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, afs_uint32 *avnoidp)
56 struct volume *tvp = 0;
59 char *cpos, *volnamep, *x;
61 afs_int32 prefetch; /* 1=>None 2=>RO 3=>BK */
62 afs_int32 mtptCell, assocCell = 0, hac = 0;
63 afs_int32 samecell, roname, len;
64 afs_uint32 volid, cellidx, vnoid = 0;
66 cpos = afs_strchr(data, ':'); /* if cell name present */
71 for (x = data; *x >= '0' && *x <= '9'; x++)
72 cellnum = (cellnum * 10) + (*x - '0');
74 tcell = afs_GetCell(cellnum, READ_LOCK);
76 tcell = afs_GetCellByName(data, READ_LOCK);
82 tcell = afs_GetCell(cellnum, READ_LOCK);
84 /*printf("No cellname %s , or cellnum %d , returning ENODEV\n",
89 /*printf("Lookup failed, returning ENODEV\n");*/
93 cellidx = tcell->cellIndex;
94 mtptCell = tcell->cellNum; /* The cell for the mountpoint */
96 hac = 1; /* has associated cell */
97 assocCell = tcell->lcellp->cellNum; /* The associated cell */
99 afs_PutCell(tcell, READ_LOCK);
101 cpos = afs_strrchr(volnamep, ':'); /* if vno present */
104 /* Look for an all-numeric volume ID */
106 for (x = volnamep; *x >= '0' && *x <= '9'; x++)
107 volid = (volid * 10) + (*x - '0');
111 if (!*x) /* allow vno with numeric volid only */
112 for (x = (cpos + 1); *x >= '0' && *x <= '9'; x++)
113 vnoid = (vnoid * 10) + (*x - '0');
119 * If the volume ID was all-numeric, and they didn't ask for a
120 * pointer to the volume structure, then just return the number
121 * as-is. This is currently only used for handling name lookups
122 * in the dynamic mount directory.
124 if (!*x && !avolpp) {
126 *acellidxp = cellidx;
135 * If the volume ID was all-numeric, and the type was '%', then
136 * assume whoever made the mount point knew what they were doing,
137 * and don't second-guess them by forcing use of a RW volume when
138 * they gave the ID of something else.
140 if (!*x && type == '%') {
141 tfid.Fid.Volume = volid; /* remember BK volume */
142 tfid.Cell = mtptCell;
143 tvp = afs_GetVolume(&tfid, areq, WRITE_LOCK); /* get the new one */
145 /*printf("afs_GetVolume failed - returning ENODEV");*/
146 return ENODEV; /* oops, can't do it */
151 /* Is volume name a "<n>.backup" or "<n>.readonly" name */
152 len = strlen(volnamep);
153 roname = ((len > 9) && (strcmp(&volnamep[len - 9], ".readonly") == 0))
154 || ((len > 7) && (strcmp(&volnamep[len - 7], ".backup") == 0));
156 /* When we cross mountpoint, do we stay in the same cell */
157 samecell = (cellnum == mtptCell) || (hac && (cellnum == assocCell));
159 /* Decide whether to prefetch the BK, or RO. Also means we want the BK or
161 * If this is a regular mountpoint with a RW volume name
162 * - If BK preference is enabled AND we remain within the same cell AND
163 * start from a BK volume, then we will want to prefetch the BK volume.
164 * - If we cross a cell boundary OR start from a RO volume, then we will
165 * want to prefetch the RO volume.
167 if ((type == '#') && !roname) {
168 if (afs_bkvolpref && samecell && (states & CBackup))
169 prefetch = 3; /* Prefetch the BK */
170 else if (!samecell || (states & CRO))
171 prefetch = 2; /* Prefetch the RO */
173 prefetch = 1; /* Do not prefetch */
175 prefetch = 1; /* Do not prefetch */
178 /* Get the volume struct. Unless this volume name has ".readonly" or
179 * ".backup" in it, this will get the volume struct for the RW volume.
180 * The RO volume will be prefetched if requested (but not returned).
182 /*printf("Calling GetVolumeByName\n");*/
183 tvp = afs_GetVolumeByName(volnamep, mtptCell, prefetch, areq, WRITE_LOCK);
185 /* If no volume was found in this cell, try the associated linked cell */
186 if (!tvp && hac && areq->volumeError) {
188 afs_GetVolumeByName(volnamep, assocCell, prefetch, areq,
192 /* Still not found. If we are looking for the RO, then perhaps the RW
193 * doesn't exist? Try adding ".readonly" to volname and look for that.
194 * Don't know why we do this. Would have still found it in above call - jpm.
196 if (!tvp && (prefetch == 2) && len < AFS_SMALLOCSIZ - 10) {
197 buf = (char *)osi_AllocSmallSpace(len + 10);
199 strcpy(buf, volnamep);
200 afs_strcat(buf, ".readonly");
202 tvp = afs_GetVolumeByName(buf, mtptCell, 1, areq, WRITE_LOCK);
204 /* Try the associated linked cell if failed */
205 if (!tvp && hac && areq->volumeError) {
206 tvp = afs_GetVolumeByName(buf, assocCell, 1, areq, WRITE_LOCK);
208 osi_FreeSmallSpace(buf);
212 /*printf("Couldn't find the volume\n");*/
213 return ENODEV; /* Couldn't find the volume */
216 /* Don't cross mountpoint from a BK to a BK volume */
217 if ((states & CBackup) && (tvp->states & VBackup)) {
218 afs_PutVolume(tvp, WRITE_LOCK);
222 /* If we want (prefetched) the BK and it exists, then drop the RW volume
224 * Otherwise, if we want (prefetched0 the RO and it exists, then drop the
225 * RW volume and get the RO.
226 * Otherwise, go with the RW.
228 if ((prefetch == 3) && tvp->backVol) {
229 tfid.Fid.Volume = tvp->backVol; /* remember BK volume */
230 tfid.Cell = tvp->cell;
231 afs_PutVolume(tvp, WRITE_LOCK); /* release old volume */
232 tvp = afs_GetVolume(&tfid, areq, WRITE_LOCK); /* get the new one */
234 return ENODEV; /* oops, can't do it */
235 } else if ((prefetch >= 2) && tvp->roVol) {
236 tfid.Fid.Volume = tvp->roVol; /* remember RO volume */
237 tfid.Cell = tvp->cell;
238 afs_PutVolume(tvp, WRITE_LOCK); /* release old volume */
239 tvp = afs_GetVolume(&tfid, areq, WRITE_LOCK); /* get the new one */
241 return ENODEV; /* oops, can't do it */
246 *acellidxp = cellidx;
248 *avolnump = tvp->volume;
254 afs_PutVolume(tvp, WRITE_LOCK);
259 EvalMountPoint(register struct vcache *avc, struct vcache *advc,
260 struct volume **avolpp, register struct vrequest *areq)
265 AFS_STATCNT(EvalMountPoint);
267 if (avc->mvid && (avc->f.states & CMValid))
268 return 0; /* done while racing */
271 code = afs_HandleLink(avc, areq);
275 /* Determine which cell and volume the mointpoint goes to */
276 code = EvalMountData(avc->linkData[0], avc->linkData + 1,
277 avc->f.states, avc->f.fid.Cell, avolpp, areq, 0, 0,
279 if (code) return code;
286 (struct VenusFid *)osi_AllocSmallSpace(sizeof(struct VenusFid));
287 avc->mvid->Cell = (*avolpp)->cell;
288 avc->mvid->Fid.Volume = (*avolpp)->volume;
289 avc->mvid->Fid.Vnode = avnoid;
290 avc->mvid->Fid.Unique = 1;
291 avc->f.states |= CMValid;
293 /* Used to: if the mount point is stored within a backup volume,
294 * then we should only update the parent pointer information if
295 * there's none already set, so as to avoid updating a volume's ..
