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>
30 #ifdef HAVE_SYS_FILE_H
35 #include <afs/afsint.h>
38 #if !defined(AFS_SGI_ENV)
41 #else /* AFS_OSF_ENV */
42 #ifdef AFS_VFSINCL_ENV
45 #include <sys/fs/ufs_fs.h>
47 #if defined(AFS_DARWIN_ENV) || defined(AFS_XBSD_ENV)
48 #include <ufs/ufs/dinode.h>
49 #include <ufs/ffs/fs.h>
54 #else /* AFS_VFSINCL_ENV */
55 #if !defined(AFS_AIX_ENV) && !defined(AFS_LINUX20_ENV) && !defined(AFS_XBSD_ENV) && !defined(AFS_DARWIN_ENV)
58 #endif /* AFS_VFSINCL_ENV */
59 #endif /* AFS_OSF_ENV */
60 #endif /* AFS_SGI_ENV */
61 #endif /* !AFS_NT40_ENV */
69 #if defined(AFS_SUN_ENV) || defined(AFS_SUN5_ENV)
71 #include <sys/mnttab.h>
72 #include <sys/mntent.h>
78 #if defined(AFS_SGI_ENV)
81 #ifndef AFS_LINUX20_ENV
82 #include <fstab.h> /* Need to find in libc 5, present in libc 6 */
85 #endif /* AFS_SGI_ENV */
87 #endif /* AFS_HPUX_ENV */
91 #include <afs/errors.h>
94 #include <afs/afssyscalls.h>
96 #include <afs/afsutil.h>
97 #include "daemon_com.h"
102 #include "partition.h"
103 #include "volume_inline.h"
108 #ifdef AFS_PTHREAD_ENV
109 pthread_mutex_t vol_glock_mutex;
110 pthread_mutex_t vol_trans_mutex;
111 pthread_cond_t vol_put_volume_cond;
112 pthread_cond_t vol_sleep_cond;
113 pthread_cond_t vol_init_attach_cond;
114 pthread_cond_t vol_vinit_cond;
115 int vol_attach_threads = 1;
116 #endif /* AFS_PTHREAD_ENV */
118 /* start-time configurable I/O parameters */
119 ih_init_params vol_io_params;
121 #ifdef AFS_DEMAND_ATTACH_FS
122 pthread_mutex_t vol_salvsync_mutex;
125 * Set this to 1 to disallow SALVSYNC communication in all threads; used
126 * during shutdown, since the salvageserver may have gone away.
128 static volatile sig_atomic_t vol_disallow_salvsync = 0;
129 #endif /* AFS_DEMAND_ATTACH_FS */
132 * has VShutdown_r been called / is VShutdown_r running?
134 static int vol_shutting_down = 0;
137 extern void *calloc(), *realloc();
140 /* Forward declarations */
141 static Volume *attach2(Error * ec, VolId volumeId, char *path,
142 struct DiskPartition64 *partp, Volume * vp,
143 int isbusy, int mode, int *acheckedOut);
144 static void ReallyFreeVolume(Volume * vp);
145 #ifdef AFS_DEMAND_ATTACH_FS
146 static void FreeVolume(Volume * vp);
147 #else /* !AFS_DEMAND_ATTACH_FS */
148 #define FreeVolume(vp) ReallyFreeVolume(vp)
149 static void VScanUpdateList(void);
150 #endif /* !AFS_DEMAND_ATTACH_FS */
151 static void VInitVolumeHeaderCache(afs_uint32 howMany);
152 static int GetVolumeHeader(Volume * vp);
153 static void ReleaseVolumeHeader(struct volHeader *hd);
154 static void FreeVolumeHeader(Volume * vp);
155 static void AddVolumeToHashTable(Volume * vp, int hashid);
156 static void DeleteVolumeFromHashTable(Volume * vp);
158 static int VHold(Volume * vp);
160 static int VHold_r(Volume * vp);
161 static void VGetBitmap_r(Error * ec, Volume * vp, VnodeClass class);
162 static void VReleaseVolumeHandles_r(Volume * vp);
163 static void VCloseVolumeHandles_r(Volume * vp);
164 static void LoadVolumeHeader(Error * ec, Volume * vp);
165 static int VCheckOffline(Volume * vp);
166 static int VCheckDetach(Volume * vp);
167 static Volume * GetVolume(Error * ec, Error * client_ec, VolId volumeId,
168 Volume * hint, const struct timespec *ts);
170 int LogLevel; /* Vice loglevel--not defined as extern so that it will be
171 * defined when not linked with vice, XXXX */
172 ProgramType programType; /* The type of program using the package */
173 static VolumePackageOptions vol_opts;
175 /* extended volume package statistics */
178 #ifdef VOL_LOCK_DEBUG
179 pthread_t vol_glock_holder = 0;
183 #define VOLUME_BITMAP_GROWSIZE 16 /* bytes, => 128vnodes */
184 /* Must be a multiple of 4 (1 word) !! */
186 /* this parameter needs to be tunable at runtime.
187 * 128 was really inadequate for largish servers -- at 16384 volumes this
188 * puts average chain length at 128, thus an average 65 deref's to find a volptr.
189 * talk about bad spatial locality...
191 * an AVL or splay tree might work a lot better, but we'll just increase
192 * the default hash table size for now
194 #define DEFAULT_VOLUME_HASH_SIZE 256 /* Must be a power of 2!! */
195 #define DEFAULT_VOLUME_HASH_MASK (DEFAULT_VOLUME_HASH_SIZE-1)
196 #define VOLUME_HASH(volumeId) (volumeId&(VolumeHashTable.Mask))
199 * turn volume hash chains into partially ordered lists.
200 * when the threshold is exceeded between two adjacent elements,
201 * perform a chain rebalancing operation.
203 * keep the threshold high in order to keep cache line invalidates
204 * low "enough" on SMPs
206 #define VOLUME_HASH_REORDER_THRESHOLD 200
209 * when possible, don't just reorder single elements, but reorder
210 * entire chains of elements at once. a chain of elements that
211 * exceed the element previous to the pivot by at least CHAIN_THRESH
212 * accesses are moved in front of the chain whose elements have at
213 * least CHAIN_THRESH less accesses than the pivot element
215 #define VOLUME_HASH_REORDER_CHAIN_THRESH (VOLUME_HASH_REORDER_THRESHOLD / 2)
217 #include "rx/rx_queue.h"
220 VolumeHashTable_t VolumeHashTable = {
221 DEFAULT_VOLUME_HASH_SIZE,
222 DEFAULT_VOLUME_HASH_MASK,
227 static void VInitVolumeHash(void);
231 /* This macro is used where an ffs() call does not exist. Was in util/ffs.c */
235 afs_int32 ffs_tmp = x;
239 for (ffs_i = 1;; ffs_i++) {
246 #endif /* !AFS_HAVE_FFS */
248 #ifdef AFS_PTHREAD_ENV
250 * disk partition queue element
252 typedef struct diskpartition_queue_t {
253 struct rx_queue queue; /**< queue header */
254 struct DiskPartition64 *diskP; /**< disk partition table entry */
255 } diskpartition_queue_t;
257 #ifndef AFS_DEMAND_ATTACH_FS
259 typedef struct vinitvolumepackage_thread_t {
260 struct rx_queue queue;
261 pthread_cond_t thread_done_cv;
262 int n_threads_complete;
263 } vinitvolumepackage_thread_t;
264 static void * VInitVolumePackageThread(void * args);
266 #else /* !AFS_DEMAND_ATTTACH_FS */
267 #define VINIT_BATCH_MAX_SIZE 512
270 * disk partition work queue
272 struct partition_queue {
273 struct rx_queue head; /**< diskpartition_queue_t queue */
274 pthread_mutex_t mutex;
279 * volumes parameters for preattach
281 struct volume_init_batch {
282 struct rx_queue queue; /**< queue header */
283 int thread; /**< posting worker thread */
284 int last; /**< indicates thread is done */
285 int size; /**< number of volume ids in batch */
286 Volume *batch[VINIT_BATCH_MAX_SIZE]; /**< volumes ids to preattach */
290 * volume parameters work queue
292 struct volume_init_queue {
293 struct rx_queue head; /**< volume_init_batch queue */
294 pthread_mutex_t mutex;
299 * volume init worker thread parameters
301 struct vinitvolumepackage_thread_param {
302 int nthreads; /**< total number of worker threads */
303 int thread; /**< thread number for this worker thread */
304 struct partition_queue *pq; /**< queue partitions to scan */
305 struct volume_init_queue *vq; /**< queue of volume to preattach */
308 static void *VInitVolumePackageThread(void *args);
309 static struct DiskPartition64 *VInitNextPartition(struct partition_queue *pq);
310 static VolId VInitNextVolumeId(DIR *dirp);
311 static int VInitPreAttachVolumes(int nthreads, struct volume_init_queue *vq);
313 #endif /* !AFS_DEMAND_ATTACH_FS */
314 #endif /* AFS_PTHREAD_ENV */
316 #ifndef AFS_DEMAND_ATTACH_FS
317 static int VAttachVolumesByPartition(struct DiskPartition64 *diskP,
318 int * nAttached, int * nUnattached);
319 #endif /* AFS_DEMAND_ATTACH_FS */
322 #ifdef AFS_DEMAND_ATTACH_FS
323 /* demand attach fileserver extensions */
326 * in the future we will support serialization of VLRU state into the fs_state
329 * these structures are the beginning of that effort
331 struct VLRU_DiskHeader {
332 struct versionStamp stamp; /* magic and structure version number */
333 afs_uint32 mtime; /* time of dump to disk */
334 afs_uint32 num_records; /* number of VLRU_DiskEntry records */
337 struct VLRU_DiskEntry {
338 afs_uint32 vid; /* volume ID */
339 afs_uint32 idx; /* generation */
340 afs_uint32 last_get; /* timestamp of last get */
343 struct VLRU_StartupQueue {
344 struct VLRU_DiskEntry * entry;
349 typedef struct vshutdown_thread_t {
351 pthread_mutex_t lock;
353 pthread_cond_t master_cv;
355 int n_threads_complete;
357 int schedule_version;
360 byte n_parts_done_pass;
361 byte part_thread_target[VOLMAXPARTS+1];
362 byte part_done_pass[VOLMAXPARTS+1];
363 struct rx_queue * part_pass_head[VOLMAXPARTS+1];
364 int stats[4][VOLMAXPARTS+1];
365 } vshutdown_thread_t;
366 static void * VShutdownThread(void * args);
369 static Volume * VAttachVolumeByVp_r(Error * ec, Volume * vp, int mode);
370 static int VCheckFree(Volume * vp);
373 static void AddVolumeToVByPList_r(Volume * vp);
374 static void DeleteVolumeFromVByPList_r(Volume * vp);
375 static void VVByPListBeginExclusive_r(struct DiskPartition64 * dp);
376 static void VVByPListEndExclusive_r(struct DiskPartition64 * dp);
377 static void VVByPListWait_r(struct DiskPartition64 * dp);
379 /* online salvager */
380 static int VCheckSalvage(Volume * vp);
381 #if defined(SALVSYNC_BUILD_CLIENT) || defined(FSSYNC_BUILD_CLIENT)
382 static int VScheduleSalvage_r(Volume * vp);
385 /* Volume hash table */
386 static void VReorderHash_r(VolumeHashChainHead * head, Volume * pp, Volume * vp);
387 static void VHashBeginExclusive_r(VolumeHashChainHead * head);
388 static void VHashEndExclusive_r(VolumeHashChainHead * head);
389 static void VHashWait_r(VolumeHashChainHead * head);
392 static int ShutdownVByPForPass_r(struct DiskPartition64 * dp, int pass);
393 static int ShutdownVolumeWalk_r(struct DiskPartition64 * dp, int pass,
394 struct rx_queue ** idx);
395 static void ShutdownController(vshutdown_thread_t * params);
396 static void ShutdownCreateSchedule(vshutdown_thread_t * params);
399 static void VLRU_ComputeConstants(void);
400 static void VInitVLRU(void);
401 static void VLRU_Init_Node_r(Volume * vp);
402 static void VLRU_Add_r(Volume * vp);
403 static void VLRU_Delete_r(Volume * vp);
404 static void VLRU_UpdateAccess_r(Volume * vp);
405 static void * VLRU_ScannerThread(void * args);
406 static void VLRU_Scan_r(int idx);
407 static void VLRU_Promote_r(int idx);
408 static void VLRU_Demote_r(int idx);
409 static void VLRU_SwitchQueues(Volume * vp, int new_idx, int append);
412 static int VCheckSoftDetach(Volume * vp, afs_uint32 thresh);
413 static int VCheckSoftDetachCandidate(Volume * vp, afs_uint32 thresh);
414 static int VSoftDetachVolume_r(Volume * vp, afs_uint32 thresh);
417 pthread_key_t VThread_key;
418 VThreadOptions_t VThread_defaults = {
419 0 /**< allow salvsync */
421 #endif /* AFS_DEMAND_ATTACH_FS */
424 struct Lock vol_listLock; /* Lock obtained when listing volumes:
425 * prevents a volume from being missed
426 * if the volume is attached during a
430 /* Common message used when the volume goes off line */
431 char *VSalvageMessage =
432 "Files in this volume are currently unavailable; call operations";
434 int VInit; /* 0 - uninitialized,
435 * 1 - initialized but not all volumes have been attached,
436 * 2 - initialized and all volumes have been attached,
437 * 3 - initialized, all volumes have been attached, and
438 * VConnectFS() has completed. */
440 static int vinit_attach_abort = 0;
442 bit32 VolumeCacheCheck; /* Incremented everytime a volume goes on line--
443 * used to stamp volume headers and in-core
444 * vnodes. When the volume goes on-line the
445 * vnode will be invalidated
446 * access only with VOL_LOCK held */
451 /***************************************************/
452 /* Startup routines */
453 /***************************************************/
455 #if defined(FAST_RESTART) && defined(AFS_DEMAND_ATTACH_FS)
456 # error FAST_RESTART and DAFS are incompatible. For the DAFS equivalent \
457 of FAST_RESTART, use the -unsafe-nosalvage fileserver argument
461 * assign default values to a VolumePackageOptions struct.
463 * Always call this on a VolumePackageOptions struct first, then set any
464 * specific options you want, then call VInitVolumePackage2.
466 * @param[in] pt caller's program type
467 * @param[out] opts volume package options
470 VOptDefaults(ProgramType pt, VolumePackageOptions *opts)
472 opts->nLargeVnodes = opts->nSmallVnodes = 5;
475 opts->canScheduleSalvage = 0;
476 opts->canUseFSSYNC = 0;
477 opts->canUseSALVSYNC = 0;
479 opts->interrupt_rxcall = NULL;
480 opts->offline_timeout = -1;
481 opts->offline_shutdown_timeout = -1;
482 opts->usage_threshold = 128;
483 opts->usage_rate_limit = 5;
486 opts->unsafe_attach = 1;
487 #else /* !FAST_RESTART */
488 opts->unsafe_attach = 0;
489 #endif /* !FAST_RESTART */
493 opts->canScheduleSalvage = 1;
494 opts->canUseSALVSYNC = 1;
498 opts->canUseFSSYNC = 1;
502 opts->nLargeVnodes = 0;
503 opts->nSmallVnodes = 0;
505 opts->canScheduleSalvage = 1;
506 opts->canUseFSSYNC = 1;
516 * Set VInit to a certain value, and signal waiters.
518 * @param[in] value the value to set VInit to
523 VSetVInit_r(int value)
526 CV_BROADCAST(&vol_vinit_cond);
530 VLogOfflineTimeout(const char *type, afs_int32 timeout)
536 Log("VInitVolumePackage: Interrupting clients accessing %s "
537 "immediately\n", type);
539 Log("VInitVolumePackage: Interrupting clients accessing %s "
540 "after %ld second%s\n", type, (long)timeout, timeout==1?"":"s");
545 VInitVolumePackage2(ProgramType pt, VolumePackageOptions * opts)
547 int errors = 0; /* Number of errors while finding vice partitions. */
552 #ifndef AFS_PTHREAD_ENV
553 if (opts->offline_timeout != -1 || opts->offline_shutdown_timeout != -1) {
554 Log("VInitVolumePackage: offline_timeout and/or "
555 "offline_shutdown_timeout was specified, but the volume package "
556 "does not support these for LWP builds\n");
560 VLogOfflineTimeout("volumes going offline", opts->offline_timeout);
561 VLogOfflineTimeout("volumes going offline during shutdown",
562 opts->offline_shutdown_timeout);
564 memset(&VStats, 0, sizeof(VStats));
565 VStats.hdr_cache_size = 200;
567 VInitPartitionPackage();
569 #ifdef AFS_DEMAND_ATTACH_FS
570 if (programType == fileServer) {
573 VLRU_SetOptions(VLRU_SET_ENABLED, 0);
575 osi_Assert(pthread_key_create(&VThread_key, NULL) == 0);
578 MUTEX_INIT(&vol_glock_mutex, "vol glock", MUTEX_DEFAULT, 0);
579 MUTEX_INIT(&vol_trans_mutex, "vol trans", MUTEX_DEFAULT, 0);
580 CV_INIT(&vol_put_volume_cond, "vol put", CV_DEFAULT, 0);
581 CV_INIT(&vol_sleep_cond, "vol sleep", CV_DEFAULT, 0);
582 CV_INIT(&vol_init_attach_cond, "vol init attach", CV_DEFAULT, 0);
583 CV_INIT(&vol_vinit_cond, "vol init", CV_DEFAULT, 0);
584 #ifndef AFS_PTHREAD_ENV
586 #endif /* AFS_PTHREAD_ENV */
587 Lock_Init(&vol_listLock);
589 srandom(time(0)); /* For VGetVolumeInfo */
591 #ifdef AFS_DEMAND_ATTACH_FS
592 MUTEX_INIT(&vol_salvsync_mutex, "salvsync", MUTEX_DEFAULT, 0);
593 #endif /* AFS_DEMAND_ATTACH_FS */
595 /* Ok, we have done enough initialization that fileserver can
596 * start accepting calls, even though the volumes may not be
597 * available just yet.
601 #if defined(AFS_DEMAND_ATTACH_FS) && defined(SALVSYNC_BUILD_SERVER)
602 if (programType == salvageServer) {
605 #endif /* AFS_DEMAND_ATTACH_FS */
606 #ifdef FSSYNC_BUILD_SERVER
607 if (programType == fileServer) {
611 #if defined(AFS_DEMAND_ATTACH_FS) && defined(SALVSYNC_BUILD_CLIENT)
612 if (VCanUseSALVSYNC()) {
613 /* establish a connection to the salvager at this point */
614 osi_Assert(VConnectSALV() != 0);
616 #endif /* AFS_DEMAND_ATTACH_FS */
618 if (opts->volcache > VStats.hdr_cache_size)
619 VStats.hdr_cache_size = opts->volcache;
620 VInitVolumeHeaderCache(VStats.hdr_cache_size);
622 VInitVnodes(vLarge, opts->nLargeVnodes);
623 VInitVnodes(vSmall, opts->nSmallVnodes);
626 errors = VAttachPartitions();
630 if (programType != fileServer) {
631 errors = VInitAttachVolumes(programType);
637 #ifdef FSSYNC_BUILD_CLIENT
638 if (VCanUseFSSYNC()) {
640 #ifdef AFS_DEMAND_ATTACH_FS
641 if (programType == salvageServer) {
642 Log("Unable to connect to file server; aborted\n");
645 #endif /* AFS_DEMAND_ATTACH_FS */
646 Log("Unable to connect to file server; will retry at need\n");
649 #endif /* FSSYNC_BUILD_CLIENT */
654 #if !defined(AFS_PTHREAD_ENV)
656 * Attach volumes in vice partitions
658 * @param[in] pt calling program type
661 * @note This is the original, non-threaded version of attach parititions.
663 * @post VInit state is 2
666 VInitAttachVolumes(ProgramType pt)
668 osi_Assert(VInit==1);
669 if (pt == fileServer) {
670 struct DiskPartition64 *diskP;
671 /* Attach all the volumes in this partition */
672 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
673 int nAttached = 0, nUnattached = 0;
674 osi_Assert(VAttachVolumesByPartition(diskP, &nAttached, &nUnattached) == 0);
678 VSetVInit_r(2); /* Initialized, and all volumes have been attached */
679 LWP_NoYieldSignal(VInitAttachVolumes);
683 #endif /* !AFS_PTHREAD_ENV */
685 #if defined(AFS_PTHREAD_ENV) && !defined(AFS_DEMAND_ATTACH_FS)
687 * Attach volumes in vice partitions
689 * @param[in] pt calling program type
692 * @note Threaded version of attach parititions.
694 * @post VInit state is 2
697 VInitAttachVolumes(ProgramType pt)
699 osi_Assert(VInit==1);
700 if (pt == fileServer) {
701 struct DiskPartition64 *diskP;
702 struct vinitvolumepackage_thread_t params;
703 struct diskpartition_queue_t * dpq;
704 int i, threads, parts;
706 pthread_attr_t attrs;
708 CV_INIT(¶ms.thread_done_cv, "thread done", CV_DEFAULT, 0);
710 params.n_threads_complete = 0;
712 /* create partition work queue */
713 for (parts=0, diskP = DiskPartitionList; diskP; diskP = diskP->next, parts++) {
714 dpq = (diskpartition_queue_t *) malloc(sizeof(struct diskpartition_queue_t));
715 osi_Assert(dpq != NULL);
717 queue_Append(¶ms,dpq);
720 threads = MIN(parts, vol_attach_threads);
723 /* spawn off a bunch of initialization threads */
724 osi_Assert(pthread_attr_init(&attrs) == 0);
725 osi_Assert(pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED) == 0);
727 Log("VInitVolumePackage: beginning parallel fileserver startup\n");
728 Log("VInitVolumePackage: using %d threads to attach volumes on %d partitions\n",
732 for (i=0; i < threads; i++) {
735 osi_Assert(pthread_create
736 (&tid, &attrs, &VInitVolumePackageThread,
738 AFS_SIGSET_RESTORE();
741 while(params.n_threads_complete < threads) {
742 VOL_CV_WAIT(¶ms.thread_done_cv);
746 osi_Assert(pthread_attr_destroy(&attrs) == 0);
748 /* if we're only going to run one init thread, don't bother creating
750 Log("VInitVolumePackage: beginning single-threaded fileserver startup\n");
751 Log("VInitVolumePackage: using 1 thread to attach volumes on %d partition(s)\n",
754 VInitVolumePackageThread(¶ms);
757 CV_DESTROY(¶ms.thread_done_cv);
760 VSetVInit_r(2); /* Initialized, and all volumes have been attached */
761 CV_BROADCAST(&vol_init_attach_cond);
767 VInitVolumePackageThread(void * args) {
769 struct DiskPartition64 *diskP;
770 struct vinitvolumepackage_thread_t * params;
771 struct diskpartition_queue_t * dpq;
773 params = (vinitvolumepackage_thread_t *) args;
777 /* Attach all the volumes in this partition */
778 while (queue_IsNotEmpty(params)) {
779 int nAttached = 0, nUnattached = 0;
781 if (vinit_attach_abort) {
782 Log("Aborting initialization\n");
786 dpq = queue_First(params,diskpartition_queue_t);
792 osi_Assert(VAttachVolumesByPartition(diskP, &nAttached, &nUnattached) == 0);
798 params->n_threads_complete++;
799 CV_SIGNAL(¶ms->thread_done_cv);
803 #endif /* AFS_PTHREAD_ENV && !AFS_DEMAND_ATTACH_FS */
805 #if defined(AFS_DEMAND_ATTACH_FS)
807 * Attach volumes in vice partitions
809 * @param[in] pt calling program type
812 * @note Threaded version of attach partitions.
814 * @post VInit state is 2
817 VInitAttachVolumes(ProgramType pt)
819 osi_Assert(VInit==1);
820 if (pt == fileServer) {
822 struct DiskPartition64 *diskP;
823 struct partition_queue pq;
824 struct volume_init_queue vq;
826 int i, threads, parts;
828 pthread_attr_t attrs;
830 /* create partition work queue */
832 CV_INIT(&(pq.cv), "partq", CV_DEFAULT, 0);
833 MUTEX_INIT(&(pq.mutex), "partq", MUTEX_DEFAULT, 0);
834 for (parts = 0, diskP = DiskPartitionList; diskP; diskP = diskP->next, parts++) {
835 struct diskpartition_queue_t *dp;
836 dp = (struct diskpartition_queue_t*)malloc(sizeof(struct diskpartition_queue_t));
837 osi_Assert(dp != NULL);
839 queue_Append(&pq, dp);
842 /* number of worker threads; at least one, not to exceed the number of partitions */
843 threads = MIN(parts, vol_attach_threads);
845 /* create volume work queue */
847 CV_INIT(&(vq.cv), "volq", CV_DEFAULT, 0);
848 MUTEX_INIT(&(vq.mutex), "volq", MUTEX_DEFAULT, 0);
850 osi_Assert(pthread_attr_init(&attrs) == 0);
851 osi_Assert(pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED) == 0);
853 Log("VInitVolumePackage: beginning parallel fileserver startup\n");
854 Log("VInitVolumePackage: using %d threads to pre-attach volumes on %d partitions\n",
857 /* create threads to scan disk partitions. */
858 for (i=0; i < threads; i++) {
859 struct vinitvolumepackage_thread_param *params;
862 params = (struct vinitvolumepackage_thread_param *)malloc(sizeof(struct vinitvolumepackage_thread_param));
866 params->nthreads = threads;
867 params->thread = i+1;
870 osi_Assert(pthread_create (&tid, &attrs, &VInitVolumePackageThread, (void*)params) == 0);
871 AFS_SIGSET_RESTORE();
874 VInitPreAttachVolumes(threads, &vq);
876 osi_Assert(pthread_attr_destroy(&attrs) == 0);
878 MUTEX_DESTROY(&pq.mutex);
880 MUTEX_DESTROY(&vq.mutex);
884 VSetVInit_r(2); /* Initialized, and all volumes have been attached */
885 CV_BROADCAST(&vol_init_attach_cond);
892 * Volume package initialization worker thread. Scan partitions for volume
893 * header files. Gather batches of volume ids and dispatch them to
894 * the main thread to be preattached. The volume preattachement is done
895 * in the main thread to avoid global volume lock contention.
898 VInitVolumePackageThread(void *args)
900 struct vinitvolumepackage_thread_param *params;
901 struct DiskPartition64 *partition;
902 struct partition_queue *pq;
903 struct volume_init_queue *vq;
904 struct volume_init_batch *vb;
907 params = (struct vinitvolumepackage_thread_param *)args;
913 vb = (struct volume_init_batch*)malloc(sizeof(struct volume_init_batch));
915 vb->thread = params->thread;
919 Log("Scanning partitions on thread %d of %d\n", params->thread, params->nthreads);
920 while((partition = VInitNextPartition(pq))) {
924 Log("Partition %s: pre-attaching volumes\n", partition->name);
925 dirp = opendir(VPartitionPath(partition));
927 Log("opendir on Partition %s failed, errno=%d!\n", partition->name, errno);
930 while ((vid = VInitNextVolumeId(dirp))) {
931 Volume *vp = (Volume*)malloc(sizeof(Volume));
933 memset(vp, 0, sizeof(Volume));
934 vp->device = partition->device;
935 vp->partition = partition;
937 queue_Init(&vp->vnode_list);
938 queue_Init(&vp->rx_call_list);
939 CV_INIT(&V_attachCV(vp), "partattach", CV_DEFAULT, 0);
941 vb->batch[vb->size++] = vp;
942 if (vb->size == VINIT_BATCH_MAX_SIZE) {
943 MUTEX_ENTER(&vq->mutex);
944 queue_Append(vq, vb);
945 CV_BROADCAST(&vq->cv);
946 MUTEX_EXIT(&vq->mutex);
948 vb = (struct volume_init_batch*)malloc(sizeof(struct volume_init_batch));
950 vb->thread = params->thread;
959 MUTEX_ENTER(&vq->mutex);
960 queue_Append(vq, vb);
961 CV_BROADCAST(&vq->cv);
962 MUTEX_EXIT(&vq->mutex);
964 Log("Partition scan thread %d of %d ended\n", params->thread, params->nthreads);
970 * Read next element from the pre-populated partition list.
972 static struct DiskPartition64*
973 VInitNextPartition(struct partition_queue *pq)
975 struct DiskPartition64 *partition;
976 struct diskpartition_queue_t *dp; /* queue element */
978 if (vinit_attach_abort) {
979 Log("Aborting volume preattach thread.\n");
983 /* get next partition to scan */
984 MUTEX_ENTER(&pq->mutex);
985 if (queue_IsEmpty(pq)) {
986 MUTEX_EXIT(&pq->mutex);
989 dp = queue_First(pq, diskpartition_queue_t);
991 MUTEX_EXIT(&pq->mutex);
994 osi_Assert(dp->diskP);
996 partition = dp->diskP;
1002 * Find next volume id on the partition.
1005 VInitNextVolumeId(DIR *dirp)
1011 while((d = readdir(dirp))) {
1012 if (vinit_attach_abort) {
1013 Log("Aborting volume preattach thread.\n");
1016 ext = strrchr(d->d_name, '.');
1017 if (d->d_name[0] == 'V' && ext && strcmp(ext, VHDREXT) == 0) {
1018 vid = VolumeNumber(d->d_name);
1022 Log("Warning: bogus volume header file: %s\n", d->d_name);
1029 * Preattach volumes in batches to avoid lock contention.
1032 VInitPreAttachVolumes(int nthreads, struct volume_init_queue *vq)
1034 struct volume_init_batch *vb;
1038 /* dequeue next volume */
1039 MUTEX_ENTER(&vq->mutex);
1040 if (queue_IsEmpty(vq)) {
1041 CV_WAIT(&vq->cv, &vq->mutex);
1043 vb = queue_First(vq, volume_init_batch);
1045 MUTEX_EXIT(&vq->mutex);
1049 for (i = 0; i<vb->size; i++) {
1055 dup = VLookupVolume_r(&ec, vp->hashid, NULL);
1057 Log("Error looking up volume, code=%d\n", ec);
1060 Log("Warning: Duplicate volume id %d detected.\n", vp->hashid);
1063 /* put pre-attached volume onto the hash table
1064 * and bring it up to the pre-attached state */
1065 AddVolumeToHashTable(vp, vp->hashid);
1066 AddVolumeToVByPList_r(vp);
1067 VLRU_Init_Node_r(vp);
1068 VChangeState_r(vp, VOL_STATE_PREATTACHED);
1081 #endif /* AFS_DEMAND_ATTACH_FS */
1083 #if !defined(AFS_DEMAND_ATTACH_FS)
1085 * attach all volumes on a given disk partition
1088 VAttachVolumesByPartition(struct DiskPartition64 *diskP, int * nAttached, int * nUnattached)
1094 Log("Partition %s: attaching volumes\n", diskP->name);
1095 dirp = opendir(VPartitionPath(diskP));
1097 Log("opendir on Partition %s failed!\n", diskP->name);
1101 while ((dp = readdir(dirp))) {
1103 p = strrchr(dp->d_name, '.');
1105 if (vinit_attach_abort) {
1106 Log("Partition %s: abort attach volumes\n", diskP->name);
1110 if (p != NULL && strcmp(p, VHDREXT) == 0) {
1113 vp = VAttachVolumeByName(&error, diskP->name, dp->d_name,
1115 (*(vp ? nAttached : nUnattached))++;
1116 if (error == VOFFLINE)
1117 Log("Volume %d stays offline (/vice/offline/%s exists)\n", VolumeNumber(dp->d_name), dp->d_name);
1118 else if (LogLevel >= 5) {
1119 Log("Partition %s: attached volume %d (%s)\n",
1120 diskP->name, VolumeNumber(dp->d_name),
1129 Log("Partition %s: attached %d volumes; %d volumes not attached\n", diskP->name, *nAttached, *nUnattached);
1134 #endif /* !AFS_DEMAND_ATTACH_FS */
1136 /***************************************************/
1137 /* Shutdown routines */
1138 /***************************************************/
1142 * highly multithreaded volume package shutdown
1144 * with the demand attach fileserver extensions,
1145 * VShutdown has been modified to be multithreaded.
1146 * In order to achieve optimal use of many threads,
1147 * the shutdown code involves one control thread and
1148 * n shutdown worker threads. The control thread
1149 * periodically examines the number of volumes available
1150 * for shutdown on each partition, and produces a worker
1151 * thread allocation schedule. The idea is to eliminate
1152 * redundant scheduling computation on the workers by
1153 * having a single master scheduler.
1155 * The scheduler's objectives are:
1157 * each partition with volumes remaining gets allocated
1158 * at least 1 thread (assuming sufficient threads)
1160 * threads are allocated proportional to the number of
1161 * volumes remaining to be offlined. This ensures that
1162 * the OS I/O scheduler has many requests to elevator
1163 * seek on partitions that will (presumably) take the
1164 * longest amount of time (from now) to finish shutdown
1165 * (3) keep threads busy
1166 * when there are extra threads, they are assigned to
1167 * partitions using a simple round-robin algorithm
1169 * In the future, we may wish to add the ability to adapt
1170 * to the relative performance patterns of each disk
1175 * multi-step shutdown process
1177 * demand attach shutdown is a four-step process. Each
1178 * shutdown "pass" shuts down increasingly more difficult
1179 * volumes. The main purpose is to achieve better cache
1180 * utilization during shutdown.
