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.target_root 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 */
70 afs_uint32 mtptCellnum;
73 if ((afs_strtoi_r(data, &endptr, &mtptCellnum) == 0) &&
75 tcell = afs_GetCell(mtptCellnum, READ_LOCK);
77 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 = 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.target_root && (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;
302 if (avc->mvid.target_root == NULL)
303 avc->mvid.target_root = osi_AllocSmallSpace(sizeof(struct VenusFid));
304 avc->mvid.target_root->Cell = (*avolpp)->cell;
305 avc->mvid.target_root->Fid.Volume = (*avolpp)->volume;
306 avc->mvid.target_root->Fid.Vnode = avnoid;
307 avc->mvid.target_root->Fid.Unique = auniq;
308 avc->f.states |= CMValid;
310 /* Used to: if the mount point is stored within a backup volume,
311 * then we should only update the parent pointer information if
312 * there's none already set, so as to avoid updating a volume's ..
313 * info with something in an OldFiles directory.
315 * Next two lines used to be under this if:
317 * if (!(avc->f.states & CBackup) || tvp->dotdot.Fid.Volume == 0)
319 * Now: update mount point back pointer on every call, so that we handle
320 * multiple mount points better. This way, when du tries to go back
321 * via chddir(".."), it will end up exactly where it started, yet
322 * cd'ing via a new path to a volume will reset the ".." pointer
325 (*avolpp)->mtpoint = avc->f.fid; /* setup back pointer to mtpoint */
328 (*avolpp)->dotdot = advc->f.fid;
336 * Must be called on an afs_fakestat_state object before calling
337 * afs_EvalFakeStat or afs_PutFakeStat. Calling afs_PutFakeStat
338 * without calling afs_EvalFakeStat is legal, as long as this
339 * function is called.
342 afs_InitFakeStat(struct afs_fakestat_state *state)
344 if (!afs_fakestat_enable)
349 state->need_release = 0;
353 * afs_EvalFakeStat_int
355 * The actual implementation of afs_EvalFakeStat and afs_TryEvalFakeStat,
356 * which is called by those wrapper functions.
358 * Only issues RPCs if canblock is non-zero.
361 afs_EvalFakeStat_int(struct vcache **avcp, struct afs_fakestat_state *state,
362 struct vrequest *areq, int canblock)
364 struct vcache *tvc, *root_vp;
365 struct volume *tvolp = NULL;
368 if (!afs_fakestat_enable)
371 osi_Assert(state->valid == 1);
372 osi_Assert(state->did_eval == 0);
376 if (tvc->mvstat != AFS_MVSTAT_MTPT)
380 /* Is the call to VerifyVCache really necessary? */
381 code = afs_VerifyVCache(tvc, areq);
385 ObtainWriteLock(&tvc->lock, 599);
386 code = EvalMountPoint(tvc, NULL, &tvolp, areq);
387 ReleaseWriteLock(&tvc->lock);
391 tvolp->dotdot = tvc->f.fid;
392 tvolp->dotdot.Fid.Vnode = tvc->f.parent.vnode;
393 tvolp->dotdot.Fid.Unique = tvc->f.parent.unique;
396 if (tvc->mvid.target_root && (tvc->f.states & CMValid)) {
402 ObtainWriteLock(&afs_xvcache, 597);
403 root_vp = afs_FindVCache(tvc->mvid.target_root, &retry, IS_WLOCK);
404 if (root_vp && retry) {
405 ReleaseWriteLock(&afs_xvcache);
406 afs_PutVCache(root_vp);
408 } while (root_vp && retry);
409 ReleaseWriteLock(&afs_xvcache);
411 root_vp = afs_GetVCache(tvc->mvid.target_root, areq, NULL, NULL);
414 code = canblock ? EIO : 0;
417 #ifdef AFS_DARWIN80_ENV
418 root_vp->f.m.Type = VDIR;
420 code = afs_darwin_finalizevnode(root_vp, NULL, NULL, 0, 0);
423 vnode_ref(AFSTOV(root_vp));
425 if (tvolp && !afs_InReadDir(root_vp)) {
426 /* Is this always kosher? Perhaps we should instead use
427 * NBObtainWriteLock to avoid potential deadlock.
429 ObtainWriteLock(&root_vp->lock, 598);
430 if (!root_vp->mvid.parent)
431 root_vp->mvid.parent = osi_AllocSmallSpace(sizeof(struct VenusFid));
432 *root_vp->mvid.parent = tvolp->dotdot;
433 ReleaseWriteLock(&root_vp->lock);
435 state->need_release = 1;
436 state->root_vp = root_vp;
440 code = canblock ? EIO : 0;
445 afs_PutVolume(tvolp, WRITE_LOCK);
452 * Automatically does the equivalent of EvalMountPoint for vcache entries
453 * which are mount points. Remembers enough state to properly release
454 * the volume root vcache when afs_PutFakeStat() is called.
456 * State variable must be initialized by afs_InitFakeState() beforehand.
458 * Returns 0 when everything succeeds and *avcp points to the vcache entry
459 * that should be used for the real vnode operation. Returns non-zero if
460 * something goes wrong and the error code should be returned to the user.
463 afs_EvalFakeStat(struct vcache **avcp, struct afs_fakestat_state *state,
464 struct vrequest *areq)
466 return afs_EvalFakeStat_int(avcp, state, areq, 1);
470 * afs_TryEvalFakeStat
472 * Same as afs_EvalFakeStat, but tries not to talk to remote servers
473 * and only evaluate the mount point if all the data is already in
476 * Returns 0 if everything succeeds and *avcp points to a valid
477 * vcache entry (possibly evaluated).
480 afs_TryEvalFakeStat(struct vcache **avcp, struct afs_fakestat_state *state,
481 struct vrequest *areq)
483 return afs_EvalFakeStat_int(avcp, state, areq, 0);
489 * Perform any necessary cleanup at the end of a vnode op, given that
490 * afs_InitFakeStat was previously called with this state.