296 * info with something in an OldFiles directory.
298 * Next two lines used to be under this if:
300 * if (!(avc->f.states & CBackup) || tvp->dotdot.Fid.Volume == 0)
302 * Now: update mount point back pointer on every call, so that we handle
303 * multiple mount points better. This way, when du tries to go back
304 * via chddir(".."), it will end up exactly where it started, yet
305 * cd'ing via a new path to a volume will reset the ".." pointer
308 (*avolpp)->mtpoint = avc->f.fid; /* setup back pointer to mtpoint */
311 (*avolpp)->dotdot = advc->f.fid;
319 * Must be called on an afs_fakestat_state object before calling
320 * afs_EvalFakeStat or afs_PutFakeStat. Calling afs_PutFakeStat
321 * without calling afs_EvalFakeStat is legal, as long as this
322 * function is called.
325 afs_InitFakeStat(struct afs_fakestat_state *state)
327 if (!afs_fakestat_enable)
332 state->need_release = 0;
336 * afs_EvalFakeStat_int
338 * The actual implementation of afs_EvalFakeStat and afs_TryEvalFakeStat,
339 * which is called by those wrapper functions.
341 * Only issues RPCs if canblock is non-zero.
344 afs_EvalFakeStat_int(struct vcache **avcp, struct afs_fakestat_state *state,
345 struct vrequest *areq, int canblock)
347 struct vcache *tvc, *root_vp;
348 struct volume *tvolp = NULL;
351 if (!afs_fakestat_enable)
354 osi_Assert(state->valid == 1);
355 osi_Assert(state->did_eval == 0);
359 if (tvc->mvstat != 1)
362 /* Is the call to VerifyVCache really necessary? */
363 code = afs_VerifyVCache(tvc, areq);
367 ObtainWriteLock(&tvc->lock, 599);
368 code = EvalMountPoint(tvc, NULL, &tvolp, areq);
369 ReleaseWriteLock(&tvc->lock);
373 tvolp->dotdot = tvc->f.fid;
374 tvolp->dotdot.Fid.Vnode = tvc->f.parent.vnode;
375 tvolp->dotdot.Fid.Unique = tvc->f.parent.unique;
378 if (tvc->mvid && (tvc->f.states & CMValid)) {
384 ObtainWriteLock(&afs_xvcache, 597);
385 root_vp = afs_FindVCache(tvc->mvid, &retry, IS_WLOCK);
386 if (root_vp && retry) {
387 ReleaseWriteLock(&afs_xvcache);
388 afs_PutVCache(root_vp);
390 } while (root_vp && retry);
391 ReleaseWriteLock(&afs_xvcache);
393 root_vp = afs_GetVCache(tvc->mvid, areq, NULL, NULL);
396 code = canblock ? ENOENT : 0;
399 #ifdef AFS_DARWIN80_ENV
400 root_vp->f.m.Type = VDIR;
402 code = afs_darwin_finalizevnode(root_vp, NULL, NULL, 0, 0);
405 vnode_ref(AFSTOV(root_vp));
407 if (tvolp && !afs_InReadDir(root_vp)) {
408 /* Is this always kosher? Perhaps we should instead use
409 * NBObtainWriteLock to avoid potential deadlock.
411 ObtainWriteLock(&root_vp->lock, 598);
413 root_vp->mvid = osi_AllocSmallSpace(sizeof(struct VenusFid));
414 *root_vp->mvid = tvolp->dotdot;
415 ReleaseWriteLock(&root_vp->lock);
417 state->need_release = 1;
418 state->root_vp = root_vp;
422 code = canblock ? ENOENT : 0;
427 afs_PutVolume(tvolp, WRITE_LOCK);
434 * Automatically does the equivalent of EvalMountPoint for vcache entries
435 * which are mount points. Remembers enough state to properly release
436 * the volume root vcache when afs_PutFakeStat() is called.
438 * State variable must be initialized by afs_InitFakeState() beforehand.
440 * Returns 0 when everything succeeds and *avcp points to the vcache entry
441 * that should be used for the real vnode operation. Returns non-zero if
442 * something goes wrong and the error code should be returned to the user.
445 afs_EvalFakeStat(struct vcache **avcp, struct afs_fakestat_state *state,
446 struct vrequest *areq)
448 return afs_EvalFakeStat_int(avcp, state, areq, 1);
452 * afs_TryEvalFakeStat
454 * Same as afs_EvalFakeStat, but tries not to talk to remote servers
455 * and only evaluate the mount point if all the data is already in
458 * Returns 0 if everything succeeds and *avcp points to a valid
459 * vcache entry (possibly evaluated).
462 afs_TryEvalFakeStat(struct vcache **avcp, struct afs_fakestat_state *state,
463 struct vrequest *areq)
465 return afs_EvalFakeStat_int(avcp, state, areq, 0);
471 * Perform any necessary cleanup at the end of a vnode op, given that
472 * afs_InitFakeStat was previously called with this state.
475 afs_PutFakeStat(struct afs_fakestat_state *state)
477 if (!afs_fakestat_enable)
480 osi_Assert(state->valid == 1);
481 if (state->need_release)
482 afs_PutVCache(state->root_vp);
487 afs_ENameOK(register char *aname)
491 AFS_STATCNT(ENameOK);
492 tlen = strlen(aname);
493 if (tlen >= 4 && strcmp(aname + tlen - 4, "@sys") == 0)
499 afs_getsysname(register struct vrequest *areq, register struct vcache *adp,
500 register char *bufp, int *num, char **sysnamelist[])
502 register struct unixuser *au;
503 register afs_int32 error;
505 AFS_STATCNT(getsysname);
507 *sysnamelist = afs_sysnamelist;
509 if (!afs_nfsexporter)
510 strcpy(bufp, (*sysnamelist)[0]);
512 au = afs_GetUser(areq->uid, adp->f.fid.Cell, 0);
514 error = EXP_SYSNAME(au->exporter, (char *)0, sysnamelist, num, 0);
516 strcpy(bufp, "@sys");
520 strcpy(bufp, (*sysnamelist)[0]);
523 strcpy(bufp, afs_sysname);
530 Check_AtSys(register struct vcache *avc, const char *aname,
531 struct sysname_info *state, struct vrequest *areq)
534 char **sysnamelist[MAXNUMSYSNAMES];
536 if (AFS_EQ_ATSYS(aname)) {
538 state->name = (char *)osi_AllocLargeSpace(MAXSYSNAME);
541 afs_getsysname(areq, avc, state->name, &num, sysnamelist);
546 state->name = (char *)aname;
551 Next_AtSys(register struct vcache *avc, struct vrequest *areq,
552 struct sysname_info *state)
554 int num = afs_sysnamecount;
555 char **sysnamelist[MAXNUMSYSNAMES];
557 if (state->index == -1)
558 return 0; /* No list */
560 /* Check for the initial state of aname != "@sys" in Check_AtSys */
561 if (state->offset == -1 && state->allocked == 0) {
562 register char *tname;
564 /* Check for .*@sys */
565 for (tname = state->name; *tname; tname++)
566 /*Move to the end of the string */ ;
568 if ((tname > state->name + 4) && (AFS_EQ_ATSYS(tname - 4))) {
569 state->offset = (tname - 4) - state->name;
570 tname = (char *)osi_AllocLargeSpace(AFS_LRALLOCSIZ);
571 strncpy(tname, state->name, state->offset);
576 afs_getsysname(areq, avc, state->name + state->offset, &num,
580 return 0; /* .*@sys doesn't match either */
582 register struct unixuser *au;
583 register afs_int32 error;
585 *sysnamelist = afs_sysnamelist;
587 if (afs_nfsexporter) {
588 au = afs_GetUser(areq->uid, avc->f.fid.Cell, 0);
591 EXP_SYSNAME(au->exporter, (char *)0, sysnamelist, &num, 0);
599 if (++(state->index) >= num || !(*sysnamelist)[(unsigned int)state->index])
600 return 0; /* end of list */
602 strcpy(state->name + state->offset, (*sysnamelist)[(unsigned int)state->index]);
606 extern int BlobScan(struct dcache * afile, afs_int32 ablob);
608 /* called with an unlocked directory and directory cookie. Areqp
609 * describes who is making the call.
