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, 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(struct vcache *avc, struct vcache *advc,
275 struct volume **avolpp, 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(char *aname)
510 AFS_STATCNT(ENameOK);
511 tlen = strlen(aname);
512 if (tlen >= 4 && strcmp(aname + tlen - 4, "@sys") == 0)
518 afs_getsysname(struct vrequest *areq, struct vcache *adp,
519 char *bufp, int *num, char **sysnamelist[])
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(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(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) {
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 */
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
654 struct vcache *lruvcp; /* vcache ptr of our goal pos in LRU queue */
655 struct dcache *dcp; /* chunk containing the dir block */
656 afs_size_t temp; /* temp for holding chunk length, &c. */
657 struct AFSFid *fidsp; /* file IDs were collecting */
658 struct AFSCallBack *cbsp; /* call back pointers */
659 struct AFSCallBack *tcbp; /* temp callback ptr */
660 struct AFSFetchStatus *statsp; /* file status info */
661 struct AFSVolSync volSync; /* vol sync return info */
662 struct vcache *tvcp; /* temp vcp */
663 struct afs_q *tq; /* temp queue variable */
664 AFSCBFids fidParm; /* file ID parm for bulk stat */
665 AFSBulkStats statParm; /* stat info parm for bulk stat */
666 int fidIndex = 0; /* which file were stating */
667 struct afs_conn *tcp = 0; /* conn for call */
668 AFSCBs cbParm; /* callback parm for bulk stat */
669 struct server *hostp = 0; /* host we got callback from */
670 long startTime; /* time we started the call,
671 * for callback expiration base
673 afs_size_t statSeqNo = 0; /* Valued of file size to detect races */
674 int code; /* error code */
675 long newIndex; /* new index in the dir */
676 struct DirEntry *dirEntryp; /* dir entry we are examining */
678 struct VenusFid afid; /* file ID we are using now */
679 struct VenusFid tfid; /* another temp. file ID */
680 afs_int32 retry; /* handle low-level SGI MP race conditions */
681 long volStates; /* flags from vol structure */
682 struct volume *volp = 0; /* volume ptr */
683 struct VenusFid dotdot = {0, {0, 0, 0}};
684 int flagIndex = 0; /* First file with bulk fetch flag set */
685 int inlinebulk = 0; /* Did we use InlineBulk RPC or not? */
688 dotdot.Fid.Unique = 0;
689 dotdot.Fid.Vnode = 0;
691 /* first compute some basic parameters. We dont want to prefetch more
692 * than a fraction of the cache in any given call, and we want to preserve
693 * a portion of the LRU queue in any event, so as to avoid thrashing
694 * the entire stat cache (we will at least leave some of it alone).
695 * presently dont stat more than 1/8 the cache in any one call. */
696 nentries = afs_cacheStats / 8;
698 /* dont bother prefetching more than one calls worth of info */
699 if (nentries > AFSCBMAX)
702 /* heuristic to make sure that things fit in 4K. This means that
703 * we shouldnt make it any bigger than 47 entries. I am typically
704 * going to keep it a little lower, since we don't want to load
705 * too much of the stat cache.
710 /* now, to reduce the stack size, well allocate two 4K blocks,
711 * one for fids and callbacks, and one for stat info. Well set
712 * up our pointers to the memory from there, too.
714 statsp = (AFSFetchStatus *)
715 osi_Alloc(AFSCBMAX * sizeof(AFSFetchStatus));
716 fidsp = (AFSFid *) osi_AllocLargeSpace(nentries * sizeof(AFSFid));
717 cbsp = (AFSCallBack *)
718 osi_Alloc(AFSCBMAX * sizeof(AFSCallBack));
720 /* next, we must iterate over the directory, starting from the specified
721 * cookie offset (dirCookie), and counting out nentries file entries.
722 * We skip files that already have stat cache entries, since we
723 * dont want to bulk stat files that are already in the cache.
726 code = afs_VerifyVCache(adp, areqp);
730 dcp = afs_GetDCache(adp, (afs_size_t) 0, areqp, &temp, &temp, 1);
736 /* lock the directory cache entry */
737 ObtainReadLock(&adp->lock);
738 ObtainReadLock(&dcp->lock);
741 * Make sure that the data in the cache is current. There are two
742 * cases we need to worry about:
743 * 1. The cache data is being fetched by another process.
744 * 2. The cache data is no longer valid
746 while ((adp->f.states & CStatd)
747 && (dcp->dflags & DFFetching)
748 && hsame(adp->f.m.DataVersion, dcp->f.versionNo)) {
749 afs_Trace4(afs_iclSetp, CM_TRACE_DCACHEWAIT, ICL_TYPE_STRING,
750 __FILE__, ICL_TYPE_INT32, __LINE__, ICL_TYPE_POINTER, dcp,
751 ICL_TYPE_INT32, dcp->dflags);
752 ReleaseReadLock(&dcp->lock);
753 ReleaseReadLock(&adp->lock);
754 afs_osi_Sleep(&dcp->validPos);
755 ObtainReadLock(&adp->lock);
756 ObtainReadLock(&dcp->lock);
758 if (!(adp->f.states & CStatd)
759 || !hsame(adp->f.m.DataVersion, dcp->f.versionNo)) {
760 ReleaseReadLock(&dcp->lock);
761 ReleaseReadLock(&adp->lock);
766 /* Generate a sequence number so we can tell whether we should
767 * store the attributes when processing the response. This number is
768 * stored in the file size when we set the CBulkFetching bit. If the
769 * CBulkFetching is still set and this value hasn't changed, then
770 * we know we were the last to set CBulkFetching bit for this file,
771 * and it is safe to set the status information for this file.