493 afs_PutFakeStat(struct afs_fakestat_state *state)
495 if (!afs_fakestat_enable)
498 osi_Assert(state->valid == 1);
499 if (state->need_release)
500 afs_PutVCache(state->root_vp);
505 afs_ENameOK(char *aname)
509 AFS_STATCNT(ENameOK);
510 tlen = strlen(aname);
511 if (tlen >= 4 && strcmp(aname + tlen - 4, "@sys") == 0)
517 afs_getsysname(struct vrequest *areq, struct vcache *adp,
518 char *bufp, int *num, char **sysnamelist[])
523 AFS_STATCNT(getsysname);
525 *sysnamelist = afs_sysnamelist;
527 if (!afs_nfsexporter)
528 strcpy(bufp, (*sysnamelist)[0]);
530 au = afs_GetUser(areq->uid, adp->f.fid.Cell, READ_LOCK);
532 error = EXP_SYSNAME(au->exporter, (char *)0, sysnamelist, num, 0);
534 strcpy(bufp, "@sys");
535 afs_PutUser(au, READ_LOCK);
538 strcpy(bufp, (*sysnamelist)[0]);
541 strcpy(bufp, afs_sysname);
542 afs_PutUser(au, READ_LOCK);
548 Check_AtSys(struct vcache *avc, const char *aname,
549 struct sysname_info *state, struct vrequest *areq)
552 char **sysnamelist[MAXNUMSYSNAMES];
554 if (AFS_EQ_ATSYS(aname)) {
556 state->name = osi_AllocLargeSpace(MAXSYSNAME);
559 afs_getsysname(areq, avc, state->name, &num, sysnamelist);
564 state->name = (char *)aname;
569 Next_AtSys(struct vcache *avc, struct vrequest *areq,
570 struct sysname_info *state)
572 int num = afs_sysnamecount;
573 char **sysnamelist[MAXNUMSYSNAMES];
575 if (state->index == -1)
576 return 0; /* No list */
578 /* Check for the initial state of aname != "@sys" in Check_AtSys */
579 if (state->offset == -1 && state->allocked == 0) {
582 /* Check for .*@sys */
583 for (tname = state->name; *tname; tname++)
584 /*Move to the end of the string */ ;
586 if ((tname > state->name + 4) && (AFS_EQ_ATSYS(tname - 4))) {
587 state->offset = (tname - 4) - state->name;
588 tname = osi_AllocLargeSpace(AFS_LRALLOCSIZ);
589 strncpy(tname, state->name, state->offset);
594 afs_getsysname(areq, avc, state->name + state->offset, &num,
598 return 0; /* .*@sys doesn't match either */
603 *sysnamelist = afs_sysnamelist;
605 if (afs_nfsexporter) {
606 au = afs_GetUser(areq->uid, avc->f.fid.Cell, READ_LOCK);
609 EXP_SYSNAME(au->exporter, (char *)0, sysnamelist, &num, 0);
611 afs_PutUser(au, READ_LOCK);
615 afs_PutUser(au, READ_LOCK);
617 if (++(state->index) >= num || !(*sysnamelist)[(unsigned int)state->index])
618 return 0; /* end of list */
620 strcpy(state->name + state->offset, (*sysnamelist)[(unsigned int)state->index]);
625 afs_CheckBulkStatus(struct afs_conn *tc, int nFids, AFSBulkStats *statParm,
631 if (statParm->AFSBulkStats_len != nFids || cbParm->AFSCBs_len != nFids) {
632 afs_warn("afs: BulkFetchStatus length %u/%u, expected %u\n",
633 (unsigned)statParm->AFSBulkStats_len,
634 (unsigned)cbParm->AFSCBs_len, nFids);
635 afs_BadFetchStatus(tc);
638 for (i = 0; i < nFids; i++) {
639 if (statParm->AFSBulkStats_val[i].errorCode) {
642 code = afs_CheckFetchStatus(tc, &statParm->AFSBulkStats_val[i]);
651 extern int BlobScan(struct dcache * afile, afs_int32 ablob, afs_int32 *ablobOut);
653 /* called with an unlocked directory and directory cookie. Areqp
654 * describes who is making the call.
655 * Scans the next N (about 30, typically) directory entries, and does
656 * a bulk stat call to stat them all.
658 * Must be very careful when merging in RPC responses, since we dont
659 * want to overwrite newer info that was added by a file system mutating
660 * call that ran concurrently with our bulk stat call.
662 * We do that, as described below, by not merging in our info (always
663 * safe to skip the merge) if the status info is valid in the vcache entry.
665 * If adapt ever implements the bulk stat RPC, then this code will need to
666 * ensure that vcaches created for failed RPC's to older servers have the
669 static struct vcache *BStvc = NULL;
672 afs_DoBulkStat(struct vcache *adp, long dirCookie, struct vrequest *areqp)
674 int nentries; /* # of entries to prefetch */
675 int nskip; /* # of slots in the LRU queue to skip */
676 #ifdef AFS_DARWIN80_ENV
680 struct vcache *lruvcp; /* vcache ptr of our goal pos in LRU queue */
681 struct dcache *dcp; /* chunk containing the dir block */
682 afs_size_t temp; /* temp for holding chunk length, &c. */
683 struct AFSFid *fidsp; /* file IDs were collecting */
684 struct AFSCallBack *cbsp; /* call back pointers */
685 struct AFSCallBack *tcbp; /* temp callback ptr */
686 struct AFSFetchStatus *statsp; /* file status info */
687 struct AFSVolSync volSync; /* vol sync return info */
688 struct vcache *tvcp; /* temp vcp */
689 struct afs_q *tq; /* temp queue variable */
690 AFSCBFids fidParm; /* file ID parm for bulk stat */
691 AFSBulkStats statParm; /* stat info parm for bulk stat */
692 int fidIndex = 0; /* which file were stating */
693 struct afs_conn *tcp = 0; /* conn for call */
694 AFSCBs cbParm; /* callback parm for bulk stat */
695 struct server *hostp = 0; /* host we got callback from */
696 long startTime; /* time we started the call,
697 * for callback expiration base
699 int ftype[4] = {VNON, VREG, VDIR, VLNK}; /* verify type is as expected */
700 afs_size_t statSeqNo = 0; /* Valued of file size to detect races */
701 int code; /* error code */
702 afs_int32 newIndex; /* new index in the dir */
703 struct DirBuffer entry; /* Buffer for dir manipulation */
704 struct DirEntry *dirEntryp; /* dir entry we are examining */
706 struct VenusFid afid; /* file ID we are using now */
707 struct VenusFid tfid; /* another temp. file ID */
708 afs_int32 retry; /* handle low-level SGI MP race conditions */
709 long volStates; /* flags from vol structure */
710 struct volume *volp = 0; /* volume ptr */
711 struct VenusFid dotdot = {0, {0, 0, 0}};
712 int flagIndex = 0; /* First file with bulk fetch flag set */
713 struct rx_connection *rxconn;
716 dotdot.Fid.Unique = 0;
717 dotdot.Fid.Vnode = 0;
719 /* first compute some basic parameters. We dont want to prefetch more
720 * than a fraction of the cache in any given call, and we want to preserve
721 * a portion of the LRU queue in any event, so as to avoid thrashing
722 * the entire stat cache (we will at least leave some of it alone).
723 * presently dont stat more than 1/8 the cache in any one call. */
724 nentries = afs_cacheStats / 8;
726 /* dont bother prefetching more than one calls worth of info */
727 if (nentries > AFSCBMAX)
730 /* heuristic to make sure that things fit in 4K. This means that
731 * we shouldnt make it any bigger than 47 entries. I am typically
732 * going to keep it a little lower, since we don't want to load
733 * too much of the stat cache.
738 /* now, to reduce the stack size, well allocate two 4K blocks,
739 * one for fids and callbacks, and one for stat info. Well set
740 * up our pointers to the memory from there, too.
742 statsp = osi_Alloc(AFSCBMAX * sizeof(AFSFetchStatus));
743 fidsp = osi_AllocLargeSpace(nentries * sizeof(AFSFid));
744 cbsp = osi_Alloc(AFSCBMAX * sizeof(AFSCallBack));
746 /* next, we must iterate over the directory, starting from the specified
747 * cookie offset (dirCookie), and counting out nentries file entries.
