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 #if defined(FAST_RESTART) && defined(AFS_DEMAND_ATTACH_FS)
494 # error FAST_RESTART and DAFS are incompatible. For the DAFS equivalent \
495 of FAST_RESTART, use the -unsafe-nosalvage fileserver argument
499 * assign default values to a VolumePackageOptions struct.
501 * Always call this on a VolumePackageOptions struct first, then set any
502 * specific options you want, then call VInitVolumePackage2.
504 * @param[in] pt caller's program type
505 * @param[out] opts volume package options
508 VOptDefaults(ProgramType pt, VolumePackageOptions *opts)
510 opts->nLargeVnodes = opts->nSmallVnodes = 5;
513 opts->canScheduleSalvage = 0;
514 opts->canUseFSSYNC = 0;
515 opts->canUseSALVSYNC = 0;
518 opts->unsafe_attach = 1;
519 #else /* !FAST_RESTART */
520 opts->unsafe_attach = 0;
521 #endif /* !FAST_RESTART */
525 opts->canScheduleSalvage = 1;
526 opts->canUseSALVSYNC = 1;
530 opts->canUseFSSYNC = 1;
534 opts->nLargeVnodes = 0;
535 opts->nSmallVnodes = 0;
537 opts->canScheduleSalvage = 1;
538 opts->canUseFSSYNC = 1;
548 VInitVolumePackage2(ProgramType pt, VolumePackageOptions * opts)
550 int errors = 0; /* Number of errors while finding vice partitions. */
555 memset(&VStats, 0, sizeof(VStats));
556 VStats.hdr_cache_size = 200;
558 VInitPartitionPackage();
560 #ifdef AFS_DEMAND_ATTACH_FS
561 if (programType == fileServer) {
564 VLRU_SetOptions(VLRU_SET_ENABLED, 0);
566 assert(pthread_key_create(&VThread_key, NULL) == 0);
569 #ifdef AFS_PTHREAD_ENV
570 assert(pthread_mutex_init(&vol_glock_mutex, NULL) == 0);
571 assert(pthread_mutex_init(&vol_trans_mutex, NULL) == 0);
572 assert(pthread_cond_init(&vol_put_volume_cond, NULL) == 0);
573 assert(pthread_cond_init(&vol_sleep_cond, NULL) == 0);
574 assert(pthread_cond_init(&vol_init_attach_cond, NULL) == 0);
575 #else /* AFS_PTHREAD_ENV */
577 #endif /* AFS_PTHREAD_ENV */
578 Lock_Init(&vol_listLock);
580 srandom(time(0)); /* For VGetVolumeInfo */
582 #ifdef AFS_DEMAND_ATTACH_FS
583 assert(pthread_mutex_init(&vol_salvsync_mutex, NULL) == 0);
584 #endif /* AFS_DEMAND_ATTACH_FS */
586 /* Ok, we have done enough initialization that fileserver can
587 * start accepting calls, even though the volumes may not be
588 * available just yet.
592 #if defined(AFS_DEMAND_ATTACH_FS) && defined(SALVSYNC_BUILD_SERVER)
593 if (programType == salvageServer) {
596 #endif /* AFS_DEMAND_ATTACH_FS */
597 #ifdef FSSYNC_BUILD_SERVER
598 if (programType == fileServer) {
602 #if defined(AFS_DEMAND_ATTACH_FS) && defined(SALVSYNC_BUILD_CLIENT)
603 if (VCanUseSALVSYNC()) {
604 /* establish a connection to the salvager at this point */
605 assert(VConnectSALV() != 0);
607 #endif /* AFS_DEMAND_ATTACH_FS */
609 if (opts->volcache > VStats.hdr_cache_size)
610 VStats.hdr_cache_size = opts->volcache;
611 VInitVolumeHeaderCache(VStats.hdr_cache_size);
613 VInitVnodes(vLarge, opts->nLargeVnodes);
614 VInitVnodes(vSmall, opts->nSmallVnodes);
617 errors = VAttachPartitions();
621 if (programType != fileServer) {
622 errors = VInitAttachVolumes(programType);
628 #ifdef FSSYNC_BUILD_CLIENT
629 if (VCanUseFSSYNC()) {
631 #ifdef AFS_DEMAND_ATTACH_FS
632 if (programType == salvageServer) {
633 Log("Unable to connect to file server; aborted\n");
636 #endif /* AFS_DEMAND_ATTACH_FS */
637 Log("Unable to connect to file server; will retry at need\n");
640 #endif /* FSSYNC_BUILD_CLIENT */
645 #if !defined(AFS_PTHREAD_ENV)
647 * Attach volumes in vice partitions
649 * @param[in] pt calling program type
652 * @note This is the original, non-threaded version of attach parititions.
654 * @post VInit state is 2
657 VInitAttachVolumes(ProgramType pt)
660 if (pt == fileServer) {
661 struct DiskPartition64 *diskP;
662 /* Attach all the volumes in this partition */
663 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
664 int nAttached = 0, nUnattached = 0;
665 assert(VAttachVolumesByPartition(diskP, &nAttached, &nUnattached) == 0);
669 VInit = 2; /* Initialized, and all volumes have been attached */
670 LWP_NoYieldSignal(VInitAttachVolumes);
674 #endif /* !AFS_PTHREAD_ENV */
676 #if defined(AFS_PTHREAD_ENV) && !defined(AFS_DEMAND_ATTACH_FS)
678 * Attach volumes in vice partitions
680 * @param[in] pt calling program type
683 * @note Threaded version of attach parititions.
685 * @post VInit state is 2
688 VInitAttachVolumes(ProgramType pt)
691 if (pt == fileServer) {
692 struct DiskPartition64 *diskP;
693 struct vinitvolumepackage_thread_t params;
694 struct diskpartition_queue_t * dpq;
695 int i, threads, parts;
697 pthread_attr_t attrs;
699 assert(pthread_cond_init(¶ms.thread_done_cv,NULL) == 0);
701 params.n_threads_complete = 0;
703 /* create partition work queue */
704 for (parts=0, diskP = DiskPartitionList; diskP; diskP = diskP->next, parts++) {
705 dpq = (diskpartition_queue_t *) malloc(sizeof(struct diskpartition_queue_t));
708 queue_Append(¶ms,dpq);
711 threads = MIN(parts, vol_attach_threads);
714 /* spawn off a bunch of initialization threads */
715 assert(pthread_attr_init(&attrs) == 0);
716 assert(pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED) == 0);
718 Log("VInitVolumePackage: beginning parallel fileserver startup\n");
719 Log("VInitVolumePackage: using %d threads to attach volumes on %d partitions\n",
723 for (i=0; i < threads; i++) {
726 assert(pthread_create
727 (&tid, &attrs, &VInitVolumePackageThread,
729 AFS_SIGSET_RESTORE();
732 while(params.n_threads_complete < threads) {
733 VOL_CV_WAIT(¶ms.thread_done_cv);
737 assert(pthread_attr_destroy(&attrs) == 0);
739 /* if we're only going to run one init thread, don't bother creating
741 Log("VInitVolumePackage: beginning single-threaded fileserver startup\n");
742 Log("VInitVolumePackage: using 1 thread to attach volumes on %d partition(s)\n",
745 VInitVolumePackageThread(¶ms);
748 assert(pthread_cond_destroy(¶ms.thread_done_cv) == 0);
751 VInit = 2; /* Initialized, and all volumes have been attached */
752 assert(pthread_cond_broadcast(&vol_init_attach_cond) == 0);
758 VInitVolumePackageThread(void * args) {
760 struct DiskPartition64 *diskP;
761 struct vinitvolumepackage_thread_t * params;
762 struct diskpartition_queue_t * dpq;
764 params = (vinitvolumepackage_thread_t *) args;
768 /* Attach all the volumes in this partition */
769 while (queue_IsNotEmpty(params)) {
770 int nAttached = 0, nUnattached = 0;
772 if (vinit_attach_abort) {
773 Log("Aborting initialization\n");
777 dpq = queue_First(params,diskpartition_queue_t);
783 assert(VAttachVolumesByPartition(diskP, &nAttached, &nUnattached) == 0);
789 params->n_threads_complete++;
790 pthread_cond_signal(¶ms->thread_done_cv);
794 #endif /* AFS_PTHREAD_ENV && !AFS_DEMAND_ATTACH_FS */
796 #if defined(AFS_DEMAND_ATTACH_FS)
798 * Attach volumes in vice partitions
800 * @param[in] pt calling program type
803 * @note Threaded version of attach partitions.
805 * @post VInit state is 2
808 VInitAttachVolumes(ProgramType pt)
811 if (pt == fileServer) {
813 struct DiskPartition64 *diskP;
814 struct partition_queue pq;
815 struct volume_init_queue vq;
817 int i, threads, parts;
819 pthread_attr_t attrs;
821 /* create partition work queue */
823 assert(pthread_cond_init(&(pq.cv), NULL) == 0);
824 assert(pthread_mutex_init(&(pq.mutex), NULL) == 0);
825 for (parts = 0, diskP = DiskPartitionList; diskP; diskP = diskP->next, parts++) {
826 struct diskpartition_queue_t *dp;
827 dp = (struct diskpartition_queue_t*)malloc(sizeof(struct diskpartition_queue_t));
830 queue_Append(&pq, dp);
833 /* number of worker threads; at least one, not to exceed the number of partitions */
834 threads = MIN(parts, vol_attach_threads);
836 /* create volume work queue */
838 assert(pthread_cond_init(&(vq.cv), NULL) == 0);
839 assert(pthread_mutex_init(&(vq.mutex), NULL) == 0);
841 assert(pthread_attr_init(&attrs) == 0);
842 assert(pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED) == 0);
844 Log("VInitVolumePackage: beginning parallel fileserver startup\n");
845 Log("VInitVolumePackage: using %d threads to pre-attach volumes on %d partitions\n",
848 /* create threads to scan disk partitions. */
849 for (i=0; i < threads; i++) {
850 struct vinitvolumepackage_thread_param *params;
853 params = (struct vinitvolumepackage_thread_param *)malloc(sizeof(struct vinitvolumepackage_thread_param));
857 params->nthreads = threads;
858 params->thread = i+1;
861 assert(pthread_create (&tid, &attrs, &VInitVolumePackageThread, (void*)params) == 0);
862 AFS_SIGSET_RESTORE();
865 VInitPreAttachVolumes(threads, &vq);
867 assert(pthread_attr_destroy(&attrs) == 0);
868 assert(pthread_cond_destroy(&pq.cv) == 0);
869 assert(pthread_mutex_destroy(&pq.mutex) == 0);
870 assert(pthread_cond_destroy(&vq.cv) == 0);
871 assert(pthread_mutex_destroy(&vq.mutex) == 0);
875 VInit = 2; /* Initialized, and all volumes have been attached */
876 assert(pthread_cond_broadcast(&vol_init_attach_cond) == 0);
883 * Volume package initialization worker thread. Scan partitions for volume
884 * header files. Gather batches of volume ids and dispatch them to
885 * the main thread to be preattached. The volume preattachement is done
886 * in the main thread to avoid global volume lock contention.
889 VInitVolumePackageThread(void *args)
891 struct vinitvolumepackage_thread_param *params;
892 struct DiskPartition64 *partition;
893 struct partition_queue *pq;
894 struct volume_init_queue *vq;
895 struct volume_init_batch *vb;
898 params = (struct vinitvolumepackage_thread_param *)args;
904 vb = (struct volume_init_batch*)malloc(sizeof(struct volume_init_batch));
906 vb->thread = params->thread;
910 Log("Scanning partitions on thread %d of %d\n", params->thread, params->nthreads);
911 while((partition = VInitNextPartition(pq))) {
915 Log("Partition %s: pre-attaching volumes\n", partition->name);
916 dirp = opendir(VPartitionPath(partition));
918 Log("opendir on Partition %s failed, errno=%d!\n", partition->name, errno);
921 while ((vid = VInitNextVolumeId(dirp))) {
922 Volume *vp = (Volume*)malloc(sizeof(Volume));
924 memset(vp, 0, sizeof(Volume));
925 vp->device = partition->device;
926 vp->partition = partition;
928 queue_Init(&vp->vnode_list);
929 assert(pthread_cond_init(&V_attachCV(vp), NULL) == 0);
931 vb->batch[vb->size++] = vp;
932 if (vb->size == VINIT_BATCH_MAX_SIZE) {
933 assert(pthread_mutex_lock(&vq->mutex) == 0);
934 queue_Append(vq, vb);
935 assert(pthread_cond_broadcast(&vq->cv) == 0);
936 assert(pthread_mutex_unlock(&vq->mutex) == 0);
938 vb = (struct volume_init_batch*)malloc(sizeof(struct volume_init_batch));
940 vb->thread = params->thread;
949 assert(pthread_mutex_lock(&vq->mutex) == 0);
950 queue_Append(vq, vb);
951 assert(pthread_cond_broadcast(&vq->cv) == 0);
952 assert(pthread_mutex_unlock(&vq->mutex) == 0);
954 Log("Partition scan thread %d of %d ended\n", params->thread, params->nthreads);
960 * Read next element from the pre-populated partition list.
962 static struct DiskPartition64*
963 VInitNextPartition(struct partition_queue *pq)
965 struct DiskPartition64 *partition;
966 struct diskpartition_queue_t *dp; /* queue element */
968 if (vinit_attach_abort) {
969 Log("Aborting volume preattach thread.\n");
973 /* get next partition to scan */
974 assert(pthread_mutex_lock(&pq->mutex) == 0);
975 if (queue_IsEmpty(pq)) {
976 assert(pthread_mutex_unlock(&pq->mutex) == 0);
979 dp = queue_First(pq, diskpartition_queue_t);
981 assert(pthread_mutex_unlock(&pq->mutex) == 0);
986 partition = dp->diskP;
992 * Find next volume id on the partition.
995 VInitNextVolumeId(DIR *dirp)
1001 while((d = readdir(dirp))) {
1002 if (vinit_attach_abort) {
1003 Log("Aborting volume preattach thread.\n");
1006 ext = strrchr(d->d_name, '.');
1007 if (d->d_name[0] == 'V' && ext && strcmp(ext, VHDREXT) == 0) {
1008 vid = VolumeNumber(d->d_name);
1012 Log("Warning: bogus volume header file: %s\n", d->d_name);
1019 * Preattach volumes in batches to avoid lock contention.
1022 VInitPreAttachVolumes(int nthreads, struct volume_init_queue *vq)
1024 struct volume_init_batch *vb;
1028 /* dequeue next volume */
1029 pthread_mutex_lock(&vq->mutex);
1030 if (queue_IsEmpty(vq)) {
1031 pthread_cond_wait(&vq->cv, &vq->mutex);
1033 vb = queue_First(vq, volume_init_batch);
1035 pthread_mutex_unlock(&vq->mutex);
1039 for (i = 0; i<vb->size; i++) {
1045 dup = VLookupVolume_r(&ec, vp->hashid, NULL);
1047 Log("Error looking up volume, code=%d\n", ec);
1050 Log("Warning: Duplicate volume id %d detected.\n", vp->hashid);
1053 /* put pre-attached volume onto the hash table
1054 * and bring it up to the pre-attached state */
1055 AddVolumeToHashTable(vp, vp->hashid);
1056 AddVolumeToVByPList_r(vp);
1057 VLRU_Init_Node_r(vp);
1058 VChangeState_r(vp, VOL_STATE_PREATTACHED);
1071 #endif /* AFS_DEMAND_ATTACH_FS */
1073 #if !defined(AFS_DEMAND_ATTACH_FS)
1075 * attach all volumes on a given disk partition
1078 VAttachVolumesByPartition(struct DiskPartition64 *diskP, int * nAttached, int * nUnattached)
1084 Log("Partition %s: attaching volumes\n", diskP->name);
1085 dirp = opendir(VPartitionPath(diskP));
1087 Log("opendir on Partition %s failed!\n", diskP->name);
1091 while ((dp = readdir(dirp))) {
1093 p = strrchr(dp->d_name, '.');
1095 if (vinit_attach_abort) {
1096 Log("Partition %s: abort attach volumes\n", diskP->name);
1100 if (p != NULL && strcmp(p, VHDREXT) == 0) {
1103 vp = VAttachVolumeByName(&error, diskP->name, dp->d_name,
1105 (*(vp ? nAttached : nUnattached))++;
1106 if (error == VOFFLINE)
1107 Log("Volume %d stays offline (/vice/offline/%s exists)\n", VolumeNumber(dp->d_name), dp->d_name);
1108 else if (LogLevel >= 5) {
1109 Log("Partition %s: attached volume %d (%s)\n",
1110 diskP->name, VolumeNumber(dp->d_name),
1119 Log("Partition %s: attached %d volumes; %d volumes not attached\n", diskP->name, *nAttached, *nUnattached);
1124 #endif /* !AFS_DEMAND_ATTACH_FS */
1126 /***************************************************/
1127 /* Shutdown routines */
1128 /***************************************************/
1132 * highly multithreaded volume package shutdown
1134 * with the demand attach fileserver extensions,
1135 * VShutdown has been modified to be multithreaded.
1136 * In order to achieve optimal use of many threads,
1137 * the shutdown code involves one control thread and
1138 * n shutdown worker threads. The control thread
1139 * periodically examines the number of volumes available
1140 * for shutdown on each partition, and produces a worker
1141 * thread allocation schedule. The idea is to eliminate
1142 * redundant scheduling computation on the workers by
1143 * having a single master scheduler.
1145 * The scheduler's objectives are:
1147 * each partition with volumes remaining gets allocated
1148 * at least 1 thread (assuming sufficient threads)
1150 * threads are allocated proportional to the number of
1151 * volumes remaining to be offlined. This ensures that
1152 * the OS I/O scheduler has many requests to elevator
1153 * seek on partitions that will (presumably) take the
1154 * longest amount of time (from now) to finish shutdown
1155 * (3) keep threads busy
1156 * when there are extra threads, they are assigned to
1157 * partitions using a simple round-robin algorithm
1159 * In the future, we may wish to add the ability to adapt
1160 * to the relative performance patterns of each disk
1165 * multi-step shutdown process
1167 * demand attach shutdown is a four-step process. Each
1168 * shutdown "pass" shuts down increasingly more difficult
1169 * volumes. The main purpose is to achieve better cache
1170 * utilization during shutdown.
1173 * shutdown volumes in the unattached, pre-attached
1176 * shutdown attached volumes with cached volume headers
1178 * shutdown all volumes in non-exclusive states
1180 * shutdown all remaining volumes
1183 #ifdef AFS_DEMAND_ATTACH_FS
1189 struct DiskPartition64 * diskP;
1190 struct diskpartition_queue_t * dpq;
1191 vshutdown_thread_t params;
1193 pthread_attr_t attrs;
1195 memset(¶ms, 0, sizeof(vshutdown_thread_t));
1198 Log("VShutdown: aborting attach volumes\n");
1199 vinit_attach_abort = 1;
1200 VOL_CV_WAIT(&vol_init_attach_cond);
1203 for (params.n_parts=0, diskP = DiskPartitionList;
1204 diskP; diskP = diskP->next, params.n_parts++);
1206 Log("VShutdown: shutting down on-line volumes on %d partition%s...\n",
1207 params.n_parts, params.n_parts > 1 ? "s" : "");
1209 if (vol_attach_threads > 1) {
1210 /* prepare for parallel shutdown */
1211 params.n_threads = vol_attach_threads;
1212 assert(pthread_mutex_init(¶ms.lock, NULL) == 0);
1213 assert(pthread_cond_init(¶ms.cv, NULL) == 0);
1214 assert(pthread_cond_init(¶ms.master_cv, NULL) == 0);
1215 assert(pthread_attr_init(&attrs) == 0);
1216 assert(pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED) == 0);
1217 queue_Init(¶ms);
1219 /* setup the basic partition information structures for
1220 * parallel shutdown */
1221 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1223 struct rx_queue * qp, * nqp;
1227 VVByPListWait_r(diskP);
1228 VVByPListBeginExclusive_r(diskP);
1231 for (queue_Scan(&diskP->vol_list, qp, nqp, rx_queue)) {
1232 vp = (Volume *)((char *)qp - offsetof(Volume, vol_list));
1236 Log("VShutdown: partition %s has %d volumes with attached headers\n",
1237 VPartitionPath(diskP), count);
1240 /* build up the pass 0 shutdown work queue */
1241 dpq = (struct diskpartition_queue_t *) malloc(sizeof(struct diskpartition_queue_t));
1242 assert(dpq != NULL);
1244 queue_Prepend(¶ms, dpq);
1246 params.part_pass_head[diskP->index] = queue_First(&diskP->vol_list, rx_queue);
1249 Log("VShutdown: beginning parallel fileserver shutdown\n");
1250 Log("VShutdown: using %d threads to offline volumes on %d partition%s\n",
1251 vol_attach_threads, params.n_parts, params.n_parts > 1 ? "s" : "" );
1253 /* do pass 0 shutdown */
1254 assert(pthread_mutex_lock(¶ms.lock) == 0);
1255 for (i=0; i < params.n_threads; i++) {
1256 assert(pthread_create
1257 (&tid, &attrs, &VShutdownThread,
1261 /* wait for all the pass 0 shutdowns to complete */
1262 while (params.n_threads_complete < params.n_threads) {
1263 assert(pthread_cond_wait(¶ms.master_cv, ¶ms.lock) == 0);
1265 params.n_threads_complete = 0;
1267 assert(pthread_cond_broadcast(¶ms.cv) == 0);
1268 assert(pthread_mutex_unlock(¶ms.lock) == 0);
1270 Log("VShutdown: pass 0 completed using the 1 thread per partition algorithm\n");
1271 Log("VShutdown: starting passes 1 through 3 using finely-granular mp-fast algorithm\n");
1273 /* run the parallel shutdown scheduler. it will drop the glock internally */
1274 ShutdownController(¶ms);
1276 /* wait for all the workers to finish pass 3 and terminate */
1277 while (params.pass < 4) {
1278 VOL_CV_WAIT(¶ms.cv);
1281 assert(pthread_attr_destroy(&attrs) == 0);
1282 assert(pthread_cond_destroy(¶ms.cv) == 0);
1283 assert(pthread_cond_destroy(¶ms.master_cv) == 0);
1284 assert(pthread_mutex_destroy(¶ms.lock) == 0);
1286 /* drop the VByPList exclusive reservations */
1287 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1288 VVByPListEndExclusive_r(diskP);
1289 Log("VShutdown: %s stats : (pass[0]=%d, pass[1]=%d, pass[2]=%d, pass[3]=%d)\n",
1290 VPartitionPath(diskP),
1291 params.stats[0][diskP->index],
1292 params.stats[1][diskP->index],
1293 params.stats[2][diskP->index],
1294 params.stats[3][diskP->index]);
1297 Log("VShutdown: shutdown finished using %d threads\n", params.n_threads);
1299 /* if we're only going to run one shutdown thread, don't bother creating
1301 Log("VShutdown: beginning single-threaded fileserver shutdown\n");
1303 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1304 VShutdownByPartition_r(diskP);
1308 Log("VShutdown: complete.\n");
1311 #else /* AFS_DEMAND_ATTACH_FS */
1317 register Volume *vp, *np;
1318 register afs_int32 code;
1321 Log("VShutdown: aborting attach volumes\n");
1322 vinit_attach_abort = 1;
1323 #ifdef AFS_PTHREAD_ENV
1324 VOL_CV_WAIT(&vol_init_attach_cond);
1326 LWP_WaitProcess(VInitAttachVolumes);
1327 #endif /* AFS_PTHREAD_ENV */
1330 Log("VShutdown: shutting down on-line volumes...\n");
1331 for (i = 0; i < VolumeHashTable.Size; i++) {
1332 /* try to hold first volume in the hash table */
1333 for (queue_Scan(&VolumeHashTable.Table[i],vp,np,Volume)) {
1337 Log("VShutdown: Attempting to take volume %u offline.\n",
1340 /* next, take the volume offline (drops reference count) */
1341 VOffline_r(vp, "File server was shut down");
1345 Log("VShutdown: complete.\n");
1347 #endif /* AFS_DEMAND_ATTACH_FS */
1360 * stop new activity (e.g. SALVSYNC) from occurring
1362 * Use this to make the volume package less busy; for example, during
1363 * shutdown. This doesn't actually shutdown/detach anything in the
1364 * volume package, but prevents certain processes from ocurring. For
1365 * example, preventing new SALVSYNC communication in DAFS. In theory, we
1366 * could also use this to prevent new volume attachment, or prevent
1367 * other programs from checking out volumes, etc.
1372 #ifdef AFS_DEMAND_ATTACH_FS
1373 /* make sure we don't try to contact the salvageserver, since it may
1374 * not be around anymore */
1375 vol_disallow_salvsync = 1;
1379 #ifdef AFS_DEMAND_ATTACH_FS
1382 * shutdown control thread
1385 ShutdownController(vshutdown_thread_t * params)
1388 struct DiskPartition64 * diskP;
1390 vshutdown_thread_t shadow;
1392 ShutdownCreateSchedule(params);
1394 while ((params->pass < 4) &&
1395 (params->n_threads_complete < params->n_threads)) {
1396 /* recompute schedule once per second */
1398 memcpy(&shadow, params, sizeof(vshutdown_thread_t));
1402 Log("ShutdownController: schedule version=%d, vol_remaining=%d, pass=%d\n",
1403 shadow.schedule_version, shadow.vol_remaining, shadow.pass);
1404 Log("ShutdownController: n_threads_complete=%d, n_parts_done_pass=%d\n",
1405 shadow.n_threads_complete, shadow.n_parts_done_pass);
1406 for (diskP = DiskPartitionList; diskP; diskP=diskP->next) {
1408 Log("ShutdownController: part[%d] : (len=%d, thread_target=%d, done_pass=%d, pass_head=%p)\n",
1410 diskP->vol_list.len,
1411 shadow.part_thread_target[id],
1412 shadow.part_done_pass[id],
1413 shadow.part_pass_head[id]);
1419 ShutdownCreateSchedule(params);
1423 /* create the shutdown thread work schedule.