1183 * shutdown volumes in the unattached, pre-attached
1186 * shutdown attached volumes with cached volume headers
1188 * shutdown all volumes in non-exclusive states
1190 * shutdown all remaining volumes
1193 #ifdef AFS_DEMAND_ATTACH_FS
1199 struct DiskPartition64 * diskP;
1200 struct diskpartition_queue_t * dpq;
1201 vshutdown_thread_t params;
1203 pthread_attr_t attrs;
1205 memset(¶ms, 0, sizeof(vshutdown_thread_t));
1208 Log("VShutdown: aborting attach volumes\n");
1209 vinit_attach_abort = 1;
1210 VOL_CV_WAIT(&vol_init_attach_cond);
1213 for (params.n_parts=0, diskP = DiskPartitionList;
1214 diskP; diskP = diskP->next, params.n_parts++);
1216 Log("VShutdown: shutting down on-line volumes on %d partition%s...\n",
1217 params.n_parts, params.n_parts > 1 ? "s" : "");
1219 vol_shutting_down = 1;
1221 if (vol_attach_threads > 1) {
1222 /* prepare for parallel shutdown */
1223 params.n_threads = vol_attach_threads;
1224 MUTEX_INIT(¶ms.lock, "params", MUTEX_DEFAULT, 0);
1225 CV_INIT(¶ms.cv, "params", CV_DEFAULT, 0);
1226 CV_INIT(¶ms.master_cv, "params master", CV_DEFAULT, 0);
1227 osi_Assert(pthread_attr_init(&attrs) == 0);
1228 osi_Assert(pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED) == 0);
1229 queue_Init(¶ms);
1231 /* setup the basic partition information structures for
1232 * parallel shutdown */
1233 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1235 struct rx_queue * qp, * nqp;
1239 VVByPListWait_r(diskP);
1240 VVByPListBeginExclusive_r(diskP);
1243 for (queue_Scan(&diskP->vol_list, qp, nqp, rx_queue)) {
1244 vp = (Volume *)((char *)qp - offsetof(Volume, vol_list));
1248 Log("VShutdown: partition %s has %d volumes with attached headers\n",
1249 VPartitionPath(diskP), count);
1252 /* build up the pass 0 shutdown work queue */
1253 dpq = (struct diskpartition_queue_t *) malloc(sizeof(struct diskpartition_queue_t));
1254 osi_Assert(dpq != NULL);
1256 queue_Prepend(¶ms, dpq);
1258 params.part_pass_head[diskP->index] = queue_First(&diskP->vol_list, rx_queue);
1261 Log("VShutdown: beginning parallel fileserver shutdown\n");
1262 Log("VShutdown: using %d threads to offline volumes on %d partition%s\n",
1263 vol_attach_threads, params.n_parts, params.n_parts > 1 ? "s" : "" );
1265 /* do pass 0 shutdown */
1266 MUTEX_ENTER(¶ms.lock);
1267 for (i=0; i < params.n_threads; i++) {
1268 osi_Assert(pthread_create
1269 (&tid, &attrs, &VShutdownThread,
1273 /* wait for all the pass 0 shutdowns to complete */
1274 while (params.n_threads_complete < params.n_threads) {
1275 CV_WAIT(¶ms.master_cv, ¶ms.lock);
1277 params.n_threads_complete = 0;
1279 CV_BROADCAST(¶ms.cv);
1280 MUTEX_EXIT(¶ms.lock);
1282 Log("VShutdown: pass 0 completed using the 1 thread per partition algorithm\n");
1283 Log("VShutdown: starting passes 1 through 3 using finely-granular mp-fast algorithm\n");
1285 /* run the parallel shutdown scheduler. it will drop the glock internally */
1286 ShutdownController(¶ms);
1288 /* wait for all the workers to finish pass 3 and terminate */
1289 while (params.pass < 4) {
1290 VOL_CV_WAIT(¶ms.cv);
1293 osi_Assert(pthread_attr_destroy(&attrs) == 0);
1294 CV_DESTROY(¶ms.cv);
1295 CV_DESTROY(¶ms.master_cv);
1296 MUTEX_DESTROY(¶ms.lock);
1298 /* drop the VByPList exclusive reservations */
1299 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1300 VVByPListEndExclusive_r(diskP);
1301 Log("VShutdown: %s stats : (pass[0]=%d, pass[1]=%d, pass[2]=%d, pass[3]=%d)\n",
1302 VPartitionPath(diskP),
1303 params.stats[0][diskP->index],
1304 params.stats[1][diskP->index],
1305 params.stats[2][diskP->index],
1306 params.stats[3][diskP->index]);
1309 Log("VShutdown: shutdown finished using %d threads\n", params.n_threads);
1311 /* if we're only going to run one shutdown thread, don't bother creating
1313 Log("VShutdown: beginning single-threaded fileserver shutdown\n");
1315 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1316 VShutdownByPartition_r(diskP);
1320 Log("VShutdown: complete.\n");
1323 #else /* AFS_DEMAND_ATTACH_FS */
1333 Log("VShutdown: aborting attach volumes\n");
1334 vinit_attach_abort = 1;
1335 #ifdef AFS_PTHREAD_ENV
1336 VOL_CV_WAIT(&vol_init_attach_cond);
1338 LWP_WaitProcess(VInitAttachVolumes);
1339 #endif /* AFS_PTHREAD_ENV */
1342 Log("VShutdown: shutting down on-line volumes...\n");
1343 vol_shutting_down = 1;
1344 for (i = 0; i < VolumeHashTable.Size; i++) {
1345 /* try to hold first volume in the hash table */
1346 for (queue_Scan(&VolumeHashTable.Table[i],vp,np,Volume)) {
1350 Log("VShutdown: Attempting to take volume %u offline.\n",
1353 /* next, take the volume offline (drops reference count) */
1354 VOffline_r(vp, "File server was shut down");
1358 Log("VShutdown: complete.\n");
1360 #endif /* AFS_DEMAND_ATTACH_FS */
1366 osi_Assert(VInit>0);
1373 * stop new activity (e.g. SALVSYNC) from occurring
1375 * Use this to make the volume package less busy; for example, during
1376 * shutdown. This doesn't actually shutdown/detach anything in the
1377 * volume package, but prevents certain processes from ocurring. For
1378 * example, preventing new SALVSYNC communication in DAFS. In theory, we
1379 * could also use this to prevent new volume attachment, or prevent
1380 * other programs from checking out volumes, etc.
1385 #ifdef AFS_DEMAND_ATTACH_FS
1386 /* make sure we don't try to contact the salvageserver, since it may
1387 * not be around anymore */
1388 vol_disallow_salvsync = 1;
1392 #ifdef AFS_DEMAND_ATTACH_FS
1395 * shutdown control thread
1398 ShutdownController(vshutdown_thread_t * params)
1401 struct DiskPartition64 * diskP;
1403 vshutdown_thread_t shadow;
1405 ShutdownCreateSchedule(params);
1407 while ((params->pass < 4) &&
1408 (params->n_threads_complete < params->n_threads)) {
1409 /* recompute schedule once per second */
1411 memcpy(&shadow, params, sizeof(vshutdown_thread_t));
1415 Log("ShutdownController: schedule version=%d, vol_remaining=%d, pass=%d\n",
1416 shadow.schedule_version, shadow.vol_remaining, shadow.pass);
1417 Log("ShutdownController: n_threads_complete=%d, n_parts_done_pass=%d\n",
1418 shadow.n_threads_complete, shadow.n_parts_done_pass);
1419 for (diskP = DiskPartitionList; diskP; diskP=diskP->next) {
1421 Log("ShutdownController: part[%d] : (len=%d, thread_target=%d, done_pass=%d, pass_head=%p)\n",
1423 diskP->vol_list.len,
1424 shadow.part_thread_target[id],
1425 shadow.part_done_pass[id],
1426 shadow.part_pass_head[id]);
1432 ShutdownCreateSchedule(params);
1436 /* create the shutdown thread work schedule.
1437 * this scheduler tries to implement fairness
1438 * by allocating at least 1 thread to each
1439 * partition with volumes to be shutdown,
1440 * and then it attempts to allocate remaining
1441 * threads based upon the amount of work left
1444 ShutdownCreateSchedule(vshutdown_thread_t * params)
1446 struct DiskPartition64 * diskP;
1447 int sum, thr_workload, thr_left;
1448 int part_residue[VOLMAXPARTS+1];
1451 /* compute the total number of outstanding volumes */
1453 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1454 sum += diskP->vol_list.len;
1457 params->schedule_version++;
1458 params->vol_remaining = sum;
1463 /* compute average per-thread workload */
1464 thr_workload = sum / params->n_threads;
1465 if (sum % params->n_threads)
1468 thr_left = params->n_threads;
1469 memset(&part_residue, 0, sizeof(part_residue));
1471 /* for fairness, give every partition with volumes remaining
1472 * at least one thread */
1473 for (diskP = DiskPartitionList; diskP && thr_left; diskP = diskP->next) {
1475 if (diskP->vol_list.len) {
1476 params->part_thread_target[id] = 1;
1479 params->part_thread_target[id] = 0;
1483 if (thr_left && thr_workload) {
1484 /* compute length-weighted workloads */
1487 for (diskP = DiskPartitionList; diskP && thr_left; diskP = diskP->next) {
1489 delta = (diskP->vol_list.len / thr_workload) -
1490 params->part_thread_target[id];
1494 if (delta < thr_left) {
1495 params->part_thread_target[id] += delta;
1498 params->part_thread_target[id] += thr_left;
1506 /* try to assign any leftover threads to partitions that
1507 * had volume lengths closer to needing thread_target+1 */
1508 int max_residue, max_id = 0;
1510 /* compute the residues */
1511 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1513 part_residue[id] = diskP->vol_list.len -
1514 (params->part_thread_target[id] * thr_workload);
1517 /* now try to allocate remaining threads to partitions with the
1518 * highest residues */
1521 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1523 if (part_residue[id] > max_residue) {
1524 max_residue = part_residue[id];
1533 params->part_thread_target[max_id]++;
1535 part_residue[max_id] = 0;
1540 /* punt and give any remaining threads equally to each partition */
1542 if (thr_left >= params->n_parts) {
1543 alloc = thr_left / params->n_parts;
1544 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1546 params->part_thread_target[id] += alloc;
1551 /* finish off the last of the threads */
1552 for (diskP = DiskPartitionList; thr_left && diskP; diskP = diskP->next) {
1554 params->part_thread_target[id]++;
1560 /* worker thread for parallel shutdown */
1562 VShutdownThread(void * args)
1564 vshutdown_thread_t * params;
1565 int found, pass, schedule_version_save, count;
1566 struct DiskPartition64 *diskP;
1567 struct diskpartition_queue_t * dpq;
1570 params = (vshutdown_thread_t *) args;
1572 /* acquire the shutdown pass 0 lock */
1573 MUTEX_ENTER(¶ms->lock);
1575 /* if there's still pass 0 work to be done,
1576 * get a work entry, and do a pass 0 shutdown */
1577 if (queue_IsNotEmpty(params)) {
1578 dpq = queue_First(params, diskpartition_queue_t);
1580 MUTEX_EXIT(¶ms->lock);
1586 while (ShutdownVolumeWalk_r(diskP, 0, ¶ms->part_pass_head[id]))
1588 params->stats[0][diskP->index] = count;
1589 MUTEX_ENTER(¶ms->lock);
1592 params->n_threads_complete++;
1593 if (params->n_threads_complete == params->n_threads) {
1594 /* notify control thread that all workers have completed pass 0 */
1595 CV_SIGNAL(¶ms->master_cv);
1597 while (params->pass == 0) {
1598 CV_WAIT(¶ms->cv, ¶ms->lock);
1602 MUTEX_EXIT(¶ms->lock);
1605 pass = params->pass;
1606 osi_Assert(pass > 0);
1608 /* now escalate through the more complicated shutdowns */
1610 schedule_version_save = params->schedule_version;
1612 /* find a disk partition to work on */
1613 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1615 if (params->part_thread_target[id] && !params->part_done_pass[id]) {
1616 params->part_thread_target[id]--;
1623 /* hmm. for some reason the controller thread couldn't find anything for
1624 * us to do. let's see if there's anything we can do */
1625 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1627 if (diskP->vol_list.len && !params->part_done_pass[id]) {
1630 } else if (!params->part_done_pass[id]) {
1631 params->part_done_pass[id] = 1;
1632 params->n_parts_done_pass++;
1634 Log("VShutdown: done shutting down volumes on partition %s.\n",
1635 VPartitionPath(diskP));
1641 /* do work on this partition until either the controller
1642 * creates a new schedule, or we run out of things to do
1643 * on this partition */
1646 while (!params->part_done_pass[id] &&
1647 (schedule_version_save == params->schedule_version)) {
1648 /* ShutdownVolumeWalk_r will drop the glock internally */
1649 if (!ShutdownVolumeWalk_r(diskP, pass, ¶ms->part_pass_head[id])) {
1650 if (!params->part_done_pass[id]) {
1651 params->part_done_pass[id] = 1;
1652 params->n_parts_done_pass++;
1654 Log("VShutdown: done shutting down volumes on partition %s.\n",
1655 VPartitionPath(diskP));
1663 params->stats[pass][id] += count;
1665 /* ok, everyone is done this pass, proceed */
1668 params->n_threads_complete++;
1669 while (params->pass == pass) {
1670 if (params->n_threads_complete == params->n_threads) {
1671 /* we are the last thread to complete, so we will
1672 * reinitialize worker pool state for the next pass */
1673 params->n_threads_complete = 0;
1674 params->n_parts_done_pass = 0;
1676 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1678 params->part_done_pass[id] = 0;
1679 params->part_pass_head[id] = queue_First(&diskP->vol_list, rx_queue);
1682 /* compute a new thread schedule before releasing all the workers */
1683 ShutdownCreateSchedule(params);
1685 /* wake up all the workers */
1686 CV_BROADCAST(¶ms->cv);
1689 Log("VShutdown: pass %d completed using %d threads on %d partitions\n",
1690 pass, params->n_threads, params->n_parts);
1693 VOL_CV_WAIT(¶ms->cv);
1696 pass = params->pass;
1710 /* shut down all volumes on a given disk partition
1712 * note that this function will not allow mp-fast
1713 * shutdown of a partition */
1715 VShutdownByPartition_r(struct DiskPartition64 * dp)
1721 /* wait for other exclusive ops to finish */
1722 VVByPListWait_r(dp);
1724 /* begin exclusive access */
1725 VVByPListBeginExclusive_r(dp);
1727 /* pick the low-hanging fruit first,
1728 * then do the complicated ones last
1729 * (has the advantage of keeping
1730 * in-use volumes up until the bitter end) */
1731 for (pass = 0, total=0; pass < 4; pass++) {
1732 pass_stats[pass] = ShutdownVByPForPass_r(dp, pass);
1733 total += pass_stats[pass];
1736 /* end exclusive access */
1737 VVByPListEndExclusive_r(dp);
1739 Log("VShutdownByPartition: shut down %d volumes on %s (pass[0]=%d, pass[1]=%d, pass[2]=%d, pass[3]=%d)\n",
1740 total, VPartitionPath(dp), pass_stats[0], pass_stats[1], pass_stats[2], pass_stats[3]);
1745 /* internal shutdown functionality
1747 * for multi-pass shutdown:
1748 * 0 to only "shutdown" {pre,un}attached and error state volumes
1749 * 1 to also shutdown attached volumes w/ volume header loaded
1750 * 2 to also shutdown attached volumes w/o volume header loaded
1751 * 3 to also shutdown exclusive state volumes
1753 * caller MUST hold exclusive access on the hash chain
1754 * because we drop vol_glock_mutex internally
1756 * this function is reentrant for passes 1--3
1757 * (e.g. multiple threads can cooperate to
1758 * shutdown a partition mp-fast)
1760 * pass 0 is not scaleable because the volume state data is
1761 * synchronized by vol_glock mutex, and the locking overhead
1762 * is too high to drop the lock long enough to do linked list
1766 ShutdownVByPForPass_r(struct DiskPartition64 * dp, int pass)
1768 struct rx_queue * q = queue_First(&dp->vol_list, rx_queue);
1771 while (ShutdownVolumeWalk_r(dp, pass, &q))
1777 /* conditionally shutdown one volume on partition dp
1778 * returns 1 if a volume was shutdown in this pass,
1781 ShutdownVolumeWalk_r(struct DiskPartition64 * dp, int pass,
1782 struct rx_queue ** idx)
1784 struct rx_queue *qp, *nqp;
1789 for (queue_ScanFrom(&dp->vol_list, qp, qp, nqp, rx_queue)) {
1790 vp = (Volume *) (((char *)qp) - offsetof(Volume, vol_list));
1794 if ((V_attachState(vp) != VOL_STATE_UNATTACHED) &&
1795 (V_attachState(vp) != VOL_STATE_ERROR) &&
1796 (V_attachState(vp) != VOL_STATE_DELETED) &&
1797 (V_attachState(vp) != VOL_STATE_PREATTACHED)) {
1801 if ((V_attachState(vp) == VOL_STATE_ATTACHED) &&
1802 (vp->header == NULL)) {
1806 if (VIsExclusiveState(V_attachState(vp))) {
1811 DeleteVolumeFromVByPList_r(vp);
1812 VShutdownVolume_r(vp);
1822 * shutdown a specific volume
1824 /* caller MUST NOT hold a heavyweight ref on vp */
1826 VShutdownVolume_r(Volume * vp)
1830 VCreateReservation_r(vp);
1832 if (LogLevel >= 5) {
1833 Log("VShutdownVolume_r: vid=%u, device=%d, state=%hu\n",
1834 vp->hashid, vp->partition->device, V_attachState(vp));
1837 /* wait for other blocking ops to finish */
1838 VWaitExclusiveState_r(vp);
1840 osi_Assert(VIsValidState(V_attachState(vp)));
1842 switch(V_attachState(vp)) {
1843 case VOL_STATE_SALVAGING:
1844 /* Leave salvaging volumes alone. Any in-progress salvages will
1845 * continue working after viced shuts down. This is intentional.
1848 case VOL_STATE_PREATTACHED:
1849 case VOL_STATE_ERROR:
1850 VChangeState_r(vp, VOL_STATE_UNATTACHED);
1851 case VOL_STATE_UNATTACHED:
1852 case VOL_STATE_DELETED:
1854 case VOL_STATE_GOING_OFFLINE:
1855 case VOL_STATE_SHUTTING_DOWN:
1856 case VOL_STATE_ATTACHED:
1860 Log("VShutdown: Attempting to take volume %u offline.\n",
1863 /* take the volume offline (drops reference count) */
1864 VOffline_r(vp, "File server was shut down");
1871 VCancelReservation_r(vp);
1875 #endif /* AFS_DEMAND_ATTACH_FS */
1878 /***************************************************/
1879 /* Header I/O routines */
1880 /***************************************************/
1882 /* open a descriptor for the inode (h),
1883 * read in an on-disk structure into buffer (to) of size (size),
1884 * verify versionstamp in structure has magic (magic) and
1885 * optionally verify version (version) if (version) is nonzero
1888 ReadHeader(Error * ec, IHandle_t * h, char *to, int size, bit32 magic,
1891 struct versionStamp *vsn;
1906 vsn = (struct versionStamp *)to;
1907 if (FDH_PREAD(fdP, to, size, 0) != size || vsn->magic != magic) {
1909 FDH_REALLYCLOSE(fdP);
1914 /* Check is conditional, in case caller wants to inspect version himself */
1915 if (version && vsn->version != version) {
1921 WriteVolumeHeader_r(Error * ec, Volume * vp)
1923 IHandle_t *h = V_diskDataHandle(vp);
1933 if (FDH_PWRITE(fdP, (char *)&V_disk(vp), sizeof(V_disk(vp)), 0)
1934 != sizeof(V_disk(vp))) {
1936 FDH_REALLYCLOSE(fdP);
1942 /* VolumeHeaderToDisk
1943 * Allows for storing 64 bit inode numbers in on-disk volume header
1946 /* convert in-memory representation of a volume header to the
1947 * on-disk representation of a volume header */
1949 VolumeHeaderToDisk(VolumeDiskHeader_t * dh, VolumeHeader_t * h)
1952 memset(dh, 0, sizeof(VolumeDiskHeader_t));
1953 dh->stamp = h->stamp;
1955 dh->parent = h->parent;
1957 #ifdef AFS_64BIT_IOPS_ENV
1958 dh->volumeInfo_lo = (afs_int32) h->volumeInfo & 0xffffffff;
1959 dh->volumeInfo_hi = (afs_int32) (h->volumeInfo >> 32) & 0xffffffff;
1960 dh->smallVnodeIndex_lo = (afs_int32) h->smallVnodeIndex & 0xffffffff;
1961 dh->smallVnodeIndex_hi =
1962 (afs_int32) (h->smallVnodeIndex >> 32) & 0xffffffff;
1963 dh->largeVnodeIndex_lo = (afs_int32) h->largeVnodeIndex & 0xffffffff;
1964 dh->largeVnodeIndex_hi =
1965 (afs_int32) (h->largeVnodeIndex >> 32) & 0xffffffff;
1966 dh->linkTable_lo = (afs_int32) h->linkTable & 0xffffffff;
1967 dh->linkTable_hi = (afs_int32) (h->linkTable >> 32) & 0xffffffff;
1969 dh->volumeInfo_lo = h->volumeInfo;
1970 dh->smallVnodeIndex_lo = h->smallVnodeIndex;
1971 dh->largeVnodeIndex_lo = h->largeVnodeIndex;
1972 dh->linkTable_lo = h->linkTable;
1976 /* DiskToVolumeHeader
1977 * Converts an on-disk representation of a volume header to
1978 * the in-memory representation of a volume header.
1980 * Makes the assumption that AFS has *always*
1981 * zero'd the volume header file so that high parts of inode
1982 * numbers are 0 in older (SGI EFS) volume header files.
1985 DiskToVolumeHeader(VolumeHeader_t * h, VolumeDiskHeader_t * dh)
1987 memset(h, 0, sizeof(VolumeHeader_t));
1988 h->stamp = dh->stamp;
1990 h->parent = dh->parent;
1992 #ifdef AFS_64BIT_IOPS_ENV
1994 (Inode) dh->volumeInfo_lo | ((Inode) dh->volumeInfo_hi << 32);
1996 h->smallVnodeIndex =
1997 (Inode) dh->smallVnodeIndex_lo | ((Inode) dh->
1998 smallVnodeIndex_hi << 32);
2000 h->largeVnodeIndex =
2001 (Inode) dh->largeVnodeIndex_lo | ((Inode) dh->
2002 largeVnodeIndex_hi << 32);
2004 (Inode) dh->linkTable_lo | ((Inode) dh->linkTable_hi << 32);
2006 h->volumeInfo = dh->volumeInfo_lo;
2007 h->smallVnodeIndex = dh->smallVnodeIndex_lo;
2008 h->largeVnodeIndex = dh->largeVnodeIndex_lo;
2009 h->linkTable = dh->linkTable_lo;
2014 /***************************************************/
2015 /* Volume Attachment routines */
2016 /***************************************************/
2018 #ifdef AFS_DEMAND_ATTACH_FS
2020 * pre-attach a volume given its path.
2022 * @param[out] ec outbound error code
2023 * @param[in] partition partition path string
2024 * @param[in] name volume id string
2026 * @return volume object pointer
2028 * @note A pre-attached volume will only have its partition
2029 * and hashid fields initialized. At first call to
2030 * VGetVolume, the volume will be fully attached.
2034 VPreAttachVolumeByName(Error * ec, char *partition, char *name)
2038 vp = VPreAttachVolumeByName_r(ec, partition, name);
2044 * pre-attach a volume given its path.
2046 * @param[out] ec outbound error code
2047 * @param[in] partition path to vice partition
2048 * @param[in] name volume id string
2050 * @return volume object pointer
2052 * @pre VOL_LOCK held
2054 * @internal volume package internal use only.
2057 VPreAttachVolumeByName_r(Error * ec, char *partition, char *name)
2059 return VPreAttachVolumeById_r(ec,
2061 VolumeNumber(name));
2065 * pre-attach a volume given its path and numeric volume id.
2067 * @param[out] ec error code return
2068 * @param[in] partition path to vice partition
2069 * @param[in] volumeId numeric volume id
2071 * @return volume object pointer
2073 * @pre VOL_LOCK held
2075 * @internal volume package internal use only.
2078 VPreAttachVolumeById_r(Error * ec,
2083 struct DiskPartition64 *partp;
2087 osi_Assert(programType == fileServer);
2089 if (!(partp = VGetPartition_r(partition, 0))) {
2091 Log("VPreAttachVolumeById_r: Error getting partition (%s)\n", partition);
2095 vp = VLookupVolume_r(ec, volumeId, NULL);
2100 return VPreAttachVolumeByVp_r(ec, partp, vp, volumeId);
2104 * preattach a volume.
2106 * @param[out] ec outbound error code
2107 * @param[in] partp pointer to partition object
2108 * @param[in] vp pointer to volume object
2109 * @param[in] vid volume id
2111 * @return volume object pointer
2113 * @pre VOL_LOCK is held.
2115 * @warning Returned volume object pointer does not have to
2116 * equal the pointer passed in as argument vp. There
2117 * are potential race conditions which can result in
2118 * the pointers having different values. It is up to
2119 * the caller to make sure that references are handled
2120 * properly in this case.
2122 * @note If there is already a volume object registered with
2123 * the same volume id, its pointer MUST be passed as
2124 * argument vp. Failure to do so will result in a silent
2125 * failure to preattach.
2127 * @internal volume package internal use only.
2130 VPreAttachVolumeByVp_r(Error * ec,
2131 struct DiskPartition64 * partp,
2139 /* check to see if pre-attach already happened */
2141 (V_attachState(vp) != VOL_STATE_UNATTACHED) &&
2142 (V_attachState(vp) != VOL_STATE_DELETED) &&
2143 (V_attachState(vp) != VOL_STATE_PREATTACHED) &&
2144 !VIsErrorState(V_attachState(vp))) {
2146 * pre-attach is a no-op in all but the following cases:
2148 * - volume is unattached
2149 * - volume is in an error state
2150 * - volume is pre-attached
2152 Log("VPreattachVolumeByVp_r: volume %u not in quiescent state\n", vid);
2155 /* we're re-attaching a volume; clear out some old state */
2156 memset(&vp->salvage, 0, sizeof(struct VolumeOnlineSalvage));
2158 if (V_partition(vp) != partp) {
2159 /* XXX potential race */
2160 DeleteVolumeFromVByPList_r(vp);
2163 /* if we need to allocate a new Volume struct,
2164 * go ahead and drop the vol glock, otherwise
2165 * do the basic setup synchronised, as it's
2166 * probably not worth dropping the lock */
2169 /* allocate the volume structure */
2170 vp = nvp = (Volume *) malloc(sizeof(Volume));
2171 osi_Assert(vp != NULL);
2172 memset(vp, 0, sizeof(Volume));
2173 queue_Init(&vp->vnode_list);
2174 queue_Init(&vp->rx_call_list);
2175 CV_INIT(&V_attachCV(vp), "vp attach", CV_DEFAULT, 0);
2178 /* link the volume with its associated vice partition */
2179 vp->device = partp->device;
2180 vp->partition = partp;
2183 vp->specialStatus = 0;
2185 /* if we dropped the lock, reacquire the lock,
2186 * check for pre-attach races, and then add
2187 * the volume to the hash table */
2190 nvp = VLookupVolume_r(ec, vid, NULL);
2195 } else if (nvp) { /* race detected */
2200 /* hack to make up for VChangeState_r() decrementing
2201 * the old state counter */
2202 VStats.state_levels[0]++;
2206 /* put pre-attached volume onto the hash table
2207 * and bring it up to the pre-attached state */
2208 AddVolumeToHashTable(vp, vp->hashid);
2209 AddVolumeToVByPList_r(vp);
2210 VLRU_Init_Node_r(vp);
2211 VChangeState_r(vp, VOL_STATE_PREATTACHED);
2214 Log("VPreAttachVolumeByVp_r: volume %u pre-attached\n", vp->hashid);
2222 #endif /* AFS_DEMAND_ATTACH_FS */
2224 /* Attach an existing volume, given its pathname, and return a
2225 pointer to the volume header information. The volume also
2226 normally goes online at this time. An offline volume
2227 must be reattached to make it go online */
2229 VAttachVolumeByName(Error * ec, char *partition, char *name, int mode)
2233 retVal = VAttachVolumeByName_r(ec, partition, name, mode);
2239 VAttachVolumeByName_r(Error * ec, char *partition, char *name, int mode)
2242 struct DiskPartition64 *partp;
2247 #ifdef AFS_DEMAND_ATTACH_FS
2248 VolumeStats stats_save;
2250 #endif /* AFS_DEMAND_ATTACH_FS */
2254 volumeId = VolumeNumber(name);
2256 if (!(partp = VGetPartition_r(partition, 0))) {
2258 Log("VAttachVolume: Error getting partition (%s)\n", partition);
2262 if (VRequiresPartLock()) {
2263 osi_Assert(VInit == 3);
2264 VLockPartition_r(partition);
2265 } else if (programType == fileServer) {
2266 #ifdef AFS_DEMAND_ATTACH_FS
2267 /* lookup the volume in the hash table */
2268 vp = VLookupVolume_r(ec, volumeId, NULL);
2274 /* save any counters that are supposed to
2275 * be monotonically increasing over the
2276 * lifetime of the fileserver */
2277 memcpy(&stats_save, &vp->stats, sizeof(VolumeStats));
2279 memset(&stats_save, 0, sizeof(VolumeStats));
2282 /* if there's something in the hash table, and it's not
2283 * in the pre-attach state, then we may need to detach
2284 * it before proceeding */
2285 if (vp && (V_attachState(vp) != VOL_STATE_PREATTACHED)) {
2286 VCreateReservation_r(vp);
2287 VWaitExclusiveState_r(vp);
2289 /* at this point state must be one of:
2299 if (vp->specialStatus == VBUSY)
2302 /* if it's already attached, see if we can return it */
2303 if (V_attachState(vp) == VOL_STATE_ATTACHED) {
2304 VGetVolumeByVp_r(ec, vp);
2305 if (V_inUse(vp) == fileServer) {
2306 VCancelReservation_r(vp);
2310 /* otherwise, we need to detach, and attempt to re-attach */
2311 VDetachVolume_r(ec, vp);
2313 Log("VAttachVolume: Error detaching old volume instance (%s)\n", name);
2316 /* if it isn't fully attached, delete from the hash tables,
2317 and let the refcounter handle the rest */
2318 DeleteVolumeFromHashTable(vp);
2319 DeleteVolumeFromVByPList_r(vp);
2322 VCancelReservation_r(vp);
2326 /* pre-attach volume if it hasn't been done yet */
2328 (V_attachState(vp) == VOL_STATE_UNATTACHED) ||
2329 (V_attachState(vp) == VOL_STATE_DELETED) ||
2330 (V_attachState(vp) == VOL_STATE_ERROR)) {
2332 vp = VPreAttachVolumeByVp_r(ec, partp, vp, volumeId);
2338 osi_Assert(vp != NULL);
2340 /* handle pre-attach races
2342 * multiple threads can race to pre-attach a volume,
2343 * but we can't let them race beyond that
2345 * our solution is to let the first thread to bring
2346 * the volume into an exclusive state win; the other
2347 * threads just wait until it finishes bringing the
2348 * volume online, and then they do a vgetvolumebyvp
2350 if (svp && (svp != vp)) {
2351 /* wait for other exclusive ops to finish */
2352 VCreateReservation_r(vp);
2353 VWaitExclusiveState_r(vp);
2355 /* get a heavyweight ref, kill the lightweight ref, and return */
2356 VGetVolumeByVp_r(ec, vp);
2357 VCancelReservation_r(vp);
2361 /* at this point, we are chosen as the thread to do
2362 * demand attachment for this volume. all other threads
2363 * doing a getvolume on vp->hashid will block until we finish */
2365 /* make sure any old header cache entries are invalidated
2366 * before proceeding */
2367 FreeVolumeHeader(vp);
2369 VChangeState_r(vp, VOL_STATE_ATTACHING);
2371 /* restore any saved counters */
2372 memcpy(&vp->stats, &stats_save, sizeof(VolumeStats));
2373 #else /* AFS_DEMAND_ATTACH_FS */
2374 vp = VGetVolume_r(ec, volumeId);
2376 if (V_inUse(vp) == fileServer)
2378 if (vp->specialStatus == VBUSY)
2380 VDetachVolume_r(ec, vp);
2382 Log("VAttachVolume: Error detaching volume (%s)\n", name);
2386 #endif /* AFS_DEMAND_ATTACH_FS */
2390 strcpy(path, VPartitionPath(partp));
2394 strcat(path, OS_DIRSEP);
2398 vp = (Volume *) calloc(1, sizeof(Volume));
2399 osi_Assert(vp != NULL);
2400 vp->hashid = volumeId;
2401 vp->device = partp->device;
2402 vp->partition = partp;
2403 queue_Init(&vp->vnode_list);
2404 queue_Init(&vp->rx_call_list);
2405 #ifdef AFS_DEMAND_ATTACH_FS
2406 CV_INIT(&V_attachCV(vp), "vp attach", CV_DEFAULT, 0);
2407 #endif /* AFS_DEMAND_ATTACH_FS */
2410 /* attach2 is entered without any locks, and returns
2411 * with vol_glock_mutex held */
2412 vp = attach2(ec, volumeId, path, partp, vp, isbusy, mode, &checkedOut);
2414 if (VCanUseFSSYNC() && vp) {
2415 #ifdef AFS_DEMAND_ATTACH_FS
2416 if ((mode == V_VOLUPD) || (VolumeWriteable(vp) && (mode == V_CLONE))) {
2417 /* mark volume header as in use so that volser crashes lead to a
2418 * salvage attempt */
2419 VUpdateVolume_r(ec, vp, 0);
2421 /* for dafs, we should tell the fileserver, except for V_PEEK
2422 * where we know it is not necessary */
2423 if (mode == V_PEEK) {
2424 vp->needsPutBack = 0;
2426 vp->needsPutBack = VOL_PUTBACK;
2428 #else /* !AFS_DEMAND_ATTACH_FS */
2429 /* duplicate computation in fssync.c about whether the server
2430 * takes the volume offline or not. If the volume isn't
2431 * offline, we must not return it when we detach the volume,
2432 * or the server will abort */
2433 if (mode == V_READONLY || mode == V_PEEK
2434 || (!VolumeWriteable(vp) && (mode == V_CLONE || mode == V_DUMP)))
2435 vp->needsPutBack = 0;
2437 vp->needsPutBack = VOL_PUTBACK;
2438 #endif /* !AFS_DEMAND_ATTACH_FS */
2440 #ifdef FSSYNC_BUILD_CLIENT
2441 /* Only give back the vol to the fileserver if we checked it out; attach2
2442 * will set checkedOut only if we successfully checked it out from the
2444 if (VCanUseFSSYNC() && vp == NULL && checkedOut) {
2446 #ifdef AFS_DEMAND_ATTACH_FS
2447 /* If we couldn't attach but we scheduled a salvage, we already
2448 * notified the fileserver; don't online it now */
2449 if (*ec != VSALVAGING)
2450 #endif /* AFS_DEMAND_ATTACH_FS */
2451 FSYNC_VolOp(volumeId, partition, FSYNC_VOL_ON, 0, NULL);
2454 if (programType == fileServer && vp) {
2455 #ifdef AFS_DEMAND_ATTACH_FS
2457 * we can get here in cases where we don't "own"
2458 * the volume (e.g. volume owned by a utility).