610 * Scans the next N (about 30, typically) directory entries, and does
611 * a bulk stat call to stat them all.
613 * Must be very careful when merging in RPC responses, since we dont
614 * want to overwrite newer info that was added by a file system mutating
615 * call that ran concurrently with our bulk stat call.
617 * We do that, as described below, by not merging in our info (always
618 * safe to skip the merge) if the status info is valid in the vcache entry.
620 * If adapt ever implements the bulk stat RPC, then this code will need to
621 * ensure that vcaches created for failed RPC's to older servers have the
624 static struct vcache *BStvc = NULL;
627 afs_DoBulkStat(struct vcache *adp, long dirCookie, struct vrequest *areqp)
629 int nentries; /* # of entries to prefetch */
630 int nskip; /* # of slots in the LRU queue to skip */
631 #ifdef AFS_DARWIN80_ENV
634 struct vcache *lruvcp; /* vcache ptr of our goal pos in LRU queue */
635 struct dcache *dcp; /* chunk containing the dir block */
636 afs_size_t temp; /* temp for holding chunk length, &c. */
637 struct AFSFid *fidsp; /* file IDs were collecting */
638 struct AFSCallBack *cbsp; /* call back pointers */
639 struct AFSCallBack *tcbp; /* temp callback ptr */
640 struct AFSFetchStatus *statsp; /* file status info */
641 struct AFSVolSync volSync; /* vol sync return info */
642 struct vcache *tvcp; /* temp vcp */
643 struct afs_q *tq; /* temp queue variable */
644 AFSCBFids fidParm; /* file ID parm for bulk stat */
645 AFSBulkStats statParm; /* stat info parm for bulk stat */
646 int fidIndex = 0; /* which file were stating */
647 struct afs_conn *tcp = 0; /* conn for call */
648 AFSCBs cbParm; /* callback parm for bulk stat */
649 struct server *hostp = 0; /* host we got callback from */
650 long startTime; /* time we started the call,
651 * for callback expiration base
653 afs_size_t statSeqNo = 0; /* Valued of file size to detect races */
654 int code; /* error code */
655 long newIndex; /* new index in the dir */
656 struct DirEntry *dirEntryp; /* dir entry we are examining */
658 struct VenusFid afid; /* file ID we are using now */
659 struct VenusFid tfid; /* another temp. file ID */
660 afs_int32 retry; /* handle low-level SGI MP race conditions */
661 long volStates; /* flags from vol structure */
662 struct volume *volp = 0; /* volume ptr */
663 struct VenusFid dotdot = {0, {0, 0, 0}};
664 int flagIndex = 0; /* First file with bulk fetch flag set */
665 int inlinebulk = 0; /* Did we use InlineBulk RPC or not? */
668 dotdot.Fid.Unique = 0;
669 dotdot.Fid.Vnode = 0;
671 /* first compute some basic parameters. We dont want to prefetch more
672 * than a fraction of the cache in any given call, and we want to preserve
673 * a portion of the LRU queue in any event, so as to avoid thrashing
674 * the entire stat cache (we will at least leave some of it alone).
675 * presently dont stat more than 1/8 the cache in any one call. */
676 nentries = afs_cacheStats / 8;
678 /* dont bother prefetching more than one calls worth of info */
679 if (nentries > AFSCBMAX)
682 /* heuristic to make sure that things fit in 4K. This means that
683 * we shouldnt make it any bigger than 47 entries. I am typically
684 * going to keep it a little lower, since we don't want to load
685 * too much of the stat cache.
690 /* now, to reduce the stack size, well allocate two 4K blocks,
691 * one for fids and callbacks, and one for stat info. Well set
692 * up our pointers to the memory from there, too.
694 statsp = (AFSFetchStatus *)
695 osi_Alloc(AFSCBMAX * sizeof(AFSFetchStatus));
696 fidsp = (AFSFid *) osi_AllocLargeSpace(nentries * sizeof(AFSFid));
697 cbsp = (AFSCallBack *)
698 osi_Alloc(AFSCBMAX * sizeof(AFSCallBack));
700 /* next, we must iterate over the directory, starting from the specified
701 * cookie offset (dirCookie), and counting out nentries file entries.
702 * We skip files that already have stat cache entries, since we
703 * dont want to bulk stat files that are already in the cache.
706 code = afs_VerifyVCache(adp, areqp);
710 dcp = afs_GetDCache(adp, (afs_size_t) 0, areqp, &temp, &temp, 1);
716 /* lock the directory cache entry */
717 ObtainReadLock(&adp->lock);
718 ObtainReadLock(&dcp->lock);
721 * Make sure that the data in the cache is current. There are two
722 * cases we need to worry about:
723 * 1. The cache data is being fetched by another process.
724 * 2. The cache data is no longer valid
726 while ((adp->f.states & CStatd)
727 && (dcp->dflags & DFFetching)
728 && hsame(adp->f.m.DataVersion, dcp->f.versionNo)) {
729 afs_Trace4(afs_iclSetp, CM_TRACE_DCACHEWAIT, ICL_TYPE_STRING,
730 __FILE__, ICL_TYPE_INT32, __LINE__, ICL_TYPE_POINTER, dcp,
731 ICL_TYPE_INT32, dcp->dflags);
732 ReleaseReadLock(&dcp->lock);
733 ReleaseReadLock(&adp->lock);
734 afs_osi_Sleep(&dcp->validPos);
735 ObtainReadLock(&adp->lock);
736 ObtainReadLock(&dcp->lock);
738 if (!(adp->f.states & CStatd)
739 || !hsame(adp->f.m.DataVersion, dcp->f.versionNo)) {
740 ReleaseReadLock(&dcp->lock);
741 ReleaseReadLock(&adp->lock);
746 /* Generate a sequence number so we can tell whether we should
747 * store the attributes when processing the response. This number is
748 * stored in the file size when we set the CBulkFetching bit. If the
749 * CBulkFetching is still set and this value hasn't changed, then
750 * we know we were the last to set CBulkFetching bit for this file,
751 * and it is safe to set the status information for this file.
