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>
26 #include <afs/afsint.h>
30 #include <sys/param.h>
31 #if !defined(AFS_SGI_ENV)
34 #else /* AFS_OSF_ENV */
35 #ifdef AFS_VFSINCL_ENV
38 #include <sys/fs/ufs_fs.h>
40 #if defined(AFS_DARWIN_ENV) || defined(AFS_XBSD_ENV)
41 #include <ufs/ufs/dinode.h>
42 #include <ufs/ffs/fs.h>
47 #else /* AFS_VFSINCL_ENV */
48 #if !defined(AFS_AIX_ENV) && !defined(AFS_LINUX20_ENV) && !defined(AFS_XBSD_ENV)
51 #endif /* AFS_VFSINCL_ENV */
52 #endif /* AFS_OSF_ENV */
53 #endif /* AFS_SGI_ENV */
54 #endif /* AFS_NT40_ENV */
72 #if defined(AFS_SUN_ENV) || defined(AFS_SUN5_ENV)
74 #include <sys/mnttab.h>
75 #include <sys/mntent.h>
81 #if defined(AFS_SGI_ENV)
86 #ifndef AFS_LINUX20_ENV
87 #include <fstab.h> /* Need to find in libc 5, present in libc 6 */
90 #endif /* AFS_SGI_ENV */
92 #endif /* AFS_HPUX_ENV */
96 #include <netinet/in.h>
100 #include <sys/time.h>
101 #endif /* ITIMER_REAL */
102 #endif /* AFS_NT40_ENV */
103 #if defined(AFS_SUN5_ENV) || defined(AFS_NT40_ENV) || defined(AFS_LINUX20_ENV)
110 #include <afs/errors.h>
113 #include <afs/afssyscalls.h>
115 #include <afs/afsutil.h>
119 #include "daemon_com.h"
121 #include "salvsync.h"
124 #include "partition.h"
125 #include "volume_inline.h"
126 #ifdef AFS_PTHREAD_ENV
128 #else /* AFS_PTHREAD_ENV */
129 #include "afs/assert.h"
130 #endif /* AFS_PTHREAD_ENV */
137 #if !defined(offsetof)
142 #define afs_stat stat64
143 #define afs_fstat fstat64
144 #define afs_open open64
145 #else /* !O_LARGEFILE */
146 #define afs_stat stat
147 #define afs_fstat fstat
148 #define afs_open open
149 #endif /* !O_LARGEFILE */
151 #ifdef AFS_PTHREAD_ENV
152 pthread_mutex_t vol_glock_mutex;
153 pthread_mutex_t vol_trans_mutex;
154 pthread_cond_t vol_put_volume_cond;
155 pthread_cond_t vol_sleep_cond;
156 pthread_cond_t vol_init_attach_cond;
157 int vol_attach_threads = 1;
158 #endif /* AFS_PTHREAD_ENV */
160 /* start-time configurable I/O parameters */
161 ih_init_params vol_io_params;
163 #ifdef AFS_DEMAND_ATTACH_FS
164 pthread_mutex_t vol_salvsync_mutex;
167 * Set this to 1 to disallow SALVSYNC communication in all threads; used
168 * during shutdown, since the salvageserver may have gone away.
170 static volatile sig_atomic_t vol_disallow_salvsync = 0;
171 #endif /* AFS_DEMAND_ATTACH_FS */
174 extern void *calloc(), *realloc();
177 /*@printflike@*/ extern void Log(const char *format, ...);
179 /* Forward declarations */
180 static Volume *attach2(Error * ec, VolId volumeId, char *path,
181 struct DiskPartition64 *partp, Volume * vp,
182 int isbusy, int mode);
183 static void ReallyFreeVolume(Volume * vp);
184 #ifdef AFS_DEMAND_ATTACH_FS
185 static void FreeVolume(Volume * vp);
186 #else /* !AFS_DEMAND_ATTACH_FS */
187 #define FreeVolume(vp) ReallyFreeVolume(vp)
188 static void VScanUpdateList(void);
189 #endif /* !AFS_DEMAND_ATTACH_FS */
190 static void VInitVolumeHeaderCache(afs_uint32 howMany);
191 static int GetVolumeHeader(register Volume * vp);
192 static void ReleaseVolumeHeader(register struct volHeader *hd);
193 static void FreeVolumeHeader(register Volume * vp);
194 static void AddVolumeToHashTable(register Volume * vp, int hashid);
195 static void DeleteVolumeFromHashTable(register Volume * vp);
197 static int VHold(Volume * vp);
199 static int VHold_r(Volume * vp);
200 static void VGetBitmap_r(Error * ec, Volume * vp, VnodeClass class);
201 static void VReleaseVolumeHandles_r(Volume * vp);
202 static void VCloseVolumeHandles_r(Volume * vp);
203 static void LoadVolumeHeader(Error * ec, Volume * vp);
204 static int VCheckOffline(register Volume * vp);
205 static int VCheckDetach(register Volume * vp);
206 static Volume * GetVolume(Error * ec, Error * client_ec, VolId volumeId, Volume * hint, int flags);
208 int LogLevel; /* Vice loglevel--not defined as extern so that it will be
209 * defined when not linked with vice, XXXX */
210 ProgramType programType; /* The type of program using the package */
211 static VolumePackageOptions vol_opts;
213 /* extended volume package statistics */
216 #ifdef VOL_LOCK_DEBUG
217 pthread_t vol_glock_holder = 0;
221 #define VOLUME_BITMAP_GROWSIZE 16 /* bytes, => 128vnodes */
222 /* Must be a multiple of 4 (1 word) !! */
224 /* this parameter needs to be tunable at runtime.
225 * 128 was really inadequate for largish servers -- at 16384 volumes this
226 * puts average chain length at 128, thus an average 65 deref's to find a volptr.
227 * talk about bad spatial locality...
229 * an AVL or splay tree might work a lot better, but we'll just increase
230 * the default hash table size for now
232 #define DEFAULT_VOLUME_HASH_SIZE 256 /* Must be a power of 2!! */
233 #define DEFAULT_VOLUME_HASH_MASK (DEFAULT_VOLUME_HASH_SIZE-1)
234 #define VOLUME_HASH(volumeId) (volumeId&(VolumeHashTable.Mask))
237 * turn volume hash chains into partially ordered lists.
238 * when the threshold is exceeded between two adjacent elements,
239 * perform a chain rebalancing operation.
241 * keep the threshold high in order to keep cache line invalidates
242 * low "enough" on SMPs
244 #define VOLUME_HASH_REORDER_THRESHOLD 200
247 * when possible, don't just reorder single elements, but reorder
248 * entire chains of elements at once. a chain of elements that
249 * exceed the element previous to the pivot by at least CHAIN_THRESH
250 * accesses are moved in front of the chain whose elements have at
251 * least CHAIN_THRESH less accesses than the pivot element
253 #define VOLUME_HASH_REORDER_CHAIN_THRESH (VOLUME_HASH_REORDER_THRESHOLD / 2)
255 #include "rx/rx_queue.h"
258 VolumeHashTable_t VolumeHashTable = {
259 DEFAULT_VOLUME_HASH_SIZE,
260 DEFAULT_VOLUME_HASH_MASK,
265 static void VInitVolumeHash(void);
269 /* This macro is used where an ffs() call does not exist. Was in util/ffs.c */
273 afs_int32 ffs_tmp = x;
277 for (ffs_i = 1;; ffs_i++) {
284 #endif /* !AFS_HAVE_FFS */
286 #ifdef AFS_PTHREAD_ENV
288 * disk partition queue element
290 typedef struct diskpartition_queue_t {
291 struct rx_queue queue; /**< queue header */
292 struct DiskPartition64 *diskP; /**< disk partition table entry */
293 } diskpartition_queue_t;
295 #ifndef AFS_DEMAND_ATTACH_FS
297 typedef struct vinitvolumepackage_thread_t {
298 struct rx_queue queue;
299 pthread_cond_t thread_done_cv;
300 int n_threads_complete;
301 } vinitvolumepackage_thread_t;
302 static void * VInitVolumePackageThread(void * args);
304 #else /* !AFS_DEMAND_ATTTACH_FS */
305 #define VINIT_BATCH_MAX_SIZE 512
308 * disk partition work queue
310 struct partition_queue {
311 struct rx_queue head; /**< diskpartition_queue_t queue */
312 pthread_mutex_t mutex;
317 * volumes parameters for preattach
319 struct volume_init_batch {
320 struct rx_queue queue; /**< queue header */
321 int thread; /**< posting worker thread */
322 int last; /**< indicates thread is done */
323 int size; /**< number of volume ids in batch */
324 Volume *batch[VINIT_BATCH_MAX_SIZE]; /**< volumes ids to preattach */
328 * volume parameters work queue
330 struct volume_init_queue {
331 struct rx_queue head; /**< volume_init_batch queue */
332 pthread_mutex_t mutex;
337 * volume init worker thread parameters
339 struct vinitvolumepackage_thread_param {
340 int nthreads; /**< total number of worker threads */
341 int thread; /**< thread number for this worker thread */
342 struct partition_queue *pq; /**< queue partitions to scan */
343 struct volume_init_queue *vq; /**< queue of volume to preattach */
346 static void *VInitVolumePackageThread(void *args);
347 static struct DiskPartition64 *VInitNextPartition(struct partition_queue *pq);
348 static VolId VInitNextVolumeId(DIR *dirp);
349 static int VInitPreAttachVolumes(int nthreads, struct volume_init_queue *vq);
351 #endif /* !AFS_DEMAND_ATTACH_FS */
352 #endif /* AFS_PTHREAD_ENV */
354 #ifndef AFS_DEMAND_ATTACH_FS
355 static int VAttachVolumesByPartition(struct DiskPartition64 *diskP,
356 int * nAttached, int * nUnattached);
357 #endif /* AFS_DEMAND_ATTACH_FS */
360 #ifdef AFS_DEMAND_ATTACH_FS
361 /* demand attach fileserver extensions */
364 * in the future we will support serialization of VLRU state into the fs_state
367 * these structures are the beginning of that effort
369 struct VLRU_DiskHeader {
370 struct versionStamp stamp; /* magic and structure version number */
371 afs_uint32 mtime; /* time of dump to disk */
372 afs_uint32 num_records; /* number of VLRU_DiskEntry records */
375 struct VLRU_DiskEntry {
376 afs_uint32 vid; /* volume ID */
377 afs_uint32 idx; /* generation */
378 afs_uint32 last_get; /* timestamp of last get */
381 struct VLRU_StartupQueue {
382 struct VLRU_DiskEntry * entry;
387 typedef struct vshutdown_thread_t {
389 pthread_mutex_t lock;
391 pthread_cond_t master_cv;
393 int n_threads_complete;
395 int schedule_version;
398 byte n_parts_done_pass;
399 byte part_thread_target[VOLMAXPARTS+1];
400 byte part_done_pass[VOLMAXPARTS+1];
401 struct rx_queue * part_pass_head[VOLMAXPARTS+1];
402 int stats[4][VOLMAXPARTS+1];
403 } vshutdown_thread_t;
404 static void * VShutdownThread(void * args);
407 static Volume * VAttachVolumeByVp_r(Error * ec, Volume * vp, int mode);
408 static int VCheckFree(Volume * vp);
411 static void AddVolumeToVByPList_r(Volume * vp);
412 static void DeleteVolumeFromVByPList_r(Volume * vp);
413 static void VVByPListBeginExclusive_r(struct DiskPartition64 * dp);
414 static void VVByPListEndExclusive_r(struct DiskPartition64 * dp);
415 static void VVByPListWait_r(struct DiskPartition64 * dp);
417 /* online salvager */
418 static int VCheckSalvage(register Volume * vp);
419 #if defined(SALVSYNC_BUILD_CLIENT) || defined(FSSYNC_BUILD_CLIENT)
420 static int VScheduleSalvage_r(Volume * vp);
423 /* Volume hash table */
424 static void VReorderHash_r(VolumeHashChainHead * head, Volume * pp, Volume * vp);
425 static void VHashBeginExclusive_r(VolumeHashChainHead * head);
426 static void VHashEndExclusive_r(VolumeHashChainHead * head);
427 static void VHashWait_r(VolumeHashChainHead * head);
430 static int ShutdownVByPForPass_r(struct DiskPartition64 * dp, int pass);
431 static int ShutdownVolumeWalk_r(struct DiskPartition64 * dp, int pass,
432 struct rx_queue ** idx);
433 static void ShutdownController(vshutdown_thread_t * params);
434 static void ShutdownCreateSchedule(vshutdown_thread_t * params);
437 static void VLRU_ComputeConstants(void);
438 static void VInitVLRU(void);
439 static void VLRU_Init_Node_r(Volume * vp);
440 static void VLRU_Add_r(Volume * vp);
441 static void VLRU_Delete_r(Volume * vp);
442 static void VLRU_UpdateAccess_r(Volume * vp);
443 static void * VLRU_ScannerThread(void * args);
444 static void VLRU_Scan_r(int idx);
445 static void VLRU_Promote_r(int idx);
446 static void VLRU_Demote_r(int idx);
447 static void VLRU_SwitchQueues(Volume * vp, int new_idx, int append);
450 static int VCheckSoftDetach(Volume * vp, afs_uint32 thresh);
451 static int VCheckSoftDetachCandidate(Volume * vp, afs_uint32 thresh);
452 static int VSoftDetachVolume_r(Volume * vp, afs_uint32 thresh);
455 pthread_key_t VThread_key;
456 VThreadOptions_t VThread_defaults = {
457 0 /**< allow salvsync */
459 #endif /* AFS_DEMAND_ATTACH_FS */
462 struct Lock vol_listLock; /* Lock obtained when listing volumes:
463 * prevents a volume from being missed
464 * if the volume is attached during a
468 /* Common message used when the volume goes off line */
469 char *VSalvageMessage =
470 "Files in this volume are currently unavailable; call operations";
472 int VInit; /* 0 - uninitialized,
473 * 1 - initialized but not all volumes have been attached,
474 * 2 - initialized and all volumes have been attached,
475 * 3 - initialized, all volumes have been attached, and
476 * VConnectFS() has completed. */
478 static int vinit_attach_abort = 0;
480 bit32 VolumeCacheCheck; /* Incremented everytime a volume goes on line--
481 * used to stamp volume headers and in-core
482 * vnodes. When the volume goes on-line the
483 * vnode will be invalidated
484 * access only with VOL_LOCK held */
489 /***************************************************/
490 /* Startup routines */
491 /***************************************************/
493 * assign default values to a VolumePackageOptions struct.
495 * Always call this on a VolumePackageOptions struct first, then set any
496 * specific options you want, then call VInitVolumePackage2.
498 * @param[in] pt caller's program type
499 * @param[out] opts volume package options
502 VOptDefaults(ProgramType pt, VolumePackageOptions *opts)
504 opts->nLargeVnodes = opts->nSmallVnodes = 5;
507 opts->canScheduleSalvage = 0;
508 opts->canUseFSSYNC = 0;
509 opts->canUseSALVSYNC = 0;
513 opts->canScheduleSalvage = 1;
514 opts->canUseSALVSYNC = 1;
518 opts->canUseFSSYNC = 1;
522 opts->nLargeVnodes = 0;
523 opts->nSmallVnodes = 0;
525 opts->canScheduleSalvage = 1;
526 opts->canUseFSSYNC = 1;
536 VInitVolumePackage2(ProgramType pt, VolumePackageOptions * opts)
538 int errors = 0; /* Number of errors while finding vice partitions. */
543 memset(&VStats, 0, sizeof(VStats));
544 VStats.hdr_cache_size = 200;
546 VInitPartitionPackage();
548 #ifdef AFS_DEMAND_ATTACH_FS
549 if (programType == fileServer) {
552 VLRU_SetOptions(VLRU_SET_ENABLED, 0);
554 assert(pthread_key_create(&VThread_key, NULL) == 0);
557 #ifdef AFS_PTHREAD_ENV
558 assert(pthread_mutex_init(&vol_glock_mutex, NULL) == 0);
559 assert(pthread_mutex_init(&vol_trans_mutex, NULL) == 0);
560 assert(pthread_cond_init(&vol_put_volume_cond, NULL) == 0);
561 assert(pthread_cond_init(&vol_sleep_cond, NULL) == 0);
562 assert(pthread_cond_init(&vol_init_attach_cond, NULL) == 0);
563 #else /* AFS_PTHREAD_ENV */
565 #endif /* AFS_PTHREAD_ENV */
566 Lock_Init(&vol_listLock);
568 srandom(time(0)); /* For VGetVolumeInfo */
570 #ifdef AFS_DEMAND_ATTACH_FS
571 assert(pthread_mutex_init(&vol_salvsync_mutex, NULL) == 0);
572 #endif /* AFS_DEMAND_ATTACH_FS */
574 /* Ok, we have done enough initialization that fileserver can
575 * start accepting calls, even though the volumes may not be
576 * available just yet.
580 #if defined(AFS_DEMAND_ATTACH_FS) && defined(SALVSYNC_BUILD_SERVER)
581 if (programType == salvageServer) {
584 #endif /* AFS_DEMAND_ATTACH_FS */
585 #ifdef FSSYNC_BUILD_SERVER
586 if (programType == fileServer) {
590 #if defined(AFS_DEMAND_ATTACH_FS) && defined(SALVSYNC_BUILD_CLIENT)
591 if (VCanUseSALVSYNC()) {
592 /* establish a connection to the salvager at this point */
593 assert(VConnectSALV() != 0);
595 #endif /* AFS_DEMAND_ATTACH_FS */
597 if (opts->volcache > VStats.hdr_cache_size)
598 VStats.hdr_cache_size = opts->volcache;
599 VInitVolumeHeaderCache(VStats.hdr_cache_size);
601 VInitVnodes(vLarge, opts->nLargeVnodes);
602 VInitVnodes(vSmall, opts->nSmallVnodes);
605 errors = VAttachPartitions();
609 if (programType != fileServer) {
610 errors = VInitAttachVolumes(programType);
616 #ifdef FSSYNC_BUILD_CLIENT
617 if (VCanUseFSSYNC()) {
619 #ifdef AFS_DEMAND_ATTACH_FS
620 if (programType == salvageServer) {
621 Log("Unable to connect to file server; aborted\n");
624 #endif /* AFS_DEMAND_ATTACH_FS */
625 Log("Unable to connect to file server; will retry at need\n");
628 #endif /* FSSYNC_BUILD_CLIENT */
633 #if !defined(AFS_PTHREAD_ENV)
635 * Attach volumes in vice partitions
637 * @param[in] pt calling program type
640 * @note This is the original, non-threaded version of attach parititions.
642 * @post VInit state is 2
645 VInitAttachVolumes(ProgramType pt)
648 if (pt == fileServer) {
649 struct DiskPartition64 *diskP;
650 /* Attach all the volumes in this partition */
651 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
652 int nAttached = 0, nUnattached = 0;
653 assert(VAttachVolumesByPartition(diskP, &nAttached, &nUnattached) == 0);
657 VInit = 2; /* Initialized, and all volumes have been attached */
658 LWP_NoYieldSignal(VInitAttachVolumes);
662 #endif /* !AFS_PTHREAD_ENV */
664 #if defined(AFS_PTHREAD_ENV) && !defined(AFS_DEMAND_ATTACH_FS)
666 * Attach volumes in vice partitions
668 * @param[in] pt calling program type
671 * @note Threaded version of attach parititions.
673 * @post VInit state is 2
676 VInitAttachVolumes(ProgramType pt)
679 if (pt == fileServer) {
680 struct DiskPartition64 *diskP;
681 struct vinitvolumepackage_thread_t params;
682 struct diskpartition_queue_t * dpq;
683 int i, threads, parts;
685 pthread_attr_t attrs;
687 assert(pthread_cond_init(¶ms.thread_done_cv,NULL) == 0);
689 params.n_threads_complete = 0;
691 /* create partition work queue */
692 for (parts=0, diskP = DiskPartitionList; diskP; diskP = diskP->next, parts++) {
693 dpq = (diskpartition_queue_t *) malloc(sizeof(struct diskpartition_queue_t));
696 queue_Append(¶ms,dpq);
699 threads = MIN(parts, vol_attach_threads);
702 /* spawn off a bunch of initialization threads */
703 assert(pthread_attr_init(&attrs) == 0);
704 assert(pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED) == 0);
706 Log("VInitVolumePackage: beginning parallel fileserver startup\n");
707 Log("VInitVolumePackage: using %d threads to attach volumes on %d partitions\n",
711 for (i=0; i < threads; i++) {
714 assert(pthread_create
715 (&tid, &attrs, &VInitVolumePackageThread,
717 AFS_SIGSET_RESTORE();
720 while(params.n_threads_complete < threads) {
721 VOL_CV_WAIT(¶ms.thread_done_cv);
725 assert(pthread_attr_destroy(&attrs) == 0);
727 /* if we're only going to run one init thread, don't bother creating
729 Log("VInitVolumePackage: beginning single-threaded fileserver startup\n");
730 Log("VInitVolumePackage: using 1 thread to attach volumes on %d partition(s)\n",
733 VInitVolumePackageThread(¶ms);
736 assert(pthread_cond_destroy(¶ms.thread_done_cv) == 0);
739 VInit = 2; /* Initialized, and all volumes have been attached */
740 assert(pthread_cond_broadcast(&vol_init_attach_cond) == 0);
746 VInitVolumePackageThread(void * args) {
748 struct DiskPartition64 *diskP;
749 struct vinitvolumepackage_thread_t * params;
750 struct diskpartition_queue_t * dpq;
752 params = (vinitvolumepackage_thread_t *) args;
756 /* Attach all the volumes in this partition */
757 while (queue_IsNotEmpty(params)) {
758 int nAttached = 0, nUnattached = 0;
760 if (vinit_attach_abort) {
761 Log("Aborting initialization\n");
765 dpq = queue_First(params,diskpartition_queue_t);
771 assert(VAttachVolumesByPartition(diskP, &nAttached, &nUnattached) == 0);
777 params->n_threads_complete++;
778 pthread_cond_signal(¶ms->thread_done_cv);
782 #endif /* AFS_PTHREAD_ENV && !AFS_DEMAND_ATTACH_FS */
784 #if defined(AFS_DEMAND_ATTACH_FS)
786 * Attach volumes in vice partitions
788 * @param[in] pt calling program type
791 * @note Threaded version of attach partitions.
793 * @post VInit state is 2
796 VInitAttachVolumes(ProgramType pt)
799 if (pt == fileServer) {
801 struct DiskPartition64 *diskP;
802 struct partition_queue pq;
803 struct volume_init_queue vq;
805 int i, threads, parts;
807 pthread_attr_t attrs;
809 /* create partition work queue */
811 assert(pthread_cond_init(&(pq.cv), NULL) == 0);
812 assert(pthread_mutex_init(&(pq.mutex), NULL) == 0);
813 for (parts = 0, diskP = DiskPartitionList; diskP; diskP = diskP->next, parts++) {
814 struct diskpartition_queue_t *dp;
815 dp = (struct diskpartition_queue_t*)malloc(sizeof(struct diskpartition_queue_t));
818 queue_Append(&pq, dp);
821 /* number of worker threads; at least one, not to exceed the number of partitions */
822 threads = MIN(parts, vol_attach_threads);
824 /* create volume work queue */
826 assert(pthread_cond_init(&(vq.cv), NULL) == 0);
827 assert(pthread_mutex_init(&(vq.mutex), NULL) == 0);
829 assert(pthread_attr_init(&attrs) == 0);
830 assert(pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED) == 0);
832 Log("VInitVolumePackage: beginning parallel fileserver startup\n");
833 Log("VInitVolumePackage: using %d threads to pre-attach volumes on %d partitions\n",
836 /* create threads to scan disk partitions. */
837 for (i=0; i < threads; i++) {
838 struct vinitvolumepackage_thread_param *params;
839 params = (struct vinitvolumepackage_thread_param *)malloc(sizeof(struct vinitvolumepackage_thread_param));
843 params->nthreads = threads;
844 params->thread = i+1;
848 assert(pthread_create (&tid, &attrs, &VInitVolumePackageThread, (void*)params) == 0);
849 AFS_SIGSET_RESTORE();
852 VInitPreAttachVolumes(threads, &vq);
854 assert(pthread_attr_destroy(&attrs) == 0);
855 assert(pthread_cond_destroy(&pq.cv) == 0);
856 assert(pthread_mutex_destroy(&pq.mutex) == 0);
857 assert(pthread_cond_destroy(&vq.cv) == 0);
858 assert(pthread_mutex_destroy(&vq.mutex) == 0);
862 VInit = 2; /* Initialized, and all volumes have been attached */
863 assert(pthread_cond_broadcast(&vol_init_attach_cond) == 0);
870 * Volume package initialization worker thread. Scan partitions for volume
871 * header files. Gather batches of volume ids and dispatch them to
872 * the main thread to be preattached. The volume preattachement is done
873 * in the main thread to avoid global volume lock contention.
876 VInitVolumePackageThread(void *args)
878 struct vinitvolumepackage_thread_param *params;
879 struct DiskPartition64 *partition;
880 struct partition_queue *pq;
881 struct volume_init_queue *vq;
882 struct volume_init_batch *vb;
885 params = (struct vinitvolumepackage_thread_param *)args;
891 vb = (struct volume_init_batch*)malloc(sizeof(struct volume_init_batch));
893 vb->thread = params->thread;
897 Log("Scanning partitions on thread %d of %d\n", params->thread, params->nthreads);
898 while((partition = VInitNextPartition(pq))) {
902 Log("Partition %s: pre-attaching volumes\n", partition->name);
903 dirp = opendir(VPartitionPath(partition));
905 Log("opendir on Partition %s failed, errno=%d!\n", partition->name, errno);
908 while ((vid = VInitNextVolumeId(dirp))) {
909 Volume *vp = (Volume*)malloc(sizeof(Volume));
911 memset(vp, 0, sizeof(Volume));
912 vp->device = partition->device;
913 vp->partition = partition;
915 queue_Init(&vp->vnode_list);
916 assert(pthread_cond_init(&V_attachCV(vp), NULL) == 0);
918 vb->batch[vb->size++] = vp;
919 if (vb->size == VINIT_BATCH_MAX_SIZE) {
920 assert(pthread_mutex_lock(&vq->mutex) == 0);
921 queue_Append(vq, vb);
922 assert(pthread_cond_broadcast(&vq->cv) == 0);
923 assert(pthread_mutex_unlock(&vq->mutex) == 0);
925 vb = (struct volume_init_batch*)malloc(sizeof(struct volume_init_batch));
927 vb->thread = params->thread;
936 assert(pthread_mutex_lock(&vq->mutex) == 0);
937 queue_Append(vq, vb);
938 assert(pthread_cond_broadcast(&vq->cv) == 0);
939 assert(pthread_mutex_unlock(&vq->mutex) == 0);
941 Log("Partition scan thread %d of %d ended\n", params->thread, params->nthreads);
947 * Read next element from the pre-populated partition list.
949 static struct DiskPartition64*
950 VInitNextPartition(struct partition_queue *pq)
952 struct DiskPartition64 *partition;
953 struct diskpartition_queue_t *dp; /* queue element */
955 if (vinit_attach_abort) {
956 Log("Aborting volume preattach thread.\n");
960 /* get next partition to scan */
961 assert(pthread_mutex_lock(&pq->mutex) == 0);
962 if (queue_IsEmpty(pq)) {
963 assert(pthread_mutex_unlock(&pq->mutex) == 0);
966 dp = queue_First(pq, diskpartition_queue_t);
968 assert(pthread_mutex_unlock(&pq->mutex) == 0);
973 partition = dp->diskP;
979 * Find next volume id on the partition.
982 VInitNextVolumeId(DIR *dirp)
988 while((d = readdir(dirp))) {
989 if (vinit_attach_abort) {
990 Log("Aborting volume preattach thread.\n");
993 ext = strrchr(d->d_name, '.');
994 if (d->d_name[0] == 'V' && ext && strcmp(ext, VHDREXT) == 0) {
995 vid = VolumeNumber(d->d_name);
999 Log("Warning: bogus volume header file: %s\n", d->d_name);
1006 * Preattach volumes in batches to avoid lock contention.
1009 VInitPreAttachVolumes(int nthreads, struct volume_init_queue *vq)
1011 struct volume_init_batch *vb;
1015 /* dequeue next volume */
1016 pthread_mutex_lock(&vq->mutex);
1017 if (queue_IsEmpty(vq)) {
1018 pthread_cond_wait(&vq->cv, &vq->mutex);
1020 vb = queue_First(vq, volume_init_batch);
1022 pthread_mutex_unlock(&vq->mutex);
1026 for (i = 0; i<vb->size; i++) {
1032 dup = VLookupVolume_r(&ec, vp->hashid, NULL);
1034 Log("Error looking up volume, code=%d\n", ec);
1037 Log("Warning: Duplicate volume id %d detected.\n", vp->hashid);
1040 /* put pre-attached volume onto the hash table
1041 * and bring it up to the pre-attached state */
1042 AddVolumeToHashTable(vp, vp->hashid);
1043 AddVolumeToVByPList_r(vp);
1044 VLRU_Init_Node_r(vp);
1045 VChangeState_r(vp, VOL_STATE_PREATTACHED);
1058 #endif /* AFS_DEMAND_ATTACH_FS */
1060 #if !defined(AFS_DEMAND_ATTACH_FS)
1062 * attach all volumes on a given disk partition
1065 VAttachVolumesByPartition(struct DiskPartition64 *diskP, int * nAttached, int * nUnattached)
1071 Log("Partition %s: attaching volumes\n", diskP->name);
1072 dirp = opendir(VPartitionPath(diskP));
1074 Log("opendir on Partition %s failed!\n", diskP->name);
1078 while ((dp = readdir(dirp))) {
1080 p = strrchr(dp->d_name, '.');
1082 if (vinit_attach_abort) {
1083 Log("Partition %s: abort attach volumes\n", diskP->name);
1087 if (p != NULL && strcmp(p, VHDREXT) == 0) {
1090 vp = VAttachVolumeByName(&error, diskP->name, dp->d_name,
1092 (*(vp ? nAttached : nUnattached))++;
1093 if (error == VOFFLINE)
1094 Log("Volume %d stays offline (/vice/offline/%s exists)\n", VolumeNumber(dp->d_name), dp->d_name);
1095 else if (LogLevel >= 5) {
1096 Log("Partition %s: attached volume %d (%s)\n",
1097 diskP->name, VolumeNumber(dp->d_name),
1106 Log("Partition %s: attached %d volumes; %d volumes not attached\n", diskP->name, *nAttached, *nUnattached);
1111 #endif /* !AFS_DEMAND_ATTACH_FS */
1113 /***************************************************/
1114 /* Shutdown routines */
1115 /***************************************************/
1119 * highly multithreaded volume package shutdown
1121 * with the demand attach fileserver extensions,
1122 * VShutdown has been modified to be multithreaded.
