */
663 AFSCBFids fidParm; /* file ID parm for bulk stat */
664 AFSBulkStats statParm; /* stat info parm for bulk stat */
665 int fidIndex = 0; /* which file were stating */
666 struct afs_conn *tcp = 0; /* conn for call */
667 AFSCBs cbParm; /* callback parm for bulk stat */
668 struct server *hostp = 0; /* host we got callback from */
669 long startTime; /* time we started the call,
670 * for callback expiration base
672 afs_size_t statSeqNo = 0; /* Valued of file size to detect races */
673 int code; /* error code */
674 long newIndex; /* new index in the dir */
675 struct DirEntry *dirEntryp; /* dir entry we are examining */
677 struct VenusFid afid; /* file ID we are using now */
678 struct VenusFid tfid; /* another temp. file ID */
679 afs_int32 retry; /* handle low-level SGI MP race conditions */
680 long volStates; /* flags from vol structure */
681 struct volume *volp = 0; /* volume ptr */
682 struct VenusFid dotdot = {0, {0, 0, 0}};
683 int flagIndex = 0; /* First file with bulk fetch flag set */
684 int inlinebulk = 0; /* Did we use InlineBulk RPC or not? */
687 dotdot.Fid.Unique = 0;
688 dotdot.Fid.Vnode = 0;
690 /* first compute some basic parameters. We dont want to prefetch more
691 * than a fraction of the cache in any given call, and we want to preserve
692 * a portion of the LRU queue in any event, so as to avoid thrashing
693 * the entire stat cache (we will at least leave some of it alone).
694 * presently dont stat more than 1/8 the cache in any one call. */
695 nentries = afs_cacheStats / 8;
697 /* dont bother prefetching more than one calls worth of info */
698 if (nentries > AFSCBMAX)
701 /* heuristic to make sure that things fit in 4K. This means that
702 * we shouldnt make it any bigger than 47 entries. I am typically
703 * going to keep it a little lower, since we don't want to load
704 * too much of the stat cache.
709 /* now, to reduce the stack size, well allocate two 4K blocks,
710 * one for fids and callbacks, and one for stat info. Well set
711 * up our pointers to the memory from there, too.
713 statsp = (AFSFetchStatus *)
714 osi_Alloc(AFSCBMAX * sizeof(AFSFetchStatus));
715 fidsp = (AFSFid *) osi_AllocLargeSpace(nentries * sizeof(AFSFid));
716 cbsp = (AFSCallBack *)
717 osi_Alloc(AFSCBMAX * sizeof(AFSCallBack));
719 /* next, we must iterate over the directory, starting from the specified
720 * cookie offset (dirCookie), and counting out nentries file entries.
721 * We skip files that already have stat cache entries, since we
722 * dont want to bulk stat files that are already in the cache.
725 code = afs_VerifyVCache(adp, areqp);
729 dcp = afs_GetDCache(adp, (afs_size_t) 0, areqp, &temp, &temp, 1);
735 /* lock the directory cache entry */
736 ObtainReadLock(&adp->lock);
737 ObtainReadLock(&dcp->lock);
740 * Make sure that the data in the cache is current. There are two
741 * cases we need to worry about:
742 * 1. The cache data is being fetched by another process.
743 * 2. The cache data is no longer valid
745 while ((adp->f.states & CStatd)
746 && (dcp->dflags & DFFetching)
747 && hsame(adp->f.m.DataVersion, dcp->f.versionNo)) {
748 afs_Trace4(afs_iclSetp, CM_TRACE_DCACHEWAIT, ICL_TYPE_STRING,
749 __FILE__, ICL_TYPE_INT32, __LINE__, ICL_TYPE_POINTER, dcp,
750 ICL_TYPE_INT32, dcp->dflags);
751 ReleaseReadLock(&dcp->lock);
752 ReleaseReadLock(&adp->lock);
753 afs_osi_Sleep(&dcp->validPos);
754 ObtainReadLock(&adp->lock);
755 ObtainReadLock(&dcp->lock);
757 if (!(adp->f.states & CStatd)
758 || !hsame(adp->f.m.DataVersion, dcp->f.versionNo)) {
759 ReleaseReadLock(&dcp->lock);
760 ReleaseReadLock(&adp->lock);
765 /* Generate a sequence number so we can tell whether we should
766 * store the attributes when processing the response. This number is
767 * stored in the file size when we set the CBulkFetching bit. If the
768 * CBulkFetching is still set and this value hasn't changed, then
769 * we know we were the last to set CBulkFetching bit for this file,
770 * and it is safe to set the status information for this file.
772 statSeqNo = bulkStatCounter++;
773 /* ensure against wrapping */
775 statSeqNo = bulkStatCounter++;
777 /* now we have dir data in the cache, so scan the dir page */
780 while (1) { /* Should probably have some constant bound */
781 /* look for first safe entry to examine in the directory. BlobScan
782 * looks for a the 1st allocated dir after the dirCookie slot.
784 newIndex = BlobScan(dcp, (dirCookie >> 5));
788 /* remember the updated directory cookie */
789 dirCookie = newIndex << 5;
791 /* get a ptr to the dir entry */
793 (struct DirEntry *)afs_dir_GetBlob(dcp, newIndex);
797 /* dont copy more than we have room for */
798 if (fidIndex >= nentries) {
799 DRelease(dirEntryp, 0);
803 /* now, if the dir entry looks good, copy it out to our list. Vnode
804 * 0 means deleted, although it should also be free were it deleted.