753 statSeqNo = bulkStatCounter++;
754 /* ensure against wrapping */
756 statSeqNo = bulkStatCounter++;
758 /* now we have dir data in the cache, so scan the dir page */
761 while (1) { /* Should probably have some constant bound */
762 /* look for first safe entry to examine in the directory. BlobScan
763 * looks for a the 1st allocated dir after the dirCookie slot.
765 newIndex = BlobScan(dcp, (dirCookie >> 5));
769 /* remember the updated directory cookie */
770 dirCookie = newIndex << 5;
772 /* get a ptr to the dir entry */
774 (struct DirEntry *)afs_dir_GetBlob(dcp, newIndex);
778 /* dont copy more than we have room for */
779 if (fidIndex >= nentries) {
780 DRelease(dirEntryp, 0);
784 /* now, if the dir entry looks good, copy it out to our list. Vnode
785 * 0 means deleted, although it should also be free were it deleted.
787 if (dirEntryp->fid.vnode != 0) {
788 /* dont copy entries we have in our cache. This check will
789 * also make us skip "." and probably "..", unless it has
790 * disappeared from the cache since we did our namei call.
792 tfid.Cell = adp->f.fid.Cell;
793 tfid.Fid.Volume = adp->f.fid.Fid.Volume;
794 tfid.Fid.Vnode = ntohl(dirEntryp->fid.vnode);
795 tfid.Fid.Unique = ntohl(dirEntryp->fid.vunique);
798 ObtainWriteLock(&afs_xvcache, 130);
799 tvcp = afs_FindVCache(&tfid, &retry, IS_WLOCK /* no stats | LRU */ );
801 ReleaseWriteLock(&afs_xvcache);
804 } while (tvcp && retry);
805 if (!tvcp) { /* otherwise, create manually */
806 tvcp = afs_NewBulkVCache(&tfid, hostp, statSeqNo);
809 ObtainWriteLock(&tvcp->lock, 505);
810 ReleaseWriteLock(&afs_xvcache);
811 afs_RemoveVCB(&tfid);
812 ReleaseWriteLock(&tvcp->lock);
814 ReleaseWriteLock(&afs_xvcache);
817 ReleaseWriteLock(&afs_xvcache);
821 DRelease(dirEntryp, 0);
822 ReleaseReadLock(&dcp->lock);
823 ReleaseReadLock(&adp->lock);
825 goto done; /* can happen if afs_NewVCache fails */
828 /* WARNING: afs_DoBulkStat uses the Length field to store a
829 * sequence number for each bulk status request. Under no
830 * circumstances should afs_DoBulkStat store a sequence number
831 * if the new length will be ignored when afs_ProcessFS is
832 * called with new stats. */
834 if (!(tvcp->f.states & CStatd)
835 && (!((tvcp->f.states & CBulkFetching) &&
836 (tvcp->f.m.Length != statSeqNo)))
837 && (tvcp->execsOrWriters <= 0)
838 && !afs_DirtyPages(tvcp)
839 && !AFS_VN_MAPPED((vnode_t *) tvcp))
841 if (!(tvcp->f.states & CStatd)
842 && (!((tvcp->f.states & CBulkFetching) &&
843 (tvcp->f.m.Length != statSeqNo)))
844 && (tvcp->execsOrWriters <= 0)
845 && !afs_DirtyPages(tvcp))
849 /* this entry doesnt exist in the cache, and is not
850 * already being fetched by someone else, so add it to the
851 * list of file IDs to obtain.
853 * We detect a callback breaking race condition by checking the
854 * CBulkFetching state bit and the value in the file size.
855 * It is safe to set the status only if the CBulkFetching
856 * flag is still set and the value in the file size does
857 * not change. NewBulkVCache sets us up.
859 * Don't fetch status for dirty files. We need to
860 * preserve the value of the file size. We could
861 * flush the pages, but it wouldn't be worthwhile.
863 memcpy((char *)(fidsp + fidIndex), (char *)&tfid.Fid,
870 /* if dir vnode has non-zero entry */
871 /* move to the next dir entry by adding in the # of entries
872 * used by this dir entry.
874 temp = afs_dir_NameBlobs(dirEntryp->name) << 5;
875 DRelease(dirEntryp, 0);
879 } /* while loop over all dir entries */
881 /* now release the dir lock and prepare to make the bulk RPC */
882 ReleaseReadLock(&dcp->lock);
883 ReleaseReadLock(&adp->lock);
885 /* release the chunk */
888 /* dont make a null call */
893 /* setup the RPC parm structures */
894 fidParm.AFSCBFids_len = fidIndex;
895 fidParm.AFSCBFids_val = fidsp;
896 statParm.AFSBulkStats_len = fidIndex;
897 statParm.AFSBulkStats_val = statsp;
898 cbParm.AFSCBs_len = fidIndex;
899 cbParm.AFSCBs_val = cbsp;
901 /* start the timer; callback expirations are relative to this */
902 startTime = osi_Time();
904 tcp = afs_Conn(&adp->f.fid, areqp, SHARED_LOCK);
906 hostp = tcp->srvr->server;
907 XSTATS_START_TIME(AFS_STATS_FS_RPCIDX_BULKSTATUS);
910 if (!(tcp->srvr->server->flags & SNO_INLINEBULK)) {
912 RXAFS_InlineBulkStatus(tcp->id, &fidParm, &statParm,
914 if (code == RXGEN_OPCODE) {
915 tcp->srvr->server->flags |= SNO_INLINEBULK;
918 RXAFS_BulkStatus(tcp->id, &fidParm, &statParm,
925 RXAFS_BulkStatus(tcp->id, &fidParm, &statParm, &cbParm,
933 (tcp, code, &adp->f.fid, areqp, AFS_STATS_FS_RPCIDX_BULKSTATUS,
936 /* now, if we didnt get the info, bail out. */
940 /* we need vol flags to create the entries properly */
941 dotdot.Fid.Volume = 0;
942 volp = afs_GetVolume(&adp->f.fid, areqp, READ_LOCK);
944 volStates = volp->states;
945 if (volp->dotdot.Fid.Volume != 0)
946 dotdot = volp->dotdot;
950 /* find the place to merge the info into We do this by skipping
951 * nskip entries in the LRU queue. The more we skip, the more
952 * we preserve, since the head of the VLRU queue is the most recently
956 nskip = afs_cacheStats / 2; /* preserved fraction of the cache */
957 ObtainReadLock(&afs_xvcache);
959 /* actually a serious error, probably should panic. Probably will
960 * panic soon, oh well. */
961 ReleaseReadLock(&afs_xvcache);
962 afs_warnuser("afs_DoBulkStat: VLRU empty!");
965 if ((VLRU.next->prev != &VLRU) || (VLRU.prev->next != &VLRU)) {
966 refpanic("Bulkstat VLRU inconsistent");
968 for (tq = VLRU.next; tq != &VLRU; tq = QNext(tq)) {
970 #ifdef AFS_DARWIN80_ENV
971 if (!(QTOV(tq)->f.states & CDeadVnode))
975 if (QNext(QPrev(tq)) != tq) {
977 refpanic("BulkStat VLRU inconsistent");
983 lruvcp = QTOV(VLRU.next);
985 /* now we have to hold this entry, so that it does not get moved
986 * into the free list while we're running. It could still get
987 * moved within the lru queue, but hopefully that will be rare; it
988 * doesn't hurt nearly as much.