1123 * In order to achieve optimal use of many threads,
1124 * the shutdown code involves one control thread and
1125 * n shutdown worker threads. The control thread
1126 * periodically examines the number of volumes available
1127 * for shutdown on each partition, and produces a worker
1128 * thread allocation schedule. The idea is to eliminate
1129 * redundant scheduling computation on the workers by
1130 * having a single master scheduler.
1132 * The scheduler's objectives are:
1134 * each partition with volumes remaining gets allocated
1135 * at least 1 thread (assuming sufficient threads)
1137 * threads are allocated proportional to the number of
1138 * volumes remaining to be offlined. This ensures that
1139 * the OS I/O scheduler has many requests to elevator
1140 * seek on partitions that will (presumably) take the
1141 * longest amount of time (from now) to finish shutdown
1142 * (3) keep threads busy
1143 * when there are extra threads, they are assigned to
1144 * partitions using a simple round-robin algorithm
1146 * In the future, we may wish to add the ability to adapt
1147 * to the relative performance patterns of each disk
1152 * multi-step shutdown process
1154 * demand attach shutdown is a four-step process. Each
1155 * shutdown "pass" shuts down increasingly more difficult
1156 * volumes. The main purpose is to achieve better cache
1157 * utilization during shutdown.
1160 * shutdown volumes in the unattached, pre-attached
1163 * shutdown attached volumes with cached volume headers
1165 * shutdown all volumes in non-exclusive states
1167 * shutdown all remaining volumes
1170 #ifdef AFS_DEMAND_ATTACH_FS
1176 struct DiskPartition64 * diskP;
1177 struct diskpartition_queue_t * dpq;
1178 vshutdown_thread_t params;
1180 pthread_attr_t attrs;
1182 memset(¶ms, 0, sizeof(vshutdown_thread_t));
1185 Log("VShutdown: aborting attach volumes\n");
1186 vinit_attach_abort = 1;
1187 VOL_CV_WAIT(&vol_init_attach_cond);
1190 for (params.n_parts=0, diskP = DiskPartitionList;
1191 diskP; diskP = diskP->next, params.n_parts++);
1193 Log("VShutdown: shutting down on-line volumes on %d partition%s...\n",
1194 params.n_parts, params.n_parts > 1 ? "s" : "");
1196 if (vol_attach_threads > 1) {
1197 /* prepare for parallel shutdown */
1198 params.n_threads = vol_attach_threads;
1199 assert(pthread_mutex_init(¶ms.lock, NULL) == 0);
1200 assert(pthread_cond_init(¶ms.cv, NULL) == 0);
1201 assert(pthread_cond_init(¶ms.master_cv, NULL) == 0);
1202 assert(pthread_attr_init(&attrs) == 0);
1203 assert(pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED) == 0);
1204 queue_Init(¶ms);
1206 /* setup the basic partition information structures for
1207 * parallel shutdown */
1208 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1210 struct rx_queue * qp, * nqp;
1214 VVByPListWait_r(diskP);
1215 VVByPListBeginExclusive_r(diskP);
1218 for (queue_Scan(&diskP->vol_list, qp, nqp, rx_queue)) {
1219 vp = (Volume *)((char *)qp - offsetof(Volume, vol_list));
1223 Log("VShutdown: partition %s has %d volumes with attached headers\n",
1224 VPartitionPath(diskP), count);
1227 /* build up the pass 0 shutdown work queue */
1228 dpq = (struct diskpartition_queue_t *) malloc(sizeof(struct diskpartition_queue_t));
1229 assert(dpq != NULL);
1231 queue_Prepend(¶ms, dpq);
1233 params.part_pass_head[diskP->index] = queue_First(&diskP->vol_list, rx_queue);
1236 Log("VShutdown: beginning parallel fileserver shutdown\n");
1237 Log("VShutdown: using %d threads to offline volumes on %d partition%s\n",
1238 vol_attach_threads, params.n_parts, params.n_parts > 1 ? "s" : "" );
1240 /* do pass 0 shutdown */
1241 assert(pthread_mutex_lock(¶ms.lock) == 0);
1242 for (i=0; i < params.n_threads; i++) {
1243 assert(pthread_create
1244 (&tid, &attrs, &VShutdownThread,
1248 /* wait for all the pass 0 shutdowns to complete */
1249 while (params.n_threads_complete < params.n_threads) {
1250 assert(pthread_cond_wait(¶ms.master_cv, ¶ms.lock) == 0);
1252 params.n_threads_complete = 0;
1254 assert(pthread_cond_broadcast(¶ms.cv) == 0);
1255 assert(pthread_mutex_unlock(¶ms.lock) == 0);
1257 Log("VShutdown: pass 0 completed using the 1 thread per partition algorithm\n");
1258 Log("VShutdown: starting passes 1 through 3 using finely-granular mp-fast algorithm\n");
1260 /* run the parallel shutdown scheduler. it will drop the glock internally */
1261 ShutdownController(¶ms);
1263 /* wait for all the workers to finish pass 3 and terminate */
1264 while (params.pass < 4) {
1265 VOL_CV_WAIT(¶ms.cv);
1268 assert(pthread_attr_destroy(&attrs) == 0);
1269 assert(pthread_cond_destroy(¶ms.cv) == 0);
1270 assert(pthread_cond_destroy(¶ms.master_cv) == 0);
1271 assert(pthread_mutex_destroy(¶ms.lock) == 0);
1273 /* drop the VByPList exclusive reservations */
1274 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1275 VVByPListEndExclusive_r(diskP);
1276 Log("VShutdown: %s stats : (pass[0]=%d, pass[1]=%d, pass[2]=%d, pass[3]=%d)\n",
1277 VPartitionPath(diskP),
1278 params.stats[0][diskP->index],
1279 params.stats[1][diskP->index],
1280 params.stats[2][diskP->index],
1281 params.stats[3][diskP->index]);
1284 Log("VShutdown: shutdown finished using %d threads\n", params.n_threads);
1286 /* if we're only going to run one shutdown thread, don't bother creating
1288 Log("VShutdown: beginning single-threaded fileserver shutdown\n");
1290 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1291 VShutdownByPartition_r(diskP);
1295 Log("VShutdown: complete.\n");
1298 #else /* AFS_DEMAND_ATTACH_FS */
1304 register Volume *vp, *np;
1305 register afs_int32 code;
1308 Log("VShutdown: aborting attach volumes\n");
1309 vinit_attach_abort = 1;
1310 #ifdef AFS_PTHREAD_ENV
1311 VOL_CV_WAIT(&vol_init_attach_cond);
1313 LWP_WaitProcess(VInitAttachVolumes);
1314 #endif /* AFS_PTHREAD_ENV */
1317 Log("VShutdown: shutting down on-line volumes...\n");
1318 for (i = 0; i < VolumeHashTable.Size; i++) {
1319 /* try to hold first volume in the hash table */
1320 for (queue_Scan(&VolumeHashTable.Table[i],vp,np,Volume)) {
1324 Log("VShutdown: Attempting to take volume %u offline.\n",
1327 /* next, take the volume offline (drops reference count) */
1328 VOffline_r(vp, "File server was shut down");
1332 Log("VShutdown: complete.\n");
1334 #endif /* AFS_DEMAND_ATTACH_FS */
1347 * stop new activity (e.g. SALVSYNC) from occurring
1349 * Use this to make the volume package less busy; for example, during
1350 * shutdown. This doesn't actually shutdown/detach anything in the
1351 * volume package, but prevents certain processes from ocurring. For
1352 * example, preventing new SALVSYNC communication in DAFS. In theory, we
1353 * could also use this to prevent new volume attachment, or prevent
1354 * other programs from checking out volumes, etc.
1359 #ifdef AFS_DEMAND_ATTACH_FS
1360 /* make sure we don't try to contact the salvageserver, since it may
1361 * not be around anymore */
1362 vol_disallow_salvsync = 1;
1366 #ifdef AFS_DEMAND_ATTACH_FS
1369 * shutdown control thread
1372 ShutdownController(vshutdown_thread_t * params)
1375 struct DiskPartition64 * diskP;
1377 vshutdown_thread_t shadow;
1379 ShutdownCreateSchedule(params);
1381 while ((params->pass < 4) &&
1382 (params->n_threads_complete < params->n_threads)) {
1383 /* recompute schedule once per second */
1385 memcpy(&shadow, params, sizeof(vshutdown_thread_t));
1389 Log("ShutdownController: schedule version=%d, vol_remaining=%d, pass=%d\n",
1390 shadow.schedule_version, shadow.vol_remaining, shadow.pass);
1391 Log("ShutdownController: n_threads_complete=%d, n_parts_done_pass=%d\n",
1392 shadow.n_threads_complete, shadow.n_parts_done_pass);
1393 for (diskP = DiskPartitionList; diskP; diskP=diskP->next) {
1395 Log("ShutdownController: part[%d] : (len=%d, thread_target=%d, done_pass=%d, pass_head=%p)\n",
1397 diskP->vol_list.len,
1398 shadow.part_thread_target[id],
1399 shadow.part_done_pass[id],
1400 shadow.part_pass_head[id]);
1406 ShutdownCreateSchedule(params);
1410 /* create the shutdown thread work schedule.
1411 * this scheduler tries to implement fairness
1412 * by allocating at least 1 thread to each
1413 * partition with volumes to be shutdown,
1414 * and then it attempts to allocate remaining
1415 * threads based upon the amount of work left
1418 ShutdownCreateSchedule(vshutdown_thread_t * params)
1420 struct DiskPartition64 * diskP;
1421 int sum, thr_workload, thr_left;
1422 int part_residue[VOLMAXPARTS+1];
1425 /* compute the total number of outstanding volumes */
1427 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1428 sum += diskP->vol_list.len;
1431 params->schedule_version++;
1432 params->vol_remaining = sum;
1437 /* compute average per-thread workload */
1438 thr_workload = sum / params->n_threads;
1439 if (sum % params->n_threads)
1442 thr_left = params->n_threads;
1443 memset(&part_residue, 0, sizeof(part_residue));
1445 /* for fairness, give every partition with volumes remaining
1446 * at least one thread */
1447 for (diskP = DiskPartitionList; diskP && thr_left; diskP = diskP->next) {
1449 if (diskP->vol_list.len) {
1450 params->part_thread_target[id] = 1;
1453 params->part_thread_target[id] = 0;
1457 if (thr_left && thr_workload) {
1458 /* compute length-weighted workloads */
1461 for (diskP = DiskPartitionList; diskP && thr_left; diskP = diskP->next) {
1463 delta = (diskP->vol_list.len / thr_workload) -
1464 params->part_thread_target[id];
1468 if (delta < thr_left) {
1469 params->part_thread_target[id] += delta;
1472 params->part_thread_target[id] += thr_left;
1480 /* try to assign any leftover threads to partitions that
1481 * had volume lengths closer to needing thread_target+1 */
1482 int max_residue, max_id = 0;
1484 /* compute the residues */
1485 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1487 part_residue[id] = diskP->vol_list.len -
1488 (params->part_thread_target[id] * thr_workload);
1491 /* now try to allocate remaining threads to partitions with the
1492 * highest residues */
1495 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1497 if (part_residue[id] > max_residue) {
1498 max_residue = part_residue[id];
1507 params->part_thread_target[max_id]++;
1509 part_residue[max_id] = 0;
1514 /* punt and give any remaining threads equally to each partition */
1516 if (thr_left >= params->n_parts) {
1517 alloc = thr_left / params->n_parts;
1518 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1520 params->part_thread_target[id] += alloc;
1525 /* finish off the last of the threads */
1526 for (diskP = DiskPartitionList; thr_left && diskP; diskP = diskP->next) {
1528 params->part_thread_target[id]++;
1534 /* worker thread for parallel shutdown */
1536 VShutdownThread(void * args)
1538 vshutdown_thread_t * params;
1539 int found, pass, schedule_version_save, count;
1540 struct DiskPartition64 *diskP;
1541 struct diskpartition_queue_t * dpq;
1544 params = (vshutdown_thread_t *) args;
1546 /* acquire the shutdown pass 0 lock */
1547 assert(pthread_mutex_lock(¶ms->lock) == 0);
1549 /* if there's still pass 0 work to be done,
1550 * get a work entry, and do a pass 0 shutdown */
1551 if (queue_IsNotEmpty(params)) {
1552 dpq = queue_First(params, diskpartition_queue_t);
1554 assert(pthread_mutex_unlock(¶ms->lock) == 0);
1560 while (ShutdownVolumeWalk_r(diskP, 0, ¶ms->part_pass_head[id]))
1562 params->stats[0][diskP->index] = count;
1563 assert(pthread_mutex_lock(¶ms->lock) == 0);
1566 params->n_threads_complete++;
1567 if (params->n_threads_complete == params->n_threads) {
1568 /* notify control thread that all workers have completed pass 0 */
1569 assert(pthread_cond_signal(¶ms->master_cv) == 0);
1571 while (params->pass == 0) {
1572 assert(pthread_cond_wait(¶ms->cv, ¶ms->lock) == 0);
1576 assert(pthread_mutex_unlock(¶ms->lock) == 0);
1579 pass = params->pass;
1582 /* now escalate through the more complicated shutdowns */
1584 schedule_version_save = params->schedule_version;
1586 /* find a disk partition to work on */
1587 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1589 if (params->part_thread_target[id] && !params->part_done_pass[id]) {
1590 params->part_thread_target[id]--;
1597 /* hmm. for some reason the controller thread couldn't find anything for
1598 * us to do. let's see if there's anything we can do */
1599 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1601 if (diskP->vol_list.len && !params->part_done_pass[id]) {
1604 } else if (!params->part_done_pass[id]) {
1605 params->part_done_pass[id] = 1;
1606 params->n_parts_done_pass++;
1608 Log("VShutdown: done shutting down volumes on partition %s.\n",
1609 VPartitionPath(diskP));
1615 /* do work on this partition until either the controller
1616 * creates a new schedule, or we run out of things to do
1617 * on this partition */
1620 while (!params->part_done_pass[id] &&
1621 (schedule_version_save == params->schedule_version)) {
1622 /* ShutdownVolumeWalk_r will drop the glock internally */
1623 if (!ShutdownVolumeWalk_r(diskP, pass, ¶ms->part_pass_head[id])) {
1624 if (!params->part_done_pass[id]) {
1625 params->part_done_pass[id] = 1;
1626 params->n_parts_done_pass++;
1628 Log("VShutdown: done shutting down volumes on partition %s.\n",
1629 VPartitionPath(diskP));
1637 params->stats[pass][id] += count;
1639 /* ok, everyone is done this pass, proceed */
1642 params->n_threads_complete++;
1643 while (params->pass == pass) {
1644 if (params->n_threads_complete == params->n_threads) {
1645 /* we are the last thread to complete, so we will
1646 * reinitialize worker pool state for the next pass */
1647 params->n_threads_complete = 0;
1648 params->n_parts_done_pass = 0;
1650 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1652 params->part_done_pass[id] = 0;
1653 params->part_pass_head[id] = queue_First(&diskP->vol_list, rx_queue);
1656 /* compute a new thread schedule before releasing all the workers */
1657 ShutdownCreateSchedule(params);
1659 /* wake up all the workers */
1660 assert(pthread_cond_broadcast(¶ms->cv) == 0);
1663 Log("VShutdown: pass %d completed using %d threads on %d partitions\n",
1664 pass, params->n_threads, params->n_parts);
1667 VOL_CV_WAIT(¶ms->cv);
1670 pass = params->pass;
1684 /* shut down all volumes on a given disk partition
1686 * note that this function will not allow mp-fast
1687 * shutdown of a partition */
1689 VShutdownByPartition_r(struct DiskPartition64 * dp)
1695 /* wait for other exclusive ops to finish */
1696 VVByPListWait_r(dp);
1698 /* begin exclusive access */
1699 VVByPListBeginExclusive_r(dp);
1701 /* pick the low-hanging fruit first,
1702 * then do the complicated ones last
1703 * (has the advantage of keeping
1704 * in-use volumes up until the bitter end) */
1705 for (pass = 0, total=0; pass < 4; pass++) {
1706 pass_stats[pass] = ShutdownVByPForPass_r(dp, pass);
1707 total += pass_stats[pass];
1710 /* end exclusive access */
1711 VVByPListEndExclusive_r(dp);
1713 Log("VShutdownByPartition: shut down %d volumes on %s (pass[0]=%d, pass[1]=%d, pass[2]=%d, pass[3]=%d)\n",
1714 total, VPartitionPath(dp), pass_stats[0], pass_stats[1], pass_stats[2], pass_stats[3]);
1719 /* internal shutdown functionality
1721 * for multi-pass shutdown:
1722 * 0 to only "shutdown" {pre,un}attached and error state volumes
1723 * 1 to also shutdown attached volumes w/ volume header loaded
1724 * 2 to also shutdown attached volumes w/o volume header loaded
1725 * 3 to also shutdown exclusive state volumes
1727 * caller MUST hold exclusive access on the hash chain
1728 * because we drop vol_glock_mutex internally
1730 * this function is reentrant for passes 1--3
1731 * (e.g. multiple threads can cooperate to
1732 * shutdown a partition mp-fast)
1734 * pass 0 is not scaleable because the volume state data is
1735 * synchronized by vol_glock mutex, and the locking overhead
1736 * is too high to drop the lock long enough to do linked list
1740 ShutdownVByPForPass_r(struct DiskPartition64 * dp, int pass)
1742 struct rx_queue * q = queue_First(&dp->vol_list, rx_queue);
1745 while (ShutdownVolumeWalk_r(dp, pass, &q))
1751 /* conditionally shutdown one volume on partition dp
1752 * returns 1 if a volume was shutdown in this pass,
1755 ShutdownVolumeWalk_r(struct DiskPartition64 * dp, int pass,
1756 struct rx_queue ** idx)
1758 struct rx_queue *qp, *nqp;
1763 for (queue_ScanFrom(&dp->vol_list, qp, qp, nqp, rx_queue)) {
1764 vp = (Volume *) (((char *)qp) - offsetof(Volume, vol_list));
1768 if ((V_attachState(vp) != VOL_STATE_UNATTACHED) &&
1769 (V_attachState(vp) != VOL_STATE_ERROR) &&
1770 (V_attachState(vp) != VOL_STATE_PREATTACHED)) {
1774 if ((V_attachState(vp) == VOL_STATE_ATTACHED) &&
1775 (vp->header == NULL)) {
1779 if (VIsExclusiveState(V_attachState(vp))) {
1784 DeleteVolumeFromVByPList_r(vp);
1785 VShutdownVolume_r(vp);
1795 * shutdown a specific volume
1797 /* caller MUST NOT hold a heavyweight ref on vp */
1799 VShutdownVolume_r(Volume * vp)
1803 VCreateReservation_r(vp);
1805 if (LogLevel >= 5) {
1806 Log("VShutdownVolume_r: vid=%u, device=%d, state=%hu\n",
1807 vp->hashid, vp->partition->device, V_attachState(vp));
1810 /* wait for other blocking ops to finish */
1811 VWaitExclusiveState_r(vp);
1813 assert(VIsValidState(V_attachState(vp)));
1815 switch(V_attachState(vp)) {
1816 case VOL_STATE_SALVAGING:
1817 /* Leave salvaging volumes alone. Any in-progress salvages will
1818 * continue working after viced shuts down. This is intentional.
1821 case VOL_STATE_PREATTACHED:
1822 case VOL_STATE_ERROR:
1823 VChangeState_r(vp, VOL_STATE_UNATTACHED);
1824 case VOL_STATE_UNATTACHED:
1826 case VOL_STATE_GOING_OFFLINE:
1827 case VOL_STATE_SHUTTING_DOWN:
1828 case VOL_STATE_ATTACHED:
1832 Log("VShutdown: Attempting to take volume %u offline.\n",
1835 /* take the volume offline (drops reference count) */
1836 VOffline_r(vp, "File server was shut down");
1843 VCancelReservation_r(vp);
1847 #endif /* AFS_DEMAND_ATTACH_FS */
1850 /***************************************************/
1851 /* Header I/O routines */
1852 /***************************************************/
1854 /* open a descriptor for the inode (h),
1855 * read in an on-disk structure into buffer (to) of size (size),
1856 * verify versionstamp in structure has magic (magic) and
1857 * optionally verify version (version) if (version) is nonzero
1860 ReadHeader(Error * ec, IHandle_t * h, char *to, int size, bit32 magic,
1863 struct versionStamp *vsn;
1878 if (FDH_SEEK(fdP, 0, SEEK_SET) < 0) {
1880 FDH_REALLYCLOSE(fdP);
1883 vsn = (struct versionStamp *)to;
1884 if (FDH_READ(fdP, to, size) != size || vsn->magic != magic) {
1886 FDH_REALLYCLOSE(fdP);
1891 /* Check is conditional, in case caller wants to inspect version himself */
1892 if (version && vsn->version != version) {
1898 WriteVolumeHeader_r(Error * ec, Volume * vp)
1900 IHandle_t *h = V_diskDataHandle(vp);
1910 if (FDH_SEEK(fdP, 0, SEEK_SET) < 0) {
1912 FDH_REALLYCLOSE(fdP);
1915 if (FDH_WRITE(fdP, (char *)&V_disk(vp), sizeof(V_disk(vp)))
1916 != sizeof(V_disk(vp))) {
1918 FDH_REALLYCLOSE(fdP);
1924 /* VolumeHeaderToDisk
1925 * Allows for storing 64 bit inode numbers in on-disk volume header
1928 /* convert in-memory representation of a volume header to the
1929 * on-disk representation of a volume header */
1931 VolumeHeaderToDisk(VolumeDiskHeader_t * dh, VolumeHeader_t * h)
1934 memset(dh, 0, sizeof(VolumeDiskHeader_t));
1935 dh->stamp = h->stamp;
1937 dh->parent = h->parent;
1939 #ifdef AFS_64BIT_IOPS_ENV
1940 dh->volumeInfo_lo = (afs_int32) h->volumeInfo & 0xffffffff;
1941 dh->volumeInfo_hi = (afs_int32) (h->volumeInfo >> 32) & 0xffffffff;
1942 dh->smallVnodeIndex_lo = (afs_int32) h->smallVnodeIndex & 0xffffffff;
1943 dh->smallVnodeIndex_hi =
1944 (afs_int32) (h->smallVnodeIndex >> 32) & 0xffffffff;
1945 dh->largeVnodeIndex_lo = (afs_int32) h->largeVnodeIndex & 0xffffffff;
1946 dh->largeVnodeIndex_hi =
1947 (afs_int32) (h->largeVnodeIndex >> 32) & 0xffffffff;
1948 dh->linkTable_lo = (afs_int32) h->linkTable & 0xffffffff;
1949 dh->linkTable_hi = (afs_int32) (h->linkTable >> 32) & 0xffffffff;
1951 dh->volumeInfo_lo = h->volumeInfo;
1952 dh->smallVnodeIndex_lo = h->smallVnodeIndex;
1953 dh->largeVnodeIndex_lo = h->largeVnodeIndex;
1954 dh->linkTable_lo = h->linkTable;
1958 /* DiskToVolumeHeader
1959 * Converts an on-disk representation of a volume header to
1960 * the in-memory representation of a volume header.
1962 * Makes the assumption that AFS has *always*
1963 * zero'd the volume header file so that high parts of inode
1964 * numbers are 0 in older (SGI EFS) volume header files.
1967 DiskToVolumeHeader(VolumeHeader_t * h, VolumeDiskHeader_t * dh)
1969 memset(h, 0, sizeof(VolumeHeader_t));
1970 h->stamp = dh->stamp;
1972 h->parent = dh->parent;
1974 #ifdef AFS_64BIT_IOPS_ENV
1976 (Inode) dh->volumeInfo_lo | ((Inode) dh->volumeInfo_hi << 32);
1978 h->smallVnodeIndex =
1979 (Inode) dh->smallVnodeIndex_lo | ((Inode) dh->
1980 smallVnodeIndex_hi << 32);
1982 h->largeVnodeIndex =
1983 (Inode) dh->largeVnodeIndex_lo | ((Inode) dh->
1984 largeVnodeIndex_hi << 32);
1986 (Inode) dh->linkTable_lo | ((Inode) dh->linkTable_hi << 32);
1988 h->volumeInfo = dh->volumeInfo_lo;
1989 h->smallVnodeIndex = dh->smallVnodeIndex_lo;
1990 h->largeVnodeIndex = dh->largeVnodeIndex_lo;
1991 h->linkTable = dh->linkTable_lo;
1996 /***************************************************/
1997 /* Volume Attachment routines */
1998 /***************************************************/
2000 #ifdef AFS_DEMAND_ATTACH_FS
2002 * pre-attach a volume given its path.
2004 * @param[out] ec outbound error code
2005 * @param[in] partition partition path string
2006 * @param[in] name volume id string
2008 * @return volume object pointer
2010 * @note A pre-attached volume will only have its partition
2011 * and hashid fields initialized. At first call to
2012 * VGetVolume, the volume will be fully attached.
2016 VPreAttachVolumeByName(Error * ec, char *partition, char *name)
2020 vp = VPreAttachVolumeByName_r(ec, partition, name);
2026 * pre-attach a volume given its path.
2028 * @param[out] ec outbound error code
2029 * @param[in] partition path to vice partition
2030 * @param[in] name volume id string
2032 * @return volume object pointer
2034 * @pre VOL_LOCK held
2036 * @internal volume package internal use only.
2039 VPreAttachVolumeByName_r(Error * ec, char *partition, char *name)
2041 return VPreAttachVolumeById_r(ec,
2043 VolumeNumber(name));
2047 * pre-attach a volume given its path and numeric volume id.
2049 * @param[out] ec error code return
2050 * @param[in] partition path to vice partition
2051 * @param[in] volumeId numeric volume id
2053 * @return volume object pointer
2055 * @pre VOL_LOCK held
2057 * @internal volume package internal use only.
2060 VPreAttachVolumeById_r(Error * ec,
2065 struct DiskPartition64 *partp;
2069 assert(programType == fileServer);
2071 if (!(partp = VGetPartition_r(partition, 0))) {
2073 Log("VPreAttachVolumeById_r: Error getting partition (%s)\n", partition);
2077 vp = VLookupVolume_r(ec, volumeId, NULL);
2082 return VPreAttachVolumeByVp_r(ec, partp, vp, volumeId);
2086 * preattach a volume.
2088 * @param[out] ec outbound error code
2089 * @param[in] partp pointer to partition object
2090 * @param[in] vp pointer to volume object
2091 * @param[in] vid volume id
2093 * @return volume object pointer
2095 * @pre VOL_LOCK is held.
2097 * @warning Returned volume object pointer does not have to
2098 * equal the pointer passed in as argument vp. There
2099 * are potential race conditions which can result in
2100 * the pointers having different values. It is up to
2101 * the caller to make sure that references are handled
2102 * properly in this case.
2104 * @note If there is already a volume object registered with
2105 * the same volume id, its pointer MUST be passed as
2106 * argument vp. Failure to do so will result in a silent
2107 * failure to preattach.
2109 * @internal volume package internal use only.