1424 * this scheduler tries to implement fairness
1425 * by allocating at least 1 thread to each
1426 * partition with volumes to be shutdown,
1427 * and then it attempts to allocate remaining
1428 * threads based upon the amount of work left
1431 ShutdownCreateSchedule(vshutdown_thread_t * params)
1433 struct DiskPartition64 * diskP;
1434 int sum, thr_workload, thr_left;
1435 int part_residue[VOLMAXPARTS+1];
1438 /* compute the total number of outstanding volumes */
1440 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1441 sum += diskP->vol_list.len;
1444 params->schedule_version++;
1445 params->vol_remaining = sum;
1450 /* compute average per-thread workload */
1451 thr_workload = sum / params->n_threads;
1452 if (sum % params->n_threads)
1455 thr_left = params->n_threads;
1456 memset(&part_residue, 0, sizeof(part_residue));
1458 /* for fairness, give every partition with volumes remaining
1459 * at least one thread */
1460 for (diskP = DiskPartitionList; diskP && thr_left; diskP = diskP->next) {
1462 if (diskP->vol_list.len) {
1463 params->part_thread_target[id] = 1;
1466 params->part_thread_target[id] = 0;
1470 if (thr_left && thr_workload) {
1471 /* compute length-weighted workloads */
1474 for (diskP = DiskPartitionList; diskP && thr_left; diskP = diskP->next) {
1476 delta = (diskP->vol_list.len / thr_workload) -
1477 params->part_thread_target[id];
1481 if (delta < thr_left) {
1482 params->part_thread_target[id] += delta;
1485 params->part_thread_target[id] += thr_left;
1493 /* try to assign any leftover threads to partitions that
1494 * had volume lengths closer to needing thread_target+1 */
1495 int max_residue, max_id = 0;
1497 /* compute the residues */
1498 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1500 part_residue[id] = diskP->vol_list.len -
1501 (params->part_thread_target[id] * thr_workload);
1504 /* now try to allocate remaining threads to partitions with the
1505 * highest residues */
1508 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1510 if (part_residue[id] > max_residue) {
1511 max_residue = part_residue[id];
1520 params->part_thread_target[max_id]++;
1522 part_residue[max_id] = 0;
1527 /* punt and give any remaining threads equally to each partition */
1529 if (thr_left >= params->n_parts) {
1530 alloc = thr_left / params->n_parts;
1531 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1533 params->part_thread_target[id] += alloc;
1538 /* finish off the last of the threads */
1539 for (diskP = DiskPartitionList; thr_left && diskP; diskP = diskP->next) {
1541 params->part_thread_target[id]++;
1547 /* worker thread for parallel shutdown */
1549 VShutdownThread(void * args)
1551 vshutdown_thread_t * params;
1552 int found, pass, schedule_version_save, count;
1553 struct DiskPartition64 *diskP;
1554 struct diskpartition_queue_t * dpq;
1557 params = (vshutdown_thread_t *) args;
1559 /* acquire the shutdown pass 0 lock */
1560 assert(pthread_mutex_lock(¶ms->lock) == 0);
1562 /* if there's still pass 0 work to be done,
1563 * get a work entry, and do a pass 0 shutdown */
1564 if (queue_IsNotEmpty(params)) {
1565 dpq = queue_First(params, diskpartition_queue_t);
1567 assert(pthread_mutex_unlock(¶ms->lock) == 0);
1573 while (ShutdownVolumeWalk_r(diskP, 0, ¶ms->part_pass_head[id]))
1575 params->stats[0][diskP->index] = count;
1576 assert(pthread_mutex_lock(¶ms->lock) == 0);
1579 params->n_threads_complete++;
1580 if (params->n_threads_complete == params->n_threads) {
1581 /* notify control thread that all workers have completed pass 0 */
1582 assert(pthread_cond_signal(¶ms->master_cv) == 0);
1584 while (params->pass == 0) {
1585 assert(pthread_cond_wait(¶ms->cv, ¶ms->lock) == 0);
1589 assert(pthread_mutex_unlock(¶ms->lock) == 0);
1592 pass = params->pass;
1595 /* now escalate through the more complicated shutdowns */
1597 schedule_version_save = params->schedule_version;
1599 /* find a disk partition to work on */
1600 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1602 if (params->part_thread_target[id] && !params->part_done_pass[id]) {
1603 params->part_thread_target[id]--;
1610 /* hmm. for some reason the controller thread couldn't find anything for
1611 * us to do. let's see if there's anything we can do */
1612 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1614 if (diskP->vol_list.len && !params->part_done_pass[id]) {
1617 } else if (!params->part_done_pass[id]) {
1618 params->part_done_pass[id] = 1;
1619 params->n_parts_done_pass++;
1621 Log("VShutdown: done shutting down volumes on partition %s.\n",
1622 VPartitionPath(diskP));
1628 /* do work on this partition until either the controller
1629 * creates a new schedule, or we run out of things to do
1630 * on this partition */
1633 while (!params->part_done_pass[id] &&
1634 (schedule_version_save == params->schedule_version)) {
1635 /* ShutdownVolumeWalk_r will drop the glock internally */
1636 if (!ShutdownVolumeWalk_r(diskP, pass, ¶ms->part_pass_head[id])) {
1637 if (!params->part_done_pass[id]) {
1638 params->part_done_pass[id] = 1;
1639 params->n_parts_done_pass++;
1641 Log("VShutdown: done shutting down volumes on partition %s.\n",
1642 VPartitionPath(diskP));
1650 params->stats[pass][id] += count;
1652 /* ok, everyone is done this pass, proceed */
1655 params->n_threads_complete++;
1656 while (params->pass == pass) {
1657 if (params->n_threads_complete == params->n_threads) {
1658 /* we are the last thread to complete, so we will
1659 * reinitialize worker pool state for the next pass */
1660 params->n_threads_complete = 0;
1661 params->n_parts_done_pass = 0;
1663 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1665 params->part_done_pass[id] = 0;
1666 params->part_pass_head[id] = queue_First(&diskP->vol_list, rx_queue);
1669 /* compute a new thread schedule before releasing all the workers */
1670 ShutdownCreateSchedule(params);
1672 /* wake up all the workers */
1673 assert(pthread_cond_broadcast(¶ms->cv) == 0);
1676 Log("VShutdown: pass %d completed using %d threads on %d partitions\n",
1677 pass, params->n_threads, params->n_parts);
1680 VOL_CV_WAIT(¶ms->cv);
1683 pass = params->pass;
1697 /* shut down all volumes on a given disk partition
1699 * note that this function will not allow mp-fast
1700 * shutdown of a partition */
1702 VShutdownByPartition_r(struct DiskPartition64 * dp)
1708 /* wait for other exclusive ops to finish */
1709 VVByPListWait_r(dp);
1711 /* begin exclusive access */
1712 VVByPListBeginExclusive_r(dp);
1714 /* pick the low-hanging fruit first,
1715 * then do the complicated ones last
1716 * (has the advantage of keeping
1717 * in-use volumes up until the bitter end) */
1718 for (pass = 0, total=0; pass < 4; pass++) {
1719 pass_stats[pass] = ShutdownVByPForPass_r(dp, pass);
1720 total += pass_stats[pass];
1723 /* end exclusive access */
1724 VVByPListEndExclusive_r(dp);
1726 Log("VShutdownByPartition: shut down %d volumes on %s (pass[0]=%d, pass[1]=%d, pass[2]=%d, pass[3]=%d)\n",
1727 total, VPartitionPath(dp), pass_stats[0], pass_stats[1], pass_stats[2], pass_stats[3]);
1732 /* internal shutdown functionality
1734 * for multi-pass shutdown:
1735 * 0 to only "shutdown" {pre,un}attached and error state volumes
1736 * 1 to also shutdown attached volumes w/ volume header loaded
1737 * 2 to also shutdown attached volumes w/o volume header loaded
1738 * 3 to also shutdown exclusive state volumes
1740 * caller MUST hold exclusive access on the hash chain
1741 * because we drop vol_glock_mutex internally
1743 * this function is reentrant for passes 1--3
1744 * (e.g. multiple threads can cooperate to
1745 * shutdown a partition mp-fast)
1747 * pass 0 is not scaleable because the volume state data is
1748 * synchronized by vol_glock mutex, and the locking overhead
1749 * is too high to drop the lock long enough to do linked list
1753 ShutdownVByPForPass_r(struct DiskPartition64 * dp, int pass)
1755 struct rx_queue * q = queue_First(&dp->vol_list, rx_queue);
1758 while (ShutdownVolumeWalk_r(dp, pass, &q))
1764 /* conditionally shutdown one volume on partition dp
1765 * returns 1 if a volume was shutdown in this pass,
1768 ShutdownVolumeWalk_r(struct DiskPartition64 * dp, int pass,
1769 struct rx_queue ** idx)
1771 struct rx_queue *qp, *nqp;
1776 for (queue_ScanFrom(&dp->vol_list, qp, qp, nqp, rx_queue)) {
1777 vp = (Volume *) (((char *)qp) - offsetof(Volume, vol_list));
1781 if ((V_attachState(vp) != VOL_STATE_UNATTACHED) &&
1782 (V_attachState(vp) != VOL_STATE_ERROR) &&
1783 (V_attachState(vp) != VOL_STATE_PREATTACHED)) {
1787 if ((V_attachState(vp) == VOL_STATE_ATTACHED) &&
1788 (vp->header == NULL)) {
1792 if (VIsExclusiveState(V_attachState(vp))) {
1797 DeleteVolumeFromVByPList_r(vp);
1798 VShutdownVolume_r(vp);
1808 * shutdown a specific volume
1810 /* caller MUST NOT hold a heavyweight ref on vp */
1812 VShutdownVolume_r(Volume * vp)
1816 VCreateReservation_r(vp);
1818 if (LogLevel >= 5) {
1819 Log("VShutdownVolume_r: vid=%u, device=%d, state=%hu\n",
1820 vp->hashid, vp->partition->device, V_attachState(vp));
1823 /* wait for other blocking ops to finish */
1824 VWaitExclusiveState_r(vp);
1826 assert(VIsValidState(V_attachState(vp)));
1828 switch(V_attachState(vp)) {
1829 case VOL_STATE_SALVAGING:
1830 /* Leave salvaging volumes alone. Any in-progress salvages will
1831 * continue working after viced shuts down. This is intentional.
1834 case VOL_STATE_PREATTACHED:
1835 case VOL_STATE_ERROR:
1836 VChangeState_r(vp, VOL_STATE_UNATTACHED);
1837 case VOL_STATE_UNATTACHED:
1839 case VOL_STATE_GOING_OFFLINE:
1840 case VOL_STATE_SHUTTING_DOWN:
1841 case VOL_STATE_ATTACHED:
1845 Log("VShutdown: Attempting to take volume %u offline.\n",
1848 /* take the volume offline (drops reference count) */
1849 VOffline_r(vp, "File server was shut down");
1856 VCancelReservation_r(vp);
1860 #endif /* AFS_DEMAND_ATTACH_FS */
1863 /***************************************************/
1864 /* Header I/O routines */
1865 /***************************************************/
1867 /* open a descriptor for the inode (h),
1868 * read in an on-disk structure into buffer (to) of size (size),
1869 * verify versionstamp in structure has magic (magic) and
1870 * optionally verify version (version) if (version) is nonzero
1873 ReadHeader(Error * ec, IHandle_t * h, char *to, int size, bit32 magic,
1876 struct versionStamp *vsn;
1891 if (FDH_SEEK(fdP, 0, SEEK_SET) < 0) {
1893 FDH_REALLYCLOSE(fdP);
1896 vsn = (struct versionStamp *)to;
1897 if (FDH_READ(fdP, to, size) != size || vsn->magic != magic) {
1899 FDH_REALLYCLOSE(fdP);
1904 /* Check is conditional, in case caller wants to inspect version himself */
1905 if (version && vsn->version != version) {
1911 WriteVolumeHeader_r(Error * ec, Volume * vp)
1913 IHandle_t *h = V_diskDataHandle(vp);
1923 if (FDH_SEEK(fdP, 0, SEEK_SET) < 0) {
1925 FDH_REALLYCLOSE(fdP);
1928 if (FDH_WRITE(fdP, (char *)&V_disk(vp), sizeof(V_disk(vp)))
1929 != sizeof(V_disk(vp))) {
1931 FDH_REALLYCLOSE(fdP);
1937 /* VolumeHeaderToDisk
1938 * Allows for storing 64 bit inode numbers in on-disk volume header
1941 /* convert in-memory representation of a volume header to the
1942 * on-disk representation of a volume header */
1944 VolumeHeaderToDisk(VolumeDiskHeader_t * dh, VolumeHeader_t * h)
1947 memset(dh, 0, sizeof(VolumeDiskHeader_t));
1948 dh->stamp = h->stamp;
1950 dh->parent = h->parent;
1952 #ifdef AFS_64BIT_IOPS_ENV
1953 dh->volumeInfo_lo = (afs_int32) h->volumeInfo & 0xffffffff;
1954 dh->volumeInfo_hi = (afs_int32) (h->volumeInfo >> 32) & 0xffffffff;
1955 dh->smallVnodeIndex_lo = (afs_int32) h->smallVnodeIndex & 0xffffffff;
1956 dh->smallVnodeIndex_hi =
1957 (afs_int32) (h->smallVnodeIndex >> 32) & 0xffffffff;
1958 dh->largeVnodeIndex_lo = (afs_int32) h->largeVnodeIndex & 0xffffffff;
1959 dh->largeVnodeIndex_hi =
1960 (afs_int32) (h->largeVnodeIndex >> 32) & 0xffffffff;
1961 dh->linkTable_lo = (afs_int32) h->linkTable & 0xffffffff;
1962 dh->linkTable_hi = (afs_int32) (h->linkTable >> 32) & 0xffffffff;
1964 dh->volumeInfo_lo = h->volumeInfo;
1965 dh->smallVnodeIndex_lo = h->smallVnodeIndex;
1966 dh->largeVnodeIndex_lo = h->largeVnodeIndex;
1967 dh->linkTable_lo = h->linkTable;
1971 /* DiskToVolumeHeader
1972 * Converts an on-disk representation of a volume header to
1973 * the in-memory representation of a volume header.
1975 * Makes the assumption that AFS has *always*
1976 * zero'd the volume header file so that high parts of inode
1977 * numbers are 0 in older (SGI EFS) volume header files.
1980 DiskToVolumeHeader(VolumeHeader_t * h, VolumeDiskHeader_t * dh)
1982 memset(h, 0, sizeof(VolumeHeader_t));
1983 h->stamp = dh->stamp;
1985 h->parent = dh->parent;
1987 #ifdef AFS_64BIT_IOPS_ENV
1989 (Inode) dh->volumeInfo_lo | ((Inode) dh->volumeInfo_hi << 32);
1991 h->smallVnodeIndex =
1992 (Inode) dh->smallVnodeIndex_lo | ((Inode) dh->
1993 smallVnodeIndex_hi << 32);
1995 h->largeVnodeIndex =
1996 (Inode) dh->largeVnodeIndex_lo | ((Inode) dh->
1997 largeVnodeIndex_hi << 32);
1999 (Inode) dh->linkTable_lo | ((Inode) dh->linkTable_hi << 32);
2001 h->volumeInfo = dh->volumeInfo_lo;
2002 h->smallVnodeIndex = dh->smallVnodeIndex_lo;
2003 h->largeVnodeIndex = dh->largeVnodeIndex_lo;
2004 h->linkTable = dh->linkTable_lo;
2009 /***************************************************/
2010 /* Volume Attachment routines */
2011 /***************************************************/
2013 #ifdef AFS_DEMAND_ATTACH_FS
2015 * pre-attach a volume given its path.
2017 * @param[out] ec outbound error code
2018 * @param[in] partition partition path string
2019 * @param[in] name volume id string
2021 * @return volume object pointer
2023 * @note A pre-attached volume will only have its partition
2024 * and hashid fields initialized. At first call to
2025 * VGetVolume, the volume will be fully attached.
2029 VPreAttachVolumeByName(Error * ec, char *partition, char *name)
2033 vp = VPreAttachVolumeByName_r(ec, partition, name);
2039 * pre-attach a volume given its path.
2041 * @param[out] ec outbound error code
2042 * @param[in] partition path to vice partition
2043 * @param[in] name volume id string
2045 * @return volume object pointer
2047 * @pre VOL_LOCK held
2049 * @internal volume package internal use only.
2052 VPreAttachVolumeByName_r(Error * ec, char *partition, char *name)
2054 return VPreAttachVolumeById_r(ec,
2056 VolumeNumber(name));
2060 * pre-attach a volume given its path and numeric volume id.
2062 * @param[out] ec error code return
2063 * @param[in] partition path to vice partition
2064 * @param[in] volumeId numeric volume id
2066 * @return volume object pointer
2068 * @pre VOL_LOCK held
2070 * @internal volume package internal use only.
2073 VPreAttachVolumeById_r(Error * ec,
2078 struct DiskPartition64 *partp;
2082 assert(programType == fileServer);
2084 if (!(partp = VGetPartition_r(partition, 0))) {
2086 Log("VPreAttachVolumeById_r: Error getting partition (%s)\n", partition);
2090 vp = VLookupVolume_r(ec, volumeId, NULL);
2095 return VPreAttachVolumeByVp_r(ec, partp, vp, volumeId);
2099 * preattach a volume.
2101 * @param[out] ec outbound error code
2102 * @param[in] partp pointer to partition object
2103 * @param[in] vp pointer to volume object
2104 * @param[in] vid volume id
2106 * @return volume object pointer
2108 * @pre VOL_LOCK is held.
2110 * @warning Returned volume object pointer does not have to
2111 * equal the pointer passed in as argument vp. There
2112 * are potential race conditions which can result in
2113 * the pointers having different values. It is up to
2114 * the caller to make sure that references are handled
2115 * properly in this case.
2117 * @note If there is already a volume object registered with
2118 * the same volume id, its pointer MUST be passed as
2119 * argument vp. Failure to do so will result in a silent
2120 * failure to preattach.
2122 * @internal volume package internal use only.
2125 VPreAttachVolumeByVp_r(Error * ec,
2126 struct DiskPartition64 * partp,
2134 /* check to see if pre-attach already happened */
2136 (V_attachState(vp) != VOL_STATE_UNATTACHED) &&
2137 (V_attachState(vp) != VOL_STATE_PREATTACHED) &&
2138 !VIsErrorState(V_attachState(vp))) {
2140 * pre-attach is a no-op in all but the following cases:
2142 * - volume is unattached
2143 * - volume is in an error state
2144 * - volume is pre-attached
2146 Log("VPreattachVolumeByVp_r: volume %u not in quiescent state\n", vid);
2149 /* we're re-attaching a volume; clear out some old state */
2150 memset(&vp->salvage, 0, sizeof(struct VolumeOnlineSalvage));
2152 if (V_partition(vp) != partp) {
2153 /* XXX potential race */
2154 DeleteVolumeFromVByPList_r(vp);
2157 /* if we need to allocate a new Volume struct,
2158 * go ahead and drop the vol glock, otherwise
2159 * do the basic setup synchronised, as it's
2160 * probably not worth dropping the lock */
2163 /* allocate the volume structure */
2164 vp = nvp = (Volume *) malloc(sizeof(Volume));
2166 memset(vp, 0, sizeof(Volume));
2167 queue_Init(&vp->vnode_list);
2168 assert(pthread_cond_init(&V_attachCV(vp), NULL) == 0);
2171 /* link the volume with its associated vice partition */
2172 vp->device = partp->device;
2173 vp->partition = partp;
2176 vp->specialStatus = 0;
2178 /* if we dropped the lock, reacquire the lock,
2179 * check for pre-attach races, and then add
2180 * the volume to the hash table */
2183 nvp = VLookupVolume_r(ec, vid, NULL);
2188 } else if (nvp) { /* race detected */
2193 /* hack to make up for VChangeState_r() decrementing
2194 * the old state counter */
2195 VStats.state_levels[0]++;
2199 /* put pre-attached volume onto the hash table
2200 * and bring it up to the pre-attached state */
2201 AddVolumeToHashTable(vp, vp->hashid);
2202 AddVolumeToVByPList_r(vp);
2203 VLRU_Init_Node_r(vp);
2204 VChangeState_r(vp, VOL_STATE_PREATTACHED);
2207 Log("VPreAttachVolumeByVp_r: volume %u pre-attached\n", vp->hashid);
2215 #endif /* AFS_DEMAND_ATTACH_FS */
2217 /* Attach an existing volume, given its pathname, and return a
2218 pointer to the volume header information. The volume also
2219 normally goes online at this time. An offline volume
2220 must be reattached to make it go online */
2222 VAttachVolumeByName(Error * ec, char *partition, char *name, int mode)
2226 retVal = VAttachVolumeByName_r(ec, partition, name, mode);
2232 VAttachVolumeByName_r(Error * ec, char *partition, char *name, int mode)
2234 register Volume *vp = NULL;
2235 struct DiskPartition64 *partp;
2239 #ifdef AFS_DEMAND_ATTACH_FS
2240 VolumeStats stats_save;
2242 #endif /* AFS_DEMAND_ATTACH_FS */
2246 volumeId = VolumeNumber(name);
2248 if (!(partp = VGetPartition_r(partition, 0))) {
2250 Log("VAttachVolume: Error getting partition (%s)\n", partition);
2254 if (VRequiresPartLock()) {
2256 VLockPartition_r(partition);
2257 } else if (programType == fileServer) {
2258 #ifdef AFS_DEMAND_ATTACH_FS
2259 /* lookup the volume in the hash table */
2260 vp = VLookupVolume_r(ec, volumeId, NULL);
2266 /* save any counters that are supposed to
2267 * be monotonically increasing over the
2268 * lifetime of the fileserver */
2269 memcpy(&stats_save, &vp->stats, sizeof(VolumeStats));
2271 memset(&stats_save, 0, sizeof(VolumeStats));
2274 /* if there's something in the hash table, and it's not
2275 * in the pre-attach state, then we may need to detach
2276 * it before proceeding */
2277 if (vp && (V_attachState(vp) != VOL_STATE_PREATTACHED)) {
2278 VCreateReservation_r(vp);
2279 VWaitExclusiveState_r(vp);
2281 /* at this point state must be one of:
2290 if (vp->specialStatus == VBUSY)
2293 /* if it's already attached, see if we can return it */
2294 if (V_attachState(vp) == VOL_STATE_ATTACHED) {
2295 VGetVolumeByVp_r(ec, vp);
2296 if (V_inUse(vp) == fileServer) {
2297 VCancelReservation_r(vp);
2301 /* otherwise, we need to detach, and attempt to re-attach */
2302 VDetachVolume_r(ec, vp);
2304 Log("VAttachVolume: Error detaching old volume instance (%s)\n", name);
2307 /* if it isn't fully attached, delete from the hash tables,
2308 and let the refcounter handle the rest */
2309 DeleteVolumeFromHashTable(vp);
2310 DeleteVolumeFromVByPList_r(vp);
2313 VCancelReservation_r(vp);
2317 /* pre-attach volume if it hasn't been done yet */
2319 (V_attachState(vp) == VOL_STATE_UNATTACHED) ||
2320 (V_attachState(vp) == VOL_STATE_ERROR)) {
2322 vp = VPreAttachVolumeByVp_r(ec, partp, vp, volumeId);
2330 /* handle pre-attach races
2332 * multiple threads can race to pre-attach a volume,
2333 * but we can't let them race beyond that
2335 * our solution is to let the first thread to bring
2336 * the volume into an exclusive state win; the other
2337 * threads just wait until it finishes bringing the
2338 * volume online, and then they do a vgetvolumebyvp
2340 if (svp && (svp != vp)) {
2341 /* wait for other exclusive ops to finish */
2342 VCreateReservation_r(vp);
2343 VWaitExclusiveState_r(vp);
2345 /* get a heavyweight ref, kill the lightweight ref, and return */
2346 VGetVolumeByVp_r(ec, vp);
2347 VCancelReservation_r(vp);
2351 /* at this point, we are chosen as the thread to do
2352 * demand attachment for this volume. all other threads
2353 * doing a getvolume on vp->hashid will block until we finish */
2355 /* make sure any old header cache entries are invalidated
2356 * before proceeding */
2357 FreeVolumeHeader(vp);
2359 VChangeState_r(vp, VOL_STATE_ATTACHING);
2361 /* restore any saved counters */
2362 memcpy(&vp->stats, &stats_save, sizeof(VolumeStats));
2363 #else /* AFS_DEMAND_ATTACH_FS */
2364 vp = VGetVolume_r(ec, volumeId);
2366 if (V_inUse(vp) == fileServer)
2368 if (vp->specialStatus == VBUSY)
2370 VDetachVolume_r(ec, vp);
2372 Log("VAttachVolume: Error detaching volume (%s)\n", name);
2376 #endif /* AFS_DEMAND_ATTACH_FS */
2380 strcpy(path, VPartitionPath(partp));
2388 vp = (Volume *) calloc(1, sizeof(Volume));
2390 vp->hashid = volumeId;
2391 vp->device = partp->device;
2392 vp->partition = partp;
2393 queue_Init(&vp->vnode_list);
2394 #ifdef AFS_DEMAND_ATTACH_FS
2395 assert(pthread_cond_init(&V_attachCV(vp), NULL) == 0);
2396 #endif /* AFS_DEMAND_ATTACH_FS */
2399 /* attach2 is entered without any locks, and returns
2400 * with vol_glock_mutex held */
2401 vp = attach2(ec, volumeId, path, partp, vp, isbusy, mode);
2403 if (VCanUseFSSYNC() && vp) {
2404 #ifdef AFS_DEMAND_ATTACH_FS
2405 if ((mode == V_VOLUPD) || (VolumeWriteable(vp) && (mode == V_CLONE))) {
2406 /* mark volume header as in use so that volser crashes lead to a
2407 * salvage attempt */
2408 VUpdateVolume_r(ec, vp, 0);
2410 /* for dafs, we should tell the fileserver, except for V_PEEK
2411 * where we know it is not necessary */
2412 if (mode == V_PEEK) {
2413 vp->needsPutBack = 0;
2415 vp->needsPutBack = 1;
2417 #else /* !AFS_DEMAND_ATTACH_FS */
2418 /* duplicate computation in fssync.c about whether the server
2419 * takes the volume offline or not. If the volume isn't
2420 * offline, we must not return it when we detach the volume,
2421 * or the server will abort */
2422 if (mode == V_READONLY || mode == V_PEEK
2423 || (!VolumeWriteable(vp) && (mode == V_CLONE || mode == V_DUMP)))
2424 vp->needsPutBack = 0;
2426 vp->needsPutBack = 1;
2427 #endif /* !AFS_DEMAND_ATTACH_FS */
2429 /* OK, there's a problem here, but one that I don't know how to
2430 * fix right now, and that I don't think should arise often.
2431 * Basically, we should only put back this volume to the server if
2432 * it was given to us by the server, but since we don't have a vp,
2433 * we can't run the VolumeWriteable function to find out as we do
2434 * above when computing vp->needsPutBack. So we send it back, but
2435 * there's a path in VAttachVolume on the server which may abort
2436 * if this volume doesn't have a header. Should be pretty rare
2437 * for all of that to happen, but if it does, probably the right
2438 * fix is for the server to allow the return of readonly volumes
2439 * that it doesn't think are really checked out. */
2440 #ifdef FSSYNC_BUILD_CLIENT
2441 if (VCanUseFSSYNC() && vp == NULL &&
2442 mode != V_SECRETLY && mode != V_PEEK) {
2444 #ifdef AFS_DEMAND_ATTACH_FS
2445 /* If we couldn't attach but we scheduled a salvage, we already
2446 * notified the fileserver; don't online it now */
2447 if (*ec != VSALVAGING)
2448 #endif /* AFS_DEMAND_ATTACH_FS */
2449 FSYNC_VolOp(volumeId, partition, FSYNC_VOL_ON, 0, NULL);
2452 if (programType == fileServer && vp) {
2453 #ifdef AFS_DEMAND_ATTACH_FS
2455 * we can get here in cases where we don't "own"
2456 * the volume (e.g. volume owned by a utility).
2457 * short circuit around potential disk header races.
2459 if (V_attachState(vp) != VOL_STATE_ATTACHED) {
2463 VUpdateVolume_r(ec, vp, 0);
2465 Log("VAttachVolume: Error updating volume\n");
2470 if (VolumeWriteable(vp) && V_dontSalvage(vp) == 0) {
2471 #ifndef AFS_DEMAND_ATTACH_FS
2472 /* This is a hack: by temporarily setting the incore
2473 * dontSalvage flag ON, the volume will be put back on the
2474 * Update list (with dontSalvage OFF again). It will then
2475 * come back in N minutes with DONT_SALVAGE eventually
2476 * set. This is the way that volumes that have never had
2477 * it set get it set; or that volumes that have been
2478 * offline without DONT SALVAGE having been set also
2479 * eventually get it set */
2480 V_dontSalvage(vp) = DONT_SALVAGE;
2481 #endif /* !AFS_DEMAND_ATTACH_FS */
2482 VAddToVolumeUpdateList_r(ec, vp);
2484 Log("VAttachVolume: Error adding volume to update list\n");
2491 Log("VOnline: volume %u (%s) attached and online\n", V_id(vp),
2496 if (VRequiresPartLock()) {
2497 VUnlockPartition_r(partition);
2500 #ifdef AFS_DEMAND_ATTACH_FS
2501 /* attach failed; make sure we're in error state */
2502 if (vp && !VIsErrorState(V_attachState(vp))) {
2503 VChangeState_r(vp, VOL_STATE_ERROR);
2505 #endif /* AFS_DEMAND_ATTACH_FS */
2512 #ifdef AFS_DEMAND_ATTACH_FS
2513 /* VAttachVolumeByVp_r
2515 * finish attaching a volume that is
2516 * in a less than fully attached state
2518 /* caller MUST hold a ref count on vp */
2520 VAttachVolumeByVp_r(Error * ec, Volume * vp, int mode)
2522 char name[VMAXPATHLEN];
2524 struct DiskPartition64 *partp;
2528 Volume * nvp = NULL;
2529 VolumeStats stats_save;
2532 /* volume utility should never call AttachByVp */
2533 assert(programType == fileServer);
2535 volumeId = vp->hashid;
2536 partp = vp->partition;
2537 VolumeExternalName_r(volumeId, name, sizeof(name));
2540 /* if another thread is performing a blocking op, wait */
2541 VWaitExclusiveState_r(vp);
2543 memcpy(&stats_save, &vp->stats, sizeof(VolumeStats));
2545 /* if it's already attached, see if we can return it */
2546 if (V_attachState(vp) == VOL_STATE_ATTACHED) {
2547 VGetVolumeByVp_r(ec, vp);
2548 if (V_inUse(vp) == fileServer) {
2551 if (vp->specialStatus == VBUSY)
2553 VDetachVolume_r(ec, vp);
2555 Log("VAttachVolume: Error detaching volume (%s)\n", name);
2561 /* pre-attach volume if it hasn't been done yet */
2563 (V_attachState(vp) == VOL_STATE_UNATTACHED) ||
2564 (V_attachState(vp) == VOL_STATE_ERROR)) {
2565 nvp = VPreAttachVolumeByVp_r(ec, partp, vp, volumeId);
2571 VCreateReservation_r(nvp);
2577 VChangeState_r(vp, VOL_STATE_ATTACHING);
2579 /* restore monotonically increasing stats */
2580 memcpy(&vp->stats, &stats_save, sizeof(VolumeStats));
2584 /* compute path to disk header */
2585 strcpy(path, VPartitionPath(partp));
2594 * NOTE: attach2 is entered without any locks, and returns
2595 * with vol_glock_mutex held */
2596 vp = attach2(ec, volumeId, path, partp, vp, isbusy, mode);
2599 * the event that an error was encountered, or
2600 * the volume was not brought to an attached state
2601 * for any reason, skip to the end. We cannot
2602 * safely call VUpdateVolume unless we "own" it.