2459 * short circuit around potential disk header races.
2461 if (V_attachState(vp) != VOL_STATE_ATTACHED) {
2465 VUpdateVolume_r(ec, vp, 0);
2467 Log("VAttachVolume: Error updating volume\n");
2472 if (VolumeWriteable(vp) && V_dontSalvage(vp) == 0) {
2473 #ifndef AFS_DEMAND_ATTACH_FS
2474 /* This is a hack: by temporarily setting the incore
2475 * dontSalvage flag ON, the volume will be put back on the
2476 * Update list (with dontSalvage OFF again). It will then
2477 * come back in N minutes with DONT_SALVAGE eventually
2478 * set. This is the way that volumes that have never had
2479 * it set get it set; or that volumes that have been
2480 * offline without DONT SALVAGE having been set also
2481 * eventually get it set */
2482 V_dontSalvage(vp) = DONT_SALVAGE;
2483 #endif /* !AFS_DEMAND_ATTACH_FS */
2484 VAddToVolumeUpdateList_r(ec, vp);
2486 Log("VAttachVolume: Error adding volume to update list\n");
2493 Log("VOnline: volume %u (%s) attached and online\n", V_id(vp),
2498 if (VRequiresPartLock()) {
2499 VUnlockPartition_r(partition);
2502 #ifdef AFS_DEMAND_ATTACH_FS
2503 /* attach failed; make sure we're in error state */
2504 if (vp && !VIsErrorState(V_attachState(vp))) {
2505 VChangeState_r(vp, VOL_STATE_ERROR);
2507 #endif /* AFS_DEMAND_ATTACH_FS */
2514 #ifdef AFS_DEMAND_ATTACH_FS
2515 /* VAttachVolumeByVp_r
2517 * finish attaching a volume that is
2518 * in a less than fully attached state
2520 /* caller MUST hold a ref count on vp */
2522 VAttachVolumeByVp_r(Error * ec, Volume * vp, int mode)
2524 char name[VMAXPATHLEN];
2526 struct DiskPartition64 *partp;
2530 Volume * nvp = NULL;
2531 VolumeStats stats_save;
2535 /* volume utility should never call AttachByVp */
2536 osi_Assert(programType == fileServer);
2538 volumeId = vp->hashid;
2539 partp = vp->partition;
2540 VolumeExternalName_r(volumeId, name, sizeof(name));
2543 /* if another thread is performing a blocking op, wait */
2544 VWaitExclusiveState_r(vp);
2546 memcpy(&stats_save, &vp->stats, sizeof(VolumeStats));
2548 /* if it's already attached, see if we can return it */
2549 if (V_attachState(vp) == VOL_STATE_ATTACHED) {
2550 VGetVolumeByVp_r(ec, vp);
2551 if (V_inUse(vp) == fileServer) {
2554 if (vp->specialStatus == VBUSY)
2556 VDetachVolume_r(ec, vp);
2558 Log("VAttachVolume: Error detaching volume (%s)\n", name);
2564 /* pre-attach volume if it hasn't been done yet */
2566 (V_attachState(vp) == VOL_STATE_UNATTACHED) ||
2567 (V_attachState(vp) == VOL_STATE_DELETED) ||
2568 (V_attachState(vp) == VOL_STATE_ERROR)) {
2569 nvp = VPreAttachVolumeByVp_r(ec, partp, vp, volumeId);
2575 VCreateReservation_r(nvp);
2580 osi_Assert(vp != NULL);
2581 VChangeState_r(vp, VOL_STATE_ATTACHING);
2583 /* restore monotonically increasing stats */
2584 memcpy(&vp->stats, &stats_save, sizeof(VolumeStats));
2588 /* compute path to disk header */
2589 strcpy(path, VPartitionPath(partp));
2593 strcat(path, OS_DIRSEP);
2598 * NOTE: attach2 is entered without any locks, and returns
2599 * with vol_glock_mutex held */
2600 vp = attach2(ec, volumeId, path, partp, vp, isbusy, mode, &checkedOut);
2603 * the event that an error was encountered, or
2604 * the volume was not brought to an attached state
2605 * for any reason, skip to the end. We cannot
2606 * safely call VUpdateVolume unless we "own" it.
2610 (V_attachState(vp) != VOL_STATE_ATTACHED)) {
2614 VUpdateVolume_r(ec, vp, 0);
2616 Log("VAttachVolume: Error updating volume %u\n", vp->hashid);
2620 if (VolumeWriteable(vp) && V_dontSalvage(vp) == 0) {
2621 #ifndef AFS_DEMAND_ATTACH_FS
2622 /* This is a hack: by temporarily setting the incore
2623 * dontSalvage flag ON, the volume will be put back on the
2624 * Update list (with dontSalvage OFF again). It will then
2625 * come back in N minutes with DONT_SALVAGE eventually
2626 * set. This is the way that volumes that have never had
2627 * it set get it set; or that volumes that have been
2628 * offline without DONT SALVAGE having been set also
2629 * eventually get it set */
2630 V_dontSalvage(vp) = DONT_SALVAGE;
2631 #endif /* !AFS_DEMAND_ATTACH_FS */
2632 VAddToVolumeUpdateList_r(ec, vp);
2634 Log("VAttachVolume: Error adding volume %u to update list\n", vp->hashid);
2641 Log("VOnline: volume %u (%s) attached and online\n", V_id(vp),
2645 VCancelReservation_r(nvp);
2648 if (*ec && (*ec != VOFFLINE) && (*ec != VSALVAGE)) {
2649 if (vp && !VIsErrorState(V_attachState(vp))) {
2650 VChangeState_r(vp, VOL_STATE_ERROR);
2659 * lock a volume on disk (non-blocking).
2661 * @param[in] vp The volume to lock
2662 * @param[in] locktype READ_LOCK or WRITE_LOCK
2664 * @return operation status
2665 * @retval 0 success, lock was obtained
2666 * @retval EBUSY a conflicting lock was held by another process
2667 * @retval EIO error acquiring lock
2669 * @pre If we're in the fileserver, vp is in an exclusive state
2671 * @pre vp is not already locked
2674 VLockVolumeNB(Volume *vp, int locktype)
2678 osi_Assert(programType != fileServer || VIsExclusiveState(V_attachState(vp)));
2679 osi_Assert(!(V_attachFlags(vp) & VOL_LOCKED));
2681 code = VLockVolumeByIdNB(vp->hashid, vp->partition, locktype);
2683 V_attachFlags(vp) |= VOL_LOCKED;
2690 * unlock a volume on disk that was locked with VLockVolumeNB.
2692 * @param[in] vp volume to unlock
2694 * @pre If we're in the fileserver, vp is in an exclusive state
2696 * @pre vp has already been locked
2699 VUnlockVolume(Volume *vp)
2701 osi_Assert(programType != fileServer || VIsExclusiveState(V_attachState(vp)));
2702 osi_Assert((V_attachFlags(vp) & VOL_LOCKED));
2704 VUnlockVolumeById(vp->hashid, vp->partition);
2706 V_attachFlags(vp) &= ~VOL_LOCKED;
2708 #endif /* AFS_DEMAND_ATTACH_FS */
2711 * read in a vol header, possibly lock the vol header, and possibly check out
2712 * the vol header from the fileserver, as part of volume attachment.
2714 * @param[out] ec error code
2715 * @param[in] vp volume pointer object
2716 * @param[in] partp disk partition object of the attaching partition
2717 * @param[in] mode attachment mode such as V_VOLUPD, V_DUMP, etc (see
2719 * @param[in] peek 1 to just try to read in the volume header and make sure
2720 * we don't try to lock the vol, or check it out from
2721 * FSSYNC or anything like that; 0 otherwise, for 'normal'
2723 * @param[out] acheckedOut If we successfully checked-out the volume from
2724 * the fileserver (if we needed to), this is set
2725 * to 1, otherwise it is untouched.
2727 * @note As part of DAFS volume attachment, the volume header may be either
2728 * read- or write-locked to ensure mutual exclusion of certain volume
2729 * operations. In some cases in order to determine whether we need to
2730 * read- or write-lock the header, we need to read in the header to see
2731 * if the volume is RW or not. So, if we read in the header under a
2732 * read-lock and determine that we actually need a write-lock on the
2733 * volume header, this function will drop the read lock, acquire a write
2734 * lock, and read the header in again.
2737 attach_volume_header(Error *ec, Volume *vp, struct DiskPartition64 *partp,
2738 int mode, int peek, int *acheckedOut)
2740 struct VolumeDiskHeader diskHeader;
2741 struct VolumeHeader header;
2744 int lock_tries = 0, checkout_tries = 0;
2746 VolumeId volid = vp->hashid;
2747 #ifdef FSSYNC_BUILD_CLIENT
2748 int checkout, done_checkout = 0;
2749 #endif /* FSSYNC_BUILD_CLIENT */
2750 #ifdef AFS_DEMAND_ATTACH_FS
2751 int locktype = 0, use_locktype = -1;
2752 #endif /* AFS_DEMAND_ATTACH_FS */
2758 if (lock_tries > VOL_MAX_CHECKOUT_RETRIES) {
2759 Log("VAttachVolume: retried too many times trying to lock header for "
2760 "vol %lu part %s; giving up\n", afs_printable_uint32_lu(volid),
2761 VPartitionPath(partp));
2765 if (checkout_tries > VOL_MAX_CHECKOUT_RETRIES) {
2766 Log("VAttachVolume: retried too many times trying to checkout "
2767 "vol %lu part %s; giving up\n", afs_printable_uint32_lu(volid),
2768 VPartitionPath(partp));
2773 if (VReadVolumeDiskHeader(volid, partp, NULL)) {
2774 /* short-circuit the 'volume does not exist' case */
2779 #ifdef FSSYNC_BUILD_CLIENT
2780 checkout = !done_checkout;
2782 if (!peek && checkout && VMustCheckoutVolume(mode)) {
2784 memset(&res, 0, sizeof(res));
2786 if (FSYNC_VolOp(volid, partp->name, FSYNC_VOL_NEEDVOLUME, mode, &res)
2789 if (res.hdr.reason == FSYNC_SALVAGE) {
2790 Log("VAttachVolume: file server says volume %lu is salvaging\n",
2791 afs_printable_uint32_lu(volid));
2794 Log("VAttachVolume: attach of volume %lu apparently denied by file server\n",
2795 afs_printable_uint32_lu(volid));
2796 *ec = VNOVOL; /* XXXX */
2804 #ifdef AFS_DEMAND_ATTACH_FS
2805 if (use_locktype < 0) {
2806 /* don't know whether vol is RO or RW; assume it's RO and we can retry
2807 * if it turns out to be RW */
2808 locktype = VVolLockType(mode, 0);
2811 /* a previous try says we should use use_locktype to lock the volume,
2813 locktype = use_locktype;
2816 if (!peek && locktype) {
2817 code = VLockVolumeNB(vp, locktype);
2819 if (code == EBUSY) {
2820 Log("VAttachVolume: another program has vol %lu locked\n",
2821 afs_printable_uint32_lu(volid));
2823 Log("VAttachVolume: error %d trying to lock vol %lu\n",
2824 code, afs_printable_uint32_lu(volid));
2831 #endif /* AFS_DEMAND_ATTACH_FS */
2833 code = VReadVolumeDiskHeader(volid, partp, &diskHeader);
2843 DiskToVolumeHeader(&header, &diskHeader);
2845 IH_INIT(vp->vnodeIndex[vLarge].handle, partp->device, header.parent,
2846 header.largeVnodeIndex);
2847 IH_INIT(vp->vnodeIndex[vSmall].handle, partp->device, header.parent,
2848 header.smallVnodeIndex);
2849 IH_INIT(vp->diskDataHandle, partp->device, header.parent,
2851 IH_INIT(vp->linkHandle, partp->device, header.parent, header.linkTable);
2854 /* only need to do this once */
2856 GetVolumeHeader(vp);
2860 #if defined(AFS_DEMAND_ATTACH_FS) && defined(FSSYNC_BUILD_CLIENT)
2861 /* demand attach changes the V_PEEK mechanism
2863 * we can now suck the current disk data structure over
2864 * the fssync interface without going to disk
2866 * (technically, we don't need to restrict this feature
2867 * to demand attach fileservers. However, I'm trying
2868 * to limit the number of common code changes)
2870 if (VCanUseFSSYNC() && (mode == V_PEEK || peek)) {
2872 res.payload.len = sizeof(VolumeDiskData);
2873 res.payload.buf = &vp->header->diskstuff;
2875 if (FSYNC_VolOp(vp->hashid,
2877 FSYNC_VOL_QUERY_HDR,
2880 goto disk_header_loaded;
2883 #endif /* AFS_DEMAND_ATTACH_FS && FSSYNC_BUILD_CLIENT */
2884 (void)ReadHeader(ec, V_diskDataHandle(vp), (char *)&V_disk(vp),
2885 sizeof(V_disk(vp)), VOLUMEINFOMAGIC, VOLUMEINFOVERSION);
2887 #ifdef AFS_DEMAND_ATTACH_FS
2890 IncUInt64(&VStats.hdr_loads);
2891 IncUInt64(&vp->stats.hdr_loads);
2893 #endif /* AFS_DEMAND_ATTACH_FS */
2896 Log("VAttachVolume: Error reading diskDataHandle header for vol %lu; "
2897 "error=%u\n", afs_printable_uint32_lu(volid), *ec);
2901 #ifdef AFS_DEMAND_ATTACH_FS
2902 # ifdef FSSYNC_BUILD_CLIENT
2904 # endif /* FSSYNC_BUILD_CLIENT */
2906 /* if the lock type we actually used to lock the volume is different than
2907 * the lock type we should have used, retry with the lock type we should
2909 use_locktype = VVolLockType(mode, VolumeWriteable(vp));
2910 if (locktype != use_locktype) {
2914 #endif /* AFS_DEMAND_ATTACH_FS */
2919 #if defined(AFS_DEMAND_ATTACH_FS) && defined(FSSYNC_BUILD_CLIENT)
2920 if (!peek && *ec == 0 && retry == 0 && VMustCheckoutVolume(mode)) {
2922 code = FSYNC_VerifyCheckout(volid, partp->name, FSYNC_VOL_NEEDVOLUME, mode);
2924 if (code == SYNC_DENIED) {
2925 /* must retry checkout; fileserver no longer thinks we have
2931 } else if (code != SYNC_OK) {
2935 #endif /* AFS_DEMAND_ATTACH_FS && FSSYNC_BUILD_CLIENT */
2938 /* either we are going to be called again for a second pass, or we
2939 * encountered an error; clean up in either case */
2941 #ifdef AFS_DEMAND_ATTACH_FS
2942 if ((V_attachFlags(vp) & VOL_LOCKED)) {
2945 #endif /* AFS_DEMAND_ATTACH_FS */
2946 if (vp->linkHandle) {
2947 IH_RELEASE(vp->vnodeIndex[vLarge].handle);
2948 IH_RELEASE(vp->vnodeIndex[vSmall].handle);
2949 IH_RELEASE(vp->diskDataHandle);
2950 IH_RELEASE(vp->linkHandle);
2963 #ifdef AFS_DEMAND_ATTACH_FS
2965 attach_check_vop(Error *ec, VolumeId volid, struct DiskPartition64 *partp,
2966 Volume *vp, int *acheckedOut)
2970 if (vp->pending_vol_op) {
2974 if (vp->pending_vol_op->vol_op_state == FSSYNC_VolOpRunningUnknown) {
2976 code = VVolOpLeaveOnlineNoHeader_r(vp, vp->pending_vol_op);
2978 vp->pending_vol_op->vol_op_state = FSSYNC_VolOpRunningOnline;
2979 } else if (code == 0) {
2980 vp->pending_vol_op->vol_op_state = FSSYNC_VolOpRunningOffline;
2983 /* we need the vol header to determine if the volume can be
2984 * left online for the vop, so... get the header */
2988 /* attach header with peek=1 to avoid checking out the volume
2989 * or locking it; we just want the header info, we're not
2990 * messing with the volume itself at all */
2991 attach_volume_header(ec, vp, partp, V_PEEK, 1, acheckedOut);
2998 if (VVolOpLeaveOnline_r(vp, vp->pending_vol_op)) {
2999 vp->pending_vol_op->vol_op_state = FSSYNC_VolOpRunningOnline;
3001 vp->pending_vol_op->vol_op_state = FSSYNC_VolOpRunningOffline;
3004 /* make sure we grab a new vol header and re-open stuff on
3005 * actual attachment; we can't keep the data we grabbed, since
3006 * it was not done under a lock and thus not safe */
3007 FreeVolumeHeader(vp);
3008 VReleaseVolumeHandles_r(vp);
3011 /* see if the pending volume op requires exclusive access */
3012 switch (vp->pending_vol_op->vol_op_state) {
3013 case FSSYNC_VolOpPending:
3014 /* this should never happen */
3015 osi_Assert(vp->pending_vol_op->vol_op_state != FSSYNC_VolOpPending);
3018 case FSSYNC_VolOpRunningUnknown:
3019 /* this should never happen; we resolved 'unknown' above */
3020 osi_Assert(vp->pending_vol_op->vol_op_state != FSSYNC_VolOpRunningUnknown);
3023 case FSSYNC_VolOpRunningOffline:
3024 /* mark the volume down */
3026 VChangeState_r(vp, VOL_STATE_UNATTACHED);
3028 /* do not set V_offlineMessage here; we don't have ownership of
3029 * the volume (and probably do not have the header loaded), so we
3030 * can't alter the disk header */
3032 /* check to see if we should set the specialStatus flag */
3033 if (VVolOpSetVBusy_r(vp, vp->pending_vol_op)) {
3034 /* don't overwrite specialStatus if it was already set to
3035 * something else (e.g. VMOVED) */
3036 if (!vp->specialStatus) {
3037 vp->specialStatus = VBUSY;
3049 #endif /* AFS_DEMAND_ATTACH_FS */
3052 * volume attachment helper function.
3054 * @param[out] ec error code
3055 * @param[in] volumeId volume ID of the attaching volume
3056 * @param[in] path full path to the volume header .vol file
3057 * @param[in] partp disk partition object for the attaching partition
3058 * @param[in] vp volume object; vp->hashid, vp->device, vp->partition,
3059 * vp->vnode_list, vp->rx_call_list, and V_attachCV (for
3060 * DAFS) should already be initialized
3061 * @param[in] isbusy 1 if vp->specialStatus should be set to VBUSY; that is,
3062 * if there is a volume operation running for this volume
3063 * that should set the volume to VBUSY during its run. 0
3064 * otherwise. (see VVolOpSetVBusy_r)
3065 * @param[in] mode attachment mode such as V_VOLUPD, V_DUMP, etc (see
3067 * @param[out] acheckedOut If we successfully checked-out the volume from
3068 * the fileserver (if we needed to), this is set
3069 * to 1, otherwise it is 0.
3071 * @return pointer to the semi-attached volume pointer
3072 * @retval NULL an error occurred (check value of *ec)
3073 * @retval vp volume successfully attaching
3075 * @pre no locks held
3077 * @post VOL_LOCK held
3080 attach2(Error * ec, VolId volumeId, char *path, struct DiskPartition64 *partp,
3081 Volume * vp, int isbusy, int mode, int *acheckedOut)
3083 /* have we read in the header successfully? */
3084 int read_header = 0;
3086 #ifdef AFS_DEMAND_ATTACH_FS
3087 /* should we FreeVolume(vp) instead of VCheckFree(vp) in the error
3091 /* in the case of an error, to what state should the volume be
3093 VolState error_state = VOL_STATE_ERROR;
3094 #endif /* AFS_DEMAND_ATTACH_FS */
3098 vp->vnodeIndex[vLarge].handle = NULL;
3099 vp->vnodeIndex[vSmall].handle = NULL;
3100 vp->diskDataHandle = NULL;
3101 vp->linkHandle = NULL;
3105 #ifdef AFS_DEMAND_ATTACH_FS
3106 attach_check_vop(ec, volumeId, partp, vp, acheckedOut);
3108 attach_volume_header(ec, vp, partp, mode, 0, acheckedOut);
3111 attach_volume_header(ec, vp, partp, mode, 0, acheckedOut);
3112 #endif /* !AFS_DEMAND_ATTACH_FS */
3114 if (*ec == VNOVOL) {
3115 /* if the volume doesn't exist, skip straight to 'error' so we don't
3116 * request a salvage */
3118 goto error_notbroken;
3124 /* ensure that we don't override specialStatus if it was set to
3125 * something else (e.g. VMOVED) */
3126 if (isbusy && !vp->specialStatus) {
3127 vp->specialStatus = VBUSY;
3129 vp->shuttingDown = 0;
3130 vp->goingOffline = 0;
3132 #ifdef AFS_DEMAND_ATTACH_FS
3133 vp->stats.last_attach = FT_ApproxTime();
3134 vp->stats.attaches++;
3138 IncUInt64(&VStats.attaches);
3139 vp->cacheCheck = ++VolumeCacheCheck;
3140 /* just in case this ever rolls over */
3141 if (!vp->cacheCheck)
3142 vp->cacheCheck = ++VolumeCacheCheck;
3145 #ifdef AFS_DEMAND_ATTACH_FS
3146 V_attachFlags(vp) |= VOL_HDR_LOADED;
3147 vp->stats.last_hdr_load = vp->stats.last_attach;
3148 #endif /* AFS_DEMAND_ATTACH_FS */
3152 struct IndexFileHeader iHead;
3154 #if OPENAFS_VOL_STATS
3156 * We just read in the diskstuff part of the header. If the detailed
3157 * volume stats area has not yet been initialized, we should bzero the
3158 * area and mark it as initialized.
3160 if (!(V_stat_initialized(vp))) {
3161 memset((V_stat_area(vp)), 0, VOL_STATS_BYTES);
3162 V_stat_initialized(vp) = 1;
3164 #endif /* OPENAFS_VOL_STATS */
3166 (void)ReadHeader(ec, vp->vnodeIndex[vSmall].handle,
3167 (char *)&iHead, sizeof(iHead),
3168 SMALLINDEXMAGIC, SMALLINDEXVERSION);
3171 Log("VAttachVolume: Error reading smallVnode vol header %s; error=%u\n", path, *ec);
3176 struct IndexFileHeader iHead;
3178 (void)ReadHeader(ec, vp->vnodeIndex[vLarge].handle,
3179 (char *)&iHead, sizeof(iHead),
3180 LARGEINDEXMAGIC, LARGEINDEXVERSION);
3183 Log("VAttachVolume: Error reading largeVnode vol header %s; error=%u\n", path, *ec);
3187 #ifdef AFS_NAMEI_ENV
3189 struct versionStamp stamp;
3191 (void)ReadHeader(ec, V_linkHandle(vp), (char *)&stamp,
3192 sizeof(stamp), LINKTABLEMAGIC, LINKTABLEVERSION);
3195 Log("VAttachVolume: Error reading namei vol header %s; error=%u\n", path, *ec);
3198 #endif /* AFS_NAMEI_ENV */
3200 #if defined(AFS_DEMAND_ATTACH_FS)
3201 if (*ec && ((*ec != VOFFLINE) || (V_attachState(vp) != VOL_STATE_UNATTACHED))) {
3203 if (!VCanScheduleSalvage()) {
3204 Log("VAttachVolume: Error attaching volume %s; volume needs salvage; error=%u\n", path, *ec);
3206 VRequestSalvage_r(ec, vp, SALVSYNC_ERROR, VOL_SALVAGE_NO_OFFLINE);
3211 /* volume operation in progress */
3213 goto error_notbroken;
3215 #else /* AFS_DEMAND_ATTACH_FS */
3217 Log("VAttachVolume: Error attaching volume %s; volume needs salvage; error=%u\n", path, *ec);
3218 goto unlocked_error;
3220 #endif /* AFS_DEMAND_ATTACH_FS */
3222 if (V_needsSalvaged(vp)) {
3223 if (vp->specialStatus)
3224 vp->specialStatus = 0;
3226 #if defined(AFS_DEMAND_ATTACH_FS)
3227 if (!VCanScheduleSalvage()) {
3228 Log("VAttachVolume: volume salvage flag is ON for %s; volume needs salvage\n", path);
3230 VRequestSalvage_r(ec, vp, SALVSYNC_NEEDED, VOL_SALVAGE_NO_OFFLINE);
3233 #else /* AFS_DEMAND_ATTACH_FS */
3235 #endif /* AFS_DEMAND_ATTACH_FS */
3241 vp->nextVnodeUnique = V_uniquifier(vp);
3243 if (VShouldCheckInUse(mode) && V_inUse(vp) && VolumeWriteable(vp)) {
3244 if (!V_needsSalvaged(vp)) {
3245 V_needsSalvaged(vp) = 1;
3246 VUpdateVolume_r(ec, vp, 0);
3248 #if defined(AFS_DEMAND_ATTACH_FS)
3249 if (!VCanScheduleSalvage()) {
3250 Log("VAttachVolume: volume %s needs to be salvaged; not attached.\n", path);
3252 VRequestSalvage_r(ec, vp, SALVSYNC_NEEDED, VOL_SALVAGE_NO_OFFLINE);
3255 #else /* AFS_DEMAND_ATTACH_FS */
3256 Log("VAttachVolume: volume %s needs to be salvaged; not attached.\n", path);
3258 #endif /* AFS_DEMAND_ATTACH_FS */
3263 if (programType == fileServer && V_destroyMe(vp) == DESTROY_ME) {
3264 /* Only check destroyMe if we are the fileserver, since the
3265 * volserver et al sometimes need to work with volumes with
3266 * destroyMe set. Examples are 'temporary' volumes the
3267 * volserver creates, and when we create a volume (destroyMe
3268 * is set on creation; sometimes a separate volserver
3269 * transaction is created to clear destroyMe).
3272 #if defined(AFS_DEMAND_ATTACH_FS)
3273 /* schedule a salvage so the volume goes away on disk */
3274 VRequestSalvage_r(ec, vp, SALVSYNC_ERROR, VOL_SALVAGE_NO_OFFLINE);
3275 VChangeState_r(vp, VOL_STATE_ERROR);
3278 #endif /* AFS_DEMAND_ATTACH_FS */
3279 Log("VAttachVolume: volume %s is junk; it should be destroyed at next salvage\n", path);
3284 vp->vnodeIndex[vSmall].bitmap = vp->vnodeIndex[vLarge].bitmap = NULL;
3285 #ifndef BITMAP_LATER
3286 if (programType == fileServer && VolumeWriteable(vp)) {
3288 for (i = 0; i < nVNODECLASSES; i++) {
3289 VGetBitmap_r(ec, vp, i);
3291 #ifdef AFS_DEMAND_ATTACH_FS
3292 VRequestSalvage_r(ec, vp, SALVSYNC_ERROR, VOL_SALVAGE_NO_OFFLINE);
3294 #endif /* AFS_DEMAND_ATTACH_FS */
3295 Log("VAttachVolume: error getting bitmap for volume (%s)\n",
3301 #endif /* BITMAP_LATER */
3303 if (VInit >= 2 && V_needsCallback(vp)) {
3304 if (V_BreakVolumeCallbacks) {
3305 Log("VAttachVolume: Volume %lu was changed externally; breaking callbacks\n",
3306 afs_printable_uint32_lu(V_id(vp)));
3307 V_needsCallback(vp) = 0;
3309 (*V_BreakVolumeCallbacks) (V_id(vp));
3312 VUpdateVolume_r(ec, vp, 0);
3314 #ifdef FSSYNC_BUILD_CLIENT
3315 else if (VCanUseFSSYNC()) {
3316 afs_int32 fsync_code;
3318 V_needsCallback(vp) = 0;
3320 fsync_code = FSYNC_VolOp(V_id(vp), NULL, FSYNC_VOL_BREAKCBKS, FSYNC_WHATEVER, NULL);
3324 V_needsCallback(vp) = 1;
3325 Log("Error trying to tell the fileserver to break callbacks for "
3326 "changed volume %lu; error code %ld\n",
3327 afs_printable_uint32_lu(V_id(vp)),
3328 afs_printable_int32_ld(fsync_code));
3330 VUpdateVolume_r(ec, vp, 0);
3333 #endif /* FSSYNC_BUILD_CLIENT */
3336 Log("VAttachVolume: error %d clearing needsCallback on volume "
3337 "%lu; needs salvage\n", (int)*ec,
3338 afs_printable_uint32_lu(V_id(vp)));
3339 #ifdef AFS_DEMAND_ATTACH_FS
3340 VRequestSalvage_r(ec, vp, SALVSYNC_ERROR, VOL_SALVAGE_NO_OFFLINE);
3342 #else /* !AFS_DEMAND_ATTACH_FS */
3344 #endif /* !AFS_DEMAND_ATTACh_FS */
3349 if (programType == fileServer) {
3350 if (vp->specialStatus)
3351 vp->specialStatus = 0;
3352 if (V_blessed(vp) && V_inService(vp) && !V_needsSalvaged(vp)) {
3353 V_inUse(vp) = fileServer;
3354 V_offlineMessage(vp)[0] = '\0';
3358 #ifdef AFS_DEMAND_ATTACH_FS
3359 /* Put the vol into PREATTACHED state, so if someone tries to
3360 * access it again, we try to attach, see that we're not blessed,
3361 * and give a VNOVOL error again. Putting it into UNATTACHED state
3362 * would result in a VOFFLINE error instead. */
3363 error_state = VOL_STATE_PREATTACHED;
3364 #endif /* AFS_DEMAND_ATTACH_FS */
3366 /* mimic e.g. GetVolume errors */
3367 if (!V_blessed(vp)) {
3368 Log("Volume %lu offline: not blessed\n", afs_printable_uint32_lu(V_id(vp)));
3369 FreeVolumeHeader(vp);
3370 } else if (!V_inService(vp)) {
3371 Log("Volume %lu offline: not in service\n", afs_printable_uint32_lu(V_id(vp)));
3372 FreeVolumeHeader(vp);
3374 Log("Volume %lu offline: needs salvage\n", afs_printable_uint32_lu(V_id(vp)));
3376 #ifdef AFS_DEMAND_ATTACH_FS
3377 error_state = VOL_STATE_ERROR;
3378 /* see if we can recover */
3379 VRequestSalvage_r(ec, vp, SALVSYNC_NEEDED, 0 /*flags*/);
3382 #ifdef AFS_DEMAND_ATTACH_FS
3388 #ifdef AFS_DEMAND_ATTACH_FS
3389 if ((mode != V_PEEK) && (mode != V_SECRETLY))
3390 V_inUse(vp) = programType;
3391 #endif /* AFS_DEMAND_ATTACH_FS */
3392 V_checkoutMode(vp) = mode;
3395 AddVolumeToHashTable(vp, V_id(vp));
3396 #ifdef AFS_DEMAND_ATTACH_FS
3397 if (VCanUnlockAttached() && (V_attachFlags(vp) & VOL_LOCKED)) {
3400 if ((programType != fileServer) ||
3401 (V_inUse(vp) == fileServer)) {
3402 AddVolumeToVByPList_r(vp);
3404 VChangeState_r(vp, VOL_STATE_ATTACHED);
3406 VChangeState_r(vp, VOL_STATE_UNATTACHED);
3412 #ifndef AFS_DEMAND_ATTACH_FS
3418 #ifdef AFS_DEMAND_ATTACH_FS
3419 if (!VIsErrorState(V_attachState(vp))) {
3420 VChangeState_r(vp, error_state);
3422 #endif /* AFS_DEMAND_ATTACH_FS */
3425 VReleaseVolumeHandles_r(vp);
3429 #ifdef AFS_DEMAND_ATTACH_FS
3436 #else /* !AFS_DEMAND_ATTACH_FS */
3438 #endif /* !AFS_DEMAND_ATTACH_FS */
3442 /* Attach an existing volume.
