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
9 * Portions Copyright (c) 2005-2008 Sine Nomine Associates
12 /* 1/1/89: NB: this stuff is all going to be replaced. Don't take it too seriously */
17 Institution: The Information Technology Center, Carnegie-Mellon University
21 #include <afsconfig.h>
22 #include <afs/param.h>
28 #include <afs/afsint.h>
31 #include <sys/param.h>
32 #if !defined(AFS_SGI_ENV)
35 #else /* AFS_OSF_ENV */
36 #ifdef AFS_VFSINCL_ENV
39 #include <sys/fs/ufs_fs.h>
41 #if defined(AFS_DARWIN_ENV) || defined(AFS_XBSD_ENV)
42 #include <ufs/ufs/dinode.h>
43 #include <ufs/ffs/fs.h>
48 #else /* AFS_VFSINCL_ENV */
49 #if !defined(AFS_AIX_ENV) && !defined(AFS_LINUX20_ENV) && !defined(AFS_XBSD_ENV)
52 #endif /* AFS_VFSINCL_ENV */
53 #endif /* AFS_OSF_ENV */
54 #endif /* AFS_SGI_ENV */
55 #endif /* AFS_NT40_ENV */
73 #if defined(AFS_SUN_ENV) || defined(AFS_SUN5_ENV)
75 #include <sys/mnttab.h>
76 #include <sys/mntent.h>
82 #if defined(AFS_SGI_ENV)
87 #ifndef AFS_LINUX20_ENV
88 #include <fstab.h> /* Need to find in libc 5, present in libc 6 */
91 #endif /* AFS_SGI_ENV */
93 #endif /* AFS_HPUX_ENV */
97 #include <netinet/in.h>
101 #include <sys/time.h>
102 #endif /* ITIMER_REAL */
103 #endif /* AFS_NT40_ENV */
104 #if defined(AFS_SUN5_ENV) || defined(AFS_NT40_ENV) || defined(AFS_LINUX20_ENV)
111 #include <afs/errors.h>
114 #include <afs/afssyscalls.h>
116 #include <afs/afsutil.h>
120 #include "daemon_com.h"
122 #include "salvsync.h"
125 #include "partition.h"
126 #include "volume_inline.h"
127 #ifdef AFS_PTHREAD_ENV
129 #else /* AFS_PTHREAD_ENV */
130 #include "afs/assert.h"
131 #endif /* AFS_PTHREAD_ENV */
138 #if !defined(offsetof)
143 #define afs_stat stat64
144 #define afs_fstat fstat64
145 #define afs_open open64
146 #else /* !O_LARGEFILE */
147 #define afs_stat stat
148 #define afs_fstat fstat
149 #define afs_open open
150 #endif /* !O_LARGEFILE */
152 #ifdef AFS_PTHREAD_ENV
153 pthread_mutex_t vol_glock_mutex;
154 pthread_mutex_t vol_trans_mutex;
155 pthread_cond_t vol_put_volume_cond;
156 pthread_cond_t vol_sleep_cond;
157 int vol_attach_threads = 1;
158 #endif /* AFS_PTHREAD_ENV */
160 #ifdef AFS_DEMAND_ATTACH_FS
161 pthread_mutex_t vol_salvsync_mutex;
162 #endif /* AFS_DEMAND_ATTACH_FS */
165 extern void *calloc(), *realloc();
168 /*@printflike@*/ extern void Log(const char *format, ...);
170 /* Forward declarations */
171 static Volume *attach2(Error * ec, VolId vid, char *path,
172 register struct VolumeHeader *header,
173 struct DiskPartition64 *partp, Volume * vp,
174 int isbusy, int mode);
175 static void ReallyFreeVolume(Volume * vp);
176 #ifdef AFS_DEMAND_ATTACH_FS
177 static void FreeVolume(Volume * vp);
178 #else /* !AFS_DEMAND_ATTACH_FS */
179 #define FreeVolume(vp) ReallyFreeVolume(vp)
180 static void VScanUpdateList(void);
181 #endif /* !AFS_DEMAND_ATTACH_FS */
182 static void VInitVolumeHeaderCache(afs_uint32 howMany);
183 static int GetVolumeHeader(register Volume * vp);
184 static void ReleaseVolumeHeader(register struct volHeader *hd);
185 static void FreeVolumeHeader(register Volume * vp);
186 static void AddVolumeToHashTable(register Volume * vp, int hashid);
187 static void DeleteVolumeFromHashTable(register Volume * vp);
188 static int VHold(Volume * vp);
189 static int VHold_r(Volume * vp);
190 static void VGetBitmap_r(Error * ec, Volume * vp, VnodeClass class);
191 static void GetVolumePath(Error * ec, VolId volumeId, char **partitionp,
193 static void VReleaseVolumeHandles_r(Volume * vp);
194 static void VCloseVolumeHandles_r(Volume * vp);
195 static void LoadVolumeHeader(Error * ec, Volume * vp);
196 static int VCheckOffline(register Volume * vp);
197 static int VCheckDetach(register Volume * vp);
198 static Volume * GetVolume(Error * ec, Error * client_ec, VolId volumeId, Volume * hint, int flags);
199 static int VolumeExternalName_r(VolumeId volumeId, char * name, size_t len);
201 int LogLevel; /* Vice loglevel--not defined as extern so that it will be
202 * defined when not linked with vice, XXXX */
203 ProgramType programType; /* The type of program using the package */
205 /* extended volume package statistics */
208 #ifdef VOL_LOCK_DEBUG
209 pthread_t vol_glock_holder = 0;
213 #define VOLUME_BITMAP_GROWSIZE 16 /* bytes, => 128vnodes */
214 /* Must be a multiple of 4 (1 word) !! */
216 /* this parameter needs to be tunable at runtime.
217 * 128 was really inadequate for largish servers -- at 16384 volumes this
218 * puts average chain length at 128, thus an average 65 deref's to find a volptr.
219 * talk about bad spatial locality...
221 * an AVL or splay tree might work a lot better, but we'll just increase
222 * the default hash table size for now
224 #define DEFAULT_VOLUME_HASH_SIZE 256 /* Must be a power of 2!! */
225 #define DEFAULT_VOLUME_HASH_MASK (DEFAULT_VOLUME_HASH_SIZE-1)
226 #define VOLUME_HASH(volumeId) (volumeId&(VolumeHashTable.Mask))
229 * turn volume hash chains into partially ordered lists.
230 * when the threshold is exceeded between two adjacent elements,
231 * perform a chain rebalancing operation.
233 * keep the threshold high in order to keep cache line invalidates
234 * low "enough" on SMPs
236 #define VOLUME_HASH_REORDER_THRESHOLD 200
239 * when possible, don't just reorder single elements, but reorder
240 * entire chains of elements at once. a chain of elements that
241 * exceed the element previous to the pivot by at least CHAIN_THRESH
242 * accesses are moved in front of the chain whose elements have at
243 * least CHAIN_THRESH less accesses than the pivot element
245 #define VOLUME_HASH_REORDER_CHAIN_THRESH (VOLUME_HASH_REORDER_THRESHOLD / 2)
247 #include "rx/rx_queue.h"
250 VolumeHashTable_t VolumeHashTable = {
251 DEFAULT_VOLUME_HASH_SIZE,
252 DEFAULT_VOLUME_HASH_MASK,
257 static void VInitVolumeHash(void);
261 /* This macro is used where an ffs() call does not exist. Was in util/ffs.c */
265 afs_int32 ffs_tmp = x;
269 for (ffs_i = 1;; ffs_i++) {
276 #endif /* !AFS_HAVE_FFS */
278 #ifdef AFS_PTHREAD_ENV
279 typedef struct diskpartition_queue_t {
280 struct rx_queue queue;
281 struct DiskPartition64 * diskP;
282 } diskpartition_queue_t;
283 typedef struct vinitvolumepackage_thread_t {
284 struct rx_queue queue;
285 pthread_cond_t thread_done_cv;
286 int n_threads_complete;
287 } vinitvolumepackage_thread_t;
288 static void * VInitVolumePackageThread(void * args);
289 #endif /* AFS_PTHREAD_ENV */
291 static int VAttachVolumesByPartition(struct DiskPartition64 *diskP,
292 int * nAttached, int * nUnattached);
295 #ifdef AFS_DEMAND_ATTACH_FS
296 /* demand attach fileserver extensions */
299 * in the future we will support serialization of VLRU state into the fs_state
302 * these structures are the beginning of that effort
304 struct VLRU_DiskHeader {
305 struct versionStamp stamp; /* magic and structure version number */
306 afs_uint32 mtime; /* time of dump to disk */
307 afs_uint32 num_records; /* number of VLRU_DiskEntry records */
310 struct VLRU_DiskEntry {
311 afs_uint32 vid; /* volume ID */
312 afs_uint32 idx; /* generation */
313 afs_uint32 last_get; /* timestamp of last get */
316 struct VLRU_StartupQueue {
317 struct VLRU_DiskEntry * entry;
322 typedef struct vshutdown_thread_t {
324 pthread_mutex_t lock;
326 pthread_cond_t master_cv;
328 int n_threads_complete;
330 int schedule_version;
333 byte n_parts_done_pass;
334 byte part_thread_target[VOLMAXPARTS+1];
335 byte part_done_pass[VOLMAXPARTS+1];
336 struct rx_queue * part_pass_head[VOLMAXPARTS+1];
337 int stats[4][VOLMAXPARTS+1];
338 } vshutdown_thread_t;
339 static void * VShutdownThread(void * args);
342 static Volume * VAttachVolumeByVp_r(Error * ec, Volume * vp, int mode);
343 static int VCheckFree(Volume * vp);
346 static void AddVolumeToVByPList_r(Volume * vp);
347 static void DeleteVolumeFromVByPList_r(Volume * vp);
348 static void VVByPListBeginExclusive_r(struct DiskPartition64 * dp);
349 static void VVByPListEndExclusive_r(struct DiskPartition64 * dp);
350 static void VVByPListWait_r(struct DiskPartition64 * dp);
352 /* online salvager */
353 static int VCheckSalvage(register Volume * vp);
354 static int VUpdateSalvagePriority_r(Volume * vp);
355 static int VScheduleSalvage_r(Volume * vp);
356 static int VCancelSalvage_r(Volume * vp, int reason);
358 /* Volume hash table */
359 static void VReorderHash_r(VolumeHashChainHead * head, Volume * pp, Volume * vp);
360 static void VHashBeginExclusive_r(VolumeHashChainHead * head);
361 static void VHashEndExclusive_r(VolumeHashChainHead * head);
362 static void VHashWait_r(VolumeHashChainHead * head);
365 static int ShutdownVByPForPass_r(struct DiskPartition64 * dp, int pass);
366 static int ShutdownVolumeWalk_r(struct DiskPartition64 * dp, int pass,
367 struct rx_queue ** idx);
368 static void ShutdownController(vshutdown_thread_t * params);
369 static void ShutdownCreateSchedule(vshutdown_thread_t * params);
372 static void VLRU_ComputeConstants(void);
373 static void VInitVLRU(void);
374 static void VLRU_Init_Node_r(volatile Volume * vp);
375 static void VLRU_Add_r(volatile Volume * vp);
376 static void VLRU_Delete_r(volatile Volume * vp);
377 static void VLRU_UpdateAccess_r(volatile Volume * vp);
378 static void * VLRU_ScannerThread(void * args);
379 static void VLRU_Scan_r(int idx);
380 static void VLRU_Promote_r(int idx);
381 static void VLRU_Demote_r(int idx);
382 static void VLRU_SwitchQueues(volatile Volume * vp, int new_idx, int append);
385 static int VCheckSoftDetach(volatile Volume * vp, afs_uint32 thresh);
386 static int VCheckSoftDetachCandidate(volatile Volume * vp, afs_uint32 thresh);
387 static int VSoftDetachVolume_r(volatile Volume * vp, afs_uint32 thresh);
388 #endif /* AFS_DEMAND_ATTACH_FS */
391 struct Lock vol_listLock; /* Lock obtained when listing volumes:
392 * prevents a volume from being missed
393 * if the volume is attached during a
397 static int TimeZoneCorrection; /* Number of seconds west of GMT */
399 /* Common message used when the volume goes off line */
400 char *VSalvageMessage =
401 "Files in this volume are currently unavailable; call operations";
403 int VInit; /* 0 - uninitialized,
404 * 1 - initialized but not all volumes have been attached,
405 * 2 - initialized and all volumes have been attached,
406 * 3 - initialized, all volumes have been attached, and
407 * VConnectFS() has completed. */
410 bit32 VolumeCacheCheck; /* Incremented everytime a volume goes on line--
411 * used to stamp volume headers and in-core
412 * vnodes. When the volume goes on-line the
413 * vnode will be invalidated
414 * access only with VOL_LOCK held */
419 /***************************************************/
420 /* Startup routines */
421 /***************************************************/
424 VInitVolumePackage(ProgramType pt, afs_uint32 nLargeVnodes, afs_uint32 nSmallVnodes,
425 int connect, afs_uint32 volcache)
427 int errors = 0; /* Number of errors while finding vice partitions. */
433 memset(&VStats, 0, sizeof(VStats));
434 VStats.hdr_cache_size = 200;
436 VInitPartitionPackage();
438 #ifdef AFS_DEMAND_ATTACH_FS
439 if (programType == fileServer) {
442 VLRU_SetOptions(VLRU_SET_ENABLED, 0);
446 #ifdef AFS_PTHREAD_ENV
447 assert(pthread_mutex_init(&vol_glock_mutex, NULL) == 0);
448 assert(pthread_mutex_init(&vol_trans_mutex, NULL) == 0);
449 assert(pthread_cond_init(&vol_put_volume_cond, NULL) == 0);
450 assert(pthread_cond_init(&vol_sleep_cond, NULL) == 0);
451 #else /* AFS_PTHREAD_ENV */
453 #endif /* AFS_PTHREAD_ENV */
454 Lock_Init(&vol_listLock);
456 srandom(time(0)); /* For VGetVolumeInfo */
457 gettimeofday(&tv, &tz);
458 TimeZoneCorrection = tz.tz_minuteswest * 60;
460 #ifdef AFS_DEMAND_ATTACH_FS
461 assert(pthread_mutex_init(&vol_salvsync_mutex, NULL) == 0);
462 #endif /* AFS_DEMAND_ATTACH_FS */
464 /* Ok, we have done enough initialization that fileserver can
465 * start accepting calls, even though the volumes may not be
466 * available just yet.
470 #if defined(AFS_DEMAND_ATTACH_FS) && defined(SALVSYNC_BUILD_SERVER)
471 if (programType == salvageServer) {
474 #endif /* AFS_DEMAND_ATTACH_FS */
475 #ifdef FSSYNC_BUILD_SERVER
476 if (programType == fileServer) {
480 #if defined(AFS_DEMAND_ATTACH_FS) && defined(SALVSYNC_BUILD_CLIENT)
481 if (programType == fileServer) {
482 /* establish a connection to the salvager at this point */
483 assert(VConnectSALV() != 0);
485 #endif /* AFS_DEMAND_ATTACH_FS */
487 if (volcache > VStats.hdr_cache_size)
488 VStats.hdr_cache_size = volcache;
489 VInitVolumeHeaderCache(VStats.hdr_cache_size);
491 VInitVnodes(vLarge, nLargeVnodes);
492 VInitVnodes(vSmall, nSmallVnodes);
495 errors = VAttachPartitions();
499 if (programType == fileServer) {
500 struct DiskPartition64 *diskP;
501 #ifdef AFS_PTHREAD_ENV
502 struct vinitvolumepackage_thread_t params;
503 struct diskpartition_queue_t * dpq;
504 int i, threads, parts;
506 pthread_attr_t attrs;
508 assert(pthread_cond_init(¶ms.thread_done_cv,NULL) == 0);
510 params.n_threads_complete = 0;
512 /* create partition work queue */
513 for (parts=0, diskP = DiskPartitionList; diskP; diskP = diskP->next, parts++) {
514 dpq = (diskpartition_queue_t *) malloc(sizeof(struct diskpartition_queue_t));
517 queue_Append(¶ms,dpq);
520 threads = MIN(parts, vol_attach_threads);
523 /* spawn off a bunch of initialization threads */
524 assert(pthread_attr_init(&attrs) == 0);
525 assert(pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED) == 0);
527 Log("VInitVolumePackage: beginning parallel fileserver startup\n");
528 #ifdef AFS_DEMAND_ATTACH_FS
529 Log("VInitVolumePackage: using %d threads to pre-attach volumes on %d partitions\n",
531 #else /* AFS_DEMAND_ATTACH_FS */
532 Log("VInitVolumePackage: using %d threads to attach volumes on %d partitions\n",
534 #endif /* AFS_DEMAND_ATTACH_FS */
537 for (i=0; i < threads; i++) {
538 assert(pthread_create
539 (&tid, &attrs, &VInitVolumePackageThread,
543 while(params.n_threads_complete < threads) {
544 VOL_CV_WAIT(¶ms.thread_done_cv);
548 assert(pthread_attr_destroy(&attrs) == 0);
550 /* if we're only going to run one init thread, don't bother creating
552 Log("VInitVolumePackage: beginning single-threaded fileserver startup\n");
553 #ifdef AFS_DEMAND_ATTACH_FS
554 Log("VInitVolumePackage: using 1 thread to pre-attach volumes on %d partition(s)\n",
556 #else /* AFS_DEMAND_ATTACH_FS */
557 Log("VInitVolumePackage: using 1 thread to attach volumes on %d partition(s)\n",
559 #endif /* AFS_DEMAND_ATTACH_FS */
561 VInitVolumePackageThread(¶ms);
564 assert(pthread_cond_destroy(¶ms.thread_done_cv) == 0);
566 #else /* AFS_PTHREAD_ENV */
570 /* Attach all the volumes in this partition */
571 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
572 int nAttached = 0, nUnattached = 0;
573 assert(VAttachVolumesByPartition(diskP, &nAttached, &nUnattached) == 0);
575 #endif /* AFS_PTHREAD_ENV */
578 VInit = 2; /* Initialized, and all volumes have been attached */
579 #ifdef FSSYNC_BUILD_CLIENT
580 if (programType == volumeUtility && connect) {
582 Log("Unable to connect to file server; aborted\n");
586 #ifdef AFS_DEMAND_ATTACH_FS
587 else if (programType == salvageServer) {
589 Log("Unable to connect to file server; aborted\n");
593 #endif /* AFS_DEMAND_ATTACH_FS */
594 #endif /* FSSYNC_BUILD_CLIENT */
598 #ifdef AFS_PTHREAD_ENV
600 VInitVolumePackageThread(void * args) {
601 int errors = 0; /* Number of errors while finding vice partitions. */
605 struct DiskPartition64 *diskP;
606 struct vinitvolumepackage_thread_t * params;
607 struct diskpartition_queue_t * dpq;
609 params = (vinitvolumepackage_thread_t *) args;
613 /* Attach all the volumes in this partition */
614 while (queue_IsNotEmpty(params)) {
615 int nAttached = 0, nUnattached = 0;
617 dpq = queue_First(params,diskpartition_queue_t);
623 assert(VAttachVolumesByPartition(diskP, &nAttached, &nUnattached) == 0);
628 params->n_threads_complete++;
629 pthread_cond_signal(¶ms->thread_done_cv);
633 #endif /* AFS_PTHREAD_ENV */
636 * attach all volumes on a given disk partition
639 VAttachVolumesByPartition(struct DiskPartition64 *diskP, int * nAttached, int * nUnattached)
645 Log("Partition %s: attaching volumes\n", diskP->name);
646 dirp = opendir(VPartitionPath(diskP));
648 Log("opendir on Partition %s failed!\n", diskP->name);
652 while ((dp = readdir(dirp))) {
654 p = strrchr(dp->d_name, '.');
655 if (p != NULL && strcmp(p, VHDREXT) == 0) {
658 #ifdef AFS_DEMAND_ATTACH_FS
659 vp = VPreAttachVolumeByName(&error, diskP->name, dp->d_name);
660 #else /* AFS_DEMAND_ATTACH_FS */
661 vp = VAttachVolumeByName(&error, diskP->name, dp->d_name,
663 #endif /* AFS_DEMAND_ATTACH_FS */
664 (*(vp ? nAttached : nUnattached))++;
665 if (error == VOFFLINE)
666 Log("Volume %d stays offline (/vice/offline/%s exists)\n", VolumeNumber(dp->d_name), dp->d_name);
667 else if (LogLevel >= 5) {
668 Log("Partition %s: attached volume %d (%s)\n",
669 diskP->name, VolumeNumber(dp->d_name),
672 #if !defined(AFS_DEMAND_ATTACH_FS)
676 #endif /* AFS_DEMAND_ATTACH_FS */
680 Log("Partition %s: attached %d volumes; %d volumes not attached\n", diskP->name, *nAttached, *nUnattached);
686 /***************************************************/
687 /* Shutdown routines */
688 /***************************************************/
692 * highly multithreaded volume package shutdown
694 * with the demand attach fileserver extensions,
695 * VShutdown has been modified to be multithreaded.
696 * In order to achieve optimal use of many threads,
697 * the shutdown code involves one control thread and
698 * n shutdown worker threads. The control thread
699 * periodically examines the number of volumes available
700 * for shutdown on each partition, and produces a worker
701 * thread allocation schedule. The idea is to eliminate
702 * redundant scheduling computation on the workers by
703 * having a single master scheduler.
705 * The scheduler's objectives are:
707 * each partition with volumes remaining gets allocated
708 * at least 1 thread (assuming sufficient threads)
710 * threads are allocated proportional to the number of
711 * volumes remaining to be offlined. This ensures that
712 * the OS I/O scheduler has many requests to elevator
713 * seek on partitions that will (presumably) take the
714 * longest amount of time (from now) to finish shutdown
715 * (3) keep threads busy
716 * when there are extra threads, they are assigned to
717 * partitions using a simple round-robin algorithm
719 * In the future, we may wish to add the ability to adapt
720 * to the relative performance patterns of each disk
725 * multi-step shutdown process
727 * demand attach shutdown is a four-step process. Each
728 * shutdown "pass" shuts down increasingly more difficult
729 * volumes. The main purpose is to achieve better cache
730 * utilization during shutdown.
733 * shutdown volumes in the unattached, pre-attached
736 * shutdown attached volumes with cached volume headers
738 * shutdown all volumes in non-exclusive states
740 * shutdown all remaining volumes
747 register Volume *vp, *np;
748 register afs_int32 code;
749 #ifdef AFS_DEMAND_ATTACH_FS
750 struct DiskPartition64 * diskP;
751 struct diskpartition_queue_t * dpq;
752 vshutdown_thread_t params;
754 pthread_attr_t attrs;
756 memset(¶ms, 0, sizeof(vshutdown_thread_t));
758 for (params.n_parts=0, diskP = DiskPartitionList;
759 diskP; diskP = diskP->next, params.n_parts++);
761 Log("VShutdown: shutting down on-line volumes on %d partition%s...\n",
762 params.n_parts, params.n_parts > 1 ? "s" : "");
764 if (vol_attach_threads > 1) {
765 /* prepare for parallel shutdown */
766 params.n_threads = vol_attach_threads;
767 assert(pthread_mutex_init(¶ms.lock, NULL) == 0);
768 assert(pthread_cond_init(¶ms.cv, NULL) == 0);
769 assert(pthread_cond_init(¶ms.master_cv, NULL) == 0);
770 assert(pthread_attr_init(&attrs) == 0);
771 assert(pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED) == 0);
774 /* setup the basic partition information structures for
775 * parallel shutdown */
776 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
778 struct rx_queue * qp, * nqp;
782 VVByPListWait_r(diskP);
783 VVByPListBeginExclusive_r(diskP);
786 for (queue_Scan(&diskP->vol_list, qp, nqp, rx_queue)) {
787 vp = (Volume *)((char *)qp - offsetof(Volume, vol_list));
791 Log("VShutdown: partition %s has %d volumes with attached headers\n",
792 VPartitionPath(diskP), count);
795 /* build up the pass 0 shutdown work queue */
796 dpq = (struct diskpartition_queue_t *) malloc(sizeof(struct diskpartition_queue_t));
799 queue_Prepend(¶ms, dpq);
801 params.part_pass_head[diskP->device] = queue_First(&diskP->vol_list, rx_queue);
804 Log("VShutdown: beginning parallel fileserver shutdown\n");
805 Log("VShutdown: using %d threads to offline volumes on %d partition%s\n",
806 vol_attach_threads, params.n_parts, params.n_parts > 1 ? "s" : "" );
808 /* do pass 0 shutdown */
809 assert(pthread_mutex_lock(¶ms.lock) == 0);
810 for (i=0; i < params.n_threads; i++) {
811 assert(pthread_create
812 (&tid, &attrs, &VShutdownThread,
816 /* wait for all the pass 0 shutdowns to complete */
817 while (params.n_threads_complete < params.n_threads) {
818 assert(pthread_cond_wait(¶ms.master_cv, ¶ms.lock) == 0);
820 params.n_threads_complete = 0;
822 assert(pthread_cond_broadcast(¶ms.cv) == 0);
823 assert(pthread_mutex_unlock(¶ms.lock) == 0);
825 Log("VShutdown: pass 0 completed using the 1 thread per partition algorithm\n");
826 Log("VShutdown: starting passes 1 through 3 using finely-granular mp-fast algorithm\n");
828 /* run the parallel shutdown scheduler. it will drop the glock internally */
829 ShutdownController(¶ms);
831 /* wait for all the workers to finish pass 3 and terminate */
832 while (params.pass < 4) {
833 VOL_CV_WAIT(¶ms.cv);
836 assert(pthread_attr_destroy(&attrs) == 0);
837 assert(pthread_cond_destroy(¶ms.cv) == 0);
838 assert(pthread_cond_destroy(¶ms.master_cv) == 0);
839 assert(pthread_mutex_destroy(¶ms.lock) == 0);
841 /* drop the VByPList exclusive reservations */
842 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
843 VVByPListEndExclusive_r(diskP);
844 Log("VShutdown: %s stats : (pass[0]=%d, pass[1]=%d, pass[2]=%d, pass[3]=%d)\n",
845 VPartitionPath(diskP),
846 params.stats[0][diskP->device],
847 params.stats[1][diskP->device],
848 params.stats[2][diskP->device],
849 params.stats[3][diskP->device]);
852 Log("VShutdown: shutdown finished using %d threads\n", params.n_threads);
854 /* if we're only going to run one shutdown thread, don't bother creating
856 Log("VShutdown: beginning single-threaded fileserver shutdown\n");
858 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
859 VShutdownByPartition_r(diskP);
863 Log("VShutdown: complete.\n");
864 #else /* AFS_DEMAND_ATTACH_FS */
865 Log("VShutdown: shutting down on-line volumes...\n");
866 for (i = 0; i < VolumeHashTable.Size; i++) {
867 /* try to hold first volume in the hash table */
868 for (queue_Scan(&VolumeHashTable.Table[i],vp,np,Volume)) {
872 Log("VShutdown: Attempting to take volume %u offline.\n",
875 /* next, take the volume offline (drops reference count) */
876 VOffline_r(vp, "File server was shut down");
880 Log("VShutdown: complete.\n");
881 #endif /* AFS_DEMAND_ATTACH_FS */
892 #ifdef AFS_DEMAND_ATTACH_FS
895 * shutdown control thread
898 ShutdownController(vshutdown_thread_t * params)
901 struct DiskPartition64 * diskP;
903 vshutdown_thread_t shadow;
905 ShutdownCreateSchedule(params);
907 while ((params->pass < 4) &&
908 (params->n_threads_complete < params->n_threads)) {
909 /* recompute schedule once per second */
911 memcpy(&shadow, params, sizeof(vshutdown_thread_t));
915 Log("ShutdownController: schedule version=%d, vol_remaining=%d, pass=%d\n",
916 shadow.schedule_version, shadow.vol_remaining, shadow.pass);
917 Log("ShutdownController: n_threads_complete=%d, n_parts_done_pass=%d\n",
918 shadow.n_threads_complete, shadow.n_parts_done_pass);
919 for (diskP = DiskPartitionList; diskP; diskP=diskP->next) {
921 Log("ShutdownController: part[%d] : (len=%d, thread_target=%d, done_pass=%d, pass_head=%p)\n",
924 shadow.part_thread_target[id],
925 shadow.part_done_pass[id],
926 shadow.part_pass_head[id]);
932 ShutdownCreateSchedule(params);
936 /* create the shutdown thread work schedule.