2112 VPreAttachVolumeByVp_r(Error * ec,
2113 struct DiskPartition64 * partp,
2121 /* check to see if pre-attach already happened */
2123 (V_attachState(vp) != VOL_STATE_UNATTACHED) &&
2124 (V_attachState(vp) != VOL_STATE_PREATTACHED) &&
2125 !VIsErrorState(V_attachState(vp))) {
2127 * pre-attach is a no-op in all but the following cases:
2129 * - volume is unattached
2130 * - volume is in an error state
2131 * - volume is pre-attached
2133 Log("VPreattachVolumeByVp_r: volume %u not in quiescent state\n", vid);
2136 /* we're re-attaching a volume; clear out some old state */
2137 memset(&vp->salvage, 0, sizeof(struct VolumeOnlineSalvage));
2139 if (V_partition(vp) != partp) {
2140 /* XXX potential race */
2141 DeleteVolumeFromVByPList_r(vp);
2144 /* if we need to allocate a new Volume struct,
2145 * go ahead and drop the vol glock, otherwise
2146 * do the basic setup synchronised, as it's
2147 * probably not worth dropping the lock */
2150 /* allocate the volume structure */
2151 vp = nvp = (Volume *) malloc(sizeof(Volume));
2153 memset(vp, 0, sizeof(Volume));
2154 queue_Init(&vp->vnode_list);
2155 assert(pthread_cond_init(&V_attachCV(vp), NULL) == 0);
2158 /* link the volume with its associated vice partition */
2159 vp->device = partp->device;
2160 vp->partition = partp;
2163 vp->specialStatus = 0;
2165 /* if we dropped the lock, reacquire the lock,
2166 * check for pre-attach races, and then add
2167 * the volume to the hash table */
2170 nvp = VLookupVolume_r(ec, vid, NULL);
2175 } else if (nvp) { /* race detected */
2180 /* hack to make up for VChangeState_r() decrementing
2181 * the old state counter */
2182 VStats.state_levels[0]++;
2186 /* put pre-attached volume onto the hash table
2187 * and bring it up to the pre-attached state */
2188 AddVolumeToHashTable(vp, vp->hashid);
2189 AddVolumeToVByPList_r(vp);
2190 VLRU_Init_Node_r(vp);
2191 VChangeState_r(vp, VOL_STATE_PREATTACHED);
2194 Log("VPreAttachVolumeByVp_r: volume %u pre-attached\n", vp->hashid);
2202 #endif /* AFS_DEMAND_ATTACH_FS */
2204 /* Attach an existing volume, given its pathname, and return a
2205 pointer to the volume header information. The volume also
2206 normally goes online at this time. An offline volume
2207 must be reattached to make it go online */
2209 VAttachVolumeByName(Error * ec, char *partition, char *name, int mode)
2213 retVal = VAttachVolumeByName_r(ec, partition, name, mode);
2219 VAttachVolumeByName_r(Error * ec, char *partition, char *name, int mode)
2221 register Volume *vp = NULL;
2222 struct DiskPartition64 *partp;
2226 #ifdef AFS_DEMAND_ATTACH_FS
2227 VolumeStats stats_save;
2229 #endif /* AFS_DEMAND_ATTACH_FS */
2233 volumeId = VolumeNumber(name);
2235 if (!(partp = VGetPartition_r(partition, 0))) {
2237 Log("VAttachVolume: Error getting partition (%s)\n", partition);
2241 if (VRequiresPartLock()) {
2243 VLockPartition_r(partition);
2244 } else if (programType == fileServer) {
2245 #ifdef AFS_DEMAND_ATTACH_FS
2246 /* lookup the volume in the hash table */
2247 vp = VLookupVolume_r(ec, volumeId, NULL);
2253 /* save any counters that are supposed to
2254 * be monotonically increasing over the
2255 * lifetime of the fileserver */
2256 memcpy(&stats_save, &vp->stats, sizeof(VolumeStats));
2258 memset(&stats_save, 0, sizeof(VolumeStats));
2261 /* if there's something in the hash table, and it's not
2262 * in the pre-attach state, then we may need to detach
2263 * it before proceeding */
2264 if (vp && (V_attachState(vp) != VOL_STATE_PREATTACHED)) {
2265 VCreateReservation_r(vp);
2266 VWaitExclusiveState_r(vp);
2268 /* at this point state must be one of:
2277 if (vp->specialStatus == VBUSY)
2280 /* if it's already attached, see if we can return it */
2281 if (V_attachState(vp) == VOL_STATE_ATTACHED) {
2282 VGetVolumeByVp_r(ec, vp);
2283 if (V_inUse(vp) == fileServer) {
2284 VCancelReservation_r(vp);
2288 /* otherwise, we need to detach, and attempt to re-attach */
2289 VDetachVolume_r(ec, vp);
2291 Log("VAttachVolume: Error detaching old volume instance (%s)\n", name);
2294 /* if it isn't fully attached, delete from the hash tables,
2295 and let the refcounter handle the rest */
2296 DeleteVolumeFromHashTable(vp);
2297 DeleteVolumeFromVByPList_r(vp);
2300 VCancelReservation_r(vp);
2304 /* pre-attach volume if it hasn't been done yet */
2306 (V_attachState(vp) == VOL_STATE_UNATTACHED) ||
2307 (V_attachState(vp) == VOL_STATE_ERROR)) {
2309 vp = VPreAttachVolumeByVp_r(ec, partp, vp, volumeId);
2317 /* handle pre-attach races
2319 * multiple threads can race to pre-attach a volume,
2320 * but we can't let them race beyond that
2322 * our solution is to let the first thread to bring
2323 * the volume into an exclusive state win; the other
2324 * threads just wait until it finishes bringing the
2325 * volume online, and then they do a vgetvolumebyvp
2327 if (svp && (svp != vp)) {
2328 /* wait for other exclusive ops to finish */
2329 VCreateReservation_r(vp);
2330 VWaitExclusiveState_r(vp);
2332 /* get a heavyweight ref, kill the lightweight ref, and return */
2333 VGetVolumeByVp_r(ec, vp);
2334 VCancelReservation_r(vp);
2338 /* at this point, we are chosen as the thread to do
2339 * demand attachment for this volume. all other threads
2340 * doing a getvolume on vp->hashid will block until we finish */
2342 /* make sure any old header cache entries are invalidated
2343 * before proceeding */
2344 FreeVolumeHeader(vp);
2346 VChangeState_r(vp, VOL_STATE_ATTACHING);
2348 /* restore any saved counters */
2349 memcpy(&vp->stats, &stats_save, sizeof(VolumeStats));
2350 #else /* AFS_DEMAND_ATTACH_FS */
2351 vp = VGetVolume_r(ec, volumeId);
2353 if (V_inUse(vp) == fileServer)
2355 if (vp->specialStatus == VBUSY)
2357 VDetachVolume_r(ec, vp);
2359 Log("VAttachVolume: Error detaching volume (%s)\n", name);
2363 #endif /* AFS_DEMAND_ATTACH_FS */
2367 strcpy(path, VPartitionPath(partp));
2375 vp = (Volume *) calloc(1, sizeof(Volume));
2377 vp->hashid = volumeId;
2378 vp->device = partp->device;
2379 vp->partition = partp;
2380 queue_Init(&vp->vnode_list);
2381 #ifdef AFS_DEMAND_ATTACH_FS
2382 assert(pthread_cond_init(&V_attachCV(vp), NULL) == 0);
2383 #endif /* AFS_DEMAND_ATTACH_FS */
2386 /* attach2 is entered without any locks, and returns
2387 * with vol_glock_mutex held */
2388 vp = attach2(ec, volumeId, path, partp, vp, isbusy, mode);
2390 if (VCanUseFSSYNC() && vp) {
2391 #ifdef AFS_DEMAND_ATTACH_FS
2392 if ((mode == V_VOLUPD) || (VolumeWriteable(vp) && (mode == V_CLONE))) {
2393 /* mark volume header as in use so that volser crashes lead to a
2394 * salvage attempt */
2395 VUpdateVolume_r(ec, vp, 0);
2397 /* for dafs, we should tell the fileserver, except for V_PEEK
2398 * where we know it is not necessary */
2399 if (mode == V_PEEK) {
2400 vp->needsPutBack = 0;
2402 vp->needsPutBack = 1;
2404 #else /* !AFS_DEMAND_ATTACH_FS */
2405 /* duplicate computation in fssync.c about whether the server
2406 * takes the volume offline or not. If the volume isn't
2407 * offline, we must not return it when we detach the volume,
2408 * or the server will abort */
2409 if (mode == V_READONLY || mode == V_PEEK
2410 || (!VolumeWriteable(vp) && (mode == V_CLONE || mode == V_DUMP)))
2411 vp->needsPutBack = 0;
2413 vp->needsPutBack = 1;
2414 #endif /* !AFS_DEMAND_ATTACH_FS */
2416 /* OK, there's a problem here, but one that I don't know how to
2417 * fix right now, and that I don't think should arise often.
2418 * Basically, we should only put back this volume to the server if
2419 * it was given to us by the server, but since we don't have a vp,
2420 * we can't run the VolumeWriteable function to find out as we do
2421 * above when computing vp->needsPutBack. So we send it back, but
2422 * there's a path in VAttachVolume on the server which may abort
2423 * if this volume doesn't have a header. Should be pretty rare
2424 * for all of that to happen, but if it does, probably the right
2425 * fix is for the server to allow the return of readonly volumes
2426 * that it doesn't think are really checked out. */
2427 #ifdef FSSYNC_BUILD_CLIENT
2428 if (VCanUseFSSYNC() && vp == NULL &&
2429 mode != V_SECRETLY && mode != V_PEEK) {
2431 #ifdef AFS_DEMAND_ATTACH_FS
2432 /* If we couldn't attach but we scheduled a salvage, we already
2433 * notified the fileserver; don't online it now */
2434 if (*ec != VSALVAGING)
2435 #endif /* AFS_DEMAND_ATTACH_FS */
2436 FSYNC_VolOp(volumeId, partition, FSYNC_VOL_ON, 0, NULL);
2439 if (programType == fileServer && vp) {
2440 #ifdef AFS_DEMAND_ATTACH_FS
2442 * we can get here in cases where we don't "own"
2443 * the volume (e.g. volume owned by a utility).
2444 * short circuit around potential disk header races.
2446 if (V_attachState(vp) != VOL_STATE_ATTACHED) {
2450 V_needsCallback(vp) = 0;
2452 if (VInit >= 2 && V_BreakVolumeCallbacks) {
2453 Log("VAttachVolume: Volume %u was changed externally; breaking callbacks\n", V_id(vp));
2454 (*V_BreakVolumeCallbacks) (V_id(vp));
2457 VUpdateVolume_r(ec, vp, 0);
2459 Log("VAttachVolume: Error updating volume\n");
2464 if (VolumeWriteable(vp) && V_dontSalvage(vp) == 0) {
2465 #ifndef AFS_DEMAND_ATTACH_FS
2466 /* This is a hack: by temporarily setting the incore
2467 * dontSalvage flag ON, the volume will be put back on the
2468 * Update list (with dontSalvage OFF again). It will then
2469 * come back in N minutes with DONT_SALVAGE eventually
2470 * set. This is the way that volumes that have never had
2471 * it set get it set; or that volumes that have been
2472 * offline without DONT SALVAGE having been set also
2473 * eventually get it set */
2474 V_dontSalvage(vp) = DONT_SALVAGE;
2475 #endif /* !AFS_DEMAND_ATTACH_FS */
2476 VAddToVolumeUpdateList_r(ec, vp);
2478 Log("VAttachVolume: Error adding volume to update list\n");
2485 Log("VOnline: volume %u (%s) attached and online\n", V_id(vp),
2490 if (VRequiresPartLock()) {
2491 VUnlockPartition_r(partition);
2494 #ifdef AFS_DEMAND_ATTACH_FS
2495 /* attach failed; make sure we're in error state */
2496 if (vp && !VIsErrorState(V_attachState(vp))) {
2497 VChangeState_r(vp, VOL_STATE_ERROR);
2499 #endif /* AFS_DEMAND_ATTACH_FS */
2506 #ifdef AFS_DEMAND_ATTACH_FS
2507 /* VAttachVolumeByVp_r
2509 * finish attaching a volume that is
2510 * in a less than fully attached state
2512 /* caller MUST hold a ref count on vp */
2514 VAttachVolumeByVp_r(Error * ec, Volume * vp, int mode)
2516 char name[VMAXPATHLEN];
2518 struct DiskPartition64 *partp;
2522 Volume * nvp = NULL;
2523 VolumeStats stats_save;
2526 /* volume utility should never call AttachByVp */
2527 assert(programType == fileServer);
2529 volumeId = vp->hashid;
2530 partp = vp->partition;
2531 VolumeExternalName_r(volumeId, name, sizeof(name));
2534 /* if another thread is performing a blocking op, wait */
2535 VWaitExclusiveState_r(vp);
2537 memcpy(&stats_save, &vp->stats, sizeof(VolumeStats));
2539 /* if it's already attached, see if we can return it */
2540 if (V_attachState(vp) == VOL_STATE_ATTACHED) {
2541 VGetVolumeByVp_r(ec, vp);
2542 if (V_inUse(vp) == fileServer) {
2545 if (vp->specialStatus == VBUSY)
2547 VDetachVolume_r(ec, vp);
2549 Log("VAttachVolume: Error detaching volume (%s)\n", name);
2555 /* pre-attach volume if it hasn't been done yet */
2557 (V_attachState(vp) == VOL_STATE_UNATTACHED) ||
2558 (V_attachState(vp) == VOL_STATE_ERROR)) {
2559 nvp = VPreAttachVolumeByVp_r(ec, partp, vp, volumeId);
2565 VCreateReservation_r(nvp);
2571 VChangeState_r(vp, VOL_STATE_ATTACHING);
2573 /* restore monotonically increasing stats */
2574 memcpy(&vp->stats, &stats_save, sizeof(VolumeStats));
2578 /* compute path to disk header */
2579 strcpy(path, VPartitionPath(partp));
2588 * NOTE: attach2 is entered without any locks, and returns
2589 * with vol_glock_mutex held */
2590 vp = attach2(ec, volumeId, path, partp, vp, isbusy, mode);
2593 * the event that an error was encountered, or
2594 * the volume was not brought to an attached state
2595 * for any reason, skip to the end. We cannot
2596 * safely call VUpdateVolume unless we "own" it.
2600 (V_attachState(vp) != VOL_STATE_ATTACHED)) {
2604 V_needsCallback(vp) = 0;
2605 VUpdateVolume_r(ec, vp, 0);
2607 Log("VAttachVolume: Error updating volume %u\n", vp->hashid);
2611 if (VolumeWriteable(vp) && V_dontSalvage(vp) == 0) {
2612 #ifndef AFS_DEMAND_ATTACH_FS
2613 /* This is a hack: by temporarily setting the incore
2614 * dontSalvage flag ON, the volume will be put back on the
2615 * Update list (with dontSalvage OFF again). It will then
2616 * come back in N minutes with DONT_SALVAGE eventually
2617 * set. This is the way that volumes that have never had
2618 * it set get it set; or that volumes that have been
2619 * offline without DONT SALVAGE having been set also
2620 * eventually get it set */
2621 V_dontSalvage(vp) = DONT_SALVAGE;
2622 #endif /* !AFS_DEMAND_ATTACH_FS */
2623 VAddToVolumeUpdateList_r(ec, vp);
2625 Log("VAttachVolume: Error adding volume %u to update list\n", vp->hashid);
2632 Log("VOnline: volume %u (%s) attached and online\n", V_id(vp),
2636 VCancelReservation_r(nvp);
2639 if (*ec && (*ec != VOFFLINE) && (*ec != VSALVAGE)) {
2640 if (vp && !VIsErrorState(V_attachState(vp))) {
2641 VChangeState_r(vp, VOL_STATE_ERROR);
2650 * lock a volume on disk (non-blocking).
2652 * @param[in] vp The volume to lock
2653 * @param[in] locktype READ_LOCK or WRITE_LOCK
2655 * @return operation status
2656 * @retval 0 success, lock was obtained
2657 * @retval EBUSY a conflicting lock was held by another process
2658 * @retval EIO error acquiring lock
2660 * @pre If we're in the fileserver, vp is in an exclusive state
2662 * @pre vp is not already locked
2665 VLockVolumeNB(Volume *vp, int locktype)
2669 assert(programType != fileServer || VIsExclusiveState(V_attachState(vp)));
2670 assert(!(V_attachFlags(vp) & VOL_LOCKED));
2672 code = VLockVolumeByIdNB(vp->hashid, vp->partition, locktype);
2674 V_attachFlags(vp) |= VOL_LOCKED;
2681 * unlock a volume on disk that was locked with VLockVolumeNB.
2683 * @param[in] vp volume to unlock
2685 * @pre If we're in the fileserver, vp is in an exclusive state
2687 * @pre vp has already been locked
2690 VUnlockVolume(Volume *vp)
2692 assert(programType != fileServer || VIsExclusiveState(V_attachState(vp)));
2693 assert((V_attachFlags(vp) & VOL_LOCKED));
2695 VUnlockVolumeById(vp->hashid, vp->partition);
2697 V_attachFlags(vp) &= ~VOL_LOCKED;
2699 #endif /* AFS_DEMAND_ATTACH_FS */
2702 * read in a vol header, possibly lock the vol header, and possibly check out
2703 * the vol header from the fileserver, as part of volume attachment.
2705 * @param[out] ec error code
2706 * @param[in] vp volume pointer object
2707 * @param[in] partp disk partition object of the attaching partition
2708 * @param[in] mode attachment mode such as V_VOLUPD, V_DUMP, etc (see
2710 * @param[in] peek 1 to just try to read in the volume header and make sure
2711 * we don't try to lock the vol, or check it out from
2712 * FSSYNC or anything like that; 0 otherwise, for 'normal'
2715 * @note As part of DAFS volume attachment, the volume header may be either
2716 * read- or write-locked to ensure mutual exclusion of certain volume
2717 * operations. In some cases in order to determine whether we need to
2718 * read- or write-lock the header, we need to read in the header to see
2719 * if the volume is RW or not. So, if we read in the header under a
2720 * read-lock and determine that we actually need a write-lock on the
2721 * volume header, this function will drop the read lock, acquire a write
2722 * lock, and read the header in again.
2725 attach_volume_header(Error *ec, Volume *vp, struct DiskPartition64 *partp,
2728 struct VolumeDiskHeader diskHeader;
2729 struct VolumeHeader header;
2732 int lock_tries = 0, checkout_tries = 0;
2734 VolumeId volid = vp->hashid;
2735 #ifdef FSSYNC_BUILD_CLIENT
2736 int checkout, done_checkout = 0;
2737 #endif /* FSSYNC_BUILD_CLIENT */
2738 #ifdef AFS_DEMAND_ATTACH_FS
2739 int locktype = 0, use_locktype = -1;
2740 #endif /* AFS_DEMAND_ATTACH_FS */
2746 if (lock_tries > VOL_MAX_CHECKOUT_RETRIES) {
2747 Log("VAttachVolume: retried too many times trying to lock header for "
2748 "vol %lu part %s; giving up\n", afs_printable_uint32_lu(volid),
2749 VPartitionPath(partp));
2753 if (checkout_tries > VOL_MAX_CHECKOUT_RETRIES) {
2754 Log("VAttachVolume: retried too many times trying to checkout "
2755 "vol %lu part %s; giving up\n", afs_printable_uint32_lu(volid),
2756 VPartitionPath(partp));
2761 if (VReadVolumeDiskHeader(volid, partp, NULL)) {
2762 /* short-circuit the 'volume does not exist' case */
2767 #ifdef FSSYNC_BUILD_CLIENT
2768 checkout = !done_checkout;
2770 if (!peek && checkout && VMustCheckoutVolume(mode)) {
2772 memset(&res, 0, sizeof(res));
2774 if (FSYNC_VolOp(volid, VPartitionPath(partp), FSYNC_VOL_NEEDVOLUME, mode, &res)
2777 if (res.hdr.reason == FSYNC_SALVAGE) {
2778 Log("VAttachVolume: file server says volume %lu is salvaging\n",
2779 afs_printable_uint32_lu(volid));
2782 Log("VAttachVolume: attach of volume %lu apparently denied by file server\n",
2783 afs_printable_uint32_lu(volid));
2784 *ec = VNOVOL; /* XXXX */
2791 #ifdef AFS_DEMAND_ATTACH_FS
2792 if (use_locktype < 0) {
2793 /* don't know whether vol is RO or RW; assume it's RO and we can retry
2794 * if it turns out to be RW */
2795 locktype = VVolLockType(mode, 0);
2798 /* a previous try says we should use use_locktype to lock the volume,
2800 locktype = use_locktype;
2803 if (!peek && locktype) {
2804 code = VLockVolumeNB(vp, locktype);
2806 if (code == EBUSY) {
2807 Log("VAttachVolume: another program has vol %lu locked\n",
2808 afs_printable_uint32_lu(volid));
2810 Log("VAttachVolume: error %d trying to lock vol %lu\n",
2811 code, afs_printable_uint32_lu(volid));
2818 #endif /* AFS_DEMAND_ATTACH_FS */
2820 code = VReadVolumeDiskHeader(volid, partp, &diskHeader);
2830 DiskToVolumeHeader(&header, &diskHeader);
2832 IH_INIT(vp->vnodeIndex[vLarge].handle, partp->device, header.parent,
2833 header.largeVnodeIndex);
2834 IH_INIT(vp->vnodeIndex[vSmall].handle, partp->device, header.parent,
2835 header.smallVnodeIndex);
2836 IH_INIT(vp->diskDataHandle, partp->device, header.parent,
2838 IH_INIT(vp->linkHandle, partp->device, header.parent, header.linkTable);
2841 /* only need to do this once */
2843 GetVolumeHeader(vp);
2847 #if defined(AFS_DEMAND_ATTACH_FS) && defined(FSSYNC_BUILD_CLIENT)
2848 /* demand attach changes the V_PEEK mechanism
2850 * we can now suck the current disk data structure over
2851 * the fssync interface without going to disk
2853 * (technically, we don't need to restrict this feature
2854 * to demand attach fileservers. However, I'm trying
2855 * to limit the number of common code changes)
2857 if (VCanUseFSSYNC() && (mode == V_PEEK || peek)) {
2859 res.payload.len = sizeof(VolumeDiskData);
2860 res.payload.buf = &vp->header->diskstuff;
2862 if (FSYNC_VolOp(vp->hashid,
2864 FSYNC_VOL_QUERY_HDR,
2867 goto disk_header_loaded;
2870 #endif /* AFS_DEMAND_ATTACH_FS && FSSYNC_BUILD_CLIENT */
2871 (void)ReadHeader(ec, V_diskDataHandle(vp), (char *)&V_disk(vp),
2872 sizeof(V_disk(vp)), VOLUMEINFOMAGIC, VOLUMEINFOVERSION);
2874 #ifdef AFS_DEMAND_ATTACH_FS
2877 IncUInt64(&VStats.hdr_loads);
2878 IncUInt64(&vp->stats.hdr_loads);
2880 #endif /* AFS_DEMAND_ATTACH_FS */
2883 Log("VAttachVolume: Error reading diskDataHandle header for vol %lu; "
2884 "error=%u\n", afs_printable_uint32_lu(volid), *ec);
2888 #ifdef AFS_DEMAND_ATTACH_FS
2889 # ifdef FSSYNC_BUILD_CLIENT
2891 # endif /* FSSYNC_BUILD_CLIENT */
2893 /* if the lock type we actually used to lock the volume is different than
2894 * the lock type we should have used, retry with the lock type we should
2896 use_locktype = VVolLockType(mode, VolumeWriteable(vp));
2897 if (locktype != use_locktype) {
2901 #endif /* AFS_DEMAND_ATTACH_FS */
2906 #if defined(AFS_DEMAND_ATTACH_FS) && defined(FSSYNC_BUILD_CLIENT)
2907 if (!peek && *ec == 0 && retry == 0 && VMustCheckoutVolume(mode)) {
2909 code = FSYNC_VerifyCheckout(volid, VPartitionPath(partp), FSYNC_VOL_NEEDVOLUME, mode);
2911 if (code == SYNC_DENIED) {
2912 /* must retry checkout; fileserver no longer thinks we have
2918 } else if (code != SYNC_OK) {
2922 #endif /* AFS_DEMAND_ATTACH_FS && FSSYNC_BUILD_CLIENT */
2925 /* either we are going to be called again for a second pass, or we
2926 * encountered an error; clean up in either case */
2928 #ifdef AFS_DEMAND_ATTACH_FS
2929 if ((V_attachFlags(vp) & VOL_LOCKED)) {
2932 #endif /* AFS_DEMAND_ATTACH_FS */
2933 if (vp->linkHandle) {
2934 IH_RELEASE(vp->vnodeIndex[vLarge].handle);
2935 IH_RELEASE(vp->vnodeIndex[vSmall].handle);
2936 IH_RELEASE(vp->diskDataHandle);
2937 IH_RELEASE(vp->linkHandle);
2950 #ifdef AFS_DEMAND_ATTACH_FS
2952 attach_check_vop(Error *ec, VolumeId volid, struct DiskPartition64 *partp,
2957 if (vp->pending_vol_op) {
2961 if (vp->pending_vol_op->vol_op_state == FSSYNC_VolOpRunningUnknown) {
2963 code = VVolOpLeaveOnlineNoHeader_r(vp, vp->pending_vol_op);
2965 vp->pending_vol_op->vol_op_state = FSSYNC_VolOpRunningOnline;
2966 } else if (code == 0) {
2967 vp->pending_vol_op->vol_op_state = FSSYNC_VolOpRunningOffline;
2970 /* we need the vol header to determine if the volume can be
2971 * left online for the vop, so... get the header */
2975 /* attach header with peek=1 to avoid checking out the volume
2976 * or locking it; we just want the header info, we're not
2977 * messing with the volume itself at all */
2978 attach_volume_header(ec, vp, partp, V_PEEK, 1);
2985 if (VVolOpLeaveOnline_r(vp, vp->pending_vol_op)) {
2986 vp->pending_vol_op->vol_op_state = FSSYNC_VolOpRunningOnline;
2988 vp->pending_vol_op->vol_op_state = FSSYNC_VolOpRunningOffline;
2991 /* make sure we grab a new vol header and re-open stuff on
2992 * actual attachment; we can't keep the data we grabbed, since
2993 * it was not done under a lock and thus not safe */
2994 FreeVolumeHeader(vp);
2995 VReleaseVolumeHandles_r(vp);
2998 /* see if the pending volume op requires exclusive access */
2999 switch (vp->pending_vol_op->vol_op_state) {
3000 case FSSYNC_VolOpPending:
3001 /* this should never happen */
3002 assert(vp->pending_vol_op->vol_op_state != FSSYNC_VolOpPending);
3005 case FSSYNC_VolOpRunningUnknown:
3006 /* this should never happen; we resolved 'unknown' above */
3007 assert(vp->pending_vol_op->vol_op_state != FSSYNC_VolOpRunningUnknown);
3010 case FSSYNC_VolOpRunningOffline:
3011 /* mark the volume down */
3013 VChangeState_r(vp, VOL_STATE_UNATTACHED);
3015 /* do not set V_offlineMessage here; we don't have ownership of
3016 * the volume (and probably do not have the header loaded), so we
3017 * can't alter the disk header */
3019 /* check to see if we should set the specialStatus flag */
3020 if (VVolOpSetVBusy_r(vp, vp->pending_vol_op)) {
3021 vp->specialStatus = VBUSY;
3032 #endif /* AFS_DEMAND_ATTACH_FS */
3035 * volume attachment helper function.
3037 * @param[out] ec error code
3038 * @param[in] volumeId volume ID of the attaching volume
3039 * @param[in] path full path to the volume header .vol file
3040 * @param[in] partp disk partition object for the attaching partition
3041 * @param[in] vp volume object; vp->hashid, vp->device, vp->partition,
3042 * vp->vnode_list, and V_attachCV (for DAFS) should already
3044 * @param[in] isbusy 1 if vp->specialStatus should be set to VBUSY; that is,
3045 * if there is a volume operation running for this volume
3046 * that should set the volume to VBUSY during its run. 0
3047 * otherwise. (see VVolOpSetVBusy_r)
3048 * @param[in] mode attachment mode such as V_VOLUPD, V_DUMP, etc (see
3051 * @return pointer to the semi-attached volume pointer
3052 * @retval NULL an error occurred (check value of *ec)
3053 * @retval vp volume successfully attaching
3055 * @pre no locks held
3057 * @post VOL_LOCK held
3060 attach2(Error * ec, VolId volumeId, char *path, struct DiskPartition64 *partp,
3061 Volume * vp, int isbusy, int mode)
3063 /* have we read in the header successfully? */
3064 int read_header = 0;
3066 /* should we FreeVolume(vp) instead of VCheckFree(vp) in the error
3072 vp->vnodeIndex[vLarge].handle = NULL;
3073 vp->vnodeIndex[vSmall].handle = NULL;
3074 vp->diskDataHandle = NULL;
3075 vp->linkHandle = NULL;
3077 #ifdef AFS_DEMAND_ATTACH_FS
3078 attach_check_vop(ec, volumeId, partp, vp);
3080 attach_volume_header(ec, vp, partp, mode, 0);
3083 attach_volume_header(ec, vp, partp, mode, 0);
3084 #endif /* !AFS_DEMAND_ATTACH_FS */
3086 if (*ec == VNOVOL) {
3087 /* if the volume doesn't exist, skip straight to 'error' so we don't
3088 * request a salvage */
3095 vp->specialStatus = (byte) (isbusy ? VBUSY : 0);
3096 vp->shuttingDown = 0;
3097 vp->goingOffline = 0;
3099 #ifdef AFS_DEMAND_ATTACH_FS
3100 vp->stats.last_attach = FT_ApproxTime();
3101 vp->stats.attaches++;
3105 IncUInt64(&VStats.attaches);
3106 vp->cacheCheck = ++VolumeCacheCheck;
3107 /* just in case this ever rolls over */
3108 if (!vp->cacheCheck)
3109 vp->cacheCheck = ++VolumeCacheCheck;
3112 #ifdef AFS_DEMAND_ATTACH_FS
3113 V_attachFlags(vp) |= VOL_HDR_LOADED;
3114 vp->stats.last_hdr_load = vp->stats.last_attach;
3115 #endif /* AFS_DEMAND_ATTACH_FS */
3119 struct IndexFileHeader iHead;
3121 #if OPENAFS_VOL_STATS
3123 * We just read in the diskstuff part of the header. If the detailed
3124 * volume stats area has not yet been initialized, we should bzero the
3125 * area and mark it as initialized.
3127 if (!(V_stat_initialized(vp))) {
3128 memset((V_stat_area(vp)), 0, VOL_STATS_BYTES);
3129 V_stat_initialized(vp) = 1;
3131 #endif /* OPENAFS_VOL_STATS */
3133 (void)ReadHeader(ec, vp->vnodeIndex[vSmall].handle,
3134 (char *)&iHead, sizeof(iHead),
3135 SMALLINDEXMAGIC, SMALLINDEXVERSION);
3138 Log("VAttachVolume: Error reading smallVnode vol header %s; error=%u\n", path, *ec);
3143 struct IndexFileHeader iHead;
3145 (void)ReadHeader(ec, vp->vnodeIndex[vLarge].handle,
3146 (char *)&iHead, sizeof(iHead),
3147 LARGEINDEXMAGIC, LARGEINDEXVERSION);
3150 Log("VAttachVolume: Error reading largeVnode vol header %s; error=%u\n", path, *ec);
3154 #ifdef AFS_NAMEI_ENV
3156 struct versionStamp stamp;
3158 (void)ReadHeader(ec, V_linkHandle(vp), (char *)&stamp,
3159 sizeof(stamp), LINKTABLEMAGIC, LINKTABLEVERSION);
3162 Log("VAttachVolume: Error reading namei vol header %s; error=%u\n", path, *ec);
3165 #endif /* AFS_NAMEI_ENV */
3167 #if defined(AFS_DEMAND_ATTACH_FS)
3168 if (*ec && ((*ec != VOFFLINE) || (V_attachState(vp) != VOL_STATE_UNATTACHED))) {
3170 if (!VCanScheduleSalvage()) {
3171 Log("VAttachVolume: Error attaching volume %s; volume needs salvage; error=%u\n", path, *ec);
3173 VRequestSalvage_r(ec, vp, SALVSYNC_ERROR, VOL_SALVAGE_INVALIDATE_HEADER);
3178 /* volume operation in progress */
3182 #else /* AFS_DEMAND_ATTACH_FS */
3184 Log("VAttachVolume: Error attaching volume %s; volume needs salvage; error=%u\n", path, *ec);
3188 #endif /* AFS_DEMAND_ATTACH_FS */
3190 if (V_needsSalvaged(vp)) {
3191 if (vp->specialStatus)
3192 vp->specialStatus = 0;
3194 #if defined(AFS_DEMAND_ATTACH_FS)
3195 if (!VCanScheduleSalvage()) {
3196 Log("VAttachVolume: volume salvage flag is ON for %s; volume needs salvage\n", path);
3198 VRequestSalvage_r(ec, vp, SALVSYNC_NEEDED, VOL_SALVAGE_INVALIDATE_HEADER);
3201 #else /* AFS_DEMAND_ATTACH_FS */
3203 #endif /* AFS_DEMAND_ATTACH_FS */
3209 vp->nextVnodeUnique = V_uniquifier(vp);
3210 if (VShouldCheckInUse(mode)) {
3211 #ifndef FAST_RESTART
3212 if (V_inUse(vp) && VolumeWriteable(vp)) {
3213 if (!V_needsSalvaged(vp)) {
3214 V_needsSalvaged(vp) = 1;
3215 VUpdateVolume_r(ec, vp, 0);
3217 #if defined(AFS_DEMAND_ATTACH_FS)
3218 if (!VCanScheduleSalvage()) {
3219 Log("VAttachVolume: volume %s needs to be salvaged; not attached.\n", path);
3221 VRequestSalvage_r(ec, vp, SALVSYNC_NEEDED, VOL_SALVAGE_INVALIDATE_HEADER);
3224 #else /* AFS_DEMAND_ATTACH_FS */
3225 Log("VAttachVolume: volume %s needs to be salvaged; not attached.\n", path);
3227 #endif /* AFS_DEMAND_ATTACH_FS */
3231 #endif /* FAST_RESTART */
3233 if (programType == fileServer && V_destroyMe(vp) == DESTROY_ME) {
3234 /* Only check destroyMe if we are the fileserver, since the
3235 * volserver et al sometimes need to work with volumes with
3236 * destroyMe set. Examples are 'temporary' volumes the
3237 * volserver creates, and when we create a volume (destroyMe
3238 * is set on creation; sometimes a separate volserver
3239 * transaction is created to clear destroyMe).
