Import Upstream version 1.8.5
[hcoop/debian/openafs.git] / src / afs / afs_volume.c
1 /*
2 * Copyright 2000, International Business Machines Corporation and others.
3 * All Rights Reserved.
4 *
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
8 */
9
10 /*
11 * Implements:
12 * afs_vtoi (local)
13 * afs_UFSGetVolSlot
14 * afs_MemGetVolSlot
15 * afs_CheckVolumeNames
16 * afs_FindVolume
17 */
18 #include <afsconfig.h>
19 #include "afs/param.h"
20
21
22 #include "afs/stds.h"
23 #include "afs/sysincludes.h" /* Standard vendor system headers */
24
25 #if !defined(UKERNEL)
26 #if !defined(AFS_LINUX20_ENV)
27 #include <net/if.h>
28 #endif
29 #include <netinet/in.h>
30
31 #ifdef AFS_SGI62_ENV
32 #include "h/hashing.h"
33 #endif
34 #if !defined(AFS_HPUX110_ENV) && !defined(AFS_LINUX20_ENV) && !defined(AFS_DARWIN_ENV)
35 #include <netinet/in_var.h>
36 #endif /* ! AFS_HPUX110_ENV */
37 #endif /* !defined(UKERNEL) */
38
39 #include "afsincludes.h" /* Afs-based standard headers */
40 #include "afs/afs_stats.h" /* afs statistics */
41 #include "afs/afs_dynroot.h"
42
43 #if defined(AFS_SUN5_ENV)
44 #include <inet/led.h>
45 #include <inet/common.h>
46 #include <netinet/ip6.h>
47 #include <inet/ip.h>
48 #endif
49
50 /* In case we don't have the vl error table yet. */
51 #ifndef ERROR_TABLE_BASE_VL
52 #define ERROR_TABLE_BASE_VL (363520L)
53 #define VL_NOENT (363524L)
54 #endif /* vlserver error base define */
55
56 /* Exported variables */
57 afs_int32 afs_volume_ttl = 0;
58 afs_dcache_id_t volumeInode; /* Inode for VolumeItems file */
59 afs_rwlock_t afs_xvolume; /** allocation lock for volumes */
60 struct volume *afs_freeVolList;
61 struct volume *afs_volumes[NVOLS];
62 afs_int32 afs_volCounter = 1; /** for allocating volume indices */
63 afs_int32 fvTable[NFENTRIES];
64
65 /* Forward declarations */
66 static struct volume *afs_NewVolumeByName(char *aname, afs_int32 acell,
67 int agood, struct vrequest *areq,
68 afs_int32 locktype);
69 static struct volume *afs_NewDynrootVolume(struct VenusFid *fid);
70 static int inVolList(struct VenusFid *fid, afs_int32 nvols, afs_int32 * vID,
71 afs_int32 * cID);
72
73
74
75 /**
76 * Convert a volume name to a number;
77 * @param aname Volume name.
78 * @return return 0 if can't parse as a number.
79 */
80 static int
81 afs_vtoi(char *aname)
82 {
83 afs_int32 temp;
84 int tc;
85 temp = 0;
86 AFS_STATCNT(afs_vtoi);
87 while ((tc = *aname++)) {
88 if (tc > '9' || tc < '0')
89 return 0; /* invalid name */
90 temp *= 10;
91 temp += tc - '0';
92 }
93 return temp;
94
95 } /*afs_vtoi */
96
97
98 /**
99 * All of the vol cache routines must be called with the afs_xvolume
100 * lock held in exclusive mode, since they use static variables.
101 * In addition, we don't want two people adding the same volume
102 * at the same time.
103 */
104
105 static struct fvolume staticFVolume;
106 afs_int32 afs_FVIndex = -1;
107
108 /*!
109 * Initialize a newly gotten volume slot.
110 *
111 * \param tv volume slot to be initialized
112 * \param tf volume item data; null if none
113 * \param volid volume id for this volume slot
114 * \param cell cell for this volume slot
115 * \return none
116 */
117 static void
118 afs_InitVolSlot(struct volume *tv, struct fvolume *tf, afs_int32 volid,
119 struct cell *tcell)
120 {
121 AFS_STATCNT(afs_InitVolSlot);
122 memset(tv, 0, sizeof(struct volume));
123 tv->cell = tcell->cellNum;
124 AFS_RWLOCK_INIT(&tv->lock, "volume lock");
125 tv->volume = volid;
126 if (tf) {
127 tv->vtix = afs_FVIndex;
128 tv->mtpoint = tf->mtpoint;
129 tv->dotdot = tf->dotdot;
130 tv->rootVnode = tf->rootVnode;
131 tv->rootUnique = tf->rootUnique;
132 } else {
133 tv->vtix = -1;
134 tv->rootVnode = tv->rootUnique = 0;
135 afs_GetDynrootMountFid(&tv->dotdot);
136 afs_GetDynrootMountFid(&tv->mtpoint);
137 tv->mtpoint.Fid.Vnode =
138 VNUM_FROM_TYPEID(VN_TYPE_MOUNT, tcell->cellIndex << 2);
139 tv->mtpoint.Fid.Unique = volid;
140 }
141 }
142
143 /**
144 * UFS specific version of afs_GetVolSlot
145 * @return
146 */
147 struct volume *
148 afs_UFSGetVolSlot(afs_int32 volid, struct cell *tcell)
149 {
150 struct volume *tv = NULL, **lv;
151 struct osi_file *tfile;
152 afs_int32 i = -1, code;
153 afs_int32 bestTime;
154 struct volume *bestVp, *oldLp = NULL, **bestLp = NULL;
155 char *oldname = NULL;
156 afs_int32 oldvtix = -2; /* Initialize to a value that doesn't occur */
157 struct fvolume *tf = NULL;
158 int j = 0;
159
160 AFS_STATCNT(afs_UFSGetVolSlot);
161 if (!afs_freeVolList) {
162 /* get free slot */
163 bestTime = 0x7fffffff;
164 bestVp = 0;
165 bestLp = 0;
166 for (i = 0; i < NVOLS; i++) {
167 lv = &afs_volumes[i];
168 for (tv = *lv; tv; lv = &tv->next, tv = *lv) {
