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 = NULL;
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, READ_LOCK);
533 error = EXP_SYSNAME(au->exporter, (char *)0, sysnamelist, num, 0);
535 strcpy(bufp, "@sys");
536 afs_PutUser(au, READ_LOCK);
539 strcpy(bufp, (*sysnamelist)[0]);
542 strcpy(bufp, afs_sysname);
543 afs_PutUser(au, READ_LOCK);
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, READ_LOCK);
610 EXP_SYSNAME(au->exporter, (char *)0, sysnamelist, &num, 0);
612 afs_PutUser(au, READ_LOCK);
616 afs_PutUser(au, READ_LOCK);
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? */
686 struct rx_connection *rxconn;
689 dotdot.Fid.Unique = 0;
690 dotdot.Fid.Vnode = 0;
692 /* first compute some basic parameters. We dont want to prefetch more
693 * than a fraction of the cache in any given call, and we want to preserve
694 * a portion of the LRU queue in any event, so as to avoid thrashing
695 * the entire stat cache (we will at least leave some of it alone).
696 * presently dont stat more than 1/8 the cache in any one call. */
697 nentries = afs_cacheStats / 8;
699 /* dont bother prefetching more than one calls worth of info */
700 if (nentries > AFSCBMAX)
703 /* heuristic to make sure that things fit in 4K. This means that
704 * we shouldnt make it any bigger than 47 entries. I am typically
705 * going to keep it a little lower, since we don't want to load
706 * too much of the stat cache.
711 /* now, to reduce the stack size, well allocate two 4K blocks,
712 * one for fids and callbacks, and one for stat info. Well set
713 * up our pointers to the memory from there, too.
715 statsp = (AFSFetchStatus *)
716 osi_Alloc(AFSCBMAX * sizeof(AFSFetchStatus));
717 fidsp = (AFSFid *) osi_AllocLargeSpace(nentries * sizeof(AFSFid));
718 cbsp = (AFSCallBack *)
719 osi_Alloc(AFSCBMAX * sizeof(AFSCallBack));
721 /* next, we must iterate over the directory, starting from the specified
722 * cookie offset (dirCookie), and counting out nentries file entries.
723 * We skip files that already have stat cache entries, since we
724 * dont want to bulk stat files that are already in the cache.
727 code = afs_VerifyVCache(adp, areqp);
731 dcp = afs_GetDCache(adp, (afs_size_t) 0, areqp, &temp, &temp, 1);
737 /* lock the directory cache entry */
738 ObtainReadLock(&adp->lock);
739 ObtainReadLock(&dcp->lock);
742 * Make sure that the data in the cache is current. There are two
743 * cases we need to worry about:
744 * 1. The cache data is being fetched by another process.
745 * 2. The cache data is no longer valid
747 while ((adp->f.states & CStatd)
748 && (dcp->dflags & DFFetching)
749 && hsame(adp->f.m.DataVersion, dcp->f.versionNo)) {
750 afs_Trace4(afs_iclSetp, CM_TRACE_DCACHEWAIT, ICL_TYPE_STRING,
751 __FILE__, ICL_TYPE_INT32, __LINE__, ICL_TYPE_POINTER, dcp,
752 ICL_TYPE_INT32, dcp->dflags);
753 ReleaseReadLock(&dcp->lock);
754 ReleaseReadLock(&adp->lock);
755 afs_osi_Sleep(&dcp->validPos);
756 ObtainReadLock(&adp->lock);
757 ObtainReadLock(&dcp->lock);
759 if (!(adp->f.states & CStatd)
760 || !hsame(adp->f.m.DataVersion, dcp->f.versionNo)) {
761 ReleaseReadLock(&dcp->lock);
762 ReleaseReadLock(&adp->lock);
767 /* Generate a sequence number so we can tell whether we should
768 * store the attributes when processing the response. This number is
769 * stored in the file size when we set the CBulkFetching bit. If the
770 * CBulkFetching is still set and this value hasn't changed, then
771 * we know we were the last to set CBulkFetching bit for this file,
772 * and it is safe to set the status information for this file.
774 statSeqNo = bulkStatCounter++;
775 /* ensure against wrapping */
777 statSeqNo = bulkStatCounter++;
779 /* now we have dir data in the cache, so scan the dir page */
782 while (1) { /* Should probably have some constant bound */
783 /* look for first safe entry to examine in the directory. BlobScan
784 * looks for a the 1st allocated dir after the dirCookie slot.
786 newIndex = BlobScan(dcp, (dirCookie >> 5));
790 /* remember the updated directory cookie */
791 dirCookie = newIndex << 5;
793 /* get a ptr to the dir entry */
795 (struct DirEntry *)afs_dir_GetBlob(dcp, newIndex);
799 /* dont copy more than we have room for */
800 if (fidIndex >= nentries) {
801 DRelease(dirEntryp, 0);
805 /* now, if the dir entry looks good, copy it out to our list. Vnode
806 * 0 means deleted, although it should also be free were it deleted.
808 if (dirEntryp->fid.vnode != 0) {
809 /* dont copy entries we have in our cache. This check will
810 * also make us skip "." and probably "..", unless it has
811 * disappeared from the cache since we did our namei call.