2606 (V_attachState(vp) != VOL_STATE_ATTACHED)) {
2610 VUpdateVolume_r(ec, vp, 0);
2612 Log("VAttachVolume: Error updating volume %u\n", vp->hashid);
2616 if (VolumeWriteable(vp) && V_dontSalvage(vp) == 0) {
2617 #ifndef AFS_DEMAND_ATTACH_FS
2618 /* This is a hack: by temporarily setting the incore
2619 * dontSalvage flag ON, the volume will be put back on the
2620 * Update list (with dontSalvage OFF again). It will then
2621 * come back in N minutes with DONT_SALVAGE eventually
2622 * set. This is the way that volumes that have never had
2623 * it set get it set; or that volumes that have been
2624 * offline without DONT SALVAGE having been set also
2625 * eventually get it set */
2626 V_dontSalvage(vp) = DONT_SALVAGE;
2627 #endif /* !AFS_DEMAND_ATTACH_FS */
2628 VAddToVolumeUpdateList_r(ec, vp);
2630 Log("VAttachVolume: Error adding volume %u to update list\n", vp->hashid);
2637 Log("VOnline: volume %u (%s) attached and online\n", V_id(vp),
2641 VCancelReservation_r(nvp);
2644 if (*ec && (*ec != VOFFLINE) && (*ec != VSALVAGE)) {
2645 if (vp && !VIsErrorState(V_attachState(vp))) {
2646 VChangeState_r(vp, VOL_STATE_ERROR);
2655 * lock a volume on disk (non-blocking).
2657 * @param[in] vp The volume to lock
2658 * @param[in] locktype READ_LOCK or WRITE_LOCK
2660 * @return operation status
2661 * @retval 0 success, lock was obtained
2662 * @retval EBUSY a conflicting lock was held by another process
2663 * @retval EIO error acquiring lock
2665 * @pre If we're in the fileserver, vp is in an exclusive state
2667 * @pre vp is not already locked
2670 VLockVolumeNB(Volume *vp, int locktype)
2674 assert(programType != fileServer || VIsExclusiveState(V_attachState(vp)));
2675 assert(!(V_attachFlags(vp) & VOL_LOCKED));
2677 code = VLockVolumeByIdNB(vp->hashid, vp->partition, locktype);
2679 V_attachFlags(vp) |= VOL_LOCKED;
2686 * unlock a volume on disk that was locked with VLockVolumeNB.
2688 * @param[in] vp volume to unlock
2690 * @pre If we're in the fileserver, vp is in an exclusive state
2692 * @pre vp has already been locked
2695 VUnlockVolume(Volume *vp)
2697 assert(programType != fileServer || VIsExclusiveState(V_attachState(vp)));
2698 assert((V_attachFlags(vp) & VOL_LOCKED));
2700 VUnlockVolumeById(vp->hashid, vp->partition);
2702 V_attachFlags(vp) &= ~VOL_LOCKED;
2704 #endif /* AFS_DEMAND_ATTACH_FS */
2707 * read in a vol header, possibly lock the vol header, and possibly check out
2708 * the vol header from the fileserver, as part of volume attachment.
2710 * @param[out] ec error code
2711 * @param[in] vp volume pointer object
2712 * @param[in] partp disk partition object of the attaching partition
2713 * @param[in] mode attachment mode such as V_VOLUPD, V_DUMP, etc (see
2715 * @param[in] peek 1 to just try to read in the volume header and make sure
2716 * we don't try to lock the vol, or check it out from
2717 * FSSYNC or anything like that; 0 otherwise, for 'normal'
2720 * @note As part of DAFS volume attachment, the volume header may be either
2721 * read- or write-locked to ensure mutual exclusion of certain volume
2722 * operations. In some cases in order to determine whether we need to
2723 * read- or write-lock the header, we need to read in the header to see
2724 * if the volume is RW or not. So, if we read in the header under a
2725 * read-lock and determine that we actually need a write-lock on the
2726 * volume header, this function will drop the read lock, acquire a write
2727 * lock, and read the header in again.
2730 attach_volume_header(Error *ec, Volume *vp, struct DiskPartition64 *partp,
2733 struct VolumeDiskHeader diskHeader;
2734 struct VolumeHeader header;
2737 int lock_tries = 0, checkout_tries = 0;
2739 VolumeId volid = vp->hashid;
2740 #ifdef FSSYNC_BUILD_CLIENT
2741 int checkout, done_checkout = 0;
2742 #endif /* FSSYNC_BUILD_CLIENT */
2743 #ifdef AFS_DEMAND_ATTACH_FS
2744 int locktype = 0, use_locktype = -1;
2745 #endif /* AFS_DEMAND_ATTACH_FS */
2751 if (lock_tries > VOL_MAX_CHECKOUT_RETRIES) {
2752 Log("VAttachVolume: retried too many times trying to lock header for "
2753 "vol %lu part %s; giving up\n", afs_printable_uint32_lu(volid),
2754 VPartitionPath(partp));
2758 if (checkout_tries > VOL_MAX_CHECKOUT_RETRIES) {
2759 Log("VAttachVolume: retried too many times trying to checkout "
2760 "vol %lu part %s; giving up\n", afs_printable_uint32_lu(volid),
2761 VPartitionPath(partp));
2766 if (VReadVolumeDiskHeader(volid, partp, NULL)) {
2767 /* short-circuit the 'volume does not exist' case */
2772 #ifdef FSSYNC_BUILD_CLIENT
2773 checkout = !done_checkout;
2775 if (!peek && checkout && VMustCheckoutVolume(mode)) {
2777 memset(&res, 0, sizeof(res));
2779 if (FSYNC_VolOp(volid, VPartitionPath(partp), FSYNC_VOL_NEEDVOLUME, mode, &res)
2782 if (res.hdr.reason == FSYNC_SALVAGE) {
2783 Log("VAttachVolume: file server says volume %lu is salvaging\n",
2784 afs_printable_uint32_lu(volid));
2787 Log("VAttachVolume: attach of volume %lu apparently denied by file server\n",
2788 afs_printable_uint32_lu(volid));
2789 *ec = VNOVOL; /* XXXX */
2796 #ifdef AFS_DEMAND_ATTACH_FS
2797 if (use_locktype < 0) {
2798 /* don't know whether vol is RO or RW; assume it's RO and we can retry
2799 * if it turns out to be RW */
2800 locktype = VVolLockType(mode, 0);
2803 /* a previous try says we should use use_locktype to lock the volume,
2805 locktype = use_locktype;
2808 if (!peek && locktype) {
2809 code = VLockVolumeNB(vp, locktype);
2811 if (code == EBUSY) {
2812 Log("VAttachVolume: another program has vol %lu locked\n",
2813 afs_printable_uint32_lu(volid));
2815 Log("VAttachVolume: error %d trying to lock vol %lu\n",
2816 code, afs_printable_uint32_lu(volid));
2823 #endif /* AFS_DEMAND_ATTACH_FS */
2825 code = VReadVolumeDiskHeader(volid, partp, &diskHeader);
2835 DiskToVolumeHeader(&header, &diskHeader);
2837 IH_INIT(vp->vnodeIndex[vLarge].handle, partp->device, header.parent,
2838 header.largeVnodeIndex);
2839 IH_INIT(vp->vnodeIndex[vSmall].handle, partp->device, header.parent,
2840 header.smallVnodeIndex);
2841 IH_INIT(vp->diskDataHandle, partp->device, header.parent,
2843 IH_INIT(vp->linkHandle, partp->device, header.parent, header.linkTable);
2846 /* only need to do this once */
2848 GetVolumeHeader(vp);
2852 #if defined(AFS_DEMAND_ATTACH_FS) && defined(FSSYNC_BUILD_CLIENT)
2853 /* demand attach changes the V_PEEK mechanism
2855 * we can now suck the current disk data structure over
2856 * the fssync interface without going to disk
2858 * (technically, we don't need to restrict this feature
2859 * to demand attach fileservers. However, I'm trying
2860 * to limit the number of common code changes)
2862 if (VCanUseFSSYNC() && (mode == V_PEEK || peek)) {
2864 res.payload.len = sizeof(VolumeDiskData);
2865 res.payload.buf = &vp->header->diskstuff;
2867 if (FSYNC_VolOp(vp->hashid,
2869 FSYNC_VOL_QUERY_HDR,
2872 goto disk_header_loaded;
2875 #endif /* AFS_DEMAND_ATTACH_FS && FSSYNC_BUILD_CLIENT */
2876 (void)ReadHeader(ec, V_diskDataHandle(vp), (char *)&V_disk(vp),
2877 sizeof(V_disk(vp)), VOLUMEINFOMAGIC, VOLUMEINFOVERSION);
2879 #ifdef AFS_DEMAND_ATTACH_FS
2882 IncUInt64(&VStats.hdr_loads);
2883 IncUInt64(&vp->stats.hdr_loads);
2885 #endif /* AFS_DEMAND_ATTACH_FS */
2888 Log("VAttachVolume: Error reading diskDataHandle header for vol %lu; "
2889 "error=%u\n", afs_printable_uint32_lu(volid), *ec);
2893 #ifdef AFS_DEMAND_ATTACH_FS
2894 # ifdef FSSYNC_BUILD_CLIENT
2896 # endif /* FSSYNC_BUILD_CLIENT */
2898 /* if the lock type we actually used to lock the volume is different than
2899 * the lock type we should have used, retry with the lock type we should
2901 use_locktype = VVolLockType(mode, VolumeWriteable(vp));
2902 if (locktype != use_locktype) {
2906 #endif /* AFS_DEMAND_ATTACH_FS */
2911 #if defined(AFS_DEMAND_ATTACH_FS) && defined(FSSYNC_BUILD_CLIENT)
2912 if (!peek && *ec == 0 && retry == 0 && VMustCheckoutVolume(mode)) {
2914 code = FSYNC_VerifyCheckout(volid, VPartitionPath(partp), FSYNC_VOL_NEEDVOLUME, mode);
2916 if (code == SYNC_DENIED) {
2917 /* must retry checkout; fileserver no longer thinks we have
2923 } else if (code != SYNC_OK) {
2927 #endif /* AFS_DEMAND_ATTACH_FS && FSSYNC_BUILD_CLIENT */
2930 /* either we are going to be called again for a second pass, or we
2931 * encountered an error; clean up in either case */
2933 #ifdef AFS_DEMAND_ATTACH_FS
2934 if ((V_attachFlags(vp) & VOL_LOCKED)) {
2937 #endif /* AFS_DEMAND_ATTACH_FS */
2938 if (vp->linkHandle) {
2939 IH_RELEASE(vp->vnodeIndex[vLarge].handle);
2940 IH_RELEASE(vp->vnodeIndex[vSmall].handle);
2941 IH_RELEASE(vp->diskDataHandle);
2942 IH_RELEASE(vp->linkHandle);
2955 #ifdef AFS_DEMAND_ATTACH_FS
2957 attach_check_vop(Error *ec, VolumeId volid, struct DiskPartition64 *partp,
2962 if (vp->pending_vol_op) {
2966 if (vp->pending_vol_op->vol_op_state == FSSYNC_VolOpRunningUnknown) {
2968 code = VVolOpLeaveOnlineNoHeader_r(vp, vp->pending_vol_op);
2970 vp->pending_vol_op->vol_op_state = FSSYNC_VolOpRunningOnline;
2971 } else if (code == 0) {
2972 vp->pending_vol_op->vol_op_state = FSSYNC_VolOpRunningOffline;
2975 /* we need the vol header to determine if the volume can be
2976 * left online for the vop, so... get the header */
2980 /* attach header with peek=1 to avoid checking out the volume
2981 * or locking it; we just want the header info, we're not
2982 * messing with the volume itself at all */
2983 attach_volume_header(ec, vp, partp, V_PEEK, 1);
2990 if (VVolOpLeaveOnline_r(vp, vp->pending_vol_op)) {
2991 vp->pending_vol_op->vol_op_state = FSSYNC_VolOpRunningOnline;
2993 vp->pending_vol_op->vol_op_state = FSSYNC_VolOpRunningOffline;
2996 /* make sure we grab a new vol header and re-open stuff on
2997 * actual attachment; we can't keep the data we grabbed, since
2998 * it was not done under a lock and thus not safe */
2999 FreeVolumeHeader(vp);
3000 VReleaseVolumeHandles_r(vp);
3003 /* see if the pending volume op requires exclusive access */
3004 switch (vp->pending_vol_op->vol_op_state) {
3005 case FSSYNC_VolOpPending:
3006 /* this should never happen */
3007 assert(vp->pending_vol_op->vol_op_state != FSSYNC_VolOpPending);
3010 case FSSYNC_VolOpRunningUnknown:
3011 /* this should never happen; we resolved 'unknown' above */
3012 assert(vp->pending_vol_op->vol_op_state != FSSYNC_VolOpRunningUnknown);
3015 case FSSYNC_VolOpRunningOffline:
3016 /* mark the volume down */
3018 VChangeState_r(vp, VOL_STATE_UNATTACHED);
3020 /* do not set V_offlineMessage here; we don't have ownership of
3021 * the volume (and probably do not have the header loaded), so we
3022 * can't alter the disk header */
3024 /* check to see if we should set the specialStatus flag */
3025 if (VVolOpSetVBusy_r(vp, vp->pending_vol_op)) {
3026 vp->specialStatus = VBUSY;
3037 #endif /* AFS_DEMAND_ATTACH_FS */
3040 * volume attachment helper function.
3042 * @param[out] ec error code
3043 * @param[in] volumeId volume ID of the attaching volume
3044 * @param[in] path full path to the volume header .vol file
3045 * @param[in] partp disk partition object for the attaching partition
3046 * @param[in] vp volume object; vp->hashid, vp->device, vp->partition,
3047 * vp->vnode_list, and V_attachCV (for DAFS) should already
3049 * @param[in] isbusy 1 if vp->specialStatus should be set to VBUSY; that is,
3050 * if there is a volume operation running for this volume
3051 * that should set the volume to VBUSY during its run. 0
3052 * otherwise. (see VVolOpSetVBusy_r)
3053 * @param[in] mode attachment mode such as V_VOLUPD, V_DUMP, etc (see
3056 * @return pointer to the semi-attached volume pointer
3057 * @retval NULL an error occurred (check value of *ec)
3058 * @retval vp volume successfully attaching
3060 * @pre no locks held
3062 * @post VOL_LOCK held
3065 attach2(Error * ec, VolId volumeId, char *path, struct DiskPartition64 *partp,
3066 Volume * vp, int isbusy, int mode)
3068 /* have we read in the header successfully? */
3069 int read_header = 0;
3071 /* should we FreeVolume(vp) instead of VCheckFree(vp) in the error
3077 vp->vnodeIndex[vLarge].handle = NULL;
3078 vp->vnodeIndex[vSmall].handle = NULL;
3079 vp->diskDataHandle = NULL;
3080 vp->linkHandle = NULL;
3082 #ifdef AFS_DEMAND_ATTACH_FS
3083 attach_check_vop(ec, volumeId, partp, vp);
3085 attach_volume_header(ec, vp, partp, mode, 0);
3088 attach_volume_header(ec, vp, partp, mode, 0);
3089 #endif /* !AFS_DEMAND_ATTACH_FS */
3091 if (*ec == VNOVOL) {
3092 /* if the volume doesn't exist, skip straight to 'error' so we don't
3093 * request a salvage */
3100 vp->specialStatus = (byte) (isbusy ? VBUSY : 0);
3101 vp->shuttingDown = 0;
3102 vp->goingOffline = 0;
3104 #ifdef AFS_DEMAND_ATTACH_FS
3105 vp->stats.last_attach = FT_ApproxTime();
3106 vp->stats.attaches++;
3110 IncUInt64(&VStats.attaches);
3111 vp->cacheCheck = ++VolumeCacheCheck;
3112 /* just in case this ever rolls over */
3113 if (!vp->cacheCheck)
3114 vp->cacheCheck = ++VolumeCacheCheck;
3117 #ifdef AFS_DEMAND_ATTACH_FS
3118 V_attachFlags(vp) |= VOL_HDR_LOADED;
3119 vp->stats.last_hdr_load = vp->stats.last_attach;
3120 #endif /* AFS_DEMAND_ATTACH_FS */
3124 struct IndexFileHeader iHead;
3126 #if OPENAFS_VOL_STATS
3128 * We just read in the diskstuff part of the header. If the detailed
3129 * volume stats area has not yet been initialized, we should bzero the
3130 * area and mark it as initialized.
3132 if (!(V_stat_initialized(vp))) {
3133 memset((V_stat_area(vp)), 0, VOL_STATS_BYTES);
3134 V_stat_initialized(vp) = 1;
3136 #endif /* OPENAFS_VOL_STATS */
3138 (void)ReadHeader(ec, vp->vnodeIndex[vSmall].handle,
3139 (char *)&iHead, sizeof(iHead),
3140 SMALLINDEXMAGIC, SMALLINDEXVERSION);
3143 Log("VAttachVolume: Error reading smallVnode vol header %s; error=%u\n", path, *ec);
3148 struct IndexFileHeader iHead;
3150 (void)ReadHeader(ec, vp->vnodeIndex[vLarge].handle,
3151 (char *)&iHead, sizeof(iHead),
3152 LARGEINDEXMAGIC, LARGEINDEXVERSION);
3155 Log("VAttachVolume: Error reading largeVnode vol header %s; error=%u\n", path, *ec);
3159 #ifdef AFS_NAMEI_ENV
3161 struct versionStamp stamp;
3163 (void)ReadHeader(ec, V_linkHandle(vp), (char *)&stamp,
3164 sizeof(stamp), LINKTABLEMAGIC, LINKTABLEVERSION);
3167 Log("VAttachVolume: Error reading namei vol header %s; error=%u\n", path, *ec);
3170 #endif /* AFS_NAMEI_ENV */
3172 #if defined(AFS_DEMAND_ATTACH_FS)
3173 if (*ec && ((*ec != VOFFLINE) || (V_attachState(vp) != VOL_STATE_UNATTACHED))) {
3175 if (!VCanScheduleSalvage()) {
3176 Log("VAttachVolume: Error attaching volume %s; volume needs salvage; error=%u\n", path, *ec);
3178 VRequestSalvage_r(ec, vp, SALVSYNC_ERROR, VOL_SALVAGE_INVALIDATE_HEADER);
3183 /* volume operation in progress */
3187 #else /* AFS_DEMAND_ATTACH_FS */
3189 Log("VAttachVolume: Error attaching volume %s; volume needs salvage; error=%u\n", path, *ec);
3193 #endif /* AFS_DEMAND_ATTACH_FS */
3195 if (V_needsSalvaged(vp)) {
3196 if (vp->specialStatus)
3197 vp->specialStatus = 0;
3199 #if defined(AFS_DEMAND_ATTACH_FS)
3200 if (!VCanScheduleSalvage()) {
3201 Log("VAttachVolume: volume salvage flag is ON for %s; volume needs salvage\n", path);
3203 VRequestSalvage_r(ec, vp, SALVSYNC_NEEDED, VOL_SALVAGE_INVALIDATE_HEADER);
3206 #else /* AFS_DEMAND_ATTACH_FS */
3208 #endif /* AFS_DEMAND_ATTACH_FS */
3214 vp->nextVnodeUnique = V_uniquifier(vp);
3216 if (VShouldCheckInUse(mode) && V_inUse(vp) && VolumeWriteable(vp)) {
3217 if (!V_needsSalvaged(vp)) {
3218 V_needsSalvaged(vp) = 1;
3219 VUpdateVolume_r(ec, vp, 0);
3221 #if defined(AFS_DEMAND_ATTACH_FS)
3222 if (!VCanScheduleSalvage()) {
3223 Log("VAttachVolume: volume %s needs to be salvaged; not attached.\n", path);
3225 VRequestSalvage_r(ec, vp, SALVSYNC_NEEDED, VOL_SALVAGE_INVALIDATE_HEADER);
3228 #else /* AFS_DEMAND_ATTACH_FS */
3229 Log("VAttachVolume: volume %s needs to be salvaged; not attached.\n", path);
3231 #endif /* AFS_DEMAND_ATTACH_FS */
3236 if (programType == fileServer && V_destroyMe(vp) == DESTROY_ME) {
3237 /* Only check destroyMe if we are the fileserver, since the
3238 * volserver et al sometimes need to work with volumes with
3239 * destroyMe set. Examples are 'temporary' volumes the
3240 * volserver creates, and when we create a volume (destroyMe
3241 * is set on creation; sometimes a separate volserver
3242 * transaction is created to clear destroyMe).
3245 #if defined(AFS_DEMAND_ATTACH_FS)
3246 /* schedule a salvage so the volume goes away on disk */
3247 VRequestSalvage_r(ec, vp, SALVSYNC_ERROR, VOL_SALVAGE_INVALIDATE_HEADER);
3248 VChangeState_r(vp, VOL_STATE_ERROR);
3250 #endif /* AFS_DEMAND_ATTACH_FS */
3251 Log("VAttachVolume: volume %s is junk; it should be destroyed at next salvage\n", path);
3257 vp->vnodeIndex[vSmall].bitmap = vp->vnodeIndex[vLarge].bitmap = NULL;
3258 #ifndef BITMAP_LATER
3259 if (programType == fileServer && VolumeWriteable(vp)) {
3261 for (i = 0; i < nVNODECLASSES; i++) {
3262 VGetBitmap_r(ec, vp, i);
3264 #ifdef AFS_DEMAND_ATTACH_FS
3265 VRequestSalvage_r(ec, vp, SALVSYNC_ERROR, VOL_SALVAGE_INVALIDATE_HEADER);
3267 #endif /* AFS_DEMAND_ATTACH_FS */
3268 Log("VAttachVolume: error getting bitmap for volume (%s)\n",
3274 #endif /* BITMAP_LATER */
3276 if (VInit >= 2 && V_needsCallback(vp)) {
3277 if (V_BreakVolumeCallbacks) {
3278 Log("VAttachVolume: Volume %lu was changed externally; breaking callbacks\n",
3279 afs_printable_uint32_lu(V_id(vp)));
3280 V_needsCallback(vp) = 0;
3282 (*V_BreakVolumeCallbacks) (V_id(vp));
3285 VUpdateVolume_r(ec, vp, 0);
3287 #ifdef FSSYNC_BUILD_CLIENT
3288 else if (VCanUseFSSYNC()) {
3289 afs_int32 fsync_code;
3291 V_needsCallback(vp) = 0;
3293 fsync_code = FSYNC_VolOp(V_id(vp), NULL, FSYNC_VOL_BREAKCBKS, FSYNC_WHATEVER, NULL);
3297 V_needsCallback(vp) = 1;
3298 Log("Error trying to tell the fileserver to break callbacks for "
3299 "changed volume %lu; error code %ld\n",
3300 afs_printable_uint32_lu(V_id(vp)),
3301 afs_printable_int32_ld(fsync_code));
3303 VUpdateVolume_r(ec, vp, 0);
3306 #endif /* FSSYNC_BUILD_CLIENT */
3309 Log("VAttachVolume: error %d clearing needsCallback on volume "
3310 "%lu; needs salvage\n", (int)*ec,
3311 afs_printable_uint32_lu(V_id(vp)));
3312 #ifdef AFS_DEMAND_ATTACH_FS
3313 VRequestSalvage_r(ec, vp, SALVSYNC_ERROR, VOL_SALVAGE_INVALIDATE_HEADER);
3315 #else /* !AFS_DEMAND_ATTACH_FS */
3317 #endif /* !AFS_DEMAND_ATTACh_FS */
3322 if (programType == fileServer) {
3323 if (vp->specialStatus)
3324 vp->specialStatus = 0;
3325 if (V_blessed(vp) && V_inService(vp) && !V_needsSalvaged(vp)) {
3326 V_inUse(vp) = fileServer;
3327 V_offlineMessage(vp)[0] = '\0';
3330 #ifdef AFS_DEMAND_ATTACH_FS
3331 if ((mode != V_PEEK) && (mode != V_SECRETLY))
3332 V_inUse(vp) = programType;
3333 #endif /* AFS_DEMAND_ATTACH_FS */
3334 V_checkoutMode(vp) = mode;
3337 AddVolumeToHashTable(vp, V_id(vp));
3338 #ifdef AFS_DEMAND_ATTACH_FS
3339 if (VCanUnlockAttached() && (V_attachFlags(vp) & VOL_LOCKED)) {
3342 if ((programType != fileServer) ||
3343 (V_inUse(vp) == fileServer)) {
3344 AddVolumeToVByPList_r(vp);
3346 VChangeState_r(vp, VOL_STATE_ATTACHED);
3348 VChangeState_r(vp, VOL_STATE_UNATTACHED);
3355 #ifdef AFS_DEMAND_ATTACH_FS
3356 if (!VIsErrorState(V_attachState(vp))) {
3357 VChangeState_r(vp, VOL_STATE_ERROR);
3359 #endif /* AFS_DEMAND_ATTACH_FS */
3362 VReleaseVolumeHandles_r(vp);
3365 #ifdef AFS_DEMAND_ATTACH_FS
3372 #else /* !AFS_DEMAND_ATTACH_FS */
3374 #endif /* !AFS_DEMAND_ATTACH_FS */
3378 /* Attach an existing volume.
3379 The volume also normally goes online at this time.
3380 An offline volume must be reattached to make it go online.
3384 VAttachVolume(Error * ec, VolumeId volumeId, int mode)
3388 retVal = VAttachVolume_r(ec, volumeId, mode);
3394 VAttachVolume_r(Error * ec, VolumeId volumeId, int mode)
3397 VGetVolumePath(ec, volumeId, &part, &name);
3399 register Volume *vp;
3401 vp = VGetVolume_r(&error, volumeId);
3403 assert(V_inUse(vp) == 0);
3404 VDetachVolume_r(ec, vp);
3408 return VAttachVolumeByName_r(ec, part, name, mode);
3411 /* Increment a reference count to a volume, sans context swaps. Requires
3412 * possibly reading the volume header in from the disk, since there's
3413 * an invariant in the volume package that nUsers>0 ==> vp->header is valid.
3415 * N.B. This call can fail if we can't read in the header!! In this case
3416 * we still guarantee we won't context swap, but the ref count won't be
3417 * incremented (otherwise we'd violate the invariant).
3419 /* NOTE: with the demand attach fileserver extensions, the global lock
3420 * is dropped within VHold */
3421 #ifdef AFS_DEMAND_ATTACH_FS
3423 VHold_r(register Volume * vp)
3427 VCreateReservation_r(vp);
3428 VWaitExclusiveState_r(vp);
3430 LoadVolumeHeader(&error, vp);
3432 VCancelReservation_r(vp);
3436 VCancelReservation_r(vp);
3439 #else /* AFS_DEMAND_ATTACH_FS */
3441 VHold_r(register Volume * vp)
3445 LoadVolumeHeader(&error, vp);
3451 #endif /* AFS_DEMAND_ATTACH_FS */
3455 VHold(register Volume * vp)
3459 retVal = VHold_r(vp);
3466 /***************************************************/
3467 /* get and put volume routines */
3468 /***************************************************/
3471 * put back a heavyweight reference to a volume object.
3473 * @param[in] vp volume object pointer
3475 * @pre VOL_LOCK held
3477 * @post heavyweight volume reference put back.
3478 * depending on state, volume may have been taken offline,
3479 * detached, salvaged, freed, etc.
3481 * @internal volume package internal use only
3484 VPutVolume_r(register Volume * vp)
3486 assert(--vp->nUsers >= 0);
3487 if (vp->nUsers == 0) {
3489 ReleaseVolumeHeader(vp->header);
3490 #ifdef AFS_DEMAND_ATTACH_FS
3491 if (!VCheckDetach(vp)) {
3495 #else /* AFS_DEMAND_ATTACH_FS */
3497 #endif /* AFS_DEMAND_ATTACH_FS */
3502 VPutVolume(register Volume * vp)
3510 /* Get a pointer to an attached volume. The pointer is returned regardless
3511 of whether or not the volume is in service or on/off line. An error
3512 code, however, is returned with an indication of the volume's status */
3514 VGetVolume(Error * ec, Error * client_ec, VolId volumeId)
3518 retVal = GetVolume(ec, client_ec, volumeId, NULL, 0);
3523 /* same as VGetVolume, but if a volume is waiting to go offline, we return
3524 * that it is actually offline, instead of waiting for it to go offline */
3526 VGetVolumeNoWait(Error * ec, Error * client_ec, VolId volumeId)
3530 retVal = GetVolume(ec, client_ec, volumeId, NULL, 1);
3536 VGetVolume_r(Error * ec, VolId volumeId)
3538 return GetVolume(ec, NULL, volumeId, NULL, 0);
3541 /* try to get a volume we've previously looked up */
3542 /* for demand attach fs, caller MUST NOT hold a ref count on vp */
3544 VGetVolumeByVp_r(Error * ec, Volume * vp)
3546 return GetVolume(ec, NULL, vp->hashid, vp, 0);
3550 * private interface for getting a volume handle
3552 * @param[out] ec error code (0 if no error)
3553 * @param[out] client_ec wire error code to be given to clients
3554 * @param[in] volumeId ID of the volume we want
3555 * @param[in] hint optional hint for hash lookups, or NULL
3556 * @param[in] nowait 0 to wait for a 'goingOffline' volume to go offline
3557 * before returning, 1 to return immediately
3559 * @return a volume handle for the specified volume
3560 * @retval NULL an error occurred, or the volume is in such a state that
3561 * we cannot load a header or return any volume struct
3563 * @note for DAFS, caller must NOT hold a ref count on 'hint'
3566 GetVolume(Error * ec, Error * client_ec, VolId volumeId, Volume * hint, int nowait)
3569 /* pull this profiling/debugging code out of regular builds */
3571 #define VGET_CTR_INC(x) x++
3572 unsigned short V0 = 0, V1 = 0, V2 = 0, V3 = 0, V5 = 0, V6 =
3573 0, V7 = 0, V8 = 0, V9 = 0;
3574 unsigned short V10 = 0, V11 = 0, V12 = 0, V13 = 0, V14 = 0, V15 = 0;
3576 #define VGET_CTR_INC(x)
3578 #ifdef AFS_DEMAND_ATTACH_FS
3579 Volume *avp, * rvp = hint;
3583 * if VInit is zero, the volume package dynamic
3584 * data structures have not been initialized yet,
3585 * and we must immediately return an error
3591 *client_ec = VOFFLINE;
3596 #ifdef AFS_DEMAND_ATTACH_FS
3598 VCreateReservation_r(rvp);
3600 #endif /* AFS_DEMAND_ATTACH_FS */
3608 vp = VLookupVolume_r(ec, volumeId, vp);
3614 #ifdef AFS_DEMAND_ATTACH_FS
3615 if (rvp && (rvp != vp)) {
3616 /* break reservation on old vp */
3617 VCancelReservation_r(rvp);
3620 #endif /* AFS_DEMAND_ATTACH_FS */
3626 /* Until we have reached an initialization level of 2
3627 * we don't know whether this volume exists or not.