991 #ifdef AFS_DARWIN80_ENV
992 lruvp = AFSTOV(lruvcp);
993 if (vnode_get(lruvp)) /* this bumps ref count */
995 else if (vnode_ref(lruvp)) {
997 /* AFSTOV(lruvcp) may be NULL */
1003 osi_vnhold(lruvcp, &retry);
1005 ReleaseReadLock(&afs_xvcache); /* could be read lock */
1009 /* otherwise, merge in the info. We have to be quite careful here,
1010 * since we need to ensure that we don't merge old info over newer
1011 * stuff in a stat cache entry. We're very conservative here: we don't
1012 * do the merge at all unless we ourselves create the stat cache
1013 * entry. That's pretty safe, and should work pretty well, since we
1014 * typically expect to do the stat cache creation ourselves.
1016 * We also have to take into account racing token revocations.
1018 for (i = 0; i < fidIndex; i++) {
1019 if ((&statsp[i])->errorCode)
1021 afid.Cell = adp->f.fid.Cell;
1022 afid.Fid.Volume = adp->f.fid.Fid.Volume;
1023 afid.Fid.Vnode = fidsp[i].Vnode;
1024 afid.Fid.Unique = fidsp[i].Unique;
1027 ObtainReadLock(&afs_xvcache);
1028 tvcp = afs_FindVCache(&afid, &retry, 0 /* !stats&!lru */ );
1029 ReleaseReadLock(&afs_xvcache);
1030 } while (tvcp && retry);
1032 /* The entry may no longer exist */
1037 /* now we have the entry held, but we need to fill it in */
1038 ObtainWriteLock(&tvcp->lock, 131);
1040 /* if CBulkFetching is not set, or if the file size no longer
1041 * matches the value we placed there when we set the CBulkFetching
1042 * flag, then someone else has done something with this node,
1043 * and we may not have the latest status information for this
1044 * file. Leave the entry alone.
1046 if (!(tvcp->f.states & CBulkFetching) || (tvcp->f.m.Length != statSeqNo)) {
1048 ReleaseWriteLock(&tvcp->lock);
1049 afs_PutVCache(tvcp);
1053 /* now copy ".." entry back out of volume structure, if necessary */
1054 if (tvcp->mvstat == 2 && (dotdot.Fid.Volume != 0)) {
1056 tvcp->mvid = (struct VenusFid *)
1057 osi_AllocSmallSpace(sizeof(struct VenusFid));
1058 *tvcp->mvid = dotdot;
1061 ObtainWriteLock(&afs_xvcache, 132);
1062 if ((VLRU.next->prev != &VLRU) || (VLRU.prev->next != &VLRU)) {
1063 refpanic("Bulkstat VLRU inconsistent2");
1065 if ((QNext(QPrev(&tvcp->vlruq)) != &tvcp->vlruq)
1066 || (QPrev(QNext(&tvcp->vlruq)) != &tvcp->vlruq)) {
1067 refpanic("Bulkstat VLRU inconsistent4");
1069 if ((QNext(QPrev(&lruvcp->vlruq)) != &lruvcp->vlruq)
1070 || (QPrev(QNext(&lruvcp->vlruq)) != &lruvcp->vlruq)) {
1071 refpanic("Bulkstat VLRU inconsistent5");
1074 if (tvcp != lruvcp) { /* if they are == don't move it, don't corrupt vlru */
1075 QRemove(&tvcp->vlruq);
1076 QAdd(&lruvcp->vlruq, &tvcp->vlruq);
1079 if ((VLRU.next->prev != &VLRU) || (VLRU.prev->next != &VLRU)) {
1080 refpanic("Bulkstat VLRU inconsistent3");
1082 if ((QNext(QPrev(&tvcp->vlruq)) != &tvcp->vlruq)
1083 || (QPrev(QNext(&tvcp->vlruq)) != &tvcp->vlruq)) {
1084 refpanic("Bulkstat VLRU inconsistent5");
1086 if ((QNext(QPrev(&lruvcp->vlruq)) != &lruvcp->vlruq)
1087 || (QPrev(QNext(&lruvcp->vlruq)) != &lruvcp->vlruq)) {
1088 refpanic("Bulkstat VLRU inconsistent6");
1090 ReleaseWriteLock(&afs_xvcache);
1092 ObtainWriteLock(&afs_xcbhash, 494);
1094 /* We need to check the flags again. We may have missed
1095 * something while we were waiting for a lock.
1097 if (!(tvcp->f.states & CBulkFetching) || (tvcp->f.m.Length != statSeqNo)) {
1099 ReleaseWriteLock(&tvcp->lock);
1100 ReleaseWriteLock(&afs_xcbhash);
1101 afs_PutVCache(tvcp);
1105 /* now merge in the resulting status back into the vnode.
1106 * We only do this if the entry looks clear.
1108 afs_ProcessFS(tvcp, &statsp[i], areqp);
1109 #if defined(AFS_LINUX22_ENV)
1110 afs_fill_inode(AFSTOV(tvcp), NULL); /* reset inode operations */
1113 /* do some accounting for bulk stats: mark this entry as
1114 * loaded, so we can tell if we use it before it gets
1117 tvcp->f.states |= CBulkStat;
1118 tvcp->f.states &= ~CBulkFetching;
1120 afs_bulkStatsDone++;
1122 /* merge in vol info */
1123 if (volStates & VRO)
1124 tvcp->f.states |= CRO;
1125 if (volStates & VBackup)
1126 tvcp->f.states |= CBackup;
1127 if (volStates & VForeign)
1128 tvcp->f.states |= CForeign;
1130 /* merge in the callback info */
1131 tvcp->f.states |= CTruth;
1133 /* get ptr to the callback we are interested in */
1136 if (tcbp->ExpirationTime != 0) {
1137 tvcp->cbExpires = tcbp->ExpirationTime + startTime;
1138 tvcp->callback = hostp;
1139 tvcp->f.states |= CStatd;
1140 afs_QueueCallback(tvcp, CBHash(tcbp->ExpirationTime), volp);
1141 } else if (tvcp->f.states & CRO) {
1142 /* ordinary callback on a read-only volume -- AFS 3.2 style */
1143 tvcp->cbExpires = 3600 + startTime;
1144 tvcp->callback = hostp;
1145 tvcp->f.states |= CStatd;
1146 afs_QueueCallback(tvcp, CBHash(3600), volp);
1149 tvcp->f.states &= ~(CStatd | CUnique);
1150 afs_DequeueCallback(tvcp);
1151 if ((tvcp->f.states & CForeign) || (vType(tvcp) == VDIR))
1152 osi_dnlc_purgedp(tvcp); /* if it (could be) a directory */
1154 ReleaseWriteLock(&afs_xcbhash);
1155 #ifdef AFS_DARWIN80_ENV
1156 /* reclaim->FlushVCache will need xcbhash */
1157 if (tvcp->f.states & CDeadVnode) {
1158 /* passing in a parent hangs getting the vnode lock */
1159 code = afs_darwin_finalizevnode(tvcp, NULL, NULL, 0, 1);
1161 /* It's gonna get recycled - shouldn't happen */
1163 tvcp->f.states &= ~(CStatd | CUnique);
1164 afs_DequeueCallback(tvcp);