813 tfid.Cell = adp->f.fid.Cell;
814 tfid.Fid.Volume = adp->f.fid.Fid.Volume;
815 tfid.Fid.Vnode = ntohl(dirEntryp->fid.vnode);
816 tfid.Fid.Unique = ntohl(dirEntryp->fid.vunique);
819 ObtainWriteLock(&afs_xvcache, 130);
820 tvcp = afs_FindVCache(&tfid, &retry, IS_WLOCK|FIND_BULKDEAD /* no stats | LRU */ );
822 ReleaseWriteLock(&afs_xvcache);
825 } while (tvcp && retry);
826 if (!tvcp) { /* otherwise, create manually */
827 tvcp = afs_NewBulkVCache(&tfid, hostp, statSeqNo);
830 ObtainWriteLock(&tvcp->lock, 505);
831 ReleaseWriteLock(&afs_xvcache);
832 afs_RemoveVCB(&tfid);
833 ReleaseWriteLock(&tvcp->lock);
835 ReleaseWriteLock(&afs_xvcache);
838 ReleaseWriteLock(&afs_xvcache);
842 DRelease(dirEntryp, 0);
843 ReleaseReadLock(&dcp->lock);
844 ReleaseReadLock(&adp->lock);
846 goto done; /* can happen if afs_NewVCache fails */
849 /* WARNING: afs_DoBulkStat uses the Length field to store a
850 * sequence number for each bulk status request. Under no
851 * circumstances should afs_DoBulkStat store a sequence number
852 * if the new length will be ignored when afs_ProcessFS is
853 * called with new stats. */
855 if (!(tvcp->f.states & CStatd)
856 && (!((tvcp->f.states & CBulkFetching) &&
857 (tvcp->f.m.Length != statSeqNo)))
858 && (tvcp->execsOrWriters <= 0)
859 && !afs_DirtyPages(tvcp)
860 && !AFS_VN_MAPPED((vnode_t *) tvcp))
862 if (!(tvcp->f.states & CStatd)
863 && (!((tvcp->f.states & CBulkFetching) &&
864 (tvcp->f.m.Length != statSeqNo)))
865 && (tvcp->execsOrWriters <= 0)
866 && !afs_DirtyPages(tvcp))
870 /* this entry doesnt exist in the cache, and is not
871 * already being fetched by someone else, so add it to the
872 * list of file IDs to obtain.
874 * We detect a callback breaking race condition by checking the
875 * CBulkFetching state bit and the value in the file size.
876 * It is safe to set the status only if the CBulkFetching
877 * flag is still set and the value in the file size does
878 * not change. NewBulkVCache sets us up for the new ones.
879 * Set up the rest here.
881 * Don't fetch status for dirty files. We need to
882 * preserve the value of the file size. We could
883 * flush the pages, but it wouldn't be worthwhile.
885 if (!(tvcp->f.states & CBulkFetching)) {
886 tvcp->f.states |= CBulkFetching;
887 tvcp->f.m.Length = statSeqNo;
889 memcpy((char *)(fidsp + fidIndex), (char *)&tfid.Fid,
896 /* if dir vnode has non-zero entry */
897 /* move to the next dir entry by adding in the # of entries
898 * used by this dir entry.
900 temp = afs_dir_NameBlobs(dirEntryp->name) << 5;
901 DRelease(dirEntryp, 0);
905 } /* while loop over all dir entries */
907 /* now release the dir lock and prepare to make the bulk RPC */
908 ReleaseReadLock(&dcp->lock);
909 ReleaseReadLock(&adp->lock);
911 /* release the chunk */
914 /* dont make a null call */
919 /* setup the RPC parm structures */
920 fidParm.AFSCBFids_len = fidIndex;
921 fidParm.AFSCBFids_val = fidsp;
922 statParm.AFSBulkStats_len = fidIndex;
923 statParm.AFSBulkStats_val = statsp;
924 cbParm.AFSCBs_len = fidIndex;
925 cbParm.AFSCBs_val = cbsp;
927 /* start the timer; callback expirations are relative to this */
928 startTime = osi_Time();
930 tcp = afs_Conn(&adp->f.fid, areqp, SHARED_LOCK, &rxconn);
932 hostp = tcp->parent->srvr->server;
934 for (i = 0; i < fidIndex; i++) {
935 /* we must set tvcp->callback before the BulkStatus call, so
936 * we can detect concurrent InitCallBackState's */
938 afid.Cell = adp->f.fid.Cell;
939 afid.Fid.Volume = adp->f.fid.Fid.Volume;
940 afid.Fid.Vnode = fidsp[i].Vnode;
941 afid.Fid.Unique = fidsp[i].Unique;
945 ObtainReadLock(&afs_xvcache);
946 tvcp = afs_FindVCache(&afid, &retry, 0 /* !stats&!lru */);
947 ReleaseReadLock(&afs_xvcache);
948 } while (tvcp && retry);
954 if ((tvcp->f.states & CBulkFetching) &&
955 (tvcp->f.m.Length == statSeqNo)) {
956 tvcp->callback = hostp;
963 XSTATS_START_TIME(AFS_STATS_FS_RPCIDX_BULKSTATUS);
965 if (!(tcp->parent->srvr->server->flags & SNO_INLINEBULK)) {
969 RXAFS_InlineBulkStatus(rxconn, &fidParm, &statParm,
972 if (code == RXGEN_OPCODE) {
973 tcp->parent->srvr->server->flags |= SNO_INLINEBULK;
977 RXAFS_BulkStatus(rxconn, &fidParm, &statParm,
982 if (!code && ((&statsp[0])->errorCode)) {
984 * If this is an error needing retry, do so.
