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
34 #ifdef AFS_PTHREAD_ENV
35 # include <opr/lock.h>
37 # include <opr/lockstub.h>
40 #include <afs/afsint.h>
42 #include <rx/rx_queue.h>
45 #if !defined(AFS_SGI_ENV)
48 #else /* AFS_OSF_ENV */
49 #ifdef AFS_VFSINCL_ENV
52 #include <sys/fs/ufs_fs.h>
54 #if defined(AFS_DARWIN_ENV) || defined(AFS_XBSD_ENV)
55 #include <ufs/ufs/dinode.h>
56 #include <ufs/ffs/fs.h>
61 #else /* AFS_VFSINCL_ENV */
62 #if !defined(AFS_AIX_ENV) && !defined(AFS_LINUX20_ENV) && !defined(AFS_XBSD_ENV) && !defined(AFS_DARWIN_ENV)
65 #endif /* AFS_VFSINCL_ENV */
66 #endif /* AFS_OSF_ENV */
67 #endif /* AFS_SGI_ENV */
68 #endif /* !AFS_NT40_ENV */
76 #if defined(AFS_SUN_ENV) || defined(AFS_SUN5_ENV)
78 #include <sys/mnttab.h>
79 #include <sys/mntent.h>
85 #if defined(AFS_SGI_ENV)
88 #ifndef AFS_LINUX20_ENV
89 #include <fstab.h> /* Need to find in libc 5, present in libc 6 */
92 #endif /* AFS_SGI_ENV */
94 #endif /* AFS_HPUX_ENV */
98 #include <afs/errors.h>
101 #include <afs/afssyscalls.h>
103 #include <afs/afsutil.h>
104 #include "daemon_com.h"
106 #include "salvsync.h"
109 #include "partition.h"
110 #include "volume_inline.h"
115 #ifdef AFS_PTHREAD_ENV
116 pthread_mutex_t vol_glock_mutex;
117 pthread_mutex_t vol_trans_mutex;
118 pthread_cond_t vol_put_volume_cond;
119 pthread_cond_t vol_sleep_cond;
120 pthread_cond_t vol_init_attach_cond;
121 pthread_cond_t vol_vinit_cond;
122 int vol_attach_threads = 1;
123 #endif /* AFS_PTHREAD_ENV */
125 /* start-time configurable I/O parameters */
126 ih_init_params vol_io_params;
128 #ifdef AFS_DEMAND_ATTACH_FS
129 pthread_mutex_t vol_salvsync_mutex;
132 * Set this to 1 to disallow SALVSYNC communication in all threads; used
133 * during shutdown, since the salvageserver may have gone away.
135 static volatile sig_atomic_t vol_disallow_salvsync = 0;
136 #endif /* AFS_DEMAND_ATTACH_FS */
139 * has VShutdown_r been called / is VShutdown_r running?
141 static int vol_shutting_down = 0;
144 extern void *calloc(), *realloc();
147 /* Forward declarations */
148 static Volume *attach2(Error * ec, VolumeId volumeId, char *path,
149 struct DiskPartition64 *partp, Volume * vp,
150 int isbusy, int mode, int *acheckedOut);
151 static void ReallyFreeVolume(Volume * vp);
152 #ifdef AFS_DEMAND_ATTACH_FS
153 static void FreeVolume(Volume * vp);
154 #else /* !AFS_DEMAND_ATTACH_FS */
155 #define FreeVolume(vp) ReallyFreeVolume(vp)
156 static void VScanUpdateList(void);
157 #endif /* !AFS_DEMAND_ATTACH_FS */
158 static void VInitVolumeHeaderCache(afs_uint32 howMany);
159 static int GetVolumeHeader(Volume * vp);
160 static void ReleaseVolumeHeader(struct volHeader *hd);
161 static void FreeVolumeHeader(Volume * vp);
162 static void AddVolumeToHashTable(Volume * vp, VolumeId hashid);
163 static void DeleteVolumeFromHashTable(Volume * vp);
165 static int VHold(Volume * vp);
167 static int VHold_r(Volume * vp);
168 static void VGetBitmap_r(Error * ec, Volume * vp, VnodeClass class);
169 static void VReleaseVolumeHandles_r(Volume * vp);
170 static void VCloseVolumeHandles_r(Volume * vp);
171 static void LoadVolumeHeader(Error * ec, Volume * vp);
172 static int VCheckOffline(Volume * vp);
173 static int VCheckDetach(Volume * vp);
174 static Volume * GetVolume(Error * ec, Error * client_ec, VolumeId volumeId,
175 Volume * hint, const struct timespec *ts);
177 int LogLevel; /* Vice loglevel--not defined as extern so that it will be
178 * defined when not linked with vice, XXXX */
179 ProgramType programType; /* The type of program using the package */
180 static VolumePackageOptions vol_opts;
182 /* extended volume package statistics */
185 #ifdef VOL_LOCK_DEBUG
186 pthread_t vol_glock_holder = 0;
190 /* this parameter needs to be tunable at runtime.
191 * 128 was really inadequate for largish servers -- at 16384 volumes this
192 * puts average chain length at 128, thus an average 65 deref's to find a volptr.
193 * talk about bad spatial locality...
195 * an AVL or splay tree might work a lot better, but we'll just increase
196 * the default hash table size for now
198 #define DEFAULT_VOLUME_HASH_SIZE 256 /* Must be a power of 2!! */
199 #define DEFAULT_VOLUME_HASH_MASK (DEFAULT_VOLUME_HASH_SIZE-1)
200 #define VOLUME_HASH(volumeId) (volumeId&(VolumeHashTable.Mask))
203 * turn volume hash chains into partially ordered lists.
204 * when the threshold is exceeded between two adjacent elements,
205 * perform a chain rebalancing operation.
207 * keep the threshold high in order to keep cache line invalidates
208 * low "enough" on SMPs
210 #define VOLUME_HASH_REORDER_THRESHOLD 200
213 * when possible, don't just reorder single elements, but reorder
214 * entire chains of elements at once. a chain of elements that
215 * exceed the element previous to the pivot by at least CHAIN_THRESH
216 * accesses are moved in front of the chain whose elements have at
217 * least CHAIN_THRESH less accesses than the pivot element
219 #define VOLUME_HASH_REORDER_CHAIN_THRESH (VOLUME_HASH_REORDER_THRESHOLD / 2)
221 #include "rx/rx_queue.h"
224 VolumeHashTable_t VolumeHashTable = {
225 DEFAULT_VOLUME_HASH_SIZE,
226 DEFAULT_VOLUME_HASH_MASK,
231 static void VInitVolumeHash(void);
235 /* This macro is used where an ffs() call does not exist. Was in util/ffs.c */
239 afs_int32 ffs_tmp = x;
243 for (ffs_i = 1;; ffs_i++) {
250 #endif /* !AFS_HAVE_FFS */
252 #ifdef AFS_PTHREAD_ENV
254 * disk partition queue element
256 typedef struct diskpartition_queue_t {
257 struct rx_queue queue; /**< queue header */
258 struct DiskPartition64 *diskP; /**< disk partition table entry */
259 } diskpartition_queue_t;
261 #ifndef AFS_DEMAND_ATTACH_FS
263 typedef struct vinitvolumepackage_thread_t {
264 struct rx_queue queue;
265 pthread_cond_t thread_done_cv;
266 int n_threads_complete;
267 } vinitvolumepackage_thread_t;
268 static void * VInitVolumePackageThread(void * args);
270 #else /* !AFS_DEMAND_ATTTACH_FS */
271 #define VINIT_BATCH_MAX_SIZE 512
274 * disk partition work queue
276 struct partition_queue {
277 struct rx_queue head; /**< diskpartition_queue_t queue */
278 pthread_mutex_t mutex;
283 * volumes parameters for preattach
285 struct volume_init_batch {
286 struct rx_queue queue; /**< queue header */
287 int thread; /**< posting worker thread */
288 int last; /**< indicates thread is done */
289 int size; /**< number of volume ids in batch */
290 Volume *batch[VINIT_BATCH_MAX_SIZE]; /**< volumes ids to preattach */
294 * volume parameters work queue
296 struct volume_init_queue {
297 struct rx_queue head; /**< volume_init_batch queue */
298 pthread_mutex_t mutex;
303 * volume init worker thread parameters
305 struct vinitvolumepackage_thread_param {
306 int nthreads; /**< total number of worker threads */
307 int thread; /**< thread number for this worker thread */
308 struct partition_queue *pq; /**< queue partitions to scan */
309 struct volume_init_queue *vq; /**< queue of volume to preattach */
312 static void *VInitVolumePackageThread(void *args);
313 static struct DiskPartition64 *VInitNextPartition(struct partition_queue *pq);
314 static VolumeId VInitNextVolumeId(DIR *dirp);
315 static int VInitPreAttachVolumes(int nthreads, struct volume_init_queue *vq);
317 #endif /* !AFS_DEMAND_ATTACH_FS */
318 #endif /* AFS_PTHREAD_ENV */
320 #ifndef AFS_DEMAND_ATTACH_FS
321 static int VAttachVolumesByPartition(struct DiskPartition64 *diskP,
322 int * nAttached, int * nUnattached);
323 #endif /* AFS_DEMAND_ATTACH_FS */
326 #ifdef AFS_DEMAND_ATTACH_FS
327 /* demand attach fileserver extensions */
330 * in the future we will support serialization of VLRU state into the fs_state
333 * these structures are the beginning of that effort
335 struct VLRU_DiskHeader {
336 struct versionStamp stamp; /* magic and structure version number */
337 afs_uint32 mtime; /* time of dump to disk */
338 afs_uint32 num_records; /* number of VLRU_DiskEntry records */
341 struct VLRU_DiskEntry {
342 VolumeId vid; /* volume ID */
343 afs_uint32 idx; /* generation */
344 afs_uint32 last_get; /* timestamp of last get */
347 struct VLRU_StartupQueue {
348 struct VLRU_DiskEntry * entry;
353 typedef struct vshutdown_thread_t {
355 pthread_mutex_t lock;
357 pthread_cond_t master_cv;
359 int n_threads_complete;
361 int schedule_version;
364 byte n_parts_done_pass;
365 byte part_thread_target[VOLMAXPARTS+1];
366 byte part_done_pass[VOLMAXPARTS+1];
367 struct rx_queue * part_pass_head[VOLMAXPARTS+1];
368 int stats[4][VOLMAXPARTS+1];
369 } vshutdown_thread_t;
370 static void * VShutdownThread(void * args);
373 static Volume * VAttachVolumeByVp_r(Error * ec, Volume * vp, int mode);
374 static int VCheckFree(Volume * vp);
377 static void AddVolumeToVByPList_r(Volume * vp);
378 static void DeleteVolumeFromVByPList_r(Volume * vp);
379 static void VVByPListBeginExclusive_r(struct DiskPartition64 * dp);
380 static void VVByPListEndExclusive_r(struct DiskPartition64 * dp);
381 static void VVByPListWait_r(struct DiskPartition64 * dp);
383 /* online salvager */
385 VCHECK_SALVAGE_OK = 0, /**< no pending salvage */
386 VCHECK_SALVAGE_SCHEDULED = 1, /**< salvage has been scheduled */
387 VCHECK_SALVAGE_ASYNC = 2, /**< salvage being scheduled */
388 VCHECK_SALVAGE_DENIED = 3, /**< salvage not scheduled; denied */
389 VCHECK_SALVAGE_FAIL = 4 /**< salvage not scheduled; failed */
391 static int VCheckSalvage(Volume * vp);
392 #if defined(SALVSYNC_BUILD_CLIENT) || defined(FSSYNC_BUILD_CLIENT)
393 static int VScheduleSalvage_r(Volume * vp);
396 /* Volume hash table */
397 static void VReorderHash_r(VolumeHashChainHead * head, Volume * pp, Volume * vp);
398 static void VHashBeginExclusive_r(VolumeHashChainHead * head);
399 static void VHashEndExclusive_r(VolumeHashChainHead * head);
400 static void VHashWait_r(VolumeHashChainHead * head);
403 static int ShutdownVByPForPass_r(struct DiskPartition64 * dp, int pass);
404 static int ShutdownVolumeWalk_r(struct DiskPartition64 * dp, int pass,
405 struct rx_queue ** idx);
406 static void ShutdownController(vshutdown_thread_t * params);
407 static void ShutdownCreateSchedule(vshutdown_thread_t * params);
410 static void VLRU_ComputeConstants(void);
411 static void VInitVLRU(void);
412 static void VLRU_Init_Node_r(Volume * vp);
413 static void VLRU_Add_r(Volume * vp);
414 static void VLRU_Delete_r(Volume * vp);
415 static void VLRU_UpdateAccess_r(Volume * vp);
416 static void * VLRU_ScannerThread(void * args);
417 static void VLRU_Scan_r(int idx);
418 static void VLRU_Promote_r(int idx);
419 static void VLRU_Demote_r(int idx);
420 static void VLRU_SwitchQueues(Volume * vp, int new_idx, int append);
423 static int VCheckSoftDetach(Volume * vp, afs_uint32 thresh);
424 static int VCheckSoftDetachCandidate(Volume * vp, afs_uint32 thresh);
425 static int VSoftDetachVolume_r(Volume * vp, afs_uint32 thresh);
428 pthread_key_t VThread_key;
429 VThreadOptions_t VThread_defaults = {
430 0 /**< allow salvsync */
432 #endif /* AFS_DEMAND_ATTACH_FS */
435 struct Lock vol_listLock; /* Lock obtained when listing volumes:
436 * prevents a volume from being missed
437 * if the volume is attached during a
441 /* Common message used when the volume goes off line */
442 char *VSalvageMessage =
443 "Files in this volume are currently unavailable; call operations";
445 int VInit; /* 0 - uninitialized,
446 * 1 - initialized but not all volumes have been attached,
447 * 2 - initialized and all volumes have been attached,
448 * 3 - initialized, all volumes have been attached, and
449 * VConnectFS() has completed. */
451 static int vinit_attach_abort = 0;
453 bit32 VolumeCacheCheck; /* Incremented everytime a volume goes on line--
454 * used to stamp volume headers and in-core
455 * vnodes. When the volume goes on-line the
456 * vnode will be invalidated
457 * access only with VOL_LOCK held */
462 /***************************************************/
463 /* Startup routines */
464 /***************************************************/
466 #if defined(FAST_RESTART) && defined(AFS_DEMAND_ATTACH_FS)
467 # error FAST_RESTART and DAFS are incompatible. For the DAFS equivalent \
468 of FAST_RESTART, use the -unsafe-nosalvage fileserver argument
472 * assign default values to a VolumePackageOptions struct.
474 * Always call this on a VolumePackageOptions struct first, then set any
475 * specific options you want, then call VInitVolumePackage2.
477 * @param[in] pt caller's program type
478 * @param[out] opts volume package options
481 VOptDefaults(ProgramType pt, VolumePackageOptions *opts)
483 opts->nLargeVnodes = opts->nSmallVnodes = 5;
486 opts->canScheduleSalvage = 0;
487 opts->canUseFSSYNC = 0;
488 opts->canUseSALVSYNC = 0;
490 opts->interrupt_rxcall = NULL;
491 opts->offline_timeout = -1;
492 opts->offline_shutdown_timeout = -1;
493 opts->usage_threshold = 128;
494 opts->usage_rate_limit = 5;
497 opts->unsafe_attach = 1;
498 #else /* !FAST_RESTART */
499 opts->unsafe_attach = 0;
500 #endif /* !FAST_RESTART */
504 opts->canScheduleSalvage = 1;
505 opts->canUseSALVSYNC = 1;
509 opts->canUseFSSYNC = 1;
513 opts->nLargeVnodes = 0;
514 opts->nSmallVnodes = 0;
516 opts->canScheduleSalvage = 1;
517 opts->canUseFSSYNC = 1;
527 * Set VInit to a certain value, and signal waiters.
529 * @param[in] value the value to set VInit to
534 VSetVInit_r(int value)
537 opr_cv_broadcast(&vol_vinit_cond);
541 VLogOfflineTimeout(const char *type, afs_int32 timeout)
547 Log("VInitVolumePackage: Interrupting clients accessing %s "
548 "immediately\n", type);
550 Log("VInitVolumePackage: Interrupting clients accessing %s "
551 "after %ld second%s\n", type, (long)timeout, timeout==1?"":"s");
556 VInitVolumePackage2(ProgramType pt, VolumePackageOptions * opts)
558 int errors = 0; /* Number of errors while finding vice partitions. */
563 #ifndef AFS_PTHREAD_ENV
564 if (opts->offline_timeout != -1 || opts->offline_shutdown_timeout != -1) {
565 Log("VInitVolumePackage: offline_timeout and/or "
566 "offline_shutdown_timeout was specified, but the volume package "
567 "does not support these for LWP builds\n");
571 VLogOfflineTimeout("volumes going offline", opts->offline_timeout);
572 VLogOfflineTimeout("volumes going offline during shutdown",
573 opts->offline_shutdown_timeout);
575 memset(&VStats, 0, sizeof(VStats));
576 VStats.hdr_cache_size = 200;
578 VInitPartitionPackage();
580 #ifdef AFS_DEMAND_ATTACH_FS
581 if (programType == fileServer) {
584 VLRU_SetOptions(VLRU_SET_ENABLED, 0);
586 opr_Verify(pthread_key_create(&VThread_key, NULL) == 0);
589 opr_mutex_init(&vol_glock_mutex);
590 opr_mutex_init(&vol_trans_mutex);
591 opr_cv_init(&vol_put_volume_cond);
592 opr_cv_init(&vol_sleep_cond);
593 opr_cv_init(&vol_init_attach_cond);
594 opr_cv_init(&vol_vinit_cond);
595 #ifndef AFS_PTHREAD_ENV
597 #endif /* AFS_PTHREAD_ENV */
598 Lock_Init(&vol_listLock);
600 srandom(time(0)); /* For VGetVolumeInfo */
602 #ifdef AFS_DEMAND_ATTACH_FS
603 opr_mutex_init(&vol_salvsync_mutex);
604 #endif /* AFS_DEMAND_ATTACH_FS */
606 /* Ok, we have done enough initialization that fileserver can
607 * start accepting calls, even though the volumes may not be
608 * available just yet.
612 #if defined(AFS_DEMAND_ATTACH_FS) && defined(SALVSYNC_BUILD_SERVER)
613 if (programType == salvageServer) {
616 #endif /* AFS_DEMAND_ATTACH_FS */
617 #ifdef FSSYNC_BUILD_SERVER
618 if (programType == fileServer) {
622 #if defined(AFS_DEMAND_ATTACH_FS) && defined(SALVSYNC_BUILD_CLIENT)
623 if (VCanUseSALVSYNC()) {
624 /* establish a connection to the salvager at this point */
625 opr_Verify(VConnectSALV() != 0);
627 #endif /* AFS_DEMAND_ATTACH_FS */
629 if (opts->volcache > VStats.hdr_cache_size)
630 VStats.hdr_cache_size = opts->volcache;
631 VInitVolumeHeaderCache(VStats.hdr_cache_size);
633 VInitVnodes(vLarge, opts->nLargeVnodes);
634 VInitVnodes(vSmall, opts->nSmallVnodes);
637 errors = VAttachPartitions();
641 if (programType != fileServer) {
642 errors = VInitAttachVolumes(programType);
648 #ifdef FSSYNC_BUILD_CLIENT
649 if (VCanUseFSSYNC()) {
651 #ifdef AFS_DEMAND_ATTACH_FS
652 if (programType == salvageServer) {
653 Log("Unable to connect to file server; aborted\n");
656 #endif /* AFS_DEMAND_ATTACH_FS */
657 Log("Unable to connect to file server; will retry at need\n");
660 #endif /* FSSYNC_BUILD_CLIENT */
665 #if !defined(AFS_PTHREAD_ENV)
667 * Attach volumes in vice partitions
669 * @param[in] pt calling program type
672 * @note This is the original, non-threaded version of attach parititions.
674 * @post VInit state is 2
677 VInitAttachVolumes(ProgramType pt)
679 opr_Assert(VInit==1);
680 if (pt == fileServer) {
681 struct DiskPartition64 *diskP;
682 /* Attach all the volumes in this partition */
683 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
684 int nAttached = 0, nUnattached = 0;
685 opr_Verify(VAttachVolumesByPartition(diskP,
686 &nAttached, &nUnattached)
691 VSetVInit_r(2); /* Initialized, and all volumes have been attached */
692 LWP_NoYieldSignal(VInitAttachVolumes);
696 #endif /* !AFS_PTHREAD_ENV */
698 #if defined(AFS_PTHREAD_ENV) && !defined(AFS_DEMAND_ATTACH_FS)
700 * Attach volumes in vice partitions
702 * @param[in] pt calling program type
705 * @note Threaded version of attach parititions.
707 * @post VInit state is 2
710 VInitAttachVolumes(ProgramType pt)
712 opr_Assert(VInit==1);
713 if (pt == fileServer) {
714 struct DiskPartition64 *diskP;
715 struct vinitvolumepackage_thread_t params;
716 struct diskpartition_queue_t * dpq;
717 int i, threads, parts;
719 pthread_attr_t attrs;
721 opr_cv_init(¶ms.thread_done_cv);
723 params.n_threads_complete = 0;
725 /* create partition work queue */
726 for (parts=0, diskP = DiskPartitionList; diskP; diskP = diskP->next, parts++) {
727 dpq = malloc(sizeof(struct diskpartition_queue_t));
728 opr_Assert(dpq != NULL);
730 queue_Append(¶ms,dpq);
733 threads = min(parts, vol_attach_threads);
736 /* spawn off a bunch of initialization threads */
737 opr_Verify(pthread_attr_init(&attrs) == 0);
738 opr_Verify(pthread_attr_setdetachstate(&attrs,
739 PTHREAD_CREATE_DETACHED)
742 Log("VInitVolumePackage: beginning parallel fileserver startup\n");
743 Log("VInitVolumePackage: using %d threads to attach volumes on %d partitions\n",
747 for (i=0; i < threads; i++) {
750 opr_Verify(pthread_create(&tid, &attrs,
751 &VInitVolumePackageThread,
753 AFS_SIGSET_RESTORE();
756 while(params.n_threads_complete < threads) {
757 VOL_CV_WAIT(¶ms.thread_done_cv);
761 opr_Verify(pthread_attr_destroy(&attrs) == 0);
763 /* if we're only going to run one init thread, don't bother creating
765 Log("VInitVolumePackage: beginning single-threaded fileserver startup\n");
766 Log("VInitVolumePackage: using 1 thread to attach volumes on %d partition(s)\n",
769 VInitVolumePackageThread(¶ms);
772 opr_cv_destroy(¶ms.thread_done_cv);
775 VSetVInit_r(2); /* Initialized, and all volumes have been attached */
776 opr_cv_broadcast(&vol_init_attach_cond);
782 VInitVolumePackageThread(void * args) {
784 struct DiskPartition64 *diskP;
785 struct vinitvolumepackage_thread_t * params;
786 struct diskpartition_queue_t * dpq;
788 params = (vinitvolumepackage_thread_t *) args;
792 /* Attach all the volumes in this partition */
793 while (queue_IsNotEmpty(params)) {
794 int nAttached = 0, nUnattached = 0;
796 if (vinit_attach_abort) {
797 Log("Aborting initialization\n");
801 dpq = queue_First(params,diskpartition_queue_t);
807 opr_Verify(VAttachVolumesByPartition(diskP, &nAttached,
814 params->n_threads_complete++;
815 opr_cv_signal(¶ms->thread_done_cv);
819 #endif /* AFS_PTHREAD_ENV && !AFS_DEMAND_ATTACH_FS */
821 #if defined(AFS_DEMAND_ATTACH_FS)
823 * Attach volumes in vice partitions
825 * @param[in] pt calling program type
828 * @note Threaded version of attach partitions.
830 * @post VInit state is 2
833 VInitAttachVolumes(ProgramType pt)
835 opr_Assert(VInit==1);
836 if (pt == fileServer) {
838 struct DiskPartition64 *diskP;
839 struct partition_queue pq;
840 struct volume_init_queue vq;
842 int i, threads, parts;
844 pthread_attr_t attrs;
846 /* create partition work queue */
849 opr_mutex_init(&pq.mutex);
850 for (parts = 0, diskP = DiskPartitionList; diskP; diskP = diskP->next, parts++) {
851 struct diskpartition_queue_t *dp;
852 dp = malloc(sizeof(struct diskpartition_queue_t));
853 opr_Assert(dp != NULL);
855 queue_Append(&pq, dp);
858 /* number of worker threads; at least one, not to exceed the number of partitions */
859 threads = min(parts, vol_attach_threads);
861 /* create volume work queue */
864 opr_mutex_init(&vq.mutex);
866 opr_Verify(pthread_attr_init(&attrs) == 0);
867 opr_Verify(pthread_attr_setdetachstate(&attrs,
868 PTHREAD_CREATE_DETACHED) == 0);
870 Log("VInitVolumePackage: beginning parallel fileserver startup\n");
871 Log("VInitVolumePackage: using %d threads to pre-attach volumes on %d partitions\n",
874 /* create threads to scan disk partitions. */
875 for (i=0; i < threads; i++) {
876 struct vinitvolumepackage_thread_param *params;
879 params = malloc(sizeof(struct vinitvolumepackage_thread_param));
883 params->nthreads = threads;
884 params->thread = i+1;
887 opr_Verify(pthread_create(&tid, &attrs,
888 &VInitVolumePackageThread,
889 (void*)params) == 0);
890 AFS_SIGSET_RESTORE();
893 VInitPreAttachVolumes(threads, &vq);
895 opr_Verify(pthread_attr_destroy(&attrs) == 0);
896 opr_cv_destroy(&pq.cv);
897 opr_mutex_destroy(&pq.mutex);
898 opr_cv_destroy(&vq.cv);
899 opr_mutex_destroy(&vq.mutex);
903 VSetVInit_r(2); /* Initialized, and all volumes have been attached */
904 opr_cv_broadcast(&vol_init_attach_cond);
911 * Volume package initialization worker thread. Scan partitions for volume
912 * header files. Gather batches of volume ids and dispatch them to
913 * the main thread to be preattached. The volume preattachement is done
914 * in the main thread to avoid global volume lock contention.
917 VInitVolumePackageThread(void *args)
919 struct vinitvolumepackage_thread_param *params;
920 struct DiskPartition64 *partition;
921 struct partition_queue *pq;
922 struct volume_init_queue *vq;
923 struct volume_init_batch *vb;
926 params = (struct vinitvolumepackage_thread_param *)args;
932 vb = malloc(sizeof(struct volume_init_batch));
934 vb->thread = params->thread;
938 Log("Scanning partitions on thread %d of %d\n", params->thread, params->nthreads);
939 while((partition = VInitNextPartition(pq))) {
943 Log("Partition %s: pre-attaching volumes\n", partition->name);
944 dirp = opendir(VPartitionPath(partition));
946 Log("opendir on Partition %s failed, errno=%d!\n", partition->name, errno);
949 while ((vid = VInitNextVolumeId(dirp))) {
950 Volume *vp = calloc(1, sizeof(Volume));
952 vp->device = partition->device;
953 vp->partition = partition;
955 queue_Init(&vp->vnode_list);
956 queue_Init(&vp->rx_call_list);
957 opr_cv_init(&V_attachCV(vp));
959 vb->batch[vb->size++] = vp;
960 if (vb->size == VINIT_BATCH_MAX_SIZE) {
961 opr_mutex_enter(&vq->mutex);
962 queue_Append(vq, vb);
963 opr_cv_broadcast(&vq->cv);
964 opr_mutex_exit(&vq->mutex);
966 vb = malloc(sizeof(struct volume_init_batch));
968 vb->thread = params->thread;
977 opr_mutex_enter(&vq->mutex);
978 queue_Append(vq, vb);
979 opr_cv_broadcast(&vq->cv);
980 opr_mutex_exit(&vq->mutex);
982 Log("Partition scan thread %d of %d ended\n", params->thread, params->nthreads);
988 * Read next element from the pre-populated partition list.
990 static struct DiskPartition64*
991 VInitNextPartition(struct partition_queue *pq)
993 struct DiskPartition64 *partition;
994 struct diskpartition_queue_t *dp; /* queue element */
996 if (vinit_attach_abort) {
997 Log("Aborting volume preattach thread.\n");
1001 /* get next partition to scan */
1002 opr_mutex_enter(&pq->mutex);
1003 if (queue_IsEmpty(pq)) {
1004 opr_mutex_exit(&pq->mutex);
1007 dp = queue_First(pq, diskpartition_queue_t);
1009 opr_mutex_exit(&pq->mutex);
1012 opr_Assert(dp->diskP);
1014 partition = dp->diskP;
1020 * Find next volume id on the partition.
1023 VInitNextVolumeId(DIR *dirp)
1029 while((d = readdir(dirp))) {
1030 if (vinit_attach_abort) {
1031 Log("Aborting volume preattach thread.\n");
1034 ext = strrchr(d->d_name, '.');
1035 if (d->d_name[0] == 'V' && ext && strcmp(ext, VHDREXT) == 0) {
1036 vid = VolumeNumber(d->d_name);
1040 Log("Warning: bogus volume header file: %s\n", d->d_name);
1047 * Preattach volumes in batches to avoid lock contention.
1050 VInitPreAttachVolumes(int nthreads, struct volume_init_queue *vq)
1052 struct volume_init_batch *vb;
1056 /* dequeue next volume */
1057 opr_mutex_enter(&vq->mutex);
1058 if (queue_IsEmpty(vq)) {
1059 opr_cv_wait(&vq->cv, &vq->mutex);
1061 vb = queue_First(vq, volume_init_batch);
1063 opr_mutex_exit(&vq->mutex);
1067 for (i = 0; i<vb->size; i++) {
1073 dup = VLookupVolume_r(&ec, vp->hashid, NULL);
1075 Log("Error looking up volume, code=%d\n", ec);
1078 Log("Warning: Duplicate volume id %" AFS_VOLID_FMT " detected.\n", afs_printable_VolumeId_lu(vp->hashid));
1081 /* put pre-attached volume onto the hash table
1082 * and bring it up to the pre-attached state */
1083 AddVolumeToHashTable(vp, vp->hashid);
1084 AddVolumeToVByPList_r(vp);
1085 VLRU_Init_Node_r(vp);
1086 VChangeState_r(vp, VOL_STATE_PREATTACHED);
1099 #endif /* AFS_DEMAND_ATTACH_FS */
1101 #if !defined(AFS_DEMAND_ATTACH_FS)
1103 * attach all volumes on a given disk partition
1106 VAttachVolumesByPartition(struct DiskPartition64 *diskP, int * nAttached, int * nUnattached)
1112 Log("Partition %s: attaching volumes\n", diskP->name);
1113 dirp = opendir(VPartitionPath(diskP));
1115 Log("opendir on Partition %s failed!\n", diskP->name);
1119 while ((dp = readdir(dirp))) {
1121 p = strrchr(dp->d_name, '.');
1123 if (vinit_attach_abort) {
1124 Log("Partition %s: abort attach volumes\n", diskP->name);
1128 if (p != NULL && strcmp(p, VHDREXT) == 0) {
1131 vp = VAttachVolumeByName(&error, diskP->name, dp->d_name,
1133 (*(vp ? nAttached : nUnattached))++;
1134 if (error == VOFFLINE)
1135 Log("Volume %d stays offline (/vice/offline/%s exists)\n", VolumeNumber(dp->d_name), dp->d_name);
1136 else if (LogLevel >= 5) {
1137 Log("Partition %s: attached volume %d (%s)\n",
1138 diskP->name, VolumeNumber(dp->d_name),
1147 Log("Partition %s: attached %d volumes; %d volumes not attached\n", diskP->name, *nAttached, *nUnattached);
1152 #endif /* !AFS_DEMAND_ATTACH_FS */
1154 /***************************************************/
1155 /* Shutdown routines */
1156 /***************************************************/
1160 * highly multithreaded volume package shutdown
1162 * with the demand attach fileserver extensions,
1163 * VShutdown has been modified to be multithreaded.
1164 * In order to achieve optimal use of many threads,
1165 * the shutdown code involves one control thread and
1166 * n shutdown worker threads. The control thread
1167 * periodically examines the number of volumes available
1168 * for shutdown on each partition, and produces a worker
1169 * thread allocation schedule. The idea is to eliminate
1170 * redundant scheduling computation on the workers by
1171 * having a single master scheduler.
1173 * The scheduler's objectives are:
1175 * each partition with volumes remaining gets allocated
1176 * at least 1 thread (assuming sufficient threads)
1178 * threads are allocated proportional to the number of
1179 * volumes remaining to be offlined. This ensures that
1180 * the OS I/O scheduler has many requests to elevator
1181 * seek on partitions that will (presumably) take the
1182 * longest amount of time (from now) to finish shutdown
1183 * (3) keep threads busy
1184 * when there are extra threads, they are assigned to
1185 * partitions using a simple round-robin algorithm
1187 * In the future, we may wish to add the ability to adapt
1188 * to the relative performance patterns of each disk
1193 * multi-step shutdown process
1195 * demand attach shutdown is a four-step process. Each
1196 * shutdown "pass" shuts down increasingly more difficult
1197 * volumes. The main purpose is to achieve better cache
1198 * utilization during shutdown.