773 statSeqNo = bulkStatCounter++;
774 /* ensure against wrapping */
776 statSeqNo = bulkStatCounter++;
778 /* now we have dir data in the cache, so scan the dir page */
781 while (1) { /* Should probably have some constant bound */
782 /* look for first safe entry to examine in the directory. BlobScan
783 * looks for a the 1st allocated dir after the dirCookie slot.
785 newIndex = BlobScan(dcp, (dirCookie >> 5));
789 /* remember the updated directory cookie */
790 dirCookie = newIndex << 5;
792 /* get a ptr to the dir entry */
794 (struct DirEntry *)afs_dir_GetBlob(dcp, newIndex);
798 /* dont copy more than we have room for */
799 if (fidIndex >= nentries) {
800 DRelease(dirEntryp, 0);
804 /* now, if the dir entry looks good, copy it out to our list. Vnode
805 * 0 means deleted, although it should also be free were it deleted.
807 if (dirEntryp->fid.vnode != 0) {
808 /* dont copy entries we have in our cache. This check will
809 * also make us skip "." and probably "..", unless it has
810 * disappeared from the cache since we did our namei call.
812 tfid.Cell = adp->f.fid.Cell;
813 tfid.Fid.Volume = adp->f.fid.Fid.Volume;
814 tfid.Fid.Vnode = ntohl(dirEntryp->fid.vnode);
815 tfid.Fid.Unique = ntohl(dirEntryp->fid.vunique);
818 ObtainWriteLock(&afs_xvcache, 130);
819 tvcp = afs_FindVCache(&tfid, &retry, IS_WLOCK /* no stats | LRU */ );
821 ReleaseWriteLock(&afs_xvcache);
824 } while (tvcp && retry);
825 if (!tvcp) { /* otherwise, create manually */
826 tvcp = afs_NewBulkVCache(&tfid, hostp, statSeqNo);
829 ObtainWriteLock(&tvcp->lock, 505);
830 ReleaseWriteLock(&afs_xvcache);
831 afs_RemoveVCB(&tfid);
832 ReleaseWriteLock(&tvcp->lock);
834 ReleaseWriteLock(&afs_xvcache);
837 ReleaseWriteLock(&afs_xvcache);
841 DRelease(dirEntryp, 0);
842 ReleaseReadLock(&dcp->lock);
843 ReleaseReadLock(&adp->lock);
845 goto done; /* can happen if afs_NewVCache fails */
848 /* WARNING: afs_DoBulkStat uses the Length field to store a
849 * sequence number for each bulk status request. Under no
850 * circumstances should afs_DoBulkStat store a sequence number
851 * if the new length will be ignored when afs_ProcessFS is
852 * called with new stats. */
854 if (!(tvcp->f.states & CStatd)
855 && (!((tvcp->f.states & CBulkFetching) &&
856 (tvcp->f.m.Length != statSeqNo)))
857 && (tvcp->execsOrWriters <= 0)
858 && !afs_DirtyPages(tvcp)
859 && !AFS_VN_MAPPED((vnode_t *) tvcp))
861 if (!(tvcp->f.states & CStatd)
862 && (!((tvcp->f.states & CBulkFetching) &&
863 (tvcp->f.m.Length != statSeqNo)))
864 && (tvcp->execsOrWriters <= 0)
865 && !afs_DirtyPages(tvcp))
869 /* this entry doesnt exist in the cache, and is not
870 * already being fetched by someone else, so add it to the
871 * list of file IDs to obtain.
873 * We detect a callback breaking race condition by checking the
874 * CBulkFetching state bit and the value in the file size.
875 * It is safe to set the status only if the CBulkFetching
876 * flag is still set and the value in the file size does
877 * not change. NewBulkVCache sets us up for the new ones.
878 * Set up the rest here.
880 * Don't fetch status for dirty files. We need to
881 * preserve the value of the file size. We could
882 * flush the pages, but it wouldn't be worthwhile.
884 if (!(tvcp->f.states & CBulkFetching)) {
885 tvcp->f.states |= CBulkFetching;
886 tvcp->f.m.Length = statSeqNo;
888 memcpy((char *)(fidsp + fidIndex), (char *)&tfid.Fid,
895 /* if dir vnode has non-zero entry */
896 /* move to the next dir entry by adding in the # of entries
897 * used by this dir entry.
899 temp = afs_dir_NameBlobs(dirEntryp->name) << 5;
900 DRelease(dirEntryp, 0);
904 } /* while loop over all dir entries */
906 /* now release the dir lock and prepare to make the bulk RPC */
907 ReleaseReadLock(&dcp->lock);
908 ReleaseReadLock(&adp->lock);
910 /* release the chunk */
913 /* dont make a null call */
918 /* setup the RPC parm structures */
919 fidParm.AFSCBFids_len = fidIndex;
920 fidParm.AFSCBFids_val = fidsp;
921 statParm.AFSBulkStats_len = fidIndex;
922 statParm.AFSBulkStats_val = statsp;
923 cbParm.AFSCBs_len = fidIndex;
924 cbParm.AFSCBs_val = cbsp;
926 /* start the timer; callback expirations are relative to this */
927 startTime = osi_Time();
929 tcp = afs_Conn(&adp->f.fid, areqp, SHARED_LOCK);
931 hostp = tcp->parent->srvr->server;
932 XSTATS_START_TIME(AFS_STATS_FS_RPCIDX_BULKSTATUS);
934 if (!(tcp->parent->srvr->server->flags & SNO_INLINEBULK)) {
938 RXAFS_InlineBulkStatus(tcp->id, &fidParm, &statParm,
941 if (code == RXGEN_OPCODE) {
942 tcp->parent->srvr->server->flags |= SNO_INLINEBULK;
946 RXAFS_BulkStatus(tcp->id, &fidParm, &statParm,
951 if (!code && ((&statsp[0])->errorCode)) {
953 * If this is an error needing retry, do so.