1165 if ((tvcp->f.states & CForeign) || (vType(tvcp) == VDIR))
1166 osi_dnlc_purgedp(tvcp); /* if it (could be) a directory */
1168 /* re-acquire the usecount that finalizevnode disposed of */
1169 vnode_ref(AFSTOV(tvcp));
1173 ReleaseWriteLock(&tvcp->lock);
1174 /* finally, we're done with the entry */
1175 afs_PutVCache(tvcp);
1176 } /* for all files we got back */
1178 /* finally return the pointer into the LRU queue */
1179 #ifdef AFS_DARWIN80_ENV
1185 afs_PutVCache(lruvcp);
1189 /* Be sure to turn off the CBulkFetching flags */
1190 for (i = flagIndex; i < fidIndex; i++) {
1191 afid.Cell = adp->f.fid.Cell;
1192 afid.Fid.Volume = adp->f.fid.Fid.Volume;
1193 afid.Fid.Vnode = fidsp[i].Vnode;
1194 afid.Fid.Unique = fidsp[i].Unique;
1197 ObtainReadLock(&afs_xvcache);
1198 tvcp = afs_FindVCache(&afid, &retry, 0 /* !stats&!lru */ );
1199 ReleaseReadLock(&afs_xvcache);
1200 } while (tvcp && retry);
1201 if (tvcp != NULL && (tvcp->f.states & CBulkFetching)
1202 && (tvcp->f.m.Length == statSeqNo)) {
1203 tvcp->f.states &= ~CBulkFetching;
1206 afs_PutVCache(tvcp);
1210 afs_PutVolume(volp, READ_LOCK);
1212 /* If we did the InlineBulk RPC pull out the return code */
1213 if (inlinebulk && code == 0) {
1214 if ((&statsp[0])->errorCode) {
1215 afs_Analyze(tcp, (&statsp[0])->errorCode, &adp->f.fid, areqp,
1216 AFS_STATS_FS_RPCIDX_BULKSTATUS, SHARED_LOCK, NULL);
1217 code = (&statsp[0])->errorCode;
1223 osi_FreeLargeSpace((char *)fidsp);
1224 osi_Free((char *)statsp, AFSCBMAX * sizeof(AFSFetchStatus));
1225 osi_Free((char *)cbsp, AFSCBMAX * sizeof(AFSCallBack));
1229 /* was: (AFS_DEC_ENV) || defined(AFS_OSF30_ENV) || defined(AFS_NCR_ENV) */
1230 static int AFSDOBULK = 1;
1233 #if defined(AFS_SUN5_ENV) || defined(AFS_SGI_ENV)
1234 afs_lookup(OSI_VC_DECL(adp), char *aname, struct vcache **avcp, struct pathname *pnp, int flags, struct vnode *rdir, afs_ucred_t *acred)
1235 #elif defined(UKERNEL)
1236 afs_lookup(OSI_VC_DECL(adp), char *aname, struct vcache **avcp, afs_ucred_t *acred, int flags)
1238 afs_lookup(OSI_VC_DECL(adp), char *aname, struct vcache **avcp, afs_ucred_t *acred)
1241 struct vrequest treq;
1243 register struct vcache *tvc = 0;
1244 register afs_int32 code;
1245 register afs_int32 bulkcode = 0;
1246 int pass = 0, hit = 0;
1247 int force_eval = afs_fakestat_enable ? 0 : 1;
1249 extern afs_int32 afs_mariner; /*Writing activity to log? */
1250 afs_hyper_t versionNo;
1251 int no_read_access = 0;
1252 struct sysname_info sysState; /* used only for @sys checking */
1253 int dynrootRetry = 1;
1254 struct afs_fakestat_state fakestate;
1255 int tryEvalOnly = 0;
1256 OSI_VC_CONVERT(adp);
1258 AFS_STATCNT(afs_lookup);
1259 afs_InitFakeStat(&fakestate);
1263 if ((code = afs_InitReq(&treq, acred)))
1266 if (afs_fakestat_enable && adp->mvstat == 1) {
1267 if (strcmp(aname, ".directory") == 0)
1271 #if defined(AFS_DARWIN_ENV)
1272 /* Workaround for MacOSX Finder, which tries to look for
1273 * .DS_Store and Contents under every directory.
1275 if (afs_fakestat_enable && adp->mvstat == 1) {
1276 if (strcmp(aname, ".DS_Store") == 0)
1278 if (strcmp(aname, "Contents") == 0)
1281 if (afs_fakestat_enable && adp->mvstat == 2) {
1282 if (strncmp(aname, "._", 2) == 0)
1288 code = afs_TryEvalFakeStat(&adp, &fakestate, &treq);
1290 code = afs_EvalFakeStat(&adp, &fakestate, &treq);
1292 /*printf("Code is %d\n", code);*/
1294 if (tryEvalOnly && adp->mvstat == 1)
1299 *avcp = NULL; /* Since some callers don't initialize it */
1301 /* come back to here if we encounter a non-existent object in a read-only
1302 * volume's directory */
1305 *avcp = NULL; /* Since some callers don't initialize it */
1308 if (!(adp->f.states & CStatd) && !afs_InReadDir(adp)) {
1309 if ((code = afs_VerifyVCache2(adp, &treq))) {
1315 /* watch for ".." in a volume root */
1316 if (adp->mvstat == 2 && aname[0] == '.' && aname[1] == '.' && !aname[2]) {
1317 /* looking up ".." in root via special hacks */
1318 if (adp->mvid == (struct VenusFid *)0 || adp->mvid->Fid.Volume == 0) {
1322 /* otherwise we have the fid here, so we use it */
1323 /*printf("Getting vcache\n");*/
1324 tvc = afs_GetVCache(adp->mvid, &treq, NULL, NULL);
1325 afs_Trace3(afs_iclSetp, CM_TRACE_GETVCDOTDOT, ICL_TYPE_FID, adp->mvid,
1326 ICL_TYPE_POINTER, tvc, ICL_TYPE_INT32, code);
1328 code = (tvc ? 0 : ENOENT);
1330 if (tvc && !VREFCOUNT_GT(tvc, 0)) {
1334 /*printf("LOOKUP GETVCDOTDOT -> %d\n", code); */
1339 /* now check the access */
1340 if (treq.uid != adp->last_looker) {
1341 if (!afs_AccessOK(adp, PRSFS_LOOKUP, &treq, CHECK_MODE_BITS)) {
1346 adp->last_looker = treq.uid;
1349 /* Check for read access as well. We need read access in order to
1350 * stat files, but not to stat subdirectories. */
1351 if (!afs_AccessOK(adp, PRSFS_READ, &treq, CHECK_MODE_BITS))
1354 /* special case lookup of ".". Can we check for it sooner in this code,
1355 * for instance, way up before "redo:" ??
1356 * I'm not fiddling with the LRUQ here, either, perhaps I should, or else
1357 * invent a lightweight version of GetVCache.
1359 if (aname[0] == '.' && !aname[1]) { /* special case */
1360 ObtainReadLock(&afs_xvcache);
1362 ReleaseReadLock(&afs_xvcache);
1363 #ifdef AFS_DARWIN80_ENV
1364 vnode_get(AFSTOV(adp));
1369 if (adp && !VREFCOUNT_GT(adp, 0)) {
1376 * Special case lookup of ".." in the dynamic mount directory.
1377 * The parent of this directory is _always_ the AFS root volume.