985 * Retryable errors are all whole-volume or
988 if (afs_Analyze(tcp, rxconn, (&statsp[0])->errorCode,
990 AFS_STATS_FS_RPCIDX_BULKSTATUS,
991 SHARED_LOCK, NULL) != 0)
999 RXAFS_BulkStatus(rxconn, &fidParm, &statParm, &cbParm,
1006 } while (afs_Analyze
1007 (tcp, rxconn, code, &adp->f.fid, areqp, AFS_STATS_FS_RPCIDX_BULKSTATUS,
1008 SHARED_LOCK, NULL));
1010 /* now, if we didnt get the info, bail out. */
1014 /* we need vol flags to create the entries properly */
1015 dotdot.Fid.Volume = 0;
1016 volp = afs_GetVolume(&adp->f.fid, areqp, READ_LOCK);
1018 volStates = volp->states;
1019 if (volp->dotdot.Fid.Volume != 0)
1020 dotdot = volp->dotdot;
1024 /* find the place to merge the info into We do this by skipping
1025 * nskip entries in the LRU queue. The more we skip, the more
1026 * we preserve, since the head of the VLRU queue is the most recently
1030 nskip = afs_cacheStats / 2; /* preserved fraction of the cache */
1031 ObtainReadLock(&afs_xvcache);
1032 #ifdef AFS_DARWIN80_ENV
1035 if (QEmpty(&VLRU)) {
1036 /* actually a serious error, probably should panic. Probably will
1037 * panic soon, oh well. */
1038 ReleaseReadLock(&afs_xvcache);
1039 afs_warnuser("afs_DoBulkStat: VLRU empty!");
1042 if ((VLRU.next->prev != &VLRU) || (VLRU.prev->next != &VLRU)) {
1043 refpanic("Bulkstat VLRU inconsistent");
1045 for (tq = VLRU.next; tq != &VLRU; tq = QNext(tq)) {
1047 #ifdef AFS_DARWIN80_ENV
1048 if ((!(QTOV(tq)->f.states & CDeadVnode)&&!(QTOV(tq)->f.states & CVInit)))
1052 if (QNext(QPrev(tq)) != tq) {
1054 refpanic("BulkStat VLRU inconsistent");
1060 lruvcp = QTOV(VLRU.next);
1062 /* now we have to hold this entry, so that it does not get moved
1063 * into the free list while we're running. It could still get
1064 * moved within the lru queue, but hopefully that will be rare; it
1065 * doesn't hurt nearly as much.
1068 #ifdef AFS_DARWIN80_ENV
1069 if (((lruvcp->f.states & CDeadVnode)||(lruvcp->f.states & CVInit))) {
1075 panic("Can't find non-dead vnode in VLRU\n");
1077 lruvp = AFSTOV(lruvcp);
1078 if (vnode_get(lruvp)) /* this bumps ref count */
1080 else if (vnode_ref(lruvp)) {
1082 /* AFSTOV(lruvcp) may be NULL */
1088 osi_vnhold(lruvcp, &retry);
1090 ReleaseReadLock(&afs_xvcache); /* could be read lock */
1094 /* otherwise, merge in the info. We have to be quite careful here,
1095 * since we need to ensure that we don't merge old info over newer
1096 * stuff in a stat cache entry. We're very conservative here: we don't
1097 * do the merge at all unless we ourselves create the stat cache
1098 * entry. That's pretty safe, and should work pretty well, since we
1099 * typically expect to do the stat cache creation ourselves.
1101 * We also have to take into account racing token revocations.
1103 for (i = 0; i < fidIndex; i++) {
1104 if ((&statsp[i])->errorCode)
1106 afid.Cell = adp->f.fid.Cell;
1107 afid.Fid.Volume = adp->f.fid.Fid.Volume;
1108 afid.Fid.Vnode = fidsp[i].Vnode;
1109 afid.Fid.Unique = fidsp[i].Unique;
1112 ObtainReadLock(&afs_xvcache);
1113 tvcp = afs_FindVCache(&afid, &retry, FIND_CDEAD /* !stats&!lru */);
1114 ReleaseReadLock(&afs_xvcache);
1115 } while (tvcp && retry);
1117 /* The entry may no longer exist */
1122 /* now we have the entry held, but we need to fill it in */
1123 ObtainWriteLock(&tvcp->lock, 131);
1125 /* if CBulkFetching is not set, or if the file size no longer
1126 * matches the value we placed there when we set the CBulkFetching
1127 * flag, then someone else has done something with this node,
1128 * and we may not have the latest status information for this
1129 * file. Leave the entry alone.