1201 * shutdown volumes in the unattached, pre-attached
1204 * shutdown attached volumes with cached volume headers
1206 * shutdown all volumes in non-exclusive states
1208 * shutdown all remaining volumes
1211 #ifdef AFS_DEMAND_ATTACH_FS
1217 struct DiskPartition64 * diskP;
1218 struct diskpartition_queue_t * dpq;
1219 vshutdown_thread_t params;
1221 pthread_attr_t attrs;
1223 memset(¶ms, 0, sizeof(vshutdown_thread_t));
1226 Log("VShutdown: aborting attach volumes\n");
1227 vinit_attach_abort = 1;
1228 VOL_CV_WAIT(&vol_init_attach_cond);
1231 for (params.n_parts=0, diskP = DiskPartitionList;
1232 diskP; diskP = diskP->next, params.n_parts++);
1234 Log("VShutdown: shutting down on-line volumes on %d partition%s...\n",
1235 params.n_parts, params.n_parts > 1 ? "s" : "");
1237 vol_shutting_down = 1;
1239 if (vol_attach_threads > 1) {
1240 /* prepare for parallel shutdown */
1241 params.n_threads = vol_attach_threads;
1242 opr_mutex_init(¶ms.lock);
1243 opr_cv_init(¶ms.cv);
1244 opr_cv_init(¶ms.master_cv);
1245 opr_Verify(pthread_attr_init(&attrs) == 0);
1246 opr_Verify(pthread_attr_setdetachstate(&attrs,
1247 PTHREAD_CREATE_DETACHED) == 0);
1248 queue_Init(¶ms);
1250 /* setup the basic partition information structures for
1251 * parallel shutdown */
1252 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1254 struct rx_queue * qp, * nqp;
1258 VVByPListWait_r(diskP);
1259 VVByPListBeginExclusive_r(diskP);
1262 for (queue_Scan(&diskP->vol_list, qp, nqp, rx_queue)) {
1263 vp = (Volume *)((char *)qp - offsetof(Volume, vol_list));
1267 Log("VShutdown: partition %s has %d volumes with attached headers\n",
1268 VPartitionPath(diskP), count);
1271 /* build up the pass 0 shutdown work queue */
1272 dpq = malloc(sizeof(struct diskpartition_queue_t));
1273 opr_Assert(dpq != NULL);
1275 queue_Prepend(¶ms, dpq);
1277 params.part_pass_head[diskP->index] = queue_First(&diskP->vol_list, rx_queue);
1280 Log("VShutdown: beginning parallel fileserver shutdown\n");
1281 Log("VShutdown: using %d threads to offline volumes on %d partition%s\n",
1282 vol_attach_threads, params.n_parts, params.n_parts > 1 ? "s" : "" );
1284 /* do pass 0 shutdown */
1285 opr_mutex_enter(¶ms.lock);
1286 for (i=0; i < params.n_threads; i++) {
1287 opr_Verify(pthread_create(&tid, &attrs, &VShutdownThread,
1291 /* wait for all the pass 0 shutdowns to complete */
1292 while (params.n_threads_complete < params.n_threads) {
1293 CV_WAIT(¶ms.master_cv, ¶ms.lock);
1295 params.n_threads_complete = 0;
1297 opr_cv_broadcast(¶ms.cv);
1298 opr_mutex_exit(¶ms.lock);
1300 Log("VShutdown: pass 0 completed using the 1 thread per partition algorithm\n");
1301 Log("VShutdown: starting passes 1 through 3 using finely-granular mp-fast algorithm\n");
1303 /* run the parallel shutdown scheduler. it will drop the glock internally */
1304 ShutdownController(¶ms);
1306 /* wait for all the workers to finish pass 3 and terminate */
1307 while (params.pass < 4) {
1308 VOL_CV_WAIT(¶ms.cv);
1311 opr_Verify(pthread_attr_destroy(&attrs) == 0);
1312 opr_cv_destroy(¶ms.cv);
1313 opr_cv_destroy(¶ms.master_cv);
1314 opr_mutex_destroy(¶ms.lock);
1316 /* drop the VByPList exclusive reservations */
1317 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1318 VVByPListEndExclusive_r(diskP);
1319 Log("VShutdown: %s stats : (pass[0]=%d, pass[1]=%d, pass[2]=%d, pass[3]=%d)\n",
1320 VPartitionPath(diskP),
1321 params.stats[0][diskP->index],
1322 params.stats[1][diskP->index],
1323 params.stats[2][diskP->index],
1324 params.stats[3][diskP->index]);
1327 Log("VShutdown: shutdown finished using %d threads\n", params.n_threads);
1329 /* if we're only going to run one shutdown thread, don't bother creating
1331 Log("VShutdown: beginning single-threaded fileserver shutdown\n");
1333 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1334 VShutdownByPartition_r(diskP);
1338 Log("VShutdown: complete.\n");
1341 #else /* AFS_DEMAND_ATTACH_FS */
1351 Log("VShutdown: aborting attach volumes\n");
1352 vinit_attach_abort = 1;
1353 #ifdef AFS_PTHREAD_ENV
1354 VOL_CV_WAIT(&vol_init_attach_cond);
1356 LWP_WaitProcess(VInitAttachVolumes);
1357 #endif /* AFS_PTHREAD_ENV */
1360 Log("VShutdown: shutting down on-line volumes...\n");
1361 vol_shutting_down = 1;
1362 for (i = 0; i < VolumeHashTable.Size; i++) {
1363 /* try to hold first volume in the hash table */
1364 for (queue_Scan(&VolumeHashTable.Table[i],vp,np,Volume)) {
1368 Log("VShutdown: Attempting to take volume %" AFS_VOLID_FMT " offline.\n",
1369 afs_printable_VolumeId_lu(vp->hashid));
1371 /* next, take the volume offline (drops reference count) */
1372 VOffline_r(vp, "File server was shut down");
1376 Log("VShutdown: complete.\n");
1378 #endif /* AFS_DEMAND_ATTACH_FS */
1384 opr_Assert(VInit>0);
1391 * stop new activity (e.g. SALVSYNC) from occurring
1393 * Use this to make the volume package less busy; for example, during
1394 * shutdown. This doesn't actually shutdown/detach anything in the
1395 * volume package, but prevents certain processes from ocurring. For
1396 * example, preventing new SALVSYNC communication in DAFS. In theory, we
1397 * could also use this to prevent new volume attachment, or prevent
1398 * other programs from checking out volumes, etc.
1403 #ifdef AFS_DEMAND_ATTACH_FS
1404 /* make sure we don't try to contact the salvageserver, since it may
1405 * not be around anymore */
1406 vol_disallow_salvsync = 1;
1410 #ifdef AFS_DEMAND_ATTACH_FS
1413 * shutdown control thread
1416 ShutdownController(vshutdown_thread_t * params)
1419 struct DiskPartition64 * diskP;
1421 vshutdown_thread_t shadow;
1423 ShutdownCreateSchedule(params);
1425 while ((params->pass < 4) &&
1426 (params->n_threads_complete < params->n_threads)) {
1427 /* recompute schedule once per second */
1429 memcpy(&shadow, params, sizeof(vshutdown_thread_t));
1433 Log("ShutdownController: schedule version=%d, vol_remaining=%d, pass=%d\n",
1434 shadow.schedule_version, shadow.vol_remaining, shadow.pass);
1435 Log("ShutdownController: n_threads_complete=%d, n_parts_done_pass=%d\n",
1436 shadow.n_threads_complete, shadow.n_parts_done_pass);
1437 for (diskP = DiskPartitionList; diskP; diskP=diskP->next) {
1439 Log("ShutdownController: part[%d] : (len=%d, thread_target=%d, done_pass=%d, pass_head=%p)\n",
1441 diskP->vol_list.len,
1442 shadow.part_thread_target[id],
1443 shadow.part_done_pass[id],
1444 shadow.part_pass_head[id]);
1450 ShutdownCreateSchedule(params);
1454 /* create the shutdown thread work schedule.
1455 * this scheduler tries to implement fairness
1456 * by allocating at least 1 thread to each
1457 * partition with volumes to be shutdown,
1458 * and then it attempts to allocate remaining
1459 * threads based upon the amount of work left
1462 ShutdownCreateSchedule(vshutdown_thread_t * params)
1464 struct DiskPartition64 * diskP;
1465 int sum, thr_workload, thr_left;
1466 int part_residue[VOLMAXPARTS+1];
1469 /* compute the total number of outstanding volumes */
1471 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1472 sum += diskP->vol_list.len;
1475 params->schedule_version++;
1476 params->vol_remaining = sum;
1481 /* compute average per-thread workload */
1482 thr_workload = sum / params->n_threads;
1483 if (sum % params->n_threads)
1486 thr_left = params->n_threads;
1487 memset(&part_residue, 0, sizeof(part_residue));
1489 /* for fairness, give every partition with volumes remaining
1490 * at least one thread */
1491 for (diskP = DiskPartitionList; diskP && thr_left; diskP = diskP->next) {
1493 if (diskP->vol_list.len) {
1494 params->part_thread_target[id] = 1;
1497 params->part_thread_target[id] = 0;
1501 if (thr_left && thr_workload) {
1502 /* compute length-weighted workloads */
1505 for (diskP = DiskPartitionList; diskP && thr_left; diskP = diskP->next) {
1507 delta = (diskP->vol_list.len / thr_workload) -
1508 params->part_thread_target[id];
1512 if (delta < thr_left) {
1513 params->part_thread_target[id] += delta;
1516 params->part_thread_target[id] += thr_left;
1524 /* try to assign any leftover threads to partitions that
1525 * had volume lengths closer to needing thread_target+1 */
1526 int max_residue, max_id = 0;
1528 /* compute the residues */
1529 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1531 part_residue[id] = diskP->vol_list.len -
1532 (params->part_thread_target[id] * thr_workload);
1535 /* now try to allocate remaining threads to partitions with the
1536 * highest residues */
1539 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1541 if (part_residue[id] > max_residue) {
1542 max_residue = part_residue[id];
1551 params->part_thread_target[max_id]++;
1553 part_residue[max_id] = 0;
1558 /* punt and give any remaining threads equally to each partition */
1560 if (thr_left >= params->n_parts) {
1561 alloc = thr_left / params->n_parts;
1562 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1564 params->part_thread_target[id] += alloc;
1569 /* finish off the last of the threads */
1570 for (diskP = DiskPartitionList; thr_left && diskP; diskP = diskP->next) {
1572 params->part_thread_target[id]++;
1578 /* worker thread for parallel shutdown */
1580 VShutdownThread(void * args)
1582 vshutdown_thread_t * params;
1583 int found, pass, schedule_version_save, count;
1584 struct DiskPartition64 *diskP;
1585 struct diskpartition_queue_t * dpq;
1588 params = (vshutdown_thread_t *) args;
1590 /* acquire the shutdown pass 0 lock */
1591 opr_mutex_enter(¶ms->lock);
1593 /* if there's still pass 0 work to be done,
1594 * get a work entry, and do a pass 0 shutdown */
1595 if (queue_IsNotEmpty(params)) {
1596 dpq = queue_First(params, diskpartition_queue_t);
1598 opr_mutex_exit(¶ms->lock);
1604 while (ShutdownVolumeWalk_r(diskP, 0, ¶ms->part_pass_head[id]))
1606 params->stats[0][diskP->index] = count;
1607 opr_mutex_enter(¶ms->lock);
1610 params->n_threads_complete++;
1611 if (params->n_threads_complete == params->n_threads) {
1612 /* notify control thread that all workers have completed pass 0 */
1613 opr_cv_signal(¶ms->master_cv);
1615 while (params->pass == 0) {
1616 opr_cv_wait(¶ms->cv, ¶ms->lock);
1620 opr_mutex_exit(¶ms->lock);
1623 pass = params->pass;
1624 opr_Assert(pass > 0);
1626 /* now escalate through the more complicated shutdowns */
1628 schedule_version_save = params->schedule_version;
1630 /* find a disk partition to work on */
1631 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1633 if (params->part_thread_target[id] && !params->part_done_pass[id]) {
1634 params->part_thread_target[id]--;
1641 /* hmm. for some reason the controller thread couldn't find anything for
1642 * us to do. let's see if there's anything we can do */
1643 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1645 if (diskP->vol_list.len && !params->part_done_pass[id]) {
1648 } else if (!params->part_done_pass[id]) {
1649 params->part_done_pass[id] = 1;
1650 params->n_parts_done_pass++;
1652 Log("VShutdown: done shutting down volumes on partition %s.\n",
1653 VPartitionPath(diskP));
1659 /* do work on this partition until either the controller
1660 * creates a new schedule, or we run out of things to do
1661 * on this partition */
1664 while (!params->part_done_pass[id] &&
1665 (schedule_version_save == params->schedule_version)) {
1666 /* ShutdownVolumeWalk_r will drop the glock internally */
1667 if (!ShutdownVolumeWalk_r(diskP, pass, ¶ms->part_pass_head[id])) {
1668 if (!params->part_done_pass[id]) {
1669 params->part_done_pass[id] = 1;
1670 params->n_parts_done_pass++;
1672 Log("VShutdown: done shutting down volumes on partition %s.\n",
1673 VPartitionPath(diskP));
1681 params->stats[pass][id] += count;
1683 /* ok, everyone is done this pass, proceed */
1686 params->n_threads_complete++;
1687 while (params->pass == pass) {
1688 if (params->n_threads_complete == params->n_threads) {
1689 /* we are the last thread to complete, so we will
1690 * reinitialize worker pool state for the next pass */
1691 params->n_threads_complete = 0;
1692 params->n_parts_done_pass = 0;
1694 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
1696 params->part_done_pass[id] = 0;
1697 params->part_pass_head[id] = queue_First(&diskP->vol_list, rx_queue);
1700 /* compute a new thread schedule before releasing all the workers */
1701 ShutdownCreateSchedule(params);
1703 /* wake up all the workers */
1704 opr_cv_broadcast(¶ms->cv);
1707 Log("VShutdown: pass %d completed using %d threads on %d partitions\n",
1708 pass, params->n_threads, params->n_parts);
1711 VOL_CV_WAIT(¶ms->cv);
1714 pass = params->pass;
1728 /* shut down all volumes on a given disk partition
1730 * note that this function will not allow mp-fast
1731 * shutdown of a partition */
1733 VShutdownByPartition_r(struct DiskPartition64 * dp)
1739 /* wait for other exclusive ops to finish */
1740 VVByPListWait_r(dp);
1742 /* begin exclusive access */
1743 VVByPListBeginExclusive_r(dp);
1745 /* pick the low-hanging fruit first,
1746 * then do the complicated ones last
1747 * (has the advantage of keeping
1748 * in-use volumes up until the bitter end) */
1749 for (pass = 0, total=0; pass < 4; pass++) {
1750 pass_stats[pass] = ShutdownVByPForPass_r(dp, pass);
1751 total += pass_stats[pass];
1754 /* end exclusive access */
1755 VVByPListEndExclusive_r(dp);
1757 Log("VShutdownByPartition: shut down %d volumes on %s (pass[0]=%d, pass[1]=%d, pass[2]=%d, pass[3]=%d)\n",
1758 total, VPartitionPath(dp), pass_stats[0], pass_stats[1], pass_stats[2], pass_stats[3]);
1763 /* internal shutdown functionality
1765 * for multi-pass shutdown:
1766 * 0 to only "shutdown" {pre,un}attached and error state volumes
1767 * 1 to also shutdown attached volumes w/ volume header loaded
1768 * 2 to also shutdown attached volumes w/o volume header loaded
1769 * 3 to also shutdown exclusive state volumes
1771 * caller MUST hold exclusive access on the hash chain
1772 * because we drop vol_glock_mutex internally
1774 * this function is reentrant for passes 1--3
1775 * (e.g. multiple threads can cooperate to
1776 * shutdown a partition mp-fast)
1778 * pass 0 is not scaleable because the volume state data is
1779 * synchronized by vol_glock mutex, and the locking overhead
1780 * is too high to drop the lock long enough to do linked list
1784 ShutdownVByPForPass_r(struct DiskPartition64 * dp, int pass)
1786 struct rx_queue * q = queue_First(&dp->vol_list, rx_queue);
1789 while (ShutdownVolumeWalk_r(dp, pass, &q))
1795 /* conditionally shutdown one volume on partition dp
1796 * returns 1 if a volume was shutdown in this pass,
1799 ShutdownVolumeWalk_r(struct DiskPartition64 * dp, int pass,
1800 struct rx_queue ** idx)
1802 struct rx_queue *qp, *nqp;
1807 for (queue_ScanFrom(&dp->vol_list, qp, qp, nqp, rx_queue)) {
1808 vp = (Volume *) (((char *)qp) - offsetof(Volume, vol_list));
1812 if ((V_attachState(vp) != VOL_STATE_UNATTACHED) &&
1813 (V_attachState(vp) != VOL_STATE_ERROR) &&
1814 (V_attachState(vp) != VOL_STATE_DELETED) &&
1815 (V_attachState(vp) != VOL_STATE_PREATTACHED)) {
1819 if ((V_attachState(vp) == VOL_STATE_ATTACHED) &&
1820 (vp->header == NULL)) {
1824 if (VIsExclusiveState(V_attachState(vp))) {
1829 DeleteVolumeFromVByPList_r(vp);
1830 VShutdownVolume_r(vp);
1840 * shutdown a specific volume
1842 /* caller MUST NOT hold a heavyweight ref on vp */
1844 VShutdownVolume_r(Volume * vp)
1848 VCreateReservation_r(vp);
1850 if (LogLevel >= 5) {
1851 Log("VShutdownVolume_r: vid=%" AFS_VOLID_FMT ", device=%d, state=%hu\n",
1852 afs_printable_VolumeId_lu(vp->hashid), vp->partition->device,
1856 /* wait for other blocking ops to finish */
1857 VWaitExclusiveState_r(vp);
1859 opr_Assert(VIsValidState(V_attachState(vp)));
1861 switch(V_attachState(vp)) {
1862 case VOL_STATE_SALVAGING:
1863 /* Leave salvaging volumes alone. Any in-progress salvages will
1864 * continue working after viced shuts down. This is intentional.
1867 case VOL_STATE_PREATTACHED:
1868 case VOL_STATE_ERROR:
1869 VChangeState_r(vp, VOL_STATE_UNATTACHED);
1870 case VOL_STATE_UNATTACHED:
1871 case VOL_STATE_DELETED:
1873 case VOL_STATE_GOING_OFFLINE:
1874 case VOL_STATE_SHUTTING_DOWN:
1875 case VOL_STATE_ATTACHED:
1879 Log("VShutdown: Attempting to take volume %" AFS_VOLID_FMT " offline.\n",
1880 afs_printable_VolumeId_lu(vp->hashid));
1882 /* take the volume offline (drops reference count) */
1883 VOffline_r(vp, "File server was shut down");
1890 VCancelReservation_r(vp);
1894 #endif /* AFS_DEMAND_ATTACH_FS */
1897 /***************************************************/
1898 /* Header I/O routines */
1899 /***************************************************/
1902 HeaderName(bit32 magic)
1905 case VOLUMEINFOMAGIC:
1906 return "volume info";
1907 case SMALLINDEXMAGIC:
1908 return "small index";
1909 case LARGEINDEXMAGIC:
1910 return "large index";
1911 case LINKTABLEMAGIC:
1912 return "link table";
1917 /* open a descriptor for the inode (h),
1918 * read in an on-disk structure into buffer (to) of size (size),
1919 * verify versionstamp in structure has magic (magic) and
1920 * optionally verify version (version) if (version) is nonzero
1923 ReadHeader(Error * ec, IHandle_t * h, char *to, int size, bit32 magic,
1926 struct versionStamp *vsn;
1928 afs_sfsize_t nbytes;
1933 Log("ReadHeader: Null inode handle argument for %s header file.\n",
1941 Log("ReadHeader: Failed to open %s header file "
1942 "(volume=%" AFS_VOLID_FMT ", inode=%s); errno=%d\n", HeaderName(magic), afs_printable_VolumeId_lu(h->ih_vid),
1943 PrintInode(stmp, h->ih_ino), errno);
1948 vsn = (struct versionStamp *)to;
1949 nbytes = FDH_PREAD(fdP, to, size, 0);
1951 Log("ReadHeader: Failed to read %s header file "
1952 "(volume=%" AFS_VOLID_FMT ", inode=%s); errno=%d\n", HeaderName(magic), afs_printable_VolumeId_lu(h->ih_vid),
1953 PrintInode(stmp, h->ih_ino), errno);
1955 FDH_REALLYCLOSE(fdP);
1958 if (nbytes != size) {
1959 Log("ReadHeader: Incorrect number of bytes read from %s header file "
1960 "(volume=%" AFS_VOLID_FMT ", inode=%s); expected=%d, read=%d\n",
1961 HeaderName(magic), afs_printable_VolumeId_lu(h->ih_vid),
1962 PrintInode(stmp, h->ih_ino), size, (int)nbytes);
1964 FDH_REALLYCLOSE(fdP);
1967 if (vsn->magic != magic) {
1968 Log("ReadHeader: Incorrect magic for %s header file "
1969 "(volume=%" AFS_VOLID_FMT ", inode=%s); expected=0x%x, read=0x%x\n",
1970 HeaderName(magic), afs_printable_VolumeId_lu(h->ih_vid),
1971 PrintInode(stmp, h->ih_ino), magic, vsn->magic);
1973 FDH_REALLYCLOSE(fdP);
1979 /* Check is conditional, in case caller wants to inspect version himself */
1980 if (version && vsn->version != version) {
1981 Log("ReadHeader: Incorrect version for %s header file "
1982 "(volume=%" AFS_VOLID_FMT ", inode=%s); expected=%x, read=%x\n",
1983 HeaderName(magic), afs_printable_VolumeId_lu(h->ih_vid), PrintInode(stmp, h->ih_ino),
1984 version, vsn->version);
1990 WriteVolumeHeader_r(Error * ec, Volume * vp)
1992 IHandle_t *h = V_diskDataHandle(vp);
2002 if (FDH_PWRITE(fdP, (char *)&V_disk(vp), sizeof(V_disk(vp)), 0)
2003 != sizeof(V_disk(vp))) {
2005 FDH_REALLYCLOSE(fdP);
2011 /* VolumeHeaderToDisk
2012 * Allows for storing 64 bit inode numbers in on-disk volume header
2015 /* convert in-memory representation of a volume header to the
2016 * on-disk representation of a volume header */
2018 VolumeHeaderToDisk(VolumeDiskHeader_t * dh, VolumeHeader_t * h)
2021 memset(dh, 0, sizeof(VolumeDiskHeader_t));
2022 dh->stamp = h->stamp;
2024 dh->parent = h->parent;
2026 #ifdef AFS_64BIT_IOPS_ENV
2027 dh->volumeInfo_lo = (afs_int32) h->volumeInfo & 0xffffffff;
2028 dh->volumeInfo_hi = (afs_int32) (h->volumeInfo >> 32) & 0xffffffff;
2029 dh->smallVnodeIndex_lo = (afs_int32) h->smallVnodeIndex & 0xffffffff;
2030 dh->smallVnodeIndex_hi =
2031 (afs_int32) (h->smallVnodeIndex >> 32) & 0xffffffff;
2032 dh->largeVnodeIndex_lo = (afs_int32) h->largeVnodeIndex & 0xffffffff;
2033 dh->largeVnodeIndex_hi =
2034 (afs_int32) (h->largeVnodeIndex >> 32) & 0xffffffff;
2035 dh->linkTable_lo = (afs_int32) h->linkTable & 0xffffffff;
2036 dh->linkTable_hi = (afs_int32) (h->linkTable >> 32) & 0xffffffff;
2038 dh->volumeInfo_lo = h->volumeInfo;
2039 dh->smallVnodeIndex_lo = h->smallVnodeIndex;
2040 dh->largeVnodeIndex_lo = h->largeVnodeIndex;
2041 dh->linkTable_lo = h->linkTable;
2045 /* DiskToVolumeHeader
2046 * Converts an on-disk representation of a volume header to
2047 * the in-memory representation of a volume header.
2049 * Makes the assumption that AFS has *always*
2050 * zero'd the volume header file so that high parts of inode
2051 * numbers are 0 in older (SGI EFS) volume header files.
2054 DiskToVolumeHeader(VolumeHeader_t * h, VolumeDiskHeader_t * dh)
2056 memset(h, 0, sizeof(VolumeHeader_t));
2057 h->stamp = dh->stamp;
2059 h->parent = dh->parent;
2061 #ifdef AFS_64BIT_IOPS_ENV
2063 (Inode) dh->volumeInfo_lo | ((Inode) dh->volumeInfo_hi << 32);
2065 h->smallVnodeIndex =
2066 (Inode) dh->smallVnodeIndex_lo | ((Inode) dh->
2067 smallVnodeIndex_hi << 32);
2069 h->largeVnodeIndex =
2070 (Inode) dh->largeVnodeIndex_lo | ((Inode) dh->
2071 largeVnodeIndex_hi << 32);
2073 (Inode) dh->linkTable_lo | ((Inode) dh->linkTable_hi << 32);
2075 h->volumeInfo = dh->volumeInfo_lo;
2076 h->smallVnodeIndex = dh->smallVnodeIndex_lo;
2077 h->largeVnodeIndex = dh->largeVnodeIndex_lo;
2078 h->linkTable = dh->linkTable_lo;
2083 /***************************************************/
2084 /* Volume Attachment routines */
2085 /***************************************************/
2087 #ifdef AFS_DEMAND_ATTACH_FS
2089 * pre-attach a volume given its path.
2091 * @param[out] ec outbound error code
2092 * @param[in] partition partition path string
2093 * @param[in] name volume id string
2095 * @return volume object pointer
2097 * @note A pre-attached volume will only have its partition
2098 * and hashid fields initialized. At first call to
2099 * VGetVolume, the volume will be fully attached.
2103 VPreAttachVolumeByName(Error * ec, char *partition, char *name)
2107 vp = VPreAttachVolumeByName_r(ec, partition, name);
2113 * pre-attach a volume given its path.
2115 * @param[out] ec outbound error code
2116 * @param[in] partition path to vice partition
2117 * @param[in] name volume id string
2119 * @return volume object pointer
2121 * @pre VOL_LOCK held
2123 * @internal volume package internal use only.
2126 VPreAttachVolumeByName_r(Error * ec, char *partition, char *name)
2128 return VPreAttachVolumeById_r(ec,
2130 VolumeNumber(name));
2134 * pre-attach a volume given its path and numeric volume id.
2136 * @param[out] ec error code return
2137 * @param[in] partition path to vice partition
2138 * @param[in] volumeId numeric volume id
2140 * @return volume object pointer
2142 * @pre VOL_LOCK held
2144 * @internal volume package internal use only.
2147 VPreAttachVolumeById_r(Error * ec,
2152 struct DiskPartition64 *partp;
2156 opr_Assert(programType == fileServer);
2158 if (!(partp = VGetPartition_r(partition, 0))) {
2160 Log("VPreAttachVolumeById_r: Error getting partition (%s)\n", partition);
2164 vp = VLookupVolume_r(ec, volumeId, NULL);
2169 return VPreAttachVolumeByVp_r(ec, partp, vp, volumeId);
2173 * preattach a volume.
2175 * @param[out] ec outbound error code
2176 * @param[in] partp pointer to partition object
2177 * @param[in] vp pointer to volume object
2178 * @param[in] vid volume id
2180 * @return volume object pointer
2182 * @pre VOL_LOCK is held.
2184 * @warning Returned volume object pointer does not have to
2185 * equal the pointer passed in as argument vp. There
2186 * are potential race conditions which can result in
2187 * the pointers having different values. It is up to
2188 * the caller to make sure that references are handled
2189 * properly in this case.
2191 * @note If there is already a volume object registered with
2192 * the same volume id, its pointer MUST be passed as
2193 * argument vp. Failure to do so will result in a silent
2194 * failure to preattach.
2196 * @internal volume package internal use only.
2199 VPreAttachVolumeByVp_r(Error * ec,
2200 struct DiskPartition64 * partp,
2208 /* check to see if pre-attach already happened */
2210 (V_attachState(vp) != VOL_STATE_UNATTACHED) &&
2211 (V_attachState(vp) != VOL_STATE_DELETED) &&
2212 (V_attachState(vp) != VOL_STATE_PREATTACHED) &&
2213 !VIsErrorState(V_attachState(vp))) {
2215 * pre-attach is a no-op in all but the following cases:
2217 * - volume is unattached
2218 * - volume is in an error state
2219 * - volume is pre-attached
2221 Log("VPreattachVolumeByVp_r: volume %" AFS_VOLID_FMT " not in quiescent state (state %u flags 0x%x)\n",
2222 afs_printable_VolumeId_lu(vid), V_attachState(vp),
2226 /* we're re-attaching a volume; clear out some old state */
2227 memset(&vp->salvage, 0, sizeof(struct VolumeOnlineSalvage));
2229 if (V_partition(vp) != partp) {
2230 /* XXX potential race */
2231 DeleteVolumeFromVByPList_r(vp);
2234 /* if we need to allocate a new Volume struct,
2235 * go ahead and drop the vol glock, otherwise
2236 * do the basic setup synchronised, as it's
2237 * probably not worth dropping the lock */
2240 /* allocate the volume structure */
2241 vp = nvp = calloc(1, sizeof(Volume));
2242 opr_Assert(vp != NULL);
2243 queue_Init(&vp->vnode_list);
2244 queue_Init(&vp->rx_call_list);
2245 opr_cv_init(&V_attachCV(vp));
2248 /* link the volume with its associated vice partition */
2249 vp->device = partp->device;
2250 vp->partition = partp;
2253 vp->specialStatus = 0;
2255 /* if we dropped the lock, reacquire the lock,
2256 * check for pre-attach races, and then add
2257 * the volume to the hash table */
2260 nvp = VLookupVolume_r(ec, vid, NULL);
2265 } else if (nvp) { /* race detected */
2270 /* hack to make up for VChangeState_r() decrementing
2271 * the old state counter */
2272 VStats.state_levels[0]++;
2276 /* put pre-attached volume onto the hash table
2277 * and bring it up to the pre-attached state */
2278 AddVolumeToHashTable(vp, vp->hashid);
2279 AddVolumeToVByPList_r(vp);
2280 VLRU_Init_Node_r(vp);
2281 VChangeState_r(vp, VOL_STATE_PREATTACHED);
2284 Log("VPreAttachVolumeByVp_r: volume %" AFS_VOLID_FMT " pre-attached\n", afs_printable_VolumeId_lu(vp->hashid));
2292 #endif /* AFS_DEMAND_ATTACH_FS */
2294 /* Attach an existing volume, given its pathname, and return a
2295 pointer to the volume header information. The volume also
2296 normally goes online at this time. An offline volume
2297 must be reattached to make it go online */
2299 VAttachVolumeByName(Error * ec, char *partition, char *name, int mode)
2303 retVal = VAttachVolumeByName_r(ec, partition, name, mode);
2309 VAttachVolumeByName_r(Error * ec, char *partition, char *name, int mode)
2312 struct DiskPartition64 *partp;
2317 #ifdef AFS_DEMAND_ATTACH_FS
2318 VolumeStats stats_save;
2320 #endif /* AFS_DEMAND_ATTACH_FS */
2324 volumeId = VolumeNumber(name);
2326 if (!(partp = VGetPartition_r(partition, 0))) {
2328 Log("VAttachVolume: Error getting partition (%s)\n", partition);
2332 if (VRequiresPartLock()) {
2333 opr_Assert(VInit == 3);
2334 VLockPartition_r(partition);
2335 } else if (programType == fileServer) {
2336 #ifdef AFS_DEMAND_ATTACH_FS
2337 /* lookup the volume in the hash table */
2338 vp = VLookupVolume_r(ec, volumeId, NULL);
2344 /* save any counters that are supposed to
2345 * be monotonically increasing over the
2346 * lifetime of the fileserver */
2347 memcpy(&stats_save, &vp->stats, sizeof(VolumeStats));
2349 memset(&stats_save, 0, sizeof(VolumeStats));
2352 /* if there's something in the hash table, and it's not
2353 * in the pre-attach state, then we may need to detach
2354 * it before proceeding */
2355 if (vp && (V_attachState(vp) != VOL_STATE_PREATTACHED)) {
2356 VCreateReservation_r(vp);
2357 VWaitExclusiveState_r(vp);
2359 /* at this point state must be one of:
2369 if (vp->specialStatus == VBUSY)
2372 /* if it's already attached, see if we can return it */
2373 if (V_attachState(vp) == VOL_STATE_ATTACHED) {
2374 VGetVolumeByVp_r(ec, vp);
2375 if (V_inUse(vp) == fileServer) {
2376 VCancelReservation_r(vp);
2380 /* otherwise, we need to detach, and attempt to re-attach */
2381 VDetachVolume_r(ec, vp);
2383 Log("VAttachVolume: Error detaching old volume instance (%s)\n", name);
2386 /* if it isn't fully attached, delete from the hash tables,
2387 and let the refcounter handle the rest */
2388 DeleteVolumeFromHashTable(vp);
2389 DeleteVolumeFromVByPList_r(vp);
2392 VCancelReservation_r(vp);
2396 /* pre-attach volume if it hasn't been done yet */
2398 (V_attachState(vp) == VOL_STATE_UNATTACHED) ||
2399 (V_attachState(vp) == VOL_STATE_DELETED) ||
2400 (V_attachState(vp) == VOL_STATE_ERROR)) {
2402 vp = VPreAttachVolumeByVp_r(ec, partp, vp, volumeId);
2408 opr_Assert(vp != NULL);
2410 /* handle pre-attach races
2412 * multiple threads can race to pre-attach a volume,
2413 * but we can't let them race beyond that
2415 * our solution is to let the first thread to bring
2416 * the volume into an exclusive state win; the other
2417 * threads just wait until it finishes bringing the
2418 * volume online, and then they do a vgetvolumebyvp
2420 if (svp && (svp != vp)) {
2421 /* wait for other exclusive ops to finish */
2422 VCreateReservation_r(vp);
2423 VWaitExclusiveState_r(vp);
2425 /* get a heavyweight ref, kill the lightweight ref, and return */
2426 VGetVolumeByVp_r(ec, vp);
2427 VCancelReservation_r(vp);
2431 /* at this point, we are chosen as the thread to do
2432 * demand attachment for this volume. all other threads
2433 * doing a getvolume on vp->hashid will block until we finish */
2435 /* make sure any old header cache entries are invalidated
2436 * before proceeding */
2437 FreeVolumeHeader(vp);
2439 VChangeState_r(vp, VOL_STATE_ATTACHING);
2441 /* restore any saved counters */
2442 memcpy(&vp->stats, &stats_save, sizeof(VolumeStats));
2443 #else /* AFS_DEMAND_ATTACH_FS */
2444 vp = VGetVolume_r(ec, volumeId);
2446 if (V_inUse(vp) == fileServer)
2448 if (vp->specialStatus == VBUSY)
2450 VDetachVolume_r(ec, vp);
2452 Log("VAttachVolume: Error detaching volume (%s)\n", name);
2456 #endif /* AFS_DEMAND_ATTACH_FS */
2460 strcpy(path, VPartitionPath(partp));
2464 strcat(path, OS_DIRSEP);
2468 vp = (Volume *) calloc(1, sizeof(Volume));
2469 opr_Assert(vp != NULL);
2470 vp->hashid = volumeId;
2471 vp->device = partp->device;
2472 vp->partition = partp;
2473 queue_Init(&vp->vnode_list);
2474 queue_Init(&vp->rx_call_list);
2475 #ifdef AFS_DEMAND_ATTACH_FS
2476 opr_cv_init(&V_attachCV(vp));
2477 #endif /* AFS_DEMAND_ATTACH_FS */
2480 /* attach2 is entered without any locks, and returns
2481 * with vol_glock_mutex held */
2482 vp = attach2(ec, volumeId, path, partp, vp, isbusy, mode, &checkedOut);
2484 if (VCanUseFSSYNC() && vp) {
2485 #ifdef AFS_DEMAND_ATTACH_FS
2486 if ((mode == V_VOLUPD) || (VolumeWriteable(vp) && (mode == V_CLONE))) {
2487 /* mark volume header as in use so that volser crashes lead to a
2488 * salvage attempt */
2489 VUpdateVolume_r(ec, vp, 0);
2491 /* for dafs, we should tell the fileserver, except for V_PEEK
2492 * where we know it is not necessary */
2493 if (mode == V_PEEK) {
2494 vp->needsPutBack = 0;
2496 vp->needsPutBack = VOL_PUTBACK;
2498 #else /* !AFS_DEMAND_ATTACH_FS */
2499 /* duplicate computation in fssync.c about whether the server
2500 * takes the volume offline or not. If the volume isn't
2501 * offline, we must not return it when we detach the volume,
2502 * or the server will abort */
2503 if (mode == V_READONLY || mode == V_PEEK
2504 || (!VolumeWriteable(vp) && (mode == V_CLONE || mode == V_DUMP)))
2505 vp->needsPutBack = 0;
2507 vp->needsPutBack = VOL_PUTBACK;
2508 #endif /* !AFS_DEMAND_ATTACH_FS */
2510 #ifdef FSSYNC_BUILD_CLIENT
2511 /* Only give back the vol to the fileserver if we checked it out; attach2
2512 * will set checkedOut only if we successfully checked it out from the
2514 if (VCanUseFSSYNC() && vp == NULL && checkedOut) {
2516 #ifdef AFS_DEMAND_ATTACH_FS
2517 /* If we couldn't attach but we scheduled a salvage, we already
2518 * notified the fileserver; don't online it now */
2519 if (*ec != VSALVAGING)
2520 #endif /* AFS_DEMAND_ATTACH_FS */
2521 FSYNC_VolOp(volumeId, partition, FSYNC_VOL_ON, 0, NULL);
2524 if (programType == fileServer && vp) {
2525 #ifdef AFS_DEMAND_ATTACH_FS
2527 * we can get here in cases where we don't "own"
2528 * the volume (e.g. volume owned by a utility).