806 if (dirEntryp->fid.vnode != 0) {
807 /* dont copy entries we have in our cache. This check will
808 * also make us skip "." and probably "..", unless it has
809 * disappeared from the cache since we did our namei call.
811 tfid.Cell = adp->f.fid.Cell;
812 tfid.Fid.Volume = adp->f.fid.Fid.Volume;
813 tfid.Fid.Vnode = ntohl(dirEntryp->fid.vnode);
814 tfid.Fid.Unique = ntohl(dirEntryp->fid.vunique);
817 ObtainWriteLock(&afs_xvcache, 130);
818 tvcp = afs_FindVCache(&tfid, &retry, IS_WLOCK /* no stats | LRU */ );
820 ReleaseWriteLock(&afs_xvcache);
823 } while (tvcp && retry);
824 if (!tvcp) { /* otherwise, create manually */
825 tvcp = afs_NewBulkVCache(&tfid, hostp, statSeqNo);
828 ObtainWriteLock(&tvcp->lock, 505);
829 ReleaseWriteLock(&afs_xvcache);
830 afs_RemoveVCB(&tfid);
831 ReleaseWriteLock(&tvcp->lock);
833 ReleaseWriteLock(&afs_xvcache);
836 ReleaseWriteLock(&afs_xvcache);
840 DRelease(dirEntryp, 0);
841 ReleaseReadLock(&dcp->lock);
842 ReleaseReadLock(&adp->lock);
844 goto done; /* can happen if afs_NewVCache fails */
847 /* WARNING: afs_DoBulkStat uses the Length field to store a
848 * sequence number for each bulk status request. Under no
849 * circumstances should afs_DoBulkStat store a sequence number
850 * if the new length will be ignored when afs_ProcessFS is
851 * called with new stats. */
853 if (!(tvcp->f.states & CStatd)
854 && (!((tvcp->f.states & CBulkFetching) &&
855 (tvcp->f.m.Length != statSeqNo)))
856 && (tvcp->execsOrWriters <= 0)
857 && !afs_DirtyPages(tvcp)
858 && !AFS_VN_MAPPED((vnode_t *) tvcp))
860 if (!(tvcp->f.states & CStatd)
861 && (!((tvcp->f.states & CBulkFetching) &&
862 (tvcp->f.m.Length != statSeqNo)))
863 && (tvcp->execsOrWriters <= 0)
864 && !afs_DirtyPages(tvcp))
868 /* this entry doesnt exist in the cache, and is not
869 * already being fetched by someone else, so add it to the
870 * list of file IDs to obtain.
872 * We detect a callback breaking race condition by checking the
873 * CBulkFetching state bit and the value in the file size.
874 * It is safe to set the status only if the CBulkFetching
875 * flag is still set and the value in the file size does
876 * not change. NewBulkVCache sets us up.
878 * Don't fetch status for dirty files. We need to
879 * preserve the value of the file size. We could
880 * flush the pages, but it wouldn't be worthwhile.
882 memcpy((char *)(fidsp + fidIndex), (char *)&tfid.Fid,
889 /* if dir vnode has non-zero entry */
890 /* move to the next dir entry by adding in the # of entries
891 * used by this dir entry.
893 temp = afs_dir_NameBlobs(dirEntryp->name) << 5;
894 DRelease(dirEntryp, 0);
898 } /* while loop over all dir entries */
900 /* now release the dir lock and prepare to make the bulk RPC */
901 ReleaseReadLock(&dcp->lock);
902 ReleaseReadLock(&adp->lock);
904 /* release the chunk */
907 /* dont make a null call */
912 /* setup the RPC parm structures */
913 fidParm.AFSCBFids_len = fidIndex;
914 fidParm.AFSCBFids_val = fidsp;
915 statParm.AFSBulkStats_len = fidIndex;
916 statParm.AFSBulkStats_val = statsp;
917 cbParm.AFSCBs_len = fidIndex;
918 cbParm.AFSCBs_val = cbsp;
920 /* start the timer; callback expirations are relative to this */
921 startTime = osi_Time();
923 tcp = afs_Conn(&adp->f.fid, areqp, SHARED_LOCK);
925 hostp = tcp->srvr->server;
926 XSTATS_START_TIME(AFS_STATS_FS_RPCIDX_BULKSTATUS);
929 if (!(tcp->srvr->server->flags & SNO_INLINEBULK)) {
931 RXAFS_InlineBulkStatus(tcp->id, &fidParm, &statParm,
933 if (code == RXGEN_OPCODE) {
934 tcp->srvr->server->flags |= SNO_INLINEBULK;
937 RXAFS_BulkStatus(tcp->id, &fidParm, &statParm,
944 RXAFS_BulkStatus(tcp->id, &fidParm, &statParm, &cbParm,
952 (tcp, code, &adp->f.fid, areqp, AFS_STATS_FS_RPCIDX_BULKSTATUS,
955 /* now, if we didnt get the info, bail out. */
959 /* we need vol flags to create the entries properly */
960 dotdot.Fid.Volume = 0;
961 volp = afs_GetVolume(&adp->f.fid, areqp, READ_LOCK);
963 volStates = volp->states;
964 if (volp->dotdot.Fid.Volume != 0)
965 dotdot = volp->dotdot;
969 /* find the place to merge the info into We do this by skipping
970 * nskip entries in the LRU queue. The more we skip, the more
971 * we preserve, since the head of the VLRU queue is the most recently
975 nskip = afs_cacheStats / 2; /* preserved fraction of the cache */
976 ObtainReadLock(&afs_xvcache);
978 /* actually a serious error, probably should panic. Probably will
979 * panic soon, oh well. */
980 ReleaseReadLock(&afs_xvcache);
981 afs_warnuser("afs_DoBulkStat: VLRU empty!");
984 if ((VLRU.next->prev != &VLRU) || (VLRU.prev->next != &VLRU)) {
985 refpanic("Bulkstat VLRU inconsistent");
987 for (tq = VLRU.next; tq != &VLRU; tq = QNext(tq)) {
989 #ifdef AFS_DARWIN80_ENV
990 if (!(QTOV(tq)->f.states & CDeadVnode))
994 if (QNext(QPrev(tq)) != tq) {
996 refpanic("BulkStat VLRU inconsistent");
1002 lruvcp = QTOV(VLRU.next);
1004 /* now we have to hold this entry, so that it does not get moved
1005 * into the free list while we're running. It could still get
1006 * moved within the lru queue, but hopefully that will be rare; it
1007 * doesn't hurt nearly as much.
