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 VReleaseVolumeHandles_r(Volume * vp);
192 static void VCloseVolumeHandles_r(Volume * vp);
193 static void LoadVolumeHeader(Error * ec, Volume * vp);
194 static int VCheckOffline(register Volume * vp);
195 static int VCheckDetach(register Volume * vp);
196 static Volume * GetVolume(Error * ec, Error * client_ec, VolId volumeId, Volume * hint, int flags);
197 static int VolumeExternalName_r(VolumeId volumeId, char * name, size_t len);
199 int LogLevel; /* Vice loglevel--not defined as extern so that it will be
200 * defined when not linked with vice, XXXX */
201 ProgramType programType; /* The type of program using the package */
203 /* extended volume package statistics */
206 #ifdef VOL_LOCK_DEBUG
207 pthread_t vol_glock_holder = 0;
211 #define VOLUME_BITMAP_GROWSIZE 16 /* bytes, => 128vnodes */
212 /* Must be a multiple of 4 (1 word) !! */
214 /* this parameter needs to be tunable at runtime.
215 * 128 was really inadequate for largish servers -- at 16384 volumes this
216 * puts average chain length at 128, thus an average 65 deref's to find a volptr.
217 * talk about bad spatial locality...
219 * an AVL or splay tree might work a lot better, but we'll just increase
220 * the default hash table size for now
222 #define DEFAULT_VOLUME_HASH_SIZE 256 /* Must be a power of 2!! */
223 #define DEFAULT_VOLUME_HASH_MASK (DEFAULT_VOLUME_HASH_SIZE-1)
224 #define VOLUME_HASH(volumeId) (volumeId&(VolumeHashTable.Mask))
227 * turn volume hash chains into partially ordered lists.
228 * when the threshold is exceeded between two adjacent elements,
229 * perform a chain rebalancing operation.
231 * keep the threshold high in order to keep cache line invalidates
232 * low "enough" on SMPs
234 #define VOLUME_HASH_REORDER_THRESHOLD 200
237 * when possible, don't just reorder single elements, but reorder
238 * entire chains of elements at once. a chain of elements that
239 * exceed the element previous to the pivot by at least CHAIN_THRESH
240 * accesses are moved in front of the chain whose elements have at
241 * least CHAIN_THRESH less accesses than the pivot element
243 #define VOLUME_HASH_REORDER_CHAIN_THRESH (VOLUME_HASH_REORDER_THRESHOLD / 2)
245 #include "rx/rx_queue.h"
248 VolumeHashTable_t VolumeHashTable = {
249 DEFAULT_VOLUME_HASH_SIZE,
250 DEFAULT_VOLUME_HASH_MASK,
255 static void VInitVolumeHash(void);
259 /* This macro is used where an ffs() call does not exist. Was in util/ffs.c */
263 afs_int32 ffs_tmp = x;
267 for (ffs_i = 1;; ffs_i++) {
274 #endif /* !AFS_HAVE_FFS */
276 #ifdef AFS_PTHREAD_ENV
277 typedef struct diskpartition_queue_t {
278 struct rx_queue queue;
279 struct DiskPartition64 * diskP;
280 } diskpartition_queue_t;
281 typedef struct vinitvolumepackage_thread_t {
282 struct rx_queue queue;
283 pthread_cond_t thread_done_cv;
284 int n_threads_complete;
285 } vinitvolumepackage_thread_t;
286 static void * VInitVolumePackageThread(void * args);
287 #endif /* AFS_PTHREAD_ENV */
289 static int VAttachVolumesByPartition(struct DiskPartition64 *diskP,
290 int * nAttached, int * nUnattached);
293 #ifdef AFS_DEMAND_ATTACH_FS
294 /* demand attach fileserver extensions */
297 * in the future we will support serialization of VLRU state into the fs_state
300 * these structures are the beginning of that effort
302 struct VLRU_DiskHeader {
303 struct versionStamp stamp; /* magic and structure version number */
304 afs_uint32 mtime; /* time of dump to disk */
305 afs_uint32 num_records; /* number of VLRU_DiskEntry records */
308 struct VLRU_DiskEntry {
309 afs_uint32 vid; /* volume ID */
310 afs_uint32 idx; /* generation */
311 afs_uint32 last_get; /* timestamp of last get */
314 struct VLRU_StartupQueue {
315 struct VLRU_DiskEntry * entry;
320 typedef struct vshutdown_thread_t {
322 pthread_mutex_t lock;
324 pthread_cond_t master_cv;
326 int n_threads_complete;
328 int schedule_version;
331 byte n_parts_done_pass;
332 byte part_thread_target[VOLMAXPARTS+1];
333 byte part_done_pass[VOLMAXPARTS+1];
334 struct rx_queue * part_pass_head[VOLMAXPARTS+1];
335 int stats[4][VOLMAXPARTS+1];
336 } vshutdown_thread_t;
337 static void * VShutdownThread(void * args);
340 static Volume * VAttachVolumeByVp_r(Error * ec, Volume * vp, int mode);
341 static int VCheckFree(Volume * vp);
344 static void AddVolumeToVByPList_r(Volume * vp);
345 static void DeleteVolumeFromVByPList_r(Volume * vp);
346 static void VVByPListBeginExclusive_r(struct DiskPartition64 * dp);
347 static void VVByPListEndExclusive_r(struct DiskPartition64 * dp);
348 static void VVByPListWait_r(struct DiskPartition64 * dp);
350 /* online salvager */
351 static int VCheckSalvage(register Volume * vp);
352 static int VUpdateSalvagePriority_r(Volume * vp);
353 static int VScheduleSalvage_r(Volume * vp);
354 static int VCancelSalvage_r(Volume * vp, int reason);
356 /* Volume hash table */
357 static void VReorderHash_r(VolumeHashChainHead * head, Volume * pp, Volume * vp);
358 static void VHashBeginExclusive_r(VolumeHashChainHead * head);
359 static void VHashEndExclusive_r(VolumeHashChainHead * head);
360 static void VHashWait_r(VolumeHashChainHead * head);
363 static int ShutdownVByPForPass_r(struct DiskPartition64 * dp, int pass);
364 static int ShutdownVolumeWalk_r(struct DiskPartition64 * dp, int pass,
365 struct rx_queue ** idx);
366 static void ShutdownController(vshutdown_thread_t * params);
367 static void ShutdownCreateSchedule(vshutdown_thread_t * params);
370 static void VLRU_ComputeConstants(void);
371 static void VInitVLRU(void);
372 static void VLRU_Init_Node_r(volatile Volume * vp);
373 static void VLRU_Add_r(volatile Volume * vp);
374 static void VLRU_Delete_r(volatile Volume * vp);
375 static void VLRU_UpdateAccess_r(volatile Volume * vp);
376 static void * VLRU_ScannerThread(void * args);
377 static void VLRU_Scan_r(int idx);
378 static void VLRU_Promote_r(int idx);
379 static void VLRU_Demote_r(int idx);
380 static void VLRU_SwitchQueues(volatile Volume * vp, int new_idx, int append);
383 static int VCheckSoftDetach(volatile Volume * vp, afs_uint32 thresh);
384 static int VCheckSoftDetachCandidate(volatile Volume * vp, afs_uint32 thresh);
385 static int VSoftDetachVolume_r(volatile Volume * vp, afs_uint32 thresh);
388 pthread_key_t VThread_key;
389 VThreadOptions_t VThread_defaults = {
390 0 /**< allow salvsync */
392 #endif /* AFS_DEMAND_ATTACH_FS */
395 struct Lock vol_listLock; /* Lock obtained when listing volumes:
396 * prevents a volume from being missed
397 * if the volume is attached during a
401 static int TimeZoneCorrection; /* Number of seconds west of GMT */
403 /* Common message used when the volume goes off line */
404 char *VSalvageMessage =
405 "Files in this volume are currently unavailable; call operations";
407 int VInit; /* 0 - uninitialized,
408 * 1 - initialized but not all volumes have been attached,
409 * 2 - initialized and all volumes have been attached,
410 * 3 - initialized, all volumes have been attached, and
411 * VConnectFS() has completed. */
414 bit32 VolumeCacheCheck; /* Incremented everytime a volume goes on line--
415 * used to stamp volume headers and in-core
416 * vnodes. When the volume goes on-line the
417 * vnode will be invalidated
418 * access only with VOL_LOCK held */
423 /***************************************************/
424 /* Startup routines */
425 /***************************************************/
428 VInitVolumePackage(ProgramType pt, afs_uint32 nLargeVnodes, afs_uint32 nSmallVnodes,
429 int connect, afs_uint32 volcache)
431 int errors = 0; /* Number of errors while finding vice partitions. */
437 memset(&VStats, 0, sizeof(VStats));
438 VStats.hdr_cache_size = 200;
440 VInitPartitionPackage();
442 #ifdef AFS_DEMAND_ATTACH_FS
443 if (programType == fileServer) {
446 VLRU_SetOptions(VLRU_SET_ENABLED, 0);
448 assert(pthread_key_create(&VThread_key, NULL) == 0);
451 #ifdef AFS_PTHREAD_ENV
452 assert(pthread_mutex_init(&vol_glock_mutex, NULL) == 0);
453 assert(pthread_mutex_init(&vol_trans_mutex, NULL) == 0);
454 assert(pthread_cond_init(&vol_put_volume_cond, NULL) == 0);
455 assert(pthread_cond_init(&vol_sleep_cond, NULL) == 0);
456 #else /* AFS_PTHREAD_ENV */
458 #endif /* AFS_PTHREAD_ENV */
459 Lock_Init(&vol_listLock);
461 srandom(time(0)); /* For VGetVolumeInfo */
462 gettimeofday(&tv, &tz);
463 TimeZoneCorrection = tz.tz_minuteswest * 60;
465 #ifdef AFS_DEMAND_ATTACH_FS
466 assert(pthread_mutex_init(&vol_salvsync_mutex, NULL) == 0);
467 #endif /* AFS_DEMAND_ATTACH_FS */
469 /* Ok, we have done enough initialization that fileserver can
470 * start accepting calls, even though the volumes may not be
471 * available just yet.
475 #if defined(AFS_DEMAND_ATTACH_FS) && defined(SALVSYNC_BUILD_SERVER)
476 if (programType == salvageServer) {
479 #endif /* AFS_DEMAND_ATTACH_FS */
480 #ifdef FSSYNC_BUILD_SERVER
481 if (programType == fileServer) {
485 #if defined(AFS_DEMAND_ATTACH_FS) && defined(SALVSYNC_BUILD_CLIENT)
486 if (programType == fileServer) {
487 /* establish a connection to the salvager at this point */
488 assert(VConnectSALV() != 0);
490 #endif /* AFS_DEMAND_ATTACH_FS */
492 if (volcache > VStats.hdr_cache_size)
493 VStats.hdr_cache_size = volcache;
494 VInitVolumeHeaderCache(VStats.hdr_cache_size);
496 VInitVnodes(vLarge, nLargeVnodes);
497 VInitVnodes(vSmall, nSmallVnodes);
500 errors = VAttachPartitions();
504 if (programType == fileServer) {
505 struct DiskPartition64 *diskP;
506 #ifdef AFS_PTHREAD_ENV
507 struct vinitvolumepackage_thread_t params;
508 struct diskpartition_queue_t * dpq;
509 int i, threads, parts;
511 pthread_attr_t attrs;
513 assert(pthread_cond_init(¶ms.thread_done_cv,NULL) == 0);
515 params.n_threads_complete = 0;
517 /* create partition work queue */
518 for (parts=0, diskP = DiskPartitionList; diskP; diskP = diskP->next, parts++) {
519 dpq = (diskpartition_queue_t *) malloc(sizeof(struct diskpartition_queue_t));
522 queue_Append(¶ms,dpq);
525 threads = MIN(parts, vol_attach_threads);
528 /* spawn off a bunch of initialization threads */
529 assert(pthread_attr_init(&attrs) == 0);
530 assert(pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED) == 0);
532 Log("VInitVolumePackage: beginning parallel fileserver startup\n");
533 #ifdef AFS_DEMAND_ATTACH_FS
534 Log("VInitVolumePackage: using %d threads to pre-attach volumes on %d partitions\n",
536 #else /* AFS_DEMAND_ATTACH_FS */
537 Log("VInitVolumePackage: using %d threads to attach volumes on %d partitions\n",
539 #endif /* AFS_DEMAND_ATTACH_FS */
542 for (i=0; i < threads; i++) {
543 assert(pthread_create
544 (&tid, &attrs, &VInitVolumePackageThread,
548 while(params.n_threads_complete < threads) {
549 VOL_CV_WAIT(¶ms.thread_done_cv);
553 assert(pthread_attr_destroy(&attrs) == 0);
555 /* if we're only going to run one init thread, don't bother creating
557 Log("VInitVolumePackage: beginning single-threaded fileserver startup\n");
558 #ifdef AFS_DEMAND_ATTACH_FS
559 Log("VInitVolumePackage: using 1 thread to pre-attach volumes on %d partition(s)\n",
561 #else /* AFS_DEMAND_ATTACH_FS */
562 Log("VInitVolumePackage: using 1 thread to attach volumes on %d partition(s)\n",
564 #endif /* AFS_DEMAND_ATTACH_FS */
566 VInitVolumePackageThread(¶ms);
569 assert(pthread_cond_destroy(¶ms.thread_done_cv) == 0);
571 #else /* AFS_PTHREAD_ENV */
575 /* Attach all the volumes in this partition */
576 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
577 int nAttached = 0, nUnattached = 0;
578 assert(VAttachVolumesByPartition(diskP, &nAttached, &nUnattached) == 0);
580 #endif /* AFS_PTHREAD_ENV */
583 VInit = 2; /* Initialized, and all volumes have been attached */
584 #ifdef FSSYNC_BUILD_CLIENT
585 if (programType == volumeUtility && connect) {
587 Log("Unable to connect to file server; will retry at need\n");
591 #ifdef AFS_DEMAND_ATTACH_FS
592 else if (programType == salvageServer) {
594 Log("Unable to connect to file server; aborted\n");
598 #endif /* AFS_DEMAND_ATTACH_FS */
599 #endif /* FSSYNC_BUILD_CLIENT */
603 #ifdef AFS_PTHREAD_ENV
605 VInitVolumePackageThread(void * args) {
606 int errors = 0; /* Number of errors while finding vice partitions. */
610 struct DiskPartition64 *diskP;
611 struct vinitvolumepackage_thread_t * params;
612 struct diskpartition_queue_t * dpq;
614 params = (vinitvolumepackage_thread_t *) args;
618 /* Attach all the volumes in this partition */
619 while (queue_IsNotEmpty(params)) {
620 int nAttached = 0, nUnattached = 0;
622 dpq = queue_First(params,diskpartition_queue_t);
628 assert(VAttachVolumesByPartition(diskP, &nAttached, &nUnattached) == 0);
633 params->n_threads_complete++;
634 pthread_cond_signal(¶ms->thread_done_cv);
638 #endif /* AFS_PTHREAD_ENV */
641 * attach all volumes on a given disk partition
644 VAttachVolumesByPartition(struct DiskPartition64 *diskP, int * nAttached, int * nUnattached)
650 Log("Partition %s: attaching volumes\n", diskP->name);
651 dirp = opendir(VPartitionPath(diskP));
653 Log("opendir on Partition %s failed!\n", diskP->name);
657 while ((dp = readdir(dirp))) {
659 p = strrchr(dp->d_name, '.');
660 if (p != NULL && strcmp(p, VHDREXT) == 0) {
663 #ifdef AFS_DEMAND_ATTACH_FS
664 vp = VPreAttachVolumeByName(&error, diskP->name, dp->d_name);
665 #else /* AFS_DEMAND_ATTACH_FS */
666 vp = VAttachVolumeByName(&error, diskP->name, dp->d_name,
668 #endif /* AFS_DEMAND_ATTACH_FS */
669 (*(vp ? nAttached : nUnattached))++;
670 if (error == VOFFLINE)
671 Log("Volume %d stays offline (/vice/offline/%s exists)\n", VolumeNumber(dp->d_name), dp->d_name);
672 else if (LogLevel >= 5) {
673 Log("Partition %s: attached volume %d (%s)\n",
674 diskP->name, VolumeNumber(dp->d_name),
677 #if !defined(AFS_DEMAND_ATTACH_FS)
681 #endif /* AFS_DEMAND_ATTACH_FS */
685 Log("Partition %s: attached %d volumes; %d volumes not attached\n", diskP->name, *nAttached, *nUnattached);
691 /***************************************************/
692 /* Shutdown routines */
693 /***************************************************/
697 * highly multithreaded volume package shutdown
699 * with the demand attach fileserver extensions,
700 * VShutdown has been modified to be multithreaded.
701 * In order to achieve optimal use of many threads,
702 * the shutdown code involves one control thread and
703 * n shutdown worker threads. The control thread
704 * periodically examines the number of volumes available
705 * for shutdown on each partition, and produces a worker
706 * thread allocation schedule. The idea is to eliminate
707 * redundant scheduling computation on the workers by
708 * having a single master scheduler.
710 * The scheduler's objectives are:
712 * each partition with volumes remaining gets allocated
713 * at least 1 thread (assuming sufficient threads)
715 * threads are allocated proportional to the number of
716 * volumes remaining to be offlined. This ensures that
717 * the OS I/O scheduler has many requests to elevator
718 * seek on partitions that will (presumably) take the
719 * longest amount of time (from now) to finish shutdown
720 * (3) keep threads busy
721 * when there are extra threads, they are assigned to
722 * partitions using a simple round-robin algorithm
724 * In the future, we may wish to add the ability to adapt
725 * to the relative performance patterns of each disk
730 * multi-step shutdown process
732 * demand attach shutdown is a four-step process. Each
733 * shutdown "pass" shuts down increasingly more difficult
734 * volumes. The main purpose is to achieve better cache
735 * utilization during shutdown.
738 * shutdown volumes in the unattached, pre-attached
741 * shutdown attached volumes with cached volume headers
743 * shutdown all volumes in non-exclusive states
745 * shutdown all remaining volumes
752 register Volume *vp, *np;
753 register afs_int32 code;
754 #ifdef AFS_DEMAND_ATTACH_FS
755 struct DiskPartition64 * diskP;
756 struct diskpartition_queue_t * dpq;
757 vshutdown_thread_t params;
759 pthread_attr_t attrs;
761 memset(¶ms, 0, sizeof(vshutdown_thread_t));
763 for (params.n_parts=0, diskP = DiskPartitionList;
764 diskP; diskP = diskP->next, params.n_parts++);
766 Log("VShutdown: shutting down on-line volumes on %d partition%s...\n",
767 params.n_parts, params.n_parts > 1 ? "s" : "");
769 if (vol_attach_threads > 1) {
770 /* prepare for parallel shutdown */
771 params.n_threads = vol_attach_threads;
772 assert(pthread_mutex_init(¶ms.lock, NULL) == 0);
773 assert(pthread_cond_init(¶ms.cv, NULL) == 0);
774 assert(pthread_cond_init(¶ms.master_cv, NULL) == 0);
775 assert(pthread_attr_init(&attrs) == 0);
776 assert(pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED) == 0);
779 /* setup the basic partition information structures for
780 * parallel shutdown */
781 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
783 struct rx_queue * qp, * nqp;
787 VVByPListWait_r(diskP);
788 VVByPListBeginExclusive_r(diskP);
791 for (queue_Scan(&diskP->vol_list, qp, nqp, rx_queue)) {
792 vp = (Volume *)((char *)qp - offsetof(Volume, vol_list));
796 Log("VShutdown: partition %s has %d volumes with attached headers\n",
797 VPartitionPath(diskP), count);
800 /* build up the pass 0 shutdown work queue */
801 dpq = (struct diskpartition_queue_t *) malloc(sizeof(struct diskpartition_queue_t));
804 queue_Prepend(¶ms, dpq);
806 params.part_pass_head[diskP->index] = queue_First(&diskP->vol_list, rx_queue);
809 Log("VShutdown: beginning parallel fileserver shutdown\n");
810 Log("VShutdown: using %d threads to offline volumes on %d partition%s\n",
811 vol_attach_threads, params.n_parts, params.n_parts > 1 ? "s" : "" );
813 /* do pass 0 shutdown */
814 assert(pthread_mutex_lock(¶ms.lock) == 0);
815 for (i=0; i < params.n_threads; i++) {
816 assert(pthread_create
817 (&tid, &attrs, &VShutdownThread,
821 /* wait for all the pass 0 shutdowns to complete */
822 while (params.n_threads_complete < params.n_threads) {
823 assert(pthread_cond_wait(¶ms.master_cv, ¶ms.lock) == 0);
825 params.n_threads_complete = 0;
827 assert(pthread_cond_broadcast(¶ms.cv) == 0);
828 assert(pthread_mutex_unlock(¶ms.lock) == 0);
830 Log("VShutdown: pass 0 completed using the 1 thread per partition algorithm\n");
831 Log("VShutdown: starting passes 1 through 3 using finely-granular mp-fast algorithm\n");
833 /* run the parallel shutdown scheduler. it will drop the glock internally */
834 ShutdownController(¶ms);
836 /* wait for all the workers to finish pass 3 and terminate */
837 while (params.pass < 4) {
838 VOL_CV_WAIT(¶ms.cv);
841 assert(pthread_attr_destroy(&attrs) == 0);
842 assert(pthread_cond_destroy(¶ms.cv) == 0);
843 assert(pthread_cond_destroy(¶ms.master_cv) == 0);
844 assert(pthread_mutex_destroy(¶ms.lock) == 0);
846 /* drop the VByPList exclusive reservations */
847 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
848 VVByPListEndExclusive_r(diskP);
849 Log("VShutdown: %s stats : (pass[0]=%d, pass[1]=%d, pass[2]=%d, pass[3]=%d)\n",
850 VPartitionPath(diskP),
851 params.stats[0][diskP->index],
852 params.stats[1][diskP->index],
853 params.stats[2][diskP->index],
854 params.stats[3][diskP->index]);
857 Log("VShutdown: shutdown finished using %d threads\n", params.n_threads);
859 /* if we're only going to run one shutdown thread, don't bother creating
861 Log("VShutdown: beginning single-threaded fileserver shutdown\n");
863 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
864 VShutdownByPartition_r(diskP);
868 Log("VShutdown: complete.\n");
869 #else /* AFS_DEMAND_ATTACH_FS */
870 Log("VShutdown: shutting down on-line volumes...\n");
871 for (i = 0; i < VolumeHashTable.Size; i++) {
872 /* try to hold first volume in the hash table */
873 for (queue_Scan(&VolumeHashTable.Table[i],vp,np,Volume)) {
877 Log("VShutdown: Attempting to take volume %u offline.\n",
880 /* next, take the volume offline (drops reference count) */
881 VOffline_r(vp, "File server was shut down");
885 Log("VShutdown: complete.\n");
886 #endif /* AFS_DEMAND_ATTACH_FS */
897 #ifdef AFS_DEMAND_ATTACH_FS
900 * shutdown control thread
903 ShutdownController(vshutdown_thread_t * params)
906 struct DiskPartition64 * diskP;
908 vshutdown_thread_t shadow;
910 ShutdownCreateSchedule(params);
912 while ((params->pass < 4) &&
913 (params->n_threads_complete < params->n_threads)) {
914 /* recompute schedule once per second */
916 memcpy(&shadow, params, sizeof(vshutdown_thread_t));
920 Log("ShutdownController: schedule version=%d, vol_remaining=%d, pass=%d\n",
921 shadow.schedule_version, shadow.vol_remaining, shadow.pass);
922 Log("ShutdownController: n_threads_complete=%d, n_parts_done_pass=%d\n",
923 shadow.n_threads_complete, shadow.n_parts_done_pass);
924 for (diskP = DiskPartitionList; diskP; diskP=diskP->next) {
926 Log("ShutdownController: part[%d] : (len=%d, thread_target=%d, done_pass=%d, pass_head=%p)\n",
929 shadow.part_thread_target[id],
930 shadow.part_done_pass[id],
931 shadow.part_pass_head[id]);
937 ShutdownCreateSchedule(params);
941 /* create the shutdown thread work schedule.