954 * Retryable errors are all whole-volume or
957 if (afs_Analyze(tcp, (&statsp[0])->errorCode,
959 AFS_STATS_FS_RPCIDX_BULKSTATUS,
960 SHARED_LOCK, NULL) != 0)
968 RXAFS_BulkStatus(tcp->id, &fidParm, &statParm, &cbParm,
976 (tcp, code, &adp->f.fid, areqp, AFS_STATS_FS_RPCIDX_BULKSTATUS,
979 /* now, if we didnt get the info, bail out. */
983 /* we need vol flags to create the entries properly */
984 dotdot.Fid.Volume = 0;
985 volp = afs_GetVolume(&adp->f.fid, areqp, READ_LOCK);
987 volStates = volp->states;
988 if (volp->dotdot.Fid.Volume != 0)
989 dotdot = volp->dotdot;
993 /* find the place to merge the info into We do this by skipping
994 * nskip entries in the LRU queue. The more we skip, the more
995 * we preserve, since the head of the VLRU queue is the most recently
999 nskip = afs_cacheStats / 2; /* preserved fraction of the cache */
1000 ObtainReadLock(&afs_xvcache);
1001 #ifdef AFS_DARWIN80_ENV
1004 if (QEmpty(&VLRU)) {
1005 /* actually a serious error, probably should panic. Probably will
1006 * panic soon, oh well. */
1007 ReleaseReadLock(&afs_xvcache);
1008 afs_warnuser("afs_DoBulkStat: VLRU empty!");
1011 if ((VLRU.next->prev != &VLRU) || (VLRU.prev->next != &VLRU)) {
1012 refpanic("Bulkstat VLRU inconsistent");
1014 for (tq = VLRU.next; tq != &VLRU; tq = QNext(tq)) {
1016 #ifdef AFS_DARWIN80_ENV
1017 if ((!(QTOV(tq)->f.states & CDeadVnode)&&!(QTOV(tq)->f.states & CVInit)))
1021 if (QNext(QPrev(tq)) != tq) {
1023 refpanic("BulkStat VLRU inconsistent");
1029 lruvcp = QTOV(VLRU.next);
1031 /* now we have to hold this entry, so that it does not get moved
1032 * into the free list while we're running. It could still get
1033 * moved within the lru queue, but hopefully that will be rare; it
1034 * doesn't hurt nearly as much.
1037 #ifdef AFS_DARWIN80_ENV
1038 if (((lruvcp->f.states & CDeadVnode)||(lruvcp->f.states & CVInit))) {
1044 panic("Can't find non-dead vnode in VLRU\n");
1046 lruvp = AFSTOV(lruvcp);
1047 if (vnode_get(lruvp)) /* this bumps ref count */
1049 else if (vnode_ref(lruvp)) {
1051 /* AFSTOV(lruvcp) may be NULL */
1057 osi_vnhold(lruvcp, &retry);
1059 ReleaseReadLock(&afs_xvcache); /* could be read lock */
1063 /* otherwise, merge in the info. We have to be quite careful here,
1064 * since we need to ensure that we don't merge old info over newer
1065 * stuff in a stat cache entry. We're very conservative here: we don't
1066 * do the merge at all unless we ourselves create the stat cache
1067 * entry. That's pretty safe, and should work pretty well, since we
1068 * typically expect to do the stat cache creation ourselves.
1070 * We also have to take into account racing token revocations.
1072 for (i = 0; i < fidIndex; i++) {
1073 if ((&statsp[i])->errorCode)
1075 afid.Cell = adp->f.fid.Cell;
1076 afid.Fid.Volume = adp->f.fid.Fid.Volume;
1077 afid.Fid.Vnode = fidsp[i].Vnode;
1078 afid.Fid.Unique = fidsp[i].Unique;
1081 ObtainReadLock(&afs_xvcache);
1082 tvcp = afs_FindVCache(&afid, &retry, FIND_CDEAD /* !stats&!lru */);
1083 ReleaseReadLock(&afs_xvcache);
1084 } while (tvcp && retry);
1086 /* The entry may no longer exist */
1091 /* now we have the entry held, but we need to fill it in */
1092 ObtainWriteLock(&tvcp->lock, 131);
1094 /* if CBulkFetching is not set, or if the file size no longer
1095 * matches the value we placed there when we set the CBulkFetching
1096 * flag, then someone else has done something with this node,
1097 * and we may not have the latest status information for this
1098 * file. Leave the entry alone.