748 * We skip files that already have stat cache entries, since we
749 * dont want to bulk stat files that are already in the cache.
752 code = afs_VerifyVCache(adp, areqp);
756 dcp = afs_GetDCache(adp, (afs_size_t) 0, areqp, &temp, &temp, 1);
762 /* lock the directory cache entry */
763 ObtainReadLock(&adp->lock);
764 ObtainReadLock(&dcp->lock);
767 * Make sure that the data in the cache is current. There are two
768 * cases we need to worry about:
769 * 1. The cache data is being fetched by another process.
770 * 2. The cache data is no longer valid
772 while ((adp->f.states & CStatd)
773 && (dcp->dflags & DFFetching)
774 && hsame(adp->f.m.DataVersion, dcp->f.versionNo)) {
775 afs_Trace4(afs_iclSetp, CM_TRACE_DCACHEWAIT, ICL_TYPE_STRING,
776 __FILE__, ICL_TYPE_INT32, __LINE__, ICL_TYPE_POINTER, dcp,
777 ICL_TYPE_INT32, dcp->dflags);
778 ReleaseReadLock(&dcp->lock);
779 ReleaseReadLock(&adp->lock);
780 afs_osi_Sleep(&dcp->validPos);
781 ObtainReadLock(&adp->lock);
782 ObtainReadLock(&dcp->lock);
784 if (!(adp->f.states & CStatd)
785 || !hsame(adp->f.m.DataVersion, dcp->f.versionNo)) {
786 ReleaseReadLock(&dcp->lock);
787 ReleaseReadLock(&adp->lock);
792 /* Generate a sequence number so we can tell whether we should
793 * store the attributes when processing the response. This number is
794 * stored in the file size when we set the CBulkFetching bit. If the
795 * CBulkFetching is still set and this value hasn't changed, then
796 * we know we were the last to set CBulkFetching bit for this file,
797 * and it is safe to set the status information for this file.
799 statSeqNo = bulkStatCounter++;
800 /* ensure against wrapping */
802 statSeqNo = bulkStatCounter++;
804 /* now we have dir data in the cache, so scan the dir page */
807 while (1) { /* Should probably have some constant bound */
808 /* look for first safe entry to examine in the directory. BlobScan
809 * looks for a the 1st allocated dir after the dirCookie slot.
811 code = BlobScan(dcp, (dirCookie >> 5), &newIndex);
812 if (code || newIndex == 0)
815 /* remember the updated directory cookie */
816 dirCookie = newIndex << 5;
818 /* get a ptr to the dir entry */
819 code = afs_dir_GetBlob(dcp, newIndex, &entry);
822 dirEntryp = (struct DirEntry *)entry.data;
824 /* dont copy more than we have room for */
825 if (fidIndex >= nentries) {
830 /* now, if the dir entry looks good, copy it out to our list. Vnode
831 * 0 means deleted, although it should also be free were it deleted.
833 if (dirEntryp->fid.vnode != 0) {
834 /* dont copy entries we have in our cache. This check will
835 * also make us skip "." and probably "..", unless it has
836 * disappeared from the cache since we did our namei call.
838 tfid.Cell = adp->f.fid.Cell;
839 tfid.Fid.Volume = adp->f.fid.Fid.Volume;
840 tfid.Fid.Vnode = ntohl(dirEntryp->fid.vnode);
841 tfid.Fid.Unique = ntohl(dirEntryp->fid.vunique);
844 ObtainWriteLock(&afs_xvcache, 130);
845 tvcp = afs_FindVCache(&tfid, &retry, IS_WLOCK /* no stats | LRU */ );
847 ReleaseWriteLock(&afs_xvcache);
850 } while (tvcp && retry);
851 if (!tvcp) { /* otherwise, create manually */
852 tvcp = afs_NewBulkVCache(&tfid, hostp, statSeqNo);
855 ObtainWriteLock(&tvcp->lock, 505);
856 #ifdef AFS_DARWIN80_ENV
857 /* use even/odd hack to guess file versus dir.
858 let links be reaped. oh well. */
859 if (dirEntryp->fid.vnode & 1)
860 tvcp->f.m.Type = VDIR;
862 tvcp->f.m.Type = VREG;
863 /* finalize to a best guess */
864 afs_darwin_finalizevnode(tvcp, AFSTOV(adp), NULL, 0, 1);
865 /* re-acquire usecount that finalizevnode disposed of */
866 vnode_ref(AFSTOV(tvcp));
868 ReleaseWriteLock(&afs_xvcache);
869 afs_RemoveVCB(&tfid);
870 ReleaseWriteLock(&tvcp->lock);
872 ReleaseWriteLock(&afs_xvcache);
875 ReleaseWriteLock(&afs_xvcache);
880 ReleaseReadLock(&dcp->lock);
881 ReleaseReadLock(&adp->lock);
883 goto done; /* can happen if afs_NewVCache fails */
886 /* WARNING: afs_DoBulkStat uses the Length field to store a
887 * sequence number for each bulk status request. Under no
888 * circumstances should afs_DoBulkStat store a sequence number
889 * if the new length will be ignored when afs_ProcessFS is
890 * called with new stats. */
892 if (!(tvcp->f.states & CStatd)
893 && (!((tvcp->f.states & CBulkFetching) &&
894 (tvcp->f.m.Length != statSeqNo)))
895 && (tvcp->execsOrWriters <= 0)
896 && !afs_DirtyPages(tvcp)
897 && !AFS_VN_MAPPED((vnode_t *) tvcp))
899 if (!(tvcp->f.states & CStatd)
900 && (!((tvcp->f.states & CBulkFetching) &&
901 (tvcp->f.m.Length != statSeqNo)))
902 && (tvcp->execsOrWriters <= 0)
903 && !afs_DirtyPages(tvcp))
907 /* this entry doesnt exist in the cache, and is not
908 * already being fetched by someone else, so add it to the
909 * list of file IDs to obtain.
911 * We detect a callback breaking race condition by checking the
912 * CBulkFetching state bit and the value in the file size.
913 * It is safe to set the status only if the CBulkFetching
914 * flag is still set and the value in the file size does
915 * not change. NewBulkVCache sets us up for the new ones.
916 * Set up the rest here.
918 * Don't fetch status for dirty files. We need to
919 * preserve the value of the file size. We could
920 * flush the pages, but it wouldn't be worthwhile.
922 if (!(tvcp->f.states & CBulkFetching)) {
923 tvcp->f.states |= CBulkFetching;
924 tvcp->f.m.Length = statSeqNo;
926 memcpy((char *)(fidsp + fidIndex), (char *)&tfid.Fid,
933 /* if dir vnode has non-zero entry */
934 /* move to the next dir entry by adding in the # of entries
935 * used by this dir entry.