942 * this scheduler tries to implement fairness
943 * by allocating at least 1 thread to each
944 * partition with volumes to be shutdown,
945 * and then it attempts to allocate remaining
946 * threads based upon the amount of work left
949 ShutdownCreateSchedule(vshutdown_thread_t * params)
951 struct DiskPartition64 * diskP;
952 int sum, thr_workload, thr_left;
953 int part_residue[VOLMAXPARTS+1];
956 /* compute the total number of outstanding volumes */
958 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
959 sum += diskP->vol_list.len;
962 params->schedule_version++;
963 params->vol_remaining = sum;
968 /* compute average per-thread workload */
969 thr_workload = sum / params->n_threads;
970 if (sum % params->n_threads)
973 thr_left = params->n_threads;
974 memset(&part_residue, 0, sizeof(part_residue));
976 /* for fairness, give every partition with volumes remaining
977 * at least one thread */
978 for (diskP = DiskPartitionList; diskP && thr_left; diskP = diskP->next) {
980 if (diskP->vol_list.len) {
981 params->part_thread_target[id] = 1;
984 params->part_thread_target[id] = 0;
988 if (thr_left && thr_workload) {
989 /* compute length-weighted workloads */
992 for (diskP = DiskPartitionList; diskP && thr_left; diskP = diskP->next) {
994 delta = (diskP->vol_list.len / thr_workload) -
995 params->part_thread_target[id];
999 if (delta < thr_left) {
1000 params->part_thread_target[id] += delta;
1003 params->part_thread_target[id] += thr_left;
1011 /* try to assign any leftover threads to partitions that
1012 * had volume lengths closer to needing thread_target+1 */
1013 int max_residue, max_id;
1015 /* compute the residues */
1016 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1018 part_residue[id] = diskP->vol_list.len -
1019 (params->part_thread_target[id] * thr_workload);
1022 /* now try to allocate remaining threads to partitions with the
1023 * highest residues */
1026 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1028 if (part_residue[id] > max_residue) {
1029 max_residue = part_residue[id];
1038 params->part_thread_target[max_id]++;
1040 part_residue[max_id] = 0;
1045 /* punt and give any remaining threads equally to each partition */
1047 if (thr_left >= params->n_parts) {
1048 alloc = thr_left / params->n_parts;
1049 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1051 params->part_thread_target[id] += alloc;
1056 /* finish off the last of the threads */
1057 for (diskP = DiskPartitionList; thr_left && diskP; diskP = diskP->next) {
1059 params->part_thread_target[id]++;
1065 /* worker thread for parallel shutdown */
1067 VShutdownThread(void * args)
1069 struct rx_queue *qp;
1071 vshutdown_thread_t * params;
1072 int part, code, found, pass, schedule_version_save, count;
1073 struct DiskPartition64 *diskP;
1074 struct diskpartition_queue_t * dpq;
1077 params = (vshutdown_thread_t *) args;
1079 /* acquire the shutdown pass 0 lock */
1080 assert(pthread_mutex_lock(¶ms->lock) == 0);
1082 /* if there's still pass 0 work to be done,
1083 * get a work entry, and do a pass 0 shutdown */
1084 if (queue_IsNotEmpty(params)) {
1085 dpq = queue_First(params, diskpartition_queue_t);
1087 assert(pthread_mutex_unlock(¶ms->lock) == 0);
1093 while (ShutdownVolumeWalk_r(diskP, 0, ¶ms->part_pass_head[id]))
1095 params->stats[0][diskP->index] = count;
1096 assert(pthread_mutex_lock(¶ms->lock) == 0);
1099 params->n_threads_complete++;
1100 if (params->n_threads_complete == params->n_threads) {
1101 /* notify control thread that all workers have completed pass 0 */
1102 assert(pthread_cond_signal(¶ms->master_cv) == 0);
1104 while (params->pass == 0) {
1105 assert(pthread_cond_wait(¶ms->cv, ¶ms->lock) == 0);
1109 assert(pthread_mutex_unlock(¶ms->lock) == 0);
1112 pass = params->pass;
1115 /* now escalate through the more complicated shutdowns */
1117 schedule_version_save = params->schedule_version;
1119 /* find a disk partition to work on */
1120 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1122 if (params->part_thread_target[id] && !params->part_done_pass[id]) {
1123 params->part_thread_target[id]--;
1130 /* hmm. for some reason the controller thread couldn't find anything for
1131 * us to do. let's see if there's anything we can do */
1132 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1134 if (diskP->vol_list.len && !params->part_done_pass[id]) {
1137 } else if (!params->part_done_pass[id]) {
1138 params->part_done_pass[id] = 1;
1139 params->n_parts_done_pass++;
1141 Log("VShutdown: done shutting down volumes on partition %s.\n",
1142 VPartitionPath(diskP));
1148 /* do work on this partition until either the controller
1149 * creates a new schedule, or we run out of things to do
1150 * on this partition */
1153 while (!params->part_done_pass[id] &&
1154 (schedule_version_save == params->schedule_version)) {
1155 /* ShutdownVolumeWalk_r will drop the glock internally */
1156 if (!ShutdownVolumeWalk_r(diskP, pass, ¶ms->part_pass_head[id])) {
1157 if (!params->part_done_pass[id]) {
1158 params->part_done_pass[id] = 1;
1159 params->n_parts_done_pass++;
1161 Log("VShutdown: done shutting down volumes on partition %s.\n",
1162 VPartitionPath(diskP));
1170 params->stats[pass][id] += count;
1172 /* ok, everyone is done this pass, proceed */
1175 params->n_threads_complete++;
1176 while (params->pass == pass) {
1177 if (params->n_threads_complete == params->n_threads) {
1178 /* we are the last thread to complete, so we will
1179 * reinitialize worker pool state for the next pass */
1180 params->n_threads_complete = 0;
1181 params->n_parts_done_pass = 0;
1183 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1185 params->part_done_pass[id] = 0;
1186 params->part_pass_head[id] = queue_First(&diskP->vol_list, rx_queue);
1189 /* compute a new thread schedule before releasing all the workers */
1190 ShutdownCreateSchedule(params);
1192 /* wake up all the workers */
1193 assert(pthread_cond_broadcast(¶ms->cv) == 0);
1196 Log("VShutdown: pass %d completed using %d threads on %d partitions\n",
1197 pass, params->n_threads, params->n_parts);
1200 VOL_CV_WAIT(¶ms->cv);
1203 pass = params->pass;
1217 /* shut down all volumes on a given disk partition
1219 * note that this function will not allow mp-fast
1220 * shutdown of a partition */
1222 VShutdownByPartition_r(struct DiskPartition64 * dp)
1228 /* wait for other exclusive ops to finish */
1229 VVByPListWait_r(dp);
1231 /* begin exclusive access */
1232 VVByPListBeginExclusive_r(dp);
1234 /* pick the low-hanging fruit first,
1235 * then do the complicated ones last
1236 * (has the advantage of keeping
1237 * in-use volumes up until the bitter end) */
1238 for (pass = 0, total=0; pass < 4; pass++) {
1239 pass_stats[pass] = ShutdownVByPForPass_r(dp, pass);
1240 total += pass_stats[pass];
1243 /* end exclusive access */
1244 VVByPListEndExclusive_r(dp);
1246 Log("VShutdownByPartition: shut down %d volumes on %s (pass[0]=%d, pass[1]=%d, pass[2]=%d, pass[3]=%d)\n",
1247 total, VPartitionPath(dp), pass_stats[0], pass_stats[1], pass_stats[2], pass_stats[3]);
1252 /* internal shutdown functionality
1254 * for multi-pass shutdown:
1255 * 0 to only "shutdown" {pre,un}attached and error state volumes
1256 * 1 to also shutdown attached volumes w/ volume header loaded
1257 * 2 to also shutdown attached volumes w/o volume header loaded
1258 * 3 to also shutdown exclusive state volumes
1260 * caller MUST hold exclusive access on the hash chain
1261 * because we drop vol_glock_mutex internally
1263 * this function is reentrant for passes 1--3
1264 * (e.g. multiple threads can cooperate to
1265 * shutdown a partition mp-fast)
1267 * pass 0 is not scaleable because the volume state data is
1268 * synchronized by vol_glock mutex, and the locking overhead
1269 * is too high to drop the lock long enough to do linked list
1273 ShutdownVByPForPass_r(struct DiskPartition64 * dp, int pass)
1275 struct rx_queue * q = queue_First(&dp->vol_list, rx_queue);
1278 while (ShutdownVolumeWalk_r(dp, pass, &q))
1284 /* conditionally shutdown one volume on partition dp
1285 * returns 1 if a volume was shutdown in this pass,
1288 ShutdownVolumeWalk_r(struct DiskPartition64 * dp, int pass,
1289 struct rx_queue ** idx)
1291 struct rx_queue *qp, *nqp;
1296 for (queue_ScanFrom(&dp->vol_list, qp, qp, nqp, rx_queue)) {
1297 vp = (Volume *) (((char *)qp) - offsetof(Volume, vol_list));
1301 if ((V_attachState(vp) != VOL_STATE_UNATTACHED) &&
1302 (V_attachState(vp) != VOL_STATE_ERROR) &&
1303 (V_attachState(vp) != VOL_STATE_PREATTACHED)) {
1307 if ((V_attachState(vp) == VOL_STATE_ATTACHED) &&
1308 (vp->header == NULL)) {
1312 if (VIsExclusiveState(V_attachState(vp))) {
1317 DeleteVolumeFromVByPList_r(vp);
1318 VShutdownVolume_r(vp);
1328 * shutdown a specific volume
1330 /* caller MUST NOT hold a heavyweight ref on vp */
1332 VShutdownVolume_r(Volume * vp)
1336 VCreateReservation_r(vp);
1338 if (LogLevel >= 5) {
1339 Log("VShutdownVolume_r: vid=%u, device=%d, state=%hu\n",
1340 vp->hashid, vp->partition->device, V_attachState(vp));
1343 /* wait for other blocking ops to finish */
1344 VWaitExclusiveState_r(vp);
1346 assert(VIsValidState(V_attachState(vp)));
1348 switch(V_attachState(vp)) {
1349 case VOL_STATE_SALVAGING:
1350 /* make sure salvager knows we don't want
1351 * the volume back */
1352 VCancelSalvage_r(vp, SALVSYNC_SHUTDOWN);
1353 case VOL_STATE_PREATTACHED:
1354 case VOL_STATE_ERROR:
1355 VChangeState_r(vp, VOL_STATE_UNATTACHED);
1356 case VOL_STATE_UNATTACHED:
1358 case VOL_STATE_GOING_OFFLINE:
1359 case VOL_STATE_SHUTTING_DOWN:
1360 case VOL_STATE_ATTACHED:
1364 Log("VShutdown: Attempting to take volume %u offline.\n",
1367 /* take the volume offline (drops reference count) */
1368 VOffline_r(vp, "File server was shut down");
1373 VCancelReservation_r(vp);
1377 #endif /* AFS_DEMAND_ATTACH_FS */
1380 /***************************************************/
1381 /* Header I/O routines */
1382 /***************************************************/
1384 /* open a descriptor for the inode (h),
1385 * read in an on-disk structure into buffer (to) of size (size),
1386 * verify versionstamp in structure has magic (magic) and
1387 * optionally verify version (version) if (version) is nonzero
1390 ReadHeader(Error * ec, IHandle_t * h, char *to, int size, bit32 magic,
1393 struct versionStamp *vsn;
1408 if (FDH_SEEK(fdP, 0, SEEK_SET) < 0) {
1410 FDH_REALLYCLOSE(fdP);
1413 vsn = (struct versionStamp *)to;
1414 if (FDH_READ(fdP, to, size) != size || vsn->magic != magic) {
1416 FDH_REALLYCLOSE(fdP);
1421 /* Check is conditional, in case caller wants to inspect version himself */
1422 if (version && vsn->version != version) {
1428 WriteVolumeHeader_r(Error * ec, Volume * vp)
1430 IHandle_t *h = V_diskDataHandle(vp);
1440 if (FDH_SEEK(fdP, 0, SEEK_SET) < 0) {
1442 FDH_REALLYCLOSE(fdP);
1445 if (FDH_WRITE(fdP, (char *)&V_disk(vp), sizeof(V_disk(vp)))
1446 != sizeof(V_disk(vp))) {
1448 FDH_REALLYCLOSE(fdP);
1454 /* VolumeHeaderToDisk
1455 * Allows for storing 64 bit inode numbers in on-disk volume header
1458 /* convert in-memory representation of a volume header to the
1459 * on-disk representation of a volume header */
1461 VolumeHeaderToDisk(VolumeDiskHeader_t * dh, VolumeHeader_t * h)
1464 memset((char *)dh, 0, sizeof(VolumeDiskHeader_t));
1465 dh->stamp = h->stamp;
1467 dh->parent = h->parent;
1469 #ifdef AFS_64BIT_IOPS_ENV
1470 dh->volumeInfo_lo = (afs_int32) h->volumeInfo & 0xffffffff;
1471 dh->volumeInfo_hi = (afs_int32) (h->volumeInfo >> 32) & 0xffffffff;
1472 dh->smallVnodeIndex_lo = (afs_int32) h->smallVnodeIndex & 0xffffffff;
1473 dh->smallVnodeIndex_hi =
1474 (afs_int32) (h->smallVnodeIndex >> 32) & 0xffffffff;
1475 dh->largeVnodeIndex_lo = (afs_int32) h->largeVnodeIndex & 0xffffffff;
1476 dh->largeVnodeIndex_hi =
1477 (afs_int32) (h->largeVnodeIndex >> 32) & 0xffffffff;
1478 dh->linkTable_lo = (afs_int32) h->linkTable & 0xffffffff;
1479 dh->linkTable_hi = (afs_int32) (h->linkTable >> 32) & 0xffffffff;
1481 dh->volumeInfo_lo = h->volumeInfo;
1482 dh->smallVnodeIndex_lo = h->smallVnodeIndex;
1483 dh->largeVnodeIndex_lo = h->largeVnodeIndex;
1484 dh->linkTable_lo = h->linkTable;
1488 /* DiskToVolumeHeader
1489 * Converts an on-disk representation of a volume header to
1490 * the in-memory representation of a volume header.
1492 * Makes the assumption that AFS has *always*
1493 * zero'd the volume header file so that high parts of inode
1494 * numbers are 0 in older (SGI EFS) volume header files.
1497 DiskToVolumeHeader(VolumeHeader_t * h, VolumeDiskHeader_t * dh)
1499 memset((char *)h, 0, sizeof(VolumeHeader_t));
1500 h->stamp = dh->stamp;
1502 h->parent = dh->parent;
1504 #ifdef AFS_64BIT_IOPS_ENV
1506 (Inode) dh->volumeInfo_lo | ((Inode) dh->volumeInfo_hi << 32);
1508 h->smallVnodeIndex =
1509 (Inode) dh->smallVnodeIndex_lo | ((Inode) dh->
1510 smallVnodeIndex_hi << 32);
1512 h->largeVnodeIndex =
1513 (Inode) dh->largeVnodeIndex_lo | ((Inode) dh->
1514 largeVnodeIndex_hi << 32);
1516 (Inode) dh->linkTable_lo | ((Inode) dh->linkTable_hi << 32);
1518 h->volumeInfo = dh->volumeInfo_lo;
1519 h->smallVnodeIndex = dh->smallVnodeIndex_lo;
1520 h->largeVnodeIndex = dh->largeVnodeIndex_lo;
1521 h->linkTable = dh->linkTable_lo;
1526 /***************************************************/
1527 /* Volume Attachment routines */
1528 /***************************************************/
1530 #ifdef AFS_DEMAND_ATTACH_FS
1532 * pre-attach a volume given its path.
1534 * @param[out] ec outbound error code
1535 * @param[in] partition partition path string
1536 * @param[in] name volume id string
1538 * @return volume object pointer
1540 * @note A pre-attached volume will only have its partition
1541 * and hashid fields initialized. At first call to
1542 * VGetVolume, the volume will be fully attached.
1546 VPreAttachVolumeByName(Error * ec, char *partition, char *name)
1550 vp = VPreAttachVolumeByName_r(ec, partition, name);
1556 * pre-attach a volume given its path.
1558 * @param[out] ec outbound error code
1559 * @param[in] partition path to vice partition
1560 * @param[in] name volume id string
1562 * @return volume object pointer
1564 * @pre VOL_LOCK held
1566 * @internal volume package internal use only.
1569 VPreAttachVolumeByName_r(Error * ec, char *partition, char *name)
1571 return VPreAttachVolumeById_r(ec,
1573 VolumeNumber(name));
1577 * pre-attach a volume given its path and numeric volume id.
1579 * @param[out] ec error code return
1580 * @param[in] partition path to vice partition
1581 * @param[in] volumeId numeric volume id
1583 * @return volume object pointer
1585 * @pre VOL_LOCK held
1587 * @internal volume package internal use only.
1590 VPreAttachVolumeById_r(Error * ec,
1595 struct DiskPartition64 *partp;
1599 assert(programType == fileServer);
1601 if (!(partp = VGetPartition_r(partition, 0))) {
1603 Log("VPreAttachVolumeById_r: Error getting partition (%s)\n", partition);
1607 vp = VLookupVolume_r(ec, volumeId, NULL);
1612 return VPreAttachVolumeByVp_r(ec, partp, vp, volumeId);
1616 * preattach a volume.
1618 * @param[out] ec outbound error code
1619 * @param[in] partp pointer to partition object
1620 * @param[in] vp pointer to volume object
1621 * @param[in] vid volume id
1623 * @return volume object pointer
1625 * @pre VOL_LOCK is held.
1627 * @warning Returned volume object pointer does not have to
1628 * equal the pointer passed in as argument vp. There
1629 * are potential race conditions which can result in
1630 * the pointers having different values. It is up to
1631 * the caller to make sure that references are handled
1632 * properly in this case.
1634 * @note If there is already a volume object registered with
1635 * the same volume id, its pointer MUST be passed as
1636 * argument vp. Failure to do so will result in a silent
1637 * failure to preattach.
1639 * @internal volume package internal use only.
1642 VPreAttachVolumeByVp_r(Error * ec,
1643 struct DiskPartition64 * partp,
1651 /* check to see if pre-attach already happened */
1653 (V_attachState(vp) != VOL_STATE_UNATTACHED) &&
1654 (V_attachState(vp) != VOL_STATE_PREATTACHED) &&
1655 !VIsErrorState(V_attachState(vp))) {
1657 * pre-attach is a no-op in all but the following cases:
1659 * - volume is unattached
1660 * - volume is in an error state
1661 * - volume is pre-attached
1663 Log("VPreattachVolumeByVp_r: volume %u not in quiescent state\n", vid);
1666 /* we're re-attaching a volume; clear out some old state */
1667 memset(&vp->salvage, 0, sizeof(struct VolumeOnlineSalvage));
1669 if (V_partition(vp) != partp) {
1670 /* XXX potential race */
1671 DeleteVolumeFromVByPList_r(vp);
1674 /* if we need to allocate a new Volume struct,
1675 * go ahead and drop the vol glock, otherwise
1676 * do the basic setup synchronised, as it's
1677 * probably not worth dropping the lock */
1680 /* allocate the volume structure */
1681 vp = nvp = (Volume *) malloc(sizeof(Volume));
1683 memset(vp, 0, sizeof(Volume));
1684 queue_Init(&vp->vnode_list);
1685 assert(pthread_cond_init(&V_attachCV(vp), NULL) == 0);
1688 /* link the volume with its associated vice partition */
1689 vp->device = partp->device;
1690 vp->partition = partp;
1693 vp->specialStatus = 0;
1695 /* if we dropped the lock, reacquire the lock,
1696 * check for pre-attach races, and then add
1697 * the volume to the hash table */
1700 nvp = VLookupVolume_r(ec, vid, NULL);
1705 } else if (nvp) { /* race detected */
1710 /* hack to make up for VChangeState_r() decrementing
1711 * the old state counter */
1712 VStats.state_levels[0]++;
1716 /* put pre-attached volume onto the hash table
1717 * and bring it up to the pre-attached state */
1718 AddVolumeToHashTable(vp, vp->hashid);
1719 AddVolumeToVByPList_r(vp);
1720 VLRU_Init_Node_r(vp);
1721 VChangeState_r(vp, VOL_STATE_PREATTACHED);
1724 Log("VPreAttachVolumeByVp_r: volume %u pre-attached\n", vp->hashid);
1732 #endif /* AFS_DEMAND_ATTACH_FS */
1734 /* Attach an existing volume, given its pathname, and return a
1735 pointer to the volume header information. The volume also
1736 normally goes online at this time. An offline volume
1737 must be reattached to make it go online */
1739 VAttachVolumeByName(Error * ec, char *partition, char *name, int mode)
1743 retVal = VAttachVolumeByName_r(ec, partition, name, mode);
1749 VAttachVolumeByName_r(Error * ec, char *partition, char *name, int mode)
1751 register Volume *vp = NULL, *svp = NULL;
1753 struct afs_stat status;
1754 struct VolumeDiskHeader diskHeader;
1755 struct VolumeHeader iheader;
1756 struct DiskPartition64 *partp;
1760 #ifdef AFS_DEMAND_ATTACH_FS
1761 VolumeStats stats_save;
1762 #endif /* AFS_DEMAND_ATTACH_FS */
1766 volumeId = VolumeNumber(name);
1768 if (!(partp = VGetPartition_r(partition, 0))) {
1770 Log("VAttachVolume: Error getting partition (%s)\n", partition);
1774 if (programType == volumeUtility) {
1776 VLockPartition_r(partition);
1777 } else if (programType == fileServer) {
1778 #ifdef AFS_DEMAND_ATTACH_FS
1779 /* lookup the volume in the hash table */
1780 vp = VLookupVolume_r(ec, volumeId, NULL);
1786 /* save any counters that are supposed to
1787 * be monotonically increasing over the
1788 * lifetime of the fileserver */
1789 memcpy(&stats_save, &vp->stats, sizeof(VolumeStats));
1791 memset(&stats_save, 0, sizeof(VolumeStats));
1794 /* if there's something in the hash table, and it's not
1795 * in the pre-attach state, then we may need to detach
1796 * it before proceeding */
1797 if (vp && (V_attachState(vp) != VOL_STATE_PREATTACHED)) {
1798 VCreateReservation_r(vp);
1799 VWaitExclusiveState_r(vp);
1801 /* at this point state must be one of:
1810 if (vp->specialStatus == VBUSY)
1813 /* if it's already attached, see if we can return it */
1814 if (V_attachState(vp) == VOL_STATE_ATTACHED) {
1815 VGetVolumeByVp_r(ec, vp);
1816 if (V_inUse(vp) == fileServer) {
1817 VCancelReservation_r(vp);
1821 /* otherwise, we need to detach, and attempt to re-attach */
1822 VDetachVolume_r(ec, vp);
1824 Log("VAttachVolume: Error detaching old volume instance (%s)\n", name);
1827 /* if it isn't fully attached, delete from the hash tables,
1828 and let the refcounter handle the rest */
1829 DeleteVolumeFromHashTable(vp);
1830 DeleteVolumeFromVByPList_r(vp);
1833 VCancelReservation_r(vp);
1837 /* pre-attach volume if it hasn't been done yet */
1839 (V_attachState(vp) == VOL_STATE_UNATTACHED) ||
1840 (V_attachState(vp) == VOL_STATE_ERROR)) {
1842 vp = VPreAttachVolumeByVp_r(ec, partp, vp, volumeId);
1850 /* handle pre-attach races
1852 * multiple threads can race to pre-attach a volume,
1853 * but we can't let them race beyond that
1855 * our solution is to let the first thread to bring
1856 * the volume into an exclusive state win; the other
1857 * threads just wait until it finishes bringing the
1858 * volume online, and then they do a vgetvolumebyvp
1860 if (svp && (svp != vp)) {
1861 /* wait for other exclusive ops to finish */
1862 VCreateReservation_r(vp);
1863 VWaitExclusiveState_r(vp);
1865 /* get a heavyweight ref, kill the lightweight ref, and return */
1866 VGetVolumeByVp_r(ec, vp);
1867 VCancelReservation_r(vp);
1871 /* at this point, we are chosen as the thread to do
1872 * demand attachment for this volume. all other threads
1873 * doing a getvolume on vp->hashid will block until we finish */
1875 /* make sure any old header cache entries are invalidated
1876 * before proceeding */
1877 FreeVolumeHeader(vp);
1879 VChangeState_r(vp, VOL_STATE_ATTACHING);
1881 /* restore any saved counters */
1882 memcpy(&vp->stats, &stats_save, sizeof(VolumeStats));
1883 #else /* AFS_DEMAND_ATTACH_FS */
1884 vp = VGetVolume_r(ec, volumeId);
1886 if (V_inUse(vp) == fileServer)
1888 if (vp->specialStatus == VBUSY)
1890 VDetachVolume_r(ec, vp);
1892 Log("VAttachVolume: Error detaching volume (%s)\n", name);
1896 #endif /* AFS_DEMAND_ATTACH_FS */
1900 strcpy(path, VPartitionPath(partp));
1906 if ((fd = afs_open(path, O_RDONLY)) == -1 || afs_fstat(fd, &status) == -1) {
1907 Log("VAttachVolume: Failed to open %s (errno %d)\n", path, errno);
1914 n = read(fd, &diskHeader, sizeof(diskHeader));
1916 if (n != sizeof(diskHeader)
1917 || diskHeader.stamp.magic != VOLUMEHEADERMAGIC) {
1918 Log("VAttachVolume: Error reading volume header %s\n", path);
1923 if (diskHeader.stamp.version != VOLUMEHEADERVERSION) {
1924 Log("VAttachVolume: Volume %s, version number is incorrect; volume needs salvaged\n", path);
1930 DiskToVolumeHeader(&iheader, &diskHeader);
1931 #ifdef FSSYNC_BUILD_CLIENT
1932 if (programType == volumeUtility && mode != V_SECRETLY && mode != V_PEEK) {
1934 if (FSYNC_VolOp(iheader.id, partition, FSYNC_VOL_NEEDVOLUME, mode, NULL)
1936 Log("VAttachVolume: attach of volume %u apparently denied by file server\n", iheader.id);
1937 *ec = VNOVOL; /* XXXX */
1945 vp = (Volume *) calloc(1, sizeof(Volume));
1947 vp->device = partp->device;
1948 vp->partition = partp;
1949 queue_Init(&vp->vnode_list);
1950 #ifdef AFS_DEMAND_ATTACH_FS
1951 assert(pthread_cond_init(&V_attachCV(vp), NULL) == 0);
1952 #endif /* AFS_DEMAND_ATTACH_FS */
1955 /* attach2 is entered without any locks, and returns
1956 * with vol_glock_mutex held */
1957 vp = attach2(ec, volumeId, path, &iheader, partp, vp, isbusy, mode);
1959 if (programType == volumeUtility && vp) {
1960 if ((mode == V_VOLUPD) || (VolumeWriteable(vp) && (mode == V_CLONE))) {
1961 /* mark volume header as in use so that volser crashes lead to a
1962 * salvage attempt */
1963 VUpdateVolume_r(ec, vp, 0);
1965 #ifdef AFS_DEMAND_ATTACH_FS
1966 /* for dafs, we should tell the fileserver, except for V_PEEK
1967 * where we know it is not necessary */
1968 if (mode == V_PEEK) {
1969 vp->needsPutBack = 0;
1971 vp->needsPutBack = 1;
1973 #else /* !AFS_DEMAND_ATTACH_FS */
1974 /* duplicate computation in fssync.c about whether the server
1975 * takes the volume offline or not. If the volume isn't
1976 * offline, we must not return it when we detach the volume,
1977 * or the server will abort */
1978 if (mode == V_READONLY || mode == V_PEEK
1979 || (!VolumeWriteable(vp) && (mode == V_CLONE || mode == V_DUMP)))
1980 vp->needsPutBack = 0;
1982 vp->needsPutBack = 1;
1983 #endif /* !AFS_DEMAND_ATTACH_FS */
1985 /* OK, there's a problem here, but one that I don't know how to
1986 * fix right now, and that I don't think should arise often.
1987 * Basically, we should only put back this volume to the server if
1988 * it was given to us by the server, but since we don't have a vp,
1989 * we can't run the VolumeWriteable function to find out as we do
1990 * above when computing vp->needsPutBack. So we send it back, but
1991 * there's a path in VAttachVolume on the server which may abort
1992 * if this volume doesn't have a header. Should be pretty rare
1993 * for all of that to happen, but if it does, probably the right
1994 * fix is for the server to allow the return of readonly volumes
1995 * that it doesn't think are really checked out. */
1996 #ifdef FSSYNC_BUILD_CLIENT
1997 if (programType == volumeUtility && vp == NULL &&
1998 mode != V_SECRETLY && mode != V_PEEK) {
1999 FSYNC_VolOp(iheader.id, partition, FSYNC_VOL_ON, 0, NULL);
2002 if (programType == fileServer && vp) {
2003 #ifdef AFS_DEMAND_ATTACH_FS
2005 * we can get here in cases where we don't "own"
2006 * the volume (e.g. volume owned by a utility).