169 if (tv->refCount == 0) { /* is this one available? */
170 if (tv->setupTime < bestTime) { /* best one available? */
171 bestTime = tv->setupTime;
172 bestLp = lv;
173 bestVp = tv;
174 }
175 }
176 }
177 }
178 if (!bestVp) {
179 afs_warn("afs_UFSGetVolSlot: no vol slots available\n");
180 goto error;
181 }
182 tv = bestVp;
183
184 oldLp = *bestLp;
185 *bestLp = tv->next;
186
187 oldname = tv->name;
188 tv->name = NULL;
189
190 oldvtix = tv->vtix;
191 /* now write out volume structure to file */
192 if (tv->vtix < 0) {
193 tv->vtix = afs_volCounter++;
194 /* now put on hash chain */
195 i = FVHash(tv->cell, tv->volume);
196 staticFVolume.next = fvTable[i];
197 fvTable[i] = tv->vtix;
198 } else {
199 /*
200 * Haul the guy in from disk so we don't overwrite hash table
201 * next chain
202 */
203 if (afs_FVIndex != tv->vtix) {
204 tfile = osi_UFSOpen(&volumeInode);
205 if (!tfile) {
206 afs_warn("afs_UFSGetVolSlot: unable to open volumeinfo\n");
207 goto error;
208 }
209 code =
210 afs_osi_Read(tfile, sizeof(struct fvolume) * tv->vtix,
211 &staticFVolume, sizeof(struct fvolume));
212 osi_UFSClose(tfile);
213 if (code != sizeof(struct fvolume)) {
214 afs_warn("afs_UFSGetVolSlot: error %d reading volumeinfo\n",
215 (int)code);
216 goto error;
217 }
218 }
219 }
220 afs_FVIndex = tv->vtix;
221 staticFVolume.volume = tv->volume;
222 staticFVolume.cell = tv->cell;
223 staticFVolume.mtpoint = tv->mtpoint;
224 staticFVolume.dotdot = tv->dotdot;
225 staticFVolume.rootVnode = tv->rootVnode;
226 staticFVolume.rootUnique = tv->rootUnique;
227 tfile = osi_UFSOpen(&volumeInode);
228 if (!tfile) {
229 afs_warn("afs_UFSGetVolSlot: unable to open volumeinfo\n");
230 goto error;
231 }
232 code =
233 afs_osi_Write(tfile, sizeof(struct fvolume) * afs_FVIndex,
234 &staticFVolume, sizeof(struct fvolume));
235 osi_UFSClose(tfile);
236 if (code != sizeof(struct fvolume)) {
237 afs_warn("afs_UFSGetVolSlot: error %d writing volumeinfo\n",
238 (int)code);
239 goto error;
240 }
241 if (oldname) {
242 afs_osi_Free(oldname, strlen(oldname) + 1);
243 oldname = NULL;
244 }
245 } else {
246 tv = afs_freeVolList;
247 afs_freeVolList = tv->next;
248 }
249
250 /* read volume item data from disk for the gotten slot */
251 for (j = fvTable[FVHash(tcell->cellNum, volid)]; j != 0; j = tf->next) {
252 if (afs_FVIndex != j) {
253 tfile = osi_UFSOpen(&volumeInode);
254 if (!tfile) {
255 afs_warn("afs_UFSGetVolSlot: unable to open volumeinfo\n");
256 code = -1; /* indicate error */
257 } else {
258 code =
259 afs_osi_Read(tfile, sizeof(struct fvolume) * j,
260 &staticFVolume, sizeof(struct fvolume));
261 osi_UFSClose(tfile);
262 if (code != sizeof(struct fvolume)) {
263 afs_warn("afs_UFSGetVolSlot: error %d reading volumeinfo\n",
264 (int)code);
265 }
266 }
267 if (code != sizeof(struct fvolume)) {
268 /* put tv back on the free list; the data in it is not valid */
269 tv->next = afs_freeVolList;
270 afs_freeVolList = tv;
271 /* staticFVolume contents are not valid */
272 afs_FVIndex = -1;
273 return NULL;
274 }
275 afs_FVIndex = j;
276 }
277 if (j != 0) { /* volume items record 0 is not used */
278 tf = &staticFVolume;
279 if (tf->cell == tcell->cellNum && tf->volume == volid) {
280 break;
281 }
282 }
283 }
284
285 afs_InitVolSlot(tv, tf, volid, tcell);
286 return tv;
287
288 error:
289 if (tv) {
290 if (oldvtix == -2) {
291 afs_warn("afs_UFSGetVolSlot: oldvtix is uninitialized\n");
292 return NULL;
293 }
294 if (oldname) {
295 tv->name = oldname;
296 oldname = NULL;
297 }
298 if (oldvtix < 0) {
299 afs_volCounter--;
300 fvTable[i] = staticFVolume.next;
301 }
302 if (bestLp) {
303 *bestLp = oldLp;
304 }
305 tv->vtix = oldvtix;
306 /* we messed with staticFVolume, so make sure someone else
307 * doesn't think it's fine to use */
308 afs_FVIndex = -1;
309 }
310 return NULL;
311 } /*afs_UFSGetVolSlot */
312
313
314 /**
315 * Get an available volume list slot. If the list does not exist,
316 * create one containing a single element.
317 * @return
318 */
319 struct volume *
320 afs_MemGetVolSlot(afs_int32 volid, struct cell *tcell)
321 {
322 struct volume *tv;
323
324 AFS_STATCNT(afs_MemGetVolSlot);
325 if (!afs_freeVolList) {
326 struct volume *newVp;
327
328 newVp = afs_osi_Alloc(sizeof(struct volume));
329 osi_Assert(newVp != NULL);
330
331 newVp->next = NULL;
332 afs_freeVolList = newVp;
333 }
334 tv = afs_freeVolList;
335 afs_freeVolList = tv->next;
336
337 afs_InitVolSlot(tv, NULL, volid, tcell);
338 return tv;
339
340 } /*afs_MemGetVolSlot */
341
342 /*!
343 * Setup a volume slot for cell:volume.
344 *
345 * Find the volume slot for the cell:volume, otherwise get
346 * and initialize a new slot.