3628 * We can't sleep and retry later because before a volume
3629 * is attached, the caller tries to get it first. Just
3630 * return VOFFLINE and the caller can choose whether to
3631 * retry the command or not. */
3641 IncUInt64(&VStats.hdr_gets);
3643 #ifdef AFS_DEMAND_ATTACH_FS
3644 /* block if someone else is performing an exclusive op on this volume */
3647 VCreateReservation_r(rvp);
3649 VWaitExclusiveState_r(vp);
3651 /* short circuit with VNOVOL in the following circumstances:
3654 * - VOL_STATE_SHUTTING_DOWN
3656 if ((V_attachState(vp) == VOL_STATE_ERROR) ||
3657 (V_attachState(vp) == VOL_STATE_SHUTTING_DOWN) ||
3658 (V_attachState(vp) == VOL_STATE_GOING_OFFLINE)) {
3665 * short circuit with VOFFLINE in the following circumstances:
3667 * - VOL_STATE_UNATTACHED
3669 if (V_attachState(vp) == VOL_STATE_UNATTACHED) {
3670 if (vp->specialStatus) {
3671 *ec = vp->specialStatus;
3679 /* allowable states:
3685 if (vp->salvage.requested) {
3686 VUpdateSalvagePriority_r(vp);
3689 if (V_attachState(vp) == VOL_STATE_PREATTACHED) {
3690 avp = VAttachVolumeByVp_r(ec, vp, 0);
3693 /* VAttachVolumeByVp_r can return a pointer
3694 * != the vp passed to it under certain
3695 * conditions; make sure we don't leak
3696 * reservations if that happens */
3698 VCancelReservation_r(rvp);
3700 VCreateReservation_r(rvp);
3710 if (!vp->pending_vol_op) {
3725 if ((V_attachState(vp) == VOL_STATE_SALVAGING) ||
3726 (*ec == VSALVAGING)) {
3728 /* see CheckVnode() in afsfileprocs.c for an explanation
3729 * of this error code logic */
3730 afs_uint32 now = FT_ApproxTime();
3731 if ((vp->stats.last_salvage + (10 * 60)) >= now) {
3734 *client_ec = VRESTARTING;
3743 #ifdef AFS_DEMAND_ATTACH_FS
3745 * this test MUST happen after VAttachVolymeByVp, so vol_op_state is
3746 * not VolOpRunningUnknown (attach2 would have converted it to Online
3750 /* only valid before/during demand attachment */
3751 assert(!vp->pending_vol_op || vp->pending_vol_op->vol_op_state != FSSYNC_VolOpRunningUnknown);
3753 /* deny getvolume due to running mutually exclusive vol op */
3754 if (vp->pending_vol_op && vp->pending_vol_op->vol_op_state==FSSYNC_VolOpRunningOffline) {
3756 * volume cannot remain online during this volume operation.
3759 if (vp->specialStatus) {
3761 * special status codes outrank normal VOFFLINE code
3763 *ec = vp->specialStatus;
3765 *client_ec = vp->specialStatus;
3769 /* see CheckVnode() in afsfileprocs.c for an explanation
3770 * of this error code logic */
3771 afs_uint32 now = FT_ApproxTime();
3772 if ((vp->stats.last_vol_op + (10 * 60)) >= now) {
3775 *client_ec = VRESTARTING;
3780 VChangeState_r(vp, VOL_STATE_UNATTACHED);
3781 FreeVolumeHeader(vp);
3785 #endif /* AFS_DEMAND_ATTACH_FS */
3787 LoadVolumeHeader(ec, vp);
3790 /* Only log the error if it was a totally unexpected error. Simply
3791 * a missing inode is likely to be caused by the volume being deleted */
3792 if (errno != ENXIO || LogLevel)
3793 Log("Volume %u: couldn't reread volume header\n",
3795 #ifdef AFS_DEMAND_ATTACH_FS
3796 if (VCanScheduleSalvage()) {
3797 VRequestSalvage_r(ec, vp, SALVSYNC_ERROR, VOL_SALVAGE_INVALIDATE_HEADER);
3802 #else /* AFS_DEMAND_ATTACH_FS */
3805 #endif /* AFS_DEMAND_ATTACH_FS */
3810 if (vp->shuttingDown) {
3817 if (programType == fileServer) {
3819 if (vp->goingOffline && !nowait) {
3821 #ifdef AFS_DEMAND_ATTACH_FS
3822 /* wait for the volume to go offline */
3823 if (V_attachState(vp) == VOL_STATE_GOING_OFFLINE) {
3824 VWaitStateChange_r(vp);
3826 #elif defined(AFS_PTHREAD_ENV)
3827 VOL_CV_WAIT(&vol_put_volume_cond);
3828 #else /* AFS_PTHREAD_ENV */
3829 LWP_WaitProcess(VPutVolume);
3830 #endif /* AFS_PTHREAD_ENV */
3833 if (vp->specialStatus) {
3835 *ec = vp->specialStatus;
3836 } else if (V_inService(vp) == 0 || V_blessed(vp) == 0) {
3839 } else if (V_inUse(vp) == 0 || vp->goingOffline) {
3850 #ifdef AFS_DEMAND_ATTACH_FS
3851 /* if no error, bump nUsers */
3854 VLRU_UpdateAccess_r(vp);
3857 VCancelReservation_r(rvp);
3860 if (client_ec && !*client_ec) {
3863 #else /* AFS_DEMAND_ATTACH_FS */
3864 /* if no error, bump nUsers */
3871 #endif /* AFS_DEMAND_ATTACH_FS */
3879 /***************************************************/
3880 /* Volume offline/detach routines */
3881 /***************************************************/
3883 /* caller MUST hold a heavyweight ref on vp */
3884 #ifdef AFS_DEMAND_ATTACH_FS
3886 VTakeOffline_r(register Volume * vp)
3890 assert(vp->nUsers > 0);
3891 assert(programType == fileServer);
3893 VCreateReservation_r(vp);
3894 VWaitExclusiveState_r(vp);
3896 vp->goingOffline = 1;
3897 V_needsSalvaged(vp) = 1;
3899 VRequestSalvage_r(&error, vp, SALVSYNC_ERROR, 0);
3900 VCancelReservation_r(vp);
3902 #else /* AFS_DEMAND_ATTACH_FS */
3904 VTakeOffline_r(register Volume * vp)
3906 assert(vp->nUsers > 0);
3907 assert(programType == fileServer);
3909 vp->goingOffline = 1;
3910 V_needsSalvaged(vp) = 1;
3912 #endif /* AFS_DEMAND_ATTACH_FS */
3915 VTakeOffline(register Volume * vp)
3923 * force a volume offline.
3925 * @param[in] vp volume object pointer
3926 * @param[in] flags flags (see note below)
3928 * @note the flag VOL_FORCEOFF_NOUPDATE is a recursion control flag
3929 * used when VUpdateVolume_r needs to call VForceOffline_r
3930 * (which in turn would normally call VUpdateVolume_r)
3932 * @see VUpdateVolume_r
3934 * @pre VOL_LOCK must be held.
3935 * for DAFS, caller must hold ref.
3937 * @note for DAFS, it _is safe_ to call this function from an
3940 * @post needsSalvaged flag is set.
3941 * for DAFS, salvage is requested.
3942 * no further references to the volume through the volume
3943 * package will be honored.
3944 * all file descriptor and vnode caches are invalidated.
3946 * @warning this is a heavy-handed interface. it results in
3947 * a volume going offline regardless of the current
3948 * reference count state.
3950 * @internal volume package internal use only
3953 VForceOffline_r(Volume * vp, int flags)
3957 #ifdef AFS_DEMAND_ATTACH_FS
3958 VChangeState_r(vp, VOL_STATE_ERROR);
3963 strcpy(V_offlineMessage(vp),
3964 "Forced offline due to internal error: volume needs to be salvaged");
3965 Log("Volume %u forced offline: it needs salvaging!\n", V_id(vp));
3968 vp->goingOffline = 0;
3969 V_needsSalvaged(vp) = 1;
3970 if (!(flags & VOL_FORCEOFF_NOUPDATE)) {
3971 VUpdateVolume_r(&error, vp, VOL_UPDATE_NOFORCEOFF);
3974 #ifdef AFS_DEMAND_ATTACH_FS
3975 VRequestSalvage_r(&error, vp, SALVSYNC_ERROR, VOL_SALVAGE_INVALIDATE_HEADER);
3976 #endif /* AFS_DEMAND_ATTACH_FS */
3978 #ifdef AFS_PTHREAD_ENV
3979 assert(pthread_cond_broadcast(&vol_put_volume_cond) == 0);
3980 #else /* AFS_PTHREAD_ENV */
3981 LWP_NoYieldSignal(VPutVolume);
3982 #endif /* AFS_PTHREAD_ENV */
3984 VReleaseVolumeHandles_r(vp);
3988 * force a volume offline.
3990 * @param[in] vp volume object pointer
3992 * @see VForceOffline_r
3995 VForceOffline(Volume * vp)
3998 VForceOffline_r(vp, 0);
4002 /* The opposite of VAttachVolume. The volume header is written to disk, with
4003 the inUse bit turned off. A copy of the header is maintained in memory,
4004 however (which is why this is VOffline, not VDetach).
4007 VOffline_r(Volume * vp, char *message)
4009 #ifndef AFS_DEMAND_ATTACH_FS
4011 VolumeId vid = V_id(vp);
4014 assert(programType != volumeUtility && programType != volumeServer);
4019 if (V_offlineMessage(vp)[0] == '\0')
4020 strncpy(V_offlineMessage(vp), message, sizeof(V_offlineMessage(vp)));
4021 V_offlineMessage(vp)[sizeof(V_offlineMessage(vp)) - 1] = '\0';
4023 vp->goingOffline = 1;
4024 #ifdef AFS_DEMAND_ATTACH_FS
4025 VChangeState_r(vp, VOL_STATE_GOING_OFFLINE);
4026 VCreateReservation_r(vp);
4029 /* wait for the volume to go offline */
4030 if (V_attachState(vp) == VOL_STATE_GOING_OFFLINE) {
4031 VWaitStateChange_r(vp);
4033 VCancelReservation_r(vp);
4034 #else /* AFS_DEMAND_ATTACH_FS */
4036 vp = VGetVolume_r(&error, vid); /* Wait for it to go offline */
4037 if (vp) /* In case it was reattached... */
4039 #endif /* AFS_DEMAND_ATTACH_FS */
4042 #ifdef AFS_DEMAND_ATTACH_FS
4044 * Take a volume offline in order to perform a volume operation.
4046 * @param[inout] ec address in which to store error code
4047 * @param[in] vp volume object pointer
4048 * @param[in] message volume offline status message
4051 * - VOL_LOCK is held
4052 * - caller MUST hold a heavyweight ref on vp
4055 * - volume is taken offline
4056 * - if possible, volume operation is promoted to running state
4057 * - on failure, *ec is set to nonzero
4059 * @note Although this function does not return any value, it may
4060 * still fail to promote our pending volume operation to
4061 * a running state. Any caller MUST check the value of *ec,
4062 * and MUST NOT blindly assume success.
4064 * @warning if the caller does not hold a lightweight ref on vp,
4065 * then it MUST NOT reference vp after this function
4066 * returns to the caller.
4068 * @internal volume package internal use only
4071 VOfflineForVolOp_r(Error *ec, Volume *vp, char *message)
4073 assert(vp->pending_vol_op);
4079 if (V_offlineMessage(vp)[0] == '\0')
4080 strncpy(V_offlineMessage(vp), message, sizeof(V_offlineMessage(vp)));
4081 V_offlineMessage(vp)[sizeof(V_offlineMessage(vp)) - 1] = '\0';
4083 vp->goingOffline = 1;
4084 VChangeState_r(vp, VOL_STATE_GOING_OFFLINE);
4085 VCreateReservation_r(vp);
4088 /* Wait for the volume to go offline */
4089 while (!VIsOfflineState(V_attachState(vp))) {
4090 /* do not give corrupted volumes to the volserver */
4091 if (vp->salvage.requested && vp->pending_vol_op->com.programType != salvageServer) {
4095 VWaitStateChange_r(vp);
4099 VCancelReservation_r(vp);
4101 #endif /* AFS_DEMAND_ATTACH_FS */
4104 VOffline(Volume * vp, char *message)
4107 VOffline_r(vp, message);
4111 /* This gets used for the most part by utility routines that don't want
4112 * to keep all the volume headers around. Generally, the file server won't
4113 * call this routine, because then the offline message in the volume header
4114 * (or other information) won't be available to clients. For NAMEI, also
4115 * close the file handles. However, the fileserver does call this during
4116 * an attach following a volume operation.
4119 VDetachVolume_r(Error * ec, Volume * vp)
4122 struct DiskPartition64 *tpartp;
4123 int notifyServer = 0;
4124 int useDone = FSYNC_VOL_ON;
4126 *ec = 0; /* always "succeeds" */
4127 if (VCanUseFSSYNC()) {
4128 notifyServer = vp->needsPutBack;
4129 if (V_destroyMe(vp) == DESTROY_ME)
4130 useDone = FSYNC_VOL_DONE;
4131 #ifdef AFS_DEMAND_ATTACH_FS
4132 else if (!V_blessed(vp) || !V_inService(vp))
4133 useDone = FSYNC_VOL_LEAVE_OFF;
4136 tpartp = vp->partition;
4138 DeleteVolumeFromHashTable(vp);
4139 vp->shuttingDown = 1;
4140 #ifdef AFS_DEMAND_ATTACH_FS
4141 DeleteVolumeFromVByPList_r(vp);
4143 VChangeState_r(vp, VOL_STATE_SHUTTING_DOWN);
4145 if (programType != fileServer)
4147 #endif /* AFS_DEMAND_ATTACH_FS */
4149 /* Will be detached sometime in the future--this is OK since volume is offline */
4151 /* XXX the following code should really be moved to VCheckDetach() since the volume
4152 * is not technically detached until the refcounts reach zero
4154 #ifdef FSSYNC_BUILD_CLIENT
4155 if (VCanUseFSSYNC() && notifyServer) {
4157 * Note: The server is not notified in the case of a bogus volume
4158 * explicitly to make it possible to create a volume, do a partial
4159 * restore, then abort the operation without ever putting the volume
4160 * online. This is essential in the case of a volume move operation
4161 * between two partitions on the same server. In that case, there
4162 * would be two instances of the same volume, one of them bogus,
4163 * which the file server would attempt to put on line
4165 FSYNC_VolOp(volume, tpartp->name, useDone, 0, NULL);
4166 /* XXX this code path is only hit by volume utilities, thus
4167 * V_BreakVolumeCallbacks will always be NULL. if we really
4168 * want to break callbacks in this path we need to use FSYNC_VolOp() */
4170 /* Dettaching it so break all callbacks on it */
4171 if (V_BreakVolumeCallbacks) {
4172 Log("volume %u detached; breaking all call backs\n", volume);
4173 (*V_BreakVolumeCallbacks) (volume);
4177 #endif /* FSSYNC_BUILD_CLIENT */
4181 VDetachVolume(Error * ec, Volume * vp)
4184 VDetachVolume_r(ec, vp);
4189 /***************************************************/
4190 /* Volume fd/inode handle closing routines */
4191 /***************************************************/
4193 /* For VDetachVolume, we close all cached file descriptors, but keep
4194 * the Inode handles in case we need to read from a busy volume.
4196 /* for demand attach, caller MUST hold ref count on vp */
4198 VCloseVolumeHandles_r(Volume * vp)
4200 #ifdef AFS_DEMAND_ATTACH_FS
4201 VolState state_save;
4203 state_save = VChangeState_r(vp, VOL_STATE_OFFLINING);
4208 * XXX need to investigate whether we can perform
4209 * DFlushVolume outside of vol_glock_mutex...
4211 * VCloseVnodeFiles_r drops the glock internally */
4212 DFlushVolume(vp->hashid);
4213 VCloseVnodeFiles_r(vp);
4215 #ifdef AFS_DEMAND_ATTACH_FS
4219 /* Too time consuming and unnecessary for the volserver */
4220 if (programType == fileServer) {
4221 IH_CONDSYNC(vp->vnodeIndex[vLarge].handle);
4222 IH_CONDSYNC(vp->vnodeIndex[vSmall].handle);
4223 IH_CONDSYNC(vp->diskDataHandle);
4225 IH_CONDSYNC(vp->linkHandle);
4226 #endif /* AFS_NT40_ENV */
4229 IH_REALLYCLOSE(vp->vnodeIndex[vLarge].handle);
4230 IH_REALLYCLOSE(vp->vnodeIndex[vSmall].handle);
4231 IH_REALLYCLOSE(vp->diskDataHandle);
4232 IH_REALLYCLOSE(vp->linkHandle);
4234 #ifdef AFS_DEMAND_ATTACH_FS
4235 if ((V_attachFlags(vp) & VOL_LOCKED)) {
4240 VChangeState_r(vp, state_save);
4244 /* For both VForceOffline and VOffline, we close all relevant handles.
4245 * For VOffline, if we re-attach the volume, the files may possible be
4246 * different than before.
4248 /* for demand attach, caller MUST hold a ref count on vp */
4250 VReleaseVolumeHandles_r(Volume * vp)
4252 #ifdef AFS_DEMAND_ATTACH_FS
4253 VolState state_save;
4255 state_save = VChangeState_r(vp, VOL_STATE_DETACHING);
4258 /* XXX need to investigate whether we can perform
4259 * DFlushVolume outside of vol_glock_mutex... */
4260 DFlushVolume(vp->hashid);
4262 VReleaseVnodeFiles_r(vp); /* releases the glock internally */
4264 #ifdef AFS_DEMAND_ATTACH_FS
4268 /* Too time consuming and unnecessary for the volserver */
4269 if (programType == fileServer) {
4270 IH_CONDSYNC(vp->vnodeIndex[vLarge].handle);
4271 IH_CONDSYNC(vp->vnodeIndex[vSmall].handle);
4272 IH_CONDSYNC(vp->diskDataHandle);
4274 IH_CONDSYNC(vp->linkHandle);
4275 #endif /* AFS_NT40_ENV */
4278 IH_RELEASE(vp->vnodeIndex[vLarge].handle);
4279 IH_RELEASE(vp->vnodeIndex[vSmall].handle);
4280 IH_RELEASE(vp->diskDataHandle);
4281 IH_RELEASE(vp->linkHandle);
4283 #ifdef AFS_DEMAND_ATTACH_FS
4284 if ((V_attachFlags(vp) & VOL_LOCKED)) {
4289 VChangeState_r(vp, state_save);
4294 /***************************************************/
4295 /* Volume write and fsync routines */
4296 /***************************************************/
4299 VUpdateVolume_r(Error * ec, Volume * vp, int flags)
4301 #ifdef AFS_DEMAND_ATTACH_FS
4302 VolState state_save;
4304 if (flags & VOL_UPDATE_WAIT) {
4305 VCreateReservation_r(vp);
4306 VWaitExclusiveState_r(vp);
4311 if (programType == fileServer)
4313 (V_inUse(vp) ? V_nextVnodeUnique(vp) +
4314 200 : V_nextVnodeUnique(vp));
4316 #ifdef AFS_DEMAND_ATTACH_FS
4317 state_save = VChangeState_r(vp, VOL_STATE_UPDATING);
4321 WriteVolumeHeader_r(ec, vp);
4323 #ifdef AFS_DEMAND_ATTACH_FS
4325 VChangeState_r(vp, state_save);
4326 if (flags & VOL_UPDATE_WAIT) {
4327 VCancelReservation_r(vp);
4332 Log("VUpdateVolume: error updating volume header, volume %u (%s)\n",
4333 V_id(vp), V_name(vp));
4334 /* try to update on-disk header,
4335 * while preventing infinite recursion */
4336 if (!(flags & VOL_UPDATE_NOFORCEOFF)) {
4337 VForceOffline_r(vp, VOL_FORCEOFF_NOUPDATE);
4343 VUpdateVolume(Error * ec, Volume * vp)
4346 VUpdateVolume_r(ec, vp, VOL_UPDATE_WAIT);
4351 VSyncVolume_r(Error * ec, Volume * vp, int flags)
4355 #ifdef AFS_DEMAND_ATTACH_FS
4356 VolState state_save;
4359 if (flags & VOL_SYNC_WAIT) {
4360 VUpdateVolume_r(ec, vp, VOL_UPDATE_WAIT);
4362 VUpdateVolume_r(ec, vp, 0);
4365 #ifdef AFS_DEMAND_ATTACH_FS
4366 state_save = VChangeState_r(vp, VOL_STATE_UPDATING);
4369 fdP = IH_OPEN(V_diskDataHandle(vp));
4370 assert(fdP != NULL);
4371 code = FDH_SYNC(fdP);
4374 #ifdef AFS_DEMAND_ATTACH_FS
4376 VChangeState_r(vp, state_save);
4382 VSyncVolume(Error * ec, Volume * vp)
4385 VSyncVolume_r(ec, vp, VOL_SYNC_WAIT);
4390 /***************************************************/
4391 /* Volume dealloaction routines */
4392 /***************************************************/
4394 #ifdef AFS_DEMAND_ATTACH_FS
4396 FreeVolume(Volume * vp)
4398 /* free the heap space, iff it's safe.
4399 * otherwise, pull it out of the hash table, so it
4400 * will get deallocated when all refs to it go away */
4401 if (!VCheckFree(vp)) {
4402 DeleteVolumeFromHashTable(vp);
4403 DeleteVolumeFromVByPList_r(vp);
4405 /* make sure we invalidate the header cache entry */
4406 FreeVolumeHeader(vp);
4409 #endif /* AFS_DEMAND_ATTACH_FS */
4412 ReallyFreeVolume(Volume * vp)
4417 #ifdef AFS_DEMAND_ATTACH_FS
4419 VChangeState_r(vp, VOL_STATE_FREED);
4420 if (vp->pending_vol_op)
4421 free(vp->pending_vol_op);
4422 #endif /* AFS_DEMAND_ATTACH_FS */
4423 for (i = 0; i < nVNODECLASSES; i++)
4424 if (vp->vnodeIndex[i].bitmap)
4425 free(vp->vnodeIndex[i].bitmap);
4426 FreeVolumeHeader(vp);
4427 #ifndef AFS_DEMAND_ATTACH_FS
4428 DeleteVolumeFromHashTable(vp);
4429 #endif /* AFS_DEMAND_ATTACH_FS */
4433 /* check to see if we should shutdown this volume
4434 * returns 1 if volume was freed, 0 otherwise */
4435 #ifdef AFS_DEMAND_ATTACH_FS
4437 VCheckDetach(register Volume * vp)
4442 if (vp->nUsers || vp->nWaiters)
4445 if (vp->shuttingDown) {
4447 if ((programType != fileServer) &&
4448 (V_inUse(vp) == programType) &&
4449 ((V_checkoutMode(vp) == V_VOLUPD) ||
4450 (V_checkoutMode(vp) == V_SECRETLY) ||
4451 ((V_checkoutMode(vp) == V_CLONE) &&
4452 (VolumeWriteable(vp))))) {
4454 VUpdateVolume_r(&ec, vp, VOL_UPDATE_NOFORCEOFF);
4456 Log("VCheckDetach: volume header update for volume %u "
4457 "failed with errno %d\n", vp->hashid, errno);
4460 VReleaseVolumeHandles_r(vp);
4462 ReallyFreeVolume(vp);
4463 if (programType == fileServer) {
4464 assert(pthread_cond_broadcast(&vol_put_volume_cond) == 0);
4469 #else /* AFS_DEMAND_ATTACH_FS */
4471 VCheckDetach(register Volume * vp)
4479 if (vp->shuttingDown) {
4481 if ((programType != fileServer) &&
4482 (V_inUse(vp) == programType) &&
4483 ((V_checkoutMode(vp) == V_VOLUPD) ||
4484 (V_checkoutMode(vp) == V_SECRETLY) ||
4485 ((V_checkoutMode(vp) == V_CLONE) &&
4486 (VolumeWriteable(vp))))) {
4488 VUpdateVolume_r(&ec, vp, VOL_UPDATE_NOFORCEOFF);
4490 Log("VCheckDetach: volume header update for volume %u failed with errno %d\n",
4494 VReleaseVolumeHandles_r(vp);
4495 ReallyFreeVolume(vp);
4496 if (programType == fileServer) {
4497 #if defined(AFS_PTHREAD_ENV)
4498 assert(pthread_cond_broadcast(&vol_put_volume_cond) == 0);
4499 #else /* AFS_PTHREAD_ENV */
4500 LWP_NoYieldSignal(VPutVolume);
4501 #endif /* AFS_PTHREAD_ENV */
4506 #endif /* AFS_DEMAND_ATTACH_FS */
4508 /* check to see if we should offline this volume
4509 * return 1 if volume went offline, 0 otherwise */
4510 #ifdef AFS_DEMAND_ATTACH_FS
4512 VCheckOffline(register Volume * vp)
4516 if (vp->goingOffline && !vp->nUsers) {
4518 assert(programType == fileServer);
4519 assert((V_attachState(vp) != VOL_STATE_ATTACHED) &&
4520 (V_attachState(vp) != VOL_STATE_FREED) &&
4521 (V_attachState(vp) != VOL_STATE_PREATTACHED) &&
4522 (V_attachState(vp) != VOL_STATE_UNATTACHED));
4526 * VOL_STATE_GOING_OFFLINE
4527 * VOL_STATE_SHUTTING_DOWN
4528 * VIsErrorState(V_attachState(vp))
4529 * VIsExclusiveState(V_attachState(vp))
4532 VCreateReservation_r(vp);
4533 VChangeState_r(vp, VOL_STATE_OFFLINING);
4536 /* must clear the goingOffline flag before we drop the glock */
4537 vp->goingOffline = 0;
4542 /* perform async operations */
4543 VUpdateVolume_r(&error, vp, 0);
4544 VCloseVolumeHandles_r(vp);
4547 if (V_offlineMessage(vp)[0]) {
4548 Log("VOffline: Volume %lu (%s) is now offline (%s)\n",
4549 afs_printable_uint32_lu(V_id(vp)), V_name(vp),
4550 V_offlineMessage(vp));
4552 Log("VOffline: Volume %lu (%s) is now offline\n",
4553 afs_printable_uint32_lu(V_id(vp)), V_name(vp));
4557 /* invalidate the volume header cache entry */
4558 FreeVolumeHeader(vp);
4560 /* if nothing changed state to error or salvaging,
4561 * drop state to unattached */
4562 if (!VIsErrorState(V_attachState(vp))) {
4563 VChangeState_r(vp, VOL_STATE_UNATTACHED);
4565 VCancelReservation_r(vp);
4566 /* no usage of vp is safe beyond this point */
4570 #else /* AFS_DEMAND_ATTACH_FS */
4572 VCheckOffline(register Volume * vp)
4576 if (vp->goingOffline && !vp->nUsers) {
4578 assert(programType == fileServer);
4581 vp->goingOffline = 0;
4583 VUpdateVolume_r(&error, vp, 0);
4584 VCloseVolumeHandles_r(vp);
4586 Log("VOffline: Volume %u (%s) is now offline", V_id(vp),
4588 if (V_offlineMessage(vp)[0])
4589 Log(" (%s)", V_offlineMessage(vp));
4592 FreeVolumeHeader(vp);
4593 #ifdef AFS_PTHREAD_ENV
4594 assert(pthread_cond_broadcast(&vol_put_volume_cond) == 0);
4595 #else /* AFS_PTHREAD_ENV */
4596 LWP_NoYieldSignal(VPutVolume);
4597 #endif /* AFS_PTHREAD_ENV */
4601 #endif /* AFS_DEMAND_ATTACH_FS */
4603 /***************************************************/
4604 /* demand attach fs ref counting routines */
4605 /***************************************************/
4607 #ifdef AFS_DEMAND_ATTACH_FS
4608 /* the following two functions handle reference counting for
4609 * asynchronous operations on volume structs.