2007 * short circuit around potential disk header races.
2009 if (V_attachState(vp) != VOL_STATE_ATTACHED) {
2013 V_needsCallback(vp) = 0;
2015 if (VInit >= 2 && V_BreakVolumeCallbacks) {
2016 Log("VAttachVolume: Volume %u was changed externally; breaking callbacks\n", V_id(vp));
2017 (*V_BreakVolumeCallbacks) (V_id(vp));
2020 VUpdateVolume_r(ec, vp, 0);
2022 Log("VAttachVolume: Error updating volume\n");
2027 if (VolumeWriteable(vp) && V_dontSalvage(vp) == 0) {
2028 #ifndef AFS_DEMAND_ATTACH_FS
2029 /* This is a hack: by temporarily setting the incore
2030 * dontSalvage flag ON, the volume will be put back on the
2031 * Update list (with dontSalvage OFF again). It will then
2032 * come back in N minutes with DONT_SALVAGE eventually
2033 * set. This is the way that volumes that have never had
2034 * it set get it set; or that volumes that have been
2035 * offline without DONT SALVAGE having been set also
2036 * eventually get it set */
2037 V_dontSalvage(vp) = DONT_SALVAGE;
2038 #endif /* !AFS_DEMAND_ATTACH_FS */
2039 VAddToVolumeUpdateList_r(ec, vp);
2041 Log("VAttachVolume: Error adding volume to update list\n");
2048 Log("VOnline: volume %u (%s) attached and online\n", V_id(vp),
2053 if (programType == volumeUtility) {
2054 VUnlockPartition_r(partition);
2057 #ifdef AFS_DEMAND_ATTACH_FS
2058 /* attach failed; make sure we're in error state */
2059 if (vp && !VIsErrorState(V_attachState(vp))) {
2060 VChangeState_r(vp, VOL_STATE_ERROR);
2062 #endif /* AFS_DEMAND_ATTACH_FS */
2069 #ifdef AFS_DEMAND_ATTACH_FS
2070 /* VAttachVolumeByVp_r
2072 * finish attaching a volume that is
2073 * in a less than fully attached state
2075 /* caller MUST hold a ref count on vp */
2077 VAttachVolumeByVp_r(Error * ec, Volume * vp, int mode)
2079 char name[VMAXPATHLEN];
2080 int fd, n, reserve = 0;
2081 struct afs_stat status;
2082 struct VolumeDiskHeader diskHeader;
2083 struct VolumeHeader iheader;
2084 struct DiskPartition64 *partp;
2089 VolumeStats stats_save;
2092 /* volume utility should never call AttachByVp */
2093 assert(programType == fileServer);
2095 volumeId = vp->hashid;
2096 partp = vp->partition;
2097 VolumeExternalName_r(volumeId, name, sizeof(name));
2100 /* if another thread is performing a blocking op, wait */
2101 VWaitExclusiveState_r(vp);
2103 memcpy(&stats_save, &vp->stats, sizeof(VolumeStats));
2105 /* if it's already attached, see if we can return it */
2106 if (V_attachState(vp) == VOL_STATE_ATTACHED) {
2107 VGetVolumeByVp_r(ec, vp);
2108 if (V_inUse(vp) == fileServer) {
2111 if (vp->specialStatus == VBUSY)
2113 VDetachVolume_r(ec, vp);
2115 Log("VAttachVolume: Error detaching volume (%s)\n", name);
2121 /* pre-attach volume if it hasn't been done yet */
2123 (V_attachState(vp) == VOL_STATE_UNATTACHED) ||
2124 (V_attachState(vp) == VOL_STATE_ERROR)) {
2125 nvp = VPreAttachVolumeByVp_r(ec, partp, vp, volumeId);
2131 VCreateReservation_r(nvp);
2137 VChangeState_r(vp, VOL_STATE_ATTACHING);
2139 /* restore monotonically increasing stats */
2140 memcpy(&vp->stats, &stats_save, sizeof(VolumeStats));
2145 /* compute path to disk header,
2147 * and verify magic and version stamps */
2148 strcpy(path, VPartitionPath(partp));
2154 if ((fd = afs_open(path, O_RDONLY)) == -1 || afs_fstat(fd, &status) == -1) {
2155 Log("VAttachVolume: Failed to open %s (errno %d)\n", path, errno);
2162 n = read(fd, &diskHeader, sizeof(diskHeader));
2164 if (n != sizeof(diskHeader)
2165 || diskHeader.stamp.magic != VOLUMEHEADERMAGIC) {
2166 Log("VAttachVolume: Error reading volume header %s\n", path);
2171 if (diskHeader.stamp.version != VOLUMEHEADERVERSION) {
2172 Log("VAttachVolume: Volume %s, version number is incorrect; volume needs salvaged\n", path);
2178 /* convert on-disk header format to in-memory header format */
2179 DiskToVolumeHeader(&iheader, &diskHeader);
2183 * NOTE: attach2 is entered without any locks, and returns
2184 * with vol_glock_mutex held */
2185 vp = attach2(ec, volumeId, path, &iheader, partp, vp, isbusy, mode);
2188 * the event that an error was encountered, or
2189 * the volume was not brought to an attached state
2190 * for any reason, skip to the end. We cannot
2191 * safely call VUpdateVolume unless we "own" it.
2195 (V_attachState(vp) != VOL_STATE_ATTACHED)) {
2199 V_needsCallback(vp) = 0;
2200 VUpdateVolume_r(ec, vp, 0);
2202 Log("VAttachVolume: Error updating volume %u\n", vp->hashid);
2206 if (VolumeWriteable(vp) && V_dontSalvage(vp) == 0) {
2207 #ifndef AFS_DEMAND_ATTACH_FS
2208 /* This is a hack: by temporarily setting the incore
2209 * dontSalvage flag ON, the volume will be put back on the
2210 * Update list (with dontSalvage OFF again). It will then
2211 * come back in N minutes with DONT_SALVAGE eventually
2212 * set. This is the way that volumes that have never had
2213 * it set get it set; or that volumes that have been
2214 * offline without DONT SALVAGE having been set also
2215 * eventually get it set */
2216 V_dontSalvage(vp) = DONT_SALVAGE;
2217 #endif /* !AFS_DEMAND_ATTACH_FS */
2218 VAddToVolumeUpdateList_r(ec, vp);
2220 Log("VAttachVolume: Error adding volume %u to update list\n", vp->hashid);
2227 Log("VOnline: volume %u (%s) attached and online\n", V_id(vp),
2231 VCancelReservation_r(nvp);
2234 if (*ec && (*ec != VOFFLINE) && (*ec != VSALVAGE)) {
2235 if (vp && !VIsErrorState(V_attachState(vp))) {
2236 VChangeState_r(vp, VOL_STATE_ERROR);
2243 #endif /* AFS_DEMAND_ATTACH_FS */
2246 * called without any locks held
2247 * returns with vol_glock_mutex held
2250 attach2(Error * ec, VolId volumeId, char *path, register struct VolumeHeader * header,
2251 struct DiskPartition64 * partp, register Volume * vp, int isbusy, int mode)
2253 vp->specialStatus = (byte) (isbusy ? VBUSY : 0);
2254 IH_INIT(vp->vnodeIndex[vLarge].handle, partp->device, header->parent,
2255 header->largeVnodeIndex);
2256 IH_INIT(vp->vnodeIndex[vSmall].handle, partp->device, header->parent,
2257 header->smallVnodeIndex);
2258 IH_INIT(vp->diskDataHandle, partp->device, header->parent,
2259 header->volumeInfo);
2260 IH_INIT(vp->linkHandle, partp->device, header->parent, header->linkTable);
2261 vp->shuttingDown = 0;
2262 vp->goingOffline = 0;
2264 #ifdef AFS_DEMAND_ATTACH_FS
2265 vp->stats.last_attach = FT_ApproxTime();
2266 vp->stats.attaches++;
2270 IncUInt64(&VStats.attaches);
2271 vp->cacheCheck = ++VolumeCacheCheck;
2272 /* just in case this ever rolls over */
2273 if (!vp->cacheCheck)
2274 vp->cacheCheck = ++VolumeCacheCheck;
2275 GetVolumeHeader(vp);
2278 #if defined(AFS_DEMAND_ATTACH_FS) && defined(FSSYNC_BUILD_CLIENT)
2279 /* demand attach changes the V_PEEK mechanism
2281 * we can now suck the current disk data structure over
2282 * the fssync interface without going to disk
2284 * (technically, we don't need to restrict this feature
2285 * to demand attach fileservers. However, I'm trying
2286 * to limit the number of common code changes)
2288 if (programType != fileServer && mode == V_PEEK) {
2290 res.payload.len = sizeof(VolumeDiskData);
2291 res.payload.buf = &vp->header->diskstuff;
2293 if (FSYNC_VolOp(volumeId,
2295 FSYNC_VOL_QUERY_HDR,
2298 goto disk_header_loaded;
2301 #endif /* AFS_DEMAND_ATTACH_FS && FSSYNC_BUILD_CLIENT */
2302 (void)ReadHeader(ec, V_diskDataHandle(vp), (char *)&V_disk(vp),
2303 sizeof(V_disk(vp)), VOLUMEINFOMAGIC, VOLUMEINFOVERSION);
2305 #ifdef AFS_DEMAND_ATTACH_FS
2308 IncUInt64(&VStats.hdr_loads);
2309 IncUInt64(&vp->stats.hdr_loads);
2311 #endif /* AFS_DEMAND_ATTACH_FS */
2314 Log("VAttachVolume: Error reading diskDataHandle vol header %s; error=%u\n", path, *ec);
2319 #ifdef AFS_DEMAND_ATTACH_FS
2322 /* check for pending volume operations */
2323 if (vp->pending_vol_op) {
2324 /* see if the pending volume op requires exclusive access */
2325 if (!VVolOpLeaveOnline_r(vp, vp->pending_vol_op)) {
2326 /* mark the volume down */
2328 VChangeState_r(vp, VOL_STATE_UNATTACHED);
2329 if (V_offlineMessage(vp)[0] == '\0')
2330 strlcpy(V_offlineMessage(vp),
2331 "A volume utility is running.",
2332 sizeof(V_offlineMessage(vp)));
2333 V_offlineMessage(vp)[sizeof(V_offlineMessage(vp)) - 1] = '\0';
2335 /* check to see if we should set the specialStatus flag */
2336 if (VVolOpSetVBusy_r(vp, vp->pending_vol_op)) {
2337 vp->specialStatus = VBUSY;
2342 V_attachFlags(vp) |= VOL_HDR_LOADED;
2343 vp->stats.last_hdr_load = vp->stats.last_attach;
2345 #endif /* AFS_DEMAND_ATTACH_FS */
2348 struct IndexFileHeader iHead;
2350 #if OPENAFS_VOL_STATS
2352 * We just read in the diskstuff part of the header. If the detailed
2353 * volume stats area has not yet been initialized, we should bzero the
2354 * area and mark it as initialized.
2356 if (!(V_stat_initialized(vp))) {
2357 memset((char *)(V_stat_area(vp)), 0, VOL_STATS_BYTES);
2358 V_stat_initialized(vp) = 1;
2360 #endif /* OPENAFS_VOL_STATS */
2362 (void)ReadHeader(ec, vp->vnodeIndex[vSmall].handle,
2363 (char *)&iHead, sizeof(iHead),
2364 SMALLINDEXMAGIC, SMALLINDEXVERSION);
2367 Log("VAttachVolume: Error reading smallVnode vol header %s; error=%u\n", path, *ec);
2372 struct IndexFileHeader iHead;
2374 (void)ReadHeader(ec, vp->vnodeIndex[vLarge].handle,
2375 (char *)&iHead, sizeof(iHead),
2376 LARGEINDEXMAGIC, LARGEINDEXVERSION);
2379 Log("VAttachVolume: Error reading largeVnode vol header %s; error=%u\n", path, *ec);
2383 #ifdef AFS_NAMEI_ENV
2385 struct versionStamp stamp;
2387 (void)ReadHeader(ec, V_linkHandle(vp), (char *)&stamp,
2388 sizeof(stamp), LINKTABLEMAGIC, LINKTABLEVERSION);
2391 Log("VAttachVolume: Error reading namei vol header %s; error=%u\n", path, *ec);
2394 #endif /* AFS_NAMEI_ENV */
2396 #if defined(AFS_DEMAND_ATTACH_FS)
2397 if (*ec && ((*ec != VOFFLINE) || (V_attachState(vp) != VOL_STATE_UNATTACHED))) {
2399 if (programType == fileServer) {
2400 VRequestSalvage_r(ec, vp, SALVSYNC_ERROR, VOL_SALVAGE_INVALIDATE_HEADER);
2403 Log("VAttachVolume: Error attaching volume %s; volume needs salvage; error=%u\n", path, *ec);
2409 /* volume operation in progress */
2413 #else /* AFS_DEMAND_ATTACH_FS */
2415 Log("VAttachVolume: Error attaching volume %s; volume needs salvage; error=%u\n", path, *ec);
2420 #endif /* AFS_DEMAND_ATTACH_FS */
2422 if (V_needsSalvaged(vp)) {
2423 if (vp->specialStatus)
2424 vp->specialStatus = 0;
2426 #if defined(AFS_DEMAND_ATTACH_FS)
2427 if (programType == fileServer) {
2428 VRequestSalvage_r(ec, vp, SALVSYNC_NEEDED, VOL_SALVAGE_INVALIDATE_HEADER);
2431 Log("VAttachVolume: volume salvage flag is ON for %s; volume needs salvage\n", path);
2435 #else /* AFS_DEMAND_ATTACH_FS */
2438 #endif /* AFS_DEMAND_ATTACH_FS */
2443 if (programType == fileServer) {
2444 #ifndef FAST_RESTART
2445 if (V_inUse(vp) && VolumeWriteable(vp)) {
2446 if (!V_needsSalvaged(vp)) {
2447 V_needsSalvaged(vp) = 1;
2448 VUpdateVolume_r(ec, vp, 0);
2450 #if defined(AFS_DEMAND_ATTACH_FS)
2451 VRequestSalvage_r(ec, vp, SALVSYNC_NEEDED, VOL_SALVAGE_INVALIDATE_HEADER);
2453 #else /* AFS_DEMAND_ATTACH_FS */
2454 Log("VAttachVolume: volume %s needs to be salvaged; not attached.\n", path);
2457 #endif /* AFS_DEMAND_ATTACH_FS */
2460 #endif /* FAST_RESTART */
2462 if (V_destroyMe(vp) == DESTROY_ME) {
2463 #if defined(AFS_DEMAND_ATTACH_FS)
2464 /* schedule a salvage so the volume goes away on disk */
2465 VRequestSalvage_r(ec, vp, SALVSYNC_ERROR, VOL_SALVAGE_INVALIDATE_HEADER);
2466 VChangeState_r(vp, VOL_STATE_ERROR);
2468 #endif /* AFS_DEMAND_ATTACH_FS */
2470 Log("VAttachVolume: volume %s is junk; it should be destroyed at next salvage\n", path);
2476 vp->nextVnodeUnique = V_uniquifier(vp);
2477 vp->vnodeIndex[vSmall].bitmap = vp->vnodeIndex[vLarge].bitmap = NULL;
2478 #ifndef BITMAP_LATER
2479 if (programType == fileServer && VolumeWriteable(vp)) {
2481 for (i = 0; i < nVNODECLASSES; i++) {
2482 VGetBitmap_r(ec, vp, i);
2484 #ifdef AFS_DEMAND_ATTACH_FS
2485 VRequestSalvage_r(ec, vp, SALVSYNC_ERROR, VOL_SALVAGE_INVALIDATE_HEADER);
2487 #else /* AFS_DEMAND_ATTACH_FS */
2489 #endif /* AFS_DEMAND_ATTACH_FS */
2490 Log("VAttachVolume: error getting bitmap for volume (%s)\n",
2496 #endif /* BITMAP_LATER */
2498 if (programType == fileServer) {
2499 if (vp->specialStatus)
2500 vp->specialStatus = 0;
2501 if (V_blessed(vp) && V_inService(vp) && !V_needsSalvaged(vp)) {
2502 V_inUse(vp) = fileServer;
2503 V_offlineMessage(vp)[0] = '\0';
2506 if ((mode != V_PEEK) && (mode != V_SECRETLY))
2507 V_inUse(vp) = programType;
2508 V_checkoutMode(vp) = mode;
2511 AddVolumeToHashTable(vp, V_id(vp));
2512 #ifdef AFS_DEMAND_ATTACH_FS
2513 if ((programType != fileServer) ||
2514 (V_inUse(vp) == fileServer)) {
2515 AddVolumeToVByPList_r(vp);
2517 VChangeState_r(vp, VOL_STATE_ATTACHED);
2519 VChangeState_r(vp, VOL_STATE_UNATTACHED);
2525 /* Attach an existing volume.
2526 The volume also normally goes online at this time.
2527 An offline volume must be reattached to make it go online.
2531 VAttachVolume(Error * ec, VolumeId volumeId, int mode)
2535 retVal = VAttachVolume_r(ec, volumeId, mode);
2541 VAttachVolume_r(Error * ec, VolumeId volumeId, int mode)
2544 VGetVolumePath(ec, volumeId, &part, &name);
2546 register Volume *vp;
2548 vp = VGetVolume_r(&error, volumeId);
2550 assert(V_inUse(vp) == 0);
2551 VDetachVolume_r(ec, vp);
2555 return VAttachVolumeByName_r(ec, part, name, mode);
2558 /* Increment a reference count to a volume, sans context swaps. Requires
2559 * possibly reading the volume header in from the disk, since there's
2560 * an invariant in the volume package that nUsers>0 ==> vp->header is valid.
2562 * N.B. This call can fail if we can't read in the header!! In this case
2563 * we still guarantee we won't context swap, but the ref count won't be
2564 * incremented (otherwise we'd violate the invariant).
2566 /* NOTE: with the demand attach fileserver extensions, the global lock
2567 * is dropped within VHold */
2568 #ifdef AFS_DEMAND_ATTACH_FS
2570 VHold_r(register Volume * vp)
2574 VCreateReservation_r(vp);
2575 VWaitExclusiveState_r(vp);
2577 LoadVolumeHeader(&error, vp);
2579 VCancelReservation_r(vp);
2583 VCancelReservation_r(vp);
2586 #else /* AFS_DEMAND_ATTACH_FS */
2588 VHold_r(register Volume * vp)
2592 LoadVolumeHeader(&error, vp);
2598 #endif /* AFS_DEMAND_ATTACH_FS */
2601 VHold(register Volume * vp)
2605 retVal = VHold_r(vp);
2611 /***************************************************/
2612 /* get and put volume routines */
2613 /***************************************************/
2616 * put back a heavyweight reference to a volume object.
2618 * @param[in] vp volume object pointer
2620 * @pre VOL_LOCK held
2622 * @post heavyweight volume reference put back.
2623 * depending on state, volume may have been taken offline,
2624 * detached, salvaged, freed, etc.
2626 * @internal volume package internal use only
2629 VPutVolume_r(register Volume * vp)
2631 assert(--vp->nUsers >= 0);
2632 if (vp->nUsers == 0) {
2634 ReleaseVolumeHeader(vp->header);
2635 #ifdef AFS_DEMAND_ATTACH_FS
2636 if (!VCheckDetach(vp)) {
2640 #else /* AFS_DEMAND_ATTACH_FS */
2642 #endif /* AFS_DEMAND_ATTACH_FS */
2647 VPutVolume(register Volume * vp)
2655 /* Get a pointer to an attached volume. The pointer is returned regardless
2656 of whether or not the volume is in service or on/off line. An error
2657 code, however, is returned with an indication of the volume's status */
2659 VGetVolume(Error * ec, Error * client_ec, VolId volumeId)
2663 retVal = GetVolume(ec, client_ec, volumeId, NULL, 0);
2669 VGetVolume_r(Error * ec, VolId volumeId)
2671 return GetVolume(ec, NULL, volumeId, NULL, 0);
2674 /* try to get a volume we've previously looked up */
2675 /* for demand attach fs, caller MUST NOT hold a ref count on vp */
2677 VGetVolumeByVp_r(Error * ec, Volume * vp)
2679 return GetVolume(ec, NULL, vp->hashid, vp, 0);
2682 /* private interface for getting a volume handle
2683 * volumeId must be provided.
2684 * hint is an optional parameter to speed up hash lookups
2685 * flags is not used at this time
2687 /* for demand attach fs, caller MUST NOT hold a ref count on hint */
2689 GetVolume(Error * ec, Error * client_ec, VolId volumeId, Volume * hint, int flags)
2692 /* pull this profiling/debugging code out of regular builds */
2694 #define VGET_CTR_INC(x) x++
2695 unsigned short V0 = 0, V1 = 0, V2 = 0, V3 = 0, V5 = 0, V6 =
2696 0, V7 = 0, V8 = 0, V9 = 0;
2697 unsigned short V10 = 0, V11 = 0, V12 = 0, V13 = 0, V14 = 0, V15 = 0;
2699 #define VGET_CTR_INC(x)
2701 #ifdef AFS_DEMAND_ATTACH_FS
2702 Volume *avp, * rvp = hint;
2706 * if VInit is zero, the volume package dynamic
2707 * data structures have not been initialized yet,
2708 * and we must immediately return an error
2714 *client_ec = VOFFLINE;
2719 #ifdef AFS_DEMAND_ATTACH_FS
2721 VCreateReservation_r(rvp);
2723 #endif /* AFS_DEMAND_ATTACH_FS */
2731 vp = VLookupVolume_r(ec, volumeId, vp);
2737 #ifdef AFS_DEMAND_ATTACH_FS
2738 if (rvp && (rvp != vp)) {
2739 /* break reservation on old vp */
2740 VCancelReservation_r(rvp);
2743 #endif /* AFS_DEMAND_ATTACH_FS */
2749 /* Until we have reached an initialization level of 2
2750 * we don't know whether this volume exists or not.
2751 * We can't sleep and retry later because before a volume
2752 * is attached, the caller tries to get it first. Just
2753 * return VOFFLINE and the caller can choose whether to
2754 * retry the command or not. */
2764 IncUInt64(&VStats.hdr_gets);
2766 #ifdef AFS_DEMAND_ATTACH_FS
2767 /* block if someone else is performing an exclusive op on this volume */
2770 VCreateReservation_r(rvp);
2772 VWaitExclusiveState_r(vp);
2774 /* short circuit with VNOVOL in the following circumstances:
2777 * - VOL_STATE_SHUTTING_DOWN
2779 if ((V_attachState(vp) == VOL_STATE_ERROR) ||
2780 (V_attachState(vp) == VOL_STATE_SHUTTING_DOWN)) {
2787 * short circuit with VOFFLINE in the following circumstances:
2789 * - VOL_STATE_UNATTACHED
2791 if (V_attachState(vp) == VOL_STATE_UNATTACHED) {
2792 if (vp->specialStatus) {
2793 *ec = vp->specialStatus;
2801 /* allowable states:
2808 if (vp->salvage.requested) {
2809 VUpdateSalvagePriority_r(vp);
2812 if (V_attachState(vp) == VOL_STATE_PREATTACHED) {
2813 avp = VAttachVolumeByVp_r(ec, vp, 0);
2816 /* VAttachVolumeByVp_r can return a pointer
2817 * != the vp passed to it under certain
2818 * conditions; make sure we don't leak
2819 * reservations if that happens */
2821 VCancelReservation_r(rvp);
2823 VCreateReservation_r(rvp);
2833 if (!vp->pending_vol_op) {
2848 if ((V_attachState(vp) == VOL_STATE_SALVAGING) ||
2849 (*ec == VSALVAGING)) {
2851 /* see CheckVnode() in afsfileprocs.c for an explanation
2852 * of this error code logic */
2853 afs_uint32 now = FT_ApproxTime();
2854 if ((vp->stats.last_salvage + (10 * 60)) >= now) {
2857 *client_ec = VRESTARTING;
2866 LoadVolumeHeader(ec, vp);
2869 /* Only log the error if it was a totally unexpected error. Simply
2870 * a missing inode is likely to be caused by the volume being deleted */
2871 if (errno != ENXIO || LogLevel)
2872 Log("Volume %u: couldn't reread volume header\n",
2874 #ifdef AFS_DEMAND_ATTACH_FS
2875 if (programType == fileServer) {
2876 VRequestSalvage_r(ec, vp, SALVSYNC_ERROR, VOL_SALVAGE_INVALIDATE_HEADER);
2881 #else /* AFS_DEMAND_ATTACH_FS */
2884 #endif /* AFS_DEMAND_ATTACH_FS */
2888 #ifdef AFS_DEMAND_ATTACH_FS
2890 * this test MUST happen after the volume header is loaded
2892 if (vp->pending_vol_op && !VVolOpLeaveOnline_r(vp, vp->pending_vol_op)) {
2894 * volume cannot remain online during this volume operation.