1131 if (!(tvcp->f.states & CBulkFetching) || (tvcp->f.m.Length != statSeqNo)) {
1133 ReleaseWriteLock(&tvcp->lock);
1134 afs_PutVCache(tvcp);
1138 /* now copy ".." entry back out of volume structure, if necessary */
1139 if (tvcp->mvstat == 2 && (dotdot.Fid.Volume != 0)) {
1141 tvcp->mvid = (struct VenusFid *)
1142 osi_AllocSmallSpace(sizeof(struct VenusFid));
1143 *tvcp->mvid = dotdot;
1146 #ifdef AFS_DARWIN80_ENV
1147 if (((lruvcp->f.states & CDeadVnode)||(lruvcp->f.states & CVInit)))
1148 panic("vlru control point went dead\n");
1151 ObtainWriteLock(&afs_xvcache, 132);
1152 if ((VLRU.next->prev != &VLRU) || (VLRU.prev->next != &VLRU)) {
1153 refpanic("Bulkstat VLRU inconsistent2");
1155 if ((QNext(QPrev(&tvcp->vlruq)) != &tvcp->vlruq)
1156 || (QPrev(QNext(&tvcp->vlruq)) != &tvcp->vlruq)) {
1157 refpanic("Bulkstat VLRU inconsistent4");
1159 if ((QNext(QPrev(&lruvcp->vlruq)) != &lruvcp->vlruq)
1160 || (QPrev(QNext(&lruvcp->vlruq)) != &lruvcp->vlruq)) {
1161 refpanic("Bulkstat VLRU inconsistent5");
1164 if (tvcp != lruvcp) { /* if they are == don't move it, don't corrupt vlru */
1165 QRemove(&tvcp->vlruq);
1166 QAdd(&lruvcp->vlruq, &tvcp->vlruq);
1169 if ((VLRU.next->prev != &VLRU) || (VLRU.prev->next != &VLRU)) {
1170 refpanic("Bulkstat VLRU inconsistent3");
1172 if ((QNext(QPrev(&tvcp->vlruq)) != &tvcp->vlruq)
1173 || (QPrev(QNext(&tvcp->vlruq)) != &tvcp->vlruq)) {
1174 refpanic("Bulkstat VLRU inconsistent5");
1176 if ((QNext(QPrev(&lruvcp->vlruq)) != &lruvcp->vlruq)
1177 || (QPrev(QNext(&lruvcp->vlruq)) != &lruvcp->vlruq)) {
1178 refpanic("Bulkstat VLRU inconsistent6");
1180 ReleaseWriteLock(&afs_xvcache);
1182 ObtainWriteLock(&afs_xcbhash, 494);
1184 /* We need to check the flags again. We may have missed
1185 * something while we were waiting for a lock.
1187 if (!(tvcp->f.states & CBulkFetching) || (tvcp->f.m.Length != statSeqNo)) {
1189 ReleaseWriteLock(&tvcp->lock);
1190 ReleaseWriteLock(&afs_xcbhash);
1191 afs_PutVCache(tvcp);
1195 /* now merge in the resulting status back into the vnode.
1196 * We only do this if the entry looks clear.
1198 afs_ProcessFS(tvcp, &statsp[i], areqp);
1199 #if defined(AFS_LINUX22_ENV)
1200 afs_fill_inode(AFSTOV(tvcp), NULL); /* reset inode operations */
1203 /* do some accounting for bulk stats: mark this entry as
1204 * loaded, so we can tell if we use it before it gets
1207 tvcp->f.states |= CBulkStat;
1208 tvcp->f.states &= ~CBulkFetching;
1210 afs_bulkStatsDone++;
1212 /* merge in vol info */
1213 if (volStates & VRO)
1214 tvcp->f.states |= CRO;
1215 if (volStates & VBackup)
1216 tvcp->f.states |= CBackup;
1217 if (volStates & VForeign)
1218 tvcp->f.states |= CForeign;
1220 /* merge in the callback info */
1221 tvcp->f.states |= CTruth;
1223 /* get ptr to the callback we are interested in */
1226 if (tcbp->ExpirationTime != 0) {
1227 tvcp->cbExpires = tcbp->ExpirationTime + startTime;
1228 tvcp->callback = hostp;
1229 tvcp->f.states |= CStatd;
1230 afs_QueueCallback(tvcp, CBHash(tcbp->ExpirationTime), volp);
1231 } else if (tvcp->f.states & CRO) {
1232 /* ordinary callback on a read-only volume -- AFS 3.2 style */
1233 tvcp->cbExpires = 3600 + startTime;
1234 tvcp->callback = hostp;
1235 tvcp->f.states |= CStatd;
1236 afs_QueueCallback(tvcp, CBHash(3600), volp);
1239 tvcp->f.states &= ~(CStatd | CUnique);
1240 afs_DequeueCallback(tvcp);
1241 if ((tvcp->f.states & CForeign) || (vType(tvcp) == VDIR))
1242 osi_dnlc_purgedp(tvcp); /* if it (could be) a directory */
1244 #ifdef AFS_DARWIN80_ENV
1245 /* reclaim->FlushVCache will need xcbhash */
1246 if (((tvcp->f.states & CDeadVnode)||(tvcp->f.states & CVInit))) {
1247 ReleaseWriteLock(&afs_xcbhash);
1248 /* passing in a parent hangs getting the vnode lock */
1249 code = afs_darwin_finalizevnode(tvcp, NULL, NULL, 0, 1);
1251 /* It's gonna get recycled - shouldn't happen */
1253 tvcp->f.states &= ~(CStatd | CUnique);
1254 afs_DequeueCallback(tvcp);
1255 if ((tvcp->f.states & CForeign) || (vType(tvcp) == VDIR))
1256 osi_dnlc_purgedp(tvcp); /* if it (could be) a directory */
1258 /* re-acquire the usecount that finalizevnode disposed of */
1259 vnode_ref(AFSTOV(tvcp));
1263 ReleaseWriteLock(&afs_xcbhash);
1265 ReleaseWriteLock(&tvcp->lock);
1266 /* finally, we're done with the entry */
1267 afs_PutVCache(tvcp);
1268 } /* for all files we got back */
1270 /* finally return the pointer into the LRU queue */
1271 #ifdef AFS_DARWIN80_ENV
1272 if (((lruvcp->f.states & CDeadVnode)||(lruvcp->f.states & CVInit)))
1273 panic("vlru control point went dead before put\n");
1279 afs_PutVCache(lruvcp);
1283 /* Be sure to turn off the CBulkFetching flags */