937 * this scheduler tries to implement fairness
938 * by allocating at least 1 thread to each
939 * partition with volumes to be shutdown,
940 * and then it attempts to allocate remaining
941 * threads based upon the amount of work left
944 ShutdownCreateSchedule(vshutdown_thread_t * params)
946 struct DiskPartition64 * diskP;
947 int sum, thr_workload, thr_left;
948 int part_residue[VOLMAXPARTS+1];
951 /* compute the total number of outstanding volumes */
953 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
954 sum += diskP->vol_list.len;
957 params->schedule_version++;
958 params->vol_remaining = sum;
963 /* compute average per-thread workload */
964 thr_workload = sum / params->n_threads;
965 if (sum % params->n_threads)
968 thr_left = params->n_threads;
969 memset(&part_residue, 0, sizeof(part_residue));
971 /* for fairness, give every partition with volumes remaining
972 * at least one thread */
973 for (diskP = DiskPartitionList; diskP && thr_left; diskP = diskP->next) {
975 if (diskP->vol_list.len) {
976 params->part_thread_target[id] = 1;
979 params->part_thread_target[id] = 0;
983 if (thr_left && thr_workload) {
984 /* compute length-weighted workloads */
987 for (diskP = DiskPartitionList; diskP && thr_left; diskP = diskP->next) {
989 delta = (diskP->vol_list.len / thr_workload) -
990 params->part_thread_target[id];
994 if (delta < thr_left) {
995 params->part_thread_target[id] += delta;
998 params->part_thread_target[id] += thr_left;
1006 /* try to assign any leftover threads to partitions that
1007 * had volume lengths closer to needing thread_target+1 */
1008 int max_residue, max_id;
1010 /* compute the residues */
1011 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1013 part_residue[id] = diskP->vol_list.len -
1014 (params->part_thread_target[id] * thr_workload);
1017 /* now try to allocate remaining threads to partitions with the
1018 * highest residues */
1021 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1023 if (part_residue[id] > max_residue) {
1024 max_residue = part_residue[id];
1033 params->part_thread_target[max_id]++;
1035 part_residue[max_id] = 0;
1040 /* punt and give any remaining threads equally to each partition */
1042 if (thr_left >= params->n_parts) {
1043 alloc = thr_left / params->n_parts;
1044 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1046 params->part_thread_target[id] += alloc;
1051 /* finish off the last of the threads */
1052 for (diskP = DiskPartitionList; thr_left && diskP; diskP = diskP->next) {
1054 params->part_thread_target[id]++;
1060 /* worker thread for parallel shutdown */
1062 VShutdownThread(void * args)
1064 struct rx_queue *qp;
1066 vshutdown_thread_t * params;
1067 int part, code, found, pass, schedule_version_save, count;
1068 struct DiskPartition64 *diskP;
1069 struct diskpartition_queue_t * dpq;
1072 params = (vshutdown_thread_t *) args;
1074 /* acquire the shutdown pass 0 lock */
1075 assert(pthread_mutex_lock(¶ms->lock) == 0);
1077 /* if there's still pass 0 work to be done,
1078 * get a work entry, and do a pass 0 shutdown */
1079 if (queue_IsNotEmpty(params)) {
1080 dpq = queue_First(params, diskpartition_queue_t);
1082 assert(pthread_mutex_unlock(¶ms->lock) == 0);
1088 while (ShutdownVolumeWalk_r(diskP, 0, ¶ms->part_pass_head[id]))
1090 params->stats[0][diskP->device] = count;
1091 assert(pthread_mutex_lock(¶ms->lock) == 0);
1094 params->n_threads_complete++;
1095 if (params->n_threads_complete == params->n_threads) {
1096 /* notify control thread that all workers have completed pass 0 */
1097 assert(pthread_cond_signal(¶ms->master_cv) == 0);
1099 while (params->pass == 0) {
1100 assert(pthread_cond_wait(¶ms->cv, ¶ms->lock) == 0);
1104 assert(pthread_mutex_unlock(¶ms->lock) == 0);
1107 pass = params->pass;
1110 /* now escalate through the more complicated shutdowns */
1112 schedule_version_save = params->schedule_version;
1114 /* find a disk partition to work on */
1115 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1117 if (params->part_thread_target[id] && !params->part_done_pass[id]) {
1118 params->part_thread_target[id]--;
1125 /* hmm. for some reason the controller thread couldn't find anything for
1126 * us to do. let's see if there's anything we can do */
1127 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1129 if (diskP->vol_list.len && !params->part_done_pass[id]) {
1132 } else if (!params->part_done_pass[id]) {
1133 params->part_done_pass[id] = 1;
1134 params->n_parts_done_pass++;
1136 Log("VShutdown: done shutting down volumes on partition %s.\n",
1137 VPartitionPath(diskP));
1143 /* do work on this partition until either the controller
1144 * creates a new schedule, or we run out of things to do
1145 * on this partition */
1148 while (!params->part_done_pass[id] &&
1149 (schedule_version_save == params->schedule_version)) {
1150 /* ShutdownVolumeWalk_r will drop the glock internally */
1151 if (!ShutdownVolumeWalk_r(diskP, pass, ¶ms->part_pass_head[id])) {
1152 if (!params->part_done_pass[id]) {
1153 params->part_done_pass[id] = 1;
1154 params->n_parts_done_pass++;
1156 Log("VShutdown: done shutting down volumes on partition %s.\n",
1157 VPartitionPath(diskP));
1165 params->stats[pass][id] += count;
1167 /* ok, everyone is done this pass, proceed */
1170 params->n_threads_complete++;
1171 while (params->pass == pass) {
1172 if (params->n_threads_complete == params->n_threads) {
1173 /* we are the last thread to complete, so we will
1174 * reinitialize worker pool state for the next pass */
1175 params->n_threads_complete = 0;
1176 params->n_parts_done_pass = 0;
1178 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1180 params->part_done_pass[id] = 0;
1181 params->part_pass_head[id] = queue_First(&diskP->vol_list, rx_queue);
1184 /* compute a new thread schedule before releasing all the workers */
1185 ShutdownCreateSchedule(params);
1187 /* wake up all the workers */
1188 assert(pthread_cond_broadcast(¶ms->cv) == 0);
1191 Log("VShutdown: pass %d completed using %d threads on %d partitions\n",
1192 pass, params->n_threads, params->n_parts);
1195 VOL_CV_WAIT(¶ms->cv);
1198 pass = params->pass;
1212 /* shut down all volumes on a given disk partition
1214 * note that this function will not allow mp-fast
1215 * shutdown of a partition */
1217 VShutdownByPartition_r(struct DiskPartition64 * dp)
1223 /* wait for other exclusive ops to finish */
1224 VVByPListWait_r(dp);
1226 /* begin exclusive access */
1227 VVByPListBeginExclusive_r(dp);
1229 /* pick the low-hanging fruit first,
1230 * then do the complicated ones last
1231 * (has the advantage of keeping
1232 * in-use volumes up until the bitter end) */
1233 for (pass = 0, total=0; pass < 4; pass++) {
1234 pass_stats[pass] = ShutdownVByPForPass_r(dp, pass);
1235 total += pass_stats[pass];
1238 /* end exclusive access */
1239 VVByPListEndExclusive_r(dp);
1241 Log("VShutdownByPartition: shut down %d volumes on %s (pass[0]=%d, pass[1]=%d, pass[2]=%d, pass[3]=%d)\n",
1242 total, VPartitionPath(dp), pass_stats[0], pass_stats[1], pass_stats[2], pass_stats[3]);
1247 /* internal shutdown functionality
1249 * for multi-pass shutdown:
1250 * 0 to only "shutdown" {pre,un}attached and error state volumes
1251 * 1 to also shutdown attached volumes w/ volume header loaded
1252 * 2 to also shutdown attached volumes w/o volume header loaded
1253 * 3 to also shutdown exclusive state volumes
1255 * caller MUST hold exclusive access on the hash chain
1256 * because we drop vol_glock_mutex internally
1258 * this function is reentrant for passes 1--3
1259 * (e.g. multiple threads can cooperate to
1260 * shutdown a partition mp-fast)
1262 * pass 0 is not scaleable because the volume state data is
1263 * synchronized by vol_glock mutex, and the locking overhead
1264 * is too high to drop the lock long enough to do linked list
1268 ShutdownVByPForPass_r(struct DiskPartition64 * dp, int pass)
1270 struct rx_queue * q = queue_First(&dp->vol_list, rx_queue);
1273 while (ShutdownVolumeWalk_r(dp, pass, &q))
1279 /* conditionally shutdown one volume on partition dp
1280 * returns 1 if a volume was shutdown in this pass,
1283 ShutdownVolumeWalk_r(struct DiskPartition64 * dp, int pass,
1284 struct rx_queue ** idx)
1286 struct rx_queue *qp, *nqp;
1291 for (queue_ScanFrom(&dp->vol_list, qp, qp, nqp, rx_queue)) {
1292 vp = (Volume *) (((char *)qp) - offsetof(Volume, vol_list));
1296 if ((V_attachState(vp) != VOL_STATE_UNATTACHED) &&
1297 (V_attachState(vp) != VOL_STATE_ERROR) &&
1298 (V_attachState(vp) != VOL_STATE_PREATTACHED)) {
1302 if ((V_attachState(vp) == VOL_STATE_ATTACHED) &&
1303 (vp->header == NULL)) {
1307 if (VIsExclusiveState(V_attachState(vp))) {
1312 DeleteVolumeFromVByPList_r(vp);
1313 VShutdownVolume_r(vp);
1323 * shutdown a specific volume
1325 /* caller MUST NOT hold a heavyweight ref on vp */
1327 VShutdownVolume_r(Volume * vp)
1331 VCreateReservation_r(vp);
1333 if (LogLevel >= 5) {
1334 Log("VShutdownVolume_r: vid=%u, device=%d, state=%hu\n",
1335 vp->hashid, vp->partition->device, V_attachState(vp));
1338 /* wait for other blocking ops to finish */
1339 VWaitExclusiveState_r(vp);
1341 assert(VIsValidState(V_attachState(vp)));
1343 switch(V_attachState(vp)) {
1344 case VOL_STATE_SALVAGING:
1345 /* make sure salvager knows we don't want
1346 * the volume back */
1347 VCancelSalvage_r(vp, SALVSYNC_SHUTDOWN);
1348 case VOL_STATE_PREATTACHED:
1349 case VOL_STATE_ERROR:
1350 VChangeState_r(vp, VOL_STATE_UNATTACHED);
1351 case VOL_STATE_UNATTACHED:
1353 case VOL_STATE_GOING_OFFLINE:
1354 case VOL_STATE_SHUTTING_DOWN:
1355 case VOL_STATE_ATTACHED:
1359 Log("VShutdown: Attempting to take volume %u offline.\n",
1362 /* take the volume offline (drops reference count) */
1363 VOffline_r(vp, "File server was shut down");
1368 VCancelReservation_r(vp);
1372 #endif /* AFS_DEMAND_ATTACH_FS */
1375 /***************************************************/
1376 /* Header I/O routines */
1377 /***************************************************/
1379 /* open a descriptor for the inode (h),
1380 * read in an on-disk structure into buffer (to) of size (size),
1381 * verify versionstamp in structure has magic (magic) and
1382 * optionally verify version (version) if (version) is nonzero
1385 ReadHeader(Error * ec, IHandle_t * h, char *to, int size, bit32 magic,
1388 struct versionStamp *vsn;
1403 if (FDH_SEEK(fdP, 0, SEEK_SET) < 0) {
1405 FDH_REALLYCLOSE(fdP);
1408 vsn = (struct versionStamp *)to;
1409 if (FDH_READ(fdP, to, size) != size || vsn->magic != magic) {
1411 FDH_REALLYCLOSE(fdP);
1416 /* Check is conditional, in case caller wants to inspect version himself */
1417 if (version && vsn->version != version) {
1423 WriteVolumeHeader_r(Error * ec, Volume * vp)
1425 IHandle_t *h = V_diskDataHandle(vp);
1435 if (FDH_SEEK(fdP, 0, SEEK_SET) < 0) {
1437 FDH_REALLYCLOSE(fdP);
1440 if (FDH_WRITE(fdP, (char *)&V_disk(vp), sizeof(V_disk(vp)))
1441 != sizeof(V_disk(vp))) {
1443 FDH_REALLYCLOSE(fdP);
1449 /* VolumeHeaderToDisk
1450 * Allows for storing 64 bit inode numbers in on-disk volume header
1453 /* convert in-memory representation of a volume header to the
1454 * on-disk representation of a volume header */
1456 VolumeHeaderToDisk(VolumeDiskHeader_t * dh, VolumeHeader_t * h)
1459 memset((char *)dh, 0, sizeof(VolumeDiskHeader_t));
1460 dh->stamp = h->stamp;
1462 dh->parent = h->parent;
1464 #ifdef AFS_64BIT_IOPS_ENV
1465 dh->volumeInfo_lo = (afs_int32) h->volumeInfo & 0xffffffff;
1466 dh->volumeInfo_hi = (afs_int32) (h->volumeInfo >> 32) & 0xffffffff;
1467 dh->smallVnodeIndex_lo = (afs_int32) h->smallVnodeIndex & 0xffffffff;
1468 dh->smallVnodeIndex_hi =
1469 (afs_int32) (h->smallVnodeIndex >> 32) & 0xffffffff;
1470 dh->largeVnodeIndex_lo = (afs_int32) h->largeVnodeIndex & 0xffffffff;
1471 dh->largeVnodeIndex_hi =
1472 (afs_int32) (h->largeVnodeIndex >> 32) & 0xffffffff;
1473 dh->linkTable_lo = (afs_int32) h->linkTable & 0xffffffff;
1474 dh->linkTable_hi = (afs_int32) (h->linkTable >> 32) & 0xffffffff;
1476 dh->volumeInfo_lo = h->volumeInfo;
1477 dh->smallVnodeIndex_lo = h->smallVnodeIndex;
1478 dh->largeVnodeIndex_lo = h->largeVnodeIndex;
1479 dh->linkTable_lo = h->linkTable;
1483 /* DiskToVolumeHeader
1484 * Converts an on-disk representation of a volume header to
1485 * the in-memory representation of a volume header.
1487 * Makes the assumption that AFS has *always*
1488 * zero'd the volume header file so that high parts of inode
1489 * numbers are 0 in older (SGI EFS) volume header files.
1492 DiskToVolumeHeader(VolumeHeader_t * h, VolumeDiskHeader_t * dh)
1494 memset((char *)h, 0, sizeof(VolumeHeader_t));
1495 h->stamp = dh->stamp;
1497 h->parent = dh->parent;
1499 #ifdef AFS_64BIT_IOPS_ENV
1501 (Inode) dh->volumeInfo_lo | ((Inode) dh->volumeInfo_hi << 32);
1503 h->smallVnodeIndex =
1504 (Inode) dh->smallVnodeIndex_lo | ((Inode) dh->
1505 smallVnodeIndex_hi << 32);
1507 h->largeVnodeIndex =
1508 (Inode) dh->largeVnodeIndex_lo | ((Inode) dh->
1509 largeVnodeIndex_hi << 32);
1511 (Inode) dh->linkTable_lo | ((Inode) dh->linkTable_hi << 32);
1513 h->volumeInfo = dh->volumeInfo_lo;
1514 h->smallVnodeIndex = dh->smallVnodeIndex_lo;
1515 h->largeVnodeIndex = dh->largeVnodeIndex_lo;
1516 h->linkTable = dh->linkTable_lo;
1521 /***************************************************/
1522 /* Volume Attachment routines */
1523 /***************************************************/
1525 #ifdef AFS_DEMAND_ATTACH_FS
1527 * pre-attach a volume given its path.
1529 * @param[out] ec outbound error code
1530 * @param[in] partition partition path string
1531 * @param[in] name volume id string
1533 * @return volume object pointer
1535 * @note A pre-attached volume will only have its partition
1536 * and hashid fields initialized. At first call to
1537 * VGetVolume, the volume will be fully attached.
1541 VPreAttachVolumeByName(Error * ec, char *partition, char *name)
1545 vp = VPreAttachVolumeByName_r(ec, partition, name);
1551 * pre-attach a volume given its path.
1553 * @param[out] ec outbound error code
1554 * @param[in] partition path to vice partition
1555 * @param[in] name volume id string
1557 * @return volume object pointer
1559 * @pre VOL_LOCK held
1561 * @internal volume package internal use only.
1564 VPreAttachVolumeByName_r(Error * ec, char *partition, char *name)
1566 return VPreAttachVolumeById_r(ec,
1568 VolumeNumber(name));
1572 * pre-attach a volume given its path and numeric volume id.
1574 * @param[out] ec error code return
1575 * @param[in] partition path to vice partition
1576 * @param[in] volumeId numeric volume id
1578 * @return volume object pointer
1580 * @pre VOL_LOCK held
1582 * @internal volume package internal use only.
1585 VPreAttachVolumeById_r(Error * ec,
1590 struct DiskPartition64 *partp;
1594 assert(programType == fileServer);
1596 if (!(partp = VGetPartition_r(partition, 0))) {
1598 Log("VPreAttachVolumeById_r: Error getting partition (%s)\n", partition);
1602 vp = VLookupVolume_r(ec, volumeId, NULL);
1607 return VPreAttachVolumeByVp_r(ec, partp, vp, volumeId);
1611 * preattach a volume.
1613 * @param[out] ec outbound error code
1614 * @param[in] partp pointer to partition object
1615 * @param[in] vp pointer to volume object
1616 * @param[in] vid volume id
1618 * @return volume object pointer
1620 * @pre VOL_LOCK is held.
1622 * @warning Returned volume object pointer does not have to
1623 * equal the pointer passed in as argument vp. There
1624 * are potential race conditions which can result in
1625 * the pointers having different values. It is up to
1626 * the caller to make sure that references are handled
1627 * properly in this case.
1629 * @note If there is already a volume object registered with
1630 * the same volume id, its pointer MUST be passed as
1631 * argument vp. Failure to do so will result in a silent
1632 * failure to preattach.
1634 * @internal volume package internal use only.
1637 VPreAttachVolumeByVp_r(Error * ec,
1638 struct DiskPartition64 * partp,
1646 /* check to see if pre-attach already happened */
1648 (V_attachState(vp) != VOL_STATE_UNATTACHED) &&
1649 (V_attachState(vp) != VOL_STATE_PREATTACHED) &&
1650 !VIsErrorState(V_attachState(vp))) {
1652 * pre-attach is a no-op in all but the following cases:
1654 * - volume is unattached
1655 * - volume is in an error state
1656 * - volume is pre-attached
1658 Log("VPreattachVolumeByVp_r: volume %u not in quiescent state\n", vid);
1661 /* we're re-attaching a volume; clear out some old state */
1662 memset(&vp->salvage, 0, sizeof(struct VolumeOnlineSalvage));
1664 if (V_partition(vp) != partp) {
1665 /* XXX potential race */
1666 DeleteVolumeFromVByPList_r(vp);
1669 /* if we need to allocate a new Volume struct,
1670 * go ahead and drop the vol glock, otherwise
1671 * do the basic setup synchronised, as it's
1672 * probably not worth dropping the lock */
1675 /* allocate the volume structure */
1676 vp = nvp = (Volume *) malloc(sizeof(Volume));
1678 memset(vp, 0, sizeof(Volume));
1679 queue_Init(&vp->vnode_list);
1680 assert(pthread_cond_init(&V_attachCV(vp), NULL) == 0);
1683 /* link the volume with its associated vice partition */
1684 vp->device = partp->device;
1685 vp->partition = partp;
1688 /* if we dropped the lock, reacquire the lock,
1689 * check for pre-attach races, and then add
1690 * the volume to the hash table */
1693 nvp = VLookupVolume_r(ec, vid, NULL);
1698 } else if (nvp) { /* race detected */
1703 /* hack to make up for VChangeState_r() decrementing
1704 * the old state counter */
1705 VStats.state_levels[0]++;
1709 /* put pre-attached volume onto the hash table
1710 * and bring it up to the pre-attached state */
1711 AddVolumeToHashTable(vp, vp->hashid);
1712 AddVolumeToVByPList_r(vp);
1713 VLRU_Init_Node_r(vp);
1714 VChangeState_r(vp, VOL_STATE_PREATTACHED);
1717 Log("VPreAttachVolumeByVp_r: volume %u pre-attached\n", vp->hashid);
1725 #endif /* AFS_DEMAND_ATTACH_FS */
1727 /* Attach an existing volume, given its pathname, and return a
1728 pointer to the volume header information. The volume also
1729 normally goes online at this time. An offline volume
1730 must be reattached to make it go online */
1732 VAttachVolumeByName(Error * ec, char *partition, char *name, int mode)
1736 retVal = VAttachVolumeByName_r(ec, partition, name, mode);
1742 VAttachVolumeByName_r(Error * ec, char *partition, char *name, int mode)
1744 register Volume *vp = NULL, *svp = NULL;
1746 struct afs_stat status;
1747 struct VolumeDiskHeader diskHeader;
1748 struct VolumeHeader iheader;
1749 struct DiskPartition64 *partp;
1753 #ifdef AFS_DEMAND_ATTACH_FS
1754 VolumeStats stats_save;
1755 #endif /* AFS_DEMAND_ATTACH_FS */
1759 volumeId = VolumeNumber(name);
1761 if (!(partp = VGetPartition_r(partition, 0))) {
1763 Log("VAttachVolume: Error getting partition (%s)\n", partition);
1767 if (programType == volumeUtility) {
1769 VLockPartition_r(partition);
1770 } else if (programType == fileServer) {
1771 #ifdef AFS_DEMAND_ATTACH_FS
1772 /* lookup the volume in the hash table */
1773 vp = VLookupVolume_r(ec, volumeId, NULL);
1779 /* save any counters that are supposed to
1780 * be monotonically increasing over the
1781 * lifetime of the fileserver */
1782 memcpy(&stats_save, &vp->stats, sizeof(VolumeStats));
1784 memset(&stats_save, 0, sizeof(VolumeStats));
1787 /* if there's something in the hash table, and it's not
1788 * in the pre-attach state, then we may need to detach
1789 * it before proceeding */
1790 if (vp && (V_attachState(vp) != VOL_STATE_PREATTACHED)) {
1791 VCreateReservation_r(vp);
1792 VWaitExclusiveState_r(vp);
1794 /* at this point state must be one of:
1803 if (vp->specialStatus == VBUSY)
1806 /* if it's already attached, see if we can return it */
1807 if (V_attachState(vp) == VOL_STATE_ATTACHED) {
1808 VGetVolumeByVp_r(ec, vp);
1810 VCancelReservation_r(vp);
1814 /* otherwise, we need to detach, and attempt to re-attach */
1815 VDetachVolume_r(ec, vp);
1817 Log("VAttachVolume: Error detaching old volume instance (%s)\n", name);
1820 /* if it isn't fully attached, delete from the hash tables,
1821 and let the refcounter handle the rest */
1822 DeleteVolumeFromHashTable(vp);
1823 DeleteVolumeFromVByPList_r(vp);
1826 VCancelReservation_r(vp);
1830 /* pre-attach volume if it hasn't been done yet */
1832 (V_attachState(vp) == VOL_STATE_UNATTACHED) ||
1833 (V_attachState(vp) == VOL_STATE_ERROR)) {
1835 vp = VPreAttachVolumeByVp_r(ec, partp, vp, volumeId);
1843 /* handle pre-attach races
1845 * multiple threads can race to pre-attach a volume,
1846 * but we can't let them race beyond that
1848 * our solution is to let the first thread to bring
1849 * the volume into an exclusive state win; the other
1850 * threads just wait until it finishes bringing the
1851 * volume online, and then they do a vgetvolumebyvp
1853 if (svp && (svp != vp)) {
1854 /* wait for other exclusive ops to finish */
1855 VCreateReservation_r(vp);
1856 VWaitExclusiveState_r(vp);
1858 /* get a heavyweight ref, kill the lightweight ref, and return */
1859 VGetVolumeByVp_r(ec, vp);
1860 VCancelReservation_r(vp);
1864 /* at this point, we are chosen as the thread to do
1865 * demand attachment for this volume. all other threads
1866 * doing a getvolume on vp->hashid will block until we finish */
1868 /* make sure any old header cache entries are invalidated
1869 * before proceeding */
1870 FreeVolumeHeader(vp);
1872 VChangeState_r(vp, VOL_STATE_ATTACHING);
1874 /* restore any saved counters */
1875 memcpy(&vp->stats, &stats_save, sizeof(VolumeStats));
1876 #else /* AFS_DEMAND_ATTACH_FS */
1877 vp = VGetVolume_r(ec, volumeId);
1881 if (vp->specialStatus == VBUSY)
1883 VDetachVolume_r(ec, vp);
1885 Log("VAttachVolume: Error detaching volume (%s)\n", name);
1889 #endif /* AFS_DEMAND_ATTACH_FS */
1893 strcpy(path, VPartitionPath(partp));
1899 if ((fd = afs_open(path, O_RDONLY)) == -1 || afs_fstat(fd, &status) == -1) {
1900 Log("VAttachVolume: Failed to open %s (errno %d)\n", path, errno);
1907 n = read(fd, &diskHeader, sizeof(diskHeader));
1909 if (n != sizeof(diskHeader)
1910 || diskHeader.stamp.magic != VOLUMEHEADERMAGIC) {
1911 Log("VAttachVolume: Error reading volume header %s\n", path);
1916 if (diskHeader.stamp.version != VOLUMEHEADERVERSION) {
1917 Log("VAttachVolume: Volume %s, version number is incorrect; volume needs salvaged\n", path);
1923 DiskToVolumeHeader(&iheader, &diskHeader);
1924 #ifdef FSSYNC_BUILD_CLIENT
1925 if (programType == volumeUtility && mode != V_SECRETLY && mode != V_PEEK) {
1927 if (FSYNC_VolOp(iheader.id, partition, FSYNC_VOL_NEEDVOLUME, mode, NULL)
1929 Log("VAttachVolume: attach of volume %u apparently denied by file server\n", iheader.id);
1930 *ec = VNOVOL; /* XXXX */
1938 vp = (Volume *) calloc(1, sizeof(Volume));
1940 vp->device = partp->device;
1941 vp->partition = partp;
1942 queue_Init(&vp->vnode_list);
1943 #ifdef AFS_DEMAND_ATTACH_FS
1944 assert(pthread_cond_init(&V_attachCV(vp), NULL) == 0);
1945 #endif /* AFS_DEMAND_ATTACH_FS */
1948 /* attach2 is entered without any locks, and returns
1949 * with vol_glock_mutex held */
1950 vp = attach2(ec, volumeId, path, &iheader, partp, vp, isbusy, mode);
1952 if (programType == volumeUtility && vp) {
1953 #ifdef AFS_DEMAND_ATTACH_FS
1954 /* for dafs, we should tell the fileserver, except for V_PEEK
1955 * where we know it is not necessary */
1956 if (mode == V_PEEK) {
1957 vp->needsPutBack = 0;
1959 vp->needsPutBack = 1;
1961 #else /* !AFS_DEMAND_ATTACH_FS */
1962 /* duplicate computation in fssync.c about whether the server
1963 * takes the volume offline or not. If the volume isn't
1964 * offline, we must not return it when we detach the volume,
1965 * or the server will abort */
1966 if (mode == V_READONLY || mode == V_PEEK
1967 || (!VolumeWriteable(vp) && (mode == V_CLONE || mode == V_DUMP)))
1968 vp->needsPutBack = 0;
1970 vp->needsPutBack = 1;
1971 #endif /* !AFS_DEMAND_ATTACH_FS */
1973 /* OK, there's a problem here, but one that I don't know how to
1974 * fix right now, and that I don't think should arise often.