1379 if (afs_IsDynrootMount(adp) &&
1380 aname[0] == '.' && aname[1] == '.' && !aname[2]) {
1382 ObtainReadLock(&afs_xvcache);
1383 osi_vnhold(afs_globalVp, 0);
1384 ReleaseReadLock(&afs_xvcache);
1385 #ifdef AFS_DARWIN80_ENV
1386 vnode_get(AFSTOV(afs_globalVp));
1389 *avcp = tvc = afs_globalVp;
1395 * Special case lookups in the dynamic mount directory.
1396 * The names here take the form cell:volume, similar to a mount point.
1397 * EvalMountData parses that and returns a cell and volume ID, which
1398 * we use to construct the appropriate dynroot Fid.
1400 if (afs_IsDynrootMount(adp)) {
1401 struct VenusFid tfid;
1402 afs_uint32 cellidx, volid, vnoid;
1404 code = EvalMountData('%', aname, 0, 0, NULL, &treq, &cellidx, &volid, &vnoid);
1407 afs_GetDynrootMountFid(&tfid);
1408 tfid.Fid.Vnode = VNUM_FROM_TYPEID(VN_TYPE_MOUNT, cellidx << 2);
1409 tfid.Fid.Unique = volid;
1410 *avcp = tvc = afs_GetVCache(&tfid, &treq, NULL, NULL);
1415 #ifdef AFS_LINUX26_ENV
1417 * Special case of the dynamic mount volume in a static root.
1418 * This is really unfortunate, but we need this for the translator.
1420 if (adp == afs_globalVp && !afs_GetDynrootEnable() &&
1421 !strcmp(aname, AFS_DYNROOT_MOUNTNAME)) {
1422 struct VenusFid tfid;
1424 afs_GetDynrootMountFid(&tfid);
1425 *avcp = tvc = afs_GetVCache(&tfid, &treq, NULL, NULL);
1432 Check_AtSys(adp, aname, &sysState, &treq);
1433 tname = sysState.name;
1435 /* 1st Check_AtSys and lookup by tname is required here, for now,
1436 * because the dnlc is *not* told to remove entries for the parent
1437 * dir of file/dir op that afs_LocalHero likes, but dnlc is informed
1438 * if the cached entry for the parent dir is invalidated for a
1440 * Otherwise, we'd be able to do a dnlc lookup on an entry ending
1441 * w/@sys and know the dnlc was consistent with reality. */
1442 tvc = osi_dnlc_lookup(adp, tname, WRITE_LOCK);
1443 *avcp = tvc; /* maybe wasn't initialized, but it is now */
1445 if (no_read_access && vType(tvc) != VDIR && vType(tvc) != VLNK) {
1446 /* need read access on dir to stat non-directory / non-link */
1447 #ifndef AFS_FBSD80_ENV
1454 #ifdef AFS_LINUX22_ENV
1455 if (tvc->mvstat == 2) { /* we don't trust the dnlc for root vcaches */
1456 AFS_RELE(AFSTOV(tvc));
1463 #else /* non - LINUX */
1467 #endif /* linux22 */
1470 { /* sub-block just to reduce stack usage */
1471 register struct dcache *tdc;
1472 afs_size_t dirOffset, dirLen;
1473 struct VenusFid tfid;
1475 /* now we have to lookup the next fid */
1476 if (afs_InReadDir(adp))
1477 tdc = adp->dcreaddir;
1479 tdc = afs_GetDCache(adp, (afs_size_t) 0, &treq,
1480 &dirOffset, &dirLen, 1);
1482 *avcp = NULL; /* redundant, but harmless */
1487 /* now we will just call dir package with appropriate inode.
1488 * Dirs are always fetched in their entirety for now */
1489 ObtainReadLock(&adp->lock);
1490 ObtainReadLock(&tdc->lock);
1493 * Make sure that the data in the cache is current. There are two
1494 * cases we need to worry about:
1495 * 1. The cache data is being fetched by another process.
1496 * 2. The cache data is no longer valid
1498 * If a readdir is in progress _in this thread_, it has a shared
1499 * lock on the vcache and has obtained current data, so we just
1500 * use that. This eliminates several possible deadlocks.
1502 if (!afs_InReadDir(adp)) {
1503 while ((adp->f.states & CStatd)
1504 && (tdc->dflags & DFFetching)
1505 && hsame(adp->f.m.DataVersion, tdc->f.versionNo)) {
1506 ReleaseReadLock(&tdc->lock);
1507 ReleaseReadLock(&adp->lock);
1508 afs_osi_Sleep(&tdc->validPos);
1509 ObtainReadLock(&adp->lock);
1510 ObtainReadLock(&tdc->lock);
1512 if (!(adp->f.states & CStatd)
1513 || !hsame(adp->f.m.DataVersion, tdc->f.versionNo)) {
1514 ReleaseReadLock(&tdc->lock);
1515 ReleaseReadLock(&adp->lock);
1517 if (tname && tname != aname)
1518 osi_FreeLargeSpace(tname);
1523 /* Save the version number for when we call osi_dnlc_enter */
1524 hset(versionNo, tdc->f.versionNo);
1527 * check for, and handle "@sys" if it's there. We should be able
1528 * to avoid the alloc and the strcpy with a little work, but it's
1529 * not pressing. If there aren't any remote users (ie, via the
1530 * NFS translator), we have a slightly easier job.
1531 * the faster way to do this is to check for *aname == '@' and if
1532 * it's there, check for @sys, otherwise, assume there's no @sys
1533 * then, if the lookup fails, check for .*@sys...
1535 /* above now implemented by Check_AtSys and Next_AtSys */
1537 /* lookup the name in the appropriate dir, and return a cache entry
1538 * on the resulting fid */
1540 afs_dir_LookupOffset(tdc, sysState.name, &tfid.Fid,
1543 /* If the first lookup doesn't succeed, maybe it's got @sys in the name */
1544 while (code == ENOENT && Next_AtSys(adp, &treq, &sysState))
1546 afs_dir_LookupOffset(tdc, sysState.name, &tfid.Fid,
1548 tname = sysState.name;
1550 ReleaseReadLock(&tdc->lock);
1551 if (!afs_InReadDir(adp))
1553 if (code == ENOENT && afs_IsDynroot(adp) && dynrootRetry && !tryEvalOnly) {
1554 ReleaseReadLock(&adp->lock);
1556 if (tname[0] == '.')
1557 afs_LookupAFSDB(tname + 1);
1559 afs_LookupAFSDB(tname);
1560 if (tname && tname != aname)
1561 osi_FreeLargeSpace(tname);
1564 ReleaseReadLock(&adp->lock);
1567 /* new fid has same cell and volume */
1568 tfid.Cell = adp->f.fid.Cell;
1569 tfid.Fid.Volume = adp->f.fid.Fid.Volume;
1570 afs_Trace4(afs_iclSetp, CM_TRACE_LOOKUP, ICL_TYPE_POINTER, adp,
1571 ICL_TYPE_STRING, tname, ICL_TYPE_FID, &tfid,
1572 ICL_TYPE_INT32, code);
1575 if (code != ENOENT) {
1576 /*printf("LOOKUP dirLookupOff -> %d\n", code);*/
1581 /* prefetch some entries, if the dir is currently open. The variable
1582 * dirCookie tells us where to start prefetching from.
1584 if (!AFS_IS_DISCONNECTED &&
1585 AFSDOBULK && adp->opens > 0 && !(adp->f.states & CForeign)
1586 && !afs_IsDynroot(adp) && !afs_InReadDir(adp)) {
1588 /* if the entry is not in the cache, or is in the cache,
1589 * but hasn't been statd, then do a bulk stat operation.