4611 * their purpose is to prevent a VDetachVolume or VShutdown
4612 * from free()ing the Volume struct during an async i/o op */
4614 /* register with the async volume op ref counter */
4615 /* VCreateReservation_r moved into inline code header because it
4616 * is now needed in vnode.c -- tkeiser 11/20/2007
4620 * decrement volume-package internal refcount.
4622 * @param vp volume object pointer
4624 * @internal volume package internal use only
4627 * @arg VOL_LOCK is held
4628 * @arg lightweight refcount held
4630 * @post volume waiters refcount is decremented; volume may
4631 * have been deallocated/shutdown/offlined/salvaged/
4632 * whatever during the process
4634 * @warning once you have tossed your last reference (you can acquire
4635 * lightweight refs recursively) it is NOT SAFE to reference
4636 * a volume object pointer ever again
4638 * @see VCreateReservation_r
4640 * @note DEMAND_ATTACH_FS only
4643 VCancelReservation_r(Volume * vp)
4645 assert(--vp->nWaiters >= 0);
4646 if (vp->nWaiters == 0) {
4648 if (!VCheckDetach(vp)) {
4655 /* check to see if we should free this volume now
4656 * return 1 if volume was freed, 0 otherwise */
4658 VCheckFree(Volume * vp)
4661 if ((vp->nUsers == 0) &&
4662 (vp->nWaiters == 0) &&
4663 !(V_attachFlags(vp) & (VOL_IN_HASH |
4667 ReallyFreeVolume(vp);
4672 #endif /* AFS_DEMAND_ATTACH_FS */
4675 /***************************************************/
4676 /* online volume operations routines */
4677 /***************************************************/
4679 #ifdef AFS_DEMAND_ATTACH_FS
4681 * register a volume operation on a given volume.
4683 * @param[in] vp volume object
4684 * @param[in] vopinfo volume operation info object
4686 * @pre VOL_LOCK is held
4688 * @post volume operation info object attached to volume object.
4689 * volume operation statistics updated.
4691 * @note by "attached" we mean a copy of the passed in object is made
4693 * @internal volume package internal use only
4696 VRegisterVolOp_r(Volume * vp, FSSYNC_VolOp_info * vopinfo)
4698 FSSYNC_VolOp_info * info;
4700 /* attach a vol op info node to the volume struct */
4701 info = (FSSYNC_VolOp_info *) malloc(sizeof(FSSYNC_VolOp_info));
4702 assert(info != NULL);
4703 memcpy(info, vopinfo, sizeof(FSSYNC_VolOp_info));
4704 vp->pending_vol_op = info;
4707 vp->stats.last_vol_op = FT_ApproxTime();
4708 vp->stats.vol_ops++;
4709 IncUInt64(&VStats.vol_ops);
4715 * deregister the volume operation attached to this volume.
4717 * @param[in] vp volume object pointer
4719 * @pre VOL_LOCK is held
4721 * @post the volume operation info object is detached from the volume object
4723 * @internal volume package internal use only
4726 VDeregisterVolOp_r(Volume * vp)
4728 if (vp->pending_vol_op) {
4729 free(vp->pending_vol_op);
4730 vp->pending_vol_op = NULL;
4734 #endif /* AFS_DEMAND_ATTACH_FS */
4737 * determine whether it is safe to leave a volume online during
4738 * the volume operation described by the vopinfo object.
4740 * @param[in] vp volume object
4741 * @param[in] vopinfo volume operation info object
4743 * @return whether it is safe to leave volume online
4744 * @retval 0 it is NOT SAFE to leave the volume online
4745 * @retval 1 it is safe to leave the volume online during the operation
4748 * @arg VOL_LOCK is held
4749 * @arg disk header attached to vp (heavyweight ref on vp will guarantee
4750 * this condition is met)
4752 * @internal volume package internal use only
4755 VVolOpLeaveOnline_r(Volume * vp, FSSYNC_VolOp_info * vopinfo)
4757 return (vopinfo->vol_op_state == FSSYNC_VolOpRunningOnline ||
4758 (vopinfo->com.command == FSYNC_VOL_NEEDVOLUME &&
4759 (vopinfo->com.reason == V_READONLY ||
4760 (!VolumeWriteable(vp) &&
4761 (vopinfo->com.reason == V_CLONE ||
4762 vopinfo->com.reason == V_DUMP)))));
4766 * same as VVolOpLeaveOnline_r, but does not require a volume with an attached
4769 * @param[in] vp volume object
4770 * @param[in] vopinfo volume operation info object
4772 * @return whether it is safe to leave volume online
4773 * @retval 0 it is NOT SAFE to leave the volume online
4774 * @retval 1 it is safe to leave the volume online during the operation
4775 * @retval -1 unsure; volume header is required in order to know whether or
4776 * not is is safe to leave the volume online
4778 * @pre VOL_LOCK is held
4780 * @internal volume package internal use only
4783 VVolOpLeaveOnlineNoHeader_r(Volume * vp, FSSYNC_VolOp_info * vopinfo)
4785 /* follow the logic in VVolOpLeaveOnline_r; this is the same, except
4786 * assume that we don't know VolumeWriteable; return -1 if the answer
4787 * depends on VolumeWriteable */
4789 if (vopinfo->vol_op_state == FSSYNC_VolOpRunningOnline) {
4792 if (vopinfo->com.command == FSYNC_VOL_NEEDVOLUME &&
4793 vopinfo->com.reason == V_READONLY) {
4797 if (vopinfo->com.command == FSYNC_VOL_NEEDVOLUME &&
4798 (vopinfo->com.reason == V_CLONE ||
4799 vopinfo->com.reason == V_DUMP)) {
4801 /* must know VolumeWriteable */
4808 * determine whether VBUSY should be set during this volume operation.
4810 * @param[in] vp volume object
4811 * @param[in] vopinfo volume operation info object
4813 * @return whether VBUSY should be set
4814 * @retval 0 VBUSY does NOT need to be set
4815 * @retval 1 VBUSY SHOULD be set
4817 * @pre VOL_LOCK is held
4819 * @internal volume package internal use only
4822 VVolOpSetVBusy_r(Volume * vp, FSSYNC_VolOp_info * vopinfo)
4824 return ((vopinfo->com.command == FSYNC_VOL_OFF &&
4825 vopinfo->com.reason == FSYNC_SALVAGE) ||
4826 (vopinfo->com.command == FSYNC_VOL_NEEDVOLUME &&
4827 (vopinfo->com.reason == V_CLONE ||
4828 vopinfo->com.reason == V_DUMP)));
4832 /***************************************************/
4833 /* online salvager routines */
4834 /***************************************************/
4835 #if defined(AFS_DEMAND_ATTACH_FS)
4837 * check whether a salvage needs to be performed on this volume.
4839 * @param[in] vp pointer to volume object
4841 * @return status code
4842 * @retval 0 no salvage scheduled
4843 * @retval 1 a salvage has been scheduled with the salvageserver
4845 * @pre VOL_LOCK is held
4847 * @post if salvage request flag is set and nUsers and nWaiters are zero,
4848 * then a salvage will be requested
4850 * @note this is one of the event handlers called by VCancelReservation_r
4852 * @see VCancelReservation_r
4854 * @internal volume package internal use only.
4857 VCheckSalvage(register Volume * vp)
4860 #if defined(SALVSYNC_BUILD_CLIENT) || defined(FSSYNC_BUILD_CLIENT)
4861 if (vp->nUsers || vp->nWaiters)
4863 if (vp->salvage.requested) {
4864 VScheduleSalvage_r(vp);
4867 #endif /* SALVSYNC_BUILD_CLIENT || FSSYNC_BUILD_CLIENT */
4872 * request volume salvage.
4874 * @param[out] ec computed client error code
4875 * @param[in] vp volume object pointer
4876 * @param[in] reason reason code (passed to salvageserver via SALVSYNC)
4877 * @param[in] flags see flags note below
4880 * VOL_SALVAGE_INVALIDATE_HEADER causes volume header cache entry
4881 * to be invalidated.
4883 * @pre VOL_LOCK is held.
4885 * @post volume state is changed.
4886 * for fileserver, salvage will be requested once refcount reaches zero.
4888 * @return operation status code
4889 * @retval 0 volume salvage will occur
4890 * @retval 1 volume salvage could not be scheduled
4894 * @note in the fileserver, this call does not synchronously schedule a volume
4895 * salvage. rather, it sets volume state so that when volume refcounts
4896 * reach zero, a volume salvage will occur. by "refcounts", we mean both
4897 * nUsers and nWaiters must be zero.
4899 * @internal volume package internal use only.
4902 VRequestSalvage_r(Error * ec, Volume * vp, int reason, int flags)
4906 * for DAFS volume utilities that are not supposed to schedule salvages,
4907 * just transition to error state instead
4909 if (!VCanScheduleSalvage()) {
4910 VChangeState_r(vp, VOL_STATE_ERROR);
4915 if (programType != fileServer && !VCanUseFSSYNC()) {
4916 VChangeState_r(vp, VOL_STATE_ERROR);
4921 if (!vp->salvage.requested) {
4922 vp->salvage.requested = 1;
4923 vp->salvage.reason = reason;
4924 vp->stats.last_salvage = FT_ApproxTime();
4926 /* Note that it is not possible for us to reach this point if a
4927 * salvage is already running on this volume (even if the fileserver
4928 * was restarted during the salvage). If a salvage were running, the
4929 * salvager would have write-locked the volume header file, so when
4930 * we tried to lock the volume header, the lock would have failed,
4931 * and we would have failed during attachment prior to calling
4932 * VRequestSalvage. So we know that we can schedule salvages without
4933 * fear of a salvage already running for this volume. */
4935 if (vp->stats.salvages < SALVAGE_COUNT_MAX) {
4936 VChangeState_r(vp, VOL_STATE_SALVAGING);
4939 Log("VRequestSalvage: volume %u online salvaged too many times; forced offline.\n", vp->hashid);
4941 /* make sure neither VScheduleSalvage_r nor
4942 * VUpdateSalvagePriority_r try to schedule another salvage */
4943 vp->salvage.requested = vp->salvage.scheduled = 0;
4945 VChangeState_r(vp, VOL_STATE_ERROR);
4949 if (flags & VOL_SALVAGE_INVALIDATE_HEADER) {
4950 /* Instead of ReleaseVolumeHeader, we do FreeVolumeHeader()
4951 so that the the next VAttachVolumeByVp_r() invocation
4952 of attach2() will pull in a cached header
4953 entry and fail, then load a fresh one from disk and attach
4956 FreeVolumeHeader(vp);
4963 * update salvageserver scheduling priority for a volume.
4965 * @param[in] vp pointer to volume object
4967 * @return operation status
4969 * @retval 1 request denied, or SALVSYNC communications failure
4971 * @pre VOL_LOCK is held.
4973 * @post in-core salvage priority counter is incremented. if at least
4974 * SALVAGE_PRIO_UPDATE_INTERVAL seconds have elapsed since the
4975 * last SALVSYNC_RAISEPRIO request, we contact the salvageserver
4976 * to update its priority queue. if no salvage is scheduled,
4977 * this function is a no-op.
4979 * @note DAFS fileserver only
4981 * @note this should be called whenever a VGetVolume fails due to a
4982 * pending salvage request
4984 * @todo should set exclusive state and drop glock around salvsync call
4986 * @internal volume package internal use only.
4989 VUpdateSalvagePriority_r(Volume * vp)
4993 #ifdef SALVSYNC_BUILD_CLIENT
4998 now = FT_ApproxTime();
5000 /* update the salvageserver priority queue occasionally so that
5001 * frequently requested volumes get moved to the head of the queue
5003 if ((vp->salvage.scheduled) &&
5004 (vp->stats.last_salvage_req < (now-SALVAGE_PRIO_UPDATE_INTERVAL))) {
5005 code = SALVSYNC_SalvageVolume(vp->hashid,
5006 VPartitionPath(vp->partition),
5011 vp->stats.last_salvage_req = now;
5012 if (code != SYNC_OK) {
5016 #endif /* SALVSYNC_BUILD_CLIENT */
5021 #if defined(SALVSYNC_BUILD_CLIENT) || defined(FSSYNC_BUILD_CLIENT)
5023 /* A couple of little helper functions. These return true if we tried to
5024 * use this mechanism to schedule a salvage, false if we haven't tried.
5025 * If we did try a salvage then the results are contained in code.
5029 try_SALVSYNC(Volume *vp, char *partName, int *code) {
5030 #ifdef SALVSYNC_BUILD_CLIENT
5031 if (VCanUseSALVSYNC()) {
5032 Log("Scheduling salvage for volume %lu on part %s over SALVSYNC\n",
5033 afs_printable_uint32_lu(vp->hashid), partName);
5035 /* can't use V_id() since there's no guarantee
5036 * we have the disk data header at this point */
5037 *code = SALVSYNC_SalvageVolume(vp->hashid,
5050 try_FSSYNC(Volume *vp, char *partName, int *code) {
5051 #ifdef FSSYNC_BUILD_CLIENT
5052 if (VCanUseFSSYNC()) {
5053 Log("Scheduling salvage for volume %lu on part %s over FSSYNC\n",
5054 afs_printable_uint32_lu(vp->hashid), partName);
5057 * If we aren't the fileserver, tell the fileserver the volume
5058 * needs to be salvaged. We could directly tell the
5059 * salvageserver, but the fileserver keeps track of some stats
5060 * related to salvages, and handles some other salvage-related
5061 * complications for us.
5063 *code = FSYNC_VolOp(vp->hashid, partName,
5064 FSYNC_VOL_FORCE_ERROR, FSYNC_SALVAGE, NULL);
5067 #endif /* FSSYNC_BUILD_CLIENT */
5072 * schedule a salvage with the salvage server or fileserver.
5074 * @param[in] vp pointer to volume object
5076 * @return operation status
5077 * @retval 0 salvage scheduled successfully
5078 * @retval 1 salvage not scheduled, or SALVSYNC/FSSYNC com error
5081 * @arg VOL_LOCK is held.
5082 * @arg nUsers and nWaiters should be zero.
5084 * @post salvageserver or fileserver is sent a salvage request
5086 * @note If we are the fileserver, the request will be sent to the salvage
5087 * server over SALVSYNC. If we are not the fileserver, the request will be
5088 * sent to the fileserver over FSSYNC (FSYNC_VOL_FORCE_ERROR/FSYNC_SALVAGE).
5092 * @internal volume package internal use only.
5095 VScheduleSalvage_r(Volume * vp)
5099 VolState state_save;
5100 VThreadOptions_t * thread_opts;
5103 assert(VCanUseSALVSYNC() || VCanUseFSSYNC());
5105 if (vp->nWaiters || vp->nUsers) {
5109 /* prevent endless salvage,attach,salvage,attach,... loops */
5110 if (vp->stats.salvages >= SALVAGE_COUNT_MAX)
5114 * don't perform salvsync ops on certain threads
5116 thread_opts = pthread_getspecific(VThread_key);
5117 if (thread_opts == NULL) {
5118 thread_opts = &VThread_defaults;
5120 if (thread_opts->disallow_salvsync || vol_disallow_salvsync) {
5125 * XXX the scheduling process should really be done asynchronously
5126 * to avoid fssync deadlocks
5128 if (!vp->salvage.scheduled) {
5129 /* if we haven't previously scheduled a salvage, do so now
5131 * set the volume to an exclusive state and drop the lock
5132 * around the SALVSYNC call
5134 * note that we do NOT acquire a reservation here -- doing so
5135 * could result in unbounded recursion
5137 strlcpy(partName, VPartitionPath(vp->partition), sizeof(partName));
5138 state_save = VChangeState_r(vp, VOL_STATE_SALVSYNC_REQ);
5141 assert(try_SALVSYNC(vp, partName, &code) ||
5142 try_FSSYNC(vp, partName, &code));
5145 VChangeState_r(vp, state_save);
5147 if (code == SYNC_OK) {
5148 vp->salvage.scheduled = 1;
5149 vp->stats.last_salvage_req = FT_ApproxTime();
5150 if (VCanUseSALVSYNC()) {
5151 /* don't record these stats for non-fileservers; let the
5152 * fileserver take care of these */
5153 vp->stats.salvages++;
5154 IncUInt64(&VStats.salvages);
5159 case SYNC_BAD_COMMAND:
5160 case SYNC_COM_ERROR:
5163 Log("VScheduleSalvage_r: Salvage request for volume %lu "
5164 "denied\n", afs_printable_uint32_lu(vp->hashid));
5167 Log("VScheduleSalvage_r: Salvage request for volume %lu "
5168 "received unknown protocol error %d\n",
5169 afs_printable_uint32_lu(vp->hashid), code);
5173 if (VCanUseFSSYNC()) {
5174 VChangeState_r(vp, VOL_STATE_ERROR);
5180 #endif /* SALVSYNC_BUILD_CLIENT || FSSYNC_BUILD_CLIENT */
5182 #ifdef SALVSYNC_BUILD_CLIENT
5185 * connect to the salvageserver SYNC service.
5187 * @return operation status
5191 * @post connection to salvageserver SYNC service established
5193 * @see VConnectSALV_r
5194 * @see VDisconnectSALV
5195 * @see VReconnectSALV
5202 retVal = VConnectSALV_r();
5208 * connect to the salvageserver SYNC service.
5210 * @return operation status
5214 * @pre VOL_LOCK is held.
5216 * @post connection to salvageserver SYNC service established
5219 * @see VDisconnectSALV_r
5220 * @see VReconnectSALV_r
5221 * @see SALVSYNC_clientInit
5223 * @internal volume package internal use only.
5226 VConnectSALV_r(void)
5228 return SALVSYNC_clientInit();
5232 * disconnect from the salvageserver SYNC service.
5234 * @return operation status
5237 * @pre client should have a live connection to the salvageserver
5239 * @post connection to salvageserver SYNC service destroyed
5241 * @see VDisconnectSALV_r
5243 * @see VReconnectSALV
5246 VDisconnectSALV(void)
5249 VDisconnectSALV_r();
5255 * disconnect from the salvageserver SYNC service.
5257 * @return operation status
5261 * @arg VOL_LOCK is held.
5262 * @arg client should have a live connection to the salvageserver.
5264 * @post connection to salvageserver SYNC service destroyed
5266 * @see VDisconnectSALV
5267 * @see VConnectSALV_r
5268 * @see VReconnectSALV_r
5269 * @see SALVSYNC_clientFinis
5271 * @internal volume package internal use only.
5274 VDisconnectSALV_r(void)
5276 return SALVSYNC_clientFinis();
5280 * disconnect and then re-connect to the salvageserver SYNC service.
5282 * @return operation status
5286 * @pre client should have a live connection to the salvageserver
5288 * @post old connection is dropped, and a new one is established
5291 * @see VDisconnectSALV
5292 * @see VReconnectSALV_r
5295 VReconnectSALV(void)
5299 retVal = VReconnectSALV_r();
5305 * disconnect and then re-connect to the salvageserver SYNC service.
5307 * @return operation status
5312 * @arg VOL_LOCK is held.
5313 * @arg client should have a live connection to the salvageserver.
5315 * @post old connection is dropped, and a new one is established
5317 * @see VConnectSALV_r
5318 * @see VDisconnectSALV
5319 * @see VReconnectSALV
5320 * @see SALVSYNC_clientReconnect
5322 * @internal volume package internal use only.
5325 VReconnectSALV_r(void)
5327 return SALVSYNC_clientReconnect();
5329 #endif /* SALVSYNC_BUILD_CLIENT */
5330 #endif /* AFS_DEMAND_ATTACH_FS */
5333 /***************************************************/
5334 /* FSSYNC routines */
5335 /***************************************************/
5337 /* This must be called by any volume utility which needs to run while the
5338 file server is also running. This is separated from VInitVolumePackage2 so
5339 that a utility can fork--and each of the children can independently
5340 initialize communication with the file server */
5341 #ifdef FSSYNC_BUILD_CLIENT
5343 * connect to the fileserver SYNC service.
5345 * @return operation status
5350 * @arg VInit must equal 2.
5351 * @arg Program Type must not be fileserver or salvager.
5353 * @post connection to fileserver SYNC service established
5356 * @see VDisconnectFS
5357 * @see VChildProcReconnectFS
5364 retVal = VConnectFS_r();
5370 * connect to the fileserver SYNC service.
5372 * @return operation status
5377 * @arg VInit must equal 2.
5378 * @arg Program Type must not be fileserver or salvager.
5379 * @arg VOL_LOCK is held.
5381 * @post connection to fileserver SYNC service established
5384 * @see VDisconnectFS_r
5385 * @see VChildProcReconnectFS_r
5387 * @internal volume package internal use only.
5393 assert((VInit == 2) &&
5394 (programType != fileServer) &&
5395 (programType != salvager));
5396 rc = FSYNC_clientInit();
5403 * disconnect from the fileserver SYNC service.
5406 * @arg client should have a live connection to the fileserver.
5407 * @arg VOL_LOCK is held.
5408 * @arg Program Type must not be fileserver or salvager.
5410 * @post connection to fileserver SYNC service destroyed
5412 * @see VDisconnectFS
5414 * @see VChildProcReconnectFS_r
5416 * @internal volume package internal use only.
5419 VDisconnectFS_r(void)
5421 assert((programType != fileServer) &&
5422 (programType != salvager));
5423 FSYNC_clientFinis();
5428 * disconnect from the fileserver SYNC service.
5431 * @arg client should have a live connection to the fileserver.
5432 * @arg Program Type must not be fileserver or salvager.
5434 * @post connection to fileserver SYNC service destroyed
5436 * @see VDisconnectFS_r
5438 * @see VChildProcReconnectFS
5449 * connect to the fileserver SYNC service from a child process following a fork.
5451 * @return operation status
5456 * @arg VOL_LOCK is held.
5457 * @arg current FSYNC handle is shared with a parent process
5459 * @post current FSYNC handle is discarded and a new connection to the
5460 * fileserver SYNC service is established
5462 * @see VChildProcReconnectFS
5464 * @see VDisconnectFS_r
5466 * @internal volume package internal use only.
5469 VChildProcReconnectFS_r(void)
5471 return FSYNC_clientChildProcReconnect();
5475 * connect to the fileserver SYNC service from a child process following a fork.
5477 * @return operation status
5481 * @pre current FSYNC handle is shared with a parent process
5483 * @post current FSYNC handle is discarded and a new connection to the
5484 * fileserver SYNC service is established
5486 * @see VChildProcReconnectFS_r
5488 * @see VDisconnectFS
5491 VChildProcReconnectFS(void)
5495 ret = VChildProcReconnectFS_r();
5499 #endif /* FSSYNC_BUILD_CLIENT */
5502 /***************************************************/
5503 /* volume bitmap routines */
5504 /***************************************************/
5507 * allocate a vnode bitmap number for the vnode
5509 * @param[out] ec error code
5510 * @param[in] vp volume object pointer
5511 * @param[in] index vnode index number for the vnode
5512 * @param[in] flags flag values described in note
5514 * @note for DAFS, flags parameter controls locking behavior.
5515 * If (flags & VOL_ALLOC_BITMAP_WAIT) is set, then this function
5516 * will create a reservation and block on any other exclusive
5517 * operations. Otherwise, this function assumes the caller
5518 * already has exclusive access to vp, and we just change the
5521 * @pre VOL_LOCK held
5523 * @return bit number allocated
5529 VAllocBitmapEntry_r(Error * ec, Volume * vp,
5530 struct vnodeIndex *index, int flags)
5533 register byte *bp, *ep;
5534 #ifdef AFS_DEMAND_ATTACH_FS
5535 VolState state_save;
5536 #endif /* AFS_DEMAND_ATTACH_FS */
5540 /* This test is probably redundant */
5541 if (!VolumeWriteable(vp)) {
5542 *ec = (bit32) VREADONLY;
5546 #ifdef AFS_DEMAND_ATTACH_FS
5547 if (flags & VOL_ALLOC_BITMAP_WAIT) {
5548 VCreateReservation_r(vp);
5549 VWaitExclusiveState_r(vp);
5551 state_save = VChangeState_r(vp, VOL_STATE_GET_BITMAP);
5552 #endif /* AFS_DEMAND_ATTACH_FS */
5555 if ((programType == fileServer) && !index->bitmap) {
5557 #ifndef AFS_DEMAND_ATTACH_FS
5558 /* demand attach fs uses the volume state to avoid races.
5559 * specialStatus field is not used at all */
5561 if (vp->specialStatus == VBUSY) {
5562 if (vp->goingOffline) { /* vos dump waiting for the volume to
5563 * go offline. We probably come here
5564 * from AddNewReadableResidency */
5567 while (vp->specialStatus == VBUSY) {
5568 #ifdef AFS_PTHREAD_ENV
5572 #else /* !AFS_PTHREAD_ENV */
5574 #endif /* !AFS_PTHREAD_ENV */
5578 #endif /* !AFS_DEMAND_ATTACH_FS */
5580 if (!index->bitmap) {
5581 #ifndef AFS_DEMAND_ATTACH_FS
5582 vp->specialStatus = VBUSY; /* Stop anyone else from using it. */
5583 #endif /* AFS_DEMAND_ATTACH_FS */
5584 for (i = 0; i < nVNODECLASSES; i++) {
5585 VGetBitmap_r(ec, vp, i);
5587 #ifdef AFS_DEMAND_ATTACH_FS
5588 VRequestSalvage_r(ec, vp, SALVSYNC_ERROR, VOL_SALVAGE_INVALIDATE_HEADER);
5589 #else /* AFS_DEMAND_ATTACH_FS */
5590 DeleteVolumeFromHashTable(vp);
5591 vp->shuttingDown = 1; /* Let who has it free it. */
5592 vp->specialStatus = 0;
5593 #endif /* AFS_DEMAND_ATTACH_FS */
5597 #ifndef AFS_DEMAND_ATTACH_FS
5599 vp->specialStatus = 0; /* Allow others to have access. */
5600 #endif /* AFS_DEMAND_ATTACH_FS */
5603 #endif /* BITMAP_LATER */
5605 #ifdef AFS_DEMAND_ATTACH_FS
5607 #endif /* AFS_DEMAND_ATTACH_FS */
5608 bp = index->bitmap + index->bitmapOffset;
5609 ep = index->bitmap + index->bitmapSize;
5611 if ((*(bit32 *) bp) != (bit32) 0xffffffff) {
5613 index->bitmapOffset = (afs_uint32) (bp - index->bitmap);
5616 o = ffs(~*bp) - 1; /* ffs is documented in BSTRING(3) */
5618 ret = ((bp - index->bitmap) * 8 + o);
5619 #ifdef AFS_DEMAND_ATTACH_FS
5621 #endif /* AFS_DEMAND_ATTACH_FS */
5624 bp += sizeof(bit32) /* i.e. 4 */ ;
5626 /* No bit map entry--must grow bitmap */
5628 realloc(index->bitmap, index->bitmapSize + VOLUME_BITMAP_GROWSIZE);
5631 bp += index->bitmapSize;
5632 memset(bp, 0, VOLUME_BITMAP_GROWSIZE);
5633 index->bitmapOffset = index->bitmapSize;
5634 index->bitmapSize += VOLUME_BITMAP_GROWSIZE;
5636 ret = index->bitmapOffset * 8;
5637 #ifdef AFS_DEMAND_ATTACH_FS
5639 #endif /* AFS_DEMAND_ATTACH_FS */
5642 #ifdef AFS_DEMAND_ATTACH_FS
5643 VChangeState_r(vp, state_save);
5644 if (flags & VOL_ALLOC_BITMAP_WAIT) {
5645 VCancelReservation_r(vp);
5647 #endif /* AFS_DEMAND_ATTACH_FS */
5652 VAllocBitmapEntry(Error * ec, Volume * vp, register struct vnodeIndex * index)
5656 retVal = VAllocBitmapEntry_r(ec, vp, index, VOL_ALLOC_BITMAP_WAIT);
5662 VFreeBitMapEntry_r(Error * ec, register struct vnodeIndex *index,
5665 unsigned int offset;
5671 #endif /* BITMAP_LATER */
5672 offset = bitNumber >> 3;
5673 if (offset >= index->bitmapSize) {
5677 if (offset < index->bitmapOffset)
5678 index->bitmapOffset = offset & ~3; /* Truncate to nearest bit32 */
5679 *(index->bitmap + offset) &= ~(1 << (bitNumber & 0x7));
5683 VFreeBitMapEntry(Error * ec, register struct vnodeIndex *index,
5687 VFreeBitMapEntry_r(ec, index, bitNumber);
5691 /* this function will drop the glock internally.