2897 if (vp->specialStatus) {
2899 * special status codes outrank normal VOFFLINE code
2901 *ec = vp->specialStatus;
2903 *client_ec = vp->specialStatus;
2907 /* see CheckVnode() in afsfileprocs.c for an explanation
2908 * of this error code logic */
2909 afs_uint32 now = FT_ApproxTime();
2910 if ((vp->stats.last_vol_op + (10 * 60)) >= now) {
2913 *client_ec = VRESTARTING;
2918 VChangeState_r(vp, VOL_STATE_UNATTACHED);
2919 FreeVolumeHeader(vp);
2923 #endif /* AFS_DEMAND_ATTACH_FS */
2926 if (vp->shuttingDown) {
2933 if (programType == fileServer) {
2935 if (vp->goingOffline) {
2937 #ifdef AFS_DEMAND_ATTACH_FS
2938 /* wait for the volume to go offline */
2939 if (V_attachState(vp) == VOL_STATE_GOING_OFFLINE) {
2940 VWaitStateChange_r(vp);
2942 #elif defined(AFS_PTHREAD_ENV)
2943 VOL_CV_WAIT(&vol_put_volume_cond);
2944 #else /* AFS_PTHREAD_ENV */
2945 LWP_WaitProcess(VPutVolume);
2946 #endif /* AFS_PTHREAD_ENV */
2949 if (vp->specialStatus) {
2951 *ec = vp->specialStatus;
2952 } else if (V_inService(vp) == 0 || V_blessed(vp) == 0) {
2955 } else if (V_inUse(vp) == 0) {
2966 #ifdef AFS_DEMAND_ATTACH_FS
2967 /* if no error, bump nUsers */
2970 VLRU_UpdateAccess_r(vp);
2973 VCancelReservation_r(rvp);
2976 if (client_ec && !*client_ec) {
2979 #else /* AFS_DEMAND_ATTACH_FS */
2980 /* if no error, bump nUsers */
2987 #endif /* AFS_DEMAND_ATTACH_FS */
2995 /***************************************************/
2996 /* Volume offline/detach routines */
2997 /***************************************************/
2999 /* caller MUST hold a heavyweight ref on vp */
3000 #ifdef AFS_DEMAND_ATTACH_FS
3002 VTakeOffline_r(register Volume * vp)
3006 assert(vp->nUsers > 0);
3007 assert(programType == fileServer);
3009 VCreateReservation_r(vp);
3010 VWaitExclusiveState_r(vp);
3012 vp->goingOffline = 1;
3013 V_needsSalvaged(vp) = 1;
3015 VRequestSalvage_r(&error, vp, SALVSYNC_ERROR, 0);
3016 VCancelReservation_r(vp);
3018 #else /* AFS_DEMAND_ATTACH_FS */
3020 VTakeOffline_r(register Volume * vp)
3022 assert(vp->nUsers > 0);
3023 assert(programType == fileServer);
3025 vp->goingOffline = 1;
3026 V_needsSalvaged(vp) = 1;
3028 #endif /* AFS_DEMAND_ATTACH_FS */
3031 VTakeOffline(register Volume * vp)
3039 * force a volume offline.
3041 * @param[in] vp volume object pointer
3042 * @param[in] flags flags (see note below)
3044 * @note the flag VOL_FORCEOFF_NOUPDATE is a recursion control flag
3045 * used when VUpdateVolume_r needs to call VForceOffline_r
3046 * (which in turn would normally call VUpdateVolume_r)
3048 * @see VUpdateVolume_r
3050 * @pre VOL_LOCK must be held.
3051 * for DAFS, caller must hold ref.
3053 * @note for DAFS, it _is safe_ to call this function from an
3056 * @post needsSalvaged flag is set.
3057 * for DAFS, salvage is requested.
3058 * no further references to the volume through the volume
3059 * package will be honored.
3060 * all file descriptor and vnode caches are invalidated.
3062 * @warning this is a heavy-handed interface. it results in
3063 * a volume going offline regardless of the current
3064 * reference count state.
3066 * @internal volume package internal use only
3069 VForceOffline_r(Volume * vp, int flags)
3073 #ifdef AFS_DEMAND_ATTACH_FS
3074 VChangeState_r(vp, VOL_STATE_ERROR);
3079 strcpy(V_offlineMessage(vp),
3080 "Forced offline due to internal error: volume needs to be salvaged");
3081 Log("Volume %u forced offline: it needs salvaging!\n", V_id(vp));
3084 vp->goingOffline = 0;
3085 V_needsSalvaged(vp) = 1;
3086 if (!(flags & VOL_FORCEOFF_NOUPDATE)) {
3087 VUpdateVolume_r(&error, vp, VOL_UPDATE_NOFORCEOFF);
3090 #ifdef AFS_DEMAND_ATTACH_FS
3091 VRequestSalvage_r(&error, vp, SALVSYNC_ERROR, VOL_SALVAGE_INVALIDATE_HEADER);
3092 #endif /* AFS_DEMAND_ATTACH_FS */
3094 #ifdef AFS_PTHREAD_ENV
3095 assert(pthread_cond_broadcast(&vol_put_volume_cond) == 0);
3096 #else /* AFS_PTHREAD_ENV */
3097 LWP_NoYieldSignal(VPutVolume);
3098 #endif /* AFS_PTHREAD_ENV */
3100 VReleaseVolumeHandles_r(vp);
3104 * force a volume offline.
3106 * @param[in] vp volume object pointer
3108 * @see VForceOffline_r
3111 VForceOffline(Volume * vp)
3114 VForceOffline_r(vp, 0);
3118 /* The opposite of VAttachVolume. The volume header is written to disk, with
3119 the inUse bit turned off. A copy of the header is maintained in memory,
3120 however (which is why this is VOffline, not VDetach).
3123 VOffline_r(Volume * vp, char *message)
3126 VolumeId vid = V_id(vp);
3128 assert(programType != volumeUtility);
3133 if (V_offlineMessage(vp)[0] == '\0')
3134 strncpy(V_offlineMessage(vp), message, sizeof(V_offlineMessage(vp)));
3135 V_offlineMessage(vp)[sizeof(V_offlineMessage(vp)) - 1] = '\0';
3137 vp->goingOffline = 1;
3138 #ifdef AFS_DEMAND_ATTACH_FS
3139 VChangeState_r(vp, VOL_STATE_GOING_OFFLINE);
3140 VCreateReservation_r(vp);
3143 /* wait for the volume to go offline */
3144 if (V_attachState(vp) == VOL_STATE_GOING_OFFLINE) {
3145 VWaitStateChange_r(vp);
3147 VCancelReservation_r(vp);
3148 #else /* AFS_DEMAND_ATTACH_FS */
3150 vp = VGetVolume_r(&error, vid); /* Wait for it to go offline */
3151 if (vp) /* In case it was reattached... */
3153 #endif /* AFS_DEMAND_ATTACH_FS */
3157 VOffline(Volume * vp, char *message)
3160 VOffline_r(vp, message);
3164 /* This gets used for the most part by utility routines that don't want
3165 * to keep all the volume headers around. Generally, the file server won't
3166 * call this routine, because then the offline message in the volume header
3167 * (or other information) won't be available to clients. For NAMEI, also
3168 * close the file handles. However, the fileserver does call this during
3169 * an attach following a volume operation.
3172 VDetachVolume_r(Error * ec, Volume * vp)
3175 struct DiskPartition64 *tpartp;
3176 int notifyServer = 0;
3177 int useDone = FSYNC_VOL_ON;
3179 *ec = 0; /* always "succeeds" */
3180 if (programType == volumeUtility) {
3181 notifyServer = vp->needsPutBack;
3182 if (V_destroyMe(vp) == DESTROY_ME)
3183 useDone = FSYNC_VOL_DONE;
3184 #ifdef AFS_DEMAND_ATTACH_FS
3185 else if (!V_blessed(vp) || !V_inService(vp))
3186 useDone = FSYNC_VOL_LEAVE_OFF;
3189 tpartp = vp->partition;
3191 DeleteVolumeFromHashTable(vp);
3192 vp->shuttingDown = 1;
3193 #ifdef AFS_DEMAND_ATTACH_FS
3194 DeleteVolumeFromVByPList_r(vp);
3196 VChangeState_r(vp, VOL_STATE_SHUTTING_DOWN);
3198 if (programType != fileServer)
3200 #endif /* AFS_DEMAND_ATTACH_FS */
3202 /* Will be detached sometime in the future--this is OK since volume is offline */
3204 /* XXX the following code should really be moved to VCheckDetach() since the volume
3205 * is not technically detached until the refcounts reach zero
3207 #ifdef FSSYNC_BUILD_CLIENT
3208 if (programType == volumeUtility && notifyServer) {
3210 * Note: The server is not notified in the case of a bogus volume
3211 * explicitly to make it possible to create a volume, do a partial
3212 * restore, then abort the operation without ever putting the volume
3213 * online. This is essential in the case of a volume move operation
3214 * between two partitions on the same server. In that case, there
3215 * would be two instances of the same volume, one of them bogus,
3216 * which the file server would attempt to put on line
3218 FSYNC_VolOp(volume, tpartp->name, useDone, 0, NULL);
3219 /* XXX this code path is only hit by volume utilities, thus
3220 * V_BreakVolumeCallbacks will always be NULL. if we really
3221 * want to break callbacks in this path we need to use FSYNC_VolOp() */
3223 /* Dettaching it so break all callbacks on it */
3224 if (V_BreakVolumeCallbacks) {
3225 Log("volume %u detached; breaking all call backs\n", volume);
3226 (*V_BreakVolumeCallbacks) (volume);
3230 #endif /* FSSYNC_BUILD_CLIENT */
3234 VDetachVolume(Error * ec, Volume * vp)
3237 VDetachVolume_r(ec, vp);
3242 /***************************************************/
3243 /* Volume fd/inode handle closing routines */
3244 /***************************************************/
3246 /* For VDetachVolume, we close all cached file descriptors, but keep
3247 * the Inode handles in case we need to read from a busy volume.
3249 /* for demand attach, caller MUST hold ref count on vp */
3251 VCloseVolumeHandles_r(Volume * vp)
3253 #ifdef AFS_DEMAND_ATTACH_FS
3254 VolState state_save;
3256 state_save = VChangeState_r(vp, VOL_STATE_OFFLINING);
3261 * XXX need to investigate whether we can perform
3262 * DFlushVolume outside of vol_glock_mutex...
3264 * VCloseVnodeFiles_r drops the glock internally */
3265 DFlushVolume(V_id(vp));
3266 VCloseVnodeFiles_r(vp);
3268 #ifdef AFS_DEMAND_ATTACH_FS
3272 /* Too time consuming and unnecessary for the volserver */
3273 if (programType != volumeUtility) {
3274 IH_CONDSYNC(vp->vnodeIndex[vLarge].handle);
3275 IH_CONDSYNC(vp->vnodeIndex[vSmall].handle);
3276 IH_CONDSYNC(vp->diskDataHandle);
3278 IH_CONDSYNC(vp->linkHandle);
3279 #endif /* AFS_NT40_ENV */
3282 IH_REALLYCLOSE(vp->vnodeIndex[vLarge].handle);
3283 IH_REALLYCLOSE(vp->vnodeIndex[vSmall].handle);
3284 IH_REALLYCLOSE(vp->diskDataHandle);
3285 IH_REALLYCLOSE(vp->linkHandle);
3287 #ifdef AFS_DEMAND_ATTACH_FS
3289 VChangeState_r(vp, state_save);
3293 /* For both VForceOffline and VOffline, we close all relevant handles.
3294 * For VOffline, if we re-attach the volume, the files may possible be
3295 * different than before.
3297 /* for demand attach, caller MUST hold a ref count on vp */
3299 VReleaseVolumeHandles_r(Volume * vp)
3301 #ifdef AFS_DEMAND_ATTACH_FS
3302 VolState state_save;
3304 state_save = VChangeState_r(vp, VOL_STATE_DETACHING);
3307 /* XXX need to investigate whether we can perform
3308 * DFlushVolume outside of vol_glock_mutex... */
3309 DFlushVolume(V_id(vp));
3311 VReleaseVnodeFiles_r(vp); /* releases the glock internally */
3313 #ifdef AFS_DEMAND_ATTACH_FS
3317 /* Too time consuming and unnecessary for the volserver */
3318 if (programType != volumeUtility) {
3319 IH_CONDSYNC(vp->vnodeIndex[vLarge].handle);
3320 IH_CONDSYNC(vp->vnodeIndex[vSmall].handle);
3321 IH_CONDSYNC(vp->diskDataHandle);
3323 IH_CONDSYNC(vp->linkHandle);
3324 #endif /* AFS_NT40_ENV */
3327 IH_RELEASE(vp->vnodeIndex[vLarge].handle);
3328 IH_RELEASE(vp->vnodeIndex[vSmall].handle);
3329 IH_RELEASE(vp->diskDataHandle);
3330 IH_RELEASE(vp->linkHandle);
3332 #ifdef AFS_DEMAND_ATTACH_FS
3334 VChangeState_r(vp, state_save);
3339 /***************************************************/
3340 /* Volume write and fsync routines */
3341 /***************************************************/
3344 VUpdateVolume_r(Error * ec, Volume * vp, int flags)
3346 #ifdef AFS_DEMAND_ATTACH_FS
3347 VolState state_save;
3349 if (flags & VOL_UPDATE_WAIT) {
3350 VCreateReservation_r(vp);
3351 VWaitExclusiveState_r(vp);
3356 if (programType == fileServer)
3358 (V_inUse(vp) ? V_nextVnodeUnique(vp) +
3359 200 : V_nextVnodeUnique(vp));
3361 #ifdef AFS_DEMAND_ATTACH_FS
3362 state_save = VChangeState_r(vp, VOL_STATE_UPDATING);
3366 WriteVolumeHeader_r(ec, vp);
3368 #ifdef AFS_DEMAND_ATTACH_FS
3370 VChangeState_r(vp, state_save);
3371 if (flags & VOL_UPDATE_WAIT) {
3372 VCancelReservation_r(vp);
3377 Log("VUpdateVolume: error updating volume header, volume %u (%s)\n",
3378 V_id(vp), V_name(vp));
3379 /* try to update on-disk header,
3380 * while preventing infinite recursion */
3381 if (!(flags & VOL_UPDATE_NOFORCEOFF)) {
3382 VForceOffline_r(vp, VOL_FORCEOFF_NOUPDATE);
3388 VUpdateVolume(Error * ec, Volume * vp)
3391 VUpdateVolume_r(ec, vp, VOL_UPDATE_WAIT);
3396 VSyncVolume_r(Error * ec, Volume * vp, int flags)
3400 #ifdef AFS_DEMAND_ATTACH_FS
3401 VolState state_save;
3404 if (flags & VOL_SYNC_WAIT) {
3405 VUpdateVolume_r(ec, vp, VOL_UPDATE_WAIT);
3407 VUpdateVolume_r(ec, vp, 0);
3410 #ifdef AFS_DEMAND_ATTACH_FS
3411 state_save = VChangeState_r(vp, VOL_STATE_UPDATING);
3414 fdP = IH_OPEN(V_diskDataHandle(vp));
3415 assert(fdP != NULL);
3416 code = FDH_SYNC(fdP);
3419 #ifdef AFS_DEMAND_ATTACH_FS
3421 VChangeState_r(vp, state_save);
3427 VSyncVolume(Error * ec, Volume * vp)
3430 VSyncVolume_r(ec, vp, VOL_SYNC_WAIT);
3435 /***************************************************/
3436 /* Volume dealloaction routines */
3437 /***************************************************/
3439 #ifdef AFS_DEMAND_ATTACH_FS
3441 FreeVolume(Volume * vp)
3443 /* free the heap space, iff it's safe.
3444 * otherwise, pull it out of the hash table, so it
3445 * will get deallocated when all refs to it go away */
3446 if (!VCheckFree(vp)) {
3447 DeleteVolumeFromHashTable(vp);
3448 DeleteVolumeFromVByPList_r(vp);
3450 /* make sure we invalidate the header cache entry */
3451 FreeVolumeHeader(vp);
3454 #endif /* AFS_DEMAND_ATTACH_FS */
3457 ReallyFreeVolume(Volume * vp)
3462 #ifdef AFS_DEMAND_ATTACH_FS
3464 VChangeState_r(vp, VOL_STATE_FREED);
3465 if (vp->pending_vol_op)
3466 free(vp->pending_vol_op);
3467 #endif /* AFS_DEMAND_ATTACH_FS */
3468 for (i = 0; i < nVNODECLASSES; i++)
3469 if (vp->vnodeIndex[i].bitmap)
3470 free(vp->vnodeIndex[i].bitmap);
3471 FreeVolumeHeader(vp);
3472 #ifndef AFS_DEMAND_ATTACH_FS
3473 DeleteVolumeFromHashTable(vp);
3474 #endif /* AFS_DEMAND_ATTACH_FS */
3478 /* check to see if we should shutdown this volume
3479 * returns 1 if volume was freed, 0 otherwise */
3480 #ifdef AFS_DEMAND_ATTACH_FS
3482 VCheckDetach(register Volume * vp)
3487 if (vp->nUsers || vp->nWaiters)
3490 if (vp->shuttingDown) {
3492 if ((programType != fileServer) &&
3493 (V_inUse(vp) == programType) &&
3494 ((V_checkoutMode(vp) == V_VOLUPD) ||
3495 ((V_checkoutMode(vp) == V_CLONE) &&
3496 (VolumeWriteable(vp))))) {
3498 VUpdateVolume_r(&ec, vp, VOL_UPDATE_NOFORCEOFF);
3500 Log("VCheckDetach: volume header update for volume %u "
3501 "failed with errno %d\n", vp->hashid, errno);
3504 VReleaseVolumeHandles_r(vp);
3506 ReallyFreeVolume(vp);
3507 if (programType == fileServer) {
3508 assert(pthread_cond_broadcast(&vol_put_volume_cond) == 0);
3513 #else /* AFS_DEMAND_ATTACH_FS */
3515 VCheckDetach(register Volume * vp)
3523 if (vp->shuttingDown) {
3525 if ((programType != fileServer) &&
3526 (V_inUse(vp) == programType) &&
3527 ((V_checkoutMode(vp) == V_VOLUPD) ||
3528 ((V_checkoutMode(vp) == V_CLONE) &&
3529 (VolumeWriteable(vp))))) {
3531 VUpdateVolume_r(&ec, vp, VOL_UPDATE_NOFORCEOFF);
3533 Log("VCheckDetach: volume header update for volume %u failed with errno %d\n",
3537 VReleaseVolumeHandles_r(vp);
3538 ReallyFreeVolume(vp);
3539 if (programType == fileServer) {
3540 #if defined(AFS_PTHREAD_ENV)
3541 assert(pthread_cond_broadcast(&vol_put_volume_cond) == 0);
3542 #else /* AFS_PTHREAD_ENV */
3543 LWP_NoYieldSignal(VPutVolume);
3544 #endif /* AFS_PTHREAD_ENV */
3549 #endif /* AFS_DEMAND_ATTACH_FS */
3551 /* check to see if we should offline this volume
3552 * return 1 if volume went offline, 0 otherwise */
3553 #ifdef AFS_DEMAND_ATTACH_FS
3555 VCheckOffline(register Volume * vp)
3557 Volume * rvp = NULL;
3560 if (vp->goingOffline && !vp->nUsers) {
3562 assert(programType == fileServer);
3563 assert((V_attachState(vp) != VOL_STATE_ATTACHED) &&
3564 (V_attachState(vp) != VOL_STATE_FREED) &&
3565 (V_attachState(vp) != VOL_STATE_PREATTACHED) &&
3566 (V_attachState(vp) != VOL_STATE_UNATTACHED));
3570 * VOL_STATE_GOING_OFFLINE
3571 * VOL_STATE_SHUTTING_DOWN
3572 * VIsErrorState(V_attachState(vp))
3573 * VIsExclusiveState(V_attachState(vp))
3576 VCreateReservation_r(vp);
3577 VChangeState_r(vp, VOL_STATE_OFFLINING);
3580 /* must clear the goingOffline flag before we drop the glock */
3581 vp->goingOffline = 0;
3586 /* perform async operations */
3587 VUpdateVolume_r(&error, vp, 0);
3588 VCloseVolumeHandles_r(vp);
3591 Log("VOffline: Volume %u (%s) is now offline", V_id(vp),
3593 if (V_offlineMessage(vp)[0])
3594 Log(" (%s)", V_offlineMessage(vp));
3598 /* invalidate the volume header cache entry */
3599 FreeVolumeHeader(vp);
3601 /* if nothing changed state to error or salvaging,
3602 * drop state to unattached */
3603 if (!VIsErrorState(V_attachState(vp))) {
3604 VChangeState_r(vp, VOL_STATE_UNATTACHED);
3606 VCancelReservation_r(vp);
3607 /* no usage of vp is safe beyond this point */
3611 #else /* AFS_DEMAND_ATTACH_FS */
3613 VCheckOffline(register Volume * vp)
3615 Volume * rvp = NULL;
3618 if (vp->goingOffline && !vp->nUsers) {
3620 assert(programType == fileServer);
3623 vp->goingOffline = 0;
3625 VUpdateVolume_r(&error, vp, 0);
3626 VCloseVolumeHandles_r(vp);
3628 Log("VOffline: Volume %u (%s) is now offline", V_id(vp),
3630 if (V_offlineMessage(vp)[0])
3631 Log(" (%s)", V_offlineMessage(vp));
3634 FreeVolumeHeader(vp);
3635 #ifdef AFS_PTHREAD_ENV
3636 assert(pthread_cond_broadcast(&vol_put_volume_cond) == 0);
3637 #else /* AFS_PTHREAD_ENV */
3638 LWP_NoYieldSignal(VPutVolume);
3639 #endif /* AFS_PTHREAD_ENV */
3643 #endif /* AFS_DEMAND_ATTACH_FS */
3645 /***************************************************/
3646 /* demand attach fs ref counting routines */
3647 /***************************************************/
3649 #ifdef AFS_DEMAND_ATTACH_FS
3650 /* the following two functions handle reference counting for
3651 * asynchronous operations on volume structs.
3653 * their purpose is to prevent a VDetachVolume or VShutdown
3654 * from free()ing the Volume struct during an async i/o op */
3656 /* register with the async volume op ref counter */
3657 /* VCreateReservation_r moved into inline code header because it
3658 * is now needed in vnode.c -- tkeiser 11/20/2007
3662 * decrement volume-package internal refcount.
3664 * @param vp volume object pointer
3666 * @internal volume package internal use only
3669 * @arg VOL_LOCK is held
3670 * @arg lightweight refcount held
3672 * @post volume waiters refcount is decremented; volume may
3673 * have been deallocated/shutdown/offlined/salvaged/
3674 * whatever during the process
3676 * @warning once you have tossed your last reference (you can acquire
3677 * lightweight refs recursively) it is NOT SAFE to reference
3678 * a volume object pointer ever again
3680 * @see VCreateReservation_r
3682 * @note DEMAND_ATTACH_FS only
3685 VCancelReservation_r(Volume * vp)
3687 assert(--vp->nWaiters >= 0);
3688 if (vp->nWaiters == 0) {
3690 if (!VCheckDetach(vp)) {
3697 /* check to see if we should free this volume now
3698 * return 1 if volume was freed, 0 otherwise */
3700 VCheckFree(Volume * vp)
3703 if ((vp->nUsers == 0) &&
3704 (vp->nWaiters == 0) &&
3705 !(V_attachFlags(vp) & (VOL_IN_HASH |
3709 ReallyFreeVolume(vp);
3714 #endif /* AFS_DEMAND_ATTACH_FS */
3717 /***************************************************/
3718 /* online volume operations routines */
3719 /***************************************************/
3721 #ifdef AFS_DEMAND_ATTACH_FS
3723 * register a volume operation on a given volume.
3725 * @param[in] vp volume object
3726 * @param[in] vopinfo volume operation info object
3728 * @pre VOL_LOCK is held
3730 * @post volume operation info object attached to volume object.
3731 * volume operation statistics updated.
3733 * @note by "attached" we mean a copy of the passed in object is made
3735 * @internal volume package internal use only
3738 VRegisterVolOp_r(Volume * vp, FSSYNC_VolOp_info * vopinfo)
3740 FSSYNC_VolOp_info * info;
3742 /* attach a vol op info node to the volume struct */
3743 info = (FSSYNC_VolOp_info *) malloc(sizeof(FSSYNC_VolOp_info));
3744 assert(info != NULL);
3745 memcpy(info, vopinfo, sizeof(FSSYNC_VolOp_info));
3746 vp->pending_vol_op = info;
3749 vp->stats.last_vol_op = FT_ApproxTime();
3750 vp->stats.vol_ops++;
3751 IncUInt64(&VStats.vol_ops);
3757 * deregister the volume operation attached to this volume.
3759 * @param[in] vp volume object pointer
3761 * @pre VOL_LOCK is held
3763 * @post the volume operation info object is detached from the volume object
3765 * @internal volume package internal use only
3768 VDeregisterVolOp_r(Volume * vp)
3770 if (vp->pending_vol_op) {
3771 free(vp->pending_vol_op);
3772 vp->pending_vol_op = NULL;
3776 #endif /* AFS_DEMAND_ATTACH_FS */
3779 * determine whether it is safe to leave a volume online during
3780 * the volume operation described by the vopinfo object.
3782 * @param[in] vp volume object
3783 * @param[in] vopinfo volume operation info object
3785 * @return whether it is safe to leave volume online
3786 * @retval 0 it is NOT SAFE to leave the volume online
3787 * @retval 1 it is safe to leave the volume online during the operation
3790 * @arg VOL_LOCK is held
3791 * @arg disk header attached to vp (heavyweight ref on vp will guarantee
3792 * this condition is met)
3794 * @internal volume package internal use only
3797 VVolOpLeaveOnline_r(Volume * vp, FSSYNC_VolOp_info * vopinfo)
3799 return (vopinfo->com.command == FSYNC_VOL_NEEDVOLUME &&
3800 (vopinfo->com.reason == V_READONLY ||
3801 (!VolumeWriteable(vp) &&
3802 (vopinfo->com.reason == V_CLONE ||
3803 vopinfo->com.reason == V_DUMP))));
3807 * determine whether VBUSY should be set during this volume operation.
3809 * @param[in] vp volume object
3810 * @param[in] vopinfo volume operation info object
3812 * @return whether VBUSY should be set
3813 * @retval 0 VBUSY does NOT need to be set
3814 * @retval 1 VBUSY SHOULD be set
3816 * @pre VOL_LOCK is held
3818 * @internal volume package internal use only
3821 VVolOpSetVBusy_r(Volume * vp, FSSYNC_VolOp_info * vopinfo)
3823 return ((vopinfo->com.command == FSYNC_VOL_OFF &&
3824 vopinfo->com.reason == FSYNC_SALVAGE) ||
3825 (vopinfo->com.command == FSYNC_VOL_NEEDVOLUME &&
3826 (vopinfo->com.reason == V_CLONE ||
3827 vopinfo->com.reason == V_DUMP)));
3831 /***************************************************/
3832 /* online salvager routines */
3833 /***************************************************/
3834 #if defined(AFS_DEMAND_ATTACH_FS)
3835 #define SALVAGE_PRIO_UPDATE_INTERVAL 3 /**< number of seconds between prio updates */
3836 #define SALVAGE_COUNT_MAX 16 /**< number of online salvages we
3837 * allow before moving the volume
3838 * into a permanent error state
3840 * once this threshold is reached,
3841 * the operator will have to manually
3842 * issue a 'bos salvage' to bring
3843 * the volume back online
3847 * check whether a salvage needs to be performed on this volume.