3443 The volume also normally goes online at this time.
3444 An offline volume must be reattached to make it go online.
3448 VAttachVolume(Error * ec, VolumeId volumeId, int mode)
3452 retVal = VAttachVolume_r(ec, volumeId, mode);
3458 VAttachVolume_r(Error * ec, VolumeId volumeId, int mode)
3461 VGetVolumePath(ec, volumeId, &part, &name);
3465 vp = VGetVolume_r(&error, volumeId);
3467 osi_Assert(V_inUse(vp) == 0);
3468 VDetachVolume_r(ec, vp);
3472 return VAttachVolumeByName_r(ec, part, name, mode);
3475 /* Increment a reference count to a volume, sans context swaps. Requires
3476 * possibly reading the volume header in from the disk, since there's
3477 * an invariant in the volume package that nUsers>0 ==> vp->header is valid.
3479 * N.B. This call can fail if we can't read in the header!! In this case
3480 * we still guarantee we won't context swap, but the ref count won't be
3481 * incremented (otherwise we'd violate the invariant).
3483 /* NOTE: with the demand attach fileserver extensions, the global lock
3484 * is dropped within VHold */
3485 #ifdef AFS_DEMAND_ATTACH_FS
3487 VHold_r(Volume * vp)
3491 VCreateReservation_r(vp);
3492 VWaitExclusiveState_r(vp);
3494 LoadVolumeHeader(&error, vp);
3496 VCancelReservation_r(vp);
3500 VCancelReservation_r(vp);
3503 #else /* AFS_DEMAND_ATTACH_FS */
3505 VHold_r(Volume * vp)
3509 LoadVolumeHeader(&error, vp);
3515 #endif /* AFS_DEMAND_ATTACH_FS */
3517 /**** volume timeout-related stuff ****/
3519 #ifdef AFS_PTHREAD_ENV
3521 static struct timespec *shutdown_timeout;
3522 static pthread_once_t shutdown_timeout_once = PTHREAD_ONCE_INIT;
3525 VTimedOut(const struct timespec *ts)
3530 if (ts->tv_sec == 0) {
3531 /* short-circuit; this will have always timed out */
3535 code = gettimeofday(&tv, NULL);
3537 Log("Error %d from gettimeofday, assuming we have not timed out\n", errno);
3538 /* assume no timeout; failure mode is we just wait longer than normal
3539 * instead of returning errors when we shouldn't */
3543 if (tv.tv_sec < ts->tv_sec ||
3544 (tv.tv_sec == ts->tv_sec && tv.tv_usec*1000 < ts->tv_nsec)) {
3553 * Calculate an absolute timeout.
3555 * @param[out] ts A timeout that is "timeout" seconds from now, if we return
3556 * NULL, the memory is not touched
3557 * @param[in] timeout How long the timeout should be from now
3559 * @return timeout to use
3560 * @retval NULL no timeout; wait forever
3561 * @retval non-NULL the given value for "ts"
3565 static struct timespec *
3566 VCalcTimeout(struct timespec *ts, afs_int32 timeout)
3576 ts->tv_sec = ts->tv_nsec = 0;
3580 code = gettimeofday(&now, NULL);
3582 Log("Error %d from gettimeofday, falling back to 'forever' timeout\n", errno);
3586 ts->tv_sec = now.tv_sec + timeout;
3587 ts->tv_nsec = now.tv_usec * 1000;
3593 * Initialize the shutdown_timeout global.
3596 VShutdownTimeoutInit(void)
3598 struct timespec *ts;
3600 ts = malloc(sizeof(*ts));
3602 shutdown_timeout = VCalcTimeout(ts, vol_opts.offline_shutdown_timeout);
3604 if (!shutdown_timeout) {
3610 * Figure out the timeout that should be used for waiting for offline volumes.
3612 * @param[out] ats Storage space for a local timeout value if needed
3614 * @return The timeout value that should be used
3615 * @retval NULL No timeout; wait forever for offlining volumes
3616 * @retval non-NULL A pointer to the absolute time that should be used as
3617 * the deadline for waiting for offlining volumes.
3619 * @note If we return non-NULL, the pointer we return may or may not be the
3622 static const struct timespec *
3623 VOfflineTimeout(struct timespec *ats)
3625 if (vol_shutting_down) {
3626 osi_Assert(pthread_once(&shutdown_timeout_once, VShutdownTimeoutInit) == 0);
3627 return shutdown_timeout;
3629 return VCalcTimeout(ats, vol_opts.offline_timeout);
3633 #else /* AFS_PTHREAD_ENV */
3635 /* Waiting a certain amount of time for offlining volumes is not supported
3636 * for LWP due to a lack of primitives. So, we never time out */
3637 # define VTimedOut(x) (0)
3638 # define VOfflineTimeout(x) (NULL)
3640 #endif /* !AFS_PTHREAD_ENV */
3648 retVal = VHold_r(vp);
3655 VIsGoingOffline_r(struct Volume *vp)
3659 if (vp->goingOffline) {
3660 if (vp->specialStatus) {
3661 code = vp->specialStatus;
3662 } else if (V_inService(vp) == 0 || V_blessed(vp) == 0) {
3673 * Tell the caller if a volume is waiting to go offline.
3675 * @param[in] vp The volume we want to know about
3677 * @return volume status
3678 * @retval 0 volume is not waiting to go offline, go ahead and use it
3679 * @retval nonzero volume is waiting to offline, and give the returned code
3680 * as an error to anyone accessing the volume
3682 * @pre VOL_LOCK is NOT held
3683 * @pre caller holds a heavyweight reference on vp
3686 VIsGoingOffline(struct Volume *vp)
3691 code = VIsGoingOffline_r(vp);
3698 * Register an RX call with a volume.
3700 * @param[inout] ec Error code; if unset when passed in, may be set if
3701 * the volume starts going offline
3702 * @param[out] client_ec @see GetVolume
3703 * @param[in] vp Volume struct
3704 * @param[in] cbv VCallByVol struct containing the RX call to register
3706 * @pre VOL_LOCK held
3707 * @pre caller holds heavy ref on vp
3712 VRegisterCall_r(Error *ec, Error *client_ec, Volume *vp, struct VCallByVol *cbv)
3715 #ifdef AFS_DEMAND_ATTACH_FS
3717 /* just in case the volume started going offline after we got the
3718 * reference to it... otherwise, if the volume started going
3719 * offline right at the end of GetVolume(), we might race with the
3720 * RX call scanner, and return success and add our cbv to the
3721 * rx_call_list _after_ the scanner has scanned the list. */
3722 *ec = VIsGoingOffline_r(vp);
3728 while (V_attachState(vp) == VOL_STATE_SCANNING_RXCALLS) {
3729 VWaitStateChange_r(vp);
3731 #endif /* AFS_DEMAND_ATTACH_FS */
3733 queue_Prepend(&vp->rx_call_list, cbv);
3738 * Deregister an RX call with a volume.
3740 * @param[in] vp Volume struct
3741 * @param[in] cbv VCallByVol struct containing the RX call to deregister
3743 * @pre VOL_LOCK held
3744 * @pre caller holds heavy ref on vp
3749 VDeregisterCall_r(Volume *vp, struct VCallByVol *cbv)
3751 if (cbv && queue_IsOnQueue(cbv)) {
3752 #ifdef AFS_DEMAND_ATTACH_FS
3753 while (V_attachState(vp) == VOL_STATE_SCANNING_RXCALLS) {
3754 VWaitStateChange_r(vp);
3756 #endif /* AFS_DEMAND_ATTACH_FS */
3762 /***************************************************/
3763 /* get and put volume routines */
3764 /***************************************************/
3767 * put back a heavyweight reference to a volume object.
3769 * @param[in] vp volume object pointer
3771 * @pre VOL_LOCK held
3773 * @post heavyweight volume reference put back.
3774 * depending on state, volume may have been taken offline,
3775 * detached, salvaged, freed, etc.
3777 * @internal volume package internal use only
3780 VPutVolume_r(Volume * vp)
3782 osi_Assert(--vp->nUsers >= 0);
3783 if (vp->nUsers == 0) {
3785 ReleaseVolumeHeader(vp->header);
3786 #ifdef AFS_DEMAND_ATTACH_FS
3787 if (!VCheckDetach(vp)) {
3791 #else /* AFS_DEMAND_ATTACH_FS */
3793 #endif /* AFS_DEMAND_ATTACH_FS */
3798 VPutVolume(Volume * vp)
3806 * Puts a volume reference obtained with VGetVolumeWithCall.
3808 * @param[in] vp Volume struct
3809 * @param[in] cbv VCallByVol struct given to VGetVolumeWithCall, or NULL if none
3811 * @pre VOL_LOCK is NOT held
3814 VPutVolumeWithCall(Volume *vp, struct VCallByVol *cbv)
3817 VDeregisterCall_r(vp, cbv);
3822 /* Get a pointer to an attached volume. The pointer is returned regardless
3823 of whether or not the volume is in service or on/off line. An error
3824 code, however, is returned with an indication of the volume's status */
3826 VGetVolume(Error * ec, Error * client_ec, VolId volumeId)
3830 retVal = GetVolume(ec, client_ec, volumeId, NULL, 0);
3836 * Get a volume reference associated with an RX call.
3838 * @param[out] ec @see GetVolume
3839 * @param[out] client_ec @see GetVolume
3840 * @param[in] volumeId @see GetVolume
3841 * @param[in] ts How long to wait for going-offline volumes (absolute time).
3842 * If NULL, wait forever. If ts->tv_sec == 0, return immediately
3843 * with an error if the volume is going offline.
3844 * @param[in] cbv Contains an RX call to be associated with this volume
3845 * reference. This call may be interrupted if the volume is
3846 * requested to go offline while we hold a ref on it. Give NULL
3847 * to not associate an RX call with this reference.
3849 * @return @see GetVolume
3851 * @note for LWP builds, ts must be NULL
3853 * @note A reference obtained with this function MUST be put back with
3854 * VPutVolumeWithCall
3857 VGetVolumeWithCall(Error * ec, Error * client_ec, VolId volumeId,
3858 const struct timespec *ts, struct VCallByVol *cbv)
3862 retVal = GetVolume(ec, client_ec, volumeId, NULL, ts);
3863 VRegisterCall_r(ec, client_ec, retVal, cbv);
3869 VGetVolume_r(Error * ec, VolId volumeId)
3871 return GetVolume(ec, NULL, volumeId, NULL, NULL);
3874 /* try to get a volume we've previously looked up */
3875 /* for demand attach fs, caller MUST NOT hold a ref count on vp */
3877 VGetVolumeByVp_r(Error * ec, Volume * vp)
3879 return GetVolume(ec, NULL, vp->hashid, vp, NULL);
3883 * private interface for getting a volume handle
3885 * @param[out] ec error code (0 if no error)
3886 * @param[out] client_ec wire error code to be given to clients
3887 * @param[in] volumeId ID of the volume we want
3888 * @param[in] hint optional hint for hash lookups, or NULL
3889 * @param[in] timeout absolute deadline for waiting for the volume to go
3890 * offline, if it is going offline. NULL to wait forever.
3892 * @return a volume handle for the specified volume
3893 * @retval NULL an error occurred, or the volume is in such a state that
3894 * we cannot load a header or return any volume struct
3896 * @note for DAFS, caller must NOT hold a ref count on 'hint'
3898 * @note 'timeout' is only checked if the volume is actually going offline; so
3899 * if you pass timeout->tv_sec = 0, this will exhibit typical
3900 * nonblocking behavior.
3902 * @note for LWP builds, 'timeout' must be NULL
3905 GetVolume(Error * ec, Error * client_ec, VolId volumeId, Volume * hint,
3906 const struct timespec *timeout)
3909 /* pull this profiling/debugging code out of regular builds */
3911 #define VGET_CTR_INC(x) x++
3912 unsigned short V0 = 0, V1 = 0, V2 = 0, V3 = 0, V5 = 0, V6 =
3913 0, V7 = 0, V8 = 0, V9 = 0;
3914 unsigned short V10 = 0, V11 = 0, V12 = 0, V13 = 0, V14 = 0, V15 = 0;
3916 #define VGET_CTR_INC(x)
3918 #ifdef AFS_DEMAND_ATTACH_FS
3919 Volume *avp, * rvp = hint;
3923 * if VInit is zero, the volume package dynamic
3924 * data structures have not been initialized yet,
3925 * and we must immediately return an error
3931 *client_ec = VOFFLINE;
3936 #ifdef AFS_DEMAND_ATTACH_FS
3938 VCreateReservation_r(rvp);
3940 #endif /* AFS_DEMAND_ATTACH_FS */
3948 vp = VLookupVolume_r(ec, volumeId, vp);
3954 #ifdef AFS_DEMAND_ATTACH_FS
3955 if (rvp && (rvp != vp)) {
3956 /* break reservation on old vp */
3957 VCancelReservation_r(rvp);
3960 #endif /* AFS_DEMAND_ATTACH_FS */
3966 /* Until we have reached an initialization level of 2
3967 * we don't know whether this volume exists or not.
3968 * We can't sleep and retry later because before a volume
3969 * is attached, the caller tries to get it first. Just
3970 * return VOFFLINE and the caller can choose whether to
3971 * retry the command or not. */
3981 IncUInt64(&VStats.hdr_gets);
3983 #ifdef AFS_DEMAND_ATTACH_FS
3984 /* block if someone else is performing an exclusive op on this volume */
3987 VCreateReservation_r(rvp);
3989 VWaitExclusiveState_r(vp);
3991 /* short circuit with VNOVOL in the following circumstances:
3994 * - VOL_STATE_SHUTTING_DOWN
3996 if ((V_attachState(vp) == VOL_STATE_ERROR) ||
3997 (V_attachState(vp) == VOL_STATE_SHUTTING_DOWN) ||
3998 (V_attachState(vp) == VOL_STATE_GOING_OFFLINE)) {
4005 * short circuit with VOFFLINE for VOL_STATE_UNATTACHED and
4006 * VNOVOL for VOL_STATE_DELETED
4008 if ((V_attachState(vp) == VOL_STATE_UNATTACHED) ||
4009 (V_attachState(vp) == VOL_STATE_DELETED)) {
4010 if (vp->specialStatus) {
4011 *ec = vp->specialStatus;
4012 } else if (V_attachState(vp) == VOL_STATE_DELETED) {
4021 /* allowable states:
4028 if (vp->salvage.requested) {
4029 VUpdateSalvagePriority_r(vp);
4032 if (V_attachState(vp) == VOL_STATE_PREATTACHED) {
4033 if (vp->specialStatus) {
4034 *ec = vp->specialStatus;
4038 avp = VAttachVolumeByVp_r(ec, vp, 0);
4041 /* VAttachVolumeByVp_r can return a pointer
4042 * != the vp passed to it under certain
4043 * conditions; make sure we don't leak
4044 * reservations if that happens */
4046 VCancelReservation_r(rvp);
4048 VCreateReservation_r(rvp);
4058 if (!vp->pending_vol_op) {
4073 if (VIsSalvaging(vp) || (*ec == VSALVAGING)) {
4075 /* see CheckVnode() in afsfileprocs.c for an explanation
4076 * of this error code logic */
4077 afs_uint32 now = FT_ApproxTime();
4078 if ((vp->stats.last_salvage + (10 * 60)) >= now) {
4081 *client_ec = VRESTARTING;
4089 if (VIsErrorState(V_attachState(vp))) {
4090 /* make sure we don't take a vp in VOL_STATE_ERROR state and use
4091 * it, or transition it out of that state */
4100 * this test MUST happen after VAttachVolymeByVp, so vol_op_state is
4101 * not VolOpRunningUnknown (attach2 would have converted it to Online
4105 /* only valid before/during demand attachment */
4106 osi_Assert(!vp->pending_vol_op || vp->pending_vol_op->vol_op_state != FSSYNC_VolOpRunningUnknown);
4108 /* deny getvolume due to running mutually exclusive vol op */
4109 if (vp->pending_vol_op && vp->pending_vol_op->vol_op_state==FSSYNC_VolOpRunningOffline) {
4111 * volume cannot remain online during this volume operation.
4114 if (vp->specialStatus) {
4116 * special status codes outrank normal VOFFLINE code
4118 *ec = vp->specialStatus;
4120 *client_ec = vp->specialStatus;
4124 /* see CheckVnode() in afsfileprocs.c for an explanation
4125 * of this error code logic */
4126 afs_uint32 now = FT_ApproxTime();
4127 if ((vp->stats.last_vol_op + (10 * 60)) >= now) {
4130 *client_ec = VRESTARTING;
4135 VChangeState_r(vp, VOL_STATE_UNATTACHED);
4136 FreeVolumeHeader(vp);
4140 #endif /* AFS_DEMAND_ATTACH_FS */
4142 LoadVolumeHeader(ec, vp);
4145 /* Only log the error if it was a totally unexpected error. Simply
4146 * a missing inode is likely to be caused by the volume being deleted */
4147 if (errno != ENXIO || LogLevel)
4148 Log("Volume %u: couldn't reread volume header\n",
4150 #ifdef AFS_DEMAND_ATTACH_FS
4151 if (VCanScheduleSalvage()) {
4152 VRequestSalvage_r(ec, vp, SALVSYNC_ERROR, 0 /*flags*/);
4157 #else /* AFS_DEMAND_ATTACH_FS */
4160 #endif /* AFS_DEMAND_ATTACH_FS */
4165 if (vp->shuttingDown) {
4172 if (programType == fileServer) {
4174 if (vp->goingOffline) {
4175 if (timeout && VTimedOut(timeout)) {
4176 /* we've timed out; don't wait for the vol */
4179 #ifdef AFS_DEMAND_ATTACH_FS
4180 /* wait for the volume to go offline */
4181 if (V_attachState(vp) == VOL_STATE_GOING_OFFLINE) {
4182 VTimedWaitStateChange_r(vp, timeout, NULL);
4184 #elif defined(AFS_PTHREAD_ENV)
4185 VOL_CV_TIMEDWAIT(&vol_put_volume_cond, timeout, NULL);
4186 #else /* AFS_PTHREAD_ENV */
4187 /* LWP has no timed wait, so the caller better not be
4189 osi_Assert(!timeout);
4190 LWP_WaitProcess(VPutVolume);
4191 #endif /* AFS_PTHREAD_ENV */
4195 if (vp->specialStatus) {
4197 *ec = vp->specialStatus;
4198 } else if (V_inService(vp) == 0 || V_blessed(vp) == 0) {
4201 } else if (V_inUse(vp) == 0 || vp->goingOffline) {
4212 #ifdef AFS_DEMAND_ATTACH_FS
4213 /* if no error, bump nUsers */
4216 VLRU_UpdateAccess_r(vp);
4219 VCancelReservation_r(rvp);
4222 if (client_ec && !*client_ec) {
4225 #else /* AFS_DEMAND_ATTACH_FS */
4226 /* if no error, bump nUsers */
4233 #endif /* AFS_DEMAND_ATTACH_FS */
4236 osi_Assert(vp || *ec);
4241 /***************************************************/
4242 /* Volume offline/detach routines */
4243 /***************************************************/
4245 /* caller MUST hold a heavyweight ref on vp */
4246 #ifdef AFS_DEMAND_ATTACH_FS
4248 VTakeOffline_r(Volume * vp)
4252 osi_Assert(vp->nUsers > 0);
4253 osi_Assert(programType == fileServer);
4255 VCreateReservation_r(vp);
4256 VWaitExclusiveState_r(vp);
4258 vp->goingOffline = 1;
4259 V_needsSalvaged(vp) = 1;
4261 VRequestSalvage_r(&error, vp, SALVSYNC_ERROR, 0);
4262 VCancelReservation_r(vp);
4264 #else /* AFS_DEMAND_ATTACH_FS */
4266 VTakeOffline_r(Volume * vp)
4268 osi_Assert(vp->nUsers > 0);
4269 osi_Assert(programType == fileServer);
4271 vp->goingOffline = 1;
4272 V_needsSalvaged(vp) = 1;
4274 #endif /* AFS_DEMAND_ATTACH_FS */
4277 VTakeOffline(Volume * vp)
4285 * force a volume offline.
4287 * @param[in] vp volume object pointer
4288 * @param[in] flags flags (see note below)
4290 * @note the flag VOL_FORCEOFF_NOUPDATE is a recursion control flag
4291 * used when VUpdateVolume_r needs to call VForceOffline_r
4292 * (which in turn would normally call VUpdateVolume_r)
4294 * @see VUpdateVolume_r
4296 * @pre VOL_LOCK must be held.
4297 * for DAFS, caller must hold ref.
4299 * @note for DAFS, it _is safe_ to call this function from an
4302 * @post needsSalvaged flag is set.
4303 * for DAFS, salvage is requested.
4304 * no further references to the volume through the volume
4305 * package will be honored.
4306 * all file descriptor and vnode caches are invalidated.
4308 * @warning this is a heavy-handed interface. it results in
4309 * a volume going offline regardless of the current
4310 * reference count state.
4312 * @internal volume package internal use only
4315 VForceOffline_r(Volume * vp, int flags)
4319 #ifdef AFS_DEMAND_ATTACH_FS
4320 VChangeState_r(vp, VOL_STATE_ERROR);
4325 strcpy(V_offlineMessage(vp),
4326 "Forced offline due to internal error: volume needs to be salvaged");
4327 Log("Volume %u forced offline: it needs salvaging!\n", V_id(vp));
4330 vp->goingOffline = 0;
4331 V_needsSalvaged(vp) = 1;
4332 if (!(flags & VOL_FORCEOFF_NOUPDATE)) {
4333 VUpdateVolume_r(&error, vp, VOL_UPDATE_NOFORCEOFF);
4336 #ifdef AFS_DEMAND_ATTACH_FS
4337 VRequestSalvage_r(&error, vp, SALVSYNC_ERROR, 0 /*flags*/);
4338 #endif /* AFS_DEMAND_ATTACH_FS */
4340 #ifdef AFS_PTHREAD_ENV
4341 CV_BROADCAST(&vol_put_volume_cond);
4342 #else /* AFS_PTHREAD_ENV */
4343 LWP_NoYieldSignal(VPutVolume);
4344 #endif /* AFS_PTHREAD_ENV */
4346 VReleaseVolumeHandles_r(vp);
4350 * force a volume offline.
4352 * @param[in] vp volume object pointer
4354 * @see VForceOffline_r
4357 VForceOffline(Volume * vp)
4360 VForceOffline_r(vp, 0);
4365 * Iterate over the RX calls associated with a volume, and interrupt them.
4367 * @param[in] vp The volume whose RX calls we want to scan
4369 * @pre VOL_LOCK held
4372 VScanCalls_r(struct Volume *vp)
4374 struct VCallByVol *cbv, *ncbv;
4376 #ifdef AFS_DEMAND_ATTACH_FS
4377 VolState state_save;
4380 if (queue_IsEmpty(&vp->rx_call_list))
4381 return; /* no calls to interrupt */
4382 if (!vol_opts.interrupt_rxcall)
4383 return; /* we have no function with which to interrupt calls */
4384 err = VIsGoingOffline_r(vp);
4386 return; /* we're not going offline anymore */
4388 #ifdef AFS_DEMAND_ATTACH_FS
4389 VWaitExclusiveState_r(vp);
4390 state_save = VChangeState_r(vp, VOL_STATE_SCANNING_RXCALLS);
4392 #endif /* AFS_DEMAND_ATTACH_FS */
4394 for(queue_Scan(&vp->rx_call_list, cbv, ncbv, VCallByVol)) {
4396 struct rx_peer *peer;
4398 peer = rx_PeerOf(rx_ConnectionOf(cbv->call));
4400 Log("Offlining volume %lu while client %s:%u is trying to read "
4401 "from it; kicking client off with error %ld\n",
4402 (long unsigned) vp->hashid,
4403 afs_inet_ntoa_r(rx_HostOf(peer), hoststr),
4404 (unsigned) ntohs(rx_PortOf(peer)),
4407 (*vol_opts.interrupt_rxcall) (cbv->call, err);
4410 #ifdef AFS_DEMAND_ATTACH_FS
4412 VChangeState_r(vp, state_save);
4413 #endif /* AFS_DEMAND_ATTACH_FS */
4416 #ifdef AFS_DEMAND_ATTACH_FS
4418 * Wait for a vp to go offline.
4420 * @param[out] ec 1 if a salvage on the volume has been requested and
4421 * salvok == 0, 0 otherwise
4422 * @param[in] vp The volume to wait for
4423 * @param[in] salvok If 0, we return immediately with *ec = 1 if the volume
4424 * has been requested to salvage. Otherwise we keep waiting
4425 * until the volume has gone offline.
4427 * @pre VOL_LOCK held
4428 * @pre caller holds a lightweight ref on vp
4433 VWaitForOfflineByVp_r(Error *ec, struct Volume *vp, int salvok)
4435 struct timespec timeout_ts;
4436 const struct timespec *ts;
4439 ts = VOfflineTimeout(&timeout_ts);
4443 while (!VIsOfflineState(V_attachState(vp)) && !timedout) {
4444 if (!salvok && vp->salvage.requested) {
4448 VTimedWaitStateChange_r(vp, ts, &timedout);
4451 /* we didn't time out, so the volume must be offline, so we're done */
4455 /* If we got here, we timed out waiting for the volume to go offline.
4456 * Kick off the accessing RX calls and wait again */
4460 while (!VIsOfflineState(V_attachState(vp))) {
4461 if (!salvok && vp->salvage.requested) {
4466 VWaitStateChange_r(vp);
4470 #else /* AFS_DEMAND_ATTACH_FS */
4473 * Wait for a volume to go offline.
4475 * @pre VOL_LOCK held
4477 * @note non-DAFS only (for DAFS, use @see WaitForOfflineByVp_r)
4480 VWaitForOffline_r(Error *ec, VolumeId volid)
4483 const struct timespec *ts;
4484 #ifdef AFS_PTHREAD_ENV
4485 struct timespec timeout_ts;
4488 ts = VOfflineTimeout(&timeout_ts);
4490 vp = GetVolume(ec, NULL, volid, NULL, ts);
4492 /* error occurred so bad that we can't even get a vp; we have no
4493 * information on the vol so we don't know whether to wait, so just
4497 if (!VIsGoingOffline_r(vp)) {
4498 /* volume is no longer going offline, so we're done */
4503 /* If we got here, we timed out waiting for the volume to go offline.
4504 * Kick off the accessing RX calls and wait again */
4510 vp = VGetVolume_r(ec, volid);
4512 /* In case it was reattached... */
4516 #endif /* !AFS_DEMAND_ATTACH_FS */
4518 /* The opposite of VAttachVolume. The volume header is written to disk, with
4519 the inUse bit turned off. A copy of the header is maintained in memory,
4520 however (which is why this is VOffline, not VDetach).
4523 VOffline_r(Volume * vp, char *message)
4526 #ifndef AFS_DEMAND_ATTACH_FS
4527 VolumeId vid = V_id(vp);
4530 osi_Assert(programType != volumeUtility && programType != volumeServer);
4535 if (V_offlineMessage(vp)[0] == '\0')
4536 strncpy(V_offlineMessage(vp), message, sizeof(V_offlineMessage(vp)));
4537 V_offlineMessage(vp)[sizeof(V_offlineMessage(vp)) - 1] = '\0';
4539 vp->goingOffline = 1;
4540 #ifdef AFS_DEMAND_ATTACH_FS
4541 VChangeState_r(vp, VOL_STATE_GOING_OFFLINE);
4542 VCreateReservation_r(vp);
4544 VWaitForOfflineByVp_r(&error, vp, 1);
4545 VCancelReservation_r(vp);
4546 #else /* AFS_DEMAND_ATTACH_FS */
4548 VWaitForOffline_r(&error, vid);
4549 #endif /* AFS_DEMAND_ATTACH_FS */
4552 #ifdef AFS_DEMAND_ATTACH_FS
4554 * Take a volume offline in order to perform a volume operation.
4556 * @param[inout] ec address in which to store error code
4557 * @param[in] vp volume object pointer
4558 * @param[in] message volume offline status message
4561 * - VOL_LOCK is held
4562 * - caller MUST hold a heavyweight ref on vp
4565 * - volume is taken offline
4566 * - if possible, volume operation is promoted to running state
4567 * - on failure, *ec is set to nonzero
4569 * @note Although this function does not return any value, it may
4570 * still fail to promote our pending volume operation to
4571 * a running state. Any caller MUST check the value of *ec,
4572 * and MUST NOT blindly assume success.
4574 * @warning if the caller does not hold a lightweight ref on vp,
4575 * then it MUST NOT reference vp after this function
4576 * returns to the caller.
4578 * @internal volume package internal use only
4581 VOfflineForVolOp_r(Error *ec, Volume *vp, char *message)
4584 osi_Assert(vp->pending_vol_op);
4590 if (V_offlineMessage(vp)[0] == '\0')
4591 strncpy(V_offlineMessage(vp), message, sizeof(V_offlineMessage(vp)));
4592 V_offlineMessage(vp)[sizeof(V_offlineMessage(vp)) - 1] = '\0';
4594 vp->goingOffline = 1;
4595 VChangeState_r(vp, VOL_STATE_GOING_OFFLINE);
4596 VCreateReservation_r(vp);
4599 if (vp->pending_vol_op->com.programType != salvageServer) {
4600 /* do not give corrupted volumes to the volserver */
4605 VWaitForOfflineByVp_r(ec, vp, salvok);
4607 VCancelReservation_r(vp);
4609 #endif /* AFS_DEMAND_ATTACH_FS */
4612 VOffline(Volume * vp, char *message)
4615 VOffline_r(vp, message);
4619 /* This gets used for the most part by utility routines that don't want
4620 * to keep all the volume headers around. Generally, the file server won't
4621 * call this routine, because then the offline message in the volume header
4622 * (or other information) won't be available to clients. For NAMEI, also
4623 * close the file handles. However, the fileserver does call this during
4624 * an attach following a volume operation.
4627 VDetachVolume_r(Error * ec, Volume * vp)
4629 #ifdef FSSYNC_BUILD_CLIENT
4631 struct DiskPartition64 *tpartp;
4632 int notifyServer = 0;
4633 int useDone = FSYNC_VOL_ON;
4635 if (VCanUseFSSYNC()) {
4636 notifyServer = vp->needsPutBack;
4637 if (V_destroyMe(vp) == DESTROY_ME)
4638 useDone = FSYNC_VOL_LEAVE_OFF;
4639 # ifdef AFS_DEMAND_ATTACH_FS
4640 else if (!V_blessed(vp) || !V_inService(vp))
4641 useDone = FSYNC_VOL_LEAVE_OFF;
4644 # ifdef AFS_DEMAND_ATTACH_FS
4645 if (V_needsSalvaged(vp)) {
4647 VRequestSalvage_r(ec, vp, SALVSYNC_NEEDED, 0);
4650 tpartp = vp->partition;
4652 #endif /* FSSYNC_BUILD_CLIENT */
4654 *ec = 0; /* always "succeeds" */
4655 DeleteVolumeFromHashTable(vp);
4656 vp->shuttingDown = 1;
4657 #ifdef AFS_DEMAND_ATTACH_FS
4658 DeleteVolumeFromVByPList_r(vp);
4660 VChangeState_r(vp, VOL_STATE_SHUTTING_DOWN);
4662 if (programType != fileServer)
4664 #endif /* AFS_DEMAND_ATTACH_FS */
4666 /* Will be detached sometime in the future--this is OK since volume is offline */
4668 /* XXX the following code should really be moved to VCheckDetach() since the volume
4669 * is not technically detached until the refcounts reach zero
4671 #ifdef FSSYNC_BUILD_CLIENT
4672 if (VCanUseFSSYNC() && notifyServer) {
4673 if (notifyServer == VOL_PUTBACK_DELETE) {
4674 /* Only send FSYNC_VOL_DONE if the volume was actually deleted.
4675 * volserver code will set needsPutBack to VOL_PUTBACK_DELETE
4676 * to signify a deleted volume. */
4677 useDone = FSYNC_VOL_DONE;
4680 * Note: The server is not notified in the case of a bogus volume
4681 * explicitly to make it possible to create a volume, do a partial
4682 * restore, then abort the operation without ever putting the volume
4683 * online. This is essential in the case of a volume move operation
4684 * between two partitions on the same server. In that case, there
4685 * would be two instances of the same volume, one of them bogus,
4686 * which the file server would attempt to put on line
4688 FSYNC_VolOp(volume, tpartp->name, useDone, 0, NULL);
4689 /* XXX this code path is only hit by volume utilities, thus
4690 * V_BreakVolumeCallbacks will always be NULL. if we really
4691 * want to break callbacks in this path we need to use FSYNC_VolOp() */
4693 /* Dettaching it so break all callbacks on it */
4694 if (V_BreakVolumeCallbacks) {
4695 Log("volume %u detached; breaking all call backs\n", volume);
4696 (*V_BreakVolumeCallbacks) (volume);
4700 #endif /* FSSYNC_BUILD_CLIENT */
4704 VDetachVolume(Error * ec, Volume * vp)
4707 VDetachVolume_r(ec, vp);
4712 /***************************************************/
4713 /* Volume fd/inode handle closing routines */
4714 /***************************************************/
4716 /* For VDetachVolume, we close all cached file descriptors, but keep
4717 * the Inode handles in case we need to read from a busy volume.