2529 * short circuit around potential disk header races.
2531 if (V_attachState(vp) != VOL_STATE_ATTACHED) {
2535 VUpdateVolume_r(ec, vp, 0);
2537 Log("VAttachVolume: Error updating volume\n");
2542 if (VolumeWriteable(vp) && V_dontSalvage(vp) == 0) {
2543 #ifndef AFS_DEMAND_ATTACH_FS
2544 /* This is a hack: by temporarily setting the incore
2545 * dontSalvage flag ON, the volume will be put back on the
2546 * Update list (with dontSalvage OFF again). It will then
2547 * come back in N minutes with DONT_SALVAGE eventually
2548 * set. This is the way that volumes that have never had
2549 * it set get it set; or that volumes that have been
2550 * offline without DONT SALVAGE having been set also
2551 * eventually get it set */
2552 V_dontSalvage(vp) = DONT_SALVAGE;
2553 #endif /* !AFS_DEMAND_ATTACH_FS */
2554 VAddToVolumeUpdateList_r(ec, vp);
2556 Log("VAttachVolume: Error adding volume to update list\n");
2563 Log("VOnline: volume %" AFS_VOLID_FMT " (%s) attached and online\n", afs_printable_VolumeId_lu(V_id(vp)),
2568 if (VRequiresPartLock()) {
2569 VUnlockPartition_r(partition);
2572 #ifdef AFS_DEMAND_ATTACH_FS
2573 /* attach failed; make sure we're in error state */
2574 if (vp && !VIsErrorState(V_attachState(vp))) {
2575 VChangeState_r(vp, VOL_STATE_ERROR);
2577 #endif /* AFS_DEMAND_ATTACH_FS */
2584 #ifdef AFS_DEMAND_ATTACH_FS
2585 /* VAttachVolumeByVp_r
2587 * finish attaching a volume that is
2588 * in a less than fully attached state
2590 /* caller MUST hold a ref count on vp */
2592 VAttachVolumeByVp_r(Error * ec, Volume * vp, int mode)
2594 char name[VMAXPATHLEN];
2596 struct DiskPartition64 *partp;
2600 Volume * nvp = NULL;
2601 VolumeStats stats_save;
2605 /* volume utility should never call AttachByVp */
2606 opr_Assert(programType == fileServer);
2608 volumeId = vp->hashid;
2609 partp = vp->partition;
2610 VolumeExternalName_r(volumeId, name, sizeof(name));
2613 /* if another thread is performing a blocking op, wait */
2614 VWaitExclusiveState_r(vp);
2616 memcpy(&stats_save, &vp->stats, sizeof(VolumeStats));
2618 /* if it's already attached, see if we can return it */
2619 if (V_attachState(vp) == VOL_STATE_ATTACHED) {
2620 VGetVolumeByVp_r(ec, vp);
2621 if (V_inUse(vp) == fileServer) {
2624 if (vp->specialStatus == VBUSY)
2626 VDetachVolume_r(ec, vp);
2628 Log("VAttachVolume: Error detaching volume (%s)\n", name);
2634 /* pre-attach volume if it hasn't been done yet */
2636 (V_attachState(vp) == VOL_STATE_UNATTACHED) ||
2637 (V_attachState(vp) == VOL_STATE_DELETED) ||
2638 (V_attachState(vp) == VOL_STATE_ERROR)) {
2639 nvp = VPreAttachVolumeByVp_r(ec, partp, vp, volumeId);
2645 VCreateReservation_r(nvp);
2650 opr_Assert(vp != NULL);
2651 VChangeState_r(vp, VOL_STATE_ATTACHING);
2653 /* restore monotonically increasing stats */
2654 memcpy(&vp->stats, &stats_save, sizeof(VolumeStats));
2658 /* compute path to disk header */
2659 strcpy(path, VPartitionPath(partp));
2663 strcat(path, OS_DIRSEP);
2668 * NOTE: attach2 is entered without any locks, and returns
2669 * with vol_glock_mutex held */
2670 vp = attach2(ec, volumeId, path, partp, vp, isbusy, mode, &checkedOut);
2673 * the event that an error was encountered, or
2674 * the volume was not brought to an attached state
2675 * for any reason, skip to the end. We cannot
2676 * safely call VUpdateVolume unless we "own" it.
2680 (V_attachState(vp) != VOL_STATE_ATTACHED)) {
2684 VUpdateVolume_r(ec, vp, 0);
2686 Log("VAttachVolume: Error updating volume %" AFS_VOLID_FMT "\n",
2687 afs_printable_VolumeId_lu(vp->hashid));
2691 if (VolumeWriteable(vp) && V_dontSalvage(vp) == 0) {
2692 #ifndef AFS_DEMAND_ATTACH_FS
2693 /* This is a hack: by temporarily setting the incore
2694 * dontSalvage flag ON, the volume will be put back on the
2695 * Update list (with dontSalvage OFF again). It will then
2696 * come back in N minutes with DONT_SALVAGE eventually
2697 * set. This is the way that volumes that have never had
2698 * it set get it set; or that volumes that have been
2699 * offline without DONT SALVAGE having been set also
2700 * eventually get it set */
2701 V_dontSalvage(vp) = DONT_SALVAGE;
2702 #endif /* !AFS_DEMAND_ATTACH_FS */
2703 VAddToVolumeUpdateList_r(ec, vp);
2705 Log("VAttachVolume: Error adding volume %" AFS_VOLID_FMT " to update list\n",
2706 afs_printable_VolumeId_lu(vp->hashid));
2713 Log("VOnline: volume %" AFS_VOLID_FMT " (%s) attached and online\n",
2714 afs_printable_VolumeId_lu(V_id(vp)), V_name(vp));
2717 VCancelReservation_r(nvp);
2720 if (*ec && (*ec != VOFFLINE) && (*ec != VSALVAGE)) {
2721 if (vp && !VIsErrorState(V_attachState(vp))) {
2722 VChangeState_r(vp, VOL_STATE_ERROR);
2731 * lock a volume on disk (non-blocking).
2733 * @param[in] vp The volume to lock
2734 * @param[in] locktype READ_LOCK or WRITE_LOCK
2736 * @return operation status
2737 * @retval 0 success, lock was obtained
2738 * @retval EBUSY a conflicting lock was held by another process
2739 * @retval EIO error acquiring lock
2741 * @pre If we're in the fileserver, vp is in an exclusive state
2743 * @pre vp is not already locked
2746 VLockVolumeNB(Volume *vp, int locktype)
2750 opr_Assert(programType != fileServer
2751 || VIsExclusiveState(V_attachState(vp)));
2752 opr_Assert(!(V_attachFlags(vp) & VOL_LOCKED));
2754 code = VLockVolumeByIdNB(vp->hashid, vp->partition, locktype);
2756 V_attachFlags(vp) |= VOL_LOCKED;
2763 * unlock a volume on disk that was locked with VLockVolumeNB.
2765 * @param[in] vp volume to unlock
2767 * @pre If we're in the fileserver, vp is in an exclusive state
2769 * @pre vp has already been locked
2772 VUnlockVolume(Volume *vp)
2774 opr_Assert(programType != fileServer
2775 || VIsExclusiveState(V_attachState(vp)));
2776 opr_Assert((V_attachFlags(vp) & VOL_LOCKED));
2778 VUnlockVolumeById(vp->hashid, vp->partition);
2780 V_attachFlags(vp) &= ~VOL_LOCKED;
2782 #endif /* AFS_DEMAND_ATTACH_FS */
2785 * read in a vol header, possibly lock the vol header, and possibly check out
2786 * the vol header from the fileserver, as part of volume attachment.
2788 * @param[out] ec error code
2789 * @param[in] vp volume pointer object
2790 * @param[in] partp disk partition object of the attaching partition
2791 * @param[in] mode attachment mode such as V_VOLUPD, V_DUMP, etc (see
2793 * @param[in] peek 1 to just try to read in the volume header and make sure
2794 * we don't try to lock the vol, or check it out from
2795 * FSSYNC or anything like that; 0 otherwise, for 'normal'
2797 * @param[out] acheckedOut If we successfully checked-out the volume from
2798 * the fileserver (if we needed to), this is set
2799 * to 1, otherwise it is untouched.
2801 * @note As part of DAFS volume attachment, the volume header may be either
2802 * read- or write-locked to ensure mutual exclusion of certain volume
2803 * operations. In some cases in order to determine whether we need to
2804 * read- or write-lock the header, we need to read in the header to see
2805 * if the volume is RW or not. So, if we read in the header under a
2806 * read-lock and determine that we actually need a write-lock on the
2807 * volume header, this function will drop the read lock, acquire a write
2808 * lock, and read the header in again.
2811 attach_volume_header(Error *ec, Volume *vp, struct DiskPartition64 *partp,
2812 int mode, int peek, int *acheckedOut)
2814 struct VolumeDiskHeader diskHeader;
2815 struct VolumeHeader header;
2818 int lock_tries = 0, checkout_tries = 0;
2820 VolumeId volid = vp->hashid;
2821 #ifdef FSSYNC_BUILD_CLIENT
2822 int checkout, done_checkout = 0;
2823 #endif /* FSSYNC_BUILD_CLIENT */
2824 #ifdef AFS_DEMAND_ATTACH_FS
2825 int locktype = 0, use_locktype = -1;
2826 #endif /* AFS_DEMAND_ATTACH_FS */
2832 if (lock_tries > VOL_MAX_CHECKOUT_RETRIES) {
2833 Log("VAttachVolume: retried too many times trying to lock header for "
2834 "vol %lu part %s; giving up\n", afs_printable_uint32_lu(volid),
2835 VPartitionPath(partp));
2839 if (checkout_tries > VOL_MAX_CHECKOUT_RETRIES) {
2840 Log("VAttachVolume: retried too many times trying to checkout "
2841 "vol %lu part %s; giving up\n", afs_printable_uint32_lu(volid),
2842 VPartitionPath(partp));
2847 if (VReadVolumeDiskHeader(volid, partp, NULL)) {
2848 /* short-circuit the 'volume does not exist' case */
2853 #ifdef FSSYNC_BUILD_CLIENT
2854 checkout = !done_checkout;
2856 if (!peek && checkout && VMustCheckoutVolume(mode)) {
2858 memset(&res, 0, sizeof(res));
2860 if (FSYNC_VolOp(volid, partp->name, FSYNC_VOL_NEEDVOLUME, mode, &res)
2863 if (res.hdr.reason == FSYNC_SALVAGE) {
2864 Log("VAttachVolume: file server says volume %lu is salvaging\n",
2865 afs_printable_uint32_lu(volid));
2868 Log("VAttachVolume: attach of volume %lu apparently denied by file server\n",
2869 afs_printable_uint32_lu(volid));
2870 *ec = VNOVOL; /* XXXX */
2878 #ifdef AFS_DEMAND_ATTACH_FS
2879 if (use_locktype < 0) {
2880 /* don't know whether vol is RO or RW; assume it's RO and we can retry
2881 * if it turns out to be RW */
2882 locktype = VVolLockType(mode, 0);
2885 /* a previous try says we should use use_locktype to lock the volume,
2887 locktype = use_locktype;
2890 if (!peek && locktype) {
2891 code = VLockVolumeNB(vp, locktype);
2893 if (code == EBUSY) {
2894 Log("VAttachVolume: another program has vol %lu locked\n",
2895 afs_printable_uint32_lu(volid));
2897 Log("VAttachVolume: error %d trying to lock vol %lu\n",
2898 code, afs_printable_uint32_lu(volid));
2905 #endif /* AFS_DEMAND_ATTACH_FS */
2907 code = VReadVolumeDiskHeader(volid, partp, &diskHeader);
2917 DiskToVolumeHeader(&header, &diskHeader);
2919 IH_INIT(vp->vnodeIndex[vLarge].handle, partp->device, header.parent,
2920 header.largeVnodeIndex);
2921 IH_INIT(vp->vnodeIndex[vSmall].handle, partp->device, header.parent,
2922 header.smallVnodeIndex);
2923 IH_INIT(vp->diskDataHandle, partp->device, header.parent,
2925 IH_INIT(vp->linkHandle, partp->device, header.parent, header.linkTable);
2928 /* only need to do this once */
2930 GetVolumeHeader(vp);
2934 #if defined(AFS_DEMAND_ATTACH_FS) && defined(FSSYNC_BUILD_CLIENT)
2935 /* demand attach changes the V_PEEK mechanism
2937 * we can now suck the current disk data structure over
2938 * the fssync interface without going to disk
2940 * (technically, we don't need to restrict this feature
2941 * to demand attach fileservers. However, I'm trying
2942 * to limit the number of common code changes)
2944 if (VCanUseFSSYNC() && (mode == V_PEEK || peek)) {
2946 res.payload.len = sizeof(VolumeDiskData);
2947 res.payload.buf = &(V_disk(vp));
2949 if (FSYNC_VolOp(vp->hashid,
2951 FSYNC_VOL_QUERY_HDR,
2954 goto disk_header_loaded;
2957 #endif /* AFS_DEMAND_ATTACH_FS && FSSYNC_BUILD_CLIENT */
2958 (void)ReadHeader(ec, V_diskDataHandle(vp), (char *)&V_disk(vp),
2959 sizeof(V_disk(vp)), VOLUMEINFOMAGIC, VOLUMEINFOVERSION);
2961 #ifdef AFS_DEMAND_ATTACH_FS
2964 IncUInt64(&VStats.hdr_loads);
2965 IncUInt64(&vp->stats.hdr_loads);
2967 #endif /* AFS_DEMAND_ATTACH_FS */
2970 Log("VAttachVolume: Error reading diskDataHandle header for vol %lu; "
2971 "error=%u\n", afs_printable_uint32_lu(volid), *ec);
2975 #ifdef AFS_DEMAND_ATTACH_FS
2976 # ifdef FSSYNC_BUILD_CLIENT
2978 # endif /* FSSYNC_BUILD_CLIENT */
2980 /* if the lock type we actually used to lock the volume is different than
2981 * the lock type we should have used, retry with the lock type we should
2983 use_locktype = VVolLockType(mode, VolumeWriteable(vp));
2984 if (locktype != use_locktype) {
2988 #endif /* AFS_DEMAND_ATTACH_FS */
2993 #if defined(AFS_DEMAND_ATTACH_FS) && defined(FSSYNC_BUILD_CLIENT)
2994 if (!peek && *ec == 0 && retry == 0 && VMustCheckoutVolume(mode)) {
2996 code = FSYNC_VerifyCheckout(volid, partp->name, FSYNC_VOL_NEEDVOLUME, mode);
2998 if (code == SYNC_DENIED) {
2999 /* must retry checkout; fileserver no longer thinks we have
3005 } else if (code != SYNC_OK) {
3009 #endif /* AFS_DEMAND_ATTACH_FS && FSSYNC_BUILD_CLIENT */
3012 /* either we are going to be called again for a second pass, or we
3013 * encountered an error; clean up in either case */
3015 #ifdef AFS_DEMAND_ATTACH_FS
3016 if ((V_attachFlags(vp) & VOL_LOCKED)) {
3019 #endif /* AFS_DEMAND_ATTACH_FS */
3020 if (vp->linkHandle) {
3021 IH_RELEASE(vp->vnodeIndex[vLarge].handle);
3022 IH_RELEASE(vp->vnodeIndex[vSmall].handle);
3023 IH_RELEASE(vp->diskDataHandle);
3024 IH_RELEASE(vp->linkHandle);
3030 FreeVolumeHeader(vp);
3040 #ifdef AFS_DEMAND_ATTACH_FS
3042 attach_check_vop(Error *ec, VolumeId volid, struct DiskPartition64 *partp,
3043 Volume *vp, int *acheckedOut)
3047 if (vp->pending_vol_op) {
3051 if (vp->pending_vol_op->vol_op_state == FSSYNC_VolOpRunningUnknown) {
3053 code = VVolOpLeaveOnlineNoHeader_r(vp, vp->pending_vol_op);
3055 vp->pending_vol_op->vol_op_state = FSSYNC_VolOpRunningOnline;
3056 } else if (code == 0) {
3057 vp->pending_vol_op->vol_op_state = FSSYNC_VolOpRunningOffline;
3060 /* we need the vol header to determine if the volume can be
3061 * left online for the vop, so... get the header */
3065 /* attach header with peek=1 to avoid checking out the volume
3066 * or locking it; we just want the header info, we're not
3067 * messing with the volume itself at all */
3068 attach_volume_header(ec, vp, partp, V_PEEK, 1, acheckedOut);
3075 if (VVolOpLeaveOnline_r(vp, vp->pending_vol_op)) {
3076 vp->pending_vol_op->vol_op_state = FSSYNC_VolOpRunningOnline;
3078 vp->pending_vol_op->vol_op_state = FSSYNC_VolOpRunningOffline;
3081 /* make sure we grab a new vol header and re-open stuff on
3082 * actual attachment; we can't keep the data we grabbed, since
3083 * it was not done under a lock and thus not safe */
3084 FreeVolumeHeader(vp);
3085 VReleaseVolumeHandles_r(vp);
3088 /* see if the pending volume op requires exclusive access */
3089 switch (vp->pending_vol_op->vol_op_state) {
3090 case FSSYNC_VolOpPending:
3091 /* this should never happen */
3092 opr_Assert(vp->pending_vol_op->vol_op_state
3093 != FSSYNC_VolOpPending);
3096 case FSSYNC_VolOpRunningUnknown:
3097 /* this should never happen; we resolved 'unknown' above */
3098 opr_Assert(vp->pending_vol_op->vol_op_state
3099 != FSSYNC_VolOpRunningUnknown);
3102 case FSSYNC_VolOpRunningOffline:
3103 /* mark the volume down */
3105 VChangeState_r(vp, VOL_STATE_UNATTACHED);
3107 /* do not set V_offlineMessage here; we don't have ownership of
3108 * the volume (and probably do not have the header loaded), so we
3109 * can't alter the disk header */
3111 /* check to see if we should set the specialStatus flag */
3112 if (VVolOpSetVBusy_r(vp, vp->pending_vol_op)) {
3113 /* don't overwrite specialStatus if it was already set to
3114 * something else (e.g. VMOVED) */
3115 if (!vp->specialStatus) {
3116 vp->specialStatus = VBUSY;
3128 #endif /* AFS_DEMAND_ATTACH_FS */
3131 * volume attachment helper function.
3133 * @param[out] ec error code
3134 * @param[in] volumeId volume ID of the attaching volume
3135 * @param[in] path full path to the volume header .vol file
3136 * @param[in] partp disk partition object for the attaching partition
3137 * @param[in] vp volume object; vp->hashid, vp->device, vp->partition,
3138 * vp->vnode_list, vp->rx_call_list, and V_attachCV (for
3139 * DAFS) should already be initialized
3140 * @param[in] isbusy 1 if vp->specialStatus should be set to VBUSY; that is,
3141 * if there is a volume operation running for this volume
3142 * that should set the volume to VBUSY during its run. 0
3143 * otherwise. (see VVolOpSetVBusy_r)
3144 * @param[in] mode attachment mode such as V_VOLUPD, V_DUMP, etc (see
3146 * @param[out] acheckedOut If we successfully checked-out the volume from
3147 * the fileserver (if we needed to), this is set
3148 * to 1, otherwise it is 0.
3150 * @return pointer to the semi-attached volume pointer
3151 * @retval NULL an error occurred (check value of *ec)
3152 * @retval vp volume successfully attaching
3154 * @pre no locks held
3156 * @post VOL_LOCK held
3159 attach2(Error * ec, VolumeId volumeId, char *path, struct DiskPartition64 *partp,
3160 Volume * vp, int isbusy, int mode, int *acheckedOut)
3162 /* have we read in the header successfully? */
3163 int read_header = 0;
3165 #ifdef AFS_DEMAND_ATTACH_FS
3166 /* should we FreeVolume(vp) instead of VCheckFree(vp) in the error
3170 /* in the case of an error, to what state should the volume be
3172 VolState error_state = VOL_STATE_ERROR;
3173 #endif /* AFS_DEMAND_ATTACH_FS */
3177 vp->vnodeIndex[vLarge].handle = NULL;
3178 vp->vnodeIndex[vSmall].handle = NULL;
3179 vp->diskDataHandle = NULL;
3180 vp->linkHandle = NULL;
3184 #ifdef AFS_DEMAND_ATTACH_FS
3185 attach_check_vop(ec, volumeId, partp, vp, acheckedOut);
3187 attach_volume_header(ec, vp, partp, mode, 0, acheckedOut);
3190 attach_volume_header(ec, vp, partp, mode, 0, acheckedOut);
3191 #endif /* !AFS_DEMAND_ATTACH_FS */
3193 if (*ec == VNOVOL) {
3194 /* if the volume doesn't exist, skip straight to 'error' so we don't
3195 * request a salvage */
3196 goto unlocked_error;
3202 /* ensure that we don't override specialStatus if it was set to
3203 * something else (e.g. VMOVED) */
3204 if (isbusy && !vp->specialStatus) {
3205 vp->specialStatus = VBUSY;
3207 vp->shuttingDown = 0;
3208 vp->goingOffline = 0;
3210 #ifdef AFS_DEMAND_ATTACH_FS
3211 vp->stats.last_attach = FT_ApproxTime();
3212 vp->stats.attaches++;
3216 IncUInt64(&VStats.attaches);
3217 vp->cacheCheck = ++VolumeCacheCheck;
3218 /* just in case this ever rolls over */
3219 if (!vp->cacheCheck)
3220 vp->cacheCheck = ++VolumeCacheCheck;
3223 #ifdef AFS_DEMAND_ATTACH_FS
3224 V_attachFlags(vp) |= VOL_HDR_LOADED;
3225 vp->stats.last_hdr_load = vp->stats.last_attach;
3226 #endif /* AFS_DEMAND_ATTACH_FS */
3230 struct IndexFileHeader iHead;
3233 * We just read in the diskstuff part of the header. If the detailed
3234 * volume stats area has not yet been initialized, we should bzero the
3235 * area and mark it as initialized.
3237 if (!(V_stat_initialized(vp))) {
3238 memset((V_stat_area(vp)), 0, VOL_STATS_BYTES);
3239 V_stat_initialized(vp) = 1;
3242 (void)ReadHeader(ec, vp->vnodeIndex[vSmall].handle,
3243 (char *)&iHead, sizeof(iHead),
3244 SMALLINDEXMAGIC, SMALLINDEXVERSION);
3247 Log("VAttachVolume: Error reading smallVnode vol header %s; error=%u\n", path, *ec);
3252 struct IndexFileHeader iHead;
3254 (void)ReadHeader(ec, vp->vnodeIndex[vLarge].handle,
3255 (char *)&iHead, sizeof(iHead),
3256 LARGEINDEXMAGIC, LARGEINDEXVERSION);
3259 Log("VAttachVolume: Error reading largeVnode vol header %s; error=%u\n", path, *ec);
3263 #ifdef AFS_NAMEI_ENV
3265 struct versionStamp stamp;
3267 (void)ReadHeader(ec, V_linkHandle(vp), (char *)&stamp,
3268 sizeof(stamp), LINKTABLEMAGIC, LINKTABLEVERSION);
3271 Log("VAttachVolume: Error reading namei vol header %s; error=%u\n", path, *ec);
3274 #endif /* AFS_NAMEI_ENV */
3276 #if defined(AFS_DEMAND_ATTACH_FS)
3277 if (*ec && ((*ec != VOFFLINE) || (V_attachState(vp) != VOL_STATE_UNATTACHED))) {
3279 if (!VCanScheduleSalvage()) {
3280 Log("VAttachVolume: Error attaching volume %s; volume needs salvage; error=%u\n", path, *ec);
3282 VRequestSalvage_r(ec, vp, SALVSYNC_ERROR, VOL_SALVAGE_NO_OFFLINE);
3287 /* volume operation in progress */
3289 /* we have already transitioned the vp away from ATTACHING state, so we
3290 * can go right to the end of attach2, and we do not need to transition
3292 goto error_notbroken;
3294 #else /* AFS_DEMAND_ATTACH_FS */
3296 Log("VAttachVolume: Error attaching volume %s; volume needs salvage; error=%u\n", path, *ec);
3297 goto unlocked_error;
3299 #endif /* AFS_DEMAND_ATTACH_FS */
3301 if (V_needsSalvaged(vp)) {
3302 if (vp->specialStatus)
3303 vp->specialStatus = 0;
3305 #if defined(AFS_DEMAND_ATTACH_FS)
3306 if (!VCanScheduleSalvage()) {
3307 Log("VAttachVolume: volume salvage flag is ON for %s; volume needs salvage\n", path);
3309 VRequestSalvage_r(ec, vp, SALVSYNC_NEEDED, VOL_SALVAGE_NO_OFFLINE);
3312 #else /* AFS_DEMAND_ATTACH_FS */
3314 #endif /* AFS_DEMAND_ATTACH_FS */
3320 vp->nextVnodeUnique = V_uniquifier(vp);
3322 if (VShouldCheckInUse(mode) && V_inUse(vp) && VolumeWriteable(vp)) {
3323 if (!V_needsSalvaged(vp)) {
3324 V_needsSalvaged(vp) = 1;
3325 VUpdateVolume_r(ec, vp, 0);
3327 #if defined(AFS_DEMAND_ATTACH_FS)
3328 if (!VCanScheduleSalvage()) {
3329 Log("VAttachVolume: volume %s needs to be salvaged; not attached.\n", path);
3331 VRequestSalvage_r(ec, vp, SALVSYNC_NEEDED, VOL_SALVAGE_NO_OFFLINE);
3334 #else /* AFS_DEMAND_ATTACH_FS */
3335 Log("VAttachVolume: volume %s needs to be salvaged; not attached.\n", path);
3337 #endif /* AFS_DEMAND_ATTACH_FS */
3342 if (programType == fileServer && V_destroyMe(vp) == DESTROY_ME) {
3343 /* Only check destroyMe if we are the fileserver, since the
3344 * volserver et al sometimes need to work with volumes with
3345 * destroyMe set. Examples are 'temporary' volumes the
3346 * volserver creates, and when we create a volume (destroyMe
3347 * is set on creation; sometimes a separate volserver
3348 * transaction is created to clear destroyMe).
3351 #if defined(AFS_DEMAND_ATTACH_FS)
3352 /* schedule a salvage so the volume goes away on disk */
3353 VRequestSalvage_r(ec, vp, SALVSYNC_ERROR, VOL_SALVAGE_NO_OFFLINE);
3354 VChangeState_r(vp, VOL_STATE_ERROR);
3357 #endif /* AFS_DEMAND_ATTACH_FS */
3358 Log("VAttachVolume: volume %s is junk; it should be destroyed at next salvage\n", path);
3363 vp->vnodeIndex[vSmall].bitmap = vp->vnodeIndex[vLarge].bitmap = NULL;
3364 #ifndef BITMAP_LATER
3365 if (programType == fileServer && VolumeWriteable(vp)) {
3367 for (i = 0; i < nVNODECLASSES; i++) {
3368 VGetBitmap_r(ec, vp, i);
3370 #ifdef AFS_DEMAND_ATTACH_FS
3371 VRequestSalvage_r(ec, vp, SALVSYNC_ERROR, VOL_SALVAGE_NO_OFFLINE);
3373 #endif /* AFS_DEMAND_ATTACH_FS */
3374 Log("VAttachVolume: error getting bitmap for volume (%s)\n",
3380 #endif /* BITMAP_LATER */
3382 if (VInit >= 2 && V_needsCallback(vp)) {
3383 if (V_BreakVolumeCallbacks) {
3384 Log("VAttachVolume: Volume %lu was changed externally; breaking callbacks\n",
3385 afs_printable_uint32_lu(V_id(vp)));
3386 V_needsCallback(vp) = 0;
3388 (*V_BreakVolumeCallbacks) (V_id(vp));
3391 VUpdateVolume_r(ec, vp, 0);
3393 #ifdef FSSYNC_BUILD_CLIENT
3394 else if (VCanUseFSSYNC()) {
3395 afs_int32 fsync_code;
3397 V_needsCallback(vp) = 0;
3399 fsync_code = FSYNC_VolOp(V_id(vp), NULL, FSYNC_VOL_BREAKCBKS, FSYNC_WHATEVER, NULL);
3403 V_needsCallback(vp) = 1;
3404 Log("Error trying to tell the fileserver to break callbacks for "
3405 "changed volume %lu; error code %ld\n",
3406 afs_printable_uint32_lu(V_id(vp)),
3407 afs_printable_int32_ld(fsync_code));
3409 VUpdateVolume_r(ec, vp, 0);
3412 #endif /* FSSYNC_BUILD_CLIENT */
3415 Log("VAttachVolume: error %d clearing needsCallback on volume "
3416 "%lu; needs salvage\n", (int)*ec,
3417 afs_printable_uint32_lu(V_id(vp)));
3418 #ifdef AFS_DEMAND_ATTACH_FS
3419 VRequestSalvage_r(ec, vp, SALVSYNC_ERROR, VOL_SALVAGE_NO_OFFLINE);
3421 #else /* !AFS_DEMAND_ATTACH_FS */
3423 #endif /* !AFS_DEMAND_ATTACh_FS */
3428 if (programType == fileServer) {
3429 if (vp->specialStatus)
3430 vp->specialStatus = 0;
3431 if (V_blessed(vp) && V_inService(vp) && !V_needsSalvaged(vp)) {
3432 V_inUse(vp) = fileServer;
3433 V_offlineMessage(vp)[0] = '\0';
3435 #ifdef AFS_DEMAND_ATTACH_FS
3436 /* check if the volume is actually usable. only do this for DAFS; for
3437 * non-DAFS, volumes that are not inService/blessed can still be
3438 * attached, even if clients cannot access them. this is relevant
3439 * because for non-DAFS, we try to attach the volume when e.g.
3440 * volserver gives us back then vol when its done with it, but
3441 * volserver may give us back a volume that is not inService/blessed. */
3445 /* Put the vol into PREATTACHED state, so if someone tries to
3446 * access it again, we try to attach, see that we're not blessed,
3447 * and give a VNOVOL error again. Putting it into UNATTACHED state
3448 * would result in a VOFFLINE error instead. */
3449 error_state = VOL_STATE_PREATTACHED;
3451 /* mimic e.g. GetVolume errors */
3452 if (!V_blessed(vp)) {
3453 Log("Volume %lu offline: not blessed\n", afs_printable_uint32_lu(V_id(vp)));
3454 FreeVolumeHeader(vp);
3455 } else if (!V_inService(vp)) {
3456 Log("Volume %lu offline: not in service\n", afs_printable_uint32_lu(V_id(vp)));
3457 FreeVolumeHeader(vp);
3459 Log("Volume %lu offline: needs salvage\n", afs_printable_uint32_lu(V_id(vp)));
3461 error_state = VOL_STATE_ERROR;
3462 /* see if we can recover */
3463 VRequestSalvage_r(ec, vp, SALVSYNC_NEEDED, VOL_SALVAGE_NO_OFFLINE);
3468 #endif /* AFS_DEMAND_ATTACH_FS */
3470 #ifdef AFS_DEMAND_ATTACH_FS
3471 if ((mode != V_PEEK) && (mode != V_SECRETLY) && (mode != V_READONLY))
3472 V_inUse(vp) = programType;
3473 #endif /* AFS_DEMAND_ATTACH_FS */
3474 V_checkoutMode(vp) = mode;
3477 AddVolumeToHashTable(vp, V_id(vp));
3478 #ifdef AFS_DEMAND_ATTACH_FS
3479 if (VCanUnlockAttached() && (V_attachFlags(vp) & VOL_LOCKED)) {
3482 if ((programType != fileServer) ||
3483 (V_inUse(vp) == fileServer)) {
3484 AddVolumeToVByPList_r(vp);
3486 VChangeState_r(vp, VOL_STATE_ATTACHED);
3488 VChangeState_r(vp, VOL_STATE_UNATTACHED);
3497 #ifdef AFS_DEMAND_ATTACH_FS
3498 if (!VIsErrorState(V_attachState(vp))) {
3499 if (VIsErrorState(error_state)) {
3500 Log("attach2: forcing vol %" AFS_VOLID_FMT " to error state (state %u flags 0x%x ec %d)\n",
3501 afs_printable_VolumeId_lu(vp->hashid), V_attachState(vp),
3502 V_attachFlags(vp), *ec);
3504 VChangeState_r(vp, error_state);
3506 #endif /* AFS_DEMAND_ATTACH_FS */
3509 VReleaseVolumeHandles_r(vp);
3512 #ifdef AFS_DEMAND_ATTACH_FS
3514 if (VCheckSalvage(vp) == VCHECK_SALVAGE_FAIL) {
3515 /* The salvage could not be scheduled with the salvage server
3516 * due to a hard error. Reset the error code to prevent retry loops by
3518 if (*ec == VSALVAGING) {
3527 #else /* !AFS_DEMAND_ATTACH_FS */
3529 #endif /* !AFS_DEMAND_ATTACH_FS */
3533 /* Attach an existing volume.