3849 * @param[in] vp pointer to volume object
3851 * @return status code
3852 * @retval 0 no salvage scheduled
3853 * @retval 1 a salvage has been scheduled with the salvageserver
3855 * @pre VOL_LOCK is held
3857 * @post if salvage request flag is set and nUsers and nWaiters are zero,
3858 * then a salvage will be requested
3860 * @note this is one of the event handlers called by VCancelReservation_r
3862 * @see VCancelReservation_r
3864 * @internal volume package internal use only.
3867 VCheckSalvage(register Volume * vp)
3870 #ifdef SALVSYNC_BUILD_CLIENT
3871 if (vp->nUsers || vp->nWaiters)
3873 if (vp->salvage.requested) {
3874 VScheduleSalvage_r(vp);
3877 #endif /* SALVSYNC_BUILD_CLIENT */
3882 * request volume salvage.
3884 * @param[out] ec computed client error code
3885 * @param[in] vp volume object pointer
3886 * @param[in] reason reason code (passed to salvageserver via SALVSYNC)
3887 * @param[in] flags see flags note below
3890 * VOL_SALVAGE_INVALIDATE_HEADER causes volume header cache entry
3891 * to be invalidated.
3893 * @pre VOL_LOCK is held.
3895 * @post volume state is changed.
3896 * for fileserver, salvage will be requested once refcount reaches zero.
3898 * @return operation status code
3899 * @retval 0 volume salvage will occur
3900 * @retval 1 volume salvage could not be scheduled
3902 * @note DAFS fileserver only
3904 * @note this call does not synchronously schedule a volume salvage. rather,
3905 * it sets volume state so that when volume refcounts reach zero, a
3906 * volume salvage will occur. by "refcounts", we mean both nUsers and
3907 * nWaiters must be zero.
3909 * @internal volume package internal use only.
3912 VRequestSalvage_r(Error * ec, Volume * vp, int reason, int flags)
3916 * for DAFS volume utilities, transition to error state
3917 * (at some point in the future, we should consider
3918 * making volser talk to salsrv)
3920 if (programType != fileServer) {
3921 VChangeState_r(vp, VOL_STATE_ERROR);
3926 if (!vp->salvage.requested) {
3927 vp->salvage.requested = 1;
3928 vp->salvage.reason = reason;
3929 vp->stats.last_salvage = FT_ApproxTime();
3930 if (flags & VOL_SALVAGE_INVALIDATE_HEADER) {
3931 /* Instead of ReleaseVolumeHeader, we do FreeVolumeHeader()
3932 so that the the next VAttachVolumeByVp_r() invocation
3933 of attach2() will pull in a cached header
3934 entry and fail, then load a fresh one from disk and attach
3937 FreeVolumeHeader(vp);
3939 if (vp->stats.salvages < SALVAGE_COUNT_MAX) {
3940 VChangeState_r(vp, VOL_STATE_SALVAGING);
3943 Log("VRequestSalvage: volume %u online salvaged too many times; forced offline.\n", vp->hashid);
3944 VChangeState_r(vp, VOL_STATE_ERROR);
3953 * update salvageserver scheduling priority for a volume.
3955 * @param[in] vp pointer to volume object
3957 * @return operation status
3959 * @retval 1 request denied, or SALVSYNC communications failure
3961 * @pre VOL_LOCK is held.
3963 * @post in-core salvage priority counter is incremented. if at least
3964 * SALVAGE_PRIO_UPDATE_INTERVAL seconds have elapsed since the
3965 * last SALVSYNC_RAISEPRIO request, we contact the salvageserver
3966 * to update its priority queue. if no salvage is scheduled,
3967 * this function is a no-op.
3969 * @note DAFS fileserver only
3971 * @note this should be called whenever a VGetVolume fails due to a
3972 * pending salvage request
3974 * @todo should set exclusive state and drop glock around salvsync call
3976 * @internal volume package internal use only.
3979 VUpdateSalvagePriority_r(Volume * vp)
3984 #ifdef SALVSYNC_BUILD_CLIENT
3986 now = FT_ApproxTime();
3988 /* update the salvageserver priority queue occasionally so that
3989 * frequently requested volumes get moved to the head of the queue
3991 if ((vp->salvage.scheduled) &&
3992 (vp->stats.last_salvage_req < (now-SALVAGE_PRIO_UPDATE_INTERVAL))) {
3993 code = SALVSYNC_SalvageVolume(vp->hashid,
3994 VPartitionPath(vp->partition),
3999 vp->stats.last_salvage_req = now;
4000 if (code != SYNC_OK) {
4004 #endif /* SALVSYNC_BUILD_CLIENT */
4010 * schedule a salvage with the salvage server.
4012 * @param[in] vp pointer to volume object
4014 * @return operation status
4015 * @retval 0 salvage scheduled successfully
4016 * @retval 1 salvage not scheduled, or SALVSYNC com error
4019 * @arg VOL_LOCK is held.
4020 * @arg nUsers and nWaiters should be zero.
4022 * @post salvageserver is sent a salvage request
4024 * @note DAFS fileserver only
4026 * @internal volume package internal use only.
4029 VScheduleSalvage_r(Volume * vp)
4032 #ifdef SALVSYNC_BUILD_CLIENT
4033 VolState state_save;
4034 VThreadOptions_t * thread_opts;
4037 if (vp->nWaiters || vp->nUsers) {
4041 /* prevent endless salvage,attach,salvage,attach,... loops */
4042 if (vp->stats.salvages >= SALVAGE_COUNT_MAX)
4046 * don't perform salvsync ops on certain threads
4048 thread_opts = pthread_getspecific(VThread_key);
4049 if (thread_opts == NULL) {
4050 thread_opts = &VThread_defaults;
4052 if (thread_opts->disallow_salvsync) {
4057 * XXX the scheduling process should really be done asynchronously
4058 * to avoid fssync deadlocks
4060 if (!vp->salvage.scheduled) {
4061 /* if we haven't previously scheduled a salvage, do so now
4063 * set the volume to an exclusive state and drop the lock
4064 * around the SALVSYNC call
4066 * note that we do NOT acquire a reservation here -- doing so
4067 * could result in unbounded recursion
4069 strlcpy(partName, VPartitionPath(vp->partition), sizeof(partName));
4070 state_save = VChangeState_r(vp, VOL_STATE_SALVSYNC_REQ);
4073 /* can't use V_id() since there's no guarantee
4074 * we have the disk data header at this point */
4075 code = SALVSYNC_SalvageVolume(vp->hashid,
4082 VChangeState_r(vp, state_save);
4084 if (code == SYNC_OK) {
4085 vp->salvage.scheduled = 1;
4086 vp->stats.salvages++;
4087 vp->stats.last_salvage_req = FT_ApproxTime();
4088 IncUInt64(&VStats.salvages);
4092 case SYNC_BAD_COMMAND:
4093 case SYNC_COM_ERROR:
4096 Log("VScheduleSalvage_r: SALVSYNC request denied\n");
4099 Log("VScheduleSalvage_r: SALVSYNC unknown protocol error\n");
4104 #endif /* SALVSYNC_BUILD_CLIENT */
4109 * ask salvageserver to cancel a scheduled salvage operation.
4111 * @param[in] vp pointer to volume object
4112 * @param[in] reason SALVSYNC protocol reason code
4114 * @return operation status
4116 * @retval 1 request failed
4118 * @pre VOL_LOCK is held.
4120 * @post salvageserver is sent a request to cancel the volume salvage.
4121 * volume is transitioned to a hard error state.
4123 * @internal volume package internal use only.
4126 VCancelSalvage_r(Volume * vp, int reason)
4130 #ifdef SALVSYNC_BUILD_CLIENT
4131 if (vp->salvage.scheduled) {
4132 VChangeState_r(vp, VOL_STATE_SALVSYNC_REQ);
4135 /* can't use V_id() since there's no guarantee
4136 * we have the disk data header at this point */
4137 code = SALVSYNC_SalvageVolume(vp->hashid,
4138 VPartitionPath(vp->partition),
4145 VChangeState_r(vp, VOL_STATE_ERROR);
4147 if (code == SYNC_OK) {
4148 vp->salvage.scheduled = 0;
4149 vp->salvage.requested = 0;
4154 #endif /* SALVSYNC_BUILD_CLIENT */
4159 #ifdef SALVSYNC_BUILD_CLIENT
4161 * connect to the salvageserver SYNC service.
4163 * @return operation status
4167 * @post connection to salvageserver SYNC service established
4169 * @see VConnectSALV_r
4170 * @see VDisconnectSALV
4171 * @see VReconnectSALV
4178 retVal = VConnectSALV_r();
4184 * connect to the salvageserver SYNC service.
4186 * @return operation status
4190 * @pre VOL_LOCK is held.
4192 * @post connection to salvageserver SYNC service established
4195 * @see VDisconnectSALV_r
4196 * @see VReconnectSALV_r
4197 * @see SALVSYNC_clientInit
4199 * @internal volume package internal use only.
4202 VConnectSALV_r(void)
4204 return SALVSYNC_clientInit();
4208 * disconnect from the salvageserver SYNC service.
4210 * @return operation status
4213 * @pre client should have a live connection to the salvageserver
4215 * @post connection to salvageserver SYNC service destroyed
4217 * @see VDisconnectSALV_r
4219 * @see VReconnectSALV
4222 VDisconnectSALV(void)
4226 VDisconnectSALV_r();
4232 * disconnect from the salvageserver SYNC service.
4234 * @return operation status
4238 * @arg VOL_LOCK is held.
4239 * @arg client should have a live connection to the salvageserver.
4241 * @post connection to salvageserver SYNC service destroyed
4243 * @see VDisconnectSALV
4244 * @see VConnectSALV_r
4245 * @see VReconnectSALV_r
4246 * @see SALVSYNC_clientFinis
4248 * @internal volume package internal use only.
4251 VDisconnectSALV_r(void)
4253 return SALVSYNC_clientFinis();
4257 * disconnect and then re-connect to the salvageserver SYNC service.
4259 * @return operation status
4263 * @pre client should have a live connection to the salvageserver
4265 * @post old connection is dropped, and a new one is established
4268 * @see VDisconnectSALV
4269 * @see VReconnectSALV_r
4272 VReconnectSALV(void)
4276 retVal = VReconnectSALV_r();
4282 * disconnect and then re-connect to the salvageserver SYNC service.
4284 * @return operation status
4289 * @arg VOL_LOCK is held.
4290 * @arg client should have a live connection to the salvageserver.
4292 * @post old connection is dropped, and a new one is established
4294 * @see VConnectSALV_r
4295 * @see VDisconnectSALV
4296 * @see VReconnectSALV
4297 * @see SALVSYNC_clientReconnect
4299 * @internal volume package internal use only.
4302 VReconnectSALV_r(void)
4304 return SALVSYNC_clientReconnect();
4306 #endif /* SALVSYNC_BUILD_CLIENT */
4307 #endif /* AFS_DEMAND_ATTACH_FS */
4310 /***************************************************/
4311 /* FSSYNC routines */
4312 /***************************************************/
4314 /* This must be called by any volume utility which needs to run while the
4315 file server is also running. This is separated from VInitVolumePackage so
4316 that a utility can fork--and each of the children can independently
4317 initialize communication with the file server */
4318 #ifdef FSSYNC_BUILD_CLIENT
4320 * connect to the fileserver SYNC service.
4322 * @return operation status
4327 * @arg VInit must equal 2.
4328 * @arg Program Type must not be fileserver or salvager.
4330 * @post connection to fileserver SYNC service established
4333 * @see VDisconnectFS
4334 * @see VChildProcReconnectFS
4341 retVal = VConnectFS_r();
4347 * connect to the fileserver SYNC service.
4349 * @return operation status
4354 * @arg VInit must equal 2.
4355 * @arg Program Type must not be fileserver or salvager.
4356 * @arg VOL_LOCK is held.
4358 * @post connection to fileserver SYNC service established
4361 * @see VDisconnectFS_r
4362 * @see VChildProcReconnectFS_r
4364 * @internal volume package internal use only.
4370 assert((VInit == 2) &&
4371 (programType != fileServer) &&
4372 (programType != salvager));
4373 rc = FSYNC_clientInit();
4380 * disconnect from the fileserver SYNC service.
4383 * @arg client should have a live connection to the fileserver.
4384 * @arg VOL_LOCK is held.
4385 * @arg Program Type must not be fileserver or salvager.
4387 * @post connection to fileserver SYNC service destroyed
4389 * @see VDisconnectFS
4391 * @see VChildProcReconnectFS_r
4393 * @internal volume package internal use only.
4396 VDisconnectFS_r(void)
4398 assert((programType != fileServer) &&
4399 (programType != salvager));
4400 FSYNC_clientFinis();
4405 * disconnect from the fileserver SYNC service.
4408 * @arg client should have a live connection to the fileserver.
4409 * @arg Program Type must not be fileserver or salvager.
4411 * @post connection to fileserver SYNC service destroyed
4413 * @see VDisconnectFS_r
4415 * @see VChildProcReconnectFS
4426 * connect to the fileserver SYNC service from a child process following a fork.
4428 * @return operation status
4433 * @arg VOL_LOCK is held.
4434 * @arg current FSYNC handle is shared with a parent process
4436 * @post current FSYNC handle is discarded and a new connection to the
4437 * fileserver SYNC service is established
4439 * @see VChildProcReconnectFS
4441 * @see VDisconnectFS_r
4443 * @internal volume package internal use only.
4446 VChildProcReconnectFS_r(void)
4448 return FSYNC_clientChildProcReconnect();
4452 * connect to the fileserver SYNC service from a child process following a fork.
4454 * @return operation status
4458 * @pre current FSYNC handle is shared with a parent process
4460 * @post current FSYNC handle is discarded and a new connection to the
4461 * fileserver SYNC service is established
4463 * @see VChildProcReconnectFS_r
4465 * @see VDisconnectFS
4468 VChildProcReconnectFS(void)
4472 ret = VChildProcReconnectFS_r();
4476 #endif /* FSSYNC_BUILD_CLIENT */
4479 /***************************************************/
4480 /* volume bitmap routines */
4481 /***************************************************/
4484 * For demand attach fs, flags parameter controls
4485 * locking behavior. If (flags & VOL_ALLOC_BITMAP_WAIT)
4486 * is set, then this function will create a reservation
4487 * and block on any other exclusive operations. Otherwise,
4488 * this function assumes the caller already has exclusive
4489 * access to vp, and we just change the volume state.
4492 VAllocBitmapEntry_r(Error * ec, Volume * vp,
4493 struct vnodeIndex *index, int flags)
4496 register byte *bp, *ep;
4497 #ifdef AFS_DEMAND_ATTACH_FS
4498 VolState state_save;
4499 #endif /* AFS_DEMAND_ATTACH_FS */
4503 /* This test is probably redundant */
4504 if (!VolumeWriteable(vp)) {
4505 *ec = (bit32) VREADONLY;
4509 #ifdef AFS_DEMAND_ATTACH_FS
4510 if (flags & VOL_ALLOC_BITMAP_WAIT) {
4511 VCreateReservation_r(vp);
4512 VWaitExclusiveState_r(vp);
4514 state_save = VChangeState_r(vp, VOL_STATE_GET_BITMAP);
4515 #endif /* AFS_DEMAND_ATTACH_FS */
4518 if ((programType == fileServer) && !index->bitmap) {
4520 #ifndef AFS_DEMAND_ATTACH_FS
4521 /* demand attach fs uses the volume state to avoid races.
4522 * specialStatus field is not used at all */
4524 if (vp->specialStatus == VBUSY) {
4525 if (vp->goingOffline) { /* vos dump waiting for the volume to
4526 * go offline. We probably come here
4527 * from AddNewReadableResidency */
4530 while (vp->specialStatus == VBUSY) {
4531 #ifdef AFS_PTHREAD_ENV
4535 #else /* !AFS_PTHREAD_ENV */
4537 #endif /* !AFS_PTHREAD_ENV */
4541 #endif /* !AFS_DEMAND_ATTACH_FS */
4543 if (!index->bitmap) {
4544 #ifndef AFS_DEMAND_ATTACH_FS
4545 vp->specialStatus = VBUSY; /* Stop anyone else from using it. */
4546 #endif /* AFS_DEMAND_ATTACH_FS */
4547 for (i = 0; i < nVNODECLASSES; i++) {
4548 VGetBitmap_r(ec, vp, i);
4550 #ifdef AFS_DEMAND_ATTACH_FS
4551 VRequestSalvage_r(ec, vp, SALVSYNC_ERROR, VOL_SALVAGE_INVALIDATE_HEADER);
4552 #else /* AFS_DEMAND_ATTACH_FS */
4553 DeleteVolumeFromHashTable(vp);
4554 vp->shuttingDown = 1; /* Let who has it free it. */
4555 vp->specialStatus = 0;
4556 #endif /* AFS_DEMAND_ATTACH_FS */
4561 #ifndef AFS_DEMAND_ATTACH_FS
4563 vp->specialStatus = 0; /* Allow others to have access. */
4564 #endif /* AFS_DEMAND_ATTACH_FS */
4567 #endif /* BITMAP_LATER */
4569 #ifdef AFS_DEMAND_ATTACH_FS
4571 #endif /* AFS_DEMAND_ATTACH_FS */
4572 bp = index->bitmap + index->bitmapOffset;
4573 ep = index->bitmap + index->bitmapSize;
4575 if ((*(bit32 *) bp) != (bit32) 0xffffffff) {
4577 index->bitmapOffset = (afs_uint32) (bp - index->bitmap);
4580 o = ffs(~*bp) - 1; /* ffs is documented in BSTRING(3) */
4582 ret = (VnodeId) ((bp - index->bitmap) * 8 + o);
4583 #ifdef AFS_DEMAND_ATTACH_FS
4585 #endif /* AFS_DEMAND_ATTACH_FS */
4588 bp += sizeof(bit32) /* i.e. 4 */ ;
4590 /* No bit map entry--must grow bitmap */
4592 realloc(index->bitmap, index->bitmapSize + VOLUME_BITMAP_GROWSIZE);
4595 bp += index->bitmapSize;
4596 memset(bp, 0, VOLUME_BITMAP_GROWSIZE);
4597 index->bitmapOffset = index->bitmapSize;
4598 index->bitmapSize += VOLUME_BITMAP_GROWSIZE;
4600 ret = index->bitmapOffset * 8;
4601 #ifdef AFS_DEMAND_ATTACH_FS
4603 #endif /* AFS_DEMAND_ATTACH_FS */
4606 #ifdef AFS_DEMAND_ATTACH_FS
4607 VChangeState_r(vp, state_save);
4608 if (flags & VOL_ALLOC_BITMAP_WAIT) {
4609 VCancelReservation_r(vp);
4611 #endif /* AFS_DEMAND_ATTACH_FS */
4616 VAllocBitmapEntry(Error * ec, Volume * vp, register struct vnodeIndex * index)
4620 retVal = VAllocBitmapEntry_r(ec, vp, index, VOL_ALLOC_BITMAP_WAIT);
4626 VFreeBitMapEntry_r(Error * ec, register struct vnodeIndex *index,
4629 unsigned int offset;
4635 #endif /* BITMAP_LATER */
4636 offset = bitNumber >> 3;
4637 if (offset >= index->bitmapSize) {
4641 if (offset < index->bitmapOffset)
4642 index->bitmapOffset = offset & ~3; /* Truncate to nearest bit32 */
4643 *(index->bitmap + offset) &= ~(1 << (bitNumber & 0x7));
4647 VFreeBitMapEntry(Error * ec, register struct vnodeIndex *index,
4651 VFreeBitMapEntry_r(ec, index, bitNumber);
4655 /* this function will drop the glock internally.
4656 * for old pthread fileservers, this is safe thanks to vbusy.
4658 * for demand attach fs, caller must have already called
4659 * VCreateReservation_r and VWaitExclusiveState_r */
4661 VGetBitmap_r(Error * ec, Volume * vp, VnodeClass class)
4663 StreamHandle_t *file;
4666 struct VnodeClassInfo *vcp = &VnodeClassInfo[class];
4667 struct vnodeIndex *vip = &vp->vnodeIndex[class];
4668 struct VnodeDiskObject *vnode;
4669 unsigned int unique = 0;
4673 #endif /* BITMAP_LATER */
4674 #ifdef AFS_DEMAND_ATTACH_FS
4675 VolState state_save;
4676 #endif /* AFS_DEMAND_ATTACH_FS */
4680 #ifdef AFS_DEMAND_ATTACH_FS
4681 state_save = VChangeState_r(vp, VOL_STATE_GET_BITMAP);
4682 #endif /* AFS_DEMAND_ATTACH_FS */
4685 fdP = IH_OPEN(vip->handle);
4686 assert(fdP != NULL);
4687 file = FDH_FDOPEN(fdP, "r");
4688 assert(file != NULL);
4689 vnode = (VnodeDiskObject *) malloc(vcp->diskSize);
4690 assert(vnode != NULL);
4691 size = OS_SIZE(fdP->fd_fd);
4693 nVnodes = (size <= vcp->diskSize ? 0 : size - vcp->diskSize)
4695 vip->bitmapSize = ((nVnodes / 8) + 10) / 4 * 4; /* The 10 is a little extra so
4696 * a few files can be created in this volume,
4697 * the whole thing is rounded up to nearest 4
4698 * bytes, because the bit map allocator likes
4701 BitMap = (byte *) calloc(1, vip->bitmapSize);
4702 assert(BitMap != NULL);
4703 #else /* BITMAP_LATER */
4704 vip->bitmap = (byte *) calloc(1, vip->bitmapSize);
4705 assert(vip->bitmap != NULL);
4706 vip->bitmapOffset = 0;
4707 #endif /* BITMAP_LATER */
4708 if (STREAM_SEEK(file, vcp->diskSize, 0) != -1) {
4710 for (bitNumber = 0; bitNumber < nVnodes + 100; bitNumber++) {
4711 if (STREAM_READ(vnode, vcp->diskSize, 1, file) != 1)
4713 if (vnode->type != vNull) {
4714 if (vnode->vnodeMagic != vcp->magic) {
4715 Log("GetBitmap: addled vnode index in volume %s; volume needs salvage\n", V_name(vp));
4720 *(BitMap + (bitNumber >> 3)) |= (1 << (bitNumber & 0x7));
4721 #else /* BITMAP_LATER */
4722 *(vip->bitmap + (bitNumber >> 3)) |= (1 << (bitNumber & 0x7));
4723 #endif /* BITMAP_LATER */
4724 if (unique <= vnode->uniquifier)
4725 unique = vnode->uniquifier + 1;
4727 #ifndef AFS_PTHREAD_ENV
4728 if ((bitNumber & 0x00ff) == 0x0ff) { /* every 256 iterations */
4731 #endif /* !AFS_PTHREAD_ENV */
4734 if (vp->nextVnodeUnique < unique) {
4735 Log("GetBitmap: bad volume uniquifier for volume %s; volume needs salvage\n", V_name(vp));
4738 /* Paranoia, partly justified--I think fclose after fdopen
4739 * doesn't seem to close fd. In any event, the documentation
4740 * doesn't specify, so it's safer to close it twice.
4748 /* There may have been a racing condition with some other thread, both
4749 * creating the bitmaps for this volume. If the other thread was faster
4750 * the pointer to bitmap should already be filled and we can free ours.
4752 if (vip->bitmap == NULL) {
4753 vip->bitmap = BitMap;
4754 vip->bitmapOffset = 0;
4756 free((byte *) BitMap);
4757 #endif /* BITMAP_LATER */
4758 #ifdef AFS_DEMAND_ATTACH_FS
4759 VChangeState_r(vp, state_save);
4760 #endif /* AFS_DEMAND_ATTACH_FS */
4764 /***************************************************/
4765 /* Volume Path and Volume Number utility routines */
4766 /***************************************************/
4769 * find the first occurrence of a volume header file and return the path.
4771 * @param[out] ec outbound error code
4772 * @param[in] volumeId volume id to find
4773 * @param[out] partitionp pointer to disk partition path string
4774 * @param[out] namep pointer to volume header file name string
4776 * @post path to first occurrence of volume header is returned in partitionp
4777 * and namep, or ec is set accordingly.
4779 * @warning this function is NOT re-entrant -- partitionp and namep point to
4780 * static data segments
4782 * @note if a volume utility inadvertently leaves behind a stale volume header
4783 * on a vice partition, it is possible for callers to get the wrong one,
4784 * depending on the order of the disk partition linked list.
4788 VGetVolumePath(Error * ec, VolId volumeId, char **partitionp, char **namep)
4790 static char partition[VMAXPATHLEN], name[VMAXPATHLEN];
4791 char path[VMAXPATHLEN];
4793 struct DiskPartition64 *dp;
4797 (void)afs_snprintf(&name[1], (sizeof name) - 1, VFORMAT, volumeId);
4798 for (dp = DiskPartitionList; dp; dp = dp->next) {
4799 struct afs_stat status;
4800 strcpy(path, VPartitionPath(dp));
4802 if (afs_stat(path, &status) == 0) {
4803 strcpy(partition, dp->name);
4810 *partitionp = *namep = NULL;
4812 *partitionp = partition;
4818 * extract a volume number from a volume header filename string.
4820 * @param[in] name volume header filename string
4822 * @return volume number
4824 * @note the string must be of the form VFORMAT. the only permissible
4825 * deviation is a leading '/' character.
4830 VolumeNumber(char *name)
4834 return atoi(name + 1);
4838 * compute the volume header filename.
4840 * @param[in] volumeId
4842 * @return volume header filename
4844 * @post volume header filename string is constructed
4846 * @warning this function is NOT re-entrant -- the returned string is
4847 * stored in a static char array. see VolumeExternalName_r
4848 * for a re-entrant equivalent.
4850 * @see VolumeExternalName_r
4852 * @deprecated due to the above re-entrancy warning, this interface should
4853 * be considered deprecated. Please use VolumeExternalName_r
4857 VolumeExternalName(VolumeId volumeId)
4859 static char name[VMAXPATHLEN];
4860 (void)afs_snprintf(name, sizeof name, VFORMAT, volumeId);
4865 * compute the volume header filename.
4867 * @param[in] volumeId
4868 * @param[inout] name array in which to store filename
4869 * @param[in] len length of name array
4871 * @return result code from afs_snprintf
4873 * @see VolumeExternalName
4876 * @note re-entrant equivalent of VolumeExternalName
4878 * @internal volume package internal use only.