1010 #ifdef AFS_DARWIN80_ENV
1011 lruvp = AFSTOV(lruvcp);
1012 if (vnode_get(lruvp)) /* this bumps ref count */
1014 else if (vnode_ref(lruvp)) {
1016 /* AFSTOV(lruvcp) may be NULL */
1022 osi_vnhold(lruvcp, &retry);
1024 ReleaseReadLock(&afs_xvcache); /* could be read lock */
1028 /* otherwise, merge in the info. We have to be quite careful here,
1029 * since we need to ensure that we don't merge old info over newer
1030 * stuff in a stat cache entry. We're very conservative here: we don't
1031 * do the merge at all unless we ourselves create the stat cache
1032 * entry. That's pretty safe, and should work pretty well, since we
1033 * typically expect to do the stat cache creation ourselves.
1035 * We also have to take into account racing token revocations.
1037 for (i = 0; i < fidIndex; i++) {
1038 if ((&statsp[i])->errorCode)
1040 afid.Cell = adp->f.fid.Cell;
1041 afid.Fid.Volume = adp->f.fid.Fid.Volume;
1042 afid.Fid.Vnode = fidsp[i].Vnode;
1043 afid.Fid.Unique = fidsp[i].Unique;
1046 ObtainReadLock(&afs_xvcache);
1047 tvcp = afs_FindVCache(&afid, &retry, 0 /* !stats&!lru */ );
1048 ReleaseReadLock(&afs_xvcache);
1049 } while (tvcp && retry);
1051 /* The entry may no longer exist */
1056 /* now we have the entry held, but we need to fill it in */
1057 ObtainWriteLock(&tvcp->lock, 131);
1059 /* if CBulkFetching is not set, or if the file size no longer
1060 * matches the value we placed there when we set the CBulkFetching
1061 * flag, then someone else has done something with this node,
1062 * and we may not have the latest status information for this
1063 * file. Leave the entry alone.
1065 if (!(tvcp->f.states & CBulkFetching) || (tvcp->f.m.Length != statSeqNo)) {
1066 #ifdef AFS_DARWIN80_ENV
1067 int isdead = (tvcp->f.states & CDeadVnode);
1070 ReleaseWriteLock(&tvcp->lock);
1071 #ifdef AFS_DARWIN80_ENV
1073 /* re-acquire the usecount that the other finalizevnode disposed of */
1074 vnode_ref(AFSTOV(tvcp));
1076 afs_PutVCache(tvcp);
1080 /* now copy ".." entry back out of volume structure, if necessary */
1081 if (tvcp->mvstat == 2 && (dotdot.Fid.Volume != 0)) {
1083 tvcp->mvid = (struct VenusFid *)
1084 osi_AllocSmallSpace(sizeof(struct VenusFid));
1085 *tvcp->mvid = dotdot;
1088 ObtainWriteLock(&afs_xvcache, 132);
1089 if ((VLRU.next->prev != &VLRU) || (VLRU.prev->next != &VLRU)) {
1090 refpanic("Bulkstat VLRU inconsistent2");
1092 if ((QNext(QPrev(&tvcp->vlruq)) != &tvcp->vlruq)
1093 || (QPrev(QNext(&tvcp->vlruq)) != &tvcp->vlruq)) {
1094 refpanic("Bulkstat VLRU inconsistent4");
1096 if ((QNext(QPrev(&lruvcp->vlruq)) != &lruvcp->vlruq)
1097 || (QPrev(QNext(&lruvcp->vlruq)) != &lruvcp->vlruq)) {
1098 refpanic("Bulkstat VLRU inconsistent5");
1101 if (tvcp != lruvcp) { /* if they are == don't move it, don't corrupt vlru */
1102 QRemove(&tvcp->vlruq);
1103 QAdd(&lruvcp->vlruq, &tvcp->vlruq);
1106 if ((VLRU.next->prev != &VLRU) || (VLRU.prev->next != &VLRU)) {
1107 refpanic("Bulkstat VLRU inconsistent3");
1109 if ((QNext(QPrev(&tvcp->vlruq)) != &tvcp->vlruq)
1110 || (QPrev(QNext(&tvcp->vlruq)) != &tvcp->vlruq)) {
1111 refpanic("Bulkstat VLRU inconsistent5");
1113 if ((QNext(QPrev(&lruvcp->vlruq)) != &lruvcp->vlruq)
1114 || (QPrev(QNext(&lruvcp->vlruq)) != &lruvcp->vlruq)) {
1115 refpanic("Bulkstat VLRU inconsistent6");
1117 ReleaseWriteLock(&afs_xvcache);
1119 ObtainWriteLock(&afs_xcbhash, 494);
1121 /* We need to check the flags again. We may have missed
1122 * something while we were waiting for a lock.