3534 The volume also normally goes online at this time.
3535 An offline volume must be reattached to make it go online.
3539 VAttachVolume(Error * ec, VolumeId volumeId, int mode)
3543 retVal = VAttachVolume_r(ec, volumeId, mode);
3549 VAttachVolume_r(Error * ec, VolumeId volumeId, int mode)
3552 VGetVolumePath(ec, volumeId, &part, &name);
3556 vp = VGetVolume_r(&error, volumeId);
3558 opr_Assert(V_inUse(vp) == 0);
3559 VDetachVolume_r(ec, vp);
3563 return VAttachVolumeByName_r(ec, part, name, mode);
3566 /* Increment a reference count to a volume, sans context swaps. Requires
3567 * possibly reading the volume header in from the disk, since there's
3568 * an invariant in the volume package that nUsers>0 ==> vp->header is valid.
3570 * N.B. This call can fail if we can't read in the header!! In this case
3571 * we still guarantee we won't context swap, but the ref count won't be
3572 * incremented (otherwise we'd violate the invariant).
3574 /* NOTE: with the demand attach fileserver extensions, the global lock
3575 * is dropped within VHold */
3576 #ifdef AFS_DEMAND_ATTACH_FS
3578 VHold_r(Volume * vp)
3582 VCreateReservation_r(vp);
3583 VWaitExclusiveState_r(vp);
3585 LoadVolumeHeader(&error, vp);
3587 VCancelReservation_r(vp);
3591 VCancelReservation_r(vp);
3594 #else /* AFS_DEMAND_ATTACH_FS */
3596 VHold_r(Volume * vp)
3600 LoadVolumeHeader(&error, vp);
3606 #endif /* AFS_DEMAND_ATTACH_FS */
3608 /**** volume timeout-related stuff ****/
3610 #ifdef AFS_PTHREAD_ENV
3612 static struct timespec *shutdown_timeout;
3613 static pthread_once_t shutdown_timeout_once = PTHREAD_ONCE_INIT;
3616 VTimedOut(const struct timespec *ts)
3621 if (ts->tv_sec == 0) {
3622 /* short-circuit; this will have always timed out */
3626 code = gettimeofday(&tv, NULL);
3628 Log("Error %d from gettimeofday, assuming we have not timed out\n", errno);
3629 /* assume no timeout; failure mode is we just wait longer than normal
3630 * instead of returning errors when we shouldn't */
3634 if (tv.tv_sec < ts->tv_sec ||
3635 (tv.tv_sec == ts->tv_sec && tv.tv_usec*1000 < ts->tv_nsec)) {
3644 * Calculate an absolute timeout.
3646 * @param[out] ts A timeout that is "timeout" seconds from now, if we return
3647 * NULL, the memory is not touched
3648 * @param[in] timeout How long the timeout should be from now
3650 * @return timeout to use
3651 * @retval NULL no timeout; wait forever
3652 * @retval non-NULL the given value for "ts"
3656 static struct timespec *
3657 VCalcTimeout(struct timespec *ts, afs_int32 timeout)
3667 ts->tv_sec = ts->tv_nsec = 0;
3671 code = gettimeofday(&now, NULL);
3673 Log("Error %d from gettimeofday, falling back to 'forever' timeout\n", errno);
3677 ts->tv_sec = now.tv_sec + timeout;
3678 ts->tv_nsec = now.tv_usec * 1000;
3684 * Initialize the shutdown_timeout global.
3687 VShutdownTimeoutInit(void)
3689 struct timespec *ts;
3691 ts = malloc(sizeof(*ts));
3693 shutdown_timeout = VCalcTimeout(ts, vol_opts.offline_shutdown_timeout);
3695 if (!shutdown_timeout) {
3701 * Figure out the timeout that should be used for waiting for offline volumes.
3703 * @param[out] ats Storage space for a local timeout value if needed
3705 * @return The timeout value that should be used
3706 * @retval NULL No timeout; wait forever for offlining volumes
3707 * @retval non-NULL A pointer to the absolute time that should be used as
3708 * the deadline for waiting for offlining volumes.
3710 * @note If we return non-NULL, the pointer we return may or may not be the
3713 static const struct timespec *
3714 VOfflineTimeout(struct timespec *ats)
3716 if (vol_shutting_down) {
3717 opr_Verify(pthread_once(&shutdown_timeout_once,
3718 VShutdownTimeoutInit) == 0);
3719 return shutdown_timeout;
3721 return VCalcTimeout(ats, vol_opts.offline_timeout);
3725 #else /* AFS_PTHREAD_ENV */
3727 /* Waiting a certain amount of time for offlining volumes is not supported
3728 * for LWP due to a lack of primitives. So, we never time out */
3729 # define VTimedOut(x) (0)
3730 # define VOfflineTimeout(x) (NULL)
3732 #endif /* !AFS_PTHREAD_ENV */
3740 retVal = VHold_r(vp);
3747 VIsGoingOffline_r(struct Volume *vp)
3751 if (vp->goingOffline) {
3752 if (vp->specialStatus) {
3753 code = vp->specialStatus;
3754 } else if (V_inService(vp) == 0 || V_blessed(vp) == 0) {
3765 * Tell the caller if a volume is waiting to go offline.
3767 * @param[in] vp The volume we want to know about
3769 * @return volume status
3770 * @retval 0 volume is not waiting to go offline, go ahead and use it
3771 * @retval nonzero volume is waiting to offline, and give the returned code
3772 * as an error to anyone accessing the volume
3774 * @pre VOL_LOCK is NOT held
3775 * @pre caller holds a heavyweight reference on vp
3778 VIsGoingOffline(struct Volume *vp)
3783 code = VIsGoingOffline_r(vp);
3790 * Register an RX call with a volume.
3792 * @param[inout] ec Error code; if unset when passed in, may be set if
3793 * the volume starts going offline
3794 * @param[out] client_ec @see GetVolume
3795 * @param[in] vp Volume struct
3796 * @param[in] cbv VCallByVol struct containing the RX call to register
3798 * @pre VOL_LOCK held
3799 * @pre caller holds heavy ref on vp
3804 VRegisterCall_r(Error *ec, Error *client_ec, Volume *vp, struct VCallByVol *cbv)
3807 #ifdef AFS_DEMAND_ATTACH_FS
3809 /* just in case the volume started going offline after we got the
3810 * reference to it... otherwise, if the volume started going
3811 * offline right at the end of GetVolume(), we might race with the
3812 * RX call scanner, and return success and add our cbv to the
3813 * rx_call_list _after_ the scanner has scanned the list. */
3814 *ec = VIsGoingOffline_r(vp);
3820 while (V_attachState(vp) == VOL_STATE_SCANNING_RXCALLS) {
3821 VWaitStateChange_r(vp);
3823 #endif /* AFS_DEMAND_ATTACH_FS */
3825 queue_Prepend(&vp->rx_call_list, cbv);
3830 * Deregister an RX call with a volume.
3832 * @param[in] vp Volume struct
3833 * @param[in] cbv VCallByVol struct containing the RX call to deregister
3835 * @pre VOL_LOCK held
3836 * @pre caller holds heavy ref on vp
3841 VDeregisterCall_r(Volume *vp, struct VCallByVol *cbv)
3843 if (cbv && queue_IsOnQueue(cbv)) {
3844 #ifdef AFS_DEMAND_ATTACH_FS
3845 while (V_attachState(vp) == VOL_STATE_SCANNING_RXCALLS) {
3846 VWaitStateChange_r(vp);
3848 #endif /* AFS_DEMAND_ATTACH_FS */
3854 /***************************************************/
3855 /* get and put volume routines */
3856 /***************************************************/
3859 * put back a heavyweight reference to a volume object.
3861 * @param[in] vp volume object pointer
3863 * @pre VOL_LOCK held
3865 * @post heavyweight volume reference put back.
3866 * depending on state, volume may have been taken offline,
3867 * detached, salvaged, freed, etc.
3869 * @internal volume package internal use only
3872 VPutVolume_r(Volume * vp)
3874 opr_Verify(--vp->nUsers >= 0);
3875 if (vp->nUsers == 0) {
3877 ReleaseVolumeHeader(vp->header);
3878 #ifdef AFS_DEMAND_ATTACH_FS
3879 if (!VCheckDetach(vp)) {
3883 #else /* AFS_DEMAND_ATTACH_FS */
3885 #endif /* AFS_DEMAND_ATTACH_FS */
3890 VPutVolume(Volume * vp)
3898 * Puts a volume reference obtained with VGetVolumeWithCall.
3900 * @param[in] vp Volume struct
3901 * @param[in] cbv VCallByVol struct given to VGetVolumeWithCall, or NULL if none
3903 * @pre VOL_LOCK is NOT held
3906 VPutVolumeWithCall(Volume *vp, struct VCallByVol *cbv)
3909 VDeregisterCall_r(vp, cbv);
3914 /* Get a pointer to an attached volume. The pointer is returned regardless
3915 of whether or not the volume is in service or on/off line. An error
3916 code, however, is returned with an indication of the volume's status */
3918 VGetVolume(Error * ec, Error * client_ec, VolumeId volumeId)
3922 retVal = GetVolume(ec, client_ec, volumeId, NULL, 0);
3928 * Get a volume reference associated with an RX call.
3930 * @param[out] ec @see GetVolume
3931 * @param[out] client_ec @see GetVolume
3932 * @param[in] volumeId @see GetVolume
3933 * @param[in] ts How long to wait for going-offline volumes (absolute time).
3934 * If NULL, wait forever. If ts->tv_sec == 0, return immediately
3935 * with an error if the volume is going offline.
3936 * @param[in] cbv Contains an RX call to be associated with this volume
3937 * reference. This call may be interrupted if the volume is
3938 * requested to go offline while we hold a ref on it. Give NULL
3939 * to not associate an RX call with this reference.
3941 * @return @see GetVolume
3943 * @note for LWP builds, ts must be NULL
3945 * @note A reference obtained with this function MUST be put back with
3946 * VPutVolumeWithCall
3949 VGetVolumeWithCall(Error * ec, Error * client_ec, VolumeId volumeId,
3950 const struct timespec *ts, struct VCallByVol *cbv)
3954 retVal = GetVolume(ec, client_ec, volumeId, NULL, ts);
3955 VRegisterCall_r(ec, client_ec, retVal, cbv);
3961 VGetVolume_r(Error * ec, VolumeId volumeId)
3963 return GetVolume(ec, NULL, volumeId, NULL, NULL);
3966 /* try to get a volume we've previously looked up */
3967 /* for demand attach fs, caller MUST NOT hold a ref count on vp */
3969 VGetVolumeByVp_r(Error * ec, Volume * vp)
3971 return GetVolume(ec, NULL, vp->hashid, vp, NULL);
3975 * private interface for getting a volume handle
3977 * @param[out] ec error code (0 if no error)
3978 * @param[out] client_ec wire error code to be given to clients
3979 * @param[in] volumeId ID of the volume we want
3980 * @param[in] hint optional hint for hash lookups, or NULL
3981 * @param[in] timeout absolute deadline for waiting for the volume to go
3982 * offline, if it is going offline. NULL to wait forever.
3984 * @return a volume handle for the specified volume
3985 * @retval NULL an error occurred, or the volume is in such a state that
3986 * we cannot load a header or return any volume struct
3988 * @note for DAFS, caller must NOT hold a ref count on 'hint'
3990 * @note 'timeout' is only checked if the volume is actually going offline; so
3991 * if you pass timeout->tv_sec = 0, this will exhibit typical
3992 * nonblocking behavior.
3994 * @note for LWP builds, 'timeout' must be NULL
3997 GetVolume(Error * ec, Error * client_ec, VolumeId volumeId, Volume * hint,
3998 const struct timespec *timeout)
4001 /* pull this profiling/debugging code out of regular builds */
4003 #define VGET_CTR_INC(x) x++
4004 unsigned short V0 = 0, V1 = 0, V2 = 0, V3 = 0, V5 = 0, V6 =
4005 0, V7 = 0, V8 = 0, V9 = 0;
4006 unsigned short V10 = 0, V11 = 0, V12 = 0, V13 = 0, V14 = 0, V15 = 0;
4008 #define VGET_CTR_INC(x)
4010 #ifdef AFS_DEMAND_ATTACH_FS
4011 Volume *avp, * rvp = hint;
4015 * if VInit is zero, the volume package dynamic
4016 * data structures have not been initialized yet,
4017 * and we must immediately return an error
4023 *client_ec = VOFFLINE;
4028 #ifdef AFS_DEMAND_ATTACH_FS
4030 VCreateReservation_r(rvp);
4032 #endif /* AFS_DEMAND_ATTACH_FS */
4040 vp = VLookupVolume_r(ec, volumeId, vp);
4046 #ifdef AFS_DEMAND_ATTACH_FS
4047 if (rvp && (rvp != vp)) {
4048 /* break reservation on old vp */
4049 VCancelReservation_r(rvp);
4052 #endif /* AFS_DEMAND_ATTACH_FS */
4058 /* Until we have reached an initialization level of 2
4059 * we don't know whether this volume exists or not.
4060 * We can't sleep and retry later because before a volume
4061 * is attached, the caller tries to get it first. Just
4062 * return VOFFLINE and the caller can choose whether to
4063 * retry the command or not. */
4073 IncUInt64(&VStats.hdr_gets);
4075 #ifdef AFS_DEMAND_ATTACH_FS
4076 /* block if someone else is performing an exclusive op on this volume */
4079 VCreateReservation_r(rvp);
4081 VWaitExclusiveState_r(vp);
4083 /* short circuit with VNOVOL in the following circumstances:
4086 * - VOL_STATE_SHUTTING_DOWN
4088 if ((V_attachState(vp) == VOL_STATE_ERROR) ||
4089 (V_attachState(vp) == VOL_STATE_SHUTTING_DOWN)) {
4096 * short circuit with VOFFLINE for VOL_STATE_UNATTACHED/GOING_OFFLINE and
4097 * VNOVOL for VOL_STATE_DELETED
4099 if ((V_attachState(vp) == VOL_STATE_UNATTACHED) ||
4100 (V_attachState(vp) == VOL_STATE_GOING_OFFLINE) ||
4101 (V_attachState(vp) == VOL_STATE_DELETED)) {
4102 if (vp->specialStatus) {
4103 *ec = vp->specialStatus;
4104 } else if (V_attachState(vp) == VOL_STATE_DELETED) {
4113 /* allowable states:
4120 if (vp->salvage.requested) {
4121 VUpdateSalvagePriority_r(vp);
4124 if (V_attachState(vp) == VOL_STATE_PREATTACHED) {
4125 if (vp->specialStatus) {
4126 *ec = vp->specialStatus;
4130 avp = VAttachVolumeByVp_r(ec, vp, 0);
4133 /* VAttachVolumeByVp_r can return a pointer
4134 * != the vp passed to it under certain
4135 * conditions; make sure we don't leak
4136 * reservations if that happens */
4138 VCancelReservation_r(rvp);
4140 VCreateReservation_r(rvp);
4151 if (vp->specialStatus) {
4152 *ec = vp->specialStatus;
4157 if (vp->specialStatus) {
4158 *ec = vp->specialStatus;
4171 if (VIsSalvaging(vp) || (*ec == VSALVAGING)) {
4173 /* see CheckVnode() in afsfileprocs.c for an explanation
4174 * of this error code logic */
4175 afs_uint32 now = FT_ApproxTime();
4176 if ((vp->stats.last_salvage + (10 * 60)) >= now) {
4179 *client_ec = VRESTARTING;
4187 if (VIsErrorState(V_attachState(vp))) {
4188 /* make sure we don't take a vp in VOL_STATE_ERROR state and use
4189 * it, or transition it out of that state */
4198 * this test MUST happen after VAttachVolymeByVp, so we have no
4199 * conflicting vol op. (attach2 would have errored out if we had one;
4200 * specifically attach_check_vop must have detected a conflicting vop)
4202 opr_Assert(!vp->pending_vol_op || vp->pending_vol_op->vol_op_state == FSSYNC_VolOpRunningOnline);
4204 #endif /* AFS_DEMAND_ATTACH_FS */
4206 LoadVolumeHeader(ec, vp);
4209 /* Only log the error if it was a totally unexpected error. Simply
4210 * a missing inode is likely to be caused by the volume being deleted */
4211 if (errno != ENXIO || LogLevel)
4212 Log("Volume %" AFS_VOLID_FMT ": couldn't reread volume header\n",
4213 afs_printable_VolumeId_lu(vp->hashid));
4214 #ifdef AFS_DEMAND_ATTACH_FS
4215 if (VCanScheduleSalvage()) {
4216 VRequestSalvage_r(ec, vp, SALVSYNC_ERROR, 0 /*flags*/);
4221 #else /* AFS_DEMAND_ATTACH_FS */
4224 #endif /* AFS_DEMAND_ATTACH_FS */
4229 if (vp->shuttingDown) {
4236 if (programType == fileServer) {
4238 if (vp->goingOffline) {
4239 if (timeout && VTimedOut(timeout)) {
4240 /* we've timed out; don't wait for the vol */
4243 #ifdef AFS_DEMAND_ATTACH_FS
4244 /* wait for the volume to go offline */
4245 if (V_attachState(vp) == VOL_STATE_GOING_OFFLINE) {
4246 VTimedWaitStateChange_r(vp, timeout, NULL);
4248 #elif defined(AFS_PTHREAD_ENV)
4249 VOL_CV_TIMEDWAIT(&vol_put_volume_cond, timeout, NULL);
4250 #else /* AFS_PTHREAD_ENV */
4251 /* LWP has no timed wait, so the caller better not be
4253 opr_Assert(!timeout);
4254 LWP_WaitProcess(VPutVolume);
4255 #endif /* AFS_PTHREAD_ENV */
4259 if (vp->specialStatus) {
4261 *ec = vp->specialStatus;
4262 } else if (V_inService(vp) == 0 || V_blessed(vp) == 0) {
4265 } else if (V_inUse(vp) == 0 || vp->goingOffline) {
4276 #ifdef AFS_DEMAND_ATTACH_FS
4277 /* if no error, bump nUsers */
4280 VLRU_UpdateAccess_r(vp);
4283 VCancelReservation_r(rvp);
4286 if (client_ec && !*client_ec) {
4289 #else /* AFS_DEMAND_ATTACH_FS */
4290 /* if no error, bump nUsers */
4297 #endif /* AFS_DEMAND_ATTACH_FS */
4300 opr_Assert(vp || *ec);
4305 /***************************************************/
4306 /* Volume offline/detach routines */
4307 /***************************************************/
4309 /* caller MUST hold a heavyweight ref on vp */
4310 #ifdef AFS_DEMAND_ATTACH_FS
4312 VTakeOffline_r(Volume * vp)
4316 opr_Assert(vp->nUsers > 0);
4317 opr_Assert(programType == fileServer);
4319 VCreateReservation_r(vp);
4320 VWaitExclusiveState_r(vp);
4322 vp->goingOffline = 1;
4323 V_needsSalvaged(vp) = 1;
4325 VRequestSalvage_r(&error, vp, SALVSYNC_ERROR, 0);
4326 VCancelReservation_r(vp);
4328 #else /* AFS_DEMAND_ATTACH_FS */
4330 VTakeOffline_r(Volume * vp)
4332 opr_Assert(vp->nUsers > 0);
4333 opr_Assert(programType == fileServer);
4335 vp->goingOffline = 1;
4336 V_needsSalvaged(vp) = 1;
4338 #endif /* AFS_DEMAND_ATTACH_FS */
4341 VTakeOffline(Volume * vp)
4349 * force a volume offline.
4351 * @param[in] vp volume object pointer
4352 * @param[in] flags flags (see note below)
4354 * @note the flag VOL_FORCEOFF_NOUPDATE is a recursion control flag
4355 * used when VUpdateVolume_r needs to call VForceOffline_r
4356 * (which in turn would normally call VUpdateVolume_r)
4358 * @see VUpdateVolume_r
4360 * @pre VOL_LOCK must be held.
4361 * for DAFS, caller must hold ref.
4363 * @note for DAFS, it _is safe_ to call this function from an
4366 * @post needsSalvaged flag is set.
4367 * for DAFS, salvage is requested.
4368 * no further references to the volume through the volume
4369 * package will be honored.
4370 * all file descriptor and vnode caches are invalidated.
4372 * @warning this is a heavy-handed interface. it results in
4373 * a volume going offline regardless of the current
4374 * reference count state.
4376 * @internal volume package internal use only
4379 VForceOffline_r(Volume * vp, int flags)
4383 #ifdef AFS_DEMAND_ATTACH_FS
4384 VChangeState_r(vp, VOL_STATE_ERROR);
4389 strcpy(V_offlineMessage(vp),
4390 "Forced offline due to internal error: volume needs to be salvaged");
4391 Log("Volume %" AFS_VOLID_FMT " forced offline: it needs salvaging!\n", afs_printable_VolumeId_lu(V_id(vp)));
4394 vp->goingOffline = 0;
4395 V_needsSalvaged(vp) = 1;
4396 if (!(flags & VOL_FORCEOFF_NOUPDATE)) {
4397 VUpdateVolume_r(&error, vp, VOL_UPDATE_NOFORCEOFF);
4400 #ifdef AFS_DEMAND_ATTACH_FS
4401 VRequestSalvage_r(&error, vp, SALVSYNC_ERROR, 0 /*flags*/);
4402 #endif /* AFS_DEMAND_ATTACH_FS */
4404 #ifdef AFS_PTHREAD_ENV
4405 opr_cv_broadcast(&vol_put_volume_cond);
4406 #else /* AFS_PTHREAD_ENV */
4407 LWP_NoYieldSignal(VPutVolume);
4408 #endif /* AFS_PTHREAD_ENV */
4410 VReleaseVolumeHandles_r(vp);
4414 * force a volume offline.
4416 * @param[in] vp volume object pointer
4418 * @see VForceOffline_r
4421 VForceOffline(Volume * vp)
4424 VForceOffline_r(vp, 0);
4429 * Iterate over the RX calls associated with a volume, and interrupt them.
4431 * @param[in] vp The volume whose RX calls we want to scan
4433 * @pre VOL_LOCK held
4436 VScanCalls_r(struct Volume *vp)
4438 struct VCallByVol *cbv, *ncbv;
4440 #ifdef AFS_DEMAND_ATTACH_FS
4441 VolState state_save;
4444 if (queue_IsEmpty(&vp->rx_call_list))
4445 return; /* no calls to interrupt */
4446 if (!vol_opts.interrupt_rxcall)
4447 return; /* we have no function with which to interrupt calls */
4448 err = VIsGoingOffline_r(vp);
4450 return; /* we're not going offline anymore */
4452 #ifdef AFS_DEMAND_ATTACH_FS
4453 VWaitExclusiveState_r(vp);
4454 state_save = VChangeState_r(vp, VOL_STATE_SCANNING_RXCALLS);
4456 #endif /* AFS_DEMAND_ATTACH_FS */
4458 for(queue_Scan(&vp->rx_call_list, cbv, ncbv, VCallByVol)) {
4460 struct rx_peer *peer;
4462 peer = rx_PeerOf(rx_ConnectionOf(cbv->call));
4464 Log("Offlining volume %" AFS_VOLID_FMT " while client %s:%u is trying to read "
4465 "from it; kicking client off with error %ld\n",
4466 afs_printable_VolumeId_lu(vp->hashid),
4467 afs_inet_ntoa_r(rx_HostOf(peer), hoststr),
4468 (unsigned) ntohs(rx_PortOf(peer)),
4471 (*vol_opts.interrupt_rxcall) (cbv->call, err);
4474 #ifdef AFS_DEMAND_ATTACH_FS
4476 VChangeState_r(vp, state_save);
4477 #endif /* AFS_DEMAND_ATTACH_FS */
4480 #ifdef AFS_DEMAND_ATTACH_FS
4482 * Wait for a vp to go offline.
4484 * @param[out] ec 1 if a salvage on the volume has been requested and
4485 * salvok == 0, 0 otherwise
4486 * @param[in] vp The volume to wait for
4487 * @param[in] salvok If 0, we return immediately with *ec = 1 if the volume
4488 * has been requested to salvage. Otherwise we keep waiting
4489 * until the volume has gone offline.
4491 * @pre VOL_LOCK held
4492 * @pre caller holds a lightweight ref on vp
4497 VWaitForOfflineByVp_r(Error *ec, struct Volume *vp, int salvok)
4499 struct timespec timeout_ts;
4500 const struct timespec *ts;
4503 ts = VOfflineTimeout(&timeout_ts);
4507 while (!VIsOfflineState(V_attachState(vp)) && !timedout) {
4508 if (!salvok && vp->salvage.requested) {
4512 VTimedWaitStateChange_r(vp, ts, &timedout);
4515 /* we didn't time out, so the volume must be offline, so we're done */
4519 /* If we got here, we timed out waiting for the volume to go offline.
4520 * Kick off the accessing RX calls and wait again */
4524 while (!VIsOfflineState(V_attachState(vp))) {
4525 if (!salvok && vp->salvage.requested) {
4530 VWaitStateChange_r(vp);
4534 #else /* AFS_DEMAND_ATTACH_FS */
4537 * Wait for a volume to go offline.
4539 * @pre VOL_LOCK held
4541 * @note non-DAFS only (for DAFS, use @see WaitForOfflineByVp_r)
4544 VWaitForOffline_r(Error *ec, VolumeId volid)
4547 const struct timespec *ts;
4548 #ifdef AFS_PTHREAD_ENV
4549 struct timespec timeout_ts;
4552 ts = VOfflineTimeout(&timeout_ts);
4554 vp = GetVolume(ec, NULL, volid, NULL, ts);
4556 /* error occurred so bad that we can't even get a vp; we have no
4557 * information on the vol so we don't know whether to wait, so just
4561 if (!VIsGoingOffline_r(vp)) {
4562 /* volume is no longer going offline, so we're done */
4567 /* If we got here, we timed out waiting for the volume to go offline.
4568 * Kick off the accessing RX calls and wait again */
4574 vp = VGetVolume_r(ec, volid);
4576 /* In case it was reattached... */
4580 #endif /* !AFS_DEMAND_ATTACH_FS */
4582 /* The opposite of VAttachVolume. The volume header is written to disk, with
4583 the inUse bit turned off. A copy of the header is maintained in memory,
4584 however (which is why this is VOffline, not VDetach).
4587 VOffline_r(Volume * vp, char *message)
4590 #ifndef AFS_DEMAND_ATTACH_FS
4591 VolumeId vid = V_id(vp);
4594 opr_Assert(programType != volumeUtility && programType != volumeServer);
4599 if (V_offlineMessage(vp)[0] == '\0')
4600 strncpy(V_offlineMessage(vp), message, sizeof(V_offlineMessage(vp)));
4601 V_offlineMessage(vp)[sizeof(V_offlineMessage(vp)) - 1] = '\0';
4603 vp->goingOffline = 1;
4604 #ifdef AFS_DEMAND_ATTACH_FS
4605 VChangeState_r(vp, VOL_STATE_GOING_OFFLINE);
4606 VCreateReservation_r(vp);
4608 VWaitForOfflineByVp_r(&error, vp, 1);
4609 VCancelReservation_r(vp);
4610 #else /* AFS_DEMAND_ATTACH_FS */
4612 VWaitForOffline_r(&error, vid);
4613 #endif /* AFS_DEMAND_ATTACH_FS */
4616 #ifdef AFS_DEMAND_ATTACH_FS
4618 * Take a volume offline in order to perform a volume operation.
4620 * @param[inout] ec address in which to store error code
4621 * @param[in] vp volume object pointer
4622 * @param[in] message volume offline status message
4625 * - VOL_LOCK is held
4626 * - caller MUST hold a heavyweight ref on vp
4629 * - volume is taken offline
4630 * - if possible, volume operation is promoted to running state
4631 * - on failure, *ec is set to nonzero
4633 * @note Although this function does not return any value, it may
4634 * still fail to promote our pending volume operation to
4635 * a running state. Any caller MUST check the value of *ec,
4636 * and MUST NOT blindly assume success.
4638 * @warning if the caller does not hold a lightweight ref on vp,
4639 * then it MUST NOT reference vp after this function
4640 * returns to the caller.
4642 * @internal volume package internal use only
4645 VOfflineForVolOp_r(Error *ec, Volume *vp, char *message)
4648 opr_Assert(vp->pending_vol_op);
4654 if (V_offlineMessage(vp)[0] == '\0')
4655 strncpy(V_offlineMessage(vp), message, sizeof(V_offlineMessage(vp)));
4656 V_offlineMessage(vp)[sizeof(V_offlineMessage(vp)) - 1] = '\0';
4658 vp->goingOffline = 1;
4659 VChangeState_r(vp, VOL_STATE_GOING_OFFLINE);
4660 VCreateReservation_r(vp);
4663 if (vp->pending_vol_op->com.programType != salvageServer) {
4664 /* do not give corrupted volumes to the volserver */
4669 VWaitForOfflineByVp_r(ec, vp, salvok);
4671 VCancelReservation_r(vp);
4673 #endif /* AFS_DEMAND_ATTACH_FS */
4676 VOffline(Volume * vp, char *message)
4679 VOffline_r(vp, message);
4683 /* This gets used for the most part by utility routines that don't want
4684 * to keep all the volume headers around. Generally, the file server won't
4685 * call this routine, because then the offline message in the volume header
4686 * (or other information) won't be available to clients. For NAMEI, also
4687 * close the file handles. However, the fileserver does call this during
4688 * an attach following a volume operation.
4691 VDetachVolume_r(Error * ec, Volume * vp)
4693 #ifdef FSSYNC_BUILD_CLIENT
4695 struct DiskPartition64 *tpartp;
4696 int notifyServer = 0;
4697 int useDone = FSYNC_VOL_ON;
4699 if (VCanUseFSSYNC()) {
4700 notifyServer = vp->needsPutBack;
4701 if (V_destroyMe(vp) == DESTROY_ME)
4702 useDone = FSYNC_VOL_LEAVE_OFF;
4703 # ifdef AFS_DEMAND_ATTACH_FS
4704 else if (!V_blessed(vp) || !V_inService(vp))
4705 useDone = FSYNC_VOL_LEAVE_OFF;
4708 # ifdef AFS_DEMAND_ATTACH_FS
4709 if (V_needsSalvaged(vp)) {
4711 VRequestSalvage_r(ec, vp, SALVSYNC_NEEDED, 0);
4714 tpartp = vp->partition;
4716 #endif /* FSSYNC_BUILD_CLIENT */
4718 *ec = 0; /* always "succeeds" */
4719 DeleteVolumeFromHashTable(vp);
4720 vp->shuttingDown = 1;
4721 #ifdef AFS_DEMAND_ATTACH_FS
4722 DeleteVolumeFromVByPList_r(vp);
4724 VChangeState_r(vp, VOL_STATE_SHUTTING_DOWN);
4726 if (programType != fileServer)
4728 #endif /* AFS_DEMAND_ATTACH_FS */
4730 /* Will be detached sometime in the future--this is OK since volume is offline */
4732 /* XXX the following code should really be moved to VCheckDetach() since the volume
4733 * is not technically detached until the refcounts reach zero
4735 #ifdef FSSYNC_BUILD_CLIENT
4736 if (VCanUseFSSYNC() && notifyServer) {
4737 if (notifyServer == VOL_PUTBACK_DELETE) {
4738 /* Only send FSYNC_VOL_DONE if the volume was actually deleted.
4739 * volserver code will set needsPutBack to VOL_PUTBACK_DELETE
4740 * to signify a deleted volume. */
4741 useDone = FSYNC_VOL_DONE;
4744 * Note: The server is not notified in the case of a bogus volume
4745 * explicitly to make it possible to create a volume, do a partial
4746 * restore, then abort the operation without ever putting the volume
4747 * online. This is essential in the case of a volume move operation
4748 * between two partitions on the same server. In that case, there
4749 * would be two instances of the same volume, one of them bogus,
4750 * which the file server would attempt to put on line
4752 FSYNC_VolOp(volume, tpartp->name, useDone, 0, NULL);
4753 /* XXX this code path is only hit by volume utilities, thus
4754 * V_BreakVolumeCallbacks will always be NULL. if we really
4755 * want to break callbacks in this path we need to use FSYNC_VolOp() */
4757 /* Dettaching it so break all callbacks on it */
4758 if (V_BreakVolumeCallbacks) {
4759 Log("volume %u detached; breaking all call backs\n", volume);
4760 (*V_BreakVolumeCallbacks) (volume);
4764 #endif /* FSSYNC_BUILD_CLIENT */
4768 VDetachVolume(Error * ec, Volume * vp)
4771 VDetachVolume_r(ec, vp);
4776 /***************************************************/
4777 /* Volume fd/inode handle closing routines */
4778 /***************************************************/
4780 /* For VDetachVolume, we close all cached file descriptors, but keep
4781 * the Inode handles in case we need to read from a busy volume.