4881 VolumeExternalName_r(VolumeId volumeId, char * name, size_t len)
4883 return afs_snprintf(name, len, VFORMAT, volumeId);
4887 /***************************************************/
4888 /* Volume Usage Statistics routines */
4889 /***************************************************/
4891 #if OPENAFS_VOL_STATS
4892 #define OneDay (86400) /* 24 hours' worth of seconds */
4894 #define OneDay (24*60*60) /* 24 hours */
4895 #endif /* OPENAFS_VOL_STATS */
4897 #define Midnight(date) ((date-TimeZoneCorrection)/OneDay*OneDay+TimeZoneCorrection)
4899 /*------------------------------------------------------------------------
4900 * [export] VAdjustVolumeStatistics
4903 * If we've passed midnight, we need to update all the day use
4904 * statistics as well as zeroing the detailed volume statistics
4905 * (if we are implementing them).
4908 * vp : Pointer to the volume structure describing the lucky
4909 * volume being considered for update.
4915 * Nothing interesting.
4919 *------------------------------------------------------------------------*/
4922 VAdjustVolumeStatistics_r(register Volume * vp)
4924 unsigned int now = FT_ApproxTime();
4926 if (now - V_dayUseDate(vp) > OneDay) {
4927 register int ndays, i;
4929 ndays = (now - V_dayUseDate(vp)) / OneDay;
4930 for (i = 6; i > ndays - 1; i--)
4931 V_weekUse(vp)[i] = V_weekUse(vp)[i - ndays];
4932 for (i = 0; i < ndays - 1 && i < 7; i++)
4933 V_weekUse(vp)[i] = 0;
4935 V_weekUse(vp)[ndays - 1] = V_dayUse(vp);
4937 V_dayUseDate(vp) = Midnight(now);
4939 #if OPENAFS_VOL_STATS
4941 * All we need to do is bzero the entire VOL_STATS_BYTES of
4942 * the detailed volume statistics area.
4944 memset((char *)(V_stat_area(vp)), 0, VOL_STATS_BYTES);
4945 #endif /* OPENAFS_VOL_STATS */
4948 /*It's been more than a day of collection */
4950 * Always return happily.
4953 } /*VAdjustVolumeStatistics */
4956 VAdjustVolumeStatistics(register Volume * vp)
4960 retVal = VAdjustVolumeStatistics_r(vp);
4966 VBumpVolumeUsage_r(register Volume * vp)
4968 unsigned int now = FT_ApproxTime();
4969 if (now - V_dayUseDate(vp) > OneDay)
4970 VAdjustVolumeStatistics_r(vp);
4972 * Save the volume header image to disk after every 128 bumps to dayUse.
4974 if ((V_dayUse(vp)++ & 127) == 0) {
4976 VUpdateVolume_r(&error, vp, VOL_UPDATE_WAIT);
4981 VBumpVolumeUsage(register Volume * vp)
4984 VBumpVolumeUsage_r(vp);
4989 VSetDiskUsage_r(void)
4991 #ifndef AFS_DEMAND_ATTACH_FS
4992 static int FifteenMinuteCounter = 0;
4996 /* NOTE: Don't attempt to access the partitions list until the
4997 * initialization level indicates that all volumes are attached,
4998 * which implies that all partitions are initialized. */
4999 #ifdef AFS_PTHREAD_ENV
5001 #else /* AFS_PTHREAD_ENV */
5003 #endif /* AFS_PTHREAD_ENV */
5006 VResetDiskUsage_r();
5008 #ifndef AFS_DEMAND_ATTACH_FS
5009 if (++FifteenMinuteCounter == 3) {
5010 FifteenMinuteCounter = 0;
5013 #endif /* !AFS_DEMAND_ATTACH_FS */
5025 /***************************************************/
5026 /* Volume Update List routines */
5027 /***************************************************/
5029 /* The number of minutes that a volume hasn't been updated before the
5030 * "Dont salvage" flag in the volume header will be turned on */
5031 #define SALVAGE_INTERVAL (10*60)
5036 * volume update list functionality has been moved into the VLRU
5037 * the DONT_SALVAGE flag is now set during VLRU demotion
5040 #ifndef AFS_DEMAND_ATTACH_FS
5041 static VolumeId *UpdateList = NULL; /* Pointer to array of Volume ID's */
5042 static int nUpdatedVolumes = 0; /* Updated with entry in UpdateList, salvage after crash flag on */
5043 static int updateSize = 0; /* number of entries possible */
5044 #define UPDATE_LIST_SIZE 128 /* initial size increment (must be a power of 2!) */
5045 #endif /* !AFS_DEMAND_ATTACH_FS */
5048 VAddToVolumeUpdateList_r(Error * ec, Volume * vp)
5051 vp->updateTime = FT_ApproxTime();
5052 if (V_dontSalvage(vp) == 0)
5054 V_dontSalvage(vp) = 0;
5055 VSyncVolume_r(ec, vp, 0);
5056 #ifdef AFS_DEMAND_ATTACH_FS
5057 V_attachFlags(vp) &= ~(VOL_HDR_DONTSALV);
5058 #else /* !AFS_DEMAND_ATTACH_FS */
5061 if (UpdateList == NULL) {
5062 updateSize = UPDATE_LIST_SIZE;
5063 UpdateList = (VolumeId *) malloc(sizeof(VolumeId) * updateSize);
5065 if (nUpdatedVolumes == updateSize) {
5067 if (updateSize > 524288) {
5068 Log("warning: there is likely a bug in the volume update scanner\n");
5072 (VolumeId *) realloc(UpdateList,
5073 sizeof(VolumeId) * updateSize);
5076 assert(UpdateList != NULL);
5077 UpdateList[nUpdatedVolumes++] = V_id(vp);
5078 #endif /* !AFS_DEMAND_ATTACH_FS */
5081 #ifndef AFS_DEMAND_ATTACH_FS
5083 VScanUpdateList(void)
5085 register int i, gap;
5086 register Volume *vp;
5088 afs_uint32 now = FT_ApproxTime();
5089 /* Be careful with this code, since it works with interleaved calls to AddToVolumeUpdateList */
5090 for (i = gap = 0; i < nUpdatedVolumes; i++) {
5092 UpdateList[i - gap] = UpdateList[i];
5094 /* XXX this routine needlessly messes up the Volume LRU by
5095 * breaking the LRU temporal-locality assumptions.....
5096 * we should use a special volume header allocator here */
5097 vp = VGetVolume_r(&error, UpdateList[i - gap] = UpdateList[i]);
5100 } else if (vp->nUsers == 1 && now - vp->updateTime > SALVAGE_INTERVAL) {
5101 V_dontSalvage(vp) = DONT_SALVAGE;
5102 VUpdateVolume_r(&error, vp, 0); /* No need to fsync--not critical */
5110 #ifndef AFS_PTHREAD_ENV
5112 #endif /* !AFS_PTHREAD_ENV */
5114 nUpdatedVolumes -= gap;
5116 #endif /* !AFS_DEMAND_ATTACH_FS */
5119 /***************************************************/
5120 /* Volume LRU routines */
5121 /***************************************************/
5126 * with demand attach fs, we attempt to soft detach(1)
5127 * volumes which have not been accessed in a long time
5128 * in order to speed up fileserver shutdown
5130 * (1) by soft detach we mean a process very similar
5131 * to VOffline, except the final state of the
5132 * Volume will be VOL_STATE_PREATTACHED, instead
5133 * of the usual VOL_STATE_UNATTACHED
5135 #ifdef AFS_DEMAND_ATTACH_FS
5137 /* implementation is reminiscent of a generational GC
5139 * queue 0 is newly attached volumes. this queue is
5140 * sorted by attach timestamp
5142 * queue 1 is volumes that have been around a bit
5143 * longer than queue 0. this queue is sorted by
5146 * queue 2 is volumes tha have been around the longest.
5147 * this queue is unsorted
5149 * queue 3 is volumes that have been marked as
5150 * candidates for soft detachment. this queue is
5153 #define VLRU_GENERATIONS 3 /**< number of generations in VLRU */
5154 #define VLRU_QUEUES 5 /**< total number of VLRU queues */
5157 * definition of a VLRU queue.
5160 volatile struct rx_queue q;
5167 * main VLRU data structure.
5170 struct VLRU_q q[VLRU_QUEUES]; /**< VLRU queues */
5173 /** time interval (in seconds) between promotion passes for
5174 * each young generation queue. */
5175 afs_uint32 promotion_interval[VLRU_GENERATIONS-1];
5177 /** time interval (in seconds) between soft detach candidate
5178 * scans for each generation queue.
5180 * scan_interval[VLRU_QUEUE_CANDIDATE] defines how frequently
5181 * we perform a soft detach pass. */
5182 afs_uint32 scan_interval[VLRU_GENERATIONS+1];
5184 /* scheduler state */
5185 int next_idx; /**< next queue to receive attention */
5186 afs_uint32 last_promotion[VLRU_GENERATIONS-1]; /**< timestamp of last promotion scan */
5187 afs_uint32 last_scan[VLRU_GENERATIONS+1]; /**< timestamp of last detach scan */
5189 int scanner_state; /**< state of scanner thread */
5190 pthread_cond_t cv; /**< state transition CV */
5193 /** global VLRU state */
5194 static struct VLRU volume_LRU;
5197 * defined states for VLRU scanner thread.
5200 VLRU_SCANNER_STATE_OFFLINE = 0, /**< vlru scanner thread is offline */
5201 VLRU_SCANNER_STATE_ONLINE = 1, /**< vlru scanner thread is online */
5202 VLRU_SCANNER_STATE_SHUTTING_DOWN = 2, /**< vlru scanner thread is shutting down */
5203 VLRU_SCANNER_STATE_PAUSING = 3, /**< vlru scanner thread is getting ready to pause */
5204 VLRU_SCANNER_STATE_PAUSED = 4 /**< vlru scanner thread is paused */
5205 } vlru_thread_state_t;
5207 /* vlru disk data header stuff */
5208 #define VLRU_DISK_MAGIC 0x7a8b9cad /**< vlru disk entry magic number */
5209 #define VLRU_DISK_VERSION 1 /**< vlru disk entry version number */
5211 /** vlru default expiration time (for eventual fs state serialization of vlru data) */
5212 #define VLRU_DUMP_EXPIRATION_TIME (60*60*24*7) /* expire vlru data after 1 week */
5215 /** minimum volume inactivity (in seconds) before a volume becomes eligible for
5216 * soft detachment. */
5217 static afs_uint32 VLRU_offline_thresh = VLRU_DEFAULT_OFFLINE_THRESH;
5219 /** time interval (in seconds) between VLRU scanner thread soft detach passes. */
5220 static afs_uint32 VLRU_offline_interval = VLRU_DEFAULT_OFFLINE_INTERVAL;
5222 /** maximum number of volumes to soft detach in a VLRU soft detach pass. */
5223 static afs_uint32 VLRU_offline_max = VLRU_DEFAULT_OFFLINE_MAX;
5225 /** VLRU control flag. non-zero value implies VLRU subsystem is activated. */
5226 static afs_uint32 VLRU_enabled = 1;
5228 /* queue synchronization routines */
5229 static void VLRU_BeginExclusive_r(struct VLRU_q * q);
5230 static void VLRU_EndExclusive_r(struct VLRU_q * q);
5231 static void VLRU_Wait_r(struct VLRU_q * q);
5234 * set VLRU subsystem tunable parameters.
5236 * @param[in] option tunable option to modify
5237 * @param[in] val new value for tunable parameter
5239 * @pre @c VInitVolumePackage has not yet been called.
5241 * @post tunable parameter is modified
5245 * @note valid option parameters are:
5246 * @arg @c VLRU_SET_THRESH
5247 * set the period of inactivity after which
5248 * volumes are eligible for soft detachment
5249 * @arg @c VLRU_SET_INTERVAL
5250 * set the time interval between calls
5251 * to the volume LRU "garbage collector"
5252 * @arg @c VLRU_SET_MAX
5253 * set the max number of volumes to deallocate
5257 VLRU_SetOptions(int option, afs_uint32 val)
5259 if (option == VLRU_SET_THRESH) {
5260 VLRU_offline_thresh = val;
5261 } else if (option == VLRU_SET_INTERVAL) {
5262 VLRU_offline_interval = val;
5263 } else if (option == VLRU_SET_MAX) {
5264 VLRU_offline_max = val;
5265 } else if (option == VLRU_SET_ENABLED) {
5268 VLRU_ComputeConstants();
5272 * compute VLRU internal timing parameters.
5274 * @post VLRU scanner thread internal timing parameters are computed
5276 * @note computes internal timing parameters based upon user-modifiable
5277 * tunable parameters.
5281 * @internal volume package internal use only.
5284 VLRU_ComputeConstants(void)
5286 afs_uint32 factor = VLRU_offline_thresh / VLRU_offline_interval;
5288 /* compute the candidate scan interval */
5289 volume_LRU.scan_interval[VLRU_QUEUE_CANDIDATE] = VLRU_offline_interval;
5291 /* compute the promotion intervals */
5292 volume_LRU.promotion_interval[VLRU_QUEUE_NEW] = VLRU_offline_thresh * 2;
5293 volume_LRU.promotion_interval[VLRU_QUEUE_MID] = VLRU_offline_thresh * 4;
5296 /* compute the gen 0 scan interval */
5297 volume_LRU.scan_interval[VLRU_QUEUE_NEW] = VLRU_offline_thresh / 8;
5299 /* compute the gen 0 scan interval */
5300 volume_LRU.scan_interval[VLRU_QUEUE_NEW] = VLRU_offline_interval * 2;
5305 * initialize VLRU subsystem.
5307 * @pre this function has not yet been called
5309 * @post VLRU subsystem is initialized and VLRU scanner thread is starting
5313 * @internal volume package internal use only.
5319 pthread_attr_t attrs;
5322 if (!VLRU_enabled) {
5323 Log("VLRU: disabled\n");
5327 /* initialize each of the VLRU queues */
5328 for (i = 0; i < VLRU_QUEUES; i++) {
5329 queue_Init(&volume_LRU.q[i]);
5330 volume_LRU.q[i].len = 0;
5331 volume_LRU.q[i].busy = 0;
5332 assert(pthread_cond_init(&volume_LRU.q[i].cv, NULL) == 0);
5335 /* setup the timing constants */
5336 VLRU_ComputeConstants();
5338 /* XXX put inside LogLevel check? */
5339 Log("VLRU: starting scanner with the following configuration parameters:\n");
5340 Log("VLRU: offlining volumes after minimum of %d seconds of inactivity\n", VLRU_offline_thresh);
5341 Log("VLRU: running VLRU soft detach pass every %d seconds\n", VLRU_offline_interval);
5342 Log("VLRU: taking up to %d volumes offline per pass\n", VLRU_offline_max);
5343 Log("VLRU: scanning generation 0 for inactive volumes every %d seconds\n", volume_LRU.scan_interval[0]);
5344 Log("VLRU: scanning for promotion/demotion between generations 0 and 1 every %d seconds\n", volume_LRU.promotion_interval[0]);
5345 Log("VLRU: scanning for promotion/demotion between generations 1 and 2 every %d seconds\n", volume_LRU.promotion_interval[1]);
5347 /* start up the VLRU scanner */
5348 volume_LRU.scanner_state = VLRU_SCANNER_STATE_OFFLINE;
5349 if (programType == fileServer) {
5350 assert(pthread_cond_init(&volume_LRU.cv, NULL) == 0);
5351 assert(pthread_attr_init(&attrs) == 0);
5352 assert(pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED) == 0);
5353 assert(pthread_create(&tid, &attrs, &VLRU_ScannerThread, NULL) == 0);
5358 * initialize the VLRU-related fields of a newly allocated volume object.
5360 * @param[in] vp pointer to volume object
5363 * @arg @c VOL_LOCK is held.
5364 * @arg volume object is not on a VLRU queue.
5366 * @post VLRU fields are initialized to indicate that volume object is not
5367 * currently registered with the VLRU subsystem
5371 * @internal volume package interal use only.
5374 VLRU_Init_Node_r(volatile Volume * vp)
5379 assert(queue_IsNotOnQueue(&vp->vlru));
5380 vp->vlru.idx = VLRU_QUEUE_INVALID;
5384 * add a volume object to a VLRU queue.
5386 * @param[in] vp pointer to volume object
5389 * @arg @c VOL_LOCK is held.
5390 * @arg caller MUST hold a lightweight ref on @p vp.
5391 * @arg caller MUST NOT hold exclusive ownership of the VLRU queue.
5393 * @post the volume object is added to the appropriate VLRU queue
5395 * @note if @c vp->vlru.idx contains the index of a valid VLRU queue,
5396 * then the volume is added to that queue. Otherwise, the value
5397 * @c VLRU_QUEUE_NEW is stored into @c vp->vlru.idx and the
5398 * volume is added to the NEW generation queue.
5400 * @note @c VOL_LOCK may be dropped internally
5402 * @note Volume state is temporarily set to @c VOL_STATE_VLRU_ADD
5403 * during the add operation, and is restored to the previous
5404 * state prior to return.
5408 * @internal volume package internal use only.
5411 VLRU_Add_r(volatile Volume * vp)
5414 VolState state_save;
5419 if (queue_IsOnQueue(&vp->vlru))
5422 state_save = VChangeState_r(vp, VOL_STATE_VLRU_ADD);
5425 if ((idx < 0) || (idx >= VLRU_QUEUE_INVALID)) {
5426 idx = VLRU_QUEUE_NEW;
5429 VLRU_Wait_r(&volume_LRU.q[idx]);
5431 /* repeat check since VLRU_Wait_r may have dropped
5433 if (queue_IsNotOnQueue(&vp->vlru)) {
5435 queue_Prepend(&volume_LRU.q[idx], &vp->vlru);
5436 volume_LRU.q[idx].len++;
5437 V_attachFlags(vp) |= VOL_ON_VLRU;
5438 vp->stats.last_promote = FT_ApproxTime();
5441 VChangeState_r(vp, state_save);
5445 * delete a volume object from a VLRU queue.
5447 * @param[in] vp pointer to volume object
5450 * @arg @c VOL_LOCK is held.
5451 * @arg caller MUST hold a lightweight ref on @p vp.
5452 * @arg caller MUST NOT hold exclusive ownership of the VLRU queue.
5454 * @post volume object is removed from the VLRU queue
5456 * @note @c VOL_LOCK may be dropped internally
5460 * @todo We should probably set volume state to something exlcusive
5461 * (as @c VLRU_Add_r does) prior to dropping @c VOL_LOCK.
5463 * @internal volume package internal use only.
5466 VLRU_Delete_r(volatile Volume * vp)
5473 if (queue_IsNotOnQueue(&vp->vlru))
5479 if (idx == VLRU_QUEUE_INVALID)
5481 VLRU_Wait_r(&volume_LRU.q[idx]);
5482 } while (idx != vp->vlru.idx);
5484 /* now remove from the VLRU and update
5485 * the appropriate counter */
5486 queue_Remove(&vp->vlru);
5487 volume_LRU.q[idx].len--;
5488 vp->vlru.idx = VLRU_QUEUE_INVALID;
5489 V_attachFlags(vp) &= ~(VOL_ON_VLRU);
5493 * tell the VLRU subsystem that a volume was just accessed.
5495 * @param[in] vp pointer to volume object
5498 * @arg @c VOL_LOCK is held
5499 * @arg caller MUST hold a lightweight ref on @p vp
5500 * @arg caller MUST NOT hold exclusive ownership of any VLRU queue
5502 * @post volume VLRU access statistics are updated. If the volume was on
5503 * the VLRU soft detach candidate queue, it is moved to the NEW
5506 * @note @c VOL_LOCK may be dropped internally
5510 * @internal volume package internal use only.
5513 VLRU_UpdateAccess_r(volatile Volume * vp)
5515 afs_uint32 live_interval;
5516 Volume * rvp = NULL;
5521 if (queue_IsNotOnQueue(&vp->vlru))
5524 assert(V_attachFlags(vp) & VOL_ON_VLRU);
5526 /* update the access timestamp */
5527 vp->stats.last_get = FT_ApproxTime();
5530 * if the volume is on the soft detach candidate
5531 * list, we need to safely move it back to a
5532 * regular generation. this has to be done
5533 * carefully so we don't race against the scanner
5537 /* if this volume is on the soft detach candidate queue,
5538 * then grab exclusive access to the necessary queues */
5539 if (vp->vlru.idx == VLRU_QUEUE_CANDIDATE) {
5541 VCreateReservation_r(rvp);
5543 VLRU_Wait_r(&volume_LRU.q[VLRU_QUEUE_NEW]);
5544 VLRU_BeginExclusive_r(&volume_LRU.q[VLRU_QUEUE_NEW]);
5545 VLRU_Wait_r(&volume_LRU.q[VLRU_QUEUE_CANDIDATE]);
5546 VLRU_BeginExclusive_r(&volume_LRU.q[VLRU_QUEUE_CANDIDATE]);
5549 /* make sure multiple threads don't race to update */
5550 if (vp->vlru.idx == VLRU_QUEUE_CANDIDATE) {
5551 VLRU_SwitchQueues(vp, VLRU_QUEUE_NEW, 1);
5555 VLRU_EndExclusive_r(&volume_LRU.q[VLRU_QUEUE_CANDIDATE]);
5556 VLRU_EndExclusive_r(&volume_LRU.q[VLRU_QUEUE_NEW]);
5557 VCancelReservation_r(rvp);
5562 * switch a volume between two VLRU queues.
5564 * @param[in] vp pointer to volume object
5565 * @param[in] new_idx index of VLRU queue onto which the volume will be moved
5566 * @param[in] append controls whether the volume will be appended or
5567 * prepended to the queue. A nonzero value means it will
5568 * be appended; zero means it will be prepended.
5570 * @pre The new (and old, if applicable) queue(s) must either be owned
5571 * exclusively by the calling thread for asynchronous manipulation,
5572 * or the queue(s) must be quiescent and VOL_LOCK must be held.
5573 * Please see VLRU_BeginExclusive_r, VLRU_EndExclusive_r and VLRU_Wait_r
5574 * for further details of the queue asynchronous processing mechanism.
5576 * @post If the volume object was already on a VLRU queue, it is
5577 * removed from the queue. Depending on the value of the append
5578 * parameter, the volume object is either appended or prepended
5579 * to the VLRU queue referenced by the new_idx parameter.
5583 * @see VLRU_BeginExclusive_r
5584 * @see VLRU_EndExclusive_r
5587 * @internal volume package internal use only.
5590 VLRU_SwitchQueues(volatile Volume * vp, int new_idx, int append)
5592 if (queue_IsNotOnQueue(&vp->vlru))
5595 queue_Remove(&vp->vlru);
5596 volume_LRU.q[vp->vlru.idx].len--;
5598 /* put the volume back on the correct generational queue */
5600 queue_Append(&volume_LRU.q[new_idx], &vp->vlru);
5602 queue_Prepend(&volume_LRU.q[new_idx], &vp->vlru);
5605 volume_LRU.q[new_idx].len++;
5606 vp->vlru.idx = new_idx;
5610 * VLRU background thread.
5612 * The VLRU Scanner Thread is responsible for periodically scanning through
5613 * each VLRU queue looking for volumes which should be moved to another
5614 * queue, or soft detached.
5616 * @param[in] args unused thread arguments parameter
5618 * @return unused thread return value
5619 * @retval NULL always
5621 * @internal volume package internal use only.
5624 VLRU_ScannerThread(void * args)
5626 afs_uint32 now, min_delay, delay;
5627 afs_uint32 next_scan[VLRU_GENERATIONS];
5628 afs_uint32 next_promotion[VLRU_GENERATIONS];
5629 int i, min_idx, min_op, overdue, state;
5631 /* set t=0 for promotion cycle to be
5632 * fileserver startup */
5633 now = FT_ApproxTime();
5634 for (i=0; i < VLRU_GENERATIONS-1; i++) {
5635 volume_LRU.last_promotion[i] = now;
5638 /* don't start the scanner until VLRU_offline_thresh
5639 * plus a small delay for VInitVolumePackage to finish
5642 sleep(VLRU_offline_thresh + 60);
5644 /* set t=0 for scan cycle to be now */
5645 now = FT_ApproxTime();
5646 for (i=0; i < VLRU_GENERATIONS+1; i++) {
5647 volume_LRU.last_scan[i] = now;
5651 if (volume_LRU.scanner_state == VLRU_SCANNER_STATE_OFFLINE) {
5652 volume_LRU.scanner_state = VLRU_SCANNER_STATE_ONLINE;
5655 while ((state = volume_LRU.scanner_state) != VLRU_SCANNER_STATE_SHUTTING_DOWN) {
5656 /* check to see if we've been asked to pause */
5657 if (volume_LRU.scanner_state == VLRU_SCANNER_STATE_PAUSING) {
5658 volume_LRU.scanner_state = VLRU_SCANNER_STATE_PAUSED;
5659 assert(pthread_cond_broadcast(&volume_LRU.cv) == 0);
5661 VOL_CV_WAIT(&volume_LRU.cv);
5662 } while (volume_LRU.scanner_state == VLRU_SCANNER_STATE_PAUSED);
5665 /* scheduling can happen outside the glock */
5668 /* figure out what is next on the schedule */
5670 /* figure out a potential schedule for the new generation first */
5672 min_delay = volume_LRU.scan_interval[0] + volume_LRU.last_scan[0] - now;
5675 if (min_delay > volume_LRU.scan_interval[0]) {
5676 /* unsigned overflow -- we're overdue to run this scan */
5681 /* if we're not overdue for gen 0, figure out schedule for candidate gen */
5683 i = VLRU_QUEUE_CANDIDATE;
5684 delay = volume_LRU.scan_interval[i] + volume_LRU.last_scan[i] - now;
5685 if (delay < min_delay) {
5689 if (delay > volume_LRU.scan_interval[i]) {
5690 /* unsigned overflow -- we're overdue to run this scan */
5697 /* if we're still not overdue for something, figure out schedules for promotions */
5698 for (i=0; !overdue && i < VLRU_GENERATIONS-1; i++) {
5699 delay = volume_LRU.promotion_interval[i] + volume_LRU.last_promotion[i] - now;
5700 if (delay < min_delay) {
5705 if (delay > volume_LRU.promotion_interval[i]) {
5706 /* unsigned overflow -- we're overdue to run this promotion */
5715 /* sleep as needed */
5720 /* do whatever is next */
5723 VLRU_Promote_r(min_idx);
5724 VLRU_Demote_r(min_idx+1);
5726 VLRU_Scan_r(min_idx);
5728 now = FT_ApproxTime();
5731 Log("VLRU scanner asked to go offline (scanner_state=%d)\n", state);
5733 /* signal that scanner is down */
5734 volume_LRU.scanner_state = VLRU_SCANNER_STATE_OFFLINE;
5735 assert(pthread_cond_broadcast(&volume_LRU.cv) == 0);
5741 * promote volumes from one VLRU generation to the next.