1124 if (!(tvcp->f.states & CBulkFetching) || (tvcp->f.m.Length != statSeqNo)) {
1126 #ifdef AFS_DARWIN80_ENV
1127 if ((tvcp->f.states & CDeadVnode) == 0)
1128 /* re-acquire the usecount that the other finalizevnode disposed of */
1129 vnode_ref(AFSTOV(tvcp));
1131 ReleaseWriteLock(&tvcp->lock);
1132 ReleaseWriteLock(&afs_xcbhash);
1133 afs_PutVCache(tvcp);
1137 /* now merge in the resulting status back into the vnode.
1138 * We only do this if the entry looks clear.
1140 afs_ProcessFS(tvcp, &statsp[i], areqp);
1141 #if defined(AFS_LINUX22_ENV)
1142 afs_fill_inode(AFSTOV(tvcp), NULL); /* reset inode operations */
1145 /* do some accounting for bulk stats: mark this entry as
1146 * loaded, so we can tell if we use it before it gets
1149 tvcp->f.states |= CBulkStat;
1150 tvcp->f.states &= ~CBulkFetching;
1152 afs_bulkStatsDone++;
1154 /* merge in vol info */
1155 if (volStates & VRO)
1156 tvcp->f.states |= CRO;
1157 if (volStates & VBackup)
1158 tvcp->f.states |= CBackup;
1159 if (volStates & VForeign)
1160 tvcp->f.states |= CForeign;
1162 /* merge in the callback info */
1163 tvcp->f.states |= CTruth;
1165 /* get ptr to the callback we are interested in */
1168 if (tcbp->ExpirationTime != 0) {
1169 tvcp->cbExpires = tcbp->ExpirationTime + startTime;
1170 tvcp->callback = hostp;
1171 tvcp->f.states |= CStatd;
1172 afs_QueueCallback(tvcp, CBHash(tcbp->ExpirationTime), volp);
1173 } else if (tvcp->f.states & CRO) {
1174 /* ordinary callback on a read-only volume -- AFS 3.2 style */
1175 tvcp->cbExpires = 3600 + startTime;
1176 tvcp->callback = hostp;
1177 tvcp->f.states |= CStatd;
1178 afs_QueueCallback(tvcp, CBHash(3600), volp);
1181 tvcp->f.states &= ~(CStatd | CUnique);
1182 afs_DequeueCallback(tvcp);
1183 if ((tvcp->f.states & CForeign) || (vType(tvcp) == VDIR))
1184 osi_dnlc_purgedp(tvcp); /* if it (could be) a directory */
1186 ReleaseWriteLock(&afs_xcbhash);
1187 #ifdef AFS_DARWIN80_ENV
1188 /* reclaim->FlushVCache will need xcbhash */
1189 if (tvcp->f.states & CDeadVnode) {
1190 /* passing in a parent hangs getting the vnode lock */
1191 code = afs_darwin_finalizevnode(tvcp, NULL, NULL, 0, 1);
1193 /* It's gonna get recycled - shouldn't happen */
1195 tvcp->f.states &= ~(CStatd | CUnique);
1196 afs_DequeueCallback(tvcp);
1197 if ((tvcp->f.states & CForeign) || (vType(tvcp) == VDIR))
1198 osi_dnlc_purgedp(tvcp); /* if it (could be) a directory */
1200 /* re-acquire the usecount that finalizevnode disposed of */
1201 vnode_ref(AFSTOV(tvcp));
1205 ReleaseWriteLock(&tvcp->lock);
1206 /* finally, we're done with the entry */
1207 afs_PutVCache(tvcp);
1208 } /* for all files we got back */
1210 /* finally return the pointer into the LRU queue */
1211 #ifdef AFS_DARWIN80_ENV
1217 afs_PutVCache(lruvcp);
1221 /* Be sure to turn off the CBulkFetching flags */
1222 for (i = flagIndex; i < fidIndex; i++) {
1223 afid.Cell = adp->f.fid.Cell;
1224 afid.Fid.Volume = adp->f.fid.Fid.Volume;
1225 afid.Fid.Vnode = fidsp[i].Vnode;
1226 afid.Fid.Unique = fidsp[i].Unique;
1229 ObtainReadLock(&afs_xvcache);
1230 tvcp = afs_FindVCache(&afid, &retry, 0 /* !stats&!lru */ );
1231 ReleaseReadLock(&afs_xvcache);
1232 } while (tvcp && retry);
1233 if (tvcp != NULL && (tvcp->f.states & CBulkFetching)
1234 && (tvcp->f.m.Length == statSeqNo)) {
1235 tvcp->f.states &= ~CBulkFetching;
1238 #ifdef AFS_DARWIN80_ENV
1239 if ((tvcp->f.states & CDeadVnode) == 0)
1240 /* re-acquire the usecount that the other finalizevnode disposed of */
1241 vnode_ref(AFSTOV(tvcp));
1243 afs_PutVCache(tvcp);
1247 afs_PutVolume(volp, READ_LOCK);
1249 /* If we did the InlineBulk RPC pull out the return code */
1250 if (inlinebulk && code == 0) {