3242 #if defined(AFS_DEMAND_ATTACH_FS)
3243 /* schedule a salvage so the volume goes away on disk */
3244 VRequestSalvage_r(ec, vp, SALVSYNC_ERROR, VOL_SALVAGE_INVALIDATE_HEADER);
3245 VChangeState_r(vp, VOL_STATE_ERROR);
3247 #endif /* AFS_DEMAND_ATTACH_FS */
3248 Log("VAttachVolume: volume %s is junk; it should be destroyed at next salvage\n", path);
3255 vp->vnodeIndex[vSmall].bitmap = vp->vnodeIndex[vLarge].bitmap = NULL;
3256 #ifndef BITMAP_LATER
3257 if (programType == fileServer && VolumeWriteable(vp)) {
3259 for (i = 0; i < nVNODECLASSES; i++) {
3260 VGetBitmap_r(ec, vp, i);
3262 #ifdef AFS_DEMAND_ATTACH_FS
3263 VRequestSalvage_r(ec, vp, SALVSYNC_ERROR, VOL_SALVAGE_INVALIDATE_HEADER);
3265 #endif /* AFS_DEMAND_ATTACH_FS */
3266 Log("VAttachVolume: error getting bitmap for volume (%s)\n",
3272 #endif /* BITMAP_LATER */
3274 if (programType == fileServer) {
3275 if (vp->specialStatus)
3276 vp->specialStatus = 0;
3277 if (V_blessed(vp) && V_inService(vp) && !V_needsSalvaged(vp)) {
3278 V_inUse(vp) = fileServer;
3279 V_offlineMessage(vp)[0] = '\0';
3282 #ifdef AFS_DEMAND_ATTACH_FS
3283 if ((mode != V_PEEK) && (mode != V_SECRETLY))
3284 V_inUse(vp) = programType;
3285 #endif /* AFS_DEMAND_ATTACH_FS */
3286 V_checkoutMode(vp) = mode;
3289 AddVolumeToHashTable(vp, V_id(vp));
3290 #ifdef AFS_DEMAND_ATTACH_FS
3291 if (VCanUnlockAttached() && (V_attachFlags(vp) & VOL_LOCKED)) {
3294 if ((programType != fileServer) ||
3295 (V_inUse(vp) == fileServer)) {
3296 AddVolumeToVByPList_r(vp);
3298 VChangeState_r(vp, VOL_STATE_ATTACHED);
3300 VChangeState_r(vp, VOL_STATE_UNATTACHED);
3307 #ifdef AFS_DEMAND_ATTACH_FS
3308 if (!VIsErrorState(V_attachState(vp))) {
3309 VChangeState_r(vp, VOL_STATE_ERROR);
3311 #endif /* AFS_DEMAND_ATTACH_FS */
3314 VReleaseVolumeHandles_r(vp);
3317 #ifdef AFS_DEMAND_ATTACH_FS
3324 #else /* !AFS_DEMAND_ATTACH_FS */
3326 #endif /* !AFS_DEMAND_ATTACH_FS */
3330 /* Attach an existing volume.
3331 The volume also normally goes online at this time.
3332 An offline volume must be reattached to make it go online.
3336 VAttachVolume(Error * ec, VolumeId volumeId, int mode)
3340 retVal = VAttachVolume_r(ec, volumeId, mode);
3346 VAttachVolume_r(Error * ec, VolumeId volumeId, int mode)
3349 VGetVolumePath(ec, volumeId, &part, &name);
3351 register Volume *vp;
3353 vp = VGetVolume_r(&error, volumeId);
3355 assert(V_inUse(vp) == 0);
3356 VDetachVolume_r(ec, vp);
3360 return VAttachVolumeByName_r(ec, part, name, mode);
3363 /* Increment a reference count to a volume, sans context swaps. Requires
3364 * possibly reading the volume header in from the disk, since there's
3365 * an invariant in the volume package that nUsers>0 ==> vp->header is valid.
3367 * N.B. This call can fail if we can't read in the header!! In this case
3368 * we still guarantee we won't context swap, but the ref count won't be
3369 * incremented (otherwise we'd violate the invariant).
3371 /* NOTE: with the demand attach fileserver extensions, the global lock
3372 * is dropped within VHold */
3373 #ifdef AFS_DEMAND_ATTACH_FS
3375 VHold_r(register Volume * vp)
3379 VCreateReservation_r(vp);
3380 VWaitExclusiveState_r(vp);
3382 LoadVolumeHeader(&error, vp);
3384 VCancelReservation_r(vp);
3388 VCancelReservation_r(vp);
3391 #else /* AFS_DEMAND_ATTACH_FS */
3393 VHold_r(register Volume * vp)
3397 LoadVolumeHeader(&error, vp);
3403 #endif /* AFS_DEMAND_ATTACH_FS */
3407 VHold(register Volume * vp)
3411 retVal = VHold_r(vp);
3418 /***************************************************/
3419 /* get and put volume routines */
3420 /***************************************************/
3423 * put back a heavyweight reference to a volume object.
3425 * @param[in] vp volume object pointer
3427 * @pre VOL_LOCK held
3429 * @post heavyweight volume reference put back.
3430 * depending on state, volume may have been taken offline,
3431 * detached, salvaged, freed, etc.
3433 * @internal volume package internal use only
3436 VPutVolume_r(register Volume * vp)
3438 assert(--vp->nUsers >= 0);
3439 if (vp->nUsers == 0) {
3441 ReleaseVolumeHeader(vp->header);
3442 #ifdef AFS_DEMAND_ATTACH_FS
3443 if (!VCheckDetach(vp)) {
3447 #else /* AFS_DEMAND_ATTACH_FS */
3449 #endif /* AFS_DEMAND_ATTACH_FS */
3454 VPutVolume(register Volume * vp)
3462 /* Get a pointer to an attached volume. The pointer is returned regardless
3463 of whether or not the volume is in service or on/off line. An error
3464 code, however, is returned with an indication of the volume's status */
3466 VGetVolume(Error * ec, Error * client_ec, VolId volumeId)
3470 retVal = GetVolume(ec, client_ec, volumeId, NULL, 0);
3476 VGetVolume_r(Error * ec, VolId volumeId)
3478 return GetVolume(ec, NULL, volumeId, NULL, 0);
3481 /* try to get a volume we've previously looked up */
3482 /* for demand attach fs, caller MUST NOT hold a ref count on vp */
3484 VGetVolumeByVp_r(Error * ec, Volume * vp)
3486 return GetVolume(ec, NULL, vp->hashid, vp, 0);
3490 * private interface for getting a volume handle
3492 * @param[out] ec error code (0 if no error)
3493 * @param[out] client_ec wire error code to be given to clients
3494 * @param[in] volumeId ID of the volume we want
3495 * @param[in] hint optional hint for hash lookups, or NULL
3496 * @param[in] flags unused; always 0
3498 * @return a volume handle for the specified volume
3499 * @retval NULL an error occurred, or the volume is in such a state that
3500 * we cannot load a header or return any volume struct
3502 * @note for DAFS, caller must NOT hold a ref count on 'hint'
3505 GetVolume(Error * ec, Error * client_ec, VolId volumeId, Volume * hint, int flags)
3508 /* pull this profiling/debugging code out of regular builds */
3510 #define VGET_CTR_INC(x) x++
3511 unsigned short V0 = 0, V1 = 0, V2 = 0, V3 = 0, V5 = 0, V6 =
3512 0, V7 = 0, V8 = 0, V9 = 0;
3513 unsigned short V10 = 0, V11 = 0, V12 = 0, V13 = 0, V14 = 0, V15 = 0;
3515 #define VGET_CTR_INC(x)
3517 #ifdef AFS_DEMAND_ATTACH_FS
3518 Volume *avp, * rvp = hint;
3522 * if VInit is zero, the volume package dynamic
3523 * data structures have not been initialized yet,
3524 * and we must immediately return an error
3530 *client_ec = VOFFLINE;
3535 #ifdef AFS_DEMAND_ATTACH_FS
3537 VCreateReservation_r(rvp);
3539 #endif /* AFS_DEMAND_ATTACH_FS */
3547 vp = VLookupVolume_r(ec, volumeId, vp);
3553 #ifdef AFS_DEMAND_ATTACH_FS
3554 if (rvp && (rvp != vp)) {
3555 /* break reservation on old vp */
3556 VCancelReservation_r(rvp);
3559 #endif /* AFS_DEMAND_ATTACH_FS */
3565 /* Until we have reached an initialization level of 2
3566 * we don't know whether this volume exists or not.
3567 * We can't sleep and retry later because before a volume
3568 * is attached, the caller tries to get it first. Just
3569 * return VOFFLINE and the caller can choose whether to
3570 * retry the command or not. */
3580 IncUInt64(&VStats.hdr_gets);
3582 #ifdef AFS_DEMAND_ATTACH_FS
3583 /* block if someone else is performing an exclusive op on this volume */
3586 VCreateReservation_r(rvp);
3588 VWaitExclusiveState_r(vp);
3590 /* short circuit with VNOVOL in the following circumstances:
3593 * - VOL_STATE_SHUTTING_DOWN
3595 if ((V_attachState(vp) == VOL_STATE_ERROR) ||
3596 (V_attachState(vp) == VOL_STATE_SHUTTING_DOWN) ||
3597 (V_attachState(vp) == VOL_STATE_GOING_OFFLINE)) {
3604 * short circuit with VOFFLINE in the following circumstances:
3606 * - VOL_STATE_UNATTACHED
3608 if (V_attachState(vp) == VOL_STATE_UNATTACHED) {
3609 if (vp->specialStatus) {
3610 *ec = vp->specialStatus;
3618 /* allowable states:
3624 if (vp->salvage.requested) {
3625 VUpdateSalvagePriority_r(vp);
3628 if (V_attachState(vp) == VOL_STATE_PREATTACHED) {
3629 avp = VAttachVolumeByVp_r(ec, vp, 0);
3632 /* VAttachVolumeByVp_r can return a pointer
3633 * != the vp passed to it under certain
3634 * conditions; make sure we don't leak
3635 * reservations if that happens */
3637 VCancelReservation_r(rvp);
3639 VCreateReservation_r(rvp);
3649 if (!vp->pending_vol_op) {
3664 if ((V_attachState(vp) == VOL_STATE_SALVAGING) ||
3665 (*ec == VSALVAGING)) {
3667 /* see CheckVnode() in afsfileprocs.c for an explanation
3668 * of this error code logic */
3669 afs_uint32 now = FT_ApproxTime();
3670 if ((vp->stats.last_salvage + (10 * 60)) >= now) {
3673 *client_ec = VRESTARTING;
3682 #ifdef AFS_DEMAND_ATTACH_FS
3684 * this test MUST happen after VAttachVolymeByVp, so vol_op_state is
3685 * not VolOpRunningUnknown (attach2 would have converted it to Online
3689 /* only valid before/during demand attachment */
3690 assert(!vp->pending_vol_op || vp->pending_vol_op->vol_op_state != FSSYNC_VolOpRunningUnknown);
3692 /* deny getvolume due to running mutually exclusive vol op */
3693 if (vp->pending_vol_op && vp->pending_vol_op->vol_op_state==FSSYNC_VolOpRunningOffline) {
3695 * volume cannot remain online during this volume operation.
3698 if (vp->specialStatus) {
3700 * special status codes outrank normal VOFFLINE code
3702 *ec = vp->specialStatus;
3704 *client_ec = vp->specialStatus;
3708 /* see CheckVnode() in afsfileprocs.c for an explanation
3709 * of this error code logic */
3710 afs_uint32 now = FT_ApproxTime();
3711 if ((vp->stats.last_vol_op + (10 * 60)) >= now) {
3714 *client_ec = VRESTARTING;
3719 VChangeState_r(vp, VOL_STATE_UNATTACHED);
3720 FreeVolumeHeader(vp);
3724 #endif /* AFS_DEMAND_ATTACH_FS */
3726 LoadVolumeHeader(ec, vp);
3729 /* Only log the error if it was a totally unexpected error. Simply
3730 * a missing inode is likely to be caused by the volume being deleted */
3731 if (errno != ENXIO || LogLevel)
3732 Log("Volume %u: couldn't reread volume header\n",
3734 #ifdef AFS_DEMAND_ATTACH_FS
3735 if (VCanScheduleSalvage()) {
3736 VRequestSalvage_r(ec, vp, SALVSYNC_ERROR, VOL_SALVAGE_INVALIDATE_HEADER);
3741 #else /* AFS_DEMAND_ATTACH_FS */
3744 #endif /* AFS_DEMAND_ATTACH_FS */
3749 if (vp->shuttingDown) {
3756 if (programType == fileServer) {
3758 if (vp->goingOffline) {
3760 #ifdef AFS_DEMAND_ATTACH_FS
3761 /* wait for the volume to go offline */
3762 if (V_attachState(vp) == VOL_STATE_GOING_OFFLINE) {
3763 VWaitStateChange_r(vp);
3765 #elif defined(AFS_PTHREAD_ENV)
3766 VOL_CV_WAIT(&vol_put_volume_cond);
3767 #else /* AFS_PTHREAD_ENV */
3768 LWP_WaitProcess(VPutVolume);
3769 #endif /* AFS_PTHREAD_ENV */
3772 if (vp->specialStatus) {
3774 *ec = vp->specialStatus;
3775 } else if (V_inService(vp) == 0 || V_blessed(vp) == 0) {
3778 } else if (V_inUse(vp) == 0) {
3789 #ifdef AFS_DEMAND_ATTACH_FS
3790 /* if no error, bump nUsers */
3793 VLRU_UpdateAccess_r(vp);
3796 VCancelReservation_r(rvp);
3799 if (client_ec && !*client_ec) {
3802 #else /* AFS_DEMAND_ATTACH_FS */
3803 /* if no error, bump nUsers */
3810 #endif /* AFS_DEMAND_ATTACH_FS */
3818 /***************************************************/
3819 /* Volume offline/detach routines */
3820 /***************************************************/
3822 /* caller MUST hold a heavyweight ref on vp */
3823 #ifdef AFS_DEMAND_ATTACH_FS
3825 VTakeOffline_r(register Volume * vp)
3829 assert(vp->nUsers > 0);
3830 assert(programType == fileServer);
3832 VCreateReservation_r(vp);
3833 VWaitExclusiveState_r(vp);
3835 vp->goingOffline = 1;
3836 V_needsSalvaged(vp) = 1;
3838 VRequestSalvage_r(&error, vp, SALVSYNC_ERROR, 0);
3839 VCancelReservation_r(vp);
3841 #else /* AFS_DEMAND_ATTACH_FS */
3843 VTakeOffline_r(register Volume * vp)
3845 assert(vp->nUsers > 0);
3846 assert(programType == fileServer);
3848 vp->goingOffline = 1;
3849 V_needsSalvaged(vp) = 1;
3851 #endif /* AFS_DEMAND_ATTACH_FS */
3854 VTakeOffline(register Volume * vp)
3862 * force a volume offline.
3864 * @param[in] vp volume object pointer
3865 * @param[in] flags flags (see note below)
3867 * @note the flag VOL_FORCEOFF_NOUPDATE is a recursion control flag
3868 * used when VUpdateVolume_r needs to call VForceOffline_r
3869 * (which in turn would normally call VUpdateVolume_r)
3871 * @see VUpdateVolume_r
3873 * @pre VOL_LOCK must be held.
3874 * for DAFS, caller must hold ref.
3876 * @note for DAFS, it _is safe_ to call this function from an
3879 * @post needsSalvaged flag is set.
3880 * for DAFS, salvage is requested.
3881 * no further references to the volume through the volume
3882 * package will be honored.
3883 * all file descriptor and vnode caches are invalidated.
3885 * @warning this is a heavy-handed interface. it results in
3886 * a volume going offline regardless of the current
3887 * reference count state.
3889 * @internal volume package internal use only
3892 VForceOffline_r(Volume * vp, int flags)
3896 #ifdef AFS_DEMAND_ATTACH_FS
3897 VChangeState_r(vp, VOL_STATE_ERROR);
3902 strcpy(V_offlineMessage(vp),
3903 "Forced offline due to internal error: volume needs to be salvaged");
3904 Log("Volume %u forced offline: it needs salvaging!\n", V_id(vp));
3907 vp->goingOffline = 0;
3908 V_needsSalvaged(vp) = 1;
3909 if (!(flags & VOL_FORCEOFF_NOUPDATE)) {
3910 VUpdateVolume_r(&error, vp, VOL_UPDATE_NOFORCEOFF);
3913 #ifdef AFS_DEMAND_ATTACH_FS
3914 VRequestSalvage_r(&error, vp, SALVSYNC_ERROR, VOL_SALVAGE_INVALIDATE_HEADER);
3915 #endif /* AFS_DEMAND_ATTACH_FS */
3917 #ifdef AFS_PTHREAD_ENV
3918 assert(pthread_cond_broadcast(&vol_put_volume_cond) == 0);
3919 #else /* AFS_PTHREAD_ENV */
3920 LWP_NoYieldSignal(VPutVolume);
3921 #endif /* AFS_PTHREAD_ENV */
3923 VReleaseVolumeHandles_r(vp);
3927 * force a volume offline.
3929 * @param[in] vp volume object pointer
3931 * @see VForceOffline_r
3934 VForceOffline(Volume * vp)
3937 VForceOffline_r(vp, 0);
3941 /* The opposite of VAttachVolume. The volume header is written to disk, with
3942 the inUse bit turned off. A copy of the header is maintained in memory,
3943 however (which is why this is VOffline, not VDetach).
3946 VOffline_r(Volume * vp, char *message)
3948 #ifndef AFS_DEMAND_ATTACH_FS
3950 VolumeId vid = V_id(vp);
3953 assert(programType != volumeUtility && programType != volumeServer);
3958 if (V_offlineMessage(vp)[0] == '\0')
3959 strncpy(V_offlineMessage(vp), message, sizeof(V_offlineMessage(vp)));
3960 V_offlineMessage(vp)[sizeof(V_offlineMessage(vp)) - 1] = '\0';
3962 vp->goingOffline = 1;
3963 #ifdef AFS_DEMAND_ATTACH_FS
3964 VChangeState_r(vp, VOL_STATE_GOING_OFFLINE);
3965 VCreateReservation_r(vp);
3968 /* wait for the volume to go offline */
3969 if (V_attachState(vp) == VOL_STATE_GOING_OFFLINE) {
3970 VWaitStateChange_r(vp);
3972 VCancelReservation_r(vp);
3973 #else /* AFS_DEMAND_ATTACH_FS */
3975 vp = VGetVolume_r(&error, vid); /* Wait for it to go offline */
3976 if (vp) /* In case it was reattached... */
3978 #endif /* AFS_DEMAND_ATTACH_FS */
3981 #ifdef AFS_DEMAND_ATTACH_FS
3983 * Take a volume offline in order to perform a volume operation.
3985 * @param[inout] ec address in which to store error code
3986 * @param[in] vp volume object pointer
3987 * @param[in] message volume offline status message
3990 * - VOL_LOCK is held
3991 * - caller MUST hold a heavyweight ref on vp
3994 * - volume is taken offline
3995 * - if possible, volume operation is promoted to running state
3996 * - on failure, *ec is set to nonzero
3998 * @note Although this function does not return any value, it may
3999 * still fail to promote our pending volume operation to
4000 * a running state. Any caller MUST check the value of *ec,
4001 * and MUST NOT blindly assume success.
4003 * @warning if the caller does not hold a lightweight ref on vp,
4004 * then it MUST NOT reference vp after this function
4005 * returns to the caller.
4007 * @internal volume package internal use only
4010 VOfflineForVolOp_r(Error *ec, Volume *vp, char *message)
4012 assert(vp->pending_vol_op);
4018 if (V_offlineMessage(vp)[0] == '\0')
4019 strncpy(V_offlineMessage(vp), message, sizeof(V_offlineMessage(vp)));
4020 V_offlineMessage(vp)[sizeof(V_offlineMessage(vp)) - 1] = '\0';
4022 vp->goingOffline = 1;
4023 VChangeState_r(vp, VOL_STATE_GOING_OFFLINE);
4024 VCreateReservation_r(vp);
4027 /* Wait for the volume to go offline */
4028 while (!VIsOfflineState(V_attachState(vp))) {
4029 /* do not give corrupted volumes to the volserver */
4030 if (vp->salvage.requested && vp->pending_vol_op->com.programType != salvageServer) {
4034 VWaitStateChange_r(vp);
4038 VCancelReservation_r(vp);
4040 #endif /* AFS_DEMAND_ATTACH_FS */
4043 VOffline(Volume * vp, char *message)
4046 VOffline_r(vp, message);
4050 /* This gets used for the most part by utility routines that don't want
4051 * to keep all the volume headers around. Generally, the file server won't
4052 * call this routine, because then the offline message in the volume header
4053 * (or other information) won't be available to clients. For NAMEI, also
4054 * close the file handles. However, the fileserver does call this during
4055 * an attach following a volume operation.
4058 VDetachVolume_r(Error * ec, Volume * vp)
4061 struct DiskPartition64 *tpartp;
4062 int notifyServer = 0;
4063 int useDone = FSYNC_VOL_ON;
4065 *ec = 0; /* always "succeeds" */
4066 if (VCanUseFSSYNC()) {
4067 notifyServer = vp->needsPutBack;
4068 if (V_destroyMe(vp) == DESTROY_ME)
4069 useDone = FSYNC_VOL_DONE;
4070 #ifdef AFS_DEMAND_ATTACH_FS
4071 else if (!V_blessed(vp) || !V_inService(vp))
4072 useDone = FSYNC_VOL_LEAVE_OFF;
4075 tpartp = vp->partition;
4077 DeleteVolumeFromHashTable(vp);
4078 vp->shuttingDown = 1;
4079 #ifdef AFS_DEMAND_ATTACH_FS
4080 DeleteVolumeFromVByPList_r(vp);
4082 VChangeState_r(vp, VOL_STATE_SHUTTING_DOWN);
4084 if (programType != fileServer)
4086 #endif /* AFS_DEMAND_ATTACH_FS */
4088 /* Will be detached sometime in the future--this is OK since volume is offline */
4090 /* XXX the following code should really be moved to VCheckDetach() since the volume
4091 * is not technically detached until the refcounts reach zero
4093 #ifdef FSSYNC_BUILD_CLIENT
4094 if (VCanUseFSSYNC() && notifyServer) {
4096 * Note: The server is not notified in the case of a bogus volume
4097 * explicitly to make it possible to create a volume, do a partial
4098 * restore, then abort the operation without ever putting the volume
4099 * online. This is essential in the case of a volume move operation
4100 * between two partitions on the same server. In that case, there
4101 * would be two instances of the same volume, one of them bogus,
4102 * which the file server would attempt to put on line
4104 FSYNC_VolOp(volume, tpartp->name, useDone, 0, NULL);
4105 /* XXX this code path is only hit by volume utilities, thus
4106 * V_BreakVolumeCallbacks will always be NULL. if we really
4107 * want to break callbacks in this path we need to use FSYNC_VolOp() */
4109 /* Dettaching it so break all callbacks on it */
4110 if (V_BreakVolumeCallbacks) {
4111 Log("volume %u detached; breaking all call backs\n", volume);
4112 (*V_BreakVolumeCallbacks) (volume);
4116 #endif /* FSSYNC_BUILD_CLIENT */
4120 VDetachVolume(Error * ec, Volume * vp)
4123 VDetachVolume_r(ec, vp);
4128 /***************************************************/
4129 /* Volume fd/inode handle closing routines */
4130 /***************************************************/
4132 /* For VDetachVolume, we close all cached file descriptors, but keep
4133 * the Inode handles in case we need to read from a busy volume.
4135 /* for demand attach, caller MUST hold ref count on vp */
4137 VCloseVolumeHandles_r(Volume * vp)
4139 #ifdef AFS_DEMAND_ATTACH_FS
4140 VolState state_save;
4142 state_save = VChangeState_r(vp, VOL_STATE_OFFLINING);
4147 * XXX need to investigate whether we can perform
4148 * DFlushVolume outside of vol_glock_mutex...
4150 * VCloseVnodeFiles_r drops the glock internally */
4151 DFlushVolume(vp->hashid);
4152 VCloseVnodeFiles_r(vp);
4154 #ifdef AFS_DEMAND_ATTACH_FS
4158 /* Too time consuming and unnecessary for the volserver */
4159 if (programType == fileServer) {
4160 IH_CONDSYNC(vp->vnodeIndex[vLarge].handle);
4161 IH_CONDSYNC(vp->vnodeIndex[vSmall].handle);
4162 IH_CONDSYNC(vp->diskDataHandle);
4164 IH_CONDSYNC(vp->linkHandle);
4165 #endif /* AFS_NT40_ENV */
4168 IH_REALLYCLOSE(vp->vnodeIndex[vLarge].handle);
4169 IH_REALLYCLOSE(vp->vnodeIndex[vSmall].handle);
4170 IH_REALLYCLOSE(vp->diskDataHandle);
4171 IH_REALLYCLOSE(vp->linkHandle);
4173 #ifdef AFS_DEMAND_ATTACH_FS
4174 if ((V_attachFlags(vp) & VOL_LOCKED)) {
4179 VChangeState_r(vp, state_save);
4183 /* For both VForceOffline and VOffline, we close all relevant handles.
4184 * For VOffline, if we re-attach the volume, the files may possible be
4185 * different than before.
4187 /* for demand attach, caller MUST hold a ref count on vp */
4189 VReleaseVolumeHandles_r(Volume * vp)
4191 #ifdef AFS_DEMAND_ATTACH_FS
4192 VolState state_save;
4194 state_save = VChangeState_r(vp, VOL_STATE_DETACHING);
4197 /* XXX need to investigate whether we can perform
4198 * DFlushVolume outside of vol_glock_mutex... */
4199 DFlushVolume(vp->hashid);
4201 VReleaseVnodeFiles_r(vp); /* releases the glock internally */
4203 #ifdef AFS_DEMAND_ATTACH_FS
4207 /* Too time consuming and unnecessary for the volserver */
4208 if (programType == fileServer) {
4209 IH_CONDSYNC(vp->vnodeIndex[vLarge].handle);
4210 IH_CONDSYNC(vp->vnodeIndex[vSmall].handle);
4211 IH_CONDSYNC(vp->diskDataHandle);
4213 IH_CONDSYNC(vp->linkHandle);
4214 #endif /* AFS_NT40_ENV */
4217 IH_RELEASE(vp->vnodeIndex[vLarge].handle);
4218 IH_RELEASE(vp->vnodeIndex[vSmall].handle);
4219 IH_RELEASE(vp->diskDataHandle);
4220 IH_RELEASE(vp->linkHandle);
4222 #ifdef AFS_DEMAND_ATTACH_FS
4223 if ((V_attachFlags(vp) & VOL_LOCKED)) {
4228 VChangeState_r(vp, state_save);
4233 /***************************************************/
4234 /* Volume write and fsync routines */
4235 /***************************************************/
4238 VUpdateVolume_r(Error * ec, Volume * vp, int flags)
4240 #ifdef AFS_DEMAND_ATTACH_FS
4241 VolState state_save;
4243 if (flags & VOL_UPDATE_WAIT) {
4244 VCreateReservation_r(vp);
4245 VWaitExclusiveState_r(vp);
4250 if (programType == fileServer)
4252 (V_inUse(vp) ? V_nextVnodeUnique(vp) +
4253 200 : V_nextVnodeUnique(vp));
4255 #ifdef AFS_DEMAND_ATTACH_FS
4256 state_save = VChangeState_r(vp, VOL_STATE_UPDATING);
4260 WriteVolumeHeader_r(ec, vp);
4262 #ifdef AFS_DEMAND_ATTACH_FS
4264 VChangeState_r(vp, state_save);
4265 if (flags & VOL_UPDATE_WAIT) {
4266 VCancelReservation_r(vp);
4271 Log("VUpdateVolume: error updating volume header, volume %u (%s)\n",
4272 V_id(vp), V_name(vp));
4273 /* try to update on-disk header,
4274 * while preventing infinite recursion */
4275 if (!(flags & VOL_UPDATE_NOFORCEOFF)) {
4276 VForceOffline_r(vp, VOL_FORCEOFF_NOUPDATE);
4282 VUpdateVolume(Error * ec, Volume * vp)
4285 VUpdateVolume_r(ec, vp, VOL_UPDATE_WAIT);
4290 VSyncVolume_r(Error * ec, Volume * vp, int flags)
4294 #ifdef AFS_DEMAND_ATTACH_FS
4295 VolState state_save;
4298 if (flags & VOL_SYNC_WAIT) {
4299 VUpdateVolume_r(ec, vp, VOL_UPDATE_WAIT);
4301 VUpdateVolume_r(ec, vp, 0);
4304 #ifdef AFS_DEMAND_ATTACH_FS
4305 state_save = VChangeState_r(vp, VOL_STATE_UPDATING);
4308 fdP = IH_OPEN(V_diskDataHandle(vp));
4309 assert(fdP != NULL);
4310 code = FDH_SYNC(fdP);
4313 #ifdef AFS_DEMAND_ATTACH_FS
4315 VChangeState_r(vp, state_save);
4321 VSyncVolume(Error * ec, Volume * vp)
4324 VSyncVolume_r(ec, vp, VOL_SYNC_WAIT);
4329 /***************************************************/
4330 /* Volume dealloaction routines */
4331 /***************************************************/
4333 #ifdef AFS_DEMAND_ATTACH_FS
4335 FreeVolume(Volume * vp)
4337 /* free the heap space, iff it's safe.