1975 * Basically, we should only put back this volume to the server if
1976 * it was given to us by the server, but since we don't have a vp,
1977 * we can't run the VolumeWriteable function to find out as we do
1978 * above when computing vp->needsPutBack. So we send it back, but
1979 * there's a path in VAttachVolume on the server which may abort
1980 * if this volume doesn't have a header. Should be pretty rare
1981 * for all of that to happen, but if it does, probably the right
1982 * fix is for the server to allow the return of readonly volumes
1983 * that it doesn't think are really checked out. */
1984 #ifdef FSSYNC_BUILD_CLIENT
1985 if (programType == volumeUtility && vp == NULL &&
1986 mode != V_SECRETLY && mode != V_PEEK) {
1987 FSYNC_VolOp(iheader.id, partition, FSYNC_VOL_ON, 0, NULL);
1990 if (programType == fileServer && vp) {
1991 #ifdef AFS_DEMAND_ATTACH_FS
1993 * we can get here in cases where we don't "own"
1994 * the volume (e.g. volume owned by a utility).
1995 * short circuit around potential disk header races.
1997 if (V_attachState(vp) != VOL_STATE_ATTACHED) {
2001 V_needsCallback(vp) = 0;
2003 if (VInit >= 2 && V_BreakVolumeCallbacks) {
2004 Log("VAttachVolume: Volume %u was changed externally; breaking callbacks\n", V_id(vp));
2005 (*V_BreakVolumeCallbacks) (V_id(vp));
2008 VUpdateVolume_r(ec, vp, 0);
2010 Log("VAttachVolume: Error updating volume\n");
2015 if (VolumeWriteable(vp) && V_dontSalvage(vp) == 0) {
2016 #ifndef AFS_DEMAND_ATTACH_FS
2017 /* This is a hack: by temporarily setting the incore
2018 * dontSalvage flag ON, the volume will be put back on the
2019 * Update list (with dontSalvage OFF again). It will then
2020 * come back in N minutes with DONT_SALVAGE eventually
2021 * set. This is the way that volumes that have never had
2022 * it set get it set; or that volumes that have been
2023 * offline without DONT SALVAGE having been set also
2024 * eventually get it set */
2025 V_dontSalvage(vp) = DONT_SALVAGE;
2026 #endif /* !AFS_DEMAND_ATTACH_FS */
2027 VAddToVolumeUpdateList_r(ec, vp);
2029 Log("VAttachVolume: Error adding volume to update list\n");
2036 Log("VOnline: volume %u (%s) attached and online\n", V_id(vp),
2041 if (programType == volumeUtility) {
2042 VUnlockPartition_r(partition);
2045 #ifdef AFS_DEMAND_ATTACH_FS
2046 /* attach failed; make sure we're in error state */
2047 if (vp && !VIsErrorState(V_attachState(vp))) {
2048 VChangeState_r(vp, VOL_STATE_ERROR);
2050 #endif /* AFS_DEMAND_ATTACH_FS */
2057 #ifdef AFS_DEMAND_ATTACH_FS
2058 /* VAttachVolumeByVp_r
2060 * finish attaching a volume that is
2061 * in a less than fully attached state
2063 /* caller MUST hold a ref count on vp */
2065 VAttachVolumeByVp_r(Error * ec, Volume * vp, int mode)
2067 char name[VMAXPATHLEN];
2068 int fd, n, reserve = 0;
2069 struct afs_stat status;
2070 struct VolumeDiskHeader diskHeader;
2071 struct VolumeHeader iheader;
2072 struct DiskPartition64 *partp;
2077 VolumeStats stats_save;
2080 /* volume utility should never call AttachByVp */
2081 assert(programType == fileServer);
2083 volumeId = vp->hashid;
2084 partp = vp->partition;
2085 VolumeExternalName_r(volumeId, name, sizeof(name));
2088 /* if another thread is performing a blocking op, wait */
2089 VWaitExclusiveState_r(vp);
2091 memcpy(&stats_save, &vp->stats, sizeof(VolumeStats));
2093 /* if it's already attached, see if we can return it */
2094 if (V_attachState(vp) == VOL_STATE_ATTACHED) {
2095 VGetVolumeByVp_r(ec, vp);
2099 if (vp->specialStatus == VBUSY)
2101 VDetachVolume_r(ec, vp);
2103 Log("VAttachVolume: Error detaching volume (%s)\n", name);
2109 /* pre-attach volume if it hasn't been done yet */
2111 (V_attachState(vp) == VOL_STATE_UNATTACHED) ||
2112 (V_attachState(vp) == VOL_STATE_ERROR)) {
2113 nvp = VPreAttachVolumeByVp_r(ec, partp, vp, volumeId);
2119 VCreateReservation_r(nvp);
2125 VChangeState_r(vp, VOL_STATE_ATTACHING);
2127 /* restore monotonically increasing stats */
2128 memcpy(&vp->stats, &stats_save, sizeof(VolumeStats));
2133 /* compute path to disk header,
2135 * and verify magic and version stamps */
2136 strcpy(path, VPartitionPath(partp));
2142 if ((fd = afs_open(path, O_RDONLY)) == -1 || afs_fstat(fd, &status) == -1) {
2143 Log("VAttachVolume: Failed to open %s (errno %d)\n", path, errno);
2150 n = read(fd, &diskHeader, sizeof(diskHeader));
2152 if (n != sizeof(diskHeader)
2153 || diskHeader.stamp.magic != VOLUMEHEADERMAGIC) {
2154 Log("VAttachVolume: Error reading volume header %s\n", path);
2159 if (diskHeader.stamp.version != VOLUMEHEADERVERSION) {
2160 Log("VAttachVolume: Volume %s, version number is incorrect; volume needs salvaged\n", path);
2166 /* convert on-disk header format to in-memory header format */
2167 DiskToVolumeHeader(&iheader, &diskHeader);
2171 * NOTE: attach2 is entered without any locks, and returns
2172 * with vol_glock_mutex held */
2173 vp = attach2(ec, volumeId, path, &iheader, partp, vp, isbusy, mode);
2176 * the event that an error was encountered, or
2177 * the volume was not brought to an attached state
2178 * for any reason, skip to the end. We cannot
2179 * safely call VUpdateVolume unless we "own" it.
2183 (V_attachState(vp) != VOL_STATE_ATTACHED)) {
2187 V_needsCallback(vp) = 0;
2188 VUpdateVolume_r(ec, vp, 0);
2190 Log("VAttachVolume: Error updating volume %u\n", vp->hashid);
2194 if (VolumeWriteable(vp) && V_dontSalvage(vp) == 0) {
2195 #ifndef AFS_DEMAND_ATTACH_FS
2196 /* This is a hack: by temporarily setting the incore
2197 * dontSalvage flag ON, the volume will be put back on the
2198 * Update list (with dontSalvage OFF again). It will then
2199 * come back in N minutes with DONT_SALVAGE eventually
2200 * set. This is the way that volumes that have never had
2201 * it set get it set; or that volumes that have been
2202 * offline without DONT SALVAGE having been set also
2203 * eventually get it set */
2204 V_dontSalvage(vp) = DONT_SALVAGE;
2205 #endif /* !AFS_DEMAND_ATTACH_FS */
2206 VAddToVolumeUpdateList_r(ec, vp);
2208 Log("VAttachVolume: Error adding volume %u to update list\n", vp->hashid);
2215 Log("VOnline: volume %u (%s) attached and online\n", V_id(vp),
2219 VCancelReservation_r(nvp);
2222 if (*ec && (*ec != VOFFLINE) && (*ec != VSALVAGE)) {
2223 if (vp && !VIsErrorState(V_attachState(vp))) {
2224 VChangeState_r(vp, VOL_STATE_ERROR);
2231 #endif /* AFS_DEMAND_ATTACH_FS */
2234 * called without any locks held
2235 * returns with vol_glock_mutex held
2238 attach2(Error * ec, VolId volumeId, char *path, register struct VolumeHeader * header,
2239 struct DiskPartition64 * partp, register Volume * vp, int isbusy, int mode)
2241 vp->specialStatus = (byte) (isbusy ? VBUSY : 0);
2242 IH_INIT(vp->vnodeIndex[vLarge].handle, partp->device, header->parent,
2243 header->largeVnodeIndex);
2244 IH_INIT(vp->vnodeIndex[vSmall].handle, partp->device, header->parent,
2245 header->smallVnodeIndex);
2246 IH_INIT(vp->diskDataHandle, partp->device, header->parent,
2247 header->volumeInfo);
2248 IH_INIT(vp->linkHandle, partp->device, header->parent, header->linkTable);
2249 vp->shuttingDown = 0;
2250 vp->goingOffline = 0;
2252 #ifdef AFS_DEMAND_ATTACH_FS
2253 vp->stats.last_attach = FT_ApproxTime();
2254 vp->stats.attaches++;
2258 IncUInt64(&VStats.attaches);
2259 vp->cacheCheck = ++VolumeCacheCheck;
2260 /* just in case this ever rolls over */
2261 if (!vp->cacheCheck)
2262 vp->cacheCheck = ++VolumeCacheCheck;
2263 GetVolumeHeader(vp);
2266 #if defined(AFS_DEMAND_ATTACH_FS) && defined(FSSYNC_BUILD_CLIENT)
2267 /* demand attach changes the V_PEEK mechanism
2269 * we can now suck the current disk data structure over
2270 * the fssync interface without going to disk
2272 * (technically, we don't need to restrict this feature
2273 * to demand attach fileservers. However, I'm trying
2274 * to limit the number of common code changes)
2276 if (programType != fileServer && mode == V_PEEK) {
2278 res.payload.len = sizeof(VolumeDiskData);
2279 res.payload.buf = &vp->header->diskstuff;
2281 if (FSYNC_VolOp(volumeId,
2282 VPartitionPath(partp),
2283 FSYNC_VOL_QUERY_HDR,
2286 goto disk_header_loaded;
2289 #endif /* AFS_DEMAND_ATTACH_FS && FSSYNC_BUILD_CLIENT */
2290 (void)ReadHeader(ec, V_diskDataHandle(vp), (char *)&V_disk(vp),
2291 sizeof(V_disk(vp)), VOLUMEINFOMAGIC, VOLUMEINFOVERSION);
2293 #ifdef AFS_DEMAND_ATTACH_FS
2296 IncUInt64(&VStats.hdr_loads);
2297 IncUInt64(&vp->stats.hdr_loads);
2299 #endif /* AFS_DEMAND_ATTACH_FS */
2302 Log("VAttachVolume: Error reading diskDataHandle vol header %s; error=%u\n", path, *ec);
2307 #ifdef AFS_DEMAND_ATTACH_FS
2310 /* check for pending volume operations */
2311 if (vp->pending_vol_op) {
2312 /* see if the pending volume op requires exclusive access */
2313 if (!VVolOpLeaveOnline_r(vp, vp->pending_vol_op)) {
2314 /* mark the volume down */
2316 VChangeState_r(vp, VOL_STATE_UNATTACHED);
2317 if (V_offlineMessage(vp)[0] == '\0')
2318 strlcpy(V_offlineMessage(vp),
2319 "A volume utility is running.",
2320 sizeof(V_offlineMessage(vp)));
2321 V_offlineMessage(vp)[sizeof(V_offlineMessage(vp)) - 1] = '\0';
2323 /* check to see if we should set the specialStatus flag */
2324 if (VVolOpSetVBusy_r(vp, vp->pending_vol_op)) {
2325 vp->specialStatus = VBUSY;
2330 V_attachFlags(vp) |= VOL_HDR_LOADED;
2331 vp->stats.last_hdr_load = vp->stats.last_attach;
2333 #endif /* AFS_DEMAND_ATTACH_FS */
2336 struct IndexFileHeader iHead;
2338 #if OPENAFS_VOL_STATS
2340 * We just read in the diskstuff part of the header. If the detailed
2341 * volume stats area has not yet been initialized, we should bzero the
2342 * area and mark it as initialized.
2344 if (!(V_stat_initialized(vp))) {
2345 memset((char *)(V_stat_area(vp)), 0, VOL_STATS_BYTES);
2346 V_stat_initialized(vp) = 1;
2348 #endif /* OPENAFS_VOL_STATS */
2350 (void)ReadHeader(ec, vp->vnodeIndex[vSmall].handle,
2351 (char *)&iHead, sizeof(iHead),
2352 SMALLINDEXMAGIC, SMALLINDEXVERSION);
2355 Log("VAttachVolume: Error reading smallVnode vol header %s; error=%u\n", path, *ec);
2360 struct IndexFileHeader iHead;
2362 (void)ReadHeader(ec, vp->vnodeIndex[vLarge].handle,
2363 (char *)&iHead, sizeof(iHead),
2364 LARGEINDEXMAGIC, LARGEINDEXVERSION);
2367 Log("VAttachVolume: Error reading largeVnode vol header %s; error=%u\n", path, *ec);
2371 #ifdef AFS_NAMEI_ENV
2373 struct versionStamp stamp;
2375 (void)ReadHeader(ec, V_linkHandle(vp), (char *)&stamp,
2376 sizeof(stamp), LINKTABLEMAGIC, LINKTABLEVERSION);
2379 Log("VAttachVolume: Error reading namei vol header %s; error=%u\n", path, *ec);
2382 #endif /* AFS_NAMEI_ENV */
2384 #if defined(AFS_DEMAND_ATTACH_FS)
2385 if (*ec && ((*ec != VOFFLINE) || (V_attachState(vp) != VOL_STATE_UNATTACHED))) {
2387 if (programType == fileServer) {
2388 VRequestSalvage_r(ec, vp, SALVSYNC_ERROR, VOL_SALVAGE_INVALIDATE_HEADER);
2391 Log("VAttachVolume: Error attaching volume %s; volume needs salvage; error=%u\n", path, *ec);
2397 /* volume operation in progress */
2401 #else /* AFS_DEMAND_ATTACH_FS */
2403 Log("VAttachVolume: Error attaching volume %s; volume needs salvage; error=%u\n", path, *ec);
2408 #endif /* AFS_DEMAND_ATTACH_FS */
2410 if (V_needsSalvaged(vp)) {
2411 if (vp->specialStatus)
2412 vp->specialStatus = 0;
2414 #if defined(AFS_DEMAND_ATTACH_FS)
2415 if (programType == fileServer) {
2416 VRequestSalvage_r(ec, vp, SALVSYNC_NEEDED, VOL_SALVAGE_INVALIDATE_HEADER);
2419 Log("VAttachVolume: volume salvage flag is ON for %s; volume needs salvage\n", path);
2423 #else /* AFS_DEMAND_ATTACH_FS */
2426 #endif /* AFS_DEMAND_ATTACH_FS */
2431 if (programType == fileServer) {
2432 #ifndef FAST_RESTART
2433 if (V_inUse(vp) && VolumeWriteable(vp)) {
2434 if (!V_needsSalvaged(vp)) {
2435 V_needsSalvaged(vp) = 1;
2436 VUpdateVolume_r(ec, vp, 0);
2438 #if defined(AFS_DEMAND_ATTACH_FS)
2439 VRequestSalvage_r(ec, vp, SALVSYNC_NEEDED, VOL_SALVAGE_INVALIDATE_HEADER);
2441 #else /* AFS_DEMAND_ATTACH_FS */
2442 Log("VAttachVolume: volume %s needs to be salvaged; not attached.\n", path);
2445 #endif /* AFS_DEMAND_ATTACH_FS */
2448 #endif /* FAST_RESTART */
2450 if (V_destroyMe(vp) == DESTROY_ME) {
2451 #if defined(AFS_DEMAND_ATTACH_FS)
2452 /* schedule a salvage so the volume goes away on disk */
2453 VRequestSalvage_r(ec, vp, SALVSYNC_ERROR, VOL_SALVAGE_INVALIDATE_HEADER);
2454 VChangeState_r(vp, VOL_STATE_ERROR);
2456 #endif /* AFS_DEMAND_ATTACH_FS */
2458 Log("VAttachVolume: volume %s is junk; it should be destroyed at next salvage\n", path);
2464 vp->nextVnodeUnique = V_uniquifier(vp);
2465 vp->vnodeIndex[vSmall].bitmap = vp->vnodeIndex[vLarge].bitmap = NULL;
2466 #ifndef BITMAP_LATER
2467 if (programType == fileServer && VolumeWriteable(vp)) {
2469 for (i = 0; i < nVNODECLASSES; i++) {
2470 VGetBitmap_r(ec, vp, i);
2472 #ifdef AFS_DEMAND_ATTACH_FS
2473 VRequestSalvage_r(ec, vp, SALVSYNC_ERROR, VOL_SALVAGE_INVALIDATE_HEADER);
2475 #else /* AFS_DEMAND_ATTACH_FS */
2477 #endif /* AFS_DEMAND_ATTACH_FS */
2478 Log("VAttachVolume: error getting bitmap for volume (%s)\n",
2484 #endif /* BITMAP_LATER */
2486 if (programType == fileServer) {
2487 if (vp->specialStatus)
2488 vp->specialStatus = 0;
2489 if (V_blessed(vp) && V_inService(vp) && !V_needsSalvaged(vp)) {
2491 V_offlineMessage(vp)[0] = '\0';
2495 AddVolumeToHashTable(vp, V_id(vp));
2496 #ifdef AFS_DEMAND_ATTACH_FS
2497 AddVolumeToVByPList_r(vp);
2499 if ((programType != fileServer) ||
2501 VChangeState_r(vp, VOL_STATE_ATTACHED);
2503 VChangeState_r(vp, VOL_STATE_UNATTACHED);
2509 /* Attach an existing volume.
2510 The volume also normally goes online at this time.
2511 An offline volume must be reattached to make it go online.
2515 VAttachVolume(Error * ec, VolumeId volumeId, int mode)
2519 retVal = VAttachVolume_r(ec, volumeId, mode);
2525 VAttachVolume_r(Error * ec, VolumeId volumeId, int mode)
2528 GetVolumePath(ec, volumeId, &part, &name);
2530 register Volume *vp;
2532 vp = VGetVolume_r(&error, volumeId);
2534 assert(V_inUse(vp) == 0);
2535 VDetachVolume_r(ec, vp);
2539 return VAttachVolumeByName_r(ec, part, name, mode);
2542 /* Increment a reference count to a volume, sans context swaps. Requires
2543 * possibly reading the volume header in from the disk, since there's
2544 * an invariant in the volume package that nUsers>0 ==> vp->header is valid.
2546 * N.B. This call can fail if we can't read in the header!! In this case
2547 * we still guarantee we won't context swap, but the ref count won't be
2548 * incremented (otherwise we'd violate the invariant).
2550 /* NOTE: with the demand attach fileserver extensions, the global lock
2551 * is dropped within VHold */
2552 #ifdef AFS_DEMAND_ATTACH_FS
2554 VHold_r(register Volume * vp)
2558 VCreateReservation_r(vp);
2559 VWaitExclusiveState_r(vp);
2561 LoadVolumeHeader(&error, vp);
2563 VCancelReservation_r(vp);
2567 VCancelReservation_r(vp);
2570 #else /* AFS_DEMAND_ATTACH_FS */
2572 VHold_r(register Volume * vp)
2576 LoadVolumeHeader(&error, vp);
2582 #endif /* AFS_DEMAND_ATTACH_FS */
2585 VHold(register Volume * vp)
2589 retVal = VHold_r(vp);
2595 /***************************************************/
2596 /* get and put volume routines */
2597 /***************************************************/
2600 * put back a heavyweight reference to a volume object.
2602 * @param[in] vp volume object pointer
2604 * @pre VOL_LOCK held
2606 * @post heavyweight volume reference put back.
2607 * depending on state, volume may have been taken offline,
2608 * detached, salvaged, freed, etc.
2610 * @internal volume package internal use only
2613 VPutVolume_r(register Volume * vp)
2615 assert(--vp->nUsers >= 0);
2616 if (vp->nUsers == 0) {
2618 ReleaseVolumeHeader(vp->header);
2619 #ifdef AFS_DEMAND_ATTACH_FS
2620 if (!VCheckDetach(vp)) {
2624 #else /* AFS_DEMAND_ATTACH_FS */
2626 #endif /* AFS_DEMAND_ATTACH_FS */
2631 VPutVolume(register Volume * vp)
2639 /* Get a pointer to an attached volume. The pointer is returned regardless
2640 of whether or not the volume is in service or on/off line. An error
2641 code, however, is returned with an indication of the volume's status */
2643 VGetVolume(Error * ec, Error * client_ec, VolId volumeId)
2647 retVal = GetVolume(ec, client_ec, volumeId, NULL, 0);
2653 VGetVolume_r(Error * ec, VolId volumeId)
2655 return GetVolume(ec, NULL, volumeId, NULL, 0);
2658 /* try to get a volume we've previously looked up */
2659 /* for demand attach fs, caller MUST NOT hold a ref count on vp */
2661 VGetVolumeByVp_r(Error * ec, Volume * vp)
2663 return GetVolume(ec, NULL, vp->hashid, vp, 0);
2666 /* private interface for getting a volume handle
2667 * volumeId must be provided.
2668 * hint is an optional parameter to speed up hash lookups
2669 * flags is not used at this time
2671 /* for demand attach fs, caller MUST NOT hold a ref count on hint */
2673 GetVolume(Error * ec, Error * client_ec, VolId volumeId, Volume * hint, int flags)
2676 /* pull this profiling/debugging code out of regular builds */
2678 #define VGET_CTR_INC(x) x++
2679 unsigned short V0 = 0, V1 = 0, V2 = 0, V3 = 0, V5 = 0, V6 =
2680 0, V7 = 0, V8 = 0, V9 = 0;
2681 unsigned short V10 = 0, V11 = 0, V12 = 0, V13 = 0, V14 = 0, V15 = 0;
2683 #define VGET_CTR_INC(x)
2685 #ifdef AFS_DEMAND_ATTACH_FS
2686 Volume *avp, * rvp = hint;
2689 #ifdef AFS_DEMAND_ATTACH_FS
2691 VCreateReservation_r(rvp);
2693 #endif /* AFS_DEMAND_ATTACH_FS */
2701 vp = VLookupVolume_r(ec, volumeId, vp);
2707 #ifdef AFS_DEMAND_ATTACH_FS
2708 if (rvp && (rvp != vp)) {
2709 /* break reservation on old vp */
2710 VCancelReservation_r(rvp);
2713 #endif /* AFS_DEMAND_ATTACH_FS */
2719 /* Until we have reached an initialization level of 2
2720 * we don't know whether this volume exists or not.