347 *
348 * \param volid volume id
349 * \param cell cell
350 * \return volume
351 */
352 static struct volume *
353 afs_SetupVolSlot(afs_int32 volid, struct cell *tcell)
354 {
355 struct volume *tv;
356 int i;
357
358 AFS_STATCNT(afs_SetupVolSlot);
359 ObtainWriteLock(&afs_xvolume, 108);
360 i = VHash(volid);
361 for (tv = afs_volumes[i]; tv; tv = tv->next) {
362 if (tv->volume == volid && tv->cell == tcell->cellNum) {
363 break;
364 }
365 }
366 if (!tv) {
367 tv = afs_GetVolSlot(volid, tcell);
368 if (!tv) {
369 ReleaseWriteLock(&afs_xvolume);
370 return NULL;
371 }
372 tv->next = afs_volumes[i]; /* thread into list */
373 afs_volumes[i] = tv;
374 }
375 tv->refCount++;
376 tv->states &= ~VRecheck; /* just checked it */
377 tv->setupTime = osi_Time();
378 ReleaseWriteLock(&afs_xvolume);
379 return tv;
380 }
381
382 /*!
383 * Reset volume information for all volume structs that
384 * point to a speicific server, skipping a given volume if provided.
385 *
386 * @param[in] srvp
387 * The server to reset volume info about
388 * @param[in] tv
389 * The volume to skip resetting info about
390 */
391 void
392 afs_ResetVolumes(struct server *srvp, struct volume *tv)
393 {
394 int j, k;
395 struct volume *vp;
396
397 /* Find any volumes residing on this server and flush their state */
398 for (j = 0; j < NVOLS; j++) {
399 for (vp = afs_volumes[j]; vp; vp = vp->next) {
400 for (k = 0; k < AFS_MAXHOSTS; k++) {
401 if (!srvp || (vp->serverHost[k] == srvp)) {
402 if (tv && tv != vp) {
403 vp->serverHost[k] = 0;
404 afs_ResetVolumeInfo(vp);
405 }
406 break;
407 }
408 }
409 }
410 }
411 }
412
413 /*!
414 * Returns non-zero if the volume information is expired.
415 *
416 * Dynroot volumes are not setup from vldb queries, so never expire.
417 * Read-only volume expiry is tied to the volume callback.
418 *
419 * Optionally, invalidate volume information after a fixed timeout.
420 * The vlservers will be periodically probed for volume information.
421 * This avoids a situation where the vldb information is cached
422 * indefinitely as long as files in the volume are accessed (and are
423 * not in the vcache) before the callback expires.
424 *
425 * \param tv volume to check
426 * \param now current time
427 *
428 * \return non-zero if the volume should be reset
429 */
430 static int
431 IsExpired(struct volume *tv, afs_int32 now)
432 {
433 if (afs_IsDynrootVolume(tv)) {
434 return 0;
435 }
436 if ((tv->states & VRO) && (tv->expireTime < (now + 10))) {
437 return 1;
438 }
439 if ((afs_volume_ttl != 0) && ((tv->setupTime + afs_volume_ttl) < now)) {
440 return 1;
441 }
442 return 0;
443 }
444
445 /**
446 * Reset volume name to volume id mapping cache.
447 * @param flags
448 */
449 void
450 afs_CheckVolumeNames(int flags)
451 {
452 afs_int32 i, j;
453 struct volume *tv;
454 afs_int32 now;
455 struct vcache *tvc;
456 afs_int32 *volumeID, *cellID, vsize, nvols;
457 #ifdef AFS_DARWIN80_ENV
458 vnode_t tvp;
459 #endif
460 AFS_STATCNT(afs_CheckVolumeNames);
461
462 nvols = 0;
463 volumeID = cellID = NULL;
464 vsize = 0;
465 ObtainReadLock(&afs_xvolume);
466 if (flags & AFS_VOLCHECK_EXPIRED) {
467 /*
468 * allocate space to hold the volumeIDs and cellIDs, only if
469 * we will be invalidating the mountpoints later on
470 */
471 for (i = 0; i < NVOLS; i++)
472 for (tv = afs_volumes[i]; tv; tv = tv->next)
473 ++vsize;
474
475 volumeID = afs_osi_Alloc(2 * vsize * sizeof(*volumeID));
476 cellID = (volumeID) ? volumeID + vsize : 0;
477 }
478
479 now = osi_Time();
480 for (i = 0; i < NVOLS; i++) {
481 for (tv = afs_volumes[i]; tv; tv = tv->next) {
482 if (flags & AFS_VOLCHECK_EXPIRED) {
483 if (IsExpired(tv, now) || (flags & AFS_VOLCHECK_FORCE)) {
484 afs_ResetVolumeInfo(tv); /* also resets status */
485 if (volumeID) {
486 volumeID[nvols] = tv->volume;
487 cellID[nvols] = tv->cell;
488 }
489 ++nvols;
490 continue;
491 }
492 }
493 /* ??? */
494 if (flags & (AFS_VOLCHECK_BUSY | AFS_VOLCHECK_FORCE)) {
495 for (j = 0; j < AFS_MAXHOSTS; j++)
496 tv->status[j] = not_busy;
497 }
498
499 }
500 }
501 ReleaseReadLock(&afs_xvolume);
502
503
504 /* next ensure all mt points are re-evaluated */
505 if (nvols || (flags & (AFS_VOLCHECK_FORCE | AFS_VOLCHECK_MTPTS))) {
506 loop:
507 ObtainReadLock(&afs_xvcache);
508 for (i = 0; i < VCSIZE; i++) {
509 for (tvc = afs_vhashT[i]; tvc; tvc = tvc->hnext) {
510
511 /* if the volume of "mvid.target_root" of the vcache entry is
512 * among the ones we found earlier, then we re-evaluate it.
513 * Also, if the force bit is set or we explicitly asked to
514 * reevaluate the mt-pts, we clean the cmvalid bit */
515
516 if ((flags & (AFS_VOLCHECK_FORCE | AFS_VOLCHECK_MTPTS))
517 || (tvc->mvid.target_root
518 && inVolList(tvc->mvid.target_root, nvols, volumeID, cellID)))
519 tvc->f.states &= ~CMValid;
520
521 /* If the volume that this file belongs to was reset earlier,
522 * then we should remove its callback.