937 temp = afs_dir_NameBlobs(dirEntryp->name) << 5;
942 } /* while loop over all dir entries */
944 /* now release the dir lock and prepare to make the bulk RPC */
945 ReleaseReadLock(&dcp->lock);
946 ReleaseReadLock(&adp->lock);
948 /* release the chunk */
951 /* dont make a null call */
956 /* setup the RPC parm structures */
957 fidParm.AFSCBFids_len = fidIndex;
958 fidParm.AFSCBFids_val = fidsp;
959 statParm.AFSBulkStats_len = fidIndex;
960 statParm.AFSBulkStats_val = statsp;
961 cbParm.AFSCBs_len = fidIndex;
962 cbParm.AFSCBs_val = cbsp;
964 /* start the timer; callback expirations are relative to this */
965 startTime = osi_Time();
967 tcp = afs_Conn(&adp->f.fid, areqp, SHARED_LOCK, &rxconn);
969 hostp = tcp->parent->srvr->server;
971 for (i = 0; i < fidIndex; i++) {
972 /* we must set tvcp->callback before the BulkStatus call, so
973 * we can detect concurrent InitCallBackState's */
975 afid.Cell = adp->f.fid.Cell;
976 afid.Fid.Volume = adp->f.fid.Fid.Volume;
977 afid.Fid.Vnode = fidsp[i].Vnode;
978 afid.Fid.Unique = fidsp[i].Unique;
982 ObtainReadLock(&afs_xvcache);
983 tvcp = afs_FindVCache(&afid, &retry, 0 /* !stats&!lru */);
984 ReleaseReadLock(&afs_xvcache);
985 } while (tvcp && retry);
991 if ((tvcp->f.states & CBulkFetching) &&
992 (tvcp->f.m.Length == statSeqNo)) {
993 tvcp->callback = hostp;
1000 XSTATS_START_TIME(AFS_STATS_FS_RPCIDX_BULKSTATUS);
1002 if (!(tcp->parent->srvr->server->flags & SNO_INLINEBULK)) {
1005 RXAFS_InlineBulkStatus(rxconn, &fidParm, &statParm,
1008 if (code == RXGEN_OPCODE) {
1009 tcp->parent->srvr->server->flags |= SNO_INLINEBULK;
1012 RXAFS_BulkStatus(rxconn, &fidParm, &statParm,
1019 RXAFS_BulkStatus(rxconn, &fidParm, &statParm, &cbParm,
1026 code = afs_CheckBulkStatus(tcp, fidIndex, &statParm, &cbParm);
1030 /* make sure we give afs_Analyze a chance to retry,
1031 * but if the RPC succeeded we may have entries to merge.
1032 * if we wipe code with one entry's status we get bogus failures.
1034 } while (afs_Analyze
1035 (tcp, rxconn, code ? code : (&statsp[0])->errorCode,
1036 &adp->f.fid, areqp, AFS_STATS_FS_RPCIDX_BULKSTATUS,
1037 SHARED_LOCK, NULL));
1039 /* now, if we didnt get the info, bail out. */
1043 /* we need vol flags to create the entries properly */
1044 dotdot.Fid.Volume = 0;
1045 volp = afs_GetVolume(&adp->f.fid, areqp, READ_LOCK);
1047 volStates = volp->states;
1048 if (volp->dotdot.Fid.Volume != 0)
1049 dotdot = volp->dotdot;
1053 /* find the place to merge the info into We do this by skipping
1054 * nskip entries in the LRU queue. The more we skip, the more
1055 * we preserve, since the head of the VLRU queue is the most recently
1059 nskip = afs_cacheStats / 2; /* preserved fraction of the cache */
1060 ObtainReadLock(&afs_xvcache);
1061 #ifdef AFS_DARWIN80_ENV
1064 if (QEmpty(&VLRU)) {
1065 /* actually a serious error, probably should panic. Probably will
1066 * panic soon, oh well. */
1067 ReleaseReadLock(&afs_xvcache);
1068 afs_warnuser("afs_DoBulkStat: VLRU empty!");
1071 if ((VLRU.next->prev != &VLRU) || (VLRU.prev->next != &VLRU)) {
1072 refpanic("Bulkstat VLRU inconsistent");
1074 for (tq = VLRU.next; tq != &VLRU; tq = QNext(tq)) {
1076 #ifdef AFS_DARWIN80_ENV
1077 if ((!(QTOV(tq)->f.states & CDeadVnode)&&!(QTOV(tq)->f.states & CVInit)))
1081 if (QNext(QPrev(tq)) != tq) {
1083 refpanic("BulkStat VLRU inconsistent");
1089 lruvcp = QTOV(VLRU.next);
1091 /* now we have to hold this entry, so that it does not get moved
1092 * into the free list while we're running. It could still get
1093 * moved within the lru queue, but hopefully that will be rare; it
1094 * doesn't hurt nearly as much.
1097 #ifdef AFS_DARWIN80_ENV
1098 if (((lruvcp->f.states & CDeadVnode)||(lruvcp->f.states & CVInit))) {
1104 panic("Can't find non-dead vnode in VLRU\n");
1106 lruvp = AFSTOV(lruvcp);
1107 if (vnode_get(lruvp)) /* this bumps ref count */
1109 else if (vnode_ref(lruvp)) {
1111 /* AFSTOV(lruvcp) may be NULL */
1117 osi_vnhold(lruvcp, &retry);
1119 ReleaseReadLock(&afs_xvcache); /* could be read lock */
1123 /* otherwise, merge in the info. We have to be quite careful here,
1124 * since we need to ensure that we don't merge old info over newer
1125 * stuff in a stat cache entry. We're very conservative here: we don't
1126 * do the merge at all unless we ourselves create the stat cache
1127 * entry. That's pretty safe, and should work pretty well, since we
1128 * typically expect to do the stat cache creation ourselves.
1130 * We also have to take into account racing token revocations.
1132 for (i = 0; i < fidIndex; i++) {
1133 if ((&statsp[i])->errorCode)
1135 afid.Cell = adp->f.fid.Cell;
1136 afid.Fid.Volume = adp->f.fid.Fid.Volume;
1137 afid.Fid.Vnode = fidsp[i].Vnode;
1138 afid.Fid.Unique = fidsp[i].Unique;
1141 ObtainReadLock(&afs_xvcache);
1142 tvcp = afs_FindVCache(&afid, &retry, 0/* !stats&!lru */);
1143 ReleaseReadLock(&afs_xvcache);
1144 } while (tvcp && retry);
1146 /* The entry may no longer exist */
1151 /* now we have the entry held, but we need to fill it in */
1152 ObtainWriteLock(&tvcp->lock, 131);
1154 /* if CBulkFetching is not set, or if the file size no longer
1155 * matches the value we placed there when we set the CBulkFetching
1156 * flag, then someone else has done something with this node,
1157 * and we may not have the latest status information for this
1158 * file. Leave the entry alone. There's also a file type
1159 * change here, for OSX bulkstat support.