1100 if (!(tvcp->f.states & CBulkFetching) || (tvcp->f.m.Length != statSeqNo)) {
1102 ReleaseWriteLock(&tvcp->lock);
1103 afs_PutVCache(tvcp);
1107 /* now copy ".." entry back out of volume structure, if necessary */
1108 if (tvcp->mvstat == 2 && (dotdot.Fid.Volume != 0)) {
1110 tvcp->mvid = (struct VenusFid *)
1111 osi_AllocSmallSpace(sizeof(struct VenusFid));
1112 *tvcp->mvid = dotdot;
1115 #ifdef AFS_DARWIN80_ENV
1116 if (((lruvcp->f.states & CDeadVnode)||(lruvcp->f.states & CVInit)))
1117 panic("vlru control point went dead\n");
1120 ObtainWriteLock(&afs_xvcache, 132);
1121 if ((VLRU.next->prev != &VLRU) || (VLRU.prev->next != &VLRU)) {
1122 refpanic("Bulkstat VLRU inconsistent2");
1124 if ((QNext(QPrev(&tvcp->vlruq)) != &tvcp->vlruq)
1125 || (QPrev(QNext(&tvcp->vlruq)) != &tvcp->vlruq)) {
1126 refpanic("Bulkstat VLRU inconsistent4");
1128 if ((QNext(QPrev(&lruvcp->vlruq)) != &lruvcp->vlruq)
1129 || (QPrev(QNext(&lruvcp->vlruq)) != &lruvcp->vlruq)) {
1130 refpanic("Bulkstat VLRU inconsistent5");
1133 if (tvcp != lruvcp) { /* if they are == don't move it, don't corrupt vlru */
1134 QRemove(&tvcp->vlruq);
1135 QAdd(&lruvcp->vlruq, &tvcp->vlruq);
1138 if ((VLRU.next->prev != &VLRU) || (VLRU.prev->next != &VLRU)) {
1139 refpanic("Bulkstat VLRU inconsistent3");
1141 if ((QNext(QPrev(&tvcp->vlruq)) != &tvcp->vlruq)
1142 || (QPrev(QNext(&tvcp->vlruq)) != &tvcp->vlruq)) {
1143 refpanic("Bulkstat VLRU inconsistent5");
1145 if ((QNext(QPrev(&lruvcp->vlruq)) != &lruvcp->vlruq)
1146 || (QPrev(QNext(&lruvcp->vlruq)) != &lruvcp->vlruq)) {
1147 refpanic("Bulkstat VLRU inconsistent6");
1149 ReleaseWriteLock(&afs_xvcache);
1151 ObtainWriteLock(&afs_xcbhash, 494);
1153 /* We need to check the flags again. We may have missed
1154 * something while we were waiting for a lock.
1156 if (!(tvcp->f.states & CBulkFetching) || (tvcp->f.m.Length != statSeqNo)) {
1158 ReleaseWriteLock(&tvcp->lock);
1159 ReleaseWriteLock(&afs_xcbhash);
1160 afs_PutVCache(tvcp);
1164 /* now merge in the resulting status back into the vnode.
1165 * We only do this if the entry looks clear.
1167 afs_ProcessFS(tvcp, &statsp[i], areqp);
1168 #if defined(AFS_LINUX22_ENV)
1169 afs_fill_inode(AFSTOV(tvcp), NULL); /* reset inode operations */
1172 /* do some accounting for bulk stats: mark this entry as
1173 * loaded, so we can tell if we use it before it gets
1176 tvcp->f.states |= CBulkStat;
1177 tvcp->f.states &= ~CBulkFetching;
1179 afs_bulkStatsDone++;
1181 /* merge in vol info */
1182 if (volStates & VRO)
1183 tvcp->f.states |= CRO;
1184 if (volStates & VBackup)
1185 tvcp->f.states |= CBackup;
1186 if (volStates & VForeign)
1187 tvcp->f.states |= CForeign;
1189 /* merge in the callback info */
1190 tvcp->f.states |= CTruth;
1192 /* get ptr to the callback we are interested in */
1195 if (tcbp->ExpirationTime != 0) {
1196 tvcp->cbExpires = tcbp->ExpirationTime + startTime;
1197 tvcp->callback = hostp;
1198 tvcp->f.states |= CStatd;
1199 afs_QueueCallback(tvcp, CBHash(tcbp->ExpirationTime), volp);
1200 } else if (tvcp->f.states & CRO) {
1201 /* ordinary callback on a read-only volume -- AFS 3.2 style */
1202 tvcp->cbExpires = 3600 + startTime;
1203 tvcp->callback = hostp;
1204 tvcp->f.states |= CStatd;
1205 afs_QueueCallback(tvcp, CBHash(3600), volp);
1208 tvcp->f.states &= ~(CStatd | CUnique);
1209 afs_DequeueCallback(tvcp);
1210 if ((tvcp->f.states & CForeign) || (vType(tvcp) == VDIR))
1211 osi_dnlc_purgedp(tvcp); /* if it (could be) a directory */
1213 #ifdef AFS_DARWIN80_ENV
1214 /* reclaim->FlushVCache will need xcbhash */
1215 if (((tvcp->f.states & CDeadVnode)||(tvcp->f.states & CVInit))) {
1216 ReleaseWriteLock(&afs_xcbhash);
1217 /* passing in a parent hangs getting the vnode lock */
1218 code = afs_darwin_finalizevnode(tvcp, NULL, NULL, 0, 1);
1220 /* It's gonna get recycled - shouldn't happen */
1222 tvcp->f.states &= ~(CStatd | CUnique);
1223 afs_DequeueCallback(tvcp);
1224 if ((tvcp->f.states & CForeign) || (vType(tvcp) == VDIR))
1225 osi_dnlc_purgedp(tvcp); /* if it (could be) a directory */
1227 /* re-acquire the usecount that finalizevnode disposed of */
1228 vnode_ref(AFSTOV(tvcp));
1232 ReleaseWriteLock(&afs_xcbhash);
1234 ReleaseWriteLock(&tvcp->lock);
1235 /* finally, we're done with the entry */
1236 afs_PutVCache(tvcp);
1237 } /* for all files we got back */
1239 /* finally return the pointer into the LRU queue */
1240 #ifdef AFS_DARWIN80_ENV
1241 if (((lruvcp->f.states & CDeadVnode)||(lruvcp->f.states & CVInit)))
1242 panic("vlru control point went dead before put\n");
1248 afs_PutVCache(lruvcp);