523 * Again, if forced, always do it.
524 */
525 if ((tvc->f.states & CRO)
526 && (inVolList(&tvc->f.fid, nvols, volumeID, cellID)
527 || (flags & AFS_VOLCHECK_FORCE))) {
528
529 if (tvc->f.states & CVInit) {
530 ReleaseReadLock(&afs_xvcache);
531 afs_osi_Sleep(&tvc->f.states);
532 goto loop;
533 }
534 #ifdef AFS_DARWIN80_ENV
535 if (tvc->f.states & CDeadVnode) {
536 ReleaseReadLock(&afs_xvcache);
537 afs_osi_Sleep(&tvc->f.states);
538 goto loop;
539 }
540 tvp = AFSTOV(tvc);
541 if (vnode_get(tvp))
542 continue;
543 if (vnode_ref(tvp)) {
544 AFS_GUNLOCK();
545 /* AFSTOV(tvc) may be NULL */
546 vnode_put(tvp);
547 AFS_GLOCK();
548 continue;
549 }
550 #else
551 AFS_FAST_HOLD(tvc);
552 #endif
553 ReleaseReadLock(&afs_xvcache);
554
555 /* LOCKXXX: We aren't holding tvc write lock? */
556 afs_StaleVCache(tvc);
557
558 #ifdef AFS_DARWIN80_ENV
559 vnode_put(AFSTOV(tvc));
560 /* our tvc ptr is still good until now */
561 AFS_FAST_RELE(tvc);
562 ObtainReadLock(&afs_xvcache);
563 #else
564 ObtainReadLock(&afs_xvcache);
565
566 /* our tvc ptr is still good until now */
567 AFS_FAST_RELE(tvc);
568 #endif
569 }
570 }
571 }
572 osi_dnlc_purge(); /* definitely overkill, but it's safer this way. */
573 ReleaseReadLock(&afs_xvcache);
574 }
575
576 if (volumeID)
577 afs_osi_Free(volumeID, 2 * vsize * sizeof(*volumeID));
578
579 } /*afs_CheckVolumeNames */
580
581
582 /**
583 * Check if volume is in the specified list.
584 * @param fid File FID.
585 * @param nvols Nomber of volumes???
586 * @param vID Array of volume IDs.
587 * @param cID Array of cache IDs.
588 * @return 1 - true, 0 - false.
589 */
590 static int
591 inVolList(struct VenusFid *fid, afs_int32 nvols, afs_int32 * vID,
592 afs_int32 * cID)
593 {
594 afs_int32 i;
595
596 /* if no arrays present, be conservative and return true */
597 if (nvols && (!vID || !cID))
598 return 1;
599
600 for (i = 0; i < nvols; ++i) {
601 if (fid->Fid.Volume == vID[i] && fid->Cell == cID[i])
602 return 1;
603 }
604 return 0;
605 }
606
607
608 /* afs_PutVolume is now a macro in afs.h */
609
610
611 /**
612 * Return volume struct if we have it cached and it's up-to-date.
613 * Environment: Must be called with afs_xvolume unlocked.
614 * @param afid Volume FID.
615 * @param locktype
616 * @return Volume or NULL if no result.
617 */
618 struct volume *
619 afs_FindVolume(struct VenusFid *afid, afs_int32 locktype)
620 {
621 struct volume *tv;
622 afs_int32 i;
623
624 if (afid == NULL)
625 return NULL;
626
627 i = VHash(afid->Fid.Volume);
628 ObtainWriteLock(&afs_xvolume, 106);
629 for (tv = afs_volumes[i]; tv; tv = tv->next) {
630 if (tv->volume == afid->Fid.Volume && tv->cell == afid->Cell
631 && (tv->states & VRecheck) == 0) {
632 tv->refCount++;
633 break;
634 }
635 }
636 ReleaseWriteLock(&afs_xvolume);
637 return tv; /* NULL if we didn't find it */
638 } /*afs_FindVolume */
639
640
641
642 /**
643 * Note that areq may be null, in which case we don't bother to set any
644 * request status information.
645 * @param afid Volume FID.
646 * @param areq Request type.
647 * @param locktype Lock to be used.
648 * @return Volume or NULL if no result.
649 */
650 struct volume *
651 afs_GetVolume(struct VenusFid *afid, struct vrequest *areq,
652 afs_int32 locktype)
653 {
654 struct volume *tv;
655 char *bp, tbuf[CVBS];
656 AFS_STATCNT(afs_GetVolume);
657
658 tv = afs_FindVolume(afid, locktype);
659 if (!tv) {
660 /* Do a dynroot check and add dynroot volume if found. */
661 if (afs_IsDynrootAnyFid(afid)) {
662 tv = afs_NewDynrootVolume(afid);
663 } else {
664 bp = afs_cv2string(&tbuf[CVBS], afid->Fid.Volume);
665 tv = afs_NewVolumeByName(bp, afid->Cell, 0, areq, locktype);
666 }
667 }
668 return tv;
669 } /*afs_GetVolume */
670
671
672
673 /**
674 *
675 * @param volid Volume ID. If it's 0, get it from the name.
676 * @param aname Volume name.
677 * @param ve Volume entry.
678 * @param tcell The cell containing this volume.
679 * @param agood
680 * @param type Type of volume.
681 * @param areq Request.
682 * @return Volume or NULL if failure.