1161 if (!(tvcp->f.states & CBulkFetching)
1162 || (tvcp->f.m.Length != statSeqNo)
1163 || (ftype[(&statsp[i])->FileType] != vType(tvcp))) {
1165 ReleaseWriteLock(&tvcp->lock);
1166 afs_PutVCache(tvcp);
1170 /* now copy ".." entry back out of volume structure, if necessary */
1171 if (tvcp->mvstat == AFS_MVSTAT_ROOT && (dotdot.Fid.Volume != 0)) {
1172 if (!tvcp->mvid.parent)
1173 tvcp->mvid.parent = osi_AllocSmallSpace(sizeof(struct VenusFid));
1174 *tvcp->mvid.parent = dotdot;
1177 #ifdef AFS_DARWIN80_ENV
1178 if (((lruvcp->f.states & CDeadVnode)||(lruvcp->f.states & CVInit)))
1179 panic("vlru control point went dead\n");
1182 ObtainWriteLock(&afs_xvcache, 132);
1183 if ((VLRU.next->prev != &VLRU) || (VLRU.prev->next != &VLRU)) {
1184 refpanic("Bulkstat VLRU inconsistent2");
1186 if ((QNext(QPrev(&tvcp->vlruq)) != &tvcp->vlruq)
1187 || (QPrev(QNext(&tvcp->vlruq)) != &tvcp->vlruq)) {
1188 refpanic("Bulkstat VLRU inconsistent4");
1190 if ((QNext(QPrev(&lruvcp->vlruq)) != &lruvcp->vlruq)
1191 || (QPrev(QNext(&lruvcp->vlruq)) != &lruvcp->vlruq)) {
1192 refpanic("Bulkstat VLRU inconsistent5");
1195 if (tvcp != lruvcp) { /* if they are == don't move it, don't corrupt vlru */
1196 QRemove(&tvcp->vlruq);
1197 QAdd(&lruvcp->vlruq, &tvcp->vlruq);
1200 if ((VLRU.next->prev != &VLRU) || (VLRU.prev->next != &VLRU)) {
1201 refpanic("Bulkstat VLRU inconsistent3");
1203 if ((QNext(QPrev(&tvcp->vlruq)) != &tvcp->vlruq)
1204 || (QPrev(QNext(&tvcp->vlruq)) != &tvcp->vlruq)) {
1205 refpanic("Bulkstat VLRU inconsistent5");
1207 if ((QNext(QPrev(&lruvcp->vlruq)) != &lruvcp->vlruq)
1208 || (QPrev(QNext(&lruvcp->vlruq)) != &lruvcp->vlruq)) {
1209 refpanic("Bulkstat VLRU inconsistent6");
1211 ReleaseWriteLock(&afs_xvcache);
1213 ObtainWriteLock(&afs_xcbhash, 494);
1215 /* We need to check the flags again. We may have missed
1216 * something while we were waiting for a lock.
1218 if (!(tvcp->f.states & CBulkFetching) || (tvcp->f.m.Length != statSeqNo)) {
1220 ReleaseWriteLock(&tvcp->lock);
1221 ReleaseWriteLock(&afs_xcbhash);
1222 afs_PutVCache(tvcp);
1226 /* now merge in the resulting status back into the vnode.
1227 * We only do this if the entry looks clear.
1229 afs_ProcessFS(tvcp, &statsp[i], areqp);
1230 #if defined(AFS_LINUX22_ENV)
1231 afs_fill_inode(AFSTOV(tvcp), NULL); /* reset inode operations */
1234 /* do some accounting for bulk stats: mark this entry as
1235 * loaded, so we can tell if we use it before it gets
1238 tvcp->f.states |= CBulkStat;
1239 tvcp->f.states &= ~CBulkFetching;
1241 afs_bulkStatsDone++;
1243 /* merge in vol info */
1244 if (volStates & VRO)
1245 tvcp->f.states |= CRO;
1246 if (volStates & VBackup)
1247 tvcp->f.states |= CBackup;
1248 if (volStates & VForeign)
1249 tvcp->f.states |= CForeign;
1251 /* merge in the callback info */
1252 tvcp->f.states |= CTruth;
1254 /* get ptr to the callback we are interested in */
1257 if (tcbp->ExpirationTime != 0) {
1258 tvcp->cbExpires = tcbp->ExpirationTime + startTime;
1259 tvcp->callback = hostp;
1260 tvcp->f.states |= CStatd;
1261 afs_QueueCallback(tvcp, CBHash(tcbp->ExpirationTime), volp);
1262 } else if (tvcp->f.states & CRO) {
1263 /* ordinary callback on a read-only volume -- AFS 3.2 style */
1264 tvcp->cbExpires = 3600 + startTime;
1265 tvcp->callback = hostp;
1266 tvcp->f.states |= CStatd;
1267 afs_QueueCallback(tvcp, CBHash(3600), volp);
1269 afs_StaleVCacheFlags(tvcp,
1270 AFS_STALEVC_CBLOCKED | AFS_STALEVC_CLEARCB,
1273 #ifdef AFS_DARWIN80_ENV
1274 /* reclaim->FlushVCache will need xcbhash */
1275 if (((tvcp->f.states & CDeadVnode)||(tvcp->f.states & CVInit))) {
1276 ReleaseWriteLock(&afs_xcbhash);
1277 /* passing in a parent hangs getting the vnode lock */
1278 code = afs_darwin_finalizevnode(tvcp, NULL, NULL, 0, 1);
1280 /* It's gonna get recycled - shouldn't happen */
1281 afs_StaleVCacheFlags(tvcp,
1282 AFS_STALEVC_CBLOCKED | AFS_STALEVC_CLEARCB,
1285 /* re-acquire the usecount that finalizevnode disposed of */
1286 vnode_ref(AFSTOV(tvcp));
1289 ReleaseWriteLock(&afs_xcbhash);
1291 ReleaseWriteLock(&tvcp->lock);
1292 /* finally, we're done with the entry */
1293 afs_PutVCache(tvcp);
1294 } /* for all files we got back */
1296 /* finally return the pointer into the LRU queue */
1297 #ifdef AFS_DARWIN80_ENV
1298 if (((lruvcp->f.states & CDeadVnode)||(lruvcp->f.states & CVInit)))
1299 panic("vlru control point went dead before put\n");
1305 afs_PutVCache(lruvcp);
1309 /* Be sure to turn off the CBulkFetching flags */
1310 for (i = flagIndex; i < fidIndex; i++) {
1311 afid.Cell = adp->f.fid.Cell;
1312 afid.Fid.Volume = adp->f.fid.Fid.Volume;
1313 afid.Fid.Vnode = fidsp[i].Vnode;
1314 afid.Fid.Unique = fidsp[i].Unique;
1317 ObtainReadLock(&afs_xvcache);
1318 tvcp = afs_FindVCache(&afid, &retry, 0 /* !stats&!lru */);
1319 ReleaseReadLock(&afs_xvcache);
1320 } while (tvcp && retry);
1322 if ((tvcp->f.states & CBulkFetching)