1284 for (i = flagIndex; i < fidIndex; i++) {
1285 afid.Cell = adp->f.fid.Cell;
1286 afid.Fid.Volume = adp->f.fid.Fid.Volume;
1287 afid.Fid.Vnode = fidsp[i].Vnode;
1288 afid.Fid.Unique = fidsp[i].Unique;
1291 ObtainReadLock(&afs_xvcache);
1292 tvcp = afs_FindVCache(&afid, &retry, FIND_CDEAD /* !stats&!lru */);
1293 ReleaseReadLock(&afs_xvcache);
1294 } while (tvcp && retry);
1296 if ((tvcp->f.states & CBulkFetching)
1297 && (tvcp->f.m.Length == statSeqNo)) {
1298 tvcp->f.states &= ~CBulkFetching;
1300 afs_PutVCache(tvcp);
1304 afs_PutVolume(volp, READ_LOCK);
1306 /* If we did the InlineBulk RPC pull out the return code */
1307 if (inlinebulk && code == 0) {
1308 if ((&statsp[0])->errorCode) {
1309 afs_Analyze(tcp, rxconn, (&statsp[0])->errorCode, &adp->f.fid, areqp,
1310 AFS_STATS_FS_RPCIDX_BULKSTATUS, SHARED_LOCK, NULL);
1311 code = (&statsp[0])->errorCode;
1317 osi_FreeLargeSpace((char *)fidsp);
1318 osi_Free((char *)statsp, AFSCBMAX * sizeof(AFSFetchStatus));
1319 osi_Free((char *)cbsp, AFSCBMAX * sizeof(AFSCallBack));
1323 /* was: (AFS_DEC_ENV) || defined(AFS_OSF30_ENV) || defined(AFS_NCR_ENV) */
1324 static int AFSDOBULK = 1;
1327 osi_lookup_isdot(const char *aname)
1331 /* in Solaris, we can get passed "" as a path component if we are the
1332 * root directory, e.g. after a call to chroot. It is equivalent to
1336 #endif /* AFS_SUN5_ENV */
1337 if (aname[0] == '.' && !aname[1]) {
1344 #if defined(AFS_SUN5_ENV) || defined(AFS_SGI_ENV)
1345 afs_lookup(OSI_VC_DECL(adp), char *aname, struct vcache **avcp, struct pathname *pnp, int flags, struct vnode *rdir, afs_ucred_t *acred)
1346 #elif defined(UKERNEL)
1347 afs_lookup(OSI_VC_DECL(adp), char *aname, struct vcache **avcp, afs_ucred_t *acred, int flags)
1349 afs_lookup(OSI_VC_DECL(adp), char *aname, struct vcache **avcp, afs_ucred_t *acred)
1352 struct vrequest treq;
1354 struct vcache *tvc = 0;
1356 afs_int32 bulkcode = 0;
1357 int pass = 0, hit = 0;
1358 int force_eval = afs_fakestat_enable ? 0 : 1;
1360 extern afs_int32 afs_mariner; /*Writing activity to log? */
1361 afs_hyper_t versionNo;
1362 int no_read_access = 0;
1363 struct sysname_info sysState; /* used only for @sys checking */
1364 int dynrootRetry = 1;
1365 struct afs_fakestat_state fakestate;
1366 int tryEvalOnly = 0;
1367 OSI_VC_CONVERT(adp);
1369 AFS_STATCNT(afs_lookup);
1370 afs_InitFakeStat(&fakestate);
1374 if ((code = afs_InitReq(&treq, acred)))
1377 if (afs_fakestat_enable && adp->mvstat == 1) {
1378 if (strcmp(aname, ".directory") == 0)
1382 #if defined(AFS_DARWIN_ENV)
1383 /* Workaround for MacOSX Finder, which tries to look for
1384 * .DS_Store and Contents under every directory.
1386 if (afs_fakestat_enable && adp->mvstat == 1) {
1387 if (strcmp(aname, ".DS_Store") == 0)
1389 if (strcmp(aname, "Contents") == 0)
1392 if (afs_fakestat_enable && adp->mvstat == 2) {
1393 if (strncmp(aname, "._", 2) == 0)
1399 code = afs_TryEvalFakeStat(&adp, &fakestate, &treq);
1401 code = afs_EvalFakeStat(&adp, &fakestate, &treq);
1403 /*printf("Code is %d\n", code);*/
1405 if (tryEvalOnly && adp->mvstat == 1)
1410 *avcp = NULL; /* Since some callers don't initialize it */
1412 /* come back to here if we encounter a non-existent object in a read-only
1413 * volume's directory */
1416 *avcp = NULL; /* Since some callers don't initialize it */
1419 if (!(adp->f.states & CStatd) && !afs_InReadDir(adp)) {
1420 if ((code = afs_VerifyVCache2(adp, &treq))) {
1426 /* watch for ".." in a volume root */
1427 if (adp->mvstat == 2 && aname[0] == '.' && aname[1] == '.' && !aname[2]) {
1428 /* looking up ".." in root via special hacks */
1429 if (adp->mvid == (struct VenusFid *)0 || adp->mvid->Fid.Volume == 0) {
1433 /* otherwise we have the fid here, so we use it */
1434 /*printf("Getting vcache\n");*/
1435 tvc = afs_GetVCache(adp->mvid, &treq, NULL, NULL);
1436 afs_Trace3(afs_iclSetp, CM_TRACE_GETVCDOTDOT, ICL_TYPE_FID, adp->mvid,
1437 ICL_TYPE_POINTER, tvc, ICL_TYPE_INT32, code);
1439 code = (tvc ? 0 : ENOENT);
1441 if (tvc && !VREFCOUNT_GT(tvc, 0)) {
1445 /*printf("LOOKUP GETVCDOTDOT -> %d\n", code); */
1450 /* now check the access */
1451 if (treq.uid != adp->last_looker) {
1452 if (!afs_AccessOK(adp, PRSFS_LOOKUP, &treq, CHECK_MODE_BITS)) {
1457 adp->last_looker = treq.uid;
1460 /* Check for read access as well. We need read access in order to
1461 * stat files, but not to stat subdirectories. */
1462 if (!afs_AccessOK(adp, PRSFS_READ, &treq, CHECK_MODE_BITS))
1465 /* special case lookup of ".". Can we check for it sooner in this code,
1466 * for instance, way up before "redo:" ??