2721 * We can't sleep and retry later because before a volume
2722 * is attached, the caller tries to get it first. Just
2723 * return VOFFLINE and the caller can choose whether to
2724 * retry the command or not. */
2734 IncUInt64(&VStats.hdr_gets);
2736 #ifdef AFS_DEMAND_ATTACH_FS
2737 /* block if someone else is performing an exclusive op on this volume */
2740 VCreateReservation_r(rvp);
2742 VWaitExclusiveState_r(vp);
2744 /* short circuit with VNOVOL in the following circumstances:
2747 * VOL_STATE_SHUTTING_DOWN
2749 if ((V_attachState(vp) == VOL_STATE_ERROR) ||
2750 (V_attachState(vp) == VOL_STATE_SHUTTING_DOWN)) {
2757 * short circuit with VOFFLINE in the following circumstances:
2759 * VOL_STATE_UNATTACHED
2761 if (V_attachState(vp) == VOL_STATE_UNATTACHED) {
2767 /* allowable states:
2775 if (vp->salvage.requested) {
2776 VUpdateSalvagePriority_r(vp);
2779 if (V_attachState(vp) == VOL_STATE_PREATTACHED) {
2780 avp = VAttachVolumeByVp_r(ec, vp, 0);
2783 /* VAttachVolumeByVp_r can return a pointer
2784 * != the vp passed to it under certain
2785 * conditions; make sure we don't leak
2786 * reservations if that happens */
2788 VCancelReservation_r(rvp);
2790 VCreateReservation_r(rvp);
2800 if (!vp->pending_vol_op) {
2815 if ((V_attachState(vp) == VOL_STATE_SALVAGING) ||
2816 (*ec == VSALVAGING)) {
2818 /* see CheckVnode() in afsfileprocs.c for an explanation
2819 * of this error code logic */
2820 afs_uint32 now = FT_ApproxTime();
2821 if ((vp->stats.last_salvage + (10 * 60)) >= now) {
2824 *client_ec = VRESTARTING;
2833 LoadVolumeHeader(ec, vp);
2836 /* Only log the error if it was a totally unexpected error. Simply
2837 * a missing inode is likely to be caused by the volume being deleted */
2838 if (errno != ENXIO || LogLevel)
2839 Log("Volume %u: couldn't reread volume header\n",
2841 #ifdef AFS_DEMAND_ATTACH_FS
2842 if (programType == fileServer) {
2843 VRequestSalvage_r(ec, vp, SALVSYNC_ERROR, VOL_SALVAGE_INVALIDATE_HEADER);
2848 #else /* AFS_DEMAND_ATTACH_FS */
2851 #endif /* AFS_DEMAND_ATTACH_FS */
2855 #ifdef AFS_DEMAND_ATTACH_FS
2857 * this test MUST happen after the volume header is loaded
2859 if (vp->pending_vol_op && !VVolOpLeaveOnline_r(vp, vp->pending_vol_op)) {
2861 /* see CheckVnode() in afsfileprocs.c for an explanation
2862 * of this error code logic */
2863 afs_uint32 now = FT_ApproxTime();
2864 if ((vp->stats.last_vol_op + (10 * 60)) >= now) {
2867 *client_ec = VRESTARTING;
2871 ReleaseVolumeHeader(vp->header);
2875 #endif /* AFS_DEMAND_ATTACH_FS */
2878 if (vp->shuttingDown) {
2885 if (programType == fileServer) {
2887 if (vp->goingOffline) {
2889 #ifdef AFS_DEMAND_ATTACH_FS
2890 /* wait for the volume to go offline */
2891 if (V_attachState(vp) == VOL_STATE_GOING_OFFLINE) {
2892 VWaitStateChange_r(vp);
2894 #elif defined(AFS_PTHREAD_ENV)
2895 VOL_CV_WAIT(&vol_put_volume_cond);
2896 #else /* AFS_PTHREAD_ENV */
2897 LWP_WaitProcess(VPutVolume);
2898 #endif /* AFS_PTHREAD_ENV */
2901 if (vp->specialStatus) {
2903 *ec = vp->specialStatus;
2904 } else if (V_inService(vp) == 0 || V_blessed(vp) == 0) {
2907 } else if (V_inUse(vp) == 0) {
2918 #ifdef AFS_DEMAND_ATTACH_FS
2919 /* if no error, bump nUsers */
2922 VLRU_UpdateAccess_r(vp);
2925 VCancelReservation_r(rvp);
2928 if (client_ec && !*client_ec) {
2931 #else /* AFS_DEMAND_ATTACH_FS */
2932 /* if no error, bump nUsers */
2939 #endif /* AFS_DEMAND_ATTACH_FS */
2946 /***************************************************/
2947 /* Volume offline/detach routines */
2948 /***************************************************/
2950 /* caller MUST hold a heavyweight ref on vp */
2951 #ifdef AFS_DEMAND_ATTACH_FS
2953 VTakeOffline_r(register Volume * vp)
2957 assert(vp->nUsers > 0);
2958 assert(programType == fileServer);
2960 VCreateReservation_r(vp);
2961 VWaitExclusiveState_r(vp);
2963 vp->goingOffline = 1;
2964 V_needsSalvaged(vp) = 1;
2966 VRequestSalvage_r(&error, vp, SALVSYNC_ERROR, 0);
2967 VCancelReservation_r(vp);
2969 #else /* AFS_DEMAND_ATTACH_FS */
2971 VTakeOffline_r(register Volume * vp)
2973 assert(vp->nUsers > 0);
2974 assert(programType == fileServer);
2976 vp->goingOffline = 1;
2977 V_needsSalvaged(vp) = 1;
2979 #endif /* AFS_DEMAND_ATTACH_FS */
2982 VTakeOffline(register Volume * vp)
2990 * force a volume offline.
2992 * @param[in] vp volume object pointer
2993 * @param[in] flags flags (see note below)
2995 * @note the flag VOL_FORCEOFF_NOUPDATE is a recursion control flag
2996 * used when VUpdateVolume_r needs to call VForceOffline_r
2997 * (which in turn would normally call VUpdateVolume_r)
2999 * @see VUpdateVolume_r
3001 * @pre VOL_LOCK must be held.
3002 * for DAFS, caller must hold ref.
3004 * @note for DAFS, it _is safe_ to call this function from an
3007 * @post needsSalvaged flag is set.
3008 * for DAFS, salvage is requested.
3009 * no further references to the volume through the volume
3010 * package will be honored.
3011 * all file descriptor and vnode caches are invalidated.
3013 * @warning this is a heavy-handed interface. it results in
3014 * a volume going offline regardless of the current
3015 * reference count state.
3017 * @internal volume package internal use only
3020 VForceOffline_r(Volume * vp, int flags)
3024 #ifdef AFS_DEMAND_ATTACH_FS
3025 VChangeState_r(vp, VOL_STATE_ERROR);
3030 strcpy(V_offlineMessage(vp),
3031 "Forced offline due to internal error: volume needs to be salvaged");
3032 Log("Volume %u forced offline: it needs salvaging!\n", V_id(vp));
3035 vp->goingOffline = 0;
3036 V_needsSalvaged(vp) = 1;
3037 if (!(flags & VOL_FORCEOFF_NOUPDATE)) {
3038 VUpdateVolume_r(&error, vp, VOL_UPDATE_NOFORCEOFF);
3041 #ifdef AFS_DEMAND_ATTACH_FS
3042 VRequestSalvage_r(&error, vp, SALVSYNC_ERROR, VOL_SALVAGE_INVALIDATE_HEADER);
3043 #endif /* AFS_DEMAND_ATTACH_FS */
3045 #ifdef AFS_PTHREAD_ENV
3046 assert(pthread_cond_broadcast(&vol_put_volume_cond) == 0);
3047 #else /* AFS_PTHREAD_ENV */
3048 LWP_NoYieldSignal(VPutVolume);
3049 #endif /* AFS_PTHREAD_ENV */
3051 VReleaseVolumeHandles_r(vp);
3055 * force a volume offline.
3057 * @param[in] vp volume object pointer
3059 * @see VForceOffline_r
3062 VForceOffline(Volume * vp)
3065 VForceOffline_r(vp, 0);
3069 /* The opposite of VAttachVolume. The volume header is written to disk, with
3070 the inUse bit turned off. A copy of the header is maintained in memory,
3071 however (which is why this is VOffline, not VDetach).
3074 VOffline_r(Volume * vp, char *message)
3077 VolumeId vid = V_id(vp);
3079 assert(programType != volumeUtility);
3084 if (V_offlineMessage(vp)[0] == '\0')
3085 strncpy(V_offlineMessage(vp), message, sizeof(V_offlineMessage(vp)));
3086 V_offlineMessage(vp)[sizeof(V_offlineMessage(vp)) - 1] = '\0';
3088 vp->goingOffline = 1;
3089 #ifdef AFS_DEMAND_ATTACH_FS
3090 VChangeState_r(vp, VOL_STATE_GOING_OFFLINE);
3091 VCreateReservation_r(vp);
3094 /* wait for the volume to go offline */
3095 if (V_attachState(vp) == VOL_STATE_GOING_OFFLINE) {
3096 VWaitStateChange_r(vp);
3098 VCancelReservation_r(vp);
3099 #else /* AFS_DEMAND_ATTACH_FS */
3101 vp = VGetVolume_r(&error, vid); /* Wait for it to go offline */
3102 if (vp) /* In case it was reattached... */
3104 #endif /* AFS_DEMAND_ATTACH_FS */
3108 VOffline(Volume * vp, char *message)
3111 VOffline_r(vp, message);
3115 /* This gets used for the most part by utility routines that don't want
3116 * to keep all the volume headers around. Generally, the file server won't
3117 * call this routine, because then the offline message in the volume header
3118 * (or other information) won't be available to clients. For NAMEI, also
3119 * close the file handles. However, the fileserver does call this during
3120 * an attach following a volume operation.
3123 VDetachVolume_r(Error * ec, Volume * vp)
3126 struct DiskPartition64 *tpartp;
3127 int notifyServer, useDone = FSYNC_VOL_ON;
3129 *ec = 0; /* always "succeeds" */
3130 if (programType == volumeUtility) {
3131 notifyServer = vp->needsPutBack;
3132 if (V_destroyMe(vp) == DESTROY_ME)
3133 useDone = FSYNC_VOL_DONE;
3134 #ifdef AFS_DEMAND_ATTACH_FS
3135 else if (!V_blessed(vp) || !V_inService(vp))
3136 useDone = FSYNC_VOL_LEAVE_OFF;
3139 tpartp = vp->partition;
3141 DeleteVolumeFromHashTable(vp);
3142 vp->shuttingDown = 1;
3143 #ifdef AFS_DEMAND_ATTACH_FS
3144 DeleteVolumeFromVByPList_r(vp);
3146 VChangeState_r(vp, VOL_STATE_SHUTTING_DOWN);
3147 #endif /* AFS_DEMAND_ATTACH_FS */
3149 /* Will be detached sometime in the future--this is OK since volume is offline */
3151 /* XXX the following code should really be moved to VCheckDetach() since the volume
3152 * is not technically detached until the refcounts reach zero
3154 #ifdef FSSYNC_BUILD_CLIENT
3155 if (programType == volumeUtility && notifyServer) {
3157 * Note: The server is not notified in the case of a bogus volume
3158 * explicitly to make it possible to create a volume, do a partial
3159 * restore, then abort the operation without ever putting the volume
3160 * online. This is essential in the case of a volume move operation
3161 * between two partitions on the same server. In that case, there
3162 * would be two instances of the same volume, one of them bogus,
3163 * which the file server would attempt to put on line
3165 FSYNC_VolOp(volume, tpartp->name, useDone, 0, NULL);
3166 /* XXX this code path is only hit by volume utilities, thus
3167 * V_BreakVolumeCallbacks will always be NULL. if we really
3168 * want to break callbacks in this path we need to use FSYNC_VolOp() */
3170 /* Dettaching it so break all callbacks on it */
3171 if (V_BreakVolumeCallbacks) {
3172 Log("volume %u detached; breaking all call backs\n", volume);
3173 (*V_BreakVolumeCallbacks) (volume);
3177 #endif /* FSSYNC_BUILD_CLIENT */
3181 VDetachVolume(Error * ec, Volume * vp)
3184 VDetachVolume_r(ec, vp);
3189 /***************************************************/
3190 /* Volume fd/inode handle closing routines */
3191 /***************************************************/
3193 /* For VDetachVolume, we close all cached file descriptors, but keep
3194 * the Inode handles in case we need to read from a busy volume.
3196 /* for demand attach, caller MUST hold ref count on vp */
3198 VCloseVolumeHandles_r(Volume * vp)
3200 #ifdef AFS_DEMAND_ATTACH_FS
3201 VolState state_save;
3203 state_save = VChangeState_r(vp, VOL_STATE_OFFLINING);
3208 * XXX need to investigate whether we can perform
3209 * DFlushVolume outside of vol_glock_mutex...
3211 * VCloseVnodeFiles_r drops the glock internally */
3212 DFlushVolume(V_id(vp));
3213 VCloseVnodeFiles_r(vp);
3215 #ifdef AFS_DEMAND_ATTACH_FS
3219 /* Too time consuming and unnecessary for the volserver */
3220 if (programType != volumeUtility) {
3221 IH_CONDSYNC(vp->vnodeIndex[vLarge].handle);
3222 IH_CONDSYNC(vp->vnodeIndex[vSmall].handle);
3223 IH_CONDSYNC(vp->diskDataHandle);
3225 IH_CONDSYNC(vp->linkHandle);
3226 #endif /* AFS_NT40_ENV */
3229 IH_REALLYCLOSE(vp->vnodeIndex[vLarge].handle);
3230 IH_REALLYCLOSE(vp->vnodeIndex[vSmall].handle);
3231 IH_REALLYCLOSE(vp->diskDataHandle);
3232 IH_REALLYCLOSE(vp->linkHandle);
3234 #ifdef AFS_DEMAND_ATTACH_FS
3236 VChangeState_r(vp, state_save);
3240 /* For both VForceOffline and VOffline, we close all relevant handles.
3241 * For VOffline, if we re-attach the volume, the files may possible be
3242 * different than before.
3244 /* for demand attach, caller MUST hold a ref count on vp */
3246 VReleaseVolumeHandles_r(Volume * vp)
3248 #ifdef AFS_DEMAND_ATTACH_FS
3249 VolState state_save;
3251 state_save = VChangeState_r(vp, VOL_STATE_DETACHING);
3254 /* XXX need to investigate whether we can perform
3255 * DFlushVolume outside of vol_glock_mutex... */
3256 DFlushVolume(V_id(vp));
3258 VReleaseVnodeFiles_r(vp); /* releases the glock internally */
3260 #ifdef AFS_DEMAND_ATTACH_FS
3264 /* Too time consuming and unnecessary for the volserver */
3265 if (programType != volumeUtility) {
3266 IH_CONDSYNC(vp->vnodeIndex[vLarge].handle);
3267 IH_CONDSYNC(vp->vnodeIndex[vSmall].handle);
3268 IH_CONDSYNC(vp->diskDataHandle);
3270 IH_CONDSYNC(vp->linkHandle);
3271 #endif /* AFS_NT40_ENV */
3274 IH_RELEASE(vp->vnodeIndex[vLarge].handle);
3275 IH_RELEASE(vp->vnodeIndex[vSmall].handle);
3276 IH_RELEASE(vp->diskDataHandle);
3277 IH_RELEASE(vp->linkHandle);
3279 #ifdef AFS_DEMAND_ATTACH_FS
3281 VChangeState_r(vp, state_save);
3286 /***************************************************/
3287 /* Volume write and fsync routines */
3288 /***************************************************/
3291 VUpdateVolume_r(Error * ec, Volume * vp, int flags)
3293 #ifdef AFS_DEMAND_ATTACH_FS
3294 VolState state_save;
3296 if (flags & VOL_UPDATE_WAIT) {
3297 VCreateReservation_r(vp);
3298 VWaitExclusiveState_r(vp);
3303 if (programType == fileServer)
3305 (V_inUse(vp) ? V_nextVnodeUnique(vp) +
3306 200 : V_nextVnodeUnique(vp));
3308 #ifdef AFS_DEMAND_ATTACH_FS
3309 state_save = VChangeState_r(vp, VOL_STATE_UPDATING);
3313 WriteVolumeHeader_r(ec, vp);
3315 #ifdef AFS_DEMAND_ATTACH_FS
3317 VChangeState_r(vp, state_save);
3318 if (flags & VOL_UPDATE_WAIT) {
3319 VCancelReservation_r(vp);
3324 Log("VUpdateVolume: error updating volume header, volume %u (%s)\n",
3325 V_id(vp), V_name(vp));
3326 /* try to update on-disk header,
3327 * while preventing infinite recursion */
3328 if (!(flags & VOL_UPDATE_NOFORCEOFF)) {
3329 VForceOffline_r(vp, VOL_FORCEOFF_NOUPDATE);
3335 VUpdateVolume(Error * ec, Volume * vp)
3338 VUpdateVolume_r(ec, vp, VOL_UPDATE_WAIT);
3343 VSyncVolume_r(Error * ec, Volume * vp, int flags)
3347 #ifdef AFS_DEMAND_ATTACH_FS
3348 VolState state_save;
3351 if (flags & VOL_SYNC_WAIT) {
3352 VUpdateVolume_r(ec, vp, VOL_UPDATE_WAIT);
3354 VUpdateVolume_r(ec, vp, 0);
3357 #ifdef AFS_DEMAND_ATTACH_FS
3358 state_save = VChangeState_r(vp, VOL_STATE_UPDATING);
3361 fdP = IH_OPEN(V_diskDataHandle(vp));
3362 assert(fdP != NULL);
3363 code = FDH_SYNC(fdP);
3366 #ifdef AFS_DEMAND_ATTACH_FS
3368 VChangeState_r(vp, state_save);
3374 VSyncVolume(Error * ec, Volume * vp)
3377 VSyncVolume_r(ec, vp, VOL_SYNC_WAIT);
3382 /***************************************************/
3383 /* Volume dealloaction routines */
3384 /***************************************************/
3386 #ifdef AFS_DEMAND_ATTACH_FS
3388 FreeVolume(Volume * vp)
3390 /* free the heap space, iff it's safe.
3391 * otherwise, pull it out of the hash table, so it
3392 * will get deallocated when all refs to it go away */
3393 if (!VCheckFree(vp)) {
3394 DeleteVolumeFromHashTable(vp);
3395 DeleteVolumeFromVByPList_r(vp);
3397 /* make sure we invalidate the header cache entry */
3398 FreeVolumeHeader(vp);
3401 #endif /* AFS_DEMAND_ATTACH_FS */
3404 ReallyFreeVolume(Volume * vp)
3409 #ifdef AFS_DEMAND_ATTACH_FS
3411 VChangeState_r(vp, VOL_STATE_FREED);
3412 if (vp->pending_vol_op)
3413 free(vp->pending_vol_op);
3414 #endif /* AFS_DEMAND_ATTACH_FS */
3415 for (i = 0; i < nVNODECLASSES; i++)
3416 if (vp->vnodeIndex[i].bitmap)
3417 free(vp->vnodeIndex[i].bitmap);
3418 FreeVolumeHeader(vp);
3419 #ifndef AFS_DEMAND_ATTACH_FS
3420 DeleteVolumeFromHashTable(vp);
3421 #endif /* AFS_DEMAND_ATTACH_FS */
3425 /* check to see if we should shutdown this volume
3426 * returns 1 if volume was freed, 0 otherwise */
3427 #ifdef AFS_DEMAND_ATTACH_FS
3429 VCheckDetach(register Volume * vp)
3433 if (vp->nUsers || vp->nWaiters)
3436 if (vp->shuttingDown) {
3438 VReleaseVolumeHandles_r(vp);
3440 ReallyFreeVolume(vp);
3441 if (programType == fileServer) {
3442 assert(pthread_cond_broadcast(&vol_put_volume_cond) == 0);
3447 #else /* AFS_DEMAND_ATTACH_FS */
3449 VCheckDetach(register Volume * vp)
3456 if (vp->shuttingDown) {
3458 VReleaseVolumeHandles_r(vp);
3459 ReallyFreeVolume(vp);
3460 if (programType == fileServer) {
3461 #if defined(AFS_PTHREAD_ENV)
3462 assert(pthread_cond_broadcast(&vol_put_volume_cond) == 0);
3463 #else /* AFS_PTHREAD_ENV */
3464 LWP_NoYieldSignal(VPutVolume);
3465 #endif /* AFS_PTHREAD_ENV */
3470 #endif /* AFS_DEMAND_ATTACH_FS */
3472 /* check to see if we should offline this volume
3473 * return 1 if volume went offline, 0 otherwise */
3474 #ifdef AFS_DEMAND_ATTACH_FS
3476 VCheckOffline(register Volume * vp)
3478 Volume * rvp = NULL;
3481 if (vp->goingOffline && !vp->nUsers) {
3483 assert(programType == fileServer);
3484 assert((V_attachState(vp) != VOL_STATE_ATTACHED) &&
3485 (V_attachState(vp) != VOL_STATE_FREED) &&
3486 (V_attachState(vp) != VOL_STATE_PREATTACHED) &&
3487 (V_attachState(vp) != VOL_STATE_UNATTACHED));
3491 * VOL_STATE_GOING_OFFLINE
3492 * VOL_STATE_SHUTTING_DOWN
3493 * VIsErrorState(V_attachState(vp))
3494 * VIsExclusiveState(V_attachState(vp))
3497 VCreateReservation_r(vp);
3498 VChangeState_r(vp, VOL_STATE_OFFLINING);
3501 /* must clear the goingOffline flag before we drop the glock */
3502 vp->goingOffline = 0;
3507 /* perform async operations */
3508 VUpdateVolume_r(&error, vp, 0);
3509 VCloseVolumeHandles_r(vp);
3512 Log("VOffline: Volume %u (%s) is now offline", V_id(vp),
3514 if (V_offlineMessage(vp)[0])
3515 Log(" (%s)", V_offlineMessage(vp));
3519 /* invalidate the volume header cache entry */
3520 FreeVolumeHeader(vp);
3522 /* if nothing changed state to error or salvaging,
3523 * drop state to unattached */
3524 if (!VIsErrorState(V_attachState(vp))) {
3525 VChangeState_r(vp, VOL_STATE_UNATTACHED);
3527 VCancelReservation_r(vp);
3528 /* no usage of vp is safe beyond this point */
3532 #else /* AFS_DEMAND_ATTACH_FS */
3534 VCheckOffline(register Volume * vp)
3536 Volume * rvp = NULL;
3539 if (vp->goingOffline && !vp->nUsers) {
3541 assert(programType == fileServer);
3544 vp->goingOffline = 0;
3546 VUpdateVolume_r(&error, vp, 0);
3547 VCloseVolumeHandles_r(vp);
3549 Log("VOffline: Volume %u (%s) is now offline", V_id(vp),
3551 if (V_offlineMessage(vp)[0])
3552 Log(" (%s)", V_offlineMessage(vp));
3555 FreeVolumeHeader(vp);
3556 #ifdef AFS_PTHREAD_ENV
3557 assert(pthread_cond_broadcast(&vol_put_volume_cond) == 0);
3558 #else /* AFS_PTHREAD_ENV */
3559 LWP_NoYieldSignal(VPutVolume);
3560 #endif /* AFS_PTHREAD_ENV */
3564 #endif /* AFS_DEMAND_ATTACH_FS */
3566 /***************************************************/
3567 /* demand attach fs ref counting routines */
3568 /***************************************************/
3570 #ifdef AFS_DEMAND_ATTACH_FS
3571 /* the following two functions handle reference counting for
3572 * asynchronous operations on volume structs.
3574 * their purpose is to prevent a VDetachVolume or VShutdown
3575 * from free()ing the Volume struct during an async i/o op */
3577 /* register with the async volume op ref counter */
3578 /* VCreateReservation_r moved into inline code header because it
3579 * is now needed in vnode.c -- tkeiser 11/20/2007
3583 * decrement volume-package internal refcount.
3585 * @param vp volume object pointer
3587 * @internal volume package internal use only
3590 * @arg VOL_LOCK is held
3591 * @arg lightweight refcount held
3593 * @post volume waiters refcount is decremented; volume may
3594 * have been deallocated/shutdown/offlined/salvaged/
3595 * whatever during the process
3597 * @warning once you have tossed your last reference (you can acquire
3598 * lightweight refs recursively) it is NOT SAFE to reference
3599 * a volume object pointer ever again
3601 * @see VCreateReservation_r
3603 * @note DEMAND_ATTACH_FS only
3606 VCancelReservation_r(Volume * vp)
3608 assert(--vp->nWaiters >= 0);
3609 if (vp->nWaiters == 0) {
3611 if (!VCheckDetach(vp)) {
3618 /* check to see if we should free this volume now
3619 * return 1 if volume was freed, 0 otherwise */
3621 VCheckFree(Volume * vp)
3624 if ((vp->nUsers == 0) &&
3625 (vp->nWaiters == 0) &&
3626 !(V_attachFlags(vp) & (VOL_IN_HASH |
3630 ReallyFreeVolume(vp);
3635 #endif /* AFS_DEMAND_ATTACH_FS */
3638 /***************************************************/
3639 /* online volume operations routines */
3640 /***************************************************/
3642 #ifdef AFS_DEMAND_ATTACH_FS
3644 * register a volume operation on a given volume.
3646 * @param[in] vp volume object
3647 * @param[in] vopinfo volume operation info object
3649 * @pre VOL_LOCK is held
3651 * @post volume operation info object attached to volume object.
3652 * volume operation statistics updated.
3654 * @note by "attached" we mean a copy of the passed in object is made
3656 * @internal volume package internal use only
3659 VRegisterVolOp_r(Volume * vp, FSSYNC_VolOp_info * vopinfo)
3661 FSSYNC_VolOp_info * info;
3663 /* attach a vol op info node to the volume struct */
3664 info = (FSSYNC_VolOp_info *) malloc(sizeof(FSSYNC_VolOp_info));
3665 assert(info != NULL);
3666 memcpy(info, vopinfo, sizeof(FSSYNC_VolOp_info));
3667 vp->pending_vol_op = info;
3670 vp->stats.last_vol_op = FT_ApproxTime();
3671 vp->stats.vol_ops++;
3672 IncUInt64(&VStats.vol_ops);
3678 * deregister the volume operation attached to this volume.