1252 /* Be sure to turn off the CBulkFetching flags */
1253 for (i = flagIndex; i < fidIndex; i++) {
1254 afid.Cell = adp->f.fid.Cell;
1255 afid.Fid.Volume = adp->f.fid.Fid.Volume;
1256 afid.Fid.Vnode = fidsp[i].Vnode;
1257 afid.Fid.Unique = fidsp[i].Unique;
1260 ObtainReadLock(&afs_xvcache);
1261 tvcp = afs_FindVCache(&afid, &retry, FIND_CDEAD /* !stats&!lru */);
1262 ReleaseReadLock(&afs_xvcache);
1263 } while (tvcp && retry);
1265 if ((tvcp->f.states & CBulkFetching)
1266 && (tvcp->f.m.Length == statSeqNo)) {
1267 tvcp->f.states &= ~CBulkFetching;
1269 afs_PutVCache(tvcp);
1273 afs_PutVolume(volp, READ_LOCK);
1275 /* If we did the InlineBulk RPC pull out the return code */
1276 if (inlinebulk && code == 0) {
1277 if ((&statsp[0])->errorCode) {
1278 afs_Analyze(tcp, (&statsp[0])->errorCode, &adp->f.fid, areqp,
1279 AFS_STATS_FS_RPCIDX_BULKSTATUS, SHARED_LOCK, NULL);
1280 code = (&statsp[0])->errorCode;
1286 osi_FreeLargeSpace((char *)fidsp);
1287 osi_Free((char *)statsp, AFSCBMAX * sizeof(AFSFetchStatus));
1288 osi_Free((char *)cbsp, AFSCBMAX * sizeof(AFSCallBack));
1292 /* was: (AFS_DEC_ENV) || defined(AFS_OSF30_ENV) || defined(AFS_NCR_ENV) */
1293 static int AFSDOBULK = 1;
1296 osi_lookup_isdot(const char *aname)
1300 /* in Solaris, we can get passed "" as a path component if we are the
1301 * root directory, e.g. after a call to chroot. It is equivalent to
1305 #endif /* AFS_SUN5_ENV */
1306 if (aname[0] == '.' && !aname[1]) {
1313 #if defined(AFS_SUN5_ENV) || defined(AFS_SGI_ENV)
1314 afs_lookup(OSI_VC_DECL(adp), char *aname, struct vcache **avcp, struct pathname *pnp, int flags, struct vnode *rdir, afs_ucred_t *acred)
1315 #elif defined(UKERNEL)
1316 afs_lookup(OSI_VC_DECL(adp), char *aname, struct vcache **avcp, afs_ucred_t *acred, int flags)
1318 afs_lookup(OSI_VC_DECL(adp), char *aname, struct vcache **avcp, afs_ucred_t *acred)
1321 struct vrequest treq;
1323 struct vcache *tvc = 0;
1325 afs_int32 bulkcode = 0;
1326 int pass = 0, hit = 0;
1327 int force_eval = afs_fakestat_enable ? 0 : 1;
1329 extern afs_int32 afs_mariner; /*Writing activity to log? */
1330 afs_hyper_t versionNo;
1331 int no_read_access = 0;
1332 struct sysname_info sysState; /* used only for @sys checking */
1333 int dynrootRetry = 1;
1334 struct afs_fakestat_state fakestate;
1335 int tryEvalOnly = 0;
1336 OSI_VC_CONVERT(adp);
1338 AFS_STATCNT(afs_lookup);
1339 afs_InitFakeStat(&fakestate);
1343 if ((code = afs_InitReq(&treq, acred)))
1346 if (afs_fakestat_enable && adp->mvstat == 1) {
1347 if (strcmp(aname, ".directory") == 0)
1351 #if defined(AFS_DARWIN_ENV)
1352 /* Workaround for MacOSX Finder, which tries to look for
1353 * .DS_Store and Contents under every directory.
1355 if (afs_fakestat_enable && adp->mvstat == 1) {
1356 if (strcmp(aname, ".DS_Store") == 0)
1358 if (strcmp(aname, "Contents") == 0)
1361 if (afs_fakestat_enable && adp->mvstat == 2) {
1362 if (strncmp(aname, "._", 2) == 0)
1368 code = afs_TryEvalFakeStat(&adp, &fakestate, &treq);
1370 code = afs_EvalFakeStat(&adp, &fakestate, &treq);
1372 /*printf("Code is %d\n", code);*/
1374 if (tryEvalOnly && adp->mvstat == 1)
1379 *avcp = NULL; /* Since some callers don't initialize it */
1381 /* come back to here if we encounter a non-existent object in a read-only
1382 * volume's directory */
1385 *avcp = NULL; /* Since some callers don't initialize it */
1388 if (!(adp->f.states & CStatd) && !afs_InReadDir(adp)) {
1389 if ((code = afs_VerifyVCache2(adp, &treq))) {
1395 /* watch for ".." in a volume root */
1396 if (adp->mvstat == 2 && aname[0] == '.' && aname[1] == '.' && !aname[2]) {
1397 /* looking up ".." in root via special hacks */
1398 if (adp->mvid == (struct VenusFid *)0 || adp->mvid->Fid.Volume == 0) {
1402 /* otherwise we have the fid here, so we use it */
1403 /*printf("Getting vcache\n");*/
1404 tvc = afs_GetVCache(adp->mvid, &treq, NULL, NULL);
1405 afs_Trace3(afs_iclSetp, CM_TRACE_GETVCDOTDOT, ICL_TYPE_FID, adp->mvid,
1406 ICL_TYPE_POINTER, tvc, ICL_TYPE_INT32, code);
1408 code = (tvc ? 0 : ENOENT);
1410 if (tvc && !VREFCOUNT_GT(tvc, 0)) {
1414 /*printf("LOOKUP GETVCDOTDOT -> %d\n", code); */
1419 /* now check the access */
1420 if (treq.uid != adp->last_looker) {
1421 if (!afs_AccessOK(adp, PRSFS_LOOKUP, &treq, CHECK_MODE_BITS)) {
1426 adp->last_looker = treq.uid;
1429 /* Check for read access as well. We need read access in order to
1430 * stat files, but not to stat subdirectories. */
1431 if (!afs_AccessOK(adp, PRSFS_READ, &treq, CHECK_MODE_BITS))
1434 /* special case lookup of ".". Can we check for it sooner in this code,
1435 * for instance, way up before "redo:" ??