1467 * I'm not fiddling with the LRUQ here, either, perhaps I should, or else
1468 * invent a lightweight version of GetVCache.
1470 if (osi_lookup_isdot(aname)) { /* special case */
1471 ObtainReadLock(&afs_xvcache);
1473 ReleaseReadLock(&afs_xvcache);
1474 #ifdef AFS_DARWIN80_ENV
1475 vnode_get(AFSTOV(adp));
1480 if (adp && !VREFCOUNT_GT(adp, 0)) {
1487 * Special case lookup of ".." in the dynamic mount directory.
1488 * The parent of this directory is _always_ the AFS root volume.
1490 if (afs_IsDynrootMount(adp) &&
1491 aname[0] == '.' && aname[1] == '.' && !aname[2]) {
1493 ObtainReadLock(&afs_xvcache);
1494 osi_vnhold(afs_globalVp, 0);
1495 ReleaseReadLock(&afs_xvcache);
1496 #ifdef AFS_DARWIN80_ENV
1497 vnode_get(AFSTOV(afs_globalVp));
1500 *avcp = tvc = afs_globalVp;
1506 * Special case lookups in the dynamic mount directory.
1507 * The names here take the form cell:volume, similar to a mount point.
1508 * EvalMountData parses that and returns a cell and volume ID, which
1509 * we use to construct the appropriate dynroot Fid.
1511 if (afs_IsDynrootMount(adp)) {
1512 struct VenusFid tfid;
1513 afs_uint32 cellidx, volid, vnoid, uniq;
1515 code = EvalMountData('%', aname, 0, 0, NULL, &treq, &cellidx, &volid, &vnoid, &uniq);
1518 /* If a vnode was returned, it's not a real mount point */
1520 struct cell *tcell = afs_GetCellByIndex(cellidx, READ_LOCK);
1521 tfid.Cell = tcell->cellNum;
1522 afs_PutCell(tcell, READ_LOCK);
1523 tfid.Fid.Vnode = vnoid;
1524 tfid.Fid.Volume = volid;
1525 tfid.Fid.Unique = uniq;
1527 afs_GetDynrootMountFid(&tfid);
1528 tfid.Fid.Vnode = VNUM_FROM_TYPEID(VN_TYPE_MOUNT, cellidx << 2);
1529 tfid.Fid.Unique = volid;
1531 *avcp = tvc = afs_GetVCache(&tfid, &treq, NULL, NULL);
1532 code = (tvc ? 0 : ENOENT);
1537 #ifdef AFS_LINUX26_ENV
1539 * Special case of the dynamic mount volume in a static root.
1540 * This is really unfortunate, but we need this for the translator.
1542 if (adp == afs_globalVp && !afs_GetDynrootEnable() &&
1543 !strcmp(aname, AFS_DYNROOT_MOUNTNAME)) {
1544 struct VenusFid tfid;
1546 afs_GetDynrootMountFid(&tfid);
1547 *avcp = tvc = afs_GetVCache(&tfid, &treq, NULL, NULL);
1554 Check_AtSys(adp, aname, &sysState, &treq);
1555 tname = sysState.name;
1557 /* 1st Check_AtSys and lookup by tname is required here, for now,
1558 * because the dnlc is *not* told to remove entries for the parent
1559 * dir of file/dir op that afs_LocalHero likes, but dnlc is informed
1560 * if the cached entry for the parent dir is invalidated for a
1562 * Otherwise, we'd be able to do a dnlc lookup on an entry ending
1563 * w/@sys and know the dnlc was consistent with reality. */
1564 tvc = osi_dnlc_lookup(adp, tname, WRITE_LOCK);
1565 *avcp = tvc; /* maybe wasn't initialized, but it is now */
1567 if (no_read_access && vType(tvc) != VDIR && vType(tvc) != VLNK) {
1568 /* need read access on dir to stat non-directory / non-link */
1574 #ifdef AFS_LINUX22_ENV
1575 if (tvc->mvstat == 2) { /* we don't trust the dnlc for root vcaches */
1576 AFS_RELE(AFSTOV(tvc));
1583 #else /* non - LINUX */
1587 #endif /* linux22 */
1590 { /* sub-block just to reduce stack usage */
1592 afs_size_t dirOffset, dirLen;
1593 struct VenusFid tfid;
1595 /* now we have to lookup the next fid */
1596 if (afs_InReadDir(adp))
1597 tdc = adp->dcreaddir;
1599 tdc = afs_GetDCache(adp, (afs_size_t) 0, &treq,
1600 &dirOffset, &dirLen, 1);
1602 *avcp = NULL; /* redundant, but harmless */
1607 /* now we will just call dir package with appropriate inode.
1608 * Dirs are always fetched in their entirety for now */
1609 ObtainReadLock(&adp->lock);
1610 ObtainReadLock(&tdc->lock);
1613 * Make sure that the data in the cache is current. There are two
1614 * cases we need to worry about:
1615 * 1. The cache data is being fetched by another process.
1616 * 2. The cache data is no longer valid
1618 * If a readdir is in progress _in this thread_, it has a shared
1619 * lock on the vcache and has obtained current data, so we just
1620 * use that. This eliminates several possible deadlocks.