4719 /* for demand attach, caller MUST hold ref count on vp */
4721 VCloseVolumeHandles_r(Volume * vp)
4723 #ifdef AFS_DEMAND_ATTACH_FS
4724 VolState state_save;
4726 state_save = VChangeState_r(vp, VOL_STATE_OFFLINING);
4731 DFlushVolume(vp->hashid);
4733 #ifdef AFS_DEMAND_ATTACH_FS
4737 /* DAFS: VCloseVnodeFiles_r drops the glock internally */
4738 VCloseVnodeFiles_r(vp);
4740 #ifdef AFS_DEMAND_ATTACH_FS
4744 /* Too time consuming and unnecessary for the volserver */
4745 if (programType == fileServer) {
4746 IH_CONDSYNC(vp->vnodeIndex[vLarge].handle);
4747 IH_CONDSYNC(vp->vnodeIndex[vSmall].handle);
4748 IH_CONDSYNC(vp->diskDataHandle);
4750 IH_CONDSYNC(vp->linkHandle);
4751 #endif /* AFS_NT40_ENV */
4754 IH_REALLYCLOSE(vp->vnodeIndex[vLarge].handle);
4755 IH_REALLYCLOSE(vp->vnodeIndex[vSmall].handle);
4756 IH_REALLYCLOSE(vp->diskDataHandle);
4757 IH_REALLYCLOSE(vp->linkHandle);
4759 #ifdef AFS_DEMAND_ATTACH_FS
4760 if ((V_attachFlags(vp) & VOL_LOCKED)) {
4765 VChangeState_r(vp, state_save);
4769 /* For both VForceOffline and VOffline, we close all relevant handles.
4770 * For VOffline, if we re-attach the volume, the files may possible be
4771 * different than before.
4773 /* for demand attach, caller MUST hold a ref count on vp */
4775 VReleaseVolumeHandles_r(Volume * vp)
4777 #ifdef AFS_DEMAND_ATTACH_FS
4778 VolState state_save;
4780 state_save = VChangeState_r(vp, VOL_STATE_DETACHING);
4785 DFlushVolume(vp->hashid);
4787 #ifdef AFS_DEMAND_ATTACH_FS
4791 VReleaseVnodeFiles_r(vp); /* DAFS: releases the glock internally */
4793 #ifdef AFS_DEMAND_ATTACH_FS
4797 /* Too time consuming and unnecessary for the volserver */
4798 if (programType == fileServer) {
4799 IH_CONDSYNC(vp->vnodeIndex[vLarge].handle);
4800 IH_CONDSYNC(vp->vnodeIndex[vSmall].handle);
4801 IH_CONDSYNC(vp->diskDataHandle);
4803 IH_CONDSYNC(vp->linkHandle);
4804 #endif /* AFS_NT40_ENV */
4807 IH_RELEASE(vp->vnodeIndex[vLarge].handle);
4808 IH_RELEASE(vp->vnodeIndex[vSmall].handle);
4809 IH_RELEASE(vp->diskDataHandle);
4810 IH_RELEASE(vp->linkHandle);
4812 #ifdef AFS_DEMAND_ATTACH_FS
4813 if ((V_attachFlags(vp) & VOL_LOCKED)) {
4818 VChangeState_r(vp, state_save);
4823 /***************************************************/
4824 /* Volume write and fsync routines */
4825 /***************************************************/
4828 VUpdateVolume_r(Error * ec, Volume * vp, int flags)
4830 #ifdef AFS_DEMAND_ATTACH_FS
4831 VolState state_save;
4833 if (flags & VOL_UPDATE_WAIT) {
4834 VCreateReservation_r(vp);
4835 VWaitExclusiveState_r(vp);
4840 if (programType == fileServer)
4842 (V_inUse(vp) ? V_nextVnodeUnique(vp) +
4843 200 : V_nextVnodeUnique(vp));
4845 #ifdef AFS_DEMAND_ATTACH_FS
4846 state_save = VChangeState_r(vp, VOL_STATE_UPDATING);
4850 WriteVolumeHeader_r(ec, vp);
4852 #ifdef AFS_DEMAND_ATTACH_FS
4854 VChangeState_r(vp, state_save);
4855 if (flags & VOL_UPDATE_WAIT) {
4856 VCancelReservation_r(vp);
4861 Log("VUpdateVolume: error updating volume header, volume %u (%s)\n",
4862 V_id(vp), V_name(vp));
4863 /* try to update on-disk header,
4864 * while preventing infinite recursion */
4865 if (!(flags & VOL_UPDATE_NOFORCEOFF)) {
4866 VForceOffline_r(vp, VOL_FORCEOFF_NOUPDATE);
4872 VUpdateVolume(Error * ec, Volume * vp)
4875 VUpdateVolume_r(ec, vp, VOL_UPDATE_WAIT);
4880 VSyncVolume_r(Error * ec, Volume * vp, int flags)
4884 #ifdef AFS_DEMAND_ATTACH_FS
4885 VolState state_save;
4888 if (flags & VOL_SYNC_WAIT) {
4889 VUpdateVolume_r(ec, vp, VOL_UPDATE_WAIT);
4891 VUpdateVolume_r(ec, vp, 0);
4894 #ifdef AFS_DEMAND_ATTACH_FS
4895 state_save = VChangeState_r(vp, VOL_STATE_UPDATING);
4898 fdP = IH_OPEN(V_diskDataHandle(vp));
4899 osi_Assert(fdP != NULL);
4900 code = FDH_SYNC(fdP);
4901 osi_Assert(code == 0);
4903 #ifdef AFS_DEMAND_ATTACH_FS
4905 VChangeState_r(vp, state_save);
4911 VSyncVolume(Error * ec, Volume * vp)
4914 VSyncVolume_r(ec, vp, VOL_SYNC_WAIT);
4919 /***************************************************/
4920 /* Volume dealloaction routines */
4921 /***************************************************/
4923 #ifdef AFS_DEMAND_ATTACH_FS
4925 FreeVolume(Volume * vp)
4927 /* free the heap space, iff it's safe.
4928 * otherwise, pull it out of the hash table, so it
4929 * will get deallocated when all refs to it go away */
4930 if (!VCheckFree(vp)) {
4931 DeleteVolumeFromHashTable(vp);
4932 DeleteVolumeFromVByPList_r(vp);
4934 /* make sure we invalidate the header cache entry */
4935 FreeVolumeHeader(vp);
4938 #endif /* AFS_DEMAND_ATTACH_FS */
4941 ReallyFreeVolume(Volume * vp)
4946 #ifdef AFS_DEMAND_ATTACH_FS
4948 VChangeState_r(vp, VOL_STATE_FREED);
4949 if (vp->pending_vol_op)
4950 free(vp->pending_vol_op);
4951 #endif /* AFS_DEMAND_ATTACH_FS */
4952 for (i = 0; i < nVNODECLASSES; i++)
4953 if (vp->vnodeIndex[i].bitmap)
4954 free(vp->vnodeIndex[i].bitmap);
4955 FreeVolumeHeader(vp);
4956 #ifndef AFS_DEMAND_ATTACH_FS
4957 DeleteVolumeFromHashTable(vp);
4958 #endif /* AFS_DEMAND_ATTACH_FS */
4962 /* check to see if we should shutdown this volume
4963 * returns 1 if volume was freed, 0 otherwise */
4964 #ifdef AFS_DEMAND_ATTACH_FS
4966 VCheckDetach(Volume * vp)
4971 if (vp->nUsers || vp->nWaiters)
4974 if (vp->shuttingDown) {
4976 if ((programType != fileServer) &&
4977 (V_inUse(vp) == programType) &&
4978 ((V_checkoutMode(vp) == V_VOLUPD) ||
4979 (V_checkoutMode(vp) == V_SECRETLY) ||
4980 ((V_checkoutMode(vp) == V_CLONE) &&
4981 (VolumeWriteable(vp))))) {
4983 VUpdateVolume_r(&ec, vp, VOL_UPDATE_NOFORCEOFF);
4985 Log("VCheckDetach: volume header update for volume %u "
4986 "failed with errno %d\n", vp->hashid, errno);
4989 VReleaseVolumeHandles_r(vp);
4991 ReallyFreeVolume(vp);
4992 if (programType == fileServer) {
4993 CV_BROADCAST(&vol_put_volume_cond);
4998 #else /* AFS_DEMAND_ATTACH_FS */
5000 VCheckDetach(Volume * vp)
5008 if (vp->shuttingDown) {
5010 if ((programType != fileServer) &&
5011 (V_inUse(vp) == programType) &&
5012 ((V_checkoutMode(vp) == V_VOLUPD) ||
5013 (V_checkoutMode(vp) == V_SECRETLY) ||
5014 ((V_checkoutMode(vp) == V_CLONE) &&
5015 (VolumeWriteable(vp))))) {
5017 VUpdateVolume_r(&ec, vp, VOL_UPDATE_NOFORCEOFF);
5019 Log("VCheckDetach: volume header update for volume %u failed with errno %d\n",
5023 VReleaseVolumeHandles_r(vp);
5024 ReallyFreeVolume(vp);
5025 if (programType == fileServer) {
5026 #if defined(AFS_PTHREAD_ENV)
5027 CV_BROADCAST(&vol_put_volume_cond);
5028 #else /* AFS_PTHREAD_ENV */
5029 LWP_NoYieldSignal(VPutVolume);
5030 #endif /* AFS_PTHREAD_ENV */
5035 #endif /* AFS_DEMAND_ATTACH_FS */
5037 /* check to see if we should offline this volume
5038 * return 1 if volume went offline, 0 otherwise */
5039 #ifdef AFS_DEMAND_ATTACH_FS
5041 VCheckOffline(Volume * vp)
5045 if (vp->goingOffline && !vp->nUsers) {
5047 osi_Assert(programType == fileServer);
5048 osi_Assert((V_attachState(vp) != VOL_STATE_ATTACHED) &&
5049 (V_attachState(vp) != VOL_STATE_FREED) &&
5050 (V_attachState(vp) != VOL_STATE_PREATTACHED) &&
5051 (V_attachState(vp) != VOL_STATE_UNATTACHED) &&
5052 (V_attachState(vp) != VOL_STATE_DELETED));
5056 * VOL_STATE_GOING_OFFLINE
5057 * VOL_STATE_SHUTTING_DOWN
5058 * VIsErrorState(V_attachState(vp))
5059 * VIsExclusiveState(V_attachState(vp))
5062 VCreateReservation_r(vp);
5063 VChangeState_r(vp, VOL_STATE_OFFLINING);
5066 /* must clear the goingOffline flag before we drop the glock */
5067 vp->goingOffline = 0;
5072 /* perform async operations */
5073 VUpdateVolume_r(&error, vp, 0);
5074 VCloseVolumeHandles_r(vp);
5077 if (V_offlineMessage(vp)[0]) {
5078 Log("VOffline: Volume %lu (%s) is now offline (%s)\n",
5079 afs_printable_uint32_lu(V_id(vp)), V_name(vp),
5080 V_offlineMessage(vp));
5082 Log("VOffline: Volume %lu (%s) is now offline\n",
5083 afs_printable_uint32_lu(V_id(vp)), V_name(vp));
5087 /* invalidate the volume header cache entry */
5088 FreeVolumeHeader(vp);
5090 /* if nothing changed state to error or salvaging,
5091 * drop state to unattached */
5092 if (!VIsErrorState(V_attachState(vp))) {
5093 VChangeState_r(vp, VOL_STATE_UNATTACHED);
5095 VCancelReservation_r(vp);
5096 /* no usage of vp is safe beyond this point */
5100 #else /* AFS_DEMAND_ATTACH_FS */
5102 VCheckOffline(Volume * vp)
5106 if (vp->goingOffline && !vp->nUsers) {
5108 osi_Assert(programType == fileServer);
5111 vp->goingOffline = 0;
5113 VUpdateVolume_r(&error, vp, 0);
5114 VCloseVolumeHandles_r(vp);
5116 if (V_offlineMessage(vp)[0]) {
5117 Log("VOffline: Volume %lu (%s) is now offline (%s)\n",
5118 afs_printable_uint32_lu(V_id(vp)), V_name(vp),
5119 V_offlineMessage(vp));
5121 Log("VOffline: Volume %lu (%s) is now offline\n",
5122 afs_printable_uint32_lu(V_id(vp)), V_name(vp));
5125 FreeVolumeHeader(vp);
5126 #ifdef AFS_PTHREAD_ENV
5127 CV_BROADCAST(&vol_put_volume_cond);
5128 #else /* AFS_PTHREAD_ENV */
5129 LWP_NoYieldSignal(VPutVolume);
5130 #endif /* AFS_PTHREAD_ENV */
5134 #endif /* AFS_DEMAND_ATTACH_FS */
5136 /***************************************************/
5137 /* demand attach fs ref counting routines */
5138 /***************************************************/
5140 #ifdef AFS_DEMAND_ATTACH_FS
5141 /* the following two functions handle reference counting for
5142 * asynchronous operations on volume structs.
5144 * their purpose is to prevent a VDetachVolume or VShutdown
5145 * from free()ing the Volume struct during an async i/o op */
5147 /* register with the async volume op ref counter */
5148 /* VCreateReservation_r moved into inline code header because it
5149 * is now needed in vnode.c -- tkeiser 11/20/2007
5153 * decrement volume-package internal refcount.
5155 * @param vp volume object pointer
5157 * @internal volume package internal use only
5160 * @arg VOL_LOCK is held
5161 * @arg lightweight refcount held
5163 * @post volume waiters refcount is decremented; volume may
5164 * have been deallocated/shutdown/offlined/salvaged/
5165 * whatever during the process
5167 * @warning once you have tossed your last reference (you can acquire
5168 * lightweight refs recursively) it is NOT SAFE to reference
5169 * a volume object pointer ever again
5171 * @see VCreateReservation_r
5173 * @note DEMAND_ATTACH_FS only
5176 VCancelReservation_r(Volume * vp)
5178 osi_Assert(--vp->nWaiters >= 0);
5179 if (vp->nWaiters == 0) {
5181 if (!VCheckDetach(vp)) {
5188 /* check to see if we should free this volume now
5189 * return 1 if volume was freed, 0 otherwise */
5191 VCheckFree(Volume * vp)
5194 if ((vp->nUsers == 0) &&
5195 (vp->nWaiters == 0) &&
5196 !(V_attachFlags(vp) & (VOL_IN_HASH |
5200 ReallyFreeVolume(vp);
5205 #endif /* AFS_DEMAND_ATTACH_FS */
5208 /***************************************************/
5209 /* online volume operations routines */
5210 /***************************************************/
5212 #ifdef AFS_DEMAND_ATTACH_FS
5214 * register a volume operation on a given volume.
5216 * @param[in] vp volume object
5217 * @param[in] vopinfo volume operation info object
5219 * @pre VOL_LOCK is held
5221 * @post volume operation info object attached to volume object.
5222 * volume operation statistics updated.
5224 * @note by "attached" we mean a copy of the passed in object is made
5226 * @internal volume package internal use only
5229 VRegisterVolOp_r(Volume * vp, FSSYNC_VolOp_info * vopinfo)
5231 FSSYNC_VolOp_info * info;
5233 /* attach a vol op info node to the volume struct */
5234 info = (FSSYNC_VolOp_info *) malloc(sizeof(FSSYNC_VolOp_info));
5235 osi_Assert(info != NULL);
5236 memcpy(info, vopinfo, sizeof(FSSYNC_VolOp_info));
5237 vp->pending_vol_op = info;
5240 vp->stats.last_vol_op = FT_ApproxTime();
5241 vp->stats.vol_ops++;
5242 IncUInt64(&VStats.vol_ops);
5248 * deregister the volume operation attached to this volume.
5250 * @param[in] vp volume object pointer
5252 * @pre VOL_LOCK is held
5254 * @post the volume operation info object is detached from the volume object
5256 * @internal volume package internal use only
5259 VDeregisterVolOp_r(Volume * vp)
5261 if (vp->pending_vol_op) {
5262 free(vp->pending_vol_op);
5263 vp->pending_vol_op = NULL;
5267 #endif /* AFS_DEMAND_ATTACH_FS */
5270 * determine whether it is safe to leave a volume online during
5271 * the volume operation described by the vopinfo object.
5273 * @param[in] vp volume object
5274 * @param[in] vopinfo volume operation info object
5276 * @return whether it is safe to leave volume online
5277 * @retval 0 it is NOT SAFE to leave the volume online
5278 * @retval 1 it is safe to leave the volume online during the operation
5281 * @arg VOL_LOCK is held
5282 * @arg disk header attached to vp (heavyweight ref on vp will guarantee
5283 * this condition is met)
5285 * @internal volume package internal use only
5288 VVolOpLeaveOnline_r(Volume * vp, FSSYNC_VolOp_info * vopinfo)
5290 return (vopinfo->vol_op_state == FSSYNC_VolOpRunningOnline ||
5291 (vopinfo->com.command == FSYNC_VOL_NEEDVOLUME &&
5292 (vopinfo->com.reason == V_READONLY ||
5293 (!VolumeWriteable(vp) &&
5294 (vopinfo->com.reason == V_CLONE ||
5295 vopinfo->com.reason == V_DUMP)))));
5299 * same as VVolOpLeaveOnline_r, but does not require a volume with an attached
5302 * @param[in] vp volume object
5303 * @param[in] vopinfo volume operation info object
5305 * @return whether it is safe to leave volume online
5306 * @retval 0 it is NOT SAFE to leave the volume online
5307 * @retval 1 it is safe to leave the volume online during the operation
5308 * @retval -1 unsure; volume header is required in order to know whether or
5309 * not is is safe to leave the volume online
5311 * @pre VOL_LOCK is held
5313 * @internal volume package internal use only
5316 VVolOpLeaveOnlineNoHeader_r(Volume * vp, FSSYNC_VolOp_info * vopinfo)
5318 /* follow the logic in VVolOpLeaveOnline_r; this is the same, except
5319 * assume that we don't know VolumeWriteable; return -1 if the answer
5320 * depends on VolumeWriteable */
5322 if (vopinfo->vol_op_state == FSSYNC_VolOpRunningOnline) {
5325 if (vopinfo->com.command == FSYNC_VOL_NEEDVOLUME &&
5326 vopinfo->com.reason == V_READONLY) {
5330 if (vopinfo->com.command == FSYNC_VOL_NEEDVOLUME &&
5331 (vopinfo->com.reason == V_CLONE ||
5332 vopinfo->com.reason == V_DUMP)) {
5334 /* must know VolumeWriteable */
5341 * determine whether VBUSY should be set during this volume operation.
5343 * @param[in] vp volume object
5344 * @param[in] vopinfo volume operation info object
5346 * @return whether VBUSY should be set
5347 * @retval 0 VBUSY does NOT need to be set
5348 * @retval 1 VBUSY SHOULD be set
5350 * @pre VOL_LOCK is held
5352 * @internal volume package internal use only
5355 VVolOpSetVBusy_r(Volume * vp, FSSYNC_VolOp_info * vopinfo)
5357 return ((vopinfo->com.command == FSYNC_VOL_OFF &&
5358 vopinfo->com.reason == FSYNC_SALVAGE) ||
5359 (vopinfo->com.command == FSYNC_VOL_NEEDVOLUME &&
5360 (vopinfo->com.reason == V_CLONE ||
5361 vopinfo->com.reason == V_DUMP)));
5365 /***************************************************/
5366 /* online salvager routines */
5367 /***************************************************/
5368 #if defined(AFS_DEMAND_ATTACH_FS)
5371 * offline a volume to let it be salvaged.
5373 * @param[in] vp Volume to offline
5375 * @return whether we offlined the volume successfully
5376 * @retval 0 volume was not offlined
5377 * @retval 1 volume is now offline
5379 * @note This is similar to VCheckOffline, but slightly different. We do not
5380 * deal with vp->goingOffline, and we try to avoid touching the volume
5381 * header except just to set needsSalvaged
5383 * @pre VOL_LOCK held
5384 * @pre vp->nUsers == 0
5385 * @pre V_attachState(vp) == VOL_STATE_SALVAGE_REQ
5388 VOfflineForSalvage_r(struct Volume *vp)
5392 VCreateReservation_r(vp);
5393 VWaitExclusiveState_r(vp);
5395 if (vp->nUsers || V_attachState(vp) == VOL_STATE_SALVAGING) {
5396 /* Someone's using the volume, or someone got to scheduling the salvage
5397 * before us. I don't think either of these should be possible, as we
5398 * should gain no new heavyweight references while we're trying to
5399 * salvage, but just to be sure... */
5400 VCancelReservation_r(vp);
5404 VChangeState_r(vp, VOL_STATE_OFFLINING);
5408 V_needsSalvaged(vp) = 1;
5409 /* ignore error; updating needsSalvaged is just best effort */
5410 VUpdateVolume_r(&error, vp, VOL_UPDATE_NOFORCEOFF);
5412 VCloseVolumeHandles_r(vp);
5414 FreeVolumeHeader(vp);
5416 /* volume has been effectively offlined; we can mark it in the SALVAGING
5417 * state now, which lets FSSYNC give it away */
5418 VChangeState_r(vp, VOL_STATE_SALVAGING);
5420 VCancelReservation_r(vp);
5426 * check whether a salvage needs to be performed on this volume.
5428 * @param[in] vp pointer to volume object
5430 * @return status code
5431 * @retval 0 no salvage scheduled
5432 * @retval 1 a salvage has been scheduled with the salvageserver
5434 * @pre VOL_LOCK is held
5436 * @post if salvage request flag is set and nUsers and nWaiters are zero,
5437 * then a salvage will be requested
5439 * @note this is one of the event handlers called by VCancelReservation_r
5441 * @note the caller must check if the volume needs to be freed after calling
5442 * this; the volume may not have any references or be on any lists after
5443 * we return, and we do not free it
5445 * @see VCancelReservation_r
5447 * @internal volume package internal use only.
5450 VCheckSalvage(Volume * vp)
5453 #if defined(SALVSYNC_BUILD_CLIENT) || defined(FSSYNC_BUILD_CLIENT)
5456 if (!vp->salvage.requested) {
5460 /* prevent recursion; some of the code below creates and removes
5461 * lightweight refs, which can call VCheckSalvage */
5462 if (vp->salvage.scheduling) {
5465 vp->salvage.scheduling = 1;
5467 if (V_attachState(vp) == VOL_STATE_SALVAGE_REQ) {
5468 if (!VOfflineForSalvage_r(vp)) {
5469 vp->salvage.scheduling = 0;
5474 if (vp->salvage.requested) {
5475 VScheduleSalvage_r(vp);
5478 vp->salvage.scheduling = 0;
5479 #endif /* SALVSYNC_BUILD_CLIENT || FSSYNC_BUILD_CLIENT */
5484 * request volume salvage.
5486 * @param[out] ec computed client error code
5487 * @param[in] vp volume object pointer
5488 * @param[in] reason reason code (passed to salvageserver via SALVSYNC)
5489 * @param[in] flags see flags note below
5492 * VOL_SALVAGE_NO_OFFLINE do not need to wait to offline the volume; it has
5493 * not been fully attached
5495 * @pre VOL_LOCK is held.
5497 * @post volume state is changed.
5498 * for fileserver, salvage will be requested once refcount reaches zero.
5500 * @return operation status code
5501 * @retval 0 volume salvage will occur
5502 * @retval 1 volume salvage could not be scheduled
5506 * @note in the fileserver, this call does not synchronously schedule a volume
5507 * salvage. rather, it sets volume state so that when volume refcounts
5508 * reach zero, a volume salvage will occur. by "refcounts", we mean both
5509 * nUsers and nWaiters must be zero.
5511 * @internal volume package internal use only.
5514 VRequestSalvage_r(Error * ec, Volume * vp, int reason, int flags)
5518 * for DAFS volume utilities that are not supposed to schedule salvages,
5519 * just transition to error state instead
5521 if (!VCanScheduleSalvage()) {
5522 VChangeState_r(vp, VOL_STATE_ERROR);
5527 if (programType != fileServer && !VCanUseFSSYNC()) {
5528 VChangeState_r(vp, VOL_STATE_ERROR);
5533 if (!vp->salvage.requested) {
5534 vp->salvage.requested = 1;
5535 vp->salvage.reason = reason;
5536 vp->stats.last_salvage = FT_ApproxTime();
5538 /* Note that it is not possible for us to reach this point if a
5539 * salvage is already running on this volume (even if the fileserver
5540 * was restarted during the salvage). If a salvage were running, the
5541 * salvager would have write-locked the volume header file, so when
5542 * we tried to lock the volume header, the lock would have failed,
5543 * and we would have failed during attachment prior to calling
5544 * VRequestSalvage. So we know that we can schedule salvages without
5545 * fear of a salvage already running for this volume. */
5547 if (vp->stats.salvages < SALVAGE_COUNT_MAX) {
5549 /* if we don't need to offline the volume, we can go directly
5550 * to SALVAGING. SALVAGING says the volume is offline and is
5551 * either salvaging or ready to be handed to the salvager.
5552 * SALVAGE_REQ says that we want to salvage the volume, but we
5553 * are waiting for it to go offline first. */
5554 if (flags & VOL_SALVAGE_NO_OFFLINE) {
5555 VChangeState_r(vp, VOL_STATE_SALVAGING);
5557 VChangeState_r(vp, VOL_STATE_SALVAGE_REQ);
5558 if (vp->nUsers == 0) {
5559 /* normally VOfflineForSalvage_r would be called from
5560 * PutVolume et al when nUsers reaches 0, but if
5561 * it's already 0, just do it ourselves, since PutVolume
5562 * isn't going to get called */
5563 VOfflineForSalvage_r(vp);
5566 /* If we are non-fileserver, we're telling the fileserver to
5567 * salvage the vol, so we don't need to give it back separately. */
5568 vp->needsPutBack = 0;
5572 Log("VRequestSalvage: volume %u online salvaged too many times; forced offline.\n", vp->hashid);
5574 /* make sure neither VScheduleSalvage_r nor
5575 * VUpdateSalvagePriority_r try to schedule another salvage */
5576 vp->salvage.requested = vp->salvage.scheduled = 0;
5578 VChangeState_r(vp, VOL_STATE_ERROR);
5587 * update salvageserver scheduling priority for a volume.
5589 * @param[in] vp pointer to volume object
5591 * @return operation status
5593 * @retval 1 request denied, or SALVSYNC communications failure
5595 * @pre VOL_LOCK is held.
5597 * @post in-core salvage priority counter is incremented. if at least
5598 * SALVAGE_PRIO_UPDATE_INTERVAL seconds have elapsed since the
5599 * last SALVSYNC_RAISEPRIO request, we contact the salvageserver
5600 * to update its priority queue. if no salvage is scheduled,
5601 * this function is a no-op.
5603 * @note DAFS fileserver only
5605 * @note this should be called whenever a VGetVolume fails due to a
5606 * pending salvage request
5608 * @todo should set exclusive state and drop glock around salvsync call
5610 * @internal volume package internal use only.
5613 VUpdateSalvagePriority_r(Volume * vp)
5617 #ifdef SALVSYNC_BUILD_CLIENT
5622 now = FT_ApproxTime();
5624 /* update the salvageserver priority queue occasionally so that
5625 * frequently requested volumes get moved to the head of the queue
5627 if ((vp->salvage.scheduled) &&
5628 (vp->stats.last_salvage_req < (now-SALVAGE_PRIO_UPDATE_INTERVAL))) {
5629 code = SALVSYNC_SalvageVolume(vp->hashid,
5630 VPartitionPath(vp->partition),
5635 vp->stats.last_salvage_req = now;
5636 if (code != SYNC_OK) {
5640 #endif /* SALVSYNC_BUILD_CLIENT */
5645 #if defined(SALVSYNC_BUILD_CLIENT) || defined(FSSYNC_BUILD_CLIENT)
5647 /* A couple of little helper functions. These return true if we tried to
5648 * use this mechanism to schedule a salvage, false if we haven't tried.
5649 * If we did try a salvage then the results are contained in code.
5653 try_SALVSYNC(Volume *vp, char *partName, int *code) {
5654 #ifdef SALVSYNC_BUILD_CLIENT
5655 if (VCanUseSALVSYNC()) {
5656 Log("Scheduling salvage for volume %lu on part %s over SALVSYNC\n",
5657 afs_printable_uint32_lu(vp->hashid), partName);
5659 /* can't use V_id() since there's no guarantee
5660 * we have the disk data header at this point */
5661 *code = SALVSYNC_SalvageVolume(vp->hashid,
5674 try_FSSYNC(Volume *vp, char *partName, int *code) {
5675 #ifdef FSSYNC_BUILD_CLIENT
5676 if (VCanUseFSSYNC()) {
5677 Log("Scheduling salvage for volume %lu on part %s over FSSYNC\n",
5678 afs_printable_uint32_lu(vp->hashid), partName);
5681 * If we aren't the fileserver, tell the fileserver the volume
5682 * needs to be salvaged. We could directly tell the
5683 * salvageserver, but the fileserver keeps track of some stats
5684 * related to salvages, and handles some other salvage-related
5685 * complications for us.
5687 *code = FSYNC_VolOp(vp->hashid, partName,
5688 FSYNC_VOL_FORCE_ERROR, FSYNC_SALVAGE, NULL);
5691 #endif /* FSSYNC_BUILD_CLIENT */
5696 * schedule a salvage with the salvage server or fileserver.
5698 * @param[in] vp pointer to volume object
5700 * @return operation status
5701 * @retval 0 salvage scheduled successfully
5702 * @retval 1 salvage not scheduled, or SALVSYNC/FSSYNC com error
5705 * @arg VOL_LOCK is held.
5706 * @arg nUsers and nWaiters should be zero.
5708 * @post salvageserver or fileserver is sent a salvage request
5710 * @note If we are the fileserver, the request will be sent to the salvage
5711 * server over SALVSYNC. If we are not the fileserver, the request will be
5712 * sent to the fileserver over FSSYNC (FSYNC_VOL_FORCE_ERROR/FSYNC_SALVAGE).
5714 * @note the caller must check if the volume needs to be freed after calling
5715 * this; the volume may not have any references or be on any lists after
5716 * we return, and we do not free it
5720 * @internal volume package internal use only.
5723 VScheduleSalvage_r(Volume * vp)
5727 VolState state_save;
5728 VThreadOptions_t * thread_opts;
5731 osi_Assert(VCanUseSALVSYNC() || VCanUseFSSYNC());
5733 if (vp->nWaiters || vp->nUsers) {
5737 /* prevent endless salvage,attach,salvage,attach,... loops */
5738 if (vp->stats.salvages >= SALVAGE_COUNT_MAX)
5742 * don't perform salvsync ops on certain threads
5744 thread_opts = pthread_getspecific(VThread_key);
5745 if (thread_opts == NULL) {
5746 thread_opts = &VThread_defaults;
5748 if (thread_opts->disallow_salvsync || vol_disallow_salvsync) {
5752 if (vp->salvage.scheduled) {
5756 VCreateReservation_r(vp);
5757 VWaitExclusiveState_r(vp);
5760 * XXX the scheduling process should really be done asynchronously
5761 * to avoid fssync deadlocks
5763 if (!vp->salvage.scheduled) {
5764 /* if we haven't previously scheduled a salvage, do so now
5766 * set the volume to an exclusive state and drop the lock
5767 * around the SALVSYNC call
5769 strlcpy(partName, vp->partition->name, sizeof(partName));
5770 state_save = VChangeState_r(vp, VOL_STATE_SALVSYNC_REQ);
5773 osi_Assert(try_SALVSYNC(vp, partName, &code) ||
5774 try_FSSYNC(vp, partName, &code));
5777 VChangeState_r(vp, state_save);
5779 if (code == SYNC_OK) {
5780 vp->salvage.scheduled = 1;
5781 vp->stats.last_salvage_req = FT_ApproxTime();
5782 if (VCanUseSALVSYNC()) {
5783 /* don't record these stats for non-fileservers; let the
5784 * fileserver take care of these */
5785 vp->stats.salvages++;
5786 IncUInt64(&VStats.salvages);
5791 case SYNC_BAD_COMMAND:
5792 case SYNC_COM_ERROR:
5795 Log("VScheduleSalvage_r: Salvage request for volume %lu "
5796 "denied\n", afs_printable_uint32_lu(vp->hashid));
5799 Log("VScheduleSalvage_r: Salvage request for volume %lu "
5800 "received unknown protocol error %d\n",
5801 afs_printable_uint32_lu(vp->hashid), code);
5805 if (VCanUseFSSYNC()) {
5806 VChangeState_r(vp, VOL_STATE_ERROR);
5811 /* NB: this is cancelling the reservation we obtained above, but we do
5812 * not call VCancelReservation_r, since that may trigger the vp dtor,
5813 * possibly free'ing the vp. We need to keep the vp around after
5814 * this, as the caller may reference vp without any refs. Instead, it
5815 * is the duty of the caller to inspect 'vp' after we return to see if
5816 * needs to be freed. */
5817 osi_Assert(--vp->nWaiters >= 0);
5820 #endif /* SALVSYNC_BUILD_CLIENT || FSSYNC_BUILD_CLIENT */
5822 #ifdef SALVSYNC_BUILD_CLIENT
5825 * connect to the salvageserver SYNC service.
5827 * @return operation status
5831 * @post connection to salvageserver SYNC service established
5833 * @see VConnectSALV_r
5834 * @see VDisconnectSALV
5835 * @see VReconnectSALV
5842 retVal = VConnectSALV_r();
5848 * connect to the salvageserver SYNC service.
5850 * @return operation status
5854 * @pre VOL_LOCK is held.
5856 * @post connection to salvageserver SYNC service established
5859 * @see VDisconnectSALV_r
5860 * @see VReconnectSALV_r
5861 * @see SALVSYNC_clientInit
5863 * @internal volume package internal use only.
5866 VConnectSALV_r(void)
5868 return SALVSYNC_clientInit();
5872 * disconnect from the salvageserver SYNC service.