1251 if ((&statsp[0])->errorCode) {
1252 afs_Analyze(tcp, (&statsp[0])->errorCode, &adp->f.fid, areqp,
1253 AFS_STATS_FS_RPCIDX_BULKSTATUS, SHARED_LOCK, NULL);
1254 code = (&statsp[0])->errorCode;
1260 osi_FreeLargeSpace((char *)fidsp);
1261 osi_Free((char *)statsp, AFSCBMAX * sizeof(AFSFetchStatus));
1262 osi_Free((char *)cbsp, AFSCBMAX * sizeof(AFSCallBack));
1266 /* was: (AFS_DEC_ENV) || defined(AFS_OSF30_ENV) || defined(AFS_NCR_ENV) */
1267 static int AFSDOBULK = 1;
1270 #if defined(AFS_SUN5_ENV) || defined(AFS_SGI_ENV)
1271 afs_lookup(OSI_VC_DECL(adp), char *aname, struct vcache **avcp, struct pathname *pnp, int flags, struct vnode *rdir, afs_ucred_t *acred)
1272 #elif defined(UKERNEL)
1273 afs_lookup(OSI_VC_DECL(adp), char *aname, struct vcache **avcp, afs_ucred_t *acred, int flags)
1275 afs_lookup(OSI_VC_DECL(adp), char *aname, struct vcache **avcp, afs_ucred_t *acred)
1278 struct vrequest treq;
1280 register struct vcache *tvc = 0;
1281 register afs_int32 code;
1282 register afs_int32 bulkcode = 0;
1283 int pass = 0, hit = 0;
1284 int force_eval = afs_fakestat_enable ? 0 : 1;
1286 extern afs_int32 afs_mariner; /*Writing activity to log? */
1287 afs_hyper_t versionNo;
1288 int no_read_access = 0;
1289 struct sysname_info sysState; /* used only for @sys checking */
1290 int dynrootRetry = 1;
1291 struct afs_fakestat_state fakestate;
1292 int tryEvalOnly = 0;
1293 OSI_VC_CONVERT(adp);
1295 AFS_STATCNT(afs_lookup);
1296 afs_InitFakeStat(&fakestate);
1300 if ((code = afs_InitReq(&treq, acred)))
1303 if (afs_fakestat_enable && adp->mvstat == 1) {
1304 if (strcmp(aname, ".directory") == 0)
1308 #if defined(AFS_DARWIN_ENV)
1309 /* Workaround for MacOSX Finder, which tries to look for
1310 * .DS_Store and Contents under every directory.
1312 if (afs_fakestat_enable && adp->mvstat == 1) {
1313 if (strcmp(aname, ".DS_Store") == 0)
1315 if (strcmp(aname, "Contents") == 0)
1318 if (afs_fakestat_enable && adp->mvstat == 2) {
1319 if (strncmp(aname, "._", 2) == 0)
1325 code = afs_TryEvalFakeStat(&adp, &fakestate, &treq);
1327 code = afs_EvalFakeStat(&adp, &fakestate, &treq);
1329 /*printf("Code is %d\n", code);*/
1331 if (tryEvalOnly && adp->mvstat == 1)
1336 *avcp = NULL; /* Since some callers don't initialize it */
1338 /* come back to here if we encounter a non-existent object in a read-only
1339 * volume's directory */
1342 *avcp = NULL; /* Since some callers don't initialize it */
1345 if (!(adp->f.states & CStatd) && !afs_InReadDir(adp)) {
1346 if ((code = afs_VerifyVCache2(adp, &treq))) {
1352 /* watch for ".." in a volume root */
1353 if (adp->mvstat == 2 && aname[0] == '.' && aname[1] == '.' && !aname[2]) {
1354 /* looking up ".." in root via special hacks */
1355 if (adp->mvid == (struct VenusFid *)0 || adp->mvid->Fid.Volume == 0) {
1359 /* otherwise we have the fid here, so we use it */
1360 /*printf("Getting vcache\n");*/
1361 tvc = afs_GetVCache(adp->mvid, &treq, NULL, NULL);
1362 afs_Trace3(afs_iclSetp, CM_TRACE_GETVCDOTDOT, ICL_TYPE_FID, adp->mvid,
1363 ICL_TYPE_POINTER, tvc, ICL_TYPE_INT32, code);
1365 code = (tvc ? 0 : ENOENT);
1367 if (tvc && !VREFCOUNT_GT(tvc, 0)) {
1371 /*printf("LOOKUP GETVCDOTDOT -> %d\n", code); */
1376 /* now check the access */
1377 if (treq.uid != adp->last_looker) {
1378 if (!afs_AccessOK(adp, PRSFS_LOOKUP, &treq, CHECK_MODE_BITS)) {
1383 adp->last_looker = treq.uid;
1386 /* Check for read access as well. We need read access in order to
1387 * stat files, but not to stat subdirectories. */
1388 if (!afs_AccessOK(adp, PRSFS_READ, &treq, CHECK_MODE_BITS))
1391 /* special case lookup of ".". Can we check for it sooner in this code,
1392 * for instance, way up before "redo:" ??