4338 * otherwise, pull it out of the hash table, so it
4339 * will get deallocated when all refs to it go away */
4340 if (!VCheckFree(vp)) {
4341 DeleteVolumeFromHashTable(vp);
4342 DeleteVolumeFromVByPList_r(vp);
4344 /* make sure we invalidate the header cache entry */
4345 FreeVolumeHeader(vp);
4348 #endif /* AFS_DEMAND_ATTACH_FS */
4351 ReallyFreeVolume(Volume * vp)
4356 #ifdef AFS_DEMAND_ATTACH_FS
4358 VChangeState_r(vp, VOL_STATE_FREED);
4359 if (vp->pending_vol_op)
4360 free(vp->pending_vol_op);
4361 #endif /* AFS_DEMAND_ATTACH_FS */
4362 for (i = 0; i < nVNODECLASSES; i++)
4363 if (vp->vnodeIndex[i].bitmap)
4364 free(vp->vnodeIndex[i].bitmap);
4365 FreeVolumeHeader(vp);
4366 #ifndef AFS_DEMAND_ATTACH_FS
4367 DeleteVolumeFromHashTable(vp);
4368 #endif /* AFS_DEMAND_ATTACH_FS */
4372 /* check to see if we should shutdown this volume
4373 * returns 1 if volume was freed, 0 otherwise */
4374 #ifdef AFS_DEMAND_ATTACH_FS
4376 VCheckDetach(register Volume * vp)
4381 if (vp->nUsers || vp->nWaiters)
4384 if (vp->shuttingDown) {
4386 if ((programType != fileServer) &&
4387 (V_inUse(vp) == programType) &&
4388 ((V_checkoutMode(vp) == V_VOLUPD) ||
4389 (V_checkoutMode(vp) == V_SECRETLY) ||
4390 ((V_checkoutMode(vp) == V_CLONE) &&
4391 (VolumeWriteable(vp))))) {
4393 VUpdateVolume_r(&ec, vp, VOL_UPDATE_NOFORCEOFF);
4395 Log("VCheckDetach: volume header update for volume %u "
4396 "failed with errno %d\n", vp->hashid, errno);
4399 VReleaseVolumeHandles_r(vp);
4401 ReallyFreeVolume(vp);
4402 if (programType == fileServer) {
4403 assert(pthread_cond_broadcast(&vol_put_volume_cond) == 0);
4408 #else /* AFS_DEMAND_ATTACH_FS */
4410 VCheckDetach(register Volume * vp)
4418 if (vp->shuttingDown) {
4420 if ((programType != fileServer) &&
4421 (V_inUse(vp) == programType) &&
4422 ((V_checkoutMode(vp) == V_VOLUPD) ||
4423 (V_checkoutMode(vp) == V_SECRETLY) ||
4424 ((V_checkoutMode(vp) == V_CLONE) &&
4425 (VolumeWriteable(vp))))) {
4427 VUpdateVolume_r(&ec, vp, VOL_UPDATE_NOFORCEOFF);
4429 Log("VCheckDetach: volume header update for volume %u failed with errno %d\n",
4433 VReleaseVolumeHandles_r(vp);
4434 ReallyFreeVolume(vp);
4435 if (programType == fileServer) {
4436 #if defined(AFS_PTHREAD_ENV)
4437 assert(pthread_cond_broadcast(&vol_put_volume_cond) == 0);
4438 #else /* AFS_PTHREAD_ENV */
4439 LWP_NoYieldSignal(VPutVolume);
4440 #endif /* AFS_PTHREAD_ENV */
4445 #endif /* AFS_DEMAND_ATTACH_FS */
4447 /* check to see if we should offline this volume
4448 * return 1 if volume went offline, 0 otherwise */
4449 #ifdef AFS_DEMAND_ATTACH_FS
4451 VCheckOffline(register Volume * vp)
4455 if (vp->goingOffline && !vp->nUsers) {
4457 assert(programType == fileServer);
4458 assert((V_attachState(vp) != VOL_STATE_ATTACHED) &&
4459 (V_attachState(vp) != VOL_STATE_FREED) &&
4460 (V_attachState(vp) != VOL_STATE_PREATTACHED) &&
4461 (V_attachState(vp) != VOL_STATE_UNATTACHED));
4465 * VOL_STATE_GOING_OFFLINE
4466 * VOL_STATE_SHUTTING_DOWN
4467 * VIsErrorState(V_attachState(vp))
4468 * VIsExclusiveState(V_attachState(vp))
4471 VCreateReservation_r(vp);
4472 VChangeState_r(vp, VOL_STATE_OFFLINING);
4475 /* must clear the goingOffline flag before we drop the glock */
4476 vp->goingOffline = 0;
4481 /* perform async operations */
4482 VUpdateVolume_r(&error, vp, 0);
4483 VCloseVolumeHandles_r(vp);
4486 if (V_offlineMessage(vp)[0]) {
4487 Log("VOffline: Volume %lu (%s) is now offline (%s)\n",
4488 afs_printable_uint32_lu(V_id(vp)), V_name(vp),
4489 V_offlineMessage(vp));
4491 Log("VOffline: Volume %lu (%s) is now offline\n",
4492 afs_printable_uint32_lu(V_id(vp)), V_name(vp));
4496 /* invalidate the volume header cache entry */
4497 FreeVolumeHeader(vp);
4499 /* if nothing changed state to error or salvaging,
4500 * drop state to unattached */
4501 if (!VIsErrorState(V_attachState(vp))) {
4502 VChangeState_r(vp, VOL_STATE_UNATTACHED);
4504 VCancelReservation_r(vp);
4505 /* no usage of vp is safe beyond this point */
4509 #else /* AFS_DEMAND_ATTACH_FS */
4511 VCheckOffline(register Volume * vp)
4515 if (vp->goingOffline && !vp->nUsers) {
4517 assert(programType == fileServer);
4520 vp->goingOffline = 0;
4522 VUpdateVolume_r(&error, vp, 0);
4523 VCloseVolumeHandles_r(vp);
4525 Log("VOffline: Volume %u (%s) is now offline", V_id(vp),
4527 if (V_offlineMessage(vp)[0])
4528 Log(" (%s)", V_offlineMessage(vp));
4531 FreeVolumeHeader(vp);
4532 #ifdef AFS_PTHREAD_ENV
4533 assert(pthread_cond_broadcast(&vol_put_volume_cond) == 0);
4534 #else /* AFS_PTHREAD_ENV */
4535 LWP_NoYieldSignal(VPutVolume);
4536 #endif /* AFS_PTHREAD_ENV */
4540 #endif /* AFS_DEMAND_ATTACH_FS */
4542 /***************************************************/
4543 /* demand attach fs ref counting routines */
4544 /***************************************************/
4546 #ifdef AFS_DEMAND_ATTACH_FS
4547 /* the following two functions handle reference counting for
4548 * asynchronous operations on volume structs.
4550 * their purpose is to prevent a VDetachVolume or VShutdown
4551 * from free()ing the Volume struct during an async i/o op */
4553 /* register with the async volume op ref counter */
4554 /* VCreateReservation_r moved into inline code header because it
4555 * is now needed in vnode.c -- tkeiser 11/20/2007
4559 * decrement volume-package internal refcount.
4561 * @param vp volume object pointer
4563 * @internal volume package internal use only
4566 * @arg VOL_LOCK is held
4567 * @arg lightweight refcount held
4569 * @post volume waiters refcount is decremented; volume may
4570 * have been deallocated/shutdown/offlined/salvaged/
4571 * whatever during the process
4573 * @warning once you have tossed your last reference (you can acquire
4574 * lightweight refs recursively) it is NOT SAFE to reference
4575 * a volume object pointer ever again
4577 * @see VCreateReservation_r
4579 * @note DEMAND_ATTACH_FS only
4582 VCancelReservation_r(Volume * vp)
4584 assert(--vp->nWaiters >= 0);
4585 if (vp->nWaiters == 0) {
4587 if (!VCheckDetach(vp)) {
4594 /* check to see if we should free this volume now
4595 * return 1 if volume was freed, 0 otherwise */
4597 VCheckFree(Volume * vp)
4600 if ((vp->nUsers == 0) &&
4601 (vp->nWaiters == 0) &&
4602 !(V_attachFlags(vp) & (VOL_IN_HASH |
4606 ReallyFreeVolume(vp);
4611 #endif /* AFS_DEMAND_ATTACH_FS */
4614 /***************************************************/
4615 /* online volume operations routines */
4616 /***************************************************/
4618 #ifdef AFS_DEMAND_ATTACH_FS
4620 * register a volume operation on a given volume.
4622 * @param[in] vp volume object
4623 * @param[in] vopinfo volume operation info object
4625 * @pre VOL_LOCK is held
4627 * @post volume operation info object attached to volume object.
4628 * volume operation statistics updated.
4630 * @note by "attached" we mean a copy of the passed in object is made
4632 * @internal volume package internal use only
4635 VRegisterVolOp_r(Volume * vp, FSSYNC_VolOp_info * vopinfo)
4637 FSSYNC_VolOp_info * info;
4639 /* attach a vol op info node to the volume struct */
4640 info = (FSSYNC_VolOp_info *) malloc(sizeof(FSSYNC_VolOp_info));
4641 assert(info != NULL);
4642 memcpy(info, vopinfo, sizeof(FSSYNC_VolOp_info));
4643 vp->pending_vol_op = info;
4646 vp->stats.last_vol_op = FT_ApproxTime();
4647 vp->stats.vol_ops++;
4648 IncUInt64(&VStats.vol_ops);
4654 * deregister the volume operation attached to this volume.
4656 * @param[in] vp volume object pointer
4658 * @pre VOL_LOCK is held
4660 * @post the volume operation info object is detached from the volume object
4662 * @internal volume package internal use only
4665 VDeregisterVolOp_r(Volume * vp)
4667 if (vp->pending_vol_op) {
4668 free(vp->pending_vol_op);
4669 vp->pending_vol_op = NULL;
4673 #endif /* AFS_DEMAND_ATTACH_FS */
4676 * determine whether it is safe to leave a volume online during
4677 * the volume operation described by the vopinfo object.
4679 * @param[in] vp volume object
4680 * @param[in] vopinfo volume operation info object
4682 * @return whether it is safe to leave volume online
4683 * @retval 0 it is NOT SAFE to leave the volume online
4684 * @retval 1 it is safe to leave the volume online during the operation
4687 * @arg VOL_LOCK is held
4688 * @arg disk header attached to vp (heavyweight ref on vp will guarantee
4689 * this condition is met)
4691 * @internal volume package internal use only
4694 VVolOpLeaveOnline_r(Volume * vp, FSSYNC_VolOp_info * vopinfo)
4696 return (vopinfo->vol_op_state == FSSYNC_VolOpRunningOnline ||
4697 (vopinfo->com.command == FSYNC_VOL_NEEDVOLUME &&
4698 (vopinfo->com.reason == V_READONLY ||
4699 (!VolumeWriteable(vp) &&
4700 (vopinfo->com.reason == V_CLONE ||
4701 vopinfo->com.reason == V_DUMP)))));
4705 * same as VVolOpLeaveOnline_r, but does not require a volume with an attached
4708 * @param[in] vp volume object
4709 * @param[in] vopinfo volume operation info object
4711 * @return whether it is safe to leave volume online
4712 * @retval 0 it is NOT SAFE to leave the volume online
4713 * @retval 1 it is safe to leave the volume online during the operation
4714 * @retval -1 unsure; volume header is required in order to know whether or
4715 * not is is safe to leave the volume online
4717 * @pre VOL_LOCK is held
4719 * @internal volume package internal use only
4722 VVolOpLeaveOnlineNoHeader_r(Volume * vp, FSSYNC_VolOp_info * vopinfo)
4724 /* follow the logic in VVolOpLeaveOnline_r; this is the same, except
4725 * assume that we don't know VolumeWriteable; return -1 if the answer
4726 * depends on VolumeWriteable */
4728 if (vopinfo->vol_op_state == FSSYNC_VolOpRunningOnline) {
4731 if (vopinfo->com.command == FSYNC_VOL_NEEDVOLUME &&
4732 vopinfo->com.reason == V_READONLY) {
4736 if (vopinfo->com.command == FSYNC_VOL_NEEDVOLUME &&
4737 (vopinfo->com.reason == V_CLONE ||
4738 vopinfo->com.reason == V_DUMP)) {
4740 /* must know VolumeWriteable */
4747 * determine whether VBUSY should be set during this volume operation.
4749 * @param[in] vp volume object
4750 * @param[in] vopinfo volume operation info object
4752 * @return whether VBUSY should be set
4753 * @retval 0 VBUSY does NOT need to be set
4754 * @retval 1 VBUSY SHOULD be set
4756 * @pre VOL_LOCK is held
4758 * @internal volume package internal use only
4761 VVolOpSetVBusy_r(Volume * vp, FSSYNC_VolOp_info * vopinfo)
4763 return ((vopinfo->com.command == FSYNC_VOL_OFF &&
4764 vopinfo->com.reason == FSYNC_SALVAGE) ||
4765 (vopinfo->com.command == FSYNC_VOL_NEEDVOLUME &&
4766 (vopinfo->com.reason == V_CLONE ||
4767 vopinfo->com.reason == V_DUMP)));
4771 /***************************************************/
4772 /* online salvager routines */
4773 /***************************************************/
4774 #if defined(AFS_DEMAND_ATTACH_FS)
4776 * check whether a salvage needs to be performed on this volume.
4778 * @param[in] vp pointer to volume object
4780 * @return status code
4781 * @retval 0 no salvage scheduled
4782 * @retval 1 a salvage has been scheduled with the salvageserver
4784 * @pre VOL_LOCK is held
4786 * @post if salvage request flag is set and nUsers and nWaiters are zero,
4787 * then a salvage will be requested
4789 * @note this is one of the event handlers called by VCancelReservation_r
4791 * @see VCancelReservation_r
4793 * @internal volume package internal use only.
4796 VCheckSalvage(register Volume * vp)
4799 #if defined(SALVSYNC_BUILD_CLIENT) || defined(FSSYNC_BUILD_CLIENT)
4800 if (vp->nUsers || vp->nWaiters)
4802 if (vp->salvage.requested) {
4803 VScheduleSalvage_r(vp);
4806 #endif /* SALVSYNC_BUILD_CLIENT || FSSYNC_BUILD_CLIENT */
4811 * request volume salvage.
4813 * @param[out] ec computed client error code
4814 * @param[in] vp volume object pointer
4815 * @param[in] reason reason code (passed to salvageserver via SALVSYNC)
4816 * @param[in] flags see flags note below
4819 * VOL_SALVAGE_INVALIDATE_HEADER causes volume header cache entry
4820 * to be invalidated.
4822 * @pre VOL_LOCK is held.
4824 * @post volume state is changed.
4825 * for fileserver, salvage will be requested once refcount reaches zero.
4827 * @return operation status code
4828 * @retval 0 volume salvage will occur
4829 * @retval 1 volume salvage could not be scheduled
4833 * @note in the fileserver, this call does not synchronously schedule a volume
4834 * salvage. rather, it sets volume state so that when volume refcounts
4835 * reach zero, a volume salvage will occur. by "refcounts", we mean both
4836 * nUsers and nWaiters must be zero.
4838 * @internal volume package internal use only.
4841 VRequestSalvage_r(Error * ec, Volume * vp, int reason, int flags)
4845 * for DAFS volume utilities that are not supposed to schedule salvages,
4846 * just transition to error state instead
4848 if (!VCanScheduleSalvage()) {
4849 VChangeState_r(vp, VOL_STATE_ERROR);
4854 if (programType != fileServer && !VCanUseFSSYNC()) {
4855 VChangeState_r(vp, VOL_STATE_ERROR);
4860 if (!vp->salvage.requested) {
4861 vp->salvage.requested = 1;
4862 vp->salvage.reason = reason;
4863 vp->stats.last_salvage = FT_ApproxTime();
4865 /* Note that it is not possible for us to reach this point if a
4866 * salvage is already running on this volume (even if the fileserver
4867 * was restarted during the salvage). If a salvage were running, the
4868 * salvager would have write-locked the volume header file, so when
4869 * we tried to lock the volume header, the lock would have failed,
4870 * and we would have failed during attachment prior to calling
4871 * VRequestSalvage. So we know that we can schedule salvages without
4872 * fear of a salvage already running for this volume. */
4874 if (vp->stats.salvages < SALVAGE_COUNT_MAX) {
4875 VChangeState_r(vp, VOL_STATE_SALVAGING);
4878 Log("VRequestSalvage: volume %u online salvaged too many times; forced offline.\n", vp->hashid);
4880 /* make sure neither VScheduleSalvage_r nor
4881 * VUpdateSalvagePriority_r try to schedule another salvage */
4882 vp->salvage.requested = vp->salvage.scheduled = 0;
4884 VChangeState_r(vp, VOL_STATE_ERROR);
4888 if (flags & VOL_SALVAGE_INVALIDATE_HEADER) {
4889 /* Instead of ReleaseVolumeHeader, we do FreeVolumeHeader()
4890 so that the the next VAttachVolumeByVp_r() invocation
4891 of attach2() will pull in a cached header
4892 entry and fail, then load a fresh one from disk and attach
4895 FreeVolumeHeader(vp);
4902 * update salvageserver scheduling priority for a volume.
4904 * @param[in] vp pointer to volume object
4906 * @return operation status
4908 * @retval 1 request denied, or SALVSYNC communications failure
4910 * @pre VOL_LOCK is held.
4912 * @post in-core salvage priority counter is incremented. if at least
4913 * SALVAGE_PRIO_UPDATE_INTERVAL seconds have elapsed since the
4914 * last SALVSYNC_RAISEPRIO request, we contact the salvageserver
4915 * to update its priority queue. if no salvage is scheduled,
4916 * this function is a no-op.
4918 * @note DAFS fileserver only
4920 * @note this should be called whenever a VGetVolume fails due to a
4921 * pending salvage request
4923 * @todo should set exclusive state and drop glock around salvsync call
4925 * @internal volume package internal use only.
4928 VUpdateSalvagePriority_r(Volume * vp)
4932 #ifdef SALVSYNC_BUILD_CLIENT
4937 now = FT_ApproxTime();
4939 /* update the salvageserver priority queue occasionally so that
4940 * frequently requested volumes get moved to the head of the queue
4942 if ((vp->salvage.scheduled) &&
4943 (vp->stats.last_salvage_req < (now-SALVAGE_PRIO_UPDATE_INTERVAL))) {
4944 code = SALVSYNC_SalvageVolume(vp->hashid,
4945 VPartitionPath(vp->partition),
4950 vp->stats.last_salvage_req = now;
4951 if (code != SYNC_OK) {
4955 #endif /* SALVSYNC_BUILD_CLIENT */
4960 #if defined(SALVSYNC_BUILD_CLIENT) || defined(FSSYNC_BUILD_CLIENT)
4962 /* A couple of little helper functions. These return true if we tried to
4963 * use this mechanism to schedule a salvage, false if we haven't tried.
4964 * If we did try a salvage then the results are contained in code.
4968 try_SALVSYNC(Volume *vp, char *partName, int *code) {
4969 #ifdef SALVSYNC_BUILD_CLIENT
4970 if (VCanUseSALVSYNC()) {
4971 /* can't use V_id() since there's no guarantee
4972 * we have the disk data header at this point */
4973 *code = SALVSYNC_SalvageVolume(vp->hashid,
4986 try_FSSYNC(Volume *vp, char *partName, int *code) {
4987 #ifdef FSSYNC_BUILD_CLIENT
4988 if (VCanUseFSSYNC()) {
4990 * If we aren't the fileserver, tell the fileserver the volume
4991 * needs to be salvaged. We could directly tell the
4992 * salvageserver, but the fileserver keeps track of some stats
4993 * related to salvages, and handles some other salvage-related
4994 * complications for us.
4996 *code = FSYNC_VolOp(vp->hashid, partName,
4997 FSYNC_VOL_FORCE_ERROR, FSYNC_SALVAGE, NULL);
5000 #endif /* FSSYNC_BUILD_CLIENT */
5005 * schedule a salvage with the salvage server or fileserver.
5007 * @param[in] vp pointer to volume object
5009 * @return operation status
5010 * @retval 0 salvage scheduled successfully
5011 * @retval 1 salvage not scheduled, or SALVSYNC/FSSYNC com error
5014 * @arg VOL_LOCK is held.
5015 * @arg nUsers and nWaiters should be zero.
5017 * @post salvageserver or fileserver is sent a salvage request
5019 * @note If we are the fileserver, the request will be sent to the salvage
5020 * server over SALVSYNC. If we are not the fileserver, the request will be
5021 * sent to the fileserver over FSSYNC (FSYNC_VOL_FORCE_ERROR/FSYNC_SALVAGE).
5025 * @internal volume package internal use only.
5028 VScheduleSalvage_r(Volume * vp)
5032 VolState state_save;
5033 VThreadOptions_t * thread_opts;
5036 assert(VCanUseSALVSYNC() || VCanUseFSSYNC());
5038 if (vp->nWaiters || vp->nUsers) {
5042 /* prevent endless salvage,attach,salvage,attach,... loops */
5043 if (vp->stats.salvages >= SALVAGE_COUNT_MAX)
5047 * don't perform salvsync ops on certain threads
5049 thread_opts = pthread_getspecific(VThread_key);
5050 if (thread_opts == NULL) {
5051 thread_opts = &VThread_defaults;
5053 if (thread_opts->disallow_salvsync || vol_disallow_salvsync) {
5058 * XXX the scheduling process should really be done asynchronously
5059 * to avoid fssync deadlocks
5061 if (!vp->salvage.scheduled) {
5062 /* if we haven't previously scheduled a salvage, do so now
5064 * set the volume to an exclusive state and drop the lock
5065 * around the SALVSYNC call
5067 * note that we do NOT acquire a reservation here -- doing so
5068 * could result in unbounded recursion
5070 strlcpy(partName, VPartitionPath(vp->partition), sizeof(partName));
5071 state_save = VChangeState_r(vp, VOL_STATE_SALVSYNC_REQ);
5074 assert(try_SALVSYNC(vp, partName, &code) ||
5075 try_FSSYNC(vp, partName, &code));
5078 VChangeState_r(vp, state_save);
5080 if (code == SYNC_OK) {
5081 vp->salvage.scheduled = 1;
5082 vp->stats.last_salvage_req = FT_ApproxTime();
5083 if (VCanUseSALVSYNC()) {
5084 /* don't record these stats for non-fileservers; let the
5085 * fileserver take care of these */
5086 vp->stats.salvages++;
5087 IncUInt64(&VStats.salvages);
5092 case SYNC_BAD_COMMAND:
5093 case SYNC_COM_ERROR:
5096 Log("VScheduleSalvage_r: Salvage request for volume %lu "
5097 "denied\n", afs_printable_uint32_lu(vp->hashid));
5100 Log("VScheduleSalvage_r: Salvage request for volume %lu "
5101 "received unknown protocol error %d\n",
5102 afs_printable_uint32_lu(vp->hashid), code);
5106 if (VCanUseFSSYNC()) {
5107 VChangeState_r(vp, VOL_STATE_ERROR);
5113 #endif /* SALVSYNC_BUILD_CLIENT || FSSYNC_BUILD_CLIENT */
5115 #ifdef SALVSYNC_BUILD_CLIENT
5118 * connect to the salvageserver SYNC service.
5120 * @return operation status
5124 * @post connection to salvageserver SYNC service established
5126 * @see VConnectSALV_r
5127 * @see VDisconnectSALV
5128 * @see VReconnectSALV
5135 retVal = VConnectSALV_r();
5141 * connect to the salvageserver SYNC service.
5143 * @return operation status
5147 * @pre VOL_LOCK is held.
5149 * @post connection to salvageserver SYNC service established
5152 * @see VDisconnectSALV_r
5153 * @see VReconnectSALV_r
5154 * @see SALVSYNC_clientInit
5156 * @internal volume package internal use only.
5159 VConnectSALV_r(void)
5161 return SALVSYNC_clientInit();
5165 * disconnect from the salvageserver SYNC service.
5167 * @return operation status
5170 * @pre client should have a live connection to the salvageserver
5172 * @post connection to salvageserver SYNC service destroyed
5174 * @see VDisconnectSALV_r
5176 * @see VReconnectSALV
5179 VDisconnectSALV(void)
5182 VDisconnectSALV_r();
5188 * disconnect from the salvageserver SYNC service.
5190 * @return operation status
5194 * @arg VOL_LOCK is held.
5195 * @arg client should have a live connection to the salvageserver.
5197 * @post connection to salvageserver SYNC service destroyed
5199 * @see VDisconnectSALV
5200 * @see VConnectSALV_r
5201 * @see VReconnectSALV_r
5202 * @see SALVSYNC_clientFinis
5204 * @internal volume package internal use only.
5207 VDisconnectSALV_r(void)
5209 return SALVSYNC_clientFinis();
5213 * disconnect and then re-connect to the salvageserver SYNC service.
5215 * @return operation status
5219 * @pre client should have a live connection to the salvageserver
5221 * @post old connection is dropped, and a new one is established
5224 * @see VDisconnectSALV
5225 * @see VReconnectSALV_r
5228 VReconnectSALV(void)
5232 retVal = VReconnectSALV_r();
5238 * disconnect and then re-connect to the salvageserver SYNC service.
5240 * @return operation status
5245 * @arg VOL_LOCK is held.
5246 * @arg client should have a live connection to the salvageserver.
5248 * @post old connection is dropped, and a new one is established
5250 * @see VConnectSALV_r
5251 * @see VDisconnectSALV
5252 * @see VReconnectSALV
5253 * @see SALVSYNC_clientReconnect
5255 * @internal volume package internal use only.
5258 VReconnectSALV_r(void)
5260 return SALVSYNC_clientReconnect();
5262 #endif /* SALVSYNC_BUILD_CLIENT */
5263 #endif /* AFS_DEMAND_ATTACH_FS */
5266 /***************************************************/
5267 /* FSSYNC routines */
5268 /***************************************************/
5270 /* This must be called by any volume utility which needs to run while the
5271 file server is also running. This is separated from VInitVolumePackage2 so
5272 that a utility can fork--and each of the children can independently
5273 initialize communication with the file server */
5274 #ifdef FSSYNC_BUILD_CLIENT
5276 * connect to the fileserver SYNC service.
5278 * @return operation status
5283 * @arg VInit must equal 2.
5284 * @arg Program Type must not be fileserver or salvager.
5286 * @post connection to fileserver SYNC service established
5289 * @see VDisconnectFS
5290 * @see VChildProcReconnectFS
5297 retVal = VConnectFS_r();
5303 * connect to the fileserver SYNC service.
5305 * @return operation status
5310 * @arg VInit must equal 2.
5311 * @arg Program Type must not be fileserver or salvager.
5312 * @arg VOL_LOCK is held.
5314 * @post connection to fileserver SYNC service established
5317 * @see VDisconnectFS_r
5318 * @see VChildProcReconnectFS_r
5320 * @internal volume package internal use only.
5326 assert((VInit == 2) &&
5327 (programType != fileServer) &&
5328 (programType != salvager));
5329 rc = FSYNC_clientInit();
5336 * disconnect from the fileserver SYNC service.
5339 * @arg client should have a live connection to the fileserver.
5340 * @arg VOL_LOCK is held.
5341 * @arg Program Type must not be fileserver or salvager.
5343 * @post connection to fileserver SYNC service destroyed
5345 * @see VDisconnectFS
5347 * @see VChildProcReconnectFS_r
5349 * @internal volume package internal use only.
5352 VDisconnectFS_r(void)
5354 assert((programType != fileServer) &&
5355 (programType != salvager));
5356 FSYNC_clientFinis();
5361 * disconnect from the fileserver SYNC service.
5364 * @arg client should have a live connection to the fileserver.
5365 * @arg Program Type must not be fileserver or salvager.
5367 * @post connection to fileserver SYNC service destroyed
5369 * @see VDisconnectFS_r
5371 * @see VChildProcReconnectFS
5382 * connect to the fileserver SYNC service from a child process following a fork.
5384 * @return operation status
5389 * @arg VOL_LOCK is held.
5390 * @arg current FSYNC handle is shared with a parent process
5392 * @post current FSYNC handle is discarded and a new connection to the
5393 * fileserver SYNC service is established
5395 * @see VChildProcReconnectFS
5397 * @see VDisconnectFS_r
5399 * @internal volume package internal use only.
5402 VChildProcReconnectFS_r(void)
5404 return FSYNC_clientChildProcReconnect();
5408 * connect to the fileserver SYNC service from a child process following a fork.
5410 * @return operation status
5414 * @pre current FSYNC handle is shared with a parent process
5416 * @post current FSYNC handle is discarded and a new connection to the
5417 * fileserver SYNC service is established
5419 * @see VChildProcReconnectFS_r
5421 * @see VDisconnectFS
5424 VChildProcReconnectFS(void)
5428 ret = VChildProcReconnectFS_r();
5432 #endif /* FSSYNC_BUILD_CLIENT */
5435 /***************************************************/
5436 /* volume bitmap routines */
5437 /***************************************************/
5440 * allocate a vnode bitmap number for the vnode
5442 * @param[out] ec error code
5443 * @param[in] vp volume object pointer
5444 * @param[in] index vnode index number for the vnode
5445 * @param[in] flags flag values described in note
5447 * @note for DAFS, flags parameter controls locking behavior.
5448 * If (flags & VOL_ALLOC_BITMAP_WAIT) is set, then this function
5449 * will create a reservation and block on any other exclusive
5450 * operations. Otherwise, this function assumes the caller
5451 * already has exclusive access to vp, and we just change the
5454 * @pre VOL_LOCK held
5456 * @return bit number allocated
5462 VAllocBitmapEntry_r(Error * ec, Volume * vp,
5463 struct vnodeIndex *index, int flags)
5466 register byte *bp, *ep;
5467 #ifdef AFS_DEMAND_ATTACH_FS
5468 VolState state_save;
5469 #endif /* AFS_DEMAND_ATTACH_FS */
5473 /* This test is probably redundant */
5474 if (!VolumeWriteable(vp)) {
5475 *ec = (bit32) VREADONLY;
5479 #ifdef AFS_DEMAND_ATTACH_FS
5480 if (flags & VOL_ALLOC_BITMAP_WAIT) {
5481 VCreateReservation_r(vp);
5482 VWaitExclusiveState_r(vp);
5484 state_save = VChangeState_r(vp, VOL_STATE_GET_BITMAP);
5485 #endif /* AFS_DEMAND_ATTACH_FS */
5488 if ((programType == fileServer) && !index->bitmap) {
5490 #ifndef AFS_DEMAND_ATTACH_FS
5491 /* demand attach fs uses the volume state to avoid races.