3680 * @param[in] vp volume object pointer
3682 * @pre VOL_LOCK is held
3684 * @post the volume operation info object is detached from the volume object
3686 * @internal volume package internal use only
3689 VDeregisterVolOp_r(Volume * vp)
3691 if (vp->pending_vol_op) {
3692 free(vp->pending_vol_op);
3693 vp->pending_vol_op = NULL;
3697 #endif /* AFS_DEMAND_ATTACH_FS */
3700 * determine whether it is safe to leave a volume online during
3701 * the volume operation described by the vopinfo object.
3703 * @param[in] vp volume object
3704 * @param[in] vopinfo volume operation info object
3706 * @return whether it is safe to leave volume online
3707 * @retval 0 it is NOT SAFE to leave the volume online
3708 * @retval 1 it is safe to leave the volume online during the operation
3711 * @arg VOL_LOCK is held
3712 * @arg disk header attached to vp (heavyweight ref on vp will guarantee
3713 * this condition is met)
3715 * @internal volume package internal use only
3718 VVolOpLeaveOnline_r(Volume * vp, FSSYNC_VolOp_info * vopinfo)
3720 return (vopinfo->com.command == FSYNC_VOL_NEEDVOLUME &&
3721 (vopinfo->com.reason == V_READONLY ||
3722 (!VolumeWriteable(vp) &&
3723 (vopinfo->com.reason == V_CLONE ||
3724 vopinfo->com.reason == V_DUMP))));
3728 * determine whether VBUSY should be set during this volume operation.
3730 * @param[in] vp volume object
3731 * @param[in] vopinfo volume operation info object
3733 * @return whether VBUSY should be set
3734 * @retval 0 VBUSY does NOT need to be set
3735 * @retval 1 VBUSY SHOULD be set
3737 * @pre VOL_LOCK is held
3739 * @internal volume package internal use only
3742 VVolOpSetVBusy_r(Volume * vp, FSSYNC_VolOp_info * vopinfo)
3744 return (vopinfo->com.command == FSYNC_VOL_NEEDVOLUME &&
3745 (vopinfo->com.reason == V_CLONE ||
3746 vopinfo->com.reason == V_DUMP));
3750 /***************************************************/
3751 /* online salvager routines */
3752 /***************************************************/
3753 #if defined(AFS_DEMAND_ATTACH_FS)
3754 #define SALVAGE_PRIO_UPDATE_INTERVAL 3 /**< number of seconds between prio updates */
3755 #define SALVAGE_COUNT_MAX 16 /**< number of online salvages we
3756 * allow before moving the volume
3757 * into a permanent error state
3759 * once this threshold is reached,
3760 * the operator will have to manually
3761 * issue a 'bos salvage' to bring
3762 * the volume back online
3766 * check whether a salvage needs to be performed on this volume.
3768 * @param[in] vp pointer to volume object
3770 * @return status code
3771 * @retval 0 no salvage scheduled
3772 * @retval 1 a salvage has been scheduled with the salvageserver
3774 * @pre VOL_LOCK is held
3776 * @post if salvage request flag is set and nUsers and nWaiters are zero,
3777 * then a salvage will be requested
3779 * @note this is one of the event handlers called by VCancelReservation_r
3781 * @see VCancelReservation_r
3783 * @internal volume package internal use only.
3786 VCheckSalvage(register Volume * vp)
3789 #ifdef SALVSYNC_BUILD_CLIENT
3790 if (vp->nUsers || vp->nWaiters)
3792 if (vp->salvage.requested) {
3793 VScheduleSalvage_r(vp);
3796 #endif /* SALVSYNC_BUILD_CLIENT */
3801 * request volume salvage.
3803 * @param[out] ec computed client error code
3804 * @param[in] vp volume object pointer
3805 * @param[in] reason reason code (passed to salvageserver via SALVSYNC)
3806 * @param[in] flags see flags note below
3809 * VOL_SALVAGE_INVALIDATE_HEADER causes volume header cache entry
3810 * to be invalidated.
3812 * @pre VOL_LOCK is held.
3814 * @post volume state is changed.
3815 * for fileserver, salvage will be requested once refcount reaches zero.
3817 * @return operation status code
3818 * @retval 0 volume salvage will occur
3819 * @retval 1 volume salvage could not be scheduled
3821 * @note DAFS fileserver only
3823 * @note this call does not synchronously schedule a volume salvage. rather,
3824 * it sets volume state so that when volume refcounts reach zero, a
3825 * volume salvage will occur. by "refcounts", we mean both nUsers and
3826 * nWaiters must be zero.
3828 * @internal volume package internal use only.
3831 VRequestSalvage_r(Error * ec, Volume * vp, int reason, int flags)
3835 * for DAFS volume utilities, transition to error state
3836 * (at some point in the future, we should consider
3837 * making volser talk to salsrv)
3839 if (programType != fileServer) {
3840 VChangeState_r(vp, VOL_STATE_ERROR);
3845 if (!vp->salvage.requested) {
3846 vp->salvage.requested = 1;
3847 vp->salvage.reason = reason;
3848 vp->stats.last_salvage = FT_ApproxTime();
3849 if (flags & VOL_SALVAGE_INVALIDATE_HEADER) {
3850 /* XXX this should likely be changed to FreeVolumeHeader() */
3851 ReleaseVolumeHeader(vp->header);
3853 if (vp->stats.salvages < SALVAGE_COUNT_MAX) {
3854 VChangeState_r(vp, VOL_STATE_SALVAGING);
3857 Log("VRequestSalvage: volume %u online salvaged too many times; forced offline.\n", vp->hashid);
3858 VChangeState_r(vp, VOL_STATE_ERROR);
3867 * update salvageserver scheduling priority for a volume.
3869 * @param[in] vp pointer to volume object
3871 * @return operation status
3873 * @retval 1 request denied, or SALVSYNC communications failure
3875 * @pre VOL_LOCK is held.
3877 * @post in-core salvage priority counter is incremented. if at least
3878 * SALVAGE_PRIO_UPDATE_INTERVAL seconds have elapsed since the
3879 * last SALVSYNC_RAISEPRIO request, we contact the salvageserver
3880 * to update its priority queue. if no salvage is scheduled,
3881 * this function is a no-op.
3883 * @note DAFS fileserver only
3885 * @note this should be called whenever a VGetVolume fails due to a
3886 * pending salvage request
3888 * @todo should set exclusive state and drop glock around salvsync call
3890 * @internal volume package internal use only.
3893 VUpdateSalvagePriority_r(Volume * vp)
3898 #ifdef SALVSYNC_BUILD_CLIENT
3900 now = FT_ApproxTime();
3902 /* update the salvageserver priority queue occasionally so that
3903 * frequently requested volumes get moved to the head of the queue
3905 if ((vp->salvage.scheduled) &&
3906 (vp->stats.last_salvage_req < (now-SALVAGE_PRIO_UPDATE_INTERVAL))) {
3907 code = SALVSYNC_SalvageVolume(vp->hashid,
3908 VPartitionPath(vp->partition),
3913 vp->stats.last_salvage_req = now;
3914 if (code != SYNC_OK) {
3918 #endif /* SALVSYNC_BUILD_CLIENT */
3924 * schedule a salvage with the salvage server.
3926 * @param[in] vp pointer to volume object
3928 * @return operation status
3929 * @retval 0 salvage scheduled successfully
3930 * @retval 1 salvage not scheduled, or SALVSYNC com error
3933 * @arg VOL_LOCK is held.
3934 * @arg nUsers and nWaiters should be zero.
3936 * @post salvageserver is sent a salvage request
3938 * @note DAFS fileserver only
3940 * @internal volume package internal use only.
3943 VScheduleSalvage_r(Volume * vp)
3946 #ifdef SALVSYNC_BUILD_CLIENT
3947 VolState state_save;
3950 if (vp->nWaiters || vp->nUsers) {
3954 /* prevent endless salvage,attach,salvage,attach,... loops */
3955 if (vp->stats.salvages >= SALVAGE_COUNT_MAX)
3958 if (!vp->salvage.scheduled) {
3959 /* if we haven't previously scheduled a salvage, do so now
3961 * set the volume to an exclusive state and drop the lock
3962 * around the SALVSYNC call
3964 * note that we do NOT acquire a reservation here -- doing so
3965 * could result in unbounded recursion
3967 strlcpy(partName, VPartitionPath(vp->partition), sizeof(partName));
3968 state_save = VChangeState_r(vp, VOL_STATE_SALVSYNC_REQ);
3969 V_attachFlags(vp) |= VOL_IS_BUSY;
3972 /* can't use V_id() since there's no guarantee
3973 * we have the disk data header at this point */
3974 code = SALVSYNC_SalvageVolume(vp->hashid,
3981 VChangeState_r(vp, state_save);
3982 V_attachFlags(vp) &= ~(VOL_IS_BUSY);
3984 if (code == SYNC_OK) {
3985 vp->salvage.scheduled = 1;
3986 vp->stats.salvages++;
3987 vp->stats.last_salvage_req = FT_ApproxTime();
3988 IncUInt64(&VStats.salvages);
3992 case SYNC_BAD_COMMAND:
3993 case SYNC_COM_ERROR:
3996 Log("VScheduleSalvage_r: SALVSYNC request denied\n");
3999 Log("VScheduleSalvage_r: SALVSYNC unknown protocol error\n");
4004 #endif /* SALVSYNC_BUILD_CLIENT */
4009 * ask salvageserver to cancel a scheduled salvage operation.
4011 * @param[in] vp pointer to volume object
4012 * @param[in] reason SALVSYNC protocol reason code
4014 * @return operation status
4016 * @retval 1 request failed
4018 * @pre VOL_LOCK is held.
4020 * @post salvageserver is sent a request to cancel the volume salvage
4022 * @todo should set exclusive state and drop glock around salvsync call
4024 * @internal volume package internal use only.
4027 VCancelSalvage_r(Volume * vp, int reason)
4031 #ifdef SALVSYNC_BUILD_CLIENT
4032 if (vp->salvage.scheduled) {
4033 code = SALVSYNC_SalvageVolume(vp->hashid,
4034 VPartitionPath(vp->partition),
4039 if (code == SYNC_OK) {
4040 vp->salvage.scheduled = 0;
4045 #endif /* SALVSYNC_BUILD_CLIENT */
4050 #ifdef SALVSYNC_BUILD_CLIENT
4052 * connect to the salvageserver SYNC service.
4054 * @return operation status
4058 * @post connection to salvageserver SYNC service established
4060 * @see VConnectSALV_r
4061 * @see VDisconnectSALV
4062 * @see VReconnectSALV
4069 retVal = VConnectSALV_r();
4075 * connect to the salvageserver SYNC service.
4077 * @return operation status
4081 * @pre VOL_LOCK is held.
4083 * @post connection to salvageserver SYNC service established
4086 * @see VDisconnectSALV_r
4087 * @see VReconnectSALV_r
4088 * @see SALVSYNC_clientInit
4090 * @internal volume package internal use only.
4093 VConnectSALV_r(void)
4095 return SALVSYNC_clientInit();
4099 * disconnect from the salvageserver SYNC service.
4101 * @return operation status
4104 * @pre client should have a live connection to the salvageserver
4106 * @post connection to salvageserver SYNC service destroyed
4108 * @see VDisconnectSALV_r
4110 * @see VReconnectSALV
4113 VDisconnectSALV(void)
4117 VDisconnectSALV_r();
4123 * disconnect from the salvageserver SYNC service.
4125 * @return operation status
4129 * @arg VOL_LOCK is held.
4130 * @arg client should have a live connection to the salvageserver.
4132 * @post connection to salvageserver SYNC service destroyed
4134 * @see VDisconnectSALV
4135 * @see VConnectSALV_r
4136 * @see VReconnectSALV_r
4137 * @see SALVSYNC_clientFinis
4139 * @internal volume package internal use only.
4142 VDisconnectSALV_r(void)
4144 return SALVSYNC_clientFinis();
4148 * disconnect and then re-connect to the salvageserver SYNC service.
4150 * @return operation status
4154 * @pre client should have a live connection to the salvageserver
4156 * @post old connection is dropped, and a new one is established
4159 * @see VDisconnectSALV
4160 * @see VReconnectSALV_r
4163 VReconnectSALV(void)
4167 retVal = VReconnectSALV_r();
4173 * disconnect and then re-connect to the salvageserver SYNC service.
4175 * @return operation status
4180 * @arg VOL_LOCK is held.
4181 * @arg client should have a live connection to the salvageserver.
4183 * @post old connection is dropped, and a new one is established
4185 * @see VConnectSALV_r
4186 * @see VDisconnectSALV
4187 * @see VReconnectSALV
4188 * @see SALVSYNC_clientReconnect
4190 * @internal volume package internal use only.
4193 VReconnectSALV_r(void)
4195 return SALVSYNC_clientReconnect();
4197 #endif /* SALVSYNC_BUILD_CLIENT */
4198 #endif /* AFS_DEMAND_ATTACH_FS */
4201 /***************************************************/
4202 /* FSSYNC routines */
4203 /***************************************************/
4205 /* This must be called by any volume utility which needs to run while the
4206 file server is also running. This is separated from VInitVolumePackage so
4207 that a utility can fork--and each of the children can independently
4208 initialize communication with the file server */
4209 #ifdef FSSYNC_BUILD_CLIENT
4211 * connect to the fileserver SYNC service.
4213 * @return operation status
4218 * @arg VInit must equal 2.
4219 * @arg Program Type must not be fileserver or salvager.
4221 * @post connection to fileserver SYNC service established
4224 * @see VDisconnectFS
4225 * @see VChildProcReconnectFS
4232 retVal = VConnectFS_r();
4238 * connect to the fileserver SYNC service.
4240 * @return operation status
4245 * @arg VInit must equal 2.
4246 * @arg Program Type must not be fileserver or salvager.
4247 * @arg VOL_LOCK is held.
4249 * @post connection to fileserver SYNC service established
4252 * @see VDisconnectFS_r
4253 * @see VChildProcReconnectFS_r
4255 * @internal volume package internal use only.
4261 assert((VInit == 2) &&
4262 (programType != fileServer) &&
4263 (programType != salvager));
4264 rc = FSYNC_clientInit();
4271 * disconnect from the fileserver SYNC service.
4274 * @arg client should have a live connection to the fileserver.
4275 * @arg VOL_LOCK is held.
4276 * @arg Program Type must not be fileserver or salvager.
4278 * @post connection to fileserver SYNC service destroyed
4280 * @see VDisconnectFS
4282 * @see VChildProcReconnectFS_r
4284 * @internal volume package internal use only.
4287 VDisconnectFS_r(void)
4289 assert((programType != fileServer) &&
4290 (programType != salvager));
4291 FSYNC_clientFinis();
4296 * disconnect from the fileserver SYNC service.
4299 * @arg client should have a live connection to the fileserver.
4300 * @arg Program Type must not be fileserver or salvager.
4302 * @post connection to fileserver SYNC service destroyed
4304 * @see VDisconnectFS_r
4306 * @see VChildProcReconnectFS
4317 * connect to the fileserver SYNC service from a child process following a fork.
4319 * @return operation status
4324 * @arg VOL_LOCK is held.
4325 * @arg current FSYNC handle is shared with a parent process
4327 * @post current FSYNC handle is discarded and a new connection to the
4328 * fileserver SYNC service is established
4330 * @see VChildProcReconnectFS
4332 * @see VDisconnectFS_r
4334 * @internal volume package internal use only.
4337 VChildProcReconnectFS_r(void)
4339 return FSYNC_clientChildProcReconnect();
4343 * connect to the fileserver SYNC service from a child process following a fork.
4345 * @return operation status
4349 * @pre current FSYNC handle is shared with a parent process
4351 * @post current FSYNC handle is discarded and a new connection to the
4352 * fileserver SYNC service is established
4354 * @see VChildProcReconnectFS_r
4356 * @see VDisconnectFS
4359 VChildProcReconnectFS(void)
4363 ret = VChildProcReconnectFS_r();
4367 #endif /* FSSYNC_BUILD_CLIENT */
4370 /***************************************************/
4371 /* volume bitmap routines */
4372 /***************************************************/
4375 * For demand attach fs, flags parameter controls
4376 * locking behavior. If (flags & VOL_ALLOC_BITMAP_WAIT)
4377 * is set, then this function will create a reservation
4378 * and block on any other exclusive operations. Otherwise,
4379 * this function assumes the caller already has exclusive
4380 * access to vp, and we just change the volume state.
4383 VAllocBitmapEntry_r(Error * ec, Volume * vp,
4384 struct vnodeIndex *index, int flags)
4387 register byte *bp, *ep;
4388 #ifdef AFS_DEMAND_ATTACH_FS
4389 VolState state_save;
4390 #endif /* AFS_DEMAND_ATTACH_FS */
4394 /* This test is probably redundant */
4395 if (!VolumeWriteable(vp)) {
4396 *ec = (bit32) VREADONLY;
4400 #ifdef AFS_DEMAND_ATTACH_FS
4401 if (flags & VOL_ALLOC_BITMAP_WAIT) {
4402 VCreateReservation_r(vp);
4403 VWaitExclusiveState_r(vp);
4405 state_save = VChangeState_r(vp, VOL_STATE_GET_BITMAP);
4406 #endif /* AFS_DEMAND_ATTACH_FS */
4409 if ((programType == fileServer) && !index->bitmap) {
4411 #ifndef AFS_DEMAND_ATTACH_FS
4412 /* demand attach fs uses the volume state to avoid races.
4413 * specialStatus field is not used at all */
4415 if (vp->specialStatus == VBUSY) {
4416 if (vp->goingOffline) { /* vos dump waiting for the volume to
4417 * go offline. We probably come here
4418 * from AddNewReadableResidency */
4421 while (vp->specialStatus == VBUSY) {
4422 #ifdef AFS_PTHREAD_ENV
4426 #else /* !AFS_PTHREAD_ENV */
4428 #endif /* !AFS_PTHREAD_ENV */
4432 #endif /* !AFS_DEMAND_ATTACH_FS */
4434 if (!index->bitmap) {
4435 #ifndef AFS_DEMAND_ATTACH_FS
4436 vp->specialStatus = VBUSY; /* Stop anyone else from using it. */
4437 #endif /* AFS_DEMAND_ATTACH_FS */
4438 for (i = 0; i < nVNODECLASSES; i++) {
4439 VGetBitmap_r(ec, vp, i);
4441 #ifdef AFS_DEMAND_ATTACH_FS
4442 VRequestSalvage_r(ec, vp, SALVSYNC_ERROR, VOL_SALVAGE_INVALIDATE_HEADER);
4443 #else /* AFS_DEMAND_ATTACH_FS */
4444 DeleteVolumeFromHashTable(vp);
4445 vp->shuttingDown = 1; /* Let who has it free it. */
4446 vp->specialStatus = 0;
4447 #endif /* AFS_DEMAND_ATTACH_FS */
4452 #ifndef AFS_DEMAND_ATTACH_FS
4454 vp->specialStatus = 0; /* Allow others to have access. */
4455 #endif /* AFS_DEMAND_ATTACH_FS */
4458 #endif /* BITMAP_LATER */
4460 #ifdef AFS_DEMAND_ATTACH_FS
4462 #endif /* AFS_DEMAND_ATTACH_FS */
4463 bp = index->bitmap + index->bitmapOffset;
4464 ep = index->bitmap + index->bitmapSize;
4466 if ((*(bit32 *) bp) != (bit32) 0xffffffff) {
4468 index->bitmapOffset = (afs_uint32) (bp - index->bitmap);
4471 o = ffs(~*bp) - 1; /* ffs is documented in BSTRING(3) */
4473 ret = (VnodeId) ((bp - index->bitmap) * 8 + o);
4474 #ifdef AFS_DEMAND_ATTACH_FS
4476 #endif /* AFS_DEMAND_ATTACH_FS */
4479 bp += sizeof(bit32) /* i.e. 4 */ ;
4481 /* No bit map entry--must grow bitmap */
4483 realloc(index->bitmap, index->bitmapSize + VOLUME_BITMAP_GROWSIZE);
4486 bp += index->bitmapSize;
4487 memset(bp, 0, VOLUME_BITMAP_GROWSIZE);
4488 index->bitmapOffset = index->bitmapSize;
4489 index->bitmapSize += VOLUME_BITMAP_GROWSIZE;
4491 ret = index->bitmapOffset * 8;
4492 #ifdef AFS_DEMAND_ATTACH_FS
4494 #endif /* AFS_DEMAND_ATTACH_FS */
4497 #ifdef AFS_DEMAND_ATTACH_FS
4498 VChangeState_r(vp, state_save);
4499 if (flags & VOL_ALLOC_BITMAP_WAIT) {
4500 VCancelReservation_r(vp);
4502 #endif /* AFS_DEMAND_ATTACH_FS */
4507 VAllocBitmapEntry(Error * ec, Volume * vp, register struct vnodeIndex * index)
4511 retVal = VAllocBitmapEntry_r(ec, vp, index, VOL_ALLOC_BITMAP_WAIT);
4517 VFreeBitMapEntry_r(Error * ec, register struct vnodeIndex *index,
4520 unsigned int offset;
4526 #endif /* BITMAP_LATER */
4527 offset = bitNumber >> 3;
4528 if (offset >= index->bitmapSize) {
4532 if (offset < index->bitmapOffset)
4533 index->bitmapOffset = offset & ~3; /* Truncate to nearest bit32 */
4534 *(index->bitmap + offset) &= ~(1 << (bitNumber & 0x7));
4538 VFreeBitMapEntry(Error * ec, register struct vnodeIndex *index,
4542 VFreeBitMapEntry_r(ec, index, bitNumber);
4546 /* this function will drop the glock internally.
4547 * for old pthread fileservers, this is safe thanks to vbusy.
4549 * for demand attach fs, caller must have already called
4550 * VCreateReservation_r and VWaitExclusiveState_r */
4552 VGetBitmap_r(Error * ec, Volume * vp, VnodeClass class)
4554 StreamHandle_t *file;
4557 struct VnodeClassInfo *vcp = &VnodeClassInfo[class];
4558 struct vnodeIndex *vip = &vp->vnodeIndex[class];
4559 struct VnodeDiskObject *vnode;
4560 unsigned int unique = 0;
4564 #endif /* BITMAP_LATER */
4565 #ifdef AFS_DEMAND_ATTACH_FS
4566 VolState state_save;
4567 #endif /* AFS_DEMAND_ATTACH_FS */
4571 #ifdef AFS_DEMAND_ATTACH_FS
4572 state_save = VChangeState_r(vp, VOL_STATE_GET_BITMAP);
4573 #endif /* AFS_DEMAND_ATTACH_FS */
4576 fdP = IH_OPEN(vip->handle);
4577 assert(fdP != NULL);
4578 file = FDH_FDOPEN(fdP, "r");
4579 assert(file != NULL);
4580 vnode = (VnodeDiskObject *) malloc(vcp->diskSize);
4581 assert(vnode != NULL);
4582 size = OS_SIZE(fdP->fd_fd);
4584 nVnodes = (size <= vcp->diskSize ? 0 : size - vcp->diskSize)
4586 vip->bitmapSize = ((nVnodes / 8) + 10) / 4 * 4; /* The 10 is a little extra so
4587 * a few files can be created in this volume,
4588 * the whole thing is rounded up to nearest 4
4589 * bytes, because the bit map allocator likes
4592 BitMap = (byte *) calloc(1, vip->bitmapSize);
4593 assert(BitMap != NULL);
4594 #else /* BITMAP_LATER */
4595 vip->bitmap = (byte *) calloc(1, vip->bitmapSize);
4596 assert(vip->bitmap != NULL);
4597 vip->bitmapOffset = 0;
4598 #endif /* BITMAP_LATER */
4599 if (STREAM_SEEK(file, vcp->diskSize, 0) != -1) {
4601 for (bitNumber = 0; bitNumber < nVnodes + 100; bitNumber++) {
4602 if (STREAM_READ(vnode, vcp->diskSize, 1, file) != 1)
4604 if (vnode->type != vNull) {
4605 if (vnode->vnodeMagic != vcp->magic) {
4606 Log("GetBitmap: addled vnode index in volume %s; volume needs salvage\n", V_name(vp));
4611 *(BitMap + (bitNumber >> 3)) |= (1 << (bitNumber & 0x7));
4612 #else /* BITMAP_LATER */
4613 *(vip->bitmap + (bitNumber >> 3)) |= (1 << (bitNumber & 0x7));
4614 #endif /* BITMAP_LATER */
4615 if (unique <= vnode->uniquifier)
4616 unique = vnode->uniquifier + 1;
4618 #ifndef AFS_PTHREAD_ENV
4619 if ((bitNumber & 0x00ff) == 0x0ff) { /* every 256 iterations */
4622 #endif /* !AFS_PTHREAD_ENV */
4625 if (vp->nextVnodeUnique < unique) {
4626 Log("GetBitmap: bad volume uniquifier for volume %s; volume needs salvage\n", V_name(vp));
4629 /* Paranoia, partly justified--I think fclose after fdopen
4630 * doesn't seem to close fd. In any event, the documentation
4631 * doesn't specify, so it's safer to close it twice.
4639 /* There may have been a racing condition with some other thread, both
4640 * creating the bitmaps for this volume. If the other thread was faster
4641 * the pointer to bitmap should already be filled and we can free ours.
4643 if (vip->bitmap == NULL) {
4644 vip->bitmap = BitMap;
4645 vip->bitmapOffset = 0;
4647 free((byte *) BitMap);
4648 #endif /* BITMAP_LATER */
4649 #ifdef AFS_DEMAND_ATTACH_FS
4650 VChangeState_r(vp, state_save);
4651 #endif /* AFS_DEMAND_ATTACH_FS */
4655 /***************************************************/
4656 /* Volume Path and Volume Number utility routines */
4657 /***************************************************/
4660 * find the first occurrence of a volume header file and return the path.