5692 * for old pthread fileservers, this is safe thanks to vbusy.
5694 * for demand attach fs, caller must have already called
5695 * VCreateReservation_r and VWaitExclusiveState_r */
5697 VGetBitmap_r(Error * ec, Volume * vp, VnodeClass class)
5699 StreamHandle_t *file;
5700 afs_sfsize_t nVnodes, size;
5701 struct VnodeClassInfo *vcp = &VnodeClassInfo[class];
5702 struct vnodeIndex *vip = &vp->vnodeIndex[class];
5703 struct VnodeDiskObject *vnode;
5704 unsigned int unique = 0;
5708 #endif /* BITMAP_LATER */
5709 #ifdef AFS_DEMAND_ATTACH_FS
5710 VolState state_save;
5711 #endif /* AFS_DEMAND_ATTACH_FS */
5715 #ifdef AFS_DEMAND_ATTACH_FS
5716 state_save = VChangeState_r(vp, VOL_STATE_GET_BITMAP);
5717 #endif /* AFS_DEMAND_ATTACH_FS */
5720 fdP = IH_OPEN(vip->handle);
5721 assert(fdP != NULL);
5722 file = FDH_FDOPEN(fdP, "r");
5723 assert(file != NULL);
5724 vnode = (VnodeDiskObject *) malloc(vcp->diskSize);
5725 assert(vnode != NULL);
5726 size = OS_SIZE(fdP->fd_fd);
5728 nVnodes = (size <= vcp->diskSize ? 0 : size - vcp->diskSize)
5730 vip->bitmapSize = ((nVnodes / 8) + 10) / 4 * 4; /* The 10 is a little extra so
5731 * a few files can be created in this volume,
5732 * the whole thing is rounded up to nearest 4
5733 * bytes, because the bit map allocator likes
5736 BitMap = (byte *) calloc(1, vip->bitmapSize);
5737 assert(BitMap != NULL);
5738 #else /* BITMAP_LATER */
5739 vip->bitmap = (byte *) calloc(1, vip->bitmapSize);
5740 assert(vip->bitmap != NULL);
5741 vip->bitmapOffset = 0;
5742 #endif /* BITMAP_LATER */
5743 if (STREAM_SEEK(file, vcp->diskSize, 0) != -1) {
5745 for (bitNumber = 0; bitNumber < nVnodes + 100; bitNumber++) {
5746 if (STREAM_READ(vnode, vcp->diskSize, 1, file) != 1)
5748 if (vnode->type != vNull) {
5749 if (vnode->vnodeMagic != vcp->magic) {
5750 Log("GetBitmap: addled vnode index in volume %s; volume needs salvage\n", V_name(vp));
5755 *(BitMap + (bitNumber >> 3)) |= (1 << (bitNumber & 0x7));
5756 #else /* BITMAP_LATER */
5757 *(vip->bitmap + (bitNumber >> 3)) |= (1 << (bitNumber & 0x7));
5758 #endif /* BITMAP_LATER */
5759 if (unique <= vnode->uniquifier)
5760 unique = vnode->uniquifier + 1;
5762 #ifndef AFS_PTHREAD_ENV
5763 if ((bitNumber & 0x00ff) == 0x0ff) { /* every 256 iterations */
5766 #endif /* !AFS_PTHREAD_ENV */
5769 if (vp->nextVnodeUnique < unique) {
5770 Log("GetBitmap: bad volume uniquifier for volume %s; volume needs salvage\n", V_name(vp));
5773 /* Paranoia, partly justified--I think fclose after fdopen
5774 * doesn't seem to close fd. In any event, the documentation
5775 * doesn't specify, so it's safer to close it twice.
5783 /* There may have been a racing condition with some other thread, both
5784 * creating the bitmaps for this volume. If the other thread was faster
5785 * the pointer to bitmap should already be filled and we can free ours.
5787 if (vip->bitmap == NULL) {
5788 vip->bitmap = BitMap;
5789 vip->bitmapOffset = 0;
5791 free((byte *) BitMap);
5792 #endif /* BITMAP_LATER */
5793 #ifdef AFS_DEMAND_ATTACH_FS
5794 VChangeState_r(vp, state_save);
5795 #endif /* AFS_DEMAND_ATTACH_FS */
5799 /***************************************************/
5800 /* Volume Path and Volume Number utility routines */
5801 /***************************************************/
5804 * find the first occurrence of a volume header file and return the path.
5806 * @param[out] ec outbound error code
5807 * @param[in] volumeId volume id to find
5808 * @param[out] partitionp pointer to disk partition path string
5809 * @param[out] namep pointer to volume header file name string
5811 * @post path to first occurrence of volume header is returned in partitionp
5812 * and namep, or ec is set accordingly.
5814 * @warning this function is NOT re-entrant -- partitionp and namep point to
5815 * static data segments
5817 * @note if a volume utility inadvertently leaves behind a stale volume header
5818 * on a vice partition, it is possible for callers to get the wrong one,
5819 * depending on the order of the disk partition linked list.
5823 VGetVolumePath(Error * ec, VolId volumeId, char **partitionp, char **namep)
5825 static char partition[VMAXPATHLEN], name[VMAXPATHLEN];
5826 char path[VMAXPATHLEN];
5828 struct DiskPartition64 *dp;
5832 (void)afs_snprintf(&name[1], (sizeof name) - 1, VFORMAT, afs_printable_uint32_lu(volumeId));
5833 for (dp = DiskPartitionList; dp; dp = dp->next) {
5834 struct afs_stat status;
5835 strcpy(path, VPartitionPath(dp));
5837 if (afs_stat(path, &status) == 0) {
5838 strcpy(partition, dp->name);
5845 *partitionp = *namep = NULL;
5847 *partitionp = partition;
5853 * extract a volume number from a volume header filename string.
5855 * @param[in] name volume header filename string
5857 * @return volume number
5859 * @note the string must be of the form VFORMAT. the only permissible
5860 * deviation is a leading '/' character.
5865 VolumeNumber(char *name)
5869 return atoi(name + 1);
5873 * compute the volume header filename.
5875 * @param[in] volumeId
5877 * @return volume header filename
5879 * @post volume header filename string is constructed
5881 * @warning this function is NOT re-entrant -- the returned string is
5882 * stored in a static char array. see VolumeExternalName_r
5883 * for a re-entrant equivalent.
5885 * @see VolumeExternalName_r
5887 * @deprecated due to the above re-entrancy warning, this interface should
5888 * be considered deprecated. Please use VolumeExternalName_r
5892 VolumeExternalName(VolumeId volumeId)
5894 static char name[VMAXPATHLEN];
5895 (void)afs_snprintf(name, sizeof name, VFORMAT, afs_printable_uint32_lu(volumeId));
5900 * compute the volume header filename.
5902 * @param[in] volumeId
5903 * @param[inout] name array in which to store filename
5904 * @param[in] len length of name array
5906 * @return result code from afs_snprintf
5908 * @see VolumeExternalName
5911 * @note re-entrant equivalent of VolumeExternalName
5914 VolumeExternalName_r(VolumeId volumeId, char * name, size_t len)
5916 return afs_snprintf(name, len, VFORMAT, afs_printable_uint32_lu(volumeId));
5920 /***************************************************/
5921 /* Volume Usage Statistics routines */
5922 /***************************************************/
5924 #if OPENAFS_VOL_STATS
5925 #define OneDay (86400) /* 24 hours' worth of seconds */
5927 #define OneDay (24*60*60) /* 24 hours */
5928 #endif /* OPENAFS_VOL_STATS */
5931 Midnight(time_t t) {
5932 struct tm local, *l;
5935 #if defined(AFS_PTHREAD_ENV) && !defined(AFS_NT40_ENV)
5936 l = localtime_r(&t, &local);
5942 /* the following is strictly speaking problematic on the
5943 switching day to daylight saving time, after the switch,
5944 as tm_isdst does not match. Similarly, on the looong day when
5945 switching back the OneDay check will not do what naively expected!
5946 The effects are minor, though, and more a matter of interpreting
5948 #ifndef AFS_PTHREAD_ENV
5951 local.tm_hour = local.tm_min=local.tm_sec = 0;
5952 midnight = mktime(&local);
5953 if (midnight != (time_t) -1) return(midnight);
5955 return( (t/OneDay)*OneDay );
5959 /*------------------------------------------------------------------------
5960 * [export] VAdjustVolumeStatistics
5963 * If we've passed midnight, we need to update all the day use
5964 * statistics as well as zeroing the detailed volume statistics
5965 * (if we are implementing them).
5968 * vp : Pointer to the volume structure describing the lucky
5969 * volume being considered for update.
5975 * Nothing interesting.
5979 *------------------------------------------------------------------------*/
5982 VAdjustVolumeStatistics_r(register Volume * vp)
5984 unsigned int now = FT_ApproxTime();
5986 if (now - V_dayUseDate(vp) > OneDay) {
5987 register int ndays, i;
5989 ndays = (now - V_dayUseDate(vp)) / OneDay;
5990 for (i = 6; i > ndays - 1; i--)
5991 V_weekUse(vp)[i] = V_weekUse(vp)[i - ndays];
5992 for (i = 0; i < ndays - 1 && i < 7; i++)
5993 V_weekUse(vp)[i] = 0;
5995 V_weekUse(vp)[ndays - 1] = V_dayUse(vp);
5997 V_dayUseDate(vp) = Midnight(now);
5999 #if OPENAFS_VOL_STATS
6001 * All we need to do is bzero the entire VOL_STATS_BYTES of
6002 * the detailed volume statistics area.
6004 memset((V_stat_area(vp)), 0, VOL_STATS_BYTES);
6005 #endif /* OPENAFS_VOL_STATS */
6008 /*It's been more than a day of collection */
6010 * Always return happily.
6013 } /*VAdjustVolumeStatistics */
6016 VAdjustVolumeStatistics(register Volume * vp)
6020 retVal = VAdjustVolumeStatistics_r(vp);
6026 VBumpVolumeUsage_r(register Volume * vp)
6028 unsigned int now = FT_ApproxTime();
6029 V_accessDate(vp) = now;
6030 if (now - V_dayUseDate(vp) > OneDay)
6031 VAdjustVolumeStatistics_r(vp);
6033 * Save the volume header image to disk after every 128 bumps to dayUse.
6035 if ((V_dayUse(vp)++ & 127) == 0) {
6037 VUpdateVolume_r(&error, vp, VOL_UPDATE_WAIT);
6042 VBumpVolumeUsage(register Volume * vp)
6045 VBumpVolumeUsage_r(vp);
6050 VSetDiskUsage_r(void)
6052 #ifndef AFS_DEMAND_ATTACH_FS
6053 static int FifteenMinuteCounter = 0;
6057 /* NOTE: Don't attempt to access the partitions list until the
6058 * initialization level indicates that all volumes are attached,
6059 * which implies that all partitions are initialized. */
6060 #ifdef AFS_PTHREAD_ENV
6062 #else /* AFS_PTHREAD_ENV */
6064 #endif /* AFS_PTHREAD_ENV */
6067 VResetDiskUsage_r();
6069 #ifndef AFS_DEMAND_ATTACH_FS
6070 if (++FifteenMinuteCounter == 3) {
6071 FifteenMinuteCounter = 0;
6074 #endif /* !AFS_DEMAND_ATTACH_FS */
6086 /***************************************************/
6087 /* Volume Update List routines */
6088 /***************************************************/
6090 /* The number of minutes that a volume hasn't been updated before the
6091 * "Dont salvage" flag in the volume header will be turned on */
6092 #define SALVAGE_INTERVAL (10*60)
6097 * volume update list functionality has been moved into the VLRU
6098 * the DONT_SALVAGE flag is now set during VLRU demotion
6101 #ifndef AFS_DEMAND_ATTACH_FS
6102 static VolumeId *UpdateList = NULL; /* Pointer to array of Volume ID's */
6103 static int nUpdatedVolumes = 0; /* Updated with entry in UpdateList, salvage after crash flag on */
6104 static int updateSize = 0; /* number of entries possible */
6105 #define UPDATE_LIST_SIZE 128 /* initial size increment (must be a power of 2!) */
6106 #endif /* !AFS_DEMAND_ATTACH_FS */
6109 VAddToVolumeUpdateList_r(Error * ec, Volume * vp)
6112 vp->updateTime = FT_ApproxTime();
6113 if (V_dontSalvage(vp) == 0)
6115 V_dontSalvage(vp) = 0;
6116 VSyncVolume_r(ec, vp, 0);
6117 #ifdef AFS_DEMAND_ATTACH_FS
6118 V_attachFlags(vp) &= ~(VOL_HDR_DONTSALV);
6119 #else /* !AFS_DEMAND_ATTACH_FS */
6122 if (UpdateList == NULL) {
6123 updateSize = UPDATE_LIST_SIZE;
6124 UpdateList = (VolumeId *) malloc(sizeof(VolumeId) * updateSize);
6126 if (nUpdatedVolumes == updateSize) {
6128 if (updateSize > 524288) {
6129 Log("warning: there is likely a bug in the volume update scanner\n");
6133 (VolumeId *) realloc(UpdateList,
6134 sizeof(VolumeId) * updateSize);
6137 assert(UpdateList != NULL);
6138 UpdateList[nUpdatedVolumes++] = V_id(vp);
6139 #endif /* !AFS_DEMAND_ATTACH_FS */
6142 #ifndef AFS_DEMAND_ATTACH_FS
6144 VScanUpdateList(void)
6146 register int i, gap;
6147 register Volume *vp;
6149 afs_uint32 now = FT_ApproxTime();
6150 /* Be careful with this code, since it works with interleaved calls to AddToVolumeUpdateList */
6151 for (i = gap = 0; i < nUpdatedVolumes; i++) {
6153 UpdateList[i - gap] = UpdateList[i];
6155 /* XXX this routine needlessly messes up the Volume LRU by
6156 * breaking the LRU temporal-locality assumptions.....
6157 * we should use a special volume header allocator here */
6158 vp = VGetVolume_r(&error, UpdateList[i - gap] = UpdateList[i]);
6161 } else if (vp->nUsers == 1 && now - vp->updateTime > SALVAGE_INTERVAL) {
6162 V_dontSalvage(vp) = DONT_SALVAGE;
6163 VUpdateVolume_r(&error, vp, 0); /* No need to fsync--not critical */
6171 #ifndef AFS_PTHREAD_ENV
6173 #endif /* !AFS_PTHREAD_ENV */
6175 nUpdatedVolumes -= gap;
6177 #endif /* !AFS_DEMAND_ATTACH_FS */
6180 /***************************************************/
6181 /* Volume LRU routines */
6182 /***************************************************/
6187 * with demand attach fs, we attempt to soft detach(1)
6188 * volumes which have not been accessed in a long time
6189 * in order to speed up fileserver shutdown
6191 * (1) by soft detach we mean a process very similar
6192 * to VOffline, except the final state of the
6193 * Volume will be VOL_STATE_PREATTACHED, instead
6194 * of the usual VOL_STATE_UNATTACHED
6196 #ifdef AFS_DEMAND_ATTACH_FS
6198 /* implementation is reminiscent of a generational GC
6200 * queue 0 is newly attached volumes. this queue is
6201 * sorted by attach timestamp
6203 * queue 1 is volumes that have been around a bit
6204 * longer than queue 0. this queue is sorted by
6207 * queue 2 is volumes tha have been around the longest.
6208 * this queue is unsorted
6210 * queue 3 is volumes that have been marked as
6211 * candidates for soft detachment. this queue is
6214 #define VLRU_GENERATIONS 3 /**< number of generations in VLRU */
6215 #define VLRU_QUEUES 5 /**< total number of VLRU queues */
6218 * definition of a VLRU queue.
6221 volatile struct rx_queue q;
6228 * main VLRU data structure.
6231 struct VLRU_q q[VLRU_QUEUES]; /**< VLRU queues */
6234 /** time interval (in seconds) between promotion passes for
6235 * each young generation queue. */
6236 afs_uint32 promotion_interval[VLRU_GENERATIONS-1];
6238 /** time interval (in seconds) between soft detach candidate
6239 * scans for each generation queue.
6241 * scan_interval[VLRU_QUEUE_CANDIDATE] defines how frequently
6242 * we perform a soft detach pass. */
6243 afs_uint32 scan_interval[VLRU_GENERATIONS+1];
6245 /* scheduler state */
6246 int next_idx; /**< next queue to receive attention */
6247 afs_uint32 last_promotion[VLRU_GENERATIONS-1]; /**< timestamp of last promotion scan */
6248 afs_uint32 last_scan[VLRU_GENERATIONS+1]; /**< timestamp of last detach scan */
6250 int scanner_state; /**< state of scanner thread */
6251 pthread_cond_t cv; /**< state transition CV */
6254 /** global VLRU state */
6255 static struct VLRU volume_LRU;
6258 * defined states for VLRU scanner thread.
6261 VLRU_SCANNER_STATE_OFFLINE = 0, /**< vlru scanner thread is offline */
6262 VLRU_SCANNER_STATE_ONLINE = 1, /**< vlru scanner thread is online */
6263 VLRU_SCANNER_STATE_SHUTTING_DOWN = 2, /**< vlru scanner thread is shutting down */
6264 VLRU_SCANNER_STATE_PAUSING = 3, /**< vlru scanner thread is getting ready to pause */
6265 VLRU_SCANNER_STATE_PAUSED = 4 /**< vlru scanner thread is paused */
6266 } vlru_thread_state_t;
6268 /* vlru disk data header stuff */
6269 #define VLRU_DISK_MAGIC 0x7a8b9cad /**< vlru disk entry magic number */
6270 #define VLRU_DISK_VERSION 1 /**< vlru disk entry version number */
6272 /** vlru default expiration time (for eventual fs state serialization of vlru data) */
6273 #define VLRU_DUMP_EXPIRATION_TIME (60*60*24*7) /* expire vlru data after 1 week */
6276 /** minimum volume inactivity (in seconds) before a volume becomes eligible for
6277 * soft detachment. */
6278 static afs_uint32 VLRU_offline_thresh = VLRU_DEFAULT_OFFLINE_THRESH;
6280 /** time interval (in seconds) between VLRU scanner thread soft detach passes. */
6281 static afs_uint32 VLRU_offline_interval = VLRU_DEFAULT_OFFLINE_INTERVAL;
6283 /** maximum number of volumes to soft detach in a VLRU soft detach pass. */
6284 static afs_uint32 VLRU_offline_max = VLRU_DEFAULT_OFFLINE_MAX;
6286 /** VLRU control flag. non-zero value implies VLRU subsystem is activated. */
6287 static afs_uint32 VLRU_enabled = 1;
6289 /* queue synchronization routines */
6290 static void VLRU_BeginExclusive_r(struct VLRU_q * q);
6291 static void VLRU_EndExclusive_r(struct VLRU_q * q);
6292 static void VLRU_Wait_r(struct VLRU_q * q);
6295 * set VLRU subsystem tunable parameters.
6297 * @param[in] option tunable option to modify
6298 * @param[in] val new value for tunable parameter
6300 * @pre @c VInitVolumePackage2 has not yet been called.
6302 * @post tunable parameter is modified
6306 * @note valid option parameters are:
6307 * @arg @c VLRU_SET_THRESH
6308 * set the period of inactivity after which
6309 * volumes are eligible for soft detachment
6310 * @arg @c VLRU_SET_INTERVAL
6311 * set the time interval between calls
6312 * to the volume LRU "garbage collector"
6313 * @arg @c VLRU_SET_MAX
6314 * set the max number of volumes to deallocate
6318 VLRU_SetOptions(int option, afs_uint32 val)
6320 if (option == VLRU_SET_THRESH) {
6321 VLRU_offline_thresh = val;
6322 } else if (option == VLRU_SET_INTERVAL) {
6323 VLRU_offline_interval = val;
6324 } else if (option == VLRU_SET_MAX) {
6325 VLRU_offline_max = val;
6326 } else if (option == VLRU_SET_ENABLED) {
6329 VLRU_ComputeConstants();
6333 * compute VLRU internal timing parameters.
6335 * @post VLRU scanner thread internal timing parameters are computed
6337 * @note computes internal timing parameters based upon user-modifiable
6338 * tunable parameters.
6342 * @internal volume package internal use only.
6345 VLRU_ComputeConstants(void)
6347 afs_uint32 factor = VLRU_offline_thresh / VLRU_offline_interval;
6349 /* compute the candidate scan interval */
6350 volume_LRU.scan_interval[VLRU_QUEUE_CANDIDATE] = VLRU_offline_interval;
6352 /* compute the promotion intervals */
6353 volume_LRU.promotion_interval[VLRU_QUEUE_NEW] = VLRU_offline_thresh * 2;
6354 volume_LRU.promotion_interval[VLRU_QUEUE_MID] = VLRU_offline_thresh * 4;
6357 /* compute the gen 0 scan interval */
6358 volume_LRU.scan_interval[VLRU_QUEUE_NEW] = VLRU_offline_thresh / 8;
6360 /* compute the gen 0 scan interval */
6361 volume_LRU.scan_interval[VLRU_QUEUE_NEW] = VLRU_offline_interval * 2;
6366 * initialize VLRU subsystem.
6368 * @pre this function has not yet been called
6370 * @post VLRU subsystem is initialized and VLRU scanner thread is starting
6374 * @internal volume package internal use only.
6380 pthread_attr_t attrs;
6383 if (!VLRU_enabled) {
6384 Log("VLRU: disabled\n");
6388 /* initialize each of the VLRU queues */
6389 for (i = 0; i < VLRU_QUEUES; i++) {
6390 queue_Init(&volume_LRU.q[i]);
6391 volume_LRU.q[i].len = 0;
6392 volume_LRU.q[i].busy = 0;
6393 assert(pthread_cond_init(&volume_LRU.q[i].cv, NULL) == 0);
6396 /* setup the timing constants */
6397 VLRU_ComputeConstants();
6399 /* XXX put inside LogLevel check? */
6400 Log("VLRU: starting scanner with the following configuration parameters:\n");
6401 Log("VLRU: offlining volumes after minimum of %d seconds of inactivity\n", VLRU_offline_thresh);
6402 Log("VLRU: running VLRU soft detach pass every %d seconds\n", VLRU_offline_interval);
6403 Log("VLRU: taking up to %d volumes offline per pass\n", VLRU_offline_max);
6404 Log("VLRU: scanning generation 0 for inactive volumes every %d seconds\n", volume_LRU.scan_interval[0]);
6405 Log("VLRU: scanning for promotion/demotion between generations 0 and 1 every %d seconds\n", volume_LRU.promotion_interval[0]);
6406 Log("VLRU: scanning for promotion/demotion between generations 1 and 2 every %d seconds\n", volume_LRU.promotion_interval[1]);
6408 /* start up the VLRU scanner */
6409 volume_LRU.scanner_state = VLRU_SCANNER_STATE_OFFLINE;
6410 if (programType == fileServer) {
6411 assert(pthread_cond_init(&volume_LRU.cv, NULL) == 0);
6412 assert(pthread_attr_init(&attrs) == 0);
6413 assert(pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED) == 0);
6414 assert(pthread_create(&tid, &attrs, &VLRU_ScannerThread, NULL) == 0);
6419 * initialize the VLRU-related fields of a newly allocated volume object.
6421 * @param[in] vp pointer to volume object
6424 * @arg @c VOL_LOCK is held.
6425 * @arg volume object is not on a VLRU queue.
6427 * @post VLRU fields are initialized to indicate that volume object is not
6428 * currently registered with the VLRU subsystem
6432 * @internal volume package interal use only.
6435 VLRU_Init_Node_r(Volume * vp)
6440 assert(queue_IsNotOnQueue(&vp->vlru));
6441 vp->vlru.idx = VLRU_QUEUE_INVALID;
6445 * add a volume object to a VLRU queue.
6447 * @param[in] vp pointer to volume object
6450 * @arg @c VOL_LOCK is held.
6451 * @arg caller MUST hold a lightweight ref on @p vp.
6452 * @arg caller MUST NOT hold exclusive ownership of the VLRU queue.
6454 * @post the volume object is added to the appropriate VLRU queue
6456 * @note if @c vp->vlru.idx contains the index of a valid VLRU queue,
6457 * then the volume is added to that queue. Otherwise, the value
6458 * @c VLRU_QUEUE_NEW is stored into @c vp->vlru.idx and the
6459 * volume is added to the NEW generation queue.
6461 * @note @c VOL_LOCK may be dropped internally
6463 * @note Volume state is temporarily set to @c VOL_STATE_VLRU_ADD
6464 * during the add operation, and is restored to the previous
6465 * state prior to return.
6469 * @internal volume package internal use only.
6472 VLRU_Add_r(Volume * vp)
6475 VolState state_save;
6480 if (queue_IsOnQueue(&vp->vlru))
6483 state_save = VChangeState_r(vp, VOL_STATE_VLRU_ADD);
6486 if ((idx < 0) || (idx >= VLRU_QUEUE_INVALID)) {
6487 idx = VLRU_QUEUE_NEW;
6490 VLRU_Wait_r(&volume_LRU.q[idx]);
6492 /* repeat check since VLRU_Wait_r may have dropped
6494 if (queue_IsNotOnQueue(&vp->vlru)) {
6496 queue_Prepend(&volume_LRU.q[idx], &vp->vlru);
6497 volume_LRU.q[idx].len++;
6498 V_attachFlags(vp) |= VOL_ON_VLRU;
6499 vp->stats.last_promote = FT_ApproxTime();
6502 VChangeState_r(vp, state_save);
6506 * delete a volume object from a VLRU queue.
6508 * @param[in] vp pointer to volume object
6511 * @arg @c VOL_LOCK is held.
6512 * @arg caller MUST hold a lightweight ref on @p vp.
6513 * @arg caller MUST NOT hold exclusive ownership of the VLRU queue.
6515 * @post volume object is removed from the VLRU queue
6517 * @note @c VOL_LOCK may be dropped internally
6521 * @todo We should probably set volume state to something exlcusive
6522 * (as @c VLRU_Add_r does) prior to dropping @c VOL_LOCK.
6524 * @internal volume package internal use only.
6527 VLRU_Delete_r(Volume * vp)
6534 if (queue_IsNotOnQueue(&vp->vlru))
6540 if (idx == VLRU_QUEUE_INVALID)
6542 VLRU_Wait_r(&volume_LRU.q[idx]);
6543 } while (idx != vp->vlru.idx);
6545 /* now remove from the VLRU and update
6546 * the appropriate counter */
6547 queue_Remove(&vp->vlru);
6548 volume_LRU.q[idx].len--;
6549 vp->vlru.idx = VLRU_QUEUE_INVALID;
6550 V_attachFlags(vp) &= ~(VOL_ON_VLRU);
6554 * tell the VLRU subsystem that a volume was just accessed.
6556 * @param[in] vp pointer to volume object
6559 * @arg @c VOL_LOCK is held
6560 * @arg caller MUST hold a lightweight ref on @p vp
6561 * @arg caller MUST NOT hold exclusive ownership of any VLRU queue
6563 * @post volume VLRU access statistics are updated. If the volume was on
6564 * the VLRU soft detach candidate queue, it is moved to the NEW
6567 * @note @c VOL_LOCK may be dropped internally
6571 * @internal volume package internal use only.
6574 VLRU_UpdateAccess_r(Volume * vp)
6576 Volume * rvp = NULL;
6581 if (queue_IsNotOnQueue(&vp->vlru))
6584 assert(V_attachFlags(vp) & VOL_ON_VLRU);
6586 /* update the access timestamp */
6587 vp->stats.last_get = FT_ApproxTime();
6590 * if the volume is on the soft detach candidate
6591 * list, we need to safely move it back to a
6592 * regular generation. this has to be done
6593 * carefully so we don't race against the scanner
6597 /* if this volume is on the soft detach candidate queue,
6598 * then grab exclusive access to the necessary queues */
6599 if (vp->vlru.idx == VLRU_QUEUE_CANDIDATE) {
6601 VCreateReservation_r(rvp);
6603 VLRU_Wait_r(&volume_LRU.q[VLRU_QUEUE_NEW]);
6604 VLRU_BeginExclusive_r(&volume_LRU.q[VLRU_QUEUE_NEW]);
6605 VLRU_Wait_r(&volume_LRU.q[VLRU_QUEUE_CANDIDATE]);
6606 VLRU_BeginExclusive_r(&volume_LRU.q[VLRU_QUEUE_CANDIDATE]);
6609 /* make sure multiple threads don't race to update */
6610 if (vp->vlru.idx == VLRU_QUEUE_CANDIDATE) {
6611 VLRU_SwitchQueues(vp, VLRU_QUEUE_NEW, 1);
6615 VLRU_EndExclusive_r(&volume_LRU.q[VLRU_QUEUE_CANDIDATE]);
6616 VLRU_EndExclusive_r(&volume_LRU.q[VLRU_QUEUE_NEW]);
6617 VCancelReservation_r(rvp);
6622 * switch a volume between two VLRU queues.
6624 * @param[in] vp pointer to volume object
6625 * @param[in] new_idx index of VLRU queue onto which the volume will be moved
6626 * @param[in] append controls whether the volume will be appended or
6627 * prepended to the queue. A nonzero value means it will
6628 * be appended; zero means it will be prepended.