4783 /* for demand attach, caller MUST hold ref count on vp */
4785 VCloseVolumeHandles_r(Volume * vp)
4787 #ifdef AFS_DEMAND_ATTACH_FS
4788 VolState state_save;
4790 state_save = VChangeState_r(vp, VOL_STATE_OFFLINING);
4795 DFlushVolume(vp->hashid);
4797 #ifdef AFS_DEMAND_ATTACH_FS
4801 /* DAFS: VCloseVnodeFiles_r drops the glock internally */
4802 VCloseVnodeFiles_r(vp);
4804 #ifdef AFS_DEMAND_ATTACH_FS
4808 /* Too time consuming and unnecessary for the volserver */
4809 if (programType == fileServer) {
4810 IH_CONDSYNC(vp->vnodeIndex[vLarge].handle);
4811 IH_CONDSYNC(vp->vnodeIndex[vSmall].handle);
4812 IH_CONDSYNC(vp->diskDataHandle);
4813 #ifdef AFS_NAMEI_ENV
4814 IH_CONDSYNC(vp->linkHandle);
4815 #endif /* AFS_NAMEI_ENV */
4818 IH_REALLYCLOSE(vp->vnodeIndex[vLarge].handle);
4819 IH_REALLYCLOSE(vp->vnodeIndex[vSmall].handle);
4820 IH_REALLYCLOSE(vp->diskDataHandle);
4821 IH_REALLYCLOSE(vp->linkHandle);
4823 #ifdef AFS_DEMAND_ATTACH_FS
4824 if ((V_attachFlags(vp) & VOL_LOCKED)) {
4829 VChangeState_r(vp, state_save);
4833 /* For both VForceOffline and VOffline, we close all relevant handles.
4834 * For VOffline, if we re-attach the volume, the files may possible be
4835 * different than before.
4837 /* for demand attach, caller MUST hold a ref count on vp */
4839 VReleaseVolumeHandles_r(Volume * vp)
4841 #ifdef AFS_DEMAND_ATTACH_FS
4842 VolState state_save;
4844 state_save = VChangeState_r(vp, VOL_STATE_DETACHING);
4849 DFlushVolume(vp->hashid);
4851 #ifdef AFS_DEMAND_ATTACH_FS
4855 VReleaseVnodeFiles_r(vp); /* DAFS: releases the glock internally */
4857 #ifdef AFS_DEMAND_ATTACH_FS
4861 /* Too time consuming and unnecessary for the volserver */
4862 if (programType == fileServer) {
4863 IH_CONDSYNC(vp->vnodeIndex[vLarge].handle);
4864 IH_CONDSYNC(vp->vnodeIndex[vSmall].handle);
4865 IH_CONDSYNC(vp->diskDataHandle);
4866 #ifdef AFS_NAMEI_ENV
4867 IH_CONDSYNC(vp->linkHandle);
4868 #endif /* AFS_NAMEI_ENV */
4871 IH_RELEASE(vp->vnodeIndex[vLarge].handle);
4872 IH_RELEASE(vp->vnodeIndex[vSmall].handle);
4873 IH_RELEASE(vp->diskDataHandle);
4874 IH_RELEASE(vp->linkHandle);
4876 #ifdef AFS_DEMAND_ATTACH_FS
4877 if ((V_attachFlags(vp) & VOL_LOCKED)) {
4882 VChangeState_r(vp, state_save);
4887 /***************************************************/
4888 /* Volume write and fsync routines */
4889 /***************************************************/
4892 VUpdateVolume_r(Error * ec, Volume * vp, int flags)
4894 #ifdef AFS_DEMAND_ATTACH_FS
4895 VolState state_save;
4897 if (flags & VOL_UPDATE_WAIT) {
4898 VCreateReservation_r(vp);
4899 VWaitExclusiveState_r(vp);
4904 if (programType == fileServer)
4906 (V_inUse(vp) ? V_nextVnodeUnique(vp) +
4907 200 : V_nextVnodeUnique(vp));
4909 #ifdef AFS_DEMAND_ATTACH_FS
4910 state_save = VChangeState_r(vp, VOL_STATE_UPDATING);
4914 WriteVolumeHeader_r(ec, vp);
4916 #ifdef AFS_DEMAND_ATTACH_FS
4918 VChangeState_r(vp, state_save);
4919 if (flags & VOL_UPDATE_WAIT) {
4920 VCancelReservation_r(vp);
4925 Log("VUpdateVolume: error updating volume header, volume %" AFS_VOLID_FMT " (%s)\n",
4926 afs_printable_VolumeId_lu(V_id(vp)), V_name(vp));
4927 /* try to update on-disk header,
4928 * while preventing infinite recursion */
4929 if (!(flags & VOL_UPDATE_NOFORCEOFF)) {
4930 VForceOffline_r(vp, VOL_FORCEOFF_NOUPDATE);
4936 VUpdateVolume(Error * ec, Volume * vp)
4939 VUpdateVolume_r(ec, vp, VOL_UPDATE_WAIT);
4944 VSyncVolume_r(Error * ec, Volume * vp, int flags)
4948 #ifdef AFS_DEMAND_ATTACH_FS
4949 VolState state_save;
4952 if (flags & VOL_SYNC_WAIT) {
4953 VUpdateVolume_r(ec, vp, VOL_UPDATE_WAIT);
4955 VUpdateVolume_r(ec, vp, 0);
4958 #ifdef AFS_DEMAND_ATTACH_FS
4959 state_save = VChangeState_r(vp, VOL_STATE_UPDATING);
4962 fdP = IH_OPEN(V_diskDataHandle(vp));
4963 opr_Assert(fdP != NULL);
4964 code = FDH_SYNC(fdP);
4965 opr_Assert(code == 0);
4967 #ifdef AFS_DEMAND_ATTACH_FS
4969 VChangeState_r(vp, state_save);
4975 VSyncVolume(Error * ec, Volume * vp)
4978 VSyncVolume_r(ec, vp, VOL_SYNC_WAIT);
4983 /***************************************************/
4984 /* Volume dealloaction routines */
4985 /***************************************************/
4987 #ifdef AFS_DEMAND_ATTACH_FS
4989 FreeVolume(Volume * vp)
4991 /* free the heap space, iff it's safe.
4992 * otherwise, pull it out of the hash table, so it
4993 * will get deallocated when all refs to it go away */
4994 if (!VCheckFree(vp)) {
4995 DeleteVolumeFromHashTable(vp);
4996 DeleteVolumeFromVByPList_r(vp);
4998 /* make sure we invalidate the header cache entry */
4999 FreeVolumeHeader(vp);
5002 #endif /* AFS_DEMAND_ATTACH_FS */
5005 ReallyFreeVolume(Volume * vp)
5010 #ifdef AFS_DEMAND_ATTACH_FS
5012 VChangeState_r(vp, VOL_STATE_FREED);
5013 if (vp->pending_vol_op)
5014 free(vp->pending_vol_op);
5015 #endif /* AFS_DEMAND_ATTACH_FS */
5016 for (i = 0; i < nVNODECLASSES; i++)
5017 if (vp->vnodeIndex[i].bitmap)
5018 free(vp->vnodeIndex[i].bitmap);
5019 FreeVolumeHeader(vp);
5020 #ifndef AFS_DEMAND_ATTACH_FS
5021 DeleteVolumeFromHashTable(vp);
5022 #endif /* AFS_DEMAND_ATTACH_FS */
5026 /* check to see if we should shutdown this volume
5027 * returns 1 if volume was freed, 0 otherwise */
5028 #ifdef AFS_DEMAND_ATTACH_FS
5030 VCheckDetach(Volume * vp)
5035 if (vp->nUsers || vp->nWaiters)
5038 if (vp->shuttingDown) {
5040 if ((programType != fileServer) &&
5041 (V_inUse(vp) == programType) &&
5042 ((V_checkoutMode(vp) == V_VOLUPD) ||
5043 (V_checkoutMode(vp) == V_SECRETLY) ||
5044 ((V_checkoutMode(vp) == V_CLONE) &&
5045 (VolumeWriteable(vp))))) {
5047 VUpdateVolume_r(&ec, vp, VOL_UPDATE_NOFORCEOFF);
5049 Log("VCheckDetach: volume header update for volume %" AFS_VOLID_FMT " "
5050 "failed with errno %d\n", afs_printable_VolumeId_lu(vp->hashid), errno);
5053 VReleaseVolumeHandles_r(vp);
5055 ReallyFreeVolume(vp);
5056 if (programType == fileServer) {
5057 opr_cv_broadcast(&vol_put_volume_cond);
5062 #else /* AFS_DEMAND_ATTACH_FS */
5064 VCheckDetach(Volume * vp)
5072 if (vp->shuttingDown) {
5074 if ((programType != fileServer) &&
5075 (V_inUse(vp) == programType) &&
5076 ((V_checkoutMode(vp) == V_VOLUPD) ||
5077 (V_checkoutMode(vp) == V_SECRETLY) ||
5078 ((V_checkoutMode(vp) == V_CLONE) &&
5079 (VolumeWriteable(vp))))) {
5081 VUpdateVolume_r(&ec, vp, VOL_UPDATE_NOFORCEOFF);
5083 Log("VCheckDetach: volume header update for volume %" AFS_VOLID_FMT " failed with errno %d\n",
5084 afs_printable_VolumeId_lu(vp->hashid), errno);
5087 VReleaseVolumeHandles_r(vp);
5088 ReallyFreeVolume(vp);
5089 if (programType == fileServer) {
5090 #if defined(AFS_PTHREAD_ENV)
5091 opr_cv_broadcast(&vol_put_volume_cond);
5092 #else /* AFS_PTHREAD_ENV */
5093 LWP_NoYieldSignal(VPutVolume);
5094 #endif /* AFS_PTHREAD_ENV */
5099 #endif /* AFS_DEMAND_ATTACH_FS */
5101 /* check to see if we should offline this volume
5102 * return 1 if volume went offline, 0 otherwise */
5103 #ifdef AFS_DEMAND_ATTACH_FS
5105 VCheckOffline(Volume * vp)
5109 if (vp->goingOffline && !vp->nUsers) {
5111 opr_Assert(programType == fileServer);
5112 opr_Assert((V_attachState(vp) != VOL_STATE_ATTACHED) &&
5113 (V_attachState(vp) != VOL_STATE_FREED) &&
5114 (V_attachState(vp) != VOL_STATE_PREATTACHED) &&
5115 (V_attachState(vp) != VOL_STATE_UNATTACHED) &&
5116 (V_attachState(vp) != VOL_STATE_DELETED));
5120 * VOL_STATE_GOING_OFFLINE
5121 * VOL_STATE_SHUTTING_DOWN
5122 * VIsErrorState(V_attachState(vp))
5123 * VIsExclusiveState(V_attachState(vp))
5126 VCreateReservation_r(vp);
5127 VChangeState_r(vp, VOL_STATE_OFFLINING);
5130 /* must clear the goingOffline flag before we drop the glock */
5131 vp->goingOffline = 0;
5136 /* perform async operations */
5137 VUpdateVolume_r(&error, vp, 0);
5138 VCloseVolumeHandles_r(vp);
5141 if (V_offlineMessage(vp)[0]) {
5142 Log("VOffline: Volume %lu (%s) is now offline (%s)\n",
5143 afs_printable_uint32_lu(V_id(vp)), V_name(vp),
5144 V_offlineMessage(vp));
5146 Log("VOffline: Volume %lu (%s) is now offline\n",
5147 afs_printable_uint32_lu(V_id(vp)), V_name(vp));
5151 /* invalidate the volume header cache entry */
5152 FreeVolumeHeader(vp);
5154 /* if nothing changed state to error or salvaging,
5155 * drop state to unattached */
5156 if (!VIsErrorState(V_attachState(vp))) {
5157 VChangeState_r(vp, VOL_STATE_UNATTACHED);
5159 VCancelReservation_r(vp);
5160 /* no usage of vp is safe beyond this point */
5164 #else /* AFS_DEMAND_ATTACH_FS */
5166 VCheckOffline(Volume * vp)
5170 if (vp->goingOffline && !vp->nUsers) {
5172 opr_Assert(programType == fileServer);
5175 vp->goingOffline = 0;
5177 VUpdateVolume_r(&error, vp, 0);
5178 VCloseVolumeHandles_r(vp);
5180 if (V_offlineMessage(vp)[0]) {
5181 Log("VOffline: Volume %lu (%s) is now offline (%s)\n",
5182 afs_printable_uint32_lu(V_id(vp)), V_name(vp),
5183 V_offlineMessage(vp));
5185 Log("VOffline: Volume %lu (%s) is now offline\n",
5186 afs_printable_uint32_lu(V_id(vp)), V_name(vp));
5189 FreeVolumeHeader(vp);
5190 #ifdef AFS_PTHREAD_ENV
5191 opr_cv_broadcast(&vol_put_volume_cond);
5192 #else /* AFS_PTHREAD_ENV */
5193 LWP_NoYieldSignal(VPutVolume);
5194 #endif /* AFS_PTHREAD_ENV */
5198 #endif /* AFS_DEMAND_ATTACH_FS */
5200 /***************************************************/
5201 /* demand attach fs ref counting routines */
5202 /***************************************************/
5204 #ifdef AFS_DEMAND_ATTACH_FS
5205 /* the following two functions handle reference counting for
5206 * asynchronous operations on volume structs.
5208 * their purpose is to prevent a VDetachVolume or VShutdown
5209 * from free()ing the Volume struct during an async i/o op */
5211 /* register with the async volume op ref counter */
5212 /* VCreateReservation_r moved into inline code header because it
5213 * is now needed in vnode.c -- tkeiser 11/20/2007
5217 * decrement volume-package internal refcount.
5219 * @param vp volume object pointer
5221 * @internal volume package internal use only
5224 * @arg VOL_LOCK is held
5225 * @arg lightweight refcount held
5227 * @post volume waiters refcount is decremented; volume may
5228 * have been deallocated/shutdown/offlined/salvaged/
5229 * whatever during the process
5231 * @warning once you have tossed your last reference (you can acquire
5232 * lightweight refs recursively) it is NOT SAFE to reference
5233 * a volume object pointer ever again
5235 * @see VCreateReservation_r
5237 * @note DEMAND_ATTACH_FS only
5240 VCancelReservation_r(Volume * vp)
5242 opr_Verify(--vp->nWaiters >= 0);
5243 if (vp->nWaiters == 0) {
5245 if (!VCheckDetach(vp)) {
5252 /* check to see if we should free this volume now
5253 * return 1 if volume was freed, 0 otherwise */
5255 VCheckFree(Volume * vp)
5258 if ((vp->nUsers == 0) &&
5259 (vp->nWaiters == 0) &&
5260 !(V_attachFlags(vp) & (VOL_IN_HASH |
5264 ReallyFreeVolume(vp);
5269 #endif /* AFS_DEMAND_ATTACH_FS */
5272 /***************************************************/
5273 /* online volume operations routines */
5274 /***************************************************/
5276 #ifdef AFS_DEMAND_ATTACH_FS
5278 * register a volume operation on a given volume.
5280 * @param[in] vp volume object
5281 * @param[in] vopinfo volume operation info object
5283 * @pre VOL_LOCK is held
5285 * @post volume operation info object attached to volume object.
5286 * volume operation statistics updated.
5288 * @note by "attached" we mean a copy of the passed in object is made
5290 * @internal volume package internal use only
5293 VRegisterVolOp_r(Volume * vp, FSSYNC_VolOp_info * vopinfo)
5295 FSSYNC_VolOp_info * info;
5297 /* attach a vol op info node to the volume struct */
5298 info = malloc(sizeof(FSSYNC_VolOp_info));
5299 opr_Assert(info != NULL);
5300 memcpy(info, vopinfo, sizeof(FSSYNC_VolOp_info));
5301 vp->pending_vol_op = info;
5304 vp->stats.last_vol_op = FT_ApproxTime();
5305 vp->stats.vol_ops++;
5306 IncUInt64(&VStats.vol_ops);
5312 * deregister the volume operation attached to this volume.
5314 * @param[in] vp volume object pointer
5316 * @pre VOL_LOCK is held
5318 * @post the volume operation info object is detached from the volume object
5320 * @internal volume package internal use only
5323 VDeregisterVolOp_r(Volume * vp)
5325 if (vp->pending_vol_op) {
5326 free(vp->pending_vol_op);
5327 vp->pending_vol_op = NULL;
5331 #endif /* AFS_DEMAND_ATTACH_FS */
5334 * determine whether it is safe to leave a volume online during
5335 * the volume operation described by the vopinfo object.
5337 * @param[in] vp volume object
5338 * @param[in] vopinfo volume operation info object
5340 * @return whether it is safe to leave volume online
5341 * @retval 0 it is NOT SAFE to leave the volume online
5342 * @retval 1 it is safe to leave the volume online during the operation
5345 * @arg VOL_LOCK is held
5346 * @arg disk header attached to vp (heavyweight ref on vp will guarantee
5347 * this condition is met)
5349 * @internal volume package internal use only
5352 VVolOpLeaveOnline_r(Volume * vp, FSSYNC_VolOp_info * vopinfo)
5354 return (vopinfo->vol_op_state == FSSYNC_VolOpRunningOnline ||
5355 (vopinfo->com.command == FSYNC_VOL_NEEDVOLUME &&
5356 (vopinfo->com.reason == V_READONLY ||
5357 (!VolumeWriteable(vp) &&
5358 (vopinfo->com.reason == V_CLONE ||
5359 vopinfo->com.reason == V_DUMP)))));
5363 * same as VVolOpLeaveOnline_r, but does not require a volume with an attached
5366 * @param[in] vp volume object
5367 * @param[in] vopinfo volume operation info object
5369 * @return whether it is safe to leave volume online
5370 * @retval 0 it is NOT SAFE to leave the volume online
5371 * @retval 1 it is safe to leave the volume online during the operation
5372 * @retval -1 unsure; volume header is required in order to know whether or
5373 * not is is safe to leave the volume online
5375 * @pre VOL_LOCK is held
5377 * @internal volume package internal use only
5380 VVolOpLeaveOnlineNoHeader_r(Volume * vp, FSSYNC_VolOp_info * vopinfo)
5382 /* follow the logic in VVolOpLeaveOnline_r; this is the same, except
5383 * assume that we don't know VolumeWriteable; return -1 if the answer
5384 * depends on VolumeWriteable */
5386 if (vopinfo->vol_op_state == FSSYNC_VolOpRunningOnline) {
5389 if (vopinfo->com.command == FSYNC_VOL_NEEDVOLUME &&
5390 vopinfo->com.reason == V_READONLY) {
5394 if (vopinfo->com.command == FSYNC_VOL_NEEDVOLUME &&
5395 (vopinfo->com.reason == V_CLONE ||
5396 vopinfo->com.reason == V_DUMP)) {
5398 /* must know VolumeWriteable */
5405 * determine whether VBUSY should be set during this volume operation.
5407 * @param[in] vp volume object
5408 * @param[in] vopinfo volume operation info object
5410 * @return whether VBUSY should be set
5411 * @retval 0 VBUSY does NOT need to be set
5412 * @retval 1 VBUSY SHOULD be set
5414 * @pre VOL_LOCK is held
5416 * @internal volume package internal use only
5419 VVolOpSetVBusy_r(Volume * vp, FSSYNC_VolOp_info * vopinfo)
5421 return ((vopinfo->com.command == FSYNC_VOL_OFF &&
5422 vopinfo->com.reason == FSYNC_SALVAGE) ||
5423 (vopinfo->com.command == FSYNC_VOL_NEEDVOLUME &&
5424 (vopinfo->com.reason == V_CLONE ||
5425 vopinfo->com.reason == V_DUMP)));
5429 /***************************************************/
5430 /* online salvager routines */
5431 /***************************************************/
5432 #if defined(AFS_DEMAND_ATTACH_FS)
5435 * offline a volume to let it be salvaged.
5437 * @param[in] vp Volume to offline
5439 * @return whether we offlined the volume successfully
5440 * @retval 0 volume was not offlined
5441 * @retval 1 volume is now offline
5443 * @note This is similar to VCheckOffline, but slightly different. We do not
5444 * deal with vp->goingOffline, and we try to avoid touching the volume
5445 * header except just to set needsSalvaged
5447 * @pre VOL_LOCK held
5448 * @pre vp->nUsers == 0
5449 * @pre V_attachState(vp) == VOL_STATE_SALVAGE_REQ
5452 VOfflineForSalvage_r(struct Volume *vp)
5456 VCreateReservation_r(vp);
5457 VWaitExclusiveState_r(vp);
5459 if (vp->nUsers || V_attachState(vp) == VOL_STATE_SALVAGING) {
5460 /* Someone's using the volume, or someone got to scheduling the salvage
5461 * before us. I don't think either of these should be possible, as we
5462 * should gain no new heavyweight references while we're trying to
5463 * salvage, but just to be sure... */
5464 VCancelReservation_r(vp);
5468 VChangeState_r(vp, VOL_STATE_OFFLINING);
5472 V_needsSalvaged(vp) = 1;
5473 /* ignore error; updating needsSalvaged is just best effort */
5474 VUpdateVolume_r(&error, vp, VOL_UPDATE_NOFORCEOFF);
5476 VCloseVolumeHandles_r(vp);
5478 FreeVolumeHeader(vp);
5480 /* volume has been effectively offlined; we can mark it in the SALVAGING
5481 * state now, which lets FSSYNC give it away */
5482 VChangeState_r(vp, VOL_STATE_SALVAGING);
5484 VCancelReservation_r(vp);
5490 * check whether a salvage needs to be performed on this volume.
5492 * @param[in] vp pointer to volume object
5494 * @return status code
5495 * @retval VCHECK_SALVAGE_OK (0) no pending salvage
5496 * @retval VCHECK_SALVAGE_SCHEDULED (1) salvage has been scheduled
5497 * @retval VCHECK_SALVAGE_ASYNC (2) salvage being scheduled
5498 * @retval VCHECK_SALVAGE_DENIED (3) salvage not scheduled; denied
5499 * @retval VCHECK_SALVAGE_FAIL (4) salvage not scheduled; failed
5501 * @pre VOL_LOCK is held
5503 * @post if salvage request flag is set and nUsers and nWaiters are zero,
5504 * then a salvage will be requested
5506 * @note this is one of the event handlers called by VCancelReservation_r
5508 * @note the caller must check if the volume needs to be freed after calling
5509 * this; the volume may not have any references or be on any lists after
5510 * we return, and we do not free it
5512 * @see VCancelReservation_r
5514 * @internal volume package internal use only.
5517 VCheckSalvage(Volume * vp)
5519 int ret = VCHECK_SALVAGE_OK;
5521 #if defined(SALVSYNC_BUILD_CLIENT) || defined(FSSYNC_BUILD_CLIENT)
5522 if (!vp->salvage.requested) {
5523 return VCHECK_SALVAGE_OK;
5526 return VCHECK_SALVAGE_ASYNC;
5529 /* prevent recursion; some of the code below creates and removes
5530 * lightweight refs, which can call VCheckSalvage */
5531 if (vp->salvage.scheduling) {
5532 return VCHECK_SALVAGE_ASYNC;
5534 vp->salvage.scheduling = 1;
5536 if (V_attachState(vp) == VOL_STATE_SALVAGE_REQ) {
5537 if (!VOfflineForSalvage_r(vp)) {
5538 vp->salvage.scheduling = 0;
5539 return VCHECK_SALVAGE_FAIL;
5543 if (vp->salvage.requested) {
5544 ret = VScheduleSalvage_r(vp);
5546 vp->salvage.scheduling = 0;
5547 #endif /* SALVSYNC_BUILD_CLIENT || FSSYNC_BUILD_CLIENT */
5552 * request volume salvage.
5554 * @param[out] ec computed client error code
5555 * @param[in] vp volume object pointer
5556 * @param[in] reason reason code (passed to salvageserver via SALVSYNC)
5557 * @param[in] flags see flags note below
5560 * VOL_SALVAGE_NO_OFFLINE do not need to wait to offline the volume; it has
5561 * not been fully attached
5563 * @pre VOL_LOCK is held.
5565 * @post volume state is changed.
5566 * for fileserver, salvage will be requested once refcount reaches zero.
5568 * @return operation status code
5569 * @retval 0 volume salvage will occur
5570 * @retval 1 volume salvage could not be scheduled
5574 * @note in the fileserver, this call does not synchronously schedule a volume
5575 * salvage. rather, it sets volume state so that when volume refcounts
5576 * reach zero, a volume salvage will occur. by "refcounts", we mean both
5577 * nUsers and nWaiters must be zero.
5579 * @internal volume package internal use only.
5582 VRequestSalvage_r(Error * ec, Volume * vp, int reason, int flags)
5586 * for DAFS volume utilities that are not supposed to schedule salvages,
5587 * just transition to error state instead
5589 if (!VCanScheduleSalvage()) {
5590 VChangeState_r(vp, VOL_STATE_ERROR);
5595 if (programType != fileServer && !VCanUseFSSYNC()) {
5596 VChangeState_r(vp, VOL_STATE_ERROR);
5601 if (!vp->salvage.requested) {
5602 vp->salvage.requested = 1;
5603 vp->salvage.reason = reason;
5604 vp->stats.last_salvage = FT_ApproxTime();
5606 /* Note that it is not possible for us to reach this point if a
5607 * salvage is already running on this volume (even if the fileserver
5608 * was restarted during the salvage). If a salvage were running, the
5609 * salvager would have write-locked the volume header file, so when
5610 * we tried to lock the volume header, the lock would have failed,
5611 * and we would have failed during attachment prior to calling
5612 * VRequestSalvage. So we know that we can schedule salvages without
5613 * fear of a salvage already running for this volume. */
5615 if (vp->stats.salvages < SALVAGE_COUNT_MAX) {
5617 /* if we don't need to offline the volume, we can go directly
5618 * to SALVAGING. SALVAGING says the volume is offline and is
5619 * either salvaging or ready to be handed to the salvager.
5620 * SALVAGE_REQ says that we want to salvage the volume, but we
5621 * are waiting for it to go offline first. */
5622 if (flags & VOL_SALVAGE_NO_OFFLINE) {
5623 VChangeState_r(vp, VOL_STATE_SALVAGING);
5625 VChangeState_r(vp, VOL_STATE_SALVAGE_REQ);
5626 if (vp->nUsers == 0) {
5627 /* normally VOfflineForSalvage_r would be called from
5628 * PutVolume et al when nUsers reaches 0, but if
5629 * it's already 0, just do it ourselves, since PutVolume
5630 * isn't going to get called */
5631 VOfflineForSalvage_r(vp);
5634 /* If we are non-fileserver, we're telling the fileserver to
5635 * salvage the vol, so we don't need to give it back separately. */
5636 vp->needsPutBack = 0;
5640 Log("VRequestSalvage: volume %" AFS_VOLID_FMT " online salvaged too many times; forced offline.\n", afs_printable_VolumeId_lu(vp->hashid));
5642 /* make sure neither VScheduleSalvage_r nor
5643 * VUpdateSalvagePriority_r try to schedule another salvage */
5644 vp->salvage.requested = vp->salvage.scheduled = 0;
5646 VChangeState_r(vp, VOL_STATE_ERROR);
5655 * update salvageserver scheduling priority for a volume.
5657 * @param[in] vp pointer to volume object
5659 * @return operation status
5661 * @retval 1 request denied, or SALVSYNC communications failure
5663 * @pre VOL_LOCK is held.
5665 * @post in-core salvage priority counter is incremented. if at least
5666 * SALVAGE_PRIO_UPDATE_INTERVAL seconds have elapsed since the
5667 * last SALVSYNC_RAISEPRIO request, we contact the salvageserver
5668 * to update its priority queue. if no salvage is scheduled,
5669 * this function is a no-op.
5671 * @note DAFS fileserver only
5673 * @note this should be called whenever a VGetVolume fails due to a
5674 * pending salvage request
5676 * @todo should set exclusive state and drop glock around salvsync call
5678 * @internal volume package internal use only.
5681 VUpdateSalvagePriority_r(Volume * vp)
5685 #ifdef SALVSYNC_BUILD_CLIENT
5690 now = FT_ApproxTime();
5692 /* update the salvageserver priority queue occasionally so that
5693 * frequently requested volumes get moved to the head of the queue
5695 if ((vp->salvage.scheduled) &&
5696 (vp->stats.last_salvage_req < (now-SALVAGE_PRIO_UPDATE_INTERVAL))) {
5697 code = SALVSYNC_SalvageVolume(vp->hashid,
5698 VPartitionPath(vp->partition),
5703 vp->stats.last_salvage_req = now;
5704 if (code != SYNC_OK) {
5708 #endif /* SALVSYNC_BUILD_CLIENT */
5713 #if defined(SALVSYNC_BUILD_CLIENT) || defined(FSSYNC_BUILD_CLIENT)
5715 /* A couple of little helper functions. These return true if we tried to
5716 * use this mechanism to schedule a salvage, false if we haven't tried.
5717 * If we did try a salvage then the results are contained in code.
5721 try_SALVSYNC(Volume *vp, char *partName, int *code) {
5722 #ifdef SALVSYNC_BUILD_CLIENT
5723 if (VCanUseSALVSYNC()) {
5724 Log("Scheduling salvage for volume %" AFS_VOLID_FMT " on part %s over SALVSYNC\n",
5725 afs_printable_VolumeId_lu(vp->hashid), partName);
5727 /* can't use V_id() since there's no guarantee
5728 * we have the disk data header at this point */
5729 *code = SALVSYNC_SalvageVolume(vp->hashid,
5742 try_FSSYNC(Volume *vp, char *partName, int *code) {
5743 #ifdef FSSYNC_BUILD_CLIENT
5744 if (VCanUseFSSYNC()) {
5745 Log("Scheduling salvage for volume %" AFS_VOLID_FMT " on part %s over FSSYNC\n",
5746 afs_printable_VolumeId_lu(vp->hashid), partName);
5749 * If we aren't the fileserver, tell the fileserver the volume
5750 * needs to be salvaged. We could directly tell the
5751 * salvageserver, but the fileserver keeps track of some stats
5752 * related to salvages, and handles some other salvage-related
5753 * complications for us.
5755 *code = FSYNC_VolOp(vp->hashid, partName,
5756 FSYNC_VOL_FORCE_ERROR, FSYNC_SALVAGE, NULL);
5759 #endif /* FSSYNC_BUILD_CLIENT */
5764 * schedule a salvage with the salvage server or fileserver.
5766 * @param[in] vp pointer to volume object
5768 * @return operation status
5769 * @retval VCHECK_SALVAGE_OK (0) no pending salvage
5770 * @retval VCHECK_SALVAGE_SCHEDULED (1) salvage has been scheduled
5771 * @retval VCHECK_SALVAGE_ASYNC (2) salvage being scheduled
5772 * @retval VCHECK_SALVAGE_DENIED (3) salvage not scheduled; denied
5773 * @retval VCHECK_SALVAGE_FAIL (4) salvage not scheduled; failed
5776 * @arg VOL_LOCK is held.
5777 * @arg nUsers and nWaiters should be zero.
5779 * @post salvageserver or fileserver is sent a salvage request
5781 * @note If we are the fileserver, the request will be sent to the salvage
5782 * server over SALVSYNC. If we are not the fileserver, the request will be
5783 * sent to the fileserver over FSSYNC (FSYNC_VOL_FORCE_ERROR/FSYNC_SALVAGE).
5785 * @note the caller must check if the volume needs to be freed after calling
5786 * this; the volume may not have any references or be on any lists after
5787 * we return, and we do not free it
5791 * @internal volume package internal use only.
5794 VScheduleSalvage_r(Volume * vp)
5796 int ret = VCHECK_SALVAGE_SCHEDULED;
5798 VolState state_save;
5799 VThreadOptions_t * thread_opts;
5802 opr_Verify(VCanUseSALVSYNC() || VCanUseFSSYNC());
5804 if (vp->nWaiters || vp->nUsers) {
5805 return VCHECK_SALVAGE_ASYNC;
5808 /* prevent endless salvage,attach,salvage,attach,... loops */
5809 if (vp->stats.salvages >= SALVAGE_COUNT_MAX) {
5810 return VCHECK_SALVAGE_FAIL;
5814 * don't perform salvsync ops on certain threads
5816 thread_opts = pthread_getspecific(VThread_key);
5817 if (thread_opts == NULL) {
5818 thread_opts = &VThread_defaults;
5820 if (thread_opts->disallow_salvsync || vol_disallow_salvsync) {
5821 return VCHECK_SALVAGE_ASYNC;
5824 if (vp->salvage.scheduled) {
5825 return VCHECK_SALVAGE_SCHEDULED;
5828 VCreateReservation_r(vp);
5829 VWaitExclusiveState_r(vp);
5832 * XXX the scheduling process should really be done asynchronously
5833 * to avoid fssync deadlocks
5835 if (vp->salvage.scheduled) {
5836 ret = VCHECK_SALVAGE_SCHEDULED;
5838 /* if we haven't previously scheduled a salvage, do so now
5840 * set the volume to an exclusive state and drop the lock
5841 * around the SALVSYNC call
5843 strlcpy(partName, vp->partition->name, sizeof(partName));
5844 state_save = VChangeState_r(vp, VOL_STATE_SALVSYNC_REQ);
5847 opr_Verify(try_SALVSYNC(vp, partName, &code)
5848 || try_FSSYNC(vp, partName, &code));
5851 VChangeState_r(vp, state_save);
5853 if (code == SYNC_OK) {
5854 ret = VCHECK_SALVAGE_SCHEDULED;
5855 vp->salvage.scheduled = 1;
5856 vp->stats.last_salvage_req = FT_ApproxTime();
5857 if (VCanUseSALVSYNC()) {
5858 /* don't record these stats for non-fileservers; let the
5859 * fileserver take care of these */
5860 vp->stats.salvages++;
5861 IncUInt64(&VStats.salvages);
5865 case SYNC_BAD_COMMAND:
5866 case SYNC_COM_ERROR:
5867 ret = VCHECK_SALVAGE_FAIL;
5870 ret = VCHECK_SALVAGE_DENIED;
5871 Log("VScheduleSalvage_r: Salvage request for volume %" AFS_VOLID_FMT " "
5872 "denied\n", afs_printable_VolumeId_lu(vp->hashid));
5875 ret = VCHECK_SALVAGE_FAIL;
5876 Log("VScheduleSalvage_r: Salvage request for volume %" AFS_VOLID_FMT " "
5877 "failed\n", afs_printable_VolumeId_lu(vp->hashid));
5880 ret = VCHECK_SALVAGE_FAIL;
5881 Log("VScheduleSalvage_r: Salvage request for volume %" AFS_VOLID_FMT " "
5882 "received unknown protocol error %d\n",
5883 afs_printable_VolumeId_lu(vp->hashid), code);
5887 if (VCanUseFSSYNC()) {
5888 VChangeState_r(vp, VOL_STATE_ERROR);
5893 /* NB: this is cancelling the reservation we obtained above, but we do
5894 * not call VCancelReservation_r, since that may trigger the vp dtor,
5895 * possibly free'ing the vp. We need to keep the vp around after
5896 * this, as the caller may reference vp without any refs. Instead, it
5897 * is the duty of the caller to inspect 'vp' after we return to see if
5898 * needs to be freed. */
5899 opr_Verify(--vp->nWaiters >= 0);
5902 #endif /* SALVSYNC_BUILD_CLIENT || FSSYNC_BUILD_CLIENT */
5904 #ifdef SALVSYNC_BUILD_CLIENT
5907 * connect to the salvageserver SYNC service.