5492 * specialStatus field is not used at all */
5494 if (vp->specialStatus == VBUSY) {
5495 if (vp->goingOffline) { /* vos dump waiting for the volume to
5496 * go offline. We probably come here
5497 * from AddNewReadableResidency */
5500 while (vp->specialStatus == VBUSY) {
5501 #ifdef AFS_PTHREAD_ENV
5505 #else /* !AFS_PTHREAD_ENV */
5507 #endif /* !AFS_PTHREAD_ENV */
5511 #endif /* !AFS_DEMAND_ATTACH_FS */
5513 if (!index->bitmap) {
5514 #ifndef AFS_DEMAND_ATTACH_FS
5515 vp->specialStatus = VBUSY; /* Stop anyone else from using it. */
5516 #endif /* AFS_DEMAND_ATTACH_FS */
5517 for (i = 0; i < nVNODECLASSES; i++) {
5518 VGetBitmap_r(ec, vp, i);
5520 #ifdef AFS_DEMAND_ATTACH_FS
5521 VRequestSalvage_r(ec, vp, SALVSYNC_ERROR, VOL_SALVAGE_INVALIDATE_HEADER);
5522 #else /* AFS_DEMAND_ATTACH_FS */
5523 DeleteVolumeFromHashTable(vp);
5524 vp->shuttingDown = 1; /* Let who has it free it. */
5525 vp->specialStatus = 0;
5526 #endif /* AFS_DEMAND_ATTACH_FS */
5530 #ifndef AFS_DEMAND_ATTACH_FS
5532 vp->specialStatus = 0; /* Allow others to have access. */
5533 #endif /* AFS_DEMAND_ATTACH_FS */
5536 #endif /* BITMAP_LATER */
5538 #ifdef AFS_DEMAND_ATTACH_FS
5540 #endif /* AFS_DEMAND_ATTACH_FS */
5541 bp = index->bitmap + index->bitmapOffset;
5542 ep = index->bitmap + index->bitmapSize;
5544 if ((*(bit32 *) bp) != (bit32) 0xffffffff) {
5546 index->bitmapOffset = (afs_uint32) (bp - index->bitmap);
5549 o = ffs(~*bp) - 1; /* ffs is documented in BSTRING(3) */
5551 ret = ((bp - index->bitmap) * 8 + o);
5552 #ifdef AFS_DEMAND_ATTACH_FS
5554 #endif /* AFS_DEMAND_ATTACH_FS */
5557 bp += sizeof(bit32) /* i.e. 4 */ ;
5559 /* No bit map entry--must grow bitmap */
5561 realloc(index->bitmap, index->bitmapSize + VOLUME_BITMAP_GROWSIZE);
5564 bp += index->bitmapSize;
5565 memset(bp, 0, VOLUME_BITMAP_GROWSIZE);
5566 index->bitmapOffset = index->bitmapSize;
5567 index->bitmapSize += VOLUME_BITMAP_GROWSIZE;
5569 ret = index->bitmapOffset * 8;
5570 #ifdef AFS_DEMAND_ATTACH_FS
5572 #endif /* AFS_DEMAND_ATTACH_FS */
5575 #ifdef AFS_DEMAND_ATTACH_FS
5576 VChangeState_r(vp, state_save);
5577 if (flags & VOL_ALLOC_BITMAP_WAIT) {
5578 VCancelReservation_r(vp);
5580 #endif /* AFS_DEMAND_ATTACH_FS */
5585 VAllocBitmapEntry(Error * ec, Volume * vp, register struct vnodeIndex * index)
5589 retVal = VAllocBitmapEntry_r(ec, vp, index, VOL_ALLOC_BITMAP_WAIT);
5595 VFreeBitMapEntry_r(Error * ec, register struct vnodeIndex *index,
5598 unsigned int offset;
5604 #endif /* BITMAP_LATER */
5605 offset = bitNumber >> 3;
5606 if (offset >= index->bitmapSize) {
5610 if (offset < index->bitmapOffset)
5611 index->bitmapOffset = offset & ~3; /* Truncate to nearest bit32 */
5612 *(index->bitmap + offset) &= ~(1 << (bitNumber & 0x7));
5616 VFreeBitMapEntry(Error * ec, register struct vnodeIndex *index,
5620 VFreeBitMapEntry_r(ec, index, bitNumber);
5624 /* this function will drop the glock internally.
5625 * for old pthread fileservers, this is safe thanks to vbusy.
5627 * for demand attach fs, caller must have already called
5628 * VCreateReservation_r and VWaitExclusiveState_r */
5630 VGetBitmap_r(Error * ec, Volume * vp, VnodeClass class)
5632 StreamHandle_t *file;
5635 struct VnodeClassInfo *vcp = &VnodeClassInfo[class];
5636 struct vnodeIndex *vip = &vp->vnodeIndex[class];
5637 struct VnodeDiskObject *vnode;
5638 unsigned int unique = 0;
5642 #endif /* BITMAP_LATER */
5643 #ifdef AFS_DEMAND_ATTACH_FS
5644 VolState state_save;
5645 #endif /* AFS_DEMAND_ATTACH_FS */
5649 #ifdef AFS_DEMAND_ATTACH_FS
5650 state_save = VChangeState_r(vp, VOL_STATE_GET_BITMAP);
5651 #endif /* AFS_DEMAND_ATTACH_FS */
5654 fdP = IH_OPEN(vip->handle);
5655 assert(fdP != NULL);
5656 file = FDH_FDOPEN(fdP, "r");
5657 assert(file != NULL);
5658 vnode = (VnodeDiskObject *) malloc(vcp->diskSize);
5659 assert(vnode != NULL);
5660 size = OS_SIZE(fdP->fd_fd);
5662 nVnodes = (size <= vcp->diskSize ? 0 : size - vcp->diskSize)
5664 vip->bitmapSize = ((nVnodes / 8) + 10) / 4 * 4; /* The 10 is a little extra so
5665 * a few files can be created in this volume,
5666 * the whole thing is rounded up to nearest 4
5667 * bytes, because the bit map allocator likes
5670 BitMap = (byte *) calloc(1, vip->bitmapSize);
5671 assert(BitMap != NULL);
5672 #else /* BITMAP_LATER */
5673 vip->bitmap = (byte *) calloc(1, vip->bitmapSize);
5674 assert(vip->bitmap != NULL);
5675 vip->bitmapOffset = 0;
5676 #endif /* BITMAP_LATER */
5677 if (STREAM_SEEK(file, vcp->diskSize, 0) != -1) {
5679 for (bitNumber = 0; bitNumber < nVnodes + 100; bitNumber++) {
5680 if (STREAM_READ(vnode, vcp->diskSize, 1, file) != 1)
5682 if (vnode->type != vNull) {
5683 if (vnode->vnodeMagic != vcp->magic) {
5684 Log("GetBitmap: addled vnode index in volume %s; volume needs salvage\n", V_name(vp));
5689 *(BitMap + (bitNumber >> 3)) |= (1 << (bitNumber & 0x7));
5690 #else /* BITMAP_LATER */
5691 *(vip->bitmap + (bitNumber >> 3)) |= (1 << (bitNumber & 0x7));
5692 #endif /* BITMAP_LATER */
5693 if (unique <= vnode->uniquifier)
5694 unique = vnode->uniquifier + 1;
5696 #ifndef AFS_PTHREAD_ENV
5697 if ((bitNumber & 0x00ff) == 0x0ff) { /* every 256 iterations */
5700 #endif /* !AFS_PTHREAD_ENV */
5703 if (vp->nextVnodeUnique < unique) {
5704 Log("GetBitmap: bad volume uniquifier for volume %s; volume needs salvage\n", V_name(vp));
5707 /* Paranoia, partly justified--I think fclose after fdopen
5708 * doesn't seem to close fd. In any event, the documentation
5709 * doesn't specify, so it's safer to close it twice.
5717 /* There may have been a racing condition with some other thread, both
5718 * creating the bitmaps for this volume. If the other thread was faster
5719 * the pointer to bitmap should already be filled and we can free ours.
5721 if (vip->bitmap == NULL) {
5722 vip->bitmap = BitMap;
5723 vip->bitmapOffset = 0;
5725 free((byte *) BitMap);
5726 #endif /* BITMAP_LATER */
5727 #ifdef AFS_DEMAND_ATTACH_FS
5728 VChangeState_r(vp, state_save);
5729 #endif /* AFS_DEMAND_ATTACH_FS */
5733 /***************************************************/
5734 /* Volume Path and Volume Number utility routines */
5735 /***************************************************/
5738 * find the first occurrence of a volume header file and return the path.
5740 * @param[out] ec outbound error code
5741 * @param[in] volumeId volume id to find
5742 * @param[out] partitionp pointer to disk partition path string
5743 * @param[out] namep pointer to volume header file name string
5745 * @post path to first occurrence of volume header is returned in partitionp
5746 * and namep, or ec is set accordingly.
5748 * @warning this function is NOT re-entrant -- partitionp and namep point to
5749 * static data segments
5751 * @note if a volume utility inadvertently leaves behind a stale volume header
5752 * on a vice partition, it is possible for callers to get the wrong one,
5753 * depending on the order of the disk partition linked list.
5757 VGetVolumePath(Error * ec, VolId volumeId, char **partitionp, char **namep)
5759 static char partition[VMAXPATHLEN], name[VMAXPATHLEN];
5760 char path[VMAXPATHLEN];
5762 struct DiskPartition64 *dp;
5766 (void)afs_snprintf(&name[1], (sizeof name) - 1, VFORMAT, afs_printable_uint32_lu(volumeId));
5767 for (dp = DiskPartitionList; dp; dp = dp->next) {
5768 struct afs_stat status;
5769 strcpy(path, VPartitionPath(dp));
5771 if (afs_stat(path, &status) == 0) {
5772 strcpy(partition, dp->name);
5779 *partitionp = *namep = NULL;
5781 *partitionp = partition;
5787 * extract a volume number from a volume header filename string.
5789 * @param[in] name volume header filename string
5791 * @return volume number
5793 * @note the string must be of the form VFORMAT. the only permissible
5794 * deviation is a leading '/' character.
5799 VolumeNumber(char *name)
5803 return atoi(name + 1);
5807 * compute the volume header filename.
5809 * @param[in] volumeId
5811 * @return volume header filename
5813 * @post volume header filename string is constructed
5815 * @warning this function is NOT re-entrant -- the returned string is
5816 * stored in a static char array. see VolumeExternalName_r
5817 * for a re-entrant equivalent.
5819 * @see VolumeExternalName_r
5821 * @deprecated due to the above re-entrancy warning, this interface should
5822 * be considered deprecated. Please use VolumeExternalName_r
5826 VolumeExternalName(VolumeId volumeId)
5828 static char name[VMAXPATHLEN];
5829 (void)afs_snprintf(name, sizeof name, VFORMAT, afs_printable_uint32_lu(volumeId));
5834 * compute the volume header filename.
5836 * @param[in] volumeId
5837 * @param[inout] name array in which to store filename
5838 * @param[in] len length of name array
5840 * @return result code from afs_snprintf
5842 * @see VolumeExternalName
5845 * @note re-entrant equivalent of VolumeExternalName
5848 VolumeExternalName_r(VolumeId volumeId, char * name, size_t len)
5850 return afs_snprintf(name, len, VFORMAT, afs_printable_uint32_lu(volumeId));
5854 /***************************************************/
5855 /* Volume Usage Statistics routines */
5856 /***************************************************/
5858 #if OPENAFS_VOL_STATS
5859 #define OneDay (86400) /* 24 hours' worth of seconds */
5861 #define OneDay (24*60*60) /* 24 hours */
5862 #endif /* OPENAFS_VOL_STATS */
5865 Midnight(time_t t) {
5866 struct tm local, *l;
5869 #if defined(AFS_PTHREAD_ENV) && !defined(AFS_NT40_ENV)
5870 l = localtime_r(&t, &local);
5876 /* the following is strictly speaking problematic on the
5877 switching day to daylight saving time, after the switch,
5878 as tm_isdst does not match. Similarly, on the looong day when
5879 switching back the OneDay check will not do what naively expected!
5880 The effects are minor, though, and more a matter of interpreting
5882 #ifndef AFS_PTHREAD_ENV
5885 local.tm_hour = local.tm_min=local.tm_sec = 0;
5886 midnight = mktime(&local);
5887 if (midnight != (time_t) -1) return(midnight);
5889 return( (t/OneDay)*OneDay );
5893 /*------------------------------------------------------------------------
5894 * [export] VAdjustVolumeStatistics
5897 * If we've passed midnight, we need to update all the day use
5898 * statistics as well as zeroing the detailed volume statistics
5899 * (if we are implementing them).
5902 * vp : Pointer to the volume structure describing the lucky
5903 * volume being considered for update.
5909 * Nothing interesting.
5913 *------------------------------------------------------------------------*/
5916 VAdjustVolumeStatistics_r(register Volume * vp)
5918 unsigned int now = FT_ApproxTime();
5920 if (now - V_dayUseDate(vp) > OneDay) {
5921 register int ndays, i;
5923 ndays = (now - V_dayUseDate(vp)) / OneDay;
5924 for (i = 6; i > ndays - 1; i--)
5925 V_weekUse(vp)[i] = V_weekUse(vp)[i - ndays];
5926 for (i = 0; i < ndays - 1 && i < 7; i++)
5927 V_weekUse(vp)[i] = 0;
5929 V_weekUse(vp)[ndays - 1] = V_dayUse(vp);
5931 V_dayUseDate(vp) = Midnight(now);
5933 #if OPENAFS_VOL_STATS
5935 * All we need to do is bzero the entire VOL_STATS_BYTES of
5936 * the detailed volume statistics area.
5938 memset((V_stat_area(vp)), 0, VOL_STATS_BYTES);
5939 #endif /* OPENAFS_VOL_STATS */
5942 /*It's been more than a day of collection */
5944 * Always return happily.
5947 } /*VAdjustVolumeStatistics */
5950 VAdjustVolumeStatistics(register Volume * vp)
5954 retVal = VAdjustVolumeStatistics_r(vp);
5960 VBumpVolumeUsage_r(register Volume * vp)
5962 unsigned int now = FT_ApproxTime();
5963 V_accessDate(vp) = now;
5964 if (now - V_dayUseDate(vp) > OneDay)
5965 VAdjustVolumeStatistics_r(vp);
5967 * Save the volume header image to disk after every 128 bumps to dayUse.
5969 if ((V_dayUse(vp)++ & 127) == 0) {
5971 VUpdateVolume_r(&error, vp, VOL_UPDATE_WAIT);
5976 VBumpVolumeUsage(register Volume * vp)
5979 VBumpVolumeUsage_r(vp);
5984 VSetDiskUsage_r(void)
5986 #ifndef AFS_DEMAND_ATTACH_FS
5987 static int FifteenMinuteCounter = 0;
5991 /* NOTE: Don't attempt to access the partitions list until the
5992 * initialization level indicates that all volumes are attached,
5993 * which implies that all partitions are initialized. */
5994 #ifdef AFS_PTHREAD_ENV
5996 #else /* AFS_PTHREAD_ENV */
5998 #endif /* AFS_PTHREAD_ENV */
6001 VResetDiskUsage_r();
6003 #ifndef AFS_DEMAND_ATTACH_FS
6004 if (++FifteenMinuteCounter == 3) {
6005 FifteenMinuteCounter = 0;
6008 #endif /* !AFS_DEMAND_ATTACH_FS */
6020 /***************************************************/
6021 /* Volume Update List routines */
6022 /***************************************************/
6024 /* The number of minutes that a volume hasn't been updated before the
6025 * "Dont salvage" flag in the volume header will be turned on */
6026 #define SALVAGE_INTERVAL (10*60)
6031 * volume update list functionality has been moved into the VLRU
6032 * the DONT_SALVAGE flag is now set during VLRU demotion
6035 #ifndef AFS_DEMAND_ATTACH_FS
6036 static VolumeId *UpdateList = NULL; /* Pointer to array of Volume ID's */
6037 static int nUpdatedVolumes = 0; /* Updated with entry in UpdateList, salvage after crash flag on */
6038 static int updateSize = 0; /* number of entries possible */
6039 #define UPDATE_LIST_SIZE 128 /* initial size increment (must be a power of 2!) */
6040 #endif /* !AFS_DEMAND_ATTACH_FS */
6043 VAddToVolumeUpdateList_r(Error * ec, Volume * vp)
6046 vp->updateTime = FT_ApproxTime();
6047 if (V_dontSalvage(vp) == 0)
6049 V_dontSalvage(vp) = 0;
6050 VSyncVolume_r(ec, vp, 0);
6051 #ifdef AFS_DEMAND_ATTACH_FS
6052 V_attachFlags(vp) &= ~(VOL_HDR_DONTSALV);
6053 #else /* !AFS_DEMAND_ATTACH_FS */
6056 if (UpdateList == NULL) {
6057 updateSize = UPDATE_LIST_SIZE;
6058 UpdateList = (VolumeId *) malloc(sizeof(VolumeId) * updateSize);
6060 if (nUpdatedVolumes == updateSize) {
6062 if (updateSize > 524288) {
6063 Log("warning: there is likely a bug in the volume update scanner\n");
6067 (VolumeId *) realloc(UpdateList,
6068 sizeof(VolumeId) * updateSize);
6071 assert(UpdateList != NULL);
6072 UpdateList[nUpdatedVolumes++] = V_id(vp);
6073 #endif /* !AFS_DEMAND_ATTACH_FS */
6076 #ifndef AFS_DEMAND_ATTACH_FS
6078 VScanUpdateList(void)
6080 register int i, gap;
6081 register Volume *vp;
6083 afs_uint32 now = FT_ApproxTime();
6084 /* Be careful with this code, since it works with interleaved calls to AddToVolumeUpdateList */
6085 for (i = gap = 0; i < nUpdatedVolumes; i++) {
6087 UpdateList[i - gap] = UpdateList[i];
6089 /* XXX this routine needlessly messes up the Volume LRU by
6090 * breaking the LRU temporal-locality assumptions.....
6091 * we should use a special volume header allocator here */
6092 vp = VGetVolume_r(&error, UpdateList[i - gap] = UpdateList[i]);
6095 } else if (vp->nUsers == 1 && now - vp->updateTime > SALVAGE_INTERVAL) {
6096 V_dontSalvage(vp) = DONT_SALVAGE;
6097 VUpdateVolume_r(&error, vp, 0); /* No need to fsync--not critical */
6105 #ifndef AFS_PTHREAD_ENV
6107 #endif /* !AFS_PTHREAD_ENV */
6109 nUpdatedVolumes -= gap;
6111 #endif /* !AFS_DEMAND_ATTACH_FS */
6114 /***************************************************/
6115 /* Volume LRU routines */
6116 /***************************************************/
6121 * with demand attach fs, we attempt to soft detach(1)
6122 * volumes which have not been accessed in a long time
6123 * in order to speed up fileserver shutdown
6125 * (1) by soft detach we mean a process very similar
6126 * to VOffline, except the final state of the
6127 * Volume will be VOL_STATE_PREATTACHED, instead
6128 * of the usual VOL_STATE_UNATTACHED
6130 #ifdef AFS_DEMAND_ATTACH_FS
6132 /* implementation is reminiscent of a generational GC
6134 * queue 0 is newly attached volumes. this queue is
6135 * sorted by attach timestamp
6137 * queue 1 is volumes that have been around a bit
6138 * longer than queue 0. this queue is sorted by
6141 * queue 2 is volumes tha have been around the longest.
6142 * this queue is unsorted
6144 * queue 3 is volumes that have been marked as
6145 * candidates for soft detachment. this queue is
6148 #define VLRU_GENERATIONS 3 /**< number of generations in VLRU */
6149 #define VLRU_QUEUES 5 /**< total number of VLRU queues */
6152 * definition of a VLRU queue.
6155 volatile struct rx_queue q;
6162 * main VLRU data structure.
6165 struct VLRU_q q[VLRU_QUEUES]; /**< VLRU queues */
6168 /** time interval (in seconds) between promotion passes for
6169 * each young generation queue. */
6170 afs_uint32 promotion_interval[VLRU_GENERATIONS-1];
6172 /** time interval (in seconds) between soft detach candidate
6173 * scans for each generation queue.
6175 * scan_interval[VLRU_QUEUE_CANDIDATE] defines how frequently
6176 * we perform a soft detach pass. */
6177 afs_uint32 scan_interval[VLRU_GENERATIONS+1];
6179 /* scheduler state */
6180 int next_idx; /**< next queue to receive attention */
6181 afs_uint32 last_promotion[VLRU_GENERATIONS-1]; /**< timestamp of last promotion scan */
6182 afs_uint32 last_scan[VLRU_GENERATIONS+1]; /**< timestamp of last detach scan */
6184 int scanner_state; /**< state of scanner thread */
6185 pthread_cond_t cv; /**< state transition CV */
6188 /** global VLRU state */
6189 static struct VLRU volume_LRU;
6192 * defined states for VLRU scanner thread.
6195 VLRU_SCANNER_STATE_OFFLINE = 0, /**< vlru scanner thread is offline */
6196 VLRU_SCANNER_STATE_ONLINE = 1, /**< vlru scanner thread is online */
6197 VLRU_SCANNER_STATE_SHUTTING_DOWN = 2, /**< vlru scanner thread is shutting down */
6198 VLRU_SCANNER_STATE_PAUSING = 3, /**< vlru scanner thread is getting ready to pause */
6199 VLRU_SCANNER_STATE_PAUSED = 4 /**< vlru scanner thread is paused */
6200 } vlru_thread_state_t;
6202 /* vlru disk data header stuff */
6203 #define VLRU_DISK_MAGIC 0x7a8b9cad /**< vlru disk entry magic number */
6204 #define VLRU_DISK_VERSION 1 /**< vlru disk entry version number */
6206 /** vlru default expiration time (for eventual fs state serialization of vlru data) */
6207 #define VLRU_DUMP_EXPIRATION_TIME (60*60*24*7) /* expire vlru data after 1 week */
6210 /** minimum volume inactivity (in seconds) before a volume becomes eligible for
6211 * soft detachment. */
6212 static afs_uint32 VLRU_offline_thresh = VLRU_DEFAULT_OFFLINE_THRESH;
6214 /** time interval (in seconds) between VLRU scanner thread soft detach passes. */
6215 static afs_uint32 VLRU_offline_interval = VLRU_DEFAULT_OFFLINE_INTERVAL;
6217 /** maximum number of volumes to soft detach in a VLRU soft detach pass. */
6218 static afs_uint32 VLRU_offline_max = VLRU_DEFAULT_OFFLINE_MAX;
6220 /** VLRU control flag. non-zero value implies VLRU subsystem is activated. */
6221 static afs_uint32 VLRU_enabled = 1;
6223 /* queue synchronization routines */
6224 static void VLRU_BeginExclusive_r(struct VLRU_q * q);
6225 static void VLRU_EndExclusive_r(struct VLRU_q * q);
6226 static void VLRU_Wait_r(struct VLRU_q * q);
6229 * set VLRU subsystem tunable parameters.
6231 * @param[in] option tunable option to modify
6232 * @param[in] val new value for tunable parameter
6234 * @pre @c VInitVolumePackage2 has not yet been called.
6236 * @post tunable parameter is modified
6240 * @note valid option parameters are:
6241 * @arg @c VLRU_SET_THRESH
6242 * set the period of inactivity after which
6243 * volumes are eligible for soft detachment
6244 * @arg @c VLRU_SET_INTERVAL
6245 * set the time interval between calls
6246 * to the volume LRU "garbage collector"
6247 * @arg @c VLRU_SET_MAX
6248 * set the max number of volumes to deallocate
6252 VLRU_SetOptions(int option, afs_uint32 val)
6254 if (option == VLRU_SET_THRESH) {
6255 VLRU_offline_thresh = val;
6256 } else if (option == VLRU_SET_INTERVAL) {
6257 VLRU_offline_interval = val;
6258 } else if (option == VLRU_SET_MAX) {
6259 VLRU_offline_max = val;
6260 } else if (option == VLRU_SET_ENABLED) {
6263 VLRU_ComputeConstants();
6267 * compute VLRU internal timing parameters.
6269 * @post VLRU scanner thread internal timing parameters are computed
6271 * @note computes internal timing parameters based upon user-modifiable
6272 * tunable parameters.
6276 * @internal volume package internal use only.
6279 VLRU_ComputeConstants(void)
6281 afs_uint32 factor = VLRU_offline_thresh / VLRU_offline_interval;
6283 /* compute the candidate scan interval */
6284 volume_LRU.scan_interval[VLRU_QUEUE_CANDIDATE] = VLRU_offline_interval;
6286 /* compute the promotion intervals */
6287 volume_LRU.promotion_interval[VLRU_QUEUE_NEW] = VLRU_offline_thresh * 2;
6288 volume_LRU.promotion_interval[VLRU_QUEUE_MID] = VLRU_offline_thresh * 4;
6291 /* compute the gen 0 scan interval */
6292 volume_LRU.scan_interval[VLRU_QUEUE_NEW] = VLRU_offline_thresh / 8;
6294 /* compute the gen 0 scan interval */
6295 volume_LRU.scan_interval[VLRU_QUEUE_NEW] = VLRU_offline_interval * 2;
6300 * initialize VLRU subsystem.
6302 * @pre this function has not yet been called
6304 * @post VLRU subsystem is initialized and VLRU scanner thread is starting
6308 * @internal volume package internal use only.
6314 pthread_attr_t attrs;
6317 if (!VLRU_enabled) {
6318 Log("VLRU: disabled\n");
6322 /* initialize each of the VLRU queues */
6323 for (i = 0; i < VLRU_QUEUES; i++) {
6324 queue_Init(&volume_LRU.q[i]);
6325 volume_LRU.q[i].len = 0;
6326 volume_LRU.q[i].busy = 0;
6327 assert(pthread_cond_init(&volume_LRU.q[i].cv, NULL) == 0);
6330 /* setup the timing constants */
6331 VLRU_ComputeConstants();
6333 /* XXX put inside LogLevel check? */
6334 Log("VLRU: starting scanner with the following configuration parameters:\n");
6335 Log("VLRU: offlining volumes after minimum of %d seconds of inactivity\n", VLRU_offline_thresh);
6336 Log("VLRU: running VLRU soft detach pass every %d seconds\n", VLRU_offline_interval);
6337 Log("VLRU: taking up to %d volumes offline per pass\n", VLRU_offline_max);
6338 Log("VLRU: scanning generation 0 for inactive volumes every %d seconds\n", volume_LRU.scan_interval[0]);
6339 Log("VLRU: scanning for promotion/demotion between generations 0 and 1 every %d seconds\n", volume_LRU.promotion_interval[0]);
6340 Log("VLRU: scanning for promotion/demotion between generations 1 and 2 every %d seconds\n", volume_LRU.promotion_interval[1]);
6342 /* start up the VLRU scanner */
6343 volume_LRU.scanner_state = VLRU_SCANNER_STATE_OFFLINE;
6344 if (programType == fileServer) {
6345 assert(pthread_cond_init(&volume_LRU.cv, NULL) == 0);
6346 assert(pthread_attr_init(&attrs) == 0);
6347 assert(pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED) == 0);
6348 assert(pthread_create(&tid, &attrs, &VLRU_ScannerThread, NULL) == 0);
6353 * initialize the VLRU-related fields of a newly allocated volume object.
6355 * @param[in] vp pointer to volume object
6358 * @arg @c VOL_LOCK is held.
6359 * @arg volume object is not on a VLRU queue.
6361 * @post VLRU fields are initialized to indicate that volume object is not
6362 * currently registered with the VLRU subsystem
6366 * @internal volume package interal use only.
6369 VLRU_Init_Node_r(Volume * vp)
6374 assert(queue_IsNotOnQueue(&vp->vlru));
6375 vp->vlru.idx = VLRU_QUEUE_INVALID;
6379 * add a volume object to a VLRU queue.
6381 * @param[in] vp pointer to volume object
6384 * @arg @c VOL_LOCK is held.
6385 * @arg caller MUST hold a lightweight ref on @p vp.
6386 * @arg caller MUST NOT hold exclusive ownership of the VLRU queue.
6388 * @post the volume object is added to the appropriate VLRU queue
6390 * @note if @c vp->vlru.idx contains the index of a valid VLRU queue,
6391 * then the volume is added to that queue. Otherwise, the value
6392 * @c VLRU_QUEUE_NEW is stored into @c vp->vlru.idx and the
6393 * volume is added to the NEW generation queue.
6395 * @note @c VOL_LOCK may be dropped internally
6397 * @note Volume state is temporarily set to @c VOL_STATE_VLRU_ADD
6398 * during the add operation, and is restored to the previous
6399 * state prior to return.
6403 * @internal volume package internal use only.
6406 VLRU_Add_r(Volume * vp)
6409 VolState state_save;
6414 if (queue_IsOnQueue(&vp->vlru))
6417 state_save = VChangeState_r(vp, VOL_STATE_VLRU_ADD);
6420 if ((idx < 0) || (idx >= VLRU_QUEUE_INVALID)) {
6421 idx = VLRU_QUEUE_NEW;
6424 VLRU_Wait_r(&volume_LRU.q[idx]);
6426 /* repeat check since VLRU_Wait_r may have dropped
6428 if (queue_IsNotOnQueue(&vp->vlru)) {
6430 queue_Prepend(&volume_LRU.q[idx], &vp->vlru);
6431 volume_LRU.q[idx].len++;
6432 V_attachFlags(vp) |= VOL_ON_VLRU;
6433 vp->stats.last_promote = FT_ApproxTime();
6436 VChangeState_r(vp, state_save);
6440 * delete a volume object from a VLRU queue.
6442 * @param[in] vp pointer to volume object
6445 * @arg @c VOL_LOCK is held.
6446 * @arg caller MUST hold a lightweight ref on @p vp.
6447 * @arg caller MUST NOT hold exclusive ownership of the VLRU queue.
6449 * @post volume object is removed from the VLRU queue
6451 * @note @c VOL_LOCK may be dropped internally
6455 * @todo We should probably set volume state to something exlcusive
6456 * (as @c VLRU_Add_r does) prior to dropping @c VOL_LOCK.