1323 && (tvcp->f.m.Length == statSeqNo)) {
1324 tvcp->f.states &= ~CBulkFetching;
1326 afs_PutVCache(tvcp);
1330 afs_PutVolume(volp, READ_LOCK);
1333 osi_FreeLargeSpace((char *)fidsp);
1334 osi_Free((char *)statsp, AFSCBMAX * sizeof(AFSFetchStatus));
1335 osi_Free((char *)cbsp, AFSCBMAX * sizeof(AFSCallBack));
1339 /* was: (AFS_DEC_ENV) || defined(AFS_OSF30_ENV) || defined(AFS_NCR_ENV) */
1340 #ifdef AFS_DARWIN80_ENV
1343 static int AFSDOBULK = 1;
1347 osi_lookup_isdot(const char *aname)
1351 /* in Solaris, we can get passed "" as a path component if we are the
1352 * root directory, e.g. after a call to chroot. It is equivalent to
1356 #endif /* AFS_SUN5_ENV */
1357 if (aname[0] == '.' && !aname[1]) {
1364 #if defined(AFS_SUN5_ENV) || defined(AFS_SGI_ENV)
1365 afs_lookup(OSI_VC_DECL(adp), char *aname, struct vcache **avcp, struct pathname *pnp, int flags, struct vnode *rdir, afs_ucred_t *acred)
1366 #elif defined(UKERNEL)
1367 afs_lookup(OSI_VC_DECL(adp), char *aname, struct vcache **avcp, afs_ucred_t *acred, int flags)
1369 afs_lookup(OSI_VC_DECL(adp), char *aname, struct vcache **avcp, afs_ucred_t *acred)
1372 struct vrequest *treq = NULL;
1374 struct vcache *tvc = 0;
1376 afs_int32 bulkcode = 0;
1377 int pass = 0, hit = 0;
1378 int force_eval = afs_fakestat_enable ? 0 : 1;
1380 afs_hyper_t versionNo;
1381 int no_read_access = 0;
1382 struct sysname_info sysState; /* used only for @sys checking */
1383 int dynrootRetry = 1;
1384 struct afs_fakestat_state fakestate;
1385 int tryEvalOnly = 0;
1386 OSI_VC_CONVERT(adp);
1388 AFS_STATCNT(afs_lookup);
1389 afs_InitFakeStat(&fakestate);
1393 if ((code = afs_CreateReq(&treq, acred)))
1396 if (afs_fakestat_enable && adp->mvstat == AFS_MVSTAT_MTPT) {
1397 if (strcmp(aname, ".directory") == 0)
1401 #if defined(AFS_DARWIN_ENV)
1402 /* Workaround for MacOSX Finder, which tries to look for
1403 * .DS_Store and Contents under every directory.
1405 if (afs_fakestat_enable && adp->mvstat == AFS_MVSTAT_MTPT) {
1406 if (strcmp(aname, ".DS_Store") == 0)
1408 if (strcmp(aname, "Contents") == 0)
1411 if (afs_fakestat_enable && adp->mvstat == AFS_MVSTAT_ROOT) {
1412 if (strncmp(aname, "._", 2) == 0)
1418 code = afs_TryEvalFakeStat(&adp, &fakestate, treq);
1420 code = afs_EvalFakeStat(&adp, &fakestate, treq);
1422 /*printf("Code is %d\n", code);*/
1424 if (tryEvalOnly && adp->mvstat == AFS_MVSTAT_MTPT)
1429 /* come back to here if we encounter a non-existent object in a read-only
1430 * volume's directory */
1432 *avcp = NULL; /* Since some callers don't initialize it */
1435 if (!(adp->f.states & CStatd) && !afs_InReadDir(adp)) {
1436 if ((code = afs_VerifyVCache2(adp, treq))) {
1442 /* watch for ".." in a volume root */
1443 if (adp->mvstat == AFS_MVSTAT_ROOT && aname[0] == '.' && aname[1] == '.' && !aname[2]) {
1444 /* looking up ".." in root via special hacks */
1445 if (adp->mvid.parent == (struct VenusFid *)0 || adp->mvid.parent->Fid.Volume == 0) {
1449 /* otherwise we have the fid here, so we use it */
1450 /*printf("Getting vcache\n");*/
1451 tvc = afs_GetVCache(adp->mvid.parent, treq, NULL, NULL);
1452 afs_Trace3(afs_iclSetp, CM_TRACE_GETVCDOTDOT, ICL_TYPE_FID, adp->mvid.parent,
1453 ICL_TYPE_POINTER, tvc, ICL_TYPE_INT32, code);
1455 code = (tvc ? 0 : EIO);
1457 if (tvc && !VREFCOUNT_GT(tvc, 0)) {
1461 /*printf("LOOKUP GETVCDOTDOT -> %d\n", code); */
1466 /* now check the access */
1467 if (treq->uid != adp->last_looker) {
1468 if (!afs_AccessOK(adp, PRSFS_LOOKUP, treq, CHECK_MODE_BITS)) {
1473 adp->last_looker = treq->uid;
1476 /* Check for read access as well. We need read access in order to
1477 * stat files, but not to stat subdirectories. */
1478 if (!afs_AccessOK(adp, PRSFS_READ, treq, CHECK_MODE_BITS))
1481 /* special case lookup of ".". Can we check for it sooner in this code,
1482 * for instance, way up before "redo:" ??
1483 * I'm not fiddling with the LRUQ here, either, perhaps I should, or else
1484 * invent a lightweight version of GetVCache.
1486 if (osi_lookup_isdot(aname)) { /* special case */
1487 ObtainReadLock(&afs_xvcache);
1489 ReleaseReadLock(&afs_xvcache);
1490 #ifdef AFS_DARWIN80_ENV
1491 vnode_get(AFSTOV(adp));
1496 if (adp && !VREFCOUNT_GT(adp, 0)) {
1503 * Special case lookup of ".." in the dynamic mount directory.
1504 * The parent of this directory is _always_ the AFS root volume.
1506 if (afs_IsDynrootMount(adp) &&
1507 aname[0] == '.' && aname[1] == '.' && !aname[2]) {
1509 ObtainReadLock(&afs_xvcache);
1510 osi_vnhold(afs_globalVp, 0);
1511 ReleaseReadLock(&afs_xvcache);
1512 #ifdef AFS_DARWIN80_ENV
1513 vnode_get(AFSTOV(afs_globalVp));
1516 *avcp = tvc = afs_globalVp;
1522 * Special case lookups in the dynamic mount directory.
1523 * The names here take the form cell:volume, similar to a mount point.
1524 * EvalMountData parses that and returns a cell and volume ID, which
1525 * we use to construct the appropriate dynroot Fid.