1436 * I'm not fiddling with the LRUQ here, either, perhaps I should, or else
1437 * invent a lightweight version of GetVCache.
1439 if (osi_lookup_isdot(aname)) { /* special case */
1440 ObtainReadLock(&afs_xvcache);
1442 ReleaseReadLock(&afs_xvcache);
1443 #ifdef AFS_DARWIN80_ENV
1444 vnode_get(AFSTOV(adp));
1449 if (adp && !VREFCOUNT_GT(adp, 0)) {
1456 * Special case lookup of ".." in the dynamic mount directory.
1457 * The parent of this directory is _always_ the AFS root volume.
1459 if (afs_IsDynrootMount(adp) &&
1460 aname[0] == '.' && aname[1] == '.' && !aname[2]) {
1462 ObtainReadLock(&afs_xvcache);
1463 osi_vnhold(afs_globalVp, 0);
1464 ReleaseReadLock(&afs_xvcache);
1465 #ifdef AFS_DARWIN80_ENV
1466 vnode_get(AFSTOV(afs_globalVp));
1469 *avcp = tvc = afs_globalVp;
1475 * Special case lookups in the dynamic mount directory.
1476 * The names here take the form cell:volume, similar to a mount point.
1477 * EvalMountData parses that and returns a cell and volume ID, which
1478 * we use to construct the appropriate dynroot Fid.
1480 if (afs_IsDynrootMount(adp)) {
1481 struct VenusFid tfid;
1482 afs_uint32 cellidx, volid, vnoid, uniq;
1484 code = EvalMountData('%', aname, 0, 0, NULL, &treq, &cellidx, &volid, &vnoid, &uniq);
1487 /* If a vnode was returned, it's not a real mount point */
1489 struct cell *tcell = afs_GetCellByIndex(cellidx, READ_LOCK);
1490 tfid.Cell = tcell->cellNum;
1491 afs_PutCell(tcell, READ_LOCK);
1492 tfid.Fid.Vnode = vnoid;
1493 tfid.Fid.Volume = volid;
1494 tfid.Fid.Unique = uniq;
1496 afs_GetDynrootMountFid(&tfid);
1497 tfid.Fid.Vnode = VNUM_FROM_TYPEID(VN_TYPE_MOUNT, cellidx << 2);
1498 tfid.Fid.Unique = volid;
1500 *avcp = tvc = afs_GetVCache(&tfid, &treq, NULL, NULL);
1501 code = (tvc ? 0 : ENOENT);
1506 #ifdef AFS_LINUX26_ENV
1508 * Special case of the dynamic mount volume in a static root.
1509 * This is really unfortunate, but we need this for the translator.
1511 if (adp == afs_globalVp && !afs_GetDynrootEnable() &&
1512 !strcmp(aname, AFS_DYNROOT_MOUNTNAME)) {
1513 struct VenusFid tfid;
1515 afs_GetDynrootMountFid(&tfid);
1516 *avcp = tvc = afs_GetVCache(&tfid, &treq, NULL, NULL);
1523 Check_AtSys(adp, aname, &sysState, &treq);
1524 tname = sysState.name;
1526 /* 1st Check_AtSys and lookup by tname is required here, for now,
1527 * because the dnlc is *not* told to remove entries for the parent
1528 * dir of file/dir op that afs_LocalHero likes, but dnlc is informed
1529 * if the cached entry for the parent dir is invalidated for a
1531 * Otherwise, we'd be able to do a dnlc lookup on an entry ending
1532 * w/@sys and know the dnlc was consistent with reality. */
1533 tvc = osi_dnlc_lookup(adp, tname, WRITE_LOCK);
1534 *avcp = tvc; /* maybe wasn't initialized, but it is now */
1536 if (no_read_access && vType(tvc) != VDIR && vType(tvc) != VLNK) {
1537 /* need read access on dir to stat non-directory / non-link */
1543 #ifdef AFS_LINUX22_ENV
1544 if (tvc->mvstat == 2) { /* we don't trust the dnlc for root vcaches */
1545 AFS_RELE(AFSTOV(tvc));
1552 #else /* non - LINUX */
1556 #endif /* linux22 */
1559 { /* sub-block just to reduce stack usage */
1561 afs_size_t dirOffset, dirLen;
1562 struct VenusFid tfid;
1564 /* now we have to lookup the next fid */
1565 if (afs_InReadDir(adp))
1566 tdc = adp->dcreaddir;
1568 tdc = afs_GetDCache(adp, (afs_size_t) 0, &treq,
1569 &dirOffset, &dirLen, 1);
1571 *avcp = NULL; /* redundant, but harmless */
1576 /* now we will just call dir package with appropriate inode.
1577 * Dirs are always fetched in their entirety for now */
1578 ObtainReadLock(&adp->lock);
1579 ObtainReadLock(&tdc->lock);
1582 * Make sure that the data in the cache is current. There are two
1583 * cases we need to worry about:
1584 * 1. The cache data is being fetched by another process.