683 */
684 static struct volume *
685 afs_SetupVolume(afs_int32 volid, char *aname, void *ve, struct cell *tcell,
686 afs_int32 agood, afs_int32 type, struct vrequest *areq)
687 {
688 struct volume *tv;
689 struct vldbentry *ove = (struct vldbentry *)ve;
690 struct nvldbentry *nve = (struct nvldbentry *)ve;
691 struct uvldbentry *uve = (struct uvldbentry *)ve;
692
693 int whichType; /* which type of volume to look for */
694 int i;
695
696 if (!volid) {
697 int len;
698 /* special hint from file server to use vlserver */
699 len = strlen(aname);
700 if (len >= 8 && strcmp(aname + len - 7, ".backup") == 0)
701 whichType = BACKVOL;
702 else if (len >= 10 && strcmp(aname + len - 9, ".readonly") == 0)
703 whichType = ROVOL;
704 else
705 whichType = RWVOL;
706
707 /* figure out which one we're really interested in (a set is returned) */
708 volid = afs_vtoi(aname);
709 if (volid == 0) {
710 if (type == 2) {
711 volid = uve->volumeId[whichType];
712 } else if (type == 1) {
713 volid = nve->volumeId[whichType];
714 } else {
715 volid = ove->volumeId[whichType];
716 }
717 } /* end of if (volid == 0) */
718 } /* end of if (!volid) */
719
720 tv = afs_SetupVolSlot(volid, tcell);
721 if (!tv) {
722 return NULL;
723 }
724
725 if (type == 2) {
726 LockAndInstallUVolumeEntry(tv, uve, tcell->cellNum, tcell, areq);
727 } else if (type == 1)
728 LockAndInstallNVolumeEntry(tv, nve, tcell->cellNum);
729 else
730 LockAndInstallVolumeEntry(tv, ove, tcell->cellNum);
731 if (agood) {
732 if (!tv->name) {
733 tv->name = afs_osi_Alloc(strlen(aname) + 1);
734 osi_Assert(tv->name != NULL);
735 strcpy(tv->name, aname);
736 }
737 }
738 for (i = 0; i < NMAXNSERVERS; i++) {
739 tv->status[i] = not_busy;
740 }
741 ReleaseWriteLock(&tv->lock);
742 return tv;
743 }
744
745
746 /**
747 * Seek volume by it's name and attributes.
748 * If volume not found, try to add one.
749 * @param aname Volume name.
750 * @param acell Cell
751 * @param agood
752 * @param areq
753 * @param locktype Type of lock to be used.
754 * @return
755 */
756 struct volume *
757 afs_GetVolumeByName(char *aname, afs_int32 acell, int agood,
758 struct vrequest *areq, afs_int32 locktype)
759 {
760 afs_int32 i;
761 struct volume *tv;
762
763 AFS_STATCNT(afs_GetVolumeByName);
764 ObtainWriteLock(&afs_xvolume, 112);
765 for (i = 0; i < NVOLS; i++) {
766 for (tv = afs_volumes[i]; tv; tv = tv->next) {
767 if (tv->name && !strcmp(aname, tv->name) && tv->cell == acell
768 && (tv->states & VRecheck) == 0) {
769 tv->refCount++;
770 ReleaseWriteLock(&afs_xvolume);
771 return tv;
772 }
773 }
774 }
775
776 ReleaseWriteLock(&afs_xvolume);
777
778 if (AFS_IS_DISCONNECTED)
779 return NULL;
780
781 tv = afs_NewVolumeByName(aname, acell, agood, areq, locktype);
782 return (tv);
783 }
784
785 /**
786 * Init a new dynroot volume.
787 * @param Volume FID.
788 * @return Volume or NULL if not found.
789 */
790 static struct volume *
791 afs_NewDynrootVolume(struct VenusFid *fid)
792 {
793 struct cell *tcell;
794 struct volume *tv;
795 struct vldbentry *tve;
796 char *bp, tbuf[CVBS];
797
798 tcell = afs_GetCell(fid->Cell, READ_LOCK);
799 if (!tcell)
800 return NULL;
801 tve = afs_osi_Alloc(sizeof(*tve));
802 osi_Assert(tve != NULL);
803 if (!(tcell->states & CHasVolRef))
804 tcell->states |= CHasVolRef;
805
806 bp = afs_cv2string(&tbuf[CVBS], fid->Fid.Volume);
807 memset(tve, 0, sizeof(*tve));
808 strcpy(tve->name, "local-dynroot");
809 tve->volumeId[ROVOL] = fid->Fid.Volume;
810 tve->flags = VLF_ROEXISTS;
811
812 tv = afs_SetupVolume(0, bp, tve, tcell, 0, 0, 0);
813 afs_PutCell(tcell, READ_LOCK);
814 afs_osi_Free(tve, sizeof(*tve));
815 return tv;
816 }
817
818 int lastnvcode;
819
820 /**
821 * @param aname Volume name.
822 * @param acell Cell id.
823 * @param agood
824 * @param areq Request type.
825 * @param locktype Type of lock to be used.
826 * @return Volume or NULL if failure.