6630 * @pre The new (and old, if applicable) queue(s) must either be owned
6631 * exclusively by the calling thread for asynchronous manipulation,
6632 * or the queue(s) must be quiescent and VOL_LOCK must be held.
6633 * Please see VLRU_BeginExclusive_r, VLRU_EndExclusive_r and VLRU_Wait_r
6634 * for further details of the queue asynchronous processing mechanism.
6636 * @post If the volume object was already on a VLRU queue, it is
6637 * removed from the queue. Depending on the value of the append
6638 * parameter, the volume object is either appended or prepended
6639 * to the VLRU queue referenced by the new_idx parameter.
6643 * @see VLRU_BeginExclusive_r
6644 * @see VLRU_EndExclusive_r
6647 * @internal volume package internal use only.
6650 VLRU_SwitchQueues(Volume * vp, int new_idx, int append)
6652 if (queue_IsNotOnQueue(&vp->vlru))
6655 queue_Remove(&vp->vlru);
6656 volume_LRU.q[vp->vlru.idx].len--;
6658 /* put the volume back on the correct generational queue */
6660 queue_Append(&volume_LRU.q[new_idx], &vp->vlru);
6662 queue_Prepend(&volume_LRU.q[new_idx], &vp->vlru);
6665 volume_LRU.q[new_idx].len++;
6666 vp->vlru.idx = new_idx;
6670 * VLRU background thread.
6672 * The VLRU Scanner Thread is responsible for periodically scanning through
6673 * each VLRU queue looking for volumes which should be moved to another
6674 * queue, or soft detached.
6676 * @param[in] args unused thread arguments parameter
6678 * @return unused thread return value
6679 * @retval NULL always
6681 * @internal volume package internal use only.
6684 VLRU_ScannerThread(void * args)
6686 afs_uint32 now, min_delay, delay;
6687 int i, min_idx, min_op, overdue, state;
6689 /* set t=0 for promotion cycle to be
6690 * fileserver startup */
6691 now = FT_ApproxTime();
6692 for (i=0; i < VLRU_GENERATIONS-1; i++) {
6693 volume_LRU.last_promotion[i] = now;
6696 /* don't start the scanner until VLRU_offline_thresh
6697 * plus a small delay for VInitVolumePackage2 to finish
6700 sleep(VLRU_offline_thresh + 60);
6702 /* set t=0 for scan cycle to be now */
6703 now = FT_ApproxTime();
6704 for (i=0; i < VLRU_GENERATIONS+1; i++) {
6705 volume_LRU.last_scan[i] = now;
6709 if (volume_LRU.scanner_state == VLRU_SCANNER_STATE_OFFLINE) {
6710 volume_LRU.scanner_state = VLRU_SCANNER_STATE_ONLINE;
6713 while ((state = volume_LRU.scanner_state) != VLRU_SCANNER_STATE_SHUTTING_DOWN) {
6714 /* check to see if we've been asked to pause */
6715 if (volume_LRU.scanner_state == VLRU_SCANNER_STATE_PAUSING) {
6716 volume_LRU.scanner_state = VLRU_SCANNER_STATE_PAUSED;
6717 assert(pthread_cond_broadcast(&volume_LRU.cv) == 0);
6719 VOL_CV_WAIT(&volume_LRU.cv);
6720 } while (volume_LRU.scanner_state == VLRU_SCANNER_STATE_PAUSED);
6723 /* scheduling can happen outside the glock */
6726 /* figure out what is next on the schedule */
6728 /* figure out a potential schedule for the new generation first */
6730 min_delay = volume_LRU.scan_interval[0] + volume_LRU.last_scan[0] - now;
6733 if (min_delay > volume_LRU.scan_interval[0]) {
6734 /* unsigned overflow -- we're overdue to run this scan */
6739 /* if we're not overdue for gen 0, figure out schedule for candidate gen */
6741 i = VLRU_QUEUE_CANDIDATE;
6742 delay = volume_LRU.scan_interval[i] + volume_LRU.last_scan[i] - now;
6743 if (delay < min_delay) {
6747 if (delay > volume_LRU.scan_interval[i]) {
6748 /* unsigned overflow -- we're overdue to run this scan */
6755 /* if we're still not overdue for something, figure out schedules for promotions */
6756 for (i=0; !overdue && i < VLRU_GENERATIONS-1; i++) {
6757 delay = volume_LRU.promotion_interval[i] + volume_LRU.last_promotion[i] - now;
6758 if (delay < min_delay) {
6763 if (delay > volume_LRU.promotion_interval[i]) {
6764 /* unsigned overflow -- we're overdue to run this promotion */
6773 /* sleep as needed */
6778 /* do whatever is next */
6781 VLRU_Promote_r(min_idx);
6782 VLRU_Demote_r(min_idx+1);
6784 VLRU_Scan_r(min_idx);
6786 now = FT_ApproxTime();
6789 Log("VLRU scanner asked to go offline (scanner_state=%d)\n", state);
6791 /* signal that scanner is down */
6792 volume_LRU.scanner_state = VLRU_SCANNER_STATE_OFFLINE;
6793 assert(pthread_cond_broadcast(&volume_LRU.cv) == 0);
6799 * promote volumes from one VLRU generation to the next.
6801 * This routine scans a VLRU generation looking for volumes which are
6802 * eligible to be promoted to the next generation. All volumes which
6803 * meet the eligibility requirement are promoted.
6805 * Promotion eligibility is based upon meeting both of the following
6808 * @arg The volume has been accessed since the last promotion:
6809 * @c (vp->stats.last_get >= vp->stats.last_promote)
6810 * @arg The last promotion occurred at least
6811 * @c volume_LRU.promotion_interval[idx] seconds ago
6813 * As a performance optimization, promotions are "globbed". In other
6814 * words, we promote arbitrarily large contiguous sublists of elements
6817 * @param[in] idx VLRU queue index to scan
6821 * @internal VLRU internal use only.
6824 VLRU_Promote_r(int idx)
6826 int len, chaining, promote;
6827 afs_uint32 now, thresh;
6828 struct rx_queue *qp, *nqp;
6829 Volume * vp, *start = NULL, *end = NULL;
6831 /* get exclusive access to two chains, and drop the glock */
6832 VLRU_Wait_r(&volume_LRU.q[idx]);
6833 VLRU_BeginExclusive_r(&volume_LRU.q[idx]);
6834 VLRU_Wait_r(&volume_LRU.q[idx+1]);
6835 VLRU_BeginExclusive_r(&volume_LRU.q[idx+1]);
6838 thresh = volume_LRU.promotion_interval[idx];
6839 now = FT_ApproxTime();
6842 for (queue_ScanBackwards(&volume_LRU.q[idx], qp, nqp, rx_queue)) {
6843 vp = (Volume *)((char *)qp - offsetof(Volume, vlru));
6844 promote = (((vp->stats.last_promote + thresh) <= now) &&
6845 (vp->stats.last_get >= vp->stats.last_promote));
6853 /* promote and prepend chain */
6854 queue_MoveChainAfter(&volume_LRU.q[idx+1], &start->vlru, &end->vlru);
6868 /* promote and prepend */
6869 queue_MoveChainAfter(&volume_LRU.q[idx+1], &start->vlru, &end->vlru);
6873 volume_LRU.q[idx].len -= len;
6874 volume_LRU.q[idx+1].len += len;
6877 /* release exclusive access to the two chains */
6879 volume_LRU.last_promotion[idx] = now;
6880 VLRU_EndExclusive_r(&volume_LRU.q[idx+1]);
6881 VLRU_EndExclusive_r(&volume_LRU.q[idx]);
6884 /* run the demotions */
6886 VLRU_Demote_r(int idx)
6889 int len, chaining, demote;
6890 afs_uint32 now, thresh;
6891 struct rx_queue *qp, *nqp;
6892 Volume * vp, *start = NULL, *end = NULL;
6893 Volume ** salv_flag_vec = NULL;
6894 int salv_vec_offset = 0;
6896 assert(idx == VLRU_QUEUE_MID || idx == VLRU_QUEUE_OLD);
6898 /* get exclusive access to two chains, and drop the glock */
6899 VLRU_Wait_r(&volume_LRU.q[idx-1]);
6900 VLRU_BeginExclusive_r(&volume_LRU.q[idx-1]);
6901 VLRU_Wait_r(&volume_LRU.q[idx]);
6902 VLRU_BeginExclusive_r(&volume_LRU.q[idx]);
6905 /* no big deal if this allocation fails */
6906 if (volume_LRU.q[idx].len) {
6907 salv_flag_vec = (Volume **) malloc(volume_LRU.q[idx].len * sizeof(Volume *));
6910 now = FT_ApproxTime();
6911 thresh = volume_LRU.promotion_interval[idx-1];
6914 for (queue_ScanBackwards(&volume_LRU.q[idx], qp, nqp, rx_queue)) {
6915 vp = (Volume *)((char *)qp - offsetof(Volume, vlru));
6916 demote = (((vp->stats.last_promote + thresh) <= now) &&
6917 (vp->stats.last_get < (now - thresh)));
6919 /* we now do volume update list DONT_SALVAGE flag setting during
6920 * demotion passes */
6921 if (salv_flag_vec &&
6922 !(V_attachFlags(vp) & VOL_HDR_DONTSALV) &&
6924 (vp->updateTime < (now - SALVAGE_INTERVAL)) &&
6925 (V_attachState(vp) == VOL_STATE_ATTACHED)) {
6926 salv_flag_vec[salv_vec_offset++] = vp;
6927 VCreateReservation_r(vp);
6936 /* demote and append chain */
6937 queue_MoveChainBefore(&volume_LRU.q[idx-1], &start->vlru, &end->vlru);
6951 queue_MoveChainBefore(&volume_LRU.q[idx-1], &start->vlru, &end->vlru);
6955 volume_LRU.q[idx].len -= len;
6956 volume_LRU.q[idx-1].len += len;
6959 /* release exclusive access to the two chains */
6961 VLRU_EndExclusive_r(&volume_LRU.q[idx]);
6962 VLRU_EndExclusive_r(&volume_LRU.q[idx-1]);
6964 /* now go back and set the DONT_SALVAGE flags as appropriate */
6965 if (salv_flag_vec) {
6967 for (i = 0; i < salv_vec_offset; i++) {
6968 vp = salv_flag_vec[i];
6969 if (!(V_attachFlags(vp) & VOL_HDR_DONTSALV) &&
6970 (vp->updateTime < (now - SALVAGE_INTERVAL)) &&
6971 (V_attachState(vp) == VOL_STATE_ATTACHED)) {
6974 V_attachFlags(vp) |= VOL_HDR_DONTSALV;
6975 V_dontSalvage(vp) = DONT_SALVAGE;
6976 VUpdateVolume_r(&ec, vp, 0);
6980 VCancelReservation_r(vp);
6982 free(salv_flag_vec);
6986 /* run a pass of the VLRU GC scanner */
6988 VLRU_Scan_r(int idx)
6990 afs_uint32 now, thresh;
6991 struct rx_queue *qp, *nqp;
6995 assert(idx == VLRU_QUEUE_NEW || idx == VLRU_QUEUE_CANDIDATE);
6997 /* gain exclusive access to the idx VLRU */
6998 VLRU_Wait_r(&volume_LRU.q[idx]);
6999 VLRU_BeginExclusive_r(&volume_LRU.q[idx]);
7001 if (idx != VLRU_QUEUE_CANDIDATE) {
7002 /* gain exclusive access to the candidate VLRU */
7003 VLRU_Wait_r(&volume_LRU.q[VLRU_QUEUE_CANDIDATE]);
7004 VLRU_BeginExclusive_r(&volume_LRU.q[VLRU_QUEUE_CANDIDATE]);
7007 now = FT_ApproxTime();
7008 thresh = now - VLRU_offline_thresh;
7010 /* perform candidate selection and soft detaching */
7011 if (idx == VLRU_QUEUE_CANDIDATE) {
7012 /* soft detach some volumes from the candidate pool */
7016 for (i=0,queue_ScanBackwards(&volume_LRU.q[idx], qp, nqp, rx_queue)) {
7017 vp = (Volume *)((char *)qp - offsetof(Volume, vlru));
7018 if (i >= VLRU_offline_max) {
7021 /* check timestamp to see if it's a candidate for soft detaching */
7022 if (vp->stats.last_get <= thresh) {
7024 if (VCheckSoftDetach(vp, thresh))
7030 /* scan for volumes to become soft detach candidates */
7031 for (i=1,queue_ScanBackwards(&volume_LRU.q[idx], qp, nqp, rx_queue),i++) {
7032 vp = (Volume *)((char *)qp - offsetof(Volume, vlru));
7034 /* check timestamp to see if it's a candidate for soft detaching */
7035 if (vp->stats.last_get <= thresh) {
7036 VCheckSoftDetachCandidate(vp, thresh);
7039 if (!(i&0x7f)) { /* lock coarsening optimization */
7047 /* relinquish exclusive access to the VLRU chains */
7051 volume_LRU.last_scan[idx] = now;
7052 if (idx != VLRU_QUEUE_CANDIDATE) {
7053 VLRU_EndExclusive_r(&volume_LRU.q[VLRU_QUEUE_CANDIDATE]);
7055 VLRU_EndExclusive_r(&volume_LRU.q[idx]);
7058 /* check whether volume is safe to soft detach
7059 * caller MUST NOT hold a ref count on vp */
7061 VCheckSoftDetach(Volume * vp, afs_uint32 thresh)
7065 if (vp->nUsers || vp->nWaiters)
7068 if (vp->stats.last_get <= thresh) {
7069 ret = VSoftDetachVolume_r(vp, thresh);
7075 /* check whether volume should be made a
7076 * soft detach candidate */
7078 VCheckSoftDetachCandidate(Volume * vp, afs_uint32 thresh)
7081 if (vp->nUsers || vp->nWaiters)
7086 assert(idx == VLRU_QUEUE_NEW);
7088 if (vp->stats.last_get <= thresh) {
7089 /* move to candidate pool */
7090 queue_Remove(&vp->vlru);
7091 volume_LRU.q[VLRU_QUEUE_NEW].len--;
7092 queue_Prepend(&volume_LRU.q[VLRU_QUEUE_CANDIDATE], &vp->vlru);
7093 vp->vlru.idx = VLRU_QUEUE_CANDIDATE;
7094 volume_LRU.q[VLRU_QUEUE_CANDIDATE].len++;
7102 /* begin exclusive access on VLRU */
7104 VLRU_BeginExclusive_r(struct VLRU_q * q)
7106 assert(q->busy == 0);
7110 /* end exclusive access on VLRU */
7112 VLRU_EndExclusive_r(struct VLRU_q * q)
7116 assert(pthread_cond_broadcast(&q->cv) == 0);
7119 /* wait for another thread to end exclusive access on VLRU */
7121 VLRU_Wait_r(struct VLRU_q * q)
7124 VOL_CV_WAIT(&q->cv);
7129 * volume soft detach
7131 * caller MUST NOT hold a ref count on vp */
7133 VSoftDetachVolume_r(Volume * vp, afs_uint32 thresh)
7138 assert(vp->vlru.idx == VLRU_QUEUE_CANDIDATE);
7140 ts_save = vp->stats.last_get;
7141 if (ts_save > thresh)
7144 if (vp->nUsers || vp->nWaiters)
7147 if (VIsExclusiveState(V_attachState(vp))) {
7151 switch (V_attachState(vp)) {
7152 case VOL_STATE_UNATTACHED:
7153 case VOL_STATE_PREATTACHED:
7154 case VOL_STATE_ERROR:
7155 case VOL_STATE_GOING_OFFLINE:
7156 case VOL_STATE_SHUTTING_DOWN:
7157 case VOL_STATE_SALVAGING:
7158 volume_LRU.q[vp->vlru.idx].len--;
7160 /* create and cancel a reservation to
7161 * give the volume an opportunity to
7163 VCreateReservation_r(vp);
7164 queue_Remove(&vp->vlru);
7165 vp->vlru.idx = VLRU_QUEUE_INVALID;
7166 V_attachFlags(vp) &= ~(VOL_ON_VLRU);
7167 VCancelReservation_r(vp);
7173 /* hold the volume and take it offline.
7174 * no need for reservations, as VHold_r
7175 * takes care of that internally. */
7176 if (VHold_r(vp) == 0) {
7177 /* vhold drops the glock, so now we should
7178 * check to make sure we aren't racing against
7179 * other threads. if we are racing, offlining vp
7180 * would be wasteful, and block the scanner for a while
7184 (vp->shuttingDown) ||
7185 (vp->goingOffline) ||
7186 (vp->stats.last_get != ts_save)) {
7187 /* looks like we're racing someone else. bail */
7191 /* pull it off the VLRU */
7192 assert(vp->vlru.idx == VLRU_QUEUE_CANDIDATE);
7193 volume_LRU.q[VLRU_QUEUE_CANDIDATE].len--;
7194 queue_Remove(&vp->vlru);
7195 vp->vlru.idx = VLRU_QUEUE_INVALID;
7196 V_attachFlags(vp) &= ~(VOL_ON_VLRU);
7198 /* take if offline */
7199 VOffline_r(vp, "volume has been soft detached");
7201 /* invalidate the volume header cache */
7202 FreeVolumeHeader(vp);
7205 IncUInt64(&VStats.soft_detaches);
7206 vp->stats.soft_detaches++;
7208 /* put in pre-attached state so demand
7209 * attacher can work on it */
7210 VChangeState_r(vp, VOL_STATE_PREATTACHED);
7216 #endif /* AFS_DEMAND_ATTACH_FS */
7219 /***************************************************/
7220 /* Volume Header Cache routines */
7221 /***************************************************/
7224 * volume header cache.
7226 struct volume_hdr_LRU_t volume_hdr_LRU;
7229 * initialize the volume header cache.
7231 * @param[in] howMany number of header cache entries to preallocate
7233 * @pre VOL_LOCK held. Function has never been called before.
7235 * @post howMany cache entries are allocated, initialized, and added
7236 * to the LRU list. Header cache statistics are initialized.
7238 * @note only applicable to fileServer program type. Should only be
7239 * called once during volume package initialization.
7241 * @internal volume package internal use only.
7244 VInitVolumeHeaderCache(afs_uint32 howMany)
7246 register struct volHeader *hp;
7247 if (programType != fileServer)
7249 queue_Init(&volume_hdr_LRU);
7250 volume_hdr_LRU.stats.free = 0;
7251 volume_hdr_LRU.stats.used = howMany;
7252 volume_hdr_LRU.stats.attached = 0;
7253 hp = (struct volHeader *)(calloc(howMany, sizeof(struct volHeader)));
7257 /* We are using ReleaseVolumeHeader to initialize the values on the header list
7258 * to ensure they have the right values
7260 ReleaseVolumeHeader(hp++);
7264 * get a volume header and attach it to the volume object.
7266 * @param[in] vp pointer to volume object
7268 * @return cache entry status
7269 * @retval 0 volume header was newly attached; cache data is invalid
7270 * @retval 1 volume header was previously attached; cache data is valid
7272 * @pre VOL_LOCK held. For DAFS, lightweight ref must be held on volume object.
7274 * @post volume header attached to volume object. if necessary, header cache
7275 * entry on LRU is synchronized to disk. Header is removed from LRU list.
7277 * @note VOL_LOCK may be dropped
7279 * @warning this interface does not load header data from disk. it merely
7280 * attaches a header object to the volume object, and may sync the old
7281 * header cache data out to disk in the process.
7283 * @internal volume package internal use only.
7286 GetVolumeHeader(register Volume * vp)
7289 register struct volHeader *hd;
7291 static int everLogged = 0;
7293 #ifdef AFS_DEMAND_ATTACH_FS
7294 VolState vp_save = 0, back_save = 0;
7296 /* XXX debug 9/19/05 we've apparently got
7297 * a ref counting bug somewhere that's
7298 * breaking the nUsers == 0 => header on LRU
7300 if (vp->header && queue_IsNotOnQueue(vp->header)) {
7301 Log("nUsers == 0, but header not on LRU\n");
7306 old = (vp->header != NULL); /* old == volume already has a header */
7308 if (programType != fileServer) {
7309 /* for volume utilities, we allocate volHeaders as needed */
7311 hd = (struct volHeader *)calloc(1, sizeof(*vp->header));
7315 #ifdef AFS_DEMAND_ATTACH_FS
7316 V_attachFlags(vp) |= VOL_HDR_ATTACHED;
7320 /* for the fileserver, we keep a volume header cache */
7322 /* the header we previously dropped in the lru is
7323 * still available. pull it off the lru and return */
7326 assert(hd->back == vp);
7327 #ifdef AFS_DEMAND_ATTACH_FS
7328 V_attachFlags(vp) &= ~(VOL_HDR_IN_LRU);
7331 /* we need to grab a new element off the LRU */
7332 if (queue_IsNotEmpty(&volume_hdr_LRU)) {
7333 /* grab an element and pull off of LRU */
7334 hd = queue_First(&volume_hdr_LRU, volHeader);
7337 /* LRU is empty, so allocate a new volHeader
7338 * this is probably indicative of a leak, so let the user know */
7339 hd = (struct volHeader *)calloc(1, sizeof(struct volHeader));
7342 Log("****Allocated more volume headers, probably leak****\n");
7345 volume_hdr_LRU.stats.free++;
7348 /* this header used to belong to someone else.
7349 * we'll need to check if the header needs to
7350 * be sync'd out to disk */
7352 #ifdef AFS_DEMAND_ATTACH_FS
7353 /* if hd->back were in an exclusive state, then
7354 * its volHeader would not be on the LRU... */
7355 assert(!VIsExclusiveState(V_attachState(hd->back)));
7358 if (hd->diskstuff.inUse) {
7359 /* volume was in use, so we'll need to sync
7360 * its header to disk */
7362 #ifdef AFS_DEMAND_ATTACH_FS
7363 back_save = VChangeState_r(hd->back, VOL_STATE_UPDATING);
7364 vp_save = VChangeState_r(vp, VOL_STATE_HDR_ATTACHING);
7365 VCreateReservation_r(hd->back);
7369 WriteVolumeHeader_r(&error, hd->back);
7370 /* Ignore errors; catch them later */
7372 #ifdef AFS_DEMAND_ATTACH_FS
7377 hd->back->header = NULL;
7378 #ifdef AFS_DEMAND_ATTACH_FS
7379 V_attachFlags(hd->back) &= ~(VOL_HDR_ATTACHED | VOL_HDR_LOADED | VOL_HDR_IN_LRU);
7381 if (hd->diskstuff.inUse) {
7382 VChangeState_r(hd->back, back_save);
7383 VCancelReservation_r(hd->back);
7384 VChangeState_r(vp, vp_save);
7388 volume_hdr_LRU.stats.attached++;
7392 #ifdef AFS_DEMAND_ATTACH_FS
7393 V_attachFlags(vp) |= VOL_HDR_ATTACHED;
7396 volume_hdr_LRU.stats.free--;
7397 volume_hdr_LRU.stats.used++;
7399 IncUInt64(&VStats.hdr_gets);
7400 #ifdef AFS_DEMAND_ATTACH_FS
7401 IncUInt64(&vp->stats.hdr_gets);
7402 vp->stats.last_hdr_get = FT_ApproxTime();
7409 * make sure volume header is attached and contains valid cache data.
7411 * @param[out] ec outbound error code
7412 * @param[in] vp pointer to volume object
7414 * @pre VOL_LOCK held. For DAFS, lightweight ref held on vp.
7416 * @post header cache entry attached, and loaded with valid data, or
7417 * *ec is nonzero, and the header is released back into the LRU.
7419 * @internal volume package internal use only.
7422 LoadVolumeHeader(Error * ec, Volume * vp)
7424 #ifdef AFS_DEMAND_ATTACH_FS
7425 VolState state_save;
7429 if (vp->nUsers == 0 && !GetVolumeHeader(vp)) {
7430 IncUInt64(&VStats.hdr_loads);
7431 state_save = VChangeState_r(vp, VOL_STATE_HDR_LOADING);
7434 ReadHeader(ec, V_diskDataHandle(vp), (char *)&V_disk(vp),
7435 sizeof(V_disk(vp)), VOLUMEINFOMAGIC,
7437 IncUInt64(&vp->stats.hdr_loads);
7438 now = FT_ApproxTime();
7442 V_attachFlags(vp) |= VOL_HDR_LOADED;
7443 vp->stats.last_hdr_load = now;
7445 VChangeState_r(vp, state_save);
7447 #else /* AFS_DEMAND_ATTACH_FS */
7449 if (vp->nUsers == 0 && !GetVolumeHeader(vp)) {
7450 IncUInt64(&VStats.hdr_loads);
7452 ReadHeader(ec, V_diskDataHandle(vp), (char *)&V_disk(vp),
7453 sizeof(V_disk(vp)), VOLUMEINFOMAGIC,
7456 #endif /* AFS_DEMAND_ATTACH_FS */
7458 /* maintain (nUsers==0) => header in LRU invariant */
7459 FreeVolumeHeader(vp);
7464 * release a header cache entry back into the LRU list.
7466 * @param[in] hd pointer to volume header cache object
7468 * @pre VOL_LOCK held.
7470 * @post header cache object appended onto end of LRU list.
7472 * @note only applicable to fileServer program type.
7474 * @note used to place a header cache entry back into the
7475 * LRU pool without invalidating it as a cache entry.
7477 * @internal volume package internal use only.
7480 ReleaseVolumeHeader(register struct volHeader *hd)
7482 if (programType != fileServer)
7484 if (!hd || queue_IsOnQueue(hd)) /* no header, or header already released */
7486 queue_Append(&volume_hdr_LRU, hd);
7487 #ifdef AFS_DEMAND_ATTACH_FS
7489 V_attachFlags(hd->back) |= VOL_HDR_IN_LRU;
7492 volume_hdr_LRU.stats.free++;
7493 volume_hdr_LRU.stats.used--;
7497 * free/invalidate a volume header cache entry.
7499 * @param[in] vp pointer to volume object
7501 * @pre VOL_LOCK is held.
7503 * @post For fileserver, header cache entry is returned to LRU, and it is
7504 * invalidated as a cache entry. For volume utilities, the header
7505 * cache entry is freed.
7507 * @note For fileserver, this should be utilized instead of ReleaseVolumeHeader
7508 * whenever it is necessary to invalidate the header cache entry.
7510 * @see ReleaseVolumeHeader
7512 * @internal volume package internal use only.
7515 FreeVolumeHeader(register Volume * vp)
7517 register struct volHeader *hd = vp->header;
7520 if (programType == fileServer) {
7521 ReleaseVolumeHeader(hd);
7526 #ifdef AFS_DEMAND_ATTACH_FS
7527 V_attachFlags(vp) &= ~(VOL_HDR_ATTACHED | VOL_HDR_IN_LRU | VOL_HDR_LOADED);
7529 volume_hdr_LRU.stats.attached--;
7534 /***************************************************/
7535 /* Volume Hash Table routines */
7536 /***************************************************/
7539 * set size of volume object hash table.
7541 * @param[in] logsize log(2) of desired hash table size
7543 * @return operation status
7545 * @retval -1 failure
7547 * @pre MUST be called prior to VInitVolumePackage2
7549 * @post Volume Hash Table will have 2^logsize buckets
7552 VSetVolHashSize(int logsize)
7554 /* 64 to 16384 hash buckets seems like a reasonable range */
7555 if ((logsize < 6 ) || (logsize > 14)) {
7560 VolumeHashTable.Size = 1 << logsize;
7561 VolumeHashTable.Mask = VolumeHashTable.Size - 1;
7563 /* we can't yet support runtime modification of this
7564 * parameter. we'll need a configuration rwlock to
7565 * make runtime modification feasible.... */
7572 * initialize dynamic data structures for volume hash table.
7574 * @post hash table is allocated, and fields are initialized.
7576 * @internal volume package internal use only.
7579 VInitVolumeHash(void)
7583 VolumeHashTable.Table = (VolumeHashChainHead *) calloc(VolumeHashTable.Size,
7584 sizeof(VolumeHashChainHead));
7585 assert(VolumeHashTable.Table != NULL);
7587 for (i=0; i < VolumeHashTable.Size; i++) {
7588 queue_Init(&VolumeHashTable.Table[i]);
7589 #ifdef AFS_DEMAND_ATTACH_FS
7590 assert(pthread_cond_init(&VolumeHashTable.Table[i].chain_busy_cv, NULL) == 0);
7591 #endif /* AFS_DEMAND_ATTACH_FS */
7596 * add a volume object to the hash table.
7598 * @param[in] vp pointer to volume object
7599 * @param[in] hashid hash of volume id
7601 * @pre VOL_LOCK is held. For DAFS, caller must hold a lightweight
7604 * @post volume is added to hash chain.
7606 * @internal volume package internal use only.
7608 * @note For DAFS, VOL_LOCK may be dropped in order to wait for an
7609 * asynchronous hash chain reordering to finish.