1585 * 2. The cache data is no longer valid
1587 * If a readdir is in progress _in this thread_, it has a shared
1588 * lock on the vcache and has obtained current data, so we just
1589 * use that. This eliminates several possible deadlocks.
1591 if (!afs_InReadDir(adp)) {
1592 while ((adp->f.states & CStatd)
1593 && (tdc->dflags & DFFetching)
1594 && hsame(adp->f.m.DataVersion, tdc->f.versionNo)) {
1595 ReleaseReadLock(&tdc->lock);
1596 ReleaseReadLock(&adp->lock);
1597 afs_osi_Sleep(&tdc->validPos);
1598 ObtainReadLock(&adp->lock);
1599 ObtainReadLock(&tdc->lock);
1601 if (!(adp->f.states & CStatd)
1602 || !hsame(adp->f.m.DataVersion, tdc->f.versionNo)) {
1603 ReleaseReadLock(&tdc->lock);
1604 ReleaseReadLock(&adp->lock);
1606 if (tname && tname != aname)
1607 osi_FreeLargeSpace(tname);
1612 /* Save the version number for when we call osi_dnlc_enter */
1613 hset(versionNo, tdc->f.versionNo);
1616 * check for, and handle "@sys" if it's there. We should be able
1617 * to avoid the alloc and the strcpy with a little work, but it's
1618 * not pressing. If there aren't any remote users (ie, via the
1619 * NFS translator), we have a slightly easier job.
1620 * the faster way to do this is to check for *aname == '@' and if
1621 * it's there, check for @sys, otherwise, assume there's no @sys
1622 * then, if the lookup fails, check for .*@sys...
1624 /* above now implemented by Check_AtSys and Next_AtSys */
1626 /* lookup the name in the appropriate dir, and return a cache entry
1627 * on the resulting fid */
1629 afs_dir_LookupOffset(tdc, sysState.name, &tfid.Fid,
1632 /* If the first lookup doesn't succeed, maybe it's got @sys in the name */
1633 while (code == ENOENT && Next_AtSys(adp, &treq, &sysState))
1635 afs_dir_LookupOffset(tdc, sysState.name, &tfid.Fid,
1637 tname = sysState.name;
1639 ReleaseReadLock(&tdc->lock);
1640 if (!afs_InReadDir(adp))
1642 if (code == ENOENT && afs_IsDynroot(adp) && dynrootRetry && !tryEvalOnly) {
1644 char *cn = (tname[0] == '.') ? tname + 1 : tname;
1645 ReleaseReadLock(&adp->lock);
1646 /* confirm it's not just hushed */
1647 tc = afs_GetCellByName(cn, WRITE_LOCK);
1649 if (tc->states & CHush) {
1650 tc->states &= ~CHush;
1651 ReleaseWriteLock(&tc->lock);
1652 afs_DynrootInvalidate();
1655 ReleaseWriteLock(&tc->lock);
1657 /* Allow a second dynroot retry if the cell was hushed before */
1659 if (tname[0] == '.')
1660 afs_LookupAFSDB(tname + 1);
1662 afs_LookupAFSDB(tname);
1663 if (tname && tname != aname)
1664 osi_FreeLargeSpace(tname);
1667 ReleaseReadLock(&adp->lock);
1670 /* new fid has same cell and volume */
1671 tfid.Cell = adp->f.fid.Cell;
1672 tfid.Fid.Volume = adp->f.fid.Fid.Volume;
1673 afs_Trace4(afs_iclSetp, CM_TRACE_LOOKUP, ICL_TYPE_POINTER, adp,
1674 ICL_TYPE_STRING, tname, ICL_TYPE_FID, &tfid,
1675 ICL_TYPE_INT32, code);
1678 if (code != ENOENT) {
1679 /*printf("LOOKUP dirLookupOff -> %d\n", code);*/
1684 /* prefetch some entries, if the dir is currently open. The variable
1685 * dirCookie tells us where to start prefetching from.
1687 if (!AFS_IS_DISCONNECTED &&
1688 AFSDOBULK && adp->opens > 0 && !(adp->f.states & CForeign)
1689 && !afs_IsDynroot(adp) && !afs_InReadDir(adp)) {
1691 /* if the entry is not in the cache, or is in the cache,
1692 * but hasn't been statd, then do a bulk stat operation.
1696 ObtainReadLock(&afs_xvcache);
1697 tvc = afs_FindVCache(&tfid, &retry, 0 /* !stats,!lru */ );
1698 ReleaseReadLock(&afs_xvcache);
1699 } while (tvc && retry);
1701 if (!tvc || !(tvc->f.states & CStatd))
1702 bulkcode = afs_DoBulkStat(adp, dirCookie, &treq);
1706 /* if the vcache isn't usable, release it */
1707 if (tvc && !(tvc->f.states & CStatd)) {
1716 /* now get the status info, if we don't already have it */
1717 /* This is kind of weird, but we might wind up accidentally calling
1718 * RXAFS_Lookup because we happened upon a file which legitimately
1719 * has a 0 uniquifier. That is the result of allowing unique to wrap
1720 * to 0. This was fixed in AFS 3.4. For CForeign, Unique == 0 means that
1721 * the file has not yet been looked up.