1622 if (!afs_InReadDir(adp)) {
1623 while ((adp->f.states & CStatd)
1624 && (tdc->dflags & DFFetching)
1625 && hsame(adp->f.m.DataVersion, tdc->f.versionNo)) {
1626 ReleaseReadLock(&tdc->lock);
1627 ReleaseReadLock(&adp->lock);
1628 afs_osi_Sleep(&tdc->validPos);
1629 ObtainReadLock(&adp->lock);
1630 ObtainReadLock(&tdc->lock);
1632 if (!(adp->f.states & CStatd)
1633 || !hsame(adp->f.m.DataVersion, tdc->f.versionNo)) {
1634 ReleaseReadLock(&tdc->lock);
1635 ReleaseReadLock(&adp->lock);
1637 if (tname && tname != aname)
1638 osi_FreeLargeSpace(tname);
1643 /* Save the version number for when we call osi_dnlc_enter */
1644 hset(versionNo, tdc->f.versionNo);
1647 * check for, and handle "@sys" if it's there. We should be able
1648 * to avoid the alloc and the strcpy with a little work, but it's
1649 * not pressing. If there aren't any remote users (ie, via the
1650 * NFS translator), we have a slightly easier job.
1651 * the faster way to do this is to check for *aname == '@' and if
1652 * it's there, check for @sys, otherwise, assume there's no @sys
1653 * then, if the lookup fails, check for .*@sys...
1655 /* above now implemented by Check_AtSys and Next_AtSys */
1657 /* lookup the name in the appropriate dir, and return a cache entry
1658 * on the resulting fid */
1660 afs_dir_LookupOffset(tdc, sysState.name, &tfid.Fid,
1663 /* If the first lookup doesn't succeed, maybe it's got @sys in the name */
1664 while (code == ENOENT && Next_AtSys(adp, &treq, &sysState))
1666 afs_dir_LookupOffset(tdc, sysState.name, &tfid.Fid,
1668 tname = sysState.name;
1670 ReleaseReadLock(&tdc->lock);
1671 if (!afs_InReadDir(adp))
1673 if (code == ENOENT && afs_IsDynroot(adp) && dynrootRetry && !tryEvalOnly) {
1675 char *cn = (tname[0] == '.') ? tname + 1 : tname;
1676 ReleaseReadLock(&adp->lock);
1677 /* confirm it's not just hushed */
1678 tc = afs_GetCellByName(cn, WRITE_LOCK);
1680 if (tc->states & CHush) {
1681 tc->states &= ~CHush;
1682 ReleaseWriteLock(&tc->lock);
1683 afs_DynrootInvalidate();
1686 ReleaseWriteLock(&tc->lock);
1688 /* Allow a second dynroot retry if the cell was hushed before */
1690 if (tname[0] == '.')
1691 afs_LookupAFSDB(tname + 1);
1693 afs_LookupAFSDB(tname);
1694 if (tname && tname != aname)
1695 osi_FreeLargeSpace(tname);
1698 ReleaseReadLock(&adp->lock);
1701 /* new fid has same cell and volume */
1702 tfid.Cell = adp->f.fid.Cell;
1703 tfid.Fid.Volume = adp->f.fid.Fid.Volume;
1704 afs_Trace4(afs_iclSetp, CM_TRACE_LOOKUP, ICL_TYPE_POINTER, adp,
1705 ICL_TYPE_STRING, tname, ICL_TYPE_FID, &tfid,
1706 ICL_TYPE_INT32, code);
1709 if (code != ENOENT) {
1710 /*printf("LOOKUP dirLookupOff -> %d\n", code);*/
1715 /* prefetch some entries, if the dir is currently open. The variable
1716 * dirCookie tells us where to start prefetching from.
1718 if (!AFS_IS_DISCONNECTED &&
1719 AFSDOBULK && adp->opens > 0 && !(adp->f.states & CForeign)
1720 && !afs_IsDynroot(adp) && !afs_InReadDir(adp)) {
1722 /* if the entry is not in the cache, or is in the cache,
1723 * but hasn't been statd, then do a bulk stat operation.
1727 ObtainReadLock(&afs_xvcache);
1728 tvc = afs_FindVCache(&tfid, &retry, 0 /* !stats,!lru */ );
1729 ReleaseReadLock(&afs_xvcache);
1730 } while (tvc && retry);
1732 if (!tvc || !(tvc->f.states & CStatd))
1733 bulkcode = afs_DoBulkStat(adp, dirCookie, &treq);
1737 /* if the vcache isn't usable, release it */
1738 if (tvc && !(tvc->f.states & CStatd)) {
1747 /* now get the status info, if we don't already have it */
1748 /* This is kind of weird, but we might wind up accidentally calling
1749 * RXAFS_Lookup because we happened upon a file which legitimately
1750 * has a 0 uniquifier. That is the result of allowing unique to wrap
1751 * to 0. This was fixed in AFS 3.4. For CForeign, Unique == 0 means that
1752 * the file has not yet been looked up.