7612 AddVolumeToHashTable(register Volume * vp, int hashid)
7614 VolumeHashChainHead * head;
7616 if (queue_IsOnQueue(vp))
7619 head = &VolumeHashTable.Table[VOLUME_HASH(hashid)];
7621 #ifdef AFS_DEMAND_ATTACH_FS
7622 /* wait for the hash chain to become available */
7625 V_attachFlags(vp) |= VOL_IN_HASH;
7626 vp->chainCacheCheck = ++head->cacheCheck;
7627 #endif /* AFS_DEMAND_ATTACH_FS */
7630 vp->hashid = hashid;
7631 queue_Append(head, vp);
7632 vp->vnodeHashOffset = VolumeHashOffset_r();
7636 * delete a volume object from the hash table.
7638 * @param[in] vp pointer to volume object
7640 * @pre VOL_LOCK is held. For DAFS, caller must hold a lightweight
7643 * @post volume is removed from hash chain.
7645 * @internal volume package internal use only.
7647 * @note For DAFS, VOL_LOCK may be dropped in order to wait for an
7648 * asynchronous hash chain reordering to finish.
7651 DeleteVolumeFromHashTable(register Volume * vp)
7653 VolumeHashChainHead * head;
7655 if (!queue_IsOnQueue(vp))
7658 head = &VolumeHashTable.Table[VOLUME_HASH(vp->hashid)];
7660 #ifdef AFS_DEMAND_ATTACH_FS
7661 /* wait for the hash chain to become available */
7664 V_attachFlags(vp) &= ~(VOL_IN_HASH);
7666 #endif /* AFS_DEMAND_ATTACH_FS */
7670 /* do NOT reset hashid to zero, as the online
7671 * salvager package may need to know the volume id
7672 * after the volume is removed from the hash */
7676 * lookup a volume object in the hash table given a volume id.
7678 * @param[out] ec error code return
7679 * @param[in] volumeId volume id
7680 * @param[in] hint volume object which we believe could be the correct
7683 * @return volume object pointer
7684 * @retval NULL no such volume id is registered with the hash table.
7686 * @pre VOL_LOCK is held. For DAFS, caller must hold a lightweight
7689 * @post volume object with the given id is returned. volume object and
7690 * hash chain access statistics are updated. hash chain may have
7693 * @note For DAFS, VOL_LOCK may be dropped in order to wait for an
7694 * asynchronous hash chain reordering operation to finish, or
7695 * in order for us to perform an asynchronous chain reordering.
7697 * @note Hash chain reorderings occur when the access count for the
7698 * volume object being looked up exceeds the sum of the previous
7699 * node's (the node ahead of it in the hash chain linked list)
7700 * access count plus the constant VOLUME_HASH_REORDER_THRESHOLD.
7702 * @note For DAFS, the hint parameter allows us to short-circuit if the
7703 * cacheCheck fields match between the hash chain head and the
7704 * hint volume object.
7707 VLookupVolume_r(Error * ec, VolId volumeId, Volume * hint)
7709 register int looks = 0;
7711 #ifdef AFS_DEMAND_ATTACH_FS
7714 VolumeHashChainHead * head;
7717 head = &VolumeHashTable.Table[VOLUME_HASH(volumeId)];
7719 #ifdef AFS_DEMAND_ATTACH_FS
7720 /* wait for the hash chain to become available */
7723 /* check to see if we can short circuit without walking the hash chain */
7724 if (hint && (hint->chainCacheCheck == head->cacheCheck)) {
7725 IncUInt64(&hint->stats.hash_short_circuits);
7728 #endif /* AFS_DEMAND_ATTACH_FS */
7730 /* someday we need to either do per-chain locks, RWlocks,
7731 * or both for volhash access.
7732 * (and move to a data structure with better cache locality) */
7734 /* search the chain for this volume id */
7735 for(queue_Scan(head, vp, np, Volume)) {
7737 if ((vp->hashid == volumeId)) {
7742 if (queue_IsEnd(head, vp)) {
7746 #ifdef AFS_DEMAND_ATTACH_FS
7747 /* update hash chain statistics */
7750 FillInt64(lks, 0, looks);
7751 AddUInt64(head->looks, lks, &head->looks);
7752 AddUInt64(VStats.hash_looks, lks, &VStats.hash_looks);
7753 IncUInt64(&head->gets);
7758 IncUInt64(&vp->stats.hash_lookups);
7760 /* for demand attach fileserver, we permit occasional hash chain reordering
7761 * so that frequently looked up volumes move towards the head of the chain */
7762 pp = queue_Prev(vp, Volume);
7763 if (!queue_IsEnd(head, pp)) {
7764 FillInt64(thresh, 0, VOLUME_HASH_REORDER_THRESHOLD);
7765 AddUInt64(thresh, pp->stats.hash_lookups, &thresh);
7766 if (GEInt64(vp->stats.hash_lookups, thresh)) {
7767 VReorderHash_r(head, pp, vp);
7771 /* update the short-circuit cache check */
7772 vp->chainCacheCheck = head->cacheCheck;
7774 #endif /* AFS_DEMAND_ATTACH_FS */
7779 #ifdef AFS_DEMAND_ATTACH_FS
7780 /* perform volume hash chain reordering.
7782 * advance a subchain beginning at vp ahead of
7783 * the adjacent subchain ending at pp */
7785 VReorderHash_r(VolumeHashChainHead * head, Volume * pp, Volume * vp)
7787 Volume *tp, *np, *lp;
7788 afs_uint64 move_thresh;
7790 /* this should never be called if the chain is already busy, so
7791 * no need to wait for other exclusive chain ops to finish */
7793 /* this is a rather heavy set of operations,
7794 * so let's set the chain busy flag and drop
7796 VHashBeginExclusive_r(head);
7799 /* scan forward in the chain from vp looking for the last element
7800 * in the chain we want to advance */
7801 FillInt64(move_thresh, 0, VOLUME_HASH_REORDER_CHAIN_THRESH);
7802 AddUInt64(move_thresh, pp->stats.hash_lookups, &move_thresh);
7803 for(queue_ScanFrom(head, vp, tp, np, Volume)) {
7804 if (LTInt64(tp->stats.hash_lookups, move_thresh)) {
7808 lp = queue_Prev(tp, Volume);
7810 /* scan backwards from pp to determine where to splice and
7811 * insert the subchain we're advancing */
7812 for(queue_ScanBackwardsFrom(head, pp, tp, np, Volume)) {
7813 if (GTInt64(tp->stats.hash_lookups, move_thresh)) {
7817 tp = queue_Next(tp, Volume);
7819 /* rebalance chain(vp,...,lp) ahead of chain(tp,...,pp) */
7820 queue_MoveChainBefore(tp,vp,lp);
7823 IncUInt64(&VStats.hash_reorders);
7825 IncUInt64(&head->reorders);
7827 /* wake up any threads waiting for the hash chain */
7828 VHashEndExclusive_r(head);
7832 /* demand-attach fs volume hash
7833 * asynchronous exclusive operations */
7836 * begin an asynchronous exclusive operation on a volume hash chain.
7838 * @param[in] head pointer to volume hash chain head object
7840 * @pre VOL_LOCK held. hash chain is quiescent.
7842 * @post hash chain marked busy.
7844 * @note this interface is used in conjunction with VHashEndExclusive_r and
7845 * VHashWait_r to perform asynchronous (wrt VOL_LOCK) operations on a
7846 * volume hash chain. Its main use case is hash chain reordering, which
7847 * has the potential to be a highly latent operation.
7849 * @see VHashEndExclusive_r
7854 * @internal volume package internal use only.
7857 VHashBeginExclusive_r(VolumeHashChainHead * head)
7859 assert(head->busy == 0);
7864 * relinquish exclusive ownership of a volume hash chain.
7866 * @param[in] head pointer to volume hash chain head object
7868 * @pre VOL_LOCK held. thread owns the hash chain exclusively.
7870 * @post hash chain is marked quiescent. threads awaiting use of
7871 * chain are awakened.
7873 * @see VHashBeginExclusive_r
7878 * @internal volume package internal use only.
7881 VHashEndExclusive_r(VolumeHashChainHead * head)
7885 assert(pthread_cond_broadcast(&head->chain_busy_cv) == 0);
7889 * wait for all asynchronous operations on a hash chain to complete.
7891 * @param[in] head pointer to volume hash chain head object
7893 * @pre VOL_LOCK held.
7895 * @post hash chain object is quiescent.
7897 * @see VHashBeginExclusive_r
7898 * @see VHashEndExclusive_r
7902 * @note This interface should be called before any attempt to
7903 * traverse the hash chain. It is permissible for a thread
7904 * to gain exclusive access to the chain, and then perform
7905 * latent operations on the chain asynchronously wrt the
7908 * @warning if waiting is necessary, VOL_LOCK is dropped
7910 * @internal volume package internal use only.
7913 VHashWait_r(VolumeHashChainHead * head)
7915 while (head->busy) {
7916 VOL_CV_WAIT(&head->chain_busy_cv);
7919 #endif /* AFS_DEMAND_ATTACH_FS */
7922 /***************************************************/
7923 /* Volume by Partition List routines */
7924 /***************************************************/
7927 * demand attach fileserver adds a
7928 * linked list of volumes to each
7929 * partition object, thus allowing
7930 * for quick enumeration of all
7931 * volumes on a partition
7934 #ifdef AFS_DEMAND_ATTACH_FS
7936 * add a volume to its disk partition VByPList.
7938 * @param[in] vp pointer to volume object
7940 * @pre either the disk partition VByPList is owned exclusively
7941 * by the calling thread, or the list is quiescent and
7944 * @post volume is added to disk partition VByPList
7948 * @warning it is the caller's responsibility to ensure list
7951 * @see VVByPListWait_r
7952 * @see VVByPListBeginExclusive_r
7953 * @see VVByPListEndExclusive_r
7955 * @internal volume package internal use only.
7958 AddVolumeToVByPList_r(Volume * vp)
7960 if (queue_IsNotOnQueue(&vp->vol_list)) {
7961 queue_Append(&vp->partition->vol_list, &vp->vol_list);
7962 V_attachFlags(vp) |= VOL_ON_VBYP_LIST;
7963 vp->partition->vol_list.len++;
7968 * delete a volume from its disk partition VByPList.
7970 * @param[in] vp pointer to volume object
7972 * @pre either the disk partition VByPList is owned exclusively
7973 * by the calling thread, or the list is quiescent and
7976 * @post volume is removed from the disk partition VByPList
7980 * @warning it is the caller's responsibility to ensure list
7983 * @see VVByPListWait_r
7984 * @see VVByPListBeginExclusive_r
7985 * @see VVByPListEndExclusive_r
7987 * @internal volume package internal use only.
7990 DeleteVolumeFromVByPList_r(Volume * vp)
7992 if (queue_IsOnQueue(&vp->vol_list)) {
7993 queue_Remove(&vp->vol_list);
7994 V_attachFlags(vp) &= ~(VOL_ON_VBYP_LIST);
7995 vp->partition->vol_list.len--;
8000 * begin an asynchronous exclusive operation on a VByPList.
8002 * @param[in] dp pointer to disk partition object
8004 * @pre VOL_LOCK held. VByPList is quiescent.
8006 * @post VByPList marked busy.
8008 * @note this interface is used in conjunction with VVByPListEndExclusive_r and
8009 * VVByPListWait_r to perform asynchronous (wrt VOL_LOCK) operations on a
8012 * @see VVByPListEndExclusive_r
8013 * @see VVByPListWait_r
8017 * @internal volume package internal use only.
8019 /* take exclusive control over the list */
8021 VVByPListBeginExclusive_r(struct DiskPartition64 * dp)
8023 assert(dp->vol_list.busy == 0);
8024 dp->vol_list.busy = 1;
8028 * relinquish exclusive ownership of a VByPList.
8030 * @param[in] dp pointer to disk partition object
8032 * @pre VOL_LOCK held. thread owns the VByPList exclusively.
8034 * @post VByPList is marked quiescent. threads awaiting use of
8035 * the list are awakened.
8037 * @see VVByPListBeginExclusive_r
8038 * @see VVByPListWait_r
8042 * @internal volume package internal use only.
8045 VVByPListEndExclusive_r(struct DiskPartition64 * dp)
8047 assert(dp->vol_list.busy);
8048 dp->vol_list.busy = 0;
8049 assert(pthread_cond_broadcast(&dp->vol_list.cv) == 0);
8053 * wait for all asynchronous operations on a VByPList to complete.
8055 * @param[in] dp pointer to disk partition object
8057 * @pre VOL_LOCK is held.
8059 * @post disk partition's VByP list is quiescent
8063 * @note This interface should be called before any attempt to
8064 * traverse the VByPList. It is permissible for a thread
8065 * to gain exclusive access to the list, and then perform
8066 * latent operations on the list asynchronously wrt the
8069 * @warning if waiting is necessary, VOL_LOCK is dropped
8071 * @see VVByPListEndExclusive_r
8072 * @see VVByPListBeginExclusive_r
8074 * @internal volume package internal use only.
8077 VVByPListWait_r(struct DiskPartition64 * dp)
8079 while (dp->vol_list.busy) {
8080 VOL_CV_WAIT(&dp->vol_list.cv);
8083 #endif /* AFS_DEMAND_ATTACH_FS */
8085 /***************************************************/
8086 /* Volume Cache Statistics routines */
8087 /***************************************************/
8090 VPrintCacheStats_r(void)
8092 afs_uint32 get_hi, get_lo, load_hi, load_lo;
8093 register struct VnodeClassInfo *vcp;
8094 vcp = &VnodeClassInfo[vLarge];
8095 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);
8096 vcp = &VnodeClassInfo[vSmall];
8097 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);
8098 SplitInt64(VStats.hdr_gets, get_hi, get_lo);
8099 SplitInt64(VStats.hdr_loads, load_hi, load_lo);
8100 Log("Volume header cache, %d entries, %d gets, %d replacements\n",
8101 VStats.hdr_cache_size, get_lo, load_lo);
8105 VPrintCacheStats(void)
8108 VPrintCacheStats_r();
8112 #ifdef AFS_DEMAND_ATTACH_FS
8114 UInt64ToDouble(afs_uint64 * x)
8116 static double c32 = 4.0 * 1.073741824 * 1000000000.0;
8118 SplitInt64(*x, h, l);
8119 return (((double)h) * c32) + ((double) l);
8123 DoubleToPrintable(double x, char * buf, int len)
8125 static double billion = 1000000000.0;
8128 y[0] = (afs_uint32) (x / (billion * billion));
8129 y[1] = (afs_uint32) ((x - (((double)y[0]) * billion * billion)) / billion);
8130 y[2] = (afs_uint32) (x - ((((double)y[0]) * billion * billion) + (((double)y[1]) * billion)));
8133 snprintf(buf, len, "%d%09d%09d", y[0], y[1], y[2]);
8135 snprintf(buf, len, "%d%09d", y[1], y[2]);
8137 snprintf(buf, len, "%d", y[2]);
8143 struct VLRUExtStatsEntry {
8147 struct VLRUExtStats {
8153 } queue_info[VLRU_QUEUE_INVALID];
8154 struct VLRUExtStatsEntry * vec;
8158 * add a 256-entry fudge factor onto the vector in case state changes
8159 * out from under us.
8161 #define VLRU_EXT_STATS_VEC_LEN_FUDGE 256
8164 * collect extended statistics for the VLRU subsystem.
8166 * @param[out] stats pointer to stats structure to be populated
8167 * @param[in] nvols number of volumes currently known to exist
8169 * @pre VOL_LOCK held
8171 * @post stats->vec allocated and populated
8173 * @return operation status
8178 VVLRUExtStats_r(struct VLRUExtStats * stats, afs_uint32 nvols)
8180 afs_uint32 cur, idx, len;
8181 struct rx_queue * qp, * nqp;
8183 struct VLRUExtStatsEntry * vec;
8185 len = nvols + VLRU_EXT_STATS_VEC_LEN_FUDGE;
8186 vec = stats->vec = calloc(len,
8187 sizeof(struct VLRUExtStatsEntry));
8193 for (idx = VLRU_QUEUE_NEW; idx < VLRU_QUEUE_INVALID; idx++) {
8194 VLRU_Wait_r(&volume_LRU.q[idx]);
8195 VLRU_BeginExclusive_r(&volume_LRU.q[idx]);
8198 stats->queue_info[idx].start = cur;
8200 for (queue_Scan(&volume_LRU.q[idx], qp, nqp, rx_queue)) {
8202 /* out of space in vec */
8205 vp = (Volume *)((char *)qp - offsetof(Volume, vlru));
8206 vec[cur].volid = vp->hashid;
8210 stats->queue_info[idx].len = cur - stats->queue_info[idx].start;
8213 VLRU_EndExclusive_r(&volume_LRU.q[idx]);
8221 #define ENUMTOSTRING(en) #en
8222 #define ENUMCASE(en) \
8224 return ENUMTOSTRING(en); \
8228 vlru_idx_to_string(int idx)
8231 ENUMCASE(VLRU_QUEUE_NEW);
8232 ENUMCASE(VLRU_QUEUE_MID);
8233 ENUMCASE(VLRU_QUEUE_OLD);
8234 ENUMCASE(VLRU_QUEUE_CANDIDATE);
8235 ENUMCASE(VLRU_QUEUE_HELD);
8236 ENUMCASE(VLRU_QUEUE_INVALID);
8238 return "**UNKNOWN**";
8243 VPrintExtendedCacheStats_r(int flags)
8246 afs_uint32 vol_sum = 0;
8253 struct stats looks, gets, reorders, len;
8254 struct stats ch_looks, ch_gets, ch_reorders;
8256 VolumeHashChainHead *head;
8258 struct VLRUExtStats vlru_stats;
8260 /* zero out stats */
8261 memset(&looks, 0, sizeof(struct stats));
8262 memset(&gets, 0, sizeof(struct stats));
8263 memset(&reorders, 0, sizeof(struct stats));
8264 memset(&len, 0, sizeof(struct stats));
8265 memset(&ch_looks, 0, sizeof(struct stats));
8266 memset(&ch_gets, 0, sizeof(struct stats));
8267 memset(&ch_reorders, 0, sizeof(struct stats));
8269 for (i = 0; i < VolumeHashTable.Size; i++) {
8270 head = &VolumeHashTable.Table[i];
8273 VHashBeginExclusive_r(head);
8276 ch_looks.sum = UInt64ToDouble(&head->looks);
8277 ch_gets.sum = UInt64ToDouble(&head->gets);
8278 ch_reorders.sum = UInt64ToDouble(&head->reorders);
8280 /* update global statistics */
8282 looks.sum += ch_looks.sum;
8283 gets.sum += ch_gets.sum;
8284 reorders.sum += ch_reorders.sum;
8285 len.sum += (double)head->len;
8286 vol_sum += head->len;
8289 len.min = (double) head->len;
8290 len.max = (double) head->len;
8291 looks.min = ch_looks.sum;
8292 looks.max = ch_looks.sum;
8293 gets.min = ch_gets.sum;
8294 gets.max = ch_gets.sum;
8295 reorders.min = ch_reorders.sum;
8296 reorders.max = ch_reorders.sum;
8298 if (((double)head->len) < len.min)
8299 len.min = (double) head->len;
8300 if (((double)head->len) > len.max)
8301 len.max = (double) head->len;
8302 if (ch_looks.sum < looks.min)
8303 looks.min = ch_looks.sum;
8304 else if (ch_looks.sum > looks.max)
8305 looks.max = ch_looks.sum;
8306 if (ch_gets.sum < gets.min)
8307 gets.min = ch_gets.sum;
8308 else if (ch_gets.sum > gets.max)
8309 gets.max = ch_gets.sum;
8310 if (ch_reorders.sum < reorders.min)
8311 reorders.min = ch_reorders.sum;
8312 else if (ch_reorders.sum > reorders.max)
8313 reorders.max = ch_reorders.sum;
8317 if ((flags & VOL_STATS_PER_CHAIN2) && queue_IsNotEmpty(head)) {
8318 /* compute detailed per-chain stats */
8319 struct stats hdr_loads, hdr_gets;
8320 double v_looks, v_loads, v_gets;
8322 /* initialize stats with data from first element in chain */
8323 vp = queue_First(head, Volume);
8324 v_looks = UInt64ToDouble(&vp->stats.hash_lookups);
8325 v_loads = UInt64ToDouble(&vp->stats.hdr_loads);
8326 v_gets = UInt64ToDouble(&vp->stats.hdr_gets);
8327 ch_gets.min = ch_gets.max = v_looks;
8328 hdr_loads.min = hdr_loads.max = v_loads;
8329 hdr_gets.min = hdr_gets.max = v_gets;
8330 hdr_loads.sum = hdr_gets.sum = 0;
8332 vp = queue_Next(vp, Volume);
8334 /* pull in stats from remaining elements in chain */
8335 for (queue_ScanFrom(head, vp, vp, np, Volume)) {
8336 v_looks = UInt64ToDouble(&vp->stats.hash_lookups);
8337 v_loads = UInt64ToDouble(&vp->stats.hdr_loads);
8338 v_gets = UInt64ToDouble(&vp->stats.hdr_gets);
8340 hdr_loads.sum += v_loads;
8341 hdr_gets.sum += v_gets;
8343 if (v_looks < ch_gets.min)
8344 ch_gets.min = v_looks;
8345 else if (v_looks > ch_gets.max)
8346 ch_gets.max = v_looks;
8348 if (v_loads < hdr_loads.min)
8349 hdr_loads.min = v_loads;
8350 else if (v_loads > hdr_loads.max)
8351 hdr_loads.max = v_loads;
8353 if (v_gets < hdr_gets.min)
8354 hdr_gets.min = v_gets;
8355 else if (v_gets > hdr_gets.max)
8356 hdr_gets.max = v_gets;
8359 /* compute per-chain averages */
8360 ch_gets.avg = ch_gets.sum / ((double)head->len);
8361 hdr_loads.avg = hdr_loads.sum / ((double)head->len);
8362 hdr_gets.avg = hdr_gets.sum / ((double)head->len);
8364 /* dump per-chain stats */
8365 Log("Volume hash chain %d : len=%d, looks=%s, reorders=%s\n",
8367 DoubleToPrintable(ch_looks.sum, pr_buf[0], sizeof(pr_buf[0])),
8368 DoubleToPrintable(ch_reorders.sum, pr_buf[1], sizeof(pr_buf[1])));
8369 Log("\tVolume gets : min=%s, max=%s, avg=%s, total=%s\n",
8370 DoubleToPrintable(ch_gets.min, pr_buf[0], sizeof(pr_buf[0])),
8371 DoubleToPrintable(ch_gets.max, pr_buf[1], sizeof(pr_buf[1])),
8372 DoubleToPrintable(ch_gets.avg, pr_buf[2], sizeof(pr_buf[2])),
8373 DoubleToPrintable(ch_gets.sum, pr_buf[3], sizeof(pr_buf[3])));
8374 Log("\tHDR gets : min=%s, max=%s, avg=%s, total=%s\n",
8375 DoubleToPrintable(hdr_gets.min, pr_buf[0], sizeof(pr_buf[0])),
8376 DoubleToPrintable(hdr_gets.max, pr_buf[1], sizeof(pr_buf[1])),
8377 DoubleToPrintable(hdr_gets.avg, pr_buf[2], sizeof(pr_buf[2])),
8378 DoubleToPrintable(hdr_gets.sum, pr_buf[3], sizeof(pr_buf[3])));
8379 Log("\tHDR loads : min=%s, max=%s, avg=%s, total=%s\n",
8380 DoubleToPrintable(hdr_loads.min, pr_buf[0], sizeof(pr_buf[0])),
8381 DoubleToPrintable(hdr_loads.max, pr_buf[1], sizeof(pr_buf[1])),
8382 DoubleToPrintable(hdr_loads.avg, pr_buf[2], sizeof(pr_buf[2])),
8383 DoubleToPrintable(hdr_loads.sum, pr_buf[3], sizeof(pr_buf[3])));
8384 } else if (flags & VOL_STATS_PER_CHAIN) {
8385 /* dump simple per-chain stats */
8386 Log("Volume hash chain %d : len=%d, looks=%s, gets=%s, reorders=%s\n",
8388 DoubleToPrintable(ch_looks.sum, pr_buf[0], sizeof(pr_buf[0])),
8389 DoubleToPrintable(ch_gets.sum, pr_buf[1], sizeof(pr_buf[1])),
8390 DoubleToPrintable(ch_reorders.sum, pr_buf[2], sizeof(pr_buf[2])));
8394 VHashEndExclusive_r(head);
8399 /* compute global averages */
8400 len.avg = len.sum / ((double)VolumeHashTable.Size);
8401 looks.avg = looks.sum / ((double)VolumeHashTable.Size);
8402 gets.avg = gets.sum / ((double)VolumeHashTable.Size);
8403 reorders.avg = reorders.sum / ((double)VolumeHashTable.Size);
8405 /* dump global stats */
8406 Log("Volume hash summary: %d buckets\n", VolumeHashTable.Size);
8407 Log(" chain length : min=%s, max=%s, avg=%s, total=%s\n",
8408 DoubleToPrintable(len.min, pr_buf[0], sizeof(pr_buf[0])),
8409 DoubleToPrintable(len.max, pr_buf[1], sizeof(pr_buf[1])),
8410 DoubleToPrintable(len.avg, pr_buf[2], sizeof(pr_buf[2])),
8411 DoubleToPrintable(len.sum, pr_buf[3], sizeof(pr_buf[3])));
8412 Log(" looks : min=%s, max=%s, avg=%s, total=%s\n",
8413 DoubleToPrintable(looks.min, pr_buf[0], sizeof(pr_buf[0])),
8414 DoubleToPrintable(looks.max, pr_buf[1], sizeof(pr_buf[1])),
8415 DoubleToPrintable(looks.avg, pr_buf[2], sizeof(pr_buf[2])),
8416 DoubleToPrintable(looks.sum, pr_buf[3], sizeof(pr_buf[3])));
8417 Log(" gets : min=%s, max=%s, avg=%s, total=%s\n",
8418 DoubleToPrintable(gets.min, pr_buf[0], sizeof(pr_buf[0])),
8419 DoubleToPrintable(gets.max, pr_buf[1], sizeof(pr_buf[1])),
8420 DoubleToPrintable(gets.avg, pr_buf[2], sizeof(pr_buf[2])),
8421 DoubleToPrintable(gets.sum, pr_buf[3], sizeof(pr_buf[3])));
8422 Log(" reorders : min=%s, max=%s, avg=%s, total=%s\n",
8423 DoubleToPrintable(reorders.min, pr_buf[0], sizeof(pr_buf[0])),
8424 DoubleToPrintable(reorders.max, pr_buf[1], sizeof(pr_buf[1])),
8425 DoubleToPrintable(reorders.avg, pr_buf[2], sizeof(pr_buf[2])),
8426 DoubleToPrintable(reorders.sum, pr_buf[3], sizeof(pr_buf[3])));
8428 /* print extended disk related statistics */
8430 struct DiskPartition64 * diskP;
8431 afs_uint32 vol_count[VOLMAXPARTS+1];
8432 byte part_exists[VOLMAXPARTS+1];
8436 memset(vol_count, 0, sizeof(vol_count));
8437 memset(part_exists, 0, sizeof(part_exists));
8441 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
8443 vol_count[id] = diskP->vol_list.len;
8444 part_exists[id] = 1;
8448 for (i = 0; i <= VOLMAXPARTS; i++) {
8449 if (part_exists[i]) {
8450 /* XXX while this is currently safe, it is a violation
8451 * of the VGetPartitionById_r interface contract. */
8452 diskP = VGetPartitionById_r(i, 0);
8454 Log("Partition %s has %d online volumes\n",
8455 VPartitionPath(diskP), diskP->vol_list.len);
8462 /* print extended VLRU statistics */
8463 if (VVLRUExtStats_r(&vlru_stats, vol_sum) == 0) {
8464 afs_uint32 idx, cur, lpos;
8469 Log("VLRU State Dump:\n\n");
8471 for (idx = VLRU_QUEUE_NEW; idx < VLRU_QUEUE_INVALID; idx++) {
8472 Log("\t%s:\n", vlru_idx_to_string(idx));
8475 for (cur = vlru_stats.queue_info[idx].start;
8476 cur < vlru_stats.queue_info[idx].len;
8478 line[lpos++] = vlru_stats.vec[cur].volid;
8480 Log("\t\t%u, %u, %u, %u, %u,\n",
8481 line[0], line[1], line[2], line[3], line[4]);
8490 Log("\t\t%u, %u, %u, %u, %u\n",
8491 line[0], line[1], line[2], line[3], line[4]);
8496 free(vlru_stats.vec);
8503 VPrintExtendedCacheStats(int flags)
8506 VPrintExtendedCacheStats_r(flags);
8509 #endif /* AFS_DEMAND_ATTACH_FS */
8512 VCanScheduleSalvage(void)
8514 return vol_opts.canScheduleSalvage;
8520 return vol_opts.canUseFSSYNC;
8524 VCanUseSALVSYNC(void)
8526 return vol_opts.canUseSALVSYNC;
8530 VCanUnsafeAttach(void)
8532 return vol_opts.unsafe_attach;