5743 * This routine scans a VLRU generation looking for volumes which are
5744 * eligible to be promoted to the next generation. All volumes which
5745 * meet the eligibility requirement are promoted.
5747 * Promotion eligibility is based upon meeting both of the following
5750 * @arg The volume has been accessed since the last promotion:
5751 * @c (vp->stats.last_get >= vp->stats.last_promote)
5752 * @arg The last promotion occurred at least
5753 * @c volume_LRU.promotion_interval[idx] seconds ago
5755 * As a performance optimization, promotions are "globbed". In other
5756 * words, we promote arbitrarily large contiguous sublists of elements
5759 * @param[in] idx VLRU queue index to scan
5763 * @internal VLRU internal use only.
5766 VLRU_Promote_r(int idx)
5768 int len, chaining, promote;
5769 afs_uint32 now, thresh;
5770 struct rx_queue *qp, *nqp;
5771 Volume * vp, *start, *end;
5773 /* get exclusive access to two chains, and drop the glock */
5774 VLRU_Wait_r(&volume_LRU.q[idx]);
5775 VLRU_BeginExclusive_r(&volume_LRU.q[idx]);
5776 VLRU_Wait_r(&volume_LRU.q[idx+1]);
5777 VLRU_BeginExclusive_r(&volume_LRU.q[idx+1]);
5780 thresh = volume_LRU.promotion_interval[idx];
5781 now = FT_ApproxTime();
5784 for (queue_ScanBackwards(&volume_LRU.q[idx], qp, nqp, rx_queue)) {
5785 vp = (Volume *)((char *)qp - offsetof(Volume, vlru));
5786 promote = (((vp->stats.last_promote + thresh) <= now) &&
5787 (vp->stats.last_get >= vp->stats.last_promote));
5795 /* promote and prepend chain */
5796 queue_MoveChainAfter(&volume_LRU.q[idx+1], &start->vlru, &end->vlru);
5810 /* promote and prepend */
5811 queue_MoveChainAfter(&volume_LRU.q[idx+1], &start->vlru, &end->vlru);
5815 volume_LRU.q[idx].len -= len;
5816 volume_LRU.q[idx+1].len += len;
5819 /* release exclusive access to the two chains */
5821 volume_LRU.last_promotion[idx] = now;
5822 VLRU_EndExclusive_r(&volume_LRU.q[idx+1]);
5823 VLRU_EndExclusive_r(&volume_LRU.q[idx]);
5826 /* run the demotions */
5828 VLRU_Demote_r(int idx)
5831 int len, chaining, demote;
5832 afs_uint32 now, thresh;
5833 struct rx_queue *qp, *nqp;
5834 Volume * vp, *start, *end;
5835 Volume ** salv_flag_vec = NULL;
5836 int salv_vec_offset = 0;
5838 assert(idx == VLRU_QUEUE_MID || idx == VLRU_QUEUE_OLD);
5840 /* get exclusive access to two chains, and drop the glock */
5841 VLRU_Wait_r(&volume_LRU.q[idx-1]);
5842 VLRU_BeginExclusive_r(&volume_LRU.q[idx-1]);
5843 VLRU_Wait_r(&volume_LRU.q[idx]);
5844 VLRU_BeginExclusive_r(&volume_LRU.q[idx]);
5847 /* no big deal if this allocation fails */
5848 if (volume_LRU.q[idx].len) {
5849 salv_flag_vec = (Volume **) malloc(volume_LRU.q[idx].len * sizeof(Volume *));
5852 now = FT_ApproxTime();
5853 thresh = volume_LRU.promotion_interval[idx-1];
5856 for (queue_ScanBackwards(&volume_LRU.q[idx], qp, nqp, rx_queue)) {
5857 vp = (Volume *)((char *)qp - offsetof(Volume, vlru));
5858 demote = (((vp->stats.last_promote + thresh) <= now) &&
5859 (vp->stats.last_get < (now - thresh)));
5861 /* we now do volume update list DONT_SALVAGE flag setting during
5862 * demotion passes */
5863 if (salv_flag_vec &&
5864 !(V_attachFlags(vp) & VOL_HDR_DONTSALV) &&
5866 (vp->updateTime < (now - SALVAGE_INTERVAL)) &&
5867 (V_attachState(vp) == VOL_STATE_ATTACHED)) {
5868 salv_flag_vec[salv_vec_offset++] = vp;
5869 VCreateReservation_r(vp);
5878 /* demote and append chain */
5879 queue_MoveChainBefore(&volume_LRU.q[idx-1], &start->vlru, &end->vlru);
5893 queue_MoveChainBefore(&volume_LRU.q[idx-1], &start->vlru, &end->vlru);
5897 volume_LRU.q[idx].len -= len;
5898 volume_LRU.q[idx-1].len += len;
5901 /* release exclusive access to the two chains */
5903 VLRU_EndExclusive_r(&volume_LRU.q[idx]);
5904 VLRU_EndExclusive_r(&volume_LRU.q[idx-1]);
5906 /* now go back and set the DONT_SALVAGE flags as appropriate */
5907 if (salv_flag_vec) {
5909 for (i = 0; i < salv_vec_offset; i++) {
5910 vp = salv_flag_vec[i];
5911 if (!(V_attachFlags(vp) & VOL_HDR_DONTSALV) &&
5912 (vp->updateTime < (now - SALVAGE_INTERVAL)) &&
5913 (V_attachState(vp) == VOL_STATE_ATTACHED)) {
5916 V_attachFlags(vp) |= VOL_HDR_DONTSALV;
5917 V_dontSalvage(vp) = DONT_SALVAGE;
5918 VUpdateVolume_r(&ec, vp, 0);
5922 VCancelReservation_r(vp);
5924 free(salv_flag_vec);
5928 /* run a pass of the VLRU GC scanner */
5930 VLRU_Scan_r(int idx)
5932 afs_uint32 now, thresh;
5933 struct rx_queue *qp, *nqp;
5934 volatile Volume * vp;
5937 assert(idx == VLRU_QUEUE_NEW || idx == VLRU_QUEUE_CANDIDATE);
5939 /* gain exclusive access to the idx VLRU */
5940 VLRU_Wait_r(&volume_LRU.q[idx]);
5941 VLRU_BeginExclusive_r(&volume_LRU.q[idx]);
5943 if (idx != VLRU_QUEUE_CANDIDATE) {
5944 /* gain exclusive access to the candidate VLRU */
5945 VLRU_Wait_r(&volume_LRU.q[VLRU_QUEUE_CANDIDATE]);
5946 VLRU_BeginExclusive_r(&volume_LRU.q[VLRU_QUEUE_CANDIDATE]);
5949 now = FT_ApproxTime();
5950 thresh = now - VLRU_offline_thresh;
5952 /* perform candidate selection and soft detaching */
5953 if (idx == VLRU_QUEUE_CANDIDATE) {
5954 /* soft detach some volumes from the candidate pool */
5958 for (i=0,queue_ScanBackwards(&volume_LRU.q[idx], qp, nqp, rx_queue)) {
5959 vp = (Volume *)((char *)qp - offsetof(Volume, vlru));
5960 if (i >= VLRU_offline_max) {
5963 /* check timestamp to see if it's a candidate for soft detaching */
5964 if (vp->stats.last_get <= thresh) {
5966 if (VCheckSoftDetach(vp, thresh))
5972 /* scan for volumes to become soft detach candidates */
5973 for (i=1,queue_ScanBackwards(&volume_LRU.q[idx], qp, nqp, rx_queue),i++) {
5974 vp = (Volume *)((char *)qp - offsetof(Volume, vlru));
5976 /* check timestamp to see if it's a candidate for soft detaching */
5977 if (vp->stats.last_get <= thresh) {
5978 VCheckSoftDetachCandidate(vp, thresh);
5981 if (!(i&0x7f)) { /* lock coarsening optimization */
5989 /* relinquish exclusive access to the VLRU chains */
5993 volume_LRU.last_scan[idx] = now;
5994 if (idx != VLRU_QUEUE_CANDIDATE) {
5995 VLRU_EndExclusive_r(&volume_LRU.q[VLRU_QUEUE_CANDIDATE]);
5997 VLRU_EndExclusive_r(&volume_LRU.q[idx]);
6000 /* check whether volume is safe to soft detach
6001 * caller MUST NOT hold a ref count on vp */
6003 VCheckSoftDetach(volatile Volume * vp, afs_uint32 thresh)
6007 if (vp->nUsers || vp->nWaiters)
6010 if (vp->stats.last_get <= thresh) {
6011 ret = VSoftDetachVolume_r(vp, thresh);
6017 /* check whether volume should be made a
6018 * soft detach candidate */
6020 VCheckSoftDetachCandidate(volatile Volume * vp, afs_uint32 thresh)
6023 if (vp->nUsers || vp->nWaiters)
6028 assert(idx == VLRU_QUEUE_NEW);
6030 if (vp->stats.last_get <= thresh) {
6031 /* move to candidate pool */
6032 queue_Remove(&vp->vlru);
6033 volume_LRU.q[VLRU_QUEUE_NEW].len--;
6034 queue_Prepend(&volume_LRU.q[VLRU_QUEUE_CANDIDATE], &vp->vlru);
6035 vp->vlru.idx = VLRU_QUEUE_CANDIDATE;
6036 volume_LRU.q[VLRU_QUEUE_CANDIDATE].len++;
6044 /* begin exclusive access on VLRU */
6046 VLRU_BeginExclusive_r(struct VLRU_q * q)
6048 assert(q->busy == 0);
6052 /* end exclusive access on VLRU */
6054 VLRU_EndExclusive_r(struct VLRU_q * q)
6058 assert(pthread_cond_broadcast(&q->cv) == 0);
6061 /* wait for another thread to end exclusive access on VLRU */
6063 VLRU_Wait_r(struct VLRU_q * q)
6066 VOL_CV_WAIT(&q->cv);
6071 * volume soft detach
6073 * caller MUST NOT hold a ref count on vp */
6075 VSoftDetachVolume_r(volatile Volume * vp, afs_uint32 thresh)
6080 assert(vp->vlru.idx == VLRU_QUEUE_CANDIDATE);
6082 ts_save = vp->stats.last_get;
6083 if (ts_save > thresh)
6086 if (vp->nUsers || vp->nWaiters)
6089 if (VIsExclusiveState(V_attachState(vp))) {
6093 switch (V_attachState(vp)) {
6094 case VOL_STATE_UNATTACHED:
6095 case VOL_STATE_PREATTACHED:
6096 case VOL_STATE_ERROR:
6097 case VOL_STATE_GOING_OFFLINE:
6098 case VOL_STATE_SHUTTING_DOWN:
6099 case VOL_STATE_SALVAGING:
6100 volume_LRU.q[vp->vlru.idx].len--;
6102 /* create and cancel a reservation to
6103 * give the volume an opportunity to
6105 VCreateReservation_r(vp);
6106 queue_Remove(&vp->vlru);
6107 vp->vlru.idx = VLRU_QUEUE_INVALID;
6108 V_attachFlags(vp) &= ~(VOL_ON_VLRU);
6109 VCancelReservation_r(vp);
6113 /* hold the volume and take it offline.
6114 * no need for reservations, as VHold_r
6115 * takes care of that internally. */
6116 if (VHold_r(vp) == 0) {
6117 /* vhold drops the glock, so now we should
6118 * check to make sure we aren't racing against
6119 * other threads. if we are racing, offlining vp
6120 * would be wasteful, and block the scanner for a while
6124 (vp->shuttingDown) ||
6125 (vp->goingOffline) ||
6126 (vp->stats.last_get != ts_save)) {
6127 /* looks like we're racing someone else. bail */
6131 /* pull it off the VLRU */
6132 assert(vp->vlru.idx == VLRU_QUEUE_CANDIDATE);
6133 volume_LRU.q[VLRU_QUEUE_CANDIDATE].len--;
6134 queue_Remove(&vp->vlru);
6135 vp->vlru.idx = VLRU_QUEUE_INVALID;
6136 V_attachFlags(vp) &= ~(VOL_ON_VLRU);
6138 /* take if offline */
6139 VOffline_r(vp, "volume has been soft detached");
6141 /* invalidate the volume header cache */
6142 FreeVolumeHeader(vp);
6145 IncUInt64(&VStats.soft_detaches);
6146 vp->stats.soft_detaches++;
6148 /* put in pre-attached state so demand
6149 * attacher can work on it */
6150 VChangeState_r(vp, VOL_STATE_PREATTACHED);
6156 #endif /* AFS_DEMAND_ATTACH_FS */
6159 /***************************************************/
6160 /* Volume Header Cache routines */
6161 /***************************************************/
6164 * volume header cache.
6166 struct volume_hdr_LRU_t volume_hdr_LRU;
6169 * initialize the volume header cache.
6171 * @param[in] howMany number of header cache entries to preallocate
6173 * @pre VOL_LOCK held. Function has never been called before.
6175 * @post howMany cache entries are allocated, initialized, and added
6176 * to the LRU list. Header cache statistics are initialized.
6178 * @note only applicable to fileServer program type. Should only be
6179 * called once during volume package initialization.
6181 * @internal volume package internal use only.
6184 VInitVolumeHeaderCache(afs_uint32 howMany)
6186 register struct volHeader *hp;
6187 if (programType != fileServer)
6189 queue_Init(&volume_hdr_LRU);
6190 volume_hdr_LRU.stats.free = 0;
6191 volume_hdr_LRU.stats.used = howMany;
6192 volume_hdr_LRU.stats.attached = 0;
6193 hp = (struct volHeader *)(calloc(howMany, sizeof(struct volHeader)));
6197 /* We are using ReleaseVolumeHeader to initialize the values on the header list
6198 * to ensure they have the right values
6200 ReleaseVolumeHeader(hp++);
6204 * get a volume header and attach it to the volume object.
6206 * @param[in] vp pointer to volume object
6208 * @return cache entry status
6209 * @retval 0 volume header was newly attached; cache data is invalid
6210 * @retval 1 volume header was previously attached; cache data is valid
6212 * @pre VOL_LOCK held. For DAFS, lightweight ref must be held on volume object.
6214 * @post volume header attached to volume object. if necessary, header cache
6215 * entry on LRU is synchronized to disk. Header is removed from LRU list.
6217 * @note VOL_LOCK may be dropped
6219 * @warning this interface does not load header data from disk. it merely
6220 * attaches a header object to the volume object, and may sync the old
6221 * header cache data out to disk in the process.
6223 * @internal volume package internal use only.
6226 GetVolumeHeader(register Volume * vp)
6229 register struct volHeader *hd;
6231 static int everLogged = 0;
6233 #ifdef AFS_DEMAND_ATTACH_FS
6234 VolState vp_save, back_save;
6236 /* XXX debug 9/19/05 we've apparently got
6237 * a ref counting bug somewhere that's
6238 * breaking the nUsers == 0 => header on LRU
6240 if (vp->header && queue_IsNotOnQueue(vp->header)) {
6241 Log("nUsers == 0, but header not on LRU\n");
6246 old = (vp->header != NULL); /* old == volume already has a header */
6248 if (programType != fileServer) {
6249 /* for volume utilities, we allocate volHeaders as needed */
6251 hd = (struct volHeader *)calloc(1, sizeof(*vp->header));
6255 #ifdef AFS_DEMAND_ATTACH_FS
6256 V_attachFlags(vp) |= VOL_HDR_ATTACHED;
6260 /* for the fileserver, we keep a volume header cache */
6262 /* the header we previously dropped in the lru is
6263 * still available. pull it off the lru and return */
6266 assert(hd->back == vp);
6268 /* we need to grab a new element off the LRU */
6269 if (queue_IsNotEmpty(&volume_hdr_LRU)) {
6270 /* grab an element and pull off of LRU */
6271 hd = queue_First(&volume_hdr_LRU, volHeader);
6274 /* LRU is empty, so allocate a new volHeader
6275 * this is probably indicative of a leak, so let the user know */
6276 hd = (struct volHeader *)calloc(1, sizeof(struct volHeader));
6279 Log("****Allocated more volume headers, probably leak****\n");
6282 volume_hdr_LRU.stats.free++;
6285 /* this header used to belong to someone else.
6286 * we'll need to check if the header needs to
6287 * be sync'd out to disk */
6289 #ifdef AFS_DEMAND_ATTACH_FS
6290 /* if hd->back were in an exclusive state, then
6291 * its volHeader would not be on the LRU... */
6292 assert(!VIsExclusiveState(V_attachState(hd->back)));
6295 if (hd->diskstuff.inUse) {
6296 /* volume was in use, so we'll need to sync
6297 * its header to disk */
6299 #ifdef AFS_DEMAND_ATTACH_FS
6300 back_save = VChangeState_r(hd->back, VOL_STATE_UPDATING);
6301 vp_save = VChangeState_r(vp, VOL_STATE_HDR_ATTACHING);
6302 VCreateReservation_r(hd->back);
6306 WriteVolumeHeader_r(&error, hd->back);
6307 /* Ignore errors; catch them later */
6309 #ifdef AFS_DEMAND_ATTACH_FS
6314 hd->back->header = NULL;
6315 #ifdef AFS_DEMAND_ATTACH_FS
6316 V_attachFlags(hd->back) &= ~(VOL_HDR_ATTACHED | VOL_HDR_LOADED | VOL_HDR_IN_LRU);
6318 if (hd->diskstuff.inUse) {
6319 VChangeState_r(hd->back, back_save);
6320 VCancelReservation_r(hd->back);
6321 VChangeState_r(vp, vp_save);
6325 volume_hdr_LRU.stats.attached++;
6329 #ifdef AFS_DEMAND_ATTACH_FS
6330 V_attachFlags(vp) |= VOL_HDR_ATTACHED;
6333 volume_hdr_LRU.stats.free--;
6334 volume_hdr_LRU.stats.used++;
6336 IncUInt64(&VStats.hdr_gets);
6337 #ifdef AFS_DEMAND_ATTACH_FS
6338 IncUInt64(&vp->stats.hdr_gets);
6339 vp->stats.last_hdr_get = FT_ApproxTime();
6346 * make sure volume header is attached and contains valid cache data.
6348 * @param[out] ec outbound error code
6349 * @param[in] vp pointer to volume object
6351 * @pre VOL_LOCK held. For DAFS, lightweight ref held on vp.
6353 * @post header cache entry attached, and loaded with valid data, or
6354 * *ec is nonzero, and the header is released back into the LRU.
6356 * @internal volume package internal use only.
6359 LoadVolumeHeader(Error * ec, Volume * vp)
6361 #ifdef AFS_DEMAND_ATTACH_FS
6362 VolState state_save;
6366 if (vp->nUsers == 0 && !GetVolumeHeader(vp)) {
6367 IncUInt64(&VStats.hdr_loads);
6368 state_save = VChangeState_r(vp, VOL_STATE_HDR_LOADING);
6371 ReadHeader(ec, V_diskDataHandle(vp), (char *)&V_disk(vp),
6372 sizeof(V_disk(vp)), VOLUMEINFOMAGIC,
6374 IncUInt64(&vp->stats.hdr_loads);
6375 now = FT_ApproxTime();
6379 V_attachFlags(vp) |= VOL_HDR_LOADED;
6380 vp->stats.last_hdr_load = now;
6382 VChangeState_r(vp, state_save);
6384 #else /* AFS_DEMAND_ATTACH_FS */
6386 if (vp->nUsers == 0 && !GetVolumeHeader(vp)) {
6387 IncUInt64(&VStats.hdr_loads);
6389 ReadHeader(ec, V_diskDataHandle(vp), (char *)&V_disk(vp),
6390 sizeof(V_disk(vp)), VOLUMEINFOMAGIC,
6393 #endif /* AFS_DEMAND_ATTACH_FS */
6395 /* maintain (nUsers==0) => header in LRU invariant */
6396 FreeVolumeHeader(vp);
6401 * release a header cache entry back into the LRU list.
6403 * @param[in] hd pointer to volume header cache object
6405 * @pre VOL_LOCK held.
6407 * @post header cache object appended onto end of LRU list.
6409 * @note only applicable to fileServer program type.
6411 * @note used to place a header cache entry back into the
6412 * LRU pool without invalidating it as a cache entry.
6414 * @internal volume package internal use only.
6417 ReleaseVolumeHeader(register struct volHeader *hd)
6419 if (programType != fileServer)
6421 if (!hd || queue_IsOnQueue(hd)) /* no header, or header already released */
6423 queue_Append(&volume_hdr_LRU, hd);
6424 #ifdef AFS_DEMAND_ATTACH_FS
6426 V_attachFlags(hd->back) |= VOL_HDR_IN_LRU;
6429 volume_hdr_LRU.stats.free++;
6430 volume_hdr_LRU.stats.used--;
6434 * free/invalidate a volume header cache entry.
6436 * @param[in] vp pointer to volume object
6438 * @pre VOL_LOCK is held.
6440 * @post For fileserver, header cache entry is returned to LRU, and it is
6441 * invalidated as a cache entry. For volume utilities, the header
6442 * cache entry is freed.
6444 * @note For fileserver, this should be utilized instead of ReleaseVolumeHeader
6445 * whenever it is necessary to invalidate the header cache entry.
6447 * @see ReleaseVolumeHeader
6449 * @internal volume package internal use only.
6452 FreeVolumeHeader(register Volume * vp)
6454 register struct volHeader *hd = vp->header;
6457 if (programType == fileServer) {
6458 ReleaseVolumeHeader(hd);
6463 #ifdef AFS_DEMAND_ATTACH_FS
6464 V_attachFlags(vp) &= ~(VOL_HDR_ATTACHED | VOL_HDR_IN_LRU | VOL_HDR_LOADED);
6466 volume_hdr_LRU.stats.attached--;
6471 /***************************************************/
6472 /* Volume Hash Table routines */
6473 /***************************************************/
6476 * set size of volume object hash table.
6478 * @param[in] logsize log(2) of desired hash table size
6480 * @return operation status
6482 * @retval -1 failure
6484 * @pre MUST be called prior to VInitVolumePackage
6486 * @post Volume Hash Table will have 2^logsize buckets
6489 VSetVolHashSize(int logsize)
6491 /* 64 to 16384 hash buckets seems like a reasonable range */
6492 if ((logsize < 6 ) || (logsize > 14)) {
6497 VolumeHashTable.Size = 1 << logsize;
6498 VolumeHashTable.Mask = VolumeHashTable.Size - 1;
6500 /* we can't yet support runtime modification of this
6501 * parameter. we'll need a configuration rwlock to
6502 * make runtime modification feasible.... */
6509 * initialize dynamic data structures for volume hash table.
6511 * @post hash table is allocated, and fields are initialized.
6513 * @internal volume package internal use only.
6516 VInitVolumeHash(void)
6520 VolumeHashTable.Table = (VolumeHashChainHead *) calloc(VolumeHashTable.Size,
6521 sizeof(VolumeHashChainHead));
6522 assert(VolumeHashTable.Table != NULL);
6524 for (i=0; i < VolumeHashTable.Size; i++) {
6525 queue_Init(&VolumeHashTable.Table[i]);
6526 #ifdef AFS_DEMAND_ATTACH_FS
6527 assert(pthread_cond_init(&VolumeHashTable.Table[i].chain_busy_cv, NULL) == 0);
6528 #endif /* AFS_DEMAND_ATTACH_FS */
6533 * add a volume object to the hash table.
6535 * @param[in] vp pointer to volume object
6536 * @param[in] hashid hash of volume id
6538 * @pre VOL_LOCK is held. For DAFS, caller must hold a lightweight
6541 * @post volume is added to hash chain.
6543 * @internal volume package internal use only.
6545 * @note For DAFS, VOL_LOCK may be dropped in order to wait for an
6546 * asynchronous hash chain reordering to finish.
6549 AddVolumeToHashTable(register Volume * vp, int hashid)
6551 VolumeHashChainHead * head;
6553 if (queue_IsOnQueue(vp))
6556 head = &VolumeHashTable.Table[VOLUME_HASH(hashid)];
6558 #ifdef AFS_DEMAND_ATTACH_FS
6559 /* wait for the hash chain to become available */
6562 V_attachFlags(vp) |= VOL_IN_HASH;
6563 vp->chainCacheCheck = ++head->cacheCheck;
6564 #endif /* AFS_DEMAND_ATTACH_FS */
6567 vp->hashid = hashid;
6568 queue_Append(head, vp);
6569 vp->vnodeHashOffset = VolumeHashOffset_r();
6573 * delete a volume object from the hash table.
6575 * @param[in] vp pointer to volume object
6577 * @pre VOL_LOCK is held. For DAFS, caller must hold a lightweight
6580 * @post volume is removed from hash chain.
6582 * @internal volume package internal use only.
6584 * @note For DAFS, VOL_LOCK may be dropped in order to wait for an
6585 * asynchronous hash chain reordering to finish.
6588 DeleteVolumeFromHashTable(register Volume * vp)
6590 VolumeHashChainHead * head;
6592 if (!queue_IsOnQueue(vp))
6595 head = &VolumeHashTable.Table[VOLUME_HASH(vp->hashid)];
6597 #ifdef AFS_DEMAND_ATTACH_FS
6598 /* wait for the hash chain to become available */
6601 V_attachFlags(vp) &= ~(VOL_IN_HASH);
6603 #endif /* AFS_DEMAND_ATTACH_FS */
6607 /* do NOT reset hashid to zero, as the online
6608 * salvager package may need to know the volume id
6609 * after the volume is removed from the hash */
6613 * lookup a volume object in the hash table given a volume id.
6615 * @param[out] ec error code return
6616 * @param[in] volumeId volume id
6617 * @param[in] hint volume object which we believe could be the correct
6620 * @return volume object pointer
6621 * @retval NULL no such volume id is registered with the hash table.
6623 * @pre VOL_LOCK is held. For DAFS, caller must hold a lightweight
6626 * @post volume object with the given id is returned. volume object and
6627 * hash chain access statistics are updated. hash chain may have
6630 * @note For DAFS, VOL_LOCK may be dropped in order to wait for an
6631 * asynchronous hash chain reordering operation to finish, or
6632 * in order for us to perform an asynchronous chain reordering.