6458 * @internal volume package internal use only.
6461 VLRU_Delete_r(Volume * vp)
6468 if (queue_IsNotOnQueue(&vp->vlru))
6474 if (idx == VLRU_QUEUE_INVALID)
6476 VLRU_Wait_r(&volume_LRU.q[idx]);
6477 } while (idx != vp->vlru.idx);
6479 /* now remove from the VLRU and update
6480 * the appropriate counter */
6481 queue_Remove(&vp->vlru);
6482 volume_LRU.q[idx].len--;
6483 vp->vlru.idx = VLRU_QUEUE_INVALID;
6484 V_attachFlags(vp) &= ~(VOL_ON_VLRU);
6488 * tell the VLRU subsystem that a volume was just accessed.
6490 * @param[in] vp pointer to volume object
6493 * @arg @c VOL_LOCK is held
6494 * @arg caller MUST hold a lightweight ref on @p vp
6495 * @arg caller MUST NOT hold exclusive ownership of any VLRU queue
6497 * @post volume VLRU access statistics are updated. If the volume was on
6498 * the VLRU soft detach candidate queue, it is moved to the NEW
6501 * @note @c VOL_LOCK may be dropped internally
6505 * @internal volume package internal use only.
6508 VLRU_UpdateAccess_r(Volume * vp)
6510 Volume * rvp = NULL;
6515 if (queue_IsNotOnQueue(&vp->vlru))
6518 assert(V_attachFlags(vp) & VOL_ON_VLRU);
6520 /* update the access timestamp */
6521 vp->stats.last_get = FT_ApproxTime();
6524 * if the volume is on the soft detach candidate
6525 * list, we need to safely move it back to a
6526 * regular generation. this has to be done
6527 * carefully so we don't race against the scanner
6531 /* if this volume is on the soft detach candidate queue,
6532 * then grab exclusive access to the necessary queues */
6533 if (vp->vlru.idx == VLRU_QUEUE_CANDIDATE) {
6535 VCreateReservation_r(rvp);
6537 VLRU_Wait_r(&volume_LRU.q[VLRU_QUEUE_NEW]);
6538 VLRU_BeginExclusive_r(&volume_LRU.q[VLRU_QUEUE_NEW]);
6539 VLRU_Wait_r(&volume_LRU.q[VLRU_QUEUE_CANDIDATE]);
6540 VLRU_BeginExclusive_r(&volume_LRU.q[VLRU_QUEUE_CANDIDATE]);
6543 /* make sure multiple threads don't race to update */
6544 if (vp->vlru.idx == VLRU_QUEUE_CANDIDATE) {
6545 VLRU_SwitchQueues(vp, VLRU_QUEUE_NEW, 1);
6549 VLRU_EndExclusive_r(&volume_LRU.q[VLRU_QUEUE_CANDIDATE]);
6550 VLRU_EndExclusive_r(&volume_LRU.q[VLRU_QUEUE_NEW]);
6551 VCancelReservation_r(rvp);
6556 * switch a volume between two VLRU queues.
6558 * @param[in] vp pointer to volume object
6559 * @param[in] new_idx index of VLRU queue onto which the volume will be moved
6560 * @param[in] append controls whether the volume will be appended or
6561 * prepended to the queue. A nonzero value means it will
6562 * be appended; zero means it will be prepended.
6564 * @pre The new (and old, if applicable) queue(s) must either be owned
6565 * exclusively by the calling thread for asynchronous manipulation,
6566 * or the queue(s) must be quiescent and VOL_LOCK must be held.
6567 * Please see VLRU_BeginExclusive_r, VLRU_EndExclusive_r and VLRU_Wait_r
6568 * for further details of the queue asynchronous processing mechanism.
6570 * @post If the volume object was already on a VLRU queue, it is
6571 * removed from the queue. Depending on the value of the append
6572 * parameter, the volume object is either appended or prepended
6573 * to the VLRU queue referenced by the new_idx parameter.
6577 * @see VLRU_BeginExclusive_r
6578 * @see VLRU_EndExclusive_r
6581 * @internal volume package internal use only.
6584 VLRU_SwitchQueues(Volume * vp, int new_idx, int append)
6586 if (queue_IsNotOnQueue(&vp->vlru))
6589 queue_Remove(&vp->vlru);
6590 volume_LRU.q[vp->vlru.idx].len--;
6592 /* put the volume back on the correct generational queue */
6594 queue_Append(&volume_LRU.q[new_idx], &vp->vlru);
6596 queue_Prepend(&volume_LRU.q[new_idx], &vp->vlru);
6599 volume_LRU.q[new_idx].len++;
6600 vp->vlru.idx = new_idx;
6604 * VLRU background thread.
6606 * The VLRU Scanner Thread is responsible for periodically scanning through
6607 * each VLRU queue looking for volumes which should be moved to another
6608 * queue, or soft detached.
6610 * @param[in] args unused thread arguments parameter
6612 * @return unused thread return value
6613 * @retval NULL always
6615 * @internal volume package internal use only.
6618 VLRU_ScannerThread(void * args)
6620 afs_uint32 now, min_delay, delay;
6621 int i, min_idx, min_op, overdue, state;
6623 /* set t=0 for promotion cycle to be
6624 * fileserver startup */
6625 now = FT_ApproxTime();
6626 for (i=0; i < VLRU_GENERATIONS-1; i++) {
6627 volume_LRU.last_promotion[i] = now;
6630 /* don't start the scanner until VLRU_offline_thresh
6631 * plus a small delay for VInitVolumePackage2 to finish
6634 sleep(VLRU_offline_thresh + 60);
6636 /* set t=0 for scan cycle to be now */
6637 now = FT_ApproxTime();
6638 for (i=0; i < VLRU_GENERATIONS+1; i++) {
6639 volume_LRU.last_scan[i] = now;
6643 if (volume_LRU.scanner_state == VLRU_SCANNER_STATE_OFFLINE) {
6644 volume_LRU.scanner_state = VLRU_SCANNER_STATE_ONLINE;
6647 while ((state = volume_LRU.scanner_state) != VLRU_SCANNER_STATE_SHUTTING_DOWN) {
6648 /* check to see if we've been asked to pause */
6649 if (volume_LRU.scanner_state == VLRU_SCANNER_STATE_PAUSING) {
6650 volume_LRU.scanner_state = VLRU_SCANNER_STATE_PAUSED;
6651 assert(pthread_cond_broadcast(&volume_LRU.cv) == 0);
6653 VOL_CV_WAIT(&volume_LRU.cv);
6654 } while (volume_LRU.scanner_state == VLRU_SCANNER_STATE_PAUSED);
6657 /* scheduling can happen outside the glock */
6660 /* figure out what is next on the schedule */
6662 /* figure out a potential schedule for the new generation first */
6664 min_delay = volume_LRU.scan_interval[0] + volume_LRU.last_scan[0] - now;
6667 if (min_delay > volume_LRU.scan_interval[0]) {
6668 /* unsigned overflow -- we're overdue to run this scan */
6673 /* if we're not overdue for gen 0, figure out schedule for candidate gen */
6675 i = VLRU_QUEUE_CANDIDATE;
6676 delay = volume_LRU.scan_interval[i] + volume_LRU.last_scan[i] - now;
6677 if (delay < min_delay) {
6681 if (delay > volume_LRU.scan_interval[i]) {
6682 /* unsigned overflow -- we're overdue to run this scan */
6689 /* if we're still not overdue for something, figure out schedules for promotions */
6690 for (i=0; !overdue && i < VLRU_GENERATIONS-1; i++) {
6691 delay = volume_LRU.promotion_interval[i] + volume_LRU.last_promotion[i] - now;
6692 if (delay < min_delay) {
6697 if (delay > volume_LRU.promotion_interval[i]) {
6698 /* unsigned overflow -- we're overdue to run this promotion */
6707 /* sleep as needed */
6712 /* do whatever is next */
6715 VLRU_Promote_r(min_idx);
6716 VLRU_Demote_r(min_idx+1);
6718 VLRU_Scan_r(min_idx);
6720 now = FT_ApproxTime();
6723 Log("VLRU scanner asked to go offline (scanner_state=%d)\n", state);
6725 /* signal that scanner is down */
6726 volume_LRU.scanner_state = VLRU_SCANNER_STATE_OFFLINE;
6727 assert(pthread_cond_broadcast(&volume_LRU.cv) == 0);
6733 * promote volumes from one VLRU generation to the next.
6735 * This routine scans a VLRU generation looking for volumes which are
6736 * eligible to be promoted to the next generation. All volumes which
6737 * meet the eligibility requirement are promoted.
6739 * Promotion eligibility is based upon meeting both of the following
6742 * @arg The volume has been accessed since the last promotion:
6743 * @c (vp->stats.last_get >= vp->stats.last_promote)
6744 * @arg The last promotion occurred at least
6745 * @c volume_LRU.promotion_interval[idx] seconds ago
6747 * As a performance optimization, promotions are "globbed". In other
6748 * words, we promote arbitrarily large contiguous sublists of elements
6751 * @param[in] idx VLRU queue index to scan
6755 * @internal VLRU internal use only.
6758 VLRU_Promote_r(int idx)
6760 int len, chaining, promote;
6761 afs_uint32 now, thresh;
6762 struct rx_queue *qp, *nqp;
6763 Volume * vp, *start = NULL, *end = NULL;
6765 /* get exclusive access to two chains, and drop the glock */
6766 VLRU_Wait_r(&volume_LRU.q[idx]);
6767 VLRU_BeginExclusive_r(&volume_LRU.q[idx]);
6768 VLRU_Wait_r(&volume_LRU.q[idx+1]);
6769 VLRU_BeginExclusive_r(&volume_LRU.q[idx+1]);
6772 thresh = volume_LRU.promotion_interval[idx];
6773 now = FT_ApproxTime();
6776 for (queue_ScanBackwards(&volume_LRU.q[idx], qp, nqp, rx_queue)) {
6777 vp = (Volume *)((char *)qp - offsetof(Volume, vlru));
6778 promote = (((vp->stats.last_promote + thresh) <= now) &&
6779 (vp->stats.last_get >= vp->stats.last_promote));
6787 /* promote and prepend chain */
6788 queue_MoveChainAfter(&volume_LRU.q[idx+1], &start->vlru, &end->vlru);
6802 /* promote and prepend */
6803 queue_MoveChainAfter(&volume_LRU.q[idx+1], &start->vlru, &end->vlru);
6807 volume_LRU.q[idx].len -= len;
6808 volume_LRU.q[idx+1].len += len;
6811 /* release exclusive access to the two chains */
6813 volume_LRU.last_promotion[idx] = now;
6814 VLRU_EndExclusive_r(&volume_LRU.q[idx+1]);
6815 VLRU_EndExclusive_r(&volume_LRU.q[idx]);
6818 /* run the demotions */
6820 VLRU_Demote_r(int idx)
6823 int len, chaining, demote;
6824 afs_uint32 now, thresh;
6825 struct rx_queue *qp, *nqp;
6826 Volume * vp, *start = NULL, *end = NULL;
6827 Volume ** salv_flag_vec = NULL;
6828 int salv_vec_offset = 0;
6830 assert(idx == VLRU_QUEUE_MID || idx == VLRU_QUEUE_OLD);
6832 /* get exclusive access to two chains, and drop the glock */
6833 VLRU_Wait_r(&volume_LRU.q[idx-1]);
6834 VLRU_BeginExclusive_r(&volume_LRU.q[idx-1]);
6835 VLRU_Wait_r(&volume_LRU.q[idx]);
6836 VLRU_BeginExclusive_r(&volume_LRU.q[idx]);
6839 /* no big deal if this allocation fails */
6840 if (volume_LRU.q[idx].len) {
6841 salv_flag_vec = (Volume **) malloc(volume_LRU.q[idx].len * sizeof(Volume *));
6844 now = FT_ApproxTime();
6845 thresh = volume_LRU.promotion_interval[idx-1];
6848 for (queue_ScanBackwards(&volume_LRU.q[idx], qp, nqp, rx_queue)) {
6849 vp = (Volume *)((char *)qp - offsetof(Volume, vlru));
6850 demote = (((vp->stats.last_promote + thresh) <= now) &&
6851 (vp->stats.last_get < (now - thresh)));
6853 /* we now do volume update list DONT_SALVAGE flag setting during
6854 * demotion passes */
6855 if (salv_flag_vec &&
6856 !(V_attachFlags(vp) & VOL_HDR_DONTSALV) &&
6858 (vp->updateTime < (now - SALVAGE_INTERVAL)) &&
6859 (V_attachState(vp) == VOL_STATE_ATTACHED)) {
6860 salv_flag_vec[salv_vec_offset++] = vp;
6861 VCreateReservation_r(vp);
6870 /* demote and append chain */
6871 queue_MoveChainBefore(&volume_LRU.q[idx-1], &start->vlru, &end->vlru);
6885 queue_MoveChainBefore(&volume_LRU.q[idx-1], &start->vlru, &end->vlru);
6889 volume_LRU.q[idx].len -= len;
6890 volume_LRU.q[idx-1].len += len;
6893 /* release exclusive access to the two chains */
6895 VLRU_EndExclusive_r(&volume_LRU.q[idx]);
6896 VLRU_EndExclusive_r(&volume_LRU.q[idx-1]);
6898 /* now go back and set the DONT_SALVAGE flags as appropriate */
6899 if (salv_flag_vec) {
6901 for (i = 0; i < salv_vec_offset; i++) {
6902 vp = salv_flag_vec[i];
6903 if (!(V_attachFlags(vp) & VOL_HDR_DONTSALV) &&
6904 (vp->updateTime < (now - SALVAGE_INTERVAL)) &&
6905 (V_attachState(vp) == VOL_STATE_ATTACHED)) {
6908 V_attachFlags(vp) |= VOL_HDR_DONTSALV;
6909 V_dontSalvage(vp) = DONT_SALVAGE;
6910 VUpdateVolume_r(&ec, vp, 0);
6914 VCancelReservation_r(vp);
6916 free(salv_flag_vec);
6920 /* run a pass of the VLRU GC scanner */
6922 VLRU_Scan_r(int idx)
6924 afs_uint32 now, thresh;
6925 struct rx_queue *qp, *nqp;
6929 assert(idx == VLRU_QUEUE_NEW || idx == VLRU_QUEUE_CANDIDATE);
6931 /* gain exclusive access to the idx VLRU */
6932 VLRU_Wait_r(&volume_LRU.q[idx]);
6933 VLRU_BeginExclusive_r(&volume_LRU.q[idx]);
6935 if (idx != VLRU_QUEUE_CANDIDATE) {
6936 /* gain exclusive access to the candidate VLRU */
6937 VLRU_Wait_r(&volume_LRU.q[VLRU_QUEUE_CANDIDATE]);
6938 VLRU_BeginExclusive_r(&volume_LRU.q[VLRU_QUEUE_CANDIDATE]);
6941 now = FT_ApproxTime();
6942 thresh = now - VLRU_offline_thresh;
6944 /* perform candidate selection and soft detaching */
6945 if (idx == VLRU_QUEUE_CANDIDATE) {
6946 /* soft detach some volumes from the candidate pool */
6950 for (i=0,queue_ScanBackwards(&volume_LRU.q[idx], qp, nqp, rx_queue)) {
6951 vp = (Volume *)((char *)qp - offsetof(Volume, vlru));
6952 if (i >= VLRU_offline_max) {
6955 /* check timestamp to see if it's a candidate for soft detaching */
6956 if (vp->stats.last_get <= thresh) {
6958 if (VCheckSoftDetach(vp, thresh))
6964 /* scan for volumes to become soft detach candidates */
6965 for (i=1,queue_ScanBackwards(&volume_LRU.q[idx], qp, nqp, rx_queue),i++) {
6966 vp = (Volume *)((char *)qp - offsetof(Volume, vlru));
6968 /* check timestamp to see if it's a candidate for soft detaching */
6969 if (vp->stats.last_get <= thresh) {
6970 VCheckSoftDetachCandidate(vp, thresh);
6973 if (!(i&0x7f)) { /* lock coarsening optimization */
6981 /* relinquish exclusive access to the VLRU chains */
6985 volume_LRU.last_scan[idx] = now;
6986 if (idx != VLRU_QUEUE_CANDIDATE) {
6987 VLRU_EndExclusive_r(&volume_LRU.q[VLRU_QUEUE_CANDIDATE]);
6989 VLRU_EndExclusive_r(&volume_LRU.q[idx]);
6992 /* check whether volume is safe to soft detach
6993 * caller MUST NOT hold a ref count on vp */
6995 VCheckSoftDetach(Volume * vp, afs_uint32 thresh)
6999 if (vp->nUsers || vp->nWaiters)
7002 if (vp->stats.last_get <= thresh) {
7003 ret = VSoftDetachVolume_r(vp, thresh);
7009 /* check whether volume should be made a
7010 * soft detach candidate */
7012 VCheckSoftDetachCandidate(Volume * vp, afs_uint32 thresh)
7015 if (vp->nUsers || vp->nWaiters)
7020 assert(idx == VLRU_QUEUE_NEW);
7022 if (vp->stats.last_get <= thresh) {
7023 /* move to candidate pool */
7024 queue_Remove(&vp->vlru);
7025 volume_LRU.q[VLRU_QUEUE_NEW].len--;
7026 queue_Prepend(&volume_LRU.q[VLRU_QUEUE_CANDIDATE], &vp->vlru);
7027 vp->vlru.idx = VLRU_QUEUE_CANDIDATE;
7028 volume_LRU.q[VLRU_QUEUE_CANDIDATE].len++;
7036 /* begin exclusive access on VLRU */
7038 VLRU_BeginExclusive_r(struct VLRU_q * q)
7040 assert(q->busy == 0);
7044 /* end exclusive access on VLRU */
7046 VLRU_EndExclusive_r(struct VLRU_q * q)
7050 assert(pthread_cond_broadcast(&q->cv) == 0);
7053 /* wait for another thread to end exclusive access on VLRU */
7055 VLRU_Wait_r(struct VLRU_q * q)
7058 VOL_CV_WAIT(&q->cv);
7063 * volume soft detach
7065 * caller MUST NOT hold a ref count on vp */
7067 VSoftDetachVolume_r(Volume * vp, afs_uint32 thresh)
7072 assert(vp->vlru.idx == VLRU_QUEUE_CANDIDATE);
7074 ts_save = vp->stats.last_get;
7075 if (ts_save > thresh)
7078 if (vp->nUsers || vp->nWaiters)
7081 if (VIsExclusiveState(V_attachState(vp))) {
7085 switch (V_attachState(vp)) {
7086 case VOL_STATE_UNATTACHED:
7087 case VOL_STATE_PREATTACHED:
7088 case VOL_STATE_ERROR:
7089 case VOL_STATE_GOING_OFFLINE:
7090 case VOL_STATE_SHUTTING_DOWN:
7091 case VOL_STATE_SALVAGING:
7092 volume_LRU.q[vp->vlru.idx].len--;
7094 /* create and cancel a reservation to
7095 * give the volume an opportunity to
7097 VCreateReservation_r(vp);
7098 queue_Remove(&vp->vlru);
7099 vp->vlru.idx = VLRU_QUEUE_INVALID;
7100 V_attachFlags(vp) &= ~(VOL_ON_VLRU);
7101 VCancelReservation_r(vp);
7107 /* hold the volume and take it offline.
7108 * no need for reservations, as VHold_r
7109 * takes care of that internally. */
7110 if (VHold_r(vp) == 0) {
7111 /* vhold drops the glock, so now we should
7112 * check to make sure we aren't racing against
7113 * other threads. if we are racing, offlining vp
7114 * would be wasteful, and block the scanner for a while
7118 (vp->shuttingDown) ||
7119 (vp->goingOffline) ||
7120 (vp->stats.last_get != ts_save)) {
7121 /* looks like we're racing someone else. bail */
7125 /* pull it off the VLRU */
7126 assert(vp->vlru.idx == VLRU_QUEUE_CANDIDATE);
7127 volume_LRU.q[VLRU_QUEUE_CANDIDATE].len--;
7128 queue_Remove(&vp->vlru);
7129 vp->vlru.idx = VLRU_QUEUE_INVALID;
7130 V_attachFlags(vp) &= ~(VOL_ON_VLRU);
7132 /* take if offline */
7133 VOffline_r(vp, "volume has been soft detached");
7135 /* invalidate the volume header cache */
7136 FreeVolumeHeader(vp);
7139 IncUInt64(&VStats.soft_detaches);
7140 vp->stats.soft_detaches++;
7142 /* put in pre-attached state so demand
7143 * attacher can work on it */
7144 VChangeState_r(vp, VOL_STATE_PREATTACHED);
7150 #endif /* AFS_DEMAND_ATTACH_FS */
7153 /***************************************************/
7154 /* Volume Header Cache routines */
7155 /***************************************************/
7158 * volume header cache.
7160 struct volume_hdr_LRU_t volume_hdr_LRU;
7163 * initialize the volume header cache.
7165 * @param[in] howMany number of header cache entries to preallocate
7167 * @pre VOL_LOCK held. Function has never been called before.
7169 * @post howMany cache entries are allocated, initialized, and added
7170 * to the LRU list. Header cache statistics are initialized.
7172 * @note only applicable to fileServer program type. Should only be
7173 * called once during volume package initialization.
7175 * @internal volume package internal use only.
7178 VInitVolumeHeaderCache(afs_uint32 howMany)
7180 register struct volHeader *hp;
7181 if (programType != fileServer)
7183 queue_Init(&volume_hdr_LRU);
7184 volume_hdr_LRU.stats.free = 0;
7185 volume_hdr_LRU.stats.used = howMany;
7186 volume_hdr_LRU.stats.attached = 0;
7187 hp = (struct volHeader *)(calloc(howMany, sizeof(struct volHeader)));
7191 /* We are using ReleaseVolumeHeader to initialize the values on the header list
7192 * to ensure they have the right values
7194 ReleaseVolumeHeader(hp++);
7198 * get a volume header and attach it to the volume object.
7200 * @param[in] vp pointer to volume object
7202 * @return cache entry status
7203 * @retval 0 volume header was newly attached; cache data is invalid
7204 * @retval 1 volume header was previously attached; cache data is valid
7206 * @pre VOL_LOCK held. For DAFS, lightweight ref must be held on volume object.
7208 * @post volume header attached to volume object. if necessary, header cache
7209 * entry on LRU is synchronized to disk. Header is removed from LRU list.
7211 * @note VOL_LOCK may be dropped
7213 * @warning this interface does not load header data from disk. it merely
7214 * attaches a header object to the volume object, and may sync the old
7215 * header cache data out to disk in the process.
7217 * @internal volume package internal use only.
7220 GetVolumeHeader(register Volume * vp)
7223 register struct volHeader *hd;
7225 static int everLogged = 0;
7227 #ifdef AFS_DEMAND_ATTACH_FS
7228 VolState vp_save = 0, back_save = 0;
7230 /* XXX debug 9/19/05 we've apparently got
7231 * a ref counting bug somewhere that's
7232 * breaking the nUsers == 0 => header on LRU
7234 if (vp->header && queue_IsNotOnQueue(vp->header)) {
7235 Log("nUsers == 0, but header not on LRU\n");
7240 old = (vp->header != NULL); /* old == volume already has a header */
7242 if (programType != fileServer) {
7243 /* for volume utilities, we allocate volHeaders as needed */
7245 hd = (struct volHeader *)calloc(1, sizeof(*vp->header));
7249 #ifdef AFS_DEMAND_ATTACH_FS
7250 V_attachFlags(vp) |= VOL_HDR_ATTACHED;
7254 /* for the fileserver, we keep a volume header cache */
7256 /* the header we previously dropped in the lru is
7257 * still available. pull it off the lru and return */
7260 assert(hd->back == vp);
7262 /* we need to grab a new element off the LRU */
7263 if (queue_IsNotEmpty(&volume_hdr_LRU)) {
7264 /* grab an element and pull off of LRU */
7265 hd = queue_First(&volume_hdr_LRU, volHeader);
7268 /* LRU is empty, so allocate a new volHeader
7269 * this is probably indicative of a leak, so let the user know */
7270 hd = (struct volHeader *)calloc(1, sizeof(struct volHeader));
7273 Log("****Allocated more volume headers, probably leak****\n");
7276 volume_hdr_LRU.stats.free++;
7279 /* this header used to belong to someone else.
7280 * we'll need to check if the header needs to
7281 * be sync'd out to disk */
7283 #ifdef AFS_DEMAND_ATTACH_FS
7284 /* if hd->back were in an exclusive state, then
7285 * its volHeader would not be on the LRU... */
7286 assert(!VIsExclusiveState(V_attachState(hd->back)));
7289 if (hd->diskstuff.inUse) {
7290 /* volume was in use, so we'll need to sync
7291 * its header to disk */
7293 #ifdef AFS_DEMAND_ATTACH_FS
7294 back_save = VChangeState_r(hd->back, VOL_STATE_UPDATING);
7295 vp_save = VChangeState_r(vp, VOL_STATE_HDR_ATTACHING);
7296 VCreateReservation_r(hd->back);
7300 WriteVolumeHeader_r(&error, hd->back);
7301 /* Ignore errors; catch them later */
7303 #ifdef AFS_DEMAND_ATTACH_FS
7308 hd->back->header = NULL;
7309 #ifdef AFS_DEMAND_ATTACH_FS
7310 V_attachFlags(hd->back) &= ~(VOL_HDR_ATTACHED | VOL_HDR_LOADED | VOL_HDR_IN_LRU);
7312 if (hd->diskstuff.inUse) {
7313 VChangeState_r(hd->back, back_save);
7314 VCancelReservation_r(hd->back);
7315 VChangeState_r(vp, vp_save);
7319 volume_hdr_LRU.stats.attached++;
7323 #ifdef AFS_DEMAND_ATTACH_FS
7324 V_attachFlags(vp) |= VOL_HDR_ATTACHED;
7327 volume_hdr_LRU.stats.free--;
7328 volume_hdr_LRU.stats.used++;
7330 IncUInt64(&VStats.hdr_gets);
7331 #ifdef AFS_DEMAND_ATTACH_FS
7332 IncUInt64(&vp->stats.hdr_gets);
7333 vp->stats.last_hdr_get = FT_ApproxTime();
7340 * make sure volume header is attached and contains valid cache data.
7342 * @param[out] ec outbound error code
7343 * @param[in] vp pointer to volume object
7345 * @pre VOL_LOCK held. For DAFS, lightweight ref held on vp.
7347 * @post header cache entry attached, and loaded with valid data, or
7348 * *ec is nonzero, and the header is released back into the LRU.
7350 * @internal volume package internal use only.
7353 LoadVolumeHeader(Error * ec, Volume * vp)
7355 #ifdef AFS_DEMAND_ATTACH_FS
7356 VolState state_save;
7360 if (vp->nUsers == 0 && !GetVolumeHeader(vp)) {
7361 IncUInt64(&VStats.hdr_loads);
7362 state_save = VChangeState_r(vp, VOL_STATE_HDR_LOADING);
7365 ReadHeader(ec, V_diskDataHandle(vp), (char *)&V_disk(vp),
7366 sizeof(V_disk(vp)), VOLUMEINFOMAGIC,
7368 IncUInt64(&vp->stats.hdr_loads);
7369 now = FT_ApproxTime();
7373 V_attachFlags(vp) |= VOL_HDR_LOADED;
7374 vp->stats.last_hdr_load = now;
7376 VChangeState_r(vp, state_save);
7378 #else /* AFS_DEMAND_ATTACH_FS */
7380 if (vp->nUsers == 0 && !GetVolumeHeader(vp)) {
7381 IncUInt64(&VStats.hdr_loads);
7383 ReadHeader(ec, V_diskDataHandle(vp), (char *)&V_disk(vp),
7384 sizeof(V_disk(vp)), VOLUMEINFOMAGIC,
7387 #endif /* AFS_DEMAND_ATTACH_FS */
7389 /* maintain (nUsers==0) => header in LRU invariant */
7390 FreeVolumeHeader(vp);
7395 * release a header cache entry back into the LRU list.
7397 * @param[in] hd pointer to volume header cache object
7399 * @pre VOL_LOCK held.
7401 * @post header cache object appended onto end of LRU list.
7403 * @note only applicable to fileServer program type.
7405 * @note used to place a header cache entry back into the
7406 * LRU pool without invalidating it as a cache entry.
7408 * @internal volume package internal use only.
7411 ReleaseVolumeHeader(register struct volHeader *hd)
7413 if (programType != fileServer)
7415 if (!hd || queue_IsOnQueue(hd)) /* no header, or header already released */
7417 queue_Append(&volume_hdr_LRU, hd);
7418 #ifdef AFS_DEMAND_ATTACH_FS
7420 V_attachFlags(hd->back) |= VOL_HDR_IN_LRU;
7423 volume_hdr_LRU.stats.free++;
7424 volume_hdr_LRU.stats.used--;
7428 * free/invalidate a volume header cache entry.
7430 * @param[in] vp pointer to volume object
7432 * @pre VOL_LOCK is held.
7434 * @post For fileserver, header cache entry is returned to LRU, and it is
7435 * invalidated as a cache entry. For volume utilities, the header
7436 * cache entry is freed.
7438 * @note For fileserver, this should be utilized instead of ReleaseVolumeHeader
7439 * whenever it is necessary to invalidate the header cache entry.
7441 * @see ReleaseVolumeHeader
7443 * @internal volume package internal use only.
7446 FreeVolumeHeader(register Volume * vp)
7448 register struct volHeader *hd = vp->header;
7451 if (programType == fileServer) {
7452 ReleaseVolumeHeader(hd);
7457 #ifdef AFS_DEMAND_ATTACH_FS
7458 V_attachFlags(vp) &= ~(VOL_HDR_ATTACHED | VOL_HDR_IN_LRU | VOL_HDR_LOADED);
7460 volume_hdr_LRU.stats.attached--;
7465 /***************************************************/
7466 /* Volume Hash Table routines */
7467 /***************************************************/
7470 * set size of volume object hash table.
7472 * @param[in] logsize log(2) of desired hash table size
7474 * @return operation status
7476 * @retval -1 failure
7478 * @pre MUST be called prior to VInitVolumePackage2
7480 * @post Volume Hash Table will have 2^logsize buckets
7483 VSetVolHashSize(int logsize)
7485 /* 64 to 16384 hash buckets seems like a reasonable range */
7486 if ((logsize < 6 ) || (logsize > 14)) {
7491 VolumeHashTable.Size = 1 << logsize;
7492 VolumeHashTable.Mask = VolumeHashTable.Size - 1;
7494 /* we can't yet support runtime modification of this
7495 * parameter. we'll need a configuration rwlock to
7496 * make runtime modification feasible.... */
7503 * initialize dynamic data structures for volume hash table.