827 */
828 static struct volume *
829 afs_NewVolumeByName(char *aname, afs_int32 acell, int agood,
830 struct vrequest *areq, afs_int32 locktype)
831 {
832 afs_int32 code, type = 0;
833 struct volume *tv, *tv1;
834 struct vldbentry *tve;
835 struct nvldbentry *ntve;
836 struct uvldbentry *utve;
837 struct cell *tcell;
838 char *tbuffer, *ve;
839 struct afs_conn *tconn;
840 struct vrequest *treq = NULL;
841 struct rx_connection *rxconn;
842
843 if (strlen(aname) > VL_MAXNAMELEN) /* Invalid volume name */
844 return NULL;
845
846 tcell = afs_GetCell(acell, READ_LOCK);
847 if (!tcell) {
848 return NULL;
849 }
850
851 code = afs_CreateReq(&treq, afs_osi_credp); /* *must* be unauth for vldb */
852 if (code) {
853 return NULL;
854 }
855
856 /* allow null request if we don't care about ENODEV/ETIMEDOUT distinction */
857 if (!areq)
858 areq = treq;
859
860
861 afs_Trace2(afs_iclSetp, CM_TRACE_GETVOL, ICL_TYPE_STRING, aname,
862 ICL_TYPE_POINTER, aname);
863 tbuffer = osi_AllocLargeSpace(AFS_LRALLOCSIZ);
864 tve = (struct vldbentry *)(tbuffer + 1024);
865 ntve = (struct nvldbentry *)tve;
866 utve = (struct uvldbentry *)tve;
867
868 do {
869 tconn =
870 afs_ConnByMHosts(tcell->cellHosts, tcell->vlport, tcell->cellNum,
871 treq, SHARED_LOCK, 0, &rxconn);
872 if (tconn) {
873 if (tconn->parent->srvr->server->flags & SNO_LHOSTS) {
874 type = 0;
875 RX_AFS_GUNLOCK();
876 code = VL_GetEntryByNameO(rxconn, aname, tve);
877 RX_AFS_GLOCK();
878 } else if (tconn->parent->srvr->server->flags & SYES_LHOSTS) {
879 type = 1;
880 RX_AFS_GUNLOCK();
881 code = VL_GetEntryByNameN(rxconn, aname, ntve);
882 RX_AFS_GLOCK();
883 } else {
884 type = 2;
885 RX_AFS_GUNLOCK();
886 code = VL_GetEntryByNameU(rxconn, aname, utve);
887 RX_AFS_GLOCK();
888 if (!(tconn->parent->srvr->server->flags & SVLSRV_UUID)) {
889 if (code == RXGEN_OPCODE) {
890 type = 1;
891 RX_AFS_GUNLOCK();
892 code = VL_GetEntryByNameN(rxconn, aname, ntve);
893 RX_AFS_GLOCK();
894 if (code == RXGEN_OPCODE) {
895 type = 0;
896 tconn->parent->srvr->server->flags |= SNO_LHOSTS;
897 RX_AFS_GUNLOCK();
898 code = VL_GetEntryByNameO(rxconn, aname, tve);
899 RX_AFS_GLOCK();
900 } else if (!code)
901 tconn->parent->srvr->server->flags |= SYES_LHOSTS;
902 } else if (!code)
903 tconn->parent->srvr->server->flags |= SVLSRV_UUID;
904 }
905 lastnvcode = code;
906 }
907 } else
908 code = -1;
909 } while (afs_Analyze(tconn, rxconn, code, NULL, treq, -1, /* no op code for this */
910 SHARED_LOCK, tcell));
911
912 if (code) {
913 /* If the client has yet to contact this cell and contact failed due
914 * to network errors, mark the VLDB servers as back up.
915 * That the client tried and failed can be determined from the
916 * fact that there was a downtime incident, but CHasVolRef is not set.
917 */
918 /* RT 48959 - unclear if this should really go */
919 #if 0
920 if (areq->networkError && !(tcell->states & CHasVolRef)) {
921 int i;
922 struct server *sp;
923 struct srvAddr *sap;
924 for (i = 0; i < AFS_MAXCELLHOSTS; i++) {
925 if ((sp = tcell->cellHosts[i]) == NULL)
926 break;
927 for (sap = sp->addr; sap; sap = sap->next_sa)
928 afs_MarkServerUpOrDown(sap, 0);
929 }
930 }
931 #endif
932 afs_CopyError(treq, areq);
933 osi_FreeLargeSpace(tbuffer);
934 afs_PutCell(tcell, READ_LOCK);
935 afs_DestroyReq(treq);
936 return NULL;
937 }
938 /*
939 * Check to see if this cell has not yet referenced a volume. If
940 * it hasn't, it's just about to change its status, and we need to mark
941 * this fact down. Note that it is remotely possible that afs_SetupVolume
942 * could fail and we would still not have a volume reference.
943 */
944 if (!(tcell->states & CHasVolRef)) {
945 tcell->states |= CHasVolRef;
946 afs_stats_cmperf.numCellsContacted++;
947 }
948 /*First time a volume in this cell has been referenced */
949 if (type == 2)
950 ve = (char *)utve;
951 else if (type == 1)
952 ve = (char *)ntve;
953 else
954 ve = (char *)tve;
955 tv = afs_SetupVolume(0, aname, ve, tcell, agood, type, treq);
956 if ((agood == 3) && tv && tv->backVol) {
957 /*
958 * This means that very soon we'll ask for the BK volume so
959 * we'll prefetch it (well we did already.)
960 */
961 tv1 =
962 afs_SetupVolume(tv->backVol, (char *)0, ve, tcell, 0, type, treq);
963 if (tv1) {
964 tv1->refCount--;
965 }
966 }
967 if ((agood >= 2) && tv && tv->roVol) {
968 /*
969 * This means that very soon we'll ask for the RO volume so
970 * we'll prefetch it (well we did already.)
971 */
972 tv1 = afs_SetupVolume(tv->roVol, NULL, ve, tcell, 0, type, treq);
973 if (tv1) {
974 tv1->refCount--;
975 }
976 }
977 osi_FreeLargeSpace(tbuffer);
978 afs_PutCell(tcell, READ_LOCK);
979 afs_DestroyReq(treq);
980 return tv;
981
982 } /*afs_NewVolumeByName */
983
984
985
986 /**
987 * Call this with the volume structure locked; used for new-style vldb requests.
988 * @param av Volume
989 * @param ve
990 * @param acell
991 */
992 void
993 LockAndInstallVolumeEntry(struct volume *av, struct vldbentry *ve, int acell)
994 {
995 struct server *ts;
996 struct cell *cellp;
997 int i, j;
998 afs_int32 mask;
999 afs_uint32 temp;
1000 char types = 0;
1001 struct server *serverHost[AFS_MAXHOSTS];
1002
1003 AFS_STATCNT(InstallVolumeEntry);
1004
1005 memset(serverHost, 0, sizeof(serverHost));
1006
1007 /* Determine the type of volume we want */
1008 if ((ve->flags & VLF_RWEXISTS) && (av->volume == ve->volumeId[RWVOL])) {
1009 mask = VLSF_RWVOL;
1010 } else if ((ve->flags & VLF_ROEXISTS)
1011 && (av->volume == ve->volumeId[ROVOL])) {
1012 mask = VLSF_ROVOL;
1013 types |= VRO;
1014 } else if ((ve->flags & VLF_BACKEXISTS)
1015 && (av->volume == ve->volumeId[BACKVOL])) {
1016 /* backup always is on the same volume as parent */
1017 mask = VLSF_RWVOL;
1018 types |= (VRO | VBackup);
1019 } else {
1020 mask = 0; /* Can't find volume in vldb entry */
1021 }
1022
1023 cellp = afs_GetCell(acell, 0);
1024
1025 /* Step through the VLDB entry making sure each server listed is there */
1026 for (i = 0, j = 0; i < ve->nServers; i++) {
1027 if (((ve->serverFlags[i] & mask) == 0)
1028 || (ve->serverFlags[i] & VLSF_DONTUSE)) {
1029 continue; /* wrong volume or don't use this volume */
1030 }
1031
1032 temp = htonl(ve->serverNumber[i]);
1033 ts = afs_GetServer(&temp, 1, acell, cellp->fsport, WRITE_LOCK,
1034 (afsUUID *) 0, 0, av);
1035 serverHost[j] = ts;
1036
1037 /*
1038 * The cell field could be 0 if the server entry was created
1039 * first with the 'fs setserverprefs' call which doesn't set
1040 * the cell field. Thus if the afs_GetServer call above
1041 * follows later on it will find the server entry thus it will
1042 * simply return without setting any fields, so we set the
1043 * field ourselves below.