5909 * @return operation status
5913 * @post connection to salvageserver SYNC service established
5915 * @see VConnectSALV_r
5916 * @see VDisconnectSALV
5917 * @see VReconnectSALV
5924 retVal = VConnectSALV_r();
5930 * connect to the salvageserver SYNC service.
5932 * @return operation status
5936 * @pre VOL_LOCK is held.
5938 * @post connection to salvageserver SYNC service established
5941 * @see VDisconnectSALV_r
5942 * @see VReconnectSALV_r
5943 * @see SALVSYNC_clientInit
5945 * @internal volume package internal use only.
5948 VConnectSALV_r(void)
5950 return SALVSYNC_clientInit();
5954 * disconnect from the salvageserver SYNC service.
5956 * @return operation status
5959 * @pre client should have a live connection to the salvageserver
5961 * @post connection to salvageserver SYNC service destroyed
5963 * @see VDisconnectSALV_r
5965 * @see VReconnectSALV
5968 VDisconnectSALV(void)
5971 VDisconnectSALV_r();
5977 * disconnect from the salvageserver SYNC service.
5979 * @return operation status
5983 * @arg VOL_LOCK is held.
5984 * @arg client should have a live connection to the salvageserver.
5986 * @post connection to salvageserver SYNC service destroyed
5988 * @see VDisconnectSALV
5989 * @see VConnectSALV_r
5990 * @see VReconnectSALV_r
5991 * @see SALVSYNC_clientFinis
5993 * @internal volume package internal use only.
5996 VDisconnectSALV_r(void)
5998 return SALVSYNC_clientFinis();
6002 * disconnect and then re-connect to the salvageserver SYNC service.
6004 * @return operation status
6008 * @pre client should have a live connection to the salvageserver
6010 * @post old connection is dropped, and a new one is established
6013 * @see VDisconnectSALV
6014 * @see VReconnectSALV_r
6017 VReconnectSALV(void)
6021 retVal = VReconnectSALV_r();
6027 * disconnect and then re-connect to the salvageserver SYNC service.
6029 * @return operation status
6034 * @arg VOL_LOCK is held.
6035 * @arg client should have a live connection to the salvageserver.
6037 * @post old connection is dropped, and a new one is established
6039 * @see VConnectSALV_r
6040 * @see VDisconnectSALV
6041 * @see VReconnectSALV
6042 * @see SALVSYNC_clientReconnect
6044 * @internal volume package internal use only.
6047 VReconnectSALV_r(void)
6049 return SALVSYNC_clientReconnect();
6051 #endif /* SALVSYNC_BUILD_CLIENT */
6052 #endif /* AFS_DEMAND_ATTACH_FS */
6055 /***************************************************/
6056 /* FSSYNC routines */
6057 /***************************************************/
6059 /* This must be called by any volume utility which needs to run while the
6060 file server is also running. This is separated from VInitVolumePackage2 so
6061 that a utility can fork--and each of the children can independently
6062 initialize communication with the file server */
6063 #ifdef FSSYNC_BUILD_CLIENT
6065 * connect to the fileserver SYNC service.
6067 * @return operation status
6072 * @arg VInit must equal 2.
6073 * @arg Program Type must not be fileserver or salvager.
6075 * @post connection to fileserver SYNC service established
6078 * @see VDisconnectFS
6079 * @see VChildProcReconnectFS
6086 retVal = VConnectFS_r();
6092 * connect to the fileserver SYNC service.
6094 * @return operation status
6099 * @arg VInit must equal 2.
6100 * @arg Program Type must not be fileserver or salvager.
6101 * @arg VOL_LOCK is held.
6103 * @post connection to fileserver SYNC service established
6106 * @see VDisconnectFS_r
6107 * @see VChildProcReconnectFS_r
6109 * @internal volume package internal use only.
6115 opr_Assert((VInit == 2) &&
6116 (programType != fileServer) &&
6117 (programType != salvager));
6118 rc = FSYNC_clientInit();
6126 * disconnect from the fileserver SYNC service.
6129 * @arg client should have a live connection to the fileserver.
6130 * @arg VOL_LOCK is held.
6131 * @arg Program Type must not be fileserver or salvager.
6133 * @post connection to fileserver SYNC service destroyed
6135 * @see VDisconnectFS
6137 * @see VChildProcReconnectFS_r
6139 * @internal volume package internal use only.
6142 VDisconnectFS_r(void)
6144 opr_Assert((programType != fileServer) &&
6145 (programType != salvager));
6146 FSYNC_clientFinis();
6151 * disconnect from the fileserver SYNC service.
6154 * @arg client should have a live connection to the fileserver.
6155 * @arg Program Type must not be fileserver or salvager.
6157 * @post connection to fileserver SYNC service destroyed
6159 * @see VDisconnectFS_r
6161 * @see VChildProcReconnectFS
6172 * connect to the fileserver SYNC service from a child process following a fork.
6174 * @return operation status
6179 * @arg VOL_LOCK is held.
6180 * @arg current FSYNC handle is shared with a parent process
6182 * @post current FSYNC handle is discarded and a new connection to the
6183 * fileserver SYNC service is established
6185 * @see VChildProcReconnectFS
6187 * @see VDisconnectFS_r
6189 * @internal volume package internal use only.
6192 VChildProcReconnectFS_r(void)
6194 return FSYNC_clientChildProcReconnect();
6198 * connect to the fileserver SYNC service from a child process following a fork.
6200 * @return operation status
6204 * @pre current FSYNC handle is shared with a parent process
6206 * @post current FSYNC handle is discarded and a new connection to the
6207 * fileserver SYNC service is established
6209 * @see VChildProcReconnectFS_r
6211 * @see VDisconnectFS
6214 VChildProcReconnectFS(void)
6218 ret = VChildProcReconnectFS_r();
6222 #endif /* FSSYNC_BUILD_CLIENT */
6225 /***************************************************/
6226 /* volume bitmap routines */
6227 /***************************************************/
6230 * Grow the bitmap by the defined increment
6233 VGrowBitmap(struct vnodeIndex *index)
6237 bp = realloc(index->bitmap, index->bitmapSize + VOLUME_BITMAP_GROWSIZE);
6238 osi_Assert(bp != NULL);
6240 bp += index->bitmapSize;
6241 memset(bp, 0, VOLUME_BITMAP_GROWSIZE);
6242 index->bitmapOffset = index->bitmapSize;
6243 index->bitmapSize += VOLUME_BITMAP_GROWSIZE;
6249 * allocate a vnode bitmap number for the vnode
6251 * @param[out] ec error code
6252 * @param[in] vp volume object pointer
6253 * @param[in] index vnode index number for the vnode
6254 * @param[in] flags flag values described in note
6256 * @note for DAFS, flags parameter controls locking behavior.
6257 * If (flags & VOL_ALLOC_BITMAP_WAIT) is set, then this function
6258 * will create a reservation and block on any other exclusive
6259 * operations. Otherwise, this function assumes the caller
6260 * already has exclusive access to vp, and we just change the
6263 * @pre VOL_LOCK held
6265 * @return bit number allocated
6271 VAllocBitmapEntry_r(Error * ec, Volume * vp,
6272 struct vnodeIndex *index, int flags)
6276 #ifdef AFS_DEMAND_ATTACH_FS
6277 VolState state_save;
6278 #endif /* AFS_DEMAND_ATTACH_FS */
6282 /* This test is probably redundant */
6283 if (!VolumeWriteable(vp)) {
6284 *ec = (bit32) VREADONLY;
6288 #ifdef AFS_DEMAND_ATTACH_FS
6289 if (flags & VOL_ALLOC_BITMAP_WAIT) {
6290 VCreateReservation_r(vp);
6291 VWaitExclusiveState_r(vp);
6293 state_save = VChangeState_r(vp, VOL_STATE_GET_BITMAP);
6294 #endif /* AFS_DEMAND_ATTACH_FS */
6297 if ((programType == fileServer) && !index->bitmap) {
6299 #ifndef AFS_DEMAND_ATTACH_FS
6300 /* demand attach fs uses the volume state to avoid races.
6301 * specialStatus field is not used at all */
6303 if (vp->specialStatus == VBUSY) {
6304 if (vp->goingOffline) { /* vos dump waiting for the volume to
6305 * go offline. We probably come here
6306 * from AddNewReadableResidency */
6309 while (vp->specialStatus == VBUSY) {
6310 #ifdef AFS_PTHREAD_ENV
6314 #else /* !AFS_PTHREAD_ENV */
6316 #endif /* !AFS_PTHREAD_ENV */
6320 #endif /* !AFS_DEMAND_ATTACH_FS */
6322 if (!index->bitmap) {
6323 #ifndef AFS_DEMAND_ATTACH_FS
6324 vp->specialStatus = VBUSY; /* Stop anyone else from using it. */
6325 #endif /* AFS_DEMAND_ATTACH_FS */
6326 for (i = 0; i < nVNODECLASSES; i++) {
6327 VGetBitmap_r(ec, vp, i);
6329 #ifdef AFS_DEMAND_ATTACH_FS
6330 VRequestSalvage_r(ec, vp, SALVSYNC_ERROR, 0 /*flags*/);
6331 #else /* AFS_DEMAND_ATTACH_FS */
6332 DeleteVolumeFromHashTable(vp);
6333 vp->shuttingDown = 1; /* Let who has it free it. */
6334 vp->specialStatus = 0;
6335 #endif /* AFS_DEMAND_ATTACH_FS */
6339 #ifndef AFS_DEMAND_ATTACH_FS
6341 vp->specialStatus = 0; /* Allow others to have access. */
6342 #endif /* AFS_DEMAND_ATTACH_FS */
6345 #endif /* BITMAP_LATER */
6347 #ifdef AFS_DEMAND_ATTACH_FS
6349 #endif /* AFS_DEMAND_ATTACH_FS */
6350 bp = index->bitmap + index->bitmapOffset;
6351 ep = index->bitmap + index->bitmapSize;
6353 if ((*(bit32 *) bp) != (bit32) 0xffffffff) {
6355 index->bitmapOffset = (afs_uint32) (bp - index->bitmap);
6358 o = ffs(~*bp) - 1; /* ffs is documented in BSTRING(3) */
6360 ret = ((bp - index->bitmap) * 8 + o);
6361 #ifdef AFS_DEMAND_ATTACH_FS
6363 #endif /* AFS_DEMAND_ATTACH_FS */
6366 bp += sizeof(bit32) /* i.e. 4 */ ;
6368 /* No bit map entry--must grow bitmap */
6370 bp = index->bitmap + index->bitmapOffset;
6372 ret = index->bitmapOffset * 8;
6373 #ifdef AFS_DEMAND_ATTACH_FS
6375 #endif /* AFS_DEMAND_ATTACH_FS */
6378 #ifdef AFS_DEMAND_ATTACH_FS
6379 VChangeState_r(vp, state_save);
6380 if (flags & VOL_ALLOC_BITMAP_WAIT) {
6381 VCancelReservation_r(vp);
6383 #endif /* AFS_DEMAND_ATTACH_FS */
6388 VAllocBitmapEntry(Error * ec, Volume * vp, struct vnodeIndex * index)
6392 retVal = VAllocBitmapEntry_r(ec, vp, index, VOL_ALLOC_BITMAP_WAIT);
6398 VFreeBitMapEntry_r(Error * ec, Volume *vp, struct vnodeIndex *index,
6399 unsigned bitNumber, int flags)
6401 unsigned int offset;
6405 #ifdef AFS_DEMAND_ATTACH_FS
6406 if (flags & VOL_FREE_BITMAP_WAIT) {
6407 /* VAllocBitmapEntry_r allocs bitmap entries under an exclusive volume
6408 * state, so ensure we're not in an exclusive volume state when we update
6410 VCreateReservation_r(vp);
6411 VWaitExclusiveState_r(vp);
6418 #endif /* BITMAP_LATER */
6420 offset = bitNumber >> 3;
6421 if (offset >= index->bitmapSize) {
6425 if (offset < index->bitmapOffset)
6426 index->bitmapOffset = offset & ~3; /* Truncate to nearest bit32 */
6427 *(index->bitmap + offset) &= ~(1 << (bitNumber & 0x7));
6430 #ifdef AFS_DEMAND_ATTACH_FS
6431 VCancelReservation_r(vp);
6433 return; /* make the compiler happy for non-DAFS */
6437 VFreeBitMapEntry(Error * ec, Volume *vp, struct vnodeIndex *index,
6441 VFreeBitMapEntry_r(ec, vp, index, bitNumber, VOL_FREE_BITMAP_WAIT);
6445 /* this function will drop the glock internally.
6446 * for old pthread fileservers, this is safe thanks to vbusy.
6448 * for demand attach fs, caller must have already called
6449 * VCreateReservation_r and VWaitExclusiveState_r */
6451 VGetBitmap_r(Error * ec, Volume * vp, VnodeClass class)
6453 StreamHandle_t *file;
6454 afs_sfsize_t nVnodes, size;
6455 struct VnodeClassInfo *vcp = &VnodeClassInfo[class];
6456 struct vnodeIndex *vip = &vp->vnodeIndex[class];
6457 struct VnodeDiskObject *vnode;
6458 unsigned int unique = 0;
6462 #endif /* BITMAP_LATER */
6463 #ifdef AFS_DEMAND_ATTACH_FS
6464 VolState state_save;
6465 #endif /* AFS_DEMAND_ATTACH_FS */
6469 #ifdef AFS_DEMAND_ATTACH_FS
6470 state_save = VChangeState_r(vp, VOL_STATE_GET_BITMAP);
6471 #endif /* AFS_DEMAND_ATTACH_FS */
6474 fdP = IH_OPEN(vip->handle);
6475 opr_Assert(fdP != NULL);
6476 file = FDH_FDOPEN(fdP, "r");
6477 opr_Assert(file != NULL);
6478 vnode = malloc(vcp->diskSize);
6479 opr_Assert(vnode != NULL);
6480 size = OS_SIZE(fdP->fd_fd);
6481 opr_Assert(size != -1);
6482 nVnodes = (size <= vcp->diskSize ? 0 : size - vcp->diskSize)
6484 vip->bitmapSize = ((nVnodes / 8) + 10) / 4 * 4; /* The 10 is a little extra so
6485 * a few files can be created in this volume,
6486 * the whole thing is rounded up to nearest 4
6487 * bytes, because the bit map allocator likes
6490 BitMap = (byte *) calloc(1, vip->bitmapSize);
6491 opr_Assert(BitMap != NULL);
6492 #else /* BITMAP_LATER */
6493 vip->bitmap = (byte *) calloc(1, vip->bitmapSize);
6494 opr_Assert(vip->bitmap != NULL);
6495 vip->bitmapOffset = 0;
6496 #endif /* BITMAP_LATER */
6497 if (STREAM_ASEEK(file, vcp->diskSize) != -1) {
6499 for (bitNumber = 0; bitNumber < nVnodes + 100; bitNumber++) {
6500 if (STREAM_READ(vnode, vcp->diskSize, 1, file) != 1)
6502 if (vnode->type != vNull) {
6503 if (vnode->vnodeMagic != vcp->magic) {
6504 Log("GetBitmap: addled vnode index in volume %s; volume needs salvage\n", V_name(vp));
6509 *(BitMap + (bitNumber >> 3)) |= (1 << (bitNumber & 0x7));
6510 #else /* BITMAP_LATER */
6511 *(vip->bitmap + (bitNumber >> 3)) |= (1 << (bitNumber & 0x7));
6512 #endif /* BITMAP_LATER */
6513 if (unique <= vnode->uniquifier)
6514 unique = vnode->uniquifier + 1;
6516 #ifndef AFS_PTHREAD_ENV
6517 if ((bitNumber & 0x00ff) == 0x0ff) { /* every 256 iterations */
6520 #endif /* !AFS_PTHREAD_ENV */
6523 if (vp->nextVnodeUnique < unique) {
6524 Log("GetBitmap: bad volume uniquifier for volume %s; volume needs salvage\n", V_name(vp));
6527 /* Paranoia, partly justified--I think fclose after fdopen
6528 * doesn't seem to close fd. In any event, the documentation
6529 * doesn't specify, so it's safer to close it twice.
6537 /* There may have been a racing condition with some other thread, both
6538 * creating the bitmaps for this volume. If the other thread was faster
6539 * the pointer to bitmap should already be filled and we can free ours.
6541 if (vip->bitmap == NULL) {
6542 vip->bitmap = BitMap;
6543 vip->bitmapOffset = 0;
6546 #endif /* BITMAP_LATER */
6547 #ifdef AFS_DEMAND_ATTACH_FS
6548 VChangeState_r(vp, state_save);
6549 #endif /* AFS_DEMAND_ATTACH_FS */
6553 /***************************************************/
6554 /* Volume Path and Volume Number utility routines */
6555 /***************************************************/
6558 * find the first occurrence of a volume header file and return the path.
6560 * @param[out] ec outbound error code
6561 * @param[in] volumeId volume id to find
6562 * @param[out] partitionp pointer to disk partition path string
6563 * @param[out] namep pointer to volume header file name string
6565 * @post path to first occurrence of volume header is returned in partitionp
6566 * and namep, or ec is set accordingly.
6568 * @warning this function is NOT re-entrant -- partitionp and namep point to
6569 * static data segments
6571 * @note if a volume utility inadvertently leaves behind a stale volume header
6572 * on a vice partition, it is possible for callers to get the wrong one,
6573 * depending on the order of the disk partition linked list.
6577 VGetVolumePath(Error * ec, VolumeId volumeId, char **partitionp, char **namep)
6579 static char partition[VMAXPATHLEN], name[VMAXPATHLEN];
6580 char path[VMAXPATHLEN];
6582 struct DiskPartition64 *dp;
6585 name[0] = OS_DIRSEPC;
6586 snprintf(&name[1], (sizeof name) - 1, VFORMAT,
6587 afs_printable_VolumeId_lu(volumeId));
6588 for (dp = DiskPartitionList; dp; dp = dp->next) {
6589 struct afs_stat_st status;
6590 strcpy(path, VPartitionPath(dp));
6592 if (afs_stat(path, &status) == 0) {
6593 strcpy(partition, dp->name);
6600 *partitionp = *namep = NULL;
6602 *partitionp = partition;
6608 * extract a volume number from a volume header filename string.
6610 * @param[in] name volume header filename string
6612 * @return volume number
6614 * @note the string must be of the form VFORMAT. the only permissible
6615 * deviation is a leading OS_DIRSEPC character.
6620 VolumeNumber(char *name)
6622 if (*name == OS_DIRSEPC)
6624 return strtoul(name + 1, NULL, 10);
6628 * compute the volume header filename.
6630 * @param[in] volumeId
6632 * @return volume header filename
6634 * @post volume header filename string is constructed
6636 * @warning this function is NOT re-entrant -- the returned string is
6637 * stored in a static char array. see VolumeExternalName_r
6638 * for a re-entrant equivalent.
6640 * @see VolumeExternalName_r
6642 * @deprecated due to the above re-entrancy warning, this interface should
6643 * be considered deprecated. Please use VolumeExternalName_r
6647 VolumeExternalName(VolumeId volumeId)
6649 static char name[VMAXPATHLEN];
6650 snprintf(name, sizeof name, VFORMAT, afs_printable_VolumeId_lu(volumeId));
6655 * compute the volume header filename.
6657 * @param[in] volumeId
6658 * @param[inout] name array in which to store filename
6659 * @param[in] len length of name array
6661 * @return result code from afs_snprintf
6663 * @see VolumeExternalName
6666 * @note re-entrant equivalent of VolumeExternalName
6669 VolumeExternalName_r(VolumeId volumeId, char * name, size_t len)
6671 return snprintf(name, len, VFORMAT, afs_printable_VolumeId_lu(volumeId));
6675 /***************************************************/
6676 /* Volume Usage Statistics routines */
6677 /***************************************************/
6679 #define OneDay (86400) /* 24 hours' worth of seconds */
6682 Midnight(time_t t) {
6683 struct tm local, *l;
6686 #if defined(AFS_PTHREAD_ENV) && !defined(AFS_NT40_ENV)
6687 l = localtime_r(&t, &local);
6693 /* the following is strictly speaking problematic on the
6694 switching day to daylight saving time, after the switch,
6695 as tm_isdst does not match. Similarly, on the looong day when
6696 switching back the OneDay check will not do what naively expected!
6697 The effects are minor, though, and more a matter of interpreting
6699 #ifndef AFS_PTHREAD_ENV
6702 local.tm_hour = local.tm_min=local.tm_sec = 0;
6703 midnight = mktime(&local);
6704 if (midnight != (time_t) -1) return(midnight);
6706 return( (t/OneDay)*OneDay );
6710 /*------------------------------------------------------------------------
6711 * [export] VAdjustVolumeStatistics
6714 * If we've passed midnight, we need to update all the day use
6715 * statistics as well as zeroing the detailed volume statistics
6716 * (if we are implementing them).
6719 * vp : Pointer to the volume structure describing the lucky
6720 * volume being considered for update.
6726 * Nothing interesting.
6730 *------------------------------------------------------------------------*/
6733 VAdjustVolumeStatistics_r(Volume * vp)
6735 unsigned int now = FT_ApproxTime();
6737 if (now - V_dayUseDate(vp) > OneDay) {
6740 ndays = (now - V_dayUseDate(vp)) / OneDay;
6741 for (i = 6; i > ndays - 1; i--)
6742 V_weekUse(vp)[i] = V_weekUse(vp)[i - ndays];
6743 for (i = 0; i < ndays - 1 && i < 7; i++)
6744 V_weekUse(vp)[i] = 0;
6746 V_weekUse(vp)[ndays - 1] = V_dayUse(vp);
6748 V_dayUseDate(vp) = Midnight(now);
6751 * All we need to do is bzero the entire VOL_STATS_BYTES of
6752 * the detailed volume statistics area.
6754 memset((V_stat_area(vp)), 0, VOL_STATS_BYTES);
6757 /*It's been more than a day of collection */
6759 * Always return happily.
6762 } /*VAdjustVolumeStatistics */
6765 VAdjustVolumeStatistics(Volume * vp)
6769 retVal = VAdjustVolumeStatistics_r(vp);
6775 VBumpVolumeUsage_r(Volume * vp)
6777 unsigned int now = FT_ApproxTime();
6778 V_accessDate(vp) = now;
6779 if (now - V_dayUseDate(vp) > OneDay)
6780 VAdjustVolumeStatistics_r(vp);
6782 * Save the volume header image to disk after a threshold of bumps to dayUse,
6783 * at most every usage_rate_limit seconds.
6786 vp->usage_bumps_outstanding++;
6787 if (vp->usage_bumps_outstanding >= vol_opts.usage_threshold
6788 && vp->usage_bumps_next_write <= now) {
6790 vp->usage_bumps_outstanding = 0;
6791 vp->usage_bumps_next_write = now + vol_opts.usage_rate_limit;
6792 VUpdateVolume_r(&error, vp, VOL_UPDATE_WAIT);
6797 VBumpVolumeUsage(Volume * vp)
6800 VBumpVolumeUsage_r(vp);
6805 VSetDiskUsage_r(void)
6807 #ifndef AFS_DEMAND_ATTACH_FS
6808 static int FifteenMinuteCounter = 0;
6812 /* NOTE: Don't attempt to access the partitions list until the
6813 * initialization level indicates that all volumes are attached,
6814 * which implies that all partitions are initialized. */
6815 #ifdef AFS_PTHREAD_ENV
6816 VOL_CV_WAIT(&vol_vinit_cond);
6817 #else /* AFS_PTHREAD_ENV */
6819 #endif /* AFS_PTHREAD_ENV */
6822 VResetDiskUsage_r();
6824 #ifndef AFS_DEMAND_ATTACH_FS
6825 if (++FifteenMinuteCounter == 3) {
6826 FifteenMinuteCounter = 0;
6829 #endif /* !AFS_DEMAND_ATTACH_FS */
6841 /***************************************************/
6842 /* Volume Update List routines */
6843 /***************************************************/
6845 /* The number of minutes that a volume hasn't been updated before the
6846 * "Dont salvage" flag in the volume header will be turned on */
6847 #define SALVAGE_INTERVAL (10*60)
6852 * volume update list functionality has been moved into the VLRU
6853 * the DONT_SALVAGE flag is now set during VLRU demotion
6856 #ifndef AFS_DEMAND_ATTACH_FS
6857 static VolumeId *UpdateList = NULL; /* Pointer to array of Volume ID's */
6858 static int nUpdatedVolumes = 0; /* Updated with entry in UpdateList, salvage after crash flag on */
6859 static int updateSize = 0; /* number of entries possible */
6860 #define UPDATE_LIST_SIZE 128 /* initial size increment (must be a power of 2!) */
6861 #endif /* !AFS_DEMAND_ATTACH_FS */
6864 VAddToVolumeUpdateList_r(Error * ec, Volume * vp)
6867 vp->updateTime = FT_ApproxTime();
6868 if (V_dontSalvage(vp) == 0)
6870 V_dontSalvage(vp) = 0;
6871 VSyncVolume_r(ec, vp, 0);
6872 #ifdef AFS_DEMAND_ATTACH_FS
6873 V_attachFlags(vp) &= ~(VOL_HDR_DONTSALV);
6874 #else /* !AFS_DEMAND_ATTACH_FS */
6877 if (UpdateList == NULL) {
6878 updateSize = UPDATE_LIST_SIZE;
6879 UpdateList = malloc(sizeof(VolumeId) * updateSize);
6881 if (nUpdatedVolumes == updateSize) {
6883 if (updateSize > 524288) {
6884 Log("warning: there is likely a bug in the volume update scanner\n");
6887 UpdateList = realloc(UpdateList,
6888 sizeof(VolumeId) * updateSize);
6891 opr_Assert(UpdateList != NULL);
6892 UpdateList[nUpdatedVolumes++] = V_id(vp);
6893 #endif /* !AFS_DEMAND_ATTACH_FS */
6896 #ifndef AFS_DEMAND_ATTACH_FS
6898 VScanUpdateList(void)
6903 afs_uint32 now = FT_ApproxTime();
6904 /* Be careful with this code, since it works with interleaved calls to AddToVolumeUpdateList */
6905 for (i = gap = 0; i < nUpdatedVolumes; i++) {
6907 UpdateList[i - gap] = UpdateList[i];
6909 /* XXX this routine needlessly messes up the Volume LRU by
6910 * breaking the LRU temporal-locality assumptions.....
6911 * we should use a special volume header allocator here */
6912 vp = VGetVolume_r(&error, UpdateList[i - gap] = UpdateList[i]);
6915 } else if (vp->nUsers == 1 && now - vp->updateTime > SALVAGE_INTERVAL) {
6916 V_dontSalvage(vp) = DONT_SALVAGE;
6917 VUpdateVolume_r(&error, vp, 0); /* No need to fsync--not critical */
6925 #ifndef AFS_PTHREAD_ENV
6927 #endif /* !AFS_PTHREAD_ENV */
6929 nUpdatedVolumes -= gap;
6931 #endif /* !AFS_DEMAND_ATTACH_FS */
6934 /***************************************************/
6935 /* Volume LRU routines */
6936 /***************************************************/
6941 * with demand attach fs, we attempt to soft detach(1)
6942 * volumes which have not been accessed in a long time
6943 * in order to speed up fileserver shutdown
6945 * (1) by soft detach we mean a process very similar
6946 * to VOffline, except the final state of the
6947 * Volume will be VOL_STATE_PREATTACHED, instead
6948 * of the usual VOL_STATE_UNATTACHED
6950 #ifdef AFS_DEMAND_ATTACH_FS
6952 /* implementation is reminiscent of a generational GC
6954 * queue 0 is newly attached volumes. this queue is
6955 * sorted by attach timestamp
6957 * queue 1 is volumes that have been around a bit
6958 * longer than queue 0. this queue is sorted by
6961 * queue 2 is volumes tha have been around the longest.
6962 * this queue is unsorted
6964 * queue 3 is volumes that have been marked as
6965 * candidates for soft detachment. this queue is
6968 #define VLRU_GENERATIONS 3 /**< number of generations in VLRU */
6969 #define VLRU_QUEUES 5 /**< total number of VLRU queues */
6972 * definition of a VLRU queue.
6975 volatile struct rx_queue q;
6982 * main VLRU data structure.
6985 struct VLRU_q q[VLRU_QUEUES]; /**< VLRU queues */
6988 /** time interval (in seconds) between promotion passes for
6989 * each young generation queue. */
6990 afs_uint32 promotion_interval[VLRU_GENERATIONS-1];
6992 /** time interval (in seconds) between soft detach candidate
6993 * scans for each generation queue.
6995 * scan_interval[VLRU_QUEUE_CANDIDATE] defines how frequently
6996 * we perform a soft detach pass. */
6997 afs_uint32 scan_interval[VLRU_GENERATIONS+1];
6999 /* scheduler state */
7000 int next_idx; /**< next queue to receive attention */
7001 afs_uint32 last_promotion[VLRU_GENERATIONS-1]; /**< timestamp of last promotion scan */
7002 afs_uint32 last_scan[VLRU_GENERATIONS+1]; /**< timestamp of last detach scan */
7004 int scanner_state; /**< state of scanner thread */
7005 pthread_cond_t cv; /**< state transition CV */
7008 /** global VLRU state */
7009 static struct VLRU volume_LRU;
7012 * defined states for VLRU scanner thread.
7015 VLRU_SCANNER_STATE_OFFLINE = 0, /**< vlru scanner thread is offline */
7016 VLRU_SCANNER_STATE_ONLINE = 1, /**< vlru scanner thread is online */
7017 VLRU_SCANNER_STATE_SHUTTING_DOWN = 2, /**< vlru scanner thread is shutting down */
7018 VLRU_SCANNER_STATE_PAUSING = 3, /**< vlru scanner thread is getting ready to pause */
7019 VLRU_SCANNER_STATE_PAUSED = 4 /**< vlru scanner thread is paused */
7020 } vlru_thread_state_t;
7022 /* vlru disk data header stuff */
7023 #define VLRU_DISK_MAGIC 0x7a8b9cad /**< vlru disk entry magic number */
7024 #define VLRU_DISK_VERSION 1 /**< vlru disk entry version number */
7026 /** vlru default expiration time (for eventual fs state serialization of vlru data) */
7027 #define VLRU_DUMP_EXPIRATION_TIME (60*60*24*7) /* expire vlru data after 1 week */
7030 /** minimum volume inactivity (in seconds) before a volume becomes eligible for
7031 * soft detachment. */
7032 static afs_uint32 VLRU_offline_thresh = VLRU_DEFAULT_OFFLINE_THRESH;
7034 /** time interval (in seconds) between VLRU scanner thread soft detach passes. */
7035 static afs_uint32 VLRU_offline_interval = VLRU_DEFAULT_OFFLINE_INTERVAL;
7037 /** maximum number of volumes to soft detach in a VLRU soft detach pass. */
7038 static afs_uint32 VLRU_offline_max = VLRU_DEFAULT_OFFLINE_MAX;
7040 /** VLRU control flag. non-zero value implies VLRU subsystem is activated. */
7041 static afs_uint32 VLRU_enabled = 1;
7043 /* queue synchronization routines */
7044 static void VLRU_BeginExclusive_r(struct VLRU_q * q);
7045 static void VLRU_EndExclusive_r(struct VLRU_q * q);
7046 static void VLRU_Wait_r(struct VLRU_q * q);
7049 * set VLRU subsystem tunable parameters.
7051 * @param[in] option tunable option to modify
7052 * @param[in] val new value for tunable parameter
7054 * @pre @c VInitVolumePackage2 has not yet been called.
7056 * @post tunable parameter is modified
7060 * @note valid option parameters are:
7061 * @arg @c VLRU_SET_THRESH
7062 * set the period of inactivity after which
7063 * volumes are eligible for soft detachment
7064 * @arg @c VLRU_SET_INTERVAL
7065 * set the time interval between calls
7066 * to the volume LRU "garbage collector"
7067 * @arg @c VLRU_SET_MAX
7068 * set the max number of volumes to deallocate
7072 VLRU_SetOptions(int option, afs_uint32 val)
7074 if (option == VLRU_SET_THRESH) {
7075 VLRU_offline_thresh = val;
7076 } else if (option == VLRU_SET_INTERVAL) {
7077 VLRU_offline_interval = val;
7078 } else if (option == VLRU_SET_MAX) {
7079 VLRU_offline_max = val;
7080 } else if (option == VLRU_SET_ENABLED) {
7083 VLRU_ComputeConstants();
7087 * compute VLRU internal timing parameters.
7089 * @post VLRU scanner thread internal timing parameters are computed
7091 * @note computes internal timing parameters based upon user-modifiable
7092 * tunable parameters.
7096 * @internal volume package internal use only.