7505 * @post hash table is allocated, and fields are initialized.
7507 * @internal volume package internal use only.
7510 VInitVolumeHash(void)
7514 VolumeHashTable.Table = (VolumeHashChainHead *) calloc(VolumeHashTable.Size,
7515 sizeof(VolumeHashChainHead));
7516 assert(VolumeHashTable.Table != NULL);
7518 for (i=0; i < VolumeHashTable.Size; i++) {
7519 queue_Init(&VolumeHashTable.Table[i]);
7520 #ifdef AFS_DEMAND_ATTACH_FS
7521 assert(pthread_cond_init(&VolumeHashTable.Table[i].chain_busy_cv, NULL) == 0);
7522 #endif /* AFS_DEMAND_ATTACH_FS */
7527 * add a volume object to the hash table.
7529 * @param[in] vp pointer to volume object
7530 * @param[in] hashid hash of volume id
7532 * @pre VOL_LOCK is held. For DAFS, caller must hold a lightweight
7535 * @post volume is added to hash chain.
7537 * @internal volume package internal use only.
7539 * @note For DAFS, VOL_LOCK may be dropped in order to wait for an
7540 * asynchronous hash chain reordering to finish.
7543 AddVolumeToHashTable(register Volume * vp, int hashid)
7545 VolumeHashChainHead * head;
7547 if (queue_IsOnQueue(vp))
7550 head = &VolumeHashTable.Table[VOLUME_HASH(hashid)];
7552 #ifdef AFS_DEMAND_ATTACH_FS
7553 /* wait for the hash chain to become available */
7556 V_attachFlags(vp) |= VOL_IN_HASH;
7557 vp->chainCacheCheck = ++head->cacheCheck;
7558 #endif /* AFS_DEMAND_ATTACH_FS */
7561 vp->hashid = hashid;
7562 queue_Append(head, vp);
7563 vp->vnodeHashOffset = VolumeHashOffset_r();
7567 * delete a volume object from the hash table.
7569 * @param[in] vp pointer to volume object
7571 * @pre VOL_LOCK is held. For DAFS, caller must hold a lightweight
7574 * @post volume is removed from hash chain.
7576 * @internal volume package internal use only.
7578 * @note For DAFS, VOL_LOCK may be dropped in order to wait for an
7579 * asynchronous hash chain reordering to finish.
7582 DeleteVolumeFromHashTable(register Volume * vp)
7584 VolumeHashChainHead * head;
7586 if (!queue_IsOnQueue(vp))
7589 head = &VolumeHashTable.Table[VOLUME_HASH(vp->hashid)];
7591 #ifdef AFS_DEMAND_ATTACH_FS
7592 /* wait for the hash chain to become available */
7595 V_attachFlags(vp) &= ~(VOL_IN_HASH);
7597 #endif /* AFS_DEMAND_ATTACH_FS */
7601 /* do NOT reset hashid to zero, as the online
7602 * salvager package may need to know the volume id
7603 * after the volume is removed from the hash */
7607 * lookup a volume object in the hash table given a volume id.
7609 * @param[out] ec error code return
7610 * @param[in] volumeId volume id
7611 * @param[in] hint volume object which we believe could be the correct
7614 * @return volume object pointer
7615 * @retval NULL no such volume id is registered with the hash table.
7617 * @pre VOL_LOCK is held. For DAFS, caller must hold a lightweight
7620 * @post volume object with the given id is returned. volume object and
7621 * hash chain access statistics are updated. hash chain may have
7624 * @note For DAFS, VOL_LOCK may be dropped in order to wait for an
7625 * asynchronous hash chain reordering operation to finish, or
7626 * in order for us to perform an asynchronous chain reordering.
7628 * @note Hash chain reorderings occur when the access count for the
7629 * volume object being looked up exceeds the sum of the previous
7630 * node's (the node ahead of it in the hash chain linked list)
7631 * access count plus the constant VOLUME_HASH_REORDER_THRESHOLD.
7633 * @note For DAFS, the hint parameter allows us to short-circuit if the
7634 * cacheCheck fields match between the hash chain head and the
7635 * hint volume object.
7638 VLookupVolume_r(Error * ec, VolId volumeId, Volume * hint)
7640 register int looks = 0;
7642 #ifdef AFS_DEMAND_ATTACH_FS
7645 VolumeHashChainHead * head;
7648 head = &VolumeHashTable.Table[VOLUME_HASH(volumeId)];
7650 #ifdef AFS_DEMAND_ATTACH_FS
7651 /* wait for the hash chain to become available */
7654 /* check to see if we can short circuit without walking the hash chain */
7655 if (hint && (hint->chainCacheCheck == head->cacheCheck)) {
7656 IncUInt64(&hint->stats.hash_short_circuits);
7659 #endif /* AFS_DEMAND_ATTACH_FS */
7661 /* someday we need to either do per-chain locks, RWlocks,
7662 * or both for volhash access.
7663 * (and move to a data structure with better cache locality) */
7665 /* search the chain for this volume id */
7666 for(queue_Scan(head, vp, np, Volume)) {
7668 if ((vp->hashid == volumeId)) {
7673 if (queue_IsEnd(head, vp)) {
7677 #ifdef AFS_DEMAND_ATTACH_FS
7678 /* update hash chain statistics */
7681 FillInt64(lks, 0, looks);
7682 AddUInt64(head->looks, lks, &head->looks);
7683 AddUInt64(VStats.hash_looks, lks, &VStats.hash_looks);
7684 IncUInt64(&head->gets);
7689 IncUInt64(&vp->stats.hash_lookups);
7691 /* for demand attach fileserver, we permit occasional hash chain reordering
7692 * so that frequently looked up volumes move towards the head of the chain */
7693 pp = queue_Prev(vp, Volume);
7694 if (!queue_IsEnd(head, pp)) {
7695 FillInt64(thresh, 0, VOLUME_HASH_REORDER_THRESHOLD);
7696 AddUInt64(thresh, pp->stats.hash_lookups, &thresh);
7697 if (GEInt64(vp->stats.hash_lookups, thresh)) {
7698 VReorderHash_r(head, pp, vp);
7702 /* update the short-circuit cache check */
7703 vp->chainCacheCheck = head->cacheCheck;
7705 #endif /* AFS_DEMAND_ATTACH_FS */
7710 #ifdef AFS_DEMAND_ATTACH_FS
7711 /* perform volume hash chain reordering.
7713 * advance a subchain beginning at vp ahead of
7714 * the adjacent subchain ending at pp */
7716 VReorderHash_r(VolumeHashChainHead * head, Volume * pp, Volume * vp)
7718 Volume *tp, *np, *lp;
7719 afs_uint64 move_thresh;
7721 /* this should never be called if the chain is already busy, so
7722 * no need to wait for other exclusive chain ops to finish */
7724 /* this is a rather heavy set of operations,
7725 * so let's set the chain busy flag and drop
7727 VHashBeginExclusive_r(head);
7730 /* scan forward in the chain from vp looking for the last element
7731 * in the chain we want to advance */
7732 FillInt64(move_thresh, 0, VOLUME_HASH_REORDER_CHAIN_THRESH);
7733 AddUInt64(move_thresh, pp->stats.hash_lookups, &move_thresh);
7734 for(queue_ScanFrom(head, vp, tp, np, Volume)) {
7735 if (LTInt64(tp->stats.hash_lookups, move_thresh)) {
7739 lp = queue_Prev(tp, Volume);
7741 /* scan backwards from pp to determine where to splice and
7742 * insert the subchain we're advancing */
7743 for(queue_ScanBackwardsFrom(head, pp, tp, np, Volume)) {
7744 if (GTInt64(tp->stats.hash_lookups, move_thresh)) {
7748 tp = queue_Next(tp, Volume);
7750 /* rebalance chain(vp,...,lp) ahead of chain(tp,...,pp) */
7751 queue_MoveChainBefore(tp,vp,lp);
7754 IncUInt64(&VStats.hash_reorders);
7756 IncUInt64(&head->reorders);
7758 /* wake up any threads waiting for the hash chain */
7759 VHashEndExclusive_r(head);
7763 /* demand-attach fs volume hash
7764 * asynchronous exclusive operations */
7767 * begin an asynchronous exclusive operation on a volume hash chain.
7769 * @param[in] head pointer to volume hash chain head object
7771 * @pre VOL_LOCK held. hash chain is quiescent.
7773 * @post hash chain marked busy.
7775 * @note this interface is used in conjunction with VHashEndExclusive_r and
7776 * VHashWait_r to perform asynchronous (wrt VOL_LOCK) operations on a
7777 * volume hash chain. Its main use case is hash chain reordering, which
7778 * has the potential to be a highly latent operation.
7780 * @see VHashEndExclusive_r
7785 * @internal volume package internal use only.
7788 VHashBeginExclusive_r(VolumeHashChainHead * head)
7790 assert(head->busy == 0);
7795 * relinquish exclusive ownership of a volume hash chain.
7797 * @param[in] head pointer to volume hash chain head object
7799 * @pre VOL_LOCK held. thread owns the hash chain exclusively.
7801 * @post hash chain is marked quiescent. threads awaiting use of
7802 * chain are awakened.
7804 * @see VHashBeginExclusive_r
7809 * @internal volume package internal use only.
7812 VHashEndExclusive_r(VolumeHashChainHead * head)
7816 assert(pthread_cond_broadcast(&head->chain_busy_cv) == 0);
7820 * wait for all asynchronous operations on a hash chain to complete.
7822 * @param[in] head pointer to volume hash chain head object
7824 * @pre VOL_LOCK held.
7826 * @post hash chain object is quiescent.
7828 * @see VHashBeginExclusive_r
7829 * @see VHashEndExclusive_r
7833 * @note This interface should be called before any attempt to
7834 * traverse the hash chain. It is permissible for a thread
7835 * to gain exclusive access to the chain, and then perform
7836 * latent operations on the chain asynchronously wrt the
7839 * @warning if waiting is necessary, VOL_LOCK is dropped
7841 * @internal volume package internal use only.
7844 VHashWait_r(VolumeHashChainHead * head)
7846 while (head->busy) {
7847 VOL_CV_WAIT(&head->chain_busy_cv);
7850 #endif /* AFS_DEMAND_ATTACH_FS */
7853 /***************************************************/
7854 /* Volume by Partition List routines */
7855 /***************************************************/
7858 * demand attach fileserver adds a
7859 * linked list of volumes to each
7860 * partition object, thus allowing
7861 * for quick enumeration of all
7862 * volumes on a partition
7865 #ifdef AFS_DEMAND_ATTACH_FS
7867 * add a volume to its disk partition VByPList.
7869 * @param[in] vp pointer to volume object
7871 * @pre either the disk partition VByPList is owned exclusively
7872 * by the calling thread, or the list is quiescent and
7875 * @post volume is added to disk partition VByPList
7879 * @warning it is the caller's responsibility to ensure list
7882 * @see VVByPListWait_r
7883 * @see VVByPListBeginExclusive_r
7884 * @see VVByPListEndExclusive_r
7886 * @internal volume package internal use only.
7889 AddVolumeToVByPList_r(Volume * vp)
7891 if (queue_IsNotOnQueue(&vp->vol_list)) {
7892 queue_Append(&vp->partition->vol_list, &vp->vol_list);
7893 V_attachFlags(vp) |= VOL_ON_VBYP_LIST;
7894 vp->partition->vol_list.len++;
7899 * delete a volume from its disk partition VByPList.
7901 * @param[in] vp pointer to volume object
7903 * @pre either the disk partition VByPList is owned exclusively
7904 * by the calling thread, or the list is quiescent and
7907 * @post volume is removed from the disk partition VByPList
7911 * @warning it is the caller's responsibility to ensure list
7914 * @see VVByPListWait_r
7915 * @see VVByPListBeginExclusive_r
7916 * @see VVByPListEndExclusive_r
7918 * @internal volume package internal use only.
7921 DeleteVolumeFromVByPList_r(Volume * vp)
7923 if (queue_IsOnQueue(&vp->vol_list)) {
7924 queue_Remove(&vp->vol_list);
7925 V_attachFlags(vp) &= ~(VOL_ON_VBYP_LIST);
7926 vp->partition->vol_list.len--;
7931 * begin an asynchronous exclusive operation on a VByPList.
7933 * @param[in] dp pointer to disk partition object
7935 * @pre VOL_LOCK held. VByPList is quiescent.
7937 * @post VByPList marked busy.
7939 * @note this interface is used in conjunction with VVByPListEndExclusive_r and
7940 * VVByPListWait_r to perform asynchronous (wrt VOL_LOCK) operations on a
7943 * @see VVByPListEndExclusive_r
7944 * @see VVByPListWait_r
7948 * @internal volume package internal use only.
7950 /* take exclusive control over the list */
7952 VVByPListBeginExclusive_r(struct DiskPartition64 * dp)
7954 assert(dp->vol_list.busy == 0);
7955 dp->vol_list.busy = 1;
7959 * relinquish exclusive ownership of a VByPList.
7961 * @param[in] dp pointer to disk partition object
7963 * @pre VOL_LOCK held. thread owns the VByPList exclusively.
7965 * @post VByPList is marked quiescent. threads awaiting use of
7966 * the list are awakened.
7968 * @see VVByPListBeginExclusive_r
7969 * @see VVByPListWait_r
7973 * @internal volume package internal use only.
7976 VVByPListEndExclusive_r(struct DiskPartition64 * dp)
7978 assert(dp->vol_list.busy);
7979 dp->vol_list.busy = 0;
7980 assert(pthread_cond_broadcast(&dp->vol_list.cv) == 0);
7984 * wait for all asynchronous operations on a VByPList to complete.
7986 * @param[in] dp pointer to disk partition object
7988 * @pre VOL_LOCK is held.
7990 * @post disk partition's VByP list is quiescent
7994 * @note This interface should be called before any attempt to
7995 * traverse the VByPList. It is permissible for a thread
7996 * to gain exclusive access to the list, and then perform
7997 * latent operations on the list asynchronously wrt the
8000 * @warning if waiting is necessary, VOL_LOCK is dropped
8002 * @see VVByPListEndExclusive_r
8003 * @see VVByPListBeginExclusive_r
8005 * @internal volume package internal use only.
8008 VVByPListWait_r(struct DiskPartition64 * dp)
8010 while (dp->vol_list.busy) {
8011 VOL_CV_WAIT(&dp->vol_list.cv);
8014 #endif /* AFS_DEMAND_ATTACH_FS */
8016 /***************************************************/
8017 /* Volume Cache Statistics routines */
8018 /***************************************************/
8021 VPrintCacheStats_r(void)
8023 afs_uint32 get_hi, get_lo, load_hi, load_lo;
8024 register struct VnodeClassInfo *vcp;
8025 vcp = &VnodeClassInfo[vLarge];
8026 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);
8027 vcp = &VnodeClassInfo[vSmall];
8028 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);
8029 SplitInt64(VStats.hdr_gets, get_hi, get_lo);
8030 SplitInt64(VStats.hdr_loads, load_hi, load_lo);
8031 Log("Volume header cache, %d entries, %d gets, %d replacements\n",
8032 VStats.hdr_cache_size, get_lo, load_lo);
8036 VPrintCacheStats(void)
8039 VPrintCacheStats_r();
8043 #ifdef AFS_DEMAND_ATTACH_FS
8045 UInt64ToDouble(afs_uint64 * x)
8047 static double c32 = 4.0 * 1.073741824 * 1000000000.0;
8049 SplitInt64(*x, h, l);
8050 return (((double)h) * c32) + ((double) l);
8054 DoubleToPrintable(double x, char * buf, int len)
8056 static double billion = 1000000000.0;
8059 y[0] = (afs_uint32) (x / (billion * billion));
8060 y[1] = (afs_uint32) ((x - (((double)y[0]) * billion * billion)) / billion);
8061 y[2] = (afs_uint32) (x - ((((double)y[0]) * billion * billion) + (((double)y[1]) * billion)));
8064 snprintf(buf, len, "%d%09d%09d", y[0], y[1], y[2]);
8066 snprintf(buf, len, "%d%09d", y[1], y[2]);
8068 snprintf(buf, len, "%d", y[2]);
8074 struct VLRUExtStatsEntry {
8078 struct VLRUExtStats {
8084 } queue_info[VLRU_QUEUE_INVALID];
8085 struct VLRUExtStatsEntry * vec;
8089 * add a 256-entry fudge factor onto the vector in case state changes
8090 * out from under us.
8092 #define VLRU_EXT_STATS_VEC_LEN_FUDGE 256
8095 * collect extended statistics for the VLRU subsystem.
8097 * @param[out] stats pointer to stats structure to be populated
8098 * @param[in] nvols number of volumes currently known to exist
8100 * @pre VOL_LOCK held
8102 * @post stats->vec allocated and populated
8104 * @return operation status
8109 VVLRUExtStats_r(struct VLRUExtStats * stats, afs_uint32 nvols)
8111 afs_uint32 cur, idx, len;
8112 struct rx_queue * qp, * nqp;
8114 struct VLRUExtStatsEntry * vec;
8116 len = nvols + VLRU_EXT_STATS_VEC_LEN_FUDGE;
8117 vec = stats->vec = calloc(len,
8118 sizeof(struct VLRUExtStatsEntry));
8124 for (idx = VLRU_QUEUE_NEW; idx < VLRU_QUEUE_INVALID; idx++) {
8125 VLRU_Wait_r(&volume_LRU.q[idx]);
8126 VLRU_BeginExclusive_r(&volume_LRU.q[idx]);
8129 stats->queue_info[idx].start = cur;
8131 for (queue_Scan(&volume_LRU.q[idx], qp, nqp, rx_queue)) {
8133 /* out of space in vec */
8136 vp = (Volume *)((char *)qp - offsetof(Volume, vlru));
8137 vec[cur].volid = vp->hashid;
8141 stats->queue_info[idx].len = cur - stats->queue_info[idx].start;
8144 VLRU_EndExclusive_r(&volume_LRU.q[idx]);
8152 #define ENUMTOSTRING(en) #en
8153 #define ENUMCASE(en) \
8155 return ENUMTOSTRING(en); \
8159 vlru_idx_to_string(int idx)
8162 ENUMCASE(VLRU_QUEUE_NEW);
8163 ENUMCASE(VLRU_QUEUE_MID);
8164 ENUMCASE(VLRU_QUEUE_OLD);
8165 ENUMCASE(VLRU_QUEUE_CANDIDATE);
8166 ENUMCASE(VLRU_QUEUE_HELD);
8167 ENUMCASE(VLRU_QUEUE_INVALID);
8169 return "**UNKNOWN**";
8174 VPrintExtendedCacheStats_r(int flags)
8177 afs_uint32 vol_sum = 0;
8184 struct stats looks, gets, reorders, len;
8185 struct stats ch_looks, ch_gets, ch_reorders;
8187 VolumeHashChainHead *head;
8189 struct VLRUExtStats vlru_stats;
8191 /* zero out stats */
8192 memset(&looks, 0, sizeof(struct stats));
8193 memset(&gets, 0, sizeof(struct stats));
8194 memset(&reorders, 0, sizeof(struct stats));
8195 memset(&len, 0, sizeof(struct stats));
8196 memset(&ch_looks, 0, sizeof(struct stats));
8197 memset(&ch_gets, 0, sizeof(struct stats));
8198 memset(&ch_reorders, 0, sizeof(struct stats));
8200 for (i = 0; i < VolumeHashTable.Size; i++) {
8201 head = &VolumeHashTable.Table[i];
8204 VHashBeginExclusive_r(head);
8207 ch_looks.sum = UInt64ToDouble(&head->looks);
8208 ch_gets.sum = UInt64ToDouble(&head->gets);
8209 ch_reorders.sum = UInt64ToDouble(&head->reorders);
8211 /* update global statistics */
8213 looks.sum += ch_looks.sum;
8214 gets.sum += ch_gets.sum;
8215 reorders.sum += ch_reorders.sum;
8216 len.sum += (double)head->len;
8217 vol_sum += head->len;
8220 len.min = (double) head->len;
8221 len.max = (double) head->len;
8222 looks.min = ch_looks.sum;
8223 looks.max = ch_looks.sum;
8224 gets.min = ch_gets.sum;
8225 gets.max = ch_gets.sum;
8226 reorders.min = ch_reorders.sum;
8227 reorders.max = ch_reorders.sum;
8229 if (((double)head->len) < len.min)
8230 len.min = (double) head->len;
8231 if (((double)head->len) > len.max)
8232 len.max = (double) head->len;
8233 if (ch_looks.sum < looks.min)
8234 looks.min = ch_looks.sum;
8235 else if (ch_looks.sum > looks.max)
8236 looks.max = ch_looks.sum;
8237 if (ch_gets.sum < gets.min)
8238 gets.min = ch_gets.sum;
8239 else if (ch_gets.sum > gets.max)
8240 gets.max = ch_gets.sum;
8241 if (ch_reorders.sum < reorders.min)
8242 reorders.min = ch_reorders.sum;
8243 else if (ch_reorders.sum > reorders.max)
8244 reorders.max = ch_reorders.sum;
8248 if ((flags & VOL_STATS_PER_CHAIN2) && queue_IsNotEmpty(head)) {
8249 /* compute detailed per-chain stats */
8250 struct stats hdr_loads, hdr_gets;
8251 double v_looks, v_loads, v_gets;
8253 /* initialize stats with data from first element in chain */
8254 vp = queue_First(head, Volume);
8255 v_looks = UInt64ToDouble(&vp->stats.hash_lookups);
8256 v_loads = UInt64ToDouble(&vp->stats.hdr_loads);
8257 v_gets = UInt64ToDouble(&vp->stats.hdr_gets);
8258 ch_gets.min = ch_gets.max = v_looks;
8259 hdr_loads.min = hdr_loads.max = v_loads;
8260 hdr_gets.min = hdr_gets.max = v_gets;
8261 hdr_loads.sum = hdr_gets.sum = 0;
8263 vp = queue_Next(vp, Volume);
8265 /* pull in stats from remaining elements in chain */
8266 for (queue_ScanFrom(head, vp, vp, np, Volume)) {
8267 v_looks = UInt64ToDouble(&vp->stats.hash_lookups);
8268 v_loads = UInt64ToDouble(&vp->stats.hdr_loads);
8269 v_gets = UInt64ToDouble(&vp->stats.hdr_gets);
8271 hdr_loads.sum += v_loads;
8272 hdr_gets.sum += v_gets;
8274 if (v_looks < ch_gets.min)
8275 ch_gets.min = v_looks;
8276 else if (v_looks > ch_gets.max)
8277 ch_gets.max = v_looks;
8279 if (v_loads < hdr_loads.min)
8280 hdr_loads.min = v_loads;
8281 else if (v_loads > hdr_loads.max)
8282 hdr_loads.max = v_loads;
8284 if (v_gets < hdr_gets.min)
8285 hdr_gets.min = v_gets;
8286 else if (v_gets > hdr_gets.max)
8287 hdr_gets.max = v_gets;
8290 /* compute per-chain averages */
8291 ch_gets.avg = ch_gets.sum / ((double)head->len);
8292 hdr_loads.avg = hdr_loads.sum / ((double)head->len);
8293 hdr_gets.avg = hdr_gets.sum / ((double)head->len);
8295 /* dump per-chain stats */
8296 Log("Volume hash chain %d : len=%d, looks=%s, reorders=%s\n",
8298 DoubleToPrintable(ch_looks.sum, pr_buf[0], sizeof(pr_buf[0])),
8299 DoubleToPrintable(ch_reorders.sum, pr_buf[1], sizeof(pr_buf[1])));
8300 Log("\tVolume gets : min=%s, max=%s, avg=%s, total=%s\n",
8301 DoubleToPrintable(ch_gets.min, pr_buf[0], sizeof(pr_buf[0])),
8302 DoubleToPrintable(ch_gets.max, pr_buf[1], sizeof(pr_buf[1])),
8303 DoubleToPrintable(ch_gets.avg, pr_buf[2], sizeof(pr_buf[2])),
8304 DoubleToPrintable(ch_gets.sum, pr_buf[3], sizeof(pr_buf[3])));
8305 Log("\tHDR gets : min=%s, max=%s, avg=%s, total=%s\n",
8306 DoubleToPrintable(hdr_gets.min, pr_buf[0], sizeof(pr_buf[0])),
8307 DoubleToPrintable(hdr_gets.max, pr_buf[1], sizeof(pr_buf[1])),
8308 DoubleToPrintable(hdr_gets.avg, pr_buf[2], sizeof(pr_buf[2])),
8309 DoubleToPrintable(hdr_gets.sum, pr_buf[3], sizeof(pr_buf[3])));
8310 Log("\tHDR loads : min=%s, max=%s, avg=%s, total=%s\n",
8311 DoubleToPrintable(hdr_loads.min, pr_buf[0], sizeof(pr_buf[0])),
8312 DoubleToPrintable(hdr_loads.max, pr_buf[1], sizeof(pr_buf[1])),
8313 DoubleToPrintable(hdr_loads.avg, pr_buf[2], sizeof(pr_buf[2])),
8314 DoubleToPrintable(hdr_loads.sum, pr_buf[3], sizeof(pr_buf[3])));
8315 } else if (flags & VOL_STATS_PER_CHAIN) {
8316 /* dump simple per-chain stats */
8317 Log("Volume hash chain %d : len=%d, looks=%s, gets=%s, reorders=%s\n",
8319 DoubleToPrintable(ch_looks.sum, pr_buf[0], sizeof(pr_buf[0])),
8320 DoubleToPrintable(ch_gets.sum, pr_buf[1], sizeof(pr_buf[1])),
8321 DoubleToPrintable(ch_reorders.sum, pr_buf[2], sizeof(pr_buf[2])));
8325 VHashEndExclusive_r(head);
8330 /* compute global averages */
8331 len.avg = len.sum / ((double)VolumeHashTable.Size);
8332 looks.avg = looks.sum / ((double)VolumeHashTable.Size);
8333 gets.avg = gets.sum / ((double)VolumeHashTable.Size);
8334 reorders.avg = reorders.sum / ((double)VolumeHashTable.Size);
8336 /* dump global stats */
8337 Log("Volume hash summary: %d buckets\n", VolumeHashTable.Size);
8338 Log(" chain length : min=%s, max=%s, avg=%s, total=%s\n",
8339 DoubleToPrintable(len.min, pr_buf[0], sizeof(pr_buf[0])),
8340 DoubleToPrintable(len.max, pr_buf[1], sizeof(pr_buf[1])),
8341 DoubleToPrintable(len.avg, pr_buf[2], sizeof(pr_buf[2])),
8342 DoubleToPrintable(len.sum, pr_buf[3], sizeof(pr_buf[3])));
8343 Log(" looks : min=%s, max=%s, avg=%s, total=%s\n",
8344 DoubleToPrintable(looks.min, pr_buf[0], sizeof(pr_buf[0])),
8345 DoubleToPrintable(looks.max, pr_buf[1], sizeof(pr_buf[1])),
8346 DoubleToPrintable(looks.avg, pr_buf[2], sizeof(pr_buf[2])),
8347 DoubleToPrintable(looks.sum, pr_buf[3], sizeof(pr_buf[3])));
8348 Log(" gets : min=%s, max=%s, avg=%s, total=%s\n",
8349 DoubleToPrintable(gets.min, pr_buf[0], sizeof(pr_buf[0])),
8350 DoubleToPrintable(gets.max, pr_buf[1], sizeof(pr_buf[1])),
8351 DoubleToPrintable(gets.avg, pr_buf[2], sizeof(pr_buf[2])),
8352 DoubleToPrintable(gets.sum, pr_buf[3], sizeof(pr_buf[3])));
8353 Log(" reorders : min=%s, max=%s, avg=%s, total=%s\n",
8354 DoubleToPrintable(reorders.min, pr_buf[0], sizeof(pr_buf[0])),
8355 DoubleToPrintable(reorders.max, pr_buf[1], sizeof(pr_buf[1])),
8356 DoubleToPrintable(reorders.avg, pr_buf[2], sizeof(pr_buf[2])),
8357 DoubleToPrintable(reorders.sum, pr_buf[3], sizeof(pr_buf[3])));
8359 /* print extended disk related statistics */
8361 struct DiskPartition64 * diskP;
8362 afs_uint32 vol_count[VOLMAXPARTS+1];
8363 byte part_exists[VOLMAXPARTS+1];
8367 memset(vol_count, 0, sizeof(vol_count));
8368 memset(part_exists, 0, sizeof(part_exists));
8372 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
8374 vol_count[id] = diskP->vol_list.len;
8375 part_exists[id] = 1;
8379 for (i = 0; i <= VOLMAXPARTS; i++) {
8380 if (part_exists[i]) {
8381 /* XXX while this is currently safe, it is a violation
8382 * of the VGetPartitionById_r interface contract. */
8383 diskP = VGetPartitionById_r(i, 0);
8385 Log("Partition %s has %d online volumes\n",
8386 VPartitionPath(diskP), diskP->vol_list.len);
8393 /* print extended VLRU statistics */
8394 if (VVLRUExtStats_r(&vlru_stats, vol_sum) == 0) {
8395 afs_uint32 idx, cur, lpos;
8399 Log("VLRU State Dump:\n\n");
8401 for (idx = VLRU_QUEUE_NEW; idx < VLRU_QUEUE_INVALID; idx++) {
8402 Log("\t%s:\n", vlru_idx_to_string(idx));
8405 for (cur = vlru_stats.queue_info[idx].start;
8406 cur < vlru_stats.queue_info[idx].len;
8408 line[lpos++] = vlru_stats.vec[cur].volid;
8410 Log("\t\t%u, %u, %u, %u, %u,\n",
8411 line[0], line[1], line[2], line[3], line[4]);
8420 Log("\t\t%u, %u, %u, %u, %u\n",
8421 line[0], line[1], line[2], line[3], line[4]);
8426 free(vlru_stats.vec);
8433 VPrintExtendedCacheStats(int flags)
8436 VPrintExtendedCacheStats_r(flags);
8439 #endif /* AFS_DEMAND_ATTACH_FS */
8442 VCanScheduleSalvage(void)
8444 return vol_opts.canScheduleSalvage;
8450 return vol_opts.canUseFSSYNC;
8454 VCanUseSALVSYNC(void)
8456 return vol_opts.canUseSALVSYNC;