1044 */
1045 if (!ts->cell)
1046 ts->cell = cellp;
1047 afs_PutServer(ts, WRITE_LOCK);
1048 j++;
1049 }
1050
1051 ObtainWriteLock(&av->lock, 109);
1052
1053 memcpy(av->serverHost, serverHost, sizeof(serverHost));
1054
1055 /* from above */
1056 av->states |= types;
1057
1058 /* fill in volume types */
1059 av->rwVol = ((ve->flags & VLF_RWEXISTS) ? ve->volumeId[RWVOL] : 0);
1060 av->roVol = ((ve->flags & VLF_ROEXISTS) ? ve->volumeId[ROVOL] : 0);
1061 av->backVol = ((ve->flags & VLF_BACKEXISTS) ? ve->volumeId[BACKVOL] : 0);
1062
1063 if (ve->flags & VLF_DFSFILESET)
1064 av->states |= VForeign;
1065
1066 afs_SortServers(av->serverHost, AFS_MAXHOSTS);
1067 } /*InstallVolumeEntry */
1068
1069
1070 void
1071 LockAndInstallNVolumeEntry(struct volume *av, struct nvldbentry *ve, int acell)
1072 {
1073 struct server *ts;
1074 struct cell *cellp;
1075 int i, j;
1076 afs_int32 mask;
1077 afs_uint32 temp;
1078 char types = 0;
1079 struct server *serverHost[AFS_MAXHOSTS];
1080
1081 AFS_STATCNT(InstallVolumeEntry);
1082
1083 memset(serverHost, 0, sizeof(serverHost));
1084
1085 /* Determine type of volume we want */
1086 if ((ve->flags & VLF_RWEXISTS) && (av->volume == ve->volumeId[RWVOL])) {
1087 mask = VLSF_RWVOL;
1088 } else if ((ve->flags & VLF_ROEXISTS)
1089 && (av->volume == ve->volumeId[ROVOL])) {
1090 mask = VLSF_ROVOL;
1091 types |= VRO;
1092 } else if ((ve->flags & VLF_BACKEXISTS)
1093 && (av->volume == ve->volumeId[BACKVOL])) {
1094 /* backup always is on the same volume as parent */
1095 mask = VLSF_RWVOL;
1096 types |= (VRO | VBackup);
1097 } else {
1098 mask = 0; /* Can't find volume in vldb entry */
1099 }
1100
1101 cellp = afs_GetCell(acell, 0);
1102
1103 /* Step through the VLDB entry making sure each server listed is there */
1104 for (i = 0, j = 0; i < ve->nServers; i++) {
1105 if (((ve->serverFlags[i] & mask) == 0)
1106 || (ve->serverFlags[i] & VLSF_DONTUSE)) {
1107 continue; /* wrong volume or don't use this volume */
1108 }
1109
1110 temp = htonl(ve->serverNumber[i]);
1111 ts = afs_GetServer(&temp, 1, acell, cellp->fsport, WRITE_LOCK,
1112 (afsUUID *) 0, 0, av);
1113 serverHost[j] = ts;
1114 /*
1115 * The cell field could be 0 if the server entry was created
1116 * first with the 'fs setserverprefs' call which doesn't set
1117 * the cell field. Thus if the afs_GetServer call above
1118 * follows later on it will find the server entry thus it will
1119 * simply return without setting any fields, so we set the
1120 * field ourselves below.
1121 */
1122 if (!ts->cell)
1123 ts->cell = cellp;
1124 afs_PutServer(ts, WRITE_LOCK);
1125 j++;
1126 }
1127
1128 ObtainWriteLock(&av->lock, 110);
1129
1130 memcpy(av->serverHost, serverHost, sizeof(serverHost));
1131
1132 /* from above */
1133 av->states |= types;
1134
1135 /* fill in volume types */
1136 av->rwVol = ((ve->flags & VLF_RWEXISTS) ? ve->volumeId[RWVOL] : 0);
1137 av->roVol = ((ve->flags & VLF_ROEXISTS) ? ve->volumeId[ROVOL] : 0);
1138 av->backVol = ((ve->flags & VLF_BACKEXISTS) ? ve->volumeId[BACKVOL] : 0);
1139
1140 if (ve->flags & VLF_DFSFILESET)
1141 av->states |= VForeign;
1142
1143 afs_SortServers(av->serverHost, AFS_MAXHOSTS);
1144 } /*InstallNVolumeEntry */
1145
1146
1147 void
1148 LockAndInstallUVolumeEntry(struct volume *av, struct uvldbentry *ve, int acell,
1149 struct cell *tcell, struct vrequest *areq)
1150 {
1151 struct server *ts;
1152 struct afs_conn *tconn;
1153 struct cell *cellp;
1154 int i, j;
1155 afs_uint32 serverid;
1156 afs_int32 mask;
1157 int k;
1158 char type = 0;
1159 struct server *serverHost[AFS_MAXHOSTS];
1160
1161 AFS_STATCNT(InstallVolumeEntry);
1162
1163 memset(serverHost, 0, sizeof(serverHost));
1164
1165 /* Determine type of volume we want */
1166 if ((ve->flags & VLF_RWEXISTS) && (av->volume == ve->volumeId[RWVOL])) {
1167 mask = VLSF_RWVOL;
1168 } else if ((ve->flags & VLF_ROEXISTS)
1169 && av->volume == ve->volumeId[ROVOL]) {
1170 mask = VLSF_ROVOL;
1171 type |= VRO;
1172 } else if ((ve->flags & VLF_BACKEXISTS)
1173 && (av->volume == ve->volumeId[BACKVOL])) {
1174 /* backup always is on the same volume as parent */
1175 mask = VLSF_RWVOL;
1176 type |= (VRO | VBackup);
1177 } else {
1178 mask = 0; /* Can't find volume in vldb entry */
1179 }
1180
1181 cellp = afs_GetCell(acell, 0);
1182
1183 /* Gather the list of servers the VLDB says the volume is on
1184 * and initialize the ve->serverHost[] array. If a server struct
1185 * is not found, then get the list of addresses for the
1186 * server, VL_GetAddrsU(), and create a server struct, afs_GetServer().