1724 afs_int32 cached = 0;
1725 if (!tfid.Fid.Unique && (adp->f.states & CForeign)) {
1726 tvc = afs_LookupVCache(&tfid, &treq, &cached, adp, tname);
1728 if (!tvc && !bulkcode) { /* lookup failed or wasn't called */
1729 tvc = afs_GetVCache(&tfid, &treq, &cached, NULL);
1732 } /* sub-block just to reduce stack usage */
1735 if (adp->f.states & CForeign)
1736 tvc->f.states |= CForeign;
1737 tvc->f.parent.vnode = adp->f.fid.Fid.Vnode;
1738 tvc->f.parent.unique = adp->f.fid.Fid.Unique;
1739 tvc->f.states &= ~CBulkStat;
1741 if (afs_fakestat_enable == 2 && tvc->mvstat == 1) {
1742 ObtainSharedLock(&tvc->lock, 680);
1743 if (!tvc->linkData) {
1744 UpgradeSToWLock(&tvc->lock, 681);
1745 code = afs_HandleLink(tvc, &treq);
1746 ConvertWToRLock(&tvc->lock);
1748 ConvertSToRLock(&tvc->lock);
1751 if (!code && !afs_strchr(tvc->linkData, ':'))
1753 ReleaseReadLock(&tvc->lock);
1755 if (tvc->mvstat == 1 && (tvc->f.states & CMValid) && tvc->mvid != NULL)
1756 force_eval = 1; /* This is now almost for free, get it correct */
1758 #if defined(UKERNEL) && defined(AFS_WEB_ENHANCEMENTS)
1759 if (!(flags & AFS_LOOKUP_NOEVAL))
1760 /* don't eval mount points */
1761 #endif /* UKERNEL && AFS_WEB_ENHANCEMENTS */
1762 if (tvc->mvstat == 1 && force_eval) {
1763 /* a mt point, possibly unevaluated */
1764 struct volume *tvolp;
1766 ObtainWriteLock(&tvc->lock, 133);
1767 code = EvalMountPoint(tvc, adp, &tvolp, &treq);
1768 ReleaseWriteLock(&tvc->lock);
1773 afs_PutVolume(tvolp, WRITE_LOCK);
1777 /* next, we want to continue using the target of the mt point */
1778 if (tvc->mvid && (tvc->f.states & CMValid)) {
1780 /* now lookup target, to set .. pointer */
1781 afs_Trace2(afs_iclSetp, CM_TRACE_LOOKUP1,
1782 ICL_TYPE_POINTER, tvc, ICL_TYPE_FID,
1784 uvc = tvc; /* remember for later */
1786 if (tvolp && (tvolp->states & VForeign)) {
1787 /* XXXX tvolp has ref cnt on but not locked! XXX */
1789 afs_GetRootVCache(tvc->mvid, &treq, NULL, tvolp);
1791 tvc = afs_GetVCache(tvc->mvid, &treq, NULL, NULL);
1793 afs_PutVCache(uvc); /* we're done with it */
1798 afs_PutVolume(tvolp, WRITE_LOCK);
1803 /* now, if we came via a new mt pt (say because of a new
1804 * release of a R/O volume), we must reevaluate the ..
1805 * ptr to point back to the appropriate place */
1807 ObtainWriteLock(&tvc->lock, 134);
1808 if (tvc->mvid == NULL) {
1809 tvc->mvid = (struct VenusFid *)
1810 osi_AllocSmallSpace(sizeof(struct VenusFid));
1812 /* setup backpointer */
1813 *tvc->mvid = tvolp->dotdot;
1814 ReleaseWriteLock(&tvc->lock);
1815 afs_PutVolume(tvolp, WRITE_LOCK);
1821 afs_PutVolume(tvolp, WRITE_LOCK);
1826 if (tvc && !VREFCOUNT_GT(tvc, 0)) {
1831 /* if we get here, we found something in a directory that couldn't
1832 * be located (a Multics "connection failure"). If the volume is
1833 * read-only, we try flushing this entry from the cache and trying
1835 if (!AFS_IS_DISCONNECTED) {
1838 tv = afs_GetVolume(&adp->f.fid, &treq, READ_LOCK);
1840 if (tv->states & VRO) {
1841 pass = 1; /* try this *once* */
1842 ObtainWriteLock(&afs_xcbhash, 495);
1843 afs_DequeueCallback(adp);
1844 /* re-stat to get later version */
1845 adp->f.states &= ~CStatd;
1846 ReleaseWriteLock(&afs_xcbhash);
1847 osi_dnlc_purgedp(adp);
1848 afs_PutVolume(tv, READ_LOCK);
1851 afs_PutVolume(tv, READ_LOCK);
1861 /* put the network buffer back, if need be */
1862 if (tname != aname && tname)
1863 osi_FreeLargeSpace(tname);
1867 afs_AddMarinerName(aname, tvc);
1869 #if defined(UKERNEL) && defined(AFS_WEB_ENHANCEMENTS)
1870 if (!(flags & AFS_LOOKUP_NOEVAL)) {
1871 /* Here we don't enter the name into the DNLC because we want the
1872 * evaluated mount dir to be there (the vcache for the mounted
1873 * volume) rather than the vc of the mount point itself. We can
1874 * still find the mount point's vc in the vcache by its fid. */
1875 #endif /* UKERNEL && AFS_WEB_ENHANCEMENTS */
1876 if (!hit && force_eval) {
1877 osi_dnlc_enter(adp, aname, tvc, &versionNo);
1879 #ifdef AFS_LINUX20_ENV
1880 /* So Linux inode cache is up to date. */
1881 code = afs_VerifyVCache(tvc, &treq);
1883 afs_PutFakeStat(&fakestate);
1884 AFS_DISCON_UNLOCK();
1885 return 0; /* can't have been any errors if hit and !code */
1888 #if defined(UKERNEL) && defined(AFS_WEB_ENHANCEMENTS)
1895 code = afs_CheckCode(code, &treq, 19);
1897 /* If there is an error, make sure *avcp is null.
1898 * Alphas panic otherwise - defect 10719.
1903 afs_PutFakeStat(&fakestate);
1904 AFS_DISCON_UNLOCK();