4662 * @param[out] ec outbound error code
4663 * @param[in] volumeId volume id to find
4664 * @param[out] partitionp pointer to disk partition path string
4665 * @param[out] namep pointer to volume header file name string
4667 * @post path to first occurrence of volume header is returned in partitionp
4668 * and namep, or ec is set accordingly.
4670 * @warning this function is NOT re-entrant -- partitionp and namep point to
4671 * static data segments
4673 * @note if a volume utility inadvertently leaves behind a stale volume header
4674 * on a vice partition, it is possible for callers to get the wrong one,
4675 * depending on the order of the disk partition linked list.
4677 * @internal volume package internal use only.
4680 GetVolumePath(Error * ec, VolId volumeId, char **partitionp, char **namep)
4682 static char partition[VMAXPATHLEN], name[VMAXPATHLEN];
4683 char path[VMAXPATHLEN];
4685 struct DiskPartition64 *dp;
4689 (void)afs_snprintf(&name[1], (sizeof name) - 1, VFORMAT, volumeId);
4690 for (dp = DiskPartitionList; dp; dp = dp->next) {
4691 struct afs_stat status;
4692 strcpy(path, VPartitionPath(dp));
4694 if (afs_stat(path, &status) == 0) {
4695 strcpy(partition, dp->name);
4702 *partitionp = *namep = NULL;
4704 *partitionp = partition;
4710 * extract a volume number from a volume header filename string.
4712 * @param[in] name volume header filename string
4714 * @return volume number
4716 * @note the string must be of the form VFORMAT. the only permissible
4717 * deviation is a leading '/' character.
4722 VolumeNumber(char *name)
4726 return atoi(name + 1);
4730 * compute the volume header filename.
4732 * @param[in] volumeId
4734 * @return volume header filename
4736 * @post volume header filename string is constructed
4738 * @warning this function is NOT re-entrant -- the returned string is
4739 * stored in a static char array. see VolumeExternalName_r
4740 * for a re-entrant equivalent.
4742 * @see VolumeExternalName_r
4744 * @deprecated due to the above re-entrancy warning, this interface should
4745 * be considered deprecated. Please use VolumeExternalName_r
4749 VolumeExternalName(VolumeId volumeId)
4751 static char name[VMAXPATHLEN];
4752 (void)afs_snprintf(name, sizeof name, VFORMAT, volumeId);
4757 * compute the volume header filename.
4759 * @param[in] volumeId
4760 * @param[inout] name array in which to store filename
4761 * @param[in] len length of name array
4763 * @return result code from afs_snprintf
4765 * @see VolumeExternalName
4768 * @note re-entrant equivalent of VolumeExternalName
4770 * @internal volume package internal use only.
4773 VolumeExternalName_r(VolumeId volumeId, char * name, size_t len)
4775 return afs_snprintf(name, len, VFORMAT, volumeId);
4779 /***************************************************/
4780 /* Volume Usage Statistics routines */
4781 /***************************************************/
4783 #if OPENAFS_VOL_STATS
4784 #define OneDay (86400) /* 24 hours' worth of seconds */
4786 #define OneDay (24*60*60) /* 24 hours */
4787 #endif /* OPENAFS_VOL_STATS */
4789 #define Midnight(date) ((date-TimeZoneCorrection)/OneDay*OneDay+TimeZoneCorrection)
4791 /*------------------------------------------------------------------------
4792 * [export] VAdjustVolumeStatistics
4795 * If we've passed midnight, we need to update all the day use
4796 * statistics as well as zeroing the detailed volume statistics
4797 * (if we are implementing them).
4800 * vp : Pointer to the volume structure describing the lucky
4801 * volume being considered for update.
4807 * Nothing interesting.
4811 *------------------------------------------------------------------------*/
4814 VAdjustVolumeStatistics_r(register Volume * vp)
4816 unsigned int now = FT_ApproxTime();
4818 if (now - V_dayUseDate(vp) > OneDay) {
4819 register int ndays, i;
4821 ndays = (now - V_dayUseDate(vp)) / OneDay;
4822 for (i = 6; i > ndays - 1; i--)
4823 V_weekUse(vp)[i] = V_weekUse(vp)[i - ndays];
4824 for (i = 0; i < ndays - 1 && i < 7; i++)
4825 V_weekUse(vp)[i] = 0;
4827 V_weekUse(vp)[ndays - 1] = V_dayUse(vp);
4829 V_dayUseDate(vp) = Midnight(now);
4831 #if OPENAFS_VOL_STATS
4833 * All we need to do is bzero the entire VOL_STATS_BYTES of
4834 * the detailed volume statistics area.
4836 memset((char *)(V_stat_area(vp)), 0, VOL_STATS_BYTES);
4837 #endif /* OPENAFS_VOL_STATS */
4840 /*It's been more than a day of collection */
4842 * Always return happily.
4845 } /*VAdjustVolumeStatistics */
4848 VAdjustVolumeStatistics(register Volume * vp)
4852 retVal = VAdjustVolumeStatistics_r(vp);
4858 VBumpVolumeUsage_r(register Volume * vp)
4860 unsigned int now = FT_ApproxTime();
4861 if (now - V_dayUseDate(vp) > OneDay)
4862 VAdjustVolumeStatistics_r(vp);
4864 * Save the volume header image to disk after every 128 bumps to dayUse.
4866 if ((V_dayUse(vp)++ & 127) == 0) {
4868 VUpdateVolume_r(&error, vp, VOL_UPDATE_WAIT);
4873 VBumpVolumeUsage(register Volume * vp)
4876 VBumpVolumeUsage_r(vp);
4881 VSetDiskUsage_r(void)
4883 #ifndef AFS_DEMAND_ATTACH_FS
4884 static int FifteenMinuteCounter = 0;
4888 /* NOTE: Don't attempt to access the partitions list until the
4889 * initialization level indicates that all volumes are attached,
4890 * which implies that all partitions are initialized. */
4891 #ifdef AFS_PTHREAD_ENV
4893 #else /* AFS_PTHREAD_ENV */
4895 #endif /* AFS_PTHREAD_ENV */
4898 VResetDiskUsage_r();
4900 #ifndef AFS_DEMAND_ATTACH_FS
4901 if (++FifteenMinuteCounter == 3) {
4902 FifteenMinuteCounter = 0;
4905 #endif /* !AFS_DEMAND_ATTACH_FS */
4917 /***************************************************/
4918 /* Volume Update List routines */
4919 /***************************************************/
4921 /* The number of minutes that a volume hasn't been updated before the
4922 * "Dont salvage" flag in the volume header will be turned on */
4923 #define SALVAGE_INTERVAL (10*60)
4928 * volume update list functionality has been moved into the VLRU
4929 * the DONT_SALVAGE flag is now set during VLRU demotion
4932 #ifndef AFS_DEMAND_ATTACH_FS
4933 static VolumeId *UpdateList = NULL; /* Pointer to array of Volume ID's */
4934 static int nUpdatedVolumes = 0; /* Updated with entry in UpdateList, salvage after crash flag on */
4935 static int updateSize = 0; /* number of entries possible */
4936 #define UPDATE_LIST_SIZE 128 /* initial size increment (must be a power of 2!) */
4937 #endif /* !AFS_DEMAND_ATTACH_FS */
4940 VAddToVolumeUpdateList_r(Error * ec, Volume * vp)
4943 vp->updateTime = FT_ApproxTime();
4944 if (V_dontSalvage(vp) == 0)
4946 V_dontSalvage(vp) = 0;
4947 VSyncVolume_r(ec, vp, 0);
4948 #ifdef AFS_DEMAND_ATTACH_FS
4949 V_attachFlags(vp) &= ~(VOL_HDR_DONTSALV);
4950 #else /* !AFS_DEMAND_ATTACH_FS */
4953 if (UpdateList == NULL) {
4954 updateSize = UPDATE_LIST_SIZE;
4955 UpdateList = (VolumeId *) malloc(sizeof(VolumeId) * updateSize);
4957 if (nUpdatedVolumes == updateSize) {
4959 if (updateSize > 524288) {
4960 Log("warning: there is likely a bug in the volume update scanner\n");
4964 (VolumeId *) realloc(UpdateList,
4965 sizeof(VolumeId) * updateSize);
4968 assert(UpdateList != NULL);
4969 UpdateList[nUpdatedVolumes++] = V_id(vp);
4970 #endif /* !AFS_DEMAND_ATTACH_FS */
4973 #ifndef AFS_DEMAND_ATTACH_FS
4975 VScanUpdateList(void)
4977 register int i, gap;
4978 register Volume *vp;
4980 afs_uint32 now = FT_ApproxTime();
4981 /* Be careful with this code, since it works with interleaved calls to AddToVolumeUpdateList */
4982 for (i = gap = 0; i < nUpdatedVolumes; i++) {
4984 UpdateList[i - gap] = UpdateList[i];
4986 /* XXX this routine needlessly messes up the Volume LRU by
4987 * breaking the LRU temporal-locality assumptions.....
4988 * we should use a special volume header allocator here */
4989 vp = VGetVolume_r(&error, UpdateList[i - gap] = UpdateList[i]);
4992 } else if (vp->nUsers == 1 && now - vp->updateTime > SALVAGE_INTERVAL) {
4993 V_dontSalvage(vp) = DONT_SALVAGE;
4994 VUpdateVolume_r(&error, vp, 0); /* No need to fsync--not critical */
5002 #ifndef AFS_PTHREAD_ENV
5004 #endif /* !AFS_PTHREAD_ENV */
5006 nUpdatedVolumes -= gap;
5008 #endif /* !AFS_DEMAND_ATTACH_FS */
5011 /***************************************************/
5012 /* Volume LRU routines */
5013 /***************************************************/
5018 * with demand attach fs, we attempt to soft detach(1)
5019 * volumes which have not been accessed in a long time
5020 * in order to speed up fileserver shutdown
5022 * (1) by soft detach we mean a process very similar
5023 * to VOffline, except the final state of the
5024 * Volume will be VOL_STATE_PREATTACHED, instead
5025 * of the usual VOL_STATE_UNATTACHED
5027 #ifdef AFS_DEMAND_ATTACH_FS
5029 /* implementation is reminiscent of a generational GC
5031 * queue 0 is newly attached volumes. this queue is
5032 * sorted by attach timestamp
5034 * queue 1 is volumes that have been around a bit
5035 * longer than queue 0. this queue is sorted by
5038 * queue 2 is volumes tha have been around the longest.
5039 * this queue is unsorted
5041 * queue 3 is volumes that have been marked as
5042 * candidates for soft detachment. this queue is
5045 #define VLRU_GENERATIONS 3 /**< number of generations in VLRU */
5046 #define VLRU_QUEUES 5 /**< total number of VLRU queues */
5049 * definition of a VLRU queue.
5052 volatile struct rx_queue q;
5059 * main VLRU data structure.
5062 struct VLRU_q q[VLRU_QUEUES]; /**< VLRU queues */
5065 /** time interval (in seconds) between promotion passes for
5066 * each young generation queue. */
5067 afs_uint32 promotion_interval[VLRU_GENERATIONS-1];
5069 /** time interval (in seconds) between soft detach candidate
5070 * scans for each generation queue.
5072 * scan_interval[VLRU_QUEUE_CANDIDATE] defines how frequently
5073 * we perform a soft detach pass. */
5074 afs_uint32 scan_interval[VLRU_GENERATIONS+1];
5076 /* scheduler state */
5077 int next_idx; /**< next queue to receive attention */
5078 afs_uint32 last_promotion[VLRU_GENERATIONS-1]; /**< timestamp of last promotion scan */
5079 afs_uint32 last_scan[VLRU_GENERATIONS+1]; /**< timestamp of last detach scan */
5081 int scanner_state; /**< state of scanner thread */
5082 pthread_cond_t cv; /**< state transition CV */
5085 /** global VLRU state */
5086 static struct VLRU volume_LRU;
5089 * defined states for VLRU scanner thread.
5092 VLRU_SCANNER_STATE_OFFLINE = 0, /**< vlru scanner thread is offline */
5093 VLRU_SCANNER_STATE_ONLINE = 1, /**< vlru scanner thread is online */
5094 VLRU_SCANNER_STATE_SHUTTING_DOWN = 2, /**< vlru scanner thread is shutting down */
5095 VLRU_SCANNER_STATE_PAUSING = 3, /**< vlru scanner thread is getting ready to pause */
5096 VLRU_SCANNER_STATE_PAUSED = 4 /**< vlru scanner thread is paused */
5097 } vlru_thread_state_t;
5099 /* vlru disk data header stuff */
5100 #define VLRU_DISK_MAGIC 0x7a8b9cad /**< vlru disk entry magic number */
5101 #define VLRU_DISK_VERSION 1 /**< vlru disk entry version number */
5103 /** vlru default expiration time (for eventual fs state serialization of vlru data) */
5104 #define VLRU_DUMP_EXPIRATION_TIME (60*60*24*7) /* expire vlru data after 1 week */
5107 /** minimum volume inactivity (in seconds) before a volume becomes eligible for
5108 * soft detachment. */
5109 static afs_uint32 VLRU_offline_thresh = VLRU_DEFAULT_OFFLINE_THRESH;
5111 /** time interval (in seconds) between VLRU scanner thread soft detach passes. */
5112 static afs_uint32 VLRU_offline_interval = VLRU_DEFAULT_OFFLINE_INTERVAL;
5114 /** maximum number of volumes to soft detach in a VLRU soft detach pass. */
5115 static afs_uint32 VLRU_offline_max = VLRU_DEFAULT_OFFLINE_MAX;
5117 /** VLRU control flag. non-zero value implies VLRU subsystem is activated. */
5118 static afs_uint32 VLRU_enabled = 1;
5120 /* queue synchronization routines */
5121 static void VLRU_BeginExclusive_r(struct VLRU_q * q);
5122 static void VLRU_EndExclusive_r(struct VLRU_q * q);
5123 static void VLRU_Wait_r(struct VLRU_q * q);
5126 * set VLRU subsystem tunable parameters.
5128 * @param[in] option tunable option to modify
5129 * @param[in] val new value for tunable parameter
5131 * @pre @c VInitVolumePackage has not yet been called.
5133 * @post tunable parameter is modified
5137 * @note valid option parameters are:
5138 * @arg @c VLRU_SET_THRESH
5139 * set the period of inactivity after which
5140 * volumes are eligible for soft detachment
5141 * @arg @c VLRU_SET_INTERVAL
5142 * set the time interval between calls
5143 * to the volume LRU "garbage collector"
5144 * @arg @c VLRU_SET_MAX
5145 * set the max number of volumes to deallocate
5149 VLRU_SetOptions(int option, afs_uint32 val)
5151 if (option == VLRU_SET_THRESH) {
5152 VLRU_offline_thresh = val;
5153 } else if (option == VLRU_SET_INTERVAL) {
5154 VLRU_offline_interval = val;
5155 } else if (option == VLRU_SET_MAX) {
5156 VLRU_offline_max = val;
5157 } else if (option == VLRU_SET_ENABLED) {
5160 VLRU_ComputeConstants();
5164 * compute VLRU internal timing parameters.
5166 * @post VLRU scanner thread internal timing parameters are computed
5168 * @note computes internal timing parameters based upon user-modifiable
5169 * tunable parameters.
5173 * @internal volume package internal use only.
5176 VLRU_ComputeConstants(void)
5178 afs_uint32 factor = VLRU_offline_thresh / VLRU_offline_interval;
5180 /* compute the candidate scan interval */
5181 volume_LRU.scan_interval[VLRU_QUEUE_CANDIDATE] = VLRU_offline_interval;
5183 /* compute the promotion intervals */
5184 volume_LRU.promotion_interval[VLRU_QUEUE_NEW] = VLRU_offline_thresh * 2;
5185 volume_LRU.promotion_interval[VLRU_QUEUE_MID] = VLRU_offline_thresh * 4;
5188 /* compute the gen 0 scan interval */
5189 volume_LRU.scan_interval[VLRU_QUEUE_NEW] = VLRU_offline_thresh / 8;
5191 /* compute the gen 0 scan interval */
5192 volume_LRU.scan_interval[VLRU_QUEUE_NEW] = VLRU_offline_interval * 2;
5197 * initialize VLRU subsystem.
5199 * @pre this function has not yet been called
5201 * @post VLRU subsystem is initialized and VLRU scanner thread is starting
5205 * @internal volume package internal use only.
5211 pthread_attr_t attrs;
5214 if (!VLRU_enabled) {
5215 Log("VLRU: disabled\n");
5219 /* initialize each of the VLRU queues */
5220 for (i = 0; i < VLRU_QUEUES; i++) {
5221 queue_Init(&volume_LRU.q[i]);
5222 volume_LRU.q[i].len = 0;
5223 volume_LRU.q[i].busy = 0;
5224 assert(pthread_cond_init(&volume_LRU.q[i].cv, NULL) == 0);
5227 /* setup the timing constants */
5228 VLRU_ComputeConstants();
5230 /* XXX put inside LogLevel check? */
5231 Log("VLRU: starting scanner with the following configuration parameters:\n");
5232 Log("VLRU: offlining volumes after minimum of %d seconds of inactivity\n", VLRU_offline_thresh);
5233 Log("VLRU: running VLRU soft detach pass every %d seconds\n", VLRU_offline_interval);
5234 Log("VLRU: taking up to %d volumes offline per pass\n", VLRU_offline_max);
5235 Log("VLRU: scanning generation 0 for inactive volumes every %d seconds\n", volume_LRU.scan_interval[0]);
5236 Log("VLRU: scanning for promotion/demotion between generations 0 and 1 every %d seconds\n", volume_LRU.promotion_interval[0]);
5237 Log("VLRU: scanning for promotion/demotion between generations 1 and 2 every %d seconds\n", volume_LRU.promotion_interval[1]);
5239 /* start up the VLRU scanner */
5240 volume_LRU.scanner_state = VLRU_SCANNER_STATE_OFFLINE;
5241 if (programType == fileServer) {
5242 assert(pthread_cond_init(&volume_LRU.cv, NULL) == 0);
5243 assert(pthread_attr_init(&attrs) == 0);
5244 assert(pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED) == 0);
5245 assert(pthread_create(&tid, &attrs, &VLRU_ScannerThread, NULL) == 0);
5250 * initialize the VLRU-related fields of a newly allocated volume object.
5252 * @param[in] vp pointer to volume object
5255 * @arg @c VOL_LOCK is held.
5256 * @arg volume object is not on a VLRU queue.
5258 * @post VLRU fields are initialized to indicate that volume object is not
5259 * currently registered with the VLRU subsystem
5263 * @internal volume package interal use only.
5266 VLRU_Init_Node_r(volatile Volume * vp)
5271 assert(queue_IsNotOnQueue(&vp->vlru));
5272 vp->vlru.idx = VLRU_QUEUE_INVALID;
5276 * add a volume object to a VLRU queue.
5278 * @param[in] vp pointer to volume object
5281 * @arg @c VOL_LOCK is held.
5282 * @arg caller MUST hold a lightweight ref on @p vp.
5283 * @arg caller MUST NOT hold exclusive ownership of the VLRU queue.
5285 * @post the volume object is added to the appropriate VLRU queue
5287 * @note if @c vp->vlru.idx contains the index of a valid VLRU queue,
5288 * then the volume is added to that queue. Otherwise, the value
5289 * @c VLRU_QUEUE_NEW is stored into @c vp->vlru.idx and the
5290 * volume is added to the NEW generation queue.
5292 * @note @c VOL_LOCK may be dropped internally
5294 * @note Volume state is temporarily set to @c VOL_STATE_VLRU_ADD
5295 * during the add operation, and is restored to the previous
5296 * state prior to return.
5300 * @internal volume package internal use only.
5303 VLRU_Add_r(volatile Volume * vp)
5306 VolState state_save;
5311 if (queue_IsOnQueue(&vp->vlru))
5314 state_save = VChangeState_r(vp, VOL_STATE_VLRU_ADD);
5317 if ((idx < 0) || (idx >= VLRU_QUEUE_INVALID)) {
5318 idx = VLRU_QUEUE_NEW;
5321 VLRU_Wait_r(&volume_LRU.q[idx]);
5323 /* repeat check since VLRU_Wait_r may have dropped
5325 if (queue_IsNotOnQueue(&vp->vlru)) {
5327 queue_Prepend(&volume_LRU.q[idx], &vp->vlru);
5328 volume_LRU.q[idx].len++;
5329 V_attachFlags(vp) |= VOL_ON_VLRU;
5330 vp->stats.last_promote = FT_ApproxTime();
5333 VChangeState_r(vp, state_save);
5337 * delete a volume object from a VLRU queue.
5339 * @param[in] vp pointer to volume object
5342 * @arg @c VOL_LOCK is held.
5343 * @arg caller MUST hold a lightweight ref on @p vp.
5344 * @arg caller MUST NOT hold exclusive ownership of the VLRU queue.
5346 * @post volume object is removed from the VLRU queue
5348 * @note @c VOL_LOCK may be dropped internally
5352 * @todo We should probably set volume state to something exlcusive
5353 * (as @c VLRU_Add_r does) prior to dropping @c VOL_LOCK.
5355 * @internal volume package internal use only.
5358 VLRU_Delete_r(volatile Volume * vp)
5365 if (queue_IsNotOnQueue(&vp->vlru))
5371 if (idx == VLRU_QUEUE_INVALID)
5373 VLRU_Wait_r(&volume_LRU.q[idx]);
5374 } while (idx != vp->vlru.idx);
5376 /* now remove from the VLRU and update
5377 * the appropriate counter */
5378 queue_Remove(&vp->vlru);
5379 volume_LRU.q[idx].len--;
5380 vp->vlru.idx = VLRU_QUEUE_INVALID;
5381 V_attachFlags(vp) &= ~(VOL_ON_VLRU);
5385 * tell the VLRU subsystem that a volume was just accessed.
5387 * @param[in] vp pointer to volume object
5390 * @arg @c VOL_LOCK is held
5391 * @arg caller MUST hold a lightweight ref on @p vp
5392 * @arg caller MUST NOT hold exclusive ownership of any VLRU queue
5394 * @post volume VLRU access statistics are updated. If the volume was on
5395 * the VLRU soft detach candidate queue, it is moved to the NEW
5398 * @note @c VOL_LOCK may be dropped internally
5402 * @internal volume package internal use only.
5405 VLRU_UpdateAccess_r(volatile Volume * vp)
5407 afs_uint32 live_interval;
5408 Volume * rvp = NULL;
5413 if (queue_IsNotOnQueue(&vp->vlru))
5416 assert(V_attachFlags(vp) & VOL_ON_VLRU);
5418 /* update the access timestamp */
5419 vp->stats.last_get = FT_ApproxTime();
5422 * if the volume is on the soft detach candidate
5423 * list, we need to safely move it back to a
5424 * regular generation. this has to be done
5425 * carefully so we don't race against the scanner
5429 /* if this volume is on the soft detach candidate queue,
5430 * then grab exclusive access to the necessary queues */
5431 if (vp->vlru.idx == VLRU_QUEUE_CANDIDATE) {
5433 VCreateReservation_r(rvp);
5435 VLRU_Wait_r(&volume_LRU.q[VLRU_QUEUE_NEW]);
5436 VLRU_BeginExclusive_r(&volume_LRU.q[VLRU_QUEUE_NEW]);
5437 VLRU_Wait_r(&volume_LRU.q[VLRU_QUEUE_CANDIDATE]);
5438 VLRU_BeginExclusive_r(&volume_LRU.q[VLRU_QUEUE_CANDIDATE]);
5441 /* make sure multiple threads don't race to update */
5442 if (vp->vlru.idx == VLRU_QUEUE_CANDIDATE) {
5443 VLRU_SwitchQueues(vp, VLRU_QUEUE_NEW, 1);
5447 VLRU_EndExclusive_r(&volume_LRU.q[VLRU_QUEUE_CANDIDATE]);
5448 VLRU_EndExclusive_r(&volume_LRU.q[VLRU_QUEUE_NEW]);
5449 VCancelReservation_r(rvp);
5454 * switch a volume between two VLRU queues.
5456 * @param[in] vp pointer to volume object
5457 * @param[in] new_idx index of VLRU queue onto which the volume will be moved
5458 * @param[in] append controls whether the volume will be appended or
5459 * prepended to the queue. A nonzero value means it will
5460 * be appended; zero means it will be prepended.
5462 * @pre The new (and old, if applicable) queue(s) must either be owned
5463 * exclusively by the calling thread for asynchronous manipulation,
5464 * or the queue(s) must be quiescent and VOL_LOCK must be held.
5465 * Please see VLRU_BeginExclusive_r, VLRU_EndExclusive_r and VLRU_Wait_r
5466 * for further details of the queue asynchronous processing mechanism.
5468 * @post If the volume object was already on a VLRU queue, it is
5469 * removed from the queue. Depending on the value of the append
5470 * parameter, the volume object is either appended or prepended
5471 * to the VLRU queue referenced by the new_idx parameter.
5475 * @see VLRU_BeginExclusive_r
5476 * @see VLRU_EndExclusive_r
5479 * @internal volume package internal use only.
5482 VLRU_SwitchQueues(volatile Volume * vp, int new_idx, int append)
5484 if (queue_IsNotOnQueue(&vp->vlru))
5487 queue_Remove(&vp->vlru);
5488 volume_LRU.q[vp->vlru.idx].len--;
5490 /* put the volume back on the correct generational queue */
5492 queue_Append(&volume_LRU.q[new_idx], &vp->vlru);
5494 queue_Prepend(&volume_LRU.q[new_idx], &vp->vlru);
5497 volume_LRU.q[new_idx].len++;
5498 vp->vlru.idx = new_idx;
5502 * VLRU background thread.
5504 * The VLRU Scanner Thread is responsible for periodically scanning through
5505 * each VLRU queue looking for volumes which should be moved to another
5506 * queue, or soft detached.
5508 * @param[in] args unused thread arguments parameter
5510 * @return unused thread return value
5511 * @retval NULL always
5513 * @internal volume package internal use only.