1393 * I'm not fiddling with the LRUQ here, either, perhaps I should, or else
1394 * invent a lightweight version of GetVCache.
1396 if (aname[0] == '.' && !aname[1]) { /* special case */
1397 ObtainReadLock(&afs_xvcache);
1399 ReleaseReadLock(&afs_xvcache);
1400 #ifdef AFS_DARWIN80_ENV
1401 vnode_get(AFSTOV(adp));
1406 if (adp && !VREFCOUNT_GT(adp, 0)) {
1413 * Special case lookup of ".." in the dynamic mount directory.
1414 * The parent of this directory is _always_ the AFS root volume.
1416 if (afs_IsDynrootMount(adp) &&
1417 aname[0] == '.' && aname[1] == '.' && !aname[2]) {
1419 ObtainReadLock(&afs_xvcache);
1420 osi_vnhold(afs_globalVp, 0);
1421 ReleaseReadLock(&afs_xvcache);
1422 #ifdef AFS_DARWIN80_ENV
1423 vnode_get(AFSTOV(afs_globalVp));
1426 *avcp = tvc = afs_globalVp;
1432 * Special case lookups in the dynamic mount directory.
1433 * The names here take the form cell:volume, similar to a mount point.
1434 * EvalMountData parses that and returns a cell and volume ID, which
1435 * we use to construct the appropriate dynroot Fid.
1437 if (afs_IsDynrootMount(adp)) {
1438 struct VenusFid tfid;
1439 afs_uint32 cellidx, volid, vnoid, uniq;
1441 code = EvalMountData('%', aname, 0, 0, NULL, &treq, &cellidx, &volid, &vnoid, &uniq);
1444 /* If a vnode was returned, it's not a real mount point */
1446 struct cell *tcell = afs_GetCellByIndex(cellidx, READ_LOCK);
1447 tfid.Cell = tcell->cellNum;
1448 afs_PutCell(tcell, READ_LOCK);
1449 tfid.Fid.Vnode = vnoid;
1450 tfid.Fid.Volume = volid;
1451 tfid.Fid.Unique = uniq;
1453 afs_GetDynrootMountFid(&tfid);
1454 tfid.Fid.Vnode = VNUM_FROM_TYPEID(VN_TYPE_MOUNT, cellidx << 2);
1455 tfid.Fid.Unique = volid;
1457 *avcp = tvc = afs_GetVCache(&tfid, &treq, NULL, NULL);
1458 code = (tvc ? 0 : ENOENT);
1463 #ifdef AFS_LINUX26_ENV
1465 * Special case of the dynamic mount volume in a static root.
1466 * This is really unfortunate, but we need this for the translator.
1468 if (adp == afs_globalVp && !afs_GetDynrootEnable() &&
1469 !strcmp(aname, AFS_DYNROOT_MOUNTNAME)) {
1470 struct VenusFid tfid;
1472 afs_GetDynrootMountFid(&tfid);
1473 *avcp = tvc = afs_GetVCache(&tfid, &treq, NULL, NULL);
1480 Check_AtSys(adp, aname, &sysState, &treq);
1481 tname = sysState.name;
1483 /* 1st Check_AtSys and lookup by tname is required here, for now,
1484 * because the dnlc is *not* told to remove entries for the parent
1485 * dir of file/dir op that afs_LocalHero likes, but dnlc is informed
1486 * if the cached entry for the parent dir is invalidated for a
1488 * Otherwise, we'd be able to do a dnlc lookup on an entry ending
1489 * w/@sys and know the dnlc was consistent with reality. */
1490 tvc = osi_dnlc_lookup(adp, tname, WRITE_LOCK);
1491 *avcp = tvc; /* maybe wasn't initialized, but it is now */
1493 if (no_read_access && vType(tvc) != VDIR && vType(tvc) != VLNK) {
1494 /* need read access on dir to stat non-directory / non-link */
1495 #ifndef AFS_FBSD80_ENV
1502 #ifdef AFS_LINUX22_ENV
1503 if (tvc->mvstat == 2) { /* we don't trust the dnlc for root vcaches */
1504 AFS_RELE(AFSTOV(tvc));
1511 #else /* non - LINUX */
1515 #endif /* linux22 */
1518 { /* sub-block just to reduce stack usage */
1519 register struct dcache *tdc;
1520 afs_size_t dirOffset, dirLen;
1521 struct VenusFid tfid;
1523 /* now we have to lookup the next fid */
1524 if (afs_InReadDir(adp))
1525 tdc = adp->dcreaddir;
1527 tdc = afs_GetDCache(adp, (afs_size_t) 0, &treq,
1528 &dirOffset, &dirLen, 1);
1530 *avcp = NULL; /* redundant, but harmless */
1535 /* now we will just call dir package with appropriate inode.
1536 * Dirs are always fetched in their entirety for now */
1537 ObtainReadLock(&adp->lock);
1538 ObtainReadLock(&tdc->lock);
1541 * Make sure that the data in the cache is current. There are two
1542 * cases we need to worry about:
1543 * 1. The cache data is being fetched by another process.
1544 * 2. The cache data is no longer valid
1546 * If a readdir is in progress _in this thread_, it has a shared
1547 * lock on the vcache and has obtained current data, so we just
1548 * use that. This eliminates several possible deadlocks.