7099 VLRU_ComputeConstants(void)
7101 afs_uint32 factor = VLRU_offline_thresh / VLRU_offline_interval;
7103 /* compute the candidate scan interval */
7104 volume_LRU.scan_interval[VLRU_QUEUE_CANDIDATE] = VLRU_offline_interval;
7106 /* compute the promotion intervals */
7107 volume_LRU.promotion_interval[VLRU_QUEUE_NEW] = VLRU_offline_thresh * 2;
7108 volume_LRU.promotion_interval[VLRU_QUEUE_MID] = VLRU_offline_thresh * 4;
7111 /* compute the gen 0 scan interval */
7112 volume_LRU.scan_interval[VLRU_QUEUE_NEW] = VLRU_offline_thresh / 8;
7114 /* compute the gen 0 scan interval */
7115 volume_LRU.scan_interval[VLRU_QUEUE_NEW] = VLRU_offline_interval * 2;
7120 * initialize VLRU subsystem.
7122 * @pre this function has not yet been called
7124 * @post VLRU subsystem is initialized and VLRU scanner thread is starting
7128 * @internal volume package internal use only.
7134 pthread_attr_t attrs;
7137 if (!VLRU_enabled) {
7138 Log("VLRU: disabled\n");
7142 /* initialize each of the VLRU queues */
7143 for (i = 0; i < VLRU_QUEUES; i++) {
7144 queue_Init(&volume_LRU.q[i]);
7145 volume_LRU.q[i].len = 0;
7146 volume_LRU.q[i].busy = 0;
7147 opr_cv_init(&volume_LRU.q[i].cv);
7150 /* setup the timing constants */
7151 VLRU_ComputeConstants();
7153 /* XXX put inside LogLevel check? */
7154 Log("VLRU: starting scanner with the following configuration parameters:\n");
7155 Log("VLRU: offlining volumes after minimum of %d seconds of inactivity\n", VLRU_offline_thresh);
7156 Log("VLRU: running VLRU soft detach pass every %d seconds\n", VLRU_offline_interval);
7157 Log("VLRU: taking up to %d volumes offline per pass\n", VLRU_offline_max);
7158 Log("VLRU: scanning generation 0 for inactive volumes every %d seconds\n", volume_LRU.scan_interval[0]);
7159 Log("VLRU: scanning for promotion/demotion between generations 0 and 1 every %d seconds\n", volume_LRU.promotion_interval[0]);
7160 Log("VLRU: scanning for promotion/demotion between generations 1 and 2 every %d seconds\n", volume_LRU.promotion_interval[1]);
7162 /* start up the VLRU scanner */
7163 volume_LRU.scanner_state = VLRU_SCANNER_STATE_OFFLINE;
7164 if (programType == fileServer) {
7165 opr_cv_init(&volume_LRU.cv);
7166 opr_Verify(pthread_attr_init(&attrs) == 0);
7167 opr_Verify(pthread_attr_setdetachstate(&attrs,
7168 PTHREAD_CREATE_DETACHED) == 0);
7169 opr_Verify(pthread_create(&tid, &attrs,
7170 &VLRU_ScannerThread, NULL) == 0);
7175 * initialize the VLRU-related fields of a newly allocated volume object.
7177 * @param[in] vp pointer to volume object
7180 * @arg @c VOL_LOCK is held.
7181 * @arg volume object is not on a VLRU queue.
7183 * @post VLRU fields are initialized to indicate that volume object is not
7184 * currently registered with the VLRU subsystem
7188 * @internal volume package interal use only.
7191 VLRU_Init_Node_r(Volume * vp)
7196 opr_Assert(queue_IsNotOnQueue(&vp->vlru));
7197 vp->vlru.idx = VLRU_QUEUE_INVALID;
7201 * add a volume object to a VLRU queue.
7203 * @param[in] vp pointer to volume object
7206 * @arg @c VOL_LOCK is held.
7207 * @arg caller MUST hold a lightweight ref on @p vp.
7208 * @arg caller MUST NOT hold exclusive ownership of the VLRU queue.
7210 * @post the volume object is added to the appropriate VLRU queue
7212 * @note if @c vp->vlru.idx contains the index of a valid VLRU queue,
7213 * then the volume is added to that queue. Otherwise, the value
7214 * @c VLRU_QUEUE_NEW is stored into @c vp->vlru.idx and the
7215 * volume is added to the NEW generation queue.
7217 * @note @c VOL_LOCK may be dropped internally
7219 * @note Volume state is temporarily set to @c VOL_STATE_VLRU_ADD
7220 * during the add operation, and is restored to the previous
7221 * state prior to return.
7225 * @internal volume package internal use only.
7228 VLRU_Add_r(Volume * vp)
7231 VolState state_save;
7236 if (queue_IsOnQueue(&vp->vlru))
7239 state_save = VChangeState_r(vp, VOL_STATE_VLRU_ADD);
7242 if ((idx < 0) || (idx >= VLRU_QUEUE_INVALID)) {
7243 idx = VLRU_QUEUE_NEW;
7246 VLRU_Wait_r(&volume_LRU.q[idx]);
7248 /* repeat check since VLRU_Wait_r may have dropped
7250 if (queue_IsNotOnQueue(&vp->vlru)) {
7252 queue_Prepend(&volume_LRU.q[idx], &vp->vlru);
7253 volume_LRU.q[idx].len++;
7254 V_attachFlags(vp) |= VOL_ON_VLRU;
7255 vp->stats.last_promote = FT_ApproxTime();
7258 VChangeState_r(vp, state_save);
7262 * delete a volume object from a VLRU queue.
7264 * @param[in] vp pointer to volume object
7267 * @arg @c VOL_LOCK is held.
7268 * @arg caller MUST hold a lightweight ref on @p vp.
7269 * @arg caller MUST NOT hold exclusive ownership of the VLRU queue.
7271 * @post volume object is removed from the VLRU queue
7273 * @note @c VOL_LOCK may be dropped internally
7277 * @todo We should probably set volume state to something exlcusive
7278 * (as @c VLRU_Add_r does) prior to dropping @c VOL_LOCK.
7280 * @internal volume package internal use only.
7283 VLRU_Delete_r(Volume * vp)
7290 if (queue_IsNotOnQueue(&vp->vlru))
7296 if (idx == VLRU_QUEUE_INVALID)
7298 VLRU_Wait_r(&volume_LRU.q[idx]);
7299 } while (idx != vp->vlru.idx);
7301 /* now remove from the VLRU and update
7302 * the appropriate counter */
7303 queue_Remove(&vp->vlru);
7304 volume_LRU.q[idx].len--;
7305 vp->vlru.idx = VLRU_QUEUE_INVALID;
7306 V_attachFlags(vp) &= ~(VOL_ON_VLRU);
7310 * tell the VLRU subsystem that a volume was just accessed.
7312 * @param[in] vp pointer to volume object
7315 * @arg @c VOL_LOCK is held
7316 * @arg caller MUST hold a lightweight ref on @p vp
7317 * @arg caller MUST NOT hold exclusive ownership of any VLRU queue
7319 * @post volume VLRU access statistics are updated. If the volume was on
7320 * the VLRU soft detach candidate queue, it is moved to the NEW
7323 * @note @c VOL_LOCK may be dropped internally
7327 * @internal volume package internal use only.
7330 VLRU_UpdateAccess_r(Volume * vp)
7332 Volume * rvp = NULL;
7337 if (queue_IsNotOnQueue(&vp->vlru))
7340 opr_Assert(V_attachFlags(vp) & VOL_ON_VLRU);
7342 /* update the access timestamp */
7343 vp->stats.last_get = FT_ApproxTime();
7346 * if the volume is on the soft detach candidate
7347 * list, we need to safely move it back to a
7348 * regular generation. this has to be done
7349 * carefully so we don't race against the scanner
7353 /* if this volume is on the soft detach candidate queue,
7354 * then grab exclusive access to the necessary queues */
7355 if (vp->vlru.idx == VLRU_QUEUE_CANDIDATE) {
7357 VCreateReservation_r(rvp);
7359 VLRU_Wait_r(&volume_LRU.q[VLRU_QUEUE_NEW]);
7360 VLRU_BeginExclusive_r(&volume_LRU.q[VLRU_QUEUE_NEW]);
7361 VLRU_Wait_r(&volume_LRU.q[VLRU_QUEUE_CANDIDATE]);
7362 VLRU_BeginExclusive_r(&volume_LRU.q[VLRU_QUEUE_CANDIDATE]);
7365 /* make sure multiple threads don't race to update */
7366 if (vp->vlru.idx == VLRU_QUEUE_CANDIDATE) {
7367 VLRU_SwitchQueues(vp, VLRU_QUEUE_NEW, 1);
7371 VLRU_EndExclusive_r(&volume_LRU.q[VLRU_QUEUE_CANDIDATE]);
7372 VLRU_EndExclusive_r(&volume_LRU.q[VLRU_QUEUE_NEW]);
7373 VCancelReservation_r(rvp);
7378 * switch a volume between two VLRU queues.
7380 * @param[in] vp pointer to volume object
7381 * @param[in] new_idx index of VLRU queue onto which the volume will be moved
7382 * @param[in] append controls whether the volume will be appended or
7383 * prepended to the queue. A nonzero value means it will
7384 * be appended; zero means it will be prepended.
7386 * @pre The new (and old, if applicable) queue(s) must either be owned
7387 * exclusively by the calling thread for asynchronous manipulation,
7388 * or the queue(s) must be quiescent and VOL_LOCK must be held.
7389 * Please see VLRU_BeginExclusive_r, VLRU_EndExclusive_r and VLRU_Wait_r
7390 * for further details of the queue asynchronous processing mechanism.
7392 * @post If the volume object was already on a VLRU queue, it is
7393 * removed from the queue. Depending on the value of the append
7394 * parameter, the volume object is either appended or prepended
7395 * to the VLRU queue referenced by the new_idx parameter.
7399 * @see VLRU_BeginExclusive_r
7400 * @see VLRU_EndExclusive_r
7403 * @internal volume package internal use only.
7406 VLRU_SwitchQueues(Volume * vp, int new_idx, int append)
7408 if (queue_IsNotOnQueue(&vp->vlru))
7411 queue_Remove(&vp->vlru);
7412 volume_LRU.q[vp->vlru.idx].len--;
7414 /* put the volume back on the correct generational queue */
7416 queue_Append(&volume_LRU.q[new_idx], &vp->vlru);
7418 queue_Prepend(&volume_LRU.q[new_idx], &vp->vlru);
7421 volume_LRU.q[new_idx].len++;
7422 vp->vlru.idx = new_idx;
7426 * VLRU background thread.
7428 * The VLRU Scanner Thread is responsible for periodically scanning through
7429 * each VLRU queue looking for volumes which should be moved to another
7430 * queue, or soft detached.
7432 * @param[in] args unused thread arguments parameter
7434 * @return unused thread return value
7435 * @retval NULL always
7437 * @internal volume package internal use only.
7440 VLRU_ScannerThread(void * args)
7442 afs_uint32 now, min_delay, delay;
7443 int i, min_idx, min_op, overdue, state;
7445 /* set t=0 for promotion cycle to be
7446 * fileserver startup */
7447 now = FT_ApproxTime();
7448 for (i=0; i < VLRU_GENERATIONS-1; i++) {
7449 volume_LRU.last_promotion[i] = now;
7452 /* don't start the scanner until VLRU_offline_thresh
7453 * plus a small delay for VInitVolumePackage2 to finish
7456 sleep(VLRU_offline_thresh + 60);
7458 /* set t=0 for scan cycle to be now */
7459 now = FT_ApproxTime();
7460 for (i=0; i < VLRU_GENERATIONS+1; i++) {
7461 volume_LRU.last_scan[i] = now;
7465 if (volume_LRU.scanner_state == VLRU_SCANNER_STATE_OFFLINE) {
7466 volume_LRU.scanner_state = VLRU_SCANNER_STATE_ONLINE;
7469 while ((state = volume_LRU.scanner_state) != VLRU_SCANNER_STATE_SHUTTING_DOWN) {
7470 /* check to see if we've been asked to pause */
7471 if (volume_LRU.scanner_state == VLRU_SCANNER_STATE_PAUSING) {
7472 volume_LRU.scanner_state = VLRU_SCANNER_STATE_PAUSED;
7473 opr_cv_broadcast(&volume_LRU.cv);
7475 VOL_CV_WAIT(&volume_LRU.cv);
7476 } while (volume_LRU.scanner_state == VLRU_SCANNER_STATE_PAUSED);
7479 /* scheduling can happen outside the glock */
7482 /* figure out what is next on the schedule */
7484 /* figure out a potential schedule for the new generation first */
7486 min_delay = volume_LRU.scan_interval[0] + volume_LRU.last_scan[0] - now;
7489 if (min_delay > volume_LRU.scan_interval[0]) {
7490 /* unsigned overflow -- we're overdue to run this scan */
7495 /* if we're not overdue for gen 0, figure out schedule for candidate gen */
7497 i = VLRU_QUEUE_CANDIDATE;
7498 delay = volume_LRU.scan_interval[i] + volume_LRU.last_scan[i] - now;
7499 if (delay < min_delay) {
7503 if (delay > volume_LRU.scan_interval[i]) {
7504 /* unsigned overflow -- we're overdue to run this scan */
7511 /* if we're still not overdue for something, figure out schedules for promotions */
7512 for (i=0; !overdue && i < VLRU_GENERATIONS-1; i++) {
7513 delay = volume_LRU.promotion_interval[i] + volume_LRU.last_promotion[i] - now;
7514 if (delay < min_delay) {
7519 if (delay > volume_LRU.promotion_interval[i]) {
7520 /* unsigned overflow -- we're overdue to run this promotion */
7529 /* sleep as needed */
7534 /* do whatever is next */
7537 VLRU_Promote_r(min_idx);
7538 VLRU_Demote_r(min_idx+1);
7540 VLRU_Scan_r(min_idx);
7542 now = FT_ApproxTime();
7545 Log("VLRU scanner asked to go offline (scanner_state=%d)\n", state);
7547 /* signal that scanner is down */
7548 volume_LRU.scanner_state = VLRU_SCANNER_STATE_OFFLINE;
7549 opr_cv_broadcast(&volume_LRU.cv);
7555 * promote volumes from one VLRU generation to the next.
7557 * This routine scans a VLRU generation looking for volumes which are
7558 * eligible to be promoted to the next generation. All volumes which
7559 * meet the eligibility requirement are promoted.
7561 * Promotion eligibility is based upon meeting both of the following
7564 * @arg The volume has been accessed since the last promotion:
7565 * @c (vp->stats.last_get >= vp->stats.last_promote)
7566 * @arg The last promotion occurred at least
7567 * @c volume_LRU.promotion_interval[idx] seconds ago
7569 * As a performance optimization, promotions are "globbed". In other
7570 * words, we promote arbitrarily large contiguous sublists of elements
7573 * @param[in] idx VLRU queue index to scan
7577 * @internal VLRU internal use only.
7580 VLRU_Promote_r(int idx)
7582 int len, chaining, promote;
7583 afs_uint32 now, thresh;
7584 struct rx_queue *qp, *nqp;
7585 Volume * vp, *start = NULL, *end = NULL;
7587 /* get exclusive access to two chains, and drop the glock */
7588 VLRU_Wait_r(&volume_LRU.q[idx]);
7589 VLRU_BeginExclusive_r(&volume_LRU.q[idx]);
7590 VLRU_Wait_r(&volume_LRU.q[idx+1]);
7591 VLRU_BeginExclusive_r(&volume_LRU.q[idx+1]);
7594 thresh = volume_LRU.promotion_interval[idx];
7595 now = FT_ApproxTime();
7598 for (queue_ScanBackwards(&volume_LRU.q[idx], qp, nqp, rx_queue)) {
7599 vp = (Volume *)((char *)qp - offsetof(Volume, vlru));
7600 promote = (((vp->stats.last_promote + thresh) <= now) &&
7601 (vp->stats.last_get >= vp->stats.last_promote));
7609 /* promote and prepend chain */
7610 queue_MoveChainAfter(&volume_LRU.q[idx+1], &start->vlru, &end->vlru);
7624 /* promote and prepend */
7625 queue_MoveChainAfter(&volume_LRU.q[idx+1], &start->vlru, &end->vlru);
7629 volume_LRU.q[idx].len -= len;
7630 volume_LRU.q[idx+1].len += len;
7633 /* release exclusive access to the two chains */
7635 volume_LRU.last_promotion[idx] = now;
7636 VLRU_EndExclusive_r(&volume_LRU.q[idx+1]);
7637 VLRU_EndExclusive_r(&volume_LRU.q[idx]);
7640 /* run the demotions */
7642 VLRU_Demote_r(int idx)
7645 int len, chaining, demote;
7646 afs_uint32 now, thresh;
7647 struct rx_queue *qp, *nqp;
7648 Volume * vp, *start = NULL, *end = NULL;
7649 Volume ** salv_flag_vec = NULL;
7650 int salv_vec_offset = 0;
7652 opr_Assert(idx == VLRU_QUEUE_MID || idx == VLRU_QUEUE_OLD);
7654 /* get exclusive access to two chains, and drop the glock */
7655 VLRU_Wait_r(&volume_LRU.q[idx-1]);
7656 VLRU_BeginExclusive_r(&volume_LRU.q[idx-1]);
7657 VLRU_Wait_r(&volume_LRU.q[idx]);
7658 VLRU_BeginExclusive_r(&volume_LRU.q[idx]);
7661 /* no big deal if this allocation fails */
7662 if (volume_LRU.q[idx].len) {
7663 salv_flag_vec = malloc(volume_LRU.q[idx].len * sizeof(Volume *));
7666 now = FT_ApproxTime();
7667 thresh = volume_LRU.promotion_interval[idx-1];
7670 for (queue_ScanBackwards(&volume_LRU.q[idx], qp, nqp, rx_queue)) {
7671 vp = (Volume *)((char *)qp - offsetof(Volume, vlru));
7672 demote = (((vp->stats.last_promote + thresh) <= now) &&
7673 (vp->stats.last_get < (now - thresh)));
7675 /* we now do volume update list DONT_SALVAGE flag setting during
7676 * demotion passes */
7677 if (salv_flag_vec &&
7678 !(V_attachFlags(vp) & VOL_HDR_DONTSALV) &&
7680 (vp->updateTime < (now - SALVAGE_INTERVAL)) &&
7681 (V_attachState(vp) == VOL_STATE_ATTACHED)) {
7682 salv_flag_vec[salv_vec_offset++] = vp;
7683 VCreateReservation_r(vp);
7692 /* demote and append chain */
7693 queue_MoveChainBefore(&volume_LRU.q[idx-1], &start->vlru, &end->vlru);
7707 queue_MoveChainBefore(&volume_LRU.q[idx-1], &start->vlru, &end->vlru);
7711 volume_LRU.q[idx].len -= len;
7712 volume_LRU.q[idx-1].len += len;
7715 /* release exclusive access to the two chains */
7717 VLRU_EndExclusive_r(&volume_LRU.q[idx]);
7718 VLRU_EndExclusive_r(&volume_LRU.q[idx-1]);
7720 /* now go back and set the DONT_SALVAGE flags as appropriate */
7721 if (salv_flag_vec) {
7723 for (i = 0; i < salv_vec_offset; i++) {
7724 vp = salv_flag_vec[i];
7725 if (!(V_attachFlags(vp) & VOL_HDR_DONTSALV) &&
7726 (vp->updateTime < (now - SALVAGE_INTERVAL)) &&
7727 (V_attachState(vp) == VOL_STATE_ATTACHED)) {
7730 V_attachFlags(vp) |= VOL_HDR_DONTSALV;
7731 V_dontSalvage(vp) = DONT_SALVAGE;
7732 VUpdateVolume_r(&ec, vp, 0);
7736 VCancelReservation_r(vp);
7738 free(salv_flag_vec);
7742 /* run a pass of the VLRU GC scanner */
7744 VLRU_Scan_r(int idx)
7746 afs_uint32 now, thresh;
7747 struct rx_queue *qp, *nqp;
7751 opr_Assert(idx == VLRU_QUEUE_NEW || idx == VLRU_QUEUE_CANDIDATE);
7753 /* gain exclusive access to the idx VLRU */
7754 VLRU_Wait_r(&volume_LRU.q[idx]);
7755 VLRU_BeginExclusive_r(&volume_LRU.q[idx]);
7757 if (idx != VLRU_QUEUE_CANDIDATE) {
7758 /* gain exclusive access to the candidate VLRU */
7759 VLRU_Wait_r(&volume_LRU.q[VLRU_QUEUE_CANDIDATE]);
7760 VLRU_BeginExclusive_r(&volume_LRU.q[VLRU_QUEUE_CANDIDATE]);
7763 now = FT_ApproxTime();
7764 thresh = now - VLRU_offline_thresh;
7766 /* perform candidate selection and soft detaching */
7767 if (idx == VLRU_QUEUE_CANDIDATE) {
7768 /* soft detach some volumes from the candidate pool */
7772 for (i=0,queue_ScanBackwards(&volume_LRU.q[idx], qp, nqp, rx_queue)) {
7773 vp = (Volume *)((char *)qp - offsetof(Volume, vlru));
7774 if (i >= VLRU_offline_max) {
7777 /* check timestamp to see if it's a candidate for soft detaching */
7778 if (vp->stats.last_get <= thresh) {
7780 if (VCheckSoftDetach(vp, thresh))
7786 /* scan for volumes to become soft detach candidates */
7787 for (i=1,queue_ScanBackwards(&volume_LRU.q[idx], qp, nqp, rx_queue),i++) {
7788 vp = (Volume *)((char *)qp - offsetof(Volume, vlru));
7790 /* check timestamp to see if it's a candidate for soft detaching */
7791 if (vp->stats.last_get <= thresh) {
7792 VCheckSoftDetachCandidate(vp, thresh);
7795 if (!(i&0x7f)) { /* lock coarsening optimization */
7803 /* relinquish exclusive access to the VLRU chains */
7807 volume_LRU.last_scan[idx] = now;
7808 if (idx != VLRU_QUEUE_CANDIDATE) {
7809 VLRU_EndExclusive_r(&volume_LRU.q[VLRU_QUEUE_CANDIDATE]);
7811 VLRU_EndExclusive_r(&volume_LRU.q[idx]);
7814 /* check whether volume is safe to soft detach
7815 * caller MUST NOT hold a ref count on vp */
7817 VCheckSoftDetach(Volume * vp, afs_uint32 thresh)
7821 if (vp->nUsers || vp->nWaiters)
7824 if (vp->stats.last_get <= thresh) {
7825 ret = VSoftDetachVolume_r(vp, thresh);
7831 /* check whether volume should be made a
7832 * soft detach candidate */
7834 VCheckSoftDetachCandidate(Volume * vp, afs_uint32 thresh)
7837 if (vp->nUsers || vp->nWaiters)
7842 opr_Assert(idx == VLRU_QUEUE_NEW);
7844 if (vp->stats.last_get <= thresh) {
7845 /* move to candidate pool */
7846 queue_Remove(&vp->vlru);
7847 volume_LRU.q[VLRU_QUEUE_NEW].len--;
7848 queue_Prepend(&volume_LRU.q[VLRU_QUEUE_CANDIDATE], &vp->vlru);
7849 vp->vlru.idx = VLRU_QUEUE_CANDIDATE;
7850 volume_LRU.q[VLRU_QUEUE_CANDIDATE].len++;
7858 /* begin exclusive access on VLRU */
7860 VLRU_BeginExclusive_r(struct VLRU_q * q)
7862 opr_Assert(q->busy == 0);
7866 /* end exclusive access on VLRU */
7868 VLRU_EndExclusive_r(struct VLRU_q * q)
7870 opr_Assert(q->busy);
7872 opr_cv_broadcast(&q->cv);
7875 /* wait for another thread to end exclusive access on VLRU */
7877 VLRU_Wait_r(struct VLRU_q * q)
7880 VOL_CV_WAIT(&q->cv);
7885 * volume soft detach
7887 * caller MUST NOT hold a ref count on vp */
7889 VSoftDetachVolume_r(Volume * vp, afs_uint32 thresh)
7894 opr_Assert(vp->vlru.idx == VLRU_QUEUE_CANDIDATE);
7896 ts_save = vp->stats.last_get;
7897 if (ts_save > thresh)
7900 if (vp->nUsers || vp->nWaiters)
7903 if (VIsExclusiveState(V_attachState(vp))) {
7907 switch (V_attachState(vp)) {
7908 case VOL_STATE_UNATTACHED:
7909 case VOL_STATE_PREATTACHED:
7910 case VOL_STATE_ERROR:
7911 case VOL_STATE_GOING_OFFLINE:
7912 case VOL_STATE_SHUTTING_DOWN:
7913 case VOL_STATE_SALVAGING:
7914 case VOL_STATE_DELETED:
7915 volume_LRU.q[vp->vlru.idx].len--;
7917 /* create and cancel a reservation to
7918 * give the volume an opportunity to
7920 VCreateReservation_r(vp);
7921 queue_Remove(&vp->vlru);
7922 vp->vlru.idx = VLRU_QUEUE_INVALID;
7923 V_attachFlags(vp) &= ~(VOL_ON_VLRU);
7924 VCancelReservation_r(vp);
7930 /* hold the volume and take it offline.
7931 * no need for reservations, as VHold_r
7932 * takes care of that internally. */
7933 if (VHold_r(vp) == 0) {
7934 /* vhold drops the glock, so now we should
7935 * check to make sure we aren't racing against
7936 * other threads. if we are racing, offlining vp
7937 * would be wasteful, and block the scanner for a while
7941 (vp->shuttingDown) ||
7942 (vp->goingOffline) ||
7943 (vp->stats.last_get != ts_save)) {
7944 /* looks like we're racing someone else. bail */
7948 /* pull it off the VLRU */
7949 opr_Assert(vp->vlru.idx == VLRU_QUEUE_CANDIDATE);
7950 volume_LRU.q[VLRU_QUEUE_CANDIDATE].len--;
7951 queue_Remove(&vp->vlru);
7952 vp->vlru.idx = VLRU_QUEUE_INVALID;
7953 V_attachFlags(vp) &= ~(VOL_ON_VLRU);
7955 /* take if offline */
7956 VOffline_r(vp, "volume has been soft detached");
7958 /* invalidate the volume header cache */
7959 FreeVolumeHeader(vp);
7962 IncUInt64(&VStats.soft_detaches);
7963 vp->stats.soft_detaches++;
7965 /* put in pre-attached state so demand
7966 * attacher can work on it */
7967 VChangeState_r(vp, VOL_STATE_PREATTACHED);
7973 #endif /* AFS_DEMAND_ATTACH_FS */
7976 /***************************************************/
7977 /* Volume Header Cache routines */
7978 /***************************************************/
7981 * volume header cache.
7983 struct volume_hdr_LRU_t volume_hdr_LRU;
7986 * initialize the volume header cache.
7988 * @param[in] howMany number of header cache entries to preallocate
7990 * @pre VOL_LOCK held. Function has never been called before.
7992 * @post howMany cache entries are allocated, initialized, and added
7993 * to the LRU list. Header cache statistics are initialized.
7995 * @note only applicable to fileServer program type. Should only be
7996 * called once during volume package initialization.
7998 * @internal volume package internal use only.
8001 VInitVolumeHeaderCache(afs_uint32 howMany)
8003 struct volHeader *hp;
8004 if (programType != fileServer)
8006 queue_Init(&volume_hdr_LRU);
8007 volume_hdr_LRU.stats.free = 0;
8008 volume_hdr_LRU.stats.used = howMany;
8009 volume_hdr_LRU.stats.attached = 0;
8010 hp = (struct volHeader *)(calloc(howMany, sizeof(struct volHeader)));
8011 opr_Assert(hp != NULL);
8014 /* We are using ReleaseVolumeHeader to initialize the values on the header list
8015 * to ensure they have the right values
8017 ReleaseVolumeHeader(hp++);
8020 /* get a volume header off of the volume header LRU.
8022 * @return volume header
8023 * @retval NULL no usable volume header is available on the LRU
8025 * @pre VOL_LOCK held
8027 * @post for DAFS, if the returned header is associated with a volume, that
8028 * volume is NOT in an exclusive state
8030 * @internal volume package internal use only.
8032 #ifdef AFS_DEMAND_ATTACH_FS
8033 static struct volHeader*
8034 GetVolHeaderFromLRU(void)
8036 struct volHeader *hd = NULL, *qh, *nqh;
8037 /* Usually, a volume in an exclusive state will not have its header on
8038 * the LRU. However, it is possible for this to occur when a salvage
8039 * request is received over FSSYNC, and possibly in other corner cases.
8040 * So just skip over headers whose volumes are in an exclusive state. We
8041 * could VWaitExclusiveState_r instead, but not waiting is faster and
8043 for (queue_Scan(&volume_hdr_LRU, qh, nqh, volHeader)) {
8044 if (!qh->back || !VIsExclusiveState(V_attachState(qh->back))) {
8052 #else /* AFS_DEMAND_ATTACH_FS */
8053 static struct volHeader*
8054 GetVolHeaderFromLRU(void)
8056 struct volHeader *hd = NULL;
8057 if (queue_IsNotEmpty(&volume_hdr_LRU)) {
8058 hd = queue_First(&volume_hdr_LRU, volHeader);
8063 #endif /* !AFS_DEMAND_ATTACH_FS */
8066 * get a volume header and attach it to the volume object.
8068 * @param[in] vp pointer to volume object
8070 * @return cache entry status
8071 * @retval 0 volume header was newly attached; cache data is invalid
8072 * @retval 1 volume header was previously attached; cache data is valid
8074 * @pre VOL_LOCK held. For DAFS, lightweight ref must be held on volume object.
8076 * @post volume header attached to volume object. if necessary, header cache
8077 * entry on LRU is synchronized to disk. Header is removed from LRU list.
8079 * @note VOL_LOCK may be dropped
8081 * @warning this interface does not load header data from disk. it merely
8082 * attaches a header object to the volume object, and may sync the old
8083 * header cache data out to disk in the process.
8085 * @internal volume package internal use only.
8088 GetVolumeHeader(Volume * vp)
8091 struct volHeader *hd;
8093 static int everLogged = 0;
8095 #ifdef AFS_DEMAND_ATTACH_FS
8096 VolState vp_save = 0, back_save = 0;
8098 /* XXX debug 9/19/05 we've apparently got
8099 * a ref counting bug somewhere that's
8100 * breaking the nUsers == 0 => header on LRU
8102 if (vp->header && queue_IsNotOnQueue(vp->header)) {
8103 Log("nUsers == 0, but header not on LRU\n");
8108 old = (vp->header != NULL); /* old == volume already has a header */
8110 if (programType != fileServer) {
8111 /* for volume utilities, we allocate volHeaders as needed */
8113 hd = calloc(1, sizeof(*vp->header));
8114 opr_Assert(hd != NULL);
8117 #ifdef AFS_DEMAND_ATTACH_FS
8118 V_attachFlags(vp) |= VOL_HDR_ATTACHED;
8122 /* for the fileserver, we keep a volume header cache */
8124 /* the header we previously dropped in the lru is
8125 * still available. pull it off the lru and return */
8128 opr_Assert(hd->back == vp);
8129 #ifdef AFS_DEMAND_ATTACH_FS
8130 V_attachFlags(vp) &= ~(VOL_HDR_IN_LRU);
8133 hd = GetVolHeaderFromLRU();
8135 /* LRU is empty, so allocate a new volHeader
8136 * this is probably indicative of a leak, so let the user know */
8137 hd = calloc(1, sizeof(struct volHeader));
8138 opr_Assert(hd != NULL);
8140 Log("****Allocated more volume headers, probably leak****\n");
8143 volume_hdr_LRU.stats.free++;
8146 /* this header used to belong to someone else.
8147 * we'll need to check if the header needs to
8148 * be sync'd out to disk */
8150 #ifdef AFS_DEMAND_ATTACH_FS
8151 /* GetVolHeaderFromLRU had better not give us back a header
8152 * with a volume in exclusive state... */
8153 opr_Assert(!VIsExclusiveState(V_attachState(hd->back)));
8156 if (hd->diskstuff.inUse) {
8157 /* volume was in use, so we'll need to sync
8158 * its header to disk */
8160 #ifdef AFS_DEMAND_ATTACH_FS
8161 back_save = VChangeState_r(hd->back, VOL_STATE_UPDATING);
8162 vp_save = VChangeState_r(vp, VOL_STATE_HDR_ATTACHING);
8163 VCreateReservation_r(hd->back);
8167 WriteVolumeHeader_r(&error, hd->back);
8168 /* Ignore errors; catch them later */
8170 #ifdef AFS_DEMAND_ATTACH_FS
8175 hd->back->header = NULL;
8176 #ifdef AFS_DEMAND_ATTACH_FS
8177 V_attachFlags(hd->back) &= ~(VOL_HDR_ATTACHED | VOL_HDR_LOADED | VOL_HDR_IN_LRU);
8179 if (hd->diskstuff.inUse) {
8180 VChangeState_r(hd->back, back_save);
8181 VCancelReservation_r(hd->back);
8182 VChangeState_r(vp, vp_save);
8186 volume_hdr_LRU.stats.attached++;
8190 #ifdef AFS_DEMAND_ATTACH_FS
8191 V_attachFlags(vp) |= VOL_HDR_ATTACHED;
8194 volume_hdr_LRU.stats.free--;
8195 volume_hdr_LRU.stats.used++;
8197 IncUInt64(&VStats.hdr_gets);
8198 #ifdef AFS_DEMAND_ATTACH_FS
8199 IncUInt64(&vp->stats.hdr_gets);
8200 vp->stats.last_hdr_get = FT_ApproxTime();
8207 * make sure volume header is attached and contains valid cache data.
8209 * @param[out] ec outbound error code
8210 * @param[in] vp pointer to volume object
8212 * @pre VOL_LOCK held. For DAFS, lightweight ref held on vp.
8214 * @post header cache entry attached, and loaded with valid data, or
8215 * *ec is nonzero, and the header is released back into the LRU.
8217 * @internal volume package internal use only.