1755 afs_int32 cached = 0;
1756 if (!tfid.Fid.Unique && (adp->f.states & CForeign)) {
1757 tvc = afs_LookupVCache(&tfid, &treq, &cached, adp, tname);
1759 if (!tvc && !bulkcode) { /* lookup failed or wasn't called */
1760 tvc = afs_GetVCache(&tfid, &treq, &cached, NULL);
1763 } /* sub-block just to reduce stack usage */
1766 if (adp->f.states & CForeign)
1767 tvc->f.states |= CForeign;
1768 tvc->f.parent.vnode = adp->f.fid.Fid.Vnode;
1769 tvc->f.parent.unique = adp->f.fid.Fid.Unique;
1770 tvc->f.states &= ~CBulkStat;
1772 if (afs_fakestat_enable == 2 && tvc->mvstat == 1) {
1773 ObtainSharedLock(&tvc->lock, 680);
1774 if (!tvc->linkData) {
1775 UpgradeSToWLock(&tvc->lock, 681);
1776 code = afs_HandleLink(tvc, &treq);
1777 ConvertWToRLock(&tvc->lock);
1779 ConvertSToRLock(&tvc->lock);
1782 if (!code && !afs_strchr(tvc->linkData, ':'))
1784 ReleaseReadLock(&tvc->lock);
1786 if (tvc->mvstat == 1 && (tvc->f.states & CMValid) && tvc->mvid != NULL)
1787 force_eval = 1; /* This is now almost for free, get it correct */
1789 #if defined(UKERNEL) && defined(AFS_WEB_ENHANCEMENTS)
1790 if (!(flags & AFS_LOOKUP_NOEVAL))
1791 /* don't eval mount points */
1792 #endif /* UKERNEL && AFS_WEB_ENHANCEMENTS */
1793 if (tvc->mvstat == 1 && force_eval) {
1794 /* a mt point, possibly unevaluated */
1795 struct volume *tvolp;
1797 ObtainWriteLock(&tvc->lock, 133);
1798 code = EvalMountPoint(tvc, adp, &tvolp, &treq);
1799 ReleaseWriteLock(&tvc->lock);
1804 afs_PutVolume(tvolp, WRITE_LOCK);
1808 /* next, we want to continue using the target of the mt point */
1809 if (tvc->mvid && (tvc->f.states & CMValid)) {
1811 /* now lookup target, to set .. pointer */
1812 afs_Trace2(afs_iclSetp, CM_TRACE_LOOKUP1,
1813 ICL_TYPE_POINTER, tvc, ICL_TYPE_FID,
1815 uvc = tvc; /* remember for later */
1817 if (tvolp && (tvolp->states & VForeign)) {
1818 /* XXXX tvolp has ref cnt on but not locked! XXX */
1820 afs_GetRootVCache(tvc->mvid, &treq, NULL, tvolp);
1822 tvc = afs_GetVCache(tvc->mvid, &treq, NULL, NULL);
1824 afs_PutVCache(uvc); /* we're done with it */
1829 afs_PutVolume(tvolp, WRITE_LOCK);
1834 /* now, if we came via a new mt pt (say because of a new
1835 * release of a R/O volume), we must reevaluate the ..
1836 * ptr to point back to the appropriate place */
1838 ObtainWriteLock(&tvc->lock, 134);
1839 if (tvc->mvid == NULL) {
1840 tvc->mvid = (struct VenusFid *)
1841 osi_AllocSmallSpace(sizeof(struct VenusFid));
1843 /* setup backpointer */
1844 *tvc->mvid = tvolp->dotdot;
1845 ReleaseWriteLock(&tvc->lock);
1846 afs_PutVolume(tvolp, WRITE_LOCK);
1852 afs_PutVolume(tvolp, WRITE_LOCK);
1857 if (tvc && !VREFCOUNT_GT(tvc, 0)) {
1862 /* if we get here, we found something in a directory that couldn't
1863 * be located (a Multics "connection failure"). If the volume is
1864 * read-only, we try flushing this entry from the cache and trying
1866 if (!AFS_IS_DISCONNECTED) {
1869 tv = afs_GetVolume(&adp->f.fid, &treq, READ_LOCK);
1871 if (tv->states & VRO) {
1872 pass = 1; /* try this *once* */
1873 ObtainWriteLock(&afs_xcbhash, 495);
1874 afs_DequeueCallback(adp);
1875 /* re-stat to get later version */
1876 adp->f.states &= ~CStatd;
1877 ReleaseWriteLock(&afs_xcbhash);
1878 osi_dnlc_purgedp(adp);
1879 afs_PutVolume(tv, READ_LOCK);
1882 afs_PutVolume(tv, READ_LOCK);
1892 /* put the network buffer back, if need be */
1893 if (tname != aname && tname)
1894 osi_FreeLargeSpace(tname);
1898 afs_AddMarinerName(aname, tvc);
1900 #if defined(UKERNEL) && defined(AFS_WEB_ENHANCEMENTS)
1901 if (!(flags & AFS_LOOKUP_NOEVAL)) {
1902 /* Here we don't enter the name into the DNLC because we want the
1903 * evaluated mount dir to be there (the vcache for the mounted
1904 * volume) rather than the vc of the mount point itself. We can
1905 * still find the mount point's vc in the vcache by its fid. */
1906 #endif /* UKERNEL && AFS_WEB_ENHANCEMENTS */
1907 if (!hit && force_eval) {
1908 osi_dnlc_enter(adp, aname, tvc, &versionNo);
1910 #ifdef AFS_LINUX20_ENV
1911 /* So Linux inode cache is up to date. */
1912 code = afs_VerifyVCache(tvc, &treq);
1914 afs_PutFakeStat(&fakestate);
1915 AFS_DISCON_UNLOCK();
1916 return 0; /* can't have been any errors if hit and !code */
1919 #if defined(UKERNEL) && defined(AFS_WEB_ENHANCEMENTS)
1926 code = afs_CheckCode(code, &treq, 19);
1928 /* If there is an error, make sure *avcp is null.
1929 * Alphas panic otherwise - defect 10719.
1934 afs_PutFakeStat(&fakestate);
1935 AFS_DISCON_UNLOCK();