1527 if (afs_IsDynrootMount(adp)) {
1528 struct VenusFid tfid;
1529 afs_uint32 cellidx, volid, vnoid, uniq;
1531 code = EvalMountData('%', aname, 0, 0, NULL, treq, &cellidx, &volid, &vnoid, &uniq);
1534 /* If a vnode was returned, it's not a real mount point */
1536 struct cell *tcell = afs_GetCellByIndex(cellidx, READ_LOCK);
1537 tfid.Cell = tcell->cellNum;
1538 afs_PutCell(tcell, READ_LOCK);
1539 tfid.Fid.Vnode = vnoid;
1540 tfid.Fid.Volume = volid;
1541 tfid.Fid.Unique = uniq;
1543 afs_GetDynrootMountFid(&tfid);
1544 tfid.Fid.Vnode = VNUM_FROM_TYPEID(VN_TYPE_MOUNT, cellidx << 2);
1545 tfid.Fid.Unique = volid;
1547 *avcp = tvc = afs_GetVCache(&tfid, treq, NULL, NULL);
1548 code = (tvc ? 0 : EIO);
1553 #ifdef AFS_LINUX26_ENV
1555 * Special case of the dynamic mount volume in a static root.
1556 * This is really unfortunate, but we need this for the translator.
1558 if (adp == afs_globalVp && !afs_GetDynrootEnable() &&
1559 !strcmp(aname, AFS_DYNROOT_MOUNTNAME)) {
1560 struct VenusFid tfid;
1562 afs_GetDynrootMountFid(&tfid);
1563 *avcp = tvc = afs_GetVCache(&tfid, treq, NULL, NULL);
1570 Check_AtSys(adp, aname, &sysState, treq);
1571 tname = sysState.name;
1573 /* 1st Check_AtSys and lookup by tname is required here, for now,
1574 * because the dnlc is *not* told to remove entries for the parent
1575 * dir of file/dir op that afs_LocalHero likes, but dnlc is informed
1576 * if the cached entry for the parent dir is invalidated for a
1578 * Otherwise, we'd be able to do a dnlc lookup on an entry ending
1579 * w/@sys and know the dnlc was consistent with reality. */
1580 tvc = osi_dnlc_lookup(adp, tname, WRITE_LOCK);
1581 *avcp = tvc; /* maybe wasn't initialized, but it is now */
1583 if (no_read_access && vType(tvc) != VDIR && vType(tvc) != VLNK) {
1584 /* need read access on dir to stat non-directory / non-link */
1590 #ifdef AFS_LINUX22_ENV
1591 if (tvc->mvstat == AFS_MVSTAT_ROOT) { /* we don't trust the dnlc for root vcaches */
1592 AFS_RELE(AFSTOV(tvc));
1599 #else /* non - LINUX */
1603 #endif /* linux22 */
1606 { /* sub-block just to reduce stack usage */
1608 afs_size_t dirOffset, dirLen;
1609 struct VenusFid tfid;
1611 /* now we have to lookup the next fid */
1612 if (afs_InReadDir(adp))
1613 tdc = adp->dcreaddir;
1615 tdc = afs_GetDCache(adp, (afs_size_t) 0, treq,
1616 &dirOffset, &dirLen, 1);
1618 *avcp = NULL; /* redundant, but harmless */
1623 /* now we will just call dir package with appropriate inode.
1624 * Dirs are always fetched in their entirety for now */
1625 ObtainReadLock(&adp->lock);
1626 ObtainReadLock(&tdc->lock);
1629 * Make sure that the data in the cache is current. There are two
1630 * cases we need to worry about:
1631 * 1. The cache data is being fetched by another process.
1632 * 2. The cache data is no longer valid
1634 * If a readdir is in progress _in this thread_, it has a shared
1635 * lock on the vcache and has obtained current data, so we just
1636 * use that. This eliminates several possible deadlocks.
1638 if (!afs_InReadDir(adp)) {
1639 while ((adp->f.states & CStatd)
1640 && (tdc->dflags & DFFetching)
1641 && hsame(adp->f.m.DataVersion, tdc->f.versionNo)) {
1642 ReleaseReadLock(&tdc->lock);
1643 ReleaseReadLock(&adp->lock);
1644 afs_osi_Sleep(&tdc->validPos);
1645 ObtainReadLock(&adp->lock);
1646 ObtainReadLock(&tdc->lock);
1648 if (!(adp->f.states & CStatd)
1649 || !hsame(adp->f.m.DataVersion, tdc->f.versionNo)) {
1650 ReleaseReadLock(&tdc->lock);
1651 ReleaseReadLock(&adp->lock);
1653 if (tname && tname != aname)
1654 osi_FreeLargeSpace(tname);
1659 /* Save the version number for when we call osi_dnlc_enter */
1660 hset(versionNo, tdc->f.versionNo);
1663 * check for, and handle "@sys" if it's there. We should be able
1664 * to avoid the alloc and the strcpy with a little work, but it's
1665 * not pressing. If there aren't any remote users (ie, via the
1666 * NFS translator), we have a slightly easier job.
1667 * the faster way to do this is to check for *aname == '@' and if
1668 * it's there, check for @sys, otherwise, assume there's no @sys
1669 * then, if the lookup fails, check for .*@sys...
1671 /* above now implemented by Check_AtSys and Next_AtSys */
1673 /* lookup the name in the appropriate dir, and return a cache entry
1674 * on the resulting fid */
1676 afs_dir_LookupOffset(tdc, sysState.name, &tfid.Fid,
1679 /* If the first lookup doesn't succeed, maybe it's got @sys in the name */
1680 while (code == ENOENT && Next_AtSys(adp, treq, &sysState))
1682 afs_dir_LookupOffset(tdc, sysState.name, &tfid.Fid,
1684 tname = sysState.name;
1686 ReleaseReadLock(&tdc->lock);
1687 if (!afs_InReadDir(adp))
1689 if (code == ENOENT && afs_IsDynroot(adp) && dynrootRetry && !tryEvalOnly) {
1691 char *cn = (tname[0] == '.') ? tname + 1 : tname;
1692 ReleaseReadLock(&adp->lock);
1693 /* confirm it's not just hushed */
1694 tc = afs_GetCellByName(cn, WRITE_LOCK);
1696 if (tc->states & CHush) {
1697 tc->states &= ~CHush;
1698 ReleaseWriteLock(&tc->lock);
1699 afs_DynrootInvalidate();
1702 ReleaseWriteLock(&tc->lock);
1704 /* Allow a second dynroot retry if the cell was hushed before */
1706 if (tname[0] == '.')
1707 afs_LookupAFSDB(tname + 1);
1709 afs_LookupAFSDB(tname);
1710 if (tname && tname != aname)
1711 osi_FreeLargeSpace(tname);
1714 ReleaseReadLock(&adp->lock);
1717 /* new fid has same cell and volume */
1718 tfid.Cell = adp->f.fid.Cell;
1719 tfid.Fid.Volume = adp->f.fid.Fid.Volume;
1720 afs_Trace4(afs_iclSetp, CM_TRACE_LOOKUP, ICL_TYPE_POINTER, adp,
1721 ICL_TYPE_STRING, tname, ICL_TYPE_FID, &tfid,
1722 ICL_TYPE_INT32, code);
1725 if (code != ENOENT) {
1726 /*printf("LOOKUP dirLookupOff -> %d\n", code);*/
1731 /* prefetch some entries, if the dir is currently open. The variable
1732 * dirCookie tells us where to start prefetching from.