5516 VLRU_ScannerThread(void * args)
5518 afs_uint32 now, min_delay, delay;
5519 afs_uint32 next_scan[VLRU_GENERATIONS];
5520 afs_uint32 next_promotion[VLRU_GENERATIONS];
5521 int i, min_idx, min_op, overdue, state;
5523 /* set t=0 for promotion cycle to be
5524 * fileserver startup */
5525 now = FT_ApproxTime();
5526 for (i=0; i < VLRU_GENERATIONS-1; i++) {
5527 volume_LRU.last_promotion[i] = now;
5530 /* don't start the scanner until VLRU_offline_thresh
5531 * plus a small delay for VInitVolumePackage to finish
5534 sleep(VLRU_offline_thresh + 60);
5536 /* set t=0 for scan cycle to be now */
5537 now = FT_ApproxTime();
5538 for (i=0; i < VLRU_GENERATIONS+1; i++) {
5539 volume_LRU.last_scan[i] = now;
5543 if (volume_LRU.scanner_state == VLRU_SCANNER_STATE_OFFLINE) {
5544 volume_LRU.scanner_state = VLRU_SCANNER_STATE_ONLINE;
5547 while ((state = volume_LRU.scanner_state) != VLRU_SCANNER_STATE_SHUTTING_DOWN) {
5548 /* check to see if we've been asked to pause */
5549 if (volume_LRU.scanner_state == VLRU_SCANNER_STATE_PAUSING) {
5550 volume_LRU.scanner_state = VLRU_SCANNER_STATE_PAUSED;
5551 assert(pthread_cond_broadcast(&volume_LRU.cv) == 0);
5553 VOL_CV_WAIT(&volume_LRU.cv);
5554 } while (volume_LRU.scanner_state == VLRU_SCANNER_STATE_PAUSED);
5557 /* scheduling can happen outside the glock */
5560 /* figure out what is next on the schedule */
5562 /* figure out a potential schedule for the new generation first */
5564 min_delay = volume_LRU.scan_interval[0] + volume_LRU.last_scan[0] - now;
5567 if (min_delay > volume_LRU.scan_interval[0]) {
5568 /* unsigned overflow -- we're overdue to run this scan */
5573 /* if we're not overdue for gen 0, figure out schedule for candidate gen */
5575 i = VLRU_QUEUE_CANDIDATE;
5576 delay = volume_LRU.scan_interval[i] + volume_LRU.last_scan[i] - now;
5577 if (delay < min_delay) {
5581 if (delay > volume_LRU.scan_interval[i]) {
5582 /* unsigned overflow -- we're overdue to run this scan */
5590 /* if we're still not overdue for something, figure out schedules for promotions */
5591 for (i=0; !overdue && i < VLRU_GENERATIONS-1; i++) {
5592 delay = volume_LRU.promotion_interval[i] + volume_LRU.last_promotion[i] - now;
5593 if (delay < min_delay) {
5598 if (delay > volume_LRU.promotion_interval[i]) {
5599 /* unsigned overflow -- we're overdue to run this promotion */
5608 /* sleep as needed */
5613 /* do whatever is next */
5616 VLRU_Promote_r(min_idx);
5617 VLRU_Demote_r(min_idx+1);
5619 VLRU_Scan_r(min_idx);
5621 now = FT_ApproxTime();
5624 Log("VLRU scanner asked to go offline (scanner_state=%d)\n", state);
5626 /* signal that scanner is down */
5627 volume_LRU.scanner_state = VLRU_SCANNER_STATE_OFFLINE;
5628 assert(pthread_cond_broadcast(&volume_LRU.cv) == 0);
5634 * promote volumes from one VLRU generation to the next.
5636 * This routine scans a VLRU generation looking for volumes which are
5637 * eligible to be promoted to the next generation. All volumes which
5638 * meet the eligibility requirement are promoted.
5640 * Promotion eligibility is based upon meeting both of the following
5643 * @arg The volume has been accessed since the last promotion:
5644 * @c (vp->stats.last_get >= vp->stats.last_promote)
5645 * @arg The last promotion occurred at least
5646 * @c volume_LRU.promotion_interval[idx] seconds ago
5648 * As a performance optimization, promotions are "globbed". In other
5649 * words, we promote arbitrarily large contiguous sublists of elements
5652 * @param[in] idx VLRU queue index to scan
5656 * @internal VLRU internal use only.
5659 VLRU_Promote_r(int idx)
5661 int len, chaining, promote;
5662 afs_uint32 now, thresh;
5663 struct rx_queue *qp, *nqp;
5664 Volume * vp, *start, *end;
5666 /* get exclusive access to two chains, and drop the glock */
5667 VLRU_Wait_r(&volume_LRU.q[idx]);
5668 VLRU_BeginExclusive_r(&volume_LRU.q[idx]);
5669 VLRU_Wait_r(&volume_LRU.q[idx+1]);
5670 VLRU_BeginExclusive_r(&volume_LRU.q[idx+1]);
5673 thresh = volume_LRU.promotion_interval[idx];
5674 now = FT_ApproxTime();
5677 for (queue_ScanBackwards(&volume_LRU.q[idx], qp, nqp, rx_queue)) {
5678 vp = (Volume *)((char *)qp - offsetof(Volume, vlru));
5679 promote = (((vp->stats.last_promote + thresh) <= now) &&
5680 (vp->stats.last_get >= vp->stats.last_promote));
5688 /* promote and prepend chain */
5689 queue_MoveChainAfter(&volume_LRU.q[idx+1], &start->vlru, &end->vlru);
5703 /* promote and prepend */
5704 queue_MoveChainAfter(&volume_LRU.q[idx+1], &start->vlru, &end->vlru);
5708 volume_LRU.q[idx].len -= len;
5709 volume_LRU.q[idx+1].len += len;
5712 /* release exclusive access to the two chains */
5714 volume_LRU.last_promotion[idx] = now;
5715 VLRU_EndExclusive_r(&volume_LRU.q[idx+1]);
5716 VLRU_EndExclusive_r(&volume_LRU.q[idx]);
5719 /* run the demotions */
5721 VLRU_Demote_r(int idx)
5724 int len, chaining, demote;
5725 afs_uint32 now, thresh;
5726 struct rx_queue *qp, *nqp;
5727 Volume * vp, *start, *end;
5728 Volume ** salv_flag_vec = NULL;
5729 int salv_vec_offset = 0;
5731 assert(idx == VLRU_QUEUE_MID || idx == VLRU_QUEUE_OLD);
5733 /* get exclusive access to two chains, and drop the glock */
5734 VLRU_Wait_r(&volume_LRU.q[idx-1]);
5735 VLRU_BeginExclusive_r(&volume_LRU.q[idx-1]);
5736 VLRU_Wait_r(&volume_LRU.q[idx]);
5737 VLRU_BeginExclusive_r(&volume_LRU.q[idx]);
5740 /* no big deal if this allocation fails */
5741 if (volume_LRU.q[idx].len) {
5742 salv_flag_vec = (Volume **) malloc(volume_LRU.q[idx].len * sizeof(Volume *));
5745 now = FT_ApproxTime();
5746 thresh = volume_LRU.promotion_interval[idx-1];
5749 for (queue_ScanBackwards(&volume_LRU.q[idx], qp, nqp, rx_queue)) {
5750 vp = (Volume *)((char *)qp - offsetof(Volume, vlru));
5751 demote = (((vp->stats.last_promote + thresh) <= now) &&
5752 (vp->stats.last_get < (now - thresh)));
5754 /* we now do volume update list DONT_SALVAGE flag setting during
5755 * demotion passes */
5756 if (salv_flag_vec &&
5757 !(V_attachFlags(vp) & VOL_HDR_DONTSALV) &&
5759 (vp->updateTime < (now - SALVAGE_INTERVAL)) &&
5760 (V_attachState(vp) == VOL_STATE_ATTACHED)) {
5761 salv_flag_vec[salv_vec_offset++] = vp;
5762 VCreateReservation_r(vp);
5771 /* demote and append chain */
5772 queue_MoveChainBefore(&volume_LRU.q[idx-1], &start->vlru, &end->vlru);
5786 queue_MoveChainBefore(&volume_LRU.q[idx-1], &start->vlru, &end->vlru);
5790 volume_LRU.q[idx].len -= len;
5791 volume_LRU.q[idx-1].len += len;
5794 /* release exclusive access to the two chains */
5796 VLRU_EndExclusive_r(&volume_LRU.q[idx]);
5797 VLRU_EndExclusive_r(&volume_LRU.q[idx-1]);
5799 /* now go back and set the DONT_SALVAGE flags as appropriate */
5800 if (salv_flag_vec) {
5802 for (i = 0; i < salv_vec_offset; i++) {
5803 vp = salv_flag_vec[i];
5804 if (!(V_attachFlags(vp) & VOL_HDR_DONTSALV) &&
5805 (vp->updateTime < (now - SALVAGE_INTERVAL)) &&
5806 (V_attachState(vp) == VOL_STATE_ATTACHED)) {
5809 V_attachFlags(vp) |= VOL_HDR_DONTSALV;
5810 V_dontSalvage(vp) = DONT_SALVAGE;
5811 VUpdateVolume_r(&ec, vp, 0);
5815 VCancelReservation_r(vp);
5817 free(salv_flag_vec);
5821 /* run a pass of the VLRU GC scanner */
5823 VLRU_Scan_r(int idx)
5825 afs_uint32 now, thresh;
5826 struct rx_queue *qp, *nqp;
5827 volatile Volume * vp;
5830 assert(idx == VLRU_QUEUE_NEW || idx == VLRU_QUEUE_CANDIDATE);
5832 /* gain exclusive access to the idx VLRU */
5833 VLRU_Wait_r(&volume_LRU.q[idx]);
5834 VLRU_BeginExclusive_r(&volume_LRU.q[idx]);
5836 if (idx != VLRU_QUEUE_CANDIDATE) {
5837 /* gain exclusive access to the candidate VLRU */
5838 VLRU_Wait_r(&volume_LRU.q[VLRU_QUEUE_CANDIDATE]);
5839 VLRU_BeginExclusive_r(&volume_LRU.q[VLRU_QUEUE_CANDIDATE]);
5842 now = FT_ApproxTime();
5843 thresh = now - VLRU_offline_thresh;
5845 /* perform candidate selection and soft detaching */
5846 if (idx == VLRU_QUEUE_CANDIDATE) {
5847 /* soft detach some volumes from the candidate pool */
5851 for (i=0,queue_ScanBackwards(&volume_LRU.q[idx], qp, nqp, rx_queue)) {
5852 vp = (Volume *)((char *)qp - offsetof(Volume, vlru));
5853 if (i >= VLRU_offline_max) {
5856 /* check timestamp to see if it's a candidate for soft detaching */
5857 if (vp->stats.last_get <= thresh) {
5859 if (VCheckSoftDetach(vp, thresh))
5865 /* scan for volumes to become soft detach candidates */
5866 for (i=1,queue_ScanBackwards(&volume_LRU.q[idx], qp, nqp, rx_queue),i++) {
5867 vp = (Volume *)((char *)qp - offsetof(Volume, vlru));
5869 /* check timestamp to see if it's a candidate for soft detaching */
5870 if (vp->stats.last_get <= thresh) {
5871 VCheckSoftDetachCandidate(vp, thresh);
5874 if (!(i&0x7f)) { /* lock coarsening optimization */
5882 /* relinquish exclusive access to the VLRU chains */
5886 volume_LRU.last_scan[idx] = now;
5887 if (idx != VLRU_QUEUE_CANDIDATE) {
5888 VLRU_EndExclusive_r(&volume_LRU.q[VLRU_QUEUE_CANDIDATE]);
5890 VLRU_EndExclusive_r(&volume_LRU.q[idx]);
5893 /* check whether volume is safe to soft detach
5894 * caller MUST NOT hold a ref count on vp */
5896 VCheckSoftDetach(volatile Volume * vp, afs_uint32 thresh)
5900 if (vp->nUsers || vp->nWaiters)
5903 if (vp->stats.last_get <= thresh) {
5904 ret = VSoftDetachVolume_r(vp, thresh);
5910 /* check whether volume should be made a
5911 * soft detach candidate */
5913 VCheckSoftDetachCandidate(volatile Volume * vp, afs_uint32 thresh)
5916 if (vp->nUsers || vp->nWaiters)
5921 assert(idx == VLRU_QUEUE_NEW);
5923 if (vp->stats.last_get <= thresh) {
5924 /* move to candidate pool */
5925 queue_Remove(&vp->vlru);
5926 volume_LRU.q[VLRU_QUEUE_NEW].len--;
5927 queue_Prepend(&volume_LRU.q[VLRU_QUEUE_CANDIDATE], &vp->vlru);
5928 vp->vlru.idx = VLRU_QUEUE_CANDIDATE;
5929 volume_LRU.q[VLRU_QUEUE_CANDIDATE].len++;
5937 /* begin exclusive access on VLRU */
5939 VLRU_BeginExclusive_r(struct VLRU_q * q)
5941 assert(q->busy == 0);
5945 /* end exclusive access on VLRU */
5947 VLRU_EndExclusive_r(struct VLRU_q * q)
5951 assert(pthread_cond_broadcast(&q->cv) == 0);
5954 /* wait for another thread to end exclusive access on VLRU */
5956 VLRU_Wait_r(struct VLRU_q * q)
5959 VOL_CV_WAIT(&q->cv);
5964 * volume soft detach
5966 * caller MUST NOT hold a ref count on vp */
5968 VSoftDetachVolume_r(volatile Volume * vp, afs_uint32 thresh)
5973 assert(vp->vlru.idx == VLRU_QUEUE_CANDIDATE);
5975 ts_save = vp->stats.last_get;
5976 if (ts_save > thresh)
5979 if (vp->nUsers || vp->nWaiters)
5982 if (VIsExclusiveState(V_attachState(vp))) {
5986 switch (V_attachState(vp)) {
5987 case VOL_STATE_UNATTACHED:
5988 case VOL_STATE_PREATTACHED:
5989 case VOL_STATE_ERROR:
5990 case VOL_STATE_GOING_OFFLINE:
5991 case VOL_STATE_SHUTTING_DOWN:
5992 case VOL_STATE_SALVAGING:
5993 volume_LRU.q[vp->vlru.idx].len--;
5995 /* create and cancel a reservation to
5996 * give the volume an opportunity to
5998 VCreateReservation_r(vp);
5999 queue_Remove(&vp->vlru);
6000 vp->vlru.idx = VLRU_QUEUE_INVALID;
6001 V_attachFlags(vp) &= ~(VOL_ON_VLRU);
6002 VCancelReservation_r(vp);
6006 /* hold the volume and take it offline.
6007 * no need for reservations, as VHold_r
6008 * takes care of that internally. */
6009 if (VHold_r(vp) == 0) {
6010 /* vhold drops the glock, so now we should
6011 * check to make sure we aren't racing against
6012 * other threads. if we are racing, offlining vp
6013 * would be wasteful, and block the scanner for a while
6017 (vp->shuttingDown) ||
6018 (vp->goingOffline) ||
6019 (vp->stats.last_get != ts_save)) {
6020 /* looks like we're racing someone else. bail */
6024 /* pull it off the VLRU */
6025 assert(vp->vlru.idx == VLRU_QUEUE_CANDIDATE);
6026 volume_LRU.q[VLRU_QUEUE_CANDIDATE].len--;
6027 queue_Remove(&vp->vlru);
6028 vp->vlru.idx = VLRU_QUEUE_INVALID;
6029 V_attachFlags(vp) &= ~(VOL_ON_VLRU);
6031 /* take if offline */
6032 VOffline_r(vp, "volume has been soft detached");
6034 /* invalidate the volume header cache */
6035 FreeVolumeHeader(vp);
6038 IncUInt64(&VStats.soft_detaches);
6039 vp->stats.soft_detaches++;
6041 /* put in pre-attached state so demand
6042 * attacher can work on it */
6043 VChangeState_r(vp, VOL_STATE_PREATTACHED);
6049 #endif /* AFS_DEMAND_ATTACH_FS */
6052 /***************************************************/
6053 /* Volume Header Cache routines */
6054 /***************************************************/
6057 * volume header cache.
6059 struct volume_hdr_LRU_t volume_hdr_LRU;
6062 * initialize the volume header cache.
6064 * @param[in] howMany number of header cache entries to preallocate
6066 * @pre VOL_LOCK held. Function has never been called before.
6068 * @post howMany cache entries are allocated, initialized, and added
6069 * to the LRU list. Header cache statistics are initialized.
6071 * @note only applicable to fileServer program type. Should only be
6072 * called once during volume package initialization.
6074 * @internal volume package internal use only.
6077 VInitVolumeHeaderCache(afs_uint32 howMany)
6079 register struct volHeader *hp;
6080 if (programType != fileServer)
6082 queue_Init(&volume_hdr_LRU);
6083 volume_hdr_LRU.stats.free = 0;
6084 volume_hdr_LRU.stats.used = howMany;
6085 volume_hdr_LRU.stats.attached = 0;
6086 hp = (struct volHeader *)(calloc(howMany, sizeof(struct volHeader)));
6088 ReleaseVolumeHeader(hp++);
6092 * get a volume header and attach it to the volume object.
6094 * @param[in] vp pointer to volume object
6096 * @return cache entry status
6097 * @retval 0 volume header was newly attached; cache data is invalid
6098 * @retval 1 volume header was previously attached; cache data is valid
6100 * @pre VOL_LOCK held. For DAFS, lightweight ref must be held on volume object.
6102 * @post volume header attached to volume object. if necessary, header cache
6103 * entry on LRU is synchronized to disk. Header is removed from LRU list.
6105 * @note VOL_LOCK may be dropped
6107 * @warning this interface does not load header data from disk. it merely
6108 * attaches a header object to the volume object, and may sync the old
6109 * header cache data out to disk in the process.
6111 * @internal volume package internal use only.
6114 GetVolumeHeader(register Volume * vp)
6117 register struct volHeader *hd;
6119 static int everLogged = 0;
6121 #ifdef AFS_DEMAND_ATTACH_FS
6122 VolState vp_save, back_save;
6124 /* XXX debug 9/19/05 we've apparently got
6125 * a ref counting bug somewhere that's
6126 * breaking the nUsers == 0 => header on LRU
6128 if (vp->header && queue_IsNotOnQueue(vp->header)) {
6129 Log("nUsers == 0, but header not on LRU\n");
6134 old = (vp->header != NULL); /* old == volume already has a header */
6136 if (programType != fileServer) {
6137 /* for volume utilities, we allocate volHeaders as needed */
6139 hd = (struct volHeader *)calloc(1, sizeof(*vp->header));
6143 #ifdef AFS_DEMAND_ATTACH_FS
6144 V_attachFlags(vp) |= VOL_HDR_ATTACHED;
6148 /* for the fileserver, we keep a volume header cache */
6150 /* the header we previously dropped in the lru is
6151 * still available. pull it off the lru and return */
6154 assert(hd->back == vp);
6156 /* we need to grab a new element off the LRU */
6157 if (queue_IsNotEmpty(&volume_hdr_LRU)) {
6158 /* grab an element and pull off of LRU */
6159 hd = queue_First(&volume_hdr_LRU, volHeader);
6162 /* LRU is empty, so allocate a new volHeader
6163 * this is probably indicative of a leak, so let the user know */
6164 hd = (struct volHeader *)calloc(1, sizeof(struct volHeader));
6167 Log("****Allocated more volume headers, probably leak****\n");
6170 volume_hdr_LRU.stats.free++;
6173 /* this header used to belong to someone else.
6174 * we'll need to check if the header needs to
6175 * be sync'd out to disk */
6177 #ifdef AFS_DEMAND_ATTACH_FS
6178 /* if hd->back were in an exclusive state, then
6179 * its volHeader would not be on the LRU... */
6180 assert(!VIsExclusiveState(V_attachState(hd->back)));
6183 if (hd->diskstuff.inUse) {
6184 /* volume was in use, so we'll need to sync
6185 * its header to disk */
6187 #ifdef AFS_DEMAND_ATTACH_FS
6188 back_save = VChangeState_r(hd->back, VOL_STATE_UPDATING);
6189 vp_save = VChangeState_r(vp, VOL_STATE_HDR_ATTACHING);
6190 VCreateReservation_r(hd->back);
6194 WriteVolumeHeader_r(&error, hd->back);
6195 /* Ignore errors; catch them later */
6197 #ifdef AFS_DEMAND_ATTACH_FS
6202 hd->back->header = NULL;
6203 #ifdef AFS_DEMAND_ATTACH_FS
6204 V_attachFlags(hd->back) &= ~(VOL_HDR_ATTACHED | VOL_HDR_LOADED | VOL_HDR_IN_LRU);
6206 if (hd->diskstuff.inUse) {
6207 VChangeState_r(hd->back, back_save);
6208 VCancelReservation_r(hd->back);
6209 VChangeState_r(vp, vp_save);
6213 volume_hdr_LRU.stats.attached++;
6217 #ifdef AFS_DEMAND_ATTACH_FS
6218 V_attachFlags(vp) |= VOL_HDR_ATTACHED;
6221 volume_hdr_LRU.stats.free--;
6222 volume_hdr_LRU.stats.used++;
6224 IncUInt64(&VStats.hdr_gets);
6225 #ifdef AFS_DEMAND_ATTACH_FS
6226 IncUInt64(&vp->stats.hdr_gets);
6227 vp->stats.last_hdr_get = FT_ApproxTime();
6234 * make sure volume header is attached and contains valid cache data.
6236 * @param[out] ec outbound error code
6237 * @param[in] vp pointer to volume object
6239 * @pre VOL_LOCK held. For DAFS, lightweight ref held on vp.
6241 * @post header cache entry attached, and loaded with valid data, or
6242 * *ec is nonzero, and the header is released back into the LRU.
6244 * @internal volume package internal use only.
6247 LoadVolumeHeader(Error * ec, Volume * vp)
6249 #ifdef AFS_DEMAND_ATTACH_FS
6250 VolState state_save;
6254 if (vp->nUsers == 0 && !GetVolumeHeader(vp)) {
6255 IncUInt64(&VStats.hdr_loads);
6256 state_save = VChangeState_r(vp, VOL_STATE_HDR_LOADING);
6259 ReadHeader(ec, V_diskDataHandle(vp), (char *)&V_disk(vp),
6260 sizeof(V_disk(vp)), VOLUMEINFOMAGIC,
6262 IncUInt64(&vp->stats.hdr_loads);
6263 now = FT_ApproxTime();
6267 V_attachFlags(vp) |= VOL_HDR_LOADED;
6268 vp->stats.last_hdr_load = now;
6270 VChangeState_r(vp, state_save);
6272 #else /* AFS_DEMAND_ATTACH_FS */
6274 if (vp->nUsers == 0 && !GetVolumeHeader(vp)) {
6275 IncUInt64(&VStats.hdr_loads);
6277 ReadHeader(ec, V_diskDataHandle(vp), (char *)&V_disk(vp),
6278 sizeof(V_disk(vp)), VOLUMEINFOMAGIC,
6281 #endif /* AFS_DEMAND_ATTACH_FS */
6283 /* maintain (nUsers==0) => header in LRU invariant */
6284 ReleaseVolumeHeader(vp->header);
6289 * release a header cache entry back into the LRU list.
6291 * @param[in] hd pointer to volume header cache object
6293 * @pre VOL_LOCK held.
6295 * @post header cache object appended onto end of LRU list.
6297 * @note only applicable to fileServer program type.
6299 * @note used to place a header cache entry back into the
6300 * LRU pool without invalidating it as a cache entry.
6302 * @internal volume package internal use only.
6305 ReleaseVolumeHeader(register struct volHeader *hd)
6307 if (programType != fileServer)
6309 if (!hd || queue_IsOnQueue(hd)) /* no header, or header already released */
6311 queue_Append(&volume_hdr_LRU, hd);
6312 #ifdef AFS_DEMAND_ATTACH_FS
6314 V_attachFlags(hd->back) |= VOL_HDR_IN_LRU;
6317 volume_hdr_LRU.stats.free++;
6318 volume_hdr_LRU.stats.used--;
6322 * free/invalidate a volume header cache entry.
6324 * @param[in] vp pointer to volume object
6326 * @pre VOL_LOCK is held.
6328 * @post For fileserver, header cache entry is returned to LRU, and it is
6329 * invalidated as a cache entry. For volume utilities, the header
6330 * cache entry is freed.
6332 * @note For fileserver, this should be utilized instead of ReleaseVolumeHeader
6333 * whenever it is necessary to invalidate the header cache entry.
6335 * @see ReleaseVolumeHeader
6337 * @internal volume package internal use only.
6340 FreeVolumeHeader(register Volume * vp)
6342 register struct volHeader *hd = vp->header;
6345 if (programType == fileServer) {
6346 ReleaseVolumeHeader(hd);
6351 #ifdef AFS_DEMAND_ATTACH_FS
6352 V_attachFlags(vp) &= ~(VOL_HDR_ATTACHED | VOL_HDR_IN_LRU | VOL_HDR_LOADED);
6354 volume_hdr_LRU.stats.attached--;
6359 /***************************************************/
6360 /* Volume Hash Table routines */
6361 /***************************************************/
6364 * set size of volume object hash table.
6366 * @param[in] logsize log(2) of desired hash table size
6368 * @return operation status
6370 * @retval -1 failure
6372 * @pre MUST be called prior to VInitVolumePackage
6374 * @post Volume Hash Table will have 2^logsize buckets
6377 VSetVolHashSize(int logsize)
6379 /* 64 to 16384 hash buckets seems like a reasonable range */
6380 if ((logsize < 6 ) || (logsize > 14)) {
6385 VolumeHashTable.Size = 1 << logsize;
6386 VolumeHashTable.Mask = VolumeHashTable.Size - 1;
6388 /* we can't yet support runtime modification of this
6389 * parameter. we'll need a configuration rwlock to
6390 * make runtime modification feasible.... */
6397 * initialize dynamic data structures for volume hash table.
6399 * @post hash table is allocated, and fields are initialized.
6401 * @internal volume package internal use only.
6404 VInitVolumeHash(void)
6408 VolumeHashTable.Table = (VolumeHashChainHead *) calloc(VolumeHashTable.Size,
6409 sizeof(VolumeHashChainHead));
6410 assert(VolumeHashTable.Table != NULL);
6412 for (i=0; i < VolumeHashTable.Size; i++) {
6413 queue_Init(&VolumeHashTable.Table[i]);
6414 #ifdef AFS_DEMAND_ATTACH_FS
6415 assert(pthread_cond_init(&VolumeHashTable.Table[i].chain_busy_cv, NULL) == 0);
6416 #endif /* AFS_DEMAND_ATTACH_FS */
6421 * add a volume object to the hash table.