8220 LoadVolumeHeader(Error * ec, Volume * vp)
8222 #ifdef AFS_DEMAND_ATTACH_FS
8223 VolState state_save;
8227 if (vp->nUsers == 0 && !GetVolumeHeader(vp)) {
8228 IncUInt64(&VStats.hdr_loads);
8229 state_save = VChangeState_r(vp, VOL_STATE_HDR_LOADING);
8232 ReadHeader(ec, V_diskDataHandle(vp), (char *)&V_disk(vp),
8233 sizeof(V_disk(vp)), VOLUMEINFOMAGIC,
8235 IncUInt64(&vp->stats.hdr_loads);
8236 now = FT_ApproxTime();
8240 V_attachFlags(vp) |= VOL_HDR_LOADED;
8241 vp->stats.last_hdr_load = now;
8243 VChangeState_r(vp, state_save);
8245 #else /* AFS_DEMAND_ATTACH_FS */
8247 if (vp->nUsers == 0 && !GetVolumeHeader(vp)) {
8248 IncUInt64(&VStats.hdr_loads);
8250 ReadHeader(ec, V_diskDataHandle(vp), (char *)&V_disk(vp),
8251 sizeof(V_disk(vp)), VOLUMEINFOMAGIC,
8254 #endif /* AFS_DEMAND_ATTACH_FS */
8256 /* maintain (nUsers==0) => header in LRU invariant */
8257 FreeVolumeHeader(vp);
8262 * release a header cache entry back into the LRU list.
8264 * @param[in] hd pointer to volume header cache object
8266 * @pre VOL_LOCK held.
8268 * @post header cache object appended onto end of LRU list.
8270 * @note only applicable to fileServer program type.
8272 * @note used to place a header cache entry back into the
8273 * LRU pool without invalidating it as a cache entry.
8275 * @internal volume package internal use only.
8278 ReleaseVolumeHeader(struct volHeader *hd)
8280 if (programType != fileServer)
8282 if (!hd || queue_IsOnQueue(hd)) /* no header, or header already released */
8284 queue_Append(&volume_hdr_LRU, hd);
8285 #ifdef AFS_DEMAND_ATTACH_FS
8287 V_attachFlags(hd->back) |= VOL_HDR_IN_LRU;
8290 volume_hdr_LRU.stats.free++;
8291 volume_hdr_LRU.stats.used--;
8295 * free/invalidate a volume header cache entry.
8297 * @param[in] vp pointer to volume object
8299 * @pre VOL_LOCK is held.
8301 * @post For fileserver, header cache entry is returned to LRU, and it is
8302 * invalidated as a cache entry. For volume utilities, the header
8303 * cache entry is freed.
8305 * @note For fileserver, this should be utilized instead of ReleaseVolumeHeader
8306 * whenever it is necessary to invalidate the header cache entry.
8308 * @see ReleaseVolumeHeader
8310 * @internal volume package internal use only.
8313 FreeVolumeHeader(Volume * vp)
8315 struct volHeader *hd = vp->header;
8318 if (programType == fileServer) {
8319 ReleaseVolumeHeader(hd);
8324 #ifdef AFS_DEMAND_ATTACH_FS
8325 V_attachFlags(vp) &= ~(VOL_HDR_ATTACHED | VOL_HDR_IN_LRU | VOL_HDR_LOADED);
8327 volume_hdr_LRU.stats.attached--;
8332 /***************************************************/
8333 /* Volume Hash Table routines */
8334 /***************************************************/
8337 * set size of volume object hash table.
8339 * @param[in] logsize log(2) of desired hash table size
8341 * @return operation status
8343 * @retval -1 failure
8345 * @pre MUST be called prior to VInitVolumePackage2
8347 * @post Volume Hash Table will have 2^logsize buckets
8350 VSetVolHashSize(int logsize)
8352 /* 64 to 268435456 hash buckets seems like a reasonable range */
8353 if ((logsize < 6 ) || (logsize > 28)) {
8358 VolumeHashTable.Size = 1 << logsize;
8359 VolumeHashTable.Mask = VolumeHashTable.Size - 1;
8361 /* we can't yet support runtime modification of this
8362 * parameter. we'll need a configuration rwlock to
8363 * make runtime modification feasible.... */
8370 * initialize dynamic data structures for volume hash table.
8372 * @post hash table is allocated, and fields are initialized.
8374 * @internal volume package internal use only.
8377 VInitVolumeHash(void)
8381 VolumeHashTable.Table = (VolumeHashChainHead *) calloc(VolumeHashTable.Size,
8382 sizeof(VolumeHashChainHead));
8383 opr_Assert(VolumeHashTable.Table != NULL);
8385 for (i=0; i < VolumeHashTable.Size; i++) {
8386 queue_Init(&VolumeHashTable.Table[i]);
8387 #ifdef AFS_DEMAND_ATTACH_FS
8388 opr_cv_init(&VolumeHashTable.Table[i].chain_busy_cv);
8389 #endif /* AFS_DEMAND_ATTACH_FS */
8394 * add a volume object to the hash table.
8396 * @param[in] vp pointer to volume object
8397 * @param[in] hashid hash of volume id
8399 * @pre VOL_LOCK is held. For DAFS, caller must hold a lightweight
8402 * @post volume is added to hash chain.
8404 * @internal volume package internal use only.
8406 * @note For DAFS, VOL_LOCK may be dropped in order to wait for an
8407 * asynchronous hash chain reordering to finish.
8410 AddVolumeToHashTable(Volume * vp, VolumeId hashid)
8412 VolumeHashChainHead * head;
8414 if (queue_IsOnQueue(vp))
8417 head = &VolumeHashTable.Table[VOLUME_HASH(hashid)];
8419 #ifdef AFS_DEMAND_ATTACH_FS
8420 /* wait for the hash chain to become available */
8423 V_attachFlags(vp) |= VOL_IN_HASH;
8424 vp->chainCacheCheck = ++head->cacheCheck;
8425 #endif /* AFS_DEMAND_ATTACH_FS */
8428 vp->hashid = hashid;
8429 queue_Append(head, vp);
8430 vp->vnodeHashOffset = VolumeHashOffset_r();
8434 * delete a volume object from the hash table.
8436 * @param[in] vp pointer to volume object
8438 * @pre VOL_LOCK is held. For DAFS, caller must hold a lightweight
8441 * @post volume is removed from hash chain.
8443 * @internal volume package internal use only.
8445 * @note For DAFS, VOL_LOCK may be dropped in order to wait for an
8446 * asynchronous hash chain reordering to finish.
8449 DeleteVolumeFromHashTable(Volume * vp)
8451 VolumeHashChainHead * head;
8453 if (!queue_IsOnQueue(vp))
8456 head = &VolumeHashTable.Table[VOLUME_HASH(vp->hashid)];
8458 #ifdef AFS_DEMAND_ATTACH_FS
8459 /* wait for the hash chain to become available */
8462 V_attachFlags(vp) &= ~(VOL_IN_HASH);
8464 #endif /* AFS_DEMAND_ATTACH_FS */
8468 /* do NOT reset hashid to zero, as the online
8469 * salvager package may need to know the volume id
8470 * after the volume is removed from the hash */
8474 * lookup a volume object in the hash table given a volume id.
8476 * @param[out] ec error code return
8477 * @param[in] volumeId volume id
8478 * @param[in] hint volume object which we believe could be the correct
8481 * @return volume object pointer
8482 * @retval NULL no such volume id is registered with the hash table.
8484 * @pre VOL_LOCK is held. For DAFS, caller must hold a lightweight
8487 * @post volume object with the given id is returned. volume object and
8488 * hash chain access statistics are updated. hash chain may have
8491 * @note For DAFS, VOL_LOCK may be dropped in order to wait for an
8492 * asynchronous hash chain reordering operation to finish, or
8493 * in order for us to perform an asynchronous chain reordering.
8495 * @note Hash chain reorderings occur when the access count for the
8496 * volume object being looked up exceeds the sum of the previous
8497 * node's (the node ahead of it in the hash chain linked list)
8498 * access count plus the constant VOLUME_HASH_REORDER_THRESHOLD.
8500 * @note For DAFS, the hint parameter allows us to short-circuit if the
8501 * cacheCheck fields match between the hash chain head and the
8502 * hint volume object.
8505 VLookupVolume_r(Error * ec, VolumeId volumeId, Volume * hint)
8509 #ifdef AFS_DEMAND_ATTACH_FS
8512 VolumeHashChainHead * head;
8515 head = &VolumeHashTable.Table[VOLUME_HASH(volumeId)];
8517 #ifdef AFS_DEMAND_ATTACH_FS
8518 /* wait for the hash chain to become available */
8521 /* check to see if we can short circuit without walking the hash chain */
8522 if (hint && (hint->chainCacheCheck == head->cacheCheck)) {
8523 IncUInt64(&hint->stats.hash_short_circuits);
8526 #endif /* AFS_DEMAND_ATTACH_FS */
8528 /* someday we need to either do per-chain locks, RWlocks,
8529 * or both for volhash access.
8530 * (and move to a data structure with better cache locality) */
8532 /* search the chain for this volume id */
8533 for(queue_Scan(head, vp, np, Volume)) {
8535 if (vp->hashid == volumeId) {
8540 if (queue_IsEnd(head, vp)) {
8544 #ifdef AFS_DEMAND_ATTACH_FS
8545 /* update hash chain statistics */
8548 FillInt64(lks, 0, looks);
8549 AddUInt64(head->looks, lks, &head->looks);
8550 AddUInt64(VStats.hash_looks, lks, &VStats.hash_looks);
8551 IncUInt64(&head->gets);
8556 IncUInt64(&vp->stats.hash_lookups);
8558 /* for demand attach fileserver, we permit occasional hash chain reordering
8559 * so that frequently looked up volumes move towards the head of the chain */
8560 pp = queue_Prev(vp, Volume);
8561 if (!queue_IsEnd(head, pp)) {
8562 FillInt64(thresh, 0, VOLUME_HASH_REORDER_THRESHOLD);
8563 AddUInt64(thresh, pp->stats.hash_lookups, &thresh);
8564 if (GEInt64(vp->stats.hash_lookups, thresh)) {
8565 VReorderHash_r(head, pp, vp);
8569 /* update the short-circuit cache check */
8570 vp->chainCacheCheck = head->cacheCheck;
8572 #endif /* AFS_DEMAND_ATTACH_FS */
8577 #ifdef AFS_DEMAND_ATTACH_FS
8578 /* perform volume hash chain reordering.
8580 * advance a subchain beginning at vp ahead of
8581 * the adjacent subchain ending at pp */
8583 VReorderHash_r(VolumeHashChainHead * head, Volume * pp, Volume * vp)
8585 Volume *tp, *np, *lp;
8586 afs_uint64 move_thresh;
8588 /* this should never be called if the chain is already busy, so
8589 * no need to wait for other exclusive chain ops to finish */
8591 /* this is a rather heavy set of operations,
8592 * so let's set the chain busy flag and drop
8594 VHashBeginExclusive_r(head);
8597 /* scan forward in the chain from vp looking for the last element
8598 * in the chain we want to advance */
8599 FillInt64(move_thresh, 0, VOLUME_HASH_REORDER_CHAIN_THRESH);
8600 AddUInt64(move_thresh, pp->stats.hash_lookups, &move_thresh);
8601 for(queue_ScanFrom(head, vp, tp, np, Volume)) {
8602 if (LTInt64(tp->stats.hash_lookups, move_thresh)) {
8606 lp = queue_Prev(tp, Volume);
8608 /* scan backwards from pp to determine where to splice and
8609 * insert the subchain we're advancing */
8610 for(queue_ScanBackwardsFrom(head, pp, tp, np, Volume)) {
8611 if (GTInt64(tp->stats.hash_lookups, move_thresh)) {
8615 tp = queue_Next(tp, Volume);
8617 /* rebalance chain(vp,...,lp) ahead of chain(tp,...,pp) */
8618 queue_MoveChainBefore(tp,vp,lp);
8621 IncUInt64(&VStats.hash_reorders);
8623 IncUInt64(&head->reorders);
8625 /* wake up any threads waiting for the hash chain */
8626 VHashEndExclusive_r(head);
8630 /* demand-attach fs volume hash
8631 * asynchronous exclusive operations */
8634 * begin an asynchronous exclusive operation on a volume hash chain.
8636 * @param[in] head pointer to volume hash chain head object
8638 * @pre VOL_LOCK held. hash chain is quiescent.
8640 * @post hash chain marked busy.
8642 * @note this interface is used in conjunction with VHashEndExclusive_r and
8643 * VHashWait_r to perform asynchronous (wrt VOL_LOCK) operations on a
8644 * volume hash chain. Its main use case is hash chain reordering, which
8645 * has the potential to be a highly latent operation.
8647 * @see VHashEndExclusive_r
8652 * @internal volume package internal use only.
8655 VHashBeginExclusive_r(VolumeHashChainHead * head)
8657 opr_Assert(head->busy == 0);
8662 * relinquish exclusive ownership of a volume hash chain.
8664 * @param[in] head pointer to volume hash chain head object
8666 * @pre VOL_LOCK held. thread owns the hash chain exclusively.
8668 * @post hash chain is marked quiescent. threads awaiting use of
8669 * chain are awakened.
8671 * @see VHashBeginExclusive_r
8676 * @internal volume package internal use only.
8679 VHashEndExclusive_r(VolumeHashChainHead * head)
8681 opr_Assert(head->busy);
8683 opr_cv_broadcast(&head->chain_busy_cv);
8687 * wait for all asynchronous operations on a hash chain to complete.
8689 * @param[in] head pointer to volume hash chain head object
8691 * @pre VOL_LOCK held.
8693 * @post hash chain object is quiescent.
8695 * @see VHashBeginExclusive_r
8696 * @see VHashEndExclusive_r
8700 * @note This interface should be called before any attempt to
8701 * traverse the hash chain. It is permissible for a thread
8702 * to gain exclusive access to the chain, and then perform
8703 * latent operations on the chain asynchronously wrt the
8706 * @warning if waiting is necessary, VOL_LOCK is dropped
8708 * @internal volume package internal use only.
8711 VHashWait_r(VolumeHashChainHead * head)
8713 while (head->busy) {
8714 VOL_CV_WAIT(&head->chain_busy_cv);
8717 #endif /* AFS_DEMAND_ATTACH_FS */
8720 /***************************************************/
8721 /* Volume by Partition List routines */
8722 /***************************************************/
8725 * demand attach fileserver adds a
8726 * linked list of volumes to each
8727 * partition object, thus allowing
8728 * for quick enumeration of all
8729 * volumes on a partition
8732 #ifdef AFS_DEMAND_ATTACH_FS
8734 * add a volume to its disk partition VByPList.
8736 * @param[in] vp pointer to volume object
8738 * @pre either the disk partition VByPList is owned exclusively
8739 * by the calling thread, or the list is quiescent and
8742 * @post volume is added to disk partition VByPList
8746 * @warning it is the caller's responsibility to ensure list
8749 * @see VVByPListWait_r
8750 * @see VVByPListBeginExclusive_r
8751 * @see VVByPListEndExclusive_r
8753 * @internal volume package internal use only.
8756 AddVolumeToVByPList_r(Volume * vp)
8758 if (queue_IsNotOnQueue(&vp->vol_list)) {
8759 queue_Append(&vp->partition->vol_list, &vp->vol_list);
8760 V_attachFlags(vp) |= VOL_ON_VBYP_LIST;
8761 vp->partition->vol_list.len++;
8766 * delete a volume from its disk partition VByPList.
8768 * @param[in] vp pointer to volume object
8770 * @pre either the disk partition VByPList is owned exclusively
8771 * by the calling thread, or the list is quiescent and
8774 * @post volume is removed from the disk partition VByPList
8778 * @warning it is the caller's responsibility to ensure list
8781 * @see VVByPListWait_r
8782 * @see VVByPListBeginExclusive_r
8783 * @see VVByPListEndExclusive_r
8785 * @internal volume package internal use only.
8788 DeleteVolumeFromVByPList_r(Volume * vp)
8790 if (queue_IsOnQueue(&vp->vol_list)) {
8791 queue_Remove(&vp->vol_list);
8792 V_attachFlags(vp) &= ~(VOL_ON_VBYP_LIST);
8793 vp->partition->vol_list.len--;
8798 * begin an asynchronous exclusive operation on a VByPList.
8800 * @param[in] dp pointer to disk partition object
8802 * @pre VOL_LOCK held. VByPList is quiescent.
8804 * @post VByPList marked busy.
8806 * @note this interface is used in conjunction with VVByPListEndExclusive_r and
8807 * VVByPListWait_r to perform asynchronous (wrt VOL_LOCK) operations on a
8810 * @see VVByPListEndExclusive_r
8811 * @see VVByPListWait_r
8815 * @internal volume package internal use only.
8817 /* take exclusive control over the list */
8819 VVByPListBeginExclusive_r(struct DiskPartition64 * dp)
8821 opr_Assert(dp->vol_list.busy == 0);
8822 dp->vol_list.busy = 1;
8826 * relinquish exclusive ownership of a VByPList.
8828 * @param[in] dp pointer to disk partition object
8830 * @pre VOL_LOCK held. thread owns the VByPList exclusively.
8832 * @post VByPList is marked quiescent. threads awaiting use of
8833 * the list are awakened.
8835 * @see VVByPListBeginExclusive_r
8836 * @see VVByPListWait_r
8840 * @internal volume package internal use only.
8843 VVByPListEndExclusive_r(struct DiskPartition64 * dp)
8845 opr_Assert(dp->vol_list.busy);
8846 dp->vol_list.busy = 0;
8847 opr_cv_broadcast(&dp->vol_list.cv);
8851 * wait for all asynchronous operations on a VByPList to complete.
8853 * @param[in] dp pointer to disk partition object
8855 * @pre VOL_LOCK is held.
8857 * @post disk partition's VByP list is quiescent
8861 * @note This interface should be called before any attempt to
8862 * traverse the VByPList. It is permissible for a thread
8863 * to gain exclusive access to the list, and then perform
8864 * latent operations on the list asynchronously wrt the
8867 * @warning if waiting is necessary, VOL_LOCK is dropped
8869 * @see VVByPListEndExclusive_r
8870 * @see VVByPListBeginExclusive_r
8872 * @internal volume package internal use only.
8875 VVByPListWait_r(struct DiskPartition64 * dp)
8877 while (dp->vol_list.busy) {
8878 VOL_CV_WAIT(&dp->vol_list.cv);
8881 #endif /* AFS_DEMAND_ATTACH_FS */
8883 /***************************************************/
8884 /* Volume Cache Statistics routines */
8885 /***************************************************/
8888 VPrintCacheStats_r(void)
8890 struct VnodeClassInfo *vcp;
8891 vcp = &VnodeClassInfo[vLarge];
8892 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);
8893 vcp = &VnodeClassInfo[vSmall];
8894 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);
8895 Log("Volume header cache, %d entries, %"AFS_INT64_FMT" gets, "
8896 "%"AFS_INT64_FMT" replacements\n",
8897 VStats.hdr_cache_size, VStats.hdr_gets, VStats.hdr_loads);
8901 VPrintCacheStats(void)
8904 VPrintCacheStats_r();
8908 #ifdef AFS_DEMAND_ATTACH_FS
8910 UInt64ToDouble(afs_uint64 * x)
8912 static double c32 = 4.0 * 1.073741824 * 1000000000.0;
8914 SplitInt64(*x, h, l);
8915 return (((double)h) * c32) + ((double) l);
8919 DoubleToPrintable(double x, char * buf, int len)
8921 static double billion = 1000000000.0;
8924 y[0] = (afs_uint32) (x / (billion * billion));
8925 y[1] = (afs_uint32) ((x - (((double)y[0]) * billion * billion)) / billion);
8926 y[2] = (afs_uint32) (x - ((((double)y[0]) * billion * billion) + (((double)y[1]) * billion)));
8929 snprintf(buf, len, "%d%09d%09d", y[0], y[1], y[2]);
8931 snprintf(buf, len, "%d%09d", y[1], y[2]);
8933 snprintf(buf, len, "%d", y[2]);
8939 struct VLRUExtStatsEntry {
8943 struct VLRUExtStats {
8949 } queue_info[VLRU_QUEUE_INVALID];
8950 struct VLRUExtStatsEntry * vec;
8954 * add a 256-entry fudge factor onto the vector in case state changes
8955 * out from under us.
8957 #define VLRU_EXT_STATS_VEC_LEN_FUDGE 256
8960 * collect extended statistics for the VLRU subsystem.
8962 * @param[out] stats pointer to stats structure to be populated
8963 * @param[in] nvols number of volumes currently known to exist
8965 * @pre VOL_LOCK held
8967 * @post stats->vec allocated and populated
8969 * @return operation status
8974 VVLRUExtStats_r(struct VLRUExtStats * stats, afs_uint32 nvols)
8976 afs_uint32 cur, idx, len;
8977 struct rx_queue * qp, * nqp;
8979 struct VLRUExtStatsEntry * vec;
8981 len = nvols + VLRU_EXT_STATS_VEC_LEN_FUDGE;
8982 vec = stats->vec = calloc(len,
8983 sizeof(struct VLRUExtStatsEntry));
8989 for (idx = VLRU_QUEUE_NEW; idx < VLRU_QUEUE_INVALID; idx++) {
8990 VLRU_Wait_r(&volume_LRU.q[idx]);
8991 VLRU_BeginExclusive_r(&volume_LRU.q[idx]);
8994 stats->queue_info[idx].start = cur;
8996 for (queue_Scan(&volume_LRU.q[idx], qp, nqp, rx_queue)) {
8998 /* out of space in vec */
9001 vp = (Volume *)((char *)qp - offsetof(Volume, vlru));
9002 vec[cur].volid = vp->hashid;
9006 stats->queue_info[idx].len = cur - stats->queue_info[idx].start;
9009 VLRU_EndExclusive_r(&volume_LRU.q[idx]);
9017 #define ENUMTOSTRING(en) #en
9018 #define ENUMCASE(en) \
9019 case en: return ENUMTOSTRING(en)
9022 vlru_idx_to_string(int idx)
9025 ENUMCASE(VLRU_QUEUE_NEW);
9026 ENUMCASE(VLRU_QUEUE_MID);
9027 ENUMCASE(VLRU_QUEUE_OLD);
9028 ENUMCASE(VLRU_QUEUE_CANDIDATE);
9029 ENUMCASE(VLRU_QUEUE_HELD);
9030 ENUMCASE(VLRU_QUEUE_INVALID);
9032 return "**UNKNOWN**";
9037 VPrintExtendedCacheStats_r(int flags)
9040 afs_uint32 vol_sum = 0;
9047 struct stats looks, gets, reorders, len;
9048 struct stats ch_looks, ch_gets, ch_reorders;
9050 VolumeHashChainHead *head;
9052 struct VLRUExtStats vlru_stats;
9054 /* zero out stats */
9055 memset(&looks, 0, sizeof(struct stats));
9056 memset(&gets, 0, sizeof(struct stats));
9057 memset(&reorders, 0, sizeof(struct stats));
9058 memset(&len, 0, sizeof(struct stats));
9059 memset(&ch_looks, 0, sizeof(struct stats));
9060 memset(&ch_gets, 0, sizeof(struct stats));
9061 memset(&ch_reorders, 0, sizeof(struct stats));
9063 for (i = 0; i < VolumeHashTable.Size; i++) {
9064 head = &VolumeHashTable.Table[i];
9067 VHashBeginExclusive_r(head);
9070 ch_looks.sum = UInt64ToDouble(&head->looks);
9071 ch_gets.sum = UInt64ToDouble(&head->gets);
9072 ch_reorders.sum = UInt64ToDouble(&head->reorders);
9074 /* update global statistics */
9076 looks.sum += ch_looks.sum;
9077 gets.sum += ch_gets.sum;
9078 reorders.sum += ch_reorders.sum;
9079 len.sum += (double)head->len;
9080 vol_sum += head->len;
9083 len.min = (double) head->len;
9084 len.max = (double) head->len;
9085 looks.min = ch_looks.sum;
9086 looks.max = ch_looks.sum;
9087 gets.min = ch_gets.sum;
9088 gets.max = ch_gets.sum;
9089 reorders.min = ch_reorders.sum;
9090 reorders.max = ch_reorders.sum;
9092 if (((double)head->len) < len.min)
9093 len.min = (double) head->len;
9094 if (((double)head->len) > len.max)
9095 len.max = (double) head->len;
9096 if (ch_looks.sum < looks.min)
9097 looks.min = ch_looks.sum;
9098 else if (ch_looks.sum > looks.max)
9099 looks.max = ch_looks.sum;
9100 if (ch_gets.sum < gets.min)
9101 gets.min = ch_gets.sum;
9102 else if (ch_gets.sum > gets.max)
9103 gets.max = ch_gets.sum;
9104 if (ch_reorders.sum < reorders.min)
9105 reorders.min = ch_reorders.sum;
9106 else if (ch_reorders.sum > reorders.max)
9107 reorders.max = ch_reorders.sum;
9111 if ((flags & VOL_STATS_PER_CHAIN2) && queue_IsNotEmpty(head)) {
9112 /* compute detailed per-chain stats */
9113 struct stats hdr_loads, hdr_gets;
9114 double v_looks, v_loads, v_gets;
9116 /* initialize stats with data from first element in chain */
9117 vp = queue_First(head, Volume);
9118 v_looks = UInt64ToDouble(&vp->stats.hash_lookups);
9119 v_loads = UInt64ToDouble(&vp->stats.hdr_loads);
9120 v_gets = UInt64ToDouble(&vp->stats.hdr_gets);
9121 ch_gets.min = ch_gets.max = v_looks;
9122 hdr_loads.min = hdr_loads.max = v_loads;
9123 hdr_gets.min = hdr_gets.max = v_gets;
9124 hdr_loads.sum = hdr_gets.sum = 0;
9126 vp = queue_Next(vp, Volume);
9128 /* pull in stats from remaining elements in chain */
9129 for (queue_ScanFrom(head, vp, vp, np, Volume)) {
9130 v_looks = UInt64ToDouble(&vp->stats.hash_lookups);
9131 v_loads = UInt64ToDouble(&vp->stats.hdr_loads);
9132 v_gets = UInt64ToDouble(&vp->stats.hdr_gets);
9134 hdr_loads.sum += v_loads;
9135 hdr_gets.sum += v_gets;
9137 if (v_looks < ch_gets.min)
9138 ch_gets.min = v_looks;
9139 else if (v_looks > ch_gets.max)
9140 ch_gets.max = v_looks;
9142 if (v_loads < hdr_loads.min)
9143 hdr_loads.min = v_loads;
9144 else if (v_loads > hdr_loads.max)
9145 hdr_loads.max = v_loads;
9147 if (v_gets < hdr_gets.min)
9148 hdr_gets.min = v_gets;
9149 else if (v_gets > hdr_gets.max)
9150 hdr_gets.max = v_gets;
9153 /* compute per-chain averages */
9154 ch_gets.avg = ch_gets.sum / ((double)head->len);
9155 hdr_loads.avg = hdr_loads.sum / ((double)head->len);
9156 hdr_gets.avg = hdr_gets.sum / ((double)head->len);
9158 /* dump per-chain stats */
9159 Log("Volume hash chain %d : len=%d, looks=%s, reorders=%s\n",
9161 DoubleToPrintable(ch_looks.sum, pr_buf[0], sizeof(pr_buf[0])),
9162 DoubleToPrintable(ch_reorders.sum, pr_buf[1], sizeof(pr_buf[1])));
9163 Log("\tVolume gets : min=%s, max=%s, avg=%s, total=%s\n",
9164 DoubleToPrintable(ch_gets.min, pr_buf[0], sizeof(pr_buf[0])),
9165 DoubleToPrintable(ch_gets.max, pr_buf[1], sizeof(pr_buf[1])),
9166 DoubleToPrintable(ch_gets.avg, pr_buf[2], sizeof(pr_buf[2])),
9167 DoubleToPrintable(ch_gets.sum, pr_buf[3], sizeof(pr_buf[3])));
9168 Log("\tHDR gets : min=%s, max=%s, avg=%s, total=%s\n",
9169 DoubleToPrintable(hdr_gets.min, pr_buf[0], sizeof(pr_buf[0])),
9170 DoubleToPrintable(hdr_gets.max, pr_buf[1], sizeof(pr_buf[1])),
9171 DoubleToPrintable(hdr_gets.avg, pr_buf[2], sizeof(pr_buf[2])),
9172 DoubleToPrintable(hdr_gets.sum, pr_buf[3], sizeof(pr_buf[3])));
9173 Log("\tHDR loads : min=%s, max=%s, avg=%s, total=%s\n",
9174 DoubleToPrintable(hdr_loads.min, pr_buf[0], sizeof(pr_buf[0])),
9175 DoubleToPrintable(hdr_loads.max, pr_buf[1], sizeof(pr_buf[1])),
9176 DoubleToPrintable(hdr_loads.avg, pr_buf[2], sizeof(pr_buf[2])),
9177 DoubleToPrintable(hdr_loads.sum, pr_buf[3], sizeof(pr_buf[3])));
9178 } else if (flags & VOL_STATS_PER_CHAIN) {
9179 /* dump simple per-chain stats */
9180 Log("Volume hash chain %d : len=%d, looks=%s, gets=%s, reorders=%s\n",
9182 DoubleToPrintable(ch_looks.sum, pr_buf[0], sizeof(pr_buf[0])),
9183 DoubleToPrintable(ch_gets.sum, pr_buf[1], sizeof(pr_buf[1])),
9184 DoubleToPrintable(ch_reorders.sum, pr_buf[2], sizeof(pr_buf[2])));
9188 VHashEndExclusive_r(head);
9193 /* compute global averages */
9194 len.avg = len.sum / ((double)VolumeHashTable.Size);
9195 looks.avg = looks.sum / ((double)VolumeHashTable.Size);
9196 gets.avg = gets.sum / ((double)VolumeHashTable.Size);
9197 reorders.avg = reorders.sum / ((double)VolumeHashTable.Size);
9199 /* dump global stats */
9200 Log("Volume hash summary: %d buckets\n", VolumeHashTable.Size);
9201 Log(" chain length : min=%s, max=%s, avg=%s, total=%s\n",
9202 DoubleToPrintable(len.min, pr_buf[0], sizeof(pr_buf[0])),
9203 DoubleToPrintable(len.max, pr_buf[1], sizeof(pr_buf[1])),
9204 DoubleToPrintable(len.avg, pr_buf[2], sizeof(pr_buf[2])),
9205 DoubleToPrintable(len.sum, pr_buf[3], sizeof(pr_buf[3])));
9206 Log(" looks : min=%s, max=%s, avg=%s, total=%s\n",
9207 DoubleToPrintable(looks.min, pr_buf[0], sizeof(pr_buf[0])),
9208 DoubleToPrintable(looks.max, pr_buf[1], sizeof(pr_buf[1])),
9209 DoubleToPrintable(looks.avg, pr_buf[2], sizeof(pr_buf[2])),
9210 DoubleToPrintable(looks.sum, pr_buf[3], sizeof(pr_buf[3])));
9211 Log(" gets : min=%s, max=%s, avg=%s, total=%s\n",
9212 DoubleToPrintable(gets.min, pr_buf[0], sizeof(pr_buf[0])),
9213 DoubleToPrintable(gets.max, pr_buf[1], sizeof(pr_buf[1])),
9214 DoubleToPrintable(gets.avg, pr_buf[2], sizeof(pr_buf[2])),
9215 DoubleToPrintable(gets.sum, pr_buf[3], sizeof(pr_buf[3])));
9216 Log(" reorders : min=%s, max=%s, avg=%s, total=%s\n",
9217 DoubleToPrintable(reorders.min, pr_buf[0], sizeof(pr_buf[0])),
9218 DoubleToPrintable(reorders.max, pr_buf[1], sizeof(pr_buf[1])),
9219 DoubleToPrintable(reorders.avg, pr_buf[2], sizeof(pr_buf[2])),
9220 DoubleToPrintable(reorders.sum, pr_buf[3], sizeof(pr_buf[3])));
9222 /* print extended disk related statistics */
9224 struct DiskPartition64 * diskP;
9225 afs_uint32 vol_count[VOLMAXPARTS+1];
9226 byte part_exists[VOLMAXPARTS+1];
9230 memset(vol_count, 0, sizeof(vol_count));
9231 memset(part_exists, 0, sizeof(part_exists));
9235 for (diskP = DiskPartitionList; diskP; diskP = diskP->next) {
9237 vol_count[id] = diskP->vol_list.len;
9238 part_exists[id] = 1;
9242 for (i = 0; i <= VOLMAXPARTS; i++) {
9243 if (part_exists[i]) {
9244 /* XXX while this is currently safe, it is a violation
9245 * of the VGetPartitionById_r interface contract. */
9246 diskP = VGetPartitionById_r(i, 0);
9248 Log("Partition %s has %d online volumes\n",
9249 VPartitionPath(diskP), diskP->vol_list.len);
9256 /* print extended VLRU statistics */
9257 if (VVLRUExtStats_r(&vlru_stats, vol_sum) == 0) {
9258 afs_uint32 idx, cur, lpos;
9263 Log("VLRU State Dump:\n\n");
9265 for (idx = VLRU_QUEUE_NEW; idx < VLRU_QUEUE_INVALID; idx++) {
9266 Log("\t%s:\n", vlru_idx_to_string(idx));
9269 for (cur = vlru_stats.queue_info[idx].start;
9270 cur < vlru_stats.queue_info[idx].len;
9272 line[lpos++] = vlru_stats.vec[cur].volid;
9274 Log("\t\t%u, %u, %u, %u, %u,\n",
9275 line[0], line[1], line[2], line[3], line[4]);
9284 Log("\t\t%u, %u, %u, %u, %u\n",
9285 line[0], line[1], line[2], line[3], line[4]);
9290 free(vlru_stats.vec);
9297 VPrintExtendedCacheStats(int flags)
9300 VPrintExtendedCacheStats_r(flags);
9303 #endif /* AFS_DEMAND_ATTACH_FS */
9306 VCanScheduleSalvage(void)
9308 return vol_opts.canScheduleSalvage;
9314 return vol_opts.canUseFSSYNC;
9318 VCanUseSALVSYNC(void)
9320 return vol_opts.canUseSALVSYNC;
9324 VCanUnsafeAttach(void)
9326 return vol_opts.unsafe_attach;