1550 if (!afs_InReadDir(adp)) {
1551 while ((adp->f.states & CStatd)
1552 && (tdc->dflags & DFFetching)
1553 && hsame(adp->f.m.DataVersion, tdc->f.versionNo)) {
1554 ReleaseReadLock(&tdc->lock);
1555 ReleaseReadLock(&adp->lock);
1556 afs_osi_Sleep(&tdc->validPos);
1557 ObtainReadLock(&adp->lock);
1558 ObtainReadLock(&tdc->lock);
1560 if (!(adp->f.states & CStatd)
1561 || !hsame(adp->f.m.DataVersion, tdc->f.versionNo)) {
1562 ReleaseReadLock(&tdc->lock);
1563 ReleaseReadLock(&adp->lock);
1565 if (tname && tname != aname)
1566 osi_FreeLargeSpace(tname);
1571 /* Save the version number for when we call osi_dnlc_enter */
1572 hset(versionNo, tdc->f.versionNo);
1575 * check for, and handle "@sys" if it's there. We should be able
1576 * to avoid the alloc and the strcpy with a little work, but it's
1577 * not pressing. If there aren't any remote users (ie, via the
1578 * NFS translator), we have a slightly easier job.
1579 * the faster way to do this is to check for *aname == '@' and if
1580 * it's there, check for @sys, otherwise, assume there's no @sys
1581 * then, if the lookup fails, check for .*@sys...
1583 /* above now implemented by Check_AtSys and Next_AtSys */
1585 /* lookup the name in the appropriate dir, and return a cache entry
1586 * on the resulting fid */
1588 afs_dir_LookupOffset(tdc, sysState.name, &tfid.Fid,
1591 /* If the first lookup doesn't succeed, maybe it's got @sys in the name */
1592 while (code == ENOENT && Next_AtSys(adp, &treq, &sysState))
1594 afs_dir_LookupOffset(tdc, sysState.name, &tfid.Fid,
1596 tname = sysState.name;
1598 ReleaseReadLock(&tdc->lock);
1599 if (!afs_InReadDir(adp))
1601 if (code == ENOENT && afs_IsDynroot(adp) && dynrootRetry && !tryEvalOnly) {
1602 ReleaseReadLock(&adp->lock);
1604 if (tname[0] == '.')
1605 afs_LookupAFSDB(tname + 1);
1607 afs_LookupAFSDB(tname);
1608 if (tname && tname != aname)
1609 osi_FreeLargeSpace(tname);
1612 ReleaseReadLock(&adp->lock);
1615 /* new fid has same cell and volume */
1616 tfid.Cell = adp->f.fid.Cell;
1617 tfid.Fid.Volume = adp->f.fid.Fid.Volume;
1618 afs_Trace4(afs_iclSetp, CM_TRACE_LOOKUP, ICL_TYPE_POINTER, adp,
1619 ICL_TYPE_STRING, tname, ICL_TYPE_FID, &tfid,
1620 ICL_TYPE_INT32, code);
1623 if (code != ENOENT) {
1624 /*printf("LOOKUP dirLookupOff -> %d\n", code);*/
1629 /* prefetch some entries, if the dir is currently open. The variable
1630 * dirCookie tells us where to start prefetching from.
1632 if (!AFS_IS_DISCONNECTED &&
1633 AFSDOBULK && adp->opens > 0 && !(adp->f.states & CForeign)
1634 && !afs_IsDynroot(adp) && !afs_InReadDir(adp)) {
1636 /* if the entry is not in the cache, or is in the cache,
1637 * but hasn't been statd, then do a bulk stat operation.
1641 ObtainReadLock(&afs_xvcache);
1642 tvc = afs_FindVCache(&tfid, &retry, 0 /* !stats,!lru */ );
1643 ReleaseReadLock(&afs_xvcache);
1644 } while (tvc && retry);
1646 if (!tvc || !(tvc->f.states & CStatd))
1647 bulkcode = afs_DoBulkStat(adp, dirCookie, &treq);
1651 /* if the vcache isn't usable, release it */
1652 if (tvc && !(tvc->f.states & CStatd)) {
1653 #ifndef AFS_FBSD80_ENV
1663 /* now get the status info, if we don't already have it */
1664 /* This is kind of weird, but we might wind up accidentally calling
1665 * RXAFS_Lookup because we happened upon a file which legitimately
1666 * has a 0 uniquifier. That is the result of allowing unique to wrap
1667 * to 0. This was fixed in AFS 3.4. For CForeign, Unique == 0 means that
1668 * the file has not yet been looked up.