6423 * @param[in] vp pointer to volume object
6424 * @param[in] hashid hash of volume id
6426 * @pre VOL_LOCK is held. For DAFS, caller must hold a lightweight
6429 * @post volume is added to hash chain.
6431 * @internal volume package internal use only.
6433 * @note For DAFS, VOL_LOCK may be dropped in order to wait for an
6434 * asynchronous hash chain reordering to finish.
6437 AddVolumeToHashTable(register Volume * vp, int hashid)
6439 VolumeHashChainHead * head;
6441 if (queue_IsOnQueue(vp))
6444 head = &VolumeHashTable.Table[VOLUME_HASH(hashid)];
6446 #ifdef AFS_DEMAND_ATTACH_FS
6447 /* wait for the hash chain to become available */
6450 V_attachFlags(vp) |= VOL_IN_HASH;
6451 vp->chainCacheCheck = ++head->cacheCheck;
6452 #endif /* AFS_DEMAND_ATTACH_FS */
6455 vp->hashid = hashid;
6456 queue_Append(head, vp);
6457 vp->vnodeHashOffset = VolumeHashOffset_r();
6461 * delete a volume object from the hash table.
6463 * @param[in] vp pointer to volume object
6465 * @pre VOL_LOCK is held. For DAFS, caller must hold a lightweight
6468 * @post volume is removed from hash chain.
6470 * @internal volume package internal use only.
6472 * @note For DAFS, VOL_LOCK may be dropped in order to wait for an
6473 * asynchronous hash chain reordering to finish.
6476 DeleteVolumeFromHashTable(register Volume * vp)
6478 VolumeHashChainHead * head;
6480 if (!queue_IsOnQueue(vp))
6483 head = &VolumeHashTable.Table[VOLUME_HASH(vp->hashid)];
6485 #ifdef AFS_DEMAND_ATTACH_FS
6486 /* wait for the hash chain to become available */
6489 V_attachFlags(vp) &= ~(VOL_IN_HASH);
6491 #endif /* AFS_DEMAND_ATTACH_FS */
6495 /* do NOT reset hashid to zero, as the online
6496 * salvager package may need to know the volume id
6497 * after the volume is removed from the hash */
6501 * lookup a volume object in the hash table given a volume id.
6503 * @param[out] ec error code return
6504 * @param[in] volumeId volume id
6505 * @param[in] hint volume object which we believe could be the correct
6508 * @return volume object pointer
6509 * @retval NULL no such volume id is registered with the hash table.
6511 * @pre VOL_LOCK is held. For DAFS, caller must hold a lightweight
6514 * @post volume object with the given id is returned. volume object and
6515 * hash chain access statistics are updated. hash chain may have
6518 * @note For DAFS, VOL_LOCK may be dropped in order to wait for an
6519 * asynchronous hash chain reordering operation to finish, or
6520 * in order for us to perform an asynchronous chain reordering.
6522 * @note Hash chain reorderings occur when the access count for the
6523 * volume object being looked up exceeds the sum of the previous
6524 * node's (the node ahead of it in the hash chain linked list)
6525 * access count plus the constant VOLUME_HASH_REORDER_THRESHOLD.
6527 * @note For DAFS, the hint parameter allows us to short-circuit if the
6528 * cacheCheck fields match between the hash chain head and the
6529 * hint volume object.
6532 VLookupVolume_r(Error * ec, VolId volumeId, Volume * hint)
6534 register int looks = 0;
6535 Volume * vp, *np, *pp;
6536 VolumeHashChainHead * head;
6539 head = &VolumeHashTable.Table[VOLUME_HASH(volumeId)];
6541 #ifdef AFS_DEMAND_ATTACH_FS
6542 /* wait for the hash chain to become available */
6545 /* check to see if we can short circuit without walking the hash chain */
6546 if (hint && (hint->chainCacheCheck == head->cacheCheck)) {
6547 IncUInt64(&hint->stats.hash_short_circuits);
6550 #endif /* AFS_DEMAND_ATTACH_FS */
6552 /* someday we need to either do per-chain locks, RWlocks,
6553 * or both for volhash access.
6554 * (and move to a data structure with better cache locality) */
6556 /* search the chain for this volume id */
6557 for(queue_Scan(head, vp, np, Volume)) {
6559 if ((vp->hashid == volumeId)) {
6564 if (queue_IsEnd(head, vp)) {
6568 #ifdef AFS_DEMAND_ATTACH_FS
6569 /* update hash chain statistics */
6572 FillInt64(lks, 0, looks);
6573 AddUInt64(head->looks, lks, &head->looks);
6574 AddUInt64(VStats.hash_looks, lks, &VStats.hash_looks);
6575 IncUInt64(&head->gets);
6580 IncUInt64(&vp->stats.hash_lookups);
6582 /* for demand attach fileserver, we permit occasional hash chain reordering
6583 * so that frequently looked up volumes move towards the head of the chain */
6584 pp = queue_Prev(vp, Volume);
6585 if (!queue_IsEnd(head, pp)) {
6586 FillInt64(thresh, 0, VOLUME_HASH_REORDER_THRESHOLD);
6587 AddUInt64(thresh, pp->stats.hash_lookups, &thresh);
6588 if (GEInt64(vp->stats.hash_lookups, thresh)) {
6589 VReorderHash_r(head, pp, vp);
6593 /* update the short-circuit cache check */
6594 vp->chainCacheCheck = head->cacheCheck;
6596 #endif /* AFS_DEMAND_ATTACH_FS */
6601 #ifdef AFS_DEMAND_ATTACH_FS
6602 /* perform volume hash chain reordering.
6604 * advance a subchain beginning at vp ahead of
6605 * the adjacent subchain ending at pp */
6607 VReorderHash_r(VolumeHashChainHead * head, Volume * pp, Volume * vp)
6609 Volume *tp, *np, *lp;
6610 afs_uint64 move_thresh;
6612 /* this should never be called if the chain is already busy, so
6613 * no need to wait for other exclusive chain ops to finish */
6615 /* this is a rather heavy set of operations,
6616 * so let's set the chain busy flag and drop
6618 VHashBeginExclusive_r(head);
6621 /* scan forward in the chain from vp looking for the last element
6622 * in the chain we want to advance */
6623 FillInt64(move_thresh, 0, VOLUME_HASH_REORDER_CHAIN_THRESH);
6624 AddUInt64(move_thresh, pp->stats.hash_lookups, &move_thresh);
6625 for(queue_ScanFrom(head, vp, tp, np, Volume)) {
6626 if (LTInt64(tp->stats.hash_lookups, move_thresh)) {
6630 lp = queue_Prev(tp, Volume);
6632 /* scan backwards from pp to determine where to splice and
6633 * insert the subchain we're advancing */
6634 for(queue_ScanBackwardsFrom(head, pp, tp, np, Volume)) {
6635 if (GTInt64(tp->stats.hash_lookups, move_thresh)) {
6639 tp = queue_Next(tp, Volume);
6641 /* rebalance chain(vp,...,lp) ahead of chain(tp,...,pp) */
6642 queue_MoveChainBefore(tp,vp,lp);
6645 IncUInt64(&VStats.hash_reorders);
6647 IncUInt64(&head->reorders);
6649 /* wake up any threads waiting for the hash chain */
6650 VHashEndExclusive_r(head);
6654 /* demand-attach fs volume hash
6655 * asynchronous exclusive operations */
6658 * begin an asynchronous exclusive operation on a volume hash chain.
6660 * @param[in] head pointer to volume hash chain head object
6662 * @pre VOL_LOCK held. hash chain is quiescent.
6664 * @post hash chain marked busy.
6666 * @note this interface is used in conjunction with VHashEndExclusive_r and
6667 * VHashWait_r to perform asynchronous (wrt VOL_LOCK) operations on a
6668 * volume hash chain. Its main use case is hash chain reordering, which
6669 * has the potential to be a highly latent operation.
6671 * @see VHashEndExclusive_r
6676 * @internal volume package internal use only.
6679 VHashBeginExclusive_r(VolumeHashChainHead * head)
6681 assert(head->busy == 0);
6686 * relinquish exclusive ownership of a volume hash chain.
6688 * @param[in] head pointer to volume hash chain head object
6690 * @pre VOL_LOCK held. thread owns the hash chain exclusively.
6692 * @post hash chain is marked quiescent. threads awaiting use of
6693 * chain are awakened.
6695 * @see VHashBeginExclusive_r
6700 * @internal volume package internal use only.
6703 VHashEndExclusive_r(VolumeHashChainHead * head)
6707 assert(pthread_cond_broadcast(&head->chain_busy_cv) == 0);
6711 * wait for all asynchronous operations on a hash chain to complete.
6713 * @param[in] head pointer to volume hash chain head object
6715 * @pre VOL_LOCK held.
6717 * @post hash chain object is quiescent.
6719 * @see VHashBeginExclusive_r
6720 * @see VHashEndExclusive_r
6724 * @note This interface should be called before any attempt to
6725 * traverse the hash chain. It is permissible for a thread
6726 * to gain exclusive access to the chain, and then perform
6727 * latent operations on the chain asynchronously wrt the
6730 * @warning if waiting is necessary, VOL_LOCK is dropped
6732 * @internal volume package internal use only.
6735 VHashWait_r(VolumeHashChainHead * head)
6737 while (head->busy) {
6738 VOL_CV_WAIT(&head->chain_busy_cv);
6741 #endif /* AFS_DEMAND_ATTACH_FS */
6744 /***************************************************/
6745 /* Volume by Partition List routines */
6746 /***************************************************/
6749 * demand attach fileserver adds a
6750 * linked list of volumes to each
6751 * partition object, thus allowing
6752 * for quick enumeration of all
6753 * volumes on a partition
6756 #ifdef AFS_DEMAND_ATTACH_FS
6758 * add a volume to its disk partition VByPList.
6760 * @param[in] vp pointer to volume object
6762 * @pre either the disk partition VByPList is owned exclusively
6763 * by the calling thread, or the list is quiescent and
6766 * @post volume is added to disk partition VByPList
6770 * @warning it is the caller's responsibility to ensure list
6773 * @see VVByPListWait_r
6774 * @see VVByPListBeginExclusive_r
6775 * @see VVByPListEndExclusive_r
6777 * @internal volume package internal use only.
6780 AddVolumeToVByPList_r(Volume * vp)
6782 if (queue_IsNotOnQueue(&vp->vol_list)) {
6783 queue_Append(&vp->partition->vol_list, &vp->vol_list);
6784 V_attachFlags(vp) |= VOL_ON_VBYP_LIST;
6785 vp->partition->vol_list.len++;
6790 * delete a volume from its disk partition VByPList.
6792 * @param[in] vp pointer to volume object
6794 * @pre either the disk partition VByPList is owned exclusively
6795 * by the calling thread, or the list is quiescent and
6798 * @post volume is removed from the disk partition VByPList
6802 * @warning it is the caller's responsibility to ensure list
6805 * @see VVByPListWait_r
6806 * @see VVByPListBeginExclusive_r
6807 * @see VVByPListEndExclusive_r
6809 * @internal volume package internal use only.
6812 DeleteVolumeFromVByPList_r(Volume * vp)
6814 if (queue_IsOnQueue(&vp->vol_list)) {
6815 queue_Remove(&vp->vol_list);
6816 V_attachFlags(vp) &= ~(VOL_ON_VBYP_LIST);
6817 vp->partition->vol_list.len--;
6822 * begin an asynchronous exclusive operation on a VByPList.
6824 * @param[in] dp pointer to disk partition object
6826 * @pre VOL_LOCK held. VByPList is quiescent.
6828 * @post VByPList marked busy.
6830 * @note this interface is used in conjunction with VVByPListEndExclusive_r and
6831 * VVByPListWait_r to perform asynchronous (wrt VOL_LOCK) operations on a
6834 * @see VVByPListEndExclusive_r
6835 * @see VVByPListWait_r
6839 * @internal volume package internal use only.
6841 /* take exclusive control over the list */
6843 VVByPListBeginExclusive_r(struct DiskPartition64 * dp)
6845 assert(dp->vol_list.busy == 0);
6846 dp->vol_list.busy = 1;
6850 * relinquish exclusive ownership of a VByPList.
6852 * @param[in] dp pointer to disk partition object
6854 * @pre VOL_LOCK held. thread owns the VByPList exclusively.
6856 * @post VByPList is marked quiescent. threads awaiting use of
6857 * the list are awakened.
6859 * @see VVByPListBeginExclusive_r
6860 * @see VVByPListWait_r
6864 * @internal volume package internal use only.
6867 VVByPListEndExclusive_r(struct DiskPartition64 * dp)
6869 assert(dp->vol_list.busy);
6870 dp->vol_list.busy = 0;
6871 assert(pthread_cond_broadcast(&dp->vol_list.cv) == 0);
6875 * wait for all asynchronous operations on a VByPList to complete.
6877 * @param[in] dp pointer to disk partition object
6879 * @pre VOL_LOCK is held.
6881 * @post disk partition's VByP list is quiescent
6885 * @note This interface should be called before any attempt to
6886 * traverse the VByPList. It is permissible for a thread
6887 * to gain exclusive access to the list, and then perform
6888 * latent operations on the list asynchronously wrt the
6891 * @warning if waiting is necessary, VOL_LOCK is dropped
6893 * @see VVByPListEndExclusive_r
6894 * @see VVByPListBeginExclusive_r
6896 * @internal volume package internal use only.
6899 VVByPListWait_r(struct DiskPartition64 * dp)
6901 while (dp->vol_list.busy) {
6902 VOL_CV_WAIT(&dp->vol_list.cv);
6905 #endif /* AFS_DEMAND_ATTACH_FS */
6907 /***************************************************/
6908 /* Volume Cache Statistics routines */
6909 /***************************************************/
6912 VPrintCacheStats_r(void)
6914 afs_uint32 get_hi, get_lo, load_hi, load_lo;
6915 register struct VnodeClassInfo *vcp;
6916 vcp = &VnodeClassInfo[vLarge];
6917 Log("Large vnode cache, %d entries, %d allocs, %d gets (%d reads), %d writes\n", vcp->cacheSize, vcp->allocs, vcp->gets, vcp->reads, vcp->writes);
6918 vcp = &VnodeClassInfo[vSmall];
6919 Log("Small vnode cache,%d entries, %d allocs, %d gets (%d reads), %d writes\n", vcp->cacheSize, vcp->allocs, vcp->gets, vcp->reads, vcp->writes);
6920 SplitInt64(VStats.hdr_gets, get_hi, get_lo);
6921 SplitInt64(VStats.hdr_loads, load_hi, load_lo);
6922 Log("Volume header cache, %d entries, %d gets, %d replacements\n",
6923 VStats.hdr_cache_size, get_lo, load_lo);
6927 VPrintCacheStats(void)
6930 VPrintCacheStats_r();
6934 #ifdef AFS_DEMAND_ATTACH_FS
6936 UInt64ToDouble(afs_uint64 * x)
6938 static double c32 = 4.0 * 1.073741824 * 1000000000.0;
6940 SplitInt64(*x, h, l);
6941 return (((double)h) * c32) + ((double) l);
6945 DoubleToPrintable(double x, char * buf, int len)
6947 static double billion = 1000000000.0;
6950 y[0] = (afs_uint32) (x / (billion * billion));
6951 y[1] = (afs_uint32) ((x - (((double)y[0]) * billion * billion)) / billion);
6952 y[2] = (afs_uint32) (x - ((((double)y[0]) * billion * billion) + (((double)y[1]) * billion)));
6955 snprintf(buf, len, "%d%09d%09d", y[0], y[1], y[2]);
6957 snprintf(buf, len, "%d%09d", y[1], y[2]);
6959 snprintf(buf, len, "%d", y[2]);
6966 VPrintExtendedCacheStats_r(int flags)
6975 struct stats looks, gets, reorders, len;
6976 struct stats ch_looks, ch_gets, ch_reorders;
6978 VolumeHashChainHead *head;
6981 /* zero out stats */
6982 memset(&looks, 0, sizeof(struct stats));
6983 memset(&gets, 0, sizeof(struct stats));
6984 memset(&reorders, 0, sizeof(struct stats));
6985 memset(&len, 0, sizeof(struct stats));
6986 memset(&ch_looks, 0, sizeof(struct stats));
6987 memset(&ch_gets, 0, sizeof(struct stats));
6988 memset(&ch_reorders, 0, sizeof(struct stats));
6990 for (i = 0; i < VolumeHashTable.Size; i++) {
6991 head = &VolumeHashTable.Table[i];
6994 VHashBeginExclusive_r(head);
6997 ch_looks.sum = UInt64ToDouble(&head->looks);
6998 ch_gets.sum = UInt64ToDouble(&head->gets);
6999 ch_reorders.sum = UInt64ToDouble(&head->reorders);
7001 /* update global statistics */
7003 looks.sum += ch_looks.sum;
7004 gets.sum += ch_gets.sum;
7005 reorders.sum += ch_reorders.sum;
7006 len.sum += (double)head->len;
7009 len.min = (double) head->len;
7010 len.max = (double) head->len;
7011 looks.min = ch_looks.sum;
7012 looks.max = ch_looks.sum;
7013 gets.min = ch_gets.sum;
7014 gets.max = ch_gets.sum;
7015 reorders.min = ch_reorders.sum;
7016 reorders.max = ch_reorders.sum;
7018 if (((double)head->len) < len.min)
7019 len.min = (double) head->len;
7020 if (((double)head->len) > len.max)
7021 len.max = (double) head->len;
7022 if (ch_looks.sum < looks.min)
7023 looks.min = ch_looks.sum;
7024 else if (ch_looks.sum > looks.max)
7025 looks.max = ch_looks.sum;
7026 if (ch_gets.sum < gets.min)
7027 gets.min = ch_gets.sum;
7028 else if (ch_gets.sum > gets.max)
7029 gets.max = ch_gets.sum;
7030 if (ch_reorders.sum < reorders.min)
7031 reorders.min = ch_reorders.sum;
7032 else if (ch_reorders.sum > reorders.max)
7033 reorders.max = ch_reorders.sum;
7037 if ((flags & VOL_STATS_PER_CHAIN2) && queue_IsNotEmpty(head)) {
7038 /* compute detailed per-chain stats */
7039 struct stats hdr_loads, hdr_gets;
7040 double v_looks, v_loads, v_gets;
7042 /* initialize stats with data from first element in chain */
7043 vp = queue_First(head, Volume);
7044 v_looks = UInt64ToDouble(&vp->stats.hash_lookups);
7045 v_loads = UInt64ToDouble(&vp->stats.hdr_loads);
7046 v_gets = UInt64ToDouble(&vp->stats.hdr_gets);
7047 ch_gets.min = ch_gets.max = v_looks;
7048 hdr_loads.min = hdr_loads.max = v_loads;
7049 hdr_gets.min = hdr_gets.max = v_gets;
7050 hdr_loads.sum = hdr_gets.sum = 0;
7052 vp = queue_Next(vp, Volume);
7054 /* pull in stats from remaining elements in chain */
7055 for (queue_ScanFrom(head, vp, vp, np, Volume)) {
7056 v_looks = UInt64ToDouble(&vp->stats.hash_lookups);
7057 v_loads = UInt64ToDouble(&vp->stats.hdr_loads);
7058 v_gets = UInt64ToDouble(&vp->stats.hdr_gets);
7060 hdr_loads.sum += v_loads;
7061 hdr_gets.sum += v_gets;
7063 if (v_looks < ch_gets.min)
7064 ch_gets.min = v_looks;
7065 else if (v_looks > ch_gets.max)
7066 ch_gets.max = v_looks;
7068 if (v_loads < hdr_loads.min)
7069 hdr_loads.min = v_loads;
7070 else if (v_loads > hdr_loads.max)
7071 hdr_loads.max = v_loads;
7073 if (v_gets < hdr_gets.min)
7074 hdr_gets.min = v_gets;
7075 else if (v_gets > hdr_gets.max)
7076 hdr_gets.max = v_gets;
7079 /* compute per-chain averages */
7080 ch_gets.avg = ch_gets.sum / ((double)head->len);
7081 hdr_loads.avg = hdr_loads.sum / ((double)head->len);
7082 hdr_gets.avg = hdr_gets.sum / ((double)head->len);
7084 /* dump per-chain stats */
7085 Log("Volume hash chain %d : len=%d, looks=%s, reorders=%s\n",
7087 DoubleToPrintable(ch_looks.sum, pr_buf[0], sizeof(pr_buf[0])),
7088 DoubleToPrintable(ch_reorders.sum, pr_buf[1], sizeof(pr_buf[1])));
7089 Log("\tVolume gets : min=%s, max=%s, avg=%s, total=%s\n",
7090 DoubleToPrintable(ch_gets.min, pr_buf[0], sizeof(pr_buf[0])),
7091 DoubleToPrintable(ch_gets.max, pr_buf[1], sizeof(pr_buf[1])),
7092 DoubleToPrintable(ch_gets.avg, pr_buf[2], sizeof(pr_buf[2])),
7093 DoubleToPrintable(ch_gets.sum, pr_buf[3], sizeof(pr_buf[3])));
7094 Log("\tHDR gets : min=%s, max=%s, avg=%s, total=%s\n",
7095 DoubleToPrintable(hdr_gets.min, pr_buf[0], sizeof(pr_buf[0])),
7096 DoubleToPrintable(hdr_gets.max, pr_buf[1], sizeof(pr_buf[1])),
7097 DoubleToPrintable(hdr_gets.avg, pr_buf[2], sizeof(pr_buf[2])),
7098 DoubleToPrintable(hdr_gets.sum, pr_buf[3], sizeof(pr_buf[3])));
7099 Log("\tHDR loads : min=%s, max=%s, avg=%s, total=%s\n",
7100 DoubleToPrintable(hdr_loads.min, pr_buf[0], sizeof(pr_buf[0])),
7101 DoubleToPrintable(hdr_loads.max, pr_buf[1], sizeof(pr_buf[1])),
7102 DoubleToPrintable(hdr_loads.avg, pr_buf[2], sizeof(pr_buf[2])),
7103 DoubleToPrintable(hdr_loads.sum, pr_buf[3], sizeof(pr_buf[3])));
7104 } else if (flags & VOL_STATS_PER_CHAIN) {
7105 /* dump simple per-chain stats */
7106 Log("Volume hash chain %d : len=%d, looks=%s, gets=%s, reorders=%s\n",
7108 DoubleToPrintable(ch_looks.sum, pr_buf[0], sizeof(pr_buf[0])),
7109 DoubleToPrintable(ch_gets.sum, pr_buf[1], sizeof(pr_buf[1])),
7110 DoubleToPrintable(ch_reorders.sum, pr_buf[2], sizeof(pr_buf[2])));
7114 VHashEndExclusive_r(head);
7119 /* compute global averages */
7120 len.avg = len.sum / ((double)VolumeHashTable.Size);
7121 looks.avg = looks.sum / ((double)VolumeHashTable.Size);
7122 gets.avg = gets.sum / ((double)VolumeHashTable.Size);
7123 reorders.avg = reorders.sum / ((double)VolumeHashTable.Size);
7125 /* dump global stats */
7126 Log("Volume hash summary: %d buckets\n", VolumeHashTable.Size);
7127 Log(" chain length : min=%s, max=%s, avg=%s, total=%s\n",
7128 DoubleToPrintable(len.min, pr_buf[0], sizeof(pr_buf[0])),
7129 DoubleToPrintable(len.max, pr_buf[1], sizeof(pr_buf[1])),
7130 DoubleToPrintable(len.avg, pr_buf[2], sizeof(pr_buf[2])),
7131 DoubleToPrintable(len.sum, pr_buf[3], sizeof(pr_buf[3])));
7132 Log(" looks : min=%s, max=%s, avg=%s, total=%s\n",
7133 DoubleToPrintable(looks.min, pr_buf[0], sizeof(pr_buf[0])),
7134 DoubleToPrintable(looks.max, pr_buf[1], sizeof(pr_buf[1])),
7135 DoubleToPrintable(looks.avg, pr_buf[2], sizeof(pr_buf[2])),
7136 DoubleToPrintable(looks.sum, pr_buf[3], sizeof(pr_buf[3])));
7137 Log(" gets : min=%s, max=%s, avg=%s, total=%s\n",
7138 DoubleToPrintable(gets.min, pr_buf[0], sizeof(pr_buf[0])),
7139 DoubleToPrintable(gets.max, pr_buf[1], sizeof(pr_buf[1])),
7140 DoubleToPrintable(gets.avg, pr_buf[2], sizeof(pr_buf[2])),
7141 DoubleToPrintable(gets.sum, pr_buf[3], sizeof(pr_buf[3])));
7142 Log(" reorders : min=%s, max=%s, avg=%s, total=%s\n",
7143 DoubleToPrintable(reorders.min, pr_buf[0], sizeof(pr_buf[0])),
7144 DoubleToPrintable(reorders.max, pr_buf[1], sizeof(pr_buf[1])),
7145 DoubleToPrintable(reorders.avg, pr_buf[2], sizeof(pr_buf[2])),
7146 DoubleToPrintable(reorders.sum, pr_buf[3], sizeof(pr_buf[3])));
7148 /* print extended disk related statistics */
7150 struct DiskPartition64 * diskP;
7151 afs_uint32 vol_count[VOLMAXPARTS+1];
7152 byte part_exists[VOLMAXPARTS+1];
7156 memset(vol_count, 0, sizeof(vol_count));
7157 memset(part_exists, 0, sizeof(part_exists));
7161 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
7163 vol_count[id] = diskP->vol_list.len;
7164 part_exists[id] = 1;
7168 for (i = 0; i <= VOLMAXPARTS; i++) {
7169 if (part_exists[i]) {
7170 diskP = VGetPartitionById_r(i, 0);
7172 Log("Partition %s has %d online volumes\n",
7173 VPartitionPath(diskP), diskP->vol_list.len);
7183 VPrintExtendedCacheStats(int flags)
7186 VPrintExtendedCacheStats_r(flags);
7189 #endif /* AFS_DEMAND_ATTACH_FS */