6634 * @note Hash chain reorderings occur when the access count for the
6635 * volume object being looked up exceeds the sum of the previous
6636 * node's (the node ahead of it in the hash chain linked list)
6637 * access count plus the constant VOLUME_HASH_REORDER_THRESHOLD.
6639 * @note For DAFS, the hint parameter allows us to short-circuit if the
6640 * cacheCheck fields match between the hash chain head and the
6641 * hint volume object.
6644 VLookupVolume_r(Error * ec, VolId volumeId, Volume * hint)
6646 register int looks = 0;
6647 Volume * vp, *np, *pp;
6648 VolumeHashChainHead * head;
6651 head = &VolumeHashTable.Table[VOLUME_HASH(volumeId)];
6653 #ifdef AFS_DEMAND_ATTACH_FS
6654 /* wait for the hash chain to become available */
6657 /* check to see if we can short circuit without walking the hash chain */
6658 if (hint && (hint->chainCacheCheck == head->cacheCheck)) {
6659 IncUInt64(&hint->stats.hash_short_circuits);
6662 #endif /* AFS_DEMAND_ATTACH_FS */
6664 /* someday we need to either do per-chain locks, RWlocks,
6665 * or both for volhash access.
6666 * (and move to a data structure with better cache locality) */
6668 /* search the chain for this volume id */
6669 for(queue_Scan(head, vp, np, Volume)) {
6671 if ((vp->hashid == volumeId)) {
6676 if (queue_IsEnd(head, vp)) {
6680 #ifdef AFS_DEMAND_ATTACH_FS
6681 /* update hash chain statistics */
6684 FillInt64(lks, 0, looks);
6685 AddUInt64(head->looks, lks, &head->looks);
6686 AddUInt64(VStats.hash_looks, lks, &VStats.hash_looks);
6687 IncUInt64(&head->gets);
6692 IncUInt64(&vp->stats.hash_lookups);
6694 /* for demand attach fileserver, we permit occasional hash chain reordering
6695 * so that frequently looked up volumes move towards the head of the chain */
6696 pp = queue_Prev(vp, Volume);
6697 if (!queue_IsEnd(head, pp)) {
6698 FillInt64(thresh, 0, VOLUME_HASH_REORDER_THRESHOLD);
6699 AddUInt64(thresh, pp->stats.hash_lookups, &thresh);
6700 if (GEInt64(vp->stats.hash_lookups, thresh)) {
6701 VReorderHash_r(head, pp, vp);
6705 /* update the short-circuit cache check */
6706 vp->chainCacheCheck = head->cacheCheck;
6708 #endif /* AFS_DEMAND_ATTACH_FS */
6713 #ifdef AFS_DEMAND_ATTACH_FS
6714 /* perform volume hash chain reordering.
6716 * advance a subchain beginning at vp ahead of
6717 * the adjacent subchain ending at pp */
6719 VReorderHash_r(VolumeHashChainHead * head, Volume * pp, Volume * vp)
6721 Volume *tp, *np, *lp;
6722 afs_uint64 move_thresh;
6724 /* this should never be called if the chain is already busy, so
6725 * no need to wait for other exclusive chain ops to finish */
6727 /* this is a rather heavy set of operations,
6728 * so let's set the chain busy flag and drop
6730 VHashBeginExclusive_r(head);
6733 /* scan forward in the chain from vp looking for the last element
6734 * in the chain we want to advance */
6735 FillInt64(move_thresh, 0, VOLUME_HASH_REORDER_CHAIN_THRESH);
6736 AddUInt64(move_thresh, pp->stats.hash_lookups, &move_thresh);
6737 for(queue_ScanFrom(head, vp, tp, np, Volume)) {
6738 if (LTInt64(tp->stats.hash_lookups, move_thresh)) {
6742 lp = queue_Prev(tp, Volume);
6744 /* scan backwards from pp to determine where to splice and
6745 * insert the subchain we're advancing */
6746 for(queue_ScanBackwardsFrom(head, pp, tp, np, Volume)) {
6747 if (GTInt64(tp->stats.hash_lookups, move_thresh)) {
6751 tp = queue_Next(tp, Volume);
6753 /* rebalance chain(vp,...,lp) ahead of chain(tp,...,pp) */
6754 queue_MoveChainBefore(tp,vp,lp);
6757 IncUInt64(&VStats.hash_reorders);
6759 IncUInt64(&head->reorders);
6761 /* wake up any threads waiting for the hash chain */
6762 VHashEndExclusive_r(head);
6766 /* demand-attach fs volume hash
6767 * asynchronous exclusive operations */
6770 * begin an asynchronous exclusive operation on a volume hash chain.
6772 * @param[in] head pointer to volume hash chain head object
6774 * @pre VOL_LOCK held. hash chain is quiescent.
6776 * @post hash chain marked busy.
6778 * @note this interface is used in conjunction with VHashEndExclusive_r and
6779 * VHashWait_r to perform asynchronous (wrt VOL_LOCK) operations on a
6780 * volume hash chain. Its main use case is hash chain reordering, which
6781 * has the potential to be a highly latent operation.
6783 * @see VHashEndExclusive_r
6788 * @internal volume package internal use only.
6791 VHashBeginExclusive_r(VolumeHashChainHead * head)
6793 assert(head->busy == 0);
6798 * relinquish exclusive ownership of a volume hash chain.
6800 * @param[in] head pointer to volume hash chain head object
6802 * @pre VOL_LOCK held. thread owns the hash chain exclusively.
6804 * @post hash chain is marked quiescent. threads awaiting use of
6805 * chain are awakened.
6807 * @see VHashBeginExclusive_r
6812 * @internal volume package internal use only.
6815 VHashEndExclusive_r(VolumeHashChainHead * head)
6819 assert(pthread_cond_broadcast(&head->chain_busy_cv) == 0);
6823 * wait for all asynchronous operations on a hash chain to complete.
6825 * @param[in] head pointer to volume hash chain head object
6827 * @pre VOL_LOCK held.
6829 * @post hash chain object is quiescent.
6831 * @see VHashBeginExclusive_r
6832 * @see VHashEndExclusive_r
6836 * @note This interface should be called before any attempt to
6837 * traverse the hash chain. It is permissible for a thread
6838 * to gain exclusive access to the chain, and then perform
6839 * latent operations on the chain asynchronously wrt the
6842 * @warning if waiting is necessary, VOL_LOCK is dropped
6844 * @internal volume package internal use only.
6847 VHashWait_r(VolumeHashChainHead * head)
6849 while (head->busy) {
6850 VOL_CV_WAIT(&head->chain_busy_cv);
6853 #endif /* AFS_DEMAND_ATTACH_FS */
6856 /***************************************************/
6857 /* Volume by Partition List routines */
6858 /***************************************************/
6861 * demand attach fileserver adds a
6862 * linked list of volumes to each
6863 * partition object, thus allowing
6864 * for quick enumeration of all
6865 * volumes on a partition
6868 #ifdef AFS_DEMAND_ATTACH_FS
6870 * add a volume to its disk partition VByPList.
6872 * @param[in] vp pointer to volume object
6874 * @pre either the disk partition VByPList is owned exclusively
6875 * by the calling thread, or the list is quiescent and
6878 * @post volume is added to disk partition VByPList
6882 * @warning it is the caller's responsibility to ensure list
6885 * @see VVByPListWait_r
6886 * @see VVByPListBeginExclusive_r
6887 * @see VVByPListEndExclusive_r
6889 * @internal volume package internal use only.
6892 AddVolumeToVByPList_r(Volume * vp)
6894 if (queue_IsNotOnQueue(&vp->vol_list)) {
6895 queue_Append(&vp->partition->vol_list, &vp->vol_list);
6896 V_attachFlags(vp) |= VOL_ON_VBYP_LIST;
6897 vp->partition->vol_list.len++;
6902 * delete a volume from its disk partition VByPList.
6904 * @param[in] vp pointer to volume object
6906 * @pre either the disk partition VByPList is owned exclusively
6907 * by the calling thread, or the list is quiescent and
6910 * @post volume is removed from the disk partition VByPList
6914 * @warning it is the caller's responsibility to ensure list
6917 * @see VVByPListWait_r
6918 * @see VVByPListBeginExclusive_r
6919 * @see VVByPListEndExclusive_r
6921 * @internal volume package internal use only.
6924 DeleteVolumeFromVByPList_r(Volume * vp)
6926 if (queue_IsOnQueue(&vp->vol_list)) {
6927 queue_Remove(&vp->vol_list);
6928 V_attachFlags(vp) &= ~(VOL_ON_VBYP_LIST);
6929 vp->partition->vol_list.len--;
6934 * begin an asynchronous exclusive operation on a VByPList.
6936 * @param[in] dp pointer to disk partition object
6938 * @pre VOL_LOCK held. VByPList is quiescent.
6940 * @post VByPList marked busy.
6942 * @note this interface is used in conjunction with VVByPListEndExclusive_r and
6943 * VVByPListWait_r to perform asynchronous (wrt VOL_LOCK) operations on a
6946 * @see VVByPListEndExclusive_r
6947 * @see VVByPListWait_r
6951 * @internal volume package internal use only.
6953 /* take exclusive control over the list */
6955 VVByPListBeginExclusive_r(struct DiskPartition64 * dp)
6957 assert(dp->vol_list.busy == 0);
6958 dp->vol_list.busy = 1;
6962 * relinquish exclusive ownership of a VByPList.
6964 * @param[in] dp pointer to disk partition object
6966 * @pre VOL_LOCK held. thread owns the VByPList exclusively.
6968 * @post VByPList is marked quiescent. threads awaiting use of
6969 * the list are awakened.
6971 * @see VVByPListBeginExclusive_r
6972 * @see VVByPListWait_r
6976 * @internal volume package internal use only.
6979 VVByPListEndExclusive_r(struct DiskPartition64 * dp)
6981 assert(dp->vol_list.busy);
6982 dp->vol_list.busy = 0;
6983 assert(pthread_cond_broadcast(&dp->vol_list.cv) == 0);
6987 * wait for all asynchronous operations on a VByPList to complete.
6989 * @param[in] dp pointer to disk partition object
6991 * @pre VOL_LOCK is held.
6993 * @post disk partition's VByP list is quiescent
6997 * @note This interface should be called before any attempt to
6998 * traverse the VByPList. It is permissible for a thread
6999 * to gain exclusive access to the list, and then perform
7000 * latent operations on the list asynchronously wrt the
7003 * @warning if waiting is necessary, VOL_LOCK is dropped
7005 * @see VVByPListEndExclusive_r
7006 * @see VVByPListBeginExclusive_r
7008 * @internal volume package internal use only.
7011 VVByPListWait_r(struct DiskPartition64 * dp)
7013 while (dp->vol_list.busy) {
7014 VOL_CV_WAIT(&dp->vol_list.cv);
7017 #endif /* AFS_DEMAND_ATTACH_FS */
7019 /***************************************************/
7020 /* Volume Cache Statistics routines */
7021 /***************************************************/
7024 VPrintCacheStats_r(void)
7026 afs_uint32 get_hi, get_lo, load_hi, load_lo;
7027 register struct VnodeClassInfo *vcp;
7028 vcp = &VnodeClassInfo[vLarge];
7029 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);
7030 vcp = &VnodeClassInfo[vSmall];
7031 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);
7032 SplitInt64(VStats.hdr_gets, get_hi, get_lo);
7033 SplitInt64(VStats.hdr_loads, load_hi, load_lo);
7034 Log("Volume header cache, %d entries, %d gets, %d replacements\n",
7035 VStats.hdr_cache_size, get_lo, load_lo);
7039 VPrintCacheStats(void)
7042 VPrintCacheStats_r();
7046 #ifdef AFS_DEMAND_ATTACH_FS
7048 UInt64ToDouble(afs_uint64 * x)
7050 static double c32 = 4.0 * 1.073741824 * 1000000000.0;
7052 SplitInt64(*x, h, l);
7053 return (((double)h) * c32) + ((double) l);
7057 DoubleToPrintable(double x, char * buf, int len)
7059 static double billion = 1000000000.0;
7062 y[0] = (afs_uint32) (x / (billion * billion));
7063 y[1] = (afs_uint32) ((x - (((double)y[0]) * billion * billion)) / billion);
7064 y[2] = (afs_uint32) (x - ((((double)y[0]) * billion * billion) + (((double)y[1]) * billion)));
7067 snprintf(buf, len, "%d%09d%09d", y[0], y[1], y[2]);
7069 snprintf(buf, len, "%d%09d", y[1], y[2]);
7071 snprintf(buf, len, "%d", y[2]);
7077 struct VLRUExtStatsEntry {
7081 struct VLRUExtStats {
7087 } queue_info[VLRU_QUEUE_INVALID];
7088 struct VLRUExtStatsEntry * vec;
7092 * add a 256-entry fudge factor onto the vector in case state changes
7093 * out from under us.
7095 #define VLRU_EXT_STATS_VEC_LEN_FUDGE 256
7098 * collect extended statistics for the VLRU subsystem.
7100 * @param[out] stats pointer to stats structure to be populated
7101 * @param[in] nvols number of volumes currently known to exist
7103 * @pre VOL_LOCK held
7105 * @post stats->vec allocated and populated
7107 * @return operation status
7112 VVLRUExtStats_r(struct VLRUExtStats * stats, afs_uint32 nvols)
7114 afs_uint32 cur, idx, len;
7115 struct rx_queue * qp, * nqp;
7117 struct VLRUExtStatsEntry * vec;
7119 len = nvols + VLRU_EXT_STATS_VEC_LEN_FUDGE;
7120 vec = stats->vec = calloc(len,
7121 sizeof(struct VLRUExtStatsEntry));
7127 for (idx = VLRU_QUEUE_NEW; idx < VLRU_QUEUE_INVALID; idx++) {
7128 VLRU_Wait_r(&volume_LRU.q[idx]);
7129 VLRU_BeginExclusive_r(&volume_LRU.q[idx]);
7132 stats->queue_info[idx].start = cur;
7134 for (queue_Scan(&volume_LRU.q[idx], qp, nqp, rx_queue)) {
7136 /* out of space in vec */
7139 vp = (Volume *)((char *)qp - offsetof(Volume, vlru));
7140 vec[cur].volid = vp->hashid;
7144 stats->queue_info[idx].len = cur - stats->queue_info[idx].start;
7147 VLRU_EndExclusive_r(&volume_LRU.q[idx]);
7155 #define ENUMTOSTRING(en) #en
7156 #define ENUMCASE(en) \
7158 return ENUMTOSTRING(en); \
7162 vlru_idx_to_string(int idx)
7165 ENUMCASE(VLRU_QUEUE_NEW);
7166 ENUMCASE(VLRU_QUEUE_MID);
7167 ENUMCASE(VLRU_QUEUE_OLD);
7168 ENUMCASE(VLRU_QUEUE_CANDIDATE);
7169 ENUMCASE(VLRU_QUEUE_HELD);
7170 ENUMCASE(VLRU_QUEUE_INVALID);
7172 return "**UNKNOWN**";
7177 VPrintExtendedCacheStats_r(int flags)
7180 afs_uint32 vol_sum = 0;
7187 struct stats looks, gets, reorders, len;
7188 struct stats ch_looks, ch_gets, ch_reorders;
7190 VolumeHashChainHead *head;
7192 struct VLRUExtStats vlru_stats;
7194 /* zero out stats */
7195 memset(&looks, 0, sizeof(struct stats));
7196 memset(&gets, 0, sizeof(struct stats));
7197 memset(&reorders, 0, sizeof(struct stats));
7198 memset(&len, 0, sizeof(struct stats));
7199 memset(&ch_looks, 0, sizeof(struct stats));
7200 memset(&ch_gets, 0, sizeof(struct stats));
7201 memset(&ch_reorders, 0, sizeof(struct stats));
7203 for (i = 0; i < VolumeHashTable.Size; i++) {
7204 head = &VolumeHashTable.Table[i];
7207 VHashBeginExclusive_r(head);
7210 ch_looks.sum = UInt64ToDouble(&head->looks);
7211 ch_gets.sum = UInt64ToDouble(&head->gets);
7212 ch_reorders.sum = UInt64ToDouble(&head->reorders);
7214 /* update global statistics */
7216 looks.sum += ch_looks.sum;
7217 gets.sum += ch_gets.sum;
7218 reorders.sum += ch_reorders.sum;
7219 len.sum += (double)head->len;
7220 vol_sum += head->len;
7223 len.min = (double) head->len;
7224 len.max = (double) head->len;
7225 looks.min = ch_looks.sum;
7226 looks.max = ch_looks.sum;
7227 gets.min = ch_gets.sum;
7228 gets.max = ch_gets.sum;
7229 reorders.min = ch_reorders.sum;
7230 reorders.max = ch_reorders.sum;
7232 if (((double)head->len) < len.min)
7233 len.min = (double) head->len;
7234 if (((double)head->len) > len.max)
7235 len.max = (double) head->len;
7236 if (ch_looks.sum < looks.min)
7237 looks.min = ch_looks.sum;
7238 else if (ch_looks.sum > looks.max)
7239 looks.max = ch_looks.sum;
7240 if (ch_gets.sum < gets.min)
7241 gets.min = ch_gets.sum;
7242 else if (ch_gets.sum > gets.max)
7243 gets.max = ch_gets.sum;
7244 if (ch_reorders.sum < reorders.min)
7245 reorders.min = ch_reorders.sum;
7246 else if (ch_reorders.sum > reorders.max)
7247 reorders.max = ch_reorders.sum;
7251 if ((flags & VOL_STATS_PER_CHAIN2) && queue_IsNotEmpty(head)) {
7252 /* compute detailed per-chain stats */
7253 struct stats hdr_loads, hdr_gets;
7254 double v_looks, v_loads, v_gets;
7256 /* initialize stats with data from first element in chain */
7257 vp = queue_First(head, Volume);
7258 v_looks = UInt64ToDouble(&vp->stats.hash_lookups);
7259 v_loads = UInt64ToDouble(&vp->stats.hdr_loads);
7260 v_gets = UInt64ToDouble(&vp->stats.hdr_gets);
7261 ch_gets.min = ch_gets.max = v_looks;
7262 hdr_loads.min = hdr_loads.max = v_loads;
7263 hdr_gets.min = hdr_gets.max = v_gets;
7264 hdr_loads.sum = hdr_gets.sum = 0;
7266 vp = queue_Next(vp, Volume);
7268 /* pull in stats from remaining elements in chain */
7269 for (queue_ScanFrom(head, vp, vp, np, Volume)) {
7270 v_looks = UInt64ToDouble(&vp->stats.hash_lookups);
7271 v_loads = UInt64ToDouble(&vp->stats.hdr_loads);
7272 v_gets = UInt64ToDouble(&vp->stats.hdr_gets);
7274 hdr_loads.sum += v_loads;
7275 hdr_gets.sum += v_gets;
7277 if (v_looks < ch_gets.min)
7278 ch_gets.min = v_looks;
7279 else if (v_looks > ch_gets.max)
7280 ch_gets.max = v_looks;
7282 if (v_loads < hdr_loads.min)
7283 hdr_loads.min = v_loads;
7284 else if (v_loads > hdr_loads.max)
7285 hdr_loads.max = v_loads;
7287 if (v_gets < hdr_gets.min)
7288 hdr_gets.min = v_gets;
7289 else if (v_gets > hdr_gets.max)
7290 hdr_gets.max = v_gets;
7293 /* compute per-chain averages */
7294 ch_gets.avg = ch_gets.sum / ((double)head->len);
7295 hdr_loads.avg = hdr_loads.sum / ((double)head->len);
7296 hdr_gets.avg = hdr_gets.sum / ((double)head->len);
7298 /* dump per-chain stats */
7299 Log("Volume hash chain %d : len=%d, looks=%s, reorders=%s\n",
7301 DoubleToPrintable(ch_looks.sum, pr_buf[0], sizeof(pr_buf[0])),
7302 DoubleToPrintable(ch_reorders.sum, pr_buf[1], sizeof(pr_buf[1])));
7303 Log("\tVolume gets : min=%s, max=%s, avg=%s, total=%s\n",
7304 DoubleToPrintable(ch_gets.min, pr_buf[0], sizeof(pr_buf[0])),
7305 DoubleToPrintable(ch_gets.max, pr_buf[1], sizeof(pr_buf[1])),
7306 DoubleToPrintable(ch_gets.avg, pr_buf[2], sizeof(pr_buf[2])),
7307 DoubleToPrintable(ch_gets.sum, pr_buf[3], sizeof(pr_buf[3])));
7308 Log("\tHDR gets : min=%s, max=%s, avg=%s, total=%s\n",
7309 DoubleToPrintable(hdr_gets.min, pr_buf[0], sizeof(pr_buf[0])),
7310 DoubleToPrintable(hdr_gets.max, pr_buf[1], sizeof(pr_buf[1])),
7311 DoubleToPrintable(hdr_gets.avg, pr_buf[2], sizeof(pr_buf[2])),
7312 DoubleToPrintable(hdr_gets.sum, pr_buf[3], sizeof(pr_buf[3])));
7313 Log("\tHDR loads : min=%s, max=%s, avg=%s, total=%s\n",
7314 DoubleToPrintable(hdr_loads.min, pr_buf[0], sizeof(pr_buf[0])),
7315 DoubleToPrintable(hdr_loads.max, pr_buf[1], sizeof(pr_buf[1])),
7316 DoubleToPrintable(hdr_loads.avg, pr_buf[2], sizeof(pr_buf[2])),
7317 DoubleToPrintable(hdr_loads.sum, pr_buf[3], sizeof(pr_buf[3])));
7318 } else if (flags & VOL_STATS_PER_CHAIN) {
7319 /* dump simple per-chain stats */
7320 Log("Volume hash chain %d : len=%d, looks=%s, gets=%s, reorders=%s\n",
7322 DoubleToPrintable(ch_looks.sum, pr_buf[0], sizeof(pr_buf[0])),
7323 DoubleToPrintable(ch_gets.sum, pr_buf[1], sizeof(pr_buf[1])),
7324 DoubleToPrintable(ch_reorders.sum, pr_buf[2], sizeof(pr_buf[2])));
7328 VHashEndExclusive_r(head);
7333 /* compute global averages */
7334 len.avg = len.sum / ((double)VolumeHashTable.Size);
7335 looks.avg = looks.sum / ((double)VolumeHashTable.Size);
7336 gets.avg = gets.sum / ((double)VolumeHashTable.Size);
7337 reorders.avg = reorders.sum / ((double)VolumeHashTable.Size);
7339 /* dump global stats */
7340 Log("Volume hash summary: %d buckets\n", VolumeHashTable.Size);
7341 Log(" chain length : min=%s, max=%s, avg=%s, total=%s\n",
7342 DoubleToPrintable(len.min, pr_buf[0], sizeof(pr_buf[0])),
7343 DoubleToPrintable(len.max, pr_buf[1], sizeof(pr_buf[1])),
7344 DoubleToPrintable(len.avg, pr_buf[2], sizeof(pr_buf[2])),
7345 DoubleToPrintable(len.sum, pr_buf[3], sizeof(pr_buf[3])));
7346 Log(" looks : min=%s, max=%s, avg=%s, total=%s\n",
7347 DoubleToPrintable(looks.min, pr_buf[0], sizeof(pr_buf[0])),
7348 DoubleToPrintable(looks.max, pr_buf[1], sizeof(pr_buf[1])),
7349 DoubleToPrintable(looks.avg, pr_buf[2], sizeof(pr_buf[2])),
7350 DoubleToPrintable(looks.sum, pr_buf[3], sizeof(pr_buf[3])));
7351 Log(" gets : min=%s, max=%s, avg=%s, total=%s\n",
7352 DoubleToPrintable(gets.min, pr_buf[0], sizeof(pr_buf[0])),
7353 DoubleToPrintable(gets.max, pr_buf[1], sizeof(pr_buf[1])),
7354 DoubleToPrintable(gets.avg, pr_buf[2], sizeof(pr_buf[2])),
7355 DoubleToPrintable(gets.sum, pr_buf[3], sizeof(pr_buf[3])));
7356 Log(" reorders : min=%s, max=%s, avg=%s, total=%s\n",
7357 DoubleToPrintable(reorders.min, pr_buf[0], sizeof(pr_buf[0])),
7358 DoubleToPrintable(reorders.max, pr_buf[1], sizeof(pr_buf[1])),
7359 DoubleToPrintable(reorders.avg, pr_buf[2], sizeof(pr_buf[2])),
7360 DoubleToPrintable(reorders.sum, pr_buf[3], sizeof(pr_buf[3])));
7362 /* print extended disk related statistics */
7364 struct DiskPartition64 * diskP;
7365 afs_uint32 vol_count[VOLMAXPARTS+1];
7366 byte part_exists[VOLMAXPARTS+1];
7370 memset(vol_count, 0, sizeof(vol_count));
7371 memset(part_exists, 0, sizeof(part_exists));
7375 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
7377 vol_count[id] = diskP->vol_list.len;
7378 part_exists[id] = 1;
7382 for (i = 0; i <= VOLMAXPARTS; i++) {
7383 if (part_exists[i]) {
7384 /* XXX while this is currently safe, it is a violation
7385 * of the VGetPartitionById_r interface contract. */
7386 diskP = VGetPartitionById_r(i, 0);
7388 Log("Partition %s has %d online volumes\n",
7389 VPartitionPath(diskP), diskP->vol_list.len);
7396 /* print extended VLRU statistics */
7397 if (VVLRUExtStats_r(&vlru_stats, vol_sum) == 0) {
7398 afs_uint32 idx, cur, lpos;
7402 Log("VLRU State Dump:\n\n");
7404 for (idx = VLRU_QUEUE_NEW; idx < VLRU_QUEUE_INVALID; idx++) {
7405 Log("\t%s:\n", vlru_idx_to_string(idx));
7408 for (cur = vlru_stats.queue_info[idx].start;
7409 cur < vlru_stats.queue_info[idx].len;
7411 line[lpos++] = vlru_stats.vec[cur].volid;
7413 Log("\t\t%u, %u, %u, %u, %u,\n",
7414 line[0], line[1], line[2], line[3], line[4]);
7423 Log("\t\t%u, %u, %u, %u, %u\n",
7424 line[0], line[1], line[2], line[3], line[4]);
7429 free(vlru_stats.vec);
7436 VPrintExtendedCacheStats(int flags)
7439 VPrintExtendedCacheStats_r(flags);
7442 #endif /* AFS_DEMAND_ATTACH_FS */