5874 * @return operation status
5877 * @pre client should have a live connection to the salvageserver
5879 * @post connection to salvageserver SYNC service destroyed
5881 * @see VDisconnectSALV_r
5883 * @see VReconnectSALV
5886 VDisconnectSALV(void)
5889 VDisconnectSALV_r();
5895 * disconnect from the salvageserver SYNC service.
5897 * @return operation status
5901 * @arg VOL_LOCK is held.
5902 * @arg client should have a live connection to the salvageserver.
5904 * @post connection to salvageserver SYNC service destroyed
5906 * @see VDisconnectSALV
5907 * @see VConnectSALV_r
5908 * @see VReconnectSALV_r
5909 * @see SALVSYNC_clientFinis
5911 * @internal volume package internal use only.
5914 VDisconnectSALV_r(void)
5916 return SALVSYNC_clientFinis();
5920 * disconnect and then re-connect to the salvageserver SYNC service.
5922 * @return operation status
5926 * @pre client should have a live connection to the salvageserver
5928 * @post old connection is dropped, and a new one is established
5931 * @see VDisconnectSALV
5932 * @see VReconnectSALV_r
5935 VReconnectSALV(void)
5939 retVal = VReconnectSALV_r();
5945 * disconnect and then re-connect to the salvageserver SYNC service.
5947 * @return operation status
5952 * @arg VOL_LOCK is held.
5953 * @arg client should have a live connection to the salvageserver.
5955 * @post old connection is dropped, and a new one is established
5957 * @see VConnectSALV_r
5958 * @see VDisconnectSALV
5959 * @see VReconnectSALV
5960 * @see SALVSYNC_clientReconnect
5962 * @internal volume package internal use only.
5965 VReconnectSALV_r(void)
5967 return SALVSYNC_clientReconnect();
5969 #endif /* SALVSYNC_BUILD_CLIENT */
5970 #endif /* AFS_DEMAND_ATTACH_FS */
5973 /***************************************************/
5974 /* FSSYNC routines */
5975 /***************************************************/
5977 /* This must be called by any volume utility which needs to run while the
5978 file server is also running. This is separated from VInitVolumePackage2 so
5979 that a utility can fork--and each of the children can independently
5980 initialize communication with the file server */
5981 #ifdef FSSYNC_BUILD_CLIENT
5983 * connect to the fileserver SYNC service.
5985 * @return operation status
5990 * @arg VInit must equal 2.
5991 * @arg Program Type must not be fileserver or salvager.
5993 * @post connection to fileserver SYNC service established
5996 * @see VDisconnectFS
5997 * @see VChildProcReconnectFS
6004 retVal = VConnectFS_r();
6010 * connect to the fileserver SYNC service.
6012 * @return operation status
6017 * @arg VInit must equal 2.
6018 * @arg Program Type must not be fileserver or salvager.
6019 * @arg VOL_LOCK is held.
6021 * @post connection to fileserver SYNC service established
6024 * @see VDisconnectFS_r
6025 * @see VChildProcReconnectFS_r
6027 * @internal volume package internal use only.
6033 osi_Assert((VInit == 2) &&
6034 (programType != fileServer) &&
6035 (programType != salvager));
6036 rc = FSYNC_clientInit();
6044 * disconnect from the fileserver SYNC service.
6047 * @arg client should have a live connection to the fileserver.
6048 * @arg VOL_LOCK is held.
6049 * @arg Program Type must not be fileserver or salvager.
6051 * @post connection to fileserver SYNC service destroyed
6053 * @see VDisconnectFS
6055 * @see VChildProcReconnectFS_r
6057 * @internal volume package internal use only.
6060 VDisconnectFS_r(void)
6062 osi_Assert((programType != fileServer) &&
6063 (programType != salvager));
6064 FSYNC_clientFinis();
6069 * disconnect from the fileserver SYNC service.
6072 * @arg client should have a live connection to the fileserver.
6073 * @arg Program Type must not be fileserver or salvager.
6075 * @post connection to fileserver SYNC service destroyed
6077 * @see VDisconnectFS_r
6079 * @see VChildProcReconnectFS
6090 * connect to the fileserver SYNC service from a child process following a fork.
6092 * @return operation status
6097 * @arg VOL_LOCK is held.
6098 * @arg current FSYNC handle is shared with a parent process
6100 * @post current FSYNC handle is discarded and a new connection to the
6101 * fileserver SYNC service is established
6103 * @see VChildProcReconnectFS
6105 * @see VDisconnectFS_r
6107 * @internal volume package internal use only.
6110 VChildProcReconnectFS_r(void)
6112 return FSYNC_clientChildProcReconnect();
6116 * connect to the fileserver SYNC service from a child process following a fork.
6118 * @return operation status
6122 * @pre current FSYNC handle is shared with a parent process
6124 * @post current FSYNC handle is discarded and a new connection to the
6125 * fileserver SYNC service is established
6127 * @see VChildProcReconnectFS_r
6129 * @see VDisconnectFS
6132 VChildProcReconnectFS(void)
6136 ret = VChildProcReconnectFS_r();
6140 #endif /* FSSYNC_BUILD_CLIENT */
6143 /***************************************************/
6144 /* volume bitmap routines */
6145 /***************************************************/
6148 * allocate a vnode bitmap number for the vnode
6150 * @param[out] ec error code
6151 * @param[in] vp volume object pointer
6152 * @param[in] index vnode index number for the vnode
6153 * @param[in] flags flag values described in note
6155 * @note for DAFS, flags parameter controls locking behavior.
6156 * If (flags & VOL_ALLOC_BITMAP_WAIT) is set, then this function
6157 * will create a reservation and block on any other exclusive
6158 * operations. Otherwise, this function assumes the caller
6159 * already has exclusive access to vp, and we just change the
6162 * @pre VOL_LOCK held
6164 * @return bit number allocated
6170 VAllocBitmapEntry_r(Error * ec, Volume * vp,
6171 struct vnodeIndex *index, int flags)
6175 #ifdef AFS_DEMAND_ATTACH_FS
6176 VolState state_save;
6177 #endif /* AFS_DEMAND_ATTACH_FS */
6181 /* This test is probably redundant */
6182 if (!VolumeWriteable(vp)) {
6183 *ec = (bit32) VREADONLY;
6187 #ifdef AFS_DEMAND_ATTACH_FS
6188 if (flags & VOL_ALLOC_BITMAP_WAIT) {
6189 VCreateReservation_r(vp);
6190 VWaitExclusiveState_r(vp);
6192 state_save = VChangeState_r(vp, VOL_STATE_GET_BITMAP);
6193 #endif /* AFS_DEMAND_ATTACH_FS */
6196 if ((programType == fileServer) && !index->bitmap) {
6198 #ifndef AFS_DEMAND_ATTACH_FS
6199 /* demand attach fs uses the volume state to avoid races.
6200 * specialStatus field is not used at all */
6202 if (vp->specialStatus == VBUSY) {
6203 if (vp->goingOffline) { /* vos dump waiting for the volume to
6204 * go offline. We probably come here
6205 * from AddNewReadableResidency */
6208 while (vp->specialStatus == VBUSY) {
6209 #ifdef AFS_PTHREAD_ENV
6213 #else /* !AFS_PTHREAD_ENV */
6215 #endif /* !AFS_PTHREAD_ENV */
6219 #endif /* !AFS_DEMAND_ATTACH_FS */
6221 if (!index->bitmap) {
6222 #ifndef AFS_DEMAND_ATTACH_FS
6223 vp->specialStatus = VBUSY; /* Stop anyone else from using it. */
6224 #endif /* AFS_DEMAND_ATTACH_FS */
6225 for (i = 0; i < nVNODECLASSES; i++) {
6226 VGetBitmap_r(ec, vp, i);
6228 #ifdef AFS_DEMAND_ATTACH_FS
6229 VRequestSalvage_r(ec, vp, SALVSYNC_ERROR, 0 /*flags*/);
6230 #else /* AFS_DEMAND_ATTACH_FS */
6231 DeleteVolumeFromHashTable(vp);
6232 vp->shuttingDown = 1; /* Let who has it free it. */
6233 vp->specialStatus = 0;
6234 #endif /* AFS_DEMAND_ATTACH_FS */
6238 #ifndef AFS_DEMAND_ATTACH_FS
6240 vp->specialStatus = 0; /* Allow others to have access. */
6241 #endif /* AFS_DEMAND_ATTACH_FS */
6244 #endif /* BITMAP_LATER */
6246 #ifdef AFS_DEMAND_ATTACH_FS
6248 #endif /* AFS_DEMAND_ATTACH_FS */
6249 bp = index->bitmap + index->bitmapOffset;
6250 ep = index->bitmap + index->bitmapSize;
6252 if ((*(bit32 *) bp) != (bit32) 0xffffffff) {
6254 index->bitmapOffset = (afs_uint32) (bp - index->bitmap);
6257 o = ffs(~*bp) - 1; /* ffs is documented in BSTRING(3) */
6259 ret = ((bp - index->bitmap) * 8 + o);
6260 #ifdef AFS_DEMAND_ATTACH_FS
6262 #endif /* AFS_DEMAND_ATTACH_FS */
6265 bp += sizeof(bit32) /* i.e. 4 */ ;
6267 /* No bit map entry--must grow bitmap */
6269 realloc(index->bitmap, index->bitmapSize + VOLUME_BITMAP_GROWSIZE);
6270 osi_Assert(bp != NULL);
6272 bp += index->bitmapSize;
6273 memset(bp, 0, VOLUME_BITMAP_GROWSIZE);
6274 index->bitmapOffset = index->bitmapSize;
6275 index->bitmapSize += VOLUME_BITMAP_GROWSIZE;
6277 ret = index->bitmapOffset * 8;
6278 #ifdef AFS_DEMAND_ATTACH_FS
6280 #endif /* AFS_DEMAND_ATTACH_FS */
6283 #ifdef AFS_DEMAND_ATTACH_FS
6284 VChangeState_r(vp, state_save);
6285 if (flags & VOL_ALLOC_BITMAP_WAIT) {
6286 VCancelReservation_r(vp);
6288 #endif /* AFS_DEMAND_ATTACH_FS */
6293 VAllocBitmapEntry(Error * ec, Volume * vp, struct vnodeIndex * index)
6297 retVal = VAllocBitmapEntry_r(ec, vp, index, VOL_ALLOC_BITMAP_WAIT);
6303 VFreeBitMapEntry_r(Error * ec, Volume *vp, struct vnodeIndex *index,
6304 unsigned bitNumber, int flags)
6306 unsigned int offset;
6310 #ifdef AFS_DEMAND_ATTACH_FS
6311 if (flags & VOL_FREE_BITMAP_WAIT) {
6312 /* VAllocBitmapEntry_r allocs bitmap entries under an exclusive volume
6313 * state, so ensure we're not in an exclusive volume state when we update
6315 VCreateReservation_r(vp);
6316 VWaitExclusiveState_r(vp);
6323 #endif /* BITMAP_LATER */
6325 offset = bitNumber >> 3;
6326 if (offset >= index->bitmapSize) {
6330 if (offset < index->bitmapOffset)
6331 index->bitmapOffset = offset & ~3; /* Truncate to nearest bit32 */
6332 *(index->bitmap + offset) &= ~(1 << (bitNumber & 0x7));
6335 #ifdef AFS_DEMAND_ATTACH_FS
6336 VCancelReservation_r(vp);
6338 return; /* make the compiler happy for non-DAFS */
6342 VFreeBitMapEntry(Error * ec, Volume *vp, struct vnodeIndex *index,
6346 VFreeBitMapEntry_r(ec, vp, index, bitNumber, VOL_FREE_BITMAP_WAIT);
6350 /* this function will drop the glock internally.
6351 * for old pthread fileservers, this is safe thanks to vbusy.
6353 * for demand attach fs, caller must have already called
6354 * VCreateReservation_r and VWaitExclusiveState_r */
6356 VGetBitmap_r(Error * ec, Volume * vp, VnodeClass class)
6358 StreamHandle_t *file;
6359 afs_sfsize_t nVnodes, size;
6360 struct VnodeClassInfo *vcp = &VnodeClassInfo[class];
6361 struct vnodeIndex *vip = &vp->vnodeIndex[class];
6362 struct VnodeDiskObject *vnode;
6363 unsigned int unique = 0;
6367 #endif /* BITMAP_LATER */
6368 #ifdef AFS_DEMAND_ATTACH_FS
6369 VolState state_save;
6370 #endif /* AFS_DEMAND_ATTACH_FS */
6374 #ifdef AFS_DEMAND_ATTACH_FS
6375 state_save = VChangeState_r(vp, VOL_STATE_GET_BITMAP);
6376 #endif /* AFS_DEMAND_ATTACH_FS */
6379 fdP = IH_OPEN(vip->handle);
6380 osi_Assert(fdP != NULL);
6381 file = FDH_FDOPEN(fdP, "r");
6382 osi_Assert(file != NULL);
6383 vnode = (VnodeDiskObject *) malloc(vcp->diskSize);
6384 osi_Assert(vnode != NULL);
6385 size = OS_SIZE(fdP->fd_fd);
6386 osi_Assert(size != -1);
6387 nVnodes = (size <= vcp->diskSize ? 0 : size - vcp->diskSize)
6389 vip->bitmapSize = ((nVnodes / 8) + 10) / 4 * 4; /* The 10 is a little extra so
6390 * a few files can be created in this volume,
6391 * the whole thing is rounded up to nearest 4
6392 * bytes, because the bit map allocator likes
6395 BitMap = (byte *) calloc(1, vip->bitmapSize);
6396 osi_Assert(BitMap != NULL);
6397 #else /* BITMAP_LATER */
6398 vip->bitmap = (byte *) calloc(1, vip->bitmapSize);
6399 osi_Assert(vip->bitmap != NULL);
6400 vip->bitmapOffset = 0;
6401 #endif /* BITMAP_LATER */
6402 if (STREAM_ASEEK(file, vcp->diskSize) != -1) {
6404 for (bitNumber = 0; bitNumber < nVnodes + 100; bitNumber++) {
6405 if (STREAM_READ(vnode, vcp->diskSize, 1, file) != 1)
6407 if (vnode->type != vNull) {
6408 if (vnode->vnodeMagic != vcp->magic) {
6409 Log("GetBitmap: addled vnode index in volume %s; volume needs salvage\n", V_name(vp));
6414 *(BitMap + (bitNumber >> 3)) |= (1 << (bitNumber & 0x7));
6415 #else /* BITMAP_LATER */
6416 *(vip->bitmap + (bitNumber >> 3)) |= (1 << (bitNumber & 0x7));
6417 #endif /* BITMAP_LATER */
6418 if (unique <= vnode->uniquifier)
6419 unique = vnode->uniquifier + 1;
6421 #ifndef AFS_PTHREAD_ENV
6422 if ((bitNumber & 0x00ff) == 0x0ff) { /* every 256 iterations */
6425 #endif /* !AFS_PTHREAD_ENV */
6428 if (vp->nextVnodeUnique < unique) {
6429 Log("GetBitmap: bad volume uniquifier for volume %s; volume needs salvage\n", V_name(vp));
6432 /* Paranoia, partly justified--I think fclose after fdopen
6433 * doesn't seem to close fd. In any event, the documentation
6434 * doesn't specify, so it's safer to close it twice.
6442 /* There may have been a racing condition with some other thread, both
6443 * creating the bitmaps for this volume. If the other thread was faster
6444 * the pointer to bitmap should already be filled and we can free ours.
6446 if (vip->bitmap == NULL) {
6447 vip->bitmap = BitMap;
6448 vip->bitmapOffset = 0;
6450 free((byte *) BitMap);
6451 #endif /* BITMAP_LATER */
6452 #ifdef AFS_DEMAND_ATTACH_FS
6453 VChangeState_r(vp, state_save);
6454 #endif /* AFS_DEMAND_ATTACH_FS */
6458 /***************************************************/
6459 /* Volume Path and Volume Number utility routines */
6460 /***************************************************/
6463 * find the first occurrence of a volume header file and return the path.
6465 * @param[out] ec outbound error code
6466 * @param[in] volumeId volume id to find
6467 * @param[out] partitionp pointer to disk partition path string
6468 * @param[out] namep pointer to volume header file name string
6470 * @post path to first occurrence of volume header is returned in partitionp
6471 * and namep, or ec is set accordingly.
6473 * @warning this function is NOT re-entrant -- partitionp and namep point to
6474 * static data segments
6476 * @note if a volume utility inadvertently leaves behind a stale volume header
6477 * on a vice partition, it is possible for callers to get the wrong one,
6478 * depending on the order of the disk partition linked list.
6482 VGetVolumePath(Error * ec, VolId volumeId, char **partitionp, char **namep)
6484 static char partition[VMAXPATHLEN], name[VMAXPATHLEN];
6485 char path[VMAXPATHLEN];
6487 struct DiskPartition64 *dp;
6490 name[0] = OS_DIRSEPC;
6491 snprintf(&name[1], (sizeof name) - 1, VFORMAT,
6492 afs_printable_uint32_lu(volumeId));
6493 for (dp = DiskPartitionList; dp; dp = dp->next) {
6494 struct afs_stat_st status;
6495 strcpy(path, VPartitionPath(dp));
6497 if (afs_stat(path, &status) == 0) {
6498 strcpy(partition, dp->name);
6505 *partitionp = *namep = NULL;
6507 *partitionp = partition;
6513 * extract a volume number from a volume header filename string.
6515 * @param[in] name volume header filename string
6517 * @return volume number
6519 * @note the string must be of the form VFORMAT. the only permissible
6520 * deviation is a leading OS_DIRSEPC character.
6525 VolumeNumber(char *name)
6527 if (*name == OS_DIRSEPC)
6529 return strtoul(name + 1, NULL, 10);
6533 * compute the volume header filename.
6535 * @param[in] volumeId
6537 * @return volume header filename
6539 * @post volume header filename string is constructed
6541 * @warning this function is NOT re-entrant -- the returned string is
6542 * stored in a static char array. see VolumeExternalName_r
6543 * for a re-entrant equivalent.
6545 * @see VolumeExternalName_r
6547 * @deprecated due to the above re-entrancy warning, this interface should
6548 * be considered deprecated. Please use VolumeExternalName_r
6552 VolumeExternalName(VolumeId volumeId)
6554 static char name[VMAXPATHLEN];
6555 snprintf(name, sizeof name, VFORMAT, afs_printable_uint32_lu(volumeId));
6560 * compute the volume header filename.
6562 * @param[in] volumeId
6563 * @param[inout] name array in which to store filename
6564 * @param[in] len length of name array
6566 * @return result code from afs_snprintf
6568 * @see VolumeExternalName
6571 * @note re-entrant equivalent of VolumeExternalName
6574 VolumeExternalName_r(VolumeId volumeId, char * name, size_t len)
6576 return snprintf(name, len, VFORMAT, afs_printable_uint32_lu(volumeId));
6580 /***************************************************/
6581 /* Volume Usage Statistics routines */
6582 /***************************************************/
6584 #if OPENAFS_VOL_STATS
6585 #define OneDay (86400) /* 24 hours' worth of seconds */
6587 #define OneDay (24*60*60) /* 24 hours */
6588 #endif /* OPENAFS_VOL_STATS */
6591 Midnight(time_t t) {
6592 struct tm local, *l;
6595 #if defined(AFS_PTHREAD_ENV) && !defined(AFS_NT40_ENV)
6596 l = localtime_r(&t, &local);
6602 /* the following is strictly speaking problematic on the
6603 switching day to daylight saving time, after the switch,
6604 as tm_isdst does not match. Similarly, on the looong day when
6605 switching back the OneDay check will not do what naively expected!
6606 The effects are minor, though, and more a matter of interpreting
6608 #ifndef AFS_PTHREAD_ENV
6611 local.tm_hour = local.tm_min=local.tm_sec = 0;
6612 midnight = mktime(&local);
6613 if (midnight != (time_t) -1) return(midnight);
6615 return( (t/OneDay)*OneDay );
6619 /*------------------------------------------------------------------------
6620 * [export] VAdjustVolumeStatistics
6623 * If we've passed midnight, we need to update all the day use
6624 * statistics as well as zeroing the detailed volume statistics
6625 * (if we are implementing them).
6628 * vp : Pointer to the volume structure describing the lucky
6629 * volume being considered for update.
6635 * Nothing interesting.
6639 *------------------------------------------------------------------------*/
6642 VAdjustVolumeStatistics_r(Volume * vp)
6644 unsigned int now = FT_ApproxTime();
6646 if (now - V_dayUseDate(vp) > OneDay) {
6649 ndays = (now - V_dayUseDate(vp)) / OneDay;
6650 for (i = 6; i > ndays - 1; i--)
6651 V_weekUse(vp)[i] = V_weekUse(vp)[i - ndays];
6652 for (i = 0; i < ndays - 1 && i < 7; i++)
6653 V_weekUse(vp)[i] = 0;
6655 V_weekUse(vp)[ndays - 1] = V_dayUse(vp);
6657 V_dayUseDate(vp) = Midnight(now);
6659 #if OPENAFS_VOL_STATS
6661 * All we need to do is bzero the entire VOL_STATS_BYTES of
6662 * the detailed volume statistics area.
6664 memset((V_stat_area(vp)), 0, VOL_STATS_BYTES);
6665 #endif /* OPENAFS_VOL_STATS */
6668 /*It's been more than a day of collection */
6670 * Always return happily.
6673 } /*VAdjustVolumeStatistics */
6676 VAdjustVolumeStatistics(Volume * vp)
6680 retVal = VAdjustVolumeStatistics_r(vp);
6686 VBumpVolumeUsage_r(Volume * vp)
6688 unsigned int now = FT_ApproxTime();
6689 V_accessDate(vp) = now;
6690 if (now - V_dayUseDate(vp) > OneDay)
6691 VAdjustVolumeStatistics_r(vp);
6693 * Save the volume header image to disk after a threshold of bumps to dayUse,
6694 * at most every usage_rate_limit seconds.
6697 vp->usage_bumps_outstanding++;
6698 if (vp->usage_bumps_outstanding >= vol_opts.usage_threshold
6699 && vp->usage_bumps_next_write <= now) {
6701 vp->usage_bumps_outstanding = 0;
6702 vp->usage_bumps_next_write = now + vol_opts.usage_rate_limit;
6703 VUpdateVolume_r(&error, vp, VOL_UPDATE_WAIT);
6708 VBumpVolumeUsage(Volume * vp)
6711 VBumpVolumeUsage_r(vp);
6716 VSetDiskUsage_r(void)
6718 #ifndef AFS_DEMAND_ATTACH_FS
6719 static int FifteenMinuteCounter = 0;
6723 /* NOTE: Don't attempt to access the partitions list until the
6724 * initialization level indicates that all volumes are attached,
6725 * which implies that all partitions are initialized. */
6726 #ifdef AFS_PTHREAD_ENV
6727 VOL_CV_WAIT(&vol_vinit_cond);
6728 #else /* AFS_PTHREAD_ENV */
6730 #endif /* AFS_PTHREAD_ENV */
6733 VResetDiskUsage_r();
6735 #ifndef AFS_DEMAND_ATTACH_FS
6736 if (++FifteenMinuteCounter == 3) {
6737 FifteenMinuteCounter = 0;
6740 #endif /* !AFS_DEMAND_ATTACH_FS */
6752 /***************************************************/
6753 /* Volume Update List routines */
6754 /***************************************************/
6756 /* The number of minutes that a volume hasn't been updated before the
6757 * "Dont salvage" flag in the volume header will be turned on */
6758 #define SALVAGE_INTERVAL (10*60)
6763 * volume update list functionality has been moved into the VLRU
6764 * the DONT_SALVAGE flag is now set during VLRU demotion
6767 #ifndef AFS_DEMAND_ATTACH_FS
6768 static VolumeId *UpdateList = NULL; /* Pointer to array of Volume ID's */
6769 static int nUpdatedVolumes = 0; /* Updated with entry in UpdateList, salvage after crash flag on */
6770 static int updateSize = 0; /* number of entries possible */
6771 #define UPDATE_LIST_SIZE 128 /* initial size increment (must be a power of 2!) */
6772 #endif /* !AFS_DEMAND_ATTACH_FS */
6775 VAddToVolumeUpdateList_r(Error * ec, Volume * vp)
6778 vp->updateTime = FT_ApproxTime();
6779 if (V_dontSalvage(vp) == 0)
6781 V_dontSalvage(vp) = 0;
6782 VSyncVolume_r(ec, vp, 0);
6783 #ifdef AFS_DEMAND_ATTACH_FS
6784 V_attachFlags(vp) &= ~(VOL_HDR_DONTSALV);
6785 #else /* !AFS_DEMAND_ATTACH_FS */
6788 if (UpdateList == NULL) {
6789 updateSize = UPDATE_LIST_SIZE;
6790 UpdateList = (VolumeId *) malloc(sizeof(VolumeId) * updateSize);
6792 if (nUpdatedVolumes == updateSize) {
6794 if (updateSize > 524288) {
6795 Log("warning: there is likely a bug in the volume update scanner\n");
6799 (VolumeId *) realloc(UpdateList,
6800 sizeof(VolumeId) * updateSize);
6803 osi_Assert(UpdateList != NULL);
6804 UpdateList[nUpdatedVolumes++] = V_id(vp);
6805 #endif /* !AFS_DEMAND_ATTACH_FS */
6808 #ifndef AFS_DEMAND_ATTACH_FS
6810 VScanUpdateList(void)
6815 afs_uint32 now = FT_ApproxTime();
6816 /* Be careful with this code, since it works with interleaved calls to AddToVolumeUpdateList */
6817 for (i = gap = 0; i < nUpdatedVolumes; i++) {
6819 UpdateList[i - gap] = UpdateList[i];
6821 /* XXX this routine needlessly messes up the Volume LRU by
6822 * breaking the LRU temporal-locality assumptions.....
6823 * we should use a special volume header allocator here */
6824 vp = VGetVolume_r(&error, UpdateList[i - gap] = UpdateList[i]);
6827 } else if (vp->nUsers == 1 && now - vp->updateTime > SALVAGE_INTERVAL) {
6828 V_dontSalvage(vp) = DONT_SALVAGE;
6829 VUpdateVolume_r(&error, vp, 0); /* No need to fsync--not critical */
6837 #ifndef AFS_PTHREAD_ENV
6839 #endif /* !AFS_PTHREAD_ENV */
6841 nUpdatedVolumes -= gap;
6843 #endif /* !AFS_DEMAND_ATTACH_FS */
6846 /***************************************************/
6847 /* Volume LRU routines */
6848 /***************************************************/
6853 * with demand attach fs, we attempt to soft detach(1)
6854 * volumes which have not been accessed in a long time
6855 * in order to speed up fileserver shutdown
6857 * (1) by soft detach we mean a process very similar
6858 * to VOffline, except the final state of the
6859 * Volume will be VOL_STATE_PREATTACHED, instead
6860 * of the usual VOL_STATE_UNATTACHED
6862 #ifdef AFS_DEMAND_ATTACH_FS
6864 /* implementation is reminiscent of a generational GC
6866 * queue 0 is newly attached volumes. this queue is
6867 * sorted by attach timestamp
6869 * queue 1 is volumes that have been around a bit
6870 * longer than queue 0. this queue is sorted by
6873 * queue 2 is volumes tha have been around the longest.
6874 * this queue is unsorted
6876 * queue 3 is volumes that have been marked as
6877 * candidates for soft detachment. this queue is
6880 #define VLRU_GENERATIONS 3 /**< number of generations in VLRU */
6881 #define VLRU_QUEUES 5 /**< total number of VLRU queues */
6884 * definition of a VLRU queue.
6887 volatile struct rx_queue q;
6894 * main VLRU data structure.
6897 struct VLRU_q q[VLRU_QUEUES]; /**< VLRU queues */
6900 /** time interval (in seconds) between promotion passes for
6901 * each young generation queue. */
6902 afs_uint32 promotion_interval[VLRU_GENERATIONS-1];
6904 /** time interval (in seconds) between soft detach candidate
6905 * scans for each generation queue.
6907 * scan_interval[VLRU_QUEUE_CANDIDATE] defines how frequently
6908 * we perform a soft detach pass. */
6909 afs_uint32 scan_interval[VLRU_GENERATIONS+1];
6911 /* scheduler state */
6912 int next_idx; /**< next queue to receive attention */
6913 afs_uint32 last_promotion[VLRU_GENERATIONS-1]; /**< timestamp of last promotion scan */
6914 afs_uint32 last_scan[VLRU_GENERATIONS+1]; /**< timestamp of last detach scan */
6916 int scanner_state; /**< state of scanner thread */
6917 pthread_cond_t cv; /**< state transition CV */
6920 /** global VLRU state */
6921 static struct VLRU volume_LRU;
6924 * defined states for VLRU scanner thread.
6927 VLRU_SCANNER_STATE_OFFLINE = 0, /**< vlru scanner thread is offline */
6928 VLRU_SCANNER_STATE_ONLINE = 1, /**< vlru scanner thread is online */
6929 VLRU_SCANNER_STATE_SHUTTING_DOWN = 2, /**< vlru scanner thread is shutting down */
6930 VLRU_SCANNER_STATE_PAUSING = 3, /**< vlru scanner thread is getting ready to pause */
6931 VLRU_SCANNER_STATE_PAUSED = 4 /**< vlru scanner thread is paused */
6932 } vlru_thread_state_t;
6934 /* vlru disk data header stuff */
6935 #define VLRU_DISK_MAGIC 0x7a8b9cad /**< vlru disk entry magic number */
6936 #define VLRU_DISK_VERSION 1 /**< vlru disk entry version number */
6938 /** vlru default expiration time (for eventual fs state serialization of vlru data) */
6939 #define VLRU_DUMP_EXPIRATION_TIME (60*60*24*7) /* expire vlru data after 1 week */
6942 /** minimum volume inactivity (in seconds) before a volume becomes eligible for
6943 * soft detachment. */
6944 static afs_uint32 VLRU_offline_thresh = VLRU_DEFAULT_OFFLINE_THRESH;
6946 /** time interval (in seconds) between VLRU scanner thread soft detach passes. */
6947 static afs_uint32 VLRU_offline_interval = VLRU_DEFAULT_OFFLINE_INTERVAL;
6949 /** maximum number of volumes to soft detach in a VLRU soft detach pass. */
6950 static afs_uint32 VLRU_offline_max = VLRU_DEFAULT_OFFLINE_MAX;
6952 /** VLRU control flag. non-zero value implies VLRU subsystem is activated. */
6953 static afs_uint32 VLRU_enabled = 1;
6955 /* queue synchronization routines */
6956 static void VLRU_BeginExclusive_r(struct VLRU_q * q);
6957 static void VLRU_EndExclusive_r(struct VLRU_q * q);
6958 static void VLRU_Wait_r(struct VLRU_q * q);
6961 * set VLRU subsystem tunable parameters.
6963 * @param[in] option tunable option to modify
6964 * @param[in] val new value for tunable parameter
6966 * @pre @c VInitVolumePackage2 has not yet been called.
6968 * @post tunable parameter is modified
6972 * @note valid option parameters are:
6973 * @arg @c VLRU_SET_THRESH
6974 * set the period of inactivity after which
6975 * volumes are eligible for soft detachment
6976 * @arg @c VLRU_SET_INTERVAL
6977 * set the time interval between calls
6978 * to the volume LRU "garbage collector"
6979 * @arg @c VLRU_SET_MAX
6980 * set the max number of volumes to deallocate
6984 VLRU_SetOptions(int option, afs_uint32 val)
6986 if (option == VLRU_SET_THRESH) {
6987 VLRU_offline_thresh = val;
6988 } else if (option == VLRU_SET_INTERVAL) {
6989 VLRU_offline_interval = val;
6990 } else if (option == VLRU_SET_MAX) {
6991 VLRU_offline_max = val;
6992 } else if (option == VLRU_SET_ENABLED) {
6995 VLRU_ComputeConstants();
6999 * compute VLRU internal timing parameters.
7001 * @post VLRU scanner thread internal timing parameters are computed
7003 * @note computes internal timing parameters based upon user-modifiable
7004 * tunable parameters.
7008 * @internal volume package internal use only.
7011 VLRU_ComputeConstants(void)
7013 afs_uint32 factor = VLRU_offline_thresh / VLRU_offline_interval;
7015 /* compute the candidate scan interval */
7016 volume_LRU.scan_interval[VLRU_QUEUE_CANDIDATE] = VLRU_offline_interval;
7018 /* compute the promotion intervals */
7019 volume_LRU.promotion_interval[VLRU_QUEUE_NEW] = VLRU_offline_thresh * 2;
7020 volume_LRU.promotion_interval[VLRU_QUEUE_MID] = VLRU_offline_thresh * 4;
7023 /* compute the gen 0 scan interval */
7024 volume_LRU.scan_interval[VLRU_QUEUE_NEW] = VLRU_offline_thresh / 8;
7026 /* compute the gen 0 scan interval */
7027 volume_LRU.scan_interval[VLRU_QUEUE_NEW] = VLRU_offline_interval * 2;
7032 * initialize VLRU subsystem.
7034 * @pre this function has not yet been called
7036 * @post VLRU subsystem is initialized and VLRU scanner thread is starting
7040 * @internal volume package internal use only.