1593 ObtainReadLock(&afs_xvcache);
1594 tvc = afs_FindVCache(&tfid, &retry, 0 /* !stats,!lru */ );
1595 ReleaseReadLock(&afs_xvcache);
1596 } while (tvc && retry);
1598 if (!tvc || !(tvc->f.states & CStatd))
1599 bulkcode = afs_DoBulkStat(adp, dirCookie, &treq);
1603 /* if the vcache isn't usable, release it */
1604 if (tvc && !(tvc->f.states & CStatd)) {
1605 #ifndef AFS_FBSD80_ENV
1615 /* now get the status info, if we don't already have it */
1616 /* This is kind of weird, but we might wind up accidentally calling
1617 * RXAFS_Lookup because we happened upon a file which legitimately
1618 * has a 0 uniquifier. That is the result of allowing unique to wrap
1619 * to 0. This was fixed in AFS 3.4. For CForeign, Unique == 0 means that
1620 * the file has not yet been looked up.
1623 afs_int32 cached = 0;
1624 if (!tfid.Fid.Unique && (adp->f.states & CForeign)) {
1625 tvc = afs_LookupVCache(&tfid, &treq, &cached, adp, tname);
1627 if (!tvc && !bulkcode) { /* lookup failed or wasn't called */
1628 tvc = afs_GetVCache(&tfid, &treq, &cached, NULL);
1631 } /* sub-block just to reduce stack usage */
1634 if (adp->f.states & CForeign)
1635 tvc->f.states |= CForeign;
1636 tvc->f.parent.vnode = adp->f.fid.Fid.Vnode;
1637 tvc->f.parent.unique = adp->f.fid.Fid.Unique;
1638 tvc->f.states &= ~CBulkStat;
1640 if (afs_fakestat_enable == 2 && tvc->mvstat == 1) {
1641 ObtainSharedLock(&tvc->lock, 680);
1642 if (!tvc->linkData) {
1643 UpgradeSToWLock(&tvc->lock, 681);
1644 code = afs_HandleLink(tvc, &treq);
1645 ConvertWToRLock(&tvc->lock);
1647 ConvertSToRLock(&tvc->lock);
1650 if (!code && !afs_strchr(tvc->linkData, ':'))
1652 ReleaseReadLock(&tvc->lock);
1654 if (tvc->mvstat == 1 && (tvc->f.states & CMValid) && tvc->mvid != NULL)
1655 force_eval = 1; /* This is now almost for free, get it correct */
1657 #if defined(UKERNEL) && defined(AFS_WEB_ENHANCEMENTS)
1658 if (!(flags & AFS_LOOKUP_NOEVAL))
1659 /* don't eval mount points */
1660 #endif /* UKERNEL && AFS_WEB_ENHANCEMENTS */
1661 if (tvc->mvstat == 1 && force_eval) {
1662 /* a mt point, possibly unevaluated */
1663 struct volume *tvolp;
1665 ObtainWriteLock(&tvc->lock, 133);
1666 code = EvalMountPoint(tvc, adp, &tvolp, &treq);
1667 ReleaseWriteLock(&tvc->lock);
1670 #ifndef AFS_FBSD80_ENV
1674 afs_PutVolume(tvolp, WRITE_LOCK);
1678 /* next, we want to continue using the target of the mt point */
1679 if (tvc->mvid && (tvc->f.states & CMValid)) {
1681 /* now lookup target, to set .. pointer */
1682 afs_Trace2(afs_iclSetp, CM_TRACE_LOOKUP1,
1683 ICL_TYPE_POINTER, tvc, ICL_TYPE_FID,
1685 uvc = tvc; /* remember for later */
1687 if (tvolp && (tvolp->states & VForeign)) {
1688 /* XXXX tvolp has ref cnt on but not locked! XXX */
1690 afs_GetRootVCache(tvc->mvid, &treq, NULL, tvolp);
1692 tvc = afs_GetVCache(tvc->mvid, &treq, NULL, NULL);
1694 #ifndef AFS_FBSD80_ENV
1695 afs_PutVCache(uvc); /* we're done with it */
1701 afs_PutVolume(tvolp, WRITE_LOCK);
1706 /* now, if we came via a new mt pt (say because of a new
1707 * release of a R/O volume), we must reevaluate the ..
1708 * ptr to point back to the appropriate place */
1710 ObtainWriteLock(&tvc->lock, 134);
1711 if (tvc->mvid == NULL) {
1712 tvc->mvid = (struct VenusFid *)
1713 osi_AllocSmallSpace(sizeof(struct VenusFid));
1715 /* setup backpointer */
1716 *tvc->mvid = tvolp->dotdot;
1717 ReleaseWriteLock(&tvc->lock);
1718 afs_PutVolume(tvolp, WRITE_LOCK);
1721 #ifndef AFS_FBSD80_ENV
1726 afs_PutVolume(tvolp, WRITE_LOCK);
1731 if (tvc && !VREFCOUNT_GT(tvc, 0)) {
1736 /* if we get here, we found something in a directory that couldn't
1737 * be located (a Multics "connection failure"). If the volume is
1738 * read-only, we try flushing this entry from the cache and trying
1740 if (!AFS_IS_DISCONNECTED) {
1743 tv = afs_GetVolume(&adp->f.fid, &treq, READ_LOCK);
1745 if (tv->states & VRO) {
1746 pass = 1; /* try this *once* */
1747 ObtainWriteLock(&afs_xcbhash, 495);
1748 afs_DequeueCallback(adp);
1749 /* re-stat to get later version */
1750 adp->f.states &= ~CStatd;
1751 ReleaseWriteLock(&afs_xcbhash);
1752 osi_dnlc_purgedp(adp);
1753 afs_PutVolume(tv, READ_LOCK);
1756 afs_PutVolume(tv, READ_LOCK);
1766 /* put the network buffer back, if need be */
1767 if (tname != aname && tname)
1768 osi_FreeLargeSpace(tname);
1772 afs_AddMarinerName(aname, tvc);
1774 #if defined(UKERNEL) && defined(AFS_WEB_ENHANCEMENTS)
1775 if (!(flags & AFS_LOOKUP_NOEVAL)) {
1776 /* Here we don't enter the name into the DNLC because we want the
1777 * evaluated mount dir to be there (the vcache for the mounted
1778 * volume) rather than the vc of the mount point itself. We can
1779 * still find the mount point's vc in the vcache by its fid. */
1780 #endif /* UKERNEL && AFS_WEB_ENHANCEMENTS */
1781 if (!hit && force_eval) {
1782 osi_dnlc_enter(adp, aname, tvc, &versionNo);
1784 #ifdef AFS_LINUX20_ENV
1785 /* So Linux inode cache is up to date. */
1786 code = afs_VerifyVCache(tvc, &treq);
1788 afs_PutFakeStat(&fakestate);
1789 AFS_DISCON_UNLOCK();
1790 return 0; /* can't have been any errors if hit and !code */
1793 #if defined(UKERNEL) && defined(AFS_WEB_ENHANCEMENTS)
1800 code = afs_CheckCode(code, &treq, 19);
1802 /* If there is an error, make sure *avcp is null.
1803 * Alphas panic otherwise - defect 10719.
1808 afs_PutFakeStat(&fakestate);
1809 AFS_DISCON_UNLOCK();