1187 */
1188 for (i = 0, j = 0; i < ve->nServers; i++) {
1189 if (((ve->serverFlags[i] & mask) == 0)
1190 || (ve->serverFlags[i] & VLSF_DONTUSE)) {
1191 continue; /* wrong volume don't use this volume */
1192 }
1193
1194 if (!(ve->serverFlags[i] & VLSF_UUID)) {
1195 /* The server has no uuid */
1196 serverid = htonl(ve->serverNumber[i].time_low);
1197 ts = afs_GetServer(&serverid, 1, acell, cellp->fsport,
1198 WRITE_LOCK, (afsUUID *) 0, 0, av);
1199 } else {
1200 ts = afs_FindServer(0, cellp->fsport, &ve->serverNumber[i], 0);
1201 if (ts && (ts->sr_addr_uniquifier == ve->serverUnique[i])
1202 && ts->addr) {
1203 /* uuid, uniquifier, and portal are the same */
1204 } else {
1205 afs_uint32 *addrp, code;
1206 afs_int32 nentries, unique;
1207 bulkaddrs addrs;
1208 ListAddrByAttributes attrs;
1209 afsUUID uuid;
1210 struct rx_connection *rxconn;
1211
1212 memset(&attrs, 0, sizeof(attrs));
1213 attrs.Mask = VLADDR_UUID;
1214 attrs.uuid = ve->serverNumber[i];
1215 memset(&uuid, 0, sizeof(uuid));
1216 memset(&addrs, 0, sizeof(addrs));
1217 do {
1218 tconn =
1219 afs_ConnByMHosts(tcell->cellHosts, tcell->vlport,
1220 tcell->cellNum, areq, SHARED_LOCK,
1221 0, &rxconn);
1222 if (tconn) {
1223 RX_AFS_GUNLOCK();
1224 code =
1225 VL_GetAddrsU(rxconn, &attrs, &uuid, &unique,
1226 &nentries, &addrs);
1227 RX_AFS_GLOCK();
1228 } else {
1229 code = -1;
1230 }
1231
1232 /* Handle corrupt VLDB (defect 7393) */
1233 if (code == 0 && nentries == 0)
1234 code = VL_NOENT;
1235
1236 } while (afs_Analyze
1237 (tconn, rxconn, code, NULL, areq, -1, SHARED_LOCK, tcell));
1238 if (code) {
1239 /* Better handing of such failures; for now we'll simply retry this call */
1240 areq->volumeError = 1;
1241 return;
1242 }
1243
1244 addrp = addrs.bulkaddrs_val;
1245 for (k = 0; k < nentries; k++) {
1246 addrp[k] = htonl(addrp[k]);
1247 }
1248 ts = afs_GetServer(addrp, nentries, acell,
1249 cellp->fsport, WRITE_LOCK,
1250 &ve->serverNumber[i],
1251 ve->serverUnique[i], av);
1252 xdr_free((xdrproc_t) xdr_bulkaddrs, &addrs);
1253 }
1254 }
1255 serverHost[j] = ts;
1256
1257 /* The cell field could be 0 if the server entry was created
1258 * first with the 'fs setserverprefs' call which doesn't set
1259 * the cell field. Thus if the afs_GetServer call above
1260 * follows later on it will find the server entry thus it will
1261 * simply return without setting any fields, so we set the
1262 * field ourselves below.
1263 */
1264 if (!ts->cell)
1265 ts->cell = cellp;
1266 afs_PutServer(ts, WRITE_LOCK);
1267 j++;
1268 }
1269
1270 ObtainWriteLock(&av->lock, 111);
1271
1272 memcpy(av->serverHost, serverHost, sizeof(serverHost));
1273
1274 /* from above */
1275 av->states |= type;
1276
1277 /* fill in volume types */
1278 av->rwVol = ((ve->flags & VLF_RWEXISTS) ? ve->volumeId[RWVOL] : 0);
1279 av->roVol = ((ve->flags & VLF_ROEXISTS) ? ve->volumeId[ROVOL] : 0);
1280 av->backVol = ((ve->flags & VLF_BACKEXISTS) ? ve->volumeId[BACKVOL] : 0);
1281
1282 if (ve->flags & VLF_DFSFILESET)
1283 av->states |= VForeign;
1284
1285 afs_SortServers(av->serverHost, AFS_MAXHOSTS);
1286 } /*InstallVolumeEntry */
1287
1288
1289 /**
1290 * Reset volume info for the specified volume strecture. Mark volume
1291 * to be rechecked next time.
1292 * @param tv
1293 */
1294 void
1295 afs_ResetVolumeInfo(struct volume *tv)
1296 {
1297 int i;
1298
1299 AFS_STATCNT(afs_ResetVolumeInfo);
1300 ObtainWriteLock(&tv->lock, 117);
1301 tv->states |= VRecheck;
1302
1303 /* the hard-mount code in afs_Analyze may not be able to reset this flag
1304 * when VRecheck is set, so clear it here to ensure it gets cleared. */
1305 tv->states &= ~VHardMount;
1306
1307 for (i = 0; i < AFS_MAXHOSTS; i++)
1308 tv->status[i] = not_busy;
1309 if (tv->name) {
1310 afs_osi_Free(tv->name, strlen(tv->name) + 1);
1311 tv->name = NULL;
1312 }
1313 ReleaseWriteLock(&tv->lock);
1314 }