7046 pthread_attr_t attrs;
7049 if (!VLRU_enabled) {
7050 Log("VLRU: disabled\n");
7054 /* initialize each of the VLRU queues */
7055 for (i = 0; i < VLRU_QUEUES; i++) {
7056 queue_Init(&volume_LRU.q[i]);
7057 volume_LRU.q[i].len = 0;
7058 volume_LRU.q[i].busy = 0;
7059 CV_INIT(&volume_LRU.q[i].cv, "vol lru", CV_DEFAULT, 0);
7062 /* setup the timing constants */
7063 VLRU_ComputeConstants();
7065 /* XXX put inside LogLevel check? */
7066 Log("VLRU: starting scanner with the following configuration parameters:\n");
7067 Log("VLRU: offlining volumes after minimum of %d seconds of inactivity\n", VLRU_offline_thresh);
7068 Log("VLRU: running VLRU soft detach pass every %d seconds\n", VLRU_offline_interval);
7069 Log("VLRU: taking up to %d volumes offline per pass\n", VLRU_offline_max);
7070 Log("VLRU: scanning generation 0 for inactive volumes every %d seconds\n", volume_LRU.scan_interval[0]);
7071 Log("VLRU: scanning for promotion/demotion between generations 0 and 1 every %d seconds\n", volume_LRU.promotion_interval[0]);
7072 Log("VLRU: scanning for promotion/demotion between generations 1 and 2 every %d seconds\n", volume_LRU.promotion_interval[1]);
7074 /* start up the VLRU scanner */
7075 volume_LRU.scanner_state = VLRU_SCANNER_STATE_OFFLINE;
7076 if (programType == fileServer) {
7077 CV_INIT(&volume_LRU.cv, "vol lru", CV_DEFAULT, 0);
7078 osi_Assert(pthread_attr_init(&attrs) == 0);
7079 osi_Assert(pthread_attr_setdetachstate(&attrs, PTHREAD_CREATE_DETACHED) == 0);
7080 osi_Assert(pthread_create(&tid, &attrs, &VLRU_ScannerThread, NULL) == 0);
7085 * initialize the VLRU-related fields of a newly allocated volume object.
7087 * @param[in] vp pointer to volume object
7090 * @arg @c VOL_LOCK is held.
7091 * @arg volume object is not on a VLRU queue.
7093 * @post VLRU fields are initialized to indicate that volume object is not
7094 * currently registered with the VLRU subsystem
7098 * @internal volume package interal use only.
7101 VLRU_Init_Node_r(Volume * vp)
7106 osi_Assert(queue_IsNotOnQueue(&vp->vlru));
7107 vp->vlru.idx = VLRU_QUEUE_INVALID;
7111 * add a volume object to a VLRU queue.
7113 * @param[in] vp pointer to volume object
7116 * @arg @c VOL_LOCK is held.
7117 * @arg caller MUST hold a lightweight ref on @p vp.
7118 * @arg caller MUST NOT hold exclusive ownership of the VLRU queue.
7120 * @post the volume object is added to the appropriate VLRU queue
7122 * @note if @c vp->vlru.idx contains the index of a valid VLRU queue,
7123 * then the volume is added to that queue. Otherwise, the value
7124 * @c VLRU_QUEUE_NEW is stored into @c vp->vlru.idx and the
7125 * volume is added to the NEW generation queue.
7127 * @note @c VOL_LOCK may be dropped internally
7129 * @note Volume state is temporarily set to @c VOL_STATE_VLRU_ADD
7130 * during the add operation, and is restored to the previous
7131 * state prior to return.
7135 * @internal volume package internal use only.
7138 VLRU_Add_r(Volume * vp)
7141 VolState state_save;
7146 if (queue_IsOnQueue(&vp->vlru))
7149 state_save = VChangeState_r(vp, VOL_STATE_VLRU_ADD);
7152 if ((idx < 0) || (idx >= VLRU_QUEUE_INVALID)) {
7153 idx = VLRU_QUEUE_NEW;
7156 VLRU_Wait_r(&volume_LRU.q[idx]);
7158 /* repeat check since VLRU_Wait_r may have dropped
7160 if (queue_IsNotOnQueue(&vp->vlru)) {
7162 queue_Prepend(&volume_LRU.q[idx], &vp->vlru);
7163 volume_LRU.q[idx].len++;
7164 V_attachFlags(vp) |= VOL_ON_VLRU;
7165 vp->stats.last_promote = FT_ApproxTime();
7168 VChangeState_r(vp, state_save);
7172 * delete a volume object from a VLRU queue.
7174 * @param[in] vp pointer to volume object
7177 * @arg @c VOL_LOCK is held.
7178 * @arg caller MUST hold a lightweight ref on @p vp.
7179 * @arg caller MUST NOT hold exclusive ownership of the VLRU queue.
7181 * @post volume object is removed from the VLRU queue
7183 * @note @c VOL_LOCK may be dropped internally
7187 * @todo We should probably set volume state to something exlcusive
7188 * (as @c VLRU_Add_r does) prior to dropping @c VOL_LOCK.
7190 * @internal volume package internal use only.
7193 VLRU_Delete_r(Volume * vp)
7200 if (queue_IsNotOnQueue(&vp->vlru))
7206 if (idx == VLRU_QUEUE_INVALID)
7208 VLRU_Wait_r(&volume_LRU.q[idx]);
7209 } while (idx != vp->vlru.idx);
7211 /* now remove from the VLRU and update
7212 * the appropriate counter */
7213 queue_Remove(&vp->vlru);
7214 volume_LRU.q[idx].len--;
7215 vp->vlru.idx = VLRU_QUEUE_INVALID;
7216 V_attachFlags(vp) &= ~(VOL_ON_VLRU);
7220 * tell the VLRU subsystem that a volume was just accessed.
7222 * @param[in] vp pointer to volume object
7225 * @arg @c VOL_LOCK is held
7226 * @arg caller MUST hold a lightweight ref on @p vp
7227 * @arg caller MUST NOT hold exclusive ownership of any VLRU queue
7229 * @post volume VLRU access statistics are updated. If the volume was on
7230 * the VLRU soft detach candidate queue, it is moved to the NEW
7233 * @note @c VOL_LOCK may be dropped internally
7237 * @internal volume package internal use only.
7240 VLRU_UpdateAccess_r(Volume * vp)
7242 Volume * rvp = NULL;
7247 if (queue_IsNotOnQueue(&vp->vlru))
7250 osi_Assert(V_attachFlags(vp) & VOL_ON_VLRU);
7252 /* update the access timestamp */
7253 vp->stats.last_get = FT_ApproxTime();
7256 * if the volume is on the soft detach candidate
7257 * list, we need to safely move it back to a
7258 * regular generation. this has to be done
7259 * carefully so we don't race against the scanner
7263 /* if this volume is on the soft detach candidate queue,
7264 * then grab exclusive access to the necessary queues */
7265 if (vp->vlru.idx == VLRU_QUEUE_CANDIDATE) {
7267 VCreateReservation_r(rvp);
7269 VLRU_Wait_r(&volume_LRU.q[VLRU_QUEUE_NEW]);
7270 VLRU_BeginExclusive_r(&volume_LRU.q[VLRU_QUEUE_NEW]);
7271 VLRU_Wait_r(&volume_LRU.q[VLRU_QUEUE_CANDIDATE]);
7272 VLRU_BeginExclusive_r(&volume_LRU.q[VLRU_QUEUE_CANDIDATE]);
7275 /* make sure multiple threads don't race to update */
7276 if (vp->vlru.idx == VLRU_QUEUE_CANDIDATE) {
7277 VLRU_SwitchQueues(vp, VLRU_QUEUE_NEW, 1);
7281 VLRU_EndExclusive_r(&volume_LRU.q[VLRU_QUEUE_CANDIDATE]);
7282 VLRU_EndExclusive_r(&volume_LRU.q[VLRU_QUEUE_NEW]);
7283 VCancelReservation_r(rvp);
7288 * switch a volume between two VLRU queues.
7290 * @param[in] vp pointer to volume object
7291 * @param[in] new_idx index of VLRU queue onto which the volume will be moved
7292 * @param[in] append controls whether the volume will be appended or
7293 * prepended to the queue. A nonzero value means it will
7294 * be appended; zero means it will be prepended.
7296 * @pre The new (and old, if applicable) queue(s) must either be owned
7297 * exclusively by the calling thread for asynchronous manipulation,
7298 * or the queue(s) must be quiescent and VOL_LOCK must be held.
7299 * Please see VLRU_BeginExclusive_r, VLRU_EndExclusive_r and VLRU_Wait_r
7300 * for further details of the queue asynchronous processing mechanism.
7302 * @post If the volume object was already on a VLRU queue, it is
7303 * removed from the queue. Depending on the value of the append
7304 * parameter, the volume object is either appended or prepended
7305 * to the VLRU queue referenced by the new_idx parameter.
7309 * @see VLRU_BeginExclusive_r
7310 * @see VLRU_EndExclusive_r
7313 * @internal volume package internal use only.
7316 VLRU_SwitchQueues(Volume * vp, int new_idx, int append)
7318 if (queue_IsNotOnQueue(&vp->vlru))
7321 queue_Remove(&vp->vlru);
7322 volume_LRU.q[vp->vlru.idx].len--;
7324 /* put the volume back on the correct generational queue */
7326 queue_Append(&volume_LRU.q[new_idx], &vp->vlru);
7328 queue_Prepend(&volume_LRU.q[new_idx], &vp->vlru);
7331 volume_LRU.q[new_idx].len++;
7332 vp->vlru.idx = new_idx;
7336 * VLRU background thread.
7338 * The VLRU Scanner Thread is responsible for periodically scanning through
7339 * each VLRU queue looking for volumes which should be moved to another
7340 * queue, or soft detached.
7342 * @param[in] args unused thread arguments parameter
7344 * @return unused thread return value
7345 * @retval NULL always
7347 * @internal volume package internal use only.
7350 VLRU_ScannerThread(void * args)
7352 afs_uint32 now, min_delay, delay;
7353 int i, min_idx, min_op, overdue, state;
7355 /* set t=0 for promotion cycle to be
7356 * fileserver startup */
7357 now = FT_ApproxTime();
7358 for (i=0; i < VLRU_GENERATIONS-1; i++) {
7359 volume_LRU.last_promotion[i] = now;
7362 /* don't start the scanner until VLRU_offline_thresh
7363 * plus a small delay for VInitVolumePackage2 to finish
7366 sleep(VLRU_offline_thresh + 60);
7368 /* set t=0 for scan cycle to be now */
7369 now = FT_ApproxTime();
7370 for (i=0; i < VLRU_GENERATIONS+1; i++) {
7371 volume_LRU.last_scan[i] = now;
7375 if (volume_LRU.scanner_state == VLRU_SCANNER_STATE_OFFLINE) {
7376 volume_LRU.scanner_state = VLRU_SCANNER_STATE_ONLINE;
7379 while ((state = volume_LRU.scanner_state) != VLRU_SCANNER_STATE_SHUTTING_DOWN) {
7380 /* check to see if we've been asked to pause */
7381 if (volume_LRU.scanner_state == VLRU_SCANNER_STATE_PAUSING) {
7382 volume_LRU.scanner_state = VLRU_SCANNER_STATE_PAUSED;
7383 CV_BROADCAST(&volume_LRU.cv);
7385 VOL_CV_WAIT(&volume_LRU.cv);
7386 } while (volume_LRU.scanner_state == VLRU_SCANNER_STATE_PAUSED);
7389 /* scheduling can happen outside the glock */
7392 /* figure out what is next on the schedule */
7394 /* figure out a potential schedule for the new generation first */
7396 min_delay = volume_LRU.scan_interval[0] + volume_LRU.last_scan[0] - now;
7399 if (min_delay > volume_LRU.scan_interval[0]) {
7400 /* unsigned overflow -- we're overdue to run this scan */
7405 /* if we're not overdue for gen 0, figure out schedule for candidate gen */
7407 i = VLRU_QUEUE_CANDIDATE;
7408 delay = volume_LRU.scan_interval[i] + volume_LRU.last_scan[i] - now;
7409 if (delay < min_delay) {
7413 if (delay > volume_LRU.scan_interval[i]) {
7414 /* unsigned overflow -- we're overdue to run this scan */
7421 /* if we're still not overdue for something, figure out schedules for promotions */
7422 for (i=0; !overdue && i < VLRU_GENERATIONS-1; i++) {
7423 delay = volume_LRU.promotion_interval[i] + volume_LRU.last_promotion[i] - now;
7424 if (delay < min_delay) {
7429 if (delay > volume_LRU.promotion_interval[i]) {
7430 /* unsigned overflow -- we're overdue to run this promotion */
7439 /* sleep as needed */
7444 /* do whatever is next */
7447 VLRU_Promote_r(min_idx);
7448 VLRU_Demote_r(min_idx+1);
7450 VLRU_Scan_r(min_idx);
7452 now = FT_ApproxTime();
7455 Log("VLRU scanner asked to go offline (scanner_state=%d)\n", state);
7457 /* signal that scanner is down */
7458 volume_LRU.scanner_state = VLRU_SCANNER_STATE_OFFLINE;
7459 CV_BROADCAST(&volume_LRU.cv);
7465 * promote volumes from one VLRU generation to the next.
7467 * This routine scans a VLRU generation looking for volumes which are
7468 * eligible to be promoted to the next generation. All volumes which
7469 * meet the eligibility requirement are promoted.
7471 * Promotion eligibility is based upon meeting both of the following
7474 * @arg The volume has been accessed since the last promotion:
7475 * @c (vp->stats.last_get >= vp->stats.last_promote)
7476 * @arg The last promotion occurred at least
7477 * @c volume_LRU.promotion_interval[idx] seconds ago
7479 * As a performance optimization, promotions are "globbed". In other
7480 * words, we promote arbitrarily large contiguous sublists of elements
7483 * @param[in] idx VLRU queue index to scan
7487 * @internal VLRU internal use only.
7490 VLRU_Promote_r(int idx)
7492 int len, chaining, promote;
7493 afs_uint32 now, thresh;
7494 struct rx_queue *qp, *nqp;
7495 Volume * vp, *start = NULL, *end = NULL;
7497 /* get exclusive access to two chains, and drop the glock */
7498 VLRU_Wait_r(&volume_LRU.q[idx]);
7499 VLRU_BeginExclusive_r(&volume_LRU.q[idx]);
7500 VLRU_Wait_r(&volume_LRU.q[idx+1]);
7501 VLRU_BeginExclusive_r(&volume_LRU.q[idx+1]);
7504 thresh = volume_LRU.promotion_interval[idx];
7505 now = FT_ApproxTime();
7508 for (queue_ScanBackwards(&volume_LRU.q[idx], qp, nqp, rx_queue)) {
7509 vp = (Volume *)((char *)qp - offsetof(Volume, vlru));
7510 promote = (((vp->stats.last_promote + thresh) <= now) &&
7511 (vp->stats.last_get >= vp->stats.last_promote));
7519 /* promote and prepend chain */
7520 queue_MoveChainAfter(&volume_LRU.q[idx+1], &start->vlru, &end->vlru);
7534 /* promote and prepend */
7535 queue_MoveChainAfter(&volume_LRU.q[idx+1], &start->vlru, &end->vlru);
7539 volume_LRU.q[idx].len -= len;
7540 volume_LRU.q[idx+1].len += len;
7543 /* release exclusive access to the two chains */
7545 volume_LRU.last_promotion[idx] = now;
7546 VLRU_EndExclusive_r(&volume_LRU.q[idx+1]);
7547 VLRU_EndExclusive_r(&volume_LRU.q[idx]);
7550 /* run the demotions */
7552 VLRU_Demote_r(int idx)
7555 int len, chaining, demote;
7556 afs_uint32 now, thresh;
7557 struct rx_queue *qp, *nqp;
7558 Volume * vp, *start = NULL, *end = NULL;
7559 Volume ** salv_flag_vec = NULL;
7560 int salv_vec_offset = 0;
7562 osi_Assert(idx == VLRU_QUEUE_MID || idx == VLRU_QUEUE_OLD);
7564 /* get exclusive access to two chains, and drop the glock */
7565 VLRU_Wait_r(&volume_LRU.q[idx-1]);
7566 VLRU_BeginExclusive_r(&volume_LRU.q[idx-1]);
7567 VLRU_Wait_r(&volume_LRU.q[idx]);
7568 VLRU_BeginExclusive_r(&volume_LRU.q[idx]);
7571 /* no big deal if this allocation fails */
7572 if (volume_LRU.q[idx].len) {
7573 salv_flag_vec = (Volume **) malloc(volume_LRU.q[idx].len * sizeof(Volume *));
7576 now = FT_ApproxTime();
7577 thresh = volume_LRU.promotion_interval[idx-1];
7580 for (queue_ScanBackwards(&volume_LRU.q[idx], qp, nqp, rx_queue)) {
7581 vp = (Volume *)((char *)qp - offsetof(Volume, vlru));
7582 demote = (((vp->stats.last_promote + thresh) <= now) &&
7583 (vp->stats.last_get < (now - thresh)));
7585 /* we now do volume update list DONT_SALVAGE flag setting during
7586 * demotion passes */
7587 if (salv_flag_vec &&
7588 !(V_attachFlags(vp) & VOL_HDR_DONTSALV) &&
7590 (vp->updateTime < (now - SALVAGE_INTERVAL)) &&
7591 (V_attachState(vp) == VOL_STATE_ATTACHED)) {
7592 salv_flag_vec[salv_vec_offset++] = vp;
7593 VCreateReservation_r(vp);
7602 /* demote and append chain */
7603 queue_MoveChainBefore(&volume_LRU.q[idx-1], &start->vlru, &end->vlru);
7617 queue_MoveChainBefore(&volume_LRU.q[idx-1], &start->vlru, &end->vlru);
7621 volume_LRU.q[idx].len -= len;
7622 volume_LRU.q[idx-1].len += len;
7625 /* release exclusive access to the two chains */
7627 VLRU_EndExclusive_r(&volume_LRU.q[idx]);
7628 VLRU_EndExclusive_r(&volume_LRU.q[idx-1]);
7630 /* now go back and set the DONT_SALVAGE flags as appropriate */
7631 if (salv_flag_vec) {
7633 for (i = 0; i < salv_vec_offset; i++) {
7634 vp = salv_flag_vec[i];
7635 if (!(V_attachFlags(vp) & VOL_HDR_DONTSALV) &&
7636 (vp->updateTime < (now - SALVAGE_INTERVAL)) &&
7637 (V_attachState(vp) == VOL_STATE_ATTACHED)) {
7640 V_attachFlags(vp) |= VOL_HDR_DONTSALV;
7641 V_dontSalvage(vp) = DONT_SALVAGE;
7642 VUpdateVolume_r(&ec, vp, 0);
7646 VCancelReservation_r(vp);
7648 free(salv_flag_vec);
7652 /* run a pass of the VLRU GC scanner */
7654 VLRU_Scan_r(int idx)
7656 afs_uint32 now, thresh;
7657 struct rx_queue *qp, *nqp;
7661 osi_Assert(idx == VLRU_QUEUE_NEW || idx == VLRU_QUEUE_CANDIDATE);
7663 /* gain exclusive access to the idx VLRU */
7664 VLRU_Wait_r(&volume_LRU.q[idx]);
7665 VLRU_BeginExclusive_r(&volume_LRU.q[idx]);
7667 if (idx != VLRU_QUEUE_CANDIDATE) {
7668 /* gain exclusive access to the candidate VLRU */
7669 VLRU_Wait_r(&volume_LRU.q[VLRU_QUEUE_CANDIDATE]);
7670 VLRU_BeginExclusive_r(&volume_LRU.q[VLRU_QUEUE_CANDIDATE]);
7673 now = FT_ApproxTime();
7674 thresh = now - VLRU_offline_thresh;
7676 /* perform candidate selection and soft detaching */
7677 if (idx == VLRU_QUEUE_CANDIDATE) {
7678 /* soft detach some volumes from the candidate pool */
7682 for (i=0,queue_ScanBackwards(&volume_LRU.q[idx], qp, nqp, rx_queue)) {
7683 vp = (Volume *)((char *)qp - offsetof(Volume, vlru));
7684 if (i >= VLRU_offline_max) {
7687 /* check timestamp to see if it's a candidate for soft detaching */
7688 if (vp->stats.last_get <= thresh) {
7690 if (VCheckSoftDetach(vp, thresh))
7696 /* scan for volumes to become soft detach candidates */
7697 for (i=1,queue_ScanBackwards(&volume_LRU.q[idx], qp, nqp, rx_queue),i++) {
7698 vp = (Volume *)((char *)qp - offsetof(Volume, vlru));
7700 /* check timestamp to see if it's a candidate for soft detaching */
7701 if (vp->stats.last_get <= thresh) {
7702 VCheckSoftDetachCandidate(vp, thresh);
7705 if (!(i&0x7f)) { /* lock coarsening optimization */
7713 /* relinquish exclusive access to the VLRU chains */
7717 volume_LRU.last_scan[idx] = now;
7718 if (idx != VLRU_QUEUE_CANDIDATE) {
7719 VLRU_EndExclusive_r(&volume_LRU.q[VLRU_QUEUE_CANDIDATE]);
7721 VLRU_EndExclusive_r(&volume_LRU.q[idx]);
7724 /* check whether volume is safe to soft detach
7725 * caller MUST NOT hold a ref count on vp */
7727 VCheckSoftDetach(Volume * vp, afs_uint32 thresh)
7731 if (vp->nUsers || vp->nWaiters)
7734 if (vp->stats.last_get <= thresh) {
7735 ret = VSoftDetachVolume_r(vp, thresh);
7741 /* check whether volume should be made a
7742 * soft detach candidate */
7744 VCheckSoftDetachCandidate(Volume * vp, afs_uint32 thresh)
7747 if (vp->nUsers || vp->nWaiters)
7752 osi_Assert(idx == VLRU_QUEUE_NEW);
7754 if (vp->stats.last_get <= thresh) {
7755 /* move to candidate pool */
7756 queue_Remove(&vp->vlru);
7757 volume_LRU.q[VLRU_QUEUE_NEW].len--;
7758 queue_Prepend(&volume_LRU.q[VLRU_QUEUE_CANDIDATE], &vp->vlru);
7759 vp->vlru.idx = VLRU_QUEUE_CANDIDATE;
7760 volume_LRU.q[VLRU_QUEUE_CANDIDATE].len++;
7768 /* begin exclusive access on VLRU */
7770 VLRU_BeginExclusive_r(struct VLRU_q * q)
7772 osi_Assert(q->busy == 0);
7776 /* end exclusive access on VLRU */
7778 VLRU_EndExclusive_r(struct VLRU_q * q)
7780 osi_Assert(q->busy);
7782 CV_BROADCAST(&q->cv);
7785 /* wait for another thread to end exclusive access on VLRU */
7787 VLRU_Wait_r(struct VLRU_q * q)
7790 VOL_CV_WAIT(&q->cv);
7795 * volume soft detach
7797 * caller MUST NOT hold a ref count on vp */
7799 VSoftDetachVolume_r(Volume * vp, afs_uint32 thresh)
7804 osi_Assert(vp->vlru.idx == VLRU_QUEUE_CANDIDATE);
7806 ts_save = vp->stats.last_get;
7807 if (ts_save > thresh)
7810 if (vp->nUsers || vp->nWaiters)
7813 if (VIsExclusiveState(V_attachState(vp))) {
7817 switch (V_attachState(vp)) {
7818 case VOL_STATE_UNATTACHED:
7819 case VOL_STATE_PREATTACHED:
7820 case VOL_STATE_ERROR:
7821 case VOL_STATE_GOING_OFFLINE:
7822 case VOL_STATE_SHUTTING_DOWN:
7823 case VOL_STATE_SALVAGING:
7824 case VOL_STATE_DELETED:
7825 volume_LRU.q[vp->vlru.idx].len--;
7827 /* create and cancel a reservation to
7828 * give the volume an opportunity to
7830 VCreateReservation_r(vp);
7831 queue_Remove(&vp->vlru);
7832 vp->vlru.idx = VLRU_QUEUE_INVALID;
7833 V_attachFlags(vp) &= ~(VOL_ON_VLRU);
7834 VCancelReservation_r(vp);
7840 /* hold the volume and take it offline.
7841 * no need for reservations, as VHold_r
7842 * takes care of that internally. */
7843 if (VHold_r(vp) == 0) {
7844 /* vhold drops the glock, so now we should
7845 * check to make sure we aren't racing against
7846 * other threads. if we are racing, offlining vp
7847 * would be wasteful, and block the scanner for a while
7851 (vp->shuttingDown) ||
7852 (vp->goingOffline) ||
7853 (vp->stats.last_get != ts_save)) {
7854 /* looks like we're racing someone else. bail */
7858 /* pull it off the VLRU */
7859 osi_Assert(vp->vlru.idx == VLRU_QUEUE_CANDIDATE);
7860 volume_LRU.q[VLRU_QUEUE_CANDIDATE].len--;
7861 queue_Remove(&vp->vlru);
7862 vp->vlru.idx = VLRU_QUEUE_INVALID;
7863 V_attachFlags(vp) &= ~(VOL_ON_VLRU);
7865 /* take if offline */
7866 VOffline_r(vp, "volume has been soft detached");
7868 /* invalidate the volume header cache */
7869 FreeVolumeHeader(vp);
7872 IncUInt64(&VStats.soft_detaches);
7873 vp->stats.soft_detaches++;
7875 /* put in pre-attached state so demand
7876 * attacher can work on it */
7877 VChangeState_r(vp, VOL_STATE_PREATTACHED);
7883 #endif /* AFS_DEMAND_ATTACH_FS */
7886 /***************************************************/
7887 /* Volume Header Cache routines */
7888 /***************************************************/
7891 * volume header cache.
7893 struct volume_hdr_LRU_t volume_hdr_LRU;
7896 * initialize the volume header cache.
7898 * @param[in] howMany number of header cache entries to preallocate
7900 * @pre VOL_LOCK held. Function has never been called before.
7902 * @post howMany cache entries are allocated, initialized, and added
7903 * to the LRU list. Header cache statistics are initialized.
7905 * @note only applicable to fileServer program type. Should only be
7906 * called once during volume package initialization.
7908 * @internal volume package internal use only.
7911 VInitVolumeHeaderCache(afs_uint32 howMany)
7913 struct volHeader *hp;
7914 if (programType != fileServer)
7916 queue_Init(&volume_hdr_LRU);
7917 volume_hdr_LRU.stats.free = 0;
7918 volume_hdr_LRU.stats.used = howMany;
7919 volume_hdr_LRU.stats.attached = 0;
7920 hp = (struct volHeader *)(calloc(howMany, sizeof(struct volHeader)));
7921 osi_Assert(hp != NULL);
7924 /* We are using ReleaseVolumeHeader to initialize the values on the header list
7925 * to ensure they have the right values
7927 ReleaseVolumeHeader(hp++);
7930 /* get a volume header off of the volume header LRU.
7932 * @return volume header
7933 * @retval NULL no usable volume header is available on the LRU
7935 * @pre VOL_LOCK held
7937 * @post for DAFS, if the returned header is associated with a volume, that
7938 * volume is NOT in an exclusive state
7940 * @internal volume package internal use only.
7942 #ifdef AFS_DEMAND_ATTACH_FS
7943 static struct volHeader*
7944 GetVolHeaderFromLRU(void)
7946 struct volHeader *hd = NULL, *qh, *nqh;
7947 /* Usually, a volume in an exclusive state will not have its header on
7948 * the LRU. However, it is possible for this to occur when a salvage
7949 * request is received over FSSYNC, and possibly in other corner cases.
7950 * So just skip over headers whose volumes are in an exclusive state. We
7951 * could VWaitExclusiveState_r instead, but not waiting is faster and
7953 for (queue_Scan(&volume_hdr_LRU, qh, nqh, volHeader)) {
7954 if (!qh->back || !VIsExclusiveState(V_attachState(qh->back))) {
7962 #else /* AFS_DEMAND_ATTACH_FS */
7963 static struct volHeader*
7964 GetVolHeaderFromLRU(void)
7966 struct volHeader *hd = NULL;
7967 if (queue_IsNotEmpty(&volume_hdr_LRU)) {
7968 hd = queue_First(&volume_hdr_LRU, volHeader);
7973 #endif /* !AFS_DEMAND_ATTACH_FS */
7976 * get a volume header and attach it to the volume object.
7978 * @param[in] vp pointer to volume object
7980 * @return cache entry status
7981 * @retval 0 volume header was newly attached; cache data is invalid
7982 * @retval 1 volume header was previously attached; cache data is valid
7984 * @pre VOL_LOCK held. For DAFS, lightweight ref must be held on volume object.
7986 * @post volume header attached to volume object. if necessary, header cache
7987 * entry on LRU is synchronized to disk. Header is removed from LRU list.
7989 * @note VOL_LOCK may be dropped
7991 * @warning this interface does not load header data from disk. it merely
7992 * attaches a header object to the volume object, and may sync the old
7993 * header cache data out to disk in the process.
7995 * @internal volume package internal use only.
7998 GetVolumeHeader(Volume * vp)
8001 struct volHeader *hd;
8003 static int everLogged = 0;
8005 #ifdef AFS_DEMAND_ATTACH_FS
8006 VolState vp_save = 0, back_save = 0;
8008 /* XXX debug 9/19/05 we've apparently got
8009 * a ref counting bug somewhere that's
8010 * breaking the nUsers == 0 => header on LRU
8012 if (vp->header && queue_IsNotOnQueue(vp->header)) {
8013 Log("nUsers == 0, but header not on LRU\n");
8018 old = (vp->header != NULL); /* old == volume already has a header */
8020 if (programType != fileServer) {
8021 /* for volume utilities, we allocate volHeaders as needed */
8023 hd = (struct volHeader *)calloc(1, sizeof(*vp->header));
8024 osi_Assert(hd != NULL);
8027 #ifdef AFS_DEMAND_ATTACH_FS
8028 V_attachFlags(vp) |= VOL_HDR_ATTACHED;
8032 /* for the fileserver, we keep a volume header cache */
8034 /* the header we previously dropped in the lru is
8035 * still available. pull it off the lru and return */
8038 osi_Assert(hd->back == vp);
8039 #ifdef AFS_DEMAND_ATTACH_FS
8040 V_attachFlags(vp) &= ~(VOL_HDR_IN_LRU);
8043 hd = GetVolHeaderFromLRU();
8045 /* LRU is empty, so allocate a new volHeader
8046 * this is probably indicative of a leak, so let the user know */
8047 hd = (struct volHeader *)calloc(1, sizeof(struct volHeader));
8048 osi_Assert(hd != NULL);
8050 Log("****Allocated more volume headers, probably leak****\n");
8053 volume_hdr_LRU.stats.free++;
8056 /* this header used to belong to someone else.
8057 * we'll need to check if the header needs to
8058 * be sync'd out to disk */
8060 #ifdef AFS_DEMAND_ATTACH_FS
8061 /* GetVolHeaderFromLRU had better not give us back a header
8062 * with a volume in exclusive state... */
8063 osi_Assert(!VIsExclusiveState(V_attachState(hd->back)));
8066 if (hd->diskstuff.inUse) {
8067 /* volume was in use, so we'll need to sync
8068 * its header to disk */
8070 #ifdef AFS_DEMAND_ATTACH_FS
8071 back_save = VChangeState_r(hd->back, VOL_STATE_UPDATING);
8072 vp_save = VChangeState_r(vp, VOL_STATE_HDR_ATTACHING);
8073 VCreateReservation_r(hd->back);
8077 WriteVolumeHeader_r(&error, hd->back);
8078 /* Ignore errors; catch them later */
8080 #ifdef AFS_DEMAND_ATTACH_FS
8085 hd->back->header = NULL;
8086 #ifdef AFS_DEMAND_ATTACH_FS
8087 V_attachFlags(hd->back) &= ~(VOL_HDR_ATTACHED | VOL_HDR_LOADED | VOL_HDR_IN_LRU);
8089 if (hd->diskstuff.inUse) {
8090 VChangeState_r(hd->back, back_save);
8091 VCancelReservation_r(hd->back);
8092 VChangeState_r(vp, vp_save);
8096 volume_hdr_LRU.stats.attached++;
8100 #ifdef AFS_DEMAND_ATTACH_FS
8101 V_attachFlags(vp) |= VOL_HDR_ATTACHED;
8104 volume_hdr_LRU.stats.free--;
8105 volume_hdr_LRU.stats.used++;
8107 IncUInt64(&VStats.hdr_gets);
8108 #ifdef AFS_DEMAND_ATTACH_FS
8109 IncUInt64(&vp->stats.hdr_gets);
8110 vp->stats.last_hdr_get = FT_ApproxTime();
8117 * make sure volume header is attached and contains valid cache data.
8119 * @param[out] ec outbound error code
8120 * @param[in] vp pointer to volume object
8122 * @pre VOL_LOCK held. For DAFS, lightweight ref held on vp.
8124 * @post header cache entry attached, and loaded with valid data, or
8125 * *ec is nonzero, and the header is released back into the LRU.
8127 * @internal volume package internal use only.
8130 LoadVolumeHeader(Error * ec, Volume * vp)
8132 #ifdef AFS_DEMAND_ATTACH_FS
8133 VolState state_save;
8137 if (vp->nUsers == 0 && !GetVolumeHeader(vp)) {
8138 IncUInt64(&VStats.hdr_loads);
8139 state_save = VChangeState_r(vp, VOL_STATE_HDR_LOADING);
8142 ReadHeader(ec, V_diskDataHandle(vp), (char *)&V_disk(vp),
8143 sizeof(V_disk(vp)), VOLUMEINFOMAGIC,
8145 IncUInt64(&vp->stats.hdr_loads);
8146 now = FT_ApproxTime();
8150 V_attachFlags(vp) |= VOL_HDR_LOADED;
8151 vp->stats.last_hdr_load = now;
8153 VChangeState_r(vp, state_save);
8155 #else /* AFS_DEMAND_ATTACH_FS */
8157 if (vp->nUsers == 0 && !GetVolumeHeader(vp)) {
8158 IncUInt64(&VStats.hdr_loads);
8160 ReadHeader(ec, V_diskDataHandle(vp), (char *)&V_disk(vp),
8161 sizeof(V_disk(vp)), VOLUMEINFOMAGIC,
8164 #endif /* AFS_DEMAND_ATTACH_FS */
8166 /* maintain (nUsers==0) => header in LRU invariant */
8167 FreeVolumeHeader(vp);
8172 * release a header cache entry back into the LRU list.