1734 if (!AFS_IS_DISCONNECTED &&
1735 AFSDOBULK && adp->opens > 0 && !(adp->f.states & CForeign)
1736 && !afs_IsDynroot(adp) && !afs_InReadDir(adp)) {
1738 /* if the entry is not in the cache, or is in the cache,
1739 * but hasn't been statd, then do a bulk stat operation.
1743 ObtainReadLock(&afs_xvcache);
1744 tvc = afs_FindVCache(&tfid, &retry, 0 /* !stats,!lru */ );
1745 ReleaseReadLock(&afs_xvcache);
1746 } while (tvc && retry);
1748 if (!tvc || !(tvc->f.states & CStatd))
1749 bulkcode = afs_DoBulkStat(adp, dirCookie, treq);
1753 /* if the vcache isn't usable, release it */
1754 if (tvc && !(tvc->f.states & CStatd)) {
1763 /* now get the status info, if we don't already have it */
1764 /* This is kind of weird, but we might wind up accidentally calling
1765 * RXAFS_Lookup because we happened upon a file which legitimately
1766 * has a 0 uniquifier. That is the result of allowing unique to wrap
1767 * to 0. This was fixed in AFS 3.4. For CForeign, Unique == 0 means that
1768 * the file has not yet been looked up.
1771 afs_int32 cached = 0;
1772 if (!tfid.Fid.Unique && (adp->f.states & CForeign)) {
1773 tvc = afs_LookupVCache(&tfid, treq, &cached, adp, tname);
1775 if (!tvc && !bulkcode) { /* lookup failed or wasn't called */
1776 tvc = afs_GetVCache(&tfid, treq, &cached, NULL);
1779 } /* sub-block just to reduce stack usage */
1782 if (adp->f.states & CForeign)
1783 tvc->f.states |= CForeign;
1784 tvc->f.parent.vnode = adp->f.fid.Fid.Vnode;
1785 tvc->f.parent.unique = adp->f.fid.Fid.Unique;
1786 tvc->f.states &= ~CBulkStat;
1788 if (afs_fakestat_enable == 2 && tvc->mvstat == AFS_MVSTAT_MTPT) {
1789 ObtainSharedLock(&tvc->lock, 680);
1790 if (!tvc->linkData) {
1791 UpgradeSToWLock(&tvc->lock, 681);
1792 code = afs_HandleLink(tvc, treq);
1793 ConvertWToRLock(&tvc->lock);
1795 ConvertSToRLock(&tvc->lock);
1798 if (!code && !afs_strchr(tvc->linkData, ':'))
1800 ReleaseReadLock(&tvc->lock);
1802 if (tvc->mvstat == AFS_MVSTAT_MTPT && (tvc->f.states & CMValid) && tvc->mvid.target_root != NULL)
1803 force_eval = 1; /* This is now almost for free, get it correct */
1805 #if defined(UKERNEL)
1806 if (!(flags & AFS_LOOKUP_NOEVAL))
1807 /* don't eval mount points */
1808 #endif /* UKERNEL */
1809 if (tvc->mvstat == AFS_MVSTAT_MTPT && force_eval) {
1810 /* a mt point, possibly unevaluated */
1811 struct volume *tvolp;
1813 ObtainWriteLock(&tvc->lock, 133);
1814 code = EvalMountPoint(tvc, adp, &tvolp, treq);
1815 ReleaseWriteLock(&tvc->lock);
1820 afs_PutVolume(tvolp, WRITE_LOCK);
1824 /* next, we want to continue using the target of the mt point */
1825 if (tvc->mvid.target_root && (tvc->f.states & CMValid)) {
1827 /* now lookup target, to set .. pointer */
1828 afs_Trace2(afs_iclSetp, CM_TRACE_LOOKUP1,
1829 ICL_TYPE_POINTER, tvc, ICL_TYPE_FID,
1831 uvc = tvc; /* remember for later */
1833 if (tvolp && (tvolp->states & VForeign)) {
1834 /* XXXX tvolp has ref cnt on but not locked! XXX */
1836 afs_GetRootVCache(tvc->mvid.target_root, treq, NULL, tvolp);
1838 tvc = afs_GetVCache(tvc->mvid.target_root, treq, NULL, NULL);
1840 afs_PutVCache(uvc); /* we're done with it */
1845 afs_PutVolume(tvolp, WRITE_LOCK);
1850 /* now, if we came via a new mt pt (say because of a new
1851 * release of a R/O volume), we must reevaluate the ..
1852 * ptr to point back to the appropriate place */
1854 ObtainWriteLock(&tvc->lock, 134);
1855 if (tvc->mvid.parent == NULL) {
1857 osi_AllocSmallSpace(sizeof(struct VenusFid));
1859 /* setup backpointer */
1860 *tvc->mvid.parent = tvolp->dotdot;
1861 ReleaseWriteLock(&tvc->lock);
1862 afs_PutVolume(tvolp, WRITE_LOCK);
1868 afs_PutVolume(tvolp, WRITE_LOCK);
1873 if (tvc && !VREFCOUNT_GT(tvc, 0)) {
1878 /* if we get here, we found something in a directory that couldn't
1879 * be located (a Multics "connection failure"). If the volume is
1880 * read-only, we try flushing this entry from the cache and trying
1882 if (!AFS_IS_DISCONNECTED) {
1885 tv = afs_GetVolume(&adp->f.fid, treq, READ_LOCK);
1887 if (tv->states & VRO) {
1888 pass = 1; /* try this *once* */
1889 /* re-stat to get later version */
1890 afs_StaleVCache(adp);
1891 afs_PutVolume(tv, READ_LOCK);
1894 afs_PutVolume(tv, READ_LOCK);
1904 /* put the network buffer back, if need be */
1905 if (tname != aname && tname)
1906 osi_FreeLargeSpace(tname);
1910 afs_AddMarinerName(aname, tvc);
1912 #if defined(UKERNEL)
1913 if (!(flags & AFS_LOOKUP_NOEVAL)) {
1914 /* Here we don't enter the name into the DNLC because we want the
1915 * evaluated mount dir to be there (the vcache for the mounted
1916 * volume) rather than the vc of the mount point itself. We can
1917 * still find the mount point's vc in the vcache by its fid. */
1918 #endif /* UKERNEL */
1919 if (!hit && (force_eval || tvc->mvstat != AFS_MVSTAT_MTPT)) {
1920 osi_dnlc_enter(adp, aname, tvc, &versionNo);
1922 #ifdef AFS_LINUX20_ENV
1923 /* So Linux inode cache is up to date. */
1924 code = afs_VerifyVCache(tvc, treq);
1926 afs_PutFakeStat(&fakestate);
1927 afs_DestroyReq(treq);
1928 AFS_DISCON_UNLOCK();
1929 return 0; /* can't have been any errors if hit and !code */
1932 #if defined(UKERNEL)
1939 code = afs_CheckCode(code, treq, 19);
1941 /* If there is an error, make sure *avcp is null.
1942 * Alphas panic otherwise - defect 10719.
1947 afs_PutFakeStat(&fakestate);
1948 afs_DestroyReq(treq);
1949 AFS_DISCON_UNLOCK();