1671 afs_int32 cached = 0;
1672 if (!tfid.Fid.Unique && (adp->f.states & CForeign)) {
1673 tvc = afs_LookupVCache(&tfid, &treq, &cached, adp, tname);
1675 if (!tvc && !bulkcode) { /* lookup failed or wasn't called */
1676 tvc = afs_GetVCache(&tfid, &treq, &cached, NULL);
1679 } /* sub-block just to reduce stack usage */
1682 if (adp->f.states & CForeign)
1683 tvc->f.states |= CForeign;
1684 tvc->f.parent.vnode = adp->f.fid.Fid.Vnode;
1685 tvc->f.parent.unique = adp->f.fid.Fid.Unique;
1686 tvc->f.states &= ~CBulkStat;
1688 if (afs_fakestat_enable == 2 && tvc->mvstat == 1) {
1689 ObtainSharedLock(&tvc->lock, 680);
1690 if (!tvc->linkData) {
1691 UpgradeSToWLock(&tvc->lock, 681);
1692 code = afs_HandleLink(tvc, &treq);
1693 ConvertWToRLock(&tvc->lock);
1695 ConvertSToRLock(&tvc->lock);
1698 if (!code && !afs_strchr(tvc->linkData, ':'))
1700 ReleaseReadLock(&tvc->lock);
1702 if (tvc->mvstat == 1 && (tvc->f.states & CMValid) && tvc->mvid != NULL)
1703 force_eval = 1; /* This is now almost for free, get it correct */
1705 #if defined(UKERNEL) && defined(AFS_WEB_ENHANCEMENTS)
1706 if (!(flags & AFS_LOOKUP_NOEVAL))
1707 /* don't eval mount points */
1708 #endif /* UKERNEL && AFS_WEB_ENHANCEMENTS */
1709 if (tvc->mvstat == 1 && force_eval) {
1710 /* a mt point, possibly unevaluated */
1711 struct volume *tvolp;
1713 ObtainWriteLock(&tvc->lock, 133);
1714 code = EvalMountPoint(tvc, adp, &tvolp, &treq);
1715 ReleaseWriteLock(&tvc->lock);
1718 #ifndef AFS_FBSD80_ENV
1722 afs_PutVolume(tvolp, WRITE_LOCK);
1726 /* next, we want to continue using the target of the mt point */
1727 if (tvc->mvid && (tvc->f.states & CMValid)) {
1729 /* now lookup target, to set .. pointer */
1730 afs_Trace2(afs_iclSetp, CM_TRACE_LOOKUP1,
1731 ICL_TYPE_POINTER, tvc, ICL_TYPE_FID,
1733 uvc = tvc; /* remember for later */
1735 if (tvolp && (tvolp->states & VForeign)) {
1736 /* XXXX tvolp has ref cnt on but not locked! XXX */
1738 afs_GetRootVCache(tvc->mvid, &treq, NULL, tvolp);
1740 tvc = afs_GetVCache(tvc->mvid, &treq, NULL, NULL);
1742 #ifndef AFS_FBSD80_ENV
1743 afs_PutVCache(uvc); /* we're done with it */
1749 afs_PutVolume(tvolp, WRITE_LOCK);
1754 /* now, if we came via a new mt pt (say because of a new
1755 * release of a R/O volume), we must reevaluate the ..
1756 * ptr to point back to the appropriate place */
1758 ObtainWriteLock(&tvc->lock, 134);
1759 if (tvc->mvid == NULL) {
1760 tvc->mvid = (struct VenusFid *)
1761 osi_AllocSmallSpace(sizeof(struct VenusFid));
1763 /* setup backpointer */
1764 *tvc->mvid = tvolp->dotdot;
1765 ReleaseWriteLock(&tvc->lock);
1766 afs_PutVolume(tvolp, WRITE_LOCK);
1769 #ifndef AFS_FBSD80_ENV
1774 afs_PutVolume(tvolp, WRITE_LOCK);
1779 if (tvc && !VREFCOUNT_GT(tvc, 0)) {
1784 /* if we get here, we found something in a directory that couldn't
1785 * be located (a Multics "connection failure"). If the volume is
1786 * read-only, we try flushing this entry from the cache and trying
1788 if (!AFS_IS_DISCONNECTED) {
1791 tv = afs_GetVolume(&adp->f.fid, &treq, READ_LOCK);
1793 if (tv->states & VRO) {
1794 pass = 1; /* try this *once* */
1795 ObtainWriteLock(&afs_xcbhash, 495);
1796 afs_DequeueCallback(adp);
1797 /* re-stat to get later version */
1798 adp->f.states &= ~CStatd;
1799 ReleaseWriteLock(&afs_xcbhash);
1800 osi_dnlc_purgedp(adp);
1801 afs_PutVolume(tv, READ_LOCK);
1804 afs_PutVolume(tv, READ_LOCK);
1814 /* put the network buffer back, if need be */
1815 if (tname != aname && tname)
1816 osi_FreeLargeSpace(tname);
1820 afs_AddMarinerName(aname, tvc);
1822 #if defined(UKERNEL) && defined(AFS_WEB_ENHANCEMENTS)
1823 if (!(flags & AFS_LOOKUP_NOEVAL)) {
1824 /* Here we don't enter the name into the DNLC because we want the
1825 * evaluated mount dir to be there (the vcache for the mounted
1826 * volume) rather than the vc of the mount point itself. We can
1827 * still find the mount point's vc in the vcache by its fid. */
1828 #endif /* UKERNEL && AFS_WEB_ENHANCEMENTS */
1829 if (!hit && force_eval) {
1830 osi_dnlc_enter(adp, aname, tvc, &versionNo);
1832 #ifdef AFS_LINUX20_ENV
1833 /* So Linux inode cache is up to date. */
1834 code = afs_VerifyVCache(tvc, &treq);
1836 afs_PutFakeStat(&fakestate);
1837 AFS_DISCON_UNLOCK();
1838 return 0; /* can't have been any errors if hit and !code */
1841 #if defined(UKERNEL) && defined(AFS_WEB_ENHANCEMENTS)
1848 code = afs_CheckCode(code, &treq, 19);
1850 /* If there is an error, make sure *avcp is null.
1851 * Alphas panic otherwise - defect 10719.
1856 afs_PutFakeStat(&fakestate);
1857 AFS_DISCON_UNLOCK();