8174 * @param[in] hd pointer to volume header cache object
8176 * @pre VOL_LOCK held.
8178 * @post header cache object appended onto end of LRU list.
8180 * @note only applicable to fileServer program type.
8182 * @note used to place a header cache entry back into the
8183 * LRU pool without invalidating it as a cache entry.
8185 * @internal volume package internal use only.
8188 ReleaseVolumeHeader(struct volHeader *hd)
8190 if (programType != fileServer)
8192 if (!hd || queue_IsOnQueue(hd)) /* no header, or header already released */
8194 queue_Append(&volume_hdr_LRU, hd);
8195 #ifdef AFS_DEMAND_ATTACH_FS
8197 V_attachFlags(hd->back) |= VOL_HDR_IN_LRU;
8200 volume_hdr_LRU.stats.free++;
8201 volume_hdr_LRU.stats.used--;
8205 * free/invalidate a volume header cache entry.
8207 * @param[in] vp pointer to volume object
8209 * @pre VOL_LOCK is held.
8211 * @post For fileserver, header cache entry is returned to LRU, and it is
8212 * invalidated as a cache entry. For volume utilities, the header
8213 * cache entry is freed.
8215 * @note For fileserver, this should be utilized instead of ReleaseVolumeHeader
8216 * whenever it is necessary to invalidate the header cache entry.
8218 * @see ReleaseVolumeHeader
8220 * @internal volume package internal use only.
8223 FreeVolumeHeader(Volume * vp)
8225 struct volHeader *hd = vp->header;
8228 if (programType == fileServer) {
8229 ReleaseVolumeHeader(hd);
8234 #ifdef AFS_DEMAND_ATTACH_FS
8235 V_attachFlags(vp) &= ~(VOL_HDR_ATTACHED | VOL_HDR_IN_LRU | VOL_HDR_LOADED);
8237 volume_hdr_LRU.stats.attached--;
8242 /***************************************************/
8243 /* Volume Hash Table routines */
8244 /***************************************************/
8247 * set size of volume object hash table.
8249 * @param[in] logsize log(2) of desired hash table size
8251 * @return operation status
8253 * @retval -1 failure
8255 * @pre MUST be called prior to VInitVolumePackage2
8257 * @post Volume Hash Table will have 2^logsize buckets
8260 VSetVolHashSize(int logsize)
8262 /* 64 to 268435456 hash buckets seems like a reasonable range */
8263 if ((logsize < 6 ) || (logsize > 28)) {
8268 VolumeHashTable.Size = 1 << logsize;
8269 VolumeHashTable.Mask = VolumeHashTable.Size - 1;
8271 /* we can't yet support runtime modification of this
8272 * parameter. we'll need a configuration rwlock to
8273 * make runtime modification feasible.... */
8280 * initialize dynamic data structures for volume hash table.
8282 * @post hash table is allocated, and fields are initialized.
8284 * @internal volume package internal use only.
8287 VInitVolumeHash(void)
8291 VolumeHashTable.Table = (VolumeHashChainHead *) calloc(VolumeHashTable.Size,
8292 sizeof(VolumeHashChainHead));
8293 osi_Assert(VolumeHashTable.Table != NULL);
8295 for (i=0; i < VolumeHashTable.Size; i++) {
8296 queue_Init(&VolumeHashTable.Table[i]);
8297 #ifdef AFS_DEMAND_ATTACH_FS
8298 CV_INIT(&VolumeHashTable.Table[i].chain_busy_cv, "vhash busy", CV_DEFAULT, 0);
8299 #endif /* AFS_DEMAND_ATTACH_FS */
8304 * add a volume object to the hash table.
8306 * @param[in] vp pointer to volume object
8307 * @param[in] hashid hash of volume id
8309 * @pre VOL_LOCK is held. For DAFS, caller must hold a lightweight
8312 * @post volume is added to hash chain.
8314 * @internal volume package internal use only.
8316 * @note For DAFS, VOL_LOCK may be dropped in order to wait for an
8317 * asynchronous hash chain reordering to finish.
8320 AddVolumeToHashTable(Volume * vp, int hashid)
8322 VolumeHashChainHead * head;
8324 if (queue_IsOnQueue(vp))
8327 head = &VolumeHashTable.Table[VOLUME_HASH(hashid)];
8329 #ifdef AFS_DEMAND_ATTACH_FS
8330 /* wait for the hash chain to become available */
8333 V_attachFlags(vp) |= VOL_IN_HASH;
8334 vp->chainCacheCheck = ++head->cacheCheck;
8335 #endif /* AFS_DEMAND_ATTACH_FS */
8338 vp->hashid = hashid;
8339 queue_Append(head, vp);
8340 vp->vnodeHashOffset = VolumeHashOffset_r();
8344 * delete a volume object from the hash table.
8346 * @param[in] vp pointer to volume object
8348 * @pre VOL_LOCK is held. For DAFS, caller must hold a lightweight
8351 * @post volume is removed from hash chain.
8353 * @internal volume package internal use only.
8355 * @note For DAFS, VOL_LOCK may be dropped in order to wait for an
8356 * asynchronous hash chain reordering to finish.
8359 DeleteVolumeFromHashTable(Volume * vp)
8361 VolumeHashChainHead * head;
8363 if (!queue_IsOnQueue(vp))
8366 head = &VolumeHashTable.Table[VOLUME_HASH(vp->hashid)];
8368 #ifdef AFS_DEMAND_ATTACH_FS
8369 /* wait for the hash chain to become available */
8372 V_attachFlags(vp) &= ~(VOL_IN_HASH);
8374 #endif /* AFS_DEMAND_ATTACH_FS */
8378 /* do NOT reset hashid to zero, as the online
8379 * salvager package may need to know the volume id
8380 * after the volume is removed from the hash */
8384 * lookup a volume object in the hash table given a volume id.
8386 * @param[out] ec error code return
8387 * @param[in] volumeId volume id
8388 * @param[in] hint volume object which we believe could be the correct
8391 * @return volume object pointer
8392 * @retval NULL no such volume id is registered with the hash table.
8394 * @pre VOL_LOCK is held. For DAFS, caller must hold a lightweight
8397 * @post volume object with the given id is returned. volume object and
8398 * hash chain access statistics are updated. hash chain may have
8401 * @note For DAFS, VOL_LOCK may be dropped in order to wait for an
8402 * asynchronous hash chain reordering operation to finish, or
8403 * in order for us to perform an asynchronous chain reordering.
8405 * @note Hash chain reorderings occur when the access count for the
8406 * volume object being looked up exceeds the sum of the previous
8407 * node's (the node ahead of it in the hash chain linked list)
8408 * access count plus the constant VOLUME_HASH_REORDER_THRESHOLD.
8410 * @note For DAFS, the hint parameter allows us to short-circuit if the
8411 * cacheCheck fields match between the hash chain head and the
8412 * hint volume object.
8415 VLookupVolume_r(Error * ec, VolId volumeId, Volume * hint)
8419 #ifdef AFS_DEMAND_ATTACH_FS
8422 VolumeHashChainHead * head;
8425 head = &VolumeHashTable.Table[VOLUME_HASH(volumeId)];
8427 #ifdef AFS_DEMAND_ATTACH_FS
8428 /* wait for the hash chain to become available */
8431 /* check to see if we can short circuit without walking the hash chain */
8432 if (hint && (hint->chainCacheCheck == head->cacheCheck)) {
8433 IncUInt64(&hint->stats.hash_short_circuits);
8436 #endif /* AFS_DEMAND_ATTACH_FS */
8438 /* someday we need to either do per-chain locks, RWlocks,
8439 * or both for volhash access.
8440 * (and move to a data structure with better cache locality) */
8442 /* search the chain for this volume id */
8443 for(queue_Scan(head, vp, np, Volume)) {
8445 if ((vp->hashid == volumeId)) {
8450 if (queue_IsEnd(head, vp)) {
8454 #ifdef AFS_DEMAND_ATTACH_FS
8455 /* update hash chain statistics */
8458 FillInt64(lks, 0, looks);
8459 AddUInt64(head->looks, lks, &head->looks);
8460 AddUInt64(VStats.hash_looks, lks, &VStats.hash_looks);
8461 IncUInt64(&head->gets);
8466 IncUInt64(&vp->stats.hash_lookups);
8468 /* for demand attach fileserver, we permit occasional hash chain reordering
8469 * so that frequently looked up volumes move towards the head of the chain */
8470 pp = queue_Prev(vp, Volume);
8471 if (!queue_IsEnd(head, pp)) {
8472 FillInt64(thresh, 0, VOLUME_HASH_REORDER_THRESHOLD);
8473 AddUInt64(thresh, pp->stats.hash_lookups, &thresh);
8474 if (GEInt64(vp->stats.hash_lookups, thresh)) {
8475 VReorderHash_r(head, pp, vp);
8479 /* update the short-circuit cache check */
8480 vp->chainCacheCheck = head->cacheCheck;
8482 #endif /* AFS_DEMAND_ATTACH_FS */
8487 #ifdef AFS_DEMAND_ATTACH_FS
8488 /* perform volume hash chain reordering.
8490 * advance a subchain beginning at vp ahead of
8491 * the adjacent subchain ending at pp */
8493 VReorderHash_r(VolumeHashChainHead * head, Volume * pp, Volume * vp)
8495 Volume *tp, *np, *lp;
8496 afs_uint64 move_thresh;
8498 /* this should never be called if the chain is already busy, so
8499 * no need to wait for other exclusive chain ops to finish */
8501 /* this is a rather heavy set of operations,
8502 * so let's set the chain busy flag and drop
8504 VHashBeginExclusive_r(head);
8507 /* scan forward in the chain from vp looking for the last element
8508 * in the chain we want to advance */
8509 FillInt64(move_thresh, 0, VOLUME_HASH_REORDER_CHAIN_THRESH);
8510 AddUInt64(move_thresh, pp->stats.hash_lookups, &move_thresh);
8511 for(queue_ScanFrom(head, vp, tp, np, Volume)) {
8512 if (LTInt64(tp->stats.hash_lookups, move_thresh)) {
8516 lp = queue_Prev(tp, Volume);
8518 /* scan backwards from pp to determine where to splice and
8519 * insert the subchain we're advancing */
8520 for(queue_ScanBackwardsFrom(head, pp, tp, np, Volume)) {
8521 if (GTInt64(tp->stats.hash_lookups, move_thresh)) {
8525 tp = queue_Next(tp, Volume);
8527 /* rebalance chain(vp,...,lp) ahead of chain(tp,...,pp) */
8528 queue_MoveChainBefore(tp,vp,lp);
8531 IncUInt64(&VStats.hash_reorders);
8533 IncUInt64(&head->reorders);
8535 /* wake up any threads waiting for the hash chain */
8536 VHashEndExclusive_r(head);
8540 /* demand-attach fs volume hash
8541 * asynchronous exclusive operations */
8544 * begin an asynchronous exclusive operation on a volume hash chain.
8546 * @param[in] head pointer to volume hash chain head object
8548 * @pre VOL_LOCK held. hash chain is quiescent.
8550 * @post hash chain marked busy.
8552 * @note this interface is used in conjunction with VHashEndExclusive_r and
8553 * VHashWait_r to perform asynchronous (wrt VOL_LOCK) operations on a
8554 * volume hash chain. Its main use case is hash chain reordering, which
8555 * has the potential to be a highly latent operation.
8557 * @see VHashEndExclusive_r
8562 * @internal volume package internal use only.
8565 VHashBeginExclusive_r(VolumeHashChainHead * head)
8567 osi_Assert(head->busy == 0);
8572 * relinquish exclusive ownership of a volume hash chain.
8574 * @param[in] head pointer to volume hash chain head object
8576 * @pre VOL_LOCK held. thread owns the hash chain exclusively.
8578 * @post hash chain is marked quiescent. threads awaiting use of
8579 * chain are awakened.
8581 * @see VHashBeginExclusive_r
8586 * @internal volume package internal use only.
8589 VHashEndExclusive_r(VolumeHashChainHead * head)
8591 osi_Assert(head->busy);
8593 CV_BROADCAST(&head->chain_busy_cv);
8597 * wait for all asynchronous operations on a hash chain to complete.
8599 * @param[in] head pointer to volume hash chain head object
8601 * @pre VOL_LOCK held.
8603 * @post hash chain object is quiescent.
8605 * @see VHashBeginExclusive_r
8606 * @see VHashEndExclusive_r
8610 * @note This interface should be called before any attempt to
8611 * traverse the hash chain. It is permissible for a thread
8612 * to gain exclusive access to the chain, and then perform
8613 * latent operations on the chain asynchronously wrt the
8616 * @warning if waiting is necessary, VOL_LOCK is dropped
8618 * @internal volume package internal use only.
8621 VHashWait_r(VolumeHashChainHead * head)
8623 while (head->busy) {
8624 VOL_CV_WAIT(&head->chain_busy_cv);
8627 #endif /* AFS_DEMAND_ATTACH_FS */
8630 /***************************************************/
8631 /* Volume by Partition List routines */
8632 /***************************************************/
8635 * demand attach fileserver adds a
8636 * linked list of volumes to each
8637 * partition object, thus allowing
8638 * for quick enumeration of all
8639 * volumes on a partition
8642 #ifdef AFS_DEMAND_ATTACH_FS
8644 * add a volume to its disk partition VByPList.
8646 * @param[in] vp pointer to volume object
8648 * @pre either the disk partition VByPList is owned exclusively
8649 * by the calling thread, or the list is quiescent and
8652 * @post volume is added to disk partition VByPList
8656 * @warning it is the caller's responsibility to ensure list
8659 * @see VVByPListWait_r
8660 * @see VVByPListBeginExclusive_r
8661 * @see VVByPListEndExclusive_r
8663 * @internal volume package internal use only.
8666 AddVolumeToVByPList_r(Volume * vp)
8668 if (queue_IsNotOnQueue(&vp->vol_list)) {
8669 queue_Append(&vp->partition->vol_list, &vp->vol_list);
8670 V_attachFlags(vp) |= VOL_ON_VBYP_LIST;
8671 vp->partition->vol_list.len++;
8676 * delete a volume from its disk partition VByPList.
8678 * @param[in] vp pointer to volume object
8680 * @pre either the disk partition VByPList is owned exclusively
8681 * by the calling thread, or the list is quiescent and
8684 * @post volume is removed from the disk partition VByPList
8688 * @warning it is the caller's responsibility to ensure list
8691 * @see VVByPListWait_r
8692 * @see VVByPListBeginExclusive_r
8693 * @see VVByPListEndExclusive_r
8695 * @internal volume package internal use only.
8698 DeleteVolumeFromVByPList_r(Volume * vp)
8700 if (queue_IsOnQueue(&vp->vol_list)) {
8701 queue_Remove(&vp->vol_list);
8702 V_attachFlags(vp) &= ~(VOL_ON_VBYP_LIST);
8703 vp->partition->vol_list.len--;
8708 * begin an asynchronous exclusive operation on a VByPList.
8710 * @param[in] dp pointer to disk partition object
8712 * @pre VOL_LOCK held. VByPList is quiescent.
8714 * @post VByPList marked busy.
8716 * @note this interface is used in conjunction with VVByPListEndExclusive_r and
8717 * VVByPListWait_r to perform asynchronous (wrt VOL_LOCK) operations on a
8720 * @see VVByPListEndExclusive_r
8721 * @see VVByPListWait_r
8725 * @internal volume package internal use only.
8727 /* take exclusive control over the list */
8729 VVByPListBeginExclusive_r(struct DiskPartition64 * dp)
8731 osi_Assert(dp->vol_list.busy == 0);
8732 dp->vol_list.busy = 1;
8736 * relinquish exclusive ownership of a VByPList.
8738 * @param[in] dp pointer to disk partition object
8740 * @pre VOL_LOCK held. thread owns the VByPList exclusively.
8742 * @post VByPList is marked quiescent. threads awaiting use of
8743 * the list are awakened.
8745 * @see VVByPListBeginExclusive_r
8746 * @see VVByPListWait_r
8750 * @internal volume package internal use only.
8753 VVByPListEndExclusive_r(struct DiskPartition64 * dp)
8755 osi_Assert(dp->vol_list.busy);
8756 dp->vol_list.busy = 0;
8757 CV_BROADCAST(&dp->vol_list.cv);
8761 * wait for all asynchronous operations on a VByPList to complete.
8763 * @param[in] dp pointer to disk partition object
8765 * @pre VOL_LOCK is held.
8767 * @post disk partition's VByP list is quiescent
8771 * @note This interface should be called before any attempt to
8772 * traverse the VByPList. It is permissible for a thread
8773 * to gain exclusive access to the list, and then perform
8774 * latent operations on the list asynchronously wrt the
8777 * @warning if waiting is necessary, VOL_LOCK is dropped
8779 * @see VVByPListEndExclusive_r
8780 * @see VVByPListBeginExclusive_r
8782 * @internal volume package internal use only.
8785 VVByPListWait_r(struct DiskPartition64 * dp)
8787 while (dp->vol_list.busy) {
8788 VOL_CV_WAIT(&dp->vol_list.cv);
8791 #endif /* AFS_DEMAND_ATTACH_FS */
8793 /***************************************************/
8794 /* Volume Cache Statistics routines */
8795 /***************************************************/
8798 VPrintCacheStats_r(void)
8800 struct VnodeClassInfo *vcp;
8801 vcp = &VnodeClassInfo[vLarge];
8802 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);
8803 vcp = &VnodeClassInfo[vSmall];
8804 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);
8805 Log("Volume header cache, %d entries, %"AFS_INT64_FMT" gets, "
8806 "%"AFS_INT64_FMT" replacements\n",
8807 VStats.hdr_cache_size, VStats.hdr_gets, VStats.hdr_loads);
8811 VPrintCacheStats(void)
8814 VPrintCacheStats_r();
8818 #ifdef AFS_DEMAND_ATTACH_FS
8820 UInt64ToDouble(afs_uint64 * x)
8822 static double c32 = 4.0 * 1.073741824 * 1000000000.0;
8824 SplitInt64(*x, h, l);
8825 return (((double)h) * c32) + ((double) l);
8829 DoubleToPrintable(double x, char * buf, int len)
8831 static double billion = 1000000000.0;
8834 y[0] = (afs_uint32) (x / (billion * billion));
8835 y[1] = (afs_uint32) ((x - (((double)y[0]) * billion * billion)) / billion);
8836 y[2] = (afs_uint32) (x - ((((double)y[0]) * billion * billion) + (((double)y[1]) * billion)));
8839 snprintf(buf, len, "%d%09d%09d", y[0], y[1], y[2]);
8841 snprintf(buf, len, "%d%09d", y[1], y[2]);
8843 snprintf(buf, len, "%d", y[2]);
8849 struct VLRUExtStatsEntry {
8853 struct VLRUExtStats {
8859 } queue_info[VLRU_QUEUE_INVALID];
8860 struct VLRUExtStatsEntry * vec;
8864 * add a 256-entry fudge factor onto the vector in case state changes
8865 * out from under us.
8867 #define VLRU_EXT_STATS_VEC_LEN_FUDGE 256
8870 * collect extended statistics for the VLRU subsystem.
8872 * @param[out] stats pointer to stats structure to be populated
8873 * @param[in] nvols number of volumes currently known to exist
8875 * @pre VOL_LOCK held
8877 * @post stats->vec allocated and populated
8879 * @return operation status
8884 VVLRUExtStats_r(struct VLRUExtStats * stats, afs_uint32 nvols)
8886 afs_uint32 cur, idx, len;
8887 struct rx_queue * qp, * nqp;
8889 struct VLRUExtStatsEntry * vec;
8891 len = nvols + VLRU_EXT_STATS_VEC_LEN_FUDGE;
8892 vec = stats->vec = calloc(len,
8893 sizeof(struct VLRUExtStatsEntry));
8899 for (idx = VLRU_QUEUE_NEW; idx < VLRU_QUEUE_INVALID; idx++) {
8900 VLRU_Wait_r(&volume_LRU.q[idx]);
8901 VLRU_BeginExclusive_r(&volume_LRU.q[idx]);
8904 stats->queue_info[idx].start = cur;
8906 for (queue_Scan(&volume_LRU.q[idx], qp, nqp, rx_queue)) {
8908 /* out of space in vec */
8911 vp = (Volume *)((char *)qp - offsetof(Volume, vlru));
8912 vec[cur].volid = vp->hashid;
8916 stats->queue_info[idx].len = cur - stats->queue_info[idx].start;
8919 VLRU_EndExclusive_r(&volume_LRU.q[idx]);
8927 #define ENUMTOSTRING(en) #en
8928 #define ENUMCASE(en) \
8929 case en: return ENUMTOSTRING(en)
8932 vlru_idx_to_string(int idx)
8935 ENUMCASE(VLRU_QUEUE_NEW);
8936 ENUMCASE(VLRU_QUEUE_MID);
8937 ENUMCASE(VLRU_QUEUE_OLD);
8938 ENUMCASE(VLRU_QUEUE_CANDIDATE);
8939 ENUMCASE(VLRU_QUEUE_HELD);
8940 ENUMCASE(VLRU_QUEUE_INVALID);
8942 return "**UNKNOWN**";
8947 VPrintExtendedCacheStats_r(int flags)
8950 afs_uint32 vol_sum = 0;
8957 struct stats looks, gets, reorders, len;
8958 struct stats ch_looks, ch_gets, ch_reorders;
8960 VolumeHashChainHead *head;
8962 struct VLRUExtStats vlru_stats;
8964 /* zero out stats */
8965 memset(&looks, 0, sizeof(struct stats));
8966 memset(&gets, 0, sizeof(struct stats));
8967 memset(&reorders, 0, sizeof(struct stats));
8968 memset(&len, 0, sizeof(struct stats));
8969 memset(&ch_looks, 0, sizeof(struct stats));
8970 memset(&ch_gets, 0, sizeof(struct stats));
8971 memset(&ch_reorders, 0, sizeof(struct stats));
8973 for (i = 0; i < VolumeHashTable.Size; i++) {
8974 head = &VolumeHashTable.Table[i];
8977 VHashBeginExclusive_r(head);
8980 ch_looks.sum = UInt64ToDouble(&head->looks);
8981 ch_gets.sum = UInt64ToDouble(&head->gets);
8982 ch_reorders.sum = UInt64ToDouble(&head->reorders);
8984 /* update global statistics */
8986 looks.sum += ch_looks.sum;
8987 gets.sum += ch_gets.sum;
8988 reorders.sum += ch_reorders.sum;
8989 len.sum += (double)head->len;
8990 vol_sum += head->len;
8993 len.min = (double) head->len;
8994 len.max = (double) head->len;
8995 looks.min = ch_looks.sum;
8996 looks.max = ch_looks.sum;
8997 gets.min = ch_gets.sum;
8998 gets.max = ch_gets.sum;
8999 reorders.min = ch_reorders.sum;
9000 reorders.max = ch_reorders.sum;
9002 if (((double)head->len) < len.min)
9003 len.min = (double) head->len;
9004 if (((double)head->len) > len.max)
9005 len.max = (double) head->len;
9006 if (ch_looks.sum < looks.min)
9007 looks.min = ch_looks.sum;
9008 else if (ch_looks.sum > looks.max)
9009 looks.max = ch_looks.sum;
9010 if (ch_gets.sum < gets.min)
9011 gets.min = ch_gets.sum;
9012 else if (ch_gets.sum > gets.max)
9013 gets.max = ch_gets.sum;
9014 if (ch_reorders.sum < reorders.min)
9015 reorders.min = ch_reorders.sum;
9016 else if (ch_reorders.sum > reorders.max)
9017 reorders.max = ch_reorders.sum;
9021 if ((flags & VOL_STATS_PER_CHAIN2) && queue_IsNotEmpty(head)) {
9022 /* compute detailed per-chain stats */
9023 struct stats hdr_loads, hdr_gets;
9024 double v_looks, v_loads, v_gets;
9026 /* initialize stats with data from first element in chain */
9027 vp = queue_First(head, Volume);
9028 v_looks = UInt64ToDouble(&vp->stats.hash_lookups);
9029 v_loads = UInt64ToDouble(&vp->stats.hdr_loads);
9030 v_gets = UInt64ToDouble(&vp->stats.hdr_gets);
9031 ch_gets.min = ch_gets.max = v_looks;
9032 hdr_loads.min = hdr_loads.max = v_loads;
9033 hdr_gets.min = hdr_gets.max = v_gets;
9034 hdr_loads.sum = hdr_gets.sum = 0;
9036 vp = queue_Next(vp, Volume);
9038 /* pull in stats from remaining elements in chain */
9039 for (queue_ScanFrom(head, vp, vp, np, Volume)) {
9040 v_looks = UInt64ToDouble(&vp->stats.hash_lookups);
9041 v_loads = UInt64ToDouble(&vp->stats.hdr_loads);
9042 v_gets = UInt64ToDouble(&vp->stats.hdr_gets);
9044 hdr_loads.sum += v_loads;
9045 hdr_gets.sum += v_gets;
9047 if (v_looks < ch_gets.min)
9048 ch_gets.min = v_looks;
9049 else if (v_looks > ch_gets.max)
9050 ch_gets.max = v_looks;
9052 if (v_loads < hdr_loads.min)
9053 hdr_loads.min = v_loads;
9054 else if (v_loads > hdr_loads.max)
9055 hdr_loads.max = v_loads;
9057 if (v_gets < hdr_gets.min)
9058 hdr_gets.min = v_gets;
9059 else if (v_gets > hdr_gets.max)
9060 hdr_gets.max = v_gets;
9063 /* compute per-chain averages */
9064 ch_gets.avg = ch_gets.sum / ((double)head->len);
9065 hdr_loads.avg = hdr_loads.sum / ((double)head->len);
9066 hdr_gets.avg = hdr_gets.sum / ((double)head->len);
9068 /* dump per-chain stats */
9069 Log("Volume hash chain %d : len=%d, looks=%s, reorders=%s\n",
9071 DoubleToPrintable(ch_looks.sum, pr_buf[0], sizeof(pr_buf[0])),
9072 DoubleToPrintable(ch_reorders.sum, pr_buf[1], sizeof(pr_buf[1])));
9073 Log("\tVolume gets : min=%s, max=%s, avg=%s, total=%s\n",
9074 DoubleToPrintable(ch_gets.min, pr_buf[0], sizeof(pr_buf[0])),
9075 DoubleToPrintable(ch_gets.max, pr_buf[1], sizeof(pr_buf[1])),
9076 DoubleToPrintable(ch_gets.avg, pr_buf[2], sizeof(pr_buf[2])),
9077 DoubleToPrintable(ch_gets.sum, pr_buf[3], sizeof(pr_buf[3])));
9078 Log("\tHDR gets : min=%s, max=%s, avg=%s, total=%s\n",
9079 DoubleToPrintable(hdr_gets.min, pr_buf[0], sizeof(pr_buf[0])),
9080 DoubleToPrintable(hdr_gets.max, pr_buf[1], sizeof(pr_buf[1])),
9081 DoubleToPrintable(hdr_gets.avg, pr_buf[2], sizeof(pr_buf[2])),
9082 DoubleToPrintable(hdr_gets.sum, pr_buf[3], sizeof(pr_buf[3])));
9083 Log("\tHDR loads : min=%s, max=%s, avg=%s, total=%s\n",
9084 DoubleToPrintable(hdr_loads.min, pr_buf[0], sizeof(pr_buf[0])),
9085 DoubleToPrintable(hdr_loads.max, pr_buf[1], sizeof(pr_buf[1])),
9086 DoubleToPrintable(hdr_loads.avg, pr_buf[2], sizeof(pr_buf[2])),
9087 DoubleToPrintable(hdr_loads.sum, pr_buf[3], sizeof(pr_buf[3])));
9088 } else if (flags & VOL_STATS_PER_CHAIN) {
9089 /* dump simple per-chain stats */
9090 Log("Volume hash chain %d : len=%d, looks=%s, gets=%s, reorders=%s\n",
9092 DoubleToPrintable(ch_looks.sum, pr_buf[0], sizeof(pr_buf[0])),
9093 DoubleToPrintable(ch_gets.sum, pr_buf[1], sizeof(pr_buf[1])),
9094 DoubleToPrintable(ch_reorders.sum, pr_buf[2], sizeof(pr_buf[2])));
9098 VHashEndExclusive_r(head);
9103 /* compute global averages */
9104 len.avg = len.sum / ((double)VolumeHashTable.Size);
9105 looks.avg = looks.sum / ((double)VolumeHashTable.Size);
9106 gets.avg = gets.sum / ((double)VolumeHashTable.Size);
9107 reorders.avg = reorders.sum / ((double)VolumeHashTable.Size);
9109 /* dump global stats */
9110 Log("Volume hash summary: %d buckets\n", VolumeHashTable.Size);
9111 Log(" chain length : min=%s, max=%s, avg=%s, total=%s\n",
9112 DoubleToPrintable(len.min, pr_buf[0], sizeof(pr_buf[0])),
9113 DoubleToPrintable(len.max, pr_buf[1], sizeof(pr_buf[1])),
9114 DoubleToPrintable(len.avg, pr_buf[2], sizeof(pr_buf[2])),
9115 DoubleToPrintable(len.sum, pr_buf[3], sizeof(pr_buf[3])));
9116 Log(" looks : min=%s, max=%s, avg=%s, total=%s\n",
9117 DoubleToPrintable(looks.min, pr_buf[0], sizeof(pr_buf[0])),
9118 DoubleToPrintable(looks.max, pr_buf[1], sizeof(pr_buf[1])),
9119 DoubleToPrintable(looks.avg, pr_buf[2], sizeof(pr_buf[2])),
9120 DoubleToPrintable(looks.sum, pr_buf[3], sizeof(pr_buf[3])));
9121 Log(" gets : min=%s, max=%s, avg=%s, total=%s\n",
9122 DoubleToPrintable(gets.min, pr_buf[0], sizeof(pr_buf[0])),
9123 DoubleToPrintable(gets.max, pr_buf[1], sizeof(pr_buf[1])),
9124 DoubleToPrintable(gets.avg, pr_buf[2], sizeof(pr_buf[2])),
9125 DoubleToPrintable(gets.sum, pr_buf[3], sizeof(pr_buf[3])));
9126 Log(" reorders : min=%s, max=%s, avg=%s, total=%s\n",
9127 DoubleToPrintable(reorders.min, pr_buf[0], sizeof(pr_buf[0])),
9128 DoubleToPrintable(reorders.max, pr_buf[1], sizeof(pr_buf[1])),
9129 DoubleToPrintable(reorders.avg, pr_buf[2], sizeof(pr_buf[2])),
9130 DoubleToPrintable(reorders.sum, pr_buf[3], sizeof(pr_buf[3])));
9132 /* print extended disk related statistics */
9134 struct DiskPartition64 * diskP;
9135 afs_uint32 vol_count[VOLMAXPARTS+1];
9136 byte part_exists[VOLMAXPARTS+1];
9140 memset(vol_count, 0, sizeof(vol_count));
9141 memset(part_exists, 0, sizeof(part_exists));
9145 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
9147 vol_count[id] = diskP->vol_list.len;
9148 part_exists[id] = 1;
9152 for (i = 0; i <= VOLMAXPARTS; i++) {
9153 if (part_exists[i]) {
9154 /* XXX while this is currently safe, it is a violation
9155 * of the VGetPartitionById_r interface contract. */
9156 diskP = VGetPartitionById_r(i, 0);
9158 Log("Partition %s has %d online volumes\n",
9159 VPartitionPath(diskP), diskP->vol_list.len);
9166 /* print extended VLRU statistics */
9167 if (VVLRUExtStats_r(&vlru_stats, vol_sum) == 0) {
9168 afs_uint32 idx, cur, lpos;
9173 Log("VLRU State Dump:\n\n");
9175 for (idx = VLRU_QUEUE_NEW; idx < VLRU_QUEUE_INVALID; idx++) {
9176 Log("\t%s:\n", vlru_idx_to_string(idx));
9179 for (cur = vlru_stats.queue_info[idx].start;
9180 cur < vlru_stats.queue_info[idx].len;
9182 line[lpos++] = vlru_stats.vec[cur].volid;
9184 Log("\t\t%u, %u, %u, %u, %u,\n",
9185 line[0], line[1], line[2], line[3], line[4]);
9194 Log("\t\t%u, %u, %u, %u, %u\n",
9195 line[0], line[1], line[2], line[3], line[4]);
9200 free(vlru_stats.vec);
9207 VPrintExtendedCacheStats(int flags)
9210 VPrintExtendedCacheStats_r(flags);
9213 #endif /* AFS_DEMAND_ATTACH_FS */
9216 VCanScheduleSalvage(void)
9218 return vol_opts.canScheduleSalvage;
9224 return vol_opts.canUseFSSYNC;
9228 VCanUseSALVSYNC(void)
9230 return vol_opts.canUseSALVSYNC;
9234 VCanUnsafeAttach(void)
9236 return vol_